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 * Goal-directed block allocation by Stephen Tweedie
16 * (sct@redhat.com), 1993, 1998
17 * Big-endian to little-endian byte-swapping/bitmaps by
18 * David S. Miller (davem@caip.rutgers.edu), 1995
19 * 64-bit file support on 64-bit platforms by Jakub Jelinek
20 * (jj@sunsite.ms.mff.cuni.cz)
22 * Assorted race fixes, rewrite of ext4_get_block() by Al Viro, 2000
25 #include <linux/module.h>
27 #include <linux/time.h>
28 #include <linux/jbd2.h>
29 #include <linux/highuid.h>
30 #include <linux/pagemap.h>
31 #include <linux/quotaops.h>
32 #include <linux/string.h>
33 #include <linux/buffer_head.h>
34 #include <linux/writeback.h>
35 #include <linux/pagevec.h>
36 #include <linux/mpage.h>
37 #include <linux/namei.h>
38 #include <linux/uio.h>
39 #include <linux/bio.h>
40 #include <linux/workqueue.h>
41 #include <linux/kernel.h>
42 #include <linux/slab.h>
43 #include <linux/ratelimit.h>
45 #include "ext4_jbd2.h"
48 #include "ext4_extents.h"
50 #include <trace/events/ext4.h>
52 #define MPAGE_DA_EXTENT_TAIL 0x01
54 static inline int ext4_begin_ordered_truncate(struct inode
*inode
,
57 trace_ext4_begin_ordered_truncate(inode
, new_size
);
59 * If jinode is zero, then we never opened the file for
60 * writing, so there's no need to call
61 * jbd2_journal_begin_ordered_truncate() since there's no
62 * outstanding writes we need to flush.
64 if (!EXT4_I(inode
)->jinode
)
66 return jbd2_journal_begin_ordered_truncate(EXT4_JOURNAL(inode
),
67 EXT4_I(inode
)->jinode
,
71 static void ext4_invalidatepage(struct page
*page
, unsigned long offset
);
72 static int noalloc_get_block_write(struct inode
*inode
, sector_t iblock
,
73 struct buffer_head
*bh_result
, int create
);
74 static int ext4_set_bh_endio(struct buffer_head
*bh
, struct inode
*inode
);
75 static void ext4_end_io_buffer_write(struct buffer_head
*bh
, int uptodate
);
76 static int __ext4_journalled_writepage(struct page
*page
, unsigned int len
);
77 static int ext4_bh_delay_or_unwritten(handle_t
*handle
, struct buffer_head
*bh
);
80 * Test whether an inode is a fast symlink.
82 static int ext4_inode_is_fast_symlink(struct inode
*inode
)
84 int ea_blocks
= EXT4_I(inode
)->i_file_acl
?
85 (inode
->i_sb
->s_blocksize
>> 9) : 0;
87 return (S_ISLNK(inode
->i_mode
) && inode
->i_blocks
- ea_blocks
== 0);
91 * Work out how many blocks we need to proceed with the next chunk of a
92 * truncate transaction.
94 static unsigned long blocks_for_truncate(struct inode
*inode
)
98 needed
= inode
->i_blocks
>> (inode
->i_sb
->s_blocksize_bits
- 9);
100 /* Give ourselves just enough room to cope with inodes in which
101 * i_blocks is corrupt: we've seen disk corruptions in the past
102 * which resulted in random data in an inode which looked enough
103 * like a regular file for ext4 to try to delete it. Things
104 * will go a bit crazy if that happens, but at least we should
105 * try not to panic the whole kernel. */
109 /* But we need to bound the transaction so we don't overflow the
111 if (needed
> EXT4_MAX_TRANS_DATA
)
112 needed
= EXT4_MAX_TRANS_DATA
;
114 return EXT4_DATA_TRANS_BLOCKS(inode
->i_sb
) + needed
;
118 * Truncate transactions can be complex and absolutely huge. So we need to
119 * be able to restart the transaction at a conventient checkpoint to make
120 * sure we don't overflow the journal.
122 * start_transaction gets us a new handle for a truncate transaction,
123 * and extend_transaction tries to extend the existing one a bit. If
124 * extend fails, we need to propagate the failure up and restart the
125 * transaction in the top-level truncate loop. --sct
127 static handle_t
*start_transaction(struct inode
*inode
)
131 result
= ext4_journal_start(inode
, blocks_for_truncate(inode
));
135 ext4_std_error(inode
->i_sb
, PTR_ERR(result
));
140 * Try to extend this transaction for the purposes of truncation.
142 * Returns 0 if we managed to create more room. If we can't create more
143 * room, and the transaction must be restarted we return 1.
145 static int try_to_extend_transaction(handle_t
*handle
, struct inode
*inode
)
147 if (!ext4_handle_valid(handle
))
149 if (ext4_handle_has_enough_credits(handle
, EXT4_RESERVE_TRANS_BLOCKS
+1))
151 if (!ext4_journal_extend(handle
, blocks_for_truncate(inode
)))
157 * Restart the transaction associated with *handle. This does a commit,
158 * so before we call here everything must be consistently dirtied against
161 int ext4_truncate_restart_trans(handle_t
*handle
, struct inode
*inode
,
167 * Drop i_data_sem to avoid deadlock with ext4_map_blocks. At this
168 * moment, get_block can be called only for blocks inside i_size since
169 * page cache has been already dropped and writes are blocked by
170 * i_mutex. So we can safely drop the i_data_sem here.
172 BUG_ON(EXT4_JOURNAL(inode
) == NULL
);
173 jbd_debug(2, "restarting handle %p\n", handle
);
174 up_write(&EXT4_I(inode
)->i_data_sem
);
175 ret
= ext4_journal_restart(handle
, blocks_for_truncate(inode
));
176 down_write(&EXT4_I(inode
)->i_data_sem
);
177 ext4_discard_preallocations(inode
);
183 * Called at the last iput() if i_nlink is zero.
185 void ext4_evict_inode(struct inode
*inode
)
190 trace_ext4_evict_inode(inode
);
191 if (inode
->i_nlink
) {
192 truncate_inode_pages(&inode
->i_data
, 0);
196 if (!is_bad_inode(inode
))
197 dquot_initialize(inode
);
199 if (ext4_should_order_data(inode
))
200 ext4_begin_ordered_truncate(inode
, 0);
201 truncate_inode_pages(&inode
->i_data
, 0);
203 if (is_bad_inode(inode
))
206 handle
= ext4_journal_start(inode
, blocks_for_truncate(inode
)+3);
207 if (IS_ERR(handle
)) {
208 ext4_std_error(inode
->i_sb
, PTR_ERR(handle
));
210 * If we're going to skip the normal cleanup, we still need to
211 * make sure that the in-core orphan linked list is properly
214 ext4_orphan_del(NULL
, inode
);
219 ext4_handle_sync(handle
);
221 err
= ext4_mark_inode_dirty(handle
, inode
);
223 ext4_warning(inode
->i_sb
,
224 "couldn't mark inode dirty (err %d)", err
);
228 ext4_truncate(inode
);
231 * ext4_ext_truncate() doesn't reserve any slop when it
232 * restarts journal transactions; therefore there may not be
233 * enough credits left in the handle to remove the inode from
234 * the orphan list and set the dtime field.
236 if (!ext4_handle_has_enough_credits(handle
, 3)) {
237 err
= ext4_journal_extend(handle
, 3);
239 err
= ext4_journal_restart(handle
, 3);
241 ext4_warning(inode
->i_sb
,
242 "couldn't extend journal (err %d)", err
);
244 ext4_journal_stop(handle
);
245 ext4_orphan_del(NULL
, inode
);
251 * Kill off the orphan record which ext4_truncate created.
252 * AKPM: I think this can be inside the above `if'.
253 * Note that ext4_orphan_del() has to be able to cope with the
254 * deletion of a non-existent orphan - this is because we don't
255 * know if ext4_truncate() actually created an orphan record.
256 * (Well, we could do this if we need to, but heck - it works)
258 ext4_orphan_del(handle
, inode
);
259 EXT4_I(inode
)->i_dtime
= get_seconds();
262 * One subtle ordering requirement: if anything has gone wrong
263 * (transaction abort, IO errors, whatever), then we can still
264 * do these next steps (the fs will already have been marked as
265 * having errors), but we can't free the inode if the mark_dirty
268 if (ext4_mark_inode_dirty(handle
, inode
))
269 /* If that failed, just do the required in-core inode clear. */
270 ext4_clear_inode(inode
);
272 ext4_free_inode(handle
, inode
);
273 ext4_journal_stop(handle
);
276 ext4_clear_inode(inode
); /* We must guarantee clearing of inode... */
282 struct buffer_head
*bh
;
285 static inline void add_chain(Indirect
*p
, struct buffer_head
*bh
, __le32
*v
)
287 p
->key
= *(p
->p
= v
);
292 * ext4_block_to_path - parse the block number into array of offsets
293 * @inode: inode in question (we are only interested in its superblock)
294 * @i_block: block number to be parsed
295 * @offsets: array to store the offsets in
296 * @boundary: set this non-zero if the referred-to block is likely to be
297 * followed (on disk) by an indirect block.
299 * To store the locations of file's data ext4 uses a data structure common
300 * for UNIX filesystems - tree of pointers anchored in the inode, with
301 * data blocks at leaves and indirect blocks in intermediate nodes.
302 * This function translates the block number into path in that tree -
303 * return value is the path length and @offsets[n] is the offset of
304 * pointer to (n+1)th node in the nth one. If @block is out of range
305 * (negative or too large) warning is printed and zero returned.
307 * Note: function doesn't find node addresses, so no IO is needed. All
308 * we need to know is the capacity of indirect blocks (taken from the
313 * Portability note: the last comparison (check that we fit into triple
314 * indirect block) is spelled differently, because otherwise on an
315 * architecture with 32-bit longs and 8Kb pages we might get into trouble
316 * if our filesystem had 8Kb blocks. We might use long long, but that would
317 * kill us on x86. Oh, well, at least the sign propagation does not matter -
318 * i_block would have to be negative in the very beginning, so we would not
322 static int ext4_block_to_path(struct inode
*inode
,
324 ext4_lblk_t offsets
[4], int *boundary
)
326 int ptrs
= EXT4_ADDR_PER_BLOCK(inode
->i_sb
);
327 int ptrs_bits
= EXT4_ADDR_PER_BLOCK_BITS(inode
->i_sb
);
328 const long direct_blocks
= EXT4_NDIR_BLOCKS
,
329 indirect_blocks
= ptrs
,
330 double_blocks
= (1 << (ptrs_bits
* 2));
334 if (i_block
< direct_blocks
) {
335 offsets
[n
++] = i_block
;
336 final
= direct_blocks
;
337 } else if ((i_block
-= direct_blocks
) < indirect_blocks
) {
338 offsets
[n
++] = EXT4_IND_BLOCK
;
339 offsets
[n
++] = i_block
;
341 } else if ((i_block
-= indirect_blocks
) < double_blocks
) {
342 offsets
[n
++] = EXT4_DIND_BLOCK
;
343 offsets
[n
++] = i_block
>> ptrs_bits
;
344 offsets
[n
++] = i_block
& (ptrs
- 1);
346 } else if (((i_block
-= double_blocks
) >> (ptrs_bits
* 2)) < ptrs
) {
347 offsets
[n
++] = EXT4_TIND_BLOCK
;
348 offsets
[n
++] = i_block
>> (ptrs_bits
* 2);
349 offsets
[n
++] = (i_block
>> ptrs_bits
) & (ptrs
- 1);
350 offsets
[n
++] = i_block
& (ptrs
- 1);
353 ext4_warning(inode
->i_sb
, "block %lu > max in inode %lu",
354 i_block
+ direct_blocks
+
355 indirect_blocks
+ double_blocks
, inode
->i_ino
);
358 *boundary
= final
- 1 - (i_block
& (ptrs
- 1));
362 static int __ext4_check_blockref(const char *function
, unsigned int line
,
364 __le32
*p
, unsigned int max
)
366 struct ext4_super_block
*es
= EXT4_SB(inode
->i_sb
)->s_es
;
370 while (bref
< p
+max
) {
371 blk
= le32_to_cpu(*bref
++);
373 unlikely(!ext4_data_block_valid(EXT4_SB(inode
->i_sb
),
375 es
->s_last_error_block
= cpu_to_le64(blk
);
376 ext4_error_inode(inode
, function
, line
, blk
,
385 #define ext4_check_indirect_blockref(inode, bh) \
386 __ext4_check_blockref(__func__, __LINE__, inode, \
387 (__le32 *)(bh)->b_data, \
388 EXT4_ADDR_PER_BLOCK((inode)->i_sb))
390 #define ext4_check_inode_blockref(inode) \
391 __ext4_check_blockref(__func__, __LINE__, inode, \
392 EXT4_I(inode)->i_data, \
396 * ext4_get_branch - read the chain of indirect blocks leading to data
397 * @inode: inode in question
398 * @depth: depth of the chain (1 - direct pointer, etc.)
399 * @offsets: offsets of pointers in inode/indirect blocks
400 * @chain: place to store the result
401 * @err: here we store the error value
403 * Function fills the array of triples <key, p, bh> and returns %NULL
404 * if everything went OK or the pointer to the last filled triple
405 * (incomplete one) otherwise. Upon the return chain[i].key contains
406 * the number of (i+1)-th block in the chain (as it is stored in memory,
407 * i.e. little-endian 32-bit), chain[i].p contains the address of that
408 * number (it points into struct inode for i==0 and into the bh->b_data
409 * for i>0) and chain[i].bh points to the buffer_head of i-th indirect
410 * block for i>0 and NULL for i==0. In other words, it holds the block
411 * numbers of the chain, addresses they were taken from (and where we can
412 * verify that chain did not change) and buffer_heads hosting these
415 * Function stops when it stumbles upon zero pointer (absent block)
416 * (pointer to last triple returned, *@err == 0)
417 * or when it gets an IO error reading an indirect block
418 * (ditto, *@err == -EIO)
419 * or when it reads all @depth-1 indirect blocks successfully and finds
420 * the whole chain, all way to the data (returns %NULL, *err == 0).
422 * Need to be called with
423 * down_read(&EXT4_I(inode)->i_data_sem)
425 static Indirect
*ext4_get_branch(struct inode
*inode
, int depth
,
426 ext4_lblk_t
*offsets
,
427 Indirect chain
[4], int *err
)
429 struct super_block
*sb
= inode
->i_sb
;
431 struct buffer_head
*bh
;
434 /* i_data is not going away, no lock needed */
435 add_chain(chain
, NULL
, EXT4_I(inode
)->i_data
+ *offsets
);
439 bh
= sb_getblk(sb
, le32_to_cpu(p
->key
));
443 if (!bh_uptodate_or_lock(bh
)) {
444 if (bh_submit_read(bh
) < 0) {
448 /* validate block references */
449 if (ext4_check_indirect_blockref(inode
, bh
)) {
455 add_chain(++p
, bh
, (__le32
*)bh
->b_data
+ *++offsets
);
469 * ext4_find_near - find a place for allocation with sufficient locality
471 * @ind: descriptor of indirect block.
473 * This function returns the preferred place for block allocation.
474 * It is used when heuristic for sequential allocation fails.
476 * + if there is a block to the left of our position - allocate near it.
477 * + if pointer will live in indirect block - allocate near that block.
478 * + if pointer will live in inode - allocate in the same
481 * In the latter case we colour the starting block by the callers PID to
482 * prevent it from clashing with concurrent allocations for a different inode
483 * in the same block group. The PID is used here so that functionally related
484 * files will be close-by on-disk.
486 * Caller must make sure that @ind is valid and will stay that way.
488 static ext4_fsblk_t
ext4_find_near(struct inode
*inode
, Indirect
*ind
)
490 struct ext4_inode_info
*ei
= EXT4_I(inode
);
491 __le32
*start
= ind
->bh
? (__le32
*) ind
->bh
->b_data
: ei
->i_data
;
493 ext4_fsblk_t bg_start
;
494 ext4_fsblk_t last_block
;
495 ext4_grpblk_t colour
;
496 ext4_group_t block_group
;
497 int flex_size
= ext4_flex_bg_size(EXT4_SB(inode
->i_sb
));
499 /* Try to find previous block */
500 for (p
= ind
->p
- 1; p
>= start
; p
--) {
502 return le32_to_cpu(*p
);
505 /* No such thing, so let's try location of indirect block */
507 return ind
->bh
->b_blocknr
;
510 * It is going to be referred to from the inode itself? OK, just put it
511 * into the same cylinder group then.
513 block_group
= ei
->i_block_group
;
514 if (flex_size
>= EXT4_FLEX_SIZE_DIR_ALLOC_SCHEME
) {
515 block_group
&= ~(flex_size
-1);
516 if (S_ISREG(inode
->i_mode
))
519 bg_start
= ext4_group_first_block_no(inode
->i_sb
, block_group
);
520 last_block
= ext4_blocks_count(EXT4_SB(inode
->i_sb
)->s_es
) - 1;
523 * If we are doing delayed allocation, we don't need take
524 * colour into account.
526 if (test_opt(inode
->i_sb
, DELALLOC
))
529 if (bg_start
+ EXT4_BLOCKS_PER_GROUP(inode
->i_sb
) <= last_block
)
530 colour
= (current
->pid
% 16) *
531 (EXT4_BLOCKS_PER_GROUP(inode
->i_sb
) / 16);
533 colour
= (current
->pid
% 16) * ((last_block
- bg_start
) / 16);
534 return bg_start
+ colour
;
538 * ext4_find_goal - find a preferred place for allocation.
540 * @block: block we want
541 * @partial: pointer to the last triple within a chain
543 * Normally this function find the preferred place for block allocation,
545 * Because this is only used for non-extent files, we limit the block nr
548 static ext4_fsblk_t
ext4_find_goal(struct inode
*inode
, ext4_lblk_t block
,
554 * XXX need to get goal block from mballoc's data structures
557 goal
= ext4_find_near(inode
, partial
);
558 goal
= goal
& EXT4_MAX_BLOCK_FILE_PHYS
;
563 * ext4_blks_to_allocate - Look up the block map and count the number
564 * of direct blocks need to be allocated for the given branch.
566 * @branch: chain of indirect blocks
567 * @k: number of blocks need for indirect blocks
568 * @blks: number of data blocks to be mapped.
569 * @blocks_to_boundary: the offset in the indirect block
571 * return the total number of blocks to be allocate, including the
572 * direct and indirect blocks.
