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 ext4_ioend_wait(inode
);
126 if (inode
->i_nlink
) {
128 * When journalling data dirty buffers are tracked only in the
129 * journal. So although mm thinks everything is clean and
130 * ready for reaping the inode might still have some pages to
131 * write in the running transaction or waiting to be
132 * checkpointed. Thus calling jbd2_journal_invalidatepage()
133 * (via truncate_inode_pages()) to discard these buffers can
134 * cause data loss. Also even if we did not discard these
135 * buffers, we would have no way to find them after the inode
136 * is reaped and thus user could see stale data if he tries to
137 * read them before the transaction is checkpointed. So be
138 * careful and force everything to disk here... We use
139 * ei->i_datasync_tid to store the newest transaction
140 * containing inode's data.
142 * Note that directories do not have this problem because they
143 * don't use page cache.
145 if (ext4_should_journal_data(inode
) &&
146 (S_ISLNK(inode
->i_mode
) || S_ISREG(inode
->i_mode
))) {
147 journal_t
*journal
= EXT4_SB(inode
->i_sb
)->s_journal
;
148 tid_t commit_tid
= EXT4_I(inode
)->i_datasync_tid
;
150 jbd2_log_start_commit(journal
, commit_tid
);
151 jbd2_log_wait_commit(journal
, commit_tid
);
152 filemap_write_and_wait(&inode
->i_data
);
154 truncate_inode_pages(&inode
->i_data
, 0);
158 if (!is_bad_inode(inode
))
159 dquot_initialize(inode
);
161 if (ext4_should_order_data(inode
))
162 ext4_begin_ordered_truncate(inode
, 0);
163 truncate_inode_pages(&inode
->i_data
, 0);
165 if (is_bad_inode(inode
))
168 handle
= ext4_journal_start(inode
, ext4_blocks_for_truncate(inode
)+3);
169 if (IS_ERR(handle
)) {
170 ext4_std_error(inode
->i_sb
, PTR_ERR(handle
));
172 * If we're going to skip the normal cleanup, we still need to
173 * make sure that the in-core orphan linked list is properly
176 ext4_orphan_del(NULL
, inode
);
181 ext4_handle_sync(handle
);
183 err
= ext4_mark_inode_dirty(handle
, inode
);
185 ext4_warning(inode
->i_sb
,
186 "couldn't mark inode dirty (err %d)", err
);
190 ext4_truncate(inode
);
193 * ext4_ext_truncate() doesn't reserve any slop when it
194 * restarts journal transactions; therefore there may not be
195 * enough credits left in the handle to remove the inode from
196 * the orphan list and set the dtime field.
198 if (!ext4_handle_has_enough_credits(handle
, 3)) {
199 err
= ext4_journal_extend(handle
, 3);
201 err
= ext4_journal_restart(handle
, 3);
203 ext4_warning(inode
->i_sb
,
204 "couldn't extend journal (err %d)", err
);
206 ext4_journal_stop(handle
);
207 ext4_orphan_del(NULL
, inode
);
213 * Kill off the orphan record which ext4_truncate created.
214 * AKPM: I think this can be inside the above `if'.
215 * Note that ext4_orphan_del() has to be able to cope with the
216 * deletion of a non-existent orphan - this is because we don't
217 * know if ext4_truncate() actually created an orphan record.
218 * (Well, we could do this if we need to, but heck - it works)
220 ext4_orphan_del(handle
, inode
);
221 EXT4_I(inode
)->i_dtime
= get_seconds();
224 * One subtle ordering requirement: if anything has gone wrong
225 * (transaction abort, IO errors, whatever), then we can still
226 * do these next steps (the fs will already have been marked as
227 * having errors), but we can't free the inode if the mark_dirty
230 if (ext4_mark_inode_dirty(handle
, inode
))
231 /* If that failed, just do the required in-core inode clear. */
232 ext4_clear_inode(inode
);
234 ext4_free_inode(handle
, inode
);
235 ext4_journal_stop(handle
);
238 ext4_clear_inode(inode
); /* We must guarantee clearing of inode... */
242 qsize_t
*ext4_get_reserved_space(struct inode
*inode
)
244 return &EXT4_I(inode
)->i_reserved_quota
;
249 * Calculate the number of metadata blocks need to reserve
250 * to allocate a block located at @lblock
252 static int ext4_calc_metadata_amount(struct inode
*inode
, ext4_lblk_t lblock
)
254 if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
))
255 return ext4_ext_calc_metadata_amount(inode
, lblock
);
257 return ext4_ind_calc_metadata_amount(inode
, lblock
);
261 * Called with i_data_sem down, which is important since we can call
262 * ext4_discard_preallocations() from here.
264 void ext4_da_update_reserve_space(struct inode
*inode
,
265 int used
, int quota_claim
)
267 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
268 struct ext4_inode_info
*ei
= EXT4_I(inode
);
270 spin_lock(&ei
->i_block_reservation_lock
);
271 trace_ext4_da_update_reserve_space(inode
, used
);
272 if (unlikely(used
> ei
->i_reserved_data_blocks
)) {
273 ext4_msg(inode
->i_sb
, KERN_NOTICE
, "%s: ino %lu, used %d "
274 "with only %d reserved data blocks\n",
275 __func__
, inode
->i_ino
, used
,
276 ei
->i_reserved_data_blocks
);
278 used
= ei
->i_reserved_data_blocks
;
281 /* Update per-inode reservations */
282 ei
->i_reserved_data_blocks
-= used
;
283 ei
->i_reserved_meta_blocks
-= ei
->i_allocated_meta_blocks
;
284 percpu_counter_sub(&sbi
->s_dirtyblocks_counter
,
285 used
+ ei
->i_allocated_meta_blocks
);
286 ei
->i_allocated_meta_blocks
= 0;
288 if (ei
->i_reserved_data_blocks
== 0) {
290 * We can release all of the reserved metadata blocks
291 * only when we have written all of the delayed
294 percpu_counter_sub(&sbi
->s_dirtyblocks_counter
,
295 ei
->i_reserved_meta_blocks
);
296 ei
->i_reserved_meta_blocks
= 0;
297 ei
->i_da_metadata_calc_len
= 0;
299 spin_unlock(&EXT4_I(inode
)->i_block_reservation_lock
);
301 /* Update quota subsystem for data blocks */
303 dquot_claim_block(inode
, used
);
306 * We did fallocate with an offset that is already delayed
307 * allocated. So on delayed allocated writeback we should
308 * not re-claim the quota for fallocated blocks.
310 dquot_release_reservation_block(inode
, used
);
314 * If we have done all the pending block allocations and if
315 * there aren't any writers on the inode, we can discard the
316 * inode's preallocations.
318 if ((ei
->i_reserved_data_blocks
== 0) &&
319 (atomic_read(&inode
->i_writecount
) == 0))
320 ext4_discard_preallocations(inode
);
323 static int __check_block_validity(struct inode
*inode
, const char *func
,
325 struct ext4_map_blocks
*map
)
327 if (!ext4_data_block_valid(EXT4_SB(inode
->i_sb
), map
->m_pblk
,
329 ext4_error_inode(inode
, func
, line
, map
->m_pblk
,
330 "lblock %lu mapped to illegal pblock "
331 "(length %d)", (unsigned long) map
->m_lblk
,
338 #define check_block_validity(inode, map) \
339 __check_block_validity((inode), __func__, __LINE__, (map))
342 * Return the number of contiguous dirty pages in a given inode
343 * starting at page frame idx.
345 static pgoff_t
ext4_num_dirty_pages(struct inode
*inode
, pgoff_t idx
,
346 unsigned int max_pages
)
348 struct address_space
*mapping
= inode
->i_mapping
;
352 int i
, nr_pages
, done
= 0;
356 pagevec_init(&pvec
, 0);
359 nr_pages
= pagevec_lookup_tag(&pvec
, mapping
, &index
,
361 (pgoff_t
)PAGEVEC_SIZE
);
364 for (i
= 0; i
< nr_pages
; i
++) {
365 struct page
*page
= pvec
.pages
[i
];
366 struct buffer_head
*bh
, *head
;
369 if (unlikely(page
->mapping
!= mapping
) ||
371 PageWriteback(page
) ||
372 page
->index
!= idx
) {
377 if (page_has_buffers(page
)) {
378 bh
= head
= page_buffers(page
);
380 if (!buffer_delay(bh
) &&
381 !buffer_unwritten(bh
))
383 bh
= bh
->b_this_page
;
384 } while (!done
&& (bh
!= head
));
391 if (num
>= max_pages
) {
396 pagevec_release(&pvec
);
402 * The ext4_map_blocks() function tries to look up the requested blocks,
403 * and returns if the blocks are already mapped.
405 * Otherwise it takes the write lock of the i_data_sem and allocate blocks
406 * and store the allocated blocks in the result buffer head and mark it
409 * If file type is extents based, it will call ext4_ext_map_blocks(),
410 * Otherwise, call with ext4_ind_map_blocks() to handle indirect mapping
413 * On success, it returns the number of blocks being mapped or allocate.
414 * if create==0 and the blocks are pre-allocated and uninitialized block,
415 * the result buffer head is unmapped. If the create ==1, it will make sure
416 * the buffer head is mapped.
418 * It returns 0 if plain look up failed (blocks have not been allocated), in
419 * that casem, buffer head is unmapped
421 * It returns the error in case of allocation failure.
423 int ext4_map_blocks(handle_t
*handle
, struct inode
*inode
,
424 struct ext4_map_blocks
*map
, int flags
)
429 ext_debug("ext4_map_blocks(): inode %lu, flag %d, max_blocks %u,"
430 "logical block %lu\n", inode
->i_ino
, flags
, map
->m_len
,
431 (unsigned long) map
->m_lblk
);
433 * Try to see if we can get the block without requesting a new
436 down_read((&EXT4_I(inode
)->i_data_sem
));
437 if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
)) {
438 retval
= ext4_ext_map_blocks(handle
, inode
, map
, 0);
440 retval
= ext4_ind_map_blocks(handle
, inode
, map
, 0);
442 up_read((&EXT4_I(inode
)->i_data_sem
));
444 if (retval
> 0 && map
->m_flags
& EXT4_MAP_MAPPED
) {
445 int ret
= check_block_validity(inode
, map
);
450 /* If it is only a block(s) look up */
451 if ((flags
& EXT4_GET_BLOCKS_CREATE
) == 0)
455 * Returns if the blocks have already allocated
457 * Note that if blocks have been preallocated
458 * ext4_ext_get_block() returns th create = 0
459 * with buffer head unmapped.
461 if (retval
> 0 && map
->m_flags
& EXT4_MAP_MAPPED
)
465 * When we call get_blocks without the create flag, the
466 * BH_Unwritten flag could have gotten set if the blocks
467 * requested were part of a uninitialized extent. We need to
468 * clear this flag now that we are committed to convert all or
469 * part of the uninitialized extent to be an initialized
470 * extent. This is because we need to avoid the combination
471 * of BH_Unwritten and BH_Mapped flags being simultaneously
472 * set on the buffer_head.
474 map
->m_flags
&= ~EXT4_MAP_UNWRITTEN
;
477 * New blocks allocate and/or writing to uninitialized extent
478 * will possibly result in updating i_data, so we take
479 * the write lock of i_data_sem, and call get_blocks()
480 * with create == 1 flag.
482 down_write((&EXT4_I(inode
)->i_data_sem
));
485 * if the caller is from delayed allocation writeout path
486 * we have already reserved fs blocks for allocation
487 * let the underlying get_block() function know to
488 * avoid double accounting
490 if (flags
& EXT4_GET_BLOCKS_DELALLOC_RESERVE
)
491 ext4_set_inode_state(inode
, EXT4_STATE_DELALLOC_RESERVED
);
493 * We need to check for EXT4 here because migrate
494 * could have changed the inode type in between
496 if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
)) {
497 retval
= ext4_ext_map_blocks(handle
, inode
, map
, flags
);
499 retval
= ext4_ind_map_blocks(handle
, inode
, map
, flags
);
501 if (retval
> 0 && map
->m_flags
& EXT4_MAP_NEW
) {
503 * We allocated new blocks which will result in
504 * i_data's format changing. Force the migrate
505 * to fail by clearing migrate flags
507 ext4_clear_inode_state(inode
, EXT4_STATE_EXT_MIGRATE
);
511 * Update reserved blocks/metadata blocks after successful
512 * block allocation which had been deferred till now. We don't
513 * support fallocate for non extent files. So we can update
514 * reserve space here.
517 (flags
& EXT4_GET_BLOCKS_DELALLOC_RESERVE
))
518 ext4_da_update_reserve_space(inode
, retval
, 1);
520 if (flags
& EXT4_GET_BLOCKS_DELALLOC_RESERVE
)
521 ext4_clear_inode_state(inode
, EXT4_STATE_DELALLOC_RESERVED
);
523 up_write((&EXT4_I(inode
)->i_data_sem
));
524 if (retval
> 0 && map
->m_flags
& EXT4_MAP_MAPPED
) {
525 int ret
= check_block_validity(inode
, map
);
532 /* Maximum number of blocks we map for direct IO at once. */
533 #define DIO_MAX_BLOCKS 4096
535 static int _ext4_get_block(struct inode
*inode
, sector_t iblock
,
536 struct buffer_head
*bh
, int flags
)
538 handle_t
*handle
= ext4_journal_current_handle();
539 struct ext4_map_blocks map
;
540 int ret
= 0, started
= 0;
544 map
.m_len
= bh
->b_size
>> inode
->i_blkbits
;
546 if (flags
&& !handle
) {
547 /* Direct IO write... */
548 if (map
.m_len
> DIO_MAX_BLOCKS
)
549 map
.m_len
= DIO_MAX_BLOCKS
;
550 dio_credits
= ext4_chunk_trans_blocks(inode
, map
.m_len
);
551 handle
= ext4_journal_start(inode
, dio_credits
);
552 if (IS_ERR(handle
)) {
553 ret
= PTR_ERR(handle
);
559 ret
= ext4_map_blocks(handle
, inode
, &map
, flags
);
561 map_bh(bh
, inode
->i_sb
, map
.m_pblk
);
562 bh
->b_state
= (bh
->b_state
& ~EXT4_MAP_FLAGS
) | map
.m_flags
;
563 bh
->b_size
= inode
->i_sb
->s_blocksize
* map
.m_len
;
567 ext4_journal_stop(handle
);
571 int ext4_get_block(struct inode
*inode
, sector_t iblock
,
572 struct buffer_head
*bh
, int create
)
574 return _ext4_get_block(inode
, iblock
, bh
,
575 create
? EXT4_GET_BLOCKS_CREATE
: 0);
579 * `handle' can be NULL if create is zero
581 struct buffer_head
*ext4_getblk(handle_t
*handle
, struct inode
*inode
,
582 ext4_lblk_t block
, int create
, int *errp
)
584 struct ext4_map_blocks map
;
585 struct buffer_head
*bh
;
588 J_ASSERT(handle
!= NULL
|| create
== 0);
592 err
= ext4_map_blocks(handle
, inode
, &map
,
593 create
? EXT4_GET_BLOCKS_CREATE
: 0);
601 bh
= sb_getblk(inode
->i_sb
, map
.m_pblk
);
606 if (map
.m_flags
& EXT4_MAP_NEW
) {
607 J_ASSERT(create
!= 0);
608 J_ASSERT(handle
!= NULL
);
611 * Now that we do not always journal data, we should
612 * keep in mind whether this should always journal the
613 * new buffer as metadata. For now, regular file
614 * writes use ext4_get_block instead, so it's not a
618 BUFFER_TRACE(bh
, "call get_create_access");
619 fatal
= ext4_journal_get_create_access(handle
, bh
);
620 if (!fatal
&& !buffer_uptodate(bh
)) {
621 memset(bh
->b_data
, 0, inode
->i_sb
->s_blocksize
);
622 set_buffer_uptodate(bh
);
625 BUFFER_TRACE(bh
, "call ext4_handle_dirty_metadata");
626 err
= ext4_handle_dirty_metadata(handle
, inode
, bh
);
630 BUFFER_TRACE(bh
, "not a new buffer");
640 struct buffer_head
*ext4_bread(handle_t
*handle
, struct inode
*inode
,
641 ext4_lblk_t block
, int create
, int *err
)
643 struct buffer_head
*bh
;
645 bh
= ext4_getblk(handle
, inode
, block
, create
, err
);
648 if (buffer_uptodate(bh
))
650 ll_rw_block(READ_META
, 1, &bh
);
652 if (buffer_uptodate(bh
))
659 static int walk_page_buffers(handle_t
*handle
,
660 struct buffer_head
*head
,
664 int (*fn
)(handle_t
*handle
,
665 struct buffer_head
*bh
))
667 struct buffer_head
*bh
;
668 unsigned block_start
, block_end
;
669 unsigned blocksize
= head
->b_size
;
671 struct buffer_head
*next
;
673 for (bh
= head
, block_start
= 0;
674 ret
== 0 && (bh
!= head
|| !block_start
);
675 block_start
= block_end
, bh
= next
) {
676 next
= bh
->b_this_page
;
677 block_end
= block_start
+ blocksize
;
678 if (block_end
<= from
|| block_start
>= to
) {
679 if (partial
&& !buffer_uptodate(bh
))
683 err
= (*fn
)(handle
, bh
);
691 * To preserve ordering, it is essential that the hole instantiation and
692 * the data write be encapsulated in a single transaction. We cannot
693 * close off a transaction and start a new one between the ext4_get_block()
694 * and the commit_write(). So doing the jbd2_journal_start at the start of
695 * prepare_write() is the right place.
697 * Also, this function can nest inside ext4_writepage() ->
698 * block_write_full_page(). In that case, we *know* that ext4_writepage()
699 * has generated enough buffer credits to do the whole page. So we won't
700 * block on the journal in that case, which is good, because the caller may
703 * By accident, ext4 can be reentered when a transaction is open via
704 * quota file writes. If we were to commit the transaction while thus
705 * reentered, there can be a deadlock - we would be holding a quota
706 * lock, and the commit would never complete if another thread had a
707 * transaction open and was blocking on the quota lock - a ranking
710 * So what we do is to rely on the fact that jbd2_journal_stop/journal_start
711 * will _not_ run commit under these circumstances because handle->h_ref
712 * is elevated. We'll still have enough credits for the tiny quotafile
715 static int do_journal_get_write_access(handle_t
*handle
,
716 struct buffer_head
*bh
)
718 int dirty
= buffer_dirty(bh
);
721 if (!buffer_mapped(bh
) || buffer_freed(bh
))
724 * __block_write_begin() could have dirtied some buffers. Clean
725 * the dirty bit as jbd2_journal_get_write_access() could complain
726 * otherwise about fs integrity issues. Setting of the dirty bit
727 * by __block_write_begin() isn't a real problem here as we clear
728 * the bit before releasing a page lock and thus writeback cannot
729 * ever write the buffer.
