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
21 #include <linux/module.h>
23 #include <linux/time.h>
24 #include <linux/jbd2.h>
25 #include <linux/highuid.h>
26 #include <linux/pagemap.h>
27 #include <linux/quotaops.h>
28 #include <linux/string.h>
29 #include <linux/buffer_head.h>
30 #include <linux/writeback.h>
31 #include <linux/pagevec.h>
32 #include <linux/mpage.h>
33 #include <linux/namei.h>
34 #include <linux/uio.h>
35 #include <linux/bio.h>
36 #include <linux/workqueue.h>
37 #include <linux/kernel.h>
38 #include <linux/printk.h>
39 #include <linux/slab.h>
40 #include <linux/ratelimit.h>
42 #include "ext4_jbd2.h"
45 #include "ext4_extents.h"
48 #include <trace/events/ext4.h>
50 #define MPAGE_DA_EXTENT_TAIL 0x01
52 static inline int ext4_begin_ordered_truncate(struct inode
*inode
,
55 trace_ext4_begin_ordered_truncate(inode
, new_size
);
57 * If jinode is zero, then we never opened the file for
58 * writing, so there's no need to call
59 * jbd2_journal_begin_ordered_truncate() since there's no
60 * outstanding writes we need to flush.
62 if (!EXT4_I(inode
)->jinode
)
64 return jbd2_journal_begin_ordered_truncate(EXT4_JOURNAL(inode
),
65 EXT4_I(inode
)->jinode
,
69 static void ext4_invalidatepage(struct page
*page
, unsigned long offset
);
70 static int noalloc_get_block_write(struct inode
*inode
, sector_t iblock
,
71 struct buffer_head
*bh_result
, int create
);
72 static int ext4_set_bh_endio(struct buffer_head
*bh
, struct inode
*inode
);
73 static void ext4_end_io_buffer_write(struct buffer_head
*bh
, int uptodate
);
74 static int __ext4_journalled_writepage(struct page
*page
, unsigned int len
);
75 static int ext4_bh_delay_or_unwritten(handle_t
*handle
, struct buffer_head
*bh
);
78 * Test whether an inode is a fast symlink.
80 static int ext4_inode_is_fast_symlink(struct inode
*inode
)
82 int ea_blocks
= EXT4_I(inode
)->i_file_acl
?
83 (inode
->i_sb
->s_blocksize
>> 9) : 0;
85 return (S_ISLNK(inode
->i_mode
) && inode
->i_blocks
- ea_blocks
== 0);
89 * Restart the transaction associated with *handle. This does a commit,
90 * so before we call here everything must be consistently dirtied against
93 int ext4_truncate_restart_trans(handle_t
*handle
, struct inode
*inode
,
99 * Drop i_data_sem to avoid deadlock with ext4_map_blocks. At this
100 * moment, get_block can be called only for blocks inside i_size since
101 * page cache has been already dropped and writes are blocked by
102 * i_mutex. So we can safely drop the i_data_sem here.
104 BUG_ON(EXT4_JOURNAL(inode
) == NULL
);
105 jbd_debug(2, "restarting handle %p\n", handle
);
106 up_write(&EXT4_I(inode
)->i_data_sem
);
107 ret
= ext4_journal_restart(handle
, nblocks
);
108 down_write(&EXT4_I(inode
)->i_data_sem
);
109 ext4_discard_preallocations(inode
);
115 * Called at the last iput() if i_nlink is zero.
117 void ext4_evict_inode(struct inode
*inode
)
122 trace_ext4_evict_inode(inode
);
124 mutex_lock(&inode
->i_mutex
);
125 ext4_flush_completed_IO(inode
);
126 mutex_unlock(&inode
->i_mutex
);
127 ext4_ioend_wait(inode
);
129 if (inode
->i_nlink
) {
131 * When journalling data dirty buffers are tracked only in the
132 * journal. So although mm thinks everything is clean and
133 * ready for reaping the inode might still have some pages to
134 * write in the running transaction or waiting to be
135 * checkpointed. Thus calling jbd2_journal_invalidatepage()
136 * (via truncate_inode_pages()) to discard these buffers can
137 * cause data loss. Also even if we did not discard these
138 * buffers, we would have no way to find them after the inode
139 * is reaped and thus user could see stale data if he tries to
140 * read them before the transaction is checkpointed. So be
141 * careful and force everything to disk here... We use
142 * ei->i_datasync_tid to store the newest transaction
143 * containing inode's data.
145 * Note that directories do not have this problem because they
146 * don't use page cache.
148 if (ext4_should_journal_data(inode
) &&
149 (S_ISLNK(inode
->i_mode
) || S_ISREG(inode
->i_mode
))) {
150 journal_t
*journal
= EXT4_SB(inode
->i_sb
)->s_journal
;
151 tid_t commit_tid
= EXT4_I(inode
)->i_datasync_tid
;
153 jbd2_log_start_commit(journal
, commit_tid
);
154 jbd2_log_wait_commit(journal
, commit_tid
);
155 filemap_write_and_wait(&inode
->i_data
);
157 truncate_inode_pages(&inode
->i_data
, 0);
161 if (!is_bad_inode(inode
))
162 dquot_initialize(inode
);
164 if (ext4_should_order_data(inode
))
165 ext4_begin_ordered_truncate(inode
, 0);
166 truncate_inode_pages(&inode
->i_data
, 0);
168 if (is_bad_inode(inode
))
171 handle
= ext4_journal_start(inode
, ext4_blocks_for_truncate(inode
)+3);
172 if (IS_ERR(handle
)) {
173 ext4_std_error(inode
->i_sb
, PTR_ERR(handle
));
175 * If we're going to skip the normal cleanup, we still need to
176 * make sure that the in-core orphan linked list is properly
179 ext4_orphan_del(NULL
, inode
);
184 ext4_handle_sync(handle
);
186 err
= ext4_mark_inode_dirty(handle
, inode
);
188 ext4_warning(inode
->i_sb
,
189 "couldn't mark inode dirty (err %d)", err
);
193 ext4_truncate(inode
);
196 * ext4_ext_truncate() doesn't reserve any slop when it
197 * restarts journal transactions; therefore there may not be
198 * enough credits left in the handle to remove the inode from
199 * the orphan list and set the dtime field.
201 if (!ext4_handle_has_enough_credits(handle
, 3)) {
202 err
= ext4_journal_extend(handle
, 3);
204 err
= ext4_journal_restart(handle
, 3);
206 ext4_warning(inode
->i_sb
,
207 "couldn't extend journal (err %d)", err
);
209 ext4_journal_stop(handle
);
210 ext4_orphan_del(NULL
, inode
);
216 * Kill off the orphan record which ext4_truncate created.
217 * AKPM: I think this can be inside the above `if'.
218 * Note that ext4_orphan_del() has to be able to cope with the
219 * deletion of a non-existent orphan - this is because we don't
220 * know if ext4_truncate() actually created an orphan record.
221 * (Well, we could do this if we need to, but heck - it works)
223 ext4_orphan_del(handle
, inode
);
224 EXT4_I(inode
)->i_dtime
= get_seconds();
227 * One subtle ordering requirement: if anything has gone wrong
228 * (transaction abort, IO errors, whatever), then we can still
229 * do these next steps (the fs will already have been marked as
230 * having errors), but we can't free the inode if the mark_dirty
233 if (ext4_mark_inode_dirty(handle
, inode
))
234 /* If that failed, just do the required in-core inode clear. */
235 ext4_clear_inode(inode
);
237 ext4_free_inode(handle
, inode
);
238 ext4_journal_stop(handle
);
241 ext4_clear_inode(inode
); /* We must guarantee clearing of inode... */
245 qsize_t
*ext4_get_reserved_space(struct inode
*inode
)
247 return &EXT4_I(inode
)->i_reserved_quota
;
252 * Calculate the number of metadata blocks need to reserve
253 * to allocate a block located at @lblock
255 static int ext4_calc_metadata_amount(struct inode
*inode
, ext4_lblk_t lblock
)
257 if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
))
258 return ext4_ext_calc_metadata_amount(inode
, lblock
);
260 return ext4_ind_calc_metadata_amount(inode
, lblock
);
264 * Called with i_data_sem down, which is important since we can call
265 * ext4_discard_preallocations() from here.
267 void ext4_da_update_reserve_space(struct inode
*inode
,
268 int used
, int quota_claim
)
270 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
271 struct ext4_inode_info
*ei
= EXT4_I(inode
);
273 spin_lock(&ei
->i_block_reservation_lock
);
274 trace_ext4_da_update_reserve_space(inode
, used
);
275 if (unlikely(used
> ei
->i_reserved_data_blocks
)) {
276 ext4_msg(inode
->i_sb
, KERN_NOTICE
, "%s: ino %lu, used %d "
277 "with only %d reserved data blocks\n",
278 __func__
, inode
->i_ino
, used
,
279 ei
->i_reserved_data_blocks
);
281 used
= ei
->i_reserved_data_blocks
;
284 /* Update per-inode reservations */
285 ei
->i_reserved_data_blocks
-= used
;
286 ei
->i_reserved_meta_blocks
-= ei
->i_allocated_meta_blocks
;
287 percpu_counter_sub(&sbi
->s_dirtyblocks_counter
,
288 used
+ ei
->i_allocated_meta_blocks
);
289 ei
->i_allocated_meta_blocks
= 0;
291 if (ei
->i_reserved_data_blocks
== 0) {
293 * We can release all of the reserved metadata blocks
294 * only when we have written all of the delayed
297 percpu_counter_sub(&sbi
->s_dirtyblocks_counter
,
298 ei
->i_reserved_meta_blocks
);
299 ei
->i_reserved_meta_blocks
= 0;
300 ei
->i_da_metadata_calc_len
= 0;
302 spin_unlock(&EXT4_I(inode
)->i_block_reservation_lock
);
304 /* Update quota subsystem for data blocks */
306 dquot_claim_block(inode
, used
);
309 * We did fallocate with an offset that is already delayed
310 * allocated. So on delayed allocated writeback we should
311 * not re-claim the quota for fallocated blocks.
313 dquot_release_reservation_block(inode
, used
);
317 * If we have done all the pending block allocations and if
318 * there aren't any writers on the inode, we can discard the
319 * inode's preallocations.
321 if ((ei
->i_reserved_data_blocks
== 0) &&
322 (atomic_read(&inode
->i_writecount
) == 0))
323 ext4_discard_preallocations(inode
);
326 static int __check_block_validity(struct inode
*inode
, const char *func
,
328 struct ext4_map_blocks
*map
)
330 if (!ext4_data_block_valid(EXT4_SB(inode
->i_sb
), map
->m_pblk
,
332 ext4_error_inode(inode
, func
, line
, map
->m_pblk
,
333 "lblock %lu mapped to illegal pblock "
334 "(length %d)", (unsigned long) map
->m_lblk
,
341 #define check_block_validity(inode, map) \
342 __check_block_validity((inode), __func__, __LINE__, (map))
345 * Return the number of contiguous dirty pages in a given inode
346 * starting at page frame idx.
348 static pgoff_t
ext4_num_dirty_pages(struct inode
*inode
, pgoff_t idx
,
349 unsigned int max_pages
)
351 struct address_space
*mapping
= inode
->i_mapping
;
355 int i
, nr_pages
, done
= 0;
359 pagevec_init(&pvec
, 0);
362 nr_pages
= pagevec_lookup_tag(&pvec
, mapping
, &index
,
364 (pgoff_t
)PAGEVEC_SIZE
);
367 for (i
= 0; i
< nr_pages
; i
++) {
368 struct page
*page
= pvec
.pages
[i
];
369 struct buffer_head
*bh
, *head
;
372 if (unlikely(page
->mapping
!= mapping
) ||
374 PageWriteback(page
) ||
375 page
->index
!= idx
) {
380 if (page_has_buffers(page
)) {
381 bh
= head
= page_buffers(page
);
383 if (!buffer_delay(bh
) &&
384 !buffer_unwritten(bh
))
386 bh
= bh
->b_this_page
;
387 } while (!done
&& (bh
!= head
));
394 if (num
>= max_pages
) {
399 pagevec_release(&pvec
);
405 * The ext4_map_blocks() function tries to look up the requested blocks,
406 * and returns if the blocks are already mapped.
408 * Otherwise it takes the write lock of the i_data_sem and allocate blocks
409 * and store the allocated blocks in the result buffer head and mark it
412 * If file type is extents based, it will call ext4_ext_map_blocks(),
413 * Otherwise, call with ext4_ind_map_blocks() to handle indirect mapping
416 * On success, it returns the number of blocks being mapped or allocate.
417 * if create==0 and the blocks are pre-allocated and uninitialized block,
418 * the result buffer head is unmapped. If the create ==1, it will make sure
419 * the buffer head is mapped.
421 * It returns 0 if plain look up failed (blocks have not been allocated), in
422 * that casem, buffer head is unmapped
424 * It returns the error in case of allocation failure.
426 int ext4_map_blocks(handle_t
*handle
, struct inode
*inode
,
427 struct ext4_map_blocks
*map
, int flags
)
432 ext_debug("ext4_map_blocks(): inode %lu, flag %d, max_blocks %u,"
433 "logical block %lu\n", inode
->i_ino
, flags
, map
->m_len
,
434 (unsigned long) map
->m_lblk
);
436 * Try to see if we can get the block without requesting a new
439 down_read((&EXT4_I(inode
)->i_data_sem
));
440 if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
)) {
441 retval
= ext4_ext_map_blocks(handle
, inode
, map
, 0);
443 retval
= ext4_ind_map_blocks(handle
, inode
, map
, 0);
445 up_read((&EXT4_I(inode
)->i_data_sem
));
447 if (retval
> 0 && map
->m_flags
& EXT4_MAP_MAPPED
) {
448 int ret
= check_block_validity(inode
, map
);
453 /* If it is only a block(s) look up */
454 if ((flags
& EXT4_GET_BLOCKS_CREATE
) == 0)
458 * Returns if the blocks have already allocated
460 * Note that if blocks have been preallocated
461 * ext4_ext_get_block() returns th create = 0
462 * with buffer head unmapped.
464 if (retval
> 0 && map
->m_flags
& EXT4_MAP_MAPPED
)
468 * When we call get_blocks without the create flag, the
469 * BH_Unwritten flag could have gotten set if the blocks
470 * requested were part of a uninitialized extent. We need to
471 * clear this flag now that we are committed to convert all or
472 * part of the uninitialized extent to be an initialized
473 * extent. This is because we need to avoid the combination
474 * of BH_Unwritten and BH_Mapped flags being simultaneously
475 * set on the buffer_head.
477 map
->m_flags
&= ~EXT4_MAP_UNWRITTEN
;
480 * New blocks allocate and/or writing to uninitialized extent
481 * will possibly result in updating i_data, so we take
482 * the write lock of i_data_sem, and call get_blocks()
483 * with create == 1 flag.
485 down_write((&EXT4_I(inode
)->i_data_sem
));
488 * if the caller is from delayed allocation writeout path
489 * we have already reserved fs blocks for allocation
490 * let the underlying get_block() function know to
491 * avoid double accounting
493 if (flags
& EXT4_GET_BLOCKS_DELALLOC_RESERVE
)
494 ext4_set_inode_state(inode
, EXT4_STATE_DELALLOC_RESERVED
);
496 * We need to check for EXT4 here because migrate
497 * could have changed the inode type in between
499 if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
)) {
500 retval
= ext4_ext_map_blocks(handle
, inode
, map
, flags
);
502 retval
= ext4_ind_map_blocks(handle
, inode
, map
, flags
);
504 if (retval
> 0 && map
->m_flags
& EXT4_MAP_NEW
) {
506 * We allocated new blocks which will result in
507 * i_data's format changing. Force the migrate
508 * to fail by clearing migrate flags
510 ext4_clear_inode_state(inode
, EXT4_STATE_EXT_MIGRATE
);
514 * Update reserved blocks/metadata blocks after successful
515 * block allocation which had been deferred till now. We don't
516 * support fallocate for non extent files. So we can update
517 * reserve space here.
520 (flags
& EXT4_GET_BLOCKS_DELALLOC_RESERVE
))
521 ext4_da_update_reserve_space(inode
, retval
, 1);
523 if (flags
& EXT4_GET_BLOCKS_DELALLOC_RESERVE
)
524 ext4_clear_inode_state(inode
, EXT4_STATE_DELALLOC_RESERVED
);
526 up_write((&EXT4_I(inode
)->i_data_sem
));
527 if (retval
> 0 && map
->m_flags
& EXT4_MAP_MAPPED
) {
528 int ret
= check_block_validity(inode
, map
);
535 /* Maximum number of blocks we map for direct IO at once. */
536 #define DIO_MAX_BLOCKS 4096
538 static int _ext4_get_block(struct inode
*inode
, sector_t iblock
,
539 struct buffer_head
*bh
, int flags
)
541 handle_t
*handle
= ext4_journal_current_handle();
542 struct ext4_map_blocks map
;
543 int ret
= 0, started
= 0;
547 map
.m_len
= bh
->b_size
>> inode
->i_blkbits
;
549 if (flags
&& !handle
) {
550 /* Direct IO write... */
551 if (map
.m_len
> DIO_MAX_BLOCKS
)
552 map
.m_len
= DIO_MAX_BLOCKS
;
553 dio_credits
= ext4_chunk_trans_blocks(inode
, map
.m_len
);
554 handle
= ext4_journal_start(inode
, dio_credits
);
555 if (IS_ERR(handle
)) {
556 ret
= PTR_ERR(handle
);
562 ret
= ext4_map_blocks(handle
, inode
, &map
, flags
);
564 map_bh(bh
, inode
->i_sb
, map
.m_pblk
);
565 bh
->b_state
= (bh
->b_state
& ~EXT4_MAP_FLAGS
) | map
.m_flags
;
566 bh
->b_size
= inode
->i_sb
->s_blocksize
* map
.m_len
;
570 ext4_journal_stop(handle
);
574 int ext4_get_block(struct inode
*inode
, sector_t iblock
,
575 struct buffer_head
*bh
, int create
)
577 return _ext4_get_block(inode
, iblock
, bh
,
578 create
? EXT4_GET_BLOCKS_CREATE
: 0);
582 * `handle' can be NULL if create is zero
584 struct buffer_head
*ext4_getblk(handle_t
*handle
, struct inode
*inode
,
585 ext4_lblk_t block
, int create
, int *errp
)
587 struct ext4_map_blocks map
;
588 struct buffer_head
*bh
;
591 J_ASSERT(handle
!= NULL
|| create
== 0);
595 err
= ext4_map_blocks(handle
, inode
, &map
,
596 create
? EXT4_GET_BLOCKS_CREATE
: 0);
604 bh
= sb_getblk(inode
->i_sb
, map
.m_pblk
);
609 if (map
.m_flags
& EXT4_MAP_NEW
) {
610 J_ASSERT(create
!= 0);
611 J_ASSERT(handle
!= NULL
);
614 * Now that we do not always journal data, we should
615 * keep in mind whether this should always journal the
616 * new buffer as metadata. For now, regular file
617 * writes use ext4_get_block instead, so it's not a
621 BUFFER_TRACE(bh
, "call get_create_access");
622 fatal
= ext4_journal_get_create_access(handle
, bh
);
623 if (!fatal
&& !buffer_uptodate(bh
)) {
624 memset(bh
->b_data
, 0, inode
->i_sb
->s_blocksize
);
625 set_buffer_uptodate(bh
);
628 BUFFER_TRACE(bh
, "call ext4_handle_dirty_metadata");
629 err
= ext4_handle_dirty_metadata(handle
, inode
, bh
);
633 BUFFER_TRACE(bh
, "not a new buffer");
643 struct buffer_head
*ext4_bread(handle_t
*handle
, struct inode
*inode
,
644 ext4_lblk_t block
, int create
, int *err
)
646 struct buffer_head
*bh
;
648 bh
= ext4_getblk(handle
, inode
, block
, create
, err
);
651 if (buffer_uptodate(bh
))
653 ll_rw_block(READ_META
, 1, &bh
);
655 if (buffer_uptodate(bh
))
662 static int walk_page_buffers(handle_t
*handle
,
663 struct buffer_head
*head
,
667 int (*fn
)(handle_t
*handle
,
668 struct buffer_head
*bh
))
670 struct buffer_head
*bh
;
671 unsigned block_start
, block_end
;
672 unsigned blocksize
= head
->b_size
;
674 struct buffer_head
*next
;
676 for (bh
= head
, block_start
= 0;
677 ret
== 0 && (bh
!= head
|| !block_start
);
678 block_start
= block_end
, bh
= next
) {
679 next
= bh
->b_this_page
;
680 block_end
= block_start
+ blocksize
;
681 if (block_end
<= from
|| block_start
>= to
) {
682 if (partial
&& !buffer_uptodate(bh
))
686 err
= (*fn
)(handle
, bh
);
694 * To preserve ordering, it is essential that the hole instantiation and
695 * the data write be encapsulated in a single transaction. We cannot
696 * close off a transaction and start a new one between the ext4_get_block()
697 * and the commit_write(). So doing the jbd2_journal_start at the start of
698 * prepare_write() is the right place.