574 static int ext4_blks_to_allocate(Indirect
*branch
, int k
, unsigned int blks
,
575 int blocks_to_boundary
)
577 unsigned int count
= 0;
580 * Simple case, [t,d]Indirect block(s) has not allocated yet
581 * then it's clear blocks on that path have not allocated
584 /* right now we don't handle cross boundary allocation */
585 if (blks
< blocks_to_boundary
+ 1)
588 count
+= blocks_to_boundary
+ 1;
593 while (count
< blks
&& count
<= blocks_to_boundary
&&
594 le32_to_cpu(*(branch
[0].p
+ count
)) == 0) {
601 * ext4_alloc_blocks: multiple allocate blocks needed for a branch
602 * @handle: handle for this transaction
603 * @inode: inode which needs allocated blocks
604 * @iblock: the logical block to start allocated at
605 * @goal: preferred physical block of allocation
606 * @indirect_blks: the number of blocks need to allocate for indirect
608 * @blks: number of desired blocks
609 * @new_blocks: on return it will store the new block numbers for
610 * the indirect blocks(if needed) and the first direct block,
611 * @err: on return it will store the error code
613 * This function will return the number of blocks allocated as
614 * requested by the passed-in parameters.
616 static int ext4_alloc_blocks(handle_t
*handle
, struct inode
*inode
,
617 ext4_lblk_t iblock
, ext4_fsblk_t goal
,
618 int indirect_blks
, int blks
,
619 ext4_fsblk_t new_blocks
[4], int *err
)
621 struct ext4_allocation_request ar
;
623 unsigned long count
= 0, blk_allocated
= 0;
625 ext4_fsblk_t current_block
= 0;
629 * Here we try to allocate the requested multiple blocks at once,
630 * on a best-effort basis.
631 * To build a branch, we should allocate blocks for
632 * the indirect blocks(if not allocated yet), and at least
633 * the first direct block of this branch. That's the
634 * minimum number of blocks need to allocate(required)
636 /* first we try to allocate the indirect blocks */
637 target
= indirect_blks
;
640 /* allocating blocks for indirect blocks and direct blocks */
641 current_block
= ext4_new_meta_blocks(handle
, inode
,
646 if (unlikely(current_block
+ count
> EXT4_MAX_BLOCK_FILE_PHYS
)) {
647 EXT4_ERROR_INODE(inode
,
648 "current_block %llu + count %lu > %d!",
649 current_block
, count
,
650 EXT4_MAX_BLOCK_FILE_PHYS
);
656 /* allocate blocks for indirect blocks */
657 while (index
< indirect_blks
&& count
) {
658 new_blocks
[index
++] = current_block
++;
663 * save the new block number
664 * for the first direct block
666 new_blocks
[index
] = current_block
;
667 printk(KERN_INFO
"%s returned more blocks than "
668 "requested\n", __func__
);
674 target
= blks
- count
;
675 blk_allocated
= count
;
678 /* Now allocate data blocks */
679 memset(&ar
, 0, sizeof(ar
));
684 if (S_ISREG(inode
->i_mode
))
685 /* enable in-core preallocation only for regular files */
686 ar
.flags
= EXT4_MB_HINT_DATA
;
688 current_block
= ext4_mb_new_blocks(handle
, &ar
, err
);
689 if (unlikely(current_block
+ ar
.len
> EXT4_MAX_BLOCK_FILE_PHYS
)) {
690 EXT4_ERROR_INODE(inode
,
691 "current_block %llu + ar.len %d > %d!",
692 current_block
, ar
.len
,
693 EXT4_MAX_BLOCK_FILE_PHYS
);
698 if (*err
&& (target
== blks
)) {
700 * if the allocation failed and we didn't allocate
706 if (target
== blks
) {
708 * save the new block number
709 * for the first direct block
711 new_blocks
[index
] = current_block
;
713 blk_allocated
+= ar
.len
;
716 /* total number of blocks allocated for direct blocks */
721 for (i
= 0; i
< index
; i
++)
722 ext4_free_blocks(handle
, inode
, 0, new_blocks
[i
], 1, 0);
727 * ext4_alloc_branch - allocate and set up a chain of blocks.
728 * @handle: handle for this transaction
730 * @indirect_blks: number of allocated indirect blocks
731 * @blks: number of allocated direct blocks
732 * @goal: preferred place for allocation
733 * @offsets: offsets (in the blocks) to store the pointers to next.
734 * @branch: place to store the chain in.
736 * This function allocates blocks, zeroes out all but the last one,
737 * links them into chain and (if we are synchronous) writes them to disk.
738 * In other words, it prepares a branch that can be spliced onto the
739 * inode. It stores the information about that chain in the branch[], in
740 * the same format as ext4_get_branch() would do. We are calling it after
741 * we had read the existing part of chain and partial points to the last
742 * triple of that (one with zero ->key). Upon the exit we have the same
743 * picture as after the successful ext4_get_block(), except that in one
744 * place chain is disconnected - *branch->p is still zero (we did not
745 * set the last link), but branch->key contains the number that should
746 * be placed into *branch->p to fill that gap.
748 * If allocation fails we free all blocks we've allocated (and forget
749 * their buffer_heads) and return the error value the from failed
750 * ext4_alloc_block() (normally -ENOSPC). Otherwise we set the chain
751 * as described above and return 0.
753 static int ext4_alloc_branch(handle_t
*handle
, struct inode
*inode
,
754 ext4_lblk_t iblock
, int indirect_blks
,
755 int *blks
, ext4_fsblk_t goal
,
756 ext4_lblk_t
*offsets
, Indirect
*branch
)
758 int blocksize
= inode
->i_sb
->s_blocksize
;
761 struct buffer_head
*bh
;
763 ext4_fsblk_t new_blocks
[4];
764 ext4_fsblk_t current_block
;
766 num
= ext4_alloc_blocks(handle
, inode
, iblock
, goal
, indirect_blks
,
767 *blks
, new_blocks
, &err
);
771 branch
[0].key
= cpu_to_le32(new_blocks
[0]);
773 * metadata blocks and data blocks are allocated.
775 for (n
= 1; n
<= indirect_blks
; n
++) {
777 * Get buffer_head for parent block, zero it out
778 * and set the pointer to new one, then send
781 bh
= sb_getblk(inode
->i_sb
, new_blocks
[n
-1]);
789 BUFFER_TRACE(bh
, "call get_create_access");
790 err
= ext4_journal_get_create_access(handle
, bh
);
792 /* Don't brelse(bh) here; it's done in
793 * ext4_journal_forget() below */
798 memset(bh
->b_data
, 0, blocksize
);
799 branch
[n
].p
= (__le32
*) bh
->b_data
+ offsets
[n
];
800 branch
[n
].key
= cpu_to_le32(new_blocks
[n
]);
801 *branch
[n
].p
= branch
[n
].key
;
802 if (n
== indirect_blks
) {
803 current_block
= new_blocks
[n
];
805 * End of chain, update the last new metablock of
806 * the chain to point to the new allocated
807 * data blocks numbers
809 for (i
= 1; i
< num
; i
++)
810 *(branch
[n
].p
+ i
) = cpu_to_le32(++current_block
);
812 BUFFER_TRACE(bh
, "marking uptodate");
813 set_buffer_uptodate(bh
);
816 BUFFER_TRACE(bh
, "call ext4_handle_dirty_metadata");
817 err
= ext4_handle_dirty_metadata(handle
, inode
, bh
);
824 /* Allocation failed, free what we already allocated */
825 ext4_free_blocks(handle
, inode
, 0, new_blocks
[0], 1, 0);
826 for (i
= 1; i
<= n
; i
++) {
828 * branch[i].bh is newly allocated, so there is no
829 * need to revoke the block, which is why we don't
830 * need to set EXT4_FREE_BLOCKS_METADATA.
832 ext4_free_blocks(handle
, inode
, 0, new_blocks
[i
], 1,
833 EXT4_FREE_BLOCKS_FORGET
);
835 for (i
= n
+1; i
< indirect_blks
; i
++)
836 ext4_free_blocks(handle
, inode
, 0, new_blocks
[i
], 1, 0);
838 ext4_free_blocks(handle
, inode
, 0, new_blocks
[i
], num
, 0);
844 * ext4_splice_branch - splice the allocated branch onto inode.
845 * @handle: handle for this transaction
847 * @block: (logical) number of block we are adding
848 * @chain: chain of indirect blocks (with a missing link - see
850 * @where: location of missing link
851 * @num: number of indirect blocks we are adding
852 * @blks: number of direct blocks we are adding
854 * This function fills the missing link and does all housekeeping needed in
855 * inode (->i_blocks, etc.). In case of success we end up with the full
856 * chain to new block and return 0.
858 static int ext4_splice_branch(handle_t
*handle
, struct inode
*inode
,
859 ext4_lblk_t block
, Indirect
*where
, int num
,
864 ext4_fsblk_t current_block
;
867 * If we're splicing into a [td]indirect block (as opposed to the
868 * inode) then we need to get write access to the [td]indirect block
872 BUFFER_TRACE(where
->bh
, "get_write_access");
873 err
= ext4_journal_get_write_access(handle
, where
->bh
);
879 *where
->p
= where
->key
;
882 * Update the host buffer_head or inode to point to more just allocated
883 * direct blocks blocks
885 if (num
== 0 && blks
> 1) {
886 current_block
= le32_to_cpu(where
->key
) + 1;
887 for (i
= 1; i
< blks
; i
++)
888 *(where
->p
+ i
) = cpu_to_le32(current_block
++);
891 /* We are done with atomic stuff, now do the rest of housekeeping */
892 /* had we spliced it onto indirect block? */
895 * If we spliced it onto an indirect block, we haven't
896 * altered the inode. Note however that if it is being spliced
897 * onto an indirect block at the very end of the file (the
898 * file is growing) then we *will* alter the inode to reflect
899 * the new i_size. But that is not done here - it is done in
900 * generic_commit_write->__mark_inode_dirty->ext4_dirty_inode.
902 jbd_debug(5, "splicing indirect only\n");
903 BUFFER_TRACE(where
->bh
, "call ext4_handle_dirty_metadata");
904 err
= ext4_handle_dirty_metadata(handle
, inode
, where
->bh
);
909 * OK, we spliced it into the inode itself on a direct block.
911 ext4_mark_inode_dirty(handle
, inode
);
912 jbd_debug(5, "splicing direct\n");
917 for (i
= 1; i
<= num
; i
++) {
919 * branch[i].bh is newly allocated, so there is no
920 * need to revoke the block, which is why we don't
921 * need to set EXT4_FREE_BLOCKS_METADATA.
923 ext4_free_blocks(handle
, inode
, where
[i
].bh
, 0, 1,
924 EXT4_FREE_BLOCKS_FORGET
);
926 ext4_free_blocks(handle
, inode
, 0, le32_to_cpu(where
[num
].key
),
933 * The ext4_ind_map_blocks() function handles non-extents inodes
934 * (i.e., using the traditional indirect/double-indirect i_blocks
935 * scheme) for ext4_map_blocks().
937 * Allocation strategy is simple: if we have to allocate something, we will
938 * have to go the whole way to leaf. So let's do it before attaching anything
939 * to tree, set linkage between the newborn blocks, write them if sync is
940 * required, recheck the path, free and repeat if check fails, otherwise
941 * set the last missing link (that will protect us from any truncate-generated
942 * removals - all blocks on the path are immune now) and possibly force the
943 * write on the parent block.
944 * That has a nice additional property: no special recovery from the failed
945 * allocations is needed - we simply release blocks and do not touch anything
946 * reachable from inode.
948 * `handle' can be NULL if create == 0.
950 * return > 0, # of blocks mapped or allocated.
951 * return = 0, if plain lookup failed.
952 * return < 0, error case.
954 * The ext4_ind_get_blocks() function should be called with
955 * down_write(&EXT4_I(inode)->i_data_sem) if allocating filesystem
956 * blocks (i.e., flags has EXT4_GET_BLOCKS_CREATE set) or
957 * down_read(&EXT4_I(inode)->i_data_sem) if not allocating file system
960 static int ext4_ind_map_blocks(handle_t
*handle
, struct inode
*inode
,
961 struct ext4_map_blocks
*map
,
965 ext4_lblk_t offsets
[4];
970 int blocks_to_boundary
= 0;
973 ext4_fsblk_t first_block
= 0;
975 J_ASSERT(!(ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
)));
976 J_ASSERT(handle
!= NULL
|| (flags
& EXT4_GET_BLOCKS_CREATE
) == 0);
977 depth
= ext4_block_to_path(inode
, map
->m_lblk
, offsets
,
978 &blocks_to_boundary
);
983 partial
= ext4_get_branch(inode
, depth
, offsets
, chain
, &err
);
985 /* Simplest case - block found, no allocation needed */
987 first_block
= le32_to_cpu(chain
[depth
- 1].key
);
990 while (count
< map
->m_len
&& count
<= blocks_to_boundary
) {
993 blk
= le32_to_cpu(*(chain
[depth
-1].p
+ count
));
995 if (blk
== first_block
+ count
)
1003 /* Next simple case - plain lookup or failed read of indirect block */
1004 if ((flags
& EXT4_GET_BLOCKS_CREATE
) == 0 || err
== -EIO
)
1008 * Okay, we need to do block allocation.
1010 goal
= ext4_find_goal(inode
, map
->m_lblk
, partial
);
1012 /* the number of blocks need to allocate for [d,t]indirect blocks */
1013 indirect_blks
= (chain
+ depth
) - partial
- 1;
1016 * Next look up the indirect map to count the totoal number of
1017 * direct blocks to allocate for this branch.
1019 count
= ext4_blks_to_allocate(partial
, indirect_blks
,
1020 map
->m_len
, blocks_to_boundary
);
1022 * Block out ext4_truncate while we alter the tree
1024 err
= ext4_alloc_branch(handle
, inode
, map
->m_lblk
, indirect_blks
,
1026 offsets
+ (partial
- chain
), partial
);
1029 * The ext4_splice_branch call will free and forget any buffers
1030 * on the new chain if there is a failure, but that risks using
1031 * up transaction credits, especially for bitmaps where the
1032 * credits cannot be returned. Can we handle this somehow? We
1033 * may need to return -EAGAIN upwards in the worst case. --sct
1036 err
= ext4_splice_branch(handle
, inode
, map
->m_lblk
,
1037 partial
, indirect_blks
, count
);
1041 map
->m_flags
|= EXT4_MAP_NEW
;
1043 ext4_update_inode_fsync_trans(handle
, inode
, 1);
1045 map
->m_flags
|= EXT4_MAP_MAPPED
;
1046 map
->m_pblk
= le32_to_cpu(chain
[depth
-1].key
);
1048 if (count
> blocks_to_boundary
)
1049 map
->m_flags
|= EXT4_MAP_BOUNDARY
;
1051 /* Clean up and exit */
1052 partial
= chain
+ depth
- 1; /* the whole chain */
1054 while (partial
> chain
) {
1055 BUFFER_TRACE(partial
->bh
, "call brelse");
1056 brelse(partial
->bh
);
1064 qsize_t
*ext4_get_reserved_space(struct inode
*inode
)
1066 return &EXT4_I(inode
)->i_reserved_quota
;
1071 * Calculate the number of metadata blocks need to reserve
1072 * to allocate a new block at @lblocks for non extent file based file
1074 static int ext4_indirect_calc_metadata_amount(struct inode
*inode
,
1077 struct ext4_inode_info
*ei
= EXT4_I(inode
);
1078 sector_t dind_mask
= ~((sector_t
)EXT4_ADDR_PER_BLOCK(inode
->i_sb
) - 1);
1081 if (lblock
< EXT4_NDIR_BLOCKS
)
1084 lblock
-= EXT4_NDIR_BLOCKS
;
1086 if (ei
->i_da_metadata_calc_len
&&
1087 (lblock
& dind_mask
) == ei
->i_da_metadata_calc_last_lblock
) {
1088 ei
->i_da_metadata_calc_len
++;
1091 ei
->i_da_metadata_calc_last_lblock
= lblock
& dind_mask
;
1092 ei
->i_da_metadata_calc_len
= 1;
1093 blk_bits
= order_base_2(lblock
);
1094 return (blk_bits
/ EXT4_ADDR_PER_BLOCK_BITS(inode
->i_sb
)) + 1;
1098 * Calculate the number of metadata blocks need to reserve
1099 * to allocate a block located at @lblock
1101 static int ext4_calc_metadata_amount(struct inode
*inode
, ext4_lblk_t lblock
)
1103 if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
))
1104 return ext4_ext_calc_metadata_amount(inode
, lblock
);
1106 return ext4_indirect_calc_metadata_amount(inode
, lblock
);
1110 * Called with i_data_sem down, which is important since we can call
1111 * ext4_discard_preallocations() from here.
1113 void ext4_da_update_reserve_space(struct inode
*inode
,
1114 int used
, int quota_claim
)
1116 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
1117 struct ext4_inode_info
*ei
= EXT4_I(inode
);
1119 spin_lock(&ei
->i_block_reservation_lock
);
1120 trace_ext4_da_update_reserve_space(inode
, used
);
1121 if (unlikely(used
> ei
->i_reserved_data_blocks
)) {
1122 ext4_msg(inode
->i_sb
, KERN_NOTICE
, "%s: ino %lu, used %d "
1123 "with only %d reserved data blocks\n",
1124 __func__
, inode
->i_ino
, used
,
1125 ei
->i_reserved_data_blocks
);
1127 used
= ei
->i_reserved_data_blocks
;
1130 /* Update per-inode reservations */
1131 ei
->i_reserved_data_blocks
-= used
;
1132 ei
->i_reserved_meta_blocks
-= ei
->i_allocated_meta_blocks
;
1133 percpu_counter_sub(&sbi
->s_dirtyblocks_counter
,
1134 used
+ ei
->i_allocated_meta_blocks
);
1135 ei
->i_allocated_meta_blocks
= 0;
1137 if (ei
->i_reserved_data_blocks
== 0) {
1139 * We can release all of the reserved metadata blocks
1140 * only when we have written all of the delayed
1141 * allocation blocks.