732 clear_buffer_dirty(bh
);
733 ret
= ext4_journal_get_write_access(handle
, bh
);
735 ret
= ext4_handle_dirty_metadata(handle
, NULL
, bh
);
739 static int ext4_get_block_write(struct inode
*inode
, sector_t iblock
,
740 struct buffer_head
*bh_result
, int create
);
741 static int ext4_write_begin(struct file
*file
, struct address_space
*mapping
,
742 loff_t pos
, unsigned len
, unsigned flags
,
743 struct page
**pagep
, void **fsdata
)
745 struct inode
*inode
= mapping
->host
;
746 int ret
, needed_blocks
;
753 trace_ext4_write_begin(inode
, pos
, len
, flags
);
755 * Reserve one block more for addition to orphan list in case
756 * we allocate blocks but write fails for some reason
758 needed_blocks
= ext4_writepage_trans_blocks(inode
) + 1;
759 index
= pos
>> PAGE_CACHE_SHIFT
;
760 from
= pos
& (PAGE_CACHE_SIZE
- 1);
764 handle
= ext4_journal_start(inode
, needed_blocks
);
765 if (IS_ERR(handle
)) {
766 ret
= PTR_ERR(handle
);
770 /* We cannot recurse into the filesystem as the transaction is already
772 flags
|= AOP_FLAG_NOFS
;
774 page
= grab_cache_page_write_begin(mapping
, index
, flags
);
776 ext4_journal_stop(handle
);
782 if (ext4_should_dioread_nolock(inode
))
783 ret
= __block_write_begin(page
, pos
, len
, ext4_get_block_write
);
785 ret
= __block_write_begin(page
, pos
, len
, ext4_get_block
);
787 if (!ret
&& ext4_should_journal_data(inode
)) {
788 ret
= walk_page_buffers(handle
, page_buffers(page
),
789 from
, to
, NULL
, do_journal_get_write_access
);
794 page_cache_release(page
);
796 * __block_write_begin may have instantiated a few blocks
797 * outside i_size. Trim these off again. Don't need
798 * i_size_read because we hold i_mutex.
800 * Add inode to orphan list in case we crash before
803 if (pos
+ len
> inode
->i_size
&& ext4_can_truncate(inode
))
804 ext4_orphan_add(handle
, inode
);
806 ext4_journal_stop(handle
);
807 if (pos
+ len
> inode
->i_size
) {
808 ext4_truncate_failed_write(inode
);
810 * If truncate failed early the inode might
811 * still be on the orphan list; we need to
812 * make sure the inode is removed from the
813 * orphan list in that case.
816 ext4_orphan_del(NULL
, inode
);
820 if (ret
== -ENOSPC
&& ext4_should_retry_alloc(inode
->i_sb
, &retries
))
826 /* For write_end() in data=journal mode */
827 static int write_end_fn(handle_t
*handle
, struct buffer_head
*bh
)
829 if (!buffer_mapped(bh
) || buffer_freed(bh
))
831 set_buffer_uptodate(bh
);
832 return ext4_handle_dirty_metadata(handle
, NULL
, bh
);
835 static int ext4_generic_write_end(struct file
*file
,
836 struct address_space
*mapping
,
837 loff_t pos
, unsigned len
, unsigned copied
,
838 struct page
*page
, void *fsdata
)
840 int i_size_changed
= 0;
841 struct inode
*inode
= mapping
->host
;
842 handle_t
*handle
= ext4_journal_current_handle();
844 copied
= block_write_end(file
, mapping
, pos
, len
, copied
, page
, fsdata
);
847 * No need to use i_size_read() here, the i_size
848 * cannot change under us because we hold i_mutex.
850 * But it's important to update i_size while still holding page lock:
851 * page writeout could otherwise come in and zero beyond i_size.
853 if (pos
+ copied
> inode
->i_size
) {
854 i_size_write(inode
, pos
+ copied
);
858 if (pos
+ copied
> EXT4_I(inode
)->i_disksize
) {
859 /* We need to mark inode dirty even if
860 * new_i_size is less that inode->i_size
861 * bu greater than i_disksize.(hint delalloc)
863 ext4_update_i_disksize(inode
, (pos
+ copied
));
867 page_cache_release(page
);
870 * Don't mark the inode dirty under page lock. First, it unnecessarily
871 * makes the holding time of page lock longer. Second, it forces lock
872 * ordering of page lock and transaction start for journaling
876 ext4_mark_inode_dirty(handle
, inode
);
882 * We need to pick up the new inode size which generic_commit_write gave us
883 * `file' can be NULL - eg, when called from page_symlink().
885 * ext4 never places buffers on inode->i_mapping->private_list. metadata
886 * buffers are managed internally.
888 static int ext4_ordered_write_end(struct file
*file
,
889 struct address_space
*mapping
,
890 loff_t pos
, unsigned len
, unsigned copied
,
891 struct page
*page
, void *fsdata
)
893 handle_t
*handle
= ext4_journal_current_handle();
894 struct inode
*inode
= mapping
->host
;
897 trace_ext4_ordered_write_end(inode
, pos
, len
, copied
);
898 ret
= ext4_jbd2_file_inode(handle
, inode
);
901 ret2
= ext4_generic_write_end(file
, mapping
, pos
, len
, copied
,
904 if (pos
+ len
> inode
->i_size
&& ext4_can_truncate(inode
))
905 /* if we have allocated more blocks and copied
906 * less. We will have blocks allocated outside
907 * inode->i_size. So truncate them
909 ext4_orphan_add(handle
, inode
);
913 ret2
= ext4_journal_stop(handle
);
917 if (pos
+ len
> inode
->i_size
) {
918 ext4_truncate_failed_write(inode
);
920 * If truncate failed early the inode might still be
921 * on the orphan list; we need to make sure the inode
922 * is removed from the orphan list in that case.
925 ext4_orphan_del(NULL
, inode
);
929 return ret
? ret
: copied
;
932 static int ext4_writeback_write_end(struct file
*file
,
933 struct address_space
*mapping
,
934 loff_t pos
, unsigned len
, unsigned copied
,
935 struct page
*page
, void *fsdata
)
937 handle_t
*handle
= ext4_journal_current_handle();
938 struct inode
*inode
= mapping
->host
;
941 trace_ext4_writeback_write_end(inode
, pos
, len
, copied
);
942 ret2
= ext4_generic_write_end(file
, mapping
, pos
, len
, copied
,
945 if (pos
+ len
> inode
->i_size
&& ext4_can_truncate(inode
))
946 /* if we have allocated more blocks and copied
947 * less. We will have blocks allocated outside
948 * inode->i_size. So truncate them
950 ext4_orphan_add(handle
, inode
);
955 ret2
= ext4_journal_stop(handle
);
959 if (pos
+ len
> inode
->i_size
) {
960 ext4_truncate_failed_write(inode
);
962 * If truncate failed early the inode might still be
963 * on the orphan list; we need to make sure the inode
964 * is removed from the orphan list in that case.
967 ext4_orphan_del(NULL
, inode
);
970 return ret
? ret
: copied
;
973 static int ext4_journalled_write_end(struct file
*file
,
974 struct address_space
*mapping
,
975 loff_t pos
, unsigned len
, unsigned copied
,
976 struct page
*page
, void *fsdata
)
978 handle_t
*handle
= ext4_journal_current_handle();
979 struct inode
*inode
= mapping
->host
;
985 trace_ext4_journalled_write_end(inode
, pos
, len
, copied
);
986 from
= pos
& (PAGE_CACHE_SIZE
- 1);
989 BUG_ON(!ext4_handle_valid(handle
));
992 if (!PageUptodate(page
))
994 page_zero_new_buffers(page
, from
+copied
, to
);
997 ret
= walk_page_buffers(handle
, page_buffers(page
), from
,
998 to
, &partial
, write_end_fn
);
1000 SetPageUptodate(page
);
1001 new_i_size
= pos
+ copied
;
1002 if (new_i_size
> inode
->i_size
)
1003 i_size_write(inode
, pos
+copied
);
1004 ext4_set_inode_state(inode
, EXT4_STATE_JDATA
);
1005 EXT4_I(inode
)->i_datasync_tid
= handle
->h_transaction
->t_tid
;
1006 if (new_i_size
> EXT4_I(inode
)->i_disksize
) {
1007 ext4_update_i_disksize(inode
, new_i_size
);
1008 ret2
= ext4_mark_inode_dirty(handle
, inode
);
1014 page_cache_release(page
);
1015 if (pos
+ len
> inode
->i_size
&& ext4_can_truncate(inode
))
1016 /* if we have allocated more blocks and copied
1017 * less. We will have blocks allocated outside
1018 * inode->i_size. So truncate them
1020 ext4_orphan_add(handle
, inode
);
1022 ret2
= ext4_journal_stop(handle
);
1025 if (pos
+ len
> inode
->i_size
) {
1026 ext4_truncate_failed_write(inode
);
1028 * If truncate failed early the inode might still be
1029 * on the orphan list; we need to make sure the inode
1030 * is removed from the orphan list in that case.
1033 ext4_orphan_del(NULL
, inode
);
1036 return ret
? ret
: copied
;
1040 * Reserve a single block located at lblock
1042 static int ext4_da_reserve_space(struct inode
*inode
, ext4_lblk_t lblock
)
1045 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
1046 struct ext4_inode_info
*ei
= EXT4_I(inode
);
1047 unsigned long md_needed
;
1051 * recalculate the amount of metadata blocks to reserve
1052 * in order to allocate nrblocks
1053 * worse case is one extent per block
1056 spin_lock(&ei
->i_block_reservation_lock
);
1057 md_needed
= ext4_calc_metadata_amount(inode
, lblock
);
1058 trace_ext4_da_reserve_space(inode
, md_needed
);
1059 spin_unlock(&ei
->i_block_reservation_lock
);
1062 * We will charge metadata quota at writeout time; this saves
1063 * us from metadata over-estimation, though we may go over by
1064 * a small amount in the end. Here we just reserve for data.
1066 ret
= dquot_reserve_block(inode
, 1);
1070 * We do still charge estimated metadata to the sb though;
1071 * we cannot afford to run out of free blocks.
1073 if (ext4_claim_free_blocks(sbi
, md_needed
+ 1, 0)) {
1074 dquot_release_reservation_block(inode
, 1);
1075 if (ext4_should_retry_alloc(inode
->i_sb
, &retries
)) {
1081 spin_lock(&ei
->i_block_reservation_lock
);
1082 ei
->i_reserved_data_blocks
++;
1083 ei
->i_reserved_meta_blocks
+= md_needed
;
1084 spin_unlock(&ei
->i_block_reservation_lock
);
1086 return 0; /* success */
1089 static void ext4_da_release_space(struct inode
*inode
, int to_free
)
1091 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
1092 struct ext4_inode_info
*ei
= EXT4_I(inode
);
1095 return; /* Nothing to release, exit */
1097 spin_lock(&EXT4_I(inode
)->i_block_reservation_lock
);
1099 trace_ext4_da_release_space(inode
, to_free
);
1100 if (unlikely(to_free
> ei
->i_reserved_data_blocks
)) {
1102 * if there aren't enough reserved blocks, then the
1103 * counter is messed up somewhere. Since this
1104 * function is called from invalidate page, it's
1105 * harmless to return without any action.
1107 ext4_msg(inode
->i_sb
, KERN_NOTICE
, "ext4_da_release_space: "
1108 "ino %lu, to_free %d with only %d reserved "
1109 "data blocks\n", inode
->i_ino
, to_free
,
1110 ei
->i_reserved_data_blocks
);
1112 to_free
= ei
->i_reserved_data_blocks
;
1114 ei
->i_reserved_data_blocks
-= to_free
;
1116 if (ei
->i_reserved_data_blocks
== 0) {
1118 * We can release all of the reserved metadata blocks
1119 * only when we have written all of the delayed
1120 * allocation blocks.
1122 percpu_counter_sub(&sbi
->s_dirtyblocks_counter
,
1123 ei
->i_reserved_meta_blocks
);
1124 ei
->i_reserved_meta_blocks
= 0;
1125 ei
->i_da_metadata_calc_len
= 0;
1128 /* update fs dirty data blocks counter */
1129 percpu_counter_sub(&sbi
->s_dirtyblocks_counter
, to_free
);
1131 spin_unlock(&EXT4_I(inode
)->i_block_reservation_lock
);
1133 dquot_release_reservation_block(inode
, to_free
);
1136 static void ext4_da_page_release_reservation(struct page
*page
,
1137 unsigned long offset
)
1140 struct buffer_head
*head
, *bh
;
1141 unsigned int curr_off
= 0;
1143 head
= page_buffers(page
);
1146 unsigned int next_off
= curr_off
+ bh
->b_size
;
1148 if ((offset
<= curr_off
) && (buffer_delay(bh
))) {
1150 clear_buffer_delay(bh
);
1152 curr_off
= next_off
;
1153 } while ((bh
= bh
->b_this_page
) != head
);
1154 ext4_da_release_space(page
->mapping
->host
, to_release
);
1158 * Delayed allocation stuff
1162 * mpage_da_submit_io - walks through extent of pages and try to write
1163 * them with writepage() call back
1165 * @mpd->inode: inode
1166 * @mpd->first_page: first page of the extent
1167 * @mpd->next_page: page after the last page of the extent
1169 * By the time mpage_da_submit_io() is called we expect all blocks
1170 * to be allocated. this may be wrong if allocation failed.
1172 * As pages are already locked by write_cache_pages(), we can't use it
1174 static int mpage_da_submit_io(struct mpage_da_data
*mpd
,
1175 struct ext4_map_blocks
*map
)
1177 struct pagevec pvec
;
1178 unsigned long index
, end
;
1179 int ret
= 0, err
, nr_pages
, i
;
1180 struct inode
*inode
= mpd
->inode
;
1181 struct address_space
*mapping
= inode
->i_mapping
;
1182 loff_t size
= i_size_read(inode
);
1183 unsigned int len
, block_start
;
1184 struct buffer_head
*bh
, *page_bufs
= NULL
;
1185 int journal_data
= ext4_should_journal_data(inode
);
1186 sector_t pblock
= 0, cur_logical
= 0;
1187 struct ext4_io_submit io_submit
;
1189 BUG_ON(mpd
->next_page
<= mpd
->first_page
);
1190 memset(&io_submit
, 0, sizeof(io_submit
));
1192 * We need to start from the first_page to the next_page - 1
1193 * to make sure we also write the mapped dirty buffer_heads.
1194 * If we look at mpd->b_blocknr we would only be looking
1195 * at the currently mapped buffer_heads.
1197 index
= mpd
->first_page
;
1198 end
= mpd
->next_page
- 1;
1200 pagevec_init(&pvec
, 0);
1201 while (index
<= end
) {
1202 nr_pages
= pagevec_lookup(&pvec
, mapping
, index
, PAGEVEC_SIZE
);
1205 for (i
= 0; i
< nr_pages
; i
++) {
1206 int commit_write
= 0, skip_page
= 0;
1207 struct page
*page
= pvec
.pages
[i
];
1209 index
= page
->index
;
1213 if (index
== size
>> PAGE_CACHE_SHIFT
)
1214 len
= size
& ~PAGE_CACHE_MASK
;
1216 len
= PAGE_CACHE_SIZE
;
1218 cur_logical
= index
<< (PAGE_CACHE_SHIFT
-
1220 pblock
= map
->m_pblk
+ (cur_logical
-
1225 BUG_ON(!PageLocked(page
));
1226 BUG_ON(PageWriteback(page
));
1229 * If the page does not have buffers (for
1230 * whatever reason), try to create them using
1231 * __block_write_begin. If this fails,
1232 * skip the page and move on.
1234 if (!page_has_buffers(page
)) {
1235 if (__block_write_begin(page
, 0, len
,
1236 noalloc_get_block_write
)) {
1244 bh
= page_bufs
= page_buffers(page
);
1249 if (map
&& (cur_logical
>= map
->m_lblk
) &&
1250 (cur_logical
<= (map
->m_lblk
+
1251 (map
->m_len
- 1)))) {
1252 if (buffer_delay(bh
)) {
1253 clear_buffer_delay(bh
);
1254 bh
->b_blocknr
= pblock
;
1256 if (buffer_unwritten(bh
) ||
1258 BUG_ON(bh
->b_blocknr
!= pblock
);
1259 if (map
->m_flags
& EXT4_MAP_UNINIT
)
1260 set_buffer_uninit(bh
);
1261 clear_buffer_unwritten(bh
);
1264 /* skip page if block allocation undone */
1265 if (buffer_delay(bh
) || buffer_unwritten(bh
))
1267 bh
= bh
->b_this_page
;
1268 block_start
+= bh
->b_size
;
1271 } while (bh
!= page_bufs
);
1277 /* mark the buffer_heads as dirty & uptodate */
1278 block_commit_write(page
, 0, len
);
1280 clear_page_dirty_for_io(page
);
1282 * Delalloc doesn't support data journalling,
1283 * but eventually maybe we'll lift this
1286 if (unlikely(journal_data
&& PageChecked(page
)))
1287 err
= __ext4_journalled_writepage(page
, len
);
1288 else if (test_opt(inode
->i_sb
, MBLK_IO_SUBMIT
))
1289 err
= ext4_bio_write_page(&io_submit
, page
,
1291 else if (buffer_uninit(page_bufs
)) {
1292 ext4_set_bh_endio(page_bufs
, inode
);
1293 err
= block_write_full_page_endio(page
,
1294 noalloc_get_block_write
,
1295 mpd
->wbc
, ext4_end_io_buffer_write
);
1297 err
= block_write_full_page(page
,
1298 noalloc_get_block_write
, mpd
->wbc
);
1301 mpd
->pages_written
++;
1303 * In error case, we have to continue because
1304 * remaining pages are still locked
1309 pagevec_release(&pvec
);
1311 ext4_io_submit(&io_submit
);
1315 static void ext4_da_block_invalidatepages(struct mpage_da_data
*mpd
)
1319 struct pagevec pvec
;
1320 struct inode
*inode
= mpd
->inode
;
1321 struct address_space
*mapping
= inode
->i_mapping
;
1323 index
= mpd
->first_page
;
1324 end
= mpd
->next_page
- 1;
1325 while (index
<= end
) {
1326 nr_pages
= pagevec_lookup(&pvec
, mapping
, index
, PAGEVEC_SIZE
);
1329 for (i
= 0; i
< nr_pages
; i
++) {
1330 struct page
*page
= pvec
.pages
[i
];
1331 if (page
->index
> end
)
1333 BUG_ON(!PageLocked(page
));
1334 BUG_ON(PageWriteback(page
));
1335 block_invalidatepage(page
, 0);
1336 ClearPageUptodate(page
);
1339 index
= pvec
.pages
[nr_pages
- 1]->index
+ 1;
1340 pagevec_release(&pvec
);
1345 static void ext4_print_free_blocks(struct inode
*inode
)
1347 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
1348 printk(KERN_CRIT
"Total free blocks count %lld\n",
1349 ext4_count_free_blocks(inode
->i_sb
));
1350 printk(KERN_CRIT
"Free/Dirty block details\n");
1351 printk(KERN_CRIT
"free_blocks=%lld\n",
1352 (long long) percpu_counter_sum(&sbi
->s_freeblocks_counter
));
1353 printk(KERN_CRIT
"dirty_blocks=%lld\n",
1354 (long long) percpu_counter_sum(&sbi
->s_dirtyblocks_counter
));
1355 printk(KERN_CRIT
"Block reservation details\n");
1356 printk(KERN_CRIT
"i_reserved_data_blocks=%u\n",
1357 EXT4_I(inode
)->i_reserved_data_blocks
);
1358 printk(KERN_CRIT
"i_reserved_meta_blocks=%u\n",
1359 EXT4_I(inode
)->i_reserved_meta_blocks
);
1364 * mpage_da_map_and_submit - go through given space, map them
1365 * if necessary, and then submit them for I/O
1367 * @mpd - bh describing space
1369 * The function skips space we know is already mapped to disk blocks.