700 * Also, this function can nest inside ext4_writepage() ->
701 * block_write_full_page(). In that case, we *know* that ext4_writepage()
702 * has generated enough buffer credits to do the whole page. So we won't
703 * block on the journal in that case, which is good, because the caller may
706 * By accident, ext4 can be reentered when a transaction is open via
707 * quota file writes. If we were to commit the transaction while thus
708 * reentered, there can be a deadlock - we would be holding a quota
709 * lock, and the commit would never complete if another thread had a
710 * transaction open and was blocking on the quota lock - a ranking
713 * So what we do is to rely on the fact that jbd2_journal_stop/journal_start
714 * will _not_ run commit under these circumstances because handle->h_ref
715 * is elevated. We'll still have enough credits for the tiny quotafile
718 static int do_journal_get_write_access(handle_t
*handle
,
719 struct buffer_head
*bh
)
721 int dirty
= buffer_dirty(bh
);
724 if (!buffer_mapped(bh
) || buffer_freed(bh
))
727 * __block_write_begin() could have dirtied some buffers. Clean
728 * the dirty bit as jbd2_journal_get_write_access() could complain
729 * otherwise about fs integrity issues. Setting of the dirty bit
730 * by __block_write_begin() isn't a real problem here as we clear
731 * the bit before releasing a page lock and thus writeback cannot
732 * ever write the buffer.
735 clear_buffer_dirty(bh
);
736 ret
= ext4_journal_get_write_access(handle
, bh
);
738 ret
= ext4_handle_dirty_metadata(handle
, NULL
, bh
);
742 static int ext4_get_block_write(struct inode
*inode
, sector_t iblock
,
743 struct buffer_head
*bh_result
, int create
);
744 static int ext4_write_begin(struct file
*file
, struct address_space
*mapping
,
745 loff_t pos
, unsigned len
, unsigned flags
,
746 struct page
**pagep
, void **fsdata
)
748 struct inode
*inode
= mapping
->host
;
749 int ret
, needed_blocks
;
756 trace_ext4_write_begin(inode
, pos
, len
, flags
);
758 * Reserve one block more for addition to orphan list in case
759 * we allocate blocks but write fails for some reason
761 needed_blocks
= ext4_writepage_trans_blocks(inode
) + 1;
762 index
= pos
>> PAGE_CACHE_SHIFT
;
763 from
= pos
& (PAGE_CACHE_SIZE
- 1);
767 handle
= ext4_journal_start(inode
, needed_blocks
);
768 if (IS_ERR(handle
)) {
769 ret
= PTR_ERR(handle
);
773 /* We cannot recurse into the filesystem as the transaction is already
775 flags
|= AOP_FLAG_NOFS
;
777 page
= grab_cache_page_write_begin(mapping
, index
, flags
);
779 ext4_journal_stop(handle
);
785 if (ext4_should_dioread_nolock(inode
))
786 ret
= __block_write_begin(page
, pos
, len
, ext4_get_block_write
);
788 ret
= __block_write_begin(page
, pos
, len
, ext4_get_block
);
790 if (!ret
&& ext4_should_journal_data(inode
)) {
791 ret
= walk_page_buffers(handle
, page_buffers(page
),
792 from
, to
, NULL
, do_journal_get_write_access
);
797 page_cache_release(page
);
799 * __block_write_begin may have instantiated a few blocks
800 * outside i_size. Trim these off again. Don't need
801 * i_size_read because we hold i_mutex.
803 * Add inode to orphan list in case we crash before
806 if (pos
+ len
> inode
->i_size
&& ext4_can_truncate(inode
))
807 ext4_orphan_add(handle
, inode
);
809 ext4_journal_stop(handle
);
810 if (pos
+ len
> inode
->i_size
) {
811 ext4_truncate_failed_write(inode
);
813 * If truncate failed early the inode might
814 * still be on the orphan list; we need to
815 * make sure the inode is removed from the
816 * orphan list in that case.
819 ext4_orphan_del(NULL
, inode
);
823 if (ret
== -ENOSPC
&& ext4_should_retry_alloc(inode
->i_sb
, &retries
))
829 /* For write_end() in data=journal mode */
830 static int write_end_fn(handle_t
*handle
, struct buffer_head
*bh
)
832 if (!buffer_mapped(bh
) || buffer_freed(bh
))
834 set_buffer_uptodate(bh
);
835 return ext4_handle_dirty_metadata(handle
, NULL
, bh
);
838 static int ext4_generic_write_end(struct file
*file
,
839 struct address_space
*mapping
,
840 loff_t pos
, unsigned len
, unsigned copied
,
841 struct page
*page
, void *fsdata
)
843 int i_size_changed
= 0;
844 struct inode
*inode
= mapping
->host
;
845 handle_t
*handle
= ext4_journal_current_handle();
847 copied
= block_write_end(file
, mapping
, pos
, len
, copied
, page
, fsdata
);
850 * No need to use i_size_read() here, the i_size
851 * cannot change under us because we hold i_mutex.
853 * But it's important to update i_size while still holding page lock:
854 * page writeout could otherwise come in and zero beyond i_size.
856 if (pos
+ copied
> inode
->i_size
) {
857 i_size_write(inode
, pos
+ copied
);
861 if (pos
+ copied
> EXT4_I(inode
)->i_disksize
) {
862 /* We need to mark inode dirty even if
863 * new_i_size is less that inode->i_size
864 * bu greater than i_disksize.(hint delalloc)
866 ext4_update_i_disksize(inode
, (pos
+ copied
));
870 page_cache_release(page
);
873 * Don't mark the inode dirty under page lock. First, it unnecessarily
874 * makes the holding time of page lock longer. Second, it forces lock
875 * ordering of page lock and transaction start for journaling
879 ext4_mark_inode_dirty(handle
, inode
);
885 * We need to pick up the new inode size which generic_commit_write gave us
886 * `file' can be NULL - eg, when called from page_symlink().
888 * ext4 never places buffers on inode->i_mapping->private_list. metadata
889 * buffers are managed internally.
891 static int ext4_ordered_write_end(struct file
*file
,
892 struct address_space
*mapping
,
893 loff_t pos
, unsigned len
, unsigned copied
,
894 struct page
*page
, void *fsdata
)
896 handle_t
*handle
= ext4_journal_current_handle();
897 struct inode
*inode
= mapping
->host
;
900 trace_ext4_ordered_write_end(inode
, pos
, len
, copied
);
901 ret
= ext4_jbd2_file_inode(handle
, inode
);
904 ret2
= ext4_generic_write_end(file
, mapping
, pos
, len
, copied
,
907 if (pos
+ len
> inode
->i_size
&& ext4_can_truncate(inode
))
908 /* if we have allocated more blocks and copied
909 * less. We will have blocks allocated outside
910 * inode->i_size. So truncate them
912 ext4_orphan_add(handle
, inode
);
916 ret2
= ext4_journal_stop(handle
);
920 if (pos
+ len
> inode
->i_size
) {
921 ext4_truncate_failed_write(inode
);
923 * If truncate failed early the inode might still be
924 * on the orphan list; we need to make sure the inode
925 * is removed from the orphan list in that case.
928 ext4_orphan_del(NULL
, inode
);
932 return ret
? ret
: copied
;
935 static int ext4_writeback_write_end(struct file
*file
,
936 struct address_space
*mapping
,
937 loff_t pos
, unsigned len
, unsigned copied
,
938 struct page
*page
, void *fsdata
)
940 handle_t
*handle
= ext4_journal_current_handle();
941 struct inode
*inode
= mapping
->host
;
944 trace_ext4_writeback_write_end(inode
, pos
, len
, copied
);
945 ret2
= ext4_generic_write_end(file
, mapping
, pos
, len
, copied
,
948 if (pos
+ len
> inode
->i_size
&& ext4_can_truncate(inode
))
949 /* if we have allocated more blocks and copied
950 * less. We will have blocks allocated outside
951 * inode->i_size. So truncate them
953 ext4_orphan_add(handle
, inode
);
958 ret2
= 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 still be
966 * on the orphan list; we need to make sure the inode
967 * is removed from the orphan list in that case.
970 ext4_orphan_del(NULL
, inode
);
973 return ret
? ret
: copied
;
976 static int ext4_journalled_write_end(struct file
*file
,
977 struct address_space
*mapping
,
978 loff_t pos
, unsigned len
, unsigned copied
,
979 struct page
*page
, void *fsdata
)
981 handle_t
*handle
= ext4_journal_current_handle();
982 struct inode
*inode
= mapping
->host
;
988 trace_ext4_journalled_write_end(inode
, pos
, len
, copied
);
989 from
= pos
& (PAGE_CACHE_SIZE
- 1);
992 BUG_ON(!ext4_handle_valid(handle
));
995 if (!PageUptodate(page
))
997 page_zero_new_buffers(page
, from
+copied
, to
);
1000 ret
= walk_page_buffers(handle
, page_buffers(page
), from
,
1001 to
, &partial
, write_end_fn
);
1003 SetPageUptodate(page
);
1004 new_i_size
= pos
+ copied
;
1005 if (new_i_size
> inode
->i_size
)
1006 i_size_write(inode
, pos
+copied
);
1007 ext4_set_inode_state(inode
, EXT4_STATE_JDATA
);
1008 EXT4_I(inode
)->i_datasync_tid
= handle
->h_transaction
->t_tid
;
1009 if (new_i_size
> EXT4_I(inode
)->i_disksize
) {
1010 ext4_update_i_disksize(inode
, new_i_size
);
1011 ret2
= ext4_mark_inode_dirty(handle
, inode
);
1017 page_cache_release(page
);
1018 if (pos
+ len
> inode
->i_size
&& ext4_can_truncate(inode
))
1019 /* if we have allocated more blocks and copied
1020 * less. We will have blocks allocated outside
1021 * inode->i_size. So truncate them
1023 ext4_orphan_add(handle
, inode
);
1025 ret2
= ext4_journal_stop(handle
);
1028 if (pos
+ len
> inode
->i_size
) {
1029 ext4_truncate_failed_write(inode
);
1031 * If truncate failed early the inode might still be
1032 * on the orphan list; we need to make sure the inode
1033 * is removed from the orphan list in that case.
1036 ext4_orphan_del(NULL
, inode
);
1039 return ret
? ret
: copied
;
1043 * Reserve a single block located at lblock
1045 static int ext4_da_reserve_space(struct inode
*inode
, ext4_lblk_t lblock
)
1048 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
1049 struct ext4_inode_info
*ei
= EXT4_I(inode
);
1050 unsigned long md_needed
;
1054 * recalculate the amount of metadata blocks to reserve
1055 * in order to allocate nrblocks
1056 * worse case is one extent per block
1059 spin_lock(&ei
->i_block_reservation_lock
);
1060 md_needed
= ext4_calc_metadata_amount(inode
, lblock
);
1061 trace_ext4_da_reserve_space(inode
, md_needed
);
1062 spin_unlock(&ei
->i_block_reservation_lock
);
1065 * We will charge metadata quota at writeout time; this saves
1066 * us from metadata over-estimation, though we may go over by
1067 * a small amount in the end. Here we just reserve for data.
1069 ret
= dquot_reserve_block(inode
, 1);
1073 * We do still charge estimated metadata to the sb though;
1074 * we cannot afford to run out of free blocks.
1076 if (ext4_claim_free_blocks(sbi
, md_needed
+ 1, 0)) {
1077 dquot_release_reservation_block(inode
, 1);
1078 if (ext4_should_retry_alloc(inode
->i_sb
, &retries
)) {
1084 spin_lock(&ei
->i_block_reservation_lock
);
1085 ei
->i_reserved_data_blocks
++;
1086 ei
->i_reserved_meta_blocks
+= md_needed
;
1087 spin_unlock(&ei
->i_block_reservation_lock
);
1089 return 0; /* success */
1092 static void ext4_da_release_space(struct inode
*inode
, int to_free
)
1094 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
1095 struct ext4_inode_info
*ei
= EXT4_I(inode
);
1098 return; /* Nothing to release, exit */
1100 spin_lock(&EXT4_I(inode
)->i_block_reservation_lock
);
1102 trace_ext4_da_release_space(inode
, to_free
);
1103 if (unlikely(to_free
> ei
->i_reserved_data_blocks
)) {
1105 * if there aren't enough reserved blocks, then the
1106 * counter is messed up somewhere. Since this
1107 * function is called from invalidate page, it's
1108 * harmless to return without any action.
1110 ext4_msg(inode
->i_sb
, KERN_NOTICE
, "ext4_da_release_space: "
1111 "ino %lu, to_free %d with only %d reserved "
1112 "data blocks\n", inode
->i_ino
, to_free
,
1113 ei
->i_reserved_data_blocks
);
1115 to_free
= ei
->i_reserved_data_blocks
;
1117 ei
->i_reserved_data_blocks
-= to_free
;
1119 if (ei
->i_reserved_data_blocks
== 0) {
1121 * We can release all of the reserved metadata blocks
1122 * only when we have written all of the delayed
1123 * allocation blocks.
1125 percpu_counter_sub(&sbi
->s_dirtyblocks_counter
,
1126 ei
->i_reserved_meta_blocks
);
1127 ei
->i_reserved_meta_blocks
= 0;
1128 ei
->i_da_metadata_calc_len
= 0;
1131 /* update fs dirty data blocks counter */
1132 percpu_counter_sub(&sbi
->s_dirtyblocks_counter
, to_free
);
1134 spin_unlock(&EXT4_I(inode
)->i_block_reservation_lock
);
1136 dquot_release_reservation_block(inode
, to_free
);
1139 static void ext4_da_page_release_reservation(struct page
*page
,
1140 unsigned long offset
)
1143 struct buffer_head
*head
, *bh
;
1144 unsigned int curr_off
= 0;
1146 head
= page_buffers(page
);
1149 unsigned int next_off
= curr_off
+ bh
->b_size
;
1151 if ((offset
<= curr_off
) && (buffer_delay(bh
))) {
1153 clear_buffer_delay(bh
);
1155 curr_off
= next_off
;
1156 } while ((bh
= bh
->b_this_page
) != head
);
1157 ext4_da_release_space(page
->mapping
->host
, to_release
);
1161 * Delayed allocation stuff
1165 * mpage_da_submit_io - walks through extent of pages and try to write
1166 * them with writepage() call back
1168 * @mpd->inode: inode
1169 * @mpd->first_page: first page of the extent
1170 * @mpd->next_page: page after the last page of the extent
1172 * By the time mpage_da_submit_io() is called we expect all blocks
1173 * to be allocated. this may be wrong if allocation failed.
1175 * As pages are already locked by write_cache_pages(), we can't use it
1177 static int mpage_da_submit_io(struct mpage_da_data
*mpd
,
1178 struct ext4_map_blocks
*map
)
1180 struct pagevec pvec
;
1181 unsigned long index
, end
;
1182 int ret
= 0, err
, nr_pages
, i
;
1183 struct inode
*inode
= mpd
->inode
;
1184 struct address_space
*mapping
= inode
->i_mapping
;
1185 loff_t size
= i_size_read(inode
);
1186 unsigned int len
, block_start
;
1187 struct buffer_head
*bh
, *page_bufs
= NULL
;
1188 int journal_data
= ext4_should_journal_data(inode
);
1189 sector_t pblock
= 0, cur_logical
= 0;
1190 struct ext4_io_submit io_submit
;
1192 BUG_ON(mpd
->next_page
<= mpd
->first_page
);
1193 memset(&io_submit
, 0, sizeof(io_submit
));
1195 * We need to start from the first_page to the next_page - 1
1196 * to make sure we also write the mapped dirty buffer_heads.
1197 * If we look at mpd->b_blocknr we would only be looking
1198 * at the currently mapped buffer_heads.
1200 index
= mpd
->first_page
;
1201 end
= mpd
->next_page
- 1;
1203 pagevec_init(&pvec
, 0);
1204 while (index
<= end
) {
1205 nr_pages
= pagevec_lookup(&pvec
, mapping
, index
, PAGEVEC_SIZE
);
1208 for (i
= 0; i
< nr_pages
; i
++) {
1209 int commit_write
= 0, skip_page
= 0;
1210 struct page
*page
= pvec
.pages
[i
];
1212 index
= page
->index
;
1216 if (index
== size
>> PAGE_CACHE_SHIFT
)
1217 len
= size
& ~PAGE_CACHE_MASK
;
1219 len
= PAGE_CACHE_SIZE
;
1221 cur_logical
= index
<< (PAGE_CACHE_SHIFT
-
1223 pblock
= map
->m_pblk
+ (cur_logical
-
1228 BUG_ON(!PageLocked(page
));
1229 BUG_ON(PageWriteback(page
));
1232 * If the page does not have buffers (for
1233 * whatever reason), try to create them using
1234 * __block_write_begin. If this fails,
1235 * skip the page and move on.