1143 percpu_counter_sub(&sbi
->s_dirtyblocks_counter
,
1144 ei
->i_reserved_meta_blocks
);
1145 ei
->i_reserved_meta_blocks
= 0;
1146 ei
->i_da_metadata_calc_len
= 0;
1148 spin_unlock(&EXT4_I(inode
)->i_block_reservation_lock
);
1150 /* Update quota subsystem for data blocks */
1152 dquot_claim_block(inode
, used
);
1155 * We did fallocate with an offset that is already delayed
1156 * allocated. So on delayed allocated writeback we should
1157 * not re-claim the quota for fallocated blocks.
1159 dquot_release_reservation_block(inode
, used
);
1163 * If we have done all the pending block allocations and if
1164 * there aren't any writers on the inode, we can discard the
1165 * inode's preallocations.
1167 if ((ei
->i_reserved_data_blocks
== 0) &&
1168 (atomic_read(&inode
->i_writecount
) == 0))
1169 ext4_discard_preallocations(inode
);
1172 static int __check_block_validity(struct inode
*inode
, const char *func
,
1174 struct ext4_map_blocks
*map
)
1176 if (!ext4_data_block_valid(EXT4_SB(inode
->i_sb
), map
->m_pblk
,
1178 ext4_error_inode(inode
, func
, line
, map
->m_pblk
,
1179 "lblock %lu mapped to illegal pblock "
1180 "(length %d)", (unsigned long) map
->m_lblk
,
1187 #define check_block_validity(inode, map) \
1188 __check_block_validity((inode), __func__, __LINE__, (map))
1191 * Return the number of contiguous dirty pages in a given inode
1192 * starting at page frame idx.
1194 static pgoff_t
ext4_num_dirty_pages(struct inode
*inode
, pgoff_t idx
,
1195 unsigned int max_pages
)
1197 struct address_space
*mapping
= inode
->i_mapping
;
1199 struct pagevec pvec
;
1201 int i
, nr_pages
, done
= 0;
1205 pagevec_init(&pvec
, 0);
1208 nr_pages
= pagevec_lookup_tag(&pvec
, mapping
, &index
,
1209 PAGECACHE_TAG_DIRTY
,
1210 (pgoff_t
)PAGEVEC_SIZE
);
1213 for (i
= 0; i
< nr_pages
; i
++) {
1214 struct page
*page
= pvec
.pages
[i
];
1215 struct buffer_head
*bh
, *head
;
1218 if (unlikely(page
->mapping
!= mapping
) ||
1220 PageWriteback(page
) ||
1221 page
->index
!= idx
) {
1226 if (page_has_buffers(page
)) {
1227 bh
= head
= page_buffers(page
);
1229 if (!buffer_delay(bh
) &&
1230 !buffer_unwritten(bh
))
1232 bh
= bh
->b_this_page
;
1233 } while (!done
&& (bh
!= head
));
1240 if (num
>= max_pages
) {
1245 pagevec_release(&pvec
);
1251 * The ext4_map_blocks() function tries to look up the requested blocks,
1252 * and returns if the blocks are already mapped.
1254 * Otherwise it takes the write lock of the i_data_sem and allocate blocks
1255 * and store the allocated blocks in the result buffer head and mark it
1258 * If file type is extents based, it will call ext4_ext_map_blocks(),
1259 * Otherwise, call with ext4_ind_map_blocks() to handle indirect mapping
1262 * On success, it returns the number of blocks being mapped or allocate.
1263 * if create==0 and the blocks are pre-allocated and uninitialized block,
1264 * the result buffer head is unmapped. If the create ==1, it will make sure
1265 * the buffer head is mapped.
1267 * It returns 0 if plain look up failed (blocks have not been allocated), in
1268 * that casem, buffer head is unmapped
1270 * It returns the error in case of allocation failure.
1272 int ext4_map_blocks(handle_t
*handle
, struct inode
*inode
,
1273 struct ext4_map_blocks
*map
, int flags
)
1278 ext_debug("ext4_map_blocks(): inode %lu, flag %d, max_blocks %u,"
1279 "logical block %lu\n", inode
->i_ino
, flags
, map
->m_len
,
1280 (unsigned long) map
->m_lblk
);
1282 * Try to see if we can get the block without requesting a new
1283 * file system block.
1285 down_read((&EXT4_I(inode
)->i_data_sem
));
1286 if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
)) {
1287 retval
= ext4_ext_map_blocks(handle
, inode
, map
, 0);
1289 retval
= ext4_ind_map_blocks(handle
, inode
, map
, 0);
1291 up_read((&EXT4_I(inode
)->i_data_sem
));
1293 if (retval
> 0 && map
->m_flags
& EXT4_MAP_MAPPED
) {
1294 int ret
= check_block_validity(inode
, map
);
1299 /* If it is only a block(s) look up */
1300 if ((flags
& EXT4_GET_BLOCKS_CREATE
) == 0)
1304 * Returns if the blocks have already allocated
1306 * Note that if blocks have been preallocated
1307 * ext4_ext_get_block() returns th create = 0
1308 * with buffer head unmapped.
1310 if (retval
> 0 && map
->m_flags
& EXT4_MAP_MAPPED
)
1314 * When we call get_blocks without the create flag, the
1315 * BH_Unwritten flag could have gotten set if the blocks
1316 * requested were part of a uninitialized extent. We need to
1317 * clear this flag now that we are committed to convert all or
1318 * part of the uninitialized extent to be an initialized
1319 * extent. This is because we need to avoid the combination
1320 * of BH_Unwritten and BH_Mapped flags being simultaneously
1321 * set on the buffer_head.
1323 map
->m_flags
&= ~EXT4_MAP_UNWRITTEN
;
1326 * New blocks allocate and/or writing to uninitialized extent
1327 * will possibly result in updating i_data, so we take
1328 * the write lock of i_data_sem, and call get_blocks()
1329 * with create == 1 flag.
1331 down_write((&EXT4_I(inode
)->i_data_sem
));
1334 * if the caller is from delayed allocation writeout path
1335 * we have already reserved fs blocks for allocation
1336 * let the underlying get_block() function know to
1337 * avoid double accounting
1339 if (flags
& EXT4_GET_BLOCKS_DELALLOC_RESERVE
)
1340 ext4_set_inode_state(inode
, EXT4_STATE_DELALLOC_RESERVED
);
1342 * We need to check for EXT4 here because migrate
1343 * could have changed the inode type in between
1345 if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
)) {
1346 retval
= ext4_ext_map_blocks(handle
, inode
, map
, flags
);
1348 retval
= ext4_ind_map_blocks(handle
, inode
, map
, flags
);
1350 if (retval
> 0 && map
->m_flags
& EXT4_MAP_NEW
) {
1352 * We allocated new blocks which will result in
1353 * i_data's format changing. Force the migrate
1354 * to fail by clearing migrate flags
1356 ext4_clear_inode_state(inode
, EXT4_STATE_EXT_MIGRATE
);
1360 * Update reserved blocks/metadata blocks after successful
1361 * block allocation which had been deferred till now. We don't
1362 * support fallocate for non extent files. So we can update
1363 * reserve space here.
1366 (flags
& EXT4_GET_BLOCKS_DELALLOC_RESERVE
))
1367 ext4_da_update_reserve_space(inode
, retval
, 1);
1369 if (flags
& EXT4_GET_BLOCKS_DELALLOC_RESERVE
)
1370 ext4_clear_inode_state(inode
, EXT4_STATE_DELALLOC_RESERVED
);
1372 up_write((&EXT4_I(inode
)->i_data_sem
));
1373 if (retval
> 0 && map
->m_flags
& EXT4_MAP_MAPPED
) {
1374 int ret
= check_block_validity(inode
, map
);
1381 /* Maximum number of blocks we map for direct IO at once. */
1382 #define DIO_MAX_BLOCKS 4096
1384 static int _ext4_get_block(struct inode
*inode
, sector_t iblock
,
1385 struct buffer_head
*bh
, int flags
)
1387 handle_t
*handle
= ext4_journal_current_handle();
1388 struct ext4_map_blocks map
;
1389 int ret
= 0, started
= 0;
1392 map
.m_lblk
= iblock
;
1393 map
.m_len
= bh
->b_size
>> inode
->i_blkbits
;
1395 if (flags
&& !handle
) {
1396 /* Direct IO write... */
1397 if (map
.m_len
> DIO_MAX_BLOCKS
)
1398 map
.m_len
= DIO_MAX_BLOCKS
;
1399 dio_credits
= ext4_chunk_trans_blocks(inode
, map
.m_len
);
1400 handle
= ext4_journal_start(inode
, dio_credits
);
1401 if (IS_ERR(handle
)) {
1402 ret
= PTR_ERR(handle
);
1408 ret
= ext4_map_blocks(handle
, inode
, &map
, flags
);
1410 map_bh(bh
, inode
->i_sb
, map
.m_pblk
);
1411 bh
->b_state
= (bh
->b_state
& ~EXT4_MAP_FLAGS
) | map
.m_flags
;
1412 bh
->b_size
= inode
->i_sb
->s_blocksize
* map
.m_len
;
1416 ext4_journal_stop(handle
);
1420 int ext4_get_block(struct inode
*inode
, sector_t iblock
,
1421 struct buffer_head
*bh
, int create
)
1423 return _ext4_get_block(inode
, iblock
, bh
,
1424 create
? EXT4_GET_BLOCKS_CREATE
: 0);
1428 * `handle' can be NULL if create is zero
1430 struct buffer_head
*ext4_getblk(handle_t
*handle
, struct inode
*inode
,
1431 ext4_lblk_t block
, int create
, int *errp
)
1433 struct ext4_map_blocks map
;
1434 struct buffer_head
*bh
;
1437 J_ASSERT(handle
!= NULL
|| create
== 0);
1441 err
= ext4_map_blocks(handle
, inode
, &map
,
1442 create
? EXT4_GET_BLOCKS_CREATE
: 0);
1450 bh
= sb_getblk(inode
->i_sb
, map
.m_pblk
);
1455 if (map
.m_flags
& EXT4_MAP_NEW
) {
1456 J_ASSERT(create
!= 0);
1457 J_ASSERT(handle
!= NULL
);
1460 * Now that we do not always journal data, we should
1461 * keep in mind whether this should always journal the
1462 * new buffer as metadata. For now, regular file
1463 * writes use ext4_get_block instead, so it's not a
1467 BUFFER_TRACE(bh
, "call get_create_access");
1468 fatal
= ext4_journal_get_create_access(handle
, bh
);
1469 if (!fatal
&& !buffer_uptodate(bh
)) {
1470 memset(bh
->b_data
, 0, inode
->i_sb
->s_blocksize
);
1471 set_buffer_uptodate(bh
);
1474 BUFFER_TRACE(bh
, "call ext4_handle_dirty_metadata");
1475 err
= ext4_handle_dirty_metadata(handle
, inode
, bh
);
1479 BUFFER_TRACE(bh
, "not a new buffer");
1489 struct buffer_head
*ext4_bread(handle_t
*handle
, struct inode
*inode
,
1490 ext4_lblk_t block
, int create
, int *err
)
1492 struct buffer_head
*bh
;
1494 bh
= ext4_getblk(handle
, inode
, block
, create
, err
);
1497 if (buffer_uptodate(bh
))
1499 ll_rw_block(READ_META
, 1, &bh
);
1501 if (buffer_uptodate(bh
))
1508 static int walk_page_buffers(handle_t
*handle
,
1509 struct buffer_head
*head
,
1513 int (*fn
)(handle_t
*handle
,
1514 struct buffer_head
*bh
))
1516 struct buffer_head
*bh
;
1517 unsigned block_start
, block_end
;
1518 unsigned blocksize
= head
->b_size
;
1520 struct buffer_head
*next
;
1522 for (bh
= head
, block_start
= 0;
1523 ret
== 0 && (bh
!= head
|| !block_start
);
1524 block_start
= block_end
, bh
= next
) {
1525 next
= bh
->b_this_page
;
1526 block_end
= block_start
+ blocksize
;
1527 if (block_end
<= from
|| block_start
>= to
) {
1528 if (partial
&& !buffer_uptodate(bh
))
1532 err
= (*fn
)(handle
, bh
);
1540 * To preserve ordering, it is essential that the hole instantiation and
1541 * the data write be encapsulated in a single transaction. We cannot
1542 * close off a transaction and start a new one between the ext4_get_block()
1543 * and the commit_write(). So doing the jbd2_journal_start at the start of
1544 * prepare_write() is the right place.
1546 * Also, this function can nest inside ext4_writepage() ->
1547 * block_write_full_page(). In that case, we *know* that ext4_writepage()
1548 * has generated enough buffer credits to do the whole page. So we won't
1549 * block on the journal in that case, which is good, because the caller may
1552 * By accident, ext4 can be reentered when a transaction is open via
1553 * quota file writes. If we were to commit the transaction while thus
1554 * reentered, there can be a deadlock - we would be holding a quota
1555 * lock, and the commit would never complete if another thread had a
1556 * transaction open and was blocking on the quota lock - a ranking
1559 * So what we do is to rely on the fact that jbd2_journal_stop/journal_start
1560 * will _not_ run commit under these circumstances because handle->h_ref
1561 * is elevated. We'll still have enough credits for the tiny quotafile
1564 static int do_journal_get_write_access(handle_t
*handle
,
1565 struct buffer_head
*bh
)
1567 int dirty
= buffer_dirty(bh
);
1570 if (!buffer_mapped(bh
) || buffer_freed(bh
))
1573 * __block_write_begin() could have dirtied some buffers. Clean
1574 * the dirty bit as jbd2_journal_get_write_access() could complain
1575 * otherwise about fs integrity issues. Setting of the dirty bit
1576 * by __block_write_begin() isn't a real problem here as we clear
1577 * the bit before releasing a page lock and thus writeback cannot
1578 * ever write the buffer.
1581 clear_buffer_dirty(bh
);
1582 ret
= ext4_journal_get_write_access(handle
, bh
);
1584 ret
= ext4_handle_dirty_metadata(handle
, NULL
, bh
);
1589 * Truncate blocks that were not used by write. We have to truncate the
1590 * pagecache as well so that corresponding buffers get properly unmapped.
1592 static void ext4_truncate_failed_write(struct inode
*inode
)
1594 truncate_inode_pages(inode
->i_mapping
, inode
->i_size
);
1595 ext4_truncate(inode
);
1598 static int ext4_get_block_write(struct inode
*inode
, sector_t iblock
,
1599 struct buffer_head
*bh_result
, int create
);
1600 static int ext4_write_begin(struct file
*file
, struct address_space
*mapping
,
1601 loff_t pos
, unsigned len
, unsigned flags
,
1602 struct page
**pagep
, void **fsdata
)
1604 struct inode
*inode
= mapping
->host
;
1605 int ret
, needed_blocks
;
1612 trace_ext4_write_begin(inode
, pos
, len
, flags
);
1614 * Reserve one block more for addition to orphan list in case
1615 * we allocate blocks but write fails for some reason
1617 needed_blocks
= ext4_writepage_trans_blocks(inode
) + 1;
1618 index
= pos
>> PAGE_CACHE_SHIFT
;
1619 from
= pos
& (PAGE_CACHE_SIZE
- 1);
1623 handle
= ext4_journal_start(inode
, needed_blocks
);
1624 if (IS_ERR(handle
)) {
1625 ret
= PTR_ERR(handle
);
1629 /* We cannot recurse into the filesystem as the transaction is already
1631 flags
|= AOP_FLAG_NOFS
;
1633 page
= grab_cache_page_write_begin(mapping
, index
, flags
);
1635 ext4_journal_stop(handle
);
1641 if (ext4_should_dioread_nolock(inode
))
1642 ret
= __block_write_begin(page
, pos
, len
, ext4_get_block_write
);
1644 ret
= __block_write_begin(page
, pos
, len
, ext4_get_block
);
1646 if (!ret
&& ext4_should_journal_data(inode
)) {
1647 ret
= walk_page_buffers(handle
, page_buffers(page
),
1648 from
, to
, NULL
, do_journal_get_write_access
);
1653 page_cache_release(page
);
1655 * __block_write_begin may have instantiated a few blocks
1656 * outside i_size. Trim these off again. Don't need
1657 * i_size_read because we hold i_mutex.
1659 * Add inode to orphan list in case we crash before
1662 if (pos
+ len
> inode
->i_size
&& ext4_can_truncate(inode
))
1663 ext4_orphan_add(handle
, inode
);
1665 ext4_journal_stop(handle
);
1666 if (pos
+ len
> inode
->i_size
) {
1667 ext4_truncate_failed_write(inode
);
1669 * If truncate failed early the inode might
1670 * still be on the orphan list; we need to
1671 * make sure the inode is removed from the
1672 * orphan list in that case.
1675 ext4_orphan_del(NULL
, inode
);
1679 if (ret
== -ENOSPC
&& ext4_should_retry_alloc(inode
->i_sb
, &retries
))
1685 /* For write_end() in data=journal mode */
1686 static int write_end_fn(handle_t
*handle
, struct buffer_head
*bh
)
1688 if (!buffer_mapped(bh
) || buffer_freed(bh
))
1690 set_buffer_uptodate(bh
);
1691 return ext4_handle_dirty_metadata(handle
, NULL
, bh
);
1694 static int ext4_generic_write_end(struct file
*file
,
1695 struct address_space
*mapping
,
1696 loff_t pos
, unsigned len
, unsigned copied
,
1697 struct page
*page
, void *fsdata
)
1699 int i_size_changed
= 0;
1700 struct inode
*inode
= mapping
->host
;
1701 handle_t
*handle
= ext4_journal_current_handle();
1703 copied
= block_write_end(file
, mapping
, pos
, len
, copied
, page
, fsdata
);
1706 * No need to use i_size_read() here, the i_size
1707 * cannot change under us because we hold i_mutex.
1709 * But it's important to update i_size while still holding page lock:
1710 * page writeout could otherwise come in and zero beyond i_size.
1712 if (pos
+ copied
> inode
->i_size
) {
1713 i_size_write(inode
, pos
+ copied
);
1717 if (pos
+ copied
> EXT4_I(inode
)->i_disksize
) {
1718 /* We need to mark inode dirty even if
1719 * new_i_size is less that inode->i_size
1720 * bu greater than i_disksize.(hint delalloc)
1722 ext4_update_i_disksize(inode
, (pos
+ copied
));
1726 page_cache_release(page
);
1729 * Don't mark the inode dirty under page lock. First, it unnecessarily
1730 * makes the holding time of page lock longer. Second, it forces lock
1731 * ordering of page lock and transaction start for journaling
1735 ext4_mark_inode_dirty(handle
, inode
);
1741 * We need to pick up the new inode size which generic_commit_write gave us
1742 * `file' can be NULL - eg, when called from page_symlink().