1372 static void mpage_da_map_and_submit(struct mpage_da_data
*mpd
)
1374 int err
, blks
, get_blocks_flags
;
1375 struct ext4_map_blocks map
, *mapp
= NULL
;
1376 sector_t next
= mpd
->b_blocknr
;
1377 unsigned max_blocks
= mpd
->b_size
>> mpd
->inode
->i_blkbits
;
1378 loff_t disksize
= EXT4_I(mpd
->inode
)->i_disksize
;
1379 handle_t
*handle
= NULL
;
1382 * If the blocks are mapped already, or we couldn't accumulate
1383 * any blocks, then proceed immediately to the submission stage.
1385 if ((mpd
->b_size
== 0) ||
1386 ((mpd
->b_state
& (1 << BH_Mapped
)) &&
1387 !(mpd
->b_state
& (1 << BH_Delay
)) &&
1388 !(mpd
->b_state
& (1 << BH_Unwritten
))))
1391 handle
= ext4_journal_current_handle();
1395 * Call ext4_map_blocks() to allocate any delayed allocation
1396 * blocks, or to convert an uninitialized extent to be
1397 * initialized (in the case where we have written into
1398 * one or more preallocated blocks).
1400 * We pass in the magic EXT4_GET_BLOCKS_DELALLOC_RESERVE to
1401 * indicate that we are on the delayed allocation path. This
1402 * affects functions in many different parts of the allocation
1403 * call path. This flag exists primarily because we don't
1404 * want to change *many* call functions, so ext4_map_blocks()
1405 * will set the EXT4_STATE_DELALLOC_RESERVED flag once the
1406 * inode's allocation semaphore is taken.
1408 * If the blocks in questions were delalloc blocks, set
1409 * EXT4_GET_BLOCKS_DELALLOC_RESERVE so the delalloc accounting
1410 * variables are updated after the blocks have been allocated.
1413 map
.m_len
= max_blocks
;
1414 get_blocks_flags
= EXT4_GET_BLOCKS_CREATE
;
1415 if (ext4_should_dioread_nolock(mpd
->inode
))
1416 get_blocks_flags
|= EXT4_GET_BLOCKS_IO_CREATE_EXT
;
1417 if (mpd
->b_state
& (1 << BH_Delay
))
1418 get_blocks_flags
|= EXT4_GET_BLOCKS_DELALLOC_RESERVE
;
1420 blks
= ext4_map_blocks(handle
, mpd
->inode
, &map
, get_blocks_flags
);
1422 struct super_block
*sb
= mpd
->inode
->i_sb
;
1426 * If get block returns EAGAIN or ENOSPC and there
1427 * appears to be free blocks we will just let
1428 * mpage_da_submit_io() unlock all of the pages.
1433 if (err
== -ENOSPC
&&
1434 ext4_count_free_blocks(sb
)) {
1440 * get block failure will cause us to loop in
1441 * writepages, because a_ops->writepage won't be able
1442 * to make progress. The page will be redirtied by
1443 * writepage and writepages will again try to write
1446 if (!(EXT4_SB(sb
)->s_mount_flags
& EXT4_MF_FS_ABORTED
)) {
1447 ext4_msg(sb
, KERN_CRIT
,
1448 "delayed block allocation failed for inode %lu "
1449 "at logical offset %llu with max blocks %zd "
1450 "with error %d", mpd
->inode
->i_ino
,
1451 (unsigned long long) next
,
1452 mpd
->b_size
>> mpd
->inode
->i_blkbits
, err
);
1453 ext4_msg(sb
, KERN_CRIT
,
1454 "This should not happen!! Data will be lost\n");
1456 ext4_print_free_blocks(mpd
->inode
);
1458 /* invalidate all the pages */
1459 ext4_da_block_invalidatepages(mpd
);
1461 /* Mark this page range as having been completed */
1468 if (map
.m_flags
& EXT4_MAP_NEW
) {
1469 struct block_device
*bdev
= mpd
->inode
->i_sb
->s_bdev
;
1472 for (i
= 0; i
< map
.m_len
; i
++)
1473 unmap_underlying_metadata(bdev
, map
.m_pblk
+ i
);
1475 if (ext4_should_order_data(mpd
->inode
)) {
1476 err
= ext4_jbd2_file_inode(handle
, mpd
->inode
);
1478 /* Only if the journal is aborted */
1484 * Update on-disk size along with block allocation.
1486 disksize
= ((loff_t
) next
+ blks
) << mpd
->inode
->i_blkbits
;
1487 if (disksize
> i_size_read(mpd
->inode
))
1488 disksize
= i_size_read(mpd
->inode
);
1489 if (disksize
> EXT4_I(mpd
->inode
)->i_disksize
) {
1490 ext4_update_i_disksize(mpd
->inode
, disksize
);
1491 err
= ext4_mark_inode_dirty(handle
, mpd
->inode
);
1493 ext4_error(mpd
->inode
->i_sb
,
1494 "Failed to mark inode %lu dirty",
1499 mpage_da_submit_io(mpd
, mapp
);
1503 #define BH_FLAGS ((1 << BH_Uptodate) | (1 << BH_Mapped) | \
1504 (1 << BH_Delay) | (1 << BH_Unwritten))
1507 * mpage_add_bh_to_extent - try to add one more block to extent of blocks
1509 * @mpd->lbh - extent of blocks
1510 * @logical - logical number of the block in the file
1511 * @bh - bh of the block (used to access block's state)
1513 * the function is used to collect contig. blocks in same state
1515 static void mpage_add_bh_to_extent(struct mpage_da_data
*mpd
,
1516 sector_t logical
, size_t b_size
,
1517 unsigned long b_state
)
1520 int nrblocks
= mpd
->b_size
>> mpd
->inode
->i_blkbits
;
1523 * XXX Don't go larger than mballoc is willing to allocate
1524 * This is a stopgap solution. We eventually need to fold
1525 * mpage_da_submit_io() into this function and then call
1526 * ext4_map_blocks() multiple times in a loop
1528 if (nrblocks
>= 8*1024*1024/mpd
->inode
->i_sb
->s_blocksize
)
1531 /* check if thereserved journal credits might overflow */
1532 if (!(ext4_test_inode_flag(mpd
->inode
, EXT4_INODE_EXTENTS
))) {
1533 if (nrblocks
>= EXT4_MAX_TRANS_DATA
) {
1535 * With non-extent format we are limited by the journal
1536 * credit available. Total credit needed to insert
1537 * nrblocks contiguous blocks is dependent on the
1538 * nrblocks. So limit nrblocks.
1541 } else if ((nrblocks
+ (b_size
>> mpd
->inode
->i_blkbits
)) >
1542 EXT4_MAX_TRANS_DATA
) {
1544 * Adding the new buffer_head would make it cross the
1545 * allowed limit for which we have journal credit
1546 * reserved. So limit the new bh->b_size
1548 b_size
= (EXT4_MAX_TRANS_DATA
- nrblocks
) <<
1549 mpd
->inode
->i_blkbits
;
1550 /* we will do mpage_da_submit_io in the next loop */
1554 * First block in the extent
1556 if (mpd
->b_size
== 0) {
1557 mpd
->b_blocknr
= logical
;
1558 mpd
->b_size
= b_size
;
1559 mpd
->b_state
= b_state
& BH_FLAGS
;
1563 next
= mpd
->b_blocknr
+ nrblocks
;
1565 * Can we merge the block to our big extent?
1567 if (logical
== next
&& (b_state
& BH_FLAGS
) == mpd
->b_state
) {
1568 mpd
->b_size
+= b_size
;
1574 * We couldn't merge the block to our extent, so we
1575 * need to flush current extent and start new one
1577 mpage_da_map_and_submit(mpd
);
1581 static int ext4_bh_delay_or_unwritten(handle_t
*handle
, struct buffer_head
*bh
)
1583 return (buffer_delay(bh
) || buffer_unwritten(bh
)) && buffer_dirty(bh
);
1587 * This is a special get_blocks_t callback which is used by
1588 * ext4_da_write_begin(). It will either return mapped block or
1589 * reserve space for a single block.
1591 * For delayed buffer_head we have BH_Mapped, BH_New, BH_Delay set.
1592 * We also have b_blocknr = -1 and b_bdev initialized properly
1594 * For unwritten buffer_head we have BH_Mapped, BH_New, BH_Unwritten set.
1595 * We also have b_blocknr = physicalblock mapping unwritten extent and b_bdev
1596 * initialized properly.
1598 static int ext4_da_get_block_prep(struct inode
*inode
, sector_t iblock
,
1599 struct buffer_head
*bh
, int create
)
1601 struct ext4_map_blocks map
;
1603 sector_t invalid_block
= ~((sector_t
) 0xffff);
1605 if (invalid_block
< ext4_blocks_count(EXT4_SB(inode
->i_sb
)->s_es
))
1608 BUG_ON(create
== 0);
1609 BUG_ON(bh
->b_size
!= inode
->i_sb
->s_blocksize
);
1611 map
.m_lblk
= iblock
;
1615 * first, we need to know whether the block is allocated already
1616 * preallocated blocks are unmapped but should treated
1617 * the same as allocated blocks.
1619 ret
= ext4_map_blocks(NULL
, inode
, &map
, 0);
1623 if (buffer_delay(bh
))
1624 return 0; /* Not sure this could or should happen */
1626 * XXX: __block_write_begin() unmaps passed block, is it OK?
1628 ret
= ext4_da_reserve_space(inode
, iblock
);
1630 /* not enough space to reserve */
1633 map_bh(bh
, inode
->i_sb
, invalid_block
);
1635 set_buffer_delay(bh
);
1639 map_bh(bh
, inode
->i_sb
, map
.m_pblk
);
1640 bh
->b_state
= (bh
->b_state
& ~EXT4_MAP_FLAGS
) | map
.m_flags
;
1642 if (buffer_unwritten(bh
)) {
1643 /* A delayed write to unwritten bh should be marked
1644 * new and mapped. Mapped ensures that we don't do
1645 * get_block multiple times when we write to the same
1646 * offset and new ensures that we do proper zero out
1647 * for partial write.
1650 set_buffer_mapped(bh
);
1656 * This function is used as a standard get_block_t calback function
1657 * when there is no desire to allocate any blocks. It is used as a
1658 * callback function for block_write_begin() and block_write_full_page().
1659 * These functions should only try to map a single block at a time.
1661 * Since this function doesn't do block allocations even if the caller
1662 * requests it by passing in create=1, it is critically important that
1663 * any caller checks to make sure that any buffer heads are returned
1664 * by this function are either all already mapped or marked for
1665 * delayed allocation before calling block_write_full_page(). Otherwise,
1666 * b_blocknr could be left unitialized, and the page write functions will
1667 * be taken by surprise.
1669 static int noalloc_get_block_write(struct inode
*inode
, sector_t iblock
,
1670 struct buffer_head
*bh_result
, int create
)
1672 BUG_ON(bh_result
->b_size
!= inode
->i_sb
->s_blocksize
);
1673 return _ext4_get_block(inode
, iblock
, bh_result
, 0);
1676 static int bget_one(handle_t
*handle
, struct buffer_head
*bh
)
1682 static int bput_one(handle_t
*handle
, struct buffer_head
*bh
)
1688 static int __ext4_journalled_writepage(struct page
*page
,
1691 struct address_space
*mapping
= page
->mapping
;
1692 struct inode
*inode
= mapping
->host
;
1693 struct buffer_head
*page_bufs
;
1694 handle_t
*handle
= NULL
;
1698 ClearPageChecked(page
);
1699 page_bufs
= page_buffers(page
);
1701 walk_page_buffers(handle
, page_bufs
, 0, len
, NULL
, bget_one
);
1702 /* As soon as we unlock the page, it can go away, but we have
1703 * references to buffers so we are safe */
1706 handle
= ext4_journal_start(inode
, ext4_writepage_trans_blocks(inode
));
1707 if (IS_ERR(handle
)) {
1708 ret
= PTR_ERR(handle
);
1712 BUG_ON(!ext4_handle_valid(handle
));
1714 ret
= walk_page_buffers(handle
, page_bufs
, 0, len
, NULL
,
1715 do_journal_get_write_access
);
1717 err
= walk_page_buffers(handle
, page_bufs
, 0, len
, NULL
,
1721 EXT4_I(inode
)->i_datasync_tid
= handle
->h_transaction
->t_tid
;
1722 err
= ext4_journal_stop(handle
);
1726 walk_page_buffers(handle
, page_bufs
, 0, len
, NULL
, bput_one
);
1727 ext4_set_inode_state(inode
, EXT4_STATE_JDATA
);
1732 static int ext4_set_bh_endio(struct buffer_head
*bh
, struct inode
*inode
);
1733 static void ext4_end_io_buffer_write(struct buffer_head
*bh
, int uptodate
);
1736 * Note that we don't need to start a transaction unless we're journaling data
1737 * because we should have holes filled from ext4_page_mkwrite(). We even don't
1738 * need to file the inode to the transaction's list in ordered mode because if
1739 * we are writing back data added by write(), the inode is already there and if
1740 * we are writing back data modified via mmap(), no one guarantees in which
1741 * transaction the data will hit the disk. In case we are journaling data, we
1742 * cannot start transaction directly because transaction start ranks above page
1743 * lock so we have to do some magic.
1745 * This function can get called via...
1746 * - ext4_da_writepages after taking page lock (have journal handle)
1747 * - journal_submit_inode_data_buffers (no journal handle)
1748 * - shrink_page_list via pdflush (no journal handle)
1749 * - grab_page_cache when doing write_begin (have journal handle)
1751 * We don't do any block allocation in this function. If we have page with
1752 * multiple blocks we need to write those buffer_heads that are mapped. This
1753 * is important for mmaped based write. So if we do with blocksize 1K
1754 * truncate(f, 1024);
1755 * a = mmap(f, 0, 4096);
1757 * truncate(f, 4096);
1758 * we have in the page first buffer_head mapped via page_mkwrite call back
1759 * but other bufer_heads would be unmapped but dirty(dirty done via the
1760 * do_wp_page). So writepage should write the first block. If we modify
1761 * the mmap area beyond 1024 we will again get a page_fault and the
1762 * page_mkwrite callback will do the block allocation and mark the
1763 * buffer_heads mapped.
1765 * We redirty the page if we have any buffer_heads that is either delay or
1766 * unwritten in the page.
1768 * We can get recursively called as show below.
1770 * ext4_writepage() -> kmalloc() -> __alloc_pages() -> page_launder() ->
1773 * But since we don't do any block allocation we should not deadlock.
1774 * Page also have the dirty flag cleared so we don't get recurive page_lock.
1776 static int ext4_writepage(struct page
*page
,
1777 struct writeback_control
*wbc
)
1779 int ret
= 0, commit_write
= 0;
1782 struct buffer_head
*page_bufs
= NULL
;
1783 struct inode
*inode
= page
->mapping
->host
;
1785 trace_ext4_writepage(page
);
1786 size
= i_size_read(inode
);
1787 if (page
->index
== size
>> PAGE_CACHE_SHIFT
)
1788 len
= size
& ~PAGE_CACHE_MASK
;
1790 len
= PAGE_CACHE_SIZE
;
1793 * If the page does not have buffers (for whatever reason),
1794 * try to create them using __block_write_begin. If this
1795 * fails, redirty the page and move on.
1797 if (!page_has_buffers(page
)) {
1798 if (__block_write_begin(page
, 0, len
,
1799 noalloc_get_block_write
)) {
1801 redirty_page_for_writepage(wbc
, page
);
1807 page_bufs
= page_buffers(page
);
1808 if (walk_page_buffers(NULL
, page_bufs
, 0, len
, NULL
,
1809 ext4_bh_delay_or_unwritten
)) {
1811 * We don't want to do block allocation, so redirty
1812 * the page and return. We may reach here when we do
1813 * a journal commit via journal_submit_inode_data_buffers.
1814 * We can also reach here via shrink_page_list
1819 /* now mark the buffer_heads as dirty and uptodate */
1820 block_commit_write(page
, 0, len
);
1822 if (PageChecked(page
) && ext4_should_journal_data(inode
))
1824 * It's mmapped pagecache. Add buffers and journal it. There
1825 * doesn't seem much point in redirtying the page here.
1827 return __ext4_journalled_writepage(page
, len
);
1829 if (buffer_uninit(page_bufs
)) {
1830 ext4_set_bh_endio(page_bufs
, inode
);
1831 ret
= block_write_full_page_endio(page
, noalloc_get_block_write
,
1832 wbc
, ext4_end_io_buffer_write
);
1834 ret
= block_write_full_page(page
, noalloc_get_block_write
,
1841 * This is called via ext4_da_writepages() to
1842 * calculate the total number of credits to reserve to fit
1843 * a single extent allocation into a single transaction,
1844 * ext4_da_writpeages() will loop calling this before
1845 * the block allocation.
1848 static int ext4_da_writepages_trans_blocks(struct inode
*inode
)
1850 int max_blocks
= EXT4_I(inode
)->i_reserved_data_blocks
;
1853 * With non-extent format the journal credit needed to
1854 * insert nrblocks contiguous block is dependent on
1855 * number of contiguous block. So we will limit
1856 * number of contiguous block to a sane value
1858 if (!(ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
)) &&
1859 (max_blocks
> EXT4_MAX_TRANS_DATA
))
1860 max_blocks
= EXT4_MAX_TRANS_DATA
;
1862 return ext4_chunk_trans_blocks(inode
, max_blocks
);
1866 * write_cache_pages_da - walk the list of dirty pages of the given
1867 * address space and accumulate pages that need writing, and call
1868 * mpage_da_map_and_submit to map a single contiguous memory region
1869 * and then write them.
1871 static int write_cache_pages_da(struct address_space
*mapping
,
1872 struct writeback_control
*wbc
,
1873 struct mpage_da_data
*mpd
,
1874 pgoff_t
*done_index
)
1876 struct buffer_head
*bh
, *head
;
1877 struct inode
*inode
= mapping
->host
;
1878 struct pagevec pvec
;
1879 unsigned int nr_pages
;
1882 long nr_to_write
= wbc
->nr_to_write
;
1883 int i
, tag
, ret
= 0;
1885 memset(mpd
, 0, sizeof(struct mpage_da_data
));
1888 pagevec_init(&pvec
, 0);
1889 index
= wbc
->range_start
>> PAGE_CACHE_SHIFT
;
1890 end
= wbc
->range_end
>> PAGE_CACHE_SHIFT
;
1892 if (wbc
->sync_mode
== WB_SYNC_ALL
|| wbc
->tagged_writepages
)
1893 tag
= PAGECACHE_TAG_TOWRITE
;
1895 tag
= PAGECACHE_TAG_DIRTY
;
1897 *done_index
= index
;
1898 while (index
<= end
) {
1899 nr_pages
= pagevec_lookup_tag(&pvec
, mapping
, &index
, tag
,
1900 min(end
- index
, (pgoff_t
)PAGEVEC_SIZE
-1) + 1);
1904 for (i
= 0; i
< nr_pages
; i
++) {
1905 struct page
*page
= pvec
.pages
[i
];
1908 * At this point, the page may be truncated or
1909 * invalidated (changing page->mapping to NULL), or
1910 * even swizzled back from swapper_space to tmpfs file
1911 * mapping. However, page->index will not change
1912 * because we have a reference on the page.