1237 if (!page_has_buffers(page
)) {
1238 if (__block_write_begin(page
, 0, len
,
1239 noalloc_get_block_write
)) {
1247 bh
= page_bufs
= page_buffers(page
);
1252 if (map
&& (cur_logical
>= map
->m_lblk
) &&
1253 (cur_logical
<= (map
->m_lblk
+
1254 (map
->m_len
- 1)))) {
1255 if (buffer_delay(bh
)) {
1256 clear_buffer_delay(bh
);
1257 bh
->b_blocknr
= pblock
;
1259 if (buffer_unwritten(bh
) ||
1261 BUG_ON(bh
->b_blocknr
!= pblock
);
1262 if (map
->m_flags
& EXT4_MAP_UNINIT
)
1263 set_buffer_uninit(bh
);
1264 clear_buffer_unwritten(bh
);
1267 /* skip page if block allocation undone */
1268 if (buffer_delay(bh
) || buffer_unwritten(bh
))
1270 bh
= bh
->b_this_page
;
1271 block_start
+= bh
->b_size
;
1274 } while (bh
!= page_bufs
);
1280 /* mark the buffer_heads as dirty & uptodate */
1281 block_commit_write(page
, 0, len
);
1283 clear_page_dirty_for_io(page
);
1285 * Delalloc doesn't support data journalling,
1286 * but eventually maybe we'll lift this
1289 if (unlikely(journal_data
&& PageChecked(page
)))
1290 err
= __ext4_journalled_writepage(page
, len
);
1291 else if (test_opt(inode
->i_sb
, MBLK_IO_SUBMIT
))
1292 err
= ext4_bio_write_page(&io_submit
, page
,
1295 err
= block_write_full_page(page
,
1296 noalloc_get_block_write
, mpd
->wbc
);
1299 mpd
->pages_written
++;
1301 * In error case, we have to continue because
1302 * remaining pages are still locked
1307 pagevec_release(&pvec
);
1309 ext4_io_submit(&io_submit
);
1313 static void ext4_da_block_invalidatepages(struct mpage_da_data
*mpd
)
1317 struct pagevec pvec
;
1318 struct inode
*inode
= mpd
->inode
;
1319 struct address_space
*mapping
= inode
->i_mapping
;
1321 index
= mpd
->first_page
;
1322 end
= mpd
->next_page
- 1;
1323 while (index
<= end
) {
1324 nr_pages
= pagevec_lookup(&pvec
, mapping
, index
, PAGEVEC_SIZE
);
1327 for (i
= 0; i
< nr_pages
; i
++) {
1328 struct page
*page
= pvec
.pages
[i
];
1329 if (page
->index
> end
)
1331 BUG_ON(!PageLocked(page
));
1332 BUG_ON(PageWriteback(page
));
1333 block_invalidatepage(page
, 0);
1334 ClearPageUptodate(page
);
1337 index
= pvec
.pages
[nr_pages
- 1]->index
+ 1;
1338 pagevec_release(&pvec
);
1343 static void ext4_print_free_blocks(struct inode
*inode
)
1345 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
1346 printk(KERN_CRIT
"Total free blocks count %lld\n",
1347 ext4_count_free_blocks(inode
->i_sb
));
1348 printk(KERN_CRIT
"Free/Dirty block details\n");
1349 printk(KERN_CRIT
"free_blocks=%lld\n",
1350 (long long) percpu_counter_sum(&sbi
->s_freeblocks_counter
));
1351 printk(KERN_CRIT
"dirty_blocks=%lld\n",
1352 (long long) percpu_counter_sum(&sbi
->s_dirtyblocks_counter
));
1353 printk(KERN_CRIT
"Block reservation details\n");
1354 printk(KERN_CRIT
"i_reserved_data_blocks=%u\n",
1355 EXT4_I(inode
)->i_reserved_data_blocks
);
1356 printk(KERN_CRIT
"i_reserved_meta_blocks=%u\n",
1357 EXT4_I(inode
)->i_reserved_meta_blocks
);
1362 * mpage_da_map_and_submit - go through given space, map them
1363 * if necessary, and then submit them for I/O
1365 * @mpd - bh describing space
1367 * The function skips space we know is already mapped to disk blocks.
1370 static void mpage_da_map_and_submit(struct mpage_da_data
*mpd
)
1372 int err
, blks
, get_blocks_flags
;
1373 struct ext4_map_blocks map
, *mapp
= NULL
;
1374 sector_t next
= mpd
->b_blocknr
;
1375 unsigned max_blocks
= mpd
->b_size
>> mpd
->inode
->i_blkbits
;
1376 loff_t disksize
= EXT4_I(mpd
->inode
)->i_disksize
;
1377 handle_t
*handle
= NULL
;
1380 * If the blocks are mapped already, or we couldn't accumulate
1381 * any blocks, then proceed immediately to the submission stage.
1383 if ((mpd
->b_size
== 0) ||
1384 ((mpd
->b_state
& (1 << BH_Mapped
)) &&
1385 !(mpd
->b_state
& (1 << BH_Delay
)) &&
1386 !(mpd
->b_state
& (1 << BH_Unwritten
))))
1389 handle
= ext4_journal_current_handle();
1393 * Call ext4_map_blocks() to allocate any delayed allocation
1394 * blocks, or to convert an uninitialized extent to be
1395 * initialized (in the case where we have written into
1396 * one or more preallocated blocks).
1398 * We pass in the magic EXT4_GET_BLOCKS_DELALLOC_RESERVE to
1399 * indicate that we are on the delayed allocation path. This
1400 * affects functions in many different parts of the allocation
1401 * call path. This flag exists primarily because we don't
1402 * want to change *many* call functions, so ext4_map_blocks()
1403 * will set the EXT4_STATE_DELALLOC_RESERVED flag once the
1404 * inode's allocation semaphore is taken.
1406 * If the blocks in questions were delalloc blocks, set
1407 * EXT4_GET_BLOCKS_DELALLOC_RESERVE so the delalloc accounting
1408 * variables are updated after the blocks have been allocated.
1411 map
.m_len
= max_blocks
;
1412 get_blocks_flags
= EXT4_GET_BLOCKS_CREATE
;
1413 if (ext4_should_dioread_nolock(mpd
->inode
))
1414 get_blocks_flags
|= EXT4_GET_BLOCKS_IO_CREATE_EXT
;
1415 if (mpd
->b_state
& (1 << BH_Delay
))
1416 get_blocks_flags
|= EXT4_GET_BLOCKS_DELALLOC_RESERVE
;
1418 blks
= ext4_map_blocks(handle
, mpd
->inode
, &map
, get_blocks_flags
);
1420 struct super_block
*sb
= mpd
->inode
->i_sb
;
1424 * If get block returns EAGAIN or ENOSPC and there
1425 * appears to be free blocks we will just let
1426 * mpage_da_submit_io() unlock all of the pages.
1431 if (err
== -ENOSPC
&&
1432 ext4_count_free_blocks(sb
)) {
1438 * get block failure will cause us to loop in
1439 * writepages, because a_ops->writepage won't be able
1440 * to make progress. The page will be redirtied by
1441 * writepage and writepages will again try to write
1444 if (!(EXT4_SB(sb
)->s_mount_flags
& EXT4_MF_FS_ABORTED
)) {
1445 ext4_msg(sb
, KERN_CRIT
,
1446 "delayed block allocation failed for inode %lu "
1447 "at logical offset %llu with max blocks %zd "
1448 "with error %d", mpd
->inode
->i_ino
,
1449 (unsigned long long) next
,
1450 mpd
->b_size
>> mpd
->inode
->i_blkbits
, err
);
1451 ext4_msg(sb
, KERN_CRIT
,
1452 "This should not happen!! Data will be lost\n");
1454 ext4_print_free_blocks(mpd
->inode
);
1456 /* invalidate all the pages */
1457 ext4_da_block_invalidatepages(mpd
);
1459 /* Mark this page range as having been completed */
1466 if (map
.m_flags
& EXT4_MAP_NEW
) {
1467 struct block_device
*bdev
= mpd
->inode
->i_sb
->s_bdev
;
1470 for (i
= 0; i
< map
.m_len
; i
++)
1471 unmap_underlying_metadata(bdev
, map
.m_pblk
+ i
);
1474 if (ext4_should_order_data(mpd
->inode
)) {
1475 err
= ext4_jbd2_file_inode(handle
, mpd
->inode
);
1477 /* This only happens if the journal is aborted */
1482 * Update on-disk size along with block allocation.
1484 disksize
= ((loff_t
) next
+ blks
) << mpd
->inode
->i_blkbits
;
1485 if (disksize
> i_size_read(mpd
->inode
))
1486 disksize
= i_size_read(mpd
->inode
);
1487 if (disksize
> EXT4_I(mpd
->inode
)->i_disksize
) {
1488 ext4_update_i_disksize(mpd
->inode
, disksize
);
1489 err
= ext4_mark_inode_dirty(handle
, mpd
->inode
);
1491 ext4_error(mpd
->inode
->i_sb
,
1492 "Failed to mark inode %lu dirty",
1497 mpage_da_submit_io(mpd
, mapp
);
1501 #define BH_FLAGS ((1 << BH_Uptodate) | (1 << BH_Mapped) | \
1502 (1 << BH_Delay) | (1 << BH_Unwritten))
1505 * mpage_add_bh_to_extent - try to add one more block to extent of blocks
1507 * @mpd->lbh - extent of blocks
1508 * @logical - logical number of the block in the file
1509 * @bh - bh of the block (used to access block's state)
1511 * the function is used to collect contig. blocks in same state
1513 static void mpage_add_bh_to_extent(struct mpage_da_data
*mpd
,
1514 sector_t logical
, size_t b_size
,
1515 unsigned long b_state
)
1518 int nrblocks
= mpd
->b_size
>> mpd
->inode
->i_blkbits
;
1521 * XXX Don't go larger than mballoc is willing to allocate
1522 * This is a stopgap solution. We eventually need to fold
1523 * mpage_da_submit_io() into this function and then call
1524 * ext4_map_blocks() multiple times in a loop
1526 if (nrblocks
>= 8*1024*1024/mpd
->inode
->i_sb
->s_blocksize
)
1529 /* check if thereserved journal credits might overflow */
1530 if (!(ext4_test_inode_flag(mpd
->inode
, EXT4_INODE_EXTENTS
))) {
1531 if (nrblocks
>= EXT4_MAX_TRANS_DATA
) {
1533 * With non-extent format we are limited by the journal
1534 * credit available. Total credit needed to insert
1535 * nrblocks contiguous blocks is dependent on the
1536 * nrblocks. So limit nrblocks.
1539 } else if ((nrblocks
+ (b_size
>> mpd
->inode
->i_blkbits
)) >
1540 EXT4_MAX_TRANS_DATA
) {
1542 * Adding the new buffer_head would make it cross the
1543 * allowed limit for which we have journal credit
1544 * reserved. So limit the new bh->b_size
1546 b_size
= (EXT4_MAX_TRANS_DATA
- nrblocks
) <<
1547 mpd
->inode
->i_blkbits
;
1548 /* we will do mpage_da_submit_io in the next loop */
1552 * First block in the extent
1554 if (mpd
->b_size
== 0) {
1555 mpd
->b_blocknr
= logical
;
1556 mpd
->b_size
= b_size
;
1557 mpd
->b_state
= b_state
& BH_FLAGS
;
1561 next
= mpd
->b_blocknr
+ nrblocks
;
1563 * Can we merge the block to our big extent?
1565 if (logical
== next
&& (b_state
& BH_FLAGS
) == mpd
->b_state
) {
1566 mpd
->b_size
+= b_size
;
1572 * We couldn't merge the block to our extent, so we
1573 * need to flush current extent and start new one
1575 mpage_da_map_and_submit(mpd
);
1579 static int ext4_bh_delay_or_unwritten(handle_t
*handle
, struct buffer_head
*bh
)
1581 return (buffer_delay(bh
) || buffer_unwritten(bh
)) && buffer_dirty(bh
);
1585 * This is a special get_blocks_t callback which is used by
1586 * ext4_da_write_begin(). It will either return mapped block or
1587 * reserve space for a single block.
1589 * For delayed buffer_head we have BH_Mapped, BH_New, BH_Delay set.
1590 * We also have b_blocknr = -1 and b_bdev initialized properly
1592 * For unwritten buffer_head we have BH_Mapped, BH_New, BH_Unwritten set.
1593 * We also have b_blocknr = physicalblock mapping unwritten extent and b_bdev
1594 * initialized properly.
1596 static int ext4_da_get_block_prep(struct inode
*inode
, sector_t iblock
,
1597 struct buffer_head
*bh
, int create
)
1599 struct ext4_map_blocks map
;
1601 sector_t invalid_block
= ~((sector_t
) 0xffff);
1603 if (invalid_block
< ext4_blocks_count(EXT4_SB(inode
->i_sb
)->s_es
))
1606 BUG_ON(create
== 0);
1607 BUG_ON(bh
->b_size
!= inode
->i_sb
->s_blocksize
);
1609 map
.m_lblk
= iblock
;
1613 * first, we need to know whether the block is allocated already
1614 * preallocated blocks are unmapped but should treated
1615 * the same as allocated blocks.
1617 ret
= ext4_map_blocks(NULL
, inode
, &map
, 0);
1621 if (buffer_delay(bh
))
1622 return 0; /* Not sure this could or should happen */
1624 * XXX: __block_write_begin() unmaps passed block, is it OK?
1626 ret
= ext4_da_reserve_space(inode
, iblock
);
1628 /* not enough space to reserve */
1631 map_bh(bh
, inode
->i_sb
, invalid_block
);
1633 set_buffer_delay(bh
);
1637 map_bh(bh
, inode
->i_sb
, map
.m_pblk
);
1638 bh
->b_state
= (bh
->b_state
& ~EXT4_MAP_FLAGS
) | map
.m_flags
;
1640 if (buffer_unwritten(bh
)) {
1641 /* A delayed write to unwritten bh should be marked
1642 * new and mapped. Mapped ensures that we don't do
1643 * get_block multiple times when we write to the same
1644 * offset and new ensures that we do proper zero out
1645 * for partial write.
1648 set_buffer_mapped(bh
);
1654 * This function is used as a standard get_block_t calback function
1655 * when there is no desire to allocate any blocks. It is used as a
1656 * callback function for block_write_begin() and block_write_full_page().
1657 * These functions should only try to map a single block at a time.
1659 * Since this function doesn't do block allocations even if the caller
1660 * requests it by passing in create=1, it is critically important that
1661 * any caller checks to make sure that any buffer heads are returned
1662 * by this function are either all already mapped or marked for
1663 * delayed allocation before calling block_write_full_page(). Otherwise,
1664 * b_blocknr could be left unitialized, and the page write functions will
1665 * be taken by surprise.
1667 static int noalloc_get_block_write(struct inode
*inode
, sector_t iblock
,
1668 struct buffer_head
*bh_result
, int create
)
1670 BUG_ON(bh_result
->b_size
!= inode
->i_sb
->s_blocksize
);
1671 return _ext4_get_block(inode
, iblock
, bh_result
, 0);
1674 static int bget_one(handle_t
*handle
, struct buffer_head
*bh
)
1680 static int bput_one(handle_t
*handle
, struct buffer_head
*bh
)
1686 static int __ext4_journalled_writepage(struct page
*page
,
1689 struct address_space
*mapping
= page
->mapping
;
1690 struct inode
*inode
= mapping
->host
;
1691 struct buffer_head
*page_bufs
;
1692 handle_t
*handle
= NULL
;
1696 ClearPageChecked(page
);
1697 page_bufs
= page_buffers(page
);
1699 walk_page_buffers(handle
, page_bufs
, 0, len
, NULL
, bget_one
);
1700 /* As soon as we unlock the page, it can go away, but we have
1701 * references to buffers so we are safe */
1704 handle
= ext4_journal_start(inode
, ext4_writepage_trans_blocks(inode
));
1705 if (IS_ERR(handle
)) {
1706 ret
= PTR_ERR(handle
);
1710 BUG_ON(!ext4_handle_valid(handle
));
1712 ret
= walk_page_buffers(handle
, page_bufs
, 0, len
, NULL
,
1713 do_journal_get_write_access
);
1715 err
= walk_page_buffers(handle
, page_bufs
, 0, len
, NULL
,
1719 EXT4_I(inode
)->i_datasync_tid
= handle
->h_transaction
->t_tid
;
1720 err
= ext4_journal_stop(handle
);
1724 walk_page_buffers(handle
, page_bufs
, 0, len
, NULL
, bput_one
);
1725 ext4_set_inode_state(inode
, EXT4_STATE_JDATA
);
1730 static int ext4_set_bh_endio(struct buffer_head
*bh
, struct inode
*inode
);
1731 static void ext4_end_io_buffer_write(struct buffer_head
*bh
, int uptodate
);
1734 * Note that we don't need to start a transaction unless we're journaling data
1735 * because we should have holes filled from ext4_page_mkwrite(). We even don't
1736 * need to file the inode to the transaction's list in ordered mode because if
1737 * we are writing back data added by write(), the inode is already there and if
1738 * we are writing back data modified via mmap(), no one guarantees in which
1739 * transaction the data will hit the disk. In case we are journaling data, we
1740 * cannot start transaction directly because transaction start ranks above page
1741 * lock so we have to do some magic.
1743 * This function can get called via...
1744 * - ext4_da_writepages after taking page lock (have journal handle)
1745 * - journal_submit_inode_data_buffers (no journal handle)
1746 * - shrink_page_list via pdflush (no journal handle)
1747 * - grab_page_cache when doing write_begin (have journal handle)
1749 * We don't do any block allocation in this function. If we have page with
1750 * multiple blocks we need to write those buffer_heads that are mapped. This
1751 * is important for mmaped based write. So if we do with blocksize 1K
1752 * truncate(f, 1024);
1753 * a = mmap(f, 0, 4096);
1755 * truncate(f, 4096);
1756 * we have in the page first buffer_head mapped via page_mkwrite call back
1757 * but other bufer_heads would be unmapped but dirty(dirty done via the
1758 * do_wp_page). So writepage should write the first block. If we modify
1759 * the mmap area beyond 1024 we will again get a page_fault and the
1760 * page_mkwrite callback will do the block allocation and mark the
1761 * buffer_heads mapped.
1763 * We redirty the page if we have any buffer_heads that is either delay or
1764 * unwritten in the page.
1766 * We can get recursively called as show below.
1768 * ext4_writepage() -> kmalloc() -> __alloc_pages() -> page_launder() ->
1771 * But since we don't do any block allocation we should not deadlock.
1772 * Page also have the dirty flag cleared so we don't get recurive page_lock.
