2 * linux/fs/ext4/inode.c
4 * Copyright (C) 1992, 1993, 1994, 1995
5 * Remy Card (card@masi.ibp.fr)
6 * Laboratoire MASI - Institut Blaise Pascal
7 * Universite Pierre et Marie Curie (Paris VI)
11 * linux/fs/minix/inode.c
13 * Copyright (C) 1991, 1992 Linus Torvalds
15 * 64-bit file support on 64-bit platforms by Jakub Jelinek
16 * (jj@sunsite.ms.mff.cuni.cz)
18 * Assorted race fixes, rewrite of ext4_get_block() by Al Viro, 2000
22 #include <linux/time.h>
23 #include <linux/highuid.h>
24 #include <linux/pagemap.h>
25 #include <linux/dax.h>
26 #include <linux/quotaops.h>
27 #include <linux/string.h>
28 #include <linux/buffer_head.h>
29 #include <linux/writeback.h>
30 #include <linux/pagevec.h>
31 #include <linux/mpage.h>
32 #include <linux/namei.h>
33 #include <linux/uio.h>
34 #include <linux/bio.h>
35 #include <linux/workqueue.h>
36 #include <linux/kernel.h>
37 #include <linux/printk.h>
38 #include <linux/slab.h>
39 #include <linux/bitops.h>
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
);
56 int offset
= offsetof(struct ext4_inode
, i_checksum_lo
);
57 unsigned int csum_size
= sizeof(dummy_csum
);
59 csum
= ext4_chksum(sbi
, ei
->i_csum_seed
, (__u8
*)raw
, offset
);
60 csum
= ext4_chksum(sbi
, csum
, (__u8
*)&dummy_csum
, csum_size
);
62 csum
= ext4_chksum(sbi
, csum
, (__u8
*)raw
+ offset
,
63 EXT4_GOOD_OLD_INODE_SIZE
- offset
);
65 if (EXT4_INODE_SIZE(inode
->i_sb
) > EXT4_GOOD_OLD_INODE_SIZE
) {
66 offset
= offsetof(struct ext4_inode
, i_checksum_hi
);
67 csum
= ext4_chksum(sbi
, csum
, (__u8
*)raw
+
68 EXT4_GOOD_OLD_INODE_SIZE
,
69 offset
- EXT4_GOOD_OLD_INODE_SIZE
);
70 if (EXT4_FITS_IN_INODE(raw
, ei
, i_checksum_hi
)) {
71 csum
= ext4_chksum(sbi
, csum
, (__u8
*)&dummy_csum
,
75 csum
= ext4_chksum(sbi
, csum
, (__u8
*)raw
+ offset
,
76 EXT4_INODE_SIZE(inode
->i_sb
) - offset
);
82 static int ext4_inode_csum_verify(struct inode
*inode
, struct ext4_inode
*raw
,
83 struct ext4_inode_info
*ei
)
85 __u32 provided
, calculated
;
87 if (EXT4_SB(inode
->i_sb
)->s_es
->s_creator_os
!=
88 cpu_to_le32(EXT4_OS_LINUX
) ||
89 !ext4_has_metadata_csum(inode
->i_sb
))
92 provided
= le16_to_cpu(raw
->i_checksum_lo
);
93 calculated
= ext4_inode_csum(inode
, raw
, ei
);
94 if (EXT4_INODE_SIZE(inode
->i_sb
) > EXT4_GOOD_OLD_INODE_SIZE
&&
95 EXT4_FITS_IN_INODE(raw
, ei
, i_checksum_hi
))
96 provided
|= ((__u32
)le16_to_cpu(raw
->i_checksum_hi
)) << 16;
100 return provided
== calculated
;
103 static void ext4_inode_csum_set(struct inode
*inode
, struct ext4_inode
*raw
,
104 struct ext4_inode_info
*ei
)
108 if (EXT4_SB(inode
->i_sb
)->s_es
->s_creator_os
!=
109 cpu_to_le32(EXT4_OS_LINUX
) ||
110 !ext4_has_metadata_csum(inode
->i_sb
))
113 csum
= ext4_inode_csum(inode
, raw
, ei
);
114 raw
->i_checksum_lo
= cpu_to_le16(csum
& 0xFFFF);
115 if (EXT4_INODE_SIZE(inode
->i_sb
) > EXT4_GOOD_OLD_INODE_SIZE
&&
116 EXT4_FITS_IN_INODE(raw
, ei
, i_checksum_hi
))
117 raw
->i_checksum_hi
= cpu_to_le16(csum
>> 16);
120 static inline int ext4_begin_ordered_truncate(struct inode
*inode
,
123 trace_ext4_begin_ordered_truncate(inode
, new_size
);
125 * If jinode is zero, then we never opened the file for
126 * writing, so there's no need to call
127 * jbd2_journal_begin_ordered_truncate() since there's no
128 * outstanding writes we need to flush.
130 if (!EXT4_I(inode
)->jinode
)
132 return jbd2_journal_begin_ordered_truncate(EXT4_JOURNAL(inode
),
133 EXT4_I(inode
)->jinode
,
137 static void ext4_invalidatepage(struct page
*page
, unsigned int offset
,
138 unsigned int length
);
139 static int __ext4_journalled_writepage(struct page
*page
, unsigned int len
);
140 static int ext4_bh_delay_or_unwritten(handle_t
*handle
, struct buffer_head
*bh
);
141 static int ext4_meta_trans_blocks(struct inode
*inode
, int lblocks
,
145 * Test whether an inode is a fast symlink.
147 int ext4_inode_is_fast_symlink(struct inode
*inode
)
149 int ea_blocks
= EXT4_I(inode
)->i_file_acl
?
150 EXT4_CLUSTER_SIZE(inode
->i_sb
) >> 9 : 0;
152 if (ext4_has_inline_data(inode
))
155 return (S_ISLNK(inode
->i_mode
) && inode
->i_blocks
- ea_blocks
== 0);
159 * Restart the transaction associated with *handle. This does a commit,
160 * so before we call here everything must be consistently dirtied against
163 int ext4_truncate_restart_trans(handle_t
*handle
, struct inode
*inode
,
169 * Drop i_data_sem to avoid deadlock with ext4_map_blocks. At this
170 * moment, get_block can be called only for blocks inside i_size since
171 * page cache has been already dropped and writes are blocked by
172 * i_mutex. So we can safely drop the i_data_sem here.
174 BUG_ON(EXT4_JOURNAL(inode
) == NULL
);
175 jbd_debug(2, "restarting handle %p\n", handle
);
176 up_write(&EXT4_I(inode
)->i_data_sem
);
177 ret
= ext4_journal_restart(handle
, nblocks
);
178 down_write(&EXT4_I(inode
)->i_data_sem
);
179 ext4_discard_preallocations(inode
);
185 * Called at the last iput() if i_nlink is zero.
187 void ext4_evict_inode(struct inode
*inode
)
192 trace_ext4_evict_inode(inode
);
194 if (inode
->i_nlink
) {
196 * When journalling data dirty buffers are tracked only in the
197 * journal. So although mm thinks everything is clean and
198 * ready for reaping the inode might still have some pages to
199 * write in the running transaction or waiting to be
200 * checkpointed. Thus calling jbd2_journal_invalidatepage()
201 * (via truncate_inode_pages()) to discard these buffers can
202 * cause data loss. Also even if we did not discard these
203 * buffers, we would have no way to find them after the inode
204 * is reaped and thus user could see stale data if he tries to
205 * read them before the transaction is checkpointed. So be
206 * careful and force everything to disk here... We use
207 * ei->i_datasync_tid to store the newest transaction
208 * containing inode's data.
210 * Note that directories do not have this problem because they
211 * don't use page cache.
213 if (inode
->i_ino
!= EXT4_JOURNAL_INO
&&
214 ext4_should_journal_data(inode
) &&
215 (S_ISLNK(inode
->i_mode
) || S_ISREG(inode
->i_mode
))) {
216 journal_t
*journal
= EXT4_SB(inode
->i_sb
)->s_journal
;
217 tid_t commit_tid
= EXT4_I(inode
)->i_datasync_tid
;
219 jbd2_complete_transaction(journal
, commit_tid
);
220 filemap_write_and_wait(&inode
->i_data
);
222 truncate_inode_pages_final(&inode
->i_data
);
224 WARN_ON(atomic_read(&EXT4_I(inode
)->i_ioend_count
));
228 if (is_bad_inode(inode
))
230 dquot_initialize(inode
);
232 if (ext4_should_order_data(inode
))
233 ext4_begin_ordered_truncate(inode
, 0);
234 truncate_inode_pages_final(&inode
->i_data
);
236 WARN_ON(atomic_read(&EXT4_I(inode
)->i_ioend_count
));
239 * Protect us against freezing - iput() caller didn't have to have any
240 * protection against it
242 sb_start_intwrite(inode
->i_sb
);
243 handle
= ext4_journal_start(inode
, EXT4_HT_TRUNCATE
,
244 ext4_blocks_for_truncate(inode
)+3);
245 if (IS_ERR(handle
)) {
246 ext4_std_error(inode
->i_sb
, PTR_ERR(handle
));
248 * If we're going to skip the normal cleanup, we still need to
249 * make sure that the in-core orphan linked list is properly
252 ext4_orphan_del(NULL
, inode
);
253 sb_end_intwrite(inode
->i_sb
);
258 ext4_handle_sync(handle
);
260 err
= ext4_mark_inode_dirty(handle
, inode
);
262 ext4_warning(inode
->i_sb
,
263 "couldn't mark inode dirty (err %d)", err
);
267 ext4_truncate(inode
);
270 * ext4_ext_truncate() doesn't reserve any slop when it
271 * restarts journal transactions; therefore there may not be
272 * enough credits left in the handle to remove the inode from
273 * the orphan list and set the dtime field.
275 if (!ext4_handle_has_enough_credits(handle
, 3)) {
276 err
= ext4_journal_extend(handle
, 3);
278 err
= ext4_journal_restart(handle
, 3);
280 ext4_warning(inode
->i_sb
,
281 "couldn't extend journal (err %d)", err
);
283 ext4_journal_stop(handle
);
284 ext4_orphan_del(NULL
, inode
);
285 sb_end_intwrite(inode
->i_sb
);
291 * Kill off the orphan record which ext4_truncate created.
292 * AKPM: I think this can be inside the above `if'.
293 * Note that ext4_orphan_del() has to be able to cope with the
294 * deletion of a non-existent orphan - this is because we don't
295 * know if ext4_truncate() actually created an orphan record.
296 * (Well, we could do this if we need to, but heck - it works)
298 ext4_orphan_del(handle
, inode
);
299 EXT4_I(inode
)->i_dtime
= get_seconds();
302 * One subtle ordering requirement: if anything has gone wrong
303 * (transaction abort, IO errors, whatever), then we can still
304 * do these next steps (the fs will already have been marked as
305 * having errors), but we can't free the inode if the mark_dirty
308 if (ext4_mark_inode_dirty(handle
, inode
))
309 /* If that failed, just do the required in-core inode clear. */
310 ext4_clear_inode(inode
);
312 ext4_free_inode(handle
, inode
);
313 ext4_journal_stop(handle
);
314 sb_end_intwrite(inode
->i_sb
);
317 ext4_clear_inode(inode
); /* We must guarantee clearing of inode... */
321 qsize_t
*ext4_get_reserved_space(struct inode
*inode
)
323 return &EXT4_I(inode
)->i_reserved_quota
;
328 * Called with i_data_sem down, which is important since we can call
329 * ext4_discard_preallocations() from here.
331 void ext4_da_update_reserve_space(struct inode
*inode
,
332 int used
, int quota_claim
)
334 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
335 struct ext4_inode_info
*ei
= EXT4_I(inode
);
337 spin_lock(&ei
->i_block_reservation_lock
);
338 trace_ext4_da_update_reserve_space(inode
, used
, quota_claim
);
339 if (unlikely(used
> ei
->i_reserved_data_blocks
)) {
340 ext4_warning(inode
->i_sb
, "%s: ino %lu, used %d "
341 "with only %d reserved data blocks",
342 __func__
, inode
->i_ino
, used
,
343 ei
->i_reserved_data_blocks
);
345 used
= ei
->i_reserved_data_blocks
;
348 /* Update per-inode reservations */
349 ei
->i_reserved_data_blocks
-= used
;
350 percpu_counter_sub(&sbi
->s_dirtyclusters_counter
, used
);
352 spin_unlock(&EXT4_I(inode
)->i_block_reservation_lock
);
354 /* Update quota subsystem for data blocks */
356 dquot_claim_block(inode
, EXT4_C2B(sbi
, used
));
359 * We did fallocate with an offset that is already delayed
360 * allocated. So on delayed allocated writeback we should
361 * not re-claim the quota for fallocated blocks.
363 dquot_release_reservation_block(inode
, EXT4_C2B(sbi
, used
));
367 * If we have done all the pending block allocations and if
368 * there aren't any writers on the inode, we can discard the
369 * inode's preallocations.
371 if ((ei
->i_reserved_data_blocks
== 0) &&
372 (atomic_read(&inode
->i_writecount
) == 0))
373 ext4_discard_preallocations(inode
);
376 static int __check_block_validity(struct inode
*inode
, const char *func
,
378 struct ext4_map_blocks
*map
)
380 if (!ext4_data_block_valid(EXT4_SB(inode
->i_sb
), map
->m_pblk
,
382 ext4_error_inode(inode
, func
, line
, map
->m_pblk
,
383 "lblock %lu mapped to illegal pblock %llu "
384 "(length %d)", (unsigned long) map
->m_lblk
,
385 map
->m_pblk
, map
->m_len
);
386 return -EFSCORRUPTED
;
391 #define check_block_validity(inode, map) \
392 __check_block_validity((inode), __func__, __LINE__, (map))
394 #ifdef ES_AGGRESSIVE_TEST
395 static void ext4_map_blocks_es_recheck(handle_t
*handle
,
397 struct ext4_map_blocks
*es_map
,
398 struct ext4_map_blocks
*map
,
405 * There is a race window that the result is not the same.
406 * e.g. xfstests #223 when dioread_nolock enables. The reason
407 * is that we lookup a block mapping in extent status tree with
408 * out taking i_data_sem. So at the time the unwritten extent
409 * could be converted.
411 if (!(flags
& EXT4_GET_BLOCKS_NO_LOCK
))
412 down_read(&EXT4_I(inode
)->i_data_sem
);
413 if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
)) {
414 retval
= ext4_ext_map_blocks(handle
, inode
, map
, flags
&
415 EXT4_GET_BLOCKS_KEEP_SIZE
);
417 retval
= ext4_ind_map_blocks(handle
, inode
, map
, flags
&
418 EXT4_GET_BLOCKS_KEEP_SIZE
);
420 if (!(flags
& EXT4_GET_BLOCKS_NO_LOCK
))
421 up_read((&EXT4_I(inode
)->i_data_sem
));
424 * We don't check m_len because extent will be collpased in status
425 * tree. So the m_len might not equal.
427 if (es_map
->m_lblk
!= map
->m_lblk
||
428 es_map
->m_flags
!= map
->m_flags
||
429 es_map
->m_pblk
!= map
->m_pblk
) {
430 printk("ES cache assertion failed for inode: %lu "
431 "es_cached ex [%d/%d/%llu/%x] != "
432 "found ex [%d/%d/%llu/%x] retval %d flags %x\n",
433 inode
->i_ino
, es_map
->m_lblk
, es_map
->m_len
,
434 es_map
->m_pblk
, es_map
->m_flags
, map
->m_lblk
,
435 map
->m_len
, map
->m_pblk
, map
->m_flags
,
439 #endif /* ES_AGGRESSIVE_TEST */
442 * The ext4_map_blocks() function tries to look up the requested blocks,
443 * and returns if the blocks are already mapped.
445 * Otherwise it takes the write lock of the i_data_sem and allocate blocks
446 * and store the allocated blocks in the result buffer head and mark it
449 * If file type is extents based, it will call ext4_ext_map_blocks(),
450 * Otherwise, call with ext4_ind_map_blocks() to handle indirect mapping
453 * On success, it returns the number of blocks being mapped or allocated.
454 * if create==0 and the blocks are pre-allocated and unwritten block,
455 * the result buffer head is unmapped. If the create ==1, it will make sure
456 * the buffer head is mapped.
458 * It returns 0 if plain look up failed (blocks have not been allocated), in
459 * that case, buffer head is unmapped
461 * It returns the error in case of allocation failure.
463 int ext4_map_blocks(handle_t
*handle
, struct inode
*inode
,
464 struct ext4_map_blocks
*map
, int flags
)
466 struct extent_status es
;
469 #ifdef ES_AGGRESSIVE_TEST
470 struct ext4_map_blocks orig_map
;
472 memcpy(&orig_map
, map
, sizeof(*map
));
476 ext_debug("ext4_map_blocks(): inode %lu, flag %d, max_blocks %u,"
477 "logical block %lu\n", inode
->i_ino
, flags
, map
->m_len
,
478 (unsigned long) map
->m_lblk
);
481 * ext4_map_blocks returns an int, and m_len is an unsigned int
483 if (unlikely(map
->m_len
> INT_MAX
))
484 map
->m_len
= INT_MAX
;
486 /* We can handle the block number less than EXT_MAX_BLOCKS */
487 if (unlikely(map
->m_lblk
>= EXT_MAX_BLOCKS
))
488 return -EFSCORRUPTED
;
490 /* Lookup extent status tree firstly */
491 if (ext4_es_lookup_extent(inode
, map
->m_lblk
, &es
)) {
492 if (ext4_es_is_written(&es
) || ext4_es_is_unwritten(&es
)) {
493 map
->m_pblk
= ext4_es_pblock(&es
) +
494 map
->m_lblk
- es
.es_lblk
;
495 map
->m_flags
|= ext4_es_is_written(&es
) ?
496 EXT4_MAP_MAPPED
: EXT4_MAP_UNWRITTEN
;
497 retval
= es
.es_len
- (map
->m_lblk
- es
.es_lblk
);
498 if (retval
> map
->m_len
)
501 } else if (ext4_es_is_delayed(&es
) || ext4_es_is_hole(&es
)) {
506 #ifdef ES_AGGRESSIVE_TEST
507 ext4_map_blocks_es_recheck(handle
, inode
, map
,
514 * Try to see if we can get the block without requesting a new
517 if (!(flags
& EXT4_GET_BLOCKS_NO_LOCK
))
518 down_read(&EXT4_I(inode
)->i_data_sem
);
519 if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
)) {
520 retval
= ext4_ext_map_blocks(handle
, inode
, map
, flags
&
521 EXT4_GET_BLOCKS_KEEP_SIZE
);
523 retval
= ext4_ind_map_blocks(handle
, inode
, map
, flags
&
524 EXT4_GET_BLOCKS_KEEP_SIZE
);
529 if (unlikely(retval
!= map
->m_len
)) {
530 ext4_warning(inode
->i_sb
,
531 "ES len assertion failed for inode "
532 "%lu: retval %d != map->m_len %d",
533 inode
->i_ino
, retval
, map
->m_len
);
537 status
= map
->m_flags
& EXT4_MAP_UNWRITTEN
?
538 EXTENT_STATUS_UNWRITTEN
: EXTENT_STATUS_WRITTEN
;
539 if (!(flags
& EXT4_GET_BLOCKS_DELALLOC_RESERVE
) &&
540 !(status
& EXTENT_STATUS_WRITTEN
) &&
541 ext4_find_delalloc_range(inode
, map
->m_lblk
,
542 map
->m_lblk
+ map
->m_len
- 1))
543 status
|= EXTENT_STATUS_DELAYED
;
544 ret
= ext4_es_insert_extent(inode
, map
->m_lblk
,
545 map
->m_len
, map
->m_pblk
, status
);
549 if (!(flags
& EXT4_GET_BLOCKS_NO_LOCK
))
550 up_read((&EXT4_I(inode
)->i_data_sem
));
553 if (retval
> 0 && map
->m_flags
& EXT4_MAP_MAPPED
) {
554 ret
= check_block_validity(inode
, map
);
559 /* If it is only a block(s) look up */
560 if ((flags
& EXT4_GET_BLOCKS_CREATE
) == 0)
564 * Returns if the blocks have already allocated
566 * Note that if blocks have been preallocated
567 * ext4_ext_get_block() returns the create = 0
568 * with buffer head unmapped.
570 if (retval
> 0 && map
->m_flags
& EXT4_MAP_MAPPED
)
572 * If we need to convert extent to unwritten
573 * we continue and do the actual work in
574 * ext4_ext_map_blocks()
576 if (!(flags
& EXT4_GET_BLOCKS_CONVERT_UNWRITTEN
))
580 * Here we clear m_flags because after allocating an new extent,
581 * it will be set again.
583 map
->m_flags
&= ~EXT4_MAP_FLAGS
;
586 * New blocks allocate and/or writing to unwritten extent
587 * will possibly result in updating i_data, so we take
588 * the write lock of i_data_sem, and call get_block()
589 * with create == 1 flag.
591 down_write(&EXT4_I(inode
)->i_data_sem
);
594 * We need to check for EXT4 here because migrate
595 * could have changed the inode type in between
597 if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
)) {
598 retval
= ext4_ext_map_blocks(handle
, inode
, map
, flags
);
600 retval
= ext4_ind_map_blocks(handle
, inode
, map
, flags
);
602 if (retval
> 0 && map
->m_flags
& EXT4_MAP_NEW
) {
604 * We allocated new blocks which will result in
605 * i_data's format changing. Force the migrate
606 * to fail by clearing migrate flags
608 ext4_clear_inode_state(inode
, EXT4_STATE_EXT_MIGRATE
);
612 * Update reserved blocks/metadata blocks after successful
613 * block allocation which had been deferred till now. We don't
614 * support fallocate for non extent files. So we can update
615 * reserve space here.
618 (flags
& EXT4_GET_BLOCKS_DELALLOC_RESERVE
))
619 ext4_da_update_reserve_space(inode
, retval
, 1);
625 if (unlikely(retval
!= map
->m_len
)) {
626 ext4_warning(inode
->i_sb
,
627 "ES len assertion failed for inode "
628 "%lu: retval %d != map->m_len %d",
629 inode
->i_ino
, retval
, map
->m_len
);
634 * If the extent has been zeroed out, we don't need to update
635 * extent status tree.
637 if ((flags
& EXT4_GET_BLOCKS_PRE_IO
) &&
638 ext4_es_lookup_extent(inode
, map
->m_lblk
, &es
)) {
639 if (ext4_es_is_written(&es
))
642 status
= map
->m_flags
& EXT4_MAP_UNWRITTEN
?
643 EXTENT_STATUS_UNWRITTEN
: EXTENT_STATUS_WRITTEN
;
644 if (!(flags
& EXT4_GET_BLOCKS_DELALLOC_RESERVE
) &&
645 !(status
& EXTENT_STATUS_WRITTEN
) &&
646 ext4_find_delalloc_range(inode
, map
->m_lblk
,
647 map
->m_lblk
+ map
->m_len
- 1))
648 status
|= EXTENT_STATUS_DELAYED
;
649 ret
= ext4_es_insert_extent(inode
, map
->m_lblk
, map
->m_len
,
650 map
->m_pblk
, status
);
656 up_write((&EXT4_I(inode
)->i_data_sem
));
657 if (retval
> 0 && map
->m_flags
& EXT4_MAP_MAPPED
) {
658 ret
= check_block_validity(inode
, map
);
663 * Inodes with freshly allocated blocks where contents will be
664 * visible after transaction commit must be on transaction's
667 if (map
->m_flags
& EXT4_MAP_NEW
&&
668 !(map
->m_flags
& EXT4_MAP_UNWRITTEN
) &&
669 !IS_NOQUOTA(inode
) &&
670 ext4_should_order_data(inode
)) {
671 ret
= ext4_jbd2_file_inode(handle
, inode
);
680 * Update EXT4_MAP_FLAGS in bh->b_state. For buffer heads attached to pages
681 * we have to be careful as someone else may be manipulating b_state as well.