1914 if (page
->index
> end
)
1917 *done_index
= page
->index
+ 1;
1920 * If we can't merge this page, and we have
1921 * accumulated an contiguous region, write it
1923 if ((mpd
->next_page
!= page
->index
) &&
1924 (mpd
->next_page
!= mpd
->first_page
)) {
1925 mpage_da_map_and_submit(mpd
);
1926 goto ret_extent_tail
;
1932 * If the page is no longer dirty, or its
1933 * mapping no longer corresponds to inode we
1934 * are writing (which means it has been
1935 * truncated or invalidated), or the page is
1936 * already under writeback and we are not
1937 * doing a data integrity writeback, skip the page
1939 if (!PageDirty(page
) ||
1940 (PageWriteback(page
) &&
1941 (wbc
->sync_mode
== WB_SYNC_NONE
)) ||
1942 unlikely(page
->mapping
!= mapping
)) {
1947 wait_on_page_writeback(page
);
1948 BUG_ON(PageWriteback(page
));
1950 if (mpd
->next_page
!= page
->index
)
1951 mpd
->first_page
= page
->index
;
1952 mpd
->next_page
= page
->index
+ 1;
1953 logical
= (sector_t
) page
->index
<<
1954 (PAGE_CACHE_SHIFT
- inode
->i_blkbits
);
1956 if (!page_has_buffers(page
)) {
1957 mpage_add_bh_to_extent(mpd
, logical
,
1959 (1 << BH_Dirty
) | (1 << BH_Uptodate
));
1961 goto ret_extent_tail
;
1964 * Page with regular buffer heads,
1965 * just add all dirty ones
1967 head
= page_buffers(page
);
1970 BUG_ON(buffer_locked(bh
));
1972 * We need to try to allocate
1973 * unmapped blocks in the same page.
1974 * Otherwise we won't make progress
1975 * with the page in ext4_writepage
1977 if (ext4_bh_delay_or_unwritten(NULL
, bh
)) {
1978 mpage_add_bh_to_extent(mpd
, logical
,
1982 goto ret_extent_tail
;
1983 } else if (buffer_dirty(bh
) && (buffer_mapped(bh
))) {
1985 * mapped dirty buffer. We need
1986 * to update the b_state
1987 * because we look at b_state
1988 * in mpage_da_map_blocks. We
1989 * don't update b_size because
1990 * if we find an unmapped
1991 * buffer_head later we need to
1992 * use the b_state flag of that
1995 if (mpd
->b_size
== 0)
1996 mpd
->b_state
= bh
->b_state
& BH_FLAGS
;
1999 } while ((bh
= bh
->b_this_page
) != head
);
2002 if (nr_to_write
> 0) {
2004 if (nr_to_write
== 0 &&
2005 wbc
->sync_mode
== WB_SYNC_NONE
)
2007 * We stop writing back only if we are
2008 * not doing integrity sync. In case of
2009 * integrity sync we have to keep going
2010 * because someone may be concurrently
2011 * dirtying pages, and we might have
2012 * synced a lot of newly appeared dirty
2013 * pages, but have not synced all of the
2019 pagevec_release(&pvec
);
2024 ret
= MPAGE_DA_EXTENT_TAIL
;
2026 pagevec_release(&pvec
);
2032 static int ext4_da_writepages(struct address_space
*mapping
,
2033 struct writeback_control
*wbc
)
2036 int range_whole
= 0;
2037 handle_t
*handle
= NULL
;
2038 struct mpage_da_data mpd
;
2039 struct inode
*inode
= mapping
->host
;
2040 int pages_written
= 0;
2041 unsigned int max_pages
;
2042 int range_cyclic
, cycled
= 1, io_done
= 0;
2043 int needed_blocks
, ret
= 0;
2044 long desired_nr_to_write
, nr_to_writebump
= 0;
2045 loff_t range_start
= wbc
->range_start
;
2046 struct ext4_sb_info
*sbi
= EXT4_SB(mapping
->host
->i_sb
);
2047 pgoff_t done_index
= 0;
2050 trace_ext4_da_writepages(inode
, wbc
);
2053 * No pages to write? This is mainly a kludge to avoid starting
2054 * a transaction for special inodes like journal inode on last iput()
2055 * because that could violate lock ordering on umount
2057 if (!mapping
->nrpages
|| !mapping_tagged(mapping
, PAGECACHE_TAG_DIRTY
))
2061 * If the filesystem has aborted, it is read-only, so return
2062 * right away instead of dumping stack traces later on that
2063 * will obscure the real source of the problem. We test
2064 * EXT4_MF_FS_ABORTED instead of sb->s_flag's MS_RDONLY because
2065 * the latter could be true if the filesystem is mounted
2066 * read-only, and in that case, ext4_da_writepages should
2067 * *never* be called, so if that ever happens, we would want
2070 if (unlikely(sbi
->s_mount_flags
& EXT4_MF_FS_ABORTED
))
2073 if (wbc
->range_start
== 0 && wbc
->range_end
== LLONG_MAX
)
2076 range_cyclic
= wbc
->range_cyclic
;
2077 if (wbc
->range_cyclic
) {
2078 index
= mapping
->writeback_index
;
2081 wbc
->range_start
= index
<< PAGE_CACHE_SHIFT
;
2082 wbc
->range_end
= LLONG_MAX
;
2083 wbc
->range_cyclic
= 0;
2086 index
= wbc
->range_start
>> PAGE_CACHE_SHIFT
;
2087 end
= wbc
->range_end
>> PAGE_CACHE_SHIFT
;
2091 * This works around two forms of stupidity. The first is in
2092 * the writeback code, which caps the maximum number of pages
2093 * written to be 1024 pages. This is wrong on multiple
2094 * levels; different architectues have a different page size,
2095 * which changes the maximum amount of data which gets
2096 * written. Secondly, 4 megabytes is way too small. XFS
2097 * forces this value to be 16 megabytes by multiplying
2098 * nr_to_write parameter by four, and then relies on its
2099 * allocator to allocate larger extents to make them
2100 * contiguous. Unfortunately this brings us to the second
2101 * stupidity, which is that ext4's mballoc code only allocates
2102 * at most 2048 blocks. So we force contiguous writes up to
2103 * the number of dirty blocks in the inode, or
2104 * sbi->max_writeback_mb_bump whichever is smaller.
2106 max_pages
= sbi
->s_max_writeback_mb_bump
<< (20 - PAGE_CACHE_SHIFT
);
2107 if (!range_cyclic
&& range_whole
) {
2108 if (wbc
->nr_to_write
== LONG_MAX
)
2109 desired_nr_to_write
= wbc
->nr_to_write
;
2111 desired_nr_to_write
= wbc
->nr_to_write
* 8;
2113 desired_nr_to_write
= ext4_num_dirty_pages(inode
, index
,
2115 if (desired_nr_to_write
> max_pages
)
2116 desired_nr_to_write
= max_pages
;
2118 if (wbc
->nr_to_write
< desired_nr_to_write
) {
2119 nr_to_writebump
= desired_nr_to_write
- wbc
->nr_to_write
;
2120 wbc
->nr_to_write
= desired_nr_to_write
;
2124 if (wbc
->sync_mode
== WB_SYNC_ALL
|| wbc
->tagged_writepages
)
2125 tag_pages_for_writeback(mapping
, index
, end
);
2127 while (!ret
&& wbc
->nr_to_write
> 0) {
2130 * we insert one extent at a time. So we need
2131 * credit needed for single extent allocation.
2132 * journalled mode is currently not supported
2135 BUG_ON(ext4_should_journal_data(inode
));
2136 needed_blocks
= ext4_da_writepages_trans_blocks(inode
);
2138 /* start a new transaction*/
2139 handle
= ext4_journal_start(inode
, needed_blocks
);
2140 if (IS_ERR(handle
)) {
2141 ret
= PTR_ERR(handle
);
2142 ext4_msg(inode
->i_sb
, KERN_CRIT
, "%s: jbd2_start: "
2143 "%ld pages, ino %lu; err %d", __func__
,
2144 wbc
->nr_to_write
, inode
->i_ino
, ret
);
2145 goto out_writepages
;
2149 * Now call write_cache_pages_da() to find the next
2150 * contiguous region of logical blocks that need
2151 * blocks to be allocated by ext4 and submit them.
2153 ret
= write_cache_pages_da(mapping
, wbc
, &mpd
, &done_index
);
2155 * If we have a contiguous extent of pages and we
2156 * haven't done the I/O yet, map the blocks and submit
2159 if (!mpd
.io_done
&& mpd
.next_page
!= mpd
.first_page
) {
2160 mpage_da_map_and_submit(&mpd
);
2161 ret
= MPAGE_DA_EXTENT_TAIL
;
2163 trace_ext4_da_write_pages(inode
, &mpd
);
2164 wbc
->nr_to_write
-= mpd
.pages_written
;
2166 ext4_journal_stop(handle
);
2168 if ((mpd
.retval
== -ENOSPC
) && sbi
->s_journal
) {
2169 /* commit the transaction which would
2170 * free blocks released in the transaction
2173 jbd2_journal_force_commit_nested(sbi
->s_journal
);
2175 } else if (ret
== MPAGE_DA_EXTENT_TAIL
) {
2177 * got one extent now try with
2180 pages_written
+= mpd
.pages_written
;
2183 } else if (wbc
->nr_to_write
)
2185 * There is no more writeout needed
2186 * or we requested for a noblocking writeout
2187 * and we found the device congested
2191 if (!io_done
&& !cycled
) {
2194 wbc
->range_start
= index
<< PAGE_CACHE_SHIFT
;
2195 wbc
->range_end
= mapping
->writeback_index
- 1;
2200 wbc
->range_cyclic
= range_cyclic
;
2201 if (wbc
->range_cyclic
|| (range_whole
&& wbc
->nr_to_write
> 0))
2203 * set the writeback_index so that range_cyclic
2204 * mode will write it back later
2206 mapping
->writeback_index
= done_index
;
2209 wbc
->nr_to_write
-= nr_to_writebump
;
2210 wbc
->range_start
= range_start
;
2211 trace_ext4_da_writepages_result(inode
, wbc
, ret
, pages_written
);
2215 #define FALL_BACK_TO_NONDELALLOC 1
2216 static int ext4_nonda_switch(struct super_block
*sb
)
2218 s64 free_blocks
, dirty_blocks
;
2219 struct ext4_sb_info
*sbi
= EXT4_SB(sb
);
2222 * switch to non delalloc mode if we are running low
2223 * on free block. The free block accounting via percpu
2224 * counters can get slightly wrong with percpu_counter_batch getting
2225 * accumulated on each CPU without updating global counters
2226 * Delalloc need an accurate free block accounting. So switch
2227 * to non delalloc when we are near to error range.
2229 free_blocks
= percpu_counter_read_positive(&sbi
->s_freeblocks_counter
);
2230 dirty_blocks
= percpu_counter_read_positive(&sbi
->s_dirtyblocks_counter
);
2231 if (2 * free_blocks
< 3 * dirty_blocks
||
2232 free_blocks
< (dirty_blocks
+ EXT4_FREEBLOCKS_WATERMARK
)) {
2234 * free block count is less than 150% of dirty blocks
2235 * or free blocks is less than watermark
2240 * Even if we don't switch but are nearing capacity,
2241 * start pushing delalloc when 1/2 of free blocks are dirty.
2243 if (free_blocks
< 2 * dirty_blocks
)
2244 writeback_inodes_sb_if_idle(sb
);
2249 static int ext4_da_write_begin(struct file
*file
, struct address_space
*mapping
,
2250 loff_t pos
, unsigned len
, unsigned flags
,
2251 struct page
**pagep
, void **fsdata
)
2253 int ret
, retries
= 0;
2256 struct inode
*inode
= mapping
->host
;
2260 index
= pos
>> PAGE_CACHE_SHIFT
;
2262 if (ext4_nonda_switch(inode
->i_sb
)) {
2263 *fsdata
= (void *)FALL_BACK_TO_NONDELALLOC
;
2264 return ext4_write_begin(file
, mapping
, pos
,
2265 len
, flags
, pagep
, fsdata
);
2267 *fsdata
= (void *)0;
2268 trace_ext4_da_write_begin(inode
, pos
, len
, flags
);
2271 * With delayed allocation, we don't log the i_disksize update
2272 * if there is delayed block allocation. But we still need
2273 * to journalling the i_disksize update if writes to the end
2274 * of file which has an already mapped buffer.
2276 handle
= ext4_journal_start(inode
, 1);
2277 if (IS_ERR(handle
)) {
2278 ret
= PTR_ERR(handle
);
2281 /* We cannot recurse into the filesystem as the transaction is already
2283 flags
|= AOP_FLAG_NOFS
;
2285 page
= grab_cache_page_write_begin(mapping
, index
, flags
);
2287 ext4_journal_stop(handle
);
2293 ret
= __block_write_begin(page
, pos
, len
, ext4_da_get_block_prep
);
2296 ext4_journal_stop(handle
);
2297 page_cache_release(page
);
2299 * block_write_begin may have instantiated a few blocks
2300 * outside i_size. Trim these off again. Don't need
2301 * i_size_read because we hold i_mutex.
2303 if (pos
+ len
> inode
->i_size
)
2304 ext4_truncate_failed_write(inode
);
2306 page_len
= pos
& (PAGE_CACHE_SIZE
- 1);
2308 ret
= ext4_discard_partial_page_buffers_no_lock(handle
,
2309 inode
, page
, pos
- page_len
, page_len
,
2310 EXT4_DISCARD_PARTIAL_PG_ZERO_UNMAPPED
);
2314 if (ret
== -ENOSPC
&& ext4_should_retry_alloc(inode
->i_sb
, &retries
))
2321 * Check if we should update i_disksize
2322 * when write to the end of file but not require block allocation
2324 static int ext4_da_should_update_i_disksize(struct page
*page
,
2325 unsigned long offset
)
2327 struct buffer_head
*bh
;
2328 struct inode
*inode
= page
->mapping
->host
;
2332 bh
= page_buffers(page
);
2333 idx
= offset
>> inode
->i_blkbits
;
2335 for (i
= 0; i
< idx
; i
++)
2336 bh
= bh
->b_this_page
;
2338 if (!buffer_mapped(bh
) || (buffer_delay(bh
)) || buffer_unwritten(bh
))
2343 static int ext4_da_write_end(struct file
*file
,
2344 struct address_space
*mapping
,
2345 loff_t pos
, unsigned len
, unsigned copied
,
2346 struct page
*page
, void *fsdata
)
2348 struct inode
*inode
= mapping
->host
;
2350 handle_t
*handle
= ext4_journal_current_handle();
2352 unsigned long start
, end
;
2353 int write_mode
= (int)(unsigned long)fsdata
;
2356 if (write_mode
== FALL_BACK_TO_NONDELALLOC
) {
2357 if (ext4_should_order_data(inode
)) {
2358 return ext4_ordered_write_end(file
, mapping
, pos
,
2359 len
, copied
, page
, fsdata
);
2360 } else if (ext4_should_writeback_data(inode
)) {
2361 return ext4_writeback_write_end(file
, mapping
, pos
,
2362 len
, copied
, page
, fsdata
);
2368 trace_ext4_da_write_end(inode
, pos
, len
, copied
);
2369 start
= pos
& (PAGE_CACHE_SIZE
- 1);
2370 end
= start
+ copied
- 1;
2373 * generic_write_end() will run mark_inode_dirty() if i_size
2374 * changes. So let's piggyback the i_disksize mark_inode_dirty
2378 new_i_size
= pos
+ copied
;
2379 if (new_i_size
> EXT4_I(inode
)->i_disksize
) {
2380 if (ext4_da_should_update_i_disksize(page
, end
)) {
2381 down_write(&EXT4_I(inode
)->i_data_sem
);
2382 if (new_i_size
> EXT4_I(inode
)->i_disksize
) {
2384 * Updating i_disksize when extending file
2385 * without needing block allocation
2387 if (ext4_should_order_data(inode
))
2388 ret
= ext4_jbd2_file_inode(handle
,
2391 EXT4_I(inode
)->i_disksize
= new_i_size
;
2393 up_write(&EXT4_I(inode
)->i_data_sem
);
2394 /* We need to mark inode dirty even if
2395 * new_i_size is less that inode->i_size
2396 * bu greater than i_disksize.(hint delalloc)
2398 ext4_mark_inode_dirty(handle
, inode
);
2401 ret2
= generic_write_end(file
, mapping
, pos
, len
, copied
,
2404 page_len
= PAGE_CACHE_SIZE
-
2405 ((pos
+ copied
- 1) & (PAGE_CACHE_SIZE
- 1));
2408 ret
= ext4_discard_partial_page_buffers_no_lock(handle
,
2409 inode
, page
, pos
+ copied
- 1, page_len
,
2410 EXT4_DISCARD_PARTIAL_PG_ZERO_UNMAPPED
);
2416 ret2
= ext4_journal_stop(handle
);
2420 return ret
? ret
: copied
;
2423 static void ext4_da_invalidatepage(struct page
*page
, unsigned long offset
)
2426 * Drop reserved blocks
2428 BUG_ON(!PageLocked(page
));
2429 if (!page_has_buffers(page
))
2432 ext4_da_page_release_reservation(page
, offset
);
2435 ext4_invalidatepage(page
, offset
);
2441 * Force all delayed allocation blocks to be allocated for a given inode.
2443 int ext4_alloc_da_blocks(struct inode
*inode
)
2445 trace_ext4_alloc_da_blocks(inode
);
2447 if (!EXT4_I(inode
)->i_reserved_data_blocks
&&
2448 !EXT4_I(inode
)->i_reserved_meta_blocks
)
2452 * We do something simple for now. The filemap_flush() will
2453 * also start triggering a write of the data blocks, which is
2454 * not strictly speaking necessary (and for users of
2455 * laptop_mode, not even desirable). However, to do otherwise
2456 * would require replicating code paths in:
2458 * ext4_da_writepages() ->
2459 * write_cache_pages() ---> (via passed in callback function)
2460 * __mpage_da_writepage() -->
2461 * mpage_add_bh_to_extent()
2462 * mpage_da_map_blocks()
2464 * The problem is that write_cache_pages(), located in
2465 * mm/page-writeback.c, marks pages clean in preparation for
2466 * doing I/O, which is not desirable if we're not planning on
2469 * We could call write_cache_pages(), and then redirty all of
2470 * the pages by calling redirty_page_for_writepage() but that
2471 * would be ugly in the extreme. So instead we would need to
2472 * replicate parts of the code in the above functions,
2473 * simplifying them because we wouldn't actually intend to
2474 * write out the pages, but rather only collect contiguous
2475 * logical block extents, call the multi-block allocator, and
2476 * then update the buffer heads with the block allocations.
2478 * For now, though, we'll cheat by calling filemap_flush(),
2479 * which will map the blocks, and start the I/O, but not
2480 * actually wait for the I/O to complete.
2482 return filemap_flush(inode
->i_mapping
);
2486 * bmap() is special. It gets used by applications such as lilo and by
2487 * the swapper to find the on-disk block of a specific piece of data.
2489 * Naturally, this is dangerous if the block concerned is still in the
2490 * journal. If somebody makes a swapfile on an ext4 data-journaling
2491 * filesystem and enables swap, then they may get a nasty shock when the
2492 * data getting swapped to that swapfile suddenly gets overwritten by
2493 * the original zero's written out previously to the journal and
2494 * awaiting writeback in the kernel's buffer cache.
2496 * So, if we see any bmap calls here on a modified, data-journaled file,
2497 * take extra steps to flush any blocks which might be in the cache.
2499 static sector_t
ext4_bmap(struct address_space
*mapping
, sector_t block
)
2501 struct inode
*inode
= mapping
->host
;
2505 if (mapping_tagged(mapping
, PAGECACHE_TAG_DIRTY
) &&
2506 test_opt(inode
->i_sb
, DELALLOC
)) {
2508 * With delalloc we want to sync the file
2509 * so that we can make sure we allocate
2512 filemap_write_and_wait(mapping
);
2515 if (EXT4_JOURNAL(inode
) &&
2516 ext4_test_inode_state(inode
, EXT4_STATE_JDATA
)) {
2518 * This is a REALLY heavyweight approach, but the use of
2519 * bmap on dirty files is expected to be extremely rare:
2520 * only if we run lilo or swapon on a freshly made file
2521 * do we expect this to happen.