1774 static int ext4_writepage(struct page
*page
,
1775 struct writeback_control
*wbc
)
1777 int ret
= 0, commit_write
= 0;
1780 struct buffer_head
*page_bufs
= NULL
;
1781 struct inode
*inode
= page
->mapping
->host
;
1783 trace_ext4_writepage(page
);
1784 size
= i_size_read(inode
);
1785 if (page
->index
== size
>> PAGE_CACHE_SHIFT
)
1786 len
= size
& ~PAGE_CACHE_MASK
;
1788 len
= PAGE_CACHE_SIZE
;
1791 * If the page does not have buffers (for whatever reason),
1792 * try to create them using __block_write_begin. If this
1793 * fails, redirty the page and move on.
1795 if (!page_has_buffers(page
)) {
1796 if (__block_write_begin(page
, 0, len
,
1797 noalloc_get_block_write
)) {
1799 redirty_page_for_writepage(wbc
, page
);
1805 page_bufs
= page_buffers(page
);
1806 if (walk_page_buffers(NULL
, page_bufs
, 0, len
, NULL
,
1807 ext4_bh_delay_or_unwritten
)) {
1809 * We don't want to do block allocation, so redirty
1810 * the page and return. We may reach here when we do
1811 * a journal commit via journal_submit_inode_data_buffers.
1812 * We can also reach here via shrink_page_list
1817 /* now mark the buffer_heads as dirty and uptodate */
1818 block_commit_write(page
, 0, len
);
1820 if (PageChecked(page
) && ext4_should_journal_data(inode
))
1822 * It's mmapped pagecache. Add buffers and journal it. There
1823 * doesn't seem much point in redirtying the page here.
1825 return __ext4_journalled_writepage(page
, len
);
1827 if (buffer_uninit(page_bufs
)) {
1828 ext4_set_bh_endio(page_bufs
, inode
);
1829 ret
= block_write_full_page_endio(page
, noalloc_get_block_write
,
1830 wbc
, ext4_end_io_buffer_write
);
1832 ret
= block_write_full_page(page
, noalloc_get_block_write
,
1839 * This is called via ext4_da_writepages() to
1840 * calculate the total number of credits to reserve to fit
1841 * a single extent allocation into a single transaction,
1842 * ext4_da_writpeages() will loop calling this before
1843 * the block allocation.
1846 static int ext4_da_writepages_trans_blocks(struct inode
*inode
)
1848 int max_blocks
= EXT4_I(inode
)->i_reserved_data_blocks
;
1851 * With non-extent format the journal credit needed to
1852 * insert nrblocks contiguous block is dependent on
1853 * number of contiguous block. So we will limit
1854 * number of contiguous block to a sane value
1856 if (!(ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
)) &&
1857 (max_blocks
> EXT4_MAX_TRANS_DATA
))
1858 max_blocks
= EXT4_MAX_TRANS_DATA
;
1860 return ext4_chunk_trans_blocks(inode
, max_blocks
);
1864 * write_cache_pages_da - walk the list of dirty pages of the given
1865 * address space and accumulate pages that need writing, and call
1866 * mpage_da_map_and_submit to map a single contiguous memory region
1867 * and then write them.
1869 static int write_cache_pages_da(struct address_space
*mapping
,
1870 struct writeback_control
*wbc
,
1871 struct mpage_da_data
*mpd
,
1872 pgoff_t
*done_index
)
1874 struct buffer_head
*bh
, *head
;
1875 struct inode
*inode
= mapping
->host
;
1876 struct pagevec pvec
;
1877 unsigned int nr_pages
;
1880 long nr_to_write
= wbc
->nr_to_write
;
1881 int i
, tag
, ret
= 0;
1883 memset(mpd
, 0, sizeof(struct mpage_da_data
));
1886 pagevec_init(&pvec
, 0);
1887 index
= wbc
->range_start
>> PAGE_CACHE_SHIFT
;
1888 end
= wbc
->range_end
>> PAGE_CACHE_SHIFT
;
1890 if (wbc
->sync_mode
== WB_SYNC_ALL
|| wbc
->tagged_writepages
)
1891 tag
= PAGECACHE_TAG_TOWRITE
;
1893 tag
= PAGECACHE_TAG_DIRTY
;
1895 *done_index
= index
;
1896 while (index
<= end
) {
1897 nr_pages
= pagevec_lookup_tag(&pvec
, mapping
, &index
, tag
,
1898 min(end
- index
, (pgoff_t
)PAGEVEC_SIZE
-1) + 1);
1902 for (i
= 0; i
< nr_pages
; i
++) {
1903 struct page
*page
= pvec
.pages
[i
];
1906 * At this point, the page may be truncated or
1907 * invalidated (changing page->mapping to NULL), or
1908 * even swizzled back from swapper_space to tmpfs file
1909 * mapping. However, page->index will not change
1910 * because we have a reference on the page.
1912 if (page
->index
> end
)
1915 *done_index
= page
->index
+ 1;
1918 * If we can't merge this page, and we have
1919 * accumulated an contiguous region, write it
1921 if ((mpd
->next_page
!= page
->index
) &&
1922 (mpd
->next_page
!= mpd
->first_page
)) {
1923 mpage_da_map_and_submit(mpd
);
1924 goto ret_extent_tail
;
1930 * If the page is no longer dirty, or its
1931 * mapping no longer corresponds to inode we
1932 * are writing (which means it has been
1933 * truncated or invalidated), or the page is
1934 * already under writeback and we are not
1935 * doing a data integrity writeback, skip the page
1937 if (!PageDirty(page
) ||
1938 (PageWriteback(page
) &&
1939 (wbc
->sync_mode
== WB_SYNC_NONE
)) ||
1940 unlikely(page
->mapping
!= mapping
)) {
1945 wait_on_page_writeback(page
);
1946 BUG_ON(PageWriteback(page
));
1948 if (mpd
->next_page
!= page
->index
)
1949 mpd
->first_page
= page
->index
;
1950 mpd
->next_page
= page
->index
+ 1;
1951 logical
= (sector_t
) page
->index
<<
1952 (PAGE_CACHE_SHIFT
- inode
->i_blkbits
);
1954 if (!page_has_buffers(page
)) {
1955 mpage_add_bh_to_extent(mpd
, logical
,
1957 (1 << BH_Dirty
) | (1 << BH_Uptodate
));
1959 goto ret_extent_tail
;
1962 * Page with regular buffer heads,
1963 * just add all dirty ones
1965 head
= page_buffers(page
);
1968 BUG_ON(buffer_locked(bh
));
1970 * We need to try to allocate
1971 * unmapped blocks in the same page.
1972 * Otherwise we won't make progress
1973 * with the page in ext4_writepage
1975 if (ext4_bh_delay_or_unwritten(NULL
, bh
)) {
1976 mpage_add_bh_to_extent(mpd
, logical
,
1980 goto ret_extent_tail
;
1981 } else if (buffer_dirty(bh
) && (buffer_mapped(bh
))) {
1983 * mapped dirty buffer. We need
1984 * to update the b_state
1985 * because we look at b_state
1986 * in mpage_da_map_blocks. We
1987 * don't update b_size because
1988 * if we find an unmapped
1989 * buffer_head later we need to
1990 * use the b_state flag of that
1993 if (mpd
->b_size
== 0)
1994 mpd
->b_state
= bh
->b_state
& BH_FLAGS
;
1997 } while ((bh
= bh
->b_this_page
) != head
);
2000 if (nr_to_write
> 0) {
2002 if (nr_to_write
== 0 &&
2003 wbc
->sync_mode
== WB_SYNC_NONE
)
2005 * We stop writing back only if we are
2006 * not doing integrity sync. In case of
2007 * integrity sync we have to keep going
2008 * because someone may be concurrently
2009 * dirtying pages, and we might have
2010 * synced a lot of newly appeared dirty
2011 * pages, but have not synced all of the
2017 pagevec_release(&pvec
);
2022 ret
= MPAGE_DA_EXTENT_TAIL
;
2024 pagevec_release(&pvec
);
2030 static int ext4_da_writepages(struct address_space
*mapping
,
2031 struct writeback_control
*wbc
)
2034 int range_whole
= 0;
2035 handle_t
*handle
= NULL
;
2036 struct mpage_da_data mpd
;
2037 struct inode
*inode
= mapping
->host
;
2038 int pages_written
= 0;
2039 unsigned int max_pages
;
2040 int range_cyclic
, cycled
= 1, io_done
= 0;
2041 int needed_blocks
, ret
= 0;
2042 long desired_nr_to_write
, nr_to_writebump
= 0;
2043 loff_t range_start
= wbc
->range_start
;
2044 struct ext4_sb_info
*sbi
= EXT4_SB(mapping
->host
->i_sb
);
2045 pgoff_t done_index
= 0;
2048 trace_ext4_da_writepages(inode
, wbc
);
2051 * No pages to write? This is mainly a kludge to avoid starting
2052 * a transaction for special inodes like journal inode on last iput()
2053 * because that could violate lock ordering on umount
2055 if (!mapping
->nrpages
|| !mapping_tagged(mapping
, PAGECACHE_TAG_DIRTY
))
2059 * If the filesystem has aborted, it is read-only, so return
2060 * right away instead of dumping stack traces later on that
2061 * will obscure the real source of the problem. We test
2062 * EXT4_MF_FS_ABORTED instead of sb->s_flag's MS_RDONLY because
2063 * the latter could be true if the filesystem is mounted
2064 * read-only, and in that case, ext4_da_writepages should
2065 * *never* be called, so if that ever happens, we would want
2068 if (unlikely(sbi
->s_mount_flags
& EXT4_MF_FS_ABORTED
))
2071 if (wbc
->range_start
== 0 && wbc
->range_end
== LLONG_MAX
)
2074 range_cyclic
= wbc
->range_cyclic
;
2075 if (wbc
->range_cyclic
) {
2076 index
= mapping
->writeback_index
;
2079 wbc
->range_start
= index
<< PAGE_CACHE_SHIFT
;
2080 wbc
->range_end
= LLONG_MAX
;
2081 wbc
->range_cyclic
= 0;
2084 index
= wbc
->range_start
>> PAGE_CACHE_SHIFT
;
2085 end
= wbc
->range_end
>> PAGE_CACHE_SHIFT
;
2089 * This works around two forms of stupidity. The first is in
2090 * the writeback code, which caps the maximum number of pages
2091 * written to be 1024 pages. This is wrong on multiple
2092 * levels; different architectues have a different page size,
2093 * which changes the maximum amount of data which gets
2094 * written. Secondly, 4 megabytes is way too small. XFS
2095 * forces this value to be 16 megabytes by multiplying
2096 * nr_to_write parameter by four, and then relies on its
2097 * allocator to allocate larger extents to make them
2098 * contiguous. Unfortunately this brings us to the second
2099 * stupidity, which is that ext4's mballoc code only allocates
2100 * at most 2048 blocks. So we force contiguous writes up to
2101 * the number of dirty blocks in the inode, or
2102 * sbi->max_writeback_mb_bump whichever is smaller.
2104 max_pages
= sbi
->s_max_writeback_mb_bump
<< (20 - PAGE_CACHE_SHIFT
);
2105 if (!range_cyclic
&& range_whole
) {
2106 if (wbc
->nr_to_write
== LONG_MAX
)
2107 desired_nr_to_write
= wbc
->nr_to_write
;
2109 desired_nr_to_write
= wbc
->nr_to_write
* 8;
2111 desired_nr_to_write
= ext4_num_dirty_pages(inode
, index
,
2113 if (desired_nr_to_write
> max_pages
)
2114 desired_nr_to_write
= max_pages
;
2116 if (wbc
->nr_to_write
< desired_nr_to_write
) {
2117 nr_to_writebump
= desired_nr_to_write
- wbc
->nr_to_write
;
2118 wbc
->nr_to_write
= desired_nr_to_write
;
2122 if (wbc
->sync_mode
== WB_SYNC_ALL
|| wbc
->tagged_writepages
)
2123 tag_pages_for_writeback(mapping
, index
, end
);
2125 while (!ret
&& wbc
->nr_to_write
> 0) {
2128 * we insert one extent at a time. So we need
2129 * credit needed for single extent allocation.
2130 * journalled mode is currently not supported
2133 BUG_ON(ext4_should_journal_data(inode
));
2134 needed_blocks
= ext4_da_writepages_trans_blocks(inode
);
2136 /* start a new transaction*/
2137 handle
= ext4_journal_start(inode
, needed_blocks
);
2138 if (IS_ERR(handle
)) {
2139 ret
= PTR_ERR(handle
);
2140 ext4_msg(inode
->i_sb
, KERN_CRIT
, "%s: jbd2_start: "
2141 "%ld pages, ino %lu; err %d", __func__
,
2142 wbc
->nr_to_write
, inode
->i_ino
, ret
);
2143 goto out_writepages
;
2147 * Now call write_cache_pages_da() to find the next
2148 * contiguous region of logical blocks that need
2149 * blocks to be allocated by ext4 and submit them.
2151 ret
= write_cache_pages_da(mapping
, wbc
, &mpd
, &done_index
);
2153 * If we have a contiguous extent of pages and we
2154 * haven't done the I/O yet, map the blocks and submit
2157 if (!mpd
.io_done
&& mpd
.next_page
!= mpd
.first_page
) {
2158 mpage_da_map_and_submit(&mpd
);
2159 ret
= MPAGE_DA_EXTENT_TAIL
;
2161 trace_ext4_da_write_pages(inode
, &mpd
);
2162 wbc
->nr_to_write
-= mpd
.pages_written
;
2164 ext4_journal_stop(handle
);
2166 if ((mpd
.retval
== -ENOSPC
) && sbi
->s_journal
) {
2167 /* commit the transaction which would
2168 * free blocks released in the transaction
2171 jbd2_journal_force_commit_nested(sbi
->s_journal
);
2173 } else if (ret
== MPAGE_DA_EXTENT_TAIL
) {
2175 * got one extent now try with
2178 pages_written
+= mpd
.pages_written
;
2181 } else if (wbc
->nr_to_write
)
2183 * There is no more writeout needed
2184 * or we requested for a noblocking writeout
2185 * and we found the device congested
2189 if (!io_done
&& !cycled
) {
2192 wbc
->range_start
= index
<< PAGE_CACHE_SHIFT
;
2193 wbc
->range_end
= mapping
->writeback_index
- 1;
2198 wbc
->range_cyclic
= range_cyclic
;
2199 if (wbc
->range_cyclic
|| (range_whole
&& wbc
->nr_to_write
> 0))
2201 * set the writeback_index so that range_cyclic
2202 * mode will write it back later
2204 mapping
->writeback_index
= done_index
;
2207 wbc
->nr_to_write
-= nr_to_writebump
;
2208 wbc
->range_start
= range_start
;
2209 trace_ext4_da_writepages_result(inode
, wbc
, ret
, pages_written
);
2213 #define FALL_BACK_TO_NONDELALLOC 1
2214 static int ext4_nonda_switch(struct super_block
*sb
)
2216 s64 free_blocks
, dirty_blocks
;
2217 struct ext4_sb_info
*sbi
= EXT4_SB(sb
);
2220 * switch to non delalloc mode if we are running low
2221 * on free block. The free block accounting via percpu
2222 * counters can get slightly wrong with percpu_counter_batch getting
2223 * accumulated on each CPU without updating global counters
2224 * Delalloc need an accurate free block accounting. So switch
2225 * to non delalloc when we are near to error range.
2227 free_blocks
= percpu_counter_read_positive(&sbi
->s_freeblocks_counter
);
2228 dirty_blocks
= percpu_counter_read_positive(&sbi
->s_dirtyblocks_counter
);
2229 if (2 * free_blocks
< 3 * dirty_blocks
||
2230 free_blocks
< (dirty_blocks
+ EXT4_FREEBLOCKS_WATERMARK
)) {
2232 * free block count is less than 150% of dirty blocks
2233 * or free blocks is less than watermark
2238 * Even if we don't switch but are nearing capacity,
2239 * start pushing delalloc when 1/2 of free blocks are dirty.
2241 if (free_blocks
< 2 * dirty_blocks
)
2242 writeback_inodes_sb_if_idle(sb
);
2247 static int ext4_da_write_begin(struct file
*file
, struct address_space
*mapping
,
2248 loff_t pos
, unsigned len
, unsigned flags
,
2249 struct page
**pagep
, void **fsdata
)
2251 int ret
, retries
= 0;
2254 struct inode
*inode
= mapping
->host
;
2257 index
= pos
>> PAGE_CACHE_SHIFT
;
2259 if (ext4_nonda_switch(inode
->i_sb
)) {
2260 *fsdata
= (void *)FALL_BACK_TO_NONDELALLOC
;
2261 return ext4_write_begin(file
, mapping
, pos
,
2262 len
, flags
, pagep
, fsdata
);
2264 *fsdata
= (void *)0;
2265 trace_ext4_da_write_begin(inode
, pos
, len
, flags
);
2268 * With delayed allocation, we don't log the i_disksize update
2269 * if there is delayed block allocation. But we still need
2270 * to journalling the i_disksize update if writes to the end
2271 * of file which has an already mapped buffer.
2273 handle
= ext4_journal_start(inode
, 1);
2274 if (IS_ERR(handle
)) {
2275 ret
= PTR_ERR(handle
);
2278 /* We cannot recurse into the filesystem as the transaction is already
2280 flags
|= AOP_FLAG_NOFS
;
2282 page
= grab_cache_page_write_begin(mapping
, index
, flags
);
2284 ext4_journal_stop(handle
);
2290 ret
= __block_write_begin(page
, pos
, len
, ext4_da_get_block_prep
);
2293 ext4_journal_stop(handle
);
2294 page_cache_release(page
);
2296 * block_write_begin may have instantiated a few blocks
2297 * outside i_size. Trim these off again. Don't need
2298 * i_size_read because we hold i_mutex.
