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/jbd2.h>
24 #include <linux/highuid.h>
25 #include <linux/pagemap.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/ratelimit.h>
40 #include <linux/aio.h>
41 #include <linux/bitops.h>
43 #include "ext4_jbd2.h"
48 #include <trace/events/ext4.h>
50 #define MPAGE_DA_EXTENT_TAIL 0x01
52 static __u32
ext4_inode_csum(struct inode
*inode
, struct ext4_inode
*raw
,
53 struct ext4_inode_info
*ei
)
55 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
60 csum_lo
= le16_to_cpu(raw
->i_checksum_lo
);
61 raw
->i_checksum_lo
= 0;
62 if (EXT4_INODE_SIZE(inode
->i_sb
) > EXT4_GOOD_OLD_INODE_SIZE
&&
63 EXT4_FITS_IN_INODE(raw
, ei
, i_checksum_hi
)) {
64 csum_hi
= le16_to_cpu(raw
->i_checksum_hi
);
65 raw
->i_checksum_hi
= 0;
68 csum
= ext4_chksum(sbi
, ei
->i_csum_seed
, (__u8
*)raw
,
69 EXT4_INODE_SIZE(inode
->i_sb
));
71 raw
->i_checksum_lo
= cpu_to_le16(csum_lo
);
72 if (EXT4_INODE_SIZE(inode
->i_sb
) > EXT4_GOOD_OLD_INODE_SIZE
&&
73 EXT4_FITS_IN_INODE(raw
, ei
, i_checksum_hi
))
74 raw
->i_checksum_hi
= cpu_to_le16(csum_hi
);
79 static int ext4_inode_csum_verify(struct inode
*inode
, struct ext4_inode
*raw
,
80 struct ext4_inode_info
*ei
)
82 __u32 provided
, calculated
;
84 if (EXT4_SB(inode
->i_sb
)->s_es
->s_creator_os
!=
85 cpu_to_le32(EXT4_OS_LINUX
) ||
86 !ext4_has_metadata_csum(inode
->i_sb
))
89 provided
= le16_to_cpu(raw
->i_checksum_lo
);
90 calculated
= ext4_inode_csum(inode
, raw
, ei
);
91 if (EXT4_INODE_SIZE(inode
->i_sb
) > EXT4_GOOD_OLD_INODE_SIZE
&&
92 EXT4_FITS_IN_INODE(raw
, ei
, i_checksum_hi
))
93 provided
|= ((__u32
)le16_to_cpu(raw
->i_checksum_hi
)) << 16;
97 return provided
== calculated
;
100 static void ext4_inode_csum_set(struct inode
*inode
, struct ext4_inode
*raw
,
101 struct ext4_inode_info
*ei
)
105 if (EXT4_SB(inode
->i_sb
)->s_es
->s_creator_os
!=
106 cpu_to_le32(EXT4_OS_LINUX
) ||
107 !ext4_has_metadata_csum(inode
->i_sb
))
110 csum
= ext4_inode_csum(inode
, raw
, ei
);
111 raw
->i_checksum_lo
= cpu_to_le16(csum
& 0xFFFF);
112 if (EXT4_INODE_SIZE(inode
->i_sb
) > EXT4_GOOD_OLD_INODE_SIZE
&&
113 EXT4_FITS_IN_INODE(raw
, ei
, i_checksum_hi
))
114 raw
->i_checksum_hi
= cpu_to_le16(csum
>> 16);
117 static inline int ext4_begin_ordered_truncate(struct inode
*inode
,
120 trace_ext4_begin_ordered_truncate(inode
, new_size
);
122 * If jinode is zero, then we never opened the file for
123 * writing, so there's no need to call
124 * jbd2_journal_begin_ordered_truncate() since there's no
125 * outstanding writes we need to flush.
127 if (!EXT4_I(inode
)->jinode
)
129 return jbd2_journal_begin_ordered_truncate(EXT4_JOURNAL(inode
),
130 EXT4_I(inode
)->jinode
,
134 static void ext4_invalidatepage(struct page
*page
, unsigned int offset
,
135 unsigned int length
);
136 static int __ext4_journalled_writepage(struct page
*page
, unsigned int len
);
137 static int ext4_bh_delay_or_unwritten(handle_t
*handle
, struct buffer_head
*bh
);
138 static int ext4_meta_trans_blocks(struct inode
*inode
, int lblocks
,
142 * Test whether an inode is a fast symlink.
144 static int ext4_inode_is_fast_symlink(struct inode
*inode
)
146 int ea_blocks
= EXT4_I(inode
)->i_file_acl
?
147 EXT4_CLUSTER_SIZE(inode
->i_sb
) >> 9 : 0;
149 if (ext4_has_inline_data(inode
))
152 return (S_ISLNK(inode
->i_mode
) && inode
->i_blocks
- ea_blocks
== 0);
156 * Restart the transaction associated with *handle. This does a commit,
157 * so before we call here everything must be consistently dirtied against
160 int ext4_truncate_restart_trans(handle_t
*handle
, struct inode
*inode
,
166 * Drop i_data_sem to avoid deadlock with ext4_map_blocks. At this
167 * moment, get_block can be called only for blocks inside i_size since
168 * page cache has been already dropped and writes are blocked by
169 * i_mutex. So we can safely drop the i_data_sem here.
171 BUG_ON(EXT4_JOURNAL(inode
) == NULL
);
172 jbd_debug(2, "restarting handle %p\n", handle
);
173 up_write(&EXT4_I(inode
)->i_data_sem
);
174 ret
= ext4_journal_restart(handle
, nblocks
);
175 down_write(&EXT4_I(inode
)->i_data_sem
);
176 ext4_discard_preallocations(inode
);
182 * Called at the last iput() if i_nlink is zero.
184 void ext4_evict_inode(struct inode
*inode
)
189 trace_ext4_evict_inode(inode
);
191 if (inode
->i_nlink
) {
193 * When journalling data dirty buffers are tracked only in the
194 * journal. So although mm thinks everything is clean and
195 * ready for reaping the inode might still have some pages to
196 * write in the running transaction or waiting to be
197 * checkpointed. Thus calling jbd2_journal_invalidatepage()
198 * (via truncate_inode_pages()) to discard these buffers can
199 * cause data loss. Also even if we did not discard these
200 * buffers, we would have no way to find them after the inode
201 * is reaped and thus user could see stale data if he tries to
202 * read them before the transaction is checkpointed. So be
203 * careful and force everything to disk here... We use
204 * ei->i_datasync_tid to store the newest transaction
205 * containing inode's data.
207 * Note that directories do not have this problem because they
208 * don't use page cache.
210 if (ext4_should_journal_data(inode
) &&
211 (S_ISLNK(inode
->i_mode
) || S_ISREG(inode
->i_mode
)) &&
212 inode
->i_ino
!= EXT4_JOURNAL_INO
) {
213 journal_t
*journal
= EXT4_SB(inode
->i_sb
)->s_journal
;
214 tid_t commit_tid
= EXT4_I(inode
)->i_datasync_tid
;
216 jbd2_complete_transaction(journal
, commit_tid
);
217 filemap_write_and_wait(&inode
->i_data
);
219 truncate_inode_pages_final(&inode
->i_data
);
221 WARN_ON(atomic_read(&EXT4_I(inode
)->i_ioend_count
));
225 if (is_bad_inode(inode
))
227 dquot_initialize(inode
);
229 if (ext4_should_order_data(inode
))
230 ext4_begin_ordered_truncate(inode
, 0);
231 truncate_inode_pages_final(&inode
->i_data
);
233 WARN_ON(atomic_read(&EXT4_I(inode
)->i_ioend_count
));
236 * Protect us against freezing - iput() caller didn't have to have any
237 * protection against it
239 sb_start_intwrite(inode
->i_sb
);
240 handle
= ext4_journal_start(inode
, EXT4_HT_TRUNCATE
,
241 ext4_blocks_for_truncate(inode
)+3);
242 if (IS_ERR(handle
)) {
243 ext4_std_error(inode
->i_sb
, PTR_ERR(handle
));
245 * If we're going to skip the normal cleanup, we still need to
246 * make sure that the in-core orphan linked list is properly
249 ext4_orphan_del(NULL
, inode
);
250 sb_end_intwrite(inode
->i_sb
);
255 ext4_handle_sync(handle
);
257 err
= ext4_mark_inode_dirty(handle
, inode
);
259 ext4_warning(inode
->i_sb
,
260 "couldn't mark inode dirty (err %d)", err
);
264 ext4_truncate(inode
);
267 * ext4_ext_truncate() doesn't reserve any slop when it
268 * restarts journal transactions; therefore there may not be
269 * enough credits left in the handle to remove the inode from
270 * the orphan list and set the dtime field.
272 if (!ext4_handle_has_enough_credits(handle
, 3)) {
273 err
= ext4_journal_extend(handle
, 3);
275 err
= ext4_journal_restart(handle
, 3);
277 ext4_warning(inode
->i_sb
,
278 "couldn't extend journal (err %d)", err
);
280 ext4_journal_stop(handle
);
281 ext4_orphan_del(NULL
, inode
);
282 sb_end_intwrite(inode
->i_sb
);
288 * Kill off the orphan record which ext4_truncate created.
289 * AKPM: I think this can be inside the above `if'.
290 * Note that ext4_orphan_del() has to be able to cope with the
291 * deletion of a non-existent orphan - this is because we don't
292 * know if ext4_truncate() actually created an orphan record.
293 * (Well, we could do this if we need to, but heck - it works)
295 ext4_orphan_del(handle
, inode
);
296 EXT4_I(inode
)->i_dtime
= get_seconds();
299 * One subtle ordering requirement: if anything has gone wrong
300 * (transaction abort, IO errors, whatever), then we can still
301 * do these next steps (the fs will already have been marked as
302 * having errors), but we can't free the inode if the mark_dirty
305 if (ext4_mark_inode_dirty(handle
, inode
))
306 /* If that failed, just do the required in-core inode clear. */
307 ext4_clear_inode(inode
);
309 ext4_free_inode(handle
, inode
);
310 ext4_journal_stop(handle
);
311 sb_end_intwrite(inode
->i_sb
);
314 ext4_clear_inode(inode
); /* We must guarantee clearing of inode... */
318 qsize_t
*ext4_get_reserved_space(struct inode
*inode
)
320 return &EXT4_I(inode
)->i_reserved_quota
;
325 * Called with i_data_sem down, which is important since we can call
326 * ext4_discard_preallocations() from here.
328 void ext4_da_update_reserve_space(struct inode
*inode
,
329 int used
, int quota_claim
)
331 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
332 struct ext4_inode_info
*ei
= EXT4_I(inode
);
334 spin_lock(&ei
->i_block_reservation_lock
);
335 trace_ext4_da_update_reserve_space(inode
, used
, quota_claim
);
336 if (unlikely(used
> ei
->i_reserved_data_blocks
)) {
337 ext4_warning(inode
->i_sb
, "%s: ino %lu, used %d "
338 "with only %d reserved data blocks",
339 __func__
, inode
->i_ino
, used
,
340 ei
->i_reserved_data_blocks
);
342 used
= ei
->i_reserved_data_blocks
;
345 /* Update per-inode reservations */
346 ei
->i_reserved_data_blocks
-= used
;
347 percpu_counter_sub(&sbi
->s_dirtyclusters_counter
, used
);
349 spin_unlock(&EXT4_I(inode
)->i_block_reservation_lock
);
351 /* Update quota subsystem for data blocks */
353 dquot_claim_block(inode
, EXT4_C2B(sbi
, used
));
356 * We did fallocate with an offset that is already delayed
357 * allocated. So on delayed allocated writeback we should
358 * not re-claim the quota for fallocated blocks.
360 dquot_release_reservation_block(inode
, EXT4_C2B(sbi
, used
));
364 * If we have done all the pending block allocations and if
365 * there aren't any writers on the inode, we can discard the
366 * inode's preallocations.
368 if ((ei
->i_reserved_data_blocks
== 0) &&
369 (atomic_read(&inode
->i_writecount
) == 0))
370 ext4_discard_preallocations(inode
);
373 static int __check_block_validity(struct inode
*inode
, const char *func
,
375 struct ext4_map_blocks
*map
)
377 if (!ext4_data_block_valid(EXT4_SB(inode
->i_sb
), map
->m_pblk
,
379 ext4_error_inode(inode
, func
, line
, map
->m_pblk
,
380 "lblock %lu mapped to illegal pblock "
381 "(length %d)", (unsigned long) map
->m_lblk
,
388 #define check_block_validity(inode, map) \
389 __check_block_validity((inode), __func__, __LINE__, (map))
391 #ifdef ES_AGGRESSIVE_TEST
392 static void ext4_map_blocks_es_recheck(handle_t
*handle
,
394 struct ext4_map_blocks
*es_map
,
395 struct ext4_map_blocks
*map
,
402 * There is a race window that the result is not the same.
403 * e.g. xfstests #223 when dioread_nolock enables. The reason
404 * is that we lookup a block mapping in extent status tree with
405 * out taking i_data_sem. So at the time the unwritten extent
406 * could be converted.
408 if (!(flags
& EXT4_GET_BLOCKS_NO_LOCK
))
409 down_read(&EXT4_I(inode
)->i_data_sem
);
410 if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
)) {
411 retval
= ext4_ext_map_blocks(handle
, inode
, map
, flags
&
412 EXT4_GET_BLOCKS_KEEP_SIZE
);
414 retval
= ext4_ind_map_blocks(handle
, inode
, map
, flags
&
415 EXT4_GET_BLOCKS_KEEP_SIZE
);
417 if (!(flags
& EXT4_GET_BLOCKS_NO_LOCK
))
418 up_read((&EXT4_I(inode
)->i_data_sem
));
421 * We don't check m_len because extent will be collpased in status
422 * tree. So the m_len might not equal.
424 if (es_map
->m_lblk
!= map
->m_lblk
||
425 es_map
->m_flags
!= map
->m_flags
||
426 es_map
->m_pblk
!= map
->m_pblk
) {
427 printk("ES cache assertion failed for inode: %lu "
428 "es_cached ex [%d/%d/%llu/%x] != "
429 "found ex [%d/%d/%llu/%x] retval %d flags %x\n",
430 inode
->i_ino
, es_map
->m_lblk
, es_map
->m_len
,
431 es_map
->m_pblk
, es_map
->m_flags
, map
->m_lblk
,
432 map
->m_len
, map
->m_pblk
, map
->m_flags
,
436 #endif /* ES_AGGRESSIVE_TEST */
439 * The ext4_map_blocks() function tries to look up the requested blocks,
440 * and returns if the blocks are already mapped.
442 * Otherwise it takes the write lock of the i_data_sem and allocate blocks
443 * and store the allocated blocks in the result buffer head and mark it
446 * If file type is extents based, it will call ext4_ext_map_blocks(),
447 * Otherwise, call with ext4_ind_map_blocks() to handle indirect mapping
450 * On success, it returns the number of blocks being mapped or allocated.
451 * if create==0 and the blocks are pre-allocated and unwritten block,
452 * the result buffer head is unmapped. If the create ==1, it will make sure
453 * the buffer head is mapped.
455 * It returns 0 if plain look up failed (blocks have not been allocated), in
456 * that case, buffer head is unmapped
458 * It returns the error in case of allocation failure.
460 int ext4_map_blocks(handle_t
*handle
, struct inode
*inode
,
461 struct ext4_map_blocks
*map
, int flags
)
463 struct extent_status es
;
466 #ifdef ES_AGGRESSIVE_TEST
467 struct ext4_map_blocks orig_map
;
469 memcpy(&orig_map
, map
, sizeof(*map
));
473 ext_debug("ext4_map_blocks(): inode %lu, flag %d, max_blocks %u,"
474 "logical block %lu\n", inode
->i_ino
, flags
, map
->m_len
,
475 (unsigned long) map
->m_lblk
);
478 * ext4_map_blocks returns an int, and m_len is an unsigned int
480 if (unlikely(map
->m_len
> INT_MAX
))
481 map
->m_len
= INT_MAX
;
483 /* We can handle the block number less than EXT_MAX_BLOCKS */
484 if (unlikely(map
->m_lblk
>= EXT_MAX_BLOCKS
))
487 /* Lookup extent status tree firstly */
488 if (ext4_es_lookup_extent(inode
, map
->m_lblk
, &es
)) {
489 ext4_es_list_add(inode
);
490 if (ext4_es_is_written(&es
) || ext4_es_is_unwritten(&es
)) {
491 map
->m_pblk
= ext4_es_pblock(&es
) +
492 map
->m_lblk
- es
.es_lblk
;
493 map
->m_flags
|= ext4_es_is_written(&es
) ?
494 EXT4_MAP_MAPPED
: EXT4_MAP_UNWRITTEN
;
495 retval
= es
.es_len
- (map
->m_lblk
- es
.es_lblk
);
496 if (retval
> map
->m_len
)
499 } else if (ext4_es_is_delayed(&es
) || ext4_es_is_hole(&es
)) {
504 #ifdef ES_AGGRESSIVE_TEST
505 ext4_map_blocks_es_recheck(handle
, inode
, map
,
512 * Try to see if we can get the block without requesting a new
515 if (!(flags
& EXT4_GET_BLOCKS_NO_LOCK
))
516 down_read(&EXT4_I(inode
)->i_data_sem
);
517 if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
)) {
518 retval
= ext4_ext_map_blocks(handle
, inode
, map
, flags
&
519 EXT4_GET_BLOCKS_KEEP_SIZE
);
521 retval
= ext4_ind_map_blocks(handle
, inode
, map
, flags
&
522 EXT4_GET_BLOCKS_KEEP_SIZE
);
527 if (unlikely(retval
!= map
->m_len
)) {
528 ext4_warning(inode
->i_sb
,
529 "ES len assertion failed for inode "
530 "%lu: retval %d != map->m_len %d",
531 inode
->i_ino
, retval
, map
->m_len
);
535 status
= map
->m_flags
& EXT4_MAP_UNWRITTEN
?
536 EXTENT_STATUS_UNWRITTEN
: EXTENT_STATUS_WRITTEN
;
537 if (!(flags
& EXT4_GET_BLOCKS_DELALLOC_RESERVE
) &&
538 ext4_find_delalloc_range(inode
, map
->m_lblk
,
539 map
->m_lblk
+ map
->m_len
- 1))
540 status
|= EXTENT_STATUS_DELAYED
;
541 ret
= ext4_es_insert_extent(inode
, map
->m_lblk
,
542 map
->m_len
, map
->m_pblk
, status
);
546 if (!(flags
& EXT4_GET_BLOCKS_NO_LOCK
))
547 up_read((&EXT4_I(inode
)->i_data_sem
));
550 if (retval
> 0 && map
->m_flags
& EXT4_MAP_MAPPED
) {
551 ret
= check_block_validity(inode
, map
);
556 /* If it is only a block(s) look up */
557 if ((flags
& EXT4_GET_BLOCKS_CREATE
) == 0)
561 * Returns if the blocks have already allocated
563 * Note that if blocks have been preallocated
564 * ext4_ext_get_block() returns the create = 0
565 * with buffer head unmapped.
567 if (retval
> 0 && map
->m_flags
& EXT4_MAP_MAPPED
)
569 * If we need to convert extent to unwritten
570 * we continue and do the actual work in
571 * ext4_ext_map_blocks()
573 if (!(flags
& EXT4_GET_BLOCKS_CONVERT_UNWRITTEN
))
577 * Here we clear m_flags because after allocating an new extent,
578 * it will be set again.
580 map
->m_flags
&= ~EXT4_MAP_FLAGS
;
583 * New blocks allocate and/or writing to unwritten extent
584 * will possibly result in updating i_data, so we take
585 * the write lock of i_data_sem, and call get_block()
586 * with create == 1 flag.
588 down_write(&EXT4_I(inode
)->i_data_sem
);
591 * We need to check for EXT4 here because migrate
592 * could have changed the inode type in between
594 if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
)) {
595 retval
= ext4_ext_map_blocks(handle
, inode
, map
, flags
);
597 retval
= ext4_ind_map_blocks(handle
, inode
, map
, flags
);
599 if (retval
> 0 && map
->m_flags
& EXT4_MAP_NEW
) {
601 * We allocated new blocks which will result in
602 * i_data's format changing. Force the migrate
603 * to fail by clearing migrate flags
605 ext4_clear_inode_state(inode
, EXT4_STATE_EXT_MIGRATE
);
609 * Update reserved blocks/metadata blocks after successful
610 * block allocation which had been deferred till now. We don't
611 * support fallocate for non extent files. So we can update
612 * reserve space here.
615 (flags
& EXT4_GET_BLOCKS_DELALLOC_RESERVE
))
616 ext4_da_update_reserve_space(inode
, retval
, 1);
622 if (unlikely(retval
!= map
->m_len
)) {
623 ext4_warning(inode
->i_sb
,
624 "ES len assertion failed for inode "
625 "%lu: retval %d != map->m_len %d",
626 inode
->i_ino
, retval
, map
->m_len
);
631 * If the extent has been zeroed out, we don't need to update
632 * extent status tree.
634 if ((flags
& EXT4_GET_BLOCKS_PRE_IO
) &&
635 ext4_es_lookup_extent(inode
, map
->m_lblk
, &es
)) {
636 if (ext4_es_is_written(&es
))
639 status
= map
->m_flags
& EXT4_MAP_UNWRITTEN
?