2523 * (bmap requires CAP_SYS_RAWIO so this does not
2524 * represent an unprivileged user DOS attack --- we'd be
2525 * in trouble if mortal users could trigger this path at
2528 * NB. EXT4_STATE_JDATA is not set on files other than
2529 * regular files. If somebody wants to bmap a directory
2530 * or symlink and gets confused because the buffer
2531 * hasn't yet been flushed to disk, they deserve
2532 * everything they get.
2535 ext4_clear_inode_state(inode
, EXT4_STATE_JDATA
);
2536 journal
= EXT4_JOURNAL(inode
);
2537 jbd2_journal_lock_updates(journal
);
2538 err
= jbd2_journal_flush(journal
);
2539 jbd2_journal_unlock_updates(journal
);
2545 return generic_block_bmap(mapping
, block
, ext4_get_block
);
2548 static int ext4_readpage(struct file
*file
, struct page
*page
)
2550 trace_ext4_readpage(page
);
2551 return mpage_readpage(page
, ext4_get_block
);
2555 ext4_readpages(struct file
*file
, struct address_space
*mapping
,
2556 struct list_head
*pages
, unsigned nr_pages
)
2558 return mpage_readpages(mapping
, pages
, nr_pages
, ext4_get_block
);
2561 static void ext4_invalidatepage_free_endio(struct page
*page
, unsigned long offset
)
2563 struct buffer_head
*head
, *bh
;
2564 unsigned int curr_off
= 0;
2566 if (!page_has_buffers(page
))
2568 head
= bh
= page_buffers(page
);
2570 if (offset
<= curr_off
&& test_clear_buffer_uninit(bh
)
2572 ext4_free_io_end(bh
->b_private
);
2573 bh
->b_private
= NULL
;
2574 bh
->b_end_io
= NULL
;
2576 curr_off
= curr_off
+ bh
->b_size
;
2577 bh
= bh
->b_this_page
;
2578 } while (bh
!= head
);
2581 static void ext4_invalidatepage(struct page
*page
, unsigned long offset
)
2583 journal_t
*journal
= EXT4_JOURNAL(page
->mapping
->host
);
2585 trace_ext4_invalidatepage(page
, offset
);
2588 * free any io_end structure allocated for buffers to be discarded
2590 if (ext4_should_dioread_nolock(page
->mapping
->host
))
2591 ext4_invalidatepage_free_endio(page
, offset
);
2593 * If it's a full truncate we just forget about the pending dirtying
2596 ClearPageChecked(page
);
2599 jbd2_journal_invalidatepage(journal
, page
, offset
);
2601 block_invalidatepage(page
, offset
);
2604 static int ext4_releasepage(struct page
*page
, gfp_t wait
)
2606 journal_t
*journal
= EXT4_JOURNAL(page
->mapping
->host
);
2608 trace_ext4_releasepage(page
);
2610 WARN_ON(PageChecked(page
));
2611 if (!page_has_buffers(page
))
2614 return jbd2_journal_try_to_free_buffers(journal
, page
, wait
);
2616 return try_to_free_buffers(page
);
2620 * ext4_get_block used when preparing for a DIO write or buffer write.
2621 * We allocate an uinitialized extent if blocks haven't been allocated.
2622 * The extent will be converted to initialized after the IO is complete.
2624 static int ext4_get_block_write(struct inode
*inode
, sector_t iblock
,
2625 struct buffer_head
*bh_result
, int create
)
2627 ext4_debug("ext4_get_block_write: inode %lu, create flag %d\n",
2628 inode
->i_ino
, create
);
2629 return _ext4_get_block(inode
, iblock
, bh_result
,
2630 EXT4_GET_BLOCKS_IO_CREATE_EXT
);
2633 static void ext4_end_io_dio(struct kiocb
*iocb
, loff_t offset
,
2634 ssize_t size
, void *private, int ret
,
2637 struct inode
*inode
= iocb
->ki_filp
->f_path
.dentry
->d_inode
;
2638 ext4_io_end_t
*io_end
= iocb
->private;
2639 struct workqueue_struct
*wq
;
2640 unsigned long flags
;
2641 struct ext4_inode_info
*ei
;
2643 /* if not async direct IO or dio with 0 bytes write, just return */
2644 if (!io_end
|| !size
)
2647 ext_debug("ext4_end_io_dio(): io_end 0x%p"
2648 "for inode %lu, iocb 0x%p, offset %llu, size %llu\n",
2649 iocb
->private, io_end
->inode
->i_ino
, iocb
, offset
,
2652 /* if not aio dio with unwritten extents, just free io and return */
2653 if (!(io_end
->flag
& EXT4_IO_END_UNWRITTEN
)) {
2654 ext4_free_io_end(io_end
);
2655 iocb
->private = NULL
;
2658 aio_complete(iocb
, ret
, 0);
2659 inode_dio_done(inode
);
2663 io_end
->offset
= offset
;
2664 io_end
->size
= size
;
2666 io_end
->iocb
= iocb
;
2667 io_end
->result
= ret
;
2669 wq
= EXT4_SB(io_end
->inode
->i_sb
)->dio_unwritten_wq
;
2671 /* Add the io_end to per-inode completed aio dio list*/
2672 ei
= EXT4_I(io_end
->inode
);
2673 spin_lock_irqsave(&ei
->i_completed_io_lock
, flags
);
2674 list_add_tail(&io_end
->list
, &ei
->i_completed_io_list
);
2675 spin_unlock_irqrestore(&ei
->i_completed_io_lock
, flags
);
2677 /* queue the work to convert unwritten extents to written */
2678 queue_work(wq
, &io_end
->work
);
2679 iocb
->private = NULL
;
2681 /* XXX: probably should move into the real I/O completion handler */
2682 inode_dio_done(inode
);
2685 static void ext4_end_io_buffer_write(struct buffer_head
*bh
, int uptodate
)
2687 ext4_io_end_t
*io_end
= bh
->b_private
;
2688 struct workqueue_struct
*wq
;
2689 struct inode
*inode
;
2690 unsigned long flags
;
2692 if (!test_clear_buffer_uninit(bh
) || !io_end
)
2695 if (!(io_end
->inode
->i_sb
->s_flags
& MS_ACTIVE
)) {
2696 printk("sb umounted, discard end_io request for inode %lu\n",
2697 io_end
->inode
->i_ino
);
2698 ext4_free_io_end(io_end
);
2703 * It may be over-defensive here to check EXT4_IO_END_UNWRITTEN now,
2704 * but being more careful is always safe for the future change.
2706 inode
= io_end
->inode
;
2707 if (!(io_end
->flag
& EXT4_IO_END_UNWRITTEN
)) {
2708 io_end
->flag
|= EXT4_IO_END_UNWRITTEN
;
2709 atomic_inc(&EXT4_I(inode
)->i_aiodio_unwritten
);
2712 /* Add the io_end to per-inode completed io list*/
2713 spin_lock_irqsave(&EXT4_I(inode
)->i_completed_io_lock
, flags
);
2714 list_add_tail(&io_end
->list
, &EXT4_I(inode
)->i_completed_io_list
);
2715 spin_unlock_irqrestore(&EXT4_I(inode
)->i_completed_io_lock
, flags
);
2717 wq
= EXT4_SB(inode
->i_sb
)->dio_unwritten_wq
;
2718 /* queue the work to convert unwritten extents to written */
2719 queue_work(wq
, &io_end
->work
);
2721 bh
->b_private
= NULL
;
2722 bh
->b_end_io
= NULL
;
2723 clear_buffer_uninit(bh
);
2724 end_buffer_async_write(bh
, uptodate
);
2727 static int ext4_set_bh_endio(struct buffer_head
*bh
, struct inode
*inode
)
2729 ext4_io_end_t
*io_end
;
2730 struct page
*page
= bh
->b_page
;
2731 loff_t offset
= (sector_t
)page
->index
<< PAGE_CACHE_SHIFT
;
2732 size_t size
= bh
->b_size
;
2735 io_end
= ext4_init_io_end(inode
, GFP_ATOMIC
);
2737 pr_warn_ratelimited("%s: allocation fail\n", __func__
);
2741 io_end
->offset
= offset
;
2742 io_end
->size
= size
;
2744 * We need to hold a reference to the page to make sure it
2745 * doesn't get evicted before ext4_end_io_work() has a chance
2746 * to convert the extent from written to unwritten.
2748 io_end
->page
= page
;
2749 get_page(io_end
->page
);
2751 bh
->b_private
= io_end
;
2752 bh
->b_end_io
= ext4_end_io_buffer_write
;
2757 * For ext4 extent files, ext4 will do direct-io write to holes,
2758 * preallocated extents, and those write extend the file, no need to
2759 * fall back to buffered IO.
2761 * For holes, we fallocate those blocks, mark them as uninitialized
2762 * If those blocks were preallocated, we mark sure they are splited, but
2763 * still keep the range to write as uninitialized.
2765 * The unwrritten extents will be converted to written when DIO is completed.
2766 * For async direct IO, since the IO may still pending when return, we
2767 * set up an end_io call back function, which will do the conversion
2768 * when async direct IO completed.
2770 * If the O_DIRECT write will extend the file then add this inode to the
2771 * orphan list. So recovery will truncate it back to the original size
2772 * if the machine crashes during the write.
2775 static ssize_t
ext4_ext_direct_IO(int rw
, struct kiocb
*iocb
,
2776 const struct iovec
*iov
, loff_t offset
,
2777 unsigned long nr_segs
)
2779 struct file
*file
= iocb
->ki_filp
;
2780 struct inode
*inode
= file
->f_mapping
->host
;
2782 size_t count
= iov_length(iov
, nr_segs
);
2784 loff_t final_size
= offset
+ count
;
2785 if (rw
== WRITE
&& final_size
<= inode
->i_size
) {
2787 * We could direct write to holes and fallocate.
2789 * Allocated blocks to fill the hole are marked as uninitialized
2790 * to prevent parallel buffered read to expose the stale data
2791 * before DIO complete the data IO.
2793 * As to previously fallocated extents, ext4 get_block
2794 * will just simply mark the buffer mapped but still
2795 * keep the extents uninitialized.
2797 * for non AIO case, we will convert those unwritten extents
2798 * to written after return back from blockdev_direct_IO.
2800 * for async DIO, the conversion needs to be defered when
2801 * the IO is completed. The ext4 end_io callback function
2802 * will be called to take care of the conversion work.
2803 * Here for async case, we allocate an io_end structure to
2806 iocb
->private = NULL
;
2807 EXT4_I(inode
)->cur_aio_dio
= NULL
;
2808 if (!is_sync_kiocb(iocb
)) {
2809 iocb
->private = ext4_init_io_end(inode
, GFP_NOFS
);
2813 * we save the io structure for current async
2814 * direct IO, so that later ext4_map_blocks()
2815 * could flag the io structure whether there
2816 * is a unwritten extents needs to be converted
2817 * when IO is completed.
2819 EXT4_I(inode
)->cur_aio_dio
= iocb
->private;
2822 ret
= __blockdev_direct_IO(rw
, iocb
, inode
,
2823 inode
->i_sb
->s_bdev
, iov
,
2825 ext4_get_block_write
,
2828 DIO_LOCKING
| DIO_SKIP_HOLES
);
2830 EXT4_I(inode
)->cur_aio_dio
= NULL
;
2832 * The io_end structure takes a reference to the inode,
2833 * that structure needs to be destroyed and the
2834 * reference to the inode need to be dropped, when IO is
2835 * complete, even with 0 byte write, or failed.
2837 * In the successful AIO DIO case, the io_end structure will be
2838 * desctroyed and the reference to the inode will be dropped
2839 * after the end_io call back function is called.
2841 * In the case there is 0 byte write, or error case, since
2842 * VFS direct IO won't invoke the end_io call back function,
2843 * we need to free the end_io structure here.
2845 if (ret
!= -EIOCBQUEUED
&& ret
<= 0 && iocb
->private) {
2846 ext4_free_io_end(iocb
->private);
2847 iocb
->private = NULL
;
2848 } else if (ret
> 0 && ext4_test_inode_state(inode
,
2849 EXT4_STATE_DIO_UNWRITTEN
)) {
2852 * for non AIO case, since the IO is already
2853 * completed, we could do the conversion right here
2855 err
= ext4_convert_unwritten_extents(inode
,
2859 ext4_clear_inode_state(inode
, EXT4_STATE_DIO_UNWRITTEN
);
2864 /* for write the the end of file case, we fall back to old way */
2865 return ext4_ind_direct_IO(rw
, iocb
, iov
, offset
, nr_segs
);
2868 static ssize_t
ext4_direct_IO(int rw
, struct kiocb
*iocb
,
2869 const struct iovec
*iov
, loff_t offset
,
2870 unsigned long nr_segs
)
2872 struct file
*file
= iocb
->ki_filp
;
2873 struct inode
*inode
= file
->f_mapping
->host
;
2877 * If we are doing data journalling we don't support O_DIRECT
2879 if (ext4_should_journal_data(inode
))
2882 trace_ext4_direct_IO_enter(inode
, offset
, iov_length(iov
, nr_segs
), rw
);
2883 if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
))
2884 ret
= ext4_ext_direct_IO(rw
, iocb
, iov
, offset
, nr_segs
);
2886 ret
= ext4_ind_direct_IO(rw
, iocb
, iov
, offset
, nr_segs
);
2887 trace_ext4_direct_IO_exit(inode
, offset
,
2888 iov_length(iov
, nr_segs
), rw
, ret
);
2893 * Pages can be marked dirty completely asynchronously from ext4's journalling
2894 * activity. By filemap_sync_pte(), try_to_unmap_one(), etc. We cannot do
2895 * much here because ->set_page_dirty is called under VFS locks. The page is
2896 * not necessarily locked.
2898 * We cannot just dirty the page and leave attached buffers clean, because the
2899 * buffers' dirty state is "definitive". We cannot just set the buffers dirty
2900 * or jbddirty because all the journalling code will explode.
2902 * So what we do is to mark the page "pending dirty" and next time writepage
2903 * is called, propagate that into the buffers appropriately.
2905 static int ext4_journalled_set_page_dirty(struct page
*page
)
2907 SetPageChecked(page
);
2908 return __set_page_dirty_nobuffers(page
);
2911 static const struct address_space_operations ext4_ordered_aops
= {
2912 .readpage
= ext4_readpage
,
2913 .readpages
= ext4_readpages
,
2914 .writepage
= ext4_writepage
,
2915 .write_begin
= ext4_write_begin
,
2916 .write_end
= ext4_ordered_write_end
,
2918 .invalidatepage
= ext4_invalidatepage
,
2919 .releasepage
= ext4_releasepage
,
2920 .direct_IO
= ext4_direct_IO
,
2921 .migratepage
= buffer_migrate_page
,
2922 .is_partially_uptodate
= block_is_partially_uptodate
,
2923 .error_remove_page
= generic_error_remove_page
,
2926 static const struct address_space_operations ext4_writeback_aops
= {
2927 .readpage
= ext4_readpage
,
2928 .readpages
= ext4_readpages
,
2929 .writepage
= ext4_writepage
,
2930 .write_begin
= ext4_write_begin
,
2931 .write_end
= ext4_writeback_write_end
,
2933 .invalidatepage
= ext4_invalidatepage
,
2934 .releasepage
= ext4_releasepage
,
2935 .direct_IO
= ext4_direct_IO
,
2936 .migratepage
= buffer_migrate_page
,
2937 .is_partially_uptodate
= block_is_partially_uptodate
,
2938 .error_remove_page
= generic_error_remove_page
,
2941 static const struct address_space_operations ext4_journalled_aops
= {
2942 .readpage
= ext4_readpage
,
2943 .readpages
= ext4_readpages
,
2944 .writepage
= ext4_writepage
,
2945 .write_begin
= ext4_write_begin
,
2946 .write_end
= ext4_journalled_write_end
,
2947 .set_page_dirty
= ext4_journalled_set_page_dirty
,
2949 .invalidatepage
= ext4_invalidatepage
,
2950 .releasepage
= ext4_releasepage
,
2951 .direct_IO
= ext4_direct_IO
,
2952 .is_partially_uptodate
= block_is_partially_uptodate
,
2953 .error_remove_page
= generic_error_remove_page
,
2956 static const struct address_space_operations ext4_da_aops
= {
2957 .readpage
= ext4_readpage
,
2958 .readpages
= ext4_readpages
,
2959 .writepage
= ext4_writepage
,
2960 .writepages
= ext4_da_writepages
,
2961 .write_begin
= ext4_da_write_begin
,
2962 .write_end
= ext4_da_write_end
,
2964 .invalidatepage
= ext4_da_invalidatepage
,
2965 .releasepage
= ext4_releasepage
,
2966 .direct_IO
= ext4_direct_IO
,
2967 .migratepage
= buffer_migrate_page
,
2968 .is_partially_uptodate
= block_is_partially_uptodate
,
2969 .error_remove_page
= generic_error_remove_page
,
2972 void ext4_set_aops(struct inode
*inode
)
2974 if (ext4_should_order_data(inode
) &&
2975 test_opt(inode
->i_sb
, DELALLOC
))
2976 inode
->i_mapping
->a_ops
= &ext4_da_aops
;
2977 else if (ext4_should_order_data(inode
))
2978 inode
->i_mapping
->a_ops
= &ext4_ordered_aops
;
2979 else if (ext4_should_writeback_data(inode
) &&
2980 test_opt(inode
->i_sb
, DELALLOC
))
2981 inode
->i_mapping
->a_ops
= &ext4_da_aops
;
2982 else if (ext4_should_writeback_data(inode
))
2983 inode
->i_mapping
->a_ops
= &ext4_writeback_aops
;
2985 inode
->i_mapping
->a_ops
= &ext4_journalled_aops
;
2990 * ext4_discard_partial_page_buffers()
2991 * Wrapper function for ext4_discard_partial_page_buffers_no_lock.
2992 * This function finds and locks the page containing the offset
2993 * "from" and passes it to ext4_discard_partial_page_buffers_no_lock.
2994 * Calling functions that already have the page locked should call
2995 * ext4_discard_partial_page_buffers_no_lock directly.
2997 int ext4_discard_partial_page_buffers(handle_t
*handle
,
2998 struct address_space
*mapping
, loff_t from
,
2999 loff_t length
, int flags
)
3001 struct inode
*inode
= mapping
->host
;
3005 page
= find_or_create_page(mapping
, from
>> PAGE_CACHE_SHIFT
,
3006 mapping_gfp_mask(mapping
) & ~__GFP_FS
);
3010 err
= ext4_discard_partial_page_buffers_no_lock(handle
, inode
, page
,
3011 from
, length
, flags
);
3014 page_cache_release(page
);
3019 * ext4_discard_partial_page_buffers_no_lock()
3020 * Zeros a page range of length 'length' starting from offset 'from'.
3021 * Buffer heads that correspond to the block aligned regions of the
3022 * zeroed range will be unmapped. Unblock aligned regions
3023 * will have the corresponding buffer head mapped if needed so that
3024 * that region of the page can be updated with the partial zero out.
3026 * This function assumes that the page has already been locked. The
3027 * The range to be discarded must be contained with in the given page.