2300 if (pos
+ len
> inode
->i_size
)
2301 ext4_truncate_failed_write(inode
);
2304 if (ret
== -ENOSPC
&& ext4_should_retry_alloc(inode
->i_sb
, &retries
))
2311 * Check if we should update i_disksize
2312 * when write to the end of file but not require block allocation
2314 static int ext4_da_should_update_i_disksize(struct page
*page
,
2315 unsigned long offset
)
2317 struct buffer_head
*bh
;
2318 struct inode
*inode
= page
->mapping
->host
;
2322 bh
= page_buffers(page
);
2323 idx
= offset
>> inode
->i_blkbits
;
2325 for (i
= 0; i
< idx
; i
++)
2326 bh
= bh
->b_this_page
;
2328 if (!buffer_mapped(bh
) || (buffer_delay(bh
)) || buffer_unwritten(bh
))
2333 static int ext4_da_write_end(struct file
*file
,
2334 struct address_space
*mapping
,
2335 loff_t pos
, unsigned len
, unsigned copied
,
2336 struct page
*page
, void *fsdata
)
2338 struct inode
*inode
= mapping
->host
;
2340 handle_t
*handle
= ext4_journal_current_handle();
2342 unsigned long start
, end
;
2343 int write_mode
= (int)(unsigned long)fsdata
;
2345 if (write_mode
== FALL_BACK_TO_NONDELALLOC
) {
2346 if (ext4_should_order_data(inode
)) {
2347 return ext4_ordered_write_end(file
, mapping
, pos
,
2348 len
, copied
, page
, fsdata
);
2349 } else if (ext4_should_writeback_data(inode
)) {
2350 return ext4_writeback_write_end(file
, mapping
, pos
,
2351 len
, copied
, page
, fsdata
);
2357 trace_ext4_da_write_end(inode
, pos
, len
, copied
);
2358 start
= pos
& (PAGE_CACHE_SIZE
- 1);
2359 end
= start
+ copied
- 1;
2362 * generic_write_end() will run mark_inode_dirty() if i_size
2363 * changes. So let's piggyback the i_disksize mark_inode_dirty
2367 new_i_size
= pos
+ copied
;
2368 if (new_i_size
> EXT4_I(inode
)->i_disksize
) {
2369 if (ext4_da_should_update_i_disksize(page
, end
)) {
2370 down_write(&EXT4_I(inode
)->i_data_sem
);
2371 if (new_i_size
> EXT4_I(inode
)->i_disksize
) {
2373 * Updating i_disksize when extending file
2374 * without needing block allocation
2376 if (ext4_should_order_data(inode
))
2377 ret
= ext4_jbd2_file_inode(handle
,
2380 EXT4_I(inode
)->i_disksize
= new_i_size
;
2382 up_write(&EXT4_I(inode
)->i_data_sem
);
2383 /* We need to mark inode dirty even if
2384 * new_i_size is less that inode->i_size
2385 * bu greater than i_disksize.(hint delalloc)
2387 ext4_mark_inode_dirty(handle
, inode
);
2390 ret2
= generic_write_end(file
, mapping
, pos
, len
, copied
,
2395 ret2
= ext4_journal_stop(handle
);
2399 return ret
? ret
: copied
;
2402 static void ext4_da_invalidatepage(struct page
*page
, unsigned long offset
)
2405 * Drop reserved blocks
2407 BUG_ON(!PageLocked(page
));
2408 if (!page_has_buffers(page
))
2411 ext4_da_page_release_reservation(page
, offset
);
2414 ext4_invalidatepage(page
, offset
);
2420 * Force all delayed allocation blocks to be allocated for a given inode.
2422 int ext4_alloc_da_blocks(struct inode
*inode
)
2424 trace_ext4_alloc_da_blocks(inode
);
2426 if (!EXT4_I(inode
)->i_reserved_data_blocks
&&
2427 !EXT4_I(inode
)->i_reserved_meta_blocks
)
2431 * We do something simple for now. The filemap_flush() will
2432 * also start triggering a write of the data blocks, which is
2433 * not strictly speaking necessary (and for users of
2434 * laptop_mode, not even desirable). However, to do otherwise
2435 * would require replicating code paths in:
2437 * ext4_da_writepages() ->
2438 * write_cache_pages() ---> (via passed in callback function)
2439 * __mpage_da_writepage() -->
2440 * mpage_add_bh_to_extent()
2441 * mpage_da_map_blocks()
2443 * The problem is that write_cache_pages(), located in
2444 * mm/page-writeback.c, marks pages clean in preparation for
2445 * doing I/O, which is not desirable if we're not planning on
2448 * We could call write_cache_pages(), and then redirty all of
2449 * the pages by calling redirty_page_for_writepage() but that
2450 * would be ugly in the extreme. So instead we would need to
2451 * replicate parts of the code in the above functions,
2452 * simplifying them because we wouldn't actually intend to
2453 * write out the pages, but rather only collect contiguous
2454 * logical block extents, call the multi-block allocator, and
2455 * then update the buffer heads with the block allocations.
2457 * For now, though, we'll cheat by calling filemap_flush(),
2458 * which will map the blocks, and start the I/O, but not
2459 * actually wait for the I/O to complete.
2461 return filemap_flush(inode
->i_mapping
);
2465 * bmap() is special. It gets used by applications such as lilo and by
2466 * the swapper to find the on-disk block of a specific piece of data.
2468 * Naturally, this is dangerous if the block concerned is still in the
2469 * journal. If somebody makes a swapfile on an ext4 data-journaling
2470 * filesystem and enables swap, then they may get a nasty shock when the
2471 * data getting swapped to that swapfile suddenly gets overwritten by
2472 * the original zero's written out previously to the journal and
2473 * awaiting writeback in the kernel's buffer cache.
2475 * So, if we see any bmap calls here on a modified, data-journaled file,
2476 * take extra steps to flush any blocks which might be in the cache.
2478 static sector_t
ext4_bmap(struct address_space
*mapping
, sector_t block
)
2480 struct inode
*inode
= mapping
->host
;
2484 if (mapping_tagged(mapping
, PAGECACHE_TAG_DIRTY
) &&
2485 test_opt(inode
->i_sb
, DELALLOC
)) {
2487 * With delalloc we want to sync the file
2488 * so that we can make sure we allocate
2491 filemap_write_and_wait(mapping
);
2494 if (EXT4_JOURNAL(inode
) &&
2495 ext4_test_inode_state(inode
, EXT4_STATE_JDATA
)) {
2497 * This is a REALLY heavyweight approach, but the use of
2498 * bmap on dirty files is expected to be extremely rare:
2499 * only if we run lilo or swapon on a freshly made file
2500 * do we expect this to happen.
2502 * (bmap requires CAP_SYS_RAWIO so this does not
2503 * represent an unprivileged user DOS attack --- we'd be
2504 * in trouble if mortal users could trigger this path at
2507 * NB. EXT4_STATE_JDATA is not set on files other than
2508 * regular files. If somebody wants to bmap a directory
2509 * or symlink and gets confused because the buffer
2510 * hasn't yet been flushed to disk, they deserve
2511 * everything they get.
2514 ext4_clear_inode_state(inode
, EXT4_STATE_JDATA
);
2515 journal
= EXT4_JOURNAL(inode
);
2516 jbd2_journal_lock_updates(journal
);
2517 err
= jbd2_journal_flush(journal
);
2518 jbd2_journal_unlock_updates(journal
);
2524 return generic_block_bmap(mapping
, block
, ext4_get_block
);
2527 static int ext4_readpage(struct file
*file
, struct page
*page
)
2529 trace_ext4_readpage(page
);
2530 return mpage_readpage(page
, ext4_get_block
);
2534 ext4_readpages(struct file
*file
, struct address_space
*mapping
,
2535 struct list_head
*pages
, unsigned nr_pages
)
2537 return mpage_readpages(mapping
, pages
, nr_pages
, ext4_get_block
);
2540 static void ext4_invalidatepage_free_endio(struct page
*page
, unsigned long offset
)
2542 struct buffer_head
*head
, *bh
;
2543 unsigned int curr_off
= 0;
2545 if (!page_has_buffers(page
))
2547 head
= bh
= page_buffers(page
);
2549 if (offset
<= curr_off
&& test_clear_buffer_uninit(bh
)
2551 ext4_free_io_end(bh
->b_private
);
2552 bh
->b_private
= NULL
;
2553 bh
->b_end_io
= NULL
;
2555 curr_off
= curr_off
+ bh
->b_size
;
2556 bh
= bh
->b_this_page
;
2557 } while (bh
!= head
);
2560 static void ext4_invalidatepage(struct page
*page
, unsigned long offset
)
2562 journal_t
*journal
= EXT4_JOURNAL(page
->mapping
->host
);
2564 trace_ext4_invalidatepage(page
, offset
);
2567 * free any io_end structure allocated for buffers to be discarded
2569 if (ext4_should_dioread_nolock(page
->mapping
->host
))
2570 ext4_invalidatepage_free_endio(page
, offset
);
2572 * If it's a full truncate we just forget about the pending dirtying
2575 ClearPageChecked(page
);
2578 jbd2_journal_invalidatepage(journal
, page
, offset
);
2580 block_invalidatepage(page
, offset
);
2583 static int ext4_releasepage(struct page
*page
, gfp_t wait
)
2585 journal_t
*journal
= EXT4_JOURNAL(page
->mapping
->host
);
2587 trace_ext4_releasepage(page
);
2589 WARN_ON(PageChecked(page
));
2590 if (!page_has_buffers(page
))
2593 return jbd2_journal_try_to_free_buffers(journal
, page
, wait
);
2595 return try_to_free_buffers(page
);
2599 * ext4_get_block used when preparing for a DIO write or buffer write.
2600 * We allocate an uinitialized extent if blocks haven't been allocated.
2601 * The extent will be converted to initialized after the IO is complete.
2603 static int ext4_get_block_write(struct inode
*inode
, sector_t iblock
,
2604 struct buffer_head
*bh_result
, int create
)
2606 ext4_debug("ext4_get_block_write: inode %lu, create flag %d\n",
2607 inode
->i_ino
, create
);
2608 return _ext4_get_block(inode
, iblock
, bh_result
,
2609 EXT4_GET_BLOCKS_IO_CREATE_EXT
);
2612 static void ext4_end_io_dio(struct kiocb
*iocb
, loff_t offset
,
2613 ssize_t size
, void *private, int ret
,
2616 struct inode
*inode
= iocb
->ki_filp
->f_path
.dentry
->d_inode
;
2617 ext4_io_end_t
*io_end
= iocb
->private;
2618 struct workqueue_struct
*wq
;
2619 unsigned long flags
;
2620 struct ext4_inode_info
*ei
;
2622 /* if not async direct IO or dio with 0 bytes write, just return */
2623 if (!io_end
|| !size
)
2626 ext_debug("ext4_end_io_dio(): io_end 0x%p"
2627 "for inode %lu, iocb 0x%p, offset %llu, size %llu\n",
2628 iocb
->private, io_end
->inode
->i_ino
, iocb
, offset
,
2631 /* if not aio dio with unwritten extents, just free io and return */
2632 if (!(io_end
->flag
& EXT4_IO_END_UNWRITTEN
)) {
2633 ext4_free_io_end(io_end
);
2634 iocb
->private = NULL
;
2637 aio_complete(iocb
, ret
, 0);
2638 inode_dio_done(inode
);
2642 io_end
->offset
= offset
;
2643 io_end
->size
= size
;
2645 io_end
->iocb
= iocb
;
2646 io_end
->result
= ret
;
2648 wq
= EXT4_SB(io_end
->inode
->i_sb
)->dio_unwritten_wq
;
2650 /* Add the io_end to per-inode completed aio dio list*/
2651 ei
= EXT4_I(io_end
->inode
);
2652 spin_lock_irqsave(&ei
->i_completed_io_lock
, flags
);
2653 list_add_tail(&io_end
->list
, &ei
->i_completed_io_list
);
2654 spin_unlock_irqrestore(&ei
->i_completed_io_lock
, flags
);
2656 /* queue the work to convert unwritten extents to written */
2657 queue_work(wq
, &io_end
->work
);
2658 iocb
->private = NULL
;
2660 /* XXX: probably should move into the real I/O completion handler */
2661 inode_dio_done(inode
);
2664 static void ext4_end_io_buffer_write(struct buffer_head
*bh
, int uptodate
)
2666 ext4_io_end_t
*io_end
= bh
->b_private
;
2667 struct workqueue_struct
*wq
;
2668 struct inode
*inode
;
2669 unsigned long flags
;
2671 if (!test_clear_buffer_uninit(bh
) || !io_end
)
2674 if (!(io_end
->inode
->i_sb
->s_flags
& MS_ACTIVE
)) {
2675 printk("sb umounted, discard end_io request for inode %lu\n",
2676 io_end
->inode
->i_ino
);
2677 ext4_free_io_end(io_end
);
2681 io_end
->flag
= EXT4_IO_END_UNWRITTEN
;
2682 inode
= io_end
->inode
;
2684 /* Add the io_end to per-inode completed io list*/
2685 spin_lock_irqsave(&EXT4_I(inode
)->i_completed_io_lock
, flags
);
2686 list_add_tail(&io_end
->list
, &EXT4_I(inode
)->i_completed_io_list
);
2687 spin_unlock_irqrestore(&EXT4_I(inode
)->i_completed_io_lock
, flags
);
2689 wq
= EXT4_SB(inode
->i_sb
)->dio_unwritten_wq
;
2690 /* queue the work to convert unwritten extents to written */
2691 queue_work(wq
, &io_end
->work
);
2693 bh
->b_private
= NULL
;
2694 bh
->b_end_io
= NULL
;
2695 clear_buffer_uninit(bh
);
2696 end_buffer_async_write(bh
, uptodate
);
2699 static int ext4_set_bh_endio(struct buffer_head
*bh
, struct inode
*inode
)
2701 ext4_io_end_t
*io_end
;
2702 struct page
*page
= bh
->b_page
;
2703 loff_t offset
= (sector_t
)page
->index
<< PAGE_CACHE_SHIFT
;
2704 size_t size
= bh
->b_size
;
2707 io_end
= ext4_init_io_end(inode
, GFP_ATOMIC
);
2709 pr_warn_ratelimited("%s: allocation fail\n", __func__
);
2713 io_end
->offset
= offset
;
2714 io_end
->size
= size
;
2716 * We need to hold a reference to the page to make sure it
2717 * doesn't get evicted before ext4_end_io_work() has a chance
2718 * to convert the extent from written to unwritten.
2720 io_end
->page
= page
;
2721 get_page(io_end
->page
);
2723 bh
->b_private
= io_end
;
2724 bh
->b_end_io
= ext4_end_io_buffer_write
;
2729 * For ext4 extent files, ext4 will do direct-io write to holes,
2730 * preallocated extents, and those write extend the file, no need to
2731 * fall back to buffered IO.
2733 * For holes, we fallocate those blocks, mark them as uninitialized
2734 * If those blocks were preallocated, we mark sure they are splited, but
2735 * still keep the range to write as uninitialized.
2737 * The unwrritten extents will be converted to written when DIO is completed.
2738 * For async direct IO, since the IO may still pending when return, we
2739 * set up an end_io call back function, which will do the conversion
2740 * when async direct IO completed.
2742 * If the O_DIRECT write will extend the file then add this inode to the
2743 * orphan list. So recovery will truncate it back to the original size
2744 * if the machine crashes during the write.
2747 static ssize_t
ext4_ext_direct_IO(int rw
, struct kiocb
*iocb
,
2748 const struct iovec
*iov
, loff_t offset
,
2749 unsigned long nr_segs
)
2751 struct file
*file
= iocb
->ki_filp
;
2752 struct inode
*inode
= file
->f_mapping
->host
;
2754 size_t count
= iov_length(iov
, nr_segs
);
2756 loff_t final_size
= offset
+ count
;
2757 if (rw
== WRITE
&& final_size
<= inode
->i_size
) {
2759 * We could direct write to holes and fallocate.
2761 * Allocated blocks to fill the hole are marked as uninitialized
2762 * to prevent parallel buffered read to expose the stale data
2763 * before DIO complete the data IO.
2765 * As to previously fallocated extents, ext4 get_block
2766 * will just simply mark the buffer mapped but still
2767 * keep the extents uninitialized.
2769 * for non AIO case, we will convert those unwritten extents
2770 * to written after return back from blockdev_direct_IO.
2772 * for async DIO, the conversion needs to be defered when
2773 * the IO is completed. The ext4 end_io callback function
2774 * will be called to take care of the conversion work.
2775 * Here for async case, we allocate an io_end structure to
2778 iocb
->private = NULL
;
2779 EXT4_I(inode
)->cur_aio_dio
= NULL
;
2780 if (!is_sync_kiocb(iocb
)) {
2781 iocb
->private = ext4_init_io_end(inode
, GFP_NOFS
);
2785 * we save the io structure for current async
2786 * direct IO, so that later ext4_map_blocks()
2787 * could flag the io structure whether there
2788 * is a unwritten extents needs to be converted
2789 * when IO is completed.
2791 EXT4_I(inode
)->cur_aio_dio
= iocb
->private;
2794 ret
= __blockdev_direct_IO(rw
, iocb
, inode
,
2795 inode
->i_sb
->s_bdev
, iov
,
2797 ext4_get_block_write
,
2800 DIO_LOCKING
| DIO_SKIP_HOLES
);
2802 EXT4_I(inode
)->cur_aio_dio
= NULL
;
2804 * The io_end structure takes a reference to the inode,
2805 * that structure needs to be destroyed and the
2806 * reference to the inode need to be dropped, when IO is
2807 * complete, even with 0 byte write, or failed.
2809 * In the successful AIO DIO case, the io_end structure will be
2810 * desctroyed and the reference to the inode will be dropped
2811 * after the end_io call back function is called.
2813 * In the case there is 0 byte write, or error case, since
2814 * VFS direct IO won't invoke the end_io call back function,
2815 * we need to free the end_io structure here.
2817 if (ret
!= -EIOCBQUEUED
&& ret
<= 0 && iocb
->private) {
2818 ext4_free_io_end(iocb
->private);
2819 iocb
->private = NULL
;
2820 } else if (ret
> 0 && ext4_test_inode_state(inode
,
2821 EXT4_STATE_DIO_UNWRITTEN
)) {
2824 * for non AIO case, since the IO is already
2825 * completed, we could do the conversion right here
2827 err
= ext4_convert_unwritten_extents(inode
,
2831 ext4_clear_inode_state(inode
, EXT4_STATE_DIO_UNWRITTEN
);
2836 /* for write the the end of file case, we fall back to old way */
2837 return ext4_ind_direct_IO(rw
, iocb
, iov
, offset
, nr_segs
);
2840 static ssize_t
ext4_direct_IO(int rw
, struct kiocb
*iocb
,
2841 const struct iovec
*iov
, loff_t offset
,
2842 unsigned long nr_segs
)
2844 struct file
*file
= iocb
->ki_filp
;
2845 struct inode
*inode
= file
->f_mapping
->host
;
2848 trace_ext4_direct_IO_enter(inode
, offset
, iov_length(iov
, nr_segs
), rw
);
2849 if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
))
2850 ret
= ext4_ext_direct_IO(rw
, iocb
, iov
, offset
, nr_segs
);
2852 ret
= ext4_ind_direct_IO(rw
, iocb
, iov
, offset
, nr_segs
);
2853 trace_ext4_direct_IO_exit(inode
, offset
,
2854 iov_length(iov
, nr_segs
), rw
, ret
);
2859 * Pages can be marked dirty completely asynchronously from ext4's journalling
2860 * activity. By filemap_sync_pte(), try_to_unmap_one(), etc. We cannot do
2861 * much here because ->set_page_dirty is called under VFS locks. The page is
2862 * not necessarily locked.
2864 * We cannot just dirty the page and leave attached buffers clean, because the
2865 * buffers' dirty state is "definitive". We cannot just set the buffers dirty
2866 * or jbddirty because all the journalling code will explode.
2868 * So what we do is to mark the page "pending dirty" and next time writepage
2869 * is called, propagate that into the buffers appropriately.