640 EXTENT_STATUS_UNWRITTEN
: EXTENT_STATUS_WRITTEN
;
641 if (!(flags
& EXT4_GET_BLOCKS_DELALLOC_RESERVE
) &&
642 ext4_find_delalloc_range(inode
, map
->m_lblk
,
643 map
->m_lblk
+ map
->m_len
- 1))
644 status
|= EXTENT_STATUS_DELAYED
;
645 ret
= ext4_es_insert_extent(inode
, map
->m_lblk
, map
->m_len
,
646 map
->m_pblk
, status
);
652 up_write((&EXT4_I(inode
)->i_data_sem
));
653 if (retval
> 0 && map
->m_flags
& EXT4_MAP_MAPPED
) {
654 ret
= check_block_validity(inode
, map
);
661 /* Maximum number of blocks we map for direct IO at once. */
662 #define DIO_MAX_BLOCKS 4096
664 static int _ext4_get_block(struct inode
*inode
, sector_t iblock
,
665 struct buffer_head
*bh
, int flags
)
667 handle_t
*handle
= ext4_journal_current_handle();
668 struct ext4_map_blocks map
;
669 int ret
= 0, started
= 0;
672 if (ext4_has_inline_data(inode
))
676 map
.m_len
= bh
->b_size
>> inode
->i_blkbits
;
678 if (flags
&& !(flags
& EXT4_GET_BLOCKS_NO_LOCK
) && !handle
) {
679 /* Direct IO write... */
680 if (map
.m_len
> DIO_MAX_BLOCKS
)
681 map
.m_len
= DIO_MAX_BLOCKS
;
682 dio_credits
= ext4_chunk_trans_blocks(inode
, map
.m_len
);
683 handle
= ext4_journal_start(inode
, EXT4_HT_MAP_BLOCKS
,
685 if (IS_ERR(handle
)) {
686 ret
= PTR_ERR(handle
);
692 ret
= ext4_map_blocks(handle
, inode
, &map
, flags
);
694 ext4_io_end_t
*io_end
= ext4_inode_aio(inode
);
696 map_bh(bh
, inode
->i_sb
, map
.m_pblk
);
697 bh
->b_state
= (bh
->b_state
& ~EXT4_MAP_FLAGS
) | map
.m_flags
;
698 if (io_end
&& io_end
->flag
& EXT4_IO_END_UNWRITTEN
)
699 set_buffer_defer_completion(bh
);
700 bh
->b_size
= inode
->i_sb
->s_blocksize
* map
.m_len
;
704 ext4_journal_stop(handle
);
708 int ext4_get_block(struct inode
*inode
, sector_t iblock
,
709 struct buffer_head
*bh
, int create
)
711 return _ext4_get_block(inode
, iblock
, bh
,
712 create
? EXT4_GET_BLOCKS_CREATE
: 0);
716 * `handle' can be NULL if create is zero
718 struct buffer_head
*ext4_getblk(handle_t
*handle
, struct inode
*inode
,
719 ext4_lblk_t block
, int create
)
721 struct ext4_map_blocks map
;
722 struct buffer_head
*bh
;
725 J_ASSERT(handle
!= NULL
|| create
== 0);
729 err
= ext4_map_blocks(handle
, inode
, &map
,
730 create
? EXT4_GET_BLOCKS_CREATE
: 0);
733 return create
? ERR_PTR(-ENOSPC
) : NULL
;
737 bh
= sb_getblk(inode
->i_sb
, map
.m_pblk
);
739 return ERR_PTR(-ENOMEM
);
740 if (map
.m_flags
& EXT4_MAP_NEW
) {
741 J_ASSERT(create
!= 0);
742 J_ASSERT(handle
!= NULL
);
745 * Now that we do not always journal data, we should
746 * keep in mind whether this should always journal the
747 * new buffer as metadata. For now, regular file
748 * writes use ext4_get_block instead, so it's not a
752 BUFFER_TRACE(bh
, "call get_create_access");
753 err
= ext4_journal_get_create_access(handle
, bh
);
758 if (!buffer_uptodate(bh
)) {
759 memset(bh
->b_data
, 0, inode
->i_sb
->s_blocksize
);
760 set_buffer_uptodate(bh
);
763 BUFFER_TRACE(bh
, "call ext4_handle_dirty_metadata");
764 err
= ext4_handle_dirty_metadata(handle
, inode
, bh
);
768 BUFFER_TRACE(bh
, "not a new buffer");
775 struct buffer_head
*ext4_bread(handle_t
*handle
, struct inode
*inode
,
776 ext4_lblk_t block
, int create
)
778 struct buffer_head
*bh
;
780 bh
= ext4_getblk(handle
, inode
, block
, create
);
783 if (!bh
|| buffer_uptodate(bh
))
785 ll_rw_block(READ
| REQ_META
| REQ_PRIO
, 1, &bh
);
787 if (buffer_uptodate(bh
))
790 return ERR_PTR(-EIO
);
793 int ext4_walk_page_buffers(handle_t
*handle
,
794 struct buffer_head
*head
,
798 int (*fn
)(handle_t
*handle
,
799 struct buffer_head
*bh
))
801 struct buffer_head
*bh
;
802 unsigned block_start
, block_end
;
803 unsigned blocksize
= head
->b_size
;
805 struct buffer_head
*next
;
807 for (bh
= head
, block_start
= 0;
808 ret
== 0 && (bh
!= head
|| !block_start
);
809 block_start
= block_end
, bh
= next
) {
810 next
= bh
->b_this_page
;
811 block_end
= block_start
+ blocksize
;
812 if (block_end
<= from
|| block_start
>= to
) {
813 if (partial
&& !buffer_uptodate(bh
))
817 err
= (*fn
)(handle
, bh
);
825 * To preserve ordering, it is essential that the hole instantiation and
826 * the data write be encapsulated in a single transaction. We cannot
827 * close off a transaction and start a new one between the ext4_get_block()
828 * and the commit_write(). So doing the jbd2_journal_start at the start of
829 * prepare_write() is the right place.
831 * Also, this function can nest inside ext4_writepage(). In that case, we
832 * *know* that ext4_writepage() has generated enough buffer credits to do the
833 * whole page. So we won't block on the journal in that case, which is good,
834 * because the caller may be PF_MEMALLOC.
836 * By accident, ext4 can be reentered when a transaction is open via
837 * quota file writes. If we were to commit the transaction while thus
838 * reentered, there can be a deadlock - we would be holding a quota
839 * lock, and the commit would never complete if another thread had a
840 * transaction open and was blocking on the quota lock - a ranking
843 * So what we do is to rely on the fact that jbd2_journal_stop/journal_start
844 * will _not_ run commit under these circumstances because handle->h_ref
845 * is elevated. We'll still have enough credits for the tiny quotafile
848 int do_journal_get_write_access(handle_t
*handle
,
849 struct buffer_head
*bh
)
851 int dirty
= buffer_dirty(bh
);
854 if (!buffer_mapped(bh
) || buffer_freed(bh
))
857 * __block_write_begin() could have dirtied some buffers. Clean
858 * the dirty bit as jbd2_journal_get_write_access() could complain
859 * otherwise about fs integrity issues. Setting of the dirty bit
860 * by __block_write_begin() isn't a real problem here as we clear
861 * the bit before releasing a page lock and thus writeback cannot
862 * ever write the buffer.
865 clear_buffer_dirty(bh
);
866 BUFFER_TRACE(bh
, "get write access");
867 ret
= ext4_journal_get_write_access(handle
, bh
);
869 ret
= ext4_handle_dirty_metadata(handle
, NULL
, bh
);
873 static int ext4_get_block_write_nolock(struct inode
*inode
, sector_t iblock
,
874 struct buffer_head
*bh_result
, int create
);
875 static int ext4_write_begin(struct file
*file
, struct address_space
*mapping
,
876 loff_t pos
, unsigned len
, unsigned flags
,
877 struct page
**pagep
, void **fsdata
)
879 struct inode
*inode
= mapping
->host
;
880 int ret
, needed_blocks
;
887 trace_ext4_write_begin(inode
, pos
, len
, flags
);
889 * Reserve one block more for addition to orphan list in case
890 * we allocate blocks but write fails for some reason
892 needed_blocks
= ext4_writepage_trans_blocks(inode
) + 1;
893 index
= pos
>> PAGE_CACHE_SHIFT
;
894 from
= pos
& (PAGE_CACHE_SIZE
- 1);
897 if (ext4_test_inode_state(inode
, EXT4_STATE_MAY_INLINE_DATA
)) {
898 ret
= ext4_try_to_write_inline_data(mapping
, inode
, pos
, len
,
907 * grab_cache_page_write_begin() can take a long time if the
908 * system is thrashing due to memory pressure, or if the page
909 * is being written back. So grab it first before we start
910 * the transaction handle. This also allows us to allocate
911 * the page (if needed) without using GFP_NOFS.
914 page
= grab_cache_page_write_begin(mapping
, index
, flags
);
920 handle
= ext4_journal_start(inode
, EXT4_HT_WRITE_PAGE
, needed_blocks
);
921 if (IS_ERR(handle
)) {
922 page_cache_release(page
);
923 return PTR_ERR(handle
);
927 if (page
->mapping
!= mapping
) {
928 /* The page got truncated from under us */
930 page_cache_release(page
);
931 ext4_journal_stop(handle
);
934 /* In case writeback began while the page was unlocked */
935 wait_for_stable_page(page
);
937 if (ext4_should_dioread_nolock(inode
))
938 ret
= __block_write_begin(page
, pos
, len
, ext4_get_block_write
);
940 ret
= __block_write_begin(page
, pos
, len
, ext4_get_block
);
942 if (!ret
&& ext4_should_journal_data(inode
)) {
943 ret
= ext4_walk_page_buffers(handle
, page_buffers(page
),
945 do_journal_get_write_access
);
951 * __block_write_begin may have instantiated a few blocks
952 * outside i_size. Trim these off again. Don't need
953 * i_size_read because we hold i_mutex.
955 * Add inode to orphan list in case we crash before
958 if (pos
+ len
> inode
->i_size
&& ext4_can_truncate(inode
))
959 ext4_orphan_add(handle
, inode
);
961 ext4_journal_stop(handle
);
962 if (pos
+ len
> inode
->i_size
) {
963 ext4_truncate_failed_write(inode
);
965 * If truncate failed early the inode might
966 * still be on the orphan list; we need to
967 * make sure the inode is removed from the
968 * orphan list in that case.
971 ext4_orphan_del(NULL
, inode
);
974 if (ret
== -ENOSPC
&&
975 ext4_should_retry_alloc(inode
->i_sb
, &retries
))
977 page_cache_release(page
);
984 /* For write_end() in data=journal mode */
985 static int write_end_fn(handle_t
*handle
, struct buffer_head
*bh
)
988 if (!buffer_mapped(bh
) || buffer_freed(bh
))
990 set_buffer_uptodate(bh
);
991 ret
= ext4_handle_dirty_metadata(handle
, NULL
, bh
);
992 clear_buffer_meta(bh
);
993 clear_buffer_prio(bh
);
998 * We need to pick up the new inode size which generic_commit_write gave us
999 * `file' can be NULL - eg, when called from page_symlink().
1001 * ext4 never places buffers on inode->i_mapping->private_list. metadata
1002 * buffers are managed internally.
1004 static int ext4_write_end(struct file
*file
,
1005 struct address_space
*mapping
,
1006 loff_t pos
, unsigned len
, unsigned copied
,
1007 struct page
*page
, void *fsdata
)
1009 handle_t
*handle
= ext4_journal_current_handle();
1010 struct inode
*inode
= mapping
->host
;
1012 int i_size_changed
= 0;
1014 trace_ext4_write_end(inode
, pos
, len
, copied
);
1015 if (ext4_test_inode_state(inode
, EXT4_STATE_ORDERED_MODE
)) {
1016 ret
= ext4_jbd2_file_inode(handle
, inode
);
1019 page_cache_release(page
);
1024 if (ext4_has_inline_data(inode
)) {
1025 ret
= ext4_write_inline_data_end(inode
, pos
, len
,
1031 copied
= block_write_end(file
, mapping
, pos
,
1032 len
, copied
, page
, fsdata
);
1034 * it's important to update i_size while still holding page lock:
1035 * page writeout could otherwise come in and zero beyond i_size.
1037 i_size_changed
= ext4_update_inode_size(inode
, pos
+ copied
);
1039 page_cache_release(page
);
1042 * Don't mark the inode dirty under page lock. First, it unnecessarily
1043 * makes the holding time of page lock longer. Second, it forces lock
1044 * ordering of page lock and transaction start for journaling
1048 ext4_mark_inode_dirty(handle
, inode
);
1050 if (pos
+ len
> inode
->i_size
&& ext4_can_truncate(inode
))
1051 /* if we have allocated more blocks and copied
1052 * less. We will have blocks allocated outside
1053 * inode->i_size. So truncate them
1055 ext4_orphan_add(handle
, inode
);
1057 ret2
= ext4_journal_stop(handle
);
1061 if (pos
+ len
> inode
->i_size
) {
1062 ext4_truncate_failed_write(inode
);
1064 * If truncate failed early the inode might still be
1065 * on the orphan list; we need to make sure the inode
1066 * is removed from the orphan list in that case.
1069 ext4_orphan_del(NULL
, inode
);
1072 return ret
? ret
: copied
;
1075 static int ext4_journalled_write_end(struct file
*file
,
1076 struct address_space
*mapping
,
1077 loff_t pos
, unsigned len
, unsigned copied
,
1078 struct page
*page
, void *fsdata
)
1080 handle_t
*handle
= ext4_journal_current_handle();
1081 struct inode
*inode
= mapping
->host
;
1085 int size_changed
= 0;
1087 trace_ext4_journalled_write_end(inode
, pos
, len
, copied
);
1088 from
= pos
& (PAGE_CACHE_SIZE
- 1);
1091 BUG_ON(!ext4_handle_valid(handle
));
1093 if (ext4_has_inline_data(inode
))
1094 copied
= ext4_write_inline_data_end(inode
, pos
, len
,
1098 if (!PageUptodate(page
))
1100 page_zero_new_buffers(page
, from
+copied
, to
);
1103 ret
= ext4_walk_page_buffers(handle
, page_buffers(page
), from
,
1104 to
, &partial
, write_end_fn
);
1106 SetPageUptodate(page
);
1108 size_changed
= ext4_update_inode_size(inode
, pos
+ copied
);
1109 ext4_set_inode_state(inode
, EXT4_STATE_JDATA
);
1110 EXT4_I(inode
)->i_datasync_tid
= handle
->h_transaction
->t_tid
;
1112 page_cache_release(page
);
1115 ret2
= ext4_mark_inode_dirty(handle
, inode
);
1120 if (pos
+ len
> inode
->i_size
&& ext4_can_truncate(inode
))
1121 /* if we have allocated more blocks and copied
1122 * less. We will have blocks allocated outside
1123 * inode->i_size. So truncate them
1125 ext4_orphan_add(handle
, inode
);
1127 ret2
= ext4_journal_stop(handle
);
1130 if (pos
+ len
> inode
->i_size
) {
1131 ext4_truncate_failed_write(inode
);
1133 * If truncate failed early the inode might still be
1134 * on the orphan list; we need to make sure the inode
1135 * is removed from the orphan list in that case.
1138 ext4_orphan_del(NULL
, inode
);
1141 return ret
? ret
: copied
;
1145 * Reserve a single cluster located at lblock
1147 static int ext4_da_reserve_space(struct inode
*inode
, ext4_lblk_t lblock
)
1149 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
1150 struct ext4_inode_info
*ei
= EXT4_I(inode
);
1151 unsigned int md_needed
;
1155 * We will charge metadata quota at writeout time; this saves
1156 * us from metadata over-estimation, though we may go over by
1157 * a small amount in the end. Here we just reserve for data.
1159 ret
= dquot_reserve_block(inode
, EXT4_C2B(sbi
, 1));
1164 * recalculate the amount of metadata blocks to reserve
1165 * in order to allocate nrblocks
1166 * worse case is one extent per block
1168 spin_lock(&ei
->i_block_reservation_lock
);
1170 * ext4_calc_metadata_amount() has side effects, which we have
1171 * to be prepared undo if we fail to claim space.
1174 trace_ext4_da_reserve_space(inode
, 0);
1176 if (ext4_claim_free_clusters(sbi
, 1, 0)) {
1177 spin_unlock(&ei
->i_block_reservation_lock
);
1178 dquot_release_reservation_block(inode
, EXT4_C2B(sbi
, 1));
1181 ei
->i_reserved_data_blocks
++;
1182 spin_unlock(&ei
->i_block_reservation_lock
);
1184 return 0; /* success */
1187 static void ext4_da_release_space(struct inode
*inode
, int to_free
)
1189 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
1190 struct ext4_inode_info
*ei
= EXT4_I(inode
);
1193 return; /* Nothing to release, exit */
1195 spin_lock(&EXT4_I(inode
)->i_block_reservation_lock
);
1197 trace_ext4_da_release_space(inode
, to_free
);
1198 if (unlikely(to_free
> ei
->i_reserved_data_blocks
)) {
1200 * if there aren't enough reserved blocks, then the
1201 * counter is messed up somewhere. Since this
1202 * function is called from invalidate page, it's
1203 * harmless to return without any action.
1205 ext4_warning(inode
->i_sb
, "ext4_da_release_space: "
1206 "ino %lu, to_free %d with only %d reserved "
1207 "data blocks", inode
->i_ino
, to_free
,
1208 ei
->i_reserved_data_blocks
);
1210 to_free
= ei
->i_reserved_data_blocks
;
1212 ei
->i_reserved_data_blocks
-= to_free
;
1214 /* update fs dirty data blocks counter */
1215 percpu_counter_sub(&sbi
->s_dirtyclusters_counter
, to_free
);
1217 spin_unlock(&EXT4_I(inode
)->i_block_reservation_lock
);
1219 dquot_release_reservation_block(inode
, EXT4_C2B(sbi
, to_free
));
1222 static void ext4_da_page_release_reservation(struct page
*page
,
1223 unsigned int offset
,
1224 unsigned int length
)
1227 struct buffer_head
*head
, *bh
;
1228 unsigned int curr_off
= 0;
1229 struct inode
*inode
= page
->mapping
->host
;
1230 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
1231 unsigned int stop
= offset
+ length
;
1235 BUG_ON(stop
> PAGE_CACHE_SIZE
|| stop
< length
);
1237 head
= page_buffers(page
);
1240 unsigned int next_off
= curr_off
+ bh
->b_size
;
1242 if (next_off
> stop
)
1245 if ((offset
<= curr_off
) && (buffer_delay(bh
))) {
1247 clear_buffer_delay(bh
);
1249 curr_off
= next_off
;
1250 } while ((bh
= bh
->b_this_page
) != head
);
1253 lblk
= page
->index
<< (PAGE_CACHE_SHIFT
- inode
->i_blkbits
);
1254 ext4_es_remove_extent(inode
, lblk
, to_release
);
1257 /* If we have released all the blocks belonging to a cluster, then we
1258 * need to release the reserved space for that cluster. */
1259 num_clusters
= EXT4_NUM_B2C(sbi
, to_release
);
1260 while (num_clusters
> 0) {
1261 lblk
= (page
->index
<< (PAGE_CACHE_SHIFT
- inode
->i_blkbits
)) +
1262 ((num_clusters
- 1) << sbi
->s_cluster_bits
);
1263 if (sbi
->s_cluster_ratio
== 1 ||
1264 !ext4_find_delalloc_cluster(inode
, lblk
))
1265 ext4_da_release_space(inode
, 1);
1272 * Delayed allocation stuff
1275 struct mpage_da_data
{
1276 struct inode
*inode
;
1277 struct writeback_control
*wbc
;
1279 pgoff_t first_page
; /* The first page to write */
1280 pgoff_t next_page
; /* Current page to examine */
1281 pgoff_t last_page
; /* Last page to examine */
1283 * Extent to map - this can be after first_page because that can be
1284 * fully mapped. We somewhat abuse m_flags to store whether the extent
1285 * is delalloc or unwritten.
1287 struct ext4_map_blocks map
;
1288 struct ext4_io_submit io_submit
; /* IO submission data */
1291 static void mpage_release_unused_pages(struct mpage_da_data
*mpd
,
1296 struct pagevec pvec
;
1297 struct inode
*inode
= mpd
->inode
;
1298 struct address_space
*mapping
= inode
->i_mapping
;
1300 /* This is necessary when next_page == 0. */
1301 if (mpd
->first_page
>= mpd
->next_page
)
1304 index
= mpd
->first_page
;
1305 end
= mpd
->next_page
- 1;
1307 ext4_lblk_t start
, last
;
1308 start
= index
<< (PAGE_CACHE_SHIFT
- inode
->i_blkbits
);
1309 last
= end
<< (PAGE_CACHE_SHIFT
- inode
->i_blkbits
);
1310 ext4_es_remove_extent(inode
, start
, last
- start
+ 1);
1313 pagevec_init(&pvec
, 0);
1314 while (index
<= end
) {
1315 nr_pages
= pagevec_lookup(&pvec
, mapping
, index
, PAGEVEC_SIZE
);
1318 for (i
= 0; i
< nr_pages
; i
++) {
1319 struct page
*page
= pvec
.pages
[i
];
1320 if (page
->index
> end
)
1322 BUG_ON(!PageLocked(page
));
1323 BUG_ON(PageWriteback(page
));
1325 block_invalidatepage(page
, 0, PAGE_CACHE_SIZE
);
1326 ClearPageUptodate(page
);
1330 index
= pvec
.pages
[nr_pages
- 1]->index
+ 1;
1331 pagevec_release(&pvec
);
1335 static void ext4_print_free_blocks(struct inode
*inode
)
1337 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
1338 struct super_block
*sb
= inode
->i_sb
;
1339 struct ext4_inode_info
*ei
= EXT4_I(inode
);
1341 ext4_msg(sb
, KERN_CRIT
, "Total free blocks count %lld",
1342 EXT4_C2B(EXT4_SB(inode
->i_sb
),
1343 ext4_count_free_clusters(sb
)));
1344 ext4_msg(sb
, KERN_CRIT
, "Free/Dirty block details");
1345 ext4_msg(sb
, KERN_CRIT
, "free_blocks=%lld",
1346 (long long) EXT4_C2B(EXT4_SB(sb
),
1347 percpu_counter_sum(&sbi
->s_freeclusters_counter
)));
1348 ext4_msg(sb
, KERN_CRIT
, "dirty_blocks=%lld",
1349 (long long) EXT4_C2B(EXT4_SB(sb
),
1350 percpu_counter_sum(&sbi
->s_dirtyclusters_counter
)));
1351 ext4_msg(sb
, KERN_CRIT
, "Block reservation details");
1352 ext4_msg(sb
, KERN_CRIT
, "i_reserved_data_blocks=%u",
1353 ei
->i_reserved_data_blocks
);
1357 static int ext4_bh_delay_or_unwritten(handle_t
*handle
, struct buffer_head
*bh
)
1359 return (buffer_delay(bh
) || buffer_unwritten(bh
)) && buffer_dirty(bh
);
1363 * This function is grabs code from the very beginning of
1364 * ext4_map_blocks, but assumes that the caller is from delayed write
1365 * time. This function looks up the requested blocks and sets the
1366 * buffer delay bit under the protection of i_data_sem.