683 static void ext4_update_bh_state(struct buffer_head
*bh
, unsigned long flags
)
685 unsigned long old_state
;
686 unsigned long new_state
;
688 flags
&= EXT4_MAP_FLAGS
;
690 /* Dummy buffer_head? Set non-atomically. */
692 bh
->b_state
= (bh
->b_state
& ~EXT4_MAP_FLAGS
) | flags
;
696 * Someone else may be modifying b_state. Be careful! This is ugly but
697 * once we get rid of using bh as a container for mapping information
698 * to pass to / from get_block functions, this can go away.
701 old_state
= READ_ONCE(bh
->b_state
);
702 new_state
= (old_state
& ~EXT4_MAP_FLAGS
) | flags
;
704 cmpxchg(&bh
->b_state
, old_state
, new_state
) != old_state
));
707 /* Maximum number of blocks we map for direct IO at once. */
708 #define DIO_MAX_BLOCKS 4096
710 static int _ext4_get_block(struct inode
*inode
, sector_t iblock
,
711 struct buffer_head
*bh
, int flags
)
713 handle_t
*handle
= ext4_journal_current_handle();
714 struct ext4_map_blocks map
;
715 int ret
= 0, started
= 0;
718 if (ext4_has_inline_data(inode
))
722 map
.m_len
= bh
->b_size
>> inode
->i_blkbits
;
724 if (flags
&& !(flags
& EXT4_GET_BLOCKS_NO_LOCK
) && !handle
) {
725 /* Direct IO write... */
726 if (map
.m_len
> DIO_MAX_BLOCKS
)
727 map
.m_len
= DIO_MAX_BLOCKS
;
728 dio_credits
= ext4_chunk_trans_blocks(inode
, map
.m_len
);
729 handle
= ext4_journal_start(inode
, EXT4_HT_MAP_BLOCKS
,
731 if (IS_ERR(handle
)) {
732 ret
= PTR_ERR(handle
);
738 ret
= ext4_map_blocks(handle
, inode
, &map
, flags
);
740 ext4_io_end_t
*io_end
= ext4_inode_aio(inode
);
742 map_bh(bh
, inode
->i_sb
, map
.m_pblk
);
743 ext4_update_bh_state(bh
, map
.m_flags
);
744 if (IS_DAX(inode
) && buffer_unwritten(bh
)) {
746 * dgc: I suspect unwritten conversion on ext4+DAX is
747 * fundamentally broken here when there are concurrent
748 * read/write in progress on this inode.
750 WARN_ON_ONCE(io_end
);
751 bh
->b_assoc_map
= inode
->i_mapping
;
752 bh
->b_private
= (void *)(unsigned long)iblock
;
754 if (io_end
&& io_end
->flag
& EXT4_IO_END_UNWRITTEN
)
755 set_buffer_defer_completion(bh
);
756 bh
->b_size
= inode
->i_sb
->s_blocksize
* map
.m_len
;
760 ext4_journal_stop(handle
);
764 int ext4_get_block(struct inode
*inode
, sector_t iblock
,
765 struct buffer_head
*bh
, int create
)
767 return _ext4_get_block(inode
, iblock
, bh
,
768 create
? EXT4_GET_BLOCKS_CREATE
: 0);
772 * `handle' can be NULL if create is zero
774 struct buffer_head
*ext4_getblk(handle_t
*handle
, struct inode
*inode
,
775 ext4_lblk_t block
, int map_flags
)
777 struct ext4_map_blocks map
;
778 struct buffer_head
*bh
;
779 int create
= map_flags
& EXT4_GET_BLOCKS_CREATE
;
782 J_ASSERT(handle
!= NULL
|| create
== 0);
786 err
= ext4_map_blocks(handle
, inode
, &map
, map_flags
);
789 return create
? ERR_PTR(-ENOSPC
) : NULL
;
793 bh
= sb_getblk(inode
->i_sb
, map
.m_pblk
);
795 return ERR_PTR(-ENOMEM
);
796 if (map
.m_flags
& EXT4_MAP_NEW
) {
797 J_ASSERT(create
!= 0);
798 J_ASSERT(handle
!= NULL
);
801 * Now that we do not always journal data, we should
802 * keep in mind whether this should always journal the
803 * new buffer as metadata. For now, regular file
804 * writes use ext4_get_block instead, so it's not a
808 BUFFER_TRACE(bh
, "call get_create_access");
809 err
= ext4_journal_get_create_access(handle
, bh
);
814 if (!buffer_uptodate(bh
)) {
815 memset(bh
->b_data
, 0, inode
->i_sb
->s_blocksize
);
816 set_buffer_uptodate(bh
);
819 BUFFER_TRACE(bh
, "call ext4_handle_dirty_metadata");
820 err
= ext4_handle_dirty_metadata(handle
, inode
, bh
);
824 BUFFER_TRACE(bh
, "not a new buffer");
831 struct buffer_head
*ext4_bread(handle_t
*handle
, struct inode
*inode
,
832 ext4_lblk_t block
, int map_flags
)
834 struct buffer_head
*bh
;
836 bh
= ext4_getblk(handle
, inode
, block
, map_flags
);
839 if (!bh
|| buffer_uptodate(bh
))
841 ll_rw_block(READ
| REQ_META
| REQ_PRIO
, 1, &bh
);
843 if (buffer_uptodate(bh
))
846 return ERR_PTR(-EIO
);
849 int ext4_walk_page_buffers(handle_t
*handle
,
850 struct buffer_head
*head
,
854 int (*fn
)(handle_t
*handle
,
855 struct buffer_head
*bh
))
857 struct buffer_head
*bh
;
858 unsigned block_start
, block_end
;
859 unsigned blocksize
= head
->b_size
;
861 struct buffer_head
*next
;
863 for (bh
= head
, block_start
= 0;
864 ret
== 0 && (bh
!= head
|| !block_start
);
865 block_start
= block_end
, bh
= next
) {
866 next
= bh
->b_this_page
;
867 block_end
= block_start
+ blocksize
;
868 if (block_end
<= from
|| block_start
>= to
) {
869 if (partial
&& !buffer_uptodate(bh
))
873 err
= (*fn
)(handle
, bh
);
881 * To preserve ordering, it is essential that the hole instantiation and
882 * the data write be encapsulated in a single transaction. We cannot
883 * close off a transaction and start a new one between the ext4_get_block()
884 * and the commit_write(). So doing the jbd2_journal_start at the start of
885 * prepare_write() is the right place.
887 * Also, this function can nest inside ext4_writepage(). In that case, we
888 * *know* that ext4_writepage() has generated enough buffer credits to do the
889 * whole page. So we won't block on the journal in that case, which is good,
890 * because the caller may be PF_MEMALLOC.
892 * By accident, ext4 can be reentered when a transaction is open via
893 * quota file writes. If we were to commit the transaction while thus
894 * reentered, there can be a deadlock - we would be holding a quota
895 * lock, and the commit would never complete if another thread had a
896 * transaction open and was blocking on the quota lock - a ranking
899 * So what we do is to rely on the fact that jbd2_journal_stop/journal_start
900 * will _not_ run commit under these circumstances because handle->h_ref
901 * is elevated. We'll still have enough credits for the tiny quotafile
904 int do_journal_get_write_access(handle_t
*handle
,
905 struct buffer_head
*bh
)
907 int dirty
= buffer_dirty(bh
);
910 if (!buffer_mapped(bh
) || buffer_freed(bh
))
913 * __block_write_begin() could have dirtied some buffers. Clean
914 * the dirty bit as jbd2_journal_get_write_access() could complain
915 * otherwise about fs integrity issues. Setting of the dirty bit
916 * by __block_write_begin() isn't a real problem here as we clear
917 * the bit before releasing a page lock and thus writeback cannot
918 * ever write the buffer.
921 clear_buffer_dirty(bh
);
922 BUFFER_TRACE(bh
, "get write access");
923 ret
= ext4_journal_get_write_access(handle
, bh
);
925 ret
= ext4_handle_dirty_metadata(handle
, NULL
, bh
);
929 static int ext4_get_block_write_nolock(struct inode
*inode
, sector_t iblock
,
930 struct buffer_head
*bh_result
, int create
);
932 #ifdef CONFIG_EXT4_FS_ENCRYPTION
933 static int ext4_block_write_begin(struct page
*page
, loff_t pos
, unsigned len
,
934 get_block_t
*get_block
)
936 unsigned from
= pos
& (PAGE_CACHE_SIZE
- 1);
937 unsigned to
= from
+ len
;
938 struct inode
*inode
= page
->mapping
->host
;
939 unsigned block_start
, block_end
;
942 unsigned blocksize
= inode
->i_sb
->s_blocksize
;
944 struct buffer_head
*bh
, *head
, *wait
[2], **wait_bh
= wait
;
945 bool decrypt
= false;
947 BUG_ON(!PageLocked(page
));
948 BUG_ON(from
> PAGE_CACHE_SIZE
);
949 BUG_ON(to
> PAGE_CACHE_SIZE
);
952 if (!page_has_buffers(page
))
953 create_empty_buffers(page
, blocksize
, 0);
954 head
= page_buffers(page
);
955 bbits
= ilog2(blocksize
);
956 block
= (sector_t
)page
->index
<< (PAGE_CACHE_SHIFT
- bbits
);
958 for (bh
= head
, block_start
= 0; bh
!= head
|| !block_start
;
959 block
++, block_start
= block_end
, bh
= bh
->b_this_page
) {
960 block_end
= block_start
+ blocksize
;
961 if (block_end
<= from
|| block_start
>= to
) {
962 if (PageUptodate(page
)) {
963 if (!buffer_uptodate(bh
))
964 set_buffer_uptodate(bh
);
969 clear_buffer_new(bh
);
970 if (!buffer_mapped(bh
)) {
971 WARN_ON(bh
->b_size
!= blocksize
);
972 err
= get_block(inode
, block
, bh
, 1);
975 if (buffer_new(bh
)) {
976 unmap_underlying_metadata(bh
->b_bdev
,
978 if (PageUptodate(page
)) {
979 clear_buffer_new(bh
);
980 set_buffer_uptodate(bh
);
981 mark_buffer_dirty(bh
);
984 if (block_end
> to
|| block_start
< from
)
985 zero_user_segments(page
, to
, block_end
,
990 if (PageUptodate(page
)) {
991 if (!buffer_uptodate(bh
))
992 set_buffer_uptodate(bh
);
995 if (!buffer_uptodate(bh
) && !buffer_delay(bh
) &&
996 !buffer_unwritten(bh
) &&
997 (block_start
< from
|| block_end
> to
)) {
998 ll_rw_block(READ
, 1, &bh
);
1000 decrypt
= ext4_encrypted_inode(inode
) &&
1001 S_ISREG(inode
->i_mode
);
1005 * If we issued read requests, let them complete.
1007 while (wait_bh
> wait
) {
1008 wait_on_buffer(*--wait_bh
);
1009 if (!buffer_uptodate(*wait_bh
))
1013 page_zero_new_buffers(page
, from
, to
);
1015 err
= ext4_decrypt(page
);
1020 static int ext4_write_begin(struct file
*file
, struct address_space
*mapping
,
1021 loff_t pos
, unsigned len
, unsigned flags
,
1022 struct page
**pagep
, void **fsdata
)
1024 struct inode
*inode
= mapping
->host
;
1025 int ret
, needed_blocks
;
1032 trace_ext4_write_begin(inode
, pos
, len
, flags
);
1034 * Reserve one block more for addition to orphan list in case
1035 * we allocate blocks but write fails for some reason
1037 needed_blocks
= ext4_writepage_trans_blocks(inode
) + 1;
1038 index
= pos
>> PAGE_CACHE_SHIFT
;
1039 from
= pos
& (PAGE_CACHE_SIZE
- 1);
1042 if (ext4_test_inode_state(inode
, EXT4_STATE_MAY_INLINE_DATA
)) {
1043 ret
= ext4_try_to_write_inline_data(mapping
, inode
, pos
, len
,
1052 * grab_cache_page_write_begin() can take a long time if the
1053 * system is thrashing due to memory pressure, or if the page
1054 * is being written back. So grab it first before we start
1055 * the transaction handle. This also allows us to allocate
1056 * the page (if needed) without using GFP_NOFS.
1059 page
= grab_cache_page_write_begin(mapping
, index
, flags
);
1065 handle
= ext4_journal_start(inode
, EXT4_HT_WRITE_PAGE
, needed_blocks
);
1066 if (IS_ERR(handle
)) {
1067 page_cache_release(page
);
1068 return PTR_ERR(handle
);
1072 if (page
->mapping
!= mapping
) {
1073 /* The page got truncated from under us */
1075 page_cache_release(page
);
1076 ext4_journal_stop(handle
);
1079 /* In case writeback began while the page was unlocked */
1080 wait_for_stable_page(page
);
1082 #ifdef CONFIG_EXT4_FS_ENCRYPTION
1083 if (ext4_should_dioread_nolock(inode
))
1084 ret
= ext4_block_write_begin(page
, pos
, len
,
1085 ext4_get_block_write
);
1087 ret
= ext4_block_write_begin(page
, pos
, len
,
1090 if (ext4_should_dioread_nolock(inode
))
1091 ret
= __block_write_begin(page
, pos
, len
, ext4_get_block_write
);
1093 ret
= __block_write_begin(page
, pos
, len
, ext4_get_block
);
1095 if (!ret
&& ext4_should_journal_data(inode
)) {
1096 ret
= ext4_walk_page_buffers(handle
, page_buffers(page
),
1098 do_journal_get_write_access
);
1104 * __block_write_begin may have instantiated a few blocks
1105 * outside i_size. Trim these off again. Don't need
1106 * i_size_read because we hold i_mutex.
1108 * Add inode to orphan list in case we crash before
1111 if (pos
+ len
> inode
->i_size
&& ext4_can_truncate(inode
))
1112 ext4_orphan_add(handle
, inode
);
1114 ext4_journal_stop(handle
);
1115 if (pos
+ len
> inode
->i_size
) {
1116 ext4_truncate_failed_write(inode
);
1118 * If truncate failed early the inode might
1119 * still be on the orphan list; we need to
1120 * make sure the inode is removed from the
1121 * orphan list in that case.
1124 ext4_orphan_del(NULL
, inode
);
1127 if (ret
== -ENOSPC
&&
1128 ext4_should_retry_alloc(inode
->i_sb
, &retries
))
1130 page_cache_release(page
);
1137 /* For write_end() in data=journal mode */
1138 static int write_end_fn(handle_t
*handle
, struct buffer_head
*bh
)
1141 if (!buffer_mapped(bh
) || buffer_freed(bh
))
1143 set_buffer_uptodate(bh
);
1144 ret
= ext4_handle_dirty_metadata(handle
, NULL
, bh
);
1145 clear_buffer_meta(bh
);
1146 clear_buffer_prio(bh
);
1151 * We need to pick up the new inode size which generic_commit_write gave us
1152 * `file' can be NULL - eg, when called from page_symlink().
1154 * ext4 never places buffers on inode->i_mapping->private_list. metadata
1155 * buffers are managed internally.
1157 static int ext4_write_end(struct file
*file
,
1158 struct address_space
*mapping
,
1159 loff_t pos
, unsigned len
, unsigned copied
,
1160 struct page
*page
, void *fsdata
)
1162 handle_t
*handle
= ext4_journal_current_handle();
1163 struct inode
*inode
= mapping
->host
;
1164 loff_t old_size
= inode
->i_size
;
1166 int i_size_changed
= 0;
1167 int inline_data
= ext4_has_inline_data(inode
);
1169 trace_ext4_write_end(inode
, pos
, len
, copied
);
1171 ret
= ext4_write_inline_data_end(inode
, pos
, len
,
1180 copied
= block_write_end(file
, mapping
, pos
,
1181 len
, copied
, page
, fsdata
);
1183 * it's important to update i_size while still holding page lock:
1184 * page writeout could otherwise come in and zero beyond i_size.
1186 i_size_changed
= ext4_update_inode_size(inode
, pos
+ copied
);
1188 page_cache_release(page
);
1191 pagecache_isize_extended(inode
, old_size
, pos
);
1193 * Don't mark the inode dirty under page lock. First, it unnecessarily
1194 * makes the holding time of page lock longer. Second, it forces lock
1195 * ordering of page lock and transaction start for journaling
1198 if (i_size_changed
|| inline_data
)
1199 ext4_mark_inode_dirty(handle
, inode
);
1201 if (pos
+ len
> inode
->i_size
&& ext4_can_truncate(inode
))
1202 /* if we have allocated more blocks and copied
1203 * less. We will have blocks allocated outside
1204 * inode->i_size. So truncate them
1206 ext4_orphan_add(handle
, inode
);
1208 ret2
= ext4_journal_stop(handle
);
1212 if (pos
+ len
> inode
->i_size
) {
1213 ext4_truncate_failed_write(inode
);
1215 * If truncate failed early the inode might still be
1216 * on the orphan list; we need to make sure the inode
1217 * is removed from the orphan list in that case.
1220 ext4_orphan_del(NULL
, inode
);
1223 return ret
? ret
: copied
;
1227 * This is a private version of page_zero_new_buffers() which doesn't
1228 * set the buffer to be dirty, since in data=journalled mode we need
1229 * to call ext4_handle_dirty_metadata() instead.
1231 static void ext4_journalled_zero_new_buffers(handle_t
*handle
,
1233 unsigned from
, unsigned to
)
1235 unsigned int block_start
= 0, block_end
;
1236 struct buffer_head
*head
, *bh
;
1238 bh
= head
= page_buffers(page
);
1240 block_end
= block_start
+ bh
->b_size
;
1241 if (buffer_new(bh
)) {
1242 if (block_end
> from
&& block_start
< to
) {
1243 if (!PageUptodate(page
)) {
1244 unsigned start
, size
;
1246 start
= max(from
, block_start
);
1247 size
= min(to
, block_end
) - start
;
1249 zero_user(page
, start
, size
);
1250 write_end_fn(handle
, bh
);
1252 clear_buffer_new(bh
);
1255 block_start
= block_end
;
1256 bh
= bh
->b_this_page
;
1257 } while (bh
!= head
);
1260 static int ext4_journalled_write_end(struct file
*file
,
1261 struct address_space
*mapping
,
1262 loff_t pos
, unsigned len
, unsigned copied
,
1263 struct page
*page
, void *fsdata
)
1265 handle_t
*handle
= ext4_journal_current_handle();
1266 struct inode
*inode
= mapping
->host
;
1267 loff_t old_size
= inode
->i_size
;
1271 int size_changed
= 0;
1272 int inline_data
= ext4_has_inline_data(inode
);
1274 trace_ext4_journalled_write_end(inode
, pos
, len
, copied
);
1275 from
= pos
& (PAGE_CACHE_SIZE
- 1);
1278 BUG_ON(!ext4_handle_valid(handle
));
1281 ret
= ext4_write_inline_data_end(inode
, pos
, len
,
1289 } else if (unlikely(copied
< len
) && !PageUptodate(page
)) {
1291 ext4_journalled_zero_new_buffers(handle
, page
, from
, to
);
1293 if (unlikely(copied
< len
))
1294 ext4_journalled_zero_new_buffers(handle
, page
,
1296 ret
= ext4_walk_page_buffers(handle
, page_buffers(page
), from
,
1297 from
+ copied
, &partial
,
1300 SetPageUptodate(page
);
1302 size_changed
= ext4_update_inode_size(inode
, pos
+ copied
);
1303 ext4_set_inode_state(inode
, EXT4_STATE_JDATA
);
1304 EXT4_I(inode
)->i_datasync_tid
= handle
->h_transaction
->t_tid
;
1306 page_cache_release(page
);
1309 pagecache_isize_extended(inode
, old_size
, pos
);
1311 if (size_changed
|| inline_data
) {
1312 ret2
= ext4_mark_inode_dirty(handle
, inode
);
1317 if (pos
+ len
> inode
->i_size
&& ext4_can_truncate(inode
))
1318 /* if we have allocated more blocks and copied
1319 * less. We will have blocks allocated outside
1320 * inode->i_size. So truncate them
1322 ext4_orphan_add(handle
, inode
);
1325 ret2
= ext4_journal_stop(handle
);
1328 if (pos
+ len
> inode
->i_size
) {
1329 ext4_truncate_failed_write(inode
);
1331 * If truncate failed early the inode might still be
1332 * on the orphan list; we need to make sure the inode
1333 * is removed from the orphan list in that case.
1336 ext4_orphan_del(NULL
, inode
);
1339 return ret
? ret
: copied
;
1343 * Reserve space for a single cluster
1345 static int ext4_da_reserve_space(struct inode
*inode
)
1347 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
1348 struct ext4_inode_info
*ei
= EXT4_I(inode
);
1352 * We will charge metadata quota at writeout time; this saves
1353 * us from metadata over-estimation, though we may go over by
1354 * a small amount in the end. Here we just reserve for data.
1356 ret
= dquot_reserve_block(inode
, EXT4_C2B(sbi
, 1));
1360 spin_lock(&ei
->i_block_reservation_lock
);
1361 if (ext4_claim_free_clusters(sbi
, 1, 0)) {
1362 spin_unlock(&ei
->i_block_reservation_lock
);
1363 dquot_release_reservation_block(inode
, EXT4_C2B(sbi
, 1));
1366 ei
->i_reserved_data_blocks
++;
1367 trace_ext4_da_reserve_space(inode
);
1368 spin_unlock(&ei
->i_block_reservation_lock
);
1370 return 0; /* success */
1373 static void ext4_da_release_space(struct inode
*inode
, int to_free
)
1375 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
1376 struct ext4_inode_info
*ei
= EXT4_I(inode
);
1379 return; /* Nothing to release, exit */
1381 spin_lock(&EXT4_I(inode
)->i_block_reservation_lock
);
1383 trace_ext4_da_release_space(inode
, to_free
);
1384 if (unlikely(to_free
> ei
->i_reserved_data_blocks
)) {
1386 * if there aren't enough reserved blocks, then the
1387 * counter is messed up somewhere. Since this
1388 * function is called from invalidate page, it's
1389 * harmless to return without any action.
1391 ext4_warning(inode
->i_sb
, "ext4_da_release_space: "
1392 "ino %lu, to_free %d with only %d reserved "
1393 "data blocks", inode
->i_ino
, to_free
,
1394 ei
->i_reserved_data_blocks
);
1396 to_free
= ei
->i_reserved_data_blocks
;
1398 ei
->i_reserved_data_blocks
-= to_free
;
1400 /* update fs dirty data blocks counter */
1401 percpu_counter_sub(&sbi
->s_dirtyclusters_counter
, to_free
);
1403 spin_unlock(&EXT4_I(inode
)->i_block_reservation_lock
);
1405 dquot_release_reservation_block(inode
, EXT4_C2B(sbi
, to_free
));
1408 static void ext4_da_page_release_reservation(struct page
*page
,
1409 unsigned int offset
,
1410 unsigned int length
)
1412 int to_release
= 0, contiguous_blks
= 0;
1413 struct buffer_head
*head
, *bh
;
1414 unsigned int curr_off
= 0;
1415 struct inode
*inode
= page
->mapping
->host
;
1416 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
1417 unsigned int stop
= offset
+ length
;
1421 BUG_ON(stop
> PAGE_CACHE_SIZE
|| stop
< length
);
1423 head
= page_buffers(page
);
1426 unsigned int next_off
= curr_off
+ bh
->b_size
;
1428 if (next_off
> stop
)
1431 if ((offset
<= curr_off
) && (buffer_delay(bh
))) {
1434 clear_buffer_delay(bh
);
1435 } else if (contiguous_blks
) {
1436 lblk
= page
->index
<<
1437 (PAGE_CACHE_SHIFT
- inode
->i_blkbits
);
1438 lblk
+= (curr_off
>> inode
->i_blkbits
) -
1440 ext4_es_remove_extent(inode
, lblk
, contiguous_blks
);
1441 contiguous_blks
= 0;
1443 curr_off
= next_off
;
1444 } while ((bh
= bh
->b_this_page
) != head
);
1446 if (contiguous_blks
) {
1447 lblk
= page
->index
<< (PAGE_CACHE_SHIFT
- inode
->i_blkbits
);
1448 lblk
+= (curr_off
>> inode
->i_blkbits
) - contiguous_blks
;
1449 ext4_es_remove_extent(inode
, lblk
, contiguous_blks
);
1452 /* If we have released all the blocks belonging to a cluster, then we
1453 * need to release the reserved space for that cluster. */
1454 num_clusters
= EXT4_NUM_B2C(sbi
, to_release
);
1455 while (num_clusters
> 0) {
1456 lblk
= (page
->index
<< (PAGE_CACHE_SHIFT
- inode
->i_blkbits
)) +
1457 ((num_clusters
- 1) << sbi
->s_cluster_bits
);
1458 if (sbi
->s_cluster_ratio
== 1 ||
1459 !ext4_find_delalloc_cluster(inode
, lblk
))
1460 ext4_da_release_space(inode
, 1);
1467 * Delayed allocation stuff
1470 struct mpage_da_data
{
1471 struct inode
*inode
;
1472 struct writeback_control
*wbc
;
1474 pgoff_t first_page
; /* The first page to write */
1475 pgoff_t next_page
; /* Current page to examine */
1476 pgoff_t last_page
; /* Last page to examine */
1478 * Extent to map - this can be after first_page because that can be
1479 * fully mapped. We somewhat abuse m_flags to store whether the extent
1480 * is delalloc or unwritten.