1744 * ext4 never places buffers on inode->i_mapping->private_list. metadata
1745 * buffers are managed internally.
1747 static int ext4_ordered_write_end(struct file
*file
,
1748 struct address_space
*mapping
,
1749 loff_t pos
, unsigned len
, unsigned copied
,
1750 struct page
*page
, void *fsdata
)
1752 handle_t
*handle
= ext4_journal_current_handle();
1753 struct inode
*inode
= mapping
->host
;
1756 trace_ext4_ordered_write_end(inode
, pos
, len
, copied
);
1757 ret
= ext4_jbd2_file_inode(handle
, inode
);
1760 ret2
= ext4_generic_write_end(file
, mapping
, pos
, len
, copied
,
1763 if (pos
+ len
> inode
->i_size
&& ext4_can_truncate(inode
))
1764 /* if we have allocated more blocks and copied
1765 * less. We will have blocks allocated outside
1766 * inode->i_size. So truncate them
1768 ext4_orphan_add(handle
, inode
);
1772 ret2
= ext4_journal_stop(handle
);
1776 if (pos
+ len
> inode
->i_size
) {
1777 ext4_truncate_failed_write(inode
);
1779 * If truncate failed early the inode might still be
1780 * on the orphan list; we need to make sure the inode
1781 * is removed from the orphan list in that case.
1784 ext4_orphan_del(NULL
, inode
);
1788 return ret
? ret
: copied
;
1791 static int ext4_writeback_write_end(struct file
*file
,
1792 struct address_space
*mapping
,
1793 loff_t pos
, unsigned len
, unsigned copied
,
1794 struct page
*page
, void *fsdata
)
1796 handle_t
*handle
= ext4_journal_current_handle();
1797 struct inode
*inode
= mapping
->host
;
1800 trace_ext4_writeback_write_end(inode
, pos
, len
, copied
);
1801 ret2
= ext4_generic_write_end(file
, mapping
, pos
, len
, copied
,
1804 if (pos
+ len
> inode
->i_size
&& ext4_can_truncate(inode
))
1805 /* if we have allocated more blocks and copied
1806 * less. We will have blocks allocated outside
1807 * inode->i_size. So truncate them
1809 ext4_orphan_add(handle
, inode
);
1814 ret2
= ext4_journal_stop(handle
);
1818 if (pos
+ len
> inode
->i_size
) {
1819 ext4_truncate_failed_write(inode
);
1821 * If truncate failed early the inode might still be
1822 * on the orphan list; we need to make sure the inode
1823 * is removed from the orphan list in that case.
1826 ext4_orphan_del(NULL
, inode
);
1829 return ret
? ret
: copied
;
1832 static int ext4_journalled_write_end(struct file
*file
,
1833 struct address_space
*mapping
,
1834 loff_t pos
, unsigned len
, unsigned copied
,
1835 struct page
*page
, void *fsdata
)
1837 handle_t
*handle
= ext4_journal_current_handle();
1838 struct inode
*inode
= mapping
->host
;
1844 trace_ext4_journalled_write_end(inode
, pos
, len
, copied
);
1845 from
= pos
& (PAGE_CACHE_SIZE
- 1);
1849 if (!PageUptodate(page
))
1851 page_zero_new_buffers(page
, from
+copied
, to
);
1854 ret
= walk_page_buffers(handle
, page_buffers(page
), from
,
1855 to
, &partial
, write_end_fn
);
1857 SetPageUptodate(page
);
1858 new_i_size
= pos
+ copied
;
1859 if (new_i_size
> inode
->i_size
)
1860 i_size_write(inode
, pos
+copied
);
1861 ext4_set_inode_state(inode
, EXT4_STATE_JDATA
);
1862 if (new_i_size
> EXT4_I(inode
)->i_disksize
) {
1863 ext4_update_i_disksize(inode
, new_i_size
);
1864 ret2
= ext4_mark_inode_dirty(handle
, inode
);
1870 page_cache_release(page
);
1871 if (pos
+ len
> inode
->i_size
&& ext4_can_truncate(inode
))
1872 /* if we have allocated more blocks and copied
1873 * less. We will have blocks allocated outside
1874 * inode->i_size. So truncate them
1876 ext4_orphan_add(handle
, inode
);
1878 ret2
= ext4_journal_stop(handle
);
1881 if (pos
+ len
> inode
->i_size
) {
1882 ext4_truncate_failed_write(inode
);
1884 * If truncate failed early the inode might still be
1885 * on the orphan list; we need to make sure the inode
1886 * is removed from the orphan list in that case.
1889 ext4_orphan_del(NULL
, inode
);
1892 return ret
? ret
: copied
;
1896 * Reserve a single block located at lblock
1898 static int ext4_da_reserve_space(struct inode
*inode
, ext4_lblk_t lblock
)
1901 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
1902 struct ext4_inode_info
*ei
= EXT4_I(inode
);
1903 unsigned long md_needed
;
1907 * recalculate the amount of metadata blocks to reserve
1908 * in order to allocate nrblocks
1909 * worse case is one extent per block
1912 spin_lock(&ei
->i_block_reservation_lock
);
1913 md_needed
= ext4_calc_metadata_amount(inode
, lblock
);
1914 trace_ext4_da_reserve_space(inode
, md_needed
);
1915 spin_unlock(&ei
->i_block_reservation_lock
);
1918 * We will charge metadata quota at writeout time; this saves
1919 * us from metadata over-estimation, though we may go over by
1920 * a small amount in the end. Here we just reserve for data.
1922 ret
= dquot_reserve_block(inode
, 1);
1926 * We do still charge estimated metadata to the sb though;
1927 * we cannot afford to run out of free blocks.
1929 if (ext4_claim_free_blocks(sbi
, md_needed
+ 1)) {
1930 dquot_release_reservation_block(inode
, 1);
1931 if (ext4_should_retry_alloc(inode
->i_sb
, &retries
)) {
1937 spin_lock(&ei
->i_block_reservation_lock
);
1938 ei
->i_reserved_data_blocks
++;
1939 ei
->i_reserved_meta_blocks
+= md_needed
;
1940 spin_unlock(&ei
->i_block_reservation_lock
);
1942 return 0; /* success */
1945 static void ext4_da_release_space(struct inode
*inode
, int to_free
)
1947 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
1948 struct ext4_inode_info
*ei
= EXT4_I(inode
);
1951 return; /* Nothing to release, exit */
1953 spin_lock(&EXT4_I(inode
)->i_block_reservation_lock
);
1955 trace_ext4_da_release_space(inode
, to_free
);
1956 if (unlikely(to_free
> ei
->i_reserved_data_blocks
)) {
1958 * if there aren't enough reserved blocks, then the
1959 * counter is messed up somewhere. Since this
1960 * function is called from invalidate page, it's
1961 * harmless to return without any action.
1963 ext4_msg(inode
->i_sb
, KERN_NOTICE
, "ext4_da_release_space: "
1964 "ino %lu, to_free %d with only %d reserved "
1965 "data blocks\n", inode
->i_ino
, to_free
,
1966 ei
->i_reserved_data_blocks
);
1968 to_free
= ei
->i_reserved_data_blocks
;
1970 ei
->i_reserved_data_blocks
-= to_free
;
1972 if (ei
->i_reserved_data_blocks
== 0) {
1974 * We can release all of the reserved metadata blocks
1975 * only when we have written all of the delayed
1976 * allocation blocks.
1978 percpu_counter_sub(&sbi
->s_dirtyblocks_counter
,
1979 ei
->i_reserved_meta_blocks
);
1980 ei
->i_reserved_meta_blocks
= 0;
1981 ei
->i_da_metadata_calc_len
= 0;
1984 /* update fs dirty data blocks counter */
1985 percpu_counter_sub(&sbi
->s_dirtyblocks_counter
, to_free
);
1987 spin_unlock(&EXT4_I(inode
)->i_block_reservation_lock
);
1989 dquot_release_reservation_block(inode
, to_free
);
1992 static void ext4_da_page_release_reservation(struct page
*page
,
1993 unsigned long offset
)
1996 struct buffer_head
*head
, *bh
;
1997 unsigned int curr_off
= 0;
1999 head
= page_buffers(page
);
2002 unsigned int next_off
= curr_off
+ bh
->b_size
;
2004 if ((offset
<= curr_off
) && (buffer_delay(bh
))) {
2006 clear_buffer_delay(bh
);
2008 curr_off
= next_off
;
2009 } while ((bh
= bh
->b_this_page
) != head
);
2010 ext4_da_release_space(page
->mapping
->host
, to_release
);
2014 * Delayed allocation stuff
2018 * mpage_da_submit_io - walks through extent of pages and try to write
2019 * them with writepage() call back
2021 * @mpd->inode: inode
2022 * @mpd->first_page: first page of the extent
2023 * @mpd->next_page: page after the last page of the extent
2025 * By the time mpage_da_submit_io() is called we expect all blocks
2026 * to be allocated. this may be wrong if allocation failed.
2028 * As pages are already locked by write_cache_pages(), we can't use it
2030 static int mpage_da_submit_io(struct mpage_da_data
*mpd
,
2031 struct ext4_map_blocks
*map
)
2033 struct pagevec pvec
;
2034 unsigned long index
, end
;
2035 int ret
= 0, err
, nr_pages
, i
;
2036 struct inode
*inode
= mpd
->inode
;
2037 struct address_space
*mapping
= inode
->i_mapping
;
2038 loff_t size
= i_size_read(inode
);
2039 unsigned int len
, block_start
;
2040 struct buffer_head
*bh
, *page_bufs
= NULL
;
2041 int journal_data
= ext4_should_journal_data(inode
);
2042 sector_t pblock
= 0, cur_logical
= 0;
2043 struct ext4_io_submit io_submit
;
2045 BUG_ON(mpd
->next_page
<= mpd
->first_page
);
2046 memset(&io_submit
, 0, sizeof(io_submit
));
2048 * We need to start from the first_page to the next_page - 1
2049 * to make sure we also write the mapped dirty buffer_heads.
2050 * If we look at mpd->b_blocknr we would only be looking
2051 * at the currently mapped buffer_heads.
2053 index
= mpd
->first_page
;
2054 end
= mpd
->next_page
- 1;
2056 pagevec_init(&pvec
, 0);
2057 while (index
<= end
) {
2058 nr_pages
= pagevec_lookup(&pvec
, mapping
, index
, PAGEVEC_SIZE
);
2061 for (i
= 0; i
< nr_pages
; i
++) {
2062 int commit_write
= 0, redirty_page
= 0;
2063 struct page
*page
= pvec
.pages
[i
];
2065 index
= page
->index
;
2069 if (index
== size
>> PAGE_CACHE_SHIFT
)
2070 len
= size
& ~PAGE_CACHE_MASK
;
2072 len
= PAGE_CACHE_SIZE
;
2074 cur_logical
= index
<< (PAGE_CACHE_SHIFT
-
2076 pblock
= map
->m_pblk
+ (cur_logical
-
2081 BUG_ON(!PageLocked(page
));
2082 BUG_ON(PageWriteback(page
));
2085 * If the page does not have buffers (for
2086 * whatever reason), try to create them using
2087 * __block_write_begin. If this fails,
2088 * redirty the page and move on.
2090 if (!page_has_buffers(page
)) {
2091 if (__block_write_begin(page
, 0, len
,
2092 noalloc_get_block_write
)) {
2094 redirty_page_for_writepage(mpd
->wbc
,
2102 bh
= page_bufs
= page_buffers(page
);
2107 if (map
&& (cur_logical
>= map
->m_lblk
) &&
2108 (cur_logical
<= (map
->m_lblk
+
2109 (map
->m_len
- 1)))) {
2110 if (buffer_delay(bh
)) {
2111 clear_buffer_delay(bh
);
2112 bh
->b_blocknr
= pblock
;
2114 if (buffer_unwritten(bh
) ||
2116 BUG_ON(bh
->b_blocknr
!= pblock
);
2117 if (map
->m_flags
& EXT4_MAP_UNINIT
)
2118 set_buffer_uninit(bh
);
2119 clear_buffer_unwritten(bh
);
2122 /* redirty page if block allocation undone */
2123 if (buffer_delay(bh
) || buffer_unwritten(bh
))
2125 bh
= bh
->b_this_page
;
2126 block_start
+= bh
->b_size
;
2129 } while (bh
!= page_bufs
);
2135 /* mark the buffer_heads as dirty & uptodate */
2136 block_commit_write(page
, 0, len
);
2139 * Delalloc doesn't support data journalling,
2140 * but eventually maybe we'll lift this
2143 if (unlikely(journal_data
&& PageChecked(page
)))
2144 err
= __ext4_journalled_writepage(page
, len
);
2145 else if (test_opt(inode
->i_sb
, MBLK_IO_SUBMIT
))
2146 err
= ext4_bio_write_page(&io_submit
, page
,
2149 err
= block_write_full_page(page
,
2150 noalloc_get_block_write
, mpd
->wbc
);
2153 mpd
->pages_written
++;
2155 * In error case, we have to continue because
2156 * remaining pages are still locked
2161 pagevec_release(&pvec
);
2163 ext4_io_submit(&io_submit
);
2167 static void ext4_da_block_invalidatepages(struct mpage_da_data
*mpd
,
2168 sector_t logical
, long blk_cnt
)
2172 struct pagevec pvec
;
2173 struct inode
*inode
= mpd
->inode
;
2174 struct address_space
*mapping
= inode
->i_mapping
;
2176 index
= logical
>> (PAGE_CACHE_SHIFT
- inode
->i_blkbits
);
2177 end
= (logical
+ blk_cnt
- 1) >>
2178 (PAGE_CACHE_SHIFT
- inode
->i_blkbits
);
2179 while (index
<= end
) {
2180 nr_pages
= pagevec_lookup(&pvec
, mapping
, index
, PAGEVEC_SIZE
);
2183 for (i
= 0; i
< nr_pages
; i
++) {
2184 struct page
*page
= pvec
.pages
[i
];
2185 if (page
->index
> end
)
2187 BUG_ON(!PageLocked(page
));
2188 BUG_ON(PageWriteback(page
));
2189 block_invalidatepage(page
, 0);
2190 ClearPageUptodate(page
);
2193 index
= pvec
.pages
[nr_pages
- 1]->index
+ 1;
2194 pagevec_release(&pvec
);
2199 static void ext4_print_free_blocks(struct inode
*inode
)
2201 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
2202 printk(KERN_CRIT
"Total free blocks count %lld\n",
2203 ext4_count_free_blocks(inode
->i_sb
));
2204 printk(KERN_CRIT
"Free/Dirty block details\n");
2205 printk(KERN_CRIT
"free_blocks=%lld\n",
2206 (long long) percpu_counter_sum(&sbi
->s_freeblocks_counter
));
2207 printk(KERN_CRIT
"dirty_blocks=%lld\n",
2208 (long long) percpu_counter_sum(&sbi
->s_dirtyblocks_counter
));
2209 printk(KERN_CRIT
"Block reservation details\n");
2210 printk(KERN_CRIT
"i_reserved_data_blocks=%u\n",
2211 EXT4_I(inode
)->i_reserved_data_blocks
);
2212 printk(KERN_CRIT
"i_reserved_meta_blocks=%u\n",
2213 EXT4_I(inode
)->i_reserved_meta_blocks
);
2218 * mpage_da_map_and_submit - go through given space, map them
2219 * if necessary, and then submit them for I/O
2221 * @mpd - bh describing space
2223 * The function skips space we know is already mapped to disk blocks.
2226 static void mpage_da_map_and_submit(struct mpage_da_data
*mpd
)
2228 int err
, blks
, get_blocks_flags
;
2229 struct ext4_map_blocks map
, *mapp
= NULL
;
2230 sector_t next
= mpd
->b_blocknr
;
2231 unsigned max_blocks
= mpd
->b_size
>> mpd
->inode
->i_blkbits
;
2232 loff_t disksize
= EXT4_I(mpd
->inode
)->i_disksize
;
2233 handle_t
*handle
= NULL
;
2236 * If the blocks are mapped already, or we couldn't accumulate
2237 * any blocks, then proceed immediately to the submission stage.
2239 if ((mpd
->b_size
== 0) ||
2240 ((mpd
->b_state
& (1 << BH_Mapped
)) &&
2241 !(mpd
->b_state
& (1 << BH_Delay
)) &&
2242 !(mpd
->b_state
& (1 << BH_Unwritten
))))
2245 handle
= ext4_journal_current_handle();
2249 * Call ext4_map_blocks() to allocate any delayed allocation
2250 * blocks, or to convert an uninitialized extent to be
2251 * initialized (in the case where we have written into
2252 * one or more preallocated blocks).
2254 * We pass in the magic EXT4_GET_BLOCKS_DELALLOC_RESERVE to
2255 * indicate that we are on the delayed allocation path. This
2256 * affects functions in many different parts of the allocation
2257 * call path. This flag exists primarily because we don't
2258 * want to change *many* call functions, so ext4_map_blocks()
2259 * will set the EXT4_STATE_DELALLOC_RESERVED flag once the
2260 * inode's allocation semaphore is taken.
2262 * If the blocks in questions were delalloc blocks, set
2263 * EXT4_GET_BLOCKS_DELALLOC_RESERVE so the delalloc accounting
2264 * variables are updated after the blocks have been allocated.
2267 map
.m_len
= max_blocks
;
2268 get_blocks_flags
= EXT4_GET_BLOCKS_CREATE
;
2269 if (ext4_should_dioread_nolock(mpd
->inode
))
2270 get_blocks_flags
|= EXT4_GET_BLOCKS_IO_CREATE_EXT
;
2271 if (mpd
->b_state
& (1 << BH_Delay
))
2272 get_blocks_flags
|= EXT4_GET_BLOCKS_DELALLOC_RESERVE
;
2274 blks
= ext4_map_blocks(handle
, mpd
->inode
, &map
, get_blocks_flags
);
2276 struct super_block
*sb
= mpd
->inode
->i_sb
;
2280 * If get block returns EAGAIN or ENOSPC and there
2281 * appears to be free blocks we will call
2282 * ext4_writepage() for all of the pages which will
2283 * just redirty the pages.