1368 static int ext4_da_map_blocks(struct inode
*inode
, sector_t iblock
,
1369 struct ext4_map_blocks
*map
,
1370 struct buffer_head
*bh
)
1372 struct extent_status es
;
1374 sector_t invalid_block
= ~((sector_t
) 0xffff);
1375 #ifdef ES_AGGRESSIVE_TEST
1376 struct ext4_map_blocks orig_map
;
1378 memcpy(&orig_map
, map
, sizeof(*map
));
1381 if (invalid_block
< ext4_blocks_count(EXT4_SB(inode
->i_sb
)->s_es
))
1385 ext_debug("ext4_da_map_blocks(): inode %lu, max_blocks %u,"
1386 "logical block %lu\n", inode
->i_ino
, map
->m_len
,
1387 (unsigned long) map
->m_lblk
);
1389 /* Lookup extent status tree firstly */
1390 if (ext4_es_lookup_extent(inode
, iblock
, &es
)) {
1391 ext4_es_list_add(inode
);
1392 if (ext4_es_is_hole(&es
)) {
1394 down_read(&EXT4_I(inode
)->i_data_sem
);
1399 * Delayed extent could be allocated by fallocate.
1400 * So we need to check it.
1402 if (ext4_es_is_delayed(&es
) && !ext4_es_is_unwritten(&es
)) {
1403 map_bh(bh
, inode
->i_sb
, invalid_block
);
1405 set_buffer_delay(bh
);
1409 map
->m_pblk
= ext4_es_pblock(&es
) + iblock
- es
.es_lblk
;
1410 retval
= es
.es_len
- (iblock
- es
.es_lblk
);
1411 if (retval
> map
->m_len
)
1412 retval
= map
->m_len
;
1413 map
->m_len
= retval
;
1414 if (ext4_es_is_written(&es
))
1415 map
->m_flags
|= EXT4_MAP_MAPPED
;
1416 else if (ext4_es_is_unwritten(&es
))
1417 map
->m_flags
|= EXT4_MAP_UNWRITTEN
;
1421 #ifdef ES_AGGRESSIVE_TEST
1422 ext4_map_blocks_es_recheck(NULL
, inode
, map
, &orig_map
, 0);
1428 * Try to see if we can get the block without requesting a new
1429 * file system block.
1431 down_read(&EXT4_I(inode
)->i_data_sem
);
1432 if (ext4_has_inline_data(inode
))
1434 else if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
))
1435 retval
= ext4_ext_map_blocks(NULL
, inode
, map
, 0);
1437 retval
= ext4_ind_map_blocks(NULL
, inode
, map
, 0);
1443 * XXX: __block_prepare_write() unmaps passed block,
1447 * If the block was allocated from previously allocated cluster,
1448 * then we don't need to reserve it again. However we still need
1449 * to reserve metadata for every block we're going to write.
1451 if (EXT4_SB(inode
->i_sb
)->s_cluster_ratio
<= 1 ||
1452 !ext4_find_delalloc_cluster(inode
, map
->m_lblk
)) {
1453 ret
= ext4_da_reserve_space(inode
, iblock
);
1455 /* not enough space to reserve */
1461 ret
= ext4_es_insert_extent(inode
, map
->m_lblk
, map
->m_len
,
1462 ~0, EXTENT_STATUS_DELAYED
);
1468 map_bh(bh
, inode
->i_sb
, invalid_block
);
1470 set_buffer_delay(bh
);
1471 } else if (retval
> 0) {
1473 unsigned int status
;
1475 if (unlikely(retval
!= map
->m_len
)) {
1476 ext4_warning(inode
->i_sb
,
1477 "ES len assertion failed for inode "
1478 "%lu: retval %d != map->m_len %d",
1479 inode
->i_ino
, retval
, map
->m_len
);
1483 status
= map
->m_flags
& EXT4_MAP_UNWRITTEN
?
1484 EXTENT_STATUS_UNWRITTEN
: EXTENT_STATUS_WRITTEN
;
1485 ret
= ext4_es_insert_extent(inode
, map
->m_lblk
, map
->m_len
,
1486 map
->m_pblk
, status
);
1492 up_read((&EXT4_I(inode
)->i_data_sem
));
1498 * This is a special get_block_t callback which is used by
1499 * ext4_da_write_begin(). It will either return mapped block or
1500 * reserve space for a single block.
1502 * For delayed buffer_head we have BH_Mapped, BH_New, BH_Delay set.
1503 * We also have b_blocknr = -1 and b_bdev initialized properly
1505 * For unwritten buffer_head we have BH_Mapped, BH_New, BH_Unwritten set.
1506 * We also have b_blocknr = physicalblock mapping unwritten extent and b_bdev
1507 * initialized properly.
1509 int ext4_da_get_block_prep(struct inode
*inode
, sector_t iblock
,
1510 struct buffer_head
*bh
, int create
)
1512 struct ext4_map_blocks map
;
1515 BUG_ON(create
== 0);
1516 BUG_ON(bh
->b_size
!= inode
->i_sb
->s_blocksize
);
1518 map
.m_lblk
= iblock
;
1522 * first, we need to know whether the block is allocated already
1523 * preallocated blocks are unmapped but should treated
1524 * the same as allocated blocks.
1526 ret
= ext4_da_map_blocks(inode
, iblock
, &map
, bh
);
1530 map_bh(bh
, inode
->i_sb
, map
.m_pblk
);
1531 bh
->b_state
= (bh
->b_state
& ~EXT4_MAP_FLAGS
) | map
.m_flags
;
1533 if (buffer_unwritten(bh
)) {
1534 /* A delayed write to unwritten bh should be marked
1535 * new and mapped. Mapped ensures that we don't do
1536 * get_block multiple times when we write to the same
1537 * offset and new ensures that we do proper zero out
1538 * for partial write.
1541 set_buffer_mapped(bh
);
1546 static int bget_one(handle_t
*handle
, struct buffer_head
*bh
)
1552 static int bput_one(handle_t
*handle
, struct buffer_head
*bh
)
1558 static int __ext4_journalled_writepage(struct page
*page
,
1561 struct address_space
*mapping
= page
->mapping
;
1562 struct inode
*inode
= mapping
->host
;
1563 struct buffer_head
*page_bufs
= NULL
;
1564 handle_t
*handle
= NULL
;
1565 int ret
= 0, err
= 0;
1566 int inline_data
= ext4_has_inline_data(inode
);
1567 struct buffer_head
*inode_bh
= NULL
;
1569 ClearPageChecked(page
);
1572 BUG_ON(page
->index
!= 0);
1573 BUG_ON(len
> ext4_get_max_inline_size(inode
));
1574 inode_bh
= ext4_journalled_write_inline_data(inode
, len
, page
);
1575 if (inode_bh
== NULL
)
1578 page_bufs
= page_buffers(page
);
1583 ext4_walk_page_buffers(handle
, page_bufs
, 0, len
,
1586 /* As soon as we unlock the page, it can go away, but we have
1587 * references to buffers so we are safe */
1590 handle
= ext4_journal_start(inode
, EXT4_HT_WRITE_PAGE
,
1591 ext4_writepage_trans_blocks(inode
));
1592 if (IS_ERR(handle
)) {
1593 ret
= PTR_ERR(handle
);
1597 BUG_ON(!ext4_handle_valid(handle
));
1600 BUFFER_TRACE(inode_bh
, "get write access");
1601 ret
= ext4_journal_get_write_access(handle
, inode_bh
);
1603 err
= ext4_handle_dirty_metadata(handle
, inode
, inode_bh
);
1606 ret
= ext4_walk_page_buffers(handle
, page_bufs
, 0, len
, NULL
,
1607 do_journal_get_write_access
);
1609 err
= ext4_walk_page_buffers(handle
, page_bufs
, 0, len
, NULL
,
1614 EXT4_I(inode
)->i_datasync_tid
= handle
->h_transaction
->t_tid
;
1615 err
= ext4_journal_stop(handle
);
1619 if (!ext4_has_inline_data(inode
))
1620 ext4_walk_page_buffers(NULL
, page_bufs
, 0, len
,
1622 ext4_set_inode_state(inode
, EXT4_STATE_JDATA
);
1629 * Note that we don't need to start a transaction unless we're journaling data
1630 * because we should have holes filled from ext4_page_mkwrite(). We even don't
1631 * need to file the inode to the transaction's list in ordered mode because if
1632 * we are writing back data added by write(), the inode is already there and if
1633 * we are writing back data modified via mmap(), no one guarantees in which
1634 * transaction the data will hit the disk. In case we are journaling data, we
1635 * cannot start transaction directly because transaction start ranks above page
1636 * lock so we have to do some magic.
1638 * This function can get called via...
1639 * - ext4_writepages after taking page lock (have journal handle)
1640 * - journal_submit_inode_data_buffers (no journal handle)
1641 * - shrink_page_list via the kswapd/direct reclaim (no journal handle)
1642 * - grab_page_cache when doing write_begin (have journal handle)
1644 * We don't do any block allocation in this function. If we have page with
1645 * multiple blocks we need to write those buffer_heads that are mapped. This
1646 * is important for mmaped based write. So if we do with blocksize 1K
1647 * truncate(f, 1024);
1648 * a = mmap(f, 0, 4096);
1650 * truncate(f, 4096);
1651 * we have in the page first buffer_head mapped via page_mkwrite call back
1652 * but other buffer_heads would be unmapped but dirty (dirty done via the
1653 * do_wp_page). So writepage should write the first block. If we modify
1654 * the mmap area beyond 1024 we will again get a page_fault and the
1655 * page_mkwrite callback will do the block allocation and mark the
1656 * buffer_heads mapped.
1658 * We redirty the page if we have any buffer_heads that is either delay or
1659 * unwritten in the page.
1661 * We can get recursively called as show below.
1663 * ext4_writepage() -> kmalloc() -> __alloc_pages() -> page_launder() ->
1666 * But since we don't do any block allocation we should not deadlock.
1667 * Page also have the dirty flag cleared so we don't get recurive page_lock.
1669 static int ext4_writepage(struct page
*page
,
1670 struct writeback_control
*wbc
)
1675 struct buffer_head
*page_bufs
= NULL
;
1676 struct inode
*inode
= page
->mapping
->host
;
1677 struct ext4_io_submit io_submit
;
1678 bool keep_towrite
= false;
1680 trace_ext4_writepage(page
);
1681 size
= i_size_read(inode
);
1682 if (page
->index
== size
>> PAGE_CACHE_SHIFT
)
1683 len
= size
& ~PAGE_CACHE_MASK
;
1685 len
= PAGE_CACHE_SIZE
;
1687 page_bufs
= page_buffers(page
);
1689 * We cannot do block allocation or other extent handling in this
1690 * function. If there are buffers needing that, we have to redirty
1691 * the page. But we may reach here when we do a journal commit via
1692 * journal_submit_inode_data_buffers() and in that case we must write
1693 * allocated buffers to achieve data=ordered mode guarantees.
1695 if (ext4_walk_page_buffers(NULL
, page_bufs
, 0, len
, NULL
,
1696 ext4_bh_delay_or_unwritten
)) {
1697 redirty_page_for_writepage(wbc
, page
);
1698 if (current
->flags
& PF_MEMALLOC
) {
1700 * For memory cleaning there's no point in writing only
1701 * some buffers. So just bail out. Warn if we came here
1702 * from direct reclaim.
1704 WARN_ON_ONCE((current
->flags
& (PF_MEMALLOC
|PF_KSWAPD
))
1709 keep_towrite
= true;
1712 if (PageChecked(page
) && ext4_should_journal_data(inode
))
1714 * It's mmapped pagecache. Add buffers and journal it. There
1715 * doesn't seem much point in redirtying the page here.
1717 return __ext4_journalled_writepage(page
, len
);
1719 ext4_io_submit_init(&io_submit
, wbc
);
1720 io_submit
.io_end
= ext4_init_io_end(inode
, GFP_NOFS
);
1721 if (!io_submit
.io_end
) {
1722 redirty_page_for_writepage(wbc
, page
);
1726 ret
= ext4_bio_write_page(&io_submit
, page
, len
, wbc
, keep_towrite
);
1727 ext4_io_submit(&io_submit
);
1728 /* Drop io_end reference we got from init */
1729 ext4_put_io_end_defer(io_submit
.io_end
);
1733 static int mpage_submit_page(struct mpage_da_data
*mpd
, struct page
*page
)
1736 loff_t size
= i_size_read(mpd
->inode
);
1739 BUG_ON(page
->index
!= mpd
->first_page
);
1740 if (page
->index
== size
>> PAGE_CACHE_SHIFT
)
1741 len
= size
& ~PAGE_CACHE_MASK
;
1743 len
= PAGE_CACHE_SIZE
;
1744 clear_page_dirty_for_io(page
);
1745 err
= ext4_bio_write_page(&mpd
->io_submit
, page
, len
, mpd
->wbc
, false);
1747 mpd
->wbc
->nr_to_write
--;
1753 #define BH_FLAGS ((1 << BH_Unwritten) | (1 << BH_Delay))
1756 * mballoc gives us at most this number of blocks...
1757 * XXX: That seems to be only a limitation of ext4_mb_normalize_request().
1758 * The rest of mballoc seems to handle chunks up to full group size.
1760 #define MAX_WRITEPAGES_EXTENT_LEN 2048
1763 * mpage_add_bh_to_extent - try to add bh to extent of blocks to map
1765 * @mpd - extent of blocks
1766 * @lblk - logical number of the block in the file
1767 * @bh - buffer head we want to add to the extent
1769 * The function is used to collect contig. blocks in the same state. If the
1770 * buffer doesn't require mapping for writeback and we haven't started the
1771 * extent of buffers to map yet, the function returns 'true' immediately - the
1772 * caller can write the buffer right away. Otherwise the function returns true
1773 * if the block has been added to the extent, false if the block couldn't be
1776 static bool mpage_add_bh_to_extent(struct mpage_da_data
*mpd
, ext4_lblk_t lblk
,
1777 struct buffer_head
*bh
)
1779 struct ext4_map_blocks
*map
= &mpd
->map
;
1781 /* Buffer that doesn't need mapping for writeback? */
1782 if (!buffer_dirty(bh
) || !buffer_mapped(bh
) ||
1783 (!buffer_delay(bh
) && !buffer_unwritten(bh
))) {
1784 /* So far no extent to map => we write the buffer right away */
1785 if (map
->m_len
== 0)
1790 /* First block in the extent? */
1791 if (map
->m_len
== 0) {
1794 map
->m_flags
= bh
->b_state
& BH_FLAGS
;
1798 /* Don't go larger than mballoc is willing to allocate */
1799 if (map
->m_len
>= MAX_WRITEPAGES_EXTENT_LEN
)
1802 /* Can we merge the block to our big extent? */
1803 if (lblk
== map
->m_lblk
+ map
->m_len
&&
1804 (bh
->b_state
& BH_FLAGS
) == map
->m_flags
) {
1812 * mpage_process_page_bufs - submit page buffers for IO or add them to extent
1814 * @mpd - extent of blocks for mapping
1815 * @head - the first buffer in the page
1816 * @bh - buffer we should start processing from
1817 * @lblk - logical number of the block in the file corresponding to @bh
1819 * Walk through page buffers from @bh upto @head (exclusive) and either submit
1820 * the page for IO if all buffers in this page were mapped and there's no
1821 * accumulated extent of buffers to map or add buffers in the page to the
1822 * extent of buffers to map. The function returns 1 if the caller can continue
1823 * by processing the next page, 0 if it should stop adding buffers to the
1824 * extent to map because we cannot extend it anymore. It can also return value
1825 * < 0 in case of error during IO submission.
1827 static int mpage_process_page_bufs(struct mpage_da_data
*mpd
,
1828 struct buffer_head
*head
,
1829 struct buffer_head
*bh
,
1832 struct inode
*inode
= mpd
->inode
;
1834 ext4_lblk_t blocks
= (i_size_read(inode
) + (1 << inode
->i_blkbits
) - 1)
1835 >> inode
->i_blkbits
;
1838 BUG_ON(buffer_locked(bh
));
1840 if (lblk
>= blocks
|| !mpage_add_bh_to_extent(mpd
, lblk
, bh
)) {
1841 /* Found extent to map? */
1844 /* Everything mapped so far and we hit EOF */
1847 } while (lblk
++, (bh
= bh
->b_this_page
) != head
);
1848 /* So far everything mapped? Submit the page for IO. */
1849 if (mpd
->map
.m_len
== 0) {
1850 err
= mpage_submit_page(mpd
, head
->b_page
);
1854 return lblk
< blocks
;
1858 * mpage_map_buffers - update buffers corresponding to changed extent and
1859 * submit fully mapped pages for IO
1861 * @mpd - description of extent to map, on return next extent to map
1863 * Scan buffers corresponding to changed extent (we expect corresponding pages
1864 * to be already locked) and update buffer state according to new extent state.
1865 * We map delalloc buffers to their physical location, clear unwritten bits,
1866 * and mark buffers as uninit when we perform writes to unwritten extents
1867 * and do extent conversion after IO is finished. If the last page is not fully
1868 * mapped, we update @map to the next extent in the last page that needs
1869 * mapping. Otherwise we submit the page for IO.
1871 static int mpage_map_and_submit_buffers(struct mpage_da_data
*mpd
)
1873 struct pagevec pvec
;
1875 struct inode
*inode
= mpd
->inode
;
1876 struct buffer_head
*head
, *bh
;
1877 int bpp_bits
= PAGE_CACHE_SHIFT
- inode
->i_blkbits
;
1883 start
= mpd
->map
.m_lblk
>> bpp_bits
;
1884 end
= (mpd
->map
.m_lblk
+ mpd
->map
.m_len
- 1) >> bpp_bits
;
1885 lblk
= start
<< bpp_bits
;
1886 pblock
= mpd
->map
.m_pblk
;
1888 pagevec_init(&pvec
, 0);
1889 while (start
<= end
) {
1890 nr_pages
= pagevec_lookup(&pvec
, inode
->i_mapping
, start
,
1894 for (i
= 0; i
< nr_pages
; i
++) {
1895 struct page
*page
= pvec
.pages
[i
];
1897 if (page
->index
> end
)
1899 /* Up to 'end' pages must be contiguous */
1900 BUG_ON(page
->index
!= start
);
1901 bh
= head
= page_buffers(page
);
1903 if (lblk
< mpd
->map
.m_lblk
)
1905 if (lblk
>= mpd
->map
.m_lblk
+ mpd
->map
.m_len
) {
1907 * Buffer after end of mapped extent.
1908 * Find next buffer in the page to map.
1911 mpd
->map
.m_flags
= 0;
1913 * FIXME: If dioread_nolock supports
1914 * blocksize < pagesize, we need to make
1915 * sure we add size mapped so far to
1916 * io_end->size as the following call
1917 * can submit the page for IO.
1919 err
= mpage_process_page_bufs(mpd
, head
,
1921 pagevec_release(&pvec
);
1926 if (buffer_delay(bh
)) {
1927 clear_buffer_delay(bh
);
1928 bh
->b_blocknr
= pblock
++;
1930 clear_buffer_unwritten(bh
);
1931 } while (lblk
++, (bh
= bh
->b_this_page
) != head
);
1934 * FIXME: This is going to break if dioread_nolock
1935 * supports blocksize < pagesize as we will try to
1936 * convert potentially unmapped parts of inode.
1938 mpd
->io_submit
.io_end
->size
+= PAGE_CACHE_SIZE
;
1939 /* Page fully mapped - let IO run! */
1940 err
= mpage_submit_page(mpd
, page
);
1942 pagevec_release(&pvec
);
1947 pagevec_release(&pvec
);
1949 /* Extent fully mapped and matches with page boundary. We are done. */
1951 mpd
->map
.m_flags
= 0;
1955 static int mpage_map_one_extent(handle_t
*handle
, struct mpage_da_data
*mpd
)
1957 struct inode
*inode
= mpd
->inode
;
1958 struct ext4_map_blocks
*map
= &mpd
->map
;
1959 int get_blocks_flags
;
1960 int err
, dioread_nolock
;
1962 trace_ext4_da_write_pages_extent(inode
, map
);
1964 * Call ext4_map_blocks() to allocate any delayed allocation blocks, or
1965 * to convert an unwritten extent to be initialized (in the case
1966 * where we have written into one or more preallocated blocks). It is
1967 * possible that we're going to need more metadata blocks than
1968 * previously reserved. However we must not fail because we're in
1969 * writeback and there is nothing we can do about it so it might result
1970 * in data loss. So use reserved blocks to allocate metadata if
1973 * We pass in the magic EXT4_GET_BLOCKS_DELALLOC_RESERVE if
1974 * the blocks in question are delalloc blocks. This indicates
1975 * that the blocks and quotas has already been checked when
1976 * the data was copied into the page cache.