1482 struct ext4_map_blocks map
;
1483 struct ext4_io_submit io_submit
; /* IO submission data */
1486 static void mpage_release_unused_pages(struct mpage_da_data
*mpd
,
1491 struct pagevec pvec
;
1492 struct inode
*inode
= mpd
->inode
;
1493 struct address_space
*mapping
= inode
->i_mapping
;
1495 /* This is necessary when next_page == 0. */
1496 if (mpd
->first_page
>= mpd
->next_page
)
1499 index
= mpd
->first_page
;
1500 end
= mpd
->next_page
- 1;
1502 ext4_lblk_t start
, last
;
1503 start
= index
<< (PAGE_CACHE_SHIFT
- inode
->i_blkbits
);
1504 last
= end
<< (PAGE_CACHE_SHIFT
- inode
->i_blkbits
);
1505 ext4_es_remove_extent(inode
, start
, last
- start
+ 1);
1508 pagevec_init(&pvec
, 0);
1509 while (index
<= end
) {
1510 nr_pages
= pagevec_lookup(&pvec
, mapping
, index
, PAGEVEC_SIZE
);
1513 for (i
= 0; i
< nr_pages
; i
++) {
1514 struct page
*page
= pvec
.pages
[i
];
1515 if (page
->index
> end
)
1517 BUG_ON(!PageLocked(page
));
1518 BUG_ON(PageWriteback(page
));
1520 if (page_mapped(page
))
1521 clear_page_dirty_for_io(page
);
1522 block_invalidatepage(page
, 0, PAGE_CACHE_SIZE
);
1523 ClearPageUptodate(page
);
1527 index
= pvec
.pages
[nr_pages
- 1]->index
+ 1;
1528 pagevec_release(&pvec
);
1532 static void ext4_print_free_blocks(struct inode
*inode
)
1534 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
1535 struct super_block
*sb
= inode
->i_sb
;
1536 struct ext4_inode_info
*ei
= EXT4_I(inode
);
1538 ext4_msg(sb
, KERN_CRIT
, "Total free blocks count %lld",
1539 EXT4_C2B(EXT4_SB(inode
->i_sb
),
1540 ext4_count_free_clusters(sb
)));
1541 ext4_msg(sb
, KERN_CRIT
, "Free/Dirty block details");
1542 ext4_msg(sb
, KERN_CRIT
, "free_blocks=%lld",
1543 (long long) EXT4_C2B(EXT4_SB(sb
),
1544 percpu_counter_sum(&sbi
->s_freeclusters_counter
)));
1545 ext4_msg(sb
, KERN_CRIT
, "dirty_blocks=%lld",
1546 (long long) EXT4_C2B(EXT4_SB(sb
),
1547 percpu_counter_sum(&sbi
->s_dirtyclusters_counter
)));
1548 ext4_msg(sb
, KERN_CRIT
, "Block reservation details");
1549 ext4_msg(sb
, KERN_CRIT
, "i_reserved_data_blocks=%u",
1550 ei
->i_reserved_data_blocks
);
1554 static int ext4_bh_delay_or_unwritten(handle_t
*handle
, struct buffer_head
*bh
)
1556 return (buffer_delay(bh
) || buffer_unwritten(bh
)) && buffer_dirty(bh
);
1560 * This function is grabs code from the very beginning of
1561 * ext4_map_blocks, but assumes that the caller is from delayed write
1562 * time. This function looks up the requested blocks and sets the
1563 * buffer delay bit under the protection of i_data_sem.
1565 static int ext4_da_map_blocks(struct inode
*inode
, sector_t iblock
,
1566 struct ext4_map_blocks
*map
,
1567 struct buffer_head
*bh
)
1569 struct extent_status es
;
1571 sector_t invalid_block
= ~((sector_t
) 0xffff);
1572 #ifdef ES_AGGRESSIVE_TEST
1573 struct ext4_map_blocks orig_map
;
1575 memcpy(&orig_map
, map
, sizeof(*map
));
1578 if (invalid_block
< ext4_blocks_count(EXT4_SB(inode
->i_sb
)->s_es
))
1582 ext_debug("ext4_da_map_blocks(): inode %lu, max_blocks %u,"
1583 "logical block %lu\n", inode
->i_ino
, map
->m_len
,
1584 (unsigned long) map
->m_lblk
);
1586 /* Lookup extent status tree firstly */
1587 if (ext4_es_lookup_extent(inode
, iblock
, &es
)) {
1588 if (ext4_es_is_hole(&es
)) {
1590 down_read(&EXT4_I(inode
)->i_data_sem
);
1595 * Delayed extent could be allocated by fallocate.
1596 * So we need to check it.
1598 if (ext4_es_is_delayed(&es
) && !ext4_es_is_unwritten(&es
)) {
1599 map_bh(bh
, inode
->i_sb
, invalid_block
);
1601 set_buffer_delay(bh
);
1605 map
->m_pblk
= ext4_es_pblock(&es
) + iblock
- es
.es_lblk
;
1606 retval
= es
.es_len
- (iblock
- es
.es_lblk
);
1607 if (retval
> map
->m_len
)
1608 retval
= map
->m_len
;
1609 map
->m_len
= retval
;
1610 if (ext4_es_is_written(&es
))
1611 map
->m_flags
|= EXT4_MAP_MAPPED
;
1612 else if (ext4_es_is_unwritten(&es
))
1613 map
->m_flags
|= EXT4_MAP_UNWRITTEN
;
1617 #ifdef ES_AGGRESSIVE_TEST
1618 ext4_map_blocks_es_recheck(NULL
, inode
, map
, &orig_map
, 0);
1624 * Try to see if we can get the block without requesting a new
1625 * file system block.
1627 down_read(&EXT4_I(inode
)->i_data_sem
);
1628 if (ext4_has_inline_data(inode
))
1630 else if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
))
1631 retval
= ext4_ext_map_blocks(NULL
, inode
, map
, 0);
1633 retval
= ext4_ind_map_blocks(NULL
, inode
, map
, 0);
1639 * XXX: __block_prepare_write() unmaps passed block,
1643 * If the block was allocated from previously allocated cluster,
1644 * then we don't need to reserve it again. However we still need
1645 * to reserve metadata for every block we're going to write.
1647 if (EXT4_SB(inode
->i_sb
)->s_cluster_ratio
== 1 ||
1648 !ext4_find_delalloc_cluster(inode
, map
->m_lblk
)) {
1649 ret
= ext4_da_reserve_space(inode
);
1651 /* not enough space to reserve */
1657 ret
= ext4_es_insert_extent(inode
, map
->m_lblk
, map
->m_len
,
1658 ~0, EXTENT_STATUS_DELAYED
);
1664 map_bh(bh
, inode
->i_sb
, invalid_block
);
1666 set_buffer_delay(bh
);
1667 } else if (retval
> 0) {
1669 unsigned int status
;
1671 if (unlikely(retval
!= map
->m_len
)) {
1672 ext4_warning(inode
->i_sb
,
1673 "ES len assertion failed for inode "
1674 "%lu: retval %d != map->m_len %d",
1675 inode
->i_ino
, retval
, map
->m_len
);
1679 status
= map
->m_flags
& EXT4_MAP_UNWRITTEN
?
1680 EXTENT_STATUS_UNWRITTEN
: EXTENT_STATUS_WRITTEN
;
1681 ret
= ext4_es_insert_extent(inode
, map
->m_lblk
, map
->m_len
,
1682 map
->m_pblk
, status
);
1688 up_read((&EXT4_I(inode
)->i_data_sem
));
1694 * This is a special get_block_t callback which is used by
1695 * ext4_da_write_begin(). It will either return mapped block or
1696 * reserve space for a single block.
1698 * For delayed buffer_head we have BH_Mapped, BH_New, BH_Delay set.
1699 * We also have b_blocknr = -1 and b_bdev initialized properly
1701 * For unwritten buffer_head we have BH_Mapped, BH_New, BH_Unwritten set.
1702 * We also have b_blocknr = physicalblock mapping unwritten extent and b_bdev
1703 * initialized properly.
1705 int ext4_da_get_block_prep(struct inode
*inode
, sector_t iblock
,
1706 struct buffer_head
*bh
, int create
)
1708 struct ext4_map_blocks map
;
1711 BUG_ON(create
== 0);
1712 BUG_ON(bh
->b_size
!= inode
->i_sb
->s_blocksize
);
1714 map
.m_lblk
= iblock
;
1718 * first, we need to know whether the block is allocated already
1719 * preallocated blocks are unmapped but should treated
1720 * the same as allocated blocks.
1722 ret
= ext4_da_map_blocks(inode
, iblock
, &map
, bh
);
1726 map_bh(bh
, inode
->i_sb
, map
.m_pblk
);
1727 ext4_update_bh_state(bh
, map
.m_flags
);
1729 if (buffer_unwritten(bh
)) {
1730 /* A delayed write to unwritten bh should be marked
1731 * new and mapped. Mapped ensures that we don't do
1732 * get_block multiple times when we write to the same
1733 * offset and new ensures that we do proper zero out
1734 * for partial write.
1737 set_buffer_mapped(bh
);
1742 static int bget_one(handle_t
*handle
, struct buffer_head
*bh
)
1748 static int bput_one(handle_t
*handle
, struct buffer_head
*bh
)
1754 static int __ext4_journalled_writepage(struct page
*page
,
1757 struct address_space
*mapping
= page
->mapping
;
1758 struct inode
*inode
= mapping
->host
;
1759 struct buffer_head
*page_bufs
= NULL
;
1760 handle_t
*handle
= NULL
;
1761 int ret
= 0, err
= 0;
1762 int inline_data
= ext4_has_inline_data(inode
);
1763 struct buffer_head
*inode_bh
= NULL
;
1765 ClearPageChecked(page
);
1768 BUG_ON(page
->index
!= 0);
1769 BUG_ON(len
> ext4_get_max_inline_size(inode
));
1770 inode_bh
= ext4_journalled_write_inline_data(inode
, len
, page
);
1771 if (inode_bh
== NULL
)
1774 page_bufs
= page_buffers(page
);
1779 ext4_walk_page_buffers(handle
, page_bufs
, 0, len
,
1783 * We need to release the page lock before we start the
1784 * journal, so grab a reference so the page won't disappear
1785 * out from under us.
1790 handle
= ext4_journal_start(inode
, EXT4_HT_WRITE_PAGE
,
1791 ext4_writepage_trans_blocks(inode
));
1792 if (IS_ERR(handle
)) {
1793 ret
= PTR_ERR(handle
);
1795 goto out_no_pagelock
;
1797 BUG_ON(!ext4_handle_valid(handle
));
1801 if (page
->mapping
!= mapping
) {
1802 /* The page got truncated from under us */
1803 ext4_journal_stop(handle
);
1809 ret
= ext4_mark_inode_dirty(handle
, inode
);
1811 ret
= ext4_walk_page_buffers(handle
, page_bufs
, 0, len
, NULL
,
1812 do_journal_get_write_access
);
1814 err
= ext4_walk_page_buffers(handle
, page_bufs
, 0, len
, NULL
,
1819 EXT4_I(inode
)->i_datasync_tid
= handle
->h_transaction
->t_tid
;
1820 err
= ext4_journal_stop(handle
);
1824 if (!ext4_has_inline_data(inode
))
1825 ext4_walk_page_buffers(NULL
, page_bufs
, 0, len
,
1827 ext4_set_inode_state(inode
, EXT4_STATE_JDATA
);
1836 * Note that we don't need to start a transaction unless we're journaling data
1837 * because we should have holes filled from ext4_page_mkwrite(). We even don't
1838 * need to file the inode to the transaction's list in ordered mode because if
1839 * we are writing back data added by write(), the inode is already there and if
1840 * we are writing back data modified via mmap(), no one guarantees in which
1841 * transaction the data will hit the disk. In case we are journaling data, we
1842 * cannot start transaction directly because transaction start ranks above page
1843 * lock so we have to do some magic.
1845 * This function can get called via...
1846 * - ext4_writepages after taking page lock (have journal handle)
1847 * - journal_submit_inode_data_buffers (no journal handle)
1848 * - shrink_page_list via the kswapd/direct reclaim (no journal handle)
1849 * - grab_page_cache when doing write_begin (have journal handle)
1851 * We don't do any block allocation in this function. If we have page with
1852 * multiple blocks we need to write those buffer_heads that are mapped. This
1853 * is important for mmaped based write. So if we do with blocksize 1K
1854 * truncate(f, 1024);
1855 * a = mmap(f, 0, 4096);
1857 * truncate(f, 4096);
1858 * we have in the page first buffer_head mapped via page_mkwrite call back
1859 * but other buffer_heads would be unmapped but dirty (dirty done via the
1860 * do_wp_page). So writepage should write the first block. If we modify
1861 * the mmap area beyond 1024 we will again get a page_fault and the
1862 * page_mkwrite callback will do the block allocation and mark the
1863 * buffer_heads mapped.
1865 * We redirty the page if we have any buffer_heads that is either delay or
1866 * unwritten in the page.
1868 * We can get recursively called as show below.
1870 * ext4_writepage() -> kmalloc() -> __alloc_pages() -> page_launder() ->
1873 * But since we don't do any block allocation we should not deadlock.
1874 * Page also have the dirty flag cleared so we don't get recurive page_lock.
1876 static int ext4_writepage(struct page
*page
,
1877 struct writeback_control
*wbc
)
1882 struct buffer_head
*page_bufs
= NULL
;
1883 struct inode
*inode
= page
->mapping
->host
;
1884 struct ext4_io_submit io_submit
;
1885 bool keep_towrite
= false;
1887 trace_ext4_writepage(page
);
1888 size
= i_size_read(inode
);
1889 if (page
->index
== size
>> PAGE_CACHE_SHIFT
)
1890 len
= size
& ~PAGE_CACHE_MASK
;
1892 len
= PAGE_CACHE_SIZE
;
1894 page_bufs
= page_buffers(page
);
1896 * We cannot do block allocation or other extent handling in this
1897 * function. If there are buffers needing that, we have to redirty
1898 * the page. But we may reach here when we do a journal commit via
1899 * journal_submit_inode_data_buffers() and in that case we must write
1900 * allocated buffers to achieve data=ordered mode guarantees.
1902 * Also, if there is only one buffer per page (the fs block
1903 * size == the page size), if one buffer needs block
1904 * allocation or needs to modify the extent tree to clear the
1905 * unwritten flag, we know that the page can't be written at
1906 * all, so we might as well refuse the write immediately.
1907 * Unfortunately if the block size != page size, we can't as
1908 * easily detect this case using ext4_walk_page_buffers(), but
1909 * for the extremely common case, this is an optimization that
1910 * skips a useless round trip through ext4_bio_write_page().
1912 if (ext4_walk_page_buffers(NULL
, page_bufs
, 0, len
, NULL
,
1913 ext4_bh_delay_or_unwritten
)) {
1914 redirty_page_for_writepage(wbc
, page
);
1915 if ((current
->flags
& PF_MEMALLOC
) ||
1916 (inode
->i_sb
->s_blocksize
== PAGE_CACHE_SIZE
)) {
1918 * For memory cleaning there's no point in writing only
1919 * some buffers. So just bail out. Warn if we came here
1920 * from direct reclaim.
1922 WARN_ON_ONCE((current
->flags
& (PF_MEMALLOC
|PF_KSWAPD
))
1927 keep_towrite
= true;
1930 if (PageChecked(page
) && ext4_should_journal_data(inode
))
1932 * It's mmapped pagecache. Add buffers and journal it. There
1933 * doesn't seem much point in redirtying the page here.
1935 return __ext4_journalled_writepage(page
, len
);
1937 ext4_io_submit_init(&io_submit
, wbc
);
1938 io_submit
.io_end
= ext4_init_io_end(inode
, GFP_NOFS
);
1939 if (!io_submit
.io_end
) {
1940 redirty_page_for_writepage(wbc
, page
);
1944 ret
= ext4_bio_write_page(&io_submit
, page
, len
, wbc
, keep_towrite
);
1945 ext4_io_submit(&io_submit
);
1946 /* Drop io_end reference we got from init */
1947 ext4_put_io_end_defer(io_submit
.io_end
);
1951 static int mpage_submit_page(struct mpage_da_data
*mpd
, struct page
*page
)
1957 BUG_ON(page
->index
!= mpd
->first_page
);
1958 clear_page_dirty_for_io(page
);
1960 * We have to be very careful here! Nothing protects writeback path
1961 * against i_size changes and the page can be writeably mapped into
1962 * page tables. So an application can be growing i_size and writing
1963 * data through mmap while writeback runs. clear_page_dirty_for_io()
1964 * write-protects our page in page tables and the page cannot get
1965 * written to again until we release page lock. So only after
1966 * clear_page_dirty_for_io() we are safe to sample i_size for
1967 * ext4_bio_write_page() to zero-out tail of the written page. We rely
1968 * on the barrier provided by TestClearPageDirty in
1969 * clear_page_dirty_for_io() to make sure i_size is really sampled only
1970 * after page tables are updated.
1972 size
= i_size_read(mpd
->inode
);
1973 if (page
->index
== size
>> PAGE_SHIFT
)
1974 len
= size
& ~PAGE_MASK
;
1977 err
= ext4_bio_write_page(&mpd
->io_submit
, page
, len
, mpd
->wbc
, false);
1979 mpd
->wbc
->nr_to_write
--;
1985 #define BH_FLAGS ((1 << BH_Unwritten) | (1 << BH_Delay))
1988 * mballoc gives us at most this number of blocks...
1989 * XXX: That seems to be only a limitation of ext4_mb_normalize_request().
1990 * The rest of mballoc seems to handle chunks up to full group size.
1992 #define MAX_WRITEPAGES_EXTENT_LEN 2048
1995 * mpage_add_bh_to_extent - try to add bh to extent of blocks to map
1997 * @mpd - extent of blocks
1998 * @lblk - logical number of the block in the file
1999 * @bh - buffer head we want to add to the extent
2001 * The function is used to collect contig. blocks in the same state. If the
2002 * buffer doesn't require mapping for writeback and we haven't started the
2003 * extent of buffers to map yet, the function returns 'true' immediately - the
2004 * caller can write the buffer right away. Otherwise the function returns true
2005 * if the block has been added to the extent, false if the block couldn't be
2008 static bool mpage_add_bh_to_extent(struct mpage_da_data
*mpd
, ext4_lblk_t lblk
,
2009 struct buffer_head
*bh
)
2011 struct ext4_map_blocks
*map
= &mpd
->map
;
2013 /* Buffer that doesn't need mapping for writeback? */
2014 if (!buffer_dirty(bh
) || !buffer_mapped(bh
) ||
2015 (!buffer_delay(bh
) && !buffer_unwritten(bh
))) {
2016 /* So far no extent to map => we write the buffer right away */
2017 if (map
->m_len
== 0)
2022 /* First block in the extent? */
2023 if (map
->m_len
== 0) {
2026 map
->m_flags
= bh
->b_state
& BH_FLAGS
;
2030 /* Don't go larger than mballoc is willing to allocate */
2031 if (map
->m_len
>= MAX_WRITEPAGES_EXTENT_LEN
)
2034 /* Can we merge the block to our big extent? */
2035 if (lblk
== map
->m_lblk
+ map
->m_len
&&
2036 (bh
->b_state
& BH_FLAGS
) == map
->m_flags
) {
2044 * mpage_process_page_bufs - submit page buffers for IO or add them to extent
2046 * @mpd - extent of blocks for mapping
2047 * @head - the first buffer in the page
2048 * @bh - buffer we should start processing from
2049 * @lblk - logical number of the block in the file corresponding to @bh
2051 * Walk through page buffers from @bh upto @head (exclusive) and either submit
2052 * the page for IO if all buffers in this page were mapped and there's no
2053 * accumulated extent of buffers to map or add buffers in the page to the
2054 * extent of buffers to map. The function returns 1 if the caller can continue
2055 * by processing the next page, 0 if it should stop adding buffers to the
2056 * extent to map because we cannot extend it anymore. It can also return value
2057 * < 0 in case of error during IO submission.
2059 static int mpage_process_page_bufs(struct mpage_da_data
*mpd
,
2060 struct buffer_head
*head
,
2061 struct buffer_head
*bh
,
2064 struct inode
*inode
= mpd
->inode
;
2066 ext4_lblk_t blocks
= (i_size_read(inode
) + i_blocksize(inode
) - 1)
2067 >> inode
->i_blkbits
;
2070 BUG_ON(buffer_locked(bh
));
2072 if (lblk
>= blocks
|| !mpage_add_bh_to_extent(mpd
, lblk
, bh
)) {
2073 /* Found extent to map? */
2076 /* Everything mapped so far and we hit EOF */
2079 } while (lblk
++, (bh
= bh
->b_this_page
) != head
);
2080 /* So far everything mapped? Submit the page for IO. */
2081 if (mpd
->map
.m_len
== 0) {
2082 err
= mpage_submit_page(mpd
, head
->b_page
);
2086 return lblk
< blocks
;
2090 * mpage_map_buffers - update buffers corresponding to changed extent and
2091 * submit fully mapped pages for IO
2093 * @mpd - description of extent to map, on return next extent to map
2095 * Scan buffers corresponding to changed extent (we expect corresponding pages
2096 * to be already locked) and update buffer state according to new extent state.
2097 * We map delalloc buffers to their physical location, clear unwritten bits,
2098 * and mark buffers as uninit when we perform writes to unwritten extents
2099 * and do extent conversion after IO is finished. If the last page is not fully
2100 * mapped, we update @map to the next extent in the last page that needs
2101 * mapping. Otherwise we submit the page for IO.
2103 static int mpage_map_and_submit_buffers(struct mpage_da_data
*mpd
)
2105 struct pagevec pvec
;
2107 struct inode
*inode
= mpd
->inode
;
2108 struct buffer_head
*head
, *bh
;
2109 int bpp_bits
= PAGE_CACHE_SHIFT
- inode
->i_blkbits
;
2115 start
= mpd
->map
.m_lblk
>> bpp_bits
;
2116 end
= (mpd
->map
.m_lblk
+ mpd
->map
.m_len
- 1) >> bpp_bits
;
2117 lblk
= start
<< bpp_bits
;
2118 pblock
= mpd
->map
.m_pblk
;
2120 pagevec_init(&pvec
, 0);
2121 while (start
<= end
) {
2122 nr_pages
= pagevec_lookup(&pvec
, inode
->i_mapping
, start
,
2126 for (i
= 0; i
< nr_pages
; i
++) {
2127 struct page
*page
= pvec
.pages
[i
];
2129 if (page
->index
> end
)
2131 /* Up to 'end' pages must be contiguous */
2132 BUG_ON(page
->index
!= start
);
2133 bh
= head
= page_buffers(page
);
2135 if (lblk
< mpd
->map
.m_lblk
)
2137 if (lblk
>= mpd
->map
.m_lblk
+ mpd
->map
.m_len
) {
2139 * Buffer after end of mapped extent.
2140 * Find next buffer in the page to map.