2288 if (err
== -ENOSPC
&&
2289 ext4_count_free_blocks(sb
)) {
2295 * get block failure will cause us to loop in
2296 * writepages, because a_ops->writepage won't be able
2297 * to make progress. The page will be redirtied by
2298 * writepage and writepages will again try to write
2301 if (!(EXT4_SB(sb
)->s_mount_flags
& EXT4_MF_FS_ABORTED
)) {
2302 ext4_msg(sb
, KERN_CRIT
,
2303 "delayed block allocation failed for inode %lu "
2304 "at logical offset %llu with max blocks %zd "
2305 "with error %d", mpd
->inode
->i_ino
,
2306 (unsigned long long) next
,
2307 mpd
->b_size
>> mpd
->inode
->i_blkbits
, err
);
2308 ext4_msg(sb
, KERN_CRIT
,
2309 "This should not happen!! Data will be lost\n");
2311 ext4_print_free_blocks(mpd
->inode
);
2313 /* invalidate all the pages */
2314 ext4_da_block_invalidatepages(mpd
, next
,
2315 mpd
->b_size
>> mpd
->inode
->i_blkbits
);
2321 if (map
.m_flags
& EXT4_MAP_NEW
) {
2322 struct block_device
*bdev
= mpd
->inode
->i_sb
->s_bdev
;
2325 for (i
= 0; i
< map
.m_len
; i
++)
2326 unmap_underlying_metadata(bdev
, map
.m_pblk
+ i
);
2329 if (ext4_should_order_data(mpd
->inode
)) {
2330 err
= ext4_jbd2_file_inode(handle
, mpd
->inode
);
2332 /* This only happens if the journal is aborted */
2337 * Update on-disk size along with block allocation.
2339 disksize
= ((loff_t
) next
+ blks
) << mpd
->inode
->i_blkbits
;
2340 if (disksize
> i_size_read(mpd
->inode
))
2341 disksize
= i_size_read(mpd
->inode
);
2342 if (disksize
> EXT4_I(mpd
->inode
)->i_disksize
) {
2343 ext4_update_i_disksize(mpd
->inode
, disksize
);
2344 err
= ext4_mark_inode_dirty(handle
, mpd
->inode
);
2346 ext4_error(mpd
->inode
->i_sb
,
2347 "Failed to mark inode %lu dirty",
2352 mpage_da_submit_io(mpd
, mapp
);
2356 #define BH_FLAGS ((1 << BH_Uptodate) | (1 << BH_Mapped) | \
2357 (1 << BH_Delay) | (1 << BH_Unwritten))
2360 * mpage_add_bh_to_extent - try to add one more block to extent of blocks
2362 * @mpd->lbh - extent of blocks
2363 * @logical - logical number of the block in the file
2364 * @bh - bh of the block (used to access block's state)
2366 * the function is used to collect contig. blocks in same state
2368 static void mpage_add_bh_to_extent(struct mpage_da_data
*mpd
,
2369 sector_t logical
, size_t b_size
,
2370 unsigned long b_state
)
2373 int nrblocks
= mpd
->b_size
>> mpd
->inode
->i_blkbits
;
2376 * XXX Don't go larger than mballoc is willing to allocate
2377 * This is a stopgap solution. We eventually need to fold
2378 * mpage_da_submit_io() into this function and then call
2379 * ext4_map_blocks() multiple times in a loop
2381 if (nrblocks
>= 8*1024*1024/mpd
->inode
->i_sb
->s_blocksize
)
2384 /* check if thereserved journal credits might overflow */
2385 if (!(ext4_test_inode_flag(mpd
->inode
, EXT4_INODE_EXTENTS
))) {
2386 if (nrblocks
>= EXT4_MAX_TRANS_DATA
) {
2388 * With non-extent format we are limited by the journal
2389 * credit available. Total credit needed to insert
2390 * nrblocks contiguous blocks is dependent on the
2391 * nrblocks. So limit nrblocks.
2394 } else if ((nrblocks
+ (b_size
>> mpd
->inode
->i_blkbits
)) >
2395 EXT4_MAX_TRANS_DATA
) {
2397 * Adding the new buffer_head would make it cross the
2398 * allowed limit for which we have journal credit
2399 * reserved. So limit the new bh->b_size
2401 b_size
= (EXT4_MAX_TRANS_DATA
- nrblocks
) <<
2402 mpd
->inode
->i_blkbits
;
2403 /* we will do mpage_da_submit_io in the next loop */
2407 * First block in the extent
2409 if (mpd
->b_size
== 0) {
2410 mpd
->b_blocknr
= logical
;
2411 mpd
->b_size
= b_size
;
2412 mpd
->b_state
= b_state
& BH_FLAGS
;
2416 next
= mpd
->b_blocknr
+ nrblocks
;
2418 * Can we merge the block to our big extent?
2420 if (logical
== next
&& (b_state
& BH_FLAGS
) == mpd
->b_state
) {
2421 mpd
->b_size
+= b_size
;
2427 * We couldn't merge the block to our extent, so we
2428 * need to flush current extent and start new one
2430 mpage_da_map_and_submit(mpd
);
2434 static int ext4_bh_delay_or_unwritten(handle_t
*handle
, struct buffer_head
*bh
)
2436 return (buffer_delay(bh
) || buffer_unwritten(bh
)) && buffer_dirty(bh
);
2440 * __mpage_da_writepage - finds extent of pages and blocks
2442 * @page: page to consider
2443 * @wbc: not used, we just follow rules
2446 * The function finds extents of pages and scan them for all blocks.
2448 static int __mpage_da_writepage(struct page
*page
,
2449 struct writeback_control
*wbc
,
2450 struct mpage_da_data
*mpd
)
2452 struct inode
*inode
= mpd
->inode
;
2453 struct buffer_head
*bh
, *head
;
2457 * Can we merge this page to current extent?
2459 if (mpd
->next_page
!= page
->index
) {
2461 * Nope, we can't. So, we map non-allocated blocks
2462 * and start IO on them
2464 if (mpd
->next_page
!= mpd
->first_page
) {
2465 mpage_da_map_and_submit(mpd
);
2467 * skip rest of the page in the page_vec
2469 redirty_page_for_writepage(wbc
, page
);
2471 return MPAGE_DA_EXTENT_TAIL
;
2475 * Start next extent of pages ...
2477 mpd
->first_page
= page
->index
;
2487 mpd
->next_page
= page
->index
+ 1;
2488 logical
= (sector_t
) page
->index
<<
2489 (PAGE_CACHE_SHIFT
- inode
->i_blkbits
);
2491 if (!page_has_buffers(page
)) {
2492 mpage_add_bh_to_extent(mpd
, logical
, PAGE_CACHE_SIZE
,
2493 (1 << BH_Dirty
) | (1 << BH_Uptodate
));
2495 return MPAGE_DA_EXTENT_TAIL
;
2498 * Page with regular buffer heads, just add all dirty ones
2500 head
= page_buffers(page
);
2503 BUG_ON(buffer_locked(bh
));
2505 * We need to try to allocate
2506 * unmapped blocks in the same page.
2507 * Otherwise we won't make progress
2508 * with the page in ext4_writepage
2510 if (ext4_bh_delay_or_unwritten(NULL
, bh
)) {
2511 mpage_add_bh_to_extent(mpd
, logical
,
2515 return MPAGE_DA_EXTENT_TAIL
;
2516 } else if (buffer_dirty(bh
) && (buffer_mapped(bh
))) {
2518 * mapped dirty buffer. We need to update
2519 * the b_state because we look at
2520 * b_state in mpage_da_map_blocks. We don't
2521 * update b_size because if we find an
2522 * unmapped buffer_head later we need to
2523 * use the b_state flag of that buffer_head.
2525 if (mpd
->b_size
== 0)
2526 mpd
->b_state
= bh
->b_state
& BH_FLAGS
;
2529 } while ((bh
= bh
->b_this_page
) != head
);
2536 * This is a special get_blocks_t callback which is used by
2537 * ext4_da_write_begin(). It will either return mapped block or
2538 * reserve space for a single block.
2540 * For delayed buffer_head we have BH_Mapped, BH_New, BH_Delay set.
2541 * We also have b_blocknr = -1 and b_bdev initialized properly
2543 * For unwritten buffer_head we have BH_Mapped, BH_New, BH_Unwritten set.
2544 * We also have b_blocknr = physicalblock mapping unwritten extent and b_bdev
2545 * initialized properly.
2547 static int ext4_da_get_block_prep(struct inode
*inode
, sector_t iblock
,
2548 struct buffer_head
*bh
, int create
)
2550 struct ext4_map_blocks map
;
2552 sector_t invalid_block
= ~((sector_t
) 0xffff);
2554 if (invalid_block
< ext4_blocks_count(EXT4_SB(inode
->i_sb
)->s_es
))
2557 BUG_ON(create
== 0);
2558 BUG_ON(bh
->b_size
!= inode
->i_sb
->s_blocksize
);
2560 map
.m_lblk
= iblock
;
2564 * first, we need to know whether the block is allocated already
2565 * preallocated blocks are unmapped but should treated
2566 * the same as allocated blocks.
2568 ret
= ext4_map_blocks(NULL
, inode
, &map
, 0);
2572 if (buffer_delay(bh
))
2573 return 0; /* Not sure this could or should happen */
2575 * XXX: __block_write_begin() unmaps passed block, is it OK?
2577 ret
= ext4_da_reserve_space(inode
, iblock
);
2579 /* not enough space to reserve */
2582 map_bh(bh
, inode
->i_sb
, invalid_block
);
2584 set_buffer_delay(bh
);
2588 map_bh(bh
, inode
->i_sb
, map
.m_pblk
);
2589 bh
->b_state
= (bh
->b_state
& ~EXT4_MAP_FLAGS
) | map
.m_flags
;
2591 if (buffer_unwritten(bh
)) {
2592 /* A delayed write to unwritten bh should be marked
2593 * new and mapped. Mapped ensures that we don't do
2594 * get_block multiple times when we write to the same
2595 * offset and new ensures that we do proper zero out
2596 * for partial write.
2599 set_buffer_mapped(bh
);
2605 * This function is used as a standard get_block_t calback function
2606 * when there is no desire to allocate any blocks. It is used as a
2607 * callback function for block_write_begin() and block_write_full_page().
2608 * These functions should only try to map a single block at a time.
2610 * Since this function doesn't do block allocations even if the caller
2611 * requests it by passing in create=1, it is critically important that
2612 * any caller checks to make sure that any buffer heads are returned
2613 * by this function are either all already mapped or marked for
2614 * delayed allocation before calling block_write_full_page(). Otherwise,
2615 * b_blocknr could be left unitialized, and the page write functions will
2616 * be taken by surprise.
2618 static int noalloc_get_block_write(struct inode
*inode
, sector_t iblock
,
2619 struct buffer_head
*bh_result
, int create
)
2621 BUG_ON(bh_result
->b_size
!= inode
->i_sb
->s_blocksize
);
2622 return _ext4_get_block(inode
, iblock
, bh_result
, 0);
2625 static int bget_one(handle_t
*handle
, struct buffer_head
*bh
)
2631 static int bput_one(handle_t
*handle
, struct buffer_head
*bh
)
2637 static int __ext4_journalled_writepage(struct page
*page
,
2640 struct address_space
*mapping
= page
->mapping
;
2641 struct inode
*inode
= mapping
->host
;
2642 struct buffer_head
*page_bufs
;
2643 handle_t
*handle
= NULL
;
2647 ClearPageChecked(page
);
2648 page_bufs
= page_buffers(page
);
2650 walk_page_buffers(handle
, page_bufs
, 0, len
, NULL
, bget_one
);
2651 /* As soon as we unlock the page, it can go away, but we have
2652 * references to buffers so we are safe */
2655 handle
= ext4_journal_start(inode
, ext4_writepage_trans_blocks(inode
));
2656 if (IS_ERR(handle
)) {
2657 ret
= PTR_ERR(handle
);
2661 ret
= walk_page_buffers(handle
, page_bufs
, 0, len
, NULL
,
2662 do_journal_get_write_access
);
2664 err
= walk_page_buffers(handle
, page_bufs
, 0, len
, NULL
,
2668 err
= ext4_journal_stop(handle
);
2672 walk_page_buffers(handle
, page_bufs
, 0, len
, NULL
, bput_one
);
2673 ext4_set_inode_state(inode
, EXT4_STATE_JDATA
);
2678 static int ext4_set_bh_endio(struct buffer_head
*bh
, struct inode
*inode
);
2679 static void ext4_end_io_buffer_write(struct buffer_head
*bh
, int uptodate
);
2682 * Note that we don't need to start a transaction unless we're journaling data
2683 * because we should have holes filled from ext4_page_mkwrite(). We even don't
2684 * need to file the inode to the transaction's list in ordered mode because if
2685 * we are writing back data added by write(), the inode is already there and if
2686 * we are writing back data modified via mmap(), noone guarantees in which
2687 * transaction the data will hit the disk. In case we are journaling data, we
2688 * cannot start transaction directly because transaction start ranks above page
2689 * lock so we have to do some magic.
2691 * This function can get called via...
2692 * - ext4_da_writepages after taking page lock (have journal handle)
2693 * - journal_submit_inode_data_buffers (no journal handle)
2694 * - shrink_page_list via pdflush (no journal handle)
2695 * - grab_page_cache when doing write_begin (have journal handle)
2697 * We don't do any block allocation in this function. If we have page with
2698 * multiple blocks we need to write those buffer_heads that are mapped. This
2699 * is important for mmaped based write. So if we do with blocksize 1K
2700 * truncate(f, 1024);
2701 * a = mmap(f, 0, 4096);
2703 * truncate(f, 4096);
2704 * we have in the page first buffer_head mapped via page_mkwrite call back
2705 * but other bufer_heads would be unmapped but dirty(dirty done via the
2706 * do_wp_page). So writepage should write the first block. If we modify
2707 * the mmap area beyond 1024 we will again get a page_fault and the
2708 * page_mkwrite callback will do the block allocation and mark the
2709 * buffer_heads mapped.
2711 * We redirty the page if we have any buffer_heads that is either delay or
2712 * unwritten in the page.
2714 * We can get recursively called as show below.
2716 * ext4_writepage() -> kmalloc() -> __alloc_pages() -> page_launder() ->
2719 * But since we don't do any block allocation we should not deadlock.
2720 * Page also have the dirty flag cleared so we don't get recurive page_lock.
2722 static int ext4_writepage(struct page
*page
,
2723 struct writeback_control
*wbc
)
2725 int ret
= 0, commit_write
= 0;
2728 struct buffer_head
*page_bufs
= NULL
;
2729 struct inode
*inode
= page
->mapping
->host
;
2731 trace_ext4_writepage(inode
, page
);
2732 size
= i_size_read(inode
);
2733 if (page
->index
== size
>> PAGE_CACHE_SHIFT
)
2734 len
= size
& ~PAGE_CACHE_MASK
;
2736 len
= PAGE_CACHE_SIZE
;
2739 * If the page does not have buffers (for whatever reason),
2740 * try to create them using __block_write_begin. If this
2741 * fails, redirty the page and move on.
2743 if (!page_has_buffers(page
)) {
2744 if (__block_write_begin(page
, 0, len
,
2745 noalloc_get_block_write
)) {
2747 redirty_page_for_writepage(wbc
, page
);
2753 page_bufs
= page_buffers(page
);
2754 if (walk_page_buffers(NULL
, page_bufs
, 0, len
, NULL
,
2755 ext4_bh_delay_or_unwritten
)) {
2757 * We don't want to do block allocation, so redirty
2758 * the page and return. We may reach here when we do
2759 * a journal commit via journal_submit_inode_data_buffers.
2760 * We can also reach here via shrink_page_list
2765 /* now mark the buffer_heads as dirty and uptodate */
2766 block_commit_write(page
, 0, len
);
2768 if (PageChecked(page
) && ext4_should_journal_data(inode
))
2770 * It's mmapped pagecache. Add buffers and journal it. There
2771 * doesn't seem much point in redirtying the page here.
2773 return __ext4_journalled_writepage(page
, len
);
2775 if (buffer_uninit(page_bufs
)) {
2776 ext4_set_bh_endio(page_bufs
, inode
);
2777 ret
= block_write_full_page_endio(page
, noalloc_get_block_write
,
2778 wbc
, ext4_end_io_buffer_write
);
2780 ret
= block_write_full_page(page
, noalloc_get_block_write
,
2787 * This is called via ext4_da_writepages() to
2788 * calulate the total number of credits to reserve to fit
2789 * a single extent allocation into a single transaction,
2790 * ext4_da_writpeages() will loop calling this before
2791 * the block allocation.
2794 static int ext4_da_writepages_trans_blocks(struct inode
*inode
)
2796 int max_blocks
= EXT4_I(inode
)->i_reserved_data_blocks
;
2799 * With non-extent format the journal credit needed to
2800 * insert nrblocks contiguous block is dependent on
2801 * number of contiguous block. So we will limit
2802 * number of contiguous block to a sane value
2804 if (!(ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
)) &&
2805 (max_blocks
> EXT4_MAX_TRANS_DATA
))
2806 max_blocks
= EXT4_MAX_TRANS_DATA
;
2808 return ext4_chunk_trans_blocks(inode
, max_blocks
);
2812 * write_cache_pages_da - walk the list of dirty pages of the given
2813 * address space and call the callback function (which usually writes
2816 * This is a forked version of write_cache_pages(). Differences:
2817 * Range cyclic is ignored.
2818 * no_nrwrite_index_update is always presumed true
2820 static int write_cache_pages_da(struct address_space
*mapping
,
2821 struct writeback_control
*wbc
,
2822 struct mpage_da_data
*mpd
,
2823 pgoff_t
*done_index
)
2827 struct pagevec pvec
;
2830 pgoff_t end
; /* Inclusive */
2831 long nr_to_write
= wbc
->nr_to_write
;
2834 pagevec_init(&pvec
, 0);
2835 index
= wbc
->range_start
>> PAGE_CACHE_SHIFT
;
2836 end
= wbc
->range_end
>> PAGE_CACHE_SHIFT
;
2838 if (wbc
->sync_mode
== WB_SYNC_ALL
)
2839 tag
= PAGECACHE_TAG_TOWRITE
;
2841 tag
= PAGECACHE_TAG_DIRTY
;
2843 *done_index
= index
;
2844 while (!done
&& (index
<= end
)) {
2847 nr_pages
= pagevec_lookup_tag(&pvec
, mapping
, &index
, tag
,
2848 min(end
- index
, (pgoff_t
)PAGEVEC_SIZE
-1) + 1);
2852 for (i
= 0; i
< nr_pages
; i
++) {
2853 struct page
*page
= pvec
.pages
[i
];
2856 * At this point, the page may be truncated or
2857 * invalidated (changing page->mapping to NULL), or
2858 * even swizzled back from swapper_space to tmpfs file
2859 * mapping. However, page->index will not change
2860 * because we have a reference on the page.