1978 get_blocks_flags
= EXT4_GET_BLOCKS_CREATE
|
1979 EXT4_GET_BLOCKS_METADATA_NOFAIL
;
1980 dioread_nolock
= ext4_should_dioread_nolock(inode
);
1982 get_blocks_flags
|= EXT4_GET_BLOCKS_IO_CREATE_EXT
;
1983 if (map
->m_flags
& (1 << BH_Delay
))
1984 get_blocks_flags
|= EXT4_GET_BLOCKS_DELALLOC_RESERVE
;
1986 err
= ext4_map_blocks(handle
, inode
, map
, get_blocks_flags
);
1989 if (dioread_nolock
&& (map
->m_flags
& EXT4_MAP_UNWRITTEN
)) {
1990 if (!mpd
->io_submit
.io_end
->handle
&&
1991 ext4_handle_valid(handle
)) {
1992 mpd
->io_submit
.io_end
->handle
= handle
->h_rsv_handle
;
1993 handle
->h_rsv_handle
= NULL
;
1995 ext4_set_io_unwritten_flag(inode
, mpd
->io_submit
.io_end
);
1998 BUG_ON(map
->m_len
== 0);
1999 if (map
->m_flags
& EXT4_MAP_NEW
) {
2000 struct block_device
*bdev
= inode
->i_sb
->s_bdev
;
2003 for (i
= 0; i
< map
->m_len
; i
++)
2004 unmap_underlying_metadata(bdev
, map
->m_pblk
+ i
);
2010 * mpage_map_and_submit_extent - map extent starting at mpd->lblk of length
2011 * mpd->len and submit pages underlying it for IO
2013 * @handle - handle for journal operations
2014 * @mpd - extent to map
2015 * @give_up_on_write - we set this to true iff there is a fatal error and there
2016 * is no hope of writing the data. The caller should discard
2017 * dirty pages to avoid infinite loops.
2019 * The function maps extent starting at mpd->lblk of length mpd->len. If it is
2020 * delayed, blocks are allocated, if it is unwritten, we may need to convert
2021 * them to initialized or split the described range from larger unwritten
2022 * extent. Note that we need not map all the described range since allocation
2023 * can return less blocks or the range is covered by more unwritten extents. We
2024 * cannot map more because we are limited by reserved transaction credits. On
2025 * the other hand we always make sure that the last touched page is fully
2026 * mapped so that it can be written out (and thus forward progress is
2027 * guaranteed). After mapping we submit all mapped pages for IO.
2029 static int mpage_map_and_submit_extent(handle_t
*handle
,
2030 struct mpage_da_data
*mpd
,
2031 bool *give_up_on_write
)
2033 struct inode
*inode
= mpd
->inode
;
2034 struct ext4_map_blocks
*map
= &mpd
->map
;
2039 mpd
->io_submit
.io_end
->offset
=
2040 ((loff_t
)map
->m_lblk
) << inode
->i_blkbits
;
2042 err
= mpage_map_one_extent(handle
, mpd
);
2044 struct super_block
*sb
= inode
->i_sb
;
2046 if (EXT4_SB(sb
)->s_mount_flags
& EXT4_MF_FS_ABORTED
)
2047 goto invalidate_dirty_pages
;
2049 * Let the uper layers retry transient errors.
2050 * In the case of ENOSPC, if ext4_count_free_blocks()
2051 * is non-zero, a commit should free up blocks.
2053 if ((err
== -ENOMEM
) ||
2054 (err
== -ENOSPC
&& ext4_count_free_clusters(sb
))) {
2056 goto update_disksize
;
2059 ext4_msg(sb
, KERN_CRIT
,
2060 "Delayed block allocation failed for "
2061 "inode %lu at logical offset %llu with"
2062 " max blocks %u with error %d",
2064 (unsigned long long)map
->m_lblk
,
2065 (unsigned)map
->m_len
, -err
);
2066 ext4_msg(sb
, KERN_CRIT
,
2067 "This should not happen!! Data will "
2070 ext4_print_free_blocks(inode
);
2071 invalidate_dirty_pages
:
2072 *give_up_on_write
= true;
2077 * Update buffer state, submit mapped pages, and get us new
2080 err
= mpage_map_and_submit_buffers(mpd
);
2082 goto update_disksize
;
2083 } while (map
->m_len
);
2087 * Update on-disk size after IO is submitted. Races with
2088 * truncate are avoided by checking i_size under i_data_sem.
2090 disksize
= ((loff_t
)mpd
->first_page
) << PAGE_CACHE_SHIFT
;
2091 if (disksize
> EXT4_I(inode
)->i_disksize
) {
2095 down_write(&EXT4_I(inode
)->i_data_sem
);
2096 i_size
= i_size_read(inode
);
2097 if (disksize
> i_size
)
2099 if (disksize
> EXT4_I(inode
)->i_disksize
)
2100 EXT4_I(inode
)->i_disksize
= disksize
;
2101 err2
= ext4_mark_inode_dirty(handle
, inode
);
2102 up_write(&EXT4_I(inode
)->i_data_sem
);
2104 ext4_error(inode
->i_sb
,
2105 "Failed to mark inode %lu dirty",
2114 * Calculate the total number of credits to reserve for one writepages
2115 * iteration. This is called from ext4_writepages(). We map an extent of
2116 * up to MAX_WRITEPAGES_EXTENT_LEN blocks and then we go on and finish mapping
2117 * the last partial page. So in total we can map MAX_WRITEPAGES_EXTENT_LEN +
2118 * bpp - 1 blocks in bpp different extents.
2120 static int ext4_da_writepages_trans_blocks(struct inode
*inode
)
2122 int bpp
= ext4_journal_blocks_per_page(inode
);
2124 return ext4_meta_trans_blocks(inode
,
2125 MAX_WRITEPAGES_EXTENT_LEN
+ bpp
- 1, bpp
);
2129 * mpage_prepare_extent_to_map - find & lock contiguous range of dirty pages
2130 * and underlying extent to map
2132 * @mpd - where to look for pages
2134 * Walk dirty pages in the mapping. If they are fully mapped, submit them for
2135 * IO immediately. When we find a page which isn't mapped we start accumulating
2136 * extent of buffers underlying these pages that needs mapping (formed by
2137 * either delayed or unwritten buffers). We also lock the pages containing
2138 * these buffers. The extent found is returned in @mpd structure (starting at
2139 * mpd->lblk with length mpd->len blocks).
2141 * Note that this function can attach bios to one io_end structure which are
2142 * neither logically nor physically contiguous. Although it may seem as an
2143 * unnecessary complication, it is actually inevitable in blocksize < pagesize
2144 * case as we need to track IO to all buffers underlying a page in one io_end.
2146 static int mpage_prepare_extent_to_map(struct mpage_da_data
*mpd
)
2148 struct address_space
*mapping
= mpd
->inode
->i_mapping
;
2149 struct pagevec pvec
;
2150 unsigned int nr_pages
;
2151 long left
= mpd
->wbc
->nr_to_write
;
2152 pgoff_t index
= mpd
->first_page
;
2153 pgoff_t end
= mpd
->last_page
;
2156 int blkbits
= mpd
->inode
->i_blkbits
;
2158 struct buffer_head
*head
;
2160 if (mpd
->wbc
->sync_mode
== WB_SYNC_ALL
|| mpd
->wbc
->tagged_writepages
)
2161 tag
= PAGECACHE_TAG_TOWRITE
;
2163 tag
= PAGECACHE_TAG_DIRTY
;
2165 pagevec_init(&pvec
, 0);
2167 mpd
->next_page
= index
;
2168 while (index
<= end
) {
2169 nr_pages
= pagevec_lookup_tag(&pvec
, mapping
, &index
, tag
,
2170 min(end
- index
, (pgoff_t
)PAGEVEC_SIZE
-1) + 1);
2174 for (i
= 0; i
< nr_pages
; i
++) {
2175 struct page
*page
= pvec
.pages
[i
];
2178 * At this point, the page may be truncated or
2179 * invalidated (changing page->mapping to NULL), or
2180 * even swizzled back from swapper_space to tmpfs file
2181 * mapping. However, page->index will not change
2182 * because we have a reference on the page.
2184 if (page
->index
> end
)
2188 * Accumulated enough dirty pages? This doesn't apply
2189 * to WB_SYNC_ALL mode. For integrity sync we have to
2190 * keep going because someone may be concurrently
2191 * dirtying pages, and we might have synced a lot of
2192 * newly appeared dirty pages, but have not synced all
2193 * of the old dirty pages.
2195 if (mpd
->wbc
->sync_mode
== WB_SYNC_NONE
&& left
<= 0)
2198 /* If we can't merge this page, we are done. */
2199 if (mpd
->map
.m_len
> 0 && mpd
->next_page
!= page
->index
)
2204 * If the page is no longer dirty, or its mapping no
2205 * longer corresponds to inode we are writing (which
2206 * means it has been truncated or invalidated), or the
2207 * page is already under writeback and we are not doing
2208 * a data integrity writeback, skip the page
2210 if (!PageDirty(page
) ||
2211 (PageWriteback(page
) &&
2212 (mpd
->wbc
->sync_mode
== WB_SYNC_NONE
)) ||
2213 unlikely(page
->mapping
!= mapping
)) {
2218 wait_on_page_writeback(page
);
2219 BUG_ON(PageWriteback(page
));
2221 if (mpd
->map
.m_len
== 0)
2222 mpd
->first_page
= page
->index
;
2223 mpd
->next_page
= page
->index
+ 1;
2224 /* Add all dirty buffers to mpd */
2225 lblk
= ((ext4_lblk_t
)page
->index
) <<
2226 (PAGE_CACHE_SHIFT
- blkbits
);
2227 head
= page_buffers(page
);
2228 err
= mpage_process_page_bufs(mpd
, head
, head
, lblk
);
2234 pagevec_release(&pvec
);
2239 pagevec_release(&pvec
);
2243 static int __writepage(struct page
*page
, struct writeback_control
*wbc
,
2246 struct address_space
*mapping
= data
;
2247 int ret
= ext4_writepage(page
, wbc
);
2248 mapping_set_error(mapping
, ret
);
2252 static int ext4_writepages(struct address_space
*mapping
,
2253 struct writeback_control
*wbc
)
2255 pgoff_t writeback_index
= 0;
2256 long nr_to_write
= wbc
->nr_to_write
;
2257 int range_whole
= 0;
2259 handle_t
*handle
= NULL
;
2260 struct mpage_da_data mpd
;
2261 struct inode
*inode
= mapping
->host
;
2262 int needed_blocks
, rsv_blocks
= 0, ret
= 0;
2263 struct ext4_sb_info
*sbi
= EXT4_SB(mapping
->host
->i_sb
);
2265 struct blk_plug plug
;
2266 bool give_up_on_write
= false;
2268 trace_ext4_writepages(inode
, wbc
);
2271 * No pages to write? This is mainly a kludge to avoid starting
2272 * a transaction for special inodes like journal inode on last iput()
2273 * because that could violate lock ordering on umount
2275 if (!mapping
->nrpages
|| !mapping_tagged(mapping
, PAGECACHE_TAG_DIRTY
))
2276 goto out_writepages
;
2278 if (ext4_should_journal_data(inode
)) {
2279 struct blk_plug plug
;
2281 blk_start_plug(&plug
);
2282 ret
= write_cache_pages(mapping
, wbc
, __writepage
, mapping
);
2283 blk_finish_plug(&plug
);
2284 goto out_writepages
;
2288 * If the filesystem has aborted, it is read-only, so return
2289 * right away instead of dumping stack traces later on that
2290 * will obscure the real source of the problem. We test
2291 * EXT4_MF_FS_ABORTED instead of sb->s_flag's MS_RDONLY because
2292 * the latter could be true if the filesystem is mounted
2293 * read-only, and in that case, ext4_writepages should
2294 * *never* be called, so if that ever happens, we would want
2297 if (unlikely(sbi
->s_mount_flags
& EXT4_MF_FS_ABORTED
)) {
2299 goto out_writepages
;
2302 if (ext4_should_dioread_nolock(inode
)) {
2304 * We may need to convert up to one extent per block in
2305 * the page and we may dirty the inode.
2307 rsv_blocks
= 1 + (PAGE_CACHE_SIZE
>> inode
->i_blkbits
);
2311 * If we have inline data and arrive here, it means that
2312 * we will soon create the block for the 1st page, so
2313 * we'd better clear the inline data here.
2315 if (ext4_has_inline_data(inode
)) {
2316 /* Just inode will be modified... */
2317 handle
= ext4_journal_start(inode
, EXT4_HT_INODE
, 1);
2318 if (IS_ERR(handle
)) {
2319 ret
= PTR_ERR(handle
);
2320 goto out_writepages
;
2322 BUG_ON(ext4_test_inode_state(inode
,
2323 EXT4_STATE_MAY_INLINE_DATA
));
2324 ext4_destroy_inline_data(handle
, inode
);
2325 ext4_journal_stop(handle
);
2328 if (wbc
->range_start
== 0 && wbc
->range_end
== LLONG_MAX
)
2331 if (wbc
->range_cyclic
) {
2332 writeback_index
= mapping
->writeback_index
;
2333 if (writeback_index
)
2335 mpd
.first_page
= writeback_index
;
2338 mpd
.first_page
= wbc
->range_start
>> PAGE_CACHE_SHIFT
;
2339 mpd
.last_page
= wbc
->range_end
>> PAGE_CACHE_SHIFT
;
2344 ext4_io_submit_init(&mpd
.io_submit
, wbc
);
2346 if (wbc
->sync_mode
== WB_SYNC_ALL
|| wbc
->tagged_writepages
)
2347 tag_pages_for_writeback(mapping
, mpd
.first_page
, mpd
.last_page
);
2349 blk_start_plug(&plug
);
2350 while (!done
&& mpd
.first_page
<= mpd
.last_page
) {
2351 /* For each extent of pages we use new io_end */
2352 mpd
.io_submit
.io_end
= ext4_init_io_end(inode
, GFP_KERNEL
);
2353 if (!mpd
.io_submit
.io_end
) {
2359 * We have two constraints: We find one extent to map and we
2360 * must always write out whole page (makes a difference when
2361 * blocksize < pagesize) so that we don't block on IO when we
2362 * try to write out the rest of the page. Journalled mode is
2363 * not supported by delalloc.
2365 BUG_ON(ext4_should_journal_data(inode
));
2366 needed_blocks
= ext4_da_writepages_trans_blocks(inode
);
2368 /* start a new transaction */
2369 handle
= ext4_journal_start_with_reserve(inode
,
2370 EXT4_HT_WRITE_PAGE
, needed_blocks
, rsv_blocks
);
2371 if (IS_ERR(handle
)) {
2372 ret
= PTR_ERR(handle
);
2373 ext4_msg(inode
->i_sb
, KERN_CRIT
, "%s: jbd2_start: "
2374 "%ld pages, ino %lu; err %d", __func__
,
2375 wbc
->nr_to_write
, inode
->i_ino
, ret
);
2376 /* Release allocated io_end */
2377 ext4_put_io_end(mpd
.io_submit
.io_end
);
2381 trace_ext4_da_write_pages(inode
, mpd
.first_page
, mpd
.wbc
);
2382 ret
= mpage_prepare_extent_to_map(&mpd
);
2385 ret
= mpage_map_and_submit_extent(handle
, &mpd
,
2389 * We scanned the whole range (or exhausted
2390 * nr_to_write), submitted what was mapped and
2391 * didn't find anything needing mapping. We are
2397 ext4_journal_stop(handle
);
2398 /* Submit prepared bio */
2399 ext4_io_submit(&mpd
.io_submit
);
2400 /* Unlock pages we didn't use */
2401 mpage_release_unused_pages(&mpd
, give_up_on_write
);
2402 /* Drop our io_end reference we got from init */
2403 ext4_put_io_end(mpd
.io_submit
.io_end
);
2405 if (ret
== -ENOSPC
&& sbi
->s_journal
) {
2407 * Commit the transaction which would
2408 * free blocks released in the transaction
2411 jbd2_journal_force_commit_nested(sbi
->s_journal
);
2415 /* Fatal error - ENOMEM, EIO... */
2419 blk_finish_plug(&plug
);
2420 if (!ret
&& !cycled
&& wbc
->nr_to_write
> 0) {
2422 mpd
.last_page
= writeback_index
- 1;
2428 if (wbc
->range_cyclic
|| (range_whole
&& wbc
->nr_to_write
> 0))
2430 * Set the writeback_index so that range_cyclic
2431 * mode will write it back later
2433 mapping
->writeback_index
= mpd
.first_page
;
2436 trace_ext4_writepages_result(inode
, wbc
, ret
,
2437 nr_to_write
- wbc
->nr_to_write
);
2441 static int ext4_nonda_switch(struct super_block
*sb
)
2443 s64 free_clusters
, dirty_clusters
;
2444 struct ext4_sb_info
*sbi
= EXT4_SB(sb
);
2447 * switch to non delalloc mode if we are running low
2448 * on free block. The free block accounting via percpu
2449 * counters can get slightly wrong with percpu_counter_batch getting
2450 * accumulated on each CPU without updating global counters
2451 * Delalloc need an accurate free block accounting. So switch
2452 * to non delalloc when we are near to error range.
2455 percpu_counter_read_positive(&sbi
->s_freeclusters_counter
);
2457 percpu_counter_read_positive(&sbi
->s_dirtyclusters_counter
);
2459 * Start pushing delalloc when 1/2 of free blocks are dirty.
2461 if (dirty_clusters
&& (free_clusters
< 2 * dirty_clusters
))
2462 try_to_writeback_inodes_sb(sb
, WB_REASON_FS_FREE_SPACE
);
2464 if (2 * free_clusters
< 3 * dirty_clusters
||
2465 free_clusters
< (dirty_clusters
+ EXT4_FREECLUSTERS_WATERMARK
)) {
2467 * free block count is less than 150% of dirty blocks
2468 * or free blocks is less than watermark
2475 /* We always reserve for an inode update; the superblock could be there too */
2476 static int ext4_da_write_credits(struct inode
*inode
, loff_t pos
, unsigned len
)
2478 if (likely(EXT4_HAS_RO_COMPAT_FEATURE(inode
->i_sb
,
2479 EXT4_FEATURE_RO_COMPAT_LARGE_FILE
)))
2482 if (pos
+ len
<= 0x7fffffffULL
)
2485 /* We might need to update the superblock to set LARGE_FILE */
2489 static int ext4_da_write_begin(struct file
*file
, struct address_space
*mapping
,
2490 loff_t pos
, unsigned len
, unsigned flags
,
2491 struct page
**pagep
, void **fsdata
)
2493 int ret
, retries
= 0;
2496 struct inode
*inode
= mapping
->host
;
2499 index
= pos
>> PAGE_CACHE_SHIFT
;
2501 if (ext4_nonda_switch(inode
->i_sb
)) {
2502 *fsdata
= (void *)FALL_BACK_TO_NONDELALLOC
;
2503 return ext4_write_begin(file
, mapping
, pos
,
2504 len
, flags
, pagep
, fsdata
);
2506 *fsdata
= (void *)0;
2507 trace_ext4_da_write_begin(inode
, pos
, len
, flags
);
2509 if (ext4_test_inode_state(inode
, EXT4_STATE_MAY_INLINE_DATA
)) {
2510 ret
= ext4_da_write_inline_data_begin(mapping
, inode
,
2520 * grab_cache_page_write_begin() can take a long time if the
2521 * system is thrashing due to memory pressure, or if the page
2522 * is being written back. So grab it first before we start
2523 * the transaction handle. This also allows us to allocate
2524 * the page (if needed) without using GFP_NOFS.
2527 page
= grab_cache_page_write_begin(mapping
, index
, flags
);
2533 * With delayed allocation, we don't log the i_disksize update
2534 * if there is delayed block allocation. But we still need
2535 * to journalling the i_disksize update if writes to the end
2536 * of file which has an already mapped buffer.
2539 handle
= ext4_journal_start(inode
, EXT4_HT_WRITE_PAGE
,
2540 ext4_da_write_credits(inode
, pos
, len
));
2541 if (IS_ERR(handle
)) {
2542 page_cache_release(page
);
2543 return PTR_ERR(handle
);
2547 if (page
->mapping
!= mapping
) {
2548 /* The page got truncated from under us */
2550 page_cache_release(page
);
2551 ext4_journal_stop(handle
);
2554 /* In case writeback began while the page was unlocked */
2555 wait_for_stable_page(page
);
2557 ret
= __block_write_begin(page
, pos
, len
, ext4_da_get_block_prep
);
2560 ext4_journal_stop(handle
);
2562 * block_write_begin may have instantiated a few blocks
2563 * outside i_size. Trim these off again. Don't need
2564 * i_size_read because we hold i_mutex.