2143 mpd
->map
.m_flags
= 0;
2145 * FIXME: If dioread_nolock supports
2146 * blocksize < pagesize, we need to make
2147 * sure we add size mapped so far to
2148 * io_end->size as the following call
2149 * can submit the page for IO.
2151 err
= mpage_process_page_bufs(mpd
, head
,
2153 pagevec_release(&pvec
);
2158 if (buffer_delay(bh
)) {
2159 clear_buffer_delay(bh
);
2160 bh
->b_blocknr
= pblock
++;
2162 clear_buffer_unwritten(bh
);
2163 } while (lblk
++, (bh
= bh
->b_this_page
) != head
);
2166 * FIXME: This is going to break if dioread_nolock
2167 * supports blocksize < pagesize as we will try to
2168 * convert potentially unmapped parts of inode.
2170 mpd
->io_submit
.io_end
->size
+= PAGE_CACHE_SIZE
;
2171 /* Page fully mapped - let IO run! */
2172 err
= mpage_submit_page(mpd
, page
);
2174 pagevec_release(&pvec
);
2179 pagevec_release(&pvec
);
2181 /* Extent fully mapped and matches with page boundary. We are done. */
2183 mpd
->map
.m_flags
= 0;
2187 static int mpage_map_one_extent(handle_t
*handle
, struct mpage_da_data
*mpd
)
2189 struct inode
*inode
= mpd
->inode
;
2190 struct ext4_map_blocks
*map
= &mpd
->map
;
2191 int get_blocks_flags
;
2192 int err
, dioread_nolock
;
2194 trace_ext4_da_write_pages_extent(inode
, map
);
2196 * Call ext4_map_blocks() to allocate any delayed allocation blocks, or
2197 * to convert an unwritten extent to be initialized (in the case
2198 * where we have written into one or more preallocated blocks). It is
2199 * possible that we're going to need more metadata blocks than
2200 * previously reserved. However we must not fail because we're in
2201 * writeback and there is nothing we can do about it so it might result
2202 * in data loss. So use reserved blocks to allocate metadata if
2205 * We pass in the magic EXT4_GET_BLOCKS_DELALLOC_RESERVE if
2206 * the blocks in question are delalloc blocks. This indicates
2207 * that the blocks and quotas has already been checked when
2208 * the data was copied into the page cache.
2210 get_blocks_flags
= EXT4_GET_BLOCKS_CREATE
|
2211 EXT4_GET_BLOCKS_METADATA_NOFAIL
;
2212 dioread_nolock
= ext4_should_dioread_nolock(inode
);
2214 get_blocks_flags
|= EXT4_GET_BLOCKS_IO_CREATE_EXT
;
2215 if (map
->m_flags
& (1 << BH_Delay
))
2216 get_blocks_flags
|= EXT4_GET_BLOCKS_DELALLOC_RESERVE
;
2218 err
= ext4_map_blocks(handle
, inode
, map
, get_blocks_flags
);
2221 if (dioread_nolock
&& (map
->m_flags
& EXT4_MAP_UNWRITTEN
)) {
2222 if (!mpd
->io_submit
.io_end
->handle
&&
2223 ext4_handle_valid(handle
)) {
2224 mpd
->io_submit
.io_end
->handle
= handle
->h_rsv_handle
;
2225 handle
->h_rsv_handle
= NULL
;
2227 ext4_set_io_unwritten_flag(inode
, mpd
->io_submit
.io_end
);
2230 BUG_ON(map
->m_len
== 0);
2231 if (map
->m_flags
& EXT4_MAP_NEW
) {
2232 struct block_device
*bdev
= inode
->i_sb
->s_bdev
;
2235 for (i
= 0; i
< map
->m_len
; i
++)
2236 unmap_underlying_metadata(bdev
, map
->m_pblk
+ i
);
2242 * mpage_map_and_submit_extent - map extent starting at mpd->lblk of length
2243 * mpd->len and submit pages underlying it for IO
2245 * @handle - handle for journal operations
2246 * @mpd - extent to map
2247 * @give_up_on_write - we set this to true iff there is a fatal error and there
2248 * is no hope of writing the data. The caller should discard
2249 * dirty pages to avoid infinite loops.
2251 * The function maps extent starting at mpd->lblk of length mpd->len. If it is
2252 * delayed, blocks are allocated, if it is unwritten, we may need to convert
2253 * them to initialized or split the described range from larger unwritten
2254 * extent. Note that we need not map all the described range since allocation
2255 * can return less blocks or the range is covered by more unwritten extents. We
2256 * cannot map more because we are limited by reserved transaction credits. On
2257 * the other hand we always make sure that the last touched page is fully
2258 * mapped so that it can be written out (and thus forward progress is
2259 * guaranteed). After mapping we submit all mapped pages for IO.
2261 static int mpage_map_and_submit_extent(handle_t
*handle
,
2262 struct mpage_da_data
*mpd
,
2263 bool *give_up_on_write
)
2265 struct inode
*inode
= mpd
->inode
;
2266 struct ext4_map_blocks
*map
= &mpd
->map
;
2271 mpd
->io_submit
.io_end
->offset
=
2272 ((loff_t
)map
->m_lblk
) << inode
->i_blkbits
;
2274 err
= mpage_map_one_extent(handle
, mpd
);
2276 struct super_block
*sb
= inode
->i_sb
;
2278 if (EXT4_SB(sb
)->s_mount_flags
& EXT4_MF_FS_ABORTED
)
2279 goto invalidate_dirty_pages
;
2281 * Let the uper layers retry transient errors.
2282 * In the case of ENOSPC, if ext4_count_free_blocks()
2283 * is non-zero, a commit should free up blocks.
2285 if ((err
== -ENOMEM
) ||
2286 (err
== -ENOSPC
&& ext4_count_free_clusters(sb
))) {
2288 goto update_disksize
;
2291 ext4_msg(sb
, KERN_CRIT
,
2292 "Delayed block allocation failed for "
2293 "inode %lu at logical offset %llu with"
2294 " max blocks %u with error %d",
2296 (unsigned long long)map
->m_lblk
,
2297 (unsigned)map
->m_len
, -err
);
2298 ext4_msg(sb
, KERN_CRIT
,
2299 "This should not happen!! Data will "
2302 ext4_print_free_blocks(inode
);
2303 invalidate_dirty_pages
:
2304 *give_up_on_write
= true;
2309 * Update buffer state, submit mapped pages, and get us new
2312 err
= mpage_map_and_submit_buffers(mpd
);
2314 goto update_disksize
;
2315 } while (map
->m_len
);
2319 * Update on-disk size after IO is submitted. Races with
2320 * truncate are avoided by checking i_size under i_data_sem.
2322 disksize
= ((loff_t
)mpd
->first_page
) << PAGE_CACHE_SHIFT
;
2323 if (disksize
> EXT4_I(inode
)->i_disksize
) {
2327 down_write(&EXT4_I(inode
)->i_data_sem
);
2328 i_size
= i_size_read(inode
);
2329 if (disksize
> i_size
)
2331 if (disksize
> EXT4_I(inode
)->i_disksize
)
2332 EXT4_I(inode
)->i_disksize
= disksize
;
2333 err2
= ext4_mark_inode_dirty(handle
, inode
);
2334 up_write(&EXT4_I(inode
)->i_data_sem
);
2336 ext4_error(inode
->i_sb
,
2337 "Failed to mark inode %lu dirty",
2346 * Calculate the total number of credits to reserve for one writepages
2347 * iteration. This is called from ext4_writepages(). We map an extent of
2348 * up to MAX_WRITEPAGES_EXTENT_LEN blocks and then we go on and finish mapping
2349 * the last partial page. So in total we can map MAX_WRITEPAGES_EXTENT_LEN +
2350 * bpp - 1 blocks in bpp different extents.
2352 static int ext4_da_writepages_trans_blocks(struct inode
*inode
)
2354 int bpp
= ext4_journal_blocks_per_page(inode
);
2356 return ext4_meta_trans_blocks(inode
,
2357 MAX_WRITEPAGES_EXTENT_LEN
+ bpp
- 1, bpp
);
2361 * mpage_prepare_extent_to_map - find & lock contiguous range of dirty pages
2362 * and underlying extent to map
2364 * @mpd - where to look for pages
2366 * Walk dirty pages in the mapping. If they are fully mapped, submit them for
2367 * IO immediately. When we find a page which isn't mapped we start accumulating
2368 * extent of buffers underlying these pages that needs mapping (formed by
2369 * either delayed or unwritten buffers). We also lock the pages containing
2370 * these buffers. The extent found is returned in @mpd structure (starting at
2371 * mpd->lblk with length mpd->len blocks).
2373 * Note that this function can attach bios to one io_end structure which are
2374 * neither logically nor physically contiguous. Although it may seem as an
2375 * unnecessary complication, it is actually inevitable in blocksize < pagesize
2376 * case as we need to track IO to all buffers underlying a page in one io_end.
2378 static int mpage_prepare_extent_to_map(struct mpage_da_data
*mpd
)
2380 struct address_space
*mapping
= mpd
->inode
->i_mapping
;
2381 struct pagevec pvec
;
2382 unsigned int nr_pages
;
2383 long left
= mpd
->wbc
->nr_to_write
;
2384 pgoff_t index
= mpd
->first_page
;
2385 pgoff_t end
= mpd
->last_page
;
2388 int blkbits
= mpd
->inode
->i_blkbits
;
2390 struct buffer_head
*head
;
2392 if (mpd
->wbc
->sync_mode
== WB_SYNC_ALL
|| mpd
->wbc
->tagged_writepages
)
2393 tag
= PAGECACHE_TAG_TOWRITE
;
2395 tag
= PAGECACHE_TAG_DIRTY
;
2397 pagevec_init(&pvec
, 0);
2399 mpd
->next_page
= index
;
2400 while (index
<= end
) {
2401 nr_pages
= pagevec_lookup_tag(&pvec
, mapping
, &index
, tag
,
2402 min(end
- index
, (pgoff_t
)PAGEVEC_SIZE
-1) + 1);
2406 for (i
= 0; i
< nr_pages
; i
++) {
2407 struct page
*page
= pvec
.pages
[i
];
2410 * At this point, the page may be truncated or
2411 * invalidated (changing page->mapping to NULL), or
2412 * even swizzled back from swapper_space to tmpfs file
2413 * mapping. However, page->index will not change
2414 * because we have a reference on the page.
2416 if (page
->index
> end
)
2420 * Accumulated enough dirty pages? This doesn't apply
2421 * to WB_SYNC_ALL mode. For integrity sync we have to
2422 * keep going because someone may be concurrently
2423 * dirtying pages, and we might have synced a lot of
2424 * newly appeared dirty pages, but have not synced all
2425 * of the old dirty pages.
2427 if (mpd
->wbc
->sync_mode
== WB_SYNC_NONE
&& left
<= 0)
2430 /* If we can't merge this page, we are done. */
2431 if (mpd
->map
.m_len
> 0 && mpd
->next_page
!= page
->index
)
2436 * If the page is no longer dirty, or its mapping no
2437 * longer corresponds to inode we are writing (which
2438 * means it has been truncated or invalidated), or the
2439 * page is already under writeback and we are not doing
2440 * a data integrity writeback, skip the page
2442 if (!PageDirty(page
) ||
2443 (PageWriteback(page
) &&
2444 (mpd
->wbc
->sync_mode
== WB_SYNC_NONE
)) ||
2445 unlikely(page
->mapping
!= mapping
)) {
2450 wait_on_page_writeback(page
);
2451 BUG_ON(PageWriteback(page
));
2453 if (mpd
->map
.m_len
== 0)
2454 mpd
->first_page
= page
->index
;
2455 mpd
->next_page
= page
->index
+ 1;
2456 /* Add all dirty buffers to mpd */
2457 lblk
= ((ext4_lblk_t
)page
->index
) <<
2458 (PAGE_CACHE_SHIFT
- blkbits
);
2459 head
= page_buffers(page
);
2460 err
= mpage_process_page_bufs(mpd
, head
, head
, lblk
);
2466 pagevec_release(&pvec
);
2471 pagevec_release(&pvec
);
2475 static int __writepage(struct page
*page
, struct writeback_control
*wbc
,
2478 struct address_space
*mapping
= data
;
2479 int ret
= ext4_writepage(page
, wbc
);
2480 mapping_set_error(mapping
, ret
);
2484 static int ext4_writepages(struct address_space
*mapping
,
2485 struct writeback_control
*wbc
)
2487 pgoff_t writeback_index
= 0;
2488 long nr_to_write
= wbc
->nr_to_write
;
2489 int range_whole
= 0;
2491 handle_t
*handle
= NULL
;
2492 struct mpage_da_data mpd
;
2493 struct inode
*inode
= mapping
->host
;
2494 int needed_blocks
, rsv_blocks
= 0, ret
= 0;
2495 struct ext4_sb_info
*sbi
= EXT4_SB(mapping
->host
->i_sb
);
2497 struct blk_plug plug
;
2498 bool give_up_on_write
= false;
2500 trace_ext4_writepages(inode
, wbc
);
2503 * No pages to write? This is mainly a kludge to avoid starting
2504 * a transaction for special inodes like journal inode on last iput()
2505 * because that could violate lock ordering on umount
2507 if (!mapping
->nrpages
|| !mapping_tagged(mapping
, PAGECACHE_TAG_DIRTY
))
2508 goto out_writepages
;
2510 if (ext4_should_journal_data(inode
)) {
2511 struct blk_plug plug
;
2513 blk_start_plug(&plug
);
2514 ret
= write_cache_pages(mapping
, wbc
, __writepage
, mapping
);
2515 blk_finish_plug(&plug
);
2516 goto out_writepages
;
2520 * If the filesystem has aborted, it is read-only, so return
2521 * right away instead of dumping stack traces later on that
2522 * will obscure the real source of the problem. We test
2523 * EXT4_MF_FS_ABORTED instead of sb->s_flag's MS_RDONLY because
2524 * the latter could be true if the filesystem is mounted
2525 * read-only, and in that case, ext4_writepages should
2526 * *never* be called, so if that ever happens, we would want
2529 if (unlikely(sbi
->s_mount_flags
& EXT4_MF_FS_ABORTED
)) {
2531 goto out_writepages
;
2534 if (ext4_should_dioread_nolock(inode
)) {
2536 * We may need to convert up to one extent per block in
2537 * the page and we may dirty the inode.
2539 rsv_blocks
= 1 + (PAGE_CACHE_SIZE
>> inode
->i_blkbits
);
2543 * If we have inline data and arrive here, it means that
2544 * we will soon create the block for the 1st page, so
2545 * we'd better clear the inline data here.
2547 if (ext4_has_inline_data(inode
)) {
2548 /* Just inode will be modified... */
2549 handle
= ext4_journal_start(inode
, EXT4_HT_INODE
, 1);
2550 if (IS_ERR(handle
)) {
2551 ret
= PTR_ERR(handle
);
2552 goto out_writepages
;
2554 BUG_ON(ext4_test_inode_state(inode
,
2555 EXT4_STATE_MAY_INLINE_DATA
));
2556 ext4_destroy_inline_data(handle
, inode
);
2557 ext4_journal_stop(handle
);
2560 if (wbc
->range_start
== 0 && wbc
->range_end
== LLONG_MAX
)
2563 if (wbc
->range_cyclic
) {
2564 writeback_index
= mapping
->writeback_index
;
2565 if (writeback_index
)
2567 mpd
.first_page
= writeback_index
;
2570 mpd
.first_page
= wbc
->range_start
>> PAGE_CACHE_SHIFT
;
2571 mpd
.last_page
= wbc
->range_end
>> PAGE_CACHE_SHIFT
;
2576 ext4_io_submit_init(&mpd
.io_submit
, wbc
);
2578 if (wbc
->sync_mode
== WB_SYNC_ALL
|| wbc
->tagged_writepages
)
2579 tag_pages_for_writeback(mapping
, mpd
.first_page
, mpd
.last_page
);
2581 blk_start_plug(&plug
);
2582 while (!done
&& mpd
.first_page
<= mpd
.last_page
) {
2583 /* For each extent of pages we use new io_end */
2584 mpd
.io_submit
.io_end
= ext4_init_io_end(inode
, GFP_KERNEL
);
2585 if (!mpd
.io_submit
.io_end
) {
2591 * We have two constraints: We find one extent to map and we
2592 * must always write out whole page (makes a difference when
2593 * blocksize < pagesize) so that we don't block on IO when we
2594 * try to write out the rest of the page. Journalled mode is
2595 * not supported by delalloc.
2597 BUG_ON(ext4_should_journal_data(inode
));
2598 needed_blocks
= ext4_da_writepages_trans_blocks(inode
);
2600 /* start a new transaction */
2601 handle
= ext4_journal_start_with_reserve(inode
,
2602 EXT4_HT_WRITE_PAGE
, needed_blocks
, rsv_blocks
);
2603 if (IS_ERR(handle
)) {
2604 ret
= PTR_ERR(handle
);
2605 ext4_msg(inode
->i_sb
, KERN_CRIT
, "%s: jbd2_start: "
2606 "%ld pages, ino %lu; err %d", __func__
,
2607 wbc
->nr_to_write
, inode
->i_ino
, ret
);
2608 /* Release allocated io_end */
2609 ext4_put_io_end(mpd
.io_submit
.io_end
);
2613 trace_ext4_da_write_pages(inode
, mpd
.first_page
, mpd
.wbc
);
2614 ret
= mpage_prepare_extent_to_map(&mpd
);
2617 ret
= mpage_map_and_submit_extent(handle
, &mpd
,
2621 * We scanned the whole range (or exhausted
2622 * nr_to_write), submitted what was mapped and
2623 * didn't find anything needing mapping. We are
2630 * Caution: If the handle is synchronous,
2631 * ext4_journal_stop() can wait for transaction commit
2632 * to finish which may depend on writeback of pages to
2633 * complete or on page lock to be released. In that
2634 * case, we have to wait until after after we have
2635 * submitted all the IO, released page locks we hold,
2636 * and dropped io_end reference (for extent conversion
2637 * to be able to complete) before stopping the handle.
2639 if (!ext4_handle_valid(handle
) || handle
->h_sync
== 0) {
2640 ext4_journal_stop(handle
);
2643 /* Submit prepared bio */
2644 ext4_io_submit(&mpd
.io_submit
);
2645 /* Unlock pages we didn't use */
2646 mpage_release_unused_pages(&mpd
, give_up_on_write
);
2648 * Drop our io_end reference we got from init. We have
2649 * to be careful and use deferred io_end finishing if
2650 * we are still holding the transaction as we can
2651 * release the last reference to io_end which may end
2652 * up doing unwritten extent conversion.
2655 ext4_put_io_end_defer(mpd
.io_submit
.io_end
);
2656 ext4_journal_stop(handle
);
2658 ext4_put_io_end(mpd
.io_submit
.io_end
);
2660 if (ret
== -ENOSPC
&& sbi
->s_journal
) {
2662 * Commit the transaction which would
2663 * free blocks released in the transaction
2666 jbd2_journal_force_commit_nested(sbi
->s_journal
);
2670 /* Fatal error - ENOMEM, EIO... */
2674 blk_finish_plug(&plug
);
2675 if (!ret
&& !cycled
&& wbc
->nr_to_write
> 0) {
2677 mpd
.last_page
= writeback_index
- 1;
2683 if (wbc
->range_cyclic
|| (range_whole
&& wbc
->nr_to_write
> 0))
2685 * Set the writeback_index so that range_cyclic
2686 * mode will write it back later
2688 mapping
->writeback_index
= mpd
.first_page
;
2691 trace_ext4_writepages_result(inode
, wbc
, ret
,
2692 nr_to_write
- wbc
->nr_to_write
);
2696 static int ext4_nonda_switch(struct super_block
*sb
)
2698 s64 free_clusters
, dirty_clusters
;
2699 struct ext4_sb_info
*sbi
= EXT4_SB(sb
);
2702 * switch to non delalloc mode if we are running low
2703 * on free block. The free block accounting via percpu
2704 * counters can get slightly wrong with percpu_counter_batch getting
2705 * accumulated on each CPU without updating global counters
2706 * Delalloc need an accurate free block accounting. So switch
2707 * to non delalloc when we are near to error range.
2710 percpu_counter_read_positive(&sbi
->s_freeclusters_counter
);
2712 percpu_counter_read_positive(&sbi
->s_dirtyclusters_counter
);
2714 * Start pushing delalloc when 1/2 of free blocks are dirty.
2716 if (dirty_clusters
&& (free_clusters
< 2 * dirty_clusters
))
2717 try_to_writeback_inodes_sb(sb
, WB_REASON_FS_FREE_SPACE
);
2719 if (2 * free_clusters
< 3 * dirty_clusters
||
2720 free_clusters
< (dirty_clusters
+ EXT4_FREECLUSTERS_WATERMARK
)) {
2722 * free block count is less than 150% of dirty blocks
2723 * or free blocks is less than watermark
2730 /* We always reserve for an inode update; the superblock could be there too */
2731 static int ext4_da_write_credits(struct inode
*inode
, loff_t pos
, unsigned len
)
2733 if (likely(ext4_has_feature_large_file(inode
->i_sb
)))
2736 if (pos
+ len
<= 0x7fffffffULL
)
2739 /* We might need to update the superblock to set LARGE_FILE */
2743 static int ext4_da_write_begin(struct file
*file
, struct address_space
*mapping
,
2744 loff_t pos
, unsigned len
, unsigned flags
,
2745 struct page
**pagep
, void **fsdata
)
2747 int ret
, retries
= 0;
2750 struct inode
*inode
= mapping
->host
;
2753 index
= pos
>> PAGE_CACHE_SHIFT
;
2755 if (ext4_nonda_switch(inode
->i_sb
)) {
2756 *fsdata
= (void *)FALL_BACK_TO_NONDELALLOC
;
2757 return ext4_write_begin(file
, mapping
, pos
,
2758 len
, flags
, pagep
, fsdata
);
2760 *fsdata
= (void *)0;
2761 trace_ext4_da_write_begin(inode
, pos
, len
, flags
);
2763 if (ext4_test_inode_state(inode
, EXT4_STATE_MAY_INLINE_DATA
)) {
2764 ret
= ext4_da_write_inline_data_begin(mapping
, inode
,
2774 * grab_cache_page_write_begin() can take a long time if the
2775 * system is thrashing due to memory pressure, or if the page
2776 * is being written back. So grab it first before we start
2777 * the transaction handle. This also allows us to allocate
2778 * the page (if needed) without using GFP_NOFS.
2781 page
= grab_cache_page_write_begin(mapping
, index
, flags
);
2787 * With delayed allocation, we don't log the i_disksize update
2788 * if there is delayed block allocation. But we still need
2789 * to journalling the i_disksize update if writes to the end
2790 * of file which has an already mapped buffer.
2793 handle
= ext4_journal_start(inode
, EXT4_HT_WRITE_PAGE
,
2794 ext4_da_write_credits(inode
, pos
, len
));
2795 if (IS_ERR(handle
)) {
2796 page_cache_release(page
);
2797 return PTR_ERR(handle
);
2801 if (page
->mapping
!= mapping
) {
2802 /* The page got truncated from under us */
2804 page_cache_release(page
);
2805 ext4_journal_stop(handle
);
2808 /* In case writeback began while the page was unlocked */
2809 wait_for_stable_page(page
);
2811 #ifdef CONFIG_EXT4_FS_ENCRYPTION
2812 ret
= ext4_block_write_begin(page
, pos
, len
,
2813 ext4_da_get_block_prep
);
2815 ret
= __block_write_begin(page
, pos
, len
, ext4_da_get_block_prep
);
2819 ext4_journal_stop(handle
);
2821 * block_write_begin may have instantiated a few blocks
2822 * outside i_size. Trim these off again. Don't need
2823 * i_size_read because we hold i_mutex.