2862 if (page
->index
> end
) {
2867 *done_index
= page
->index
+ 1;
2872 * Page truncated or invalidated. We can freely skip it
2873 * then, even for data integrity operations: the page
2874 * has disappeared concurrently, so there could be no
2875 * real expectation of this data interity operation
2876 * even if there is now a new, dirty page at the same
2877 * pagecache address.
2879 if (unlikely(page
->mapping
!= mapping
)) {
2885 if (!PageDirty(page
)) {
2886 /* someone wrote it for us */
2887 goto continue_unlock
;
2890 if (PageWriteback(page
)) {
2891 if (wbc
->sync_mode
!= WB_SYNC_NONE
)
2892 wait_on_page_writeback(page
);
2894 goto continue_unlock
;
2897 BUG_ON(PageWriteback(page
));
2898 if (!clear_page_dirty_for_io(page
))
2899 goto continue_unlock
;
2901 ret
= __mpage_da_writepage(page
, wbc
, mpd
);
2902 if (unlikely(ret
)) {
2903 if (ret
== AOP_WRITEPAGE_ACTIVATE
) {
2912 if (nr_to_write
> 0) {
2914 if (nr_to_write
== 0 &&
2915 wbc
->sync_mode
== WB_SYNC_NONE
) {
2917 * We stop writing back only if we are
2918 * not doing integrity sync. In case of
2919 * integrity sync we have to keep going
2920 * because someone may be concurrently
2921 * dirtying pages, and we might have
2922 * synced a lot of newly appeared dirty
2923 * pages, but have not synced all of the
2931 pagevec_release(&pvec
);
2938 static int ext4_da_writepages(struct address_space
*mapping
,
2939 struct writeback_control
*wbc
)
2942 int range_whole
= 0;
2943 handle_t
*handle
= NULL
;
2944 struct mpage_da_data mpd
;
2945 struct inode
*inode
= mapping
->host
;
2946 int pages_written
= 0;
2948 unsigned int max_pages
;
2949 int range_cyclic
, cycled
= 1, io_done
= 0;
2950 int needed_blocks
, ret
= 0;
2951 long desired_nr_to_write
, nr_to_writebump
= 0;
2952 loff_t range_start
= wbc
->range_start
;
2953 struct ext4_sb_info
*sbi
= EXT4_SB(mapping
->host
->i_sb
);
2954 pgoff_t done_index
= 0;
2957 trace_ext4_da_writepages(inode
, wbc
);
2960 * No pages to write? This is mainly a kludge to avoid starting
2961 * a transaction for special inodes like journal inode on last iput()
2962 * because that could violate lock ordering on umount
2964 if (!mapping
->nrpages
|| !mapping_tagged(mapping
, PAGECACHE_TAG_DIRTY
))
2968 * If the filesystem has aborted, it is read-only, so return
2969 * right away instead of dumping stack traces later on that
2970 * will obscure the real source of the problem. We test
2971 * EXT4_MF_FS_ABORTED instead of sb->s_flag's MS_RDONLY because
2972 * the latter could be true if the filesystem is mounted
2973 * read-only, and in that case, ext4_da_writepages should
2974 * *never* be called, so if that ever happens, we would want
2977 if (unlikely(sbi
->s_mount_flags
& EXT4_MF_FS_ABORTED
))
2980 if (wbc
->range_start
== 0 && wbc
->range_end
== LLONG_MAX
)
2983 range_cyclic
= wbc
->range_cyclic
;
2984 if (wbc
->range_cyclic
) {
2985 index
= mapping
->writeback_index
;
2988 wbc
->range_start
= index
<< PAGE_CACHE_SHIFT
;
2989 wbc
->range_end
= LLONG_MAX
;
2990 wbc
->range_cyclic
= 0;
2993 index
= wbc
->range_start
>> PAGE_CACHE_SHIFT
;
2994 end
= wbc
->range_end
>> PAGE_CACHE_SHIFT
;
2998 * This works around two forms of stupidity. The first is in
2999 * the writeback code, which caps the maximum number of pages
3000 * written to be 1024 pages. This is wrong on multiple
3001 * levels; different architectues have a different page size,
3002 * which changes the maximum amount of data which gets
3003 * written. Secondly, 4 megabytes is way too small. XFS
3004 * forces this value to be 16 megabytes by multiplying
3005 * nr_to_write parameter by four, and then relies on its
3006 * allocator to allocate larger extents to make them
3007 * contiguous. Unfortunately this brings us to the second
3008 * stupidity, which is that ext4's mballoc code only allocates
3009 * at most 2048 blocks. So we force contiguous writes up to
3010 * the number of dirty blocks in the inode, or
3011 * sbi->max_writeback_mb_bump whichever is smaller.
3013 max_pages
= sbi
->s_max_writeback_mb_bump
<< (20 - PAGE_CACHE_SHIFT
);
3014 if (!range_cyclic
&& range_whole
) {
3015 if (wbc
->nr_to_write
== LONG_MAX
)
3016 desired_nr_to_write
= wbc
->nr_to_write
;
3018 desired_nr_to_write
= wbc
->nr_to_write
* 8;
3020 desired_nr_to_write
= ext4_num_dirty_pages(inode
, index
,
3022 if (desired_nr_to_write
> max_pages
)
3023 desired_nr_to_write
= max_pages
;
3025 if (wbc
->nr_to_write
< desired_nr_to_write
) {
3026 nr_to_writebump
= desired_nr_to_write
- wbc
->nr_to_write
;
3027 wbc
->nr_to_write
= desired_nr_to_write
;
3031 mpd
.inode
= mapping
->host
;
3033 pages_skipped
= wbc
->pages_skipped
;
3036 if (wbc
->sync_mode
== WB_SYNC_ALL
)
3037 tag_pages_for_writeback(mapping
, index
, end
);
3039 while (!ret
&& wbc
->nr_to_write
> 0) {
3042 * we insert one extent at a time. So we need
3043 * credit needed for single extent allocation.
3044 * journalled mode is currently not supported
3047 BUG_ON(ext4_should_journal_data(inode
));
3048 needed_blocks
= ext4_da_writepages_trans_blocks(inode
);
3050 /* start a new transaction*/
3051 handle
= ext4_journal_start(inode
, needed_blocks
);
3052 if (IS_ERR(handle
)) {
3053 ret
= PTR_ERR(handle
);
3054 ext4_msg(inode
->i_sb
, KERN_CRIT
, "%s: jbd2_start: "
3055 "%ld pages, ino %lu; err %d", __func__
,
3056 wbc
->nr_to_write
, inode
->i_ino
, ret
);
3057 goto out_writepages
;
3061 * Now call __mpage_da_writepage to find the next
3062 * contiguous region of logical blocks that need
3063 * blocks to be allocated by ext4. We don't actually
3064 * submit the blocks for I/O here, even though
3065 * write_cache_pages thinks it will, and will set the
3066 * pages as clean for write before calling
3067 * __mpage_da_writepage().
3075 mpd
.pages_written
= 0;
3077 ret
= write_cache_pages_da(mapping
, wbc
, &mpd
, &done_index
);
3079 * If we have a contiguous extent of pages and we
3080 * haven't done the I/O yet, map the blocks and submit
3083 if (!mpd
.io_done
&& mpd
.next_page
!= mpd
.first_page
) {
3084 mpage_da_map_and_submit(&mpd
);
3085 ret
= MPAGE_DA_EXTENT_TAIL
;
3087 trace_ext4_da_write_pages(inode
, &mpd
);
3088 wbc
->nr_to_write
-= mpd
.pages_written
;
3090 ext4_journal_stop(handle
);
3092 if ((mpd
.retval
== -ENOSPC
) && sbi
->s_journal
) {
3093 /* commit the transaction which would
3094 * free blocks released in the transaction
3097 jbd2_journal_force_commit_nested(sbi
->s_journal
);
3098 wbc
->pages_skipped
= pages_skipped
;
3100 } else if (ret
== MPAGE_DA_EXTENT_TAIL
) {
3102 * got one extent now try with
3105 pages_written
+= mpd
.pages_written
;
3106 wbc
->pages_skipped
= pages_skipped
;
3109 } else if (wbc
->nr_to_write
)
3111 * There is no more writeout needed
3112 * or we requested for a noblocking writeout
3113 * and we found the device congested
3117 if (!io_done
&& !cycled
) {
3120 wbc
->range_start
= index
<< PAGE_CACHE_SHIFT
;
3121 wbc
->range_end
= mapping
->writeback_index
- 1;
3124 if (pages_skipped
!= wbc
->pages_skipped
)
3125 ext4_msg(inode
->i_sb
, KERN_CRIT
,
3126 "This should not happen leaving %s "
3127 "with nr_to_write = %ld ret = %d",
3128 __func__
, wbc
->nr_to_write
, ret
);
3131 wbc
->range_cyclic
= range_cyclic
;
3132 if (wbc
->range_cyclic
|| (range_whole
&& wbc
->nr_to_write
> 0))
3134 * set the writeback_index so that range_cyclic
3135 * mode will write it back later
3137 mapping
->writeback_index
= done_index
;
3140 wbc
->nr_to_write
-= nr_to_writebump
;
3141 wbc
->range_start
= range_start
;
3142 trace_ext4_da_writepages_result(inode
, wbc
, ret
, pages_written
);
3146 #define FALL_BACK_TO_NONDELALLOC 1
3147 static int ext4_nonda_switch(struct super_block
*sb
)
3149 s64 free_blocks
, dirty_blocks
;
3150 struct ext4_sb_info
*sbi
= EXT4_SB(sb
);
3153 * switch to non delalloc mode if we are running low
3154 * on free block. The free block accounting via percpu
3155 * counters can get slightly wrong with percpu_counter_batch getting
3156 * accumulated on each CPU without updating global counters
3157 * Delalloc need an accurate free block accounting. So switch
3158 * to non delalloc when we are near to error range.
3160 free_blocks
= percpu_counter_read_positive(&sbi
->s_freeblocks_counter
);
3161 dirty_blocks
= percpu_counter_read_positive(&sbi
->s_dirtyblocks_counter
);
3162 if (2 * free_blocks
< 3 * dirty_blocks
||
3163 free_blocks
< (dirty_blocks
+ EXT4_FREEBLOCKS_WATERMARK
)) {
3165 * free block count is less than 150% of dirty blocks
3166 * or free blocks is less than watermark
3171 * Even if we don't switch but are nearing capacity,
3172 * start pushing delalloc when 1/2 of free blocks are dirty.
3174 if (free_blocks
< 2 * dirty_blocks
)
3175 writeback_inodes_sb_if_idle(sb
);
3180 static int ext4_da_write_begin(struct file
*file
, struct address_space
*mapping
,
3181 loff_t pos
, unsigned len
, unsigned flags
,
3182 struct page
**pagep
, void **fsdata
)
3184 int ret
, retries
= 0;
3187 struct inode
*inode
= mapping
->host
;
3190 index
= pos
>> PAGE_CACHE_SHIFT
;
3192 if (ext4_nonda_switch(inode
->i_sb
)) {
3193 *fsdata
= (void *)FALL_BACK_TO_NONDELALLOC
;
3194 return ext4_write_begin(file
, mapping
, pos
,
3195 len
, flags
, pagep
, fsdata
);
3197 *fsdata
= (void *)0;
3198 trace_ext4_da_write_begin(inode
, pos
, len
, flags
);
3201 * With delayed allocation, we don't log the i_disksize update
3202 * if there is delayed block allocation. But we still need
3203 * to journalling the i_disksize update if writes to the end
3204 * of file which has an already mapped buffer.
3206 handle
= ext4_journal_start(inode
, 1);
3207 if (IS_ERR(handle
)) {
3208 ret
= PTR_ERR(handle
);
3211 /* We cannot recurse into the filesystem as the transaction is already
3213 flags
|= AOP_FLAG_NOFS
;
3215 page
= grab_cache_page_write_begin(mapping
, index
, flags
);
3217 ext4_journal_stop(handle
);
3223 ret
= __block_write_begin(page
, pos
, len
, ext4_da_get_block_prep
);
3226 ext4_journal_stop(handle
);
3227 page_cache_release(page
);
3229 * block_write_begin may have instantiated a few blocks
3230 * outside i_size. Trim these off again. Don't need
3231 * i_size_read because we hold i_mutex.
3233 if (pos
+ len
> inode
->i_size
)
3234 ext4_truncate_failed_write(inode
);
3237 if (ret
== -ENOSPC
&& ext4_should_retry_alloc(inode
->i_sb
, &retries
))
3244 * Check if we should update i_disksize
3245 * when write to the end of file but not require block allocation
3247 static int ext4_da_should_update_i_disksize(struct page
*page
,
3248 unsigned long offset
)
3250 struct buffer_head
*bh
;
3251 struct inode
*inode
= page
->mapping
->host
;
3255 bh
= page_buffers(page
);
3256 idx
= offset
>> inode
->i_blkbits
;
3258 for (i
= 0; i
< idx
; i
++)
3259 bh
= bh
->b_this_page
;
3261 if (!buffer_mapped(bh
) || (buffer_delay(bh
)) || buffer_unwritten(bh
))
3266 static int ext4_da_write_end(struct file
*file
,
3267 struct address_space
*mapping
,
3268 loff_t pos
, unsigned len
, unsigned copied
,
3269 struct page
*page
, void *fsdata
)
3271 struct inode
*inode
= mapping
->host
;
3273 handle_t
*handle
= ext4_journal_current_handle();
3275 unsigned long start
, end
;
3276 int write_mode
= (int)(unsigned long)fsdata
;
3278 if (write_mode
== FALL_BACK_TO_NONDELALLOC
) {
3279 if (ext4_should_order_data(inode
)) {
3280 return ext4_ordered_write_end(file
, mapping
, pos
,
3281 len
, copied
, page
, fsdata
);
3282 } else if (ext4_should_writeback_data(inode
)) {
3283 return ext4_writeback_write_end(file
, mapping
, pos
,
3284 len
, copied
, page
, fsdata
);
3290 trace_ext4_da_write_end(inode
, pos
, len
, copied
);
3291 start
= pos
& (PAGE_CACHE_SIZE
- 1);
3292 end
= start
+ copied
- 1;
3295 * generic_write_end() will run mark_inode_dirty() if i_size
3296 * changes. So let's piggyback the i_disksize mark_inode_dirty
3300 new_i_size
= pos
+ copied
;
3301 if (new_i_size
> EXT4_I(inode
)->i_disksize
) {
3302 if (ext4_da_should_update_i_disksize(page
, end
)) {
3303 down_write(&EXT4_I(inode
)->i_data_sem
);
3304 if (new_i_size
> EXT4_I(inode
)->i_disksize
) {
3306 * Updating i_disksize when extending file
3307 * without needing block allocation
3309 if (ext4_should_order_data(inode
))
3310 ret
= ext4_jbd2_file_inode(handle
,
3313 EXT4_I(inode
)->i_disksize
= new_i_size
;
3315 up_write(&EXT4_I(inode
)->i_data_sem
);
3316 /* We need to mark inode dirty even if
3317 * new_i_size is less that inode->i_size
3318 * bu greater than i_disksize.(hint delalloc)
3320 ext4_mark_inode_dirty(handle
, inode
);
3323 ret2
= generic_write_end(file
, mapping
, pos
, len
, copied
,
3328 ret2
= ext4_journal_stop(handle
);
3332 return ret
? ret
: copied
;
3335 static void ext4_da_invalidatepage(struct page
*page
, unsigned long offset
)
3338 * Drop reserved blocks
3340 BUG_ON(!PageLocked(page
));
3341 if (!page_has_buffers(page
))
3344 ext4_da_page_release_reservation(page
, offset
);
3347 ext4_invalidatepage(page
, offset
);
3353 * Force all delayed allocation blocks to be allocated for a given inode.
3355 int ext4_alloc_da_blocks(struct inode
*inode
)
3357 trace_ext4_alloc_da_blocks(inode
);
3359 if (!EXT4_I(inode
)->i_reserved_data_blocks
&&
3360 !EXT4_I(inode
)->i_reserved_meta_blocks
)
3364 * We do something simple for now. The filemap_flush() will
3365 * also start triggering a write of the data blocks, which is
3366 * not strictly speaking necessary (and for users of
3367 * laptop_mode, not even desirable). However, to do otherwise
3368 * would require replicating code paths in:
3370 * ext4_da_writepages() ->
3371 * write_cache_pages() ---> (via passed in callback function)
3372 * __mpage_da_writepage() -->
3373 * mpage_add_bh_to_extent()
3374 * mpage_da_map_blocks()
3376 * The problem is that write_cache_pages(), located in
3377 * mm/page-writeback.c, marks pages clean in preparation for
3378 * doing I/O, which is not desirable if we're not planning on
3381 * We could call write_cache_pages(), and then redirty all of
3382 * the pages by calling redirty_page_for_writeback() but that
3383 * would be ugly in the extreme. So instead we would need to
3384 * replicate parts of the code in the above functions,
3385 * simplifying them becuase we wouldn't actually intend to
3386 * write out the pages, but rather only collect contiguous
3387 * logical block extents, call the multi-block allocator, and
3388 * then update the buffer heads with the block allocations.
3390 * For now, though, we'll cheat by calling filemap_flush(),
3391 * which will map the blocks, and start the I/O, but not
3392 * actually wait for the I/O to complete.
3394 return filemap_flush(inode
->i_mapping
);
3398 * bmap() is special. It gets used by applications such as lilo and by
3399 * the swapper to find the on-disk block of a specific piece of data.
3401 * Naturally, this is dangerous if the block concerned is still in the
3402 * journal. If somebody makes a swapfile on an ext4 data-journaling
3403 * filesystem and enables swap, then they may get a nasty shock when the
3404 * data getting swapped to that swapfile suddenly gets overwritten by
3405 * the original zero's written out previously to the journal and
3406 * awaiting writeback in the kernel's buffer cache.
3408 * So, if we see any bmap calls here on a modified, data-journaled file,
3409 * take extra steps to flush any blocks which might be in the cache.