2566 if (pos
+ len
> inode
->i_size
)
2567 ext4_truncate_failed_write(inode
);
2569 if (ret
== -ENOSPC
&&
2570 ext4_should_retry_alloc(inode
->i_sb
, &retries
))
2573 page_cache_release(page
);
2582 * Check if we should update i_disksize
2583 * when write to the end of file but not require block allocation
2585 static int ext4_da_should_update_i_disksize(struct page
*page
,
2586 unsigned long offset
)
2588 struct buffer_head
*bh
;
2589 struct inode
*inode
= page
->mapping
->host
;
2593 bh
= page_buffers(page
);
2594 idx
= offset
>> inode
->i_blkbits
;
2596 for (i
= 0; i
< idx
; i
++)
2597 bh
= bh
->b_this_page
;
2599 if (!buffer_mapped(bh
) || (buffer_delay(bh
)) || buffer_unwritten(bh
))
2604 static int ext4_da_write_end(struct file
*file
,
2605 struct address_space
*mapping
,
2606 loff_t pos
, unsigned len
, unsigned copied
,
2607 struct page
*page
, void *fsdata
)
2609 struct inode
*inode
= mapping
->host
;
2611 handle_t
*handle
= ext4_journal_current_handle();
2613 unsigned long start
, end
;
2614 int write_mode
= (int)(unsigned long)fsdata
;
2616 if (write_mode
== FALL_BACK_TO_NONDELALLOC
)
2617 return ext4_write_end(file
, mapping
, pos
,
2618 len
, copied
, page
, fsdata
);
2620 trace_ext4_da_write_end(inode
, pos
, len
, copied
);
2621 start
= pos
& (PAGE_CACHE_SIZE
- 1);
2622 end
= start
+ copied
- 1;
2625 * generic_write_end() will run mark_inode_dirty() if i_size
2626 * changes. So let's piggyback the i_disksize mark_inode_dirty
2629 new_i_size
= pos
+ copied
;
2630 if (copied
&& new_i_size
> EXT4_I(inode
)->i_disksize
) {
2631 if (ext4_has_inline_data(inode
) ||
2632 ext4_da_should_update_i_disksize(page
, end
)) {
2633 ext4_update_i_disksize(inode
, new_i_size
);
2634 /* We need to mark inode dirty even if
2635 * new_i_size is less that inode->i_size
2636 * bu greater than i_disksize.(hint delalloc)
2638 ext4_mark_inode_dirty(handle
, inode
);
2642 if (write_mode
!= CONVERT_INLINE_DATA
&&
2643 ext4_test_inode_state(inode
, EXT4_STATE_MAY_INLINE_DATA
) &&
2644 ext4_has_inline_data(inode
))
2645 ret2
= ext4_da_write_inline_data_end(inode
, pos
, len
, copied
,
2648 ret2
= generic_write_end(file
, mapping
, pos
, len
, copied
,
2654 ret2
= ext4_journal_stop(handle
);
2658 return ret
? ret
: copied
;
2661 static void ext4_da_invalidatepage(struct page
*page
, unsigned int offset
,
2662 unsigned int length
)
2665 * Drop reserved blocks
2667 BUG_ON(!PageLocked(page
));
2668 if (!page_has_buffers(page
))
2671 ext4_da_page_release_reservation(page
, offset
, length
);
2674 ext4_invalidatepage(page
, offset
, length
);
2680 * Force all delayed allocation blocks to be allocated for a given inode.
2682 int ext4_alloc_da_blocks(struct inode
*inode
)
2684 trace_ext4_alloc_da_blocks(inode
);
2686 if (!EXT4_I(inode
)->i_reserved_data_blocks
)
2690 * We do something simple for now. The filemap_flush() will
2691 * also start triggering a write of the data blocks, which is
2692 * not strictly speaking necessary (and for users of
2693 * laptop_mode, not even desirable). However, to do otherwise
2694 * would require replicating code paths in:
2696 * ext4_writepages() ->
2697 * write_cache_pages() ---> (via passed in callback function)
2698 * __mpage_da_writepage() -->
2699 * mpage_add_bh_to_extent()
2700 * mpage_da_map_blocks()
2702 * The problem is that write_cache_pages(), located in
2703 * mm/page-writeback.c, marks pages clean in preparation for
2704 * doing I/O, which is not desirable if we're not planning on
2707 * We could call write_cache_pages(), and then redirty all of
2708 * the pages by calling redirty_page_for_writepage() but that
2709 * would be ugly in the extreme. So instead we would need to
2710 * replicate parts of the code in the above functions,
2711 * simplifying them because we wouldn't actually intend to
2712 * write out the pages, but rather only collect contiguous
2713 * logical block extents, call the multi-block allocator, and
2714 * then update the buffer heads with the block allocations.
2716 * For now, though, we'll cheat by calling filemap_flush(),
2717 * which will map the blocks, and start the I/O, but not
2718 * actually wait for the I/O to complete.
2720 return filemap_flush(inode
->i_mapping
);
2724 * bmap() is special. It gets used by applications such as lilo and by
2725 * the swapper to find the on-disk block of a specific piece of data.
2727 * Naturally, this is dangerous if the block concerned is still in the
2728 * journal. If somebody makes a swapfile on an ext4 data-journaling
2729 * filesystem and enables swap, then they may get a nasty shock when the
2730 * data getting swapped to that swapfile suddenly gets overwritten by
2731 * the original zero's written out previously to the journal and
2732 * awaiting writeback in the kernel's buffer cache.
2734 * So, if we see any bmap calls here on a modified, data-journaled file,
2735 * take extra steps to flush any blocks which might be in the cache.
2737 static sector_t
ext4_bmap(struct address_space
*mapping
, sector_t block
)
2739 struct inode
*inode
= mapping
->host
;
2744 * We can get here for an inline file via the FIBMAP ioctl
2746 if (ext4_has_inline_data(inode
))
2749 if (mapping_tagged(mapping
, PAGECACHE_TAG_DIRTY
) &&
2750 test_opt(inode
->i_sb
, DELALLOC
)) {
2752 * With delalloc we want to sync the file
2753 * so that we can make sure we allocate
2756 filemap_write_and_wait(mapping
);
2759 if (EXT4_JOURNAL(inode
) &&
2760 ext4_test_inode_state(inode
, EXT4_STATE_JDATA
)) {
2762 * This is a REALLY heavyweight approach, but the use of
2763 * bmap on dirty files is expected to be extremely rare:
2764 * only if we run lilo or swapon on a freshly made file
2765 * do we expect this to happen.
2767 * (bmap requires CAP_SYS_RAWIO so this does not
2768 * represent an unprivileged user DOS attack --- we'd be
2769 * in trouble if mortal users could trigger this path at
2772 * NB. EXT4_STATE_JDATA is not set on files other than
2773 * regular files. If somebody wants to bmap a directory
2774 * or symlink and gets confused because the buffer
2775 * hasn't yet been flushed to disk, they deserve
2776 * everything they get.
2779 ext4_clear_inode_state(inode
, EXT4_STATE_JDATA
);
2780 journal
= EXT4_JOURNAL(inode
);
2781 jbd2_journal_lock_updates(journal
);
2782 err
= jbd2_journal_flush(journal
);
2783 jbd2_journal_unlock_updates(journal
);
2789 return generic_block_bmap(mapping
, block
, ext4_get_block
);
2792 static int ext4_readpage(struct file
*file
, struct page
*page
)
2795 struct inode
*inode
= page
->mapping
->host
;
2797 trace_ext4_readpage(page
);
2799 if (ext4_has_inline_data(inode
))
2800 ret
= ext4_readpage_inline(inode
, page
);
2803 return mpage_readpage(page
, ext4_get_block
);
2809 ext4_readpages(struct file
*file
, struct address_space
*mapping
,
2810 struct list_head
*pages
, unsigned nr_pages
)
2812 struct inode
*inode
= mapping
->host
;
2814 /* If the file has inline data, no need to do readpages. */
2815 if (ext4_has_inline_data(inode
))
2818 return mpage_readpages(mapping
, pages
, nr_pages
, ext4_get_block
);
2821 static void ext4_invalidatepage(struct page
*page
, unsigned int offset
,
2822 unsigned int length
)
2824 trace_ext4_invalidatepage(page
, offset
, length
);
2826 /* No journalling happens on data buffers when this function is used */
2827 WARN_ON(page_has_buffers(page
) && buffer_jbd(page_buffers(page
)));
2829 block_invalidatepage(page
, offset
, length
);
2832 static int __ext4_journalled_invalidatepage(struct page
*page
,
2833 unsigned int offset
,
2834 unsigned int length
)
2836 journal_t
*journal
= EXT4_JOURNAL(page
->mapping
->host
);
2838 trace_ext4_journalled_invalidatepage(page
, offset
, length
);
2841 * If it's a full truncate we just forget about the pending dirtying
2843 if (offset
== 0 && length
== PAGE_CACHE_SIZE
)
2844 ClearPageChecked(page
);
2846 return jbd2_journal_invalidatepage(journal
, page
, offset
, length
);
2849 /* Wrapper for aops... */
2850 static void ext4_journalled_invalidatepage(struct page
*page
,
2851 unsigned int offset
,
2852 unsigned int length
)
2854 WARN_ON(__ext4_journalled_invalidatepage(page
, offset
, length
) < 0);
2857 static int ext4_releasepage(struct page
*page
, gfp_t wait
)
2859 journal_t
*journal
= EXT4_JOURNAL(page
->mapping
->host
);
2861 trace_ext4_releasepage(page
);
2863 /* Page has dirty journalled data -> cannot release */
2864 if (PageChecked(page
))
2867 return jbd2_journal_try_to_free_buffers(journal
, page
, wait
);
2869 return try_to_free_buffers(page
);
2873 * ext4_get_block used when preparing for a DIO write or buffer write.
2874 * We allocate an uinitialized extent if blocks haven't been allocated.
2875 * The extent will be converted to initialized after the IO is complete.
2877 int ext4_get_block_write(struct inode
*inode
, sector_t iblock
,
2878 struct buffer_head
*bh_result
, int create
)
2880 ext4_debug("ext4_get_block_write: inode %lu, create flag %d\n",
2881 inode
->i_ino
, create
);
2882 return _ext4_get_block(inode
, iblock
, bh_result
,
2883 EXT4_GET_BLOCKS_IO_CREATE_EXT
);
2886 static int ext4_get_block_write_nolock(struct inode
*inode
, sector_t iblock
,
2887 struct buffer_head
*bh_result
, int create
)
2889 ext4_debug("ext4_get_block_write_nolock: inode %lu, create flag %d\n",
2890 inode
->i_ino
, create
);
2891 return _ext4_get_block(inode
, iblock
, bh_result
,
2892 EXT4_GET_BLOCKS_NO_LOCK
);
2895 static void ext4_end_io_dio(struct kiocb
*iocb
, loff_t offset
,
2896 ssize_t size
, void *private)
2898 ext4_io_end_t
*io_end
= iocb
->private;
2900 /* if not async direct IO just return */
2904 ext_debug("ext4_end_io_dio(): io_end 0x%p "
2905 "for inode %lu, iocb 0x%p, offset %llu, size %zd\n",
2906 iocb
->private, io_end
->inode
->i_ino
, iocb
, offset
,
2909 iocb
->private = NULL
;
2910 io_end
->offset
= offset
;
2911 io_end
->size
= size
;
2912 ext4_put_io_end(io_end
);
2916 * For ext4 extent files, ext4 will do direct-io write to holes,
2917 * preallocated extents, and those write extend the file, no need to
2918 * fall back to buffered IO.
2920 * For holes, we fallocate those blocks, mark them as unwritten
2921 * If those blocks were preallocated, we mark sure they are split, but
2922 * still keep the range to write as unwritten.
2924 * The unwritten extents will be converted to written when DIO is completed.
2925 * For async direct IO, since the IO may still pending when return, we
2926 * set up an end_io call back function, which will do the conversion
2927 * when async direct IO completed.
2929 * If the O_DIRECT write will extend the file then add this inode to the
2930 * orphan list. So recovery will truncate it back to the original size
2931 * if the machine crashes during the write.
2934 static ssize_t
ext4_ext_direct_IO(int rw
, struct kiocb
*iocb
,
2935 struct iov_iter
*iter
, loff_t offset
)
2937 struct file
*file
= iocb
->ki_filp
;
2938 struct inode
*inode
= file
->f_mapping
->host
;
2940 size_t count
= iov_iter_count(iter
);
2942 get_block_t
*get_block_func
= NULL
;
2944 loff_t final_size
= offset
+ count
;
2945 ext4_io_end_t
*io_end
= NULL
;
2947 /* Use the old path for reads and writes beyond i_size. */
2948 if (rw
!= WRITE
|| final_size
> inode
->i_size
)
2949 return ext4_ind_direct_IO(rw
, iocb
, iter
, offset
);
2951 BUG_ON(iocb
->private == NULL
);
2954 * Make all waiters for direct IO properly wait also for extent
2955 * conversion. This also disallows race between truncate() and
2956 * overwrite DIO as i_dio_count needs to be incremented under i_mutex.
2959 atomic_inc(&inode
->i_dio_count
);
2961 /* If we do a overwrite dio, i_mutex locking can be released */
2962 overwrite
= *((int *)iocb
->private);
2965 down_read(&EXT4_I(inode
)->i_data_sem
);
2966 mutex_unlock(&inode
->i_mutex
);
2970 * We could direct write to holes and fallocate.
2972 * Allocated blocks to fill the hole are marked as
2973 * unwritten to prevent parallel buffered read to expose
2974 * the stale data before DIO complete the data IO.
2976 * As to previously fallocated extents, ext4 get_block will
2977 * just simply mark the buffer mapped but still keep the
2978 * extents unwritten.
2980 * For non AIO case, we will convert those unwritten extents
2981 * to written after return back from blockdev_direct_IO.
2983 * For async DIO, the conversion needs to be deferred when the
2984 * IO is completed. The ext4 end_io callback function will be
2985 * called to take care of the conversion work. Here for async
2986 * case, we allocate an io_end structure to hook to the iocb.
2988 iocb
->private = NULL
;
2989 ext4_inode_aio_set(inode
, NULL
);
2990 if (!is_sync_kiocb(iocb
)) {
2991 io_end
= ext4_init_io_end(inode
, GFP_NOFS
);
2997 * Grab reference for DIO. Will be dropped in ext4_end_io_dio()
2999 iocb
->private = ext4_get_io_end(io_end
);
3001 * we save the io structure for current async direct
3002 * IO, so that later ext4_map_blocks() could flag the
3003 * io structure whether there is a unwritten extents
3004 * needs to be converted when IO is completed.
3006 ext4_inode_aio_set(inode
, io_end
);
3010 get_block_func
= ext4_get_block_write_nolock
;
3012 get_block_func
= ext4_get_block_write
;
3013 dio_flags
= DIO_LOCKING
;
3015 ret
= __blockdev_direct_IO(rw
, iocb
, inode
,
3016 inode
->i_sb
->s_bdev
, iter
,
3024 * Put our reference to io_end. This can free the io_end structure e.g.
3025 * in sync IO case or in case of error. It can even perform extent
3026 * conversion if all bios we submitted finished before we got here.
3027 * Note that in that case iocb->private can be already set to NULL
3031 ext4_inode_aio_set(inode
, NULL
);
3032 ext4_put_io_end(io_end
);
3034 * When no IO was submitted ext4_end_io_dio() was not
3035 * called so we have to put iocb's reference.
3037 if (ret
<= 0 && ret
!= -EIOCBQUEUED
&& iocb
->private) {
3038 WARN_ON(iocb
->private != io_end
);
3039 WARN_ON(io_end
->flag
& EXT4_IO_END_UNWRITTEN
);
3040 ext4_put_io_end(io_end
);
3041 iocb
->private = NULL
;
3044 if (ret
> 0 && !overwrite
&& ext4_test_inode_state(inode
,
3045 EXT4_STATE_DIO_UNWRITTEN
)) {
3048 * for non AIO case, since the IO is already
3049 * completed, we could do the conversion right here
3051 err
= ext4_convert_unwritten_extents(NULL
, inode
,
3055 ext4_clear_inode_state(inode
, EXT4_STATE_DIO_UNWRITTEN
);
3060 inode_dio_done(inode
);
3061 /* take i_mutex locking again if we do a ovewrite dio */
3063 up_read(&EXT4_I(inode
)->i_data_sem
);
3064 mutex_lock(&inode
->i_mutex
);
3070 static ssize_t
ext4_direct_IO(int rw
, struct kiocb
*iocb
,
3071 struct iov_iter
*iter
, loff_t offset
)
3073 struct file
*file
= iocb
->ki_filp
;
3074 struct inode
*inode
= file
->f_mapping
->host
;
3075 size_t count
= iov_iter_count(iter
);
3079 * If we are doing data journalling we don't support O_DIRECT
3081 if (ext4_should_journal_data(inode
))
3084 /* Let buffer I/O handle the inline data case. */
3085 if (ext4_has_inline_data(inode
))
3088 trace_ext4_direct_IO_enter(inode
, offset
, count
, rw
);
3089 if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
))
3090 ret
= ext4_ext_direct_IO(rw
, iocb
, iter
, offset
);
3092 ret
= ext4_ind_direct_IO(rw
, iocb
, iter
, offset
);
3093 trace_ext4_direct_IO_exit(inode
, offset
, count
, rw
, ret
);
3098 * Pages can be marked dirty completely asynchronously from ext4's journalling
3099 * activity. By filemap_sync_pte(), try_to_unmap_one(), etc. We cannot do
3100 * much here because ->set_page_dirty is called under VFS locks. The page is
3101 * not necessarily locked.
3103 * We cannot just dirty the page and leave attached buffers clean, because the
3104 * buffers' dirty state is "definitive". We cannot just set the buffers dirty
3105 * or jbddirty because all the journalling code will explode.
3107 * So what we do is to mark the page "pending dirty" and next time writepage
3108 * is called, propagate that into the buffers appropriately.
3110 static int ext4_journalled_set_page_dirty(struct page
*page
)
3112 SetPageChecked(page
);
3113 return __set_page_dirty_nobuffers(page
);
3116 static const struct address_space_operations ext4_aops
= {
3117 .readpage
= ext4_readpage
,
3118 .readpages
= ext4_readpages
,
3119 .writepage
= ext4_writepage
,
3120 .writepages
= ext4_writepages
,
3121 .write_begin
= ext4_write_begin
,
3122 .write_end
= ext4_write_end
,
3124 .invalidatepage
= ext4_invalidatepage
,
3125 .releasepage
= ext4_releasepage
,
3126 .direct_IO
= ext4_direct_IO
,
3127 .migratepage
= buffer_migrate_page
,
3128 .is_partially_uptodate
= block_is_partially_uptodate
,
3129 .error_remove_page
= generic_error_remove_page
,
3132 static const struct address_space_operations ext4_journalled_aops
= {
3133 .readpage
= ext4_readpage
,
3134 .readpages
= ext4_readpages
,
3135 .writepage
= ext4_writepage
,
3136 .writepages
= ext4_writepages
,
3137 .write_begin
= ext4_write_begin
,
3138 .write_end
= ext4_journalled_write_end
,
3139 .set_page_dirty
= ext4_journalled_set_page_dirty
,
3141 .invalidatepage
= ext4_journalled_invalidatepage
,
3142 .releasepage
= ext4_releasepage
,
3143 .direct_IO
= ext4_direct_IO
,
3144 .is_partially_uptodate
= block_is_partially_uptodate
,
3145 .error_remove_page
= generic_error_remove_page
,
3148 static const struct address_space_operations ext4_da_aops
= {
3149 .readpage
= ext4_readpage
,
3150 .readpages
= ext4_readpages
,
3151 .writepage
= ext4_writepage
,
3152 .writepages
= ext4_writepages
,
3153 .write_begin
= ext4_da_write_begin
,
3154 .write_end
= ext4_da_write_end
,
3156 .invalidatepage
= ext4_da_invalidatepage
,
3157 .releasepage
= ext4_releasepage
,
3158 .direct_IO
= ext4_direct_IO
,
3159 .migratepage
= buffer_migrate_page
,
3160 .is_partially_uptodate
= block_is_partially_uptodate
,
3161 .error_remove_page
= generic_error_remove_page
,
3164 void ext4_set_aops(struct inode
*inode
)
3166 switch (ext4_inode_journal_mode(inode
)) {
3167 case EXT4_INODE_ORDERED_DATA_MODE
:
3168 ext4_set_inode_state(inode
, EXT4_STATE_ORDERED_MODE
);
3170 case EXT4_INODE_WRITEBACK_DATA_MODE
:
3171 ext4_clear_inode_state(inode
, EXT4_STATE_ORDERED_MODE
);
3173 case EXT4_INODE_JOURNAL_DATA_MODE
:
3174 inode
->i_mapping
->a_ops
= &ext4_journalled_aops
;
3179 if (test_opt(inode
->i_sb
, DELALLOC
))
3180 inode
->i_mapping
->a_ops
= &ext4_da_aops
;
3182 inode
->i_mapping
->a_ops
= &ext4_aops
;
3186 * ext4_block_zero_page_range() zeros out a mapping of length 'length'
3187 * starting from file offset 'from'. The range to be zero'd must
3188 * be contained with in one block. If the specified range exceeds
3189 * the end of the block it will be shortened to end of the block
3190 * that cooresponds to 'from'
3192 static int ext4_block_zero_page_range(handle_t
*handle
,
3193 struct address_space
*mapping
, loff_t from
, loff_t length
)
3195 ext4_fsblk_t index
= from
>> PAGE_CACHE_SHIFT
;
3196 unsigned offset
= from
& (PAGE_CACHE_SIZE
-1);
3197 unsigned blocksize
, max
, pos
;
3199 struct inode
*inode
= mapping
->host
;
3200 struct buffer_head
*bh
;
3204 page
= find_or_create_page(mapping
, from
>> PAGE_CACHE_SHIFT
,
3205 mapping_gfp_mask(mapping
) & ~__GFP_FS
);
3209 blocksize
= inode
->i_sb
->s_blocksize
;
3210 max
= blocksize
- (offset
& (blocksize
- 1));
3213 * correct length if it does not fall between
3214 * 'from' and the end of the block
3216 if (length
> max
|| length
< 0)
3219 iblock
= index
<< (PAGE_CACHE_SHIFT
- inode
->i_sb
->s_blocksize_bits
);
3221 if (!page_has_buffers(page
))
3222 create_empty_buffers(page
, blocksize
, 0);
3224 /* Find the buffer that contains "offset" */
3225 bh
= page_buffers(page
);
3227 while (offset
>= pos
) {
3228 bh
= bh
->b_this_page
;
3232 if (buffer_freed(bh
)) {
3233 BUFFER_TRACE(bh
, "freed: skip");
3236 if (!buffer_mapped(bh
)) {
3237 BUFFER_TRACE(bh
, "unmapped");
3238 ext4_get_block(inode
, iblock
, bh
, 0);
3239 /* unmapped? It's a hole - nothing to do */
3240 if (!buffer_mapped(bh
)) {
3241 BUFFER_TRACE(bh
, "still unmapped");
3246 /* Ok, it's mapped. Make sure it's up-to-date */
3247 if (PageUptodate(page
))
3248 set_buffer_uptodate(bh
);
3250 if (!buffer_uptodate(bh
)) {
3252 ll_rw_block(READ
, 1, &bh
);
3254 /* Uhhuh. Read error. Complain and punt. */
3255 if (!buffer_uptodate(bh
))
3258 if (ext4_should_journal_data(inode
)) {
3259 BUFFER_TRACE(bh
, "get write access");
3260 err
= ext4_journal_get_write_access(handle
, bh
);
3264 zero_user(page
, offset
, length
);
3265 BUFFER_TRACE(bh
, "zeroed end of block");
3267 if (ext4_should_journal_data(inode
)) {
3268 err
= ext4_handle_dirty_metadata(handle
, inode
, bh
);
3271 mark_buffer_dirty(bh
);
3272 if (ext4_test_inode_state(inode
, EXT4_STATE_ORDERED_MODE
))
3273 err
= ext4_jbd2_file_inode(handle
, inode
);
3278 page_cache_release(page
);
3283 * ext4_block_truncate_page() zeroes out a mapping from file offset `from'
3284 * up to the end of the block which corresponds to `from'.