3028 * If the specified range exceeds the end of the page it will be shortened
3029 * to the end of the page that corresponds to 'from'. This function is
3030 * appropriate for updating a page and it buffer heads to be unmapped and
3031 * zeroed for blocks that have been either released, or are going to be
3034 * handle: The journal handle
3035 * inode: The files inode
3036 * page: A locked page that contains the offset "from"
3037 * from: The starting byte offset (from the begining of the file)
3038 * to begin discarding
3039 * len: The length of bytes to discard
3040 * flags: Optional flags that may be used:
3042 * EXT4_DISCARD_PARTIAL_PG_ZERO_UNMAPPED
3043 * Only zero the regions of the page whose buffer heads
3044 * have already been unmapped. This flag is appropriate
3045 * for updateing the contents of a page whose blocks may
3046 * have already been released, and we only want to zero
3047 * out the regions that correspond to those released blocks.
3049 * Returns zero on sucess or negative on failure.
3051 int ext4_discard_partial_page_buffers_no_lock(handle_t
*handle
,
3052 struct inode
*inode
, struct page
*page
, loff_t from
,
3053 loff_t length
, int flags
)
3055 ext4_fsblk_t index
= from
>> PAGE_CACHE_SHIFT
;
3056 unsigned int offset
= from
& (PAGE_CACHE_SIZE
-1);
3057 unsigned int blocksize
, max
, pos
;
3058 unsigned int end_of_block
, range_to_discard
;
3060 struct buffer_head
*bh
;
3063 blocksize
= inode
->i_sb
->s_blocksize
;
3064 max
= PAGE_CACHE_SIZE
- offset
;
3066 if (index
!= page
->index
)
3070 * correct length if it does not fall between
3071 * 'from' and the end of the page
3073 if (length
> max
|| length
< 0)
3076 iblock
= index
<< (PAGE_CACHE_SHIFT
- inode
->i_sb
->s_blocksize_bits
);
3078 if (!page_has_buffers(page
)) {
3080 * If the range to be discarded covers a partial block
3081 * we need to get the page buffers. This is because
3082 * partial blocks cannot be released and the page needs
3083 * to be updated with the contents of the block before
3084 * we write the zeros on top of it.
3086 if (!(from
& (blocksize
- 1)) ||
3087 !((from
+ length
) & (blocksize
- 1))) {
3088 create_empty_buffers(page
, blocksize
, 0);
3091 * If there are no partial blocks,
3092 * there is nothing to update,
3093 * so we can return now
3099 /* Find the buffer that contains "offset" */
3100 bh
= page_buffers(page
);
3102 while (offset
>= pos
) {
3103 bh
= bh
->b_this_page
;
3109 while (pos
< offset
+ length
) {
3112 /* The length of space left to zero and unmap */
3113 range_to_discard
= offset
+ length
- pos
;
3115 /* The length of space until the end of the block */
3116 end_of_block
= blocksize
- (pos
& (blocksize
-1));
3119 * Do not unmap or zero past end of block
3120 * for this buffer head
3122 if (range_to_discard
> end_of_block
)
3123 range_to_discard
= end_of_block
;
3127 * Skip this buffer head if we are only zeroing unampped
3128 * regions of the page
3130 if (flags
& EXT4_DISCARD_PARTIAL_PG_ZERO_UNMAPPED
&&
3134 /* If the range is block aligned, unmap */
3135 if (range_to_discard
== blocksize
) {
3136 clear_buffer_dirty(bh
);
3138 clear_buffer_mapped(bh
);
3139 clear_buffer_req(bh
);
3140 clear_buffer_new(bh
);
3141 clear_buffer_delay(bh
);
3142 clear_buffer_unwritten(bh
);
3143 clear_buffer_uptodate(bh
);
3144 zero_user(page
, pos
, range_to_discard
);
3145 BUFFER_TRACE(bh
, "Buffer discarded");
3150 * If this block is not completely contained in the range
3151 * to be discarded, then it is not going to be released. Because
3152 * we need to keep this block, we need to make sure this part
3153 * of the page is uptodate before we modify it by writeing
3154 * partial zeros on it.
3156 if (!buffer_mapped(bh
)) {
3158 * Buffer head must be mapped before we can read
3161 BUFFER_TRACE(bh
, "unmapped");
3162 ext4_get_block(inode
, iblock
, bh
, 0);
3163 /* unmapped? It's a hole - nothing to do */
3164 if (!buffer_mapped(bh
)) {
3165 BUFFER_TRACE(bh
, "still unmapped");
3170 /* Ok, it's mapped. Make sure it's up-to-date */
3171 if (PageUptodate(page
))
3172 set_buffer_uptodate(bh
);
3174 if (!buffer_uptodate(bh
)) {
3176 ll_rw_block(READ
, 1, &bh
);
3178 /* Uhhuh. Read error. Complain and punt.*/
3179 if (!buffer_uptodate(bh
))
3183 if (ext4_should_journal_data(inode
)) {
3184 BUFFER_TRACE(bh
, "get write access");
3185 err
= ext4_journal_get_write_access(handle
, bh
);
3190 zero_user(page
, pos
, range_to_discard
);
3193 if (ext4_should_journal_data(inode
)) {
3194 err
= ext4_handle_dirty_metadata(handle
, inode
, bh
);
3196 mark_buffer_dirty(bh
);
3198 BUFFER_TRACE(bh
, "Partial buffer zeroed");
3200 bh
= bh
->b_this_page
;
3202 pos
+= range_to_discard
;
3209 * ext4_block_truncate_page() zeroes out a mapping from file offset `from'
3210 * up to the end of the block which corresponds to `from'.
3211 * This required during truncate. We need to physically zero the tail end
3212 * of that block so it doesn't yield old data if the file is later grown.
3214 int ext4_block_truncate_page(handle_t
*handle
,
3215 struct address_space
*mapping
, loff_t from
)
3217 unsigned offset
= from
& (PAGE_CACHE_SIZE
-1);
3220 struct inode
*inode
= mapping
->host
;
3222 blocksize
= inode
->i_sb
->s_blocksize
;
3223 length
= blocksize
- (offset
& (blocksize
- 1));
3225 return ext4_block_zero_page_range(handle
, mapping
, from
, length
);
3229 * ext4_block_zero_page_range() zeros out a mapping of length 'length'
3230 * starting from file offset 'from'. The range to be zero'd must
3231 * be contained with in one block. If the specified range exceeds
3232 * the end of the block it will be shortened to end of the block
3233 * that cooresponds to 'from'
3235 int ext4_block_zero_page_range(handle_t
*handle
,
3236 struct address_space
*mapping
, loff_t from
, loff_t length
)
3238 ext4_fsblk_t index
= from
>> PAGE_CACHE_SHIFT
;
3239 unsigned offset
= from
& (PAGE_CACHE_SIZE
-1);
3240 unsigned blocksize
, max
, pos
;
3242 struct inode
*inode
= mapping
->host
;
3243 struct buffer_head
*bh
;
3247 page
= find_or_create_page(mapping
, from
>> PAGE_CACHE_SHIFT
,
3248 mapping_gfp_mask(mapping
) & ~__GFP_FS
);
3252 blocksize
= inode
->i_sb
->s_blocksize
;
3253 max
= blocksize
- (offset
& (blocksize
- 1));
3256 * correct length if it does not fall between
3257 * 'from' and the end of the block
3259 if (length
> max
|| length
< 0)
3262 iblock
= index
<< (PAGE_CACHE_SHIFT
- inode
->i_sb
->s_blocksize_bits
);
3264 if (!page_has_buffers(page
))
3265 create_empty_buffers(page
, blocksize
, 0);
3267 /* Find the buffer that contains "offset" */
3268 bh
= page_buffers(page
);
3270 while (offset
>= pos
) {
3271 bh
= bh
->b_this_page
;
3277 if (buffer_freed(bh
)) {
3278 BUFFER_TRACE(bh
, "freed: skip");
3282 if (!buffer_mapped(bh
)) {
3283 BUFFER_TRACE(bh
, "unmapped");
3284 ext4_get_block(inode
, iblock
, bh
, 0);
3285 /* unmapped? It's a hole - nothing to do */
3286 if (!buffer_mapped(bh
)) {
3287 BUFFER_TRACE(bh
, "still unmapped");
3292 /* Ok, it's mapped. Make sure it's up-to-date */
3293 if (PageUptodate(page
))
3294 set_buffer_uptodate(bh
);
3296 if (!buffer_uptodate(bh
)) {
3298 ll_rw_block(READ
, 1, &bh
);
3300 /* Uhhuh. Read error. Complain and punt. */
3301 if (!buffer_uptodate(bh
))
3305 if (ext4_should_journal_data(inode
)) {
3306 BUFFER_TRACE(bh
, "get write access");
3307 err
= ext4_journal_get_write_access(handle
, bh
);
3312 zero_user(page
, offset
, length
);
3314 BUFFER_TRACE(bh
, "zeroed end of block");
3317 if (ext4_should_journal_data(inode
)) {
3318 err
= ext4_handle_dirty_metadata(handle
, inode
, bh
);
3320 mark_buffer_dirty(bh
);
3324 page_cache_release(page
);
3328 int ext4_can_truncate(struct inode
*inode
)
3330 if (S_ISREG(inode
->i_mode
))
3332 if (S_ISDIR(inode
->i_mode
))
3334 if (S_ISLNK(inode
->i_mode
))
3335 return !ext4_inode_is_fast_symlink(inode
);
3340 * ext4_punch_hole: punches a hole in a file by releaseing the blocks
3341 * associated with the given offset and length
3343 * @inode: File inode
3344 * @offset: The offset where the hole will begin
3345 * @len: The length of the hole
3347 * Returns: 0 on sucess or negative on failure
3350 int ext4_punch_hole(struct file
*file
, loff_t offset
, loff_t length
)
3352 struct inode
*inode
= file
->f_path
.dentry
->d_inode
;
3353 if (!S_ISREG(inode
->i_mode
))
3356 if (!ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
)) {
3357 /* TODO: Add support for non extent hole punching */
3361 if (EXT4_SB(inode
->i_sb
)->s_cluster_ratio
> 1) {
3362 /* TODO: Add support for bigalloc file systems */
3366 return ext4_ext_punch_hole(file
, offset
, length
);
3372 * We block out ext4_get_block() block instantiations across the entire
3373 * transaction, and VFS/VM ensures that ext4_truncate() cannot run
3374 * simultaneously on behalf of the same inode.
3376 * As we work through the truncate and commmit bits of it to the journal there
3377 * is one core, guiding principle: the file's tree must always be consistent on
3378 * disk. We must be able to restart the truncate after a crash.
3380 * The file's tree may be transiently inconsistent in memory (although it
3381 * probably isn't), but whenever we close off and commit a journal transaction,
3382 * the contents of (the filesystem + the journal) must be consistent and
3383 * restartable. It's pretty simple, really: bottom up, right to left (although
3384 * left-to-right works OK too).
3386 * Note that at recovery time, journal replay occurs *before* the restart of
3387 * truncate against the orphan inode list.
3389 * The committed inode has the new, desired i_size (which is the same as
3390 * i_disksize in this case). After a crash, ext4_orphan_cleanup() will see
3391 * that this inode's truncate did not complete and it will again call
3392 * ext4_truncate() to have another go. So there will be instantiated blocks
3393 * to the right of the truncation point in a crashed ext4 filesystem. But
3394 * that's fine - as long as they are linked from the inode, the post-crash
3395 * ext4_truncate() run will find them and release them.
3397 void ext4_truncate(struct inode
*inode
)
3399 trace_ext4_truncate_enter(inode
);
3401 if (!ext4_can_truncate(inode
))
3404 ext4_clear_inode_flag(inode
, EXT4_INODE_EOFBLOCKS
);
3406 if (inode
->i_size
== 0 && !test_opt(inode
->i_sb
, NO_AUTO_DA_ALLOC
))
3407 ext4_set_inode_state(inode
, EXT4_STATE_DA_ALLOC_CLOSE
);
3409 if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
))
3410 ext4_ext_truncate(inode
);
3412 ext4_ind_truncate(inode
);
3414 trace_ext4_truncate_exit(inode
);
3418 * ext4_get_inode_loc returns with an extra refcount against the inode's
3419 * underlying buffer_head on success. If 'in_mem' is true, we have all
3420 * data in memory that is needed to recreate the on-disk version of this
3423 static int __ext4_get_inode_loc(struct inode
*inode
,
3424 struct ext4_iloc
*iloc
, int in_mem
)
3426 struct ext4_group_desc
*gdp
;
3427 struct buffer_head
*bh
;
3428 struct super_block
*sb
= inode
->i_sb
;
3430 int inodes_per_block
, inode_offset
;
3433 if (!ext4_valid_inum(sb
, inode
->i_ino
))
3436 iloc
->block_group
= (inode
->i_ino
- 1) / EXT4_INODES_PER_GROUP(sb
);
3437 gdp
= ext4_get_group_desc(sb
, iloc
->block_group
, NULL
);
3442 * Figure out the offset within the block group inode table
3444 inodes_per_block
= EXT4_SB(sb
)->s_inodes_per_block
;
3445 inode_offset
= ((inode
->i_ino
- 1) %
3446 EXT4_INODES_PER_GROUP(sb
));
3447 block
= ext4_inode_table(sb
, gdp
) + (inode_offset
/ inodes_per_block
);
3448 iloc
->offset
= (inode_offset
% inodes_per_block
) * EXT4_INODE_SIZE(sb
);
3450 bh
= sb_getblk(sb
, block
);
3452 EXT4_ERROR_INODE_BLOCK(inode
, block
,
3453 "unable to read itable block");
3456 if (!buffer_uptodate(bh
)) {
3460 * If the buffer has the write error flag, we have failed
3461 * to write out another inode in the same block. In this
3462 * case, we don't have to read the block because we may
3463 * read the old inode data successfully.
3465 if (buffer_write_io_error(bh
) && !buffer_uptodate(bh
))
3466 set_buffer_uptodate(bh
);
3468 if (buffer_uptodate(bh
)) {
3469 /* someone brought it uptodate while we waited */
3475 * If we have all information of the inode in memory and this
3476 * is the only valid inode in the block, we need not read the
3480 struct buffer_head
*bitmap_bh
;
3483 start
= inode_offset
& ~(inodes_per_block
- 1);
3485 /* Is the inode bitmap in cache? */
3486 bitmap_bh
= sb_getblk(sb
, ext4_inode_bitmap(sb
, gdp
));
3491 * If the inode bitmap isn't in cache then the
3492 * optimisation may end up performing two reads instead
3493 * of one, so skip it.
3495 if (!buffer_uptodate(bitmap_bh
)) {
3499 for (i
= start
; i
< start
+ inodes_per_block
; i
++) {
3500 if (i
== inode_offset
)
3502 if (ext4_test_bit(i
, bitmap_bh
->b_data
))
3506 if (i
== start
+ inodes_per_block
) {
3507 /* all other inodes are free, so skip I/O */
3508 memset(bh
->b_data
, 0, bh
->b_size
);
3509 set_buffer_uptodate(bh
);
3517 * If we need to do any I/O, try to pre-readahead extra
3518 * blocks from the inode table.
3520 if (EXT4_SB(sb
)->s_inode_readahead_blks
) {
3521 ext4_fsblk_t b
, end
, table
;
3524 table
= ext4_inode_table(sb
, gdp
);
3525 /* s_inode_readahead_blks is always a power of 2 */
3526 b
= block
& ~(EXT4_SB(sb
)->s_inode_readahead_blks
-1);
3529 end
= b
+ EXT4_SB(sb
)->s_inode_readahead_blks
;
3530 num
= EXT4_INODES_PER_GROUP(sb
);
3531 if (EXT4_HAS_RO_COMPAT_FEATURE(sb
,
3532 EXT4_FEATURE_RO_COMPAT_GDT_CSUM
))
3533 num
-= ext4_itable_unused_count(sb
, gdp
);
3534 table
+= num
/ inodes_per_block
;
3538 sb_breadahead(sb
, b
++);
3542 * There are other valid inodes in the buffer, this inode
3543 * has in-inode xattrs, or we don't have this inode in memory.
3544 * Read the block from disk.
3546 trace_ext4_load_inode(inode
);
3548 bh
->b_end_io
= end_buffer_read_sync
;
3549 submit_bh(READ_META
, bh
);
3551 if (!buffer_uptodate(bh
)) {
3552 EXT4_ERROR_INODE_BLOCK(inode
, block
,
3553 "unable to read itable block");
3563 int ext4_get_inode_loc(struct inode
*inode
, struct ext4_iloc
*iloc
)
3565 /* We have all inode data except xattrs in memory here. */
3566 return __ext4_get_inode_loc(inode
, iloc
,
3567 !ext4_test_inode_state(inode
, EXT4_STATE_XATTR
));
3570 void ext4_set_inode_flags(struct inode
*inode
)
3572 unsigned int flags
= EXT4_I(inode
)->i_flags
;
3574 inode
->i_flags
&= ~(S_SYNC
|S_APPEND
|S_IMMUTABLE
|S_NOATIME
|S_DIRSYNC
);
3575 if (flags
& EXT4_SYNC_FL
)
3576 inode
->i_flags
|= S_SYNC
;
3577 if (flags
& EXT4_APPEND_FL
)
3578 inode
->i_flags
|= S_APPEND
;
3579 if (flags
& EXT4_IMMUTABLE_FL
)
3580 inode
->i_flags
|= S_IMMUTABLE
;
3581 if (flags
& EXT4_NOATIME_FL
)
3582 inode
->i_flags
|= S_NOATIME
;
3583 if (flags
& EXT4_DIRSYNC_FL
)
3584 inode
->i_flags
|= S_DIRSYNC
;
3587 /* Propagate flags from i_flags to EXT4_I(inode)->i_flags */
3588 void ext4_get_inode_flags(struct ext4_inode_info
*ei
)
3590 unsigned int vfs_fl
;
3591 unsigned long old_fl
, new_fl
;
3594 vfs_fl
= ei
->vfs_inode
.i_flags
;
3595 old_fl
= ei
->i_flags
;
3596 new_fl
= old_fl
& ~(EXT4_SYNC_FL
|EXT4_APPEND_FL
|
3597 EXT4_IMMUTABLE_FL
|EXT4_NOATIME_FL
|
3599 if (vfs_fl
& S_SYNC
)
3600 new_fl
|= EXT4_SYNC_FL
;
3601 if (vfs_fl
& S_APPEND
)
3602 new_fl
|= EXT4_APPEND_FL
;
3603 if (vfs_fl
& S_IMMUTABLE
)
3604 new_fl
|= EXT4_IMMUTABLE_FL
;
3605 if (vfs_fl
& S_NOATIME
)
3606 new_fl
|= EXT4_NOATIME_FL
;
3607 if (vfs_fl
& S_DIRSYNC
)
3608 new_fl
|= EXT4_DIRSYNC_FL
;
3609 } while (cmpxchg(&ei
->i_flags
, old_fl
, new_fl
) != old_fl
);
3612 static blkcnt_t
ext4_inode_blocks(struct ext4_inode
*raw_inode
,
3613 struct ext4_inode_info
*ei
)
3616 struct inode
*inode
= &(ei
->vfs_inode
);
3617 struct super_block
*sb
= inode
->i_sb
;
3619 if (EXT4_HAS_RO_COMPAT_FEATURE(sb
,
3620 EXT4_FEATURE_RO_COMPAT_HUGE_FILE
)) {
3621 /* we are using combined 48 bit field */
3622 i_blocks
= ((u64
)le16_to_cpu(raw_inode
->i_blocks_high
)) << 32 |
3623 le32_to_cpu(raw_inode
->i_blocks_lo
);
3624 if (ext4_test_inode_flag(inode
, EXT4_INODE_HUGE_FILE
)) {
3625 /* i_blocks represent file system block size */
3626 return i_blocks
<< (inode
->i_blkbits
- 9);
3631 return le32_to_cpu(raw_inode
->i_blocks_lo
);
3635 struct inode
*ext4_iget(struct super_block
*sb
, unsigned long ino
)
3637 struct ext4_iloc iloc
;
3638 struct ext4_inode
*raw_inode
;
3639 struct ext4_inode_info
*ei
;
3640 struct inode
*inode
;
3641 journal_t
*journal
= EXT4_SB(sb
)->s_journal
;
3645 inode
= iget_locked(sb
, ino
);
3647 return ERR_PTR(-ENOMEM
);
3648 if (!(inode
->i_state
& I_NEW
))
3654 ret
= __ext4_get_inode_loc(inode
, &iloc
, 0);
3657 raw_inode
= ext4_raw_inode(&iloc
);
3658 inode
->i_mode
= le16_to_cpu(raw_inode
->i_mode
);
3659 inode
->i_uid
= (uid_t
)le16_to_cpu(raw_inode
->i_uid_low
);
3660 inode
->i_gid
= (gid_t
)le16_to_cpu(raw_inode
->i_gid_low
);
3661 if (!(test_opt(inode
->i_sb
, NO_UID32
))) {
3662 inode
->i_uid
|= le16_to_cpu(raw_inode
->i_uid_high
) << 16;
3663 inode
->i_gid
|= le16_to_cpu(raw_inode
->i_gid_high
) << 16;
3665 inode
->i_nlink
= le16_to_cpu(raw_inode
->i_links_count
);
3667 ext4_clear_state_flags(ei
); /* Only relevant on 32-bit archs */
3668 ei
->i_dir_start_lookup
= 0;
3669 ei
->i_dtime
= le32_to_cpu(raw_inode
->i_dtime
);
3670 /* We now have enough fields to check if the inode was active or not.