3411 static sector_t
ext4_bmap(struct address_space
*mapping
, sector_t block
)
3413 struct inode
*inode
= mapping
->host
;
3417 if (mapping_tagged(mapping
, PAGECACHE_TAG_DIRTY
) &&
3418 test_opt(inode
->i_sb
, DELALLOC
)) {
3420 * With delalloc we want to sync the file
3421 * so that we can make sure we allocate
3424 filemap_write_and_wait(mapping
);
3427 if (EXT4_JOURNAL(inode
) &&
3428 ext4_test_inode_state(inode
, EXT4_STATE_JDATA
)) {
3430 * This is a REALLY heavyweight approach, but the use of
3431 * bmap on dirty files is expected to be extremely rare:
3432 * only if we run lilo or swapon on a freshly made file
3433 * do we expect this to happen.
3435 * (bmap requires CAP_SYS_RAWIO so this does not
3436 * represent an unprivileged user DOS attack --- we'd be
3437 * in trouble if mortal users could trigger this path at
3440 * NB. EXT4_STATE_JDATA is not set on files other than
3441 * regular files. If somebody wants to bmap a directory
3442 * or symlink and gets confused because the buffer
3443 * hasn't yet been flushed to disk, they deserve
3444 * everything they get.
3447 ext4_clear_inode_state(inode
, EXT4_STATE_JDATA
);
3448 journal
= EXT4_JOURNAL(inode
);
3449 jbd2_journal_lock_updates(journal
);
3450 err
= jbd2_journal_flush(journal
);
3451 jbd2_journal_unlock_updates(journal
);
3457 return generic_block_bmap(mapping
, block
, ext4_get_block
);
3460 static int ext4_readpage(struct file
*file
, struct page
*page
)
3462 return mpage_readpage(page
, ext4_get_block
);
3466 ext4_readpages(struct file
*file
, struct address_space
*mapping
,
3467 struct list_head
*pages
, unsigned nr_pages
)
3469 return mpage_readpages(mapping
, pages
, nr_pages
, ext4_get_block
);
3472 static void ext4_invalidatepage_free_endio(struct page
*page
, unsigned long offset
)
3474 struct buffer_head
*head
, *bh
;
3475 unsigned int curr_off
= 0;
3477 if (!page_has_buffers(page
))
3479 head
= bh
= page_buffers(page
);
3481 if (offset
<= curr_off
&& test_clear_buffer_uninit(bh
)
3483 ext4_free_io_end(bh
->b_private
);
3484 bh
->b_private
= NULL
;
3485 bh
->b_end_io
= NULL
;
3487 curr_off
= curr_off
+ bh
->b_size
;
3488 bh
= bh
->b_this_page
;
3489 } while (bh
!= head
);
3492 static void ext4_invalidatepage(struct page
*page
, unsigned long offset
)
3494 journal_t
*journal
= EXT4_JOURNAL(page
->mapping
->host
);
3497 * free any io_end structure allocated for buffers to be discarded
3499 if (ext4_should_dioread_nolock(page
->mapping
->host
))
3500 ext4_invalidatepage_free_endio(page
, offset
);
3502 * If it's a full truncate we just forget about the pending dirtying
3505 ClearPageChecked(page
);
3508 jbd2_journal_invalidatepage(journal
, page
, offset
);
3510 block_invalidatepage(page
, offset
);
3513 static int ext4_releasepage(struct page
*page
, gfp_t wait
)
3515 journal_t
*journal
= EXT4_JOURNAL(page
->mapping
->host
);
3517 WARN_ON(PageChecked(page
));
3518 if (!page_has_buffers(page
))
3521 return jbd2_journal_try_to_free_buffers(journal
, page
, wait
);
3523 return try_to_free_buffers(page
);
3527 * O_DIRECT for ext3 (or indirect map) based files
3529 * If the O_DIRECT write will extend the file then add this inode to the
3530 * orphan list. So recovery will truncate it back to the original size
3531 * if the machine crashes during the write.
3533 * If the O_DIRECT write is intantiating holes inside i_size and the machine
3534 * crashes then stale disk data _may_ be exposed inside the file. But current
3535 * VFS code falls back into buffered path in that case so we are safe.
3537 static ssize_t
ext4_ind_direct_IO(int rw
, struct kiocb
*iocb
,
3538 const struct iovec
*iov
, loff_t offset
,
3539 unsigned long nr_segs
)
3541 struct file
*file
= iocb
->ki_filp
;
3542 struct inode
*inode
= file
->f_mapping
->host
;
3543 struct ext4_inode_info
*ei
= EXT4_I(inode
);
3547 size_t count
= iov_length(iov
, nr_segs
);
3551 loff_t final_size
= offset
+ count
;
3553 if (final_size
> inode
->i_size
) {
3554 /* Credits for sb + inode write */
3555 handle
= ext4_journal_start(inode
, 2);
3556 if (IS_ERR(handle
)) {
3557 ret
= PTR_ERR(handle
);
3560 ret
= ext4_orphan_add(handle
, inode
);
3562 ext4_journal_stop(handle
);
3566 ei
->i_disksize
= inode
->i_size
;
3567 ext4_journal_stop(handle
);
3572 if (rw
== READ
&& ext4_should_dioread_nolock(inode
))
3573 ret
= __blockdev_direct_IO(rw
, iocb
, inode
,
3574 inode
->i_sb
->s_bdev
, iov
,
3576 ext4_get_block
, NULL
, NULL
, 0);
3578 ret
= blockdev_direct_IO(rw
, iocb
, inode
,
3579 inode
->i_sb
->s_bdev
, iov
,
3581 ext4_get_block
, NULL
);
3583 if (unlikely((rw
& WRITE
) && ret
< 0)) {
3584 loff_t isize
= i_size_read(inode
);
3585 loff_t end
= offset
+ iov_length(iov
, nr_segs
);
3588 vmtruncate(inode
, isize
);
3591 if (ret
== -ENOSPC
&& ext4_should_retry_alloc(inode
->i_sb
, &retries
))
3597 /* Credits for sb + inode write */
3598 handle
= ext4_journal_start(inode
, 2);
3599 if (IS_ERR(handle
)) {
3600 /* This is really bad luck. We've written the data
3601 * but cannot extend i_size. Bail out and pretend
3602 * the write failed... */
3603 ret
= PTR_ERR(handle
);
3605 ext4_orphan_del(NULL
, inode
);
3610 ext4_orphan_del(handle
, inode
);
3612 loff_t end
= offset
+ ret
;
3613 if (end
> inode
->i_size
) {
3614 ei
->i_disksize
= end
;
3615 i_size_write(inode
, end
);
3617 * We're going to return a positive `ret'
3618 * here due to non-zero-length I/O, so there's
3619 * no way of reporting error returns from
3620 * ext4_mark_inode_dirty() to userspace. So
3623 ext4_mark_inode_dirty(handle
, inode
);
3626 err
= ext4_journal_stop(handle
);
3635 * ext4_get_block used when preparing for a DIO write or buffer write.
3636 * We allocate an uinitialized extent if blocks haven't been allocated.
3637 * The extent will be converted to initialized after the IO is complete.
3639 static int ext4_get_block_write(struct inode
*inode
, sector_t iblock
,
3640 struct buffer_head
*bh_result
, int create
)
3642 ext4_debug("ext4_get_block_write: inode %lu, create flag %d\n",
3643 inode
->i_ino
, create
);
3644 return _ext4_get_block(inode
, iblock
, bh_result
,
3645 EXT4_GET_BLOCKS_IO_CREATE_EXT
);
3648 static void ext4_end_io_dio(struct kiocb
*iocb
, loff_t offset
,
3649 ssize_t size
, void *private, int ret
,
3652 ext4_io_end_t
*io_end
= iocb
->private;
3653 struct workqueue_struct
*wq
;
3654 unsigned long flags
;
3655 struct ext4_inode_info
*ei
;
3657 /* if not async direct IO or dio with 0 bytes write, just return */
3658 if (!io_end
|| !size
)
3661 ext_debug("ext4_end_io_dio(): io_end 0x%p"
3662 "for inode %lu, iocb 0x%p, offset %llu, size %llu\n",
3663 iocb
->private, io_end
->inode
->i_ino
, iocb
, offset
,
3666 /* if not aio dio with unwritten extents, just free io and return */
3667 if (!(io_end
->flag
& EXT4_IO_END_UNWRITTEN
)) {
3668 ext4_free_io_end(io_end
);
3669 iocb
->private = NULL
;
3672 aio_complete(iocb
, ret
, 0);
3676 io_end
->offset
= offset
;
3677 io_end
->size
= size
;
3679 io_end
->iocb
= iocb
;
3680 io_end
->result
= ret
;
3682 wq
= EXT4_SB(io_end
->inode
->i_sb
)->dio_unwritten_wq
;
3684 /* Add the io_end to per-inode completed aio dio list*/
3685 ei
= EXT4_I(io_end
->inode
);
3686 spin_lock_irqsave(&ei
->i_completed_io_lock
, flags
);
3687 list_add_tail(&io_end
->list
, &ei
->i_completed_io_list
);
3688 spin_unlock_irqrestore(&ei
->i_completed_io_lock
, flags
);
3690 /* queue the work to convert unwritten extents to written */
3691 queue_work(wq
, &io_end
->work
);
3692 iocb
->private = NULL
;
3695 static void ext4_end_io_buffer_write(struct buffer_head
*bh
, int uptodate
)
3697 ext4_io_end_t
*io_end
= bh
->b_private
;
3698 struct workqueue_struct
*wq
;
3699 struct inode
*inode
;
3700 unsigned long flags
;
3702 if (!test_clear_buffer_uninit(bh
) || !io_end
)
3705 if (!(io_end
->inode
->i_sb
->s_flags
& MS_ACTIVE
)) {
3706 printk("sb umounted, discard end_io request for inode %lu\n",
3707 io_end
->inode
->i_ino
);
3708 ext4_free_io_end(io_end
);
3712 io_end
->flag
= EXT4_IO_END_UNWRITTEN
;
3713 inode
= io_end
->inode
;
3715 /* Add the io_end to per-inode completed io list*/
3716 spin_lock_irqsave(&EXT4_I(inode
)->i_completed_io_lock
, flags
);
3717 list_add_tail(&io_end
->list
, &EXT4_I(inode
)->i_completed_io_list
);
3718 spin_unlock_irqrestore(&EXT4_I(inode
)->i_completed_io_lock
, flags
);
3720 wq
= EXT4_SB(inode
->i_sb
)->dio_unwritten_wq
;
3721 /* queue the work to convert unwritten extents to written */
3722 queue_work(wq
, &io_end
->work
);
3724 bh
->b_private
= NULL
;
3725 bh
->b_end_io
= NULL
;
3726 clear_buffer_uninit(bh
);
3727 end_buffer_async_write(bh
, uptodate
);
3730 static int ext4_set_bh_endio(struct buffer_head
*bh
, struct inode
*inode
)
3732 ext4_io_end_t
*io_end
;
3733 struct page
*page
= bh
->b_page
;
3734 loff_t offset
= (sector_t
)page
->index
<< PAGE_CACHE_SHIFT
;
3735 size_t size
= bh
->b_size
;
3738 io_end
= ext4_init_io_end(inode
, GFP_ATOMIC
);
3740 pr_warning_ratelimited("%s: allocation fail\n", __func__
);
3744 io_end
->offset
= offset
;
3745 io_end
->size
= size
;
3747 * We need to hold a reference to the page to make sure it
3748 * doesn't get evicted before ext4_end_io_work() has a chance
3749 * to convert the extent from written to unwritten.
3751 io_end
->page
= page
;
3752 get_page(io_end
->page
);
3754 bh
->b_private
= io_end
;
3755 bh
->b_end_io
= ext4_end_io_buffer_write
;
3760 * For ext4 extent files, ext4 will do direct-io write to holes,
3761 * preallocated extents, and those write extend the file, no need to
3762 * fall back to buffered IO.
3764 * For holes, we fallocate those blocks, mark them as unintialized
3765 * If those blocks were preallocated, we mark sure they are splited, but
3766 * still keep the range to write as unintialized.
3768 * The unwrritten extents will be converted to written when DIO is completed.
3769 * For async direct IO, since the IO may still pending when return, we
3770 * set up an end_io call back function, which will do the convertion
3771 * when async direct IO completed.
3773 * If the O_DIRECT write will extend the file then add this inode to the
3774 * orphan list. So recovery will truncate it back to the original size
3775 * if the machine crashes during the write.
3778 static ssize_t
ext4_ext_direct_IO(int rw
, struct kiocb
*iocb
,
3779 const struct iovec
*iov
, loff_t offset
,
3780 unsigned long nr_segs
)
3782 struct file
*file
= iocb
->ki_filp
;
3783 struct inode
*inode
= file
->f_mapping
->host
;
3785 size_t count
= iov_length(iov
, nr_segs
);
3787 loff_t final_size
= offset
+ count
;
3788 if (rw
== WRITE
&& final_size
<= inode
->i_size
) {
3790 * We could direct write to holes and fallocate.
3792 * Allocated blocks to fill the hole are marked as uninitialized
3793 * to prevent paralel buffered read to expose the stale data
3794 * before DIO complete the data IO.
3796 * As to previously fallocated extents, ext4 get_block
3797 * will just simply mark the buffer mapped but still
3798 * keep the extents uninitialized.
3800 * for non AIO case, we will convert those unwritten extents
3801 * to written after return back from blockdev_direct_IO.
3803 * for async DIO, the conversion needs to be defered when
3804 * the IO is completed. The ext4 end_io callback function
3805 * will be called to take care of the conversion work.
3806 * Here for async case, we allocate an io_end structure to
3809 iocb
->private = NULL
;
3810 EXT4_I(inode
)->cur_aio_dio
= NULL
;
3811 if (!is_sync_kiocb(iocb
)) {
3812 iocb
->private = ext4_init_io_end(inode
, GFP_NOFS
);
3816 * we save the io structure for current async
3817 * direct IO, so that later ext4_map_blocks()
3818 * could flag the io structure whether there
3819 * is a unwritten extents needs to be converted
3820 * when IO is completed.
3822 EXT4_I(inode
)->cur_aio_dio
= iocb
->private;
3825 ret
= blockdev_direct_IO(rw
, iocb
, inode
,
3826 inode
->i_sb
->s_bdev
, iov
,
3828 ext4_get_block_write
,
3831 EXT4_I(inode
)->cur_aio_dio
= NULL
;
3833 * The io_end structure takes a reference to the inode,
3834 * that structure needs to be destroyed and the
3835 * reference to the inode need to be dropped, when IO is
3836 * complete, even with 0 byte write, or failed.
3838 * In the successful AIO DIO case, the io_end structure will be
3839 * desctroyed and the reference to the inode will be dropped
3840 * after the end_io call back function is called.
3842 * In the case there is 0 byte write, or error case, since
3843 * VFS direct IO won't invoke the end_io call back function,
3844 * we need to free the end_io structure here.
3846 if (ret
!= -EIOCBQUEUED
&& ret
<= 0 && iocb
->private) {
3847 ext4_free_io_end(iocb
->private);
3848 iocb
->private = NULL
;
3849 } else if (ret
> 0 && ext4_test_inode_state(inode
,
3850 EXT4_STATE_DIO_UNWRITTEN
)) {
3853 * for non AIO case, since the IO is already
3854 * completed, we could do the convertion right here
3856 err
= ext4_convert_unwritten_extents(inode
,
3860 ext4_clear_inode_state(inode
, EXT4_STATE_DIO_UNWRITTEN
);
3865 /* for write the the end of file case, we fall back to old way */
3866 return ext4_ind_direct_IO(rw
, iocb
, iov
, offset
, nr_segs
);
3869 static ssize_t
ext4_direct_IO(int rw
, struct kiocb
*iocb
,
3870 const struct iovec
*iov
, loff_t offset
,
3871 unsigned long nr_segs
)
3873 struct file
*file
= iocb
->ki_filp
;
3874 struct inode
*inode
= file
->f_mapping
->host
;
3876 if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
))
3877 return ext4_ext_direct_IO(rw
, iocb
, iov
, offset
, nr_segs
);
3879 return ext4_ind_direct_IO(rw
, iocb
, iov
, offset
, nr_segs
);
3883 * Pages can be marked dirty completely asynchronously from ext4's journalling
3884 * activity. By filemap_sync_pte(), try_to_unmap_one(), etc. We cannot do
3885 * much here because ->set_page_dirty is called under VFS locks. The page is
3886 * not necessarily locked.
3888 * We cannot just dirty the page and leave attached buffers clean, because the
3889 * buffers' dirty state is "definitive". We cannot just set the buffers dirty
3890 * or jbddirty because all the journalling code will explode.
3892 * So what we do is to mark the page "pending dirty" and next time writepage
3893 * is called, propagate that into the buffers appropriately.