2871 static int ext4_journalled_set_page_dirty(struct page
*page
)
2873 SetPageChecked(page
);
2874 return __set_page_dirty_nobuffers(page
);
2877 static const struct address_space_operations ext4_ordered_aops
= {
2878 .readpage
= ext4_readpage
,
2879 .readpages
= ext4_readpages
,
2880 .writepage
= ext4_writepage
,
2881 .write_begin
= ext4_write_begin
,
2882 .write_end
= ext4_ordered_write_end
,
2884 .invalidatepage
= ext4_invalidatepage
,
2885 .releasepage
= ext4_releasepage
,
2886 .direct_IO
= ext4_direct_IO
,
2887 .migratepage
= buffer_migrate_page
,
2888 .is_partially_uptodate
= block_is_partially_uptodate
,
2889 .error_remove_page
= generic_error_remove_page
,
2892 static const struct address_space_operations ext4_writeback_aops
= {
2893 .readpage
= ext4_readpage
,
2894 .readpages
= ext4_readpages
,
2895 .writepage
= ext4_writepage
,
2896 .write_begin
= ext4_write_begin
,
2897 .write_end
= ext4_writeback_write_end
,
2899 .invalidatepage
= ext4_invalidatepage
,
2900 .releasepage
= ext4_releasepage
,
2901 .direct_IO
= ext4_direct_IO
,
2902 .migratepage
= buffer_migrate_page
,
2903 .is_partially_uptodate
= block_is_partially_uptodate
,
2904 .error_remove_page
= generic_error_remove_page
,
2907 static const struct address_space_operations ext4_journalled_aops
= {
2908 .readpage
= ext4_readpage
,
2909 .readpages
= ext4_readpages
,
2910 .writepage
= ext4_writepage
,
2911 .write_begin
= ext4_write_begin
,
2912 .write_end
= ext4_journalled_write_end
,
2913 .set_page_dirty
= ext4_journalled_set_page_dirty
,
2915 .invalidatepage
= ext4_invalidatepage
,
2916 .releasepage
= ext4_releasepage
,
2917 .is_partially_uptodate
= block_is_partially_uptodate
,
2918 .error_remove_page
= generic_error_remove_page
,
2921 static const struct address_space_operations ext4_da_aops
= {
2922 .readpage
= ext4_readpage
,
2923 .readpages
= ext4_readpages
,
2924 .writepage
= ext4_writepage
,
2925 .writepages
= ext4_da_writepages
,
2926 .write_begin
= ext4_da_write_begin
,
2927 .write_end
= ext4_da_write_end
,
2929 .invalidatepage
= ext4_da_invalidatepage
,
2930 .releasepage
= ext4_releasepage
,
2931 .direct_IO
= ext4_direct_IO
,
2932 .migratepage
= buffer_migrate_page
,
2933 .is_partially_uptodate
= block_is_partially_uptodate
,
2934 .error_remove_page
= generic_error_remove_page
,
2937 void ext4_set_aops(struct inode
*inode
)
2939 if (ext4_should_order_data(inode
) &&
2940 test_opt(inode
->i_sb
, DELALLOC
))
2941 inode
->i_mapping
->a_ops
= &ext4_da_aops
;
2942 else if (ext4_should_order_data(inode
))
2943 inode
->i_mapping
->a_ops
= &ext4_ordered_aops
;
2944 else if (ext4_should_writeback_data(inode
) &&
2945 test_opt(inode
->i_sb
, DELALLOC
))
2946 inode
->i_mapping
->a_ops
= &ext4_da_aops
;
2947 else if (ext4_should_writeback_data(inode
))
2948 inode
->i_mapping
->a_ops
= &ext4_writeback_aops
;
2950 inode
->i_mapping
->a_ops
= &ext4_journalled_aops
;
2954 * ext4_block_truncate_page() zeroes out a mapping from file offset `from'
2955 * up to the end of the block which corresponds to `from'.
2956 * This required during truncate. We need to physically zero the tail end
2957 * of that block so it doesn't yield old data if the file is later grown.
2959 int ext4_block_truncate_page(handle_t
*handle
,
2960 struct address_space
*mapping
, loff_t from
)
2962 unsigned offset
= from
& (PAGE_CACHE_SIZE
-1);
2965 struct inode
*inode
= mapping
->host
;
2967 blocksize
= inode
->i_sb
->s_blocksize
;
2968 length
= blocksize
- (offset
& (blocksize
- 1));
2970 return ext4_block_zero_page_range(handle
, mapping
, from
, length
);
2974 * ext4_block_zero_page_range() zeros out a mapping of length 'length'
2975 * starting from file offset 'from'. The range to be zero'd must
2976 * be contained with in one block. If the specified range exceeds
2977 * the end of the block it will be shortened to end of the block
2978 * that cooresponds to 'from'
2980 int ext4_block_zero_page_range(handle_t
*handle
,
2981 struct address_space
*mapping
, loff_t from
, loff_t length
)
2983 ext4_fsblk_t index
= from
>> PAGE_CACHE_SHIFT
;
2984 unsigned offset
= from
& (PAGE_CACHE_SIZE
-1);
2985 unsigned blocksize
, max
, pos
;
2987 struct inode
*inode
= mapping
->host
;
2988 struct buffer_head
*bh
;
2992 page
= find_or_create_page(mapping
, from
>> PAGE_CACHE_SHIFT
,
2993 mapping_gfp_mask(mapping
) & ~__GFP_FS
);
2997 blocksize
= inode
->i_sb
->s_blocksize
;
2998 max
= blocksize
- (offset
& (blocksize
- 1));
3001 * correct length if it does not fall between
3002 * 'from' and the end of the block
3004 if (length
> max
|| length
< 0)
3007 iblock
= index
<< (PAGE_CACHE_SHIFT
- inode
->i_sb
->s_blocksize_bits
);
3009 if (!page_has_buffers(page
))
3010 create_empty_buffers(page
, blocksize
, 0);
3012 /* Find the buffer that contains "offset" */
3013 bh
= page_buffers(page
);
3015 while (offset
>= pos
) {
3016 bh
= bh
->b_this_page
;
3022 if (buffer_freed(bh
)) {
3023 BUFFER_TRACE(bh
, "freed: skip");
3027 if (!buffer_mapped(bh
)) {
3028 BUFFER_TRACE(bh
, "unmapped");
3029 ext4_get_block(inode
, iblock
, bh
, 0);
3030 /* unmapped? It's a hole - nothing to do */
3031 if (!buffer_mapped(bh
)) {
3032 BUFFER_TRACE(bh
, "still unmapped");
3037 /* Ok, it's mapped. Make sure it's up-to-date */
3038 if (PageUptodate(page
))
3039 set_buffer_uptodate(bh
);
3041 if (!buffer_uptodate(bh
)) {
3043 ll_rw_block(READ
, 1, &bh
);
3045 /* Uhhuh. Read error. Complain and punt. */
3046 if (!buffer_uptodate(bh
))
3050 if (ext4_should_journal_data(inode
)) {
3051 BUFFER_TRACE(bh
, "get write access");
3052 err
= ext4_journal_get_write_access(handle
, bh
);
3057 zero_user(page
, offset
, length
);
3059 BUFFER_TRACE(bh
, "zeroed end of block");
3062 if (ext4_should_journal_data(inode
)) {
3063 err
= ext4_handle_dirty_metadata(handle
, inode
, bh
);
3065 if (ext4_should_order_data(inode
) && EXT4_I(inode
)->jinode
)
3066 err
= ext4_jbd2_file_inode(handle
, inode
);
3067 mark_buffer_dirty(bh
);
3072 page_cache_release(page
);
3076 int ext4_can_truncate(struct inode
*inode
)
3078 if (S_ISREG(inode
->i_mode
))
3080 if (S_ISDIR(inode
->i_mode
))
3082 if (S_ISLNK(inode
->i_mode
))
3083 return !ext4_inode_is_fast_symlink(inode
);
3088 * ext4_punch_hole: punches a hole in a file by releaseing the blocks
3089 * associated with the given offset and length
3091 * @inode: File inode
3092 * @offset: The offset where the hole will begin
3093 * @len: The length of the hole
3095 * Returns: 0 on sucess or negative on failure
3098 int ext4_punch_hole(struct file
*file
, loff_t offset
, loff_t length
)
3100 struct inode
*inode
= file
->f_path
.dentry
->d_inode
;
3101 if (!S_ISREG(inode
->i_mode
))
3104 if (!ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
)) {
3105 /* TODO: Add support for non extent hole punching */
3109 return ext4_ext_punch_hole(file
, offset
, length
);
3115 * We block out ext4_get_block() block instantiations across the entire
3116 * transaction, and VFS/VM ensures that ext4_truncate() cannot run
3117 * simultaneously on behalf of the same inode.
3119 * As we work through the truncate and commmit bits of it to the journal there
3120 * is one core, guiding principle: the file's tree must always be consistent on
3121 * disk. We must be able to restart the truncate after a crash.
3123 * The file's tree may be transiently inconsistent in memory (although it
3124 * probably isn't), but whenever we close off and commit a journal transaction,
3125 * the contents of (the filesystem + the journal) must be consistent and
3126 * restartable. It's pretty simple, really: bottom up, right to left (although
3127 * left-to-right works OK too).
3129 * Note that at recovery time, journal replay occurs *before* the restart of
3130 * truncate against the orphan inode list.
3132 * The committed inode has the new, desired i_size (which is the same as
3133 * i_disksize in this case). After a crash, ext4_orphan_cleanup() will see
3134 * that this inode's truncate did not complete and it will again call
3135 * ext4_truncate() to have another go. So there will be instantiated blocks
3136 * to the right of the truncation point in a crashed ext4 filesystem. But
3137 * that's fine - as long as they are linked from the inode, the post-crash
3138 * ext4_truncate() run will find them and release them.
3140 void ext4_truncate(struct inode
*inode
)
3142 trace_ext4_truncate_enter(inode
);
3144 if (!ext4_can_truncate(inode
))
3147 ext4_clear_inode_flag(inode
, EXT4_INODE_EOFBLOCKS
);
3149 if (inode
->i_size
== 0 && !test_opt(inode
->i_sb
, NO_AUTO_DA_ALLOC
))
3150 ext4_set_inode_state(inode
, EXT4_STATE_DA_ALLOC_CLOSE
);
3152 if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
))
3153 ext4_ext_truncate(inode
);
3155 ext4_ind_truncate(inode
);
3157 trace_ext4_truncate_exit(inode
);
3161 * ext4_get_inode_loc returns with an extra refcount against the inode's
3162 * underlying buffer_head on success. If 'in_mem' is true, we have all
3163 * data in memory that is needed to recreate the on-disk version of this
3166 static int __ext4_get_inode_loc(struct inode
*inode
,
3167 struct ext4_iloc
*iloc
, int in_mem
)
3169 struct ext4_group_desc
*gdp
;
3170 struct buffer_head
*bh
;
3171 struct super_block
*sb
= inode
->i_sb
;
3173 int inodes_per_block
, inode_offset
;
3176 if (!ext4_valid_inum(sb
, inode
->i_ino
))
3179 iloc
->block_group
= (inode
->i_ino
- 1) / EXT4_INODES_PER_GROUP(sb
);
3180 gdp
= ext4_get_group_desc(sb
, iloc
->block_group
, NULL
);
3185 * Figure out the offset within the block group inode table
3187 inodes_per_block
= EXT4_SB(sb
)->s_inodes_per_block
;
3188 inode_offset
= ((inode
->i_ino
- 1) %
3189 EXT4_INODES_PER_GROUP(sb
));
3190 block
= ext4_inode_table(sb
, gdp
) + (inode_offset
/ inodes_per_block
);
3191 iloc
->offset
= (inode_offset
% inodes_per_block
) * EXT4_INODE_SIZE(sb
);
3193 bh
= sb_getblk(sb
, block
);
3195 EXT4_ERROR_INODE_BLOCK(inode
, block
,
3196 "unable to read itable block");
3199 if (!buffer_uptodate(bh
)) {
3203 * If the buffer has the write error flag, we have failed
3204 * to write out another inode in the same block. In this
3205 * case, we don't have to read the block because we may
3206 * read the old inode data successfully.
3208 if (buffer_write_io_error(bh
) && !buffer_uptodate(bh
))
3209 set_buffer_uptodate(bh
);
3211 if (buffer_uptodate(bh
)) {
3212 /* someone brought it uptodate while we waited */
3218 * If we have all information of the inode in memory and this
3219 * is the only valid inode in the block, we need not read the
3223 struct buffer_head
*bitmap_bh
;
3226 start
= inode_offset
& ~(inodes_per_block
- 1);
3228 /* Is the inode bitmap in cache? */
3229 bitmap_bh
= sb_getblk(sb
, ext4_inode_bitmap(sb
, gdp
));
3234 * If the inode bitmap isn't in cache then the
3235 * optimisation may end up performing two reads instead
3236 * of one, so skip it.
3238 if (!buffer_uptodate(bitmap_bh
)) {
3242 for (i
= start
; i
< start
+ inodes_per_block
; i
++) {
3243 if (i
== inode_offset
)
3245 if (ext4_test_bit(i
, bitmap_bh
->b_data
))
3249 if (i
== start
+ inodes_per_block
) {
3250 /* all other inodes are free, so skip I/O */
3251 memset(bh
->b_data
, 0, bh
->b_size
);
3252 set_buffer_uptodate(bh
);
3260 * If we need to do any I/O, try to pre-readahead extra
3261 * blocks from the inode table.
3263 if (EXT4_SB(sb
)->s_inode_readahead_blks
) {
3264 ext4_fsblk_t b
, end
, table
;
3267 table
= ext4_inode_table(sb
, gdp
);
3268 /* s_inode_readahead_blks is always a power of 2 */
3269 b
= block
& ~(EXT4_SB(sb
)->s_inode_readahead_blks
-1);
3272 end
= b
+ EXT4_SB(sb
)->s_inode_readahead_blks
;
3273 num
= EXT4_INODES_PER_GROUP(sb
);
3274 if (EXT4_HAS_RO_COMPAT_FEATURE(sb
,
3275 EXT4_FEATURE_RO_COMPAT_GDT_CSUM
))
3276 num
-= ext4_itable_unused_count(sb
, gdp
);
3277 table
+= num
/ inodes_per_block
;
3281 sb_breadahead(sb
, b
++);
3285 * There are other valid inodes in the buffer, this inode
3286 * has in-inode xattrs, or we don't have this inode in memory.
3287 * Read the block from disk.
3289 trace_ext4_load_inode(inode
);
3291 bh
->b_end_io
= end_buffer_read_sync
;
3292 submit_bh(READ_META
, bh
);
3294 if (!buffer_uptodate(bh
)) {
3295 EXT4_ERROR_INODE_BLOCK(inode
, block
,
3296 "unable to read itable block");
3306 int ext4_get_inode_loc(struct inode
*inode
, struct ext4_iloc
*iloc
)
3308 /* We have all inode data except xattrs in memory here. */
3309 return __ext4_get_inode_loc(inode
, iloc
,
3310 !ext4_test_inode_state(inode
, EXT4_STATE_XATTR
));
3313 void ext4_set_inode_flags(struct inode
*inode
)
3315 unsigned int flags
= EXT4_I(inode
)->i_flags
;
3317 inode
->i_flags
&= ~(S_SYNC
|S_APPEND
|S_IMMUTABLE
|S_NOATIME
|S_DIRSYNC
);
3318 if (flags
& EXT4_SYNC_FL
)
3319 inode
->i_flags
|= S_SYNC
;
3320 if (flags
& EXT4_APPEND_FL
)
3321 inode
->i_flags
|= S_APPEND
;
3322 if (flags
& EXT4_IMMUTABLE_FL
)
3323 inode
->i_flags
|= S_IMMUTABLE
;
3324 if (flags
& EXT4_NOATIME_FL
)
3325 inode
->i_flags
|= S_NOATIME
;
3326 if (flags
& EXT4_DIRSYNC_FL
)
3327 inode
->i_flags
|= S_DIRSYNC
;
3330 /* Propagate flags from i_flags to EXT4_I(inode)->i_flags */
3331 void ext4_get_inode_flags(struct ext4_inode_info
*ei
)
3333 unsigned int vfs_fl
;
3334 unsigned long old_fl
, new_fl
;
3337 vfs_fl
= ei
->vfs_inode
.i_flags
;
3338 old_fl
= ei
->i_flags
;
3339 new_fl
= old_fl
& ~(EXT4_SYNC_FL
|EXT4_APPEND_FL
|
3340 EXT4_IMMUTABLE_FL
|EXT4_NOATIME_FL
|
3342 if (vfs_fl
& S_SYNC
)
3343 new_fl
|= EXT4_SYNC_FL
;
3344 if (vfs_fl
& S_APPEND
)
3345 new_fl
|= EXT4_APPEND_FL
;
3346 if (vfs_fl
& S_IMMUTABLE
)
3347 new_fl
|= EXT4_IMMUTABLE_FL
;
3348 if (vfs_fl
& S_NOATIME
)
3349 new_fl
|= EXT4_NOATIME_FL
;
3350 if (vfs_fl
& S_DIRSYNC
)
3351 new_fl
|= EXT4_DIRSYNC_FL
;
3352 } while (cmpxchg(&ei
->i_flags
, old_fl
, new_fl
) != old_fl
);
3355 static blkcnt_t
ext4_inode_blocks(struct ext4_inode
*raw_inode
,
3356 struct ext4_inode_info
*ei
)
3359 struct inode
*inode
= &(ei
->vfs_inode
);
3360 struct super_block
*sb
= inode
->i_sb
;
3362 if (EXT4_HAS_RO_COMPAT_FEATURE(sb
,
3363 EXT4_FEATURE_RO_COMPAT_HUGE_FILE
)) {
3364 /* we are using combined 48 bit field */
3365 i_blocks
= ((u64
)le16_to_cpu(raw_inode
->i_blocks_high
)) << 32 |
3366 le32_to_cpu(raw_inode
->i_blocks_lo
);
3367 if (ext4_test_inode_flag(inode
, EXT4_INODE_HUGE_FILE
)) {
3368 /* i_blocks represent file system block size */
3369 return i_blocks
<< (inode
->i_blkbits
- 9);
3374 return le32_to_cpu(raw_inode
->i_blocks_lo
);
3378 struct inode
*ext4_iget(struct super_block
*sb
, unsigned long ino
)
3380 struct ext4_iloc iloc
;
3381 struct ext4_inode
*raw_inode
;
3382 struct ext4_inode_info
*ei
;
3383 struct inode
*inode
;
3384 journal_t
*journal
= EXT4_SB(sb
)->s_journal
;
3388 inode
= iget_locked(sb
, ino
);
3390 return ERR_PTR(-ENOMEM
);
3391 if (!(inode
->i_state
& I_NEW
))
3397 ret
= __ext4_get_inode_loc(inode
, &iloc
, 0);
3400 raw_inode
= ext4_raw_inode(&iloc
);
3401 inode
->i_mode
= le16_to_cpu(raw_inode
->i_mode
);
3402 inode
->i_uid
= (uid_t
)le16_to_cpu(raw_inode
->i_uid_low
);
3403 inode
->i_gid
= (gid_t
)le16_to_cpu(raw_inode
->i_gid_low
);
3404 if (!(test_opt(inode
->i_sb
, NO_UID32
))) {
3405 inode
->i_uid
|= le16_to_cpu(raw_inode
->i_uid_high
) << 16;
3406 inode
->i_gid
|= le16_to_cpu(raw_inode
->i_gid_high
) << 16;
3408 inode
->i_nlink
= le16_to_cpu(raw_inode
->i_links_count
);
3410 ext4_clear_state_flags(ei
); /* Only relevant on 32-bit archs */
3411 ei
->i_dir_start_lookup
= 0;
3412 ei
->i_dtime
= le32_to_cpu(raw_inode
->i_dtime
);
3413 /* We now have enough fields to check if the inode was active or not.