3285 * This required during truncate. We need to physically zero the tail end
3286 * of that block so it doesn't yield old data if the file is later grown.
3288 static int ext4_block_truncate_page(handle_t
*handle
,
3289 struct address_space
*mapping
, loff_t from
)
3291 unsigned offset
= from
& (PAGE_CACHE_SIZE
-1);
3294 struct inode
*inode
= mapping
->host
;
3296 blocksize
= inode
->i_sb
->s_blocksize
;
3297 length
= blocksize
- (offset
& (blocksize
- 1));
3299 return ext4_block_zero_page_range(handle
, mapping
, from
, length
);
3302 int ext4_zero_partial_blocks(handle_t
*handle
, struct inode
*inode
,
3303 loff_t lstart
, loff_t length
)
3305 struct super_block
*sb
= inode
->i_sb
;
3306 struct address_space
*mapping
= inode
->i_mapping
;
3307 unsigned partial_start
, partial_end
;
3308 ext4_fsblk_t start
, end
;
3309 loff_t byte_end
= (lstart
+ length
- 1);
3312 partial_start
= lstart
& (sb
->s_blocksize
- 1);
3313 partial_end
= byte_end
& (sb
->s_blocksize
- 1);
3315 start
= lstart
>> sb
->s_blocksize_bits
;
3316 end
= byte_end
>> sb
->s_blocksize_bits
;
3318 /* Handle partial zero within the single block */
3320 (partial_start
|| (partial_end
!= sb
->s_blocksize
- 1))) {
3321 err
= ext4_block_zero_page_range(handle
, mapping
,
3325 /* Handle partial zero out on the start of the range */
3326 if (partial_start
) {
3327 err
= ext4_block_zero_page_range(handle
, mapping
,
3328 lstart
, sb
->s_blocksize
);
3332 /* Handle partial zero out on the end of the range */
3333 if (partial_end
!= sb
->s_blocksize
- 1)
3334 err
= ext4_block_zero_page_range(handle
, mapping
,
3335 byte_end
- partial_end
,
3340 int ext4_can_truncate(struct inode
*inode
)
3342 if (S_ISREG(inode
->i_mode
))
3344 if (S_ISDIR(inode
->i_mode
))
3346 if (S_ISLNK(inode
->i_mode
))
3347 return !ext4_inode_is_fast_symlink(inode
);
3352 * ext4_punch_hole: punches a hole in a file by releaseing the blocks
3353 * associated with the given offset and length
3355 * @inode: File inode
3356 * @offset: The offset where the hole will begin
3357 * @len: The length of the hole
3359 * Returns: 0 on success or negative on failure
3362 int ext4_punch_hole(struct inode
*inode
, loff_t offset
, loff_t length
)
3364 struct super_block
*sb
= inode
->i_sb
;
3365 ext4_lblk_t first_block
, stop_block
;
3366 struct address_space
*mapping
= inode
->i_mapping
;
3367 loff_t first_block_offset
, last_block_offset
;
3369 unsigned int credits
;
3372 if (!S_ISREG(inode
->i_mode
))
3375 trace_ext4_punch_hole(inode
, offset
, length
, 0);
3378 * Write out all dirty pages to avoid race conditions
3379 * Then release them.
3381 if (mapping
->nrpages
&& mapping_tagged(mapping
, PAGECACHE_TAG_DIRTY
)) {
3382 ret
= filemap_write_and_wait_range(mapping
, offset
,
3383 offset
+ length
- 1);
3388 mutex_lock(&inode
->i_mutex
);
3390 /* No need to punch hole beyond i_size */
3391 if (offset
>= inode
->i_size
)
3395 * If the hole extends beyond i_size, set the hole
3396 * to end after the page that contains i_size
3398 if (offset
+ length
> inode
->i_size
) {
3399 length
= inode
->i_size
+
3400 PAGE_CACHE_SIZE
- (inode
->i_size
& (PAGE_CACHE_SIZE
- 1)) -
3404 if (offset
& (sb
->s_blocksize
- 1) ||
3405 (offset
+ length
) & (sb
->s_blocksize
- 1)) {
3407 * Attach jinode to inode for jbd2 if we do any zeroing of
3410 ret
= ext4_inode_attach_jinode(inode
);
3416 first_block_offset
= round_up(offset
, sb
->s_blocksize
);
3417 last_block_offset
= round_down((offset
+ length
), sb
->s_blocksize
) - 1;
3419 /* Now release the pages and zero block aligned part of pages*/
3420 if (last_block_offset
> first_block_offset
)
3421 truncate_pagecache_range(inode
, first_block_offset
,
3424 /* Wait all existing dio workers, newcomers will block on i_mutex */
3425 ext4_inode_block_unlocked_dio(inode
);
3426 inode_dio_wait(inode
);
3428 if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
))
3429 credits
= ext4_writepage_trans_blocks(inode
);
3431 credits
= ext4_blocks_for_truncate(inode
);
3432 handle
= ext4_journal_start(inode
, EXT4_HT_TRUNCATE
, credits
);
3433 if (IS_ERR(handle
)) {
3434 ret
= PTR_ERR(handle
);
3435 ext4_std_error(sb
, ret
);
3439 ret
= ext4_zero_partial_blocks(handle
, inode
, offset
,
3444 first_block
= (offset
+ sb
->s_blocksize
- 1) >>
3445 EXT4_BLOCK_SIZE_BITS(sb
);
3446 stop_block
= (offset
+ length
) >> EXT4_BLOCK_SIZE_BITS(sb
);
3448 /* If there are no blocks to remove, return now */
3449 if (first_block
>= stop_block
)
3452 down_write(&EXT4_I(inode
)->i_data_sem
);
3453 ext4_discard_preallocations(inode
);
3455 ret
= ext4_es_remove_extent(inode
, first_block
,
3456 stop_block
- first_block
);
3458 up_write(&EXT4_I(inode
)->i_data_sem
);
3462 if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
))
3463 ret
= ext4_ext_remove_space(inode
, first_block
,
3466 ret
= ext4_ind_remove_space(handle
, inode
, first_block
,
3469 up_write(&EXT4_I(inode
)->i_data_sem
);
3471 ext4_handle_sync(handle
);
3473 /* Now release the pages again to reduce race window */
3474 if (last_block_offset
> first_block_offset
)
3475 truncate_pagecache_range(inode
, first_block_offset
,
3478 inode
->i_mtime
= inode
->i_ctime
= ext4_current_time(inode
);
3479 ext4_mark_inode_dirty(handle
, inode
);
3481 ext4_journal_stop(handle
);
3483 ext4_inode_resume_unlocked_dio(inode
);
3485 mutex_unlock(&inode
->i_mutex
);
3489 int ext4_inode_attach_jinode(struct inode
*inode
)
3491 struct ext4_inode_info
*ei
= EXT4_I(inode
);
3492 struct jbd2_inode
*jinode
;
3494 if (ei
->jinode
|| !EXT4_SB(inode
->i_sb
)->s_journal
)
3497 jinode
= jbd2_alloc_inode(GFP_KERNEL
);
3498 spin_lock(&inode
->i_lock
);
3501 spin_unlock(&inode
->i_lock
);
3504 ei
->jinode
= jinode
;
3505 jbd2_journal_init_jbd_inode(ei
->jinode
, inode
);
3508 spin_unlock(&inode
->i_lock
);
3509 if (unlikely(jinode
!= NULL
))
3510 jbd2_free_inode(jinode
);
3517 * We block out ext4_get_block() block instantiations across the entire
3518 * transaction, and VFS/VM ensures that ext4_truncate() cannot run
3519 * simultaneously on behalf of the same inode.
3521 * As we work through the truncate and commit bits of it to the journal there
3522 * is one core, guiding principle: the file's tree must always be consistent on
3523 * disk. We must be able to restart the truncate after a crash.
3525 * The file's tree may be transiently inconsistent in memory (although it
3526 * probably isn't), but whenever we close off and commit a journal transaction,
3527 * the contents of (the filesystem + the journal) must be consistent and
3528 * restartable. It's pretty simple, really: bottom up, right to left (although
3529 * left-to-right works OK too).
3531 * Note that at recovery time, journal replay occurs *before* the restart of
3532 * truncate against the orphan inode list.
3534 * The committed inode has the new, desired i_size (which is the same as
3535 * i_disksize in this case). After a crash, ext4_orphan_cleanup() will see
3536 * that this inode's truncate did not complete and it will again call
3537 * ext4_truncate() to have another go. So there will be instantiated blocks
3538 * to the right of the truncation point in a crashed ext4 filesystem. But
3539 * that's fine - as long as they are linked from the inode, the post-crash
3540 * ext4_truncate() run will find them and release them.
3542 void ext4_truncate(struct inode
*inode
)
3544 struct ext4_inode_info
*ei
= EXT4_I(inode
);
3545 unsigned int credits
;
3547 struct address_space
*mapping
= inode
->i_mapping
;
3550 * There is a possibility that we're either freeing the inode
3551 * or it's a completely new inode. In those cases we might not
3552 * have i_mutex locked because it's not necessary.
3554 if (!(inode
->i_state
& (I_NEW
|I_FREEING
)))
3555 WARN_ON(!mutex_is_locked(&inode
->i_mutex
));
3556 trace_ext4_truncate_enter(inode
);
3558 if (!ext4_can_truncate(inode
))
3561 ext4_clear_inode_flag(inode
, EXT4_INODE_EOFBLOCKS
);
3563 if (inode
->i_size
== 0 && !test_opt(inode
->i_sb
, NO_AUTO_DA_ALLOC
))
3564 ext4_set_inode_state(inode
, EXT4_STATE_DA_ALLOC_CLOSE
);
3566 if (ext4_has_inline_data(inode
)) {
3569 ext4_inline_data_truncate(inode
, &has_inline
);
3574 /* If we zero-out tail of the page, we have to create jinode for jbd2 */
3575 if (inode
->i_size
& (inode
->i_sb
->s_blocksize
- 1)) {
3576 if (ext4_inode_attach_jinode(inode
) < 0)
3580 if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
))
3581 credits
= ext4_writepage_trans_blocks(inode
);
3583 credits
= ext4_blocks_for_truncate(inode
);
3585 handle
= ext4_journal_start(inode
, EXT4_HT_TRUNCATE
, credits
);
3586 if (IS_ERR(handle
)) {
3587 ext4_std_error(inode
->i_sb
, PTR_ERR(handle
));
3591 if (inode
->i_size
& (inode
->i_sb
->s_blocksize
- 1))
3592 ext4_block_truncate_page(handle
, mapping
, inode
->i_size
);
3595 * We add the inode to the orphan list, so that if this
3596 * truncate spans multiple transactions, and we crash, we will
3597 * resume the truncate when the filesystem recovers. It also
3598 * marks the inode dirty, to catch the new size.
3600 * Implication: the file must always be in a sane, consistent
3601 * truncatable state while each transaction commits.
3603 if (ext4_orphan_add(handle
, inode
))
3606 down_write(&EXT4_I(inode
)->i_data_sem
);
3608 ext4_discard_preallocations(inode
);
3610 if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
))
3611 ext4_ext_truncate(handle
, inode
);
3613 ext4_ind_truncate(handle
, inode
);
3615 up_write(&ei
->i_data_sem
);
3618 ext4_handle_sync(handle
);
3622 * If this was a simple ftruncate() and the file will remain alive,
3623 * then we need to clear up the orphan record which we created above.
3624 * However, if this was a real unlink then we were called by
3625 * ext4_delete_inode(), and we allow that function to clean up the
3626 * orphan info for us.
3629 ext4_orphan_del(handle
, inode
);
3631 inode
->i_mtime
= inode
->i_ctime
= ext4_current_time(inode
);
3632 ext4_mark_inode_dirty(handle
, inode
);
3633 ext4_journal_stop(handle
);
3635 trace_ext4_truncate_exit(inode
);
3639 * ext4_get_inode_loc returns with an extra refcount against the inode's
3640 * underlying buffer_head on success. If 'in_mem' is true, we have all
3641 * data in memory that is needed to recreate the on-disk version of this
3644 static int __ext4_get_inode_loc(struct inode
*inode
,
3645 struct ext4_iloc
*iloc
, int in_mem
)
3647 struct ext4_group_desc
*gdp
;
3648 struct buffer_head
*bh
;
3649 struct super_block
*sb
= inode
->i_sb
;
3651 int inodes_per_block
, inode_offset
;
3654 if (!ext4_valid_inum(sb
, inode
->i_ino
))
3657 iloc
->block_group
= (inode
->i_ino
- 1) / EXT4_INODES_PER_GROUP(sb
);
3658 gdp
= ext4_get_group_desc(sb
, iloc
->block_group
, NULL
);
3663 * Figure out the offset within the block group inode table
3665 inodes_per_block
= EXT4_SB(sb
)->s_inodes_per_block
;
3666 inode_offset
= ((inode
->i_ino
- 1) %
3667 EXT4_INODES_PER_GROUP(sb
));
3668 block
= ext4_inode_table(sb
, gdp
) + (inode_offset
/ inodes_per_block
);
3669 iloc
->offset
= (inode_offset
% inodes_per_block
) * EXT4_INODE_SIZE(sb
);
3671 bh
= sb_getblk(sb
, block
);
3674 if (!buffer_uptodate(bh
)) {
3678 * If the buffer has the write error flag, we have failed
3679 * to write out another inode in the same block. In this
3680 * case, we don't have to read the block because we may
3681 * read the old inode data successfully.
3683 if (buffer_write_io_error(bh
) && !buffer_uptodate(bh
))
3684 set_buffer_uptodate(bh
);
3686 if (buffer_uptodate(bh
)) {
3687 /* someone brought it uptodate while we waited */
3693 * If we have all information of the inode in memory and this
3694 * is the only valid inode in the block, we need not read the
3698 struct buffer_head
*bitmap_bh
;
3701 start
= inode_offset
& ~(inodes_per_block
- 1);
3703 /* Is the inode bitmap in cache? */
3704 bitmap_bh
= sb_getblk(sb
, ext4_inode_bitmap(sb
, gdp
));
3705 if (unlikely(!bitmap_bh
))
3709 * If the inode bitmap isn't in cache then the
3710 * optimisation may end up performing two reads instead
3711 * of one, so skip it.
3713 if (!buffer_uptodate(bitmap_bh
)) {
3717 for (i
= start
; i
< start
+ inodes_per_block
; i
++) {
3718 if (i
== inode_offset
)
3720 if (ext4_test_bit(i
, bitmap_bh
->b_data
))
3724 if (i
== start
+ inodes_per_block
) {
3725 /* all other inodes are free, so skip I/O */
3726 memset(bh
->b_data
, 0, bh
->b_size
);
3727 set_buffer_uptodate(bh
);
3735 * If we need to do any I/O, try to pre-readahead extra
3736 * blocks from the inode table.
3738 if (EXT4_SB(sb
)->s_inode_readahead_blks
) {
3739 ext4_fsblk_t b
, end
, table
;
3741 __u32 ra_blks
= EXT4_SB(sb
)->s_inode_readahead_blks
;
3743 table
= ext4_inode_table(sb
, gdp
);
3744 /* s_inode_readahead_blks is always a power of 2 */
3745 b
= block
& ~((ext4_fsblk_t
) ra_blks
- 1);
3749 num
= EXT4_INODES_PER_GROUP(sb
);
3750 if (ext4_has_group_desc_csum(sb
))
3751 num
-= ext4_itable_unused_count(sb
, gdp
);
3752 table
+= num
/ inodes_per_block
;
3756 sb_breadahead(sb
, b
++);
3760 * There are other valid inodes in the buffer, this inode
3761 * has in-inode xattrs, or we don't have this inode in memory.
3762 * Read the block from disk.
3764 trace_ext4_load_inode(inode
);
3766 bh
->b_end_io
= end_buffer_read_sync
;
3767 submit_bh(READ
| REQ_META
| REQ_PRIO
, bh
);
3769 if (!buffer_uptodate(bh
)) {
3770 EXT4_ERROR_INODE_BLOCK(inode
, block
,
3771 "unable to read itable block");
3781 int ext4_get_inode_loc(struct inode
*inode
, struct ext4_iloc
*iloc
)
3783 /* We have all inode data except xattrs in memory here. */
3784 return __ext4_get_inode_loc(inode
, iloc
,
3785 !ext4_test_inode_state(inode
, EXT4_STATE_XATTR
));
3788 void ext4_set_inode_flags(struct inode
*inode
)
3790 unsigned int flags
= EXT4_I(inode
)->i_flags
;
3791 unsigned int new_fl
= 0;
3793 if (flags
& EXT4_SYNC_FL
)
3795 if (flags
& EXT4_APPEND_FL
)
3797 if (flags
& EXT4_IMMUTABLE_FL
)
3798 new_fl
|= S_IMMUTABLE
;
3799 if (flags
& EXT4_NOATIME_FL
)
3800 new_fl
|= S_NOATIME
;
3801 if (flags
& EXT4_DIRSYNC_FL
)
3802 new_fl
|= S_DIRSYNC
;
3803 inode_set_flags(inode
, new_fl
,
3804 S_SYNC
|S_APPEND
|S_IMMUTABLE
|S_NOATIME
|S_DIRSYNC
);
3807 /* Propagate flags from i_flags to EXT4_I(inode)->i_flags */
3808 void ext4_get_inode_flags(struct ext4_inode_info
*ei
)
3810 unsigned int vfs_fl
;
3811 unsigned long old_fl
, new_fl
;
3814 vfs_fl
= ei
->vfs_inode
.i_flags
;
3815 old_fl
= ei
->i_flags
;
3816 new_fl
= old_fl
& ~(EXT4_SYNC_FL
|EXT4_APPEND_FL
|
3817 EXT4_IMMUTABLE_FL
|EXT4_NOATIME_FL
|
3819 if (vfs_fl
& S_SYNC
)
3820 new_fl
|= EXT4_SYNC_FL
;
3821 if (vfs_fl
& S_APPEND
)
3822 new_fl
|= EXT4_APPEND_FL
;
3823 if (vfs_fl
& S_IMMUTABLE
)
3824 new_fl
|= EXT4_IMMUTABLE_FL
;
3825 if (vfs_fl
& S_NOATIME
)
3826 new_fl
|= EXT4_NOATIME_FL
;
3827 if (vfs_fl
& S_DIRSYNC
)
3828 new_fl
|= EXT4_DIRSYNC_FL
;
3829 } while (cmpxchg(&ei
->i_flags
, old_fl
, new_fl
) != old_fl
);
3832 static blkcnt_t
ext4_inode_blocks(struct ext4_inode
*raw_inode
,
3833 struct ext4_inode_info
*ei
)
3836 struct inode
*inode
= &(ei
->vfs_inode
);
3837 struct super_block
*sb
= inode
->i_sb
;
3839 if (EXT4_HAS_RO_COMPAT_FEATURE(sb
,
3840 EXT4_FEATURE_RO_COMPAT_HUGE_FILE
)) {
3841 /* we are using combined 48 bit field */
3842 i_blocks
= ((u64
)le16_to_cpu(raw_inode
->i_blocks_high
)) << 32 |
3843 le32_to_cpu(raw_inode
->i_blocks_lo
);
3844 if (ext4_test_inode_flag(inode
, EXT4_INODE_HUGE_FILE
)) {
3845 /* i_blocks represent file system block size */
3846 return i_blocks
<< (inode
->i_blkbits
- 9);
3851 return le32_to_cpu(raw_inode
->i_blocks_lo
);
3855 static inline void ext4_iget_extra_inode(struct inode
*inode
,
3856 struct ext4_inode
*raw_inode
,
3857 struct ext4_inode_info
*ei
)
3859 __le32
*magic
= (void *)raw_inode
+
3860 EXT4_GOOD_OLD_INODE_SIZE
+ ei
->i_extra_isize
;
3861 if (*magic
== cpu_to_le32(EXT4_XATTR_MAGIC
)) {
3862 ext4_set_inode_state(inode
, EXT4_STATE_XATTR
);
3863 ext4_find_inline_data_nolock(inode
);
3865 EXT4_I(inode
)->i_inline_off
= 0;
3868 struct inode
*ext4_iget(struct super_block
*sb
, unsigned long ino
)
3870 struct ext4_iloc iloc
;
3871 struct ext4_inode
*raw_inode
;
3872 struct ext4_inode_info
*ei
;
3873 struct inode
*inode
;
3874 journal_t
*journal
= EXT4_SB(sb
)->s_journal
;
3880 inode
= iget_locked(sb
, ino
);
3882 return ERR_PTR(-ENOMEM
);
3883 if (!(inode
->i_state
& I_NEW
))
3889 ret
= __ext4_get_inode_loc(inode
, &iloc
, 0);
3892 raw_inode
= ext4_raw_inode(&iloc
);
3894 if (EXT4_INODE_SIZE(inode
->i_sb
) > EXT4_GOOD_OLD_INODE_SIZE
) {
3895 ei
->i_extra_isize
= le16_to_cpu(raw_inode
->i_extra_isize
);
3896 if (EXT4_GOOD_OLD_INODE_SIZE
+ ei
->i_extra_isize
>
3897 EXT4_INODE_SIZE(inode
->i_sb
)) {
3898 EXT4_ERROR_INODE(inode
, "bad extra_isize (%u != %u)",
3899 EXT4_GOOD_OLD_INODE_SIZE
+ ei
->i_extra_isize
,
3900 EXT4_INODE_SIZE(inode
->i_sb
));
3905 ei
->i_extra_isize
= 0;
3907 /* Precompute checksum seed for inode metadata */
3908 if (ext4_has_metadata_csum(sb
)) {
3909 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
3911 __le32 inum
= cpu_to_le32(inode
->i_ino
);
3912 __le32 gen
= raw_inode
->i_generation
;
3913 csum
= ext4_chksum(sbi
, sbi
->s_csum_seed
, (__u8
*)&inum
,
3915 ei
->i_csum_seed
= ext4_chksum(sbi
, csum
, (__u8
*)&gen
,
3919 if (!ext4_inode_csum_verify(inode
, raw_inode
, ei
)) {
3920 EXT4_ERROR_INODE(inode
, "checksum invalid");
3925 inode
->i_mode
= le16_to_cpu(raw_inode
->i_mode
);
3926 i_uid
= (uid_t
)le16_to_cpu(raw_inode
->i_uid_low
);
3927 i_gid
= (gid_t
)le16_to_cpu(raw_inode
->i_gid_low
);
3928 if (!(test_opt(inode
->i_sb
, NO_UID32
))) {
3929 i_uid
|= le16_to_cpu(raw_inode
->i_uid_high
) << 16;
3930 i_gid
|= le16_to_cpu(raw_inode
->i_gid_high
) << 16;
3932 i_uid_write(inode
, i_uid
);
3933 i_gid_write(inode
, i_gid
);
3934 set_nlink(inode
, le16_to_cpu(raw_inode
->i_links_count
));
3936 ext4_clear_state_flags(ei
); /* Only relevant on 32-bit archs */
3937 ei
->i_inline_off
= 0;
3938 ei
->i_dir_start_lookup
= 0;
3939 ei
->i_dtime
= le32_to_cpu(raw_inode
->i_dtime
);
3940 /* We now have enough fields to check if the inode was active or not.