3671 * This is needed because nfsd might try to access dead inodes
3672 * the test is that same one that e2fsck uses
3673 * NeilBrown 1999oct15
3675 if (inode
->i_nlink
== 0) {
3676 if (inode
->i_mode
== 0 ||
3677 !(EXT4_SB(inode
->i_sb
)->s_mount_state
& EXT4_ORPHAN_FS
)) {
3678 /* this inode is deleted */
3682 /* The only unlinked inodes we let through here have
3683 * valid i_mode and are being read by the orphan
3684 * recovery code: that's fine, we're about to complete
3685 * the process of deleting those. */
3687 ei
->i_flags
= le32_to_cpu(raw_inode
->i_flags
);
3688 inode
->i_blocks
= ext4_inode_blocks(raw_inode
, ei
);
3689 ei
->i_file_acl
= le32_to_cpu(raw_inode
->i_file_acl_lo
);
3690 if (EXT4_HAS_INCOMPAT_FEATURE(sb
, EXT4_FEATURE_INCOMPAT_64BIT
))
3692 ((__u64
)le16_to_cpu(raw_inode
->i_file_acl_high
)) << 32;
3693 inode
->i_size
= ext4_isize(raw_inode
);
3694 ei
->i_disksize
= inode
->i_size
;
3696 ei
->i_reserved_quota
= 0;
3698 inode
->i_generation
= le32_to_cpu(raw_inode
->i_generation
);
3699 ei
->i_block_group
= iloc
.block_group
;
3700 ei
->i_last_alloc_group
= ~0;
3702 * NOTE! The in-memory inode i_data array is in little-endian order
3703 * even on big-endian machines: we do NOT byteswap the block numbers!
3705 for (block
= 0; block
< EXT4_N_BLOCKS
; block
++)
3706 ei
->i_data
[block
] = raw_inode
->i_block
[block
];
3707 INIT_LIST_HEAD(&ei
->i_orphan
);
3710 * Set transaction id's of transactions that have to be committed
3711 * to finish f[data]sync. We set them to currently running transaction
3712 * as we cannot be sure that the inode or some of its metadata isn't
3713 * part of the transaction - the inode could have been reclaimed and
3714 * now it is reread from disk.
3717 transaction_t
*transaction
;
3720 read_lock(&journal
->j_state_lock
);
3721 if (journal
->j_running_transaction
)
3722 transaction
= journal
->j_running_transaction
;
3724 transaction
= journal
->j_committing_transaction
;
3726 tid
= transaction
->t_tid
;
3728 tid
= journal
->j_commit_sequence
;
3729 read_unlock(&journal
->j_state_lock
);
3730 ei
->i_sync_tid
= tid
;
3731 ei
->i_datasync_tid
= tid
;
3734 if (EXT4_INODE_SIZE(inode
->i_sb
) > EXT4_GOOD_OLD_INODE_SIZE
) {
3735 ei
->i_extra_isize
= le16_to_cpu(raw_inode
->i_extra_isize
);
3736 if (EXT4_GOOD_OLD_INODE_SIZE
+ ei
->i_extra_isize
>
3737 EXT4_INODE_SIZE(inode
->i_sb
)) {
3741 if (ei
->i_extra_isize
== 0) {
3742 /* The extra space is currently unused. Use it. */
3743 ei
->i_extra_isize
= sizeof(struct ext4_inode
) -
3744 EXT4_GOOD_OLD_INODE_SIZE
;
3746 __le32
*magic
= (void *)raw_inode
+
3747 EXT4_GOOD_OLD_INODE_SIZE
+
3749 if (*magic
== cpu_to_le32(EXT4_XATTR_MAGIC
))
3750 ext4_set_inode_state(inode
, EXT4_STATE_XATTR
);
3753 ei
->i_extra_isize
= 0;
3755 EXT4_INODE_GET_XTIME(i_ctime
, inode
, raw_inode
);
3756 EXT4_INODE_GET_XTIME(i_mtime
, inode
, raw_inode
);
3757 EXT4_INODE_GET_XTIME(i_atime
, inode
, raw_inode
);
3758 EXT4_EINODE_GET_XTIME(i_crtime
, ei
, raw_inode
);
3760 inode
->i_version
= le32_to_cpu(raw_inode
->i_disk_version
);
3761 if (EXT4_INODE_SIZE(inode
->i_sb
) > EXT4_GOOD_OLD_INODE_SIZE
) {
3762 if (EXT4_FITS_IN_INODE(raw_inode
, ei
, i_version_hi
))
3764 (__u64
)(le32_to_cpu(raw_inode
->i_version_hi
)) << 32;
3768 if (ei
->i_file_acl
&&
3769 !ext4_data_block_valid(EXT4_SB(sb
), ei
->i_file_acl
, 1)) {
3770 EXT4_ERROR_INODE(inode
, "bad extended attribute block %llu",
3774 } else if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
)) {
3775 if (S_ISREG(inode
->i_mode
) || S_ISDIR(inode
->i_mode
) ||
3776 (S_ISLNK(inode
->i_mode
) &&
3777 !ext4_inode_is_fast_symlink(inode
)))
3778 /* Validate extent which is part of inode */
3779 ret
= ext4_ext_check_inode(inode
);
3780 } else if (S_ISREG(inode
->i_mode
) || S_ISDIR(inode
->i_mode
) ||
3781 (S_ISLNK(inode
->i_mode
) &&
3782 !ext4_inode_is_fast_symlink(inode
))) {
3783 /* Validate block references which are part of inode */
3784 ret
= ext4_ind_check_inode(inode
);
3789 if (S_ISREG(inode
->i_mode
)) {
3790 inode
->i_op
= &ext4_file_inode_operations
;
3791 inode
->i_fop
= &ext4_file_operations
;
3792 ext4_set_aops(inode
);
3793 } else if (S_ISDIR(inode
->i_mode
)) {
3794 inode
->i_op
= &ext4_dir_inode_operations
;
3795 inode
->i_fop
= &ext4_dir_operations
;
3796 } else if (S_ISLNK(inode
->i_mode
)) {
3797 if (ext4_inode_is_fast_symlink(inode
)) {
3798 inode
->i_op
= &ext4_fast_symlink_inode_operations
;
3799 nd_terminate_link(ei
->i_data
, inode
->i_size
,
3800 sizeof(ei
->i_data
) - 1);
3802 inode
->i_op
= &ext4_symlink_inode_operations
;
3803 ext4_set_aops(inode
);
3805 } else if (S_ISCHR(inode
->i_mode
) || S_ISBLK(inode
->i_mode
) ||
3806 S_ISFIFO(inode
->i_mode
) || S_ISSOCK(inode
->i_mode
)) {
3807 inode
->i_op
= &ext4_special_inode_operations
;
3808 if (raw_inode
->i_block
[0])
3809 init_special_inode(inode
, inode
->i_mode
,
3810 old_decode_dev(le32_to_cpu(raw_inode
->i_block
[0])));
3812 init_special_inode(inode
, inode
->i_mode
,
3813 new_decode_dev(le32_to_cpu(raw_inode
->i_block
[1])));
3816 EXT4_ERROR_INODE(inode
, "bogus i_mode (%o)", inode
->i_mode
);
3820 ext4_set_inode_flags(inode
);
3821 unlock_new_inode(inode
);
3827 return ERR_PTR(ret
);
3830 static int ext4_inode_blocks_set(handle_t
*handle
,
3831 struct ext4_inode
*raw_inode
,
3832 struct ext4_inode_info
*ei
)
3834 struct inode
*inode
= &(ei
->vfs_inode
);
3835 u64 i_blocks
= inode
->i_blocks
;
3836 struct super_block
*sb
= inode
->i_sb
;
3838 if (i_blocks
<= ~0U) {
3840 * i_blocks can be represnted in a 32 bit variable
3841 * as multiple of 512 bytes
3843 raw_inode
->i_blocks_lo
= cpu_to_le32(i_blocks
);
3844 raw_inode
->i_blocks_high
= 0;
3845 ext4_clear_inode_flag(inode
, EXT4_INODE_HUGE_FILE
);
3848 if (!EXT4_HAS_RO_COMPAT_FEATURE(sb
, EXT4_FEATURE_RO_COMPAT_HUGE_FILE
))
3851 if (i_blocks
<= 0xffffffffffffULL
) {
3853 * i_blocks can be represented in a 48 bit variable
3854 * as multiple of 512 bytes
3856 raw_inode
->i_blocks_lo
= cpu_to_le32(i_blocks
);
3857 raw_inode
->i_blocks_high
= cpu_to_le16(i_blocks
>> 32);
3858 ext4_clear_inode_flag(inode
, EXT4_INODE_HUGE_FILE
);
3860 ext4_set_inode_flag(inode
, EXT4_INODE_HUGE_FILE
);
3861 /* i_block is stored in file system block size */
3862 i_blocks
= i_blocks
>> (inode
->i_blkbits
- 9);
3863 raw_inode
->i_blocks_lo
= cpu_to_le32(i_blocks
);
3864 raw_inode
->i_blocks_high
= cpu_to_le16(i_blocks
>> 32);
3870 * Post the struct inode info into an on-disk inode location in the
3871 * buffer-cache. This gobbles the caller's reference to the
3872 * buffer_head in the inode location struct.
3874 * The caller must have write access to iloc->bh.
3876 static int ext4_do_update_inode(handle_t
*handle
,
3877 struct inode
*inode
,
3878 struct ext4_iloc
*iloc
)
3880 struct ext4_inode
*raw_inode
= ext4_raw_inode(iloc
);
3881 struct ext4_inode_info
*ei
= EXT4_I(inode
);
3882 struct buffer_head
*bh
= iloc
->bh
;
3883 int err
= 0, rc
, block
;
3885 /* For fields not not tracking in the in-memory inode,
3886 * initialise them to zero for new inodes. */
3887 if (ext4_test_inode_state(inode
, EXT4_STATE_NEW
))
3888 memset(raw_inode
, 0, EXT4_SB(inode
->i_sb
)->s_inode_size
);
3890 ext4_get_inode_flags(ei
);
3891 raw_inode
->i_mode
= cpu_to_le16(inode
->i_mode
);
3892 if (!(test_opt(inode
->i_sb
, NO_UID32
))) {
3893 raw_inode
->i_uid_low
= cpu_to_le16(low_16_bits(inode
->i_uid
));
3894 raw_inode
->i_gid_low
= cpu_to_le16(low_16_bits(inode
->i_gid
));
3896 * Fix up interoperability with old kernels. Otherwise, old inodes get
3897 * re-used with the upper 16 bits of the uid/gid intact
3900 raw_inode
->i_uid_high
=
3901 cpu_to_le16(high_16_bits(inode
->i_uid
));
3902 raw_inode
->i_gid_high
=
3903 cpu_to_le16(high_16_bits(inode
->i_gid
));
3905 raw_inode
->i_uid_high
= 0;
3906 raw_inode
->i_gid_high
= 0;
3909 raw_inode
->i_uid_low
=
3910 cpu_to_le16(fs_high2lowuid(inode
->i_uid
));
3911 raw_inode
->i_gid_low
=
3912 cpu_to_le16(fs_high2lowgid(inode
->i_gid
));
3913 raw_inode
->i_uid_high
= 0;
3914 raw_inode
->i_gid_high
= 0;
3916 raw_inode
->i_links_count
= cpu_to_le16(inode
->i_nlink
);
3918 EXT4_INODE_SET_XTIME(i_ctime
, inode
, raw_inode
);
3919 EXT4_INODE_SET_XTIME(i_mtime
, inode
, raw_inode
);
3920 EXT4_INODE_SET_XTIME(i_atime
, inode
, raw_inode
);
3921 EXT4_EINODE_SET_XTIME(i_crtime
, ei
, raw_inode
);
3923 if (ext4_inode_blocks_set(handle
, raw_inode
, ei
))
3925 raw_inode
->i_dtime
= cpu_to_le32(ei
->i_dtime
);
3926 raw_inode
->i_flags
= cpu_to_le32(ei
->i_flags
& 0xFFFFFFFF);
3927 if (EXT4_SB(inode
->i_sb
)->s_es
->s_creator_os
!=
3928 cpu_to_le32(EXT4_OS_HURD
))
3929 raw_inode
->i_file_acl_high
=
3930 cpu_to_le16(ei
->i_file_acl
>> 32);
3931 raw_inode
->i_file_acl_lo
= cpu_to_le32(ei
->i_file_acl
);
3932 ext4_isize_set(raw_inode
, ei
->i_disksize
);
3933 if (ei
->i_disksize
> 0x7fffffffULL
) {
3934 struct super_block
*sb
= inode
->i_sb
;
3935 if (!EXT4_HAS_RO_COMPAT_FEATURE(sb
,
3936 EXT4_FEATURE_RO_COMPAT_LARGE_FILE
) ||
3937 EXT4_SB(sb
)->s_es
->s_rev_level
==
3938 cpu_to_le32(EXT4_GOOD_OLD_REV
)) {
3939 /* If this is the first large file
3940 * created, add a flag to the superblock.
3942 err
= ext4_journal_get_write_access(handle
,
3943 EXT4_SB(sb
)->s_sbh
);
3946 ext4_update_dynamic_rev(sb
);
3947 EXT4_SET_RO_COMPAT_FEATURE(sb
,
3948 EXT4_FEATURE_RO_COMPAT_LARGE_FILE
);
3950 ext4_handle_sync(handle
);
3951 err
= ext4_handle_dirty_metadata(handle
, NULL
,
3952 EXT4_SB(sb
)->s_sbh
);
3955 raw_inode
->i_generation
= cpu_to_le32(inode
->i_generation
);
3956 if (S_ISCHR(inode
->i_mode
) || S_ISBLK(inode
->i_mode
)) {
3957 if (old_valid_dev(inode
->i_rdev
)) {
3958 raw_inode
->i_block
[0] =
3959 cpu_to_le32(old_encode_dev(inode
->i_rdev
));
3960 raw_inode
->i_block
[1] = 0;
3962 raw_inode
->i_block
[0] = 0;
3963 raw_inode
->i_block
[1] =
3964 cpu_to_le32(new_encode_dev(inode
->i_rdev
));
3965 raw_inode
->i_block
[2] = 0;
3968 for (block
= 0; block
< EXT4_N_BLOCKS
; block
++)
3969 raw_inode
->i_block
[block
] = ei
->i_data
[block
];
3971 raw_inode
->i_disk_version
= cpu_to_le32(inode
->i_version
);
3972 if (ei
->i_extra_isize
) {
3973 if (EXT4_FITS_IN_INODE(raw_inode
, ei
, i_version_hi
))
3974 raw_inode
->i_version_hi
=
3975 cpu_to_le32(inode
->i_version
>> 32);
3976 raw_inode
->i_extra_isize
= cpu_to_le16(ei
->i_extra_isize
);
3979 BUFFER_TRACE(bh
, "call ext4_handle_dirty_metadata");
3980 rc
= ext4_handle_dirty_metadata(handle
, NULL
, bh
);
3983 ext4_clear_inode_state(inode
, EXT4_STATE_NEW
);
3985 ext4_update_inode_fsync_trans(handle
, inode
, 0);
3988 ext4_std_error(inode
->i_sb
, err
);
3993 * ext4_write_inode()
3995 * We are called from a few places:
3997 * - Within generic_file_write() for O_SYNC files.
3998 * Here, there will be no transaction running. We wait for any running
3999 * trasnaction to commit.
4001 * - Within sys_sync(), kupdate and such.
4002 * We wait on commit, if tol to.
4004 * - Within prune_icache() (PF_MEMALLOC == true)
4005 * Here we simply return. We can't afford to block kswapd on the
4008 * In all cases it is actually safe for us to return without doing anything,
4009 * because the inode has been copied into a raw inode buffer in
4010 * ext4_mark_inode_dirty(). This is a correctness thing for O_SYNC and for
4013 * Note that we are absolutely dependent upon all inode dirtiers doing the
4014 * right thing: they *must* call mark_inode_dirty() after dirtying info in
4015 * which we are interested.
4017 * It would be a bug for them to not do this. The code:
4019 * mark_inode_dirty(inode)
4021 * inode->i_size = expr;
4023 * is in error because a kswapd-driven write_inode() could occur while
4024 * `stuff()' is running, and the new i_size will be lost. Plus the inode
4025 * will no longer be on the superblock's dirty inode list.
4027 int ext4_write_inode(struct inode
*inode
, struct writeback_control
*wbc
)
4031 if (current
->flags
& PF_MEMALLOC
)
4034 if (EXT4_SB(inode
->i_sb
)->s_journal
) {
4035 if (ext4_journal_current_handle()) {
4036 jbd_debug(1, "called recursively, non-PF_MEMALLOC!\n");
4041 if (wbc
->sync_mode
!= WB_SYNC_ALL
)
4044 err
= ext4_force_commit(inode
->i_sb
);
4046 struct ext4_iloc iloc
;
4048 err
= __ext4_get_inode_loc(inode
, &iloc
, 0);
4051 if (wbc
->sync_mode
== WB_SYNC_ALL
)
4052 sync_dirty_buffer(iloc
.bh
);
4053 if (buffer_req(iloc
.bh
) && !buffer_uptodate(iloc
.bh
)) {
4054 EXT4_ERROR_INODE_BLOCK(inode
, iloc
.bh
->b_blocknr
,
4055 "IO error syncing inode");
4066 * Called from notify_change.