2825 if (pos
+ len
> inode
->i_size
)
2826 ext4_truncate_failed_write(inode
);
2828 if (ret
== -ENOSPC
&&
2829 ext4_should_retry_alloc(inode
->i_sb
, &retries
))
2832 page_cache_release(page
);
2841 * Check if we should update i_disksize
2842 * when write to the end of file but not require block allocation
2844 static int ext4_da_should_update_i_disksize(struct page
*page
,
2845 unsigned long offset
)
2847 struct buffer_head
*bh
;
2848 struct inode
*inode
= page
->mapping
->host
;
2852 bh
= page_buffers(page
);
2853 idx
= offset
>> inode
->i_blkbits
;
2855 for (i
= 0; i
< idx
; i
++)
2856 bh
= bh
->b_this_page
;
2858 if (!buffer_mapped(bh
) || (buffer_delay(bh
)) || buffer_unwritten(bh
))
2863 static int ext4_da_write_end(struct file
*file
,
2864 struct address_space
*mapping
,
2865 loff_t pos
, unsigned len
, unsigned copied
,
2866 struct page
*page
, void *fsdata
)
2868 struct inode
*inode
= mapping
->host
;
2870 handle_t
*handle
= ext4_journal_current_handle();
2872 unsigned long start
, end
;
2873 int write_mode
= (int)(unsigned long)fsdata
;
2875 if (write_mode
== FALL_BACK_TO_NONDELALLOC
)
2876 return ext4_write_end(file
, mapping
, pos
,
2877 len
, copied
, page
, fsdata
);
2879 trace_ext4_da_write_end(inode
, pos
, len
, copied
);
2880 start
= pos
& (PAGE_CACHE_SIZE
- 1);
2881 end
= start
+ copied
- 1;
2884 * generic_write_end() will run mark_inode_dirty() if i_size
2885 * changes. So let's piggyback the i_disksize mark_inode_dirty
2888 new_i_size
= pos
+ copied
;
2889 if (copied
&& new_i_size
> EXT4_I(inode
)->i_disksize
) {
2890 if (ext4_has_inline_data(inode
) ||
2891 ext4_da_should_update_i_disksize(page
, end
)) {
2892 ext4_update_i_disksize(inode
, new_i_size
);
2893 /* We need to mark inode dirty even if
2894 * new_i_size is less that inode->i_size
2895 * bu greater than i_disksize.(hint delalloc)
2897 ext4_mark_inode_dirty(handle
, inode
);
2901 if (write_mode
!= CONVERT_INLINE_DATA
&&
2902 ext4_test_inode_state(inode
, EXT4_STATE_MAY_INLINE_DATA
) &&
2903 ext4_has_inline_data(inode
))
2904 ret2
= ext4_da_write_inline_data_end(inode
, pos
, len
, copied
,
2907 ret2
= generic_write_end(file
, mapping
, pos
, len
, copied
,
2913 ret2
= ext4_journal_stop(handle
);
2917 return ret
? ret
: copied
;
2920 static void ext4_da_invalidatepage(struct page
*page
, unsigned int offset
,
2921 unsigned int length
)
2924 * Drop reserved blocks
2926 BUG_ON(!PageLocked(page
));
2927 if (!page_has_buffers(page
))
2930 ext4_da_page_release_reservation(page
, offset
, length
);
2933 ext4_invalidatepage(page
, offset
, length
);
2939 * Force all delayed allocation blocks to be allocated for a given inode.
2941 int ext4_alloc_da_blocks(struct inode
*inode
)
2943 trace_ext4_alloc_da_blocks(inode
);
2945 if (!EXT4_I(inode
)->i_reserved_data_blocks
)
2949 * We do something simple for now. The filemap_flush() will
2950 * also start triggering a write of the data blocks, which is
2951 * not strictly speaking necessary (and for users of
2952 * laptop_mode, not even desirable). However, to do otherwise
2953 * would require replicating code paths in:
2955 * ext4_writepages() ->
2956 * write_cache_pages() ---> (via passed in callback function)
2957 * __mpage_da_writepage() -->
2958 * mpage_add_bh_to_extent()
2959 * mpage_da_map_blocks()
2961 * The problem is that write_cache_pages(), located in
2962 * mm/page-writeback.c, marks pages clean in preparation for
2963 * doing I/O, which is not desirable if we're not planning on
2966 * We could call write_cache_pages(), and then redirty all of
2967 * the pages by calling redirty_page_for_writepage() but that
2968 * would be ugly in the extreme. So instead we would need to
2969 * replicate parts of the code in the above functions,
2970 * simplifying them because we wouldn't actually intend to
2971 * write out the pages, but rather only collect contiguous
2972 * logical block extents, call the multi-block allocator, and
2973 * then update the buffer heads with the block allocations.
2975 * For now, though, we'll cheat by calling filemap_flush(),
2976 * which will map the blocks, and start the I/O, but not
2977 * actually wait for the I/O to complete.
2979 return filemap_flush(inode
->i_mapping
);
2983 * bmap() is special. It gets used by applications such as lilo and by
2984 * the swapper to find the on-disk block of a specific piece of data.
2986 * Naturally, this is dangerous if the block concerned is still in the
2987 * journal. If somebody makes a swapfile on an ext4 data-journaling
2988 * filesystem and enables swap, then they may get a nasty shock when the
2989 * data getting swapped to that swapfile suddenly gets overwritten by
2990 * the original zero's written out previously to the journal and
2991 * awaiting writeback in the kernel's buffer cache.
2993 * So, if we see any bmap calls here on a modified, data-journaled file,
2994 * take extra steps to flush any blocks which might be in the cache.
2996 static sector_t
ext4_bmap(struct address_space
*mapping
, sector_t block
)
2998 struct inode
*inode
= mapping
->host
;
3003 * We can get here for an inline file via the FIBMAP ioctl
3005 if (ext4_has_inline_data(inode
))
3008 if (mapping_tagged(mapping
, PAGECACHE_TAG_DIRTY
) &&
3009 test_opt(inode
->i_sb
, DELALLOC
)) {
3011 * With delalloc we want to sync the file
3012 * so that we can make sure we allocate
3015 filemap_write_and_wait(mapping
);
3018 if (EXT4_JOURNAL(inode
) &&
3019 ext4_test_inode_state(inode
, EXT4_STATE_JDATA
)) {
3021 * This is a REALLY heavyweight approach, but the use of
3022 * bmap on dirty files is expected to be extremely rare:
3023 * only if we run lilo or swapon on a freshly made file
3024 * do we expect this to happen.
3026 * (bmap requires CAP_SYS_RAWIO so this does not
3027 * represent an unprivileged user DOS attack --- we'd be
3028 * in trouble if mortal users could trigger this path at
3031 * NB. EXT4_STATE_JDATA is not set on files other than
3032 * regular files. If somebody wants to bmap a directory
3033 * or symlink and gets confused because the buffer
3034 * hasn't yet been flushed to disk, they deserve
3035 * everything they get.
3038 ext4_clear_inode_state(inode
, EXT4_STATE_JDATA
);
3039 journal
= EXT4_JOURNAL(inode
);
3040 jbd2_journal_lock_updates(journal
);
3041 err
= jbd2_journal_flush(journal
);
3042 jbd2_journal_unlock_updates(journal
);
3048 return generic_block_bmap(mapping
, block
, ext4_get_block
);
3051 static int ext4_readpage(struct file
*file
, struct page
*page
)
3054 struct inode
*inode
= page
->mapping
->host
;
3056 trace_ext4_readpage(page
);
3058 if (ext4_has_inline_data(inode
))
3059 ret
= ext4_readpage_inline(inode
, page
);
3062 return ext4_mpage_readpages(page
->mapping
, NULL
, page
, 1);
3068 ext4_readpages(struct file
*file
, struct address_space
*mapping
,
3069 struct list_head
*pages
, unsigned nr_pages
)
3071 struct inode
*inode
= mapping
->host
;
3073 /* If the file has inline data, no need to do readpages. */
3074 if (ext4_has_inline_data(inode
))
3077 return ext4_mpage_readpages(mapping
, pages
, NULL
, nr_pages
);
3080 static void ext4_invalidatepage(struct page
*page
, unsigned int offset
,
3081 unsigned int length
)
3083 trace_ext4_invalidatepage(page
, offset
, length
);
3085 /* No journalling happens on data buffers when this function is used */
3086 WARN_ON(page_has_buffers(page
) && buffer_jbd(page_buffers(page
)));
3088 block_invalidatepage(page
, offset
, length
);
3091 static int __ext4_journalled_invalidatepage(struct page
*page
,
3092 unsigned int offset
,
3093 unsigned int length
)
3095 journal_t
*journal
= EXT4_JOURNAL(page
->mapping
->host
);
3097 trace_ext4_journalled_invalidatepage(page
, offset
, length
);
3100 * If it's a full truncate we just forget about the pending dirtying
3102 if (offset
== 0 && length
== PAGE_CACHE_SIZE
)
3103 ClearPageChecked(page
);
3105 return jbd2_journal_invalidatepage(journal
, page
, offset
, length
);
3108 /* Wrapper for aops... */
3109 static void ext4_journalled_invalidatepage(struct page
*page
,
3110 unsigned int offset
,
3111 unsigned int length
)
3113 WARN_ON(__ext4_journalled_invalidatepage(page
, offset
, length
) < 0);
3116 static int ext4_releasepage(struct page
*page
, gfp_t wait
)
3118 journal_t
*journal
= EXT4_JOURNAL(page
->mapping
->host
);
3120 trace_ext4_releasepage(page
);
3122 /* Page has dirty journalled data -> cannot release */
3123 if (PageChecked(page
))
3126 return jbd2_journal_try_to_free_buffers(journal
, page
, wait
);
3128 return try_to_free_buffers(page
);
3132 * ext4_get_block used when preparing for a DIO write or buffer write.
3133 * We allocate an uinitialized extent if blocks haven't been allocated.
3134 * The extent will be converted to initialized after the IO is complete.
3136 int ext4_get_block_write(struct inode
*inode
, sector_t iblock
,
3137 struct buffer_head
*bh_result
, int create
)
3139 ext4_debug("ext4_get_block_write: inode %lu, create flag %d\n",
3140 inode
->i_ino
, create
);
3141 return _ext4_get_block(inode
, iblock
, bh_result
,
3142 EXT4_GET_BLOCKS_IO_CREATE_EXT
);
3145 static int ext4_get_block_write_nolock(struct inode
*inode
, sector_t iblock
,
3146 struct buffer_head
*bh_result
, int create
)
3148 ext4_debug("ext4_get_block_write_nolock: inode %lu, create flag %d\n",
3149 inode
->i_ino
, create
);
3150 return _ext4_get_block(inode
, iblock
, bh_result
,
3151 EXT4_GET_BLOCKS_NO_LOCK
);
3154 int ext4_get_block_dax(struct inode
*inode
, sector_t iblock
,
3155 struct buffer_head
*bh_result
, int create
)
3157 int flags
= EXT4_GET_BLOCKS_PRE_IO
| EXT4_GET_BLOCKS_UNWRIT_EXT
;
3159 flags
|= EXT4_GET_BLOCKS_CREATE
;
3160 ext4_debug("ext4_get_block_dax: inode %lu, create flag %d\n",
3161 inode
->i_ino
, create
);
3162 return _ext4_get_block(inode
, iblock
, bh_result
, flags
);
3165 static void ext4_end_io_dio(struct kiocb
*iocb
, loff_t offset
,
3166 ssize_t size
, void *private)
3168 ext4_io_end_t
*io_end
= iocb
->private;
3170 /* if not async direct IO just return */
3174 ext_debug("ext4_end_io_dio(): io_end 0x%p "
3175 "for inode %lu, iocb 0x%p, offset %llu, size %zd\n",
3176 iocb
->private, io_end
->inode
->i_ino
, iocb
, offset
,
3179 iocb
->private = NULL
;
3180 io_end
->offset
= offset
;
3181 io_end
->size
= size
;
3182 ext4_put_io_end(io_end
);
3186 * For ext4 extent files, ext4 will do direct-io write to holes,
3187 * preallocated extents, and those write extend the file, no need to
3188 * fall back to buffered IO.
3190 * For holes, we fallocate those blocks, mark them as unwritten
3191 * If those blocks were preallocated, we mark sure they are split, but
3192 * still keep the range to write as unwritten.
3194 * The unwritten extents will be converted to written when DIO is completed.
3195 * For async direct IO, since the IO may still pending when return, we
3196 * set up an end_io call back function, which will do the conversion
3197 * when async direct IO completed.
3199 * If the O_DIRECT write will extend the file then add this inode to the
3200 * orphan list. So recovery will truncate it back to the original size
3201 * if the machine crashes during the write.
3204 static ssize_t
ext4_ext_direct_IO(struct kiocb
*iocb
, struct iov_iter
*iter
,
3207 struct file
*file
= iocb
->ki_filp
;
3208 struct inode
*inode
= file
->f_mapping
->host
;
3210 size_t count
= iov_iter_count(iter
);
3212 get_block_t
*get_block_func
= NULL
;
3214 loff_t final_size
= offset
+ count
;
3215 ext4_io_end_t
*io_end
= NULL
;
3217 /* Use the old path for reads and writes beyond i_size. */
3218 if (iov_iter_rw(iter
) != WRITE
|| final_size
> inode
->i_size
)
3219 return ext4_ind_direct_IO(iocb
, iter
, offset
);
3221 BUG_ON(iocb
->private == NULL
);
3224 * Make all waiters for direct IO properly wait also for extent
3225 * conversion. This also disallows race between truncate() and
3226 * overwrite DIO as i_dio_count needs to be incremented under i_mutex.
3228 if (iov_iter_rw(iter
) == WRITE
)
3229 inode_dio_begin(inode
);
3231 /* If we do a overwrite dio, i_mutex locking can be released */
3232 overwrite
= *((int *)iocb
->private);
3235 down_read(&EXT4_I(inode
)->i_data_sem
);
3236 mutex_unlock(&inode
->i_mutex
);
3240 * We could direct write to holes and fallocate.
3242 * Allocated blocks to fill the hole are marked as
3243 * unwritten to prevent parallel buffered read to expose
3244 * the stale data before DIO complete the data IO.
3246 * As to previously fallocated extents, ext4 get_block will
3247 * just simply mark the buffer mapped but still keep the
3248 * extents unwritten.
3250 * For non AIO case, we will convert those unwritten extents
3251 * to written after return back from blockdev_direct_IO.
3253 * For async DIO, the conversion needs to be deferred when the
3254 * IO is completed. The ext4 end_io callback function will be
3255 * called to take care of the conversion work. Here for async
3256 * case, we allocate an io_end structure to hook to the iocb.
3258 iocb
->private = NULL
;
3260 get_block_func
= ext4_get_block_write_nolock
;
3262 ext4_inode_aio_set(inode
, NULL
);
3263 if (!is_sync_kiocb(iocb
)) {
3264 io_end
= ext4_init_io_end(inode
, GFP_NOFS
);
3270 * Grab reference for DIO. Will be dropped in
3273 iocb
->private = ext4_get_io_end(io_end
);
3275 * we save the io structure for current async direct
3276 * IO, so that later ext4_map_blocks() could flag the
3277 * io structure whether there is a unwritten extents
3278 * needs to be converted when IO is completed.
3280 ext4_inode_aio_set(inode
, io_end
);
3282 get_block_func
= ext4_get_block_write
;
3283 dio_flags
= DIO_LOCKING
;
3285 #ifdef CONFIG_EXT4_FS_ENCRYPTION
3286 BUG_ON(ext4_encrypted_inode(inode
) && S_ISREG(inode
->i_mode
));
3289 ret
= dax_do_io(iocb
, inode
, iter
, offset
, get_block_func
,
3290 ext4_end_io_dio
, dio_flags
);
3292 ret
= __blockdev_direct_IO(iocb
, inode
,
3293 inode
->i_sb
->s_bdev
, iter
, offset
,
3295 ext4_end_io_dio
, NULL
, dio_flags
);
3298 * Put our reference to io_end. This can free the io_end structure e.g.
3299 * in sync IO case or in case of error. It can even perform extent
3300 * conversion if all bios we submitted finished before we got here.
3301 * Note that in that case iocb->private can be already set to NULL
3305 ext4_inode_aio_set(inode
, NULL
);
3306 ext4_put_io_end(io_end
);
3308 * When no IO was submitted ext4_end_io_dio() was not
3309 * called so we have to put iocb's reference.
3311 if (ret
<= 0 && ret
!= -EIOCBQUEUED
&& iocb
->private) {
3312 WARN_ON(iocb
->private != io_end
);
3313 WARN_ON(io_end
->flag
& EXT4_IO_END_UNWRITTEN
);
3314 ext4_put_io_end(io_end
);
3315 iocb
->private = NULL
;
3318 if (ret
> 0 && !overwrite
&& ext4_test_inode_state(inode
,
3319 EXT4_STATE_DIO_UNWRITTEN
)) {
3322 * for non AIO case, since the IO is already
3323 * completed, we could do the conversion right here
3325 err
= ext4_convert_unwritten_extents(NULL
, inode
,
3329 ext4_clear_inode_state(inode
, EXT4_STATE_DIO_UNWRITTEN
);
3333 if (iov_iter_rw(iter
) == WRITE
)
3334 inode_dio_end(inode
);
3335 /* take i_mutex locking again if we do a ovewrite dio */
3337 up_read(&EXT4_I(inode
)->i_data_sem
);
3338 mutex_lock(&inode
->i_mutex
);
3344 static ssize_t
ext4_direct_IO(struct kiocb
*iocb
, struct iov_iter
*iter
,
3347 struct file
*file
= iocb
->ki_filp
;
3348 struct inode
*inode
= file
->f_mapping
->host
;
3349 size_t count
= iov_iter_count(iter
);
3352 #ifdef CONFIG_EXT4_FS_ENCRYPTION
3353 if (ext4_encrypted_inode(inode
) && S_ISREG(inode
->i_mode
))
3358 * If we are doing data journalling we don't support O_DIRECT
3360 if (ext4_should_journal_data(inode
))
3363 /* Let buffer I/O handle the inline data case. */
3364 if (ext4_has_inline_data(inode
))
3367 trace_ext4_direct_IO_enter(inode
, offset
, count
, iov_iter_rw(iter
));
3368 if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
))
3369 ret
= ext4_ext_direct_IO(iocb
, iter
, offset
);
3371 ret
= ext4_ind_direct_IO(iocb
, iter
, offset
);
3372 trace_ext4_direct_IO_exit(inode
, offset
, count
, iov_iter_rw(iter
), ret
);
3377 * Pages can be marked dirty completely asynchronously from ext4's journalling
3378 * activity. By filemap_sync_pte(), try_to_unmap_one(), etc. We cannot do
3379 * much here because ->set_page_dirty is called under VFS locks. The page is
3380 * not necessarily locked.
3382 * We cannot just dirty the page and leave attached buffers clean, because the
3383 * buffers' dirty state is "definitive". We cannot just set the buffers dirty
3384 * or jbddirty because all the journalling code will explode.
3386 * So what we do is to mark the page "pending dirty" and next time writepage
3387 * is called, propagate that into the buffers appropriately.
3389 static int ext4_journalled_set_page_dirty(struct page
*page
)
3391 SetPageChecked(page
);
3392 return __set_page_dirty_nobuffers(page
);
3395 static const struct address_space_operations ext4_aops
= {
3396 .readpage
= ext4_readpage
,
3397 .readpages
= ext4_readpages
,
3398 .writepage
= ext4_writepage
,
3399 .writepages
= ext4_writepages
,
3400 .write_begin
= ext4_write_begin
,
3401 .write_end
= ext4_write_end
,
3403 .invalidatepage
= ext4_invalidatepage
,
3404 .releasepage
= ext4_releasepage
,
3405 .direct_IO
= ext4_direct_IO
,
3406 .migratepage
= buffer_migrate_page
,
3407 .is_partially_uptodate
= block_is_partially_uptodate
,
3408 .error_remove_page
= generic_error_remove_page
,
3411 static const struct address_space_operations ext4_journalled_aops
= {
3412 .readpage
= ext4_readpage
,
3413 .readpages
= ext4_readpages
,
3414 .writepage
= ext4_writepage
,
3415 .writepages
= ext4_writepages
,
3416 .write_begin
= ext4_write_begin
,
3417 .write_end
= ext4_journalled_write_end
,
3418 .set_page_dirty
= ext4_journalled_set_page_dirty
,
3420 .invalidatepage
= ext4_journalled_invalidatepage
,
3421 .releasepage
= ext4_releasepage
,
3422 .direct_IO
= ext4_direct_IO
,
3423 .is_partially_uptodate
= block_is_partially_uptodate
,
3424 .error_remove_page
= generic_error_remove_page
,
3427 static const struct address_space_operations ext4_da_aops
= {
3428 .readpage
= ext4_readpage
,
3429 .readpages
= ext4_readpages
,
3430 .writepage
= ext4_writepage
,
3431 .writepages
= ext4_writepages
,
3432 .write_begin
= ext4_da_write_begin
,
3433 .write_end
= ext4_da_write_end
,
3435 .invalidatepage
= ext4_da_invalidatepage
,
3436 .releasepage
= ext4_releasepage
,
3437 .direct_IO
= ext4_direct_IO
,
3438 .migratepage
= buffer_migrate_page
,
3439 .is_partially_uptodate
= block_is_partially_uptodate
,
3440 .error_remove_page
= generic_error_remove_page
,
3443 void ext4_set_aops(struct inode
*inode
)
3445 switch (ext4_inode_journal_mode(inode
)) {
3446 case EXT4_INODE_ORDERED_DATA_MODE
:
3447 ext4_set_inode_state(inode
, EXT4_STATE_ORDERED_MODE
);
3449 case EXT4_INODE_WRITEBACK_DATA_MODE
:
3450 ext4_clear_inode_state(inode
, EXT4_STATE_ORDERED_MODE
);
3452 case EXT4_INODE_JOURNAL_DATA_MODE
:
3453 inode
->i_mapping
->a_ops
= &ext4_journalled_aops
;
3458 if (test_opt(inode
->i_sb
, DELALLOC
))
3459 inode
->i_mapping
->a_ops
= &ext4_da_aops
;
3461 inode
->i_mapping
->a_ops
= &ext4_aops
;
3464 static int __ext4_block_zero_page_range(handle_t
*handle
,
3465 struct address_space
*mapping
, loff_t from
, loff_t length
)
3467 ext4_fsblk_t index
= from
>> PAGE_CACHE_SHIFT
;
3468 unsigned offset
= from
& (PAGE_CACHE_SIZE
-1);
3469 unsigned blocksize
, pos
;
3471 struct inode
*inode
= mapping
->host
;
3472 struct buffer_head
*bh
;
3476 page
= find_or_create_page(mapping
, from
>> PAGE_CACHE_SHIFT
,
3477 mapping_gfp_constraint(mapping
, ~__GFP_FS
));
3481 blocksize
= inode
->i_sb
->s_blocksize
;
3483 iblock
= index
<< (PAGE_CACHE_SHIFT
- inode
->i_sb
->s_blocksize_bits
);
3485 if (!page_has_buffers(page
))
3486 create_empty_buffers(page
, blocksize
, 0);
3488 /* Find the buffer that contains "offset" */
3489 bh
= page_buffers(page
);
3491 while (offset
>= pos
) {
3492 bh
= bh
->b_this_page
;
3496 if (buffer_freed(bh
)) {
3497 BUFFER_TRACE(bh
, "freed: skip");
3500 if (!buffer_mapped(bh
)) {
3501 BUFFER_TRACE(bh
, "unmapped");
3502 ext4_get_block(inode
, iblock
, bh
, 0);
3503 /* unmapped? It's a hole - nothing to do */
3504 if (!buffer_mapped(bh
)) {
3505 BUFFER_TRACE(bh
, "still unmapped");
3510 /* Ok, it's mapped. Make sure it's up-to-date */
3511 if (PageUptodate(page
))
3512 set_buffer_uptodate(bh
);
3514 if (!buffer_uptodate(bh
)) {
3516 ll_rw_block(READ
, 1, &bh
);
3518 /* Uhhuh. Read error. Complain and punt. */
3519 if (!buffer_uptodate(bh
))
3521 if (S_ISREG(inode
->i_mode
) &&
3522 ext4_encrypted_inode(inode
)) {
3523 /* We expect the key to be set. */
3524 BUG_ON(!ext4_has_encryption_key(inode
));
3525 BUG_ON(blocksize
!= PAGE_CACHE_SIZE
);
3526 WARN_ON_ONCE(ext4_decrypt(page
));
3529 if (ext4_should_journal_data(inode
)) {
3530 BUFFER_TRACE(bh
, "get write access");
3531 err
= ext4_journal_get_write_access(handle
, bh
);
3535 zero_user(page
, offset
, length
);
3536 BUFFER_TRACE(bh
, "zeroed end of block");
3538 if (ext4_should_journal_data(inode
)) {
3539 err
= ext4_handle_dirty_metadata(handle
, inode
, bh
);
3542 mark_buffer_dirty(bh
);
3543 if (ext4_test_inode_state(inode
, EXT4_STATE_ORDERED_MODE
))
3544 err
= ext4_jbd2_file_inode(handle
, inode
);
3549 page_cache_release(page
);
3554 * ext4_block_zero_page_range() zeros out a mapping of length 'length'
3555 * starting from file offset 'from'. The range to be zero'd must
3556 * be contained with in one block. If the specified range exceeds
3557 * the end of the block it will be shortened to end of the block
3558 * that cooresponds to 'from'
3560 static int ext4_block_zero_page_range(handle_t
*handle
,
3561 struct address_space
*mapping
, loff_t from
, loff_t length
)
3563 struct inode
*inode
= mapping
->host
;
3564 unsigned offset
= from
& (PAGE_CACHE_SIZE
-1);
3565 unsigned blocksize
= inode
->i_sb
->s_blocksize
;
3566 unsigned max
= blocksize
- (offset
& (blocksize
- 1));
3569 * correct length if it does not fall between
3570 * 'from' and the end of the block
3572 if (length
> max
|| length
< 0)
3576 return dax_zero_page_range(inode
, from
, length
, ext4_get_block
);
3577 return __ext4_block_zero_page_range(handle
, mapping
, from
, length
);
3581 * ext4_block_truncate_page() zeroes out a mapping from file offset `from'
3582 * up to the end of the block which corresponds to `from'.