3941 * This is needed because nfsd might try to access dead inodes
3942 * the test is that same one that e2fsck uses
3943 * NeilBrown 1999oct15
3945 if (inode
->i_nlink
== 0) {
3946 if ((inode
->i_mode
== 0 ||
3947 !(EXT4_SB(inode
->i_sb
)->s_mount_state
& EXT4_ORPHAN_FS
)) &&
3948 ino
!= EXT4_BOOT_LOADER_INO
) {
3949 /* this inode is deleted */
3953 /* The only unlinked inodes we let through here have
3954 * valid i_mode and are being read by the orphan
3955 * recovery code: that's fine, we're about to complete
3956 * the process of deleting those.
3957 * OR it is the EXT4_BOOT_LOADER_INO which is
3958 * not initialized on a new filesystem. */
3960 ei
->i_flags
= le32_to_cpu(raw_inode
->i_flags
);
3961 inode
->i_blocks
= ext4_inode_blocks(raw_inode
, ei
);
3962 ei
->i_file_acl
= le32_to_cpu(raw_inode
->i_file_acl_lo
);
3963 if (EXT4_HAS_INCOMPAT_FEATURE(sb
, EXT4_FEATURE_INCOMPAT_64BIT
))
3965 ((__u64
)le16_to_cpu(raw_inode
->i_file_acl_high
)) << 32;
3966 inode
->i_size
= ext4_isize(raw_inode
);
3967 ei
->i_disksize
= inode
->i_size
;
3969 ei
->i_reserved_quota
= 0;
3971 inode
->i_generation
= le32_to_cpu(raw_inode
->i_generation
);
3972 ei
->i_block_group
= iloc
.block_group
;
3973 ei
->i_last_alloc_group
= ~0;
3975 * NOTE! The in-memory inode i_data array is in little-endian order
3976 * even on big-endian machines: we do NOT byteswap the block numbers!
3978 for (block
= 0; block
< EXT4_N_BLOCKS
; block
++)
3979 ei
->i_data
[block
] = raw_inode
->i_block
[block
];
3980 INIT_LIST_HEAD(&ei
->i_orphan
);
3983 * Set transaction id's of transactions that have to be committed
3984 * to finish f[data]sync. We set them to currently running transaction
3985 * as we cannot be sure that the inode or some of its metadata isn't
3986 * part of the transaction - the inode could have been reclaimed and
3987 * now it is reread from disk.
3990 transaction_t
*transaction
;
3993 read_lock(&journal
->j_state_lock
);
3994 if (journal
->j_running_transaction
)
3995 transaction
= journal
->j_running_transaction
;
3997 transaction
= journal
->j_committing_transaction
;
3999 tid
= transaction
->t_tid
;
4001 tid
= journal
->j_commit_sequence
;
4002 read_unlock(&journal
->j_state_lock
);
4003 ei
->i_sync_tid
= tid
;
4004 ei
->i_datasync_tid
= tid
;
4007 if (EXT4_INODE_SIZE(inode
->i_sb
) > EXT4_GOOD_OLD_INODE_SIZE
) {
4008 if (ei
->i_extra_isize
== 0) {
4009 /* The extra space is currently unused. Use it. */
4010 ei
->i_extra_isize
= sizeof(struct ext4_inode
) -
4011 EXT4_GOOD_OLD_INODE_SIZE
;
4013 ext4_iget_extra_inode(inode
, raw_inode
, ei
);
4017 EXT4_INODE_GET_XTIME(i_ctime
, inode
, raw_inode
);
4018 EXT4_INODE_GET_XTIME(i_mtime
, inode
, raw_inode
);
4019 EXT4_INODE_GET_XTIME(i_atime
, inode
, raw_inode
);
4020 EXT4_EINODE_GET_XTIME(i_crtime
, ei
, raw_inode
);
4022 if (likely(!test_opt2(inode
->i_sb
, HURD_COMPAT
))) {
4023 inode
->i_version
= le32_to_cpu(raw_inode
->i_disk_version
);
4024 if (EXT4_INODE_SIZE(inode
->i_sb
) > EXT4_GOOD_OLD_INODE_SIZE
) {
4025 if (EXT4_FITS_IN_INODE(raw_inode
, ei
, i_version_hi
))
4027 (__u64
)(le32_to_cpu(raw_inode
->i_version_hi
)) << 32;
4032 if (ei
->i_file_acl
&&
4033 !ext4_data_block_valid(EXT4_SB(sb
), ei
->i_file_acl
, 1)) {
4034 EXT4_ERROR_INODE(inode
, "bad extended attribute block %llu",
4038 } else if (!ext4_has_inline_data(inode
)) {
4039 if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
)) {
4040 if ((S_ISREG(inode
->i_mode
) || S_ISDIR(inode
->i_mode
) ||
4041 (S_ISLNK(inode
->i_mode
) &&
4042 !ext4_inode_is_fast_symlink(inode
))))
4043 /* Validate extent which is part of inode */
4044 ret
= ext4_ext_check_inode(inode
);
4045 } else if (S_ISREG(inode
->i_mode
) || S_ISDIR(inode
->i_mode
) ||
4046 (S_ISLNK(inode
->i_mode
) &&
4047 !ext4_inode_is_fast_symlink(inode
))) {
4048 /* Validate block references which are part of inode */
4049 ret
= ext4_ind_check_inode(inode
);
4055 if (S_ISREG(inode
->i_mode
)) {
4056 inode
->i_op
= &ext4_file_inode_operations
;
4057 inode
->i_fop
= &ext4_file_operations
;
4058 ext4_set_aops(inode
);
4059 } else if (S_ISDIR(inode
->i_mode
)) {
4060 inode
->i_op
= &ext4_dir_inode_operations
;
4061 inode
->i_fop
= &ext4_dir_operations
;
4062 } else if (S_ISLNK(inode
->i_mode
)) {
4063 if (ext4_inode_is_fast_symlink(inode
)) {
4064 inode
->i_op
= &ext4_fast_symlink_inode_operations
;
4065 nd_terminate_link(ei
->i_data
, inode
->i_size
,
4066 sizeof(ei
->i_data
) - 1);
4068 inode
->i_op
= &ext4_symlink_inode_operations
;
4069 ext4_set_aops(inode
);
4071 } else if (S_ISCHR(inode
->i_mode
) || S_ISBLK(inode
->i_mode
) ||
4072 S_ISFIFO(inode
->i_mode
) || S_ISSOCK(inode
->i_mode
)) {
4073 inode
->i_op
= &ext4_special_inode_operations
;
4074 if (raw_inode
->i_block
[0])
4075 init_special_inode(inode
, inode
->i_mode
,
4076 old_decode_dev(le32_to_cpu(raw_inode
->i_block
[0])));
4078 init_special_inode(inode
, inode
->i_mode
,
4079 new_decode_dev(le32_to_cpu(raw_inode
->i_block
[1])));
4080 } else if (ino
== EXT4_BOOT_LOADER_INO
) {
4081 make_bad_inode(inode
);
4084 EXT4_ERROR_INODE(inode
, "bogus i_mode (%o)", inode
->i_mode
);
4088 ext4_set_inode_flags(inode
);
4089 unlock_new_inode(inode
);
4095 return ERR_PTR(ret
);
4098 struct inode
*ext4_iget_normal(struct super_block
*sb
, unsigned long ino
)
4100 if (ino
< EXT4_FIRST_INO(sb
) && ino
!= EXT4_ROOT_INO
)
4101 return ERR_PTR(-EIO
);
4102 return ext4_iget(sb
, ino
);
4105 static int ext4_inode_blocks_set(handle_t
*handle
,
4106 struct ext4_inode
*raw_inode
,
4107 struct ext4_inode_info
*ei
)
4109 struct inode
*inode
= &(ei
->vfs_inode
);
4110 u64 i_blocks
= inode
->i_blocks
;
4111 struct super_block
*sb
= inode
->i_sb
;
4113 if (i_blocks
<= ~0U) {
4115 * i_blocks can be represented in a 32 bit variable
4116 * as multiple of 512 bytes
4118 raw_inode
->i_blocks_lo
= cpu_to_le32(i_blocks
);
4119 raw_inode
->i_blocks_high
= 0;
4120 ext4_clear_inode_flag(inode
, EXT4_INODE_HUGE_FILE
);
4123 if (!EXT4_HAS_RO_COMPAT_FEATURE(sb
, EXT4_FEATURE_RO_COMPAT_HUGE_FILE
))
4126 if (i_blocks
<= 0xffffffffffffULL
) {
4128 * i_blocks can be represented in a 48 bit variable
4129 * as multiple of 512 bytes
4131 raw_inode
->i_blocks_lo
= cpu_to_le32(i_blocks
);
4132 raw_inode
->i_blocks_high
= cpu_to_le16(i_blocks
>> 32);
4133 ext4_clear_inode_flag(inode
, EXT4_INODE_HUGE_FILE
);
4135 ext4_set_inode_flag(inode
, EXT4_INODE_HUGE_FILE
);
4136 /* i_block is stored in file system block size */
4137 i_blocks
= i_blocks
>> (inode
->i_blkbits
- 9);
4138 raw_inode
->i_blocks_lo
= cpu_to_le32(i_blocks
);
4139 raw_inode
->i_blocks_high
= cpu_to_le16(i_blocks
>> 32);
4145 * Post the struct inode info into an on-disk inode location in the
4146 * buffer-cache. This gobbles the caller's reference to the
4147 * buffer_head in the inode location struct.
4149 * The caller must have write access to iloc->bh.
4151 static int ext4_do_update_inode(handle_t
*handle
,
4152 struct inode
*inode
,
4153 struct ext4_iloc
*iloc
)
4155 struct ext4_inode
*raw_inode
= ext4_raw_inode(iloc
);
4156 struct ext4_inode_info
*ei
= EXT4_I(inode
);
4157 struct buffer_head
*bh
= iloc
->bh
;
4158 struct super_block
*sb
= inode
->i_sb
;
4159 int err
= 0, rc
, block
;
4160 int need_datasync
= 0, set_large_file
= 0;
4164 spin_lock(&ei
->i_raw_lock
);
4166 /* For fields not tracked in the in-memory inode,
4167 * initialise them to zero for new inodes. */
4168 if (ext4_test_inode_state(inode
, EXT4_STATE_NEW
))
4169 memset(raw_inode
, 0, EXT4_SB(inode
->i_sb
)->s_inode_size
);
4171 ext4_get_inode_flags(ei
);
4172 raw_inode
->i_mode
= cpu_to_le16(inode
->i_mode
);
4173 i_uid
= i_uid_read(inode
);
4174 i_gid
= i_gid_read(inode
);
4175 if (!(test_opt(inode
->i_sb
, NO_UID32
))) {
4176 raw_inode
->i_uid_low
= cpu_to_le16(low_16_bits(i_uid
));
4177 raw_inode
->i_gid_low
= cpu_to_le16(low_16_bits(i_gid
));
4179 * Fix up interoperability with old kernels. Otherwise, old inodes get
4180 * re-used with the upper 16 bits of the uid/gid intact
4183 raw_inode
->i_uid_high
=
4184 cpu_to_le16(high_16_bits(i_uid
));
4185 raw_inode
->i_gid_high
=
4186 cpu_to_le16(high_16_bits(i_gid
));
4188 raw_inode
->i_uid_high
= 0;
4189 raw_inode
->i_gid_high
= 0;
4192 raw_inode
->i_uid_low
= cpu_to_le16(fs_high2lowuid(i_uid
));
4193 raw_inode
->i_gid_low
= cpu_to_le16(fs_high2lowgid(i_gid
));
4194 raw_inode
->i_uid_high
= 0;
4195 raw_inode
->i_gid_high
= 0;
4197 raw_inode
->i_links_count
= cpu_to_le16(inode
->i_nlink
);
4199 EXT4_INODE_SET_XTIME(i_ctime
, inode
, raw_inode
);
4200 EXT4_INODE_SET_XTIME(i_mtime
, inode
, raw_inode
);
4201 EXT4_INODE_SET_XTIME(i_atime
, inode
, raw_inode
);
4202 EXT4_EINODE_SET_XTIME(i_crtime
, ei
, raw_inode
);
4204 err
= ext4_inode_blocks_set(handle
, raw_inode
, ei
);
4206 spin_unlock(&ei
->i_raw_lock
);
4209 raw_inode
->i_dtime
= cpu_to_le32(ei
->i_dtime
);
4210 raw_inode
->i_flags
= cpu_to_le32(ei
->i_flags
& 0xFFFFFFFF);
4211 if (likely(!test_opt2(inode
->i_sb
, HURD_COMPAT
)))
4212 raw_inode
->i_file_acl_high
=
4213 cpu_to_le16(ei
->i_file_acl
>> 32);
4214 raw_inode
->i_file_acl_lo
= cpu_to_le32(ei
->i_file_acl
);
4215 if (ei
->i_disksize
!= ext4_isize(raw_inode
)) {
4216 ext4_isize_set(raw_inode
, ei
->i_disksize
);
4219 if (ei
->i_disksize
> 0x7fffffffULL
) {
4220 if (!EXT4_HAS_RO_COMPAT_FEATURE(sb
,
4221 EXT4_FEATURE_RO_COMPAT_LARGE_FILE
) ||
4222 EXT4_SB(sb
)->s_es
->s_rev_level
==
4223 cpu_to_le32(EXT4_GOOD_OLD_REV
))
4226 raw_inode
->i_generation
= cpu_to_le32(inode
->i_generation
);
4227 if (S_ISCHR(inode
->i_mode
) || S_ISBLK(inode
->i_mode
)) {
4228 if (old_valid_dev(inode
->i_rdev
)) {
4229 raw_inode
->i_block
[0] =
4230 cpu_to_le32(old_encode_dev(inode
->i_rdev
));
4231 raw_inode
->i_block
[1] = 0;
4233 raw_inode
->i_block
[0] = 0;
4234 raw_inode
->i_block
[1] =
4235 cpu_to_le32(new_encode_dev(inode
->i_rdev
));
4236 raw_inode
->i_block
[2] = 0;
4238 } else if (!ext4_has_inline_data(inode
)) {
4239 for (block
= 0; block
< EXT4_N_BLOCKS
; block
++)
4240 raw_inode
->i_block
[block
] = ei
->i_data
[block
];
4243 if (likely(!test_opt2(inode
->i_sb
, HURD_COMPAT
))) {
4244 raw_inode
->i_disk_version
= cpu_to_le32(inode
->i_version
);
4245 if (ei
->i_extra_isize
) {
4246 if (EXT4_FITS_IN_INODE(raw_inode
, ei
, i_version_hi
))
4247 raw_inode
->i_version_hi
=
4248 cpu_to_le32(inode
->i_version
>> 32);
4249 raw_inode
->i_extra_isize
=
4250 cpu_to_le16(ei
->i_extra_isize
);
4254 ext4_inode_csum_set(inode
, raw_inode
, ei
);
4256 spin_unlock(&ei
->i_raw_lock
);
4258 BUFFER_TRACE(bh
, "call ext4_handle_dirty_metadata");
4259 rc
= ext4_handle_dirty_metadata(handle
, NULL
, bh
);
4262 ext4_clear_inode_state(inode
, EXT4_STATE_NEW
);
4263 if (set_large_file
) {
4264 BUFFER_TRACE(EXT4_SB(sb
)->s_sbh
, "get write access");
4265 err
= ext4_journal_get_write_access(handle
, EXT4_SB(sb
)->s_sbh
);
4268 ext4_update_dynamic_rev(sb
);
4269 EXT4_SET_RO_COMPAT_FEATURE(sb
,
4270 EXT4_FEATURE_RO_COMPAT_LARGE_FILE
);
4271 ext4_handle_sync(handle
);
4272 err
= ext4_handle_dirty_super(handle
, sb
);
4274 ext4_update_inode_fsync_trans(handle
, inode
, need_datasync
);
4277 ext4_std_error(inode
->i_sb
, err
);
4282 * ext4_write_inode()
4284 * We are called from a few places:
4286 * - Within generic_file_aio_write() -> generic_write_sync() for O_SYNC files.
4287 * Here, there will be no transaction running. We wait for any running
4288 * transaction to commit.
4290 * - Within flush work (sys_sync(), kupdate and such).
4291 * We wait on commit, if told to.
4293 * - Within iput_final() -> write_inode_now()
4294 * We wait on commit, if told to.
4296 * In all cases it is actually safe for us to return without doing anything,
4297 * because the inode has been copied into a raw inode buffer in
4298 * ext4_mark_inode_dirty(). This is a correctness thing for WB_SYNC_ALL
4301 * Note that we are absolutely dependent upon all inode dirtiers doing the
4302 * right thing: they *must* call mark_inode_dirty() after dirtying info in
4303 * which we are interested.
4305 * It would be a bug for them to not do this. The code:
4307 * mark_inode_dirty(inode)
4309 * inode->i_size = expr;
4311 * is in error because write_inode() could occur while `stuff()' is running,
4312 * and the new i_size will be lost. Plus the inode will no longer be on the
4313 * superblock's dirty inode list.
4315 int ext4_write_inode(struct inode
*inode
, struct writeback_control
*wbc
)
4319 if (WARN_ON_ONCE(current
->flags
& PF_MEMALLOC
))
4322 if (EXT4_SB(inode
->i_sb
)->s_journal
) {
4323 if (ext4_journal_current_handle()) {
4324 jbd_debug(1, "called recursively, non-PF_MEMALLOC!\n");
4330 * No need to force transaction in WB_SYNC_NONE mode. Also
4331 * ext4_sync_fs() will force the commit after everything is
4334 if (wbc
->sync_mode
!= WB_SYNC_ALL
|| wbc
->for_sync
)
4337 err
= ext4_force_commit(inode
->i_sb
);
4339 struct ext4_iloc iloc
;
4341 err
= __ext4_get_inode_loc(inode
, &iloc
, 0);
4345 * sync(2) will flush the whole buffer cache. No need to do
4346 * it here separately for each inode.
4348 if (wbc
->sync_mode
== WB_SYNC_ALL
&& !wbc
->for_sync
)
4349 sync_dirty_buffer(iloc
.bh
);
4350 if (buffer_req(iloc
.bh
) && !buffer_uptodate(iloc
.bh
)) {
4351 EXT4_ERROR_INODE_BLOCK(inode
, iloc
.bh
->b_blocknr
,
4352 "IO error syncing inode");
4361 * In data=journal mode ext4_journalled_invalidatepage() may fail to invalidate
4362 * buffers that are attached to a page stradding i_size and are undergoing
4363 * commit. In that case we have to wait for commit to finish and try again.
4365 static void ext4_wait_for_tail_page_commit(struct inode
*inode
)
4369 journal_t
*journal
= EXT4_SB(inode
->i_sb
)->s_journal
;
4370 tid_t commit_tid
= 0;
4373 offset
= inode
->i_size
& (PAGE_CACHE_SIZE
- 1);
4375 * All buffers in the last page remain valid? Then there's nothing to
4376 * do. We do the check mainly to optimize the common PAGE_CACHE_SIZE ==
4379 if (offset
> PAGE_CACHE_SIZE
- (1 << inode
->i_blkbits
))
4382 page
= find_lock_page(inode
->i_mapping
,
4383 inode
->i_size
>> PAGE_CACHE_SHIFT
);
4386 ret
= __ext4_journalled_invalidatepage(page
, offset
,
4387 PAGE_CACHE_SIZE
- offset
);
4389 page_cache_release(page
);
4393 read_lock(&journal
->j_state_lock
);
4394 if (journal
->j_committing_transaction
)
4395 commit_tid
= journal
->j_committing_transaction
->t_tid
;
4396 read_unlock(&journal
->j_state_lock
);
4398 jbd2_log_wait_commit(journal
, commit_tid
);
4405 * Called from notify_change.