3414 * This is needed because nfsd might try to access dead inodes
3415 * the test is that same one that e2fsck uses
3416 * NeilBrown 1999oct15
3418 if (inode
->i_nlink
== 0) {
3419 if (inode
->i_mode
== 0 ||
3420 !(EXT4_SB(inode
->i_sb
)->s_mount_state
& EXT4_ORPHAN_FS
)) {
3421 /* this inode is deleted */
3425 /* The only unlinked inodes we let through here have
3426 * valid i_mode and are being read by the orphan
3427 * recovery code: that's fine, we're about to complete
3428 * the process of deleting those. */
3430 ei
->i_flags
= le32_to_cpu(raw_inode
->i_flags
);
3431 inode
->i_blocks
= ext4_inode_blocks(raw_inode
, ei
);
3432 ei
->i_file_acl
= le32_to_cpu(raw_inode
->i_file_acl_lo
);
3433 if (EXT4_HAS_INCOMPAT_FEATURE(sb
, EXT4_FEATURE_INCOMPAT_64BIT
))
3435 ((__u64
)le16_to_cpu(raw_inode
->i_file_acl_high
)) << 32;
3436 inode
->i_size
= ext4_isize(raw_inode
);
3437 ei
->i_disksize
= inode
->i_size
;
3439 ei
->i_reserved_quota
= 0;
3441 inode
->i_generation
= le32_to_cpu(raw_inode
->i_generation
);
3442 ei
->i_block_group
= iloc
.block_group
;
3443 ei
->i_last_alloc_group
= ~0;
3445 * NOTE! The in-memory inode i_data array is in little-endian order
3446 * even on big-endian machines: we do NOT byteswap the block numbers!
3448 for (block
= 0; block
< EXT4_N_BLOCKS
; block
++)
3449 ei
->i_data
[block
] = raw_inode
->i_block
[block
];
3450 INIT_LIST_HEAD(&ei
->i_orphan
);
3453 * Set transaction id's of transactions that have to be committed
3454 * to finish f[data]sync. We set them to currently running transaction
3455 * as we cannot be sure that the inode or some of its metadata isn't
3456 * part of the transaction - the inode could have been reclaimed and
3457 * now it is reread from disk.
3460 transaction_t
*transaction
;
3463 read_lock(&journal
->j_state_lock
);
3464 if (journal
->j_running_transaction
)
3465 transaction
= journal
->j_running_transaction
;
3467 transaction
= journal
->j_committing_transaction
;
3469 tid
= transaction
->t_tid
;
3471 tid
= journal
->j_commit_sequence
;
3472 read_unlock(&journal
->j_state_lock
);
3473 ei
->i_sync_tid
= tid
;
3474 ei
->i_datasync_tid
= tid
;
3477 if (EXT4_INODE_SIZE(inode
->i_sb
) > EXT4_GOOD_OLD_INODE_SIZE
) {
3478 ei
->i_extra_isize
= le16_to_cpu(raw_inode
->i_extra_isize
);
3479 if (EXT4_GOOD_OLD_INODE_SIZE
+ ei
->i_extra_isize
>
3480 EXT4_INODE_SIZE(inode
->i_sb
)) {
3484 if (ei
->i_extra_isize
== 0) {
3485 /* The extra space is currently unused. Use it. */
3486 ei
->i_extra_isize
= sizeof(struct ext4_inode
) -
3487 EXT4_GOOD_OLD_INODE_SIZE
;
3489 __le32
*magic
= (void *)raw_inode
+
3490 EXT4_GOOD_OLD_INODE_SIZE
+
3492 if (*magic
== cpu_to_le32(EXT4_XATTR_MAGIC
))
3493 ext4_set_inode_state(inode
, EXT4_STATE_XATTR
);
3496 ei
->i_extra_isize
= 0;
3498 EXT4_INODE_GET_XTIME(i_ctime
, inode
, raw_inode
);
3499 EXT4_INODE_GET_XTIME(i_mtime
, inode
, raw_inode
);
3500 EXT4_INODE_GET_XTIME(i_atime
, inode
, raw_inode
);
3501 EXT4_EINODE_GET_XTIME(i_crtime
, ei
, raw_inode
);
3503 inode
->i_version
= le32_to_cpu(raw_inode
->i_disk_version
);
3504 if (EXT4_INODE_SIZE(inode
->i_sb
) > EXT4_GOOD_OLD_INODE_SIZE
) {
3505 if (EXT4_FITS_IN_INODE(raw_inode
, ei
, i_version_hi
))
3507 (__u64
)(le32_to_cpu(raw_inode
->i_version_hi
)) << 32;
3511 if (ei
->i_file_acl
&&
3512 !ext4_data_block_valid(EXT4_SB(sb
), ei
->i_file_acl
, 1)) {
3513 EXT4_ERROR_INODE(inode
, "bad extended attribute block %llu",
3517 } else if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
)) {
3518 if (S_ISREG(inode
->i_mode
) || S_ISDIR(inode
->i_mode
) ||
3519 (S_ISLNK(inode
->i_mode
) &&
3520 !ext4_inode_is_fast_symlink(inode
)))
3521 /* Validate extent which is part of inode */
3522 ret
= ext4_ext_check_inode(inode
);
3523 } else if (S_ISREG(inode
->i_mode
) || S_ISDIR(inode
->i_mode
) ||
3524 (S_ISLNK(inode
->i_mode
) &&
3525 !ext4_inode_is_fast_symlink(inode
))) {
3526 /* Validate block references which are part of inode */
3527 ret
= ext4_ind_check_inode(inode
);
3532 if (S_ISREG(inode
->i_mode
)) {
3533 inode
->i_op
= &ext4_file_inode_operations
;
3534 inode
->i_fop
= &ext4_file_operations
;
3535 ext4_set_aops(inode
);
3536 } else if (S_ISDIR(inode
->i_mode
)) {
3537 inode
->i_op
= &ext4_dir_inode_operations
;
3538 inode
->i_fop
= &ext4_dir_operations
;
3539 } else if (S_ISLNK(inode
->i_mode
)) {
3540 if (ext4_inode_is_fast_symlink(inode
)) {
3541 inode
->i_op
= &ext4_fast_symlink_inode_operations
;
3542 nd_terminate_link(ei
->i_data
, inode
->i_size
,
3543 sizeof(ei
->i_data
) - 1);
3545 inode
->i_op
= &ext4_symlink_inode_operations
;
3546 ext4_set_aops(inode
);
3548 } else if (S_ISCHR(inode
->i_mode
) || S_ISBLK(inode
->i_mode
) ||
3549 S_ISFIFO(inode
->i_mode
) || S_ISSOCK(inode
->i_mode
)) {
3550 inode
->i_op
= &ext4_special_inode_operations
;
3551 if (raw_inode
->i_block
[0])
3552 init_special_inode(inode
, inode
->i_mode
,
3553 old_decode_dev(le32_to_cpu(raw_inode
->i_block
[0])));
3555 init_special_inode(inode
, inode
->i_mode
,
3556 new_decode_dev(le32_to_cpu(raw_inode
->i_block
[1])));
3559 EXT4_ERROR_INODE(inode
, "bogus i_mode (%o)", inode
->i_mode
);
3563 ext4_set_inode_flags(inode
);
3564 unlock_new_inode(inode
);
3570 return ERR_PTR(ret
);
3573 static int ext4_inode_blocks_set(handle_t
*handle
,
3574 struct ext4_inode
*raw_inode
,
3575 struct ext4_inode_info
*ei
)
3577 struct inode
*inode
= &(ei
->vfs_inode
);
3578 u64 i_blocks
= inode
->i_blocks
;
3579 struct super_block
*sb
= inode
->i_sb
;
3581 if (i_blocks
<= ~0U) {
3583 * i_blocks can be represnted in a 32 bit variable
3584 * as multiple of 512 bytes
3586 raw_inode
->i_blocks_lo
= cpu_to_le32(i_blocks
);
3587 raw_inode
->i_blocks_high
= 0;
3588 ext4_clear_inode_flag(inode
, EXT4_INODE_HUGE_FILE
);
3591 if (!EXT4_HAS_RO_COMPAT_FEATURE(sb
, EXT4_FEATURE_RO_COMPAT_HUGE_FILE
))
3594 if (i_blocks
<= 0xffffffffffffULL
) {
3596 * i_blocks can be represented in a 48 bit variable
3597 * as multiple of 512 bytes
3599 raw_inode
->i_blocks_lo
= cpu_to_le32(i_blocks
);
3600 raw_inode
->i_blocks_high
= cpu_to_le16(i_blocks
>> 32);
3601 ext4_clear_inode_flag(inode
, EXT4_INODE_HUGE_FILE
);
3603 ext4_set_inode_flag(inode
, EXT4_INODE_HUGE_FILE
);
3604 /* i_block is stored in file system block size */
3605 i_blocks
= i_blocks
>> (inode
->i_blkbits
- 9);
3606 raw_inode
->i_blocks_lo
= cpu_to_le32(i_blocks
);
3607 raw_inode
->i_blocks_high
= cpu_to_le16(i_blocks
>> 32);
3613 * Post the struct inode info into an on-disk inode location in the
3614 * buffer-cache. This gobbles the caller's reference to the
3615 * buffer_head in the inode location struct.
3617 * The caller must have write access to iloc->bh.
3619 static int ext4_do_update_inode(handle_t
*handle
,
3620 struct inode
*inode
,
3621 struct ext4_iloc
*iloc
)
3623 struct ext4_inode
*raw_inode
= ext4_raw_inode(iloc
);
3624 struct ext4_inode_info
*ei
= EXT4_I(inode
);
3625 struct buffer_head
*bh
= iloc
->bh
;
3626 int err
= 0, rc
, block
;
3628 /* For fields not not tracking in the in-memory inode,
3629 * initialise them to zero for new inodes. */
3630 if (ext4_test_inode_state(inode
, EXT4_STATE_NEW
))
3631 memset(raw_inode
, 0, EXT4_SB(inode
->i_sb
)->s_inode_size
);
3633 ext4_get_inode_flags(ei
);
3634 raw_inode
->i_mode
= cpu_to_le16(inode
->i_mode
);
3635 if (!(test_opt(inode
->i_sb
, NO_UID32
))) {
3636 raw_inode
->i_uid_low
= cpu_to_le16(low_16_bits(inode
->i_uid
));
3637 raw_inode
->i_gid_low
= cpu_to_le16(low_16_bits(inode
->i_gid
));
3639 * Fix up interoperability with old kernels. Otherwise, old inodes get
3640 * re-used with the upper 16 bits of the uid/gid intact
3643 raw_inode
->i_uid_high
=
3644 cpu_to_le16(high_16_bits(inode
->i_uid
));
3645 raw_inode
->i_gid_high
=
3646 cpu_to_le16(high_16_bits(inode
->i_gid
));
3648 raw_inode
->i_uid_high
= 0;
3649 raw_inode
->i_gid_high
= 0;
3652 raw_inode
->i_uid_low
=
3653 cpu_to_le16(fs_high2lowuid(inode
->i_uid
));
3654 raw_inode
->i_gid_low
=
3655 cpu_to_le16(fs_high2lowgid(inode
->i_gid
));
3656 raw_inode
->i_uid_high
= 0;
3657 raw_inode
->i_gid_high
= 0;
3659 raw_inode
->i_links_count
= cpu_to_le16(inode
->i_nlink
);
3661 EXT4_INODE_SET_XTIME(i_ctime
, inode
, raw_inode
);
3662 EXT4_INODE_SET_XTIME(i_mtime
, inode
, raw_inode
);
3663 EXT4_INODE_SET_XTIME(i_atime
, inode
, raw_inode
);
3664 EXT4_EINODE_SET_XTIME(i_crtime
, ei
, raw_inode
);
3666 if (ext4_inode_blocks_set(handle
, raw_inode
, ei
))
3668 raw_inode
->i_dtime
= cpu_to_le32(ei
->i_dtime
);
3669 raw_inode
->i_flags
= cpu_to_le32(ei
->i_flags
& 0xFFFFFFFF);
3670 if (EXT4_SB(inode
->i_sb
)->s_es
->s_creator_os
!=
3671 cpu_to_le32(EXT4_OS_HURD
))
3672 raw_inode
->i_file_acl_high
=
3673 cpu_to_le16(ei
->i_file_acl
>> 32);
3674 raw_inode
->i_file_acl_lo
= cpu_to_le32(ei
->i_file_acl
);
3675 ext4_isize_set(raw_inode
, ei
->i_disksize
);
3676 if (ei
->i_disksize
> 0x7fffffffULL
) {
3677 struct super_block
*sb
= inode
->i_sb
;
3678 if (!EXT4_HAS_RO_COMPAT_FEATURE(sb
,
3679 EXT4_FEATURE_RO_COMPAT_LARGE_FILE
) ||
3680 EXT4_SB(sb
)->s_es
->s_rev_level
==
3681 cpu_to_le32(EXT4_GOOD_OLD_REV
)) {
3682 /* If this is the first large file
3683 * created, add a flag to the superblock.
3685 err
= ext4_journal_get_write_access(handle
,
3686 EXT4_SB(sb
)->s_sbh
);
3689 ext4_update_dynamic_rev(sb
);
3690 EXT4_SET_RO_COMPAT_FEATURE(sb
,
3691 EXT4_FEATURE_RO_COMPAT_LARGE_FILE
);
3693 ext4_handle_sync(handle
);
3694 err
= ext4_handle_dirty_metadata(handle
, NULL
,
3695 EXT4_SB(sb
)->s_sbh
);
3698 raw_inode
->i_generation
= cpu_to_le32(inode
->i_generation
);
3699 if (S_ISCHR(inode
->i_mode
) || S_ISBLK(inode
->i_mode
)) {
3700 if (old_valid_dev(inode
->i_rdev
)) {
3701 raw_inode
->i_block
[0] =
3702 cpu_to_le32(old_encode_dev(inode
->i_rdev
));
3703 raw_inode
->i_block
[1] = 0;
3705 raw_inode
->i_block
[0] = 0;
3706 raw_inode
->i_block
[1] =
3707 cpu_to_le32(new_encode_dev(inode
->i_rdev
));
3708 raw_inode
->i_block
[2] = 0;
3711 for (block
= 0; block
< EXT4_N_BLOCKS
; block
++)
3712 raw_inode
->i_block
[block
] = ei
->i_data
[block
];
3714 raw_inode
->i_disk_version
= cpu_to_le32(inode
->i_version
);
3715 if (ei
->i_extra_isize
) {
3716 if (EXT4_FITS_IN_INODE(raw_inode
, ei
, i_version_hi
))
3717 raw_inode
->i_version_hi
=
3718 cpu_to_le32(inode
->i_version
>> 32);
3719 raw_inode
->i_extra_isize
= cpu_to_le16(ei
->i_extra_isize
);
3722 BUFFER_TRACE(bh
, "call ext4_handle_dirty_metadata");
3723 rc
= ext4_handle_dirty_metadata(handle
, NULL
, bh
);
3726 ext4_clear_inode_state(inode
, EXT4_STATE_NEW
);
3728 ext4_update_inode_fsync_trans(handle
, inode
, 0);
3731 ext4_std_error(inode
->i_sb
, err
);
3736 * ext4_write_inode()
3738 * We are called from a few places:
3740 * - Within generic_file_write() for O_SYNC files.
3741 * Here, there will be no transaction running. We wait for any running
3742 * trasnaction to commit.
3744 * - Within sys_sync(), kupdate and such.
3745 * We wait on commit, if tol to.
3747 * - Within prune_icache() (PF_MEMALLOC == true)
3748 * Here we simply return. We can't afford to block kswapd on the
3751 * In all cases it is actually safe for us to return without doing anything,
3752 * because the inode has been copied into a raw inode buffer in
3753 * ext4_mark_inode_dirty(). This is a correctness thing for O_SYNC and for
3756 * Note that we are absolutely dependent upon all inode dirtiers doing the
3757 * right thing: they *must* call mark_inode_dirty() after dirtying info in
3758 * which we are interested.
3760 * It would be a bug for them to not do this. The code:
3762 * mark_inode_dirty(inode)
3764 * inode->i_size = expr;
3766 * is in error because a kswapd-driven write_inode() could occur while
3767 * `stuff()' is running, and the new i_size will be lost. Plus the inode
3768 * will no longer be on the superblock's dirty inode list.
3770 int ext4_write_inode(struct inode
*inode
, struct writeback_control
*wbc
)
3774 if (current
->flags
& PF_MEMALLOC
)
3777 if (EXT4_SB(inode
->i_sb
)->s_journal
) {
3778 if (ext4_journal_current_handle()) {
3779 jbd_debug(1, "called recursively, non-PF_MEMALLOC!\n");
3784 if (wbc
->sync_mode
!= WB_SYNC_ALL
)
3787 err
= ext4_force_commit(inode
->i_sb
);
3789 struct ext4_iloc iloc
;
3791 err
= __ext4_get_inode_loc(inode
, &iloc
, 0);
3794 if (wbc
->sync_mode
== WB_SYNC_ALL
)
3795 sync_dirty_buffer(iloc
.bh
);
3796 if (buffer_req(iloc
.bh
) && !buffer_uptodate(iloc
.bh
)) {
3797 EXT4_ERROR_INODE_BLOCK(inode
, iloc
.bh
->b_blocknr
,
3798 "IO error syncing inode");
3809 * Called from notify_change.
3811 * We want to trap VFS attempts to truncate the file as soon as
3812 * possible. In particular, we want to make sure that when the VFS
3813 * shrinks i_size, we put the inode on the orphan list and modify
3814 * i_disksize immediately, so that during the subsequent flushing of
3815 * dirty pages and freeing of disk blocks, we can guarantee that any
3816 * commit will leave the blocks being flushed in an unused state on
3817 * disk. (On recovery, the inode will get truncated and the blocks will
3818 * be freed, so we have a strong guarantee that no future commit will
3819 * leave these blocks visible to the user.)
3821 * Another thing we have to assure is that if we are in ordered mode
3822 * and inode is still attached to the committing transaction, we must
3823 * we start writeout of all the dirty pages which are being truncated.
3824 * This way we are sure that all the data written in the previous
3825 * transaction are already on disk (truncate waits for pages under
3828 * Called with inode->i_mutex down.