3583 * This required during truncate. We need to physically zero the tail end
3584 * of that block so it doesn't yield old data if the file is later grown.
3586 static int ext4_block_truncate_page(handle_t
*handle
,
3587 struct address_space
*mapping
, loff_t from
)
3589 unsigned offset
= from
& (PAGE_CACHE_SIZE
-1);
3592 struct inode
*inode
= mapping
->host
;
3594 /* If we are processing an encrypted inode during orphan list handling */
3595 if (ext4_encrypted_inode(inode
) && !ext4_has_encryption_key(inode
))
3598 blocksize
= inode
->i_sb
->s_blocksize
;
3599 length
= blocksize
- (offset
& (blocksize
- 1));
3601 return ext4_block_zero_page_range(handle
, mapping
, from
, length
);
3604 int ext4_zero_partial_blocks(handle_t
*handle
, struct inode
*inode
,
3605 loff_t lstart
, loff_t length
)
3607 struct super_block
*sb
= inode
->i_sb
;
3608 struct address_space
*mapping
= inode
->i_mapping
;
3609 unsigned partial_start
, partial_end
;
3610 ext4_fsblk_t start
, end
;
3611 loff_t byte_end
= (lstart
+ length
- 1);
3614 partial_start
= lstart
& (sb
->s_blocksize
- 1);
3615 partial_end
= byte_end
& (sb
->s_blocksize
- 1);
3617 start
= lstart
>> sb
->s_blocksize_bits
;
3618 end
= byte_end
>> sb
->s_blocksize_bits
;
3620 /* Handle partial zero within the single block */
3622 (partial_start
|| (partial_end
!= sb
->s_blocksize
- 1))) {
3623 err
= ext4_block_zero_page_range(handle
, mapping
,
3627 /* Handle partial zero out on the start of the range */
3628 if (partial_start
) {
3629 err
= ext4_block_zero_page_range(handle
, mapping
,
3630 lstart
, sb
->s_blocksize
);
3634 /* Handle partial zero out on the end of the range */
3635 if (partial_end
!= sb
->s_blocksize
- 1)
3636 err
= ext4_block_zero_page_range(handle
, mapping
,
3637 byte_end
- partial_end
,
3642 int ext4_can_truncate(struct inode
*inode
)
3644 if (S_ISREG(inode
->i_mode
))
3646 if (S_ISDIR(inode
->i_mode
))
3648 if (S_ISLNK(inode
->i_mode
))
3649 return !ext4_inode_is_fast_symlink(inode
);
3654 * We have to make sure i_disksize gets properly updated before we truncate
3655 * page cache due to hole punching or zero range. Otherwise i_disksize update
3656 * can get lost as it may have been postponed to submission of writeback but
3657 * that will never happen after we truncate page cache.
3659 int ext4_update_disksize_before_punch(struct inode
*inode
, loff_t offset
,
3663 loff_t size
= i_size_read(inode
);
3665 WARN_ON(!mutex_is_locked(&inode
->i_mutex
));
3666 if (offset
> size
|| offset
+ len
< size
)
3669 if (EXT4_I(inode
)->i_disksize
>= size
)
3672 handle
= ext4_journal_start(inode
, EXT4_HT_MISC
, 1);
3674 return PTR_ERR(handle
);
3675 ext4_update_i_disksize(inode
, size
);
3676 ext4_mark_inode_dirty(handle
, inode
);
3677 ext4_journal_stop(handle
);
3683 * ext4_punch_hole: punches a hole in a file by releasing the blocks
3684 * associated with the given offset and length
3686 * @inode: File inode
3687 * @offset: The offset where the hole will begin
3688 * @len: The length of the hole
3690 * Returns: 0 on success or negative on failure
3693 int ext4_punch_hole(struct inode
*inode
, loff_t offset
, loff_t length
)
3695 struct super_block
*sb
= inode
->i_sb
;
3696 ext4_lblk_t first_block
, stop_block
;
3697 struct address_space
*mapping
= inode
->i_mapping
;
3698 loff_t first_block_offset
, last_block_offset
;
3700 unsigned int credits
;
3703 if (!S_ISREG(inode
->i_mode
))
3706 trace_ext4_punch_hole(inode
, offset
, length
, 0);
3709 * Write out all dirty pages to avoid race conditions
3710 * Then release them.
3712 if (mapping_tagged(mapping
, PAGECACHE_TAG_DIRTY
)) {
3713 ret
= filemap_write_and_wait_range(mapping
, offset
,
3714 offset
+ length
- 1);
3719 mutex_lock(&inode
->i_mutex
);
3721 /* No need to punch hole beyond i_size */
3722 if (offset
>= inode
->i_size
)
3726 * If the hole extends beyond i_size, set the hole
3727 * to end after the page that contains i_size
3729 if (offset
+ length
> inode
->i_size
) {
3730 length
= inode
->i_size
+
3731 PAGE_CACHE_SIZE
- (inode
->i_size
& (PAGE_CACHE_SIZE
- 1)) -
3735 if (offset
& (sb
->s_blocksize
- 1) ||
3736 (offset
+ length
) & (sb
->s_blocksize
- 1)) {
3738 * Attach jinode to inode for jbd2 if we do any zeroing of
3741 ret
= ext4_inode_attach_jinode(inode
);
3747 /* Wait all existing dio workers, newcomers will block on i_mutex */
3748 ext4_inode_block_unlocked_dio(inode
);
3749 inode_dio_wait(inode
);
3752 * Prevent page faults from reinstantiating pages we have released from
3755 down_write(&EXT4_I(inode
)->i_mmap_sem
);
3756 first_block_offset
= round_up(offset
, sb
->s_blocksize
);
3757 last_block_offset
= round_down((offset
+ length
), sb
->s_blocksize
) - 1;
3759 /* Now release the pages and zero block aligned part of pages*/
3760 if (last_block_offset
> first_block_offset
) {
3761 ret
= ext4_update_disksize_before_punch(inode
, offset
, length
);
3764 truncate_pagecache_range(inode
, first_block_offset
,
3768 if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
))
3769 credits
= ext4_writepage_trans_blocks(inode
);
3771 credits
= ext4_blocks_for_truncate(inode
);
3772 handle
= ext4_journal_start(inode
, EXT4_HT_TRUNCATE
, credits
);
3773 if (IS_ERR(handle
)) {
3774 ret
= PTR_ERR(handle
);
3775 ext4_std_error(sb
, ret
);
3779 ret
= ext4_zero_partial_blocks(handle
, inode
, offset
,
3784 first_block
= (offset
+ sb
->s_blocksize
- 1) >>
3785 EXT4_BLOCK_SIZE_BITS(sb
);
3786 stop_block
= (offset
+ length
) >> EXT4_BLOCK_SIZE_BITS(sb
);
3788 /* If there are blocks to remove, do it */
3789 if (stop_block
> first_block
) {
3791 down_write(&EXT4_I(inode
)->i_data_sem
);
3792 ext4_discard_preallocations(inode
);
3794 ret
= ext4_es_remove_extent(inode
, first_block
,
3795 stop_block
- first_block
);
3797 up_write(&EXT4_I(inode
)->i_data_sem
);
3801 if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
))
3802 ret
= ext4_ext_remove_space(inode
, first_block
,
3805 ret
= ext4_ind_remove_space(handle
, inode
, first_block
,
3808 up_write(&EXT4_I(inode
)->i_data_sem
);
3811 ext4_handle_sync(handle
);
3813 inode
->i_mtime
= inode
->i_ctime
= ext4_current_time(inode
);
3814 ext4_mark_inode_dirty(handle
, inode
);
3816 ext4_update_inode_fsync_trans(handle
, inode
, 1);
3818 ext4_journal_stop(handle
);
3820 up_write(&EXT4_I(inode
)->i_mmap_sem
);
3821 ext4_inode_resume_unlocked_dio(inode
);
3823 mutex_unlock(&inode
->i_mutex
);
3827 int ext4_inode_attach_jinode(struct inode
*inode
)
3829 struct ext4_inode_info
*ei
= EXT4_I(inode
);
3830 struct jbd2_inode
*jinode
;
3832 if (ei
->jinode
|| !EXT4_SB(inode
->i_sb
)->s_journal
)
3835 jinode
= jbd2_alloc_inode(GFP_KERNEL
);
3836 spin_lock(&inode
->i_lock
);
3839 spin_unlock(&inode
->i_lock
);
3842 ei
->jinode
= jinode
;
3843 jbd2_journal_init_jbd_inode(ei
->jinode
, inode
);
3846 spin_unlock(&inode
->i_lock
);
3847 if (unlikely(jinode
!= NULL
))
3848 jbd2_free_inode(jinode
);
3855 * We block out ext4_get_block() block instantiations across the entire
3856 * transaction, and VFS/VM ensures that ext4_truncate() cannot run
3857 * simultaneously on behalf of the same inode.
3859 * As we work through the truncate and commit bits of it to the journal there
3860 * is one core, guiding principle: the file's tree must always be consistent on
3861 * disk. We must be able to restart the truncate after a crash.
3863 * The file's tree may be transiently inconsistent in memory (although it
3864 * probably isn't), but whenever we close off and commit a journal transaction,
3865 * the contents of (the filesystem + the journal) must be consistent and
3866 * restartable. It's pretty simple, really: bottom up, right to left (although
3867 * left-to-right works OK too).
3869 * Note that at recovery time, journal replay occurs *before* the restart of
3870 * truncate against the orphan inode list.
3872 * The committed inode has the new, desired i_size (which is the same as
3873 * i_disksize in this case). After a crash, ext4_orphan_cleanup() will see
3874 * that this inode's truncate did not complete and it will again call
3875 * ext4_truncate() to have another go. So there will be instantiated blocks
3876 * to the right of the truncation point in a crashed ext4 filesystem. But
3877 * that's fine - as long as they are linked from the inode, the post-crash
3878 * ext4_truncate() run will find them and release them.
3880 void ext4_truncate(struct inode
*inode
)
3882 struct ext4_inode_info
*ei
= EXT4_I(inode
);
3883 unsigned int credits
;
3885 struct address_space
*mapping
= inode
->i_mapping
;
3888 * There is a possibility that we're either freeing the inode
3889 * or it's a completely new inode. In those cases we might not
3890 * have i_mutex locked because it's not necessary.
3892 if (!(inode
->i_state
& (I_NEW
|I_FREEING
)))
3893 WARN_ON(!mutex_is_locked(&inode
->i_mutex
));
3894 trace_ext4_truncate_enter(inode
);
3896 if (!ext4_can_truncate(inode
))
3899 ext4_clear_inode_flag(inode
, EXT4_INODE_EOFBLOCKS
);
3901 if (inode
->i_size
== 0 && !test_opt(inode
->i_sb
, NO_AUTO_DA_ALLOC
))
3902 ext4_set_inode_state(inode
, EXT4_STATE_DA_ALLOC_CLOSE
);
3904 if (ext4_has_inline_data(inode
)) {
3907 ext4_inline_data_truncate(inode
, &has_inline
);
3912 /* If we zero-out tail of the page, we have to create jinode for jbd2 */
3913 if (inode
->i_size
& (inode
->i_sb
->s_blocksize
- 1)) {
3914 if (ext4_inode_attach_jinode(inode
) < 0)
3918 if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
))
3919 credits
= ext4_writepage_trans_blocks(inode
);
3921 credits
= ext4_blocks_for_truncate(inode
);
3923 handle
= ext4_journal_start(inode
, EXT4_HT_TRUNCATE
, credits
);
3924 if (IS_ERR(handle
)) {
3925 ext4_std_error(inode
->i_sb
, PTR_ERR(handle
));
3929 if (inode
->i_size
& (inode
->i_sb
->s_blocksize
- 1))
3930 ext4_block_truncate_page(handle
, mapping
, inode
->i_size
);
3933 * We add the inode to the orphan list, so that if this
3934 * truncate spans multiple transactions, and we crash, we will
3935 * resume the truncate when the filesystem recovers. It also
3936 * marks the inode dirty, to catch the new size.
3938 * Implication: the file must always be in a sane, consistent
3939 * truncatable state while each transaction commits.
3941 if (ext4_orphan_add(handle
, inode
))
3944 down_write(&EXT4_I(inode
)->i_data_sem
);
3946 ext4_discard_preallocations(inode
);
3948 if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
))
3949 ext4_ext_truncate(handle
, inode
);
3951 ext4_ind_truncate(handle
, inode
);
3953 up_write(&ei
->i_data_sem
);
3956 ext4_handle_sync(handle
);
3960 * If this was a simple ftruncate() and the file will remain alive,
3961 * then we need to clear up the orphan record which we created above.
3962 * However, if this was a real unlink then we were called by
3963 * ext4_evict_inode(), and we allow that function to clean up the
3964 * orphan info for us.
3967 ext4_orphan_del(handle
, inode
);
3969 inode
->i_mtime
= inode
->i_ctime
= ext4_current_time(inode
);
3970 ext4_mark_inode_dirty(handle
, inode
);
3971 ext4_journal_stop(handle
);
3973 trace_ext4_truncate_exit(inode
);
3977 * ext4_get_inode_loc returns with an extra refcount against the inode's
3978 * underlying buffer_head on success. If 'in_mem' is true, we have all
3979 * data in memory that is needed to recreate the on-disk version of this
3982 static int __ext4_get_inode_loc(struct inode
*inode
,
3983 struct ext4_iloc
*iloc
, int in_mem
)
3985 struct ext4_group_desc
*gdp
;
3986 struct buffer_head
*bh
;
3987 struct super_block
*sb
= inode
->i_sb
;
3989 int inodes_per_block
, inode_offset
;
3992 if (inode
->i_ino
< EXT4_ROOT_INO
||
3993 inode
->i_ino
> le32_to_cpu(EXT4_SB(sb
)->s_es
->s_inodes_count
))
3994 return -EFSCORRUPTED
;
3996 iloc
->block_group
= (inode
->i_ino
- 1) / EXT4_INODES_PER_GROUP(sb
);
3997 gdp
= ext4_get_group_desc(sb
, iloc
->block_group
, NULL
);
4002 * Figure out the offset within the block group inode table
4004 inodes_per_block
= EXT4_SB(sb
)->s_inodes_per_block
;
4005 inode_offset
= ((inode
->i_ino
- 1) %
4006 EXT4_INODES_PER_GROUP(sb
));
4007 block
= ext4_inode_table(sb
, gdp
) + (inode_offset
/ inodes_per_block
);
4008 iloc
->offset
= (inode_offset
% inodes_per_block
) * EXT4_INODE_SIZE(sb
);
4010 bh
= sb_getblk(sb
, block
);
4013 if (!buffer_uptodate(bh
)) {
4017 * If the buffer has the write error flag, we have failed
4018 * to write out another inode in the same block. In this
4019 * case, we don't have to read the block because we may
4020 * read the old inode data successfully.
4022 if (buffer_write_io_error(bh
) && !buffer_uptodate(bh
))
4023 set_buffer_uptodate(bh
);
4025 if (buffer_uptodate(bh
)) {
4026 /* someone brought it uptodate while we waited */
4032 * If we have all information of the inode in memory and this
4033 * is the only valid inode in the block, we need not read the
4037 struct buffer_head
*bitmap_bh
;
4040 start
= inode_offset
& ~(inodes_per_block
- 1);
4042 /* Is the inode bitmap in cache? */
4043 bitmap_bh
= sb_getblk(sb
, ext4_inode_bitmap(sb
, gdp
));
4044 if (unlikely(!bitmap_bh
))
4048 * If the inode bitmap isn't in cache then the
4049 * optimisation may end up performing two reads instead
4050 * of one, so skip it.
4052 if (!buffer_uptodate(bitmap_bh
)) {
4056 for (i
= start
; i
< start
+ inodes_per_block
; i
++) {
4057 if (i
== inode_offset
)
4059 if (ext4_test_bit(i
, bitmap_bh
->b_data
))
4063 if (i
== start
+ inodes_per_block
) {
4064 /* all other inodes are free, so skip I/O */
4065 memset(bh
->b_data
, 0, bh
->b_size
);
4066 set_buffer_uptodate(bh
);
4074 * If we need to do any I/O, try to pre-readahead extra
4075 * blocks from the inode table.
4077 if (EXT4_SB(sb
)->s_inode_readahead_blks
) {
4078 ext4_fsblk_t b
, end
, table
;
4080 __u32 ra_blks
= EXT4_SB(sb
)->s_inode_readahead_blks
;
4082 table
= ext4_inode_table(sb
, gdp
);
4083 /* s_inode_readahead_blks is always a power of 2 */
4084 b
= block
& ~((ext4_fsblk_t
) ra_blks
- 1);
4088 num
= EXT4_INODES_PER_GROUP(sb
);
4089 if (ext4_has_group_desc_csum(sb
))
4090 num
-= ext4_itable_unused_count(sb
, gdp
);
4091 table
+= num
/ inodes_per_block
;
4095 sb_breadahead(sb
, b
++);
4099 * There are other valid inodes in the buffer, this inode
4100 * has in-inode xattrs, or we don't have this inode in memory.
4101 * Read the block from disk.
4103 trace_ext4_load_inode(inode
);
4105 bh
->b_end_io
= end_buffer_read_sync
;
4106 submit_bh(READ
| REQ_META
| REQ_PRIO
, bh
);
4108 if (!buffer_uptodate(bh
)) {
4109 EXT4_ERROR_INODE_BLOCK(inode
, block
,
4110 "unable to read itable block");
4120 int ext4_get_inode_loc(struct inode
*inode
, struct ext4_iloc
*iloc
)
4122 /* We have all inode data except xattrs in memory here. */
4123 return __ext4_get_inode_loc(inode
, iloc
,
4124 !ext4_test_inode_state(inode
, EXT4_STATE_XATTR
));
4127 void ext4_set_inode_flags(struct inode
*inode
)
4129 unsigned int flags
= EXT4_I(inode
)->i_flags
;
4130 unsigned int new_fl
= 0;
4132 if (flags
& EXT4_SYNC_FL
)
4134 if (flags
& EXT4_APPEND_FL
)
4136 if (flags
& EXT4_IMMUTABLE_FL
)
4137 new_fl
|= S_IMMUTABLE
;
4138 if (flags
& EXT4_NOATIME_FL
)
4139 new_fl
|= S_NOATIME
;
4140 if (flags
& EXT4_DIRSYNC_FL
)
4141 new_fl
|= S_DIRSYNC
;
4142 if (test_opt(inode
->i_sb
, DAX
))
4144 inode_set_flags(inode
, new_fl
,
4145 S_SYNC
|S_APPEND
|S_IMMUTABLE
|S_NOATIME
|S_DIRSYNC
|S_DAX
);
4148 /* Propagate flags from i_flags to EXT4_I(inode)->i_flags */
4149 void ext4_get_inode_flags(struct ext4_inode_info
*ei
)
4151 unsigned int vfs_fl
;
4152 unsigned long old_fl
, new_fl
;
4155 vfs_fl
= ei
->vfs_inode
.i_flags
;
4156 old_fl
= ei
->i_flags
;
4157 new_fl
= old_fl
& ~(EXT4_SYNC_FL
|EXT4_APPEND_FL
|
4158 EXT4_IMMUTABLE_FL
|EXT4_NOATIME_FL
|
4160 if (vfs_fl
& S_SYNC
)
4161 new_fl
|= EXT4_SYNC_FL
;
4162 if (vfs_fl
& S_APPEND
)
4163 new_fl
|= EXT4_APPEND_FL
;
4164 if (vfs_fl
& S_IMMUTABLE
)
4165 new_fl
|= EXT4_IMMUTABLE_FL
;
4166 if (vfs_fl
& S_NOATIME
)
4167 new_fl
|= EXT4_NOATIME_FL
;
4168 if (vfs_fl
& S_DIRSYNC
)
4169 new_fl
|= EXT4_DIRSYNC_FL
;
4170 } while (cmpxchg(&ei
->i_flags
, old_fl
, new_fl
) != old_fl
);
4173 static blkcnt_t
ext4_inode_blocks(struct ext4_inode
*raw_inode
,
4174 struct ext4_inode_info
*ei
)
4177 struct inode
*inode
= &(ei
->vfs_inode
);
4178 struct super_block
*sb
= inode
->i_sb
;
4180 if (ext4_has_feature_huge_file(sb
)) {
4181 /* we are using combined 48 bit field */
4182 i_blocks
= ((u64
)le16_to_cpu(raw_inode
->i_blocks_high
)) << 32 |
4183 le32_to_cpu(raw_inode
->i_blocks_lo
);
4184 if (ext4_test_inode_flag(inode
, EXT4_INODE_HUGE_FILE
)) {
4185 /* i_blocks represent file system block size */
4186 return i_blocks
<< (inode
->i_blkbits
- 9);
4191 return le32_to_cpu(raw_inode
->i_blocks_lo
);
4195 static inline void ext4_iget_extra_inode(struct inode
*inode
,
4196 struct ext4_inode
*raw_inode
,
4197 struct ext4_inode_info
*ei
)
4199 __le32
*magic
= (void *)raw_inode
+
4200 EXT4_GOOD_OLD_INODE_SIZE
+ ei
->i_extra_isize
;
4201 if (*magic
== cpu_to_le32(EXT4_XATTR_MAGIC
)) {
4202 ext4_set_inode_state(inode
, EXT4_STATE_XATTR
);
4203 ext4_find_inline_data_nolock(inode
);
4205 EXT4_I(inode
)->i_inline_off
= 0;
4208 struct inode
*ext4_iget(struct super_block
*sb
, unsigned long ino
)
4210 struct ext4_iloc iloc
;
4211 struct ext4_inode
*raw_inode
;
4212 struct ext4_inode_info
*ei
;
4213 struct inode
*inode
;
4214 journal_t
*journal
= EXT4_SB(sb
)->s_journal
;
4221 inode
= iget_locked(sb
, ino
);
4223 return ERR_PTR(-ENOMEM
);
4224 if (!(inode
->i_state
& I_NEW
))
4230 ret
= __ext4_get_inode_loc(inode
, &iloc
, 0);
4233 raw_inode
= ext4_raw_inode(&iloc
);
4235 if ((ino
== EXT4_ROOT_INO
) && (raw_inode
->i_links_count
== 0)) {
4236 EXT4_ERROR_INODE(inode
, "root inode unallocated");
4237 ret
= -EFSCORRUPTED
;
4241 if (EXT4_INODE_SIZE(inode
->i_sb
) > EXT4_GOOD_OLD_INODE_SIZE
) {
4242 ei
->i_extra_isize
= le16_to_cpu(raw_inode
->i_extra_isize
);
4243 if (EXT4_GOOD_OLD_INODE_SIZE
+ ei
->i_extra_isize
>
4244 EXT4_INODE_SIZE(inode
->i_sb
)) {
4245 EXT4_ERROR_INODE(inode
, "bad extra_isize (%u != %u)",
4246 EXT4_GOOD_OLD_INODE_SIZE
+ ei
->i_extra_isize
,
4247 EXT4_INODE_SIZE(inode
->i_sb
));
4248 ret
= -EFSCORRUPTED
;
4252 ei
->i_extra_isize
= 0;
4254 /* Precompute checksum seed for inode metadata */
4255 if (ext4_has_metadata_csum(sb
)) {
4256 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
4258 __le32 inum
= cpu_to_le32(inode
->i_ino
);
4259 __le32 gen
= raw_inode
->i_generation
;
4260 csum
= ext4_chksum(sbi
, sbi
->s_csum_seed
, (__u8
*)&inum
,
4262 ei
->i_csum_seed
= ext4_chksum(sbi
, csum
, (__u8
*)&gen
,
4266 if (!ext4_inode_csum_verify(inode
, raw_inode
, ei
)) {
4267 EXT4_ERROR_INODE(inode
, "checksum invalid");
4272 inode
->i_mode
= le16_to_cpu(raw_inode
->i_mode
);
4273 i_uid
= (uid_t
)le16_to_cpu(raw_inode
->i_uid_low
);
4274 i_gid
= (gid_t
)le16_to_cpu(raw_inode
->i_gid_low
);
4275 if (!(test_opt(inode
->i_sb
, NO_UID32
))) {
4276 i_uid
|= le16_to_cpu(raw_inode
->i_uid_high
) << 16;
4277 i_gid
|= le16_to_cpu(raw_inode
->i_gid_high
) << 16;
4279 i_uid_write(inode
, i_uid
);
4280 i_gid_write(inode
, i_gid
);
4281 set_nlink(inode
, le16_to_cpu(raw_inode
->i_links_count
));
4283 ext4_clear_state_flags(ei
); /* Only relevant on 32-bit archs */
4284 ei
->i_inline_off
= 0;
4285 ei
->i_dir_start_lookup
= 0;
4286 ei
->i_dtime
= le32_to_cpu(raw_inode
->i_dtime
);
4287 /* We now have enough fields to check if the inode was active or not.