4407 * We want to trap VFS attempts to truncate the file as soon as
4408 * possible. In particular, we want to make sure that when the VFS
4409 * shrinks i_size, we put the inode on the orphan list and modify
4410 * i_disksize immediately, so that during the subsequent flushing of
4411 * dirty pages and freeing of disk blocks, we can guarantee that any
4412 * commit will leave the blocks being flushed in an unused state on
4413 * disk. (On recovery, the inode will get truncated and the blocks will
4414 * be freed, so we have a strong guarantee that no future commit will
4415 * leave these blocks visible to the user.)
4417 * Another thing we have to assure is that if we are in ordered mode
4418 * and inode is still attached to the committing transaction, we must
4419 * we start writeout of all the dirty pages which are being truncated.
4420 * This way we are sure that all the data written in the previous
4421 * transaction are already on disk (truncate waits for pages under
4424 * Called with inode->i_mutex down.
4426 int ext4_setattr(struct dentry
*dentry
, struct iattr
*attr
)
4428 struct inode
*inode
= dentry
->d_inode
;
4431 const unsigned int ia_valid
= attr
->ia_valid
;
4433 error
= inode_change_ok(inode
, attr
);
4437 if (is_quota_modification(inode
, attr
))
4438 dquot_initialize(inode
);
4439 if ((ia_valid
& ATTR_UID
&& !uid_eq(attr
->ia_uid
, inode
->i_uid
)) ||
4440 (ia_valid
& ATTR_GID
&& !gid_eq(attr
->ia_gid
, inode
->i_gid
))) {
4443 /* (user+group)*(old+new) structure, inode write (sb,
4444 * inode block, ? - but truncate inode update has it) */
4445 handle
= ext4_journal_start(inode
, EXT4_HT_QUOTA
,
4446 (EXT4_MAXQUOTAS_INIT_BLOCKS(inode
->i_sb
) +
4447 EXT4_MAXQUOTAS_DEL_BLOCKS(inode
->i_sb
)) + 3);
4448 if (IS_ERR(handle
)) {
4449 error
= PTR_ERR(handle
);
4452 error
= dquot_transfer(inode
, attr
);
4454 ext4_journal_stop(handle
);
4457 /* Update corresponding info in inode so that everything is in
4458 * one transaction */
4459 if (attr
->ia_valid
& ATTR_UID
)
4460 inode
->i_uid
= attr
->ia_uid
;
4461 if (attr
->ia_valid
& ATTR_GID
)
4462 inode
->i_gid
= attr
->ia_gid
;
4463 error
= ext4_mark_inode_dirty(handle
, inode
);
4464 ext4_journal_stop(handle
);
4467 if (attr
->ia_valid
& ATTR_SIZE
&& attr
->ia_size
!= inode
->i_size
) {
4470 if (!(ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
))) {
4471 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
4473 if (attr
->ia_size
> sbi
->s_bitmap_maxbytes
)
4477 if (IS_I_VERSION(inode
) && attr
->ia_size
!= inode
->i_size
)
4478 inode_inc_iversion(inode
);
4480 if (S_ISREG(inode
->i_mode
) &&
4481 (attr
->ia_size
< inode
->i_size
)) {
4482 if (ext4_should_order_data(inode
)) {
4483 error
= ext4_begin_ordered_truncate(inode
,
4488 handle
= ext4_journal_start(inode
, EXT4_HT_INODE
, 3);
4489 if (IS_ERR(handle
)) {
4490 error
= PTR_ERR(handle
);
4493 if (ext4_handle_valid(handle
)) {
4494 error
= ext4_orphan_add(handle
, inode
);
4497 down_write(&EXT4_I(inode
)->i_data_sem
);
4498 EXT4_I(inode
)->i_disksize
= attr
->ia_size
;
4499 rc
= ext4_mark_inode_dirty(handle
, inode
);
4503 * We have to update i_size under i_data_sem together
4504 * with i_disksize to avoid races with writeback code
4505 * running ext4_wb_update_i_disksize().
4508 i_size_write(inode
, attr
->ia_size
);
4509 up_write(&EXT4_I(inode
)->i_data_sem
);
4510 ext4_journal_stop(handle
);
4512 ext4_orphan_del(NULL
, inode
);
4516 loff_t oldsize
= inode
->i_size
;
4518 i_size_write(inode
, attr
->ia_size
);
4519 pagecache_isize_extended(inode
, oldsize
, inode
->i_size
);
4523 * Blocks are going to be removed from the inode. Wait
4524 * for dio in flight. Temporarily disable
4525 * dioread_nolock to prevent livelock.
4528 if (!ext4_should_journal_data(inode
)) {
4529 ext4_inode_block_unlocked_dio(inode
);
4530 inode_dio_wait(inode
);
4531 ext4_inode_resume_unlocked_dio(inode
);
4533 ext4_wait_for_tail_page_commit(inode
);
4536 * Truncate pagecache after we've waited for commit
4537 * in data=journal mode to make pages freeable.
4539 truncate_pagecache(inode
, inode
->i_size
);
4542 * We want to call ext4_truncate() even if attr->ia_size ==
4543 * inode->i_size for cases like truncation of fallocated space
4545 if (attr
->ia_valid
& ATTR_SIZE
)
4546 ext4_truncate(inode
);
4549 setattr_copy(inode
, attr
);
4550 mark_inode_dirty(inode
);
4554 * If the call to ext4_truncate failed to get a transaction handle at
4555 * all, we need to clean up the in-core orphan list manually.
4557 if (orphan
&& inode
->i_nlink
)
4558 ext4_orphan_del(NULL
, inode
);
4560 if (!rc
&& (ia_valid
& ATTR_MODE
))
4561 rc
= posix_acl_chmod(inode
, inode
->i_mode
);
4564 ext4_std_error(inode
->i_sb
, error
);
4570 int ext4_getattr(struct vfsmount
*mnt
, struct dentry
*dentry
,
4573 struct inode
*inode
;
4574 unsigned long long delalloc_blocks
;
4576 inode
= dentry
->d_inode
;
4577 generic_fillattr(inode
, stat
);
4580 * If there is inline data in the inode, the inode will normally not
4581 * have data blocks allocated (it may have an external xattr block).
4582 * Report at least one sector for such files, so tools like tar, rsync,
4583 * others doen't incorrectly think the file is completely sparse.
4585 if (unlikely(ext4_has_inline_data(inode
)))
4586 stat
->blocks
+= (stat
->size
+ 511) >> 9;
4589 * We can't update i_blocks if the block allocation is delayed
4590 * otherwise in the case of system crash before the real block
4591 * allocation is done, we will have i_blocks inconsistent with
4592 * on-disk file blocks.
4593 * We always keep i_blocks updated together with real
4594 * allocation. But to not confuse with user, stat
4595 * will return the blocks that include the delayed allocation
4596 * blocks for this file.
4598 delalloc_blocks
= EXT4_C2B(EXT4_SB(inode
->i_sb
),
4599 EXT4_I(inode
)->i_reserved_data_blocks
);
4600 stat
->blocks
+= delalloc_blocks
<< (inode
->i_sb
->s_blocksize_bits
- 9);
4604 static int ext4_index_trans_blocks(struct inode
*inode
, int lblocks
,
4607 if (!(ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
)))
4608 return ext4_ind_trans_blocks(inode
, lblocks
);
4609 return ext4_ext_index_trans_blocks(inode
, pextents
);
4613 * Account for index blocks, block groups bitmaps and block group
4614 * descriptor blocks if modify datablocks and index blocks
4615 * worse case, the indexs blocks spread over different block groups
4617 * If datablocks are discontiguous, they are possible to spread over
4618 * different block groups too. If they are contiguous, with flexbg,
4619 * they could still across block group boundary.
4621 * Also account for superblock, inode, quota and xattr blocks
4623 static int ext4_meta_trans_blocks(struct inode
*inode
, int lblocks
,
4626 ext4_group_t groups
, ngroups
= ext4_get_groups_count(inode
->i_sb
);
4632 * How many index blocks need to touch to map @lblocks logical blocks
4633 * to @pextents physical extents?
4635 idxblocks
= ext4_index_trans_blocks(inode
, lblocks
, pextents
);
4640 * Now let's see how many group bitmaps and group descriptors need
4643 groups
= idxblocks
+ pextents
;
4645 if (groups
> ngroups
)
4647 if (groups
> EXT4_SB(inode
->i_sb
)->s_gdb_count
)
4648 gdpblocks
= EXT4_SB(inode
->i_sb
)->s_gdb_count
;
4650 /* bitmaps and block group descriptor blocks */
4651 ret
+= groups
+ gdpblocks
;
4653 /* Blocks for super block, inode, quota and xattr blocks */
4654 ret
+= EXT4_META_TRANS_BLOCKS(inode
->i_sb
);
4660 * Calculate the total number of credits to reserve to fit
4661 * the modification of a single pages into a single transaction,
4662 * which may include multiple chunks of block allocations.
4664 * This could be called via ext4_write_begin()
4666 * We need to consider the worse case, when
4667 * one new block per extent.
4669 int ext4_writepage_trans_blocks(struct inode
*inode
)
4671 int bpp
= ext4_journal_blocks_per_page(inode
);
4674 ret
= ext4_meta_trans_blocks(inode
, bpp
, bpp
);
4676 /* Account for data blocks for journalled mode */
4677 if (ext4_should_journal_data(inode
))
4683 * Calculate the journal credits for a chunk of data modification.
4685 * This is called from DIO, fallocate or whoever calling
4686 * ext4_map_blocks() to map/allocate a chunk of contiguous disk blocks.
4688 * journal buffers for data blocks are not included here, as DIO
4689 * and fallocate do no need to journal data buffers.
4691 int ext4_chunk_trans_blocks(struct inode
*inode
, int nrblocks
)
4693 return ext4_meta_trans_blocks(inode
, nrblocks
, 1);
4697 * The caller must have previously called ext4_reserve_inode_write().
4698 * Give this, we know that the caller already has write access to iloc->bh.
4700 int ext4_mark_iloc_dirty(handle_t
*handle
,
4701 struct inode
*inode
, struct ext4_iloc
*iloc
)
4705 if (IS_I_VERSION(inode
))
4706 inode_inc_iversion(inode
);
4708 /* the do_update_inode consumes one bh->b_count */
4711 /* ext4_do_update_inode() does jbd2_journal_dirty_metadata */
4712 err
= ext4_do_update_inode(handle
, inode
, iloc
);
4718 * On success, We end up with an outstanding reference count against
4719 * iloc->bh. This _must_ be cleaned up later.
4723 ext4_reserve_inode_write(handle_t
*handle
, struct inode
*inode
,
4724 struct ext4_iloc
*iloc
)
4728 err
= ext4_get_inode_loc(inode
, iloc
);
4730 BUFFER_TRACE(iloc
->bh
, "get_write_access");
4731 err
= ext4_journal_get_write_access(handle
, iloc
->bh
);
4737 ext4_std_error(inode
->i_sb
, err
);
4742 * Expand an inode by new_extra_isize bytes.
4743 * Returns 0 on success or negative error number on failure.
4745 static int ext4_expand_extra_isize(struct inode
*inode
,
4746 unsigned int new_extra_isize
,
4747 struct ext4_iloc iloc
,
4750 struct ext4_inode
*raw_inode
;
4751 struct ext4_xattr_ibody_header
*header
;
4753 if (EXT4_I(inode
)->i_extra_isize
>= new_extra_isize
)
4756 raw_inode
= ext4_raw_inode(&iloc
);
4758 header
= IHDR(inode
, raw_inode
);
4760 /* No extended attributes present */
4761 if (!ext4_test_inode_state(inode
, EXT4_STATE_XATTR
) ||
4762 header
->h_magic
!= cpu_to_le32(EXT4_XATTR_MAGIC
)) {
4763 memset((void *)raw_inode
+ EXT4_GOOD_OLD_INODE_SIZE
, 0,
4765 EXT4_I(inode
)->i_extra_isize
= new_extra_isize
;
4769 /* try to expand with EAs present */
4770 return ext4_expand_extra_isize_ea(inode
, new_extra_isize
,
4775 * What we do here is to mark the in-core inode as clean with respect to inode
4776 * dirtiness (it may still be data-dirty).
4777 * This means that the in-core inode may be reaped by prune_icache
4778 * without having to perform any I/O. This is a very good thing,
4779 * because *any* task may call prune_icache - even ones which
4780 * have a transaction open against a different journal.
4782 * Is this cheating? Not really. Sure, we haven't written the
4783 * inode out, but prune_icache isn't a user-visible syncing function.
4784 * Whenever the user wants stuff synced (sys_sync, sys_msync, sys_fsync)
4785 * we start and wait on commits.
4787 int ext4_mark_inode_dirty(handle_t
*handle
, struct inode
*inode
)
4789 struct ext4_iloc iloc
;
4790 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
4791 static unsigned int mnt_count
;
4795 trace_ext4_mark_inode_dirty(inode
, _RET_IP_
);
4796 err
= ext4_reserve_inode_write(handle
, inode
, &iloc
);
4797 if (ext4_handle_valid(handle
) &&
4798 EXT4_I(inode
)->i_extra_isize
< sbi
->s_want_extra_isize
&&
4799 !ext4_test_inode_state(inode
, EXT4_STATE_NO_EXPAND
)) {
4801 * We need extra buffer credits since we may write into EA block
4802 * with this same handle. If journal_extend fails, then it will
4803 * only result in a minor loss of functionality for that inode.
4804 * If this is felt to be critical, then e2fsck should be run to
4805 * force a large enough s_min_extra_isize.
4807 if ((jbd2_journal_extend(handle
,
4808 EXT4_DATA_TRANS_BLOCKS(inode
->i_sb
))) == 0) {
4809 ret
= ext4_expand_extra_isize(inode
,
4810 sbi
->s_want_extra_isize
,
4813 ext4_set_inode_state(inode
,
4814 EXT4_STATE_NO_EXPAND
);
4816 le16_to_cpu(sbi
->s_es
->s_mnt_count
)) {
4817 ext4_warning(inode
->i_sb
,
4818 "Unable to expand inode %lu. Delete"
4819 " some EAs or run e2fsck.",
4822 le16_to_cpu(sbi
->s_es
->s_mnt_count
);
4828 err
= ext4_mark_iloc_dirty(handle
, inode
, &iloc
);
4833 * ext4_dirty_inode() is called from __mark_inode_dirty()
4835 * We're really interested in the case where a file is being extended.
4836 * i_size has been changed by generic_commit_write() and we thus need
4837 * to include the updated inode in the current transaction.
4839 * Also, dquot_alloc_block() will always dirty the inode when blocks
4840 * are allocated to the file.
4842 * If the inode is marked synchronous, we don't honour that here - doing
4843 * so would cause a commit on atime updates, which we don't bother doing.
4844 * We handle synchronous inodes at the highest possible level.
4846 void ext4_dirty_inode(struct inode
*inode
, int flags
)
4850 handle
= ext4_journal_start(inode
, EXT4_HT_INODE
, 2);
4854 ext4_mark_inode_dirty(handle
, inode
);
4856 ext4_journal_stop(handle
);
4863 * Bind an inode's backing buffer_head into this transaction, to prevent
4864 * it from being flushed to disk early. Unlike
4865 * ext4_reserve_inode_write, this leaves behind no bh reference and
4866 * returns no iloc structure, so the caller needs to repeat the iloc
4867 * lookup to mark the inode dirty later.
4869 static int ext4_pin_inode(handle_t
*handle
, struct inode
*inode
)
4871 struct ext4_iloc iloc
;
4875 err
= ext4_get_inode_loc(inode
, &iloc
);
4877 BUFFER_TRACE(iloc
.bh
, "get_write_access");
4878 err
= jbd2_journal_get_write_access(handle
, iloc
.bh
);
4880 err
= ext4_handle_dirty_metadata(handle
,
4886 ext4_std_error(inode
->i_sb
, err
);
4891 int ext4_change_inode_journal_flag(struct inode
*inode
, int val
)
4898 * We have to be very careful here: changing a data block's
4899 * journaling status dynamically is dangerous. If we write a
4900 * data block to the journal, change the status and then delete
4901 * that block, we risk forgetting to revoke the old log record
4902 * from the journal and so a subsequent replay can corrupt data.
4903 * So, first we make sure that the journal is empty and that
4904 * nobody is changing anything.
4907 journal
= EXT4_JOURNAL(inode
);
4910 if (is_journal_aborted(journal
))
4912 /* We have to allocate physical blocks for delalloc blocks
4913 * before flushing journal. otherwise delalloc blocks can not
4914 * be allocated any more. even more truncate on delalloc blocks
4915 * could trigger BUG by flushing delalloc blocks in journal.
4916 * There is no delalloc block in non-journal data mode.
4918 if (val
&& test_opt(inode
->i_sb
, DELALLOC
)) {
4919 err
= ext4_alloc_da_blocks(inode
);
4924 /* Wait for all existing dio workers */
4925 ext4_inode_block_unlocked_dio(inode
);
4926 inode_dio_wait(inode
);
4928 jbd2_journal_lock_updates(journal
);
4931 * OK, there are no updates running now, and all cached data is
4932 * synced to disk. We are now in a completely consistent state
4933 * which doesn't have anything in the journal, and we know that
4934 * no filesystem updates are running, so it is safe to modify
4935 * the inode's in-core data-journaling state flag now.
4939 ext4_set_inode_flag(inode
, EXT4_INODE_JOURNAL_DATA
);
4941 err
= jbd2_journal_flush(journal
);
4943 jbd2_journal_unlock_updates(journal
);
4944 ext4_inode_resume_unlocked_dio(inode
);
4947 ext4_clear_inode_flag(inode
, EXT4_INODE_JOURNAL_DATA
);
4949 ext4_set_aops(inode
);
4951 jbd2_journal_unlock_updates(journal
);
4952 ext4_inode_resume_unlocked_dio(inode
);
4954 /* Finally we can mark the inode as dirty. */
4956 handle
= ext4_journal_start(inode
, EXT4_HT_INODE
, 1);
4958 return PTR_ERR(handle
);
4960 err
= ext4_mark_inode_dirty(handle
, inode
);
4961 ext4_handle_sync(handle
);
4962 ext4_journal_stop(handle
);
4963 ext4_std_error(inode
->i_sb
, err
);
4968 static int ext4_bh_unmapped(handle_t
*handle
, struct buffer_head
*bh
)
4970 return !buffer_mapped(bh
);
4973 int ext4_page_mkwrite(struct vm_area_struct
*vma
, struct vm_fault
*vmf
)
4975 struct page
*page
= vmf
->page
;
4979 struct file
*file
= vma
->vm_file
;
4980 struct inode
*inode
= file_inode(file
);
4981 struct address_space
*mapping
= inode
->i_mapping
;
4983 get_block_t
*get_block
;
4986 sb_start_pagefault(inode
->i_sb
);
4987 file_update_time(vma
->vm_file
);
4988 /* Delalloc case is easy... */
4989 if (test_opt(inode
->i_sb
, DELALLOC
) &&
4990 !ext4_should_journal_data(inode
) &&
4991 !ext4_nonda_switch(inode
->i_sb
)) {
4993 ret
= __block_page_mkwrite(vma
, vmf
,
4994 ext4_da_get_block_prep
);
4995 } while (ret
== -ENOSPC
&&
4996 ext4_should_retry_alloc(inode
->i_sb
, &retries
));
5001 size
= i_size_read(inode
);
5002 /* Page got truncated from under us? */
5003 if (page
->mapping
!= mapping
|| page_offset(page
) > size
) {
5005 ret
= VM_FAULT_NOPAGE
;
5009 if (page
->index
== size
>> PAGE_CACHE_SHIFT
)
5010 len
= size
& ~PAGE_CACHE_MASK
;
5012 len
= PAGE_CACHE_SIZE
;
5014 * Return if we have all the buffers mapped. This avoids the need to do
5015 * journal_start/journal_stop which can block and take a long time
5017 if (page_has_buffers(page
)) {
5018 if (!ext4_walk_page_buffers(NULL
, page_buffers(page
),
5020 ext4_bh_unmapped
)) {
5021 /* Wait so that we don't change page under IO */
5022 wait_for_stable_page(page
);
5023 ret
= VM_FAULT_LOCKED
;
5028 /* OK, we need to fill the hole... */
5029 if (ext4_should_dioread_nolock(inode
))
5030 get_block
= ext4_get_block_write
;
5032 get_block
= ext4_get_block
;
5034 handle
= ext4_journal_start(inode
, EXT4_HT_WRITE_PAGE
,
5035 ext4_writepage_trans_blocks(inode
));
5036 if (IS_ERR(handle
)) {
5037 ret
= VM_FAULT_SIGBUS
;
5040 ret
= __block_page_mkwrite(vma
, vmf
, get_block
);
5041 if (!ret
&& ext4_should_journal_data(inode
)) {
5042 if (ext4_walk_page_buffers(handle
, page_buffers(page
), 0,
5043 PAGE_CACHE_SIZE
, NULL
, do_journal_get_write_access
)) {
5045 ret
= VM_FAULT_SIGBUS
;
5046 ext4_journal_stop(handle
);
5049 ext4_set_inode_state(inode
, EXT4_STATE_JDATA
);
5051 ext4_journal_stop(handle
);
5052 if (ret
== -ENOSPC
&& ext4_should_retry_alloc(inode
->i_sb
, &retries
))
5055 ret
= block_page_mkwrite_return(ret
);
5057 sb_end_pagefault(inode
->i_sb
);