4068 * We want to trap VFS attempts to truncate the file as soon as
4069 * possible. In particular, we want to make sure that when the VFS
4070 * shrinks i_size, we put the inode on the orphan list and modify
4071 * i_disksize immediately, so that during the subsequent flushing of
4072 * dirty pages and freeing of disk blocks, we can guarantee that any
4073 * commit will leave the blocks being flushed in an unused state on
4074 * disk. (On recovery, the inode will get truncated and the blocks will
4075 * be freed, so we have a strong guarantee that no future commit will
4076 * leave these blocks visible to the user.)
4078 * Another thing we have to assure is that if we are in ordered mode
4079 * and inode is still attached to the committing transaction, we must
4080 * we start writeout of all the dirty pages which are being truncated.
4081 * This way we are sure that all the data written in the previous
4082 * transaction are already on disk (truncate waits for pages under
4085 * Called with inode->i_mutex down.
4087 int ext4_setattr(struct dentry
*dentry
, struct iattr
*attr
)
4089 struct inode
*inode
= dentry
->d_inode
;
4092 const unsigned int ia_valid
= attr
->ia_valid
;
4094 error
= inode_change_ok(inode
, attr
);
4098 if (is_quota_modification(inode
, attr
))
4099 dquot_initialize(inode
);
4100 if ((ia_valid
& ATTR_UID
&& attr
->ia_uid
!= inode
->i_uid
) ||
4101 (ia_valid
& ATTR_GID
&& attr
->ia_gid
!= inode
->i_gid
)) {
4104 /* (user+group)*(old+new) structure, inode write (sb,
4105 * inode block, ? - but truncate inode update has it) */
4106 handle
= ext4_journal_start(inode
, (EXT4_MAXQUOTAS_INIT_BLOCKS(inode
->i_sb
)+
4107 EXT4_MAXQUOTAS_DEL_BLOCKS(inode
->i_sb
))+3);
4108 if (IS_ERR(handle
)) {
4109 error
= PTR_ERR(handle
);
4112 error
= dquot_transfer(inode
, attr
);
4114 ext4_journal_stop(handle
);
4117 /* Update corresponding info in inode so that everything is in
4118 * one transaction */
4119 if (attr
->ia_valid
& ATTR_UID
)
4120 inode
->i_uid
= attr
->ia_uid
;
4121 if (attr
->ia_valid
& ATTR_GID
)
4122 inode
->i_gid
= attr
->ia_gid
;
4123 error
= ext4_mark_inode_dirty(handle
, inode
);
4124 ext4_journal_stop(handle
);
4127 if (attr
->ia_valid
& ATTR_SIZE
) {
4128 inode_dio_wait(inode
);
4130 if (!(ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
))) {
4131 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
4133 if (attr
->ia_size
> sbi
->s_bitmap_maxbytes
)
4138 if (S_ISREG(inode
->i_mode
) &&
4139 attr
->ia_valid
& ATTR_SIZE
&&
4140 (attr
->ia_size
< inode
->i_size
)) {
4143 handle
= ext4_journal_start(inode
, 3);
4144 if (IS_ERR(handle
)) {
4145 error
= PTR_ERR(handle
);
4148 if (ext4_handle_valid(handle
)) {
4149 error
= ext4_orphan_add(handle
, inode
);
4152 EXT4_I(inode
)->i_disksize
= attr
->ia_size
;
4153 rc
= ext4_mark_inode_dirty(handle
, inode
);
4156 ext4_journal_stop(handle
);
4158 if (ext4_should_order_data(inode
)) {
4159 error
= ext4_begin_ordered_truncate(inode
,
4162 /* Do as much error cleanup as possible */
4163 handle
= ext4_journal_start(inode
, 3);
4164 if (IS_ERR(handle
)) {
4165 ext4_orphan_del(NULL
, inode
);
4168 ext4_orphan_del(handle
, inode
);
4170 ext4_journal_stop(handle
);
4176 if (attr
->ia_valid
& ATTR_SIZE
) {
4177 if (attr
->ia_size
!= i_size_read(inode
)) {
4178 truncate_setsize(inode
, attr
->ia_size
);
4179 ext4_truncate(inode
);
4180 } else if (ext4_test_inode_flag(inode
, EXT4_INODE_EOFBLOCKS
))
4181 ext4_truncate(inode
);
4185 setattr_copy(inode
, attr
);
4186 mark_inode_dirty(inode
);
4190 * If the call to ext4_truncate failed to get a transaction handle at
4191 * all, we need to clean up the in-core orphan list manually.
4193 if (orphan
&& inode
->i_nlink
)
4194 ext4_orphan_del(NULL
, inode
);
4196 if (!rc
&& (ia_valid
& ATTR_MODE
))
4197 rc
= ext4_acl_chmod(inode
);
4200 ext4_std_error(inode
->i_sb
, error
);
4206 int ext4_getattr(struct vfsmount
*mnt
, struct dentry
*dentry
,
4209 struct inode
*inode
;
4210 unsigned long delalloc_blocks
;
4212 inode
= dentry
->d_inode
;
4213 generic_fillattr(inode
, stat
);
4216 * We can't update i_blocks if the block allocation is delayed
4217 * otherwise in the case of system crash before the real block
4218 * allocation is done, we will have i_blocks inconsistent with
4219 * on-disk file blocks.
4220 * We always keep i_blocks updated together with real
4221 * allocation. But to not confuse with user, stat
4222 * will return the blocks that include the delayed allocation
4223 * blocks for this file.
4225 delalloc_blocks
= EXT4_I(inode
)->i_reserved_data_blocks
;
4227 stat
->blocks
+= (delalloc_blocks
<< inode
->i_sb
->s_blocksize_bits
)>>9;
4231 static int ext4_index_trans_blocks(struct inode
*inode
, int nrblocks
, int chunk
)
4233 if (!(ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
)))
4234 return ext4_ind_trans_blocks(inode
, nrblocks
, chunk
);
4235 return ext4_ext_index_trans_blocks(inode
, nrblocks
, chunk
);
4239 * Account for index blocks, block groups bitmaps and block group
4240 * descriptor blocks if modify datablocks and index blocks
4241 * worse case, the indexs blocks spread over different block groups
4243 * If datablocks are discontiguous, they are possible to spread over
4244 * different block groups too. If they are contiuguous, with flexbg,
4245 * they could still across block group boundary.
4247 * Also account for superblock, inode, quota and xattr blocks
4249 static int ext4_meta_trans_blocks(struct inode
*inode
, int nrblocks
, int chunk
)
4251 ext4_group_t groups
, ngroups
= ext4_get_groups_count(inode
->i_sb
);
4257 * How many index blocks need to touch to modify nrblocks?
4258 * The "Chunk" flag indicating whether the nrblocks is
4259 * physically contiguous on disk
4261 * For Direct IO and fallocate, they calls get_block to allocate
4262 * one single extent at a time, so they could set the "Chunk" flag
4264 idxblocks
= ext4_index_trans_blocks(inode
, nrblocks
, chunk
);
4269 * Now let's see how many group bitmaps and group descriptors need
4279 if (groups
> ngroups
)
4281 if (groups
> EXT4_SB(inode
->i_sb
)->s_gdb_count
)
4282 gdpblocks
= EXT4_SB(inode
->i_sb
)->s_gdb_count
;
4284 /* bitmaps and block group descriptor blocks */
4285 ret
+= groups
+ gdpblocks
;
4287 /* Blocks for super block, inode, quota and xattr blocks */
4288 ret
+= EXT4_META_TRANS_BLOCKS(inode
->i_sb
);
4294 * Calculate the total number of credits to reserve to fit
4295 * the modification of a single pages into a single transaction,
4296 * which may include multiple chunks of block allocations.
4298 * This could be called via ext4_write_begin()
4300 * We need to consider the worse case, when
4301 * one new block per extent.
4303 int ext4_writepage_trans_blocks(struct inode
*inode
)
4305 int bpp
= ext4_journal_blocks_per_page(inode
);
4308 ret
= ext4_meta_trans_blocks(inode
, bpp
, 0);
4310 /* Account for data blocks for journalled mode */
4311 if (ext4_should_journal_data(inode
))
4317 * Calculate the journal credits for a chunk of data modification.
4319 * This is called from DIO, fallocate or whoever calling
4320 * ext4_map_blocks() to map/allocate a chunk of contiguous disk blocks.
4322 * journal buffers for data blocks are not included here, as DIO
4323 * and fallocate do no need to journal data buffers.
4325 int ext4_chunk_trans_blocks(struct inode
*inode
, int nrblocks
)
4327 return ext4_meta_trans_blocks(inode
, nrblocks
, 1);
4331 * The caller must have previously called ext4_reserve_inode_write().
4332 * Give this, we know that the caller already has write access to iloc->bh.
4334 int ext4_mark_iloc_dirty(handle_t
*handle
,
4335 struct inode
*inode
, struct ext4_iloc
*iloc
)
4339 if (test_opt(inode
->i_sb
, I_VERSION
))
4340 inode_inc_iversion(inode
);
4342 /* the do_update_inode consumes one bh->b_count */
4345 /* ext4_do_update_inode() does jbd2_journal_dirty_metadata */
4346 err
= ext4_do_update_inode(handle
, inode
, iloc
);
4352 * On success, We end up with an outstanding reference count against
4353 * iloc->bh. This _must_ be cleaned up later.
4357 ext4_reserve_inode_write(handle_t
*handle
, struct inode
*inode
,
4358 struct ext4_iloc
*iloc
)
4362 err
= ext4_get_inode_loc(inode
, iloc
);
4364 BUFFER_TRACE(iloc
->bh
, "get_write_access");
4365 err
= ext4_journal_get_write_access(handle
, iloc
->bh
);
4371 ext4_std_error(inode
->i_sb
, err
);
4376 * Expand an inode by new_extra_isize bytes.
4377 * Returns 0 on success or negative error number on failure.
4379 static int ext4_expand_extra_isize(struct inode
*inode
,
4380 unsigned int new_extra_isize
,
4381 struct ext4_iloc iloc
,
4384 struct ext4_inode
*raw_inode
;
4385 struct ext4_xattr_ibody_header
*header
;
4387 if (EXT4_I(inode
)->i_extra_isize
>= new_extra_isize
)
4390 raw_inode
= ext4_raw_inode(&iloc
);
4392 header
= IHDR(inode
, raw_inode
);
4394 /* No extended attributes present */
4395 if (!ext4_test_inode_state(inode
, EXT4_STATE_XATTR
) ||
4396 header
->h_magic
!= cpu_to_le32(EXT4_XATTR_MAGIC
)) {
4397 memset((void *)raw_inode
+ EXT4_GOOD_OLD_INODE_SIZE
, 0,
4399 EXT4_I(inode
)->i_extra_isize
= new_extra_isize
;
4403 /* try to expand with EAs present */
4404 return ext4_expand_extra_isize_ea(inode
, new_extra_isize
,
4409 * What we do here is to mark the in-core inode as clean with respect to inode
4410 * dirtiness (it may still be data-dirty).
4411 * This means that the in-core inode may be reaped by prune_icache
4412 * without having to perform any I/O. This is a very good thing,
4413 * because *any* task may call prune_icache - even ones which
4414 * have a transaction open against a different journal.
4416 * Is this cheating? Not really. Sure, we haven't written the
4417 * inode out, but prune_icache isn't a user-visible syncing function.
4418 * Whenever the user wants stuff synced (sys_sync, sys_msync, sys_fsync)
4419 * we start and wait on commits.
4421 * Is this efficient/effective? Well, we're being nice to the system
4422 * by cleaning up our inodes proactively so they can be reaped
4423 * without I/O. But we are potentially leaving up to five seconds'
4424 * worth of inodes floating about which prune_icache wants us to
4425 * write out. One way to fix that would be to get prune_icache()
4426 * to do a write_super() to free up some memory. It has the desired
4429 int ext4_mark_inode_dirty(handle_t
*handle
, struct inode
*inode
)
4431 struct ext4_iloc iloc
;
4432 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
4433 static unsigned int mnt_count
;
4437 trace_ext4_mark_inode_dirty(inode
, _RET_IP_
);
4438 err
= ext4_reserve_inode_write(handle
, inode
, &iloc
);
4439 if (ext4_handle_valid(handle
) &&
4440 EXT4_I(inode
)->i_extra_isize
< sbi
->s_want_extra_isize
&&
4441 !ext4_test_inode_state(inode
, EXT4_STATE_NO_EXPAND
)) {
4443 * We need extra buffer credits since we may write into EA block
4444 * with this same handle. If journal_extend fails, then it will
4445 * only result in a minor loss of functionality for that inode.
4446 * If this is felt to be critical, then e2fsck should be run to
4447 * force a large enough s_min_extra_isize.
4449 if ((jbd2_journal_extend(handle
,
4450 EXT4_DATA_TRANS_BLOCKS(inode
->i_sb
))) == 0) {
4451 ret
= ext4_expand_extra_isize(inode
,
4452 sbi
->s_want_extra_isize
,
4455 ext4_set_inode_state(inode
,
4456 EXT4_STATE_NO_EXPAND
);
4458 le16_to_cpu(sbi
->s_es
->s_mnt_count
)) {
4459 ext4_warning(inode
->i_sb
,
4460 "Unable to expand inode %lu. Delete"
4461 " some EAs or run e2fsck.",
4464 le16_to_cpu(sbi
->s_es
->s_mnt_count
);
4470 err
= ext4_mark_iloc_dirty(handle
, inode
, &iloc
);
4475 * ext4_dirty_inode() is called from __mark_inode_dirty()
4477 * We're really interested in the case where a file is being extended.
4478 * i_size has been changed by generic_commit_write() and we thus need
4479 * to include the updated inode in the current transaction.
4481 * Also, dquot_alloc_block() will always dirty the inode when blocks
4482 * are allocated to the file.
4484 * If the inode is marked synchronous, we don't honour that here - doing
4485 * so would cause a commit on atime updates, which we don't bother doing.
4486 * We handle synchronous inodes at the highest possible level.
4488 void ext4_dirty_inode(struct inode
*inode
, int flags
)
4492 handle
= ext4_journal_start(inode
, 2);
4496 ext4_mark_inode_dirty(handle
, inode
);
4498 ext4_journal_stop(handle
);
4505 * Bind an inode's backing buffer_head into this transaction, to prevent
4506 * it from being flushed to disk early. Unlike
4507 * ext4_reserve_inode_write, this leaves behind no bh reference and
4508 * returns no iloc structure, so the caller needs to repeat the iloc
4509 * lookup to mark the inode dirty later.
4511 static int ext4_pin_inode(handle_t
*handle
, struct inode
*inode
)
4513 struct ext4_iloc iloc
;
4517 err
= ext4_get_inode_loc(inode
, &iloc
);
4519 BUFFER_TRACE(iloc
.bh
, "get_write_access");
4520 err
= jbd2_journal_get_write_access(handle
, iloc
.bh
);
4522 err
= ext4_handle_dirty_metadata(handle
,
4528 ext4_std_error(inode
->i_sb
, err
);
4533 int ext4_change_inode_journal_flag(struct inode
*inode
, int val
)
4540 * We have to be very careful here: changing a data block's
4541 * journaling status dynamically is dangerous. If we write a
4542 * data block to the journal, change the status and then delete
4543 * that block, we risk forgetting to revoke the old log record
4544 * from the journal and so a subsequent replay can corrupt data.
4545 * So, first we make sure that the journal is empty and that
4546 * nobody is changing anything.
4549 journal
= EXT4_JOURNAL(inode
);
4552 if (is_journal_aborted(journal
))
4555 jbd2_journal_lock_updates(journal
);
4556 jbd2_journal_flush(journal
);
4559 * OK, there are no updates running now, and all cached data is
4560 * synced to disk. We are now in a completely consistent state
4561 * which doesn't have anything in the journal, and we know that
4562 * no filesystem updates are running, so it is safe to modify
4563 * the inode's in-core data-journaling state flag now.
4567 ext4_set_inode_flag(inode
, EXT4_INODE_JOURNAL_DATA
);
4569 ext4_clear_inode_flag(inode
, EXT4_INODE_JOURNAL_DATA
);
4570 ext4_set_aops(inode
);
4572 jbd2_journal_unlock_updates(journal
);
4574 /* Finally we can mark the inode as dirty. */
4576 handle
= ext4_journal_start(inode
, 1);
4578 return PTR_ERR(handle
);
4580 err
= ext4_mark_inode_dirty(handle
, inode
);
4581 ext4_handle_sync(handle
);
4582 ext4_journal_stop(handle
);
4583 ext4_std_error(inode
->i_sb
, err
);
4588 static int ext4_bh_unmapped(handle_t
*handle
, struct buffer_head
*bh
)
4590 return !buffer_mapped(bh
);
4593 int ext4_page_mkwrite(struct vm_area_struct
*vma
, struct vm_fault
*vmf
)
4595 struct page
*page
= vmf
->page
;
4599 struct file
*file
= vma
->vm_file
;
4600 struct inode
*inode
= file
->f_path
.dentry
->d_inode
;
4601 struct address_space
*mapping
= inode
->i_mapping
;
4603 get_block_t
*get_block
;
4607 * This check is racy but catches the common case. We rely on
4608 * __block_page_mkwrite() to do a reliable check.
4610 vfs_check_frozen(inode
->i_sb
, SB_FREEZE_WRITE
);
4611 /* Delalloc case is easy... */
4612 if (test_opt(inode
->i_sb
, DELALLOC
) &&
4613 !ext4_should_journal_data(inode
) &&
4614 !ext4_nonda_switch(inode
->i_sb
)) {
4616 ret
= __block_page_mkwrite(vma
, vmf
,
4617 ext4_da_get_block_prep
);
4618 } while (ret
== -ENOSPC
&&
4619 ext4_should_retry_alloc(inode
->i_sb
, &retries
));
4624 size
= i_size_read(inode
);
4625 /* Page got truncated from under us? */
4626 if (page
->mapping
!= mapping
|| page_offset(page
) > size
) {
4628 ret
= VM_FAULT_NOPAGE
;
4632 if (page
->index
== size
>> PAGE_CACHE_SHIFT
)
4633 len
= size
& ~PAGE_CACHE_MASK
;
4635 len
= PAGE_CACHE_SIZE
;
4637 * Return if we have all the buffers mapped. This avoids the need to do
4638 * journal_start/journal_stop which can block and take a long time
4640 if (page_has_buffers(page
)) {
4641 if (!walk_page_buffers(NULL
, page_buffers(page
), 0, len
, NULL
,
4642 ext4_bh_unmapped
)) {
4643 /* Wait so that we don't change page under IO */
4644 wait_on_page_writeback(page
);
4645 ret
= VM_FAULT_LOCKED
;
4650 /* OK, we need to fill the hole... */
4651 if (ext4_should_dioread_nolock(inode
))
4652 get_block
= ext4_get_block_write
;
4654 get_block
= ext4_get_block
;
4656 handle
= ext4_journal_start(inode
, ext4_writepage_trans_blocks(inode
));
4657 if (IS_ERR(handle
)) {
4658 ret
= VM_FAULT_SIGBUS
;
4661 ret
= __block_page_mkwrite(vma
, vmf
, get_block
);
4662 if (!ret
&& ext4_should_journal_data(inode
)) {
4663 if (walk_page_buffers(handle
, page_buffers(page
), 0,
4664 PAGE_CACHE_SIZE
, NULL
, do_journal_get_write_access
)) {
4666 ret
= VM_FAULT_SIGBUS
;
4669 ext4_set_inode_state(inode
, EXT4_STATE_JDATA
);
4671 ext4_journal_stop(handle
);
4672 if (ret
== -ENOSPC
&& ext4_should_retry_alloc(inode
->i_sb
, &retries
))
4675 ret
= block_page_mkwrite_return(ret
);