4288 * This is needed because nfsd might try to access dead inodes
4289 * the test is that same one that e2fsck uses
4290 * NeilBrown 1999oct15
4292 if (inode
->i_nlink
== 0) {
4293 if ((inode
->i_mode
== 0 ||
4294 !(EXT4_SB(inode
->i_sb
)->s_mount_state
& EXT4_ORPHAN_FS
)) &&
4295 ino
!= EXT4_BOOT_LOADER_INO
) {
4296 /* this inode is deleted */
4300 /* The only unlinked inodes we let through here have
4301 * valid i_mode and are being read by the orphan
4302 * recovery code: that's fine, we're about to complete
4303 * the process of deleting those.
4304 * OR it is the EXT4_BOOT_LOADER_INO which is
4305 * not initialized on a new filesystem. */
4307 ei
->i_flags
= le32_to_cpu(raw_inode
->i_flags
);
4308 inode
->i_blocks
= ext4_inode_blocks(raw_inode
, ei
);
4309 ei
->i_file_acl
= le32_to_cpu(raw_inode
->i_file_acl_lo
);
4310 if (ext4_has_feature_64bit(sb
))
4312 ((__u64
)le16_to_cpu(raw_inode
->i_file_acl_high
)) << 32;
4313 inode
->i_size
= ext4_isize(raw_inode
);
4314 if ((size
= i_size_read(inode
)) < 0) {
4315 EXT4_ERROR_INODE(inode
, "bad i_size value: %lld", size
);
4316 ret
= -EFSCORRUPTED
;
4319 ei
->i_disksize
= inode
->i_size
;
4321 ei
->i_reserved_quota
= 0;
4323 inode
->i_generation
= le32_to_cpu(raw_inode
->i_generation
);
4324 ei
->i_block_group
= iloc
.block_group
;
4325 ei
->i_last_alloc_group
= ~0;
4327 * NOTE! The in-memory inode i_data array is in little-endian order
4328 * even on big-endian machines: we do NOT byteswap the block numbers!
4330 for (block
= 0; block
< EXT4_N_BLOCKS
; block
++)
4331 ei
->i_data
[block
] = raw_inode
->i_block
[block
];
4332 INIT_LIST_HEAD(&ei
->i_orphan
);
4335 * Set transaction id's of transactions that have to be committed
4336 * to finish f[data]sync. We set them to currently running transaction
4337 * as we cannot be sure that the inode or some of its metadata isn't
4338 * part of the transaction - the inode could have been reclaimed and
4339 * now it is reread from disk.
4342 transaction_t
*transaction
;
4345 read_lock(&journal
->j_state_lock
);
4346 if (journal
->j_running_transaction
)
4347 transaction
= journal
->j_running_transaction
;
4349 transaction
= journal
->j_committing_transaction
;
4351 tid
= transaction
->t_tid
;
4353 tid
= journal
->j_commit_sequence
;
4354 read_unlock(&journal
->j_state_lock
);
4355 ei
->i_sync_tid
= tid
;
4356 ei
->i_datasync_tid
= tid
;
4359 if (EXT4_INODE_SIZE(inode
->i_sb
) > EXT4_GOOD_OLD_INODE_SIZE
) {
4360 if (ei
->i_extra_isize
== 0) {
4361 /* The extra space is currently unused. Use it. */
4362 ei
->i_extra_isize
= sizeof(struct ext4_inode
) -
4363 EXT4_GOOD_OLD_INODE_SIZE
;
4365 ext4_iget_extra_inode(inode
, raw_inode
, ei
);
4369 EXT4_INODE_GET_XTIME(i_ctime
, inode
, raw_inode
);
4370 EXT4_INODE_GET_XTIME(i_mtime
, inode
, raw_inode
);
4371 EXT4_INODE_GET_XTIME(i_atime
, inode
, raw_inode
);
4372 EXT4_EINODE_GET_XTIME(i_crtime
, ei
, raw_inode
);
4374 if (likely(!test_opt2(inode
->i_sb
, HURD_COMPAT
))) {
4375 inode
->i_version
= le32_to_cpu(raw_inode
->i_disk_version
);
4376 if (EXT4_INODE_SIZE(inode
->i_sb
) > EXT4_GOOD_OLD_INODE_SIZE
) {
4377 if (EXT4_FITS_IN_INODE(raw_inode
, ei
, i_version_hi
))
4379 (__u64
)(le32_to_cpu(raw_inode
->i_version_hi
)) << 32;
4384 if (ei
->i_file_acl
&&
4385 !ext4_data_block_valid(EXT4_SB(sb
), ei
->i_file_acl
, 1)) {
4386 EXT4_ERROR_INODE(inode
, "bad extended attribute block %llu",
4388 ret
= -EFSCORRUPTED
;
4390 } else if (!ext4_has_inline_data(inode
)) {
4391 if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
)) {
4392 if ((S_ISREG(inode
->i_mode
) || S_ISDIR(inode
->i_mode
) ||
4393 (S_ISLNK(inode
->i_mode
) &&
4394 !ext4_inode_is_fast_symlink(inode
))))
4395 /* Validate extent which is part of inode */
4396 ret
= ext4_ext_check_inode(inode
);
4397 } else if (S_ISREG(inode
->i_mode
) || S_ISDIR(inode
->i_mode
) ||
4398 (S_ISLNK(inode
->i_mode
) &&
4399 !ext4_inode_is_fast_symlink(inode
))) {
4400 /* Validate block references which are part of inode */
4401 ret
= ext4_ind_check_inode(inode
);
4407 if (S_ISREG(inode
->i_mode
)) {
4408 inode
->i_op
= &ext4_file_inode_operations
;
4409 inode
->i_fop
= &ext4_file_operations
;
4410 ext4_set_aops(inode
);
4411 } else if (S_ISDIR(inode
->i_mode
)) {
4412 inode
->i_op
= &ext4_dir_inode_operations
;
4413 inode
->i_fop
= &ext4_dir_operations
;
4414 } else if (S_ISLNK(inode
->i_mode
)) {
4415 if (ext4_encrypted_inode(inode
)) {
4416 inode
->i_op
= &ext4_encrypted_symlink_inode_operations
;
4417 ext4_set_aops(inode
);
4418 } else if (ext4_inode_is_fast_symlink(inode
)) {
4419 inode
->i_link
= (char *)ei
->i_data
;
4420 inode
->i_op
= &ext4_fast_symlink_inode_operations
;
4421 nd_terminate_link(ei
->i_data
, inode
->i_size
,
4422 sizeof(ei
->i_data
) - 1);
4424 inode
->i_op
= &ext4_symlink_inode_operations
;
4425 ext4_set_aops(inode
);
4427 inode_nohighmem(inode
);
4428 } else if (S_ISCHR(inode
->i_mode
) || S_ISBLK(inode
->i_mode
) ||
4429 S_ISFIFO(inode
->i_mode
) || S_ISSOCK(inode
->i_mode
)) {
4430 inode
->i_op
= &ext4_special_inode_operations
;
4431 if (raw_inode
->i_block
[0])
4432 init_special_inode(inode
, inode
->i_mode
,
4433 old_decode_dev(le32_to_cpu(raw_inode
->i_block
[0])));
4435 init_special_inode(inode
, inode
->i_mode
,
4436 new_decode_dev(le32_to_cpu(raw_inode
->i_block
[1])));
4437 } else if (ino
== EXT4_BOOT_LOADER_INO
) {
4438 make_bad_inode(inode
);
4440 ret
= -EFSCORRUPTED
;
4441 EXT4_ERROR_INODE(inode
, "bogus i_mode (%o)", inode
->i_mode
);
4445 ext4_set_inode_flags(inode
);
4446 unlock_new_inode(inode
);
4452 return ERR_PTR(ret
);
4455 struct inode
*ext4_iget_normal(struct super_block
*sb
, unsigned long ino
)
4457 if (ino
< EXT4_FIRST_INO(sb
) && ino
!= EXT4_ROOT_INO
)
4458 return ERR_PTR(-EFSCORRUPTED
);
4459 return ext4_iget(sb
, ino
);
4462 static int ext4_inode_blocks_set(handle_t
*handle
,
4463 struct ext4_inode
*raw_inode
,
4464 struct ext4_inode_info
*ei
)
4466 struct inode
*inode
= &(ei
->vfs_inode
);
4467 u64 i_blocks
= inode
->i_blocks
;
4468 struct super_block
*sb
= inode
->i_sb
;
4470 if (i_blocks
<= ~0U) {
4472 * i_blocks can be represented in a 32 bit variable
4473 * as multiple of 512 bytes
4475 raw_inode
->i_blocks_lo
= cpu_to_le32(i_blocks
);
4476 raw_inode
->i_blocks_high
= 0;
4477 ext4_clear_inode_flag(inode
, EXT4_INODE_HUGE_FILE
);
4480 if (!ext4_has_feature_huge_file(sb
))
4483 if (i_blocks
<= 0xffffffffffffULL
) {
4485 * i_blocks can be represented in a 48 bit variable
4486 * as multiple of 512 bytes
4488 raw_inode
->i_blocks_lo
= cpu_to_le32(i_blocks
);
4489 raw_inode
->i_blocks_high
= cpu_to_le16(i_blocks
>> 32);
4490 ext4_clear_inode_flag(inode
, EXT4_INODE_HUGE_FILE
);
4492 ext4_set_inode_flag(inode
, EXT4_INODE_HUGE_FILE
);
4493 /* i_block is stored in file system block size */
4494 i_blocks
= i_blocks
>> (inode
->i_blkbits
- 9);
4495 raw_inode
->i_blocks_lo
= cpu_to_le32(i_blocks
);
4496 raw_inode
->i_blocks_high
= cpu_to_le16(i_blocks
>> 32);
4501 struct other_inode
{
4502 unsigned long orig_ino
;
4503 struct ext4_inode
*raw_inode
;
4506 static int other_inode_match(struct inode
* inode
, unsigned long ino
,
4509 struct other_inode
*oi
= (struct other_inode
*) data
;
4511 if ((inode
->i_ino
!= ino
) ||
4512 (inode
->i_state
& (I_FREEING
| I_WILL_FREE
| I_NEW
|
4513 I_DIRTY_SYNC
| I_DIRTY_DATASYNC
)) ||
4514 ((inode
->i_state
& I_DIRTY_TIME
) == 0))
4516 spin_lock(&inode
->i_lock
);
4517 if (((inode
->i_state
& (I_FREEING
| I_WILL_FREE
| I_NEW
|
4518 I_DIRTY_SYNC
| I_DIRTY_DATASYNC
)) == 0) &&
4519 (inode
->i_state
& I_DIRTY_TIME
)) {
4520 struct ext4_inode_info
*ei
= EXT4_I(inode
);
4522 inode
->i_state
&= ~(I_DIRTY_TIME
| I_DIRTY_TIME_EXPIRED
);
4523 spin_unlock(&inode
->i_lock
);
4525 spin_lock(&ei
->i_raw_lock
);
4526 EXT4_INODE_SET_XTIME(i_ctime
, inode
, oi
->raw_inode
);
4527 EXT4_INODE_SET_XTIME(i_mtime
, inode
, oi
->raw_inode
);
4528 EXT4_INODE_SET_XTIME(i_atime
, inode
, oi
->raw_inode
);
4529 ext4_inode_csum_set(inode
, oi
->raw_inode
, ei
);
4530 spin_unlock(&ei
->i_raw_lock
);
4531 trace_ext4_other_inode_update_time(inode
, oi
->orig_ino
);
4534 spin_unlock(&inode
->i_lock
);
4539 * Opportunistically update the other time fields for other inodes in
4540 * the same inode table block.
4542 static void ext4_update_other_inodes_time(struct super_block
*sb
,
4543 unsigned long orig_ino
, char *buf
)
4545 struct other_inode oi
;
4547 int i
, inodes_per_block
= EXT4_SB(sb
)->s_inodes_per_block
;
4548 int inode_size
= EXT4_INODE_SIZE(sb
);
4550 oi
.orig_ino
= orig_ino
;
4552 * Calculate the first inode in the inode table block. Inode
4553 * numbers are one-based. That is, the first inode in a block
4554 * (assuming 4k blocks and 256 byte inodes) is (n*16 + 1).
4556 ino
= ((orig_ino
- 1) & ~(inodes_per_block
- 1)) + 1;
4557 for (i
= 0; i
< inodes_per_block
; i
++, ino
++, buf
+= inode_size
) {
4558 if (ino
== orig_ino
)
4560 oi
.raw_inode
= (struct ext4_inode
*) buf
;
4561 (void) find_inode_nowait(sb
, ino
, other_inode_match
, &oi
);
4566 * Post the struct inode info into an on-disk inode location in the
4567 * buffer-cache. This gobbles the caller's reference to the
4568 * buffer_head in the inode location struct.
4570 * The caller must have write access to iloc->bh.
4572 static int ext4_do_update_inode(handle_t
*handle
,
4573 struct inode
*inode
,
4574 struct ext4_iloc
*iloc
)
4576 struct ext4_inode
*raw_inode
= ext4_raw_inode(iloc
);
4577 struct ext4_inode_info
*ei
= EXT4_I(inode
);
4578 struct buffer_head
*bh
= iloc
->bh
;
4579 struct super_block
*sb
= inode
->i_sb
;
4580 int err
= 0, rc
, block
;
4581 int need_datasync
= 0, set_large_file
= 0;
4585 spin_lock(&ei
->i_raw_lock
);
4587 /* For fields not tracked in the in-memory inode,
4588 * initialise them to zero for new inodes. */
4589 if (ext4_test_inode_state(inode
, EXT4_STATE_NEW
))
4590 memset(raw_inode
, 0, EXT4_SB(inode
->i_sb
)->s_inode_size
);
4592 ext4_get_inode_flags(ei
);
4593 raw_inode
->i_mode
= cpu_to_le16(inode
->i_mode
);
4594 i_uid
= i_uid_read(inode
);
4595 i_gid
= i_gid_read(inode
);
4596 if (!(test_opt(inode
->i_sb
, NO_UID32
))) {
4597 raw_inode
->i_uid_low
= cpu_to_le16(low_16_bits(i_uid
));
4598 raw_inode
->i_gid_low
= cpu_to_le16(low_16_bits(i_gid
));
4600 * Fix up interoperability with old kernels. Otherwise, old inodes get
4601 * re-used with the upper 16 bits of the uid/gid intact
4603 if (ei
->i_dtime
&& list_empty(&ei
->i_orphan
)) {
4604 raw_inode
->i_uid_high
= 0;
4605 raw_inode
->i_gid_high
= 0;
4607 raw_inode
->i_uid_high
=
4608 cpu_to_le16(high_16_bits(i_uid
));
4609 raw_inode
->i_gid_high
=
4610 cpu_to_le16(high_16_bits(i_gid
));
4613 raw_inode
->i_uid_low
= cpu_to_le16(fs_high2lowuid(i_uid
));
4614 raw_inode
->i_gid_low
= cpu_to_le16(fs_high2lowgid(i_gid
));
4615 raw_inode
->i_uid_high
= 0;
4616 raw_inode
->i_gid_high
= 0;
4618 raw_inode
->i_links_count
= cpu_to_le16(inode
->i_nlink
);
4620 EXT4_INODE_SET_XTIME(i_ctime
, inode
, raw_inode
);
4621 EXT4_INODE_SET_XTIME(i_mtime
, inode
, raw_inode
);
4622 EXT4_INODE_SET_XTIME(i_atime
, inode
, raw_inode
);
4623 EXT4_EINODE_SET_XTIME(i_crtime
, ei
, raw_inode
);
4625 err
= ext4_inode_blocks_set(handle
, raw_inode
, ei
);
4627 spin_unlock(&ei
->i_raw_lock
);
4630 raw_inode
->i_dtime
= cpu_to_le32(ei
->i_dtime
);
4631 raw_inode
->i_flags
= cpu_to_le32(ei
->i_flags
& 0xFFFFFFFF);
4632 if (likely(!test_opt2(inode
->i_sb
, HURD_COMPAT
)))
4633 raw_inode
->i_file_acl_high
=
4634 cpu_to_le16(ei
->i_file_acl
>> 32);
4635 raw_inode
->i_file_acl_lo
= cpu_to_le32(ei
->i_file_acl
);
4636 if (ei
->i_disksize
!= ext4_isize(raw_inode
)) {
4637 ext4_isize_set(raw_inode
, ei
->i_disksize
);
4640 if (ei
->i_disksize
> 0x7fffffffULL
) {
4641 if (!ext4_has_feature_large_file(sb
) ||
4642 EXT4_SB(sb
)->s_es
->s_rev_level
==
4643 cpu_to_le32(EXT4_GOOD_OLD_REV
))
4646 raw_inode
->i_generation
= cpu_to_le32(inode
->i_generation
);
4647 if (S_ISCHR(inode
->i_mode
) || S_ISBLK(inode
->i_mode
)) {
4648 if (old_valid_dev(inode
->i_rdev
)) {
4649 raw_inode
->i_block
[0] =
4650 cpu_to_le32(old_encode_dev(inode
->i_rdev
));
4651 raw_inode
->i_block
[1] = 0;
4653 raw_inode
->i_block
[0] = 0;
4654 raw_inode
->i_block
[1] =
4655 cpu_to_le32(new_encode_dev(inode
->i_rdev
));
4656 raw_inode
->i_block
[2] = 0;
4658 } else if (!ext4_has_inline_data(inode
)) {
4659 for (block
= 0; block
< EXT4_N_BLOCKS
; block
++)
4660 raw_inode
->i_block
[block
] = ei
->i_data
[block
];
4663 if (likely(!test_opt2(inode
->i_sb
, HURD_COMPAT
))) {
4664 raw_inode
->i_disk_version
= cpu_to_le32(inode
->i_version
);
4665 if (ei
->i_extra_isize
) {
4666 if (EXT4_FITS_IN_INODE(raw_inode
, ei
, i_version_hi
))
4667 raw_inode
->i_version_hi
=
4668 cpu_to_le32(inode
->i_version
>> 32);
4669 raw_inode
->i_extra_isize
=
4670 cpu_to_le16(ei
->i_extra_isize
);
4673 ext4_inode_csum_set(inode
, raw_inode
, ei
);
4674 spin_unlock(&ei
->i_raw_lock
);
4675 if (inode
->i_sb
->s_flags
& MS_LAZYTIME
)
4676 ext4_update_other_inodes_time(inode
->i_sb
, inode
->i_ino
,
4679 BUFFER_TRACE(bh
, "call ext4_handle_dirty_metadata");
4680 rc
= ext4_handle_dirty_metadata(handle
, NULL
, bh
);
4683 ext4_clear_inode_state(inode
, EXT4_STATE_NEW
);
4684 if (set_large_file
) {
4685 BUFFER_TRACE(EXT4_SB(sb
)->s_sbh
, "get write access");
4686 err
= ext4_journal_get_write_access(handle
, EXT4_SB(sb
)->s_sbh
);
4689 ext4_update_dynamic_rev(sb
);
4690 ext4_set_feature_large_file(sb
);
4691 ext4_handle_sync(handle
);
4692 err
= ext4_handle_dirty_super(handle
, sb
);
4694 ext4_update_inode_fsync_trans(handle
, inode
, need_datasync
);
4697 ext4_std_error(inode
->i_sb
, err
);
4702 * ext4_write_inode()
4704 * We are called from a few places:
4706 * - Within generic_file_aio_write() -> generic_write_sync() for O_SYNC files.
4707 * Here, there will be no transaction running. We wait for any running
4708 * transaction to commit.
4710 * - Within flush work (sys_sync(), kupdate and such).
4711 * We wait on commit, if told to.
4713 * - Within iput_final() -> write_inode_now()
4714 * We wait on commit, if told to.
4716 * In all cases it is actually safe for us to return without doing anything,
4717 * because the inode has been copied into a raw inode buffer in
4718 * ext4_mark_inode_dirty(). This is a correctness thing for WB_SYNC_ALL
4721 * Note that we are absolutely dependent upon all inode dirtiers doing the
4722 * right thing: they *must* call mark_inode_dirty() after dirtying info in
4723 * which we are interested.
4725 * It would be a bug for them to not do this. The code:
4727 * mark_inode_dirty(inode)
4729 * inode->i_size = expr;
4731 * is in error because write_inode() could occur while `stuff()' is running,
4732 * and the new i_size will be lost. Plus the inode will no longer be on the
4733 * superblock's dirty inode list.
4735 int ext4_write_inode(struct inode
*inode
, struct writeback_control
*wbc
)
4739 if (WARN_ON_ONCE(current
->flags
& PF_MEMALLOC
))
4742 if (EXT4_SB(inode
->i_sb
)->s_journal
) {
4743 if (ext4_journal_current_handle()) {
4744 jbd_debug(1, "called recursively, non-PF_MEMALLOC!\n");
4750 * No need to force transaction in WB_SYNC_NONE mode. Also
4751 * ext4_sync_fs() will force the commit after everything is
4754 if (wbc
->sync_mode
!= WB_SYNC_ALL
|| wbc
->for_sync
)
4757 err
= ext4_force_commit(inode
->i_sb
);
4759 struct ext4_iloc iloc
;
4761 err
= __ext4_get_inode_loc(inode
, &iloc
, 0);
4765 * sync(2) will flush the whole buffer cache. No need to do
4766 * it here separately for each inode.