3830 int ext4_setattr(struct dentry
*dentry
, struct iattr
*attr
)
3832 struct inode
*inode
= dentry
->d_inode
;
3835 const unsigned int ia_valid
= attr
->ia_valid
;
3837 error
= inode_change_ok(inode
, attr
);
3841 if (is_quota_modification(inode
, attr
))
3842 dquot_initialize(inode
);
3843 if ((ia_valid
& ATTR_UID
&& attr
->ia_uid
!= inode
->i_uid
) ||
3844 (ia_valid
& ATTR_GID
&& attr
->ia_gid
!= inode
->i_gid
)) {
3847 /* (user+group)*(old+new) structure, inode write (sb,
3848 * inode block, ? - but truncate inode update has it) */
3849 handle
= ext4_journal_start(inode
, (EXT4_MAXQUOTAS_INIT_BLOCKS(inode
->i_sb
)+
3850 EXT4_MAXQUOTAS_DEL_BLOCKS(inode
->i_sb
))+3);
3851 if (IS_ERR(handle
)) {
3852 error
= PTR_ERR(handle
);
3855 error
= dquot_transfer(inode
, attr
);
3857 ext4_journal_stop(handle
);
3860 /* Update corresponding info in inode so that everything is in
3861 * one transaction */
3862 if (attr
->ia_valid
& ATTR_UID
)
3863 inode
->i_uid
= attr
->ia_uid
;
3864 if (attr
->ia_valid
& ATTR_GID
)
3865 inode
->i_gid
= attr
->ia_gid
;
3866 error
= ext4_mark_inode_dirty(handle
, inode
);
3867 ext4_journal_stop(handle
);
3870 if (attr
->ia_valid
& ATTR_SIZE
) {
3871 inode_dio_wait(inode
);
3873 if (!(ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
))) {
3874 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
3876 if (attr
->ia_size
> sbi
->s_bitmap_maxbytes
)
3881 if (S_ISREG(inode
->i_mode
) &&
3882 attr
->ia_valid
& ATTR_SIZE
&&
3883 (attr
->ia_size
< inode
->i_size
)) {
3886 handle
= ext4_journal_start(inode
, 3);
3887 if (IS_ERR(handle
)) {
3888 error
= PTR_ERR(handle
);
3891 if (ext4_handle_valid(handle
)) {
3892 error
= ext4_orphan_add(handle
, inode
);
3895 EXT4_I(inode
)->i_disksize
= attr
->ia_size
;
3896 rc
= ext4_mark_inode_dirty(handle
, inode
);
3899 ext4_journal_stop(handle
);
3901 if (ext4_should_order_data(inode
)) {
3902 error
= ext4_begin_ordered_truncate(inode
,
3905 /* Do as much error cleanup as possible */
3906 handle
= ext4_journal_start(inode
, 3);
3907 if (IS_ERR(handle
)) {
3908 ext4_orphan_del(NULL
, inode
);
3911 ext4_orphan_del(handle
, inode
);
3913 ext4_journal_stop(handle
);
3919 if (attr
->ia_valid
& ATTR_SIZE
) {
3920 if (attr
->ia_size
!= i_size_read(inode
)) {
3921 truncate_setsize(inode
, attr
->ia_size
);
3922 ext4_truncate(inode
);
3923 } else if (ext4_test_inode_flag(inode
, EXT4_INODE_EOFBLOCKS
))
3924 ext4_truncate(inode
);
3928 setattr_copy(inode
, attr
);
3929 mark_inode_dirty(inode
);
3933 * If the call to ext4_truncate failed to get a transaction handle at
3934 * all, we need to clean up the in-core orphan list manually.
3936 if (orphan
&& inode
->i_nlink
)
3937 ext4_orphan_del(NULL
, inode
);
3939 if (!rc
&& (ia_valid
& ATTR_MODE
))
3940 rc
= ext4_acl_chmod(inode
);
3943 ext4_std_error(inode
->i_sb
, error
);
3949 int ext4_getattr(struct vfsmount
*mnt
, struct dentry
*dentry
,
3952 struct inode
*inode
;
3953 unsigned long delalloc_blocks
;
3955 inode
= dentry
->d_inode
;
3956 generic_fillattr(inode
, stat
);
3959 * We can't update i_blocks if the block allocation is delayed
3960 * otherwise in the case of system crash before the real block
3961 * allocation is done, we will have i_blocks inconsistent with
3962 * on-disk file blocks.
3963 * We always keep i_blocks updated together with real
3964 * allocation. But to not confuse with user, stat
3965 * will return the blocks that include the delayed allocation
3966 * blocks for this file.
3968 delalloc_blocks
= EXT4_I(inode
)->i_reserved_data_blocks
;
3970 stat
->blocks
+= (delalloc_blocks
<< inode
->i_sb
->s_blocksize_bits
)>>9;
3974 static int ext4_index_trans_blocks(struct inode
*inode
, int nrblocks
, int chunk
)
3976 if (!(ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
)))
3977 return ext4_ind_trans_blocks(inode
, nrblocks
, chunk
);
3978 return ext4_ext_index_trans_blocks(inode
, nrblocks
, chunk
);
3982 * Account for index blocks, block groups bitmaps and block group
3983 * descriptor blocks if modify datablocks and index blocks
3984 * worse case, the indexs blocks spread over different block groups
3986 * If datablocks are discontiguous, they are possible to spread over
3987 * different block groups too. If they are contiuguous, with flexbg,
3988 * they could still across block group boundary.
3990 * Also account for superblock, inode, quota and xattr blocks
3992 static int ext4_meta_trans_blocks(struct inode
*inode
, int nrblocks
, int chunk
)
3994 ext4_group_t groups
, ngroups
= ext4_get_groups_count(inode
->i_sb
);
4000 * How many index blocks need to touch to modify nrblocks?
4001 * The "Chunk" flag indicating whether the nrblocks is
4002 * physically contiguous on disk
4004 * For Direct IO and fallocate, they calls get_block to allocate
4005 * one single extent at a time, so they could set the "Chunk" flag
4007 idxblocks
= ext4_index_trans_blocks(inode
, nrblocks
, chunk
);
4012 * Now let's see how many group bitmaps and group descriptors need
4022 if (groups
> ngroups
)
4024 if (groups
> EXT4_SB(inode
->i_sb
)->s_gdb_count
)
4025 gdpblocks
= EXT4_SB(inode
->i_sb
)->s_gdb_count
;
4027 /* bitmaps and block group descriptor blocks */
4028 ret
+= groups
+ gdpblocks
;
4030 /* Blocks for super block, inode, quota and xattr blocks */
4031 ret
+= EXT4_META_TRANS_BLOCKS(inode
->i_sb
);
4037 * Calculate the total number of credits to reserve to fit
4038 * the modification of a single pages into a single transaction,
4039 * which may include multiple chunks of block allocations.
4041 * This could be called via ext4_write_begin()
4043 * We need to consider the worse case, when
4044 * one new block per extent.
4046 int ext4_writepage_trans_blocks(struct inode
*inode
)
4048 int bpp
= ext4_journal_blocks_per_page(inode
);
4051 ret
= ext4_meta_trans_blocks(inode
, bpp
, 0);
4053 /* Account for data blocks for journalled mode */
4054 if (ext4_should_journal_data(inode
))
4060 * Calculate the journal credits for a chunk of data modification.
4062 * This is called from DIO, fallocate or whoever calling
4063 * ext4_map_blocks() to map/allocate a chunk of contiguous disk blocks.
4065 * journal buffers for data blocks are not included here, as DIO
4066 * and fallocate do no need to journal data buffers.
4068 int ext4_chunk_trans_blocks(struct inode
*inode
, int nrblocks
)
4070 return ext4_meta_trans_blocks(inode
, nrblocks
, 1);
4074 * The caller must have previously called ext4_reserve_inode_write().
4075 * Give this, we know that the caller already has write access to iloc->bh.
4077 int ext4_mark_iloc_dirty(handle_t
*handle
,
4078 struct inode
*inode
, struct ext4_iloc
*iloc
)
4082 if (test_opt(inode
->i_sb
, I_VERSION
))
4083 inode_inc_iversion(inode
);
4085 /* the do_update_inode consumes one bh->b_count */
4088 /* ext4_do_update_inode() does jbd2_journal_dirty_metadata */
4089 err
= ext4_do_update_inode(handle
, inode
, iloc
);
4095 * On success, We end up with an outstanding reference count against
4096 * iloc->bh. This _must_ be cleaned up later.
4100 ext4_reserve_inode_write(handle_t
*handle
, struct inode
*inode
,
4101 struct ext4_iloc
*iloc
)
4105 err
= ext4_get_inode_loc(inode
, iloc
);
4107 BUFFER_TRACE(iloc
->bh
, "get_write_access");
4108 err
= ext4_journal_get_write_access(handle
, iloc
->bh
);
4114 ext4_std_error(inode
->i_sb
, err
);
4119 * Expand an inode by new_extra_isize bytes.
4120 * Returns 0 on success or negative error number on failure.
4122 static int ext4_expand_extra_isize(struct inode
*inode
,
4123 unsigned int new_extra_isize
,
4124 struct ext4_iloc iloc
,
4127 struct ext4_inode
*raw_inode
;
4128 struct ext4_xattr_ibody_header
*header
;
4130 if (EXT4_I(inode
)->i_extra_isize
>= new_extra_isize
)
4133 raw_inode
= ext4_raw_inode(&iloc
);
4135 header
= IHDR(inode
, raw_inode
);
4137 /* No extended attributes present */
4138 if (!ext4_test_inode_state(inode
, EXT4_STATE_XATTR
) ||
4139 header
->h_magic
!= cpu_to_le32(EXT4_XATTR_MAGIC
)) {
4140 memset((void *)raw_inode
+ EXT4_GOOD_OLD_INODE_SIZE
, 0,
4142 EXT4_I(inode
)->i_extra_isize
= new_extra_isize
;
4146 /* try to expand with EAs present */
4147 return ext4_expand_extra_isize_ea(inode
, new_extra_isize
,
4152 * What we do here is to mark the in-core inode as clean with respect to inode
4153 * dirtiness (it may still be data-dirty).
4154 * This means that the in-core inode may be reaped by prune_icache
4155 * without having to perform any I/O. This is a very good thing,
4156 * because *any* task may call prune_icache - even ones which
4157 * have a transaction open against a different journal.
4159 * Is this cheating? Not really. Sure, we haven't written the
4160 * inode out, but prune_icache isn't a user-visible syncing function.
4161 * Whenever the user wants stuff synced (sys_sync, sys_msync, sys_fsync)
4162 * we start and wait on commits.
4164 * Is this efficient/effective? Well, we're being nice to the system
4165 * by cleaning up our inodes proactively so they can be reaped
4166 * without I/O. But we are potentially leaving up to five seconds'
4167 * worth of inodes floating about which prune_icache wants us to
4168 * write out. One way to fix that would be to get prune_icache()
4169 * to do a write_super() to free up some memory. It has the desired
4172 int ext4_mark_inode_dirty(handle_t
*handle
, struct inode
*inode
)
4174 struct ext4_iloc iloc
;
4175 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
4176 static unsigned int mnt_count
;
4180 trace_ext4_mark_inode_dirty(inode
, _RET_IP_
);
4181 err
= ext4_reserve_inode_write(handle
, inode
, &iloc
);
4182 if (ext4_handle_valid(handle
) &&
4183 EXT4_I(inode
)->i_extra_isize
< sbi
->s_want_extra_isize
&&
4184 !ext4_test_inode_state(inode
, EXT4_STATE_NO_EXPAND
)) {
4186 * We need extra buffer credits since we may write into EA block
4187 * with this same handle. If journal_extend fails, then it will
4188 * only result in a minor loss of functionality for that inode.
4189 * If this is felt to be critical, then e2fsck should be run to
4190 * force a large enough s_min_extra_isize.
4192 if ((jbd2_journal_extend(handle
,
4193 EXT4_DATA_TRANS_BLOCKS(inode
->i_sb
))) == 0) {
4194 ret
= ext4_expand_extra_isize(inode
,
4195 sbi
->s_want_extra_isize
,
4198 ext4_set_inode_state(inode
,
4199 EXT4_STATE_NO_EXPAND
);
4201 le16_to_cpu(sbi
->s_es
->s_mnt_count
)) {
4202 ext4_warning(inode
->i_sb
,
4203 "Unable to expand inode %lu. Delete"
4204 " some EAs or run e2fsck.",
4207 le16_to_cpu(sbi
->s_es
->s_mnt_count
);
4213 err
= ext4_mark_iloc_dirty(handle
, inode
, &iloc
);
4218 * ext4_dirty_inode() is called from __mark_inode_dirty()
4220 * We're really interested in the case where a file is being extended.
4221 * i_size has been changed by generic_commit_write() and we thus need
4222 * to include the updated inode in the current transaction.
4224 * Also, dquot_alloc_block() will always dirty the inode when blocks
4225 * are allocated to the file.
4227 * If the inode is marked synchronous, we don't honour that here - doing
4228 * so would cause a commit on atime updates, which we don't bother doing.
4229 * We handle synchronous inodes at the highest possible level.
4231 void ext4_dirty_inode(struct inode
*inode
, int flags
)
4235 handle
= ext4_journal_start(inode
, 2);
4239 ext4_mark_inode_dirty(handle
, inode
);
4241 ext4_journal_stop(handle
);
4248 * Bind an inode's backing buffer_head into this transaction, to prevent
4249 * it from being flushed to disk early. Unlike
4250 * ext4_reserve_inode_write, this leaves behind no bh reference and
4251 * returns no iloc structure, so the caller needs to repeat the iloc
4252 * lookup to mark the inode dirty later.
4254 static int ext4_pin_inode(handle_t
*handle
, struct inode
*inode
)
4256 struct ext4_iloc iloc
;
4260 err
= ext4_get_inode_loc(inode
, &iloc
);
4262 BUFFER_TRACE(iloc
.bh
, "get_write_access");
4263 err
= jbd2_journal_get_write_access(handle
, iloc
.bh
);
4265 err
= ext4_handle_dirty_metadata(handle
,
4271 ext4_std_error(inode
->i_sb
, err
);
4276 int ext4_change_inode_journal_flag(struct inode
*inode
, int val
)
4283 * We have to be very careful here: changing a data block's
4284 * journaling status dynamically is dangerous. If we write a
4285 * data block to the journal, change the status and then delete
4286 * that block, we risk forgetting to revoke the old log record
4287 * from the journal and so a subsequent replay can corrupt data.
4288 * So, first we make sure that the journal is empty and that
4289 * nobody is changing anything.
4292 journal
= EXT4_JOURNAL(inode
);
4295 if (is_journal_aborted(journal
))
4298 jbd2_journal_lock_updates(journal
);
4299 jbd2_journal_flush(journal
);
4302 * OK, there are no updates running now, and all cached data is
4303 * synced to disk. We are now in a completely consistent state
4304 * which doesn't have anything in the journal, and we know that
4305 * no filesystem updates are running, so it is safe to modify
4306 * the inode's in-core data-journaling state flag now.
4310 ext4_set_inode_flag(inode
, EXT4_INODE_JOURNAL_DATA
);
4312 ext4_clear_inode_flag(inode
, EXT4_INODE_JOURNAL_DATA
);
4313 ext4_set_aops(inode
);
4315 jbd2_journal_unlock_updates(journal
);
4317 /* Finally we can mark the inode as dirty. */
4319 handle
= ext4_journal_start(inode
, 1);
4321 return PTR_ERR(handle
);
4323 err
= ext4_mark_inode_dirty(handle
, inode
);
4324 ext4_handle_sync(handle
);
4325 ext4_journal_stop(handle
);
4326 ext4_std_error(inode
->i_sb
, err
);
4331 static int ext4_bh_unmapped(handle_t
*handle
, struct buffer_head
*bh
)
4333 return !buffer_mapped(bh
);
4336 int ext4_page_mkwrite(struct vm_area_struct
*vma
, struct vm_fault
*vmf
)
4338 struct page
*page
= vmf
->page
;
4342 struct file
*file
= vma
->vm_file
;
4343 struct inode
*inode
= file
->f_path
.dentry
->d_inode
;
4344 struct address_space
*mapping
= inode
->i_mapping
;
4346 get_block_t
*get_block
;
4350 * This check is racy but catches the common case. We rely on
4351 * __block_page_mkwrite() to do a reliable check.
4353 vfs_check_frozen(inode
->i_sb
, SB_FREEZE_WRITE
);
4354 /* Delalloc case is easy... */
4355 if (test_opt(inode
->i_sb
, DELALLOC
) &&
4356 !ext4_should_journal_data(inode
) &&
4357 !ext4_nonda_switch(inode
->i_sb
)) {
4359 ret
= __block_page_mkwrite(vma
, vmf
,
4360 ext4_da_get_block_prep
);
4361 } while (ret
== -ENOSPC
&&
4362 ext4_should_retry_alloc(inode
->i_sb
, &retries
));
4367 size
= i_size_read(inode
);
4368 /* Page got truncated from under us? */
4369 if (page
->mapping
!= mapping
|| page_offset(page
) > size
) {
4371 ret
= VM_FAULT_NOPAGE
;
4375 if (page
->index
== size
>> PAGE_CACHE_SHIFT
)
4376 len
= size
& ~PAGE_CACHE_MASK
;
4378 len
= PAGE_CACHE_SIZE
;
4380 * Return if we have all the buffers mapped. This avoids the need to do
4381 * journal_start/journal_stop which can block and take a long time
4383 if (page_has_buffers(page
)) {
4384 if (!walk_page_buffers(NULL
, page_buffers(page
), 0, len
, NULL
,
4385 ext4_bh_unmapped
)) {
4386 /* Wait so that we don't change page under IO */
4387 wait_on_page_writeback(page
);
4388 ret
= VM_FAULT_LOCKED
;
4393 /* OK, we need to fill the hole... */
4394 if (ext4_should_dioread_nolock(inode
))
4395 get_block
= ext4_get_block_write
;
4397 get_block
= ext4_get_block
;
4399 handle
= ext4_journal_start(inode
, ext4_writepage_trans_blocks(inode
));
4400 if (IS_ERR(handle
)) {
4401 ret
= VM_FAULT_SIGBUS
;
4404 ret
= __block_page_mkwrite(vma
, vmf
, get_block
);
4405 if (!ret
&& ext4_should_journal_data(inode
)) {
4406 if (walk_page_buffers(handle
, page_buffers(page
), 0,
4407 PAGE_CACHE_SIZE
, NULL
, do_journal_get_write_access
)) {
4409 ret
= VM_FAULT_SIGBUS
;
4412 ext4_set_inode_state(inode
, EXT4_STATE_JDATA
);
4414 ext4_journal_stop(handle
);
4415 if (ret
== -ENOSPC
&& ext4_should_retry_alloc(inode
->i_sb
, &retries
))
4418 ret
= block_page_mkwrite_return(ret
);