4768 if (wbc
->sync_mode
== WB_SYNC_ALL
&& !wbc
->for_sync
)
4769 sync_dirty_buffer(iloc
.bh
);
4770 if (buffer_req(iloc
.bh
) && !buffer_uptodate(iloc
.bh
)) {
4771 EXT4_ERROR_INODE_BLOCK(inode
, iloc
.bh
->b_blocknr
,
4772 "IO error syncing inode");
4781 * In data=journal mode ext4_journalled_invalidatepage() may fail to invalidate
4782 * buffers that are attached to a page stradding i_size and are undergoing
4783 * commit. In that case we have to wait for commit to finish and try again.
4785 static void ext4_wait_for_tail_page_commit(struct inode
*inode
)
4789 journal_t
*journal
= EXT4_SB(inode
->i_sb
)->s_journal
;
4790 tid_t commit_tid
= 0;
4793 offset
= inode
->i_size
& (PAGE_CACHE_SIZE
- 1);
4795 * All buffers in the last page remain valid? Then there's nothing to
4796 * do. We do the check mainly to optimize the common PAGE_CACHE_SIZE ==
4799 if (offset
> PAGE_CACHE_SIZE
- (1 << inode
->i_blkbits
))
4802 page
= find_lock_page(inode
->i_mapping
,
4803 inode
->i_size
>> PAGE_CACHE_SHIFT
);
4806 ret
= __ext4_journalled_invalidatepage(page
, offset
,
4807 PAGE_CACHE_SIZE
- offset
);
4809 page_cache_release(page
);
4813 read_lock(&journal
->j_state_lock
);
4814 if (journal
->j_committing_transaction
)
4815 commit_tid
= journal
->j_committing_transaction
->t_tid
;
4816 read_unlock(&journal
->j_state_lock
);
4818 jbd2_log_wait_commit(journal
, commit_tid
);
4825 * Called from notify_change.
4827 * We want to trap VFS attempts to truncate the file as soon as
4828 * possible. In particular, we want to make sure that when the VFS
4829 * shrinks i_size, we put the inode on the orphan list and modify
4830 * i_disksize immediately, so that during the subsequent flushing of
4831 * dirty pages and freeing of disk blocks, we can guarantee that any
4832 * commit will leave the blocks being flushed in an unused state on
4833 * disk. (On recovery, the inode will get truncated and the blocks will
4834 * be freed, so we have a strong guarantee that no future commit will
4835 * leave these blocks visible to the user.)
4837 * Another thing we have to assure is that if we are in ordered mode
4838 * and inode is still attached to the committing transaction, we must
4839 * we start writeout of all the dirty pages which are being truncated.
4840 * This way we are sure that all the data written in the previous
4841 * transaction are already on disk (truncate waits for pages under
4844 * Called with inode->i_mutex down.
4846 int ext4_setattr(struct dentry
*dentry
, struct iattr
*attr
)
4848 struct inode
*inode
= d_inode(dentry
);
4851 const unsigned int ia_valid
= attr
->ia_valid
;
4853 error
= inode_change_ok(inode
, attr
);
4857 if (is_quota_modification(inode
, attr
)) {
4858 error
= dquot_initialize(inode
);
4862 if ((ia_valid
& ATTR_UID
&& !uid_eq(attr
->ia_uid
, inode
->i_uid
)) ||
4863 (ia_valid
& ATTR_GID
&& !gid_eq(attr
->ia_gid
, inode
->i_gid
))) {
4866 /* (user+group)*(old+new) structure, inode write (sb,
4867 * inode block, ? - but truncate inode update has it) */
4868 handle
= ext4_journal_start(inode
, EXT4_HT_QUOTA
,
4869 (EXT4_MAXQUOTAS_INIT_BLOCKS(inode
->i_sb
) +
4870 EXT4_MAXQUOTAS_DEL_BLOCKS(inode
->i_sb
)) + 3);
4871 if (IS_ERR(handle
)) {
4872 error
= PTR_ERR(handle
);
4875 error
= dquot_transfer(inode
, attr
);
4877 ext4_journal_stop(handle
);
4880 /* Update corresponding info in inode so that everything is in
4881 * one transaction */
4882 if (attr
->ia_valid
& ATTR_UID
)
4883 inode
->i_uid
= attr
->ia_uid
;
4884 if (attr
->ia_valid
& ATTR_GID
)
4885 inode
->i_gid
= attr
->ia_gid
;
4886 error
= ext4_mark_inode_dirty(handle
, inode
);
4887 ext4_journal_stop(handle
);
4890 if (attr
->ia_valid
& ATTR_SIZE
) {
4892 loff_t oldsize
= inode
->i_size
;
4893 int shrink
= (attr
->ia_size
<= inode
->i_size
);
4895 if (!(ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
))) {
4896 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
4898 if (attr
->ia_size
> sbi
->s_bitmap_maxbytes
)
4901 if (!S_ISREG(inode
->i_mode
))
4904 if (IS_I_VERSION(inode
) && attr
->ia_size
!= inode
->i_size
)
4905 inode_inc_iversion(inode
);
4907 if (ext4_should_order_data(inode
) &&
4908 (attr
->ia_size
< inode
->i_size
)) {
4909 error
= ext4_begin_ordered_truncate(inode
,
4914 if (attr
->ia_size
!= inode
->i_size
) {
4915 handle
= ext4_journal_start(inode
, EXT4_HT_INODE
, 3);
4916 if (IS_ERR(handle
)) {
4917 error
= PTR_ERR(handle
);
4920 if (ext4_handle_valid(handle
) && shrink
) {
4921 error
= ext4_orphan_add(handle
, inode
);
4925 * Update c/mtime on truncate up, ext4_truncate() will
4926 * update c/mtime in shrink case below
4929 inode
->i_mtime
= ext4_current_time(inode
);
4930 inode
->i_ctime
= inode
->i_mtime
;
4932 down_write(&EXT4_I(inode
)->i_data_sem
);
4933 EXT4_I(inode
)->i_disksize
= attr
->ia_size
;
4934 rc
= ext4_mark_inode_dirty(handle
, inode
);
4938 * We have to update i_size under i_data_sem together
4939 * with i_disksize to avoid races with writeback code
4940 * running ext4_wb_update_i_disksize().
4943 i_size_write(inode
, attr
->ia_size
);
4944 up_write(&EXT4_I(inode
)->i_data_sem
);
4945 ext4_journal_stop(handle
);
4948 ext4_orphan_del(NULL
, inode
);
4953 pagecache_isize_extended(inode
, oldsize
, inode
->i_size
);
4956 * Blocks are going to be removed from the inode. Wait
4957 * for dio in flight. Temporarily disable
4958 * dioread_nolock to prevent livelock.
4961 if (!ext4_should_journal_data(inode
)) {
4962 ext4_inode_block_unlocked_dio(inode
);
4963 inode_dio_wait(inode
);
4964 ext4_inode_resume_unlocked_dio(inode
);
4966 ext4_wait_for_tail_page_commit(inode
);
4968 down_write(&EXT4_I(inode
)->i_mmap_sem
);
4970 * Truncate pagecache after we've waited for commit
4971 * in data=journal mode to make pages freeable.
4973 truncate_pagecache(inode
, inode
->i_size
);
4975 ext4_truncate(inode
);
4976 up_write(&EXT4_I(inode
)->i_mmap_sem
);
4980 setattr_copy(inode
, attr
);
4981 mark_inode_dirty(inode
);
4985 * If the call to ext4_truncate failed to get a transaction handle at
4986 * all, we need to clean up the in-core orphan list manually.
4988 if (orphan
&& inode
->i_nlink
)
4989 ext4_orphan_del(NULL
, inode
);
4991 if (!rc
&& (ia_valid
& ATTR_MODE
))
4992 rc
= posix_acl_chmod(inode
, inode
->i_mode
);
4995 ext4_std_error(inode
->i_sb
, error
);
5001 int ext4_getattr(struct vfsmount
*mnt
, struct dentry
*dentry
,
5004 struct inode
*inode
;
5005 unsigned long long delalloc_blocks
;
5007 inode
= d_inode(dentry
);
5008 generic_fillattr(inode
, stat
);
5011 * If there is inline data in the inode, the inode will normally not
5012 * have data blocks allocated (it may have an external xattr block).
5013 * Report at least one sector for such files, so tools like tar, rsync,
5014 * others doen't incorrectly think the file is completely sparse.
5016 if (unlikely(ext4_has_inline_data(inode
)))
5017 stat
->blocks
+= (stat
->size
+ 511) >> 9;
5020 * We can't update i_blocks if the block allocation is delayed
5021 * otherwise in the case of system crash before the real block
5022 * allocation is done, we will have i_blocks inconsistent with
5023 * on-disk file blocks.
5024 * We always keep i_blocks updated together with real
5025 * allocation. But to not confuse with user, stat
5026 * will return the blocks that include the delayed allocation
5027 * blocks for this file.
5029 delalloc_blocks
= EXT4_C2B(EXT4_SB(inode
->i_sb
),
5030 EXT4_I(inode
)->i_reserved_data_blocks
);
5031 stat
->blocks
+= delalloc_blocks
<< (inode
->i_sb
->s_blocksize_bits
- 9);
5035 static int ext4_index_trans_blocks(struct inode
*inode
, int lblocks
,
5038 if (!(ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
)))
5039 return ext4_ind_trans_blocks(inode
, lblocks
);
5040 return ext4_ext_index_trans_blocks(inode
, pextents
);
5044 * Account for index blocks, block groups bitmaps and block group
5045 * descriptor blocks if modify datablocks and index blocks
5046 * worse case, the indexs blocks spread over different block groups
5048 * If datablocks are discontiguous, they are possible to spread over
5049 * different block groups too. If they are contiguous, with flexbg,
5050 * they could still across block group boundary.
5052 * Also account for superblock, inode, quota and xattr blocks
5054 static int ext4_meta_trans_blocks(struct inode
*inode
, int lblocks
,
5057 ext4_group_t groups
, ngroups
= ext4_get_groups_count(inode
->i_sb
);
5063 * How many index blocks need to touch to map @lblocks logical blocks
5064 * to @pextents physical extents?
5066 idxblocks
= ext4_index_trans_blocks(inode
, lblocks
, pextents
);
5071 * Now let's see how many group bitmaps and group descriptors need
5074 groups
= idxblocks
+ pextents
;
5076 if (groups
> ngroups
)
5078 if (groups
> EXT4_SB(inode
->i_sb
)->s_gdb_count
)
5079 gdpblocks
= EXT4_SB(inode
->i_sb
)->s_gdb_count
;
5081 /* bitmaps and block group descriptor blocks */
5082 ret
+= groups
+ gdpblocks
;
5084 /* Blocks for super block, inode, quota and xattr blocks */
5085 ret
+= EXT4_META_TRANS_BLOCKS(inode
->i_sb
);
5091 * Calculate the total number of credits to reserve to fit
5092 * the modification of a single pages into a single transaction,
5093 * which may include multiple chunks of block allocations.
5095 * This could be called via ext4_write_begin()
5097 * We need to consider the worse case, when
5098 * one new block per extent.
5100 int ext4_writepage_trans_blocks(struct inode
*inode
)
5102 int bpp
= ext4_journal_blocks_per_page(inode
);
5105 ret
= ext4_meta_trans_blocks(inode
, bpp
, bpp
);
5107 /* Account for data blocks for journalled mode */
5108 if (ext4_should_journal_data(inode
))
5114 * Calculate the journal credits for a chunk of data modification.
5116 * This is called from DIO, fallocate or whoever calling
5117 * ext4_map_blocks() to map/allocate a chunk of contiguous disk blocks.
5119 * journal buffers for data blocks are not included here, as DIO
5120 * and fallocate do no need to journal data buffers.
5122 int ext4_chunk_trans_blocks(struct inode
*inode
, int nrblocks
)
5124 return ext4_meta_trans_blocks(inode
, nrblocks
, 1);
5128 * The caller must have previously called ext4_reserve_inode_write().
5129 * Give this, we know that the caller already has write access to iloc->bh.
5131 int ext4_mark_iloc_dirty(handle_t
*handle
,
5132 struct inode
*inode
, struct ext4_iloc
*iloc
)
5136 if (IS_I_VERSION(inode
))
5137 inode_inc_iversion(inode
);
5139 /* the do_update_inode consumes one bh->b_count */
5142 /* ext4_do_update_inode() does jbd2_journal_dirty_metadata */
5143 err
= ext4_do_update_inode(handle
, inode
, iloc
);
5149 * On success, We end up with an outstanding reference count against
5150 * iloc->bh. This _must_ be cleaned up later.
5154 ext4_reserve_inode_write(handle_t
*handle
, struct inode
*inode
,
5155 struct ext4_iloc
*iloc
)
5159 err
= ext4_get_inode_loc(inode
, iloc
);
5161 BUFFER_TRACE(iloc
->bh
, "get_write_access");
5162 err
= ext4_journal_get_write_access(handle
, iloc
->bh
);
5168 ext4_std_error(inode
->i_sb
, err
);
5173 * Expand an inode by new_extra_isize bytes.
5174 * Returns 0 on success or negative error number on failure.
5176 static int ext4_expand_extra_isize(struct inode
*inode
,
5177 unsigned int new_extra_isize
,
5178 struct ext4_iloc iloc
,
5181 struct ext4_inode
*raw_inode
;
5182 struct ext4_xattr_ibody_header
*header
;
5184 if (EXT4_I(inode
)->i_extra_isize
>= new_extra_isize
)
5187 raw_inode
= ext4_raw_inode(&iloc
);
5189 header
= IHDR(inode
, raw_inode
);
5191 /* No extended attributes present */
5192 if (!ext4_test_inode_state(inode
, EXT4_STATE_XATTR
) ||
5193 header
->h_magic
!= cpu_to_le32(EXT4_XATTR_MAGIC
)) {
5194 memset((void *)raw_inode
+ EXT4_GOOD_OLD_INODE_SIZE
+
5195 EXT4_I(inode
)->i_extra_isize
, 0,
5196 new_extra_isize
- EXT4_I(inode
)->i_extra_isize
);
5197 EXT4_I(inode
)->i_extra_isize
= new_extra_isize
;
5201 /* try to expand with EAs present */
5202 return ext4_expand_extra_isize_ea(inode
, new_extra_isize
,
5207 * What we do here is to mark the in-core inode as clean with respect to inode
5208 * dirtiness (it may still be data-dirty).
5209 * This means that the in-core inode may be reaped by prune_icache
5210 * without having to perform any I/O. This is a very good thing,
5211 * because *any* task may call prune_icache - even ones which
5212 * have a transaction open against a different journal.
5214 * Is this cheating? Not really. Sure, we haven't written the
5215 * inode out, but prune_icache isn't a user-visible syncing function.
5216 * Whenever the user wants stuff synced (sys_sync, sys_msync, sys_fsync)
5217 * we start and wait on commits.
5219 int ext4_mark_inode_dirty(handle_t
*handle
, struct inode
*inode
)
5221 struct ext4_iloc iloc
;
5222 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
5223 static unsigned int mnt_count
;
5227 trace_ext4_mark_inode_dirty(inode
, _RET_IP_
);
5228 err
= ext4_reserve_inode_write(handle
, inode
, &iloc
);
5231 if (ext4_handle_valid(handle
) &&
5232 EXT4_I(inode
)->i_extra_isize
< sbi
->s_want_extra_isize
&&
5233 !ext4_test_inode_state(inode
, EXT4_STATE_NO_EXPAND
)) {
5235 * We need extra buffer credits since we may write into EA block
5236 * with this same handle. If journal_extend fails, then it will
5237 * only result in a minor loss of functionality for that inode.
5238 * If this is felt to be critical, then e2fsck should be run to
5239 * force a large enough s_min_extra_isize.
5241 if ((jbd2_journal_extend(handle
,
5242 EXT4_DATA_TRANS_BLOCKS(inode
->i_sb
))) == 0) {
5243 ret
= ext4_expand_extra_isize(inode
,
5244 sbi
->s_want_extra_isize
,
5248 le16_to_cpu(sbi
->s_es
->s_mnt_count
)) {
5249 ext4_warning(inode
->i_sb
,
5250 "Unable to expand inode %lu. Delete"
5251 " some EAs or run e2fsck.",
5254 le16_to_cpu(sbi
->s_es
->s_mnt_count
);
5259 return ext4_mark_iloc_dirty(handle
, inode
, &iloc
);
5263 * ext4_dirty_inode() is called from __mark_inode_dirty()
5265 * We're really interested in the case where a file is being extended.
5266 * i_size has been changed by generic_commit_write() and we thus need
5267 * to include the updated inode in the current transaction.
5269 * Also, dquot_alloc_block() will always dirty the inode when blocks
5270 * are allocated to the file.
5272 * If the inode is marked synchronous, we don't honour that here - doing
5273 * so would cause a commit on atime updates, which we don't bother doing.
5274 * We handle synchronous inodes at the highest possible level.
5276 * If only the I_DIRTY_TIME flag is set, we can skip everything. If
5277 * I_DIRTY_TIME and I_DIRTY_SYNC is set, the only inode fields we need
5278 * to copy into the on-disk inode structure are the timestamp files.
5280 void ext4_dirty_inode(struct inode
*inode
, int flags
)
5284 if (flags
== I_DIRTY_TIME
)
5286 handle
= ext4_journal_start(inode
, EXT4_HT_INODE
, 2);
5290 ext4_mark_inode_dirty(handle
, inode
);
5292 ext4_journal_stop(handle
);
5299 * Bind an inode's backing buffer_head into this transaction, to prevent
5300 * it from being flushed to disk early. Unlike
5301 * ext4_reserve_inode_write, this leaves behind no bh reference and
5302 * returns no iloc structure, so the caller needs to repeat the iloc
5303 * lookup to mark the inode dirty later.
5305 static int ext4_pin_inode(handle_t
*handle
, struct inode
*inode
)
5307 struct ext4_iloc iloc
;
5311 err
= ext4_get_inode_loc(inode
, &iloc
);
5313 BUFFER_TRACE(iloc
.bh
, "get_write_access");
5314 err
= jbd2_journal_get_write_access(handle
, iloc
.bh
);
5316 err
= ext4_handle_dirty_metadata(handle
,
5322 ext4_std_error(inode
->i_sb
, err
);
5327 int ext4_change_inode_journal_flag(struct inode
*inode
, int val
)
5334 * We have to be very careful here: changing a data block's
5335 * journaling status dynamically is dangerous. If we write a
5336 * data block to the journal, change the status and then delete
5337 * that block, we risk forgetting to revoke the old log record
5338 * from the journal and so a subsequent replay can corrupt data.
5339 * So, first we make sure that the journal is empty and that
5340 * nobody is changing anything.
5343 journal
= EXT4_JOURNAL(inode
);
5346 if (is_journal_aborted(journal
))
5348 /* We have to allocate physical blocks for delalloc blocks
5349 * before flushing journal. otherwise delalloc blocks can not
5350 * be allocated any more. even more truncate on delalloc blocks
5351 * could trigger BUG by flushing delalloc blocks in journal.
5352 * There is no delalloc block in non-journal data mode.
5354 if (val
&& test_opt(inode
->i_sb
, DELALLOC
)) {
5355 err
= ext4_alloc_da_blocks(inode
);
5360 /* Wait for all existing dio workers */
5361 ext4_inode_block_unlocked_dio(inode
);
5362 inode_dio_wait(inode
);
5364 jbd2_journal_lock_updates(journal
);
5367 * OK, there are no updates running now, and all cached data is
5368 * synced to disk. We are now in a completely consistent state
5369 * which doesn't have anything in the journal, and we know that
5370 * no filesystem updates are running, so it is safe to modify
5371 * the inode's in-core data-journaling state flag now.
5375 ext4_set_inode_flag(inode
, EXT4_INODE_JOURNAL_DATA
);
5377 err
= jbd2_journal_flush(journal
);
5379 jbd2_journal_unlock_updates(journal
);
5380 ext4_inode_resume_unlocked_dio(inode
);
5383 ext4_clear_inode_flag(inode
, EXT4_INODE_JOURNAL_DATA
);
5385 ext4_set_aops(inode
);
5387 jbd2_journal_unlock_updates(journal
);
5388 ext4_inode_resume_unlocked_dio(inode
);
5390 /* Finally we can mark the inode as dirty. */
5392 handle
= ext4_journal_start(inode
, EXT4_HT_INODE
, 1);
5394 return PTR_ERR(handle
);
5396 err
= ext4_mark_inode_dirty(handle
, inode
);
5397 ext4_handle_sync(handle
);
5398 ext4_journal_stop(handle
);
5399 ext4_std_error(inode
->i_sb
, err
);
5404 static int ext4_bh_unmapped(handle_t
*handle
, struct buffer_head
*bh
)
5406 return !buffer_mapped(bh
);
5409 int ext4_page_mkwrite(struct vm_area_struct
*vma
, struct vm_fault
*vmf
)
5411 struct page
*page
= vmf
->page
;
5415 struct file
*file
= vma
->vm_file
;
5416 struct inode
*inode
= file_inode(file
);
5417 struct address_space
*mapping
= inode
->i_mapping
;
5419 get_block_t
*get_block
;
5422 sb_start_pagefault(inode
->i_sb
);
5423 file_update_time(vma
->vm_file
);
5425 down_read(&EXT4_I(inode
)->i_mmap_sem
);
5427 ret
= ext4_convert_inline_data(inode
);
5431 /* Delalloc case is easy... */
5432 if (test_opt(inode
->i_sb
, DELALLOC
) &&
5433 !ext4_should_journal_data(inode
) &&
5434 !ext4_nonda_switch(inode
->i_sb
)) {
5436 ret
= block_page_mkwrite(vma
, vmf
,
5437 ext4_da_get_block_prep
);
5438 } while (ret
== -ENOSPC
&&
5439 ext4_should_retry_alloc(inode
->i_sb
, &retries
));
5444 size
= i_size_read(inode
);
5445 /* Page got truncated from under us? */
5446 if (page
->mapping
!= mapping
|| page_offset(page
) > size
) {
5448 ret
= VM_FAULT_NOPAGE
;
5452 if (page
->index
== size
>> PAGE_CACHE_SHIFT
)
5453 len
= size
& ~PAGE_CACHE_MASK
;
5455 len
= PAGE_CACHE_SIZE
;
5457 * Return if we have all the buffers mapped. This avoids the need to do
5458 * journal_start/journal_stop which can block and take a long time
5460 if (page_has_buffers(page
)) {
5461 if (!ext4_walk_page_buffers(NULL
, page_buffers(page
),
5463 ext4_bh_unmapped
)) {
5464 /* Wait so that we don't change page under IO */
5465 wait_for_stable_page(page
);
5466 ret
= VM_FAULT_LOCKED
;
5471 /* OK, we need to fill the hole... */
5472 if (ext4_should_dioread_nolock(inode
))
5473 get_block
= ext4_get_block_write
;
5475 get_block
= ext4_get_block
;
5477 handle
= ext4_journal_start(inode
, EXT4_HT_WRITE_PAGE
,
5478 ext4_writepage_trans_blocks(inode
));
5479 if (IS_ERR(handle
)) {
5480 ret
= VM_FAULT_SIGBUS
;
5483 ret
= block_page_mkwrite(vma
, vmf
, get_block
);
5484 if (!ret
&& ext4_should_journal_data(inode
)) {
5485 if (ext4_walk_page_buffers(handle
, page_buffers(page
), 0,
5486 PAGE_CACHE_SIZE
, NULL
, do_journal_get_write_access
)) {
5488 ret
= VM_FAULT_SIGBUS
;
5489 ext4_journal_stop(handle
);
5492 ext4_set_inode_state(inode
, EXT4_STATE_JDATA
);
5494 ext4_journal_stop(handle
);
5495 if (ret
== -ENOSPC
&& ext4_should_retry_alloc(inode
->i_sb
, &retries
))
5498 ret
= block_page_mkwrite_return(ret
);
5500 up_read(&EXT4_I(inode
)->i_mmap_sem
);
5501 sb_end_pagefault(inode
->i_sb
);
5505 int ext4_filemap_fault(struct vm_area_struct
*vma
, struct vm_fault
*vmf
)
5507 struct inode
*inode
= file_inode(vma
->vm_file
);
5510 down_read(&EXT4_I(inode
)->i_mmap_sem
);
5511 err
= filemap_fault(vma
, vmf
);
5512 up_read(&EXT4_I(inode
)->i_mmap_sem
);