3895 static int ext4_journalled_set_page_dirty(struct page
*page
)
3897 SetPageChecked(page
);
3898 return __set_page_dirty_nobuffers(page
);
3901 static const struct address_space_operations ext4_ordered_aops
= {
3902 .readpage
= ext4_readpage
,
3903 .readpages
= ext4_readpages
,
3904 .writepage
= ext4_writepage
,
3905 .sync_page
= block_sync_page
,
3906 .write_begin
= ext4_write_begin
,
3907 .write_end
= ext4_ordered_write_end
,
3909 .invalidatepage
= ext4_invalidatepage
,
3910 .releasepage
= ext4_releasepage
,
3911 .direct_IO
= ext4_direct_IO
,
3912 .migratepage
= buffer_migrate_page
,
3913 .is_partially_uptodate
= block_is_partially_uptodate
,
3914 .error_remove_page
= generic_error_remove_page
,
3917 static const struct address_space_operations ext4_writeback_aops
= {
3918 .readpage
= ext4_readpage
,
3919 .readpages
= ext4_readpages
,
3920 .writepage
= ext4_writepage
,
3921 .sync_page
= block_sync_page
,
3922 .write_begin
= ext4_write_begin
,
3923 .write_end
= ext4_writeback_write_end
,
3925 .invalidatepage
= ext4_invalidatepage
,
3926 .releasepage
= ext4_releasepage
,
3927 .direct_IO
= ext4_direct_IO
,
3928 .migratepage
= buffer_migrate_page
,
3929 .is_partially_uptodate
= block_is_partially_uptodate
,
3930 .error_remove_page
= generic_error_remove_page
,
3933 static const struct address_space_operations ext4_journalled_aops
= {
3934 .readpage
= ext4_readpage
,
3935 .readpages
= ext4_readpages
,
3936 .writepage
= ext4_writepage
,
3937 .sync_page
= block_sync_page
,
3938 .write_begin
= ext4_write_begin
,
3939 .write_end
= ext4_journalled_write_end
,
3940 .set_page_dirty
= ext4_journalled_set_page_dirty
,
3942 .invalidatepage
= ext4_invalidatepage
,
3943 .releasepage
= ext4_releasepage
,
3944 .is_partially_uptodate
= block_is_partially_uptodate
,
3945 .error_remove_page
= generic_error_remove_page
,
3948 static const struct address_space_operations ext4_da_aops
= {
3949 .readpage
= ext4_readpage
,
3950 .readpages
= ext4_readpages
,
3951 .writepage
= ext4_writepage
,
3952 .writepages
= ext4_da_writepages
,
3953 .sync_page
= block_sync_page
,
3954 .write_begin
= ext4_da_write_begin
,
3955 .write_end
= ext4_da_write_end
,
3957 .invalidatepage
= ext4_da_invalidatepage
,
3958 .releasepage
= ext4_releasepage
,
3959 .direct_IO
= ext4_direct_IO
,
3960 .migratepage
= buffer_migrate_page
,
3961 .is_partially_uptodate
= block_is_partially_uptodate
,
3962 .error_remove_page
= generic_error_remove_page
,
3965 void ext4_set_aops(struct inode
*inode
)
3967 if (ext4_should_order_data(inode
) &&
3968 test_opt(inode
->i_sb
, DELALLOC
))
3969 inode
->i_mapping
->a_ops
= &ext4_da_aops
;
3970 else if (ext4_should_order_data(inode
))
3971 inode
->i_mapping
->a_ops
= &ext4_ordered_aops
;
3972 else if (ext4_should_writeback_data(inode
) &&
3973 test_opt(inode
->i_sb
, DELALLOC
))
3974 inode
->i_mapping
->a_ops
= &ext4_da_aops
;
3975 else if (ext4_should_writeback_data(inode
))
3976 inode
->i_mapping
->a_ops
= &ext4_writeback_aops
;
3978 inode
->i_mapping
->a_ops
= &ext4_journalled_aops
;
3982 * ext4_block_truncate_page() zeroes out a mapping from file offset `from'
3983 * up to the end of the block which corresponds to `from'.
3984 * This required during truncate. We need to physically zero the tail end
3985 * of that block so it doesn't yield old data if the file is later grown.
3987 int ext4_block_truncate_page(handle_t
*handle
,
3988 struct address_space
*mapping
, loff_t from
)
3990 ext4_fsblk_t index
= from
>> PAGE_CACHE_SHIFT
;
3991 unsigned offset
= from
& (PAGE_CACHE_SIZE
-1);
3992 unsigned blocksize
, length
, pos
;
3994 struct inode
*inode
= mapping
->host
;
3995 struct buffer_head
*bh
;
3999 page
= find_or_create_page(mapping
, from
>> PAGE_CACHE_SHIFT
,
4000 mapping_gfp_mask(mapping
) & ~__GFP_FS
);
4004 blocksize
= inode
->i_sb
->s_blocksize
;
4005 length
= blocksize
- (offset
& (blocksize
- 1));
4006 iblock
= index
<< (PAGE_CACHE_SHIFT
- inode
->i_sb
->s_blocksize_bits
);
4008 if (!page_has_buffers(page
))
4009 create_empty_buffers(page
, blocksize
, 0);
4011 /* Find the buffer that contains "offset" */
4012 bh
= page_buffers(page
);
4014 while (offset
>= pos
) {
4015 bh
= bh
->b_this_page
;
4021 if (buffer_freed(bh
)) {
4022 BUFFER_TRACE(bh
, "freed: skip");
4026 if (!buffer_mapped(bh
)) {
4027 BUFFER_TRACE(bh
, "unmapped");
4028 ext4_get_block(inode
, iblock
, bh
, 0);
4029 /* unmapped? It's a hole - nothing to do */
4030 if (!buffer_mapped(bh
)) {
4031 BUFFER_TRACE(bh
, "still unmapped");
4036 /* Ok, it's mapped. Make sure it's up-to-date */
4037 if (PageUptodate(page
))
4038 set_buffer_uptodate(bh
);
4040 if (!buffer_uptodate(bh
)) {
4042 ll_rw_block(READ
, 1, &bh
);
4044 /* Uhhuh. Read error. Complain and punt. */
4045 if (!buffer_uptodate(bh
))
4049 if (ext4_should_journal_data(inode
)) {
4050 BUFFER_TRACE(bh
, "get write access");
4051 err
= ext4_journal_get_write_access(handle
, bh
);
4056 zero_user(page
, offset
, length
);
4058 BUFFER_TRACE(bh
, "zeroed end of block");
4061 if (ext4_should_journal_data(inode
)) {
4062 err
= ext4_handle_dirty_metadata(handle
, inode
, bh
);
4064 if (ext4_should_order_data(inode
) && EXT4_I(inode
)->jinode
)
4065 err
= ext4_jbd2_file_inode(handle
, inode
);
4066 mark_buffer_dirty(bh
);
4071 page_cache_release(page
);
4076 * Probably it should be a library function... search for first non-zero word
4077 * or memcmp with zero_page, whatever is better for particular architecture.
4080 static inline int all_zeroes(__le32
*p
, __le32
*q
)
4089 * ext4_find_shared - find the indirect blocks for partial truncation.
4090 * @inode: inode in question
4091 * @depth: depth of the affected branch
4092 * @offsets: offsets of pointers in that branch (see ext4_block_to_path)
4093 * @chain: place to store the pointers to partial indirect blocks
4094 * @top: place to the (detached) top of branch
4096 * This is a helper function used by ext4_truncate().
4098 * When we do truncate() we may have to clean the ends of several
4099 * indirect blocks but leave the blocks themselves alive. Block is
4100 * partially truncated if some data below the new i_size is refered
4101 * from it (and it is on the path to the first completely truncated
4102 * data block, indeed). We have to free the top of that path along
4103 * with everything to the right of the path. Since no allocation
4104 * past the truncation point is possible until ext4_truncate()
4105 * finishes, we may safely do the latter, but top of branch may
4106 * require special attention - pageout below the truncation point
4107 * might try to populate it.
4109 * We atomically detach the top of branch from the tree, store the
4110 * block number of its root in *@top, pointers to buffer_heads of
4111 * partially truncated blocks - in @chain[].bh and pointers to
4112 * their last elements that should not be removed - in
4113 * @chain[].p. Return value is the pointer to last filled element
4116 * The work left to caller to do the actual freeing of subtrees:
4117 * a) free the subtree starting from *@top
4118 * b) free the subtrees whose roots are stored in
4119 * (@chain[i].p+1 .. end of @chain[i].bh->b_data)
4120 * c) free the subtrees growing from the inode past the @chain[0].
4121 * (no partially truncated stuff there). */
4123 static Indirect
*ext4_find_shared(struct inode
*inode
, int depth
,
4124 ext4_lblk_t offsets
[4], Indirect chain
[4],
4127 Indirect
*partial
, *p
;
4131 /* Make k index the deepest non-null offset + 1 */
4132 for (k
= depth
; k
> 1 && !offsets
[k
-1]; k
--)
4134 partial
= ext4_get_branch(inode
, k
, offsets
, chain
, &err
);
4135 /* Writer: pointers */
4137 partial
= chain
+ k
-1;
4139 * If the branch acquired continuation since we've looked at it -
4140 * fine, it should all survive and (new) top doesn't belong to us.
4142 if (!partial
->key
&& *partial
->p
)
4145 for (p
= partial
; (p
> chain
) && all_zeroes((__le32
*) p
->bh
->b_data
, p
->p
); p
--)
4148 * OK, we've found the last block that must survive. The rest of our
4149 * branch should be detached before unlocking. However, if that rest
4150 * of branch is all ours and does not grow immediately from the inode
4151 * it's easier to cheat and just decrement partial->p.
4153 if (p
== chain
+ k
- 1 && p
> chain
) {
4157 /* Nope, don't do this in ext4. Must leave the tree intact */
4164 while (partial
> p
) {
4165 brelse(partial
->bh
);
4173 * Zero a number of block pointers in either an inode or an indirect block.
4174 * If we restart the transaction we must again get write access to the
4175 * indirect block for further modification.
4177 * We release `count' blocks on disk, but (last - first) may be greater
4178 * than `count' because there can be holes in there.
4180 static int ext4_clear_blocks(handle_t
*handle
, struct inode
*inode
,
4181 struct buffer_head
*bh
,
4182 ext4_fsblk_t block_to_free
,
4183 unsigned long count
, __le32
*first
,
4187 int flags
= EXT4_FREE_BLOCKS_FORGET
| EXT4_FREE_BLOCKS_VALIDATED
;
4190 if (S_ISDIR(inode
->i_mode
) || S_ISLNK(inode
->i_mode
))
4191 flags
|= EXT4_FREE_BLOCKS_METADATA
;
4193 if (!ext4_data_block_valid(EXT4_SB(inode
->i_sb
), block_to_free
,
4195 EXT4_ERROR_INODE(inode
, "attempt to clear invalid "
4196 "blocks %llu len %lu",
4197 (unsigned long long) block_to_free
, count
);
4201 if (try_to_extend_transaction(handle
, inode
)) {
4203 BUFFER_TRACE(bh
, "call ext4_handle_dirty_metadata");
4204 err
= ext4_handle_dirty_metadata(handle
, inode
, bh
);
4205 if (unlikely(err
)) {
4206 ext4_std_error(inode
->i_sb
, err
);
4210 err
= ext4_mark_inode_dirty(handle
, inode
);
4211 if (unlikely(err
)) {
4212 ext4_std_error(inode
->i_sb
, err
);
4215 err
= ext4_truncate_restart_trans(handle
, inode
,
4216 blocks_for_truncate(inode
));
4217 if (unlikely(err
)) {
4218 ext4_std_error(inode
->i_sb
, err
);
4222 BUFFER_TRACE(bh
, "retaking write access");
4223 ext4_journal_get_write_access(handle
, bh
);
4227 for (p
= first
; p
< last
; p
++)
4230 ext4_free_blocks(handle
, inode
, 0, block_to_free
, count
, flags
);
4235 * ext4_free_data - free a list of data blocks
4236 * @handle: handle for this transaction
4237 * @inode: inode we are dealing with
4238 * @this_bh: indirect buffer_head which contains *@first and *@last
4239 * @first: array of block numbers
4240 * @last: points immediately past the end of array
4242 * We are freeing all blocks refered from that array (numbers are stored as
4243 * little-endian 32-bit) and updating @inode->i_blocks appropriately.
4245 * We accumulate contiguous runs of blocks to free. Conveniently, if these
4246 * blocks are contiguous then releasing them at one time will only affect one
4247 * or two bitmap blocks (+ group descriptor(s) and superblock) and we won't
4248 * actually use a lot of journal space.
4250 * @this_bh will be %NULL if @first and @last point into the inode's direct
4253 static void ext4_free_data(handle_t
*handle
, struct inode
*inode
,
4254 struct buffer_head
*this_bh
,
4255 __le32
*first
, __le32
*last
)
4257 ext4_fsblk_t block_to_free
= 0; /* Starting block # of a run */
4258 unsigned long count
= 0; /* Number of blocks in the run */
4259 __le32
*block_to_free_p
= NULL
; /* Pointer into inode/ind
4262 ext4_fsblk_t nr
; /* Current block # */
4263 __le32
*p
; /* Pointer into inode/ind
4264 for current block */
4267 if (this_bh
) { /* For indirect block */
4268 BUFFER_TRACE(this_bh
, "get_write_access");
4269 err
= ext4_journal_get_write_access(handle
, this_bh
);
4270 /* Important: if we can't update the indirect pointers
4271 * to the blocks, we can't free them. */
4276 for (p
= first
; p
< last
; p
++) {
4277 nr
= le32_to_cpu(*p
);
4279 /* accumulate blocks to free if they're contiguous */
4282 block_to_free_p
= p
;
4284 } else if (nr
== block_to_free
+ count
) {
4287 if (ext4_clear_blocks(handle
, inode
, this_bh
,
4288 block_to_free
, count
,
4289 block_to_free_p
, p
))
4292 block_to_free_p
= p
;
4299 ext4_clear_blocks(handle
, inode
, this_bh
, block_to_free
,
4300 count
, block_to_free_p
, p
);
4303 BUFFER_TRACE(this_bh
, "call ext4_handle_dirty_metadata");
4306 * The buffer head should have an attached journal head at this
4307 * point. However, if the data is corrupted and an indirect
4308 * block pointed to itself, it would have been detached when
4309 * the block was cleared. Check for this instead of OOPSing.
4311 if ((EXT4_JOURNAL(inode
) == NULL
) || bh2jh(this_bh
))
4312 ext4_handle_dirty_metadata(handle
, inode
, this_bh
);
4314 EXT4_ERROR_INODE(inode
,
4315 "circular indirect block detected at "
4317 (unsigned long long) this_bh
->b_blocknr
);
4322 * ext4_free_branches - free an array of branches
4323 * @handle: JBD handle for this transaction
4324 * @inode: inode we are dealing with
4325 * @parent_bh: the buffer_head which contains *@first and *@last
4326 * @first: array of block numbers
4327 * @last: pointer immediately past the end of array
4328 * @depth: depth of the branches to free
4330 * We are freeing all blocks refered from these branches (numbers are
4331 * stored as little-endian 32-bit) and updating @inode->i_blocks
4334 static void ext4_free_branches(handle_t
*handle
, struct inode
*inode
,
4335 struct buffer_head
*parent_bh
,
4336 __le32
*first
, __le32
*last
, int depth
)
4341 if (ext4_handle_is_aborted(handle
))
4345 struct buffer_head
*bh
;
4346 int addr_per_block
= EXT4_ADDR_PER_BLOCK(inode
->i_sb
);
4348 while (--p
>= first
) {
4349 nr
= le32_to_cpu(*p
);
4351 continue; /* A hole */
4353 if (!ext4_data_block_valid(EXT4_SB(inode
->i_sb
),
4355 EXT4_ERROR_INODE(inode
,
4356 "invalid indirect mapped "
4357 "block %lu (level %d)",
4358 (unsigned long) nr
, depth
);
4362 /* Go read the buffer for the next level down */
4363 bh
= sb_bread(inode
->i_sb
, nr
);
4366 * A read failure? Report error and clear slot
4370 EXT4_ERROR_INODE_BLOCK(inode
, nr
,
4375 /* This zaps the entire block. Bottom up. */
4376 BUFFER_TRACE(bh
, "free child branches");
4377 ext4_free_branches(handle
, inode
, bh
,
4378 (__le32
*) bh
->b_data
,
4379 (__le32
*) bh
->b_data
+ addr_per_block
,
4384 * Everything below this this pointer has been
4385 * released. Now let this top-of-subtree go.
4387 * We want the freeing of this indirect block to be
4388 * atomic in the journal with the updating of the
4389 * bitmap block which owns it. So make some room in
4392 * We zero the parent pointer *after* freeing its
4393 * pointee in the bitmaps, so if extend_transaction()
4394 * for some reason fails to put the bitmap changes and
4395 * the release into the same transaction, recovery
4396 * will merely complain about releasing a free block,
4397 * rather than leaking blocks.
4399 if (ext4_handle_is_aborted(handle
))
4401 if (try_to_extend_transaction(handle
, inode
)) {
4402 ext4_mark_inode_dirty(handle
, inode
);
4403 ext4_truncate_restart_trans(handle
, inode
,
4404 blocks_for_truncate(inode
));
4408 * The forget flag here is critical because if
4409 * we are journaling (and not doing data
4410 * journaling), we have to make sure a revoke
4411 * record is written to prevent the journal
4412 * replay from overwriting the (former)
4413 * indirect block if it gets reallocated as a
4414 * data block. This must happen in the same
4415 * transaction where the data blocks are
4418 ext4_free_blocks(handle
, inode
, 0, nr
, 1,
4419 EXT4_FREE_BLOCKS_METADATA
|
4420 EXT4_FREE_BLOCKS_FORGET
);
4424 * The block which we have just freed is
4425 * pointed to by an indirect block: journal it
4427 BUFFER_TRACE(parent_bh
, "get_write_access");
4428 if (!ext4_journal_get_write_access(handle
,
4431 BUFFER_TRACE(parent_bh
,
4432 "call ext4_handle_dirty_metadata");
4433 ext4_handle_dirty_metadata(handle
,
4440 /* We have reached the bottom of the tree. */
4441 BUFFER_TRACE(parent_bh
, "free data blocks");
4442 ext4_free_data(handle
, inode
, parent_bh
, first
, last
);
4446 int ext4_can_truncate(struct inode
*inode
)
4448 if (IS_APPEND(inode
) || IS_IMMUTABLE(inode
))
4450 if (S_ISREG(inode
->i_mode
))
4452 if (S_ISDIR(inode
->i_mode
))
4454 if (S_ISLNK(inode
->i_mode
))
4455 return !ext4_inode_is_fast_symlink(inode
);
4462 * We block out ext4_get_block() block instantiations across the entire
4463 * transaction, and VFS/VM ensures that ext4_truncate() cannot run
4464 * simultaneously on behalf of the same inode.
4466 * As we work through the truncate and commmit bits of it to the journal there
4467 * is one core, guiding principle: the file's tree must always be consistent on
4468 * disk. We must be able to restart the truncate after a crash.
4470 * The file's tree may be transiently inconsistent in memory (although it
4471 * probably isn't), but whenever we close off and commit a journal transaction,
4472 * the contents of (the filesystem + the journal) must be consistent and
4473 * restartable. It's pretty simple, really: bottom up, right to left (although
4474 * left-to-right works OK too).
4476 * Note that at recovery time, journal replay occurs *before* the restart of
4477 * truncate against the orphan inode list.
4479 * The committed inode has the new, desired i_size (which is the same as
4480 * i_disksize in this case). After a crash, ext4_orphan_cleanup() will see
4481 * that this inode's truncate did not complete and it will again call
4482 * ext4_truncate() to have another go. So there will be instantiated blocks
4483 * to the right of the truncation point in a crashed ext4 filesystem. But
4484 * that's fine - as long as they are linked from the inode, the post-crash
4485 * ext4_truncate() run will find them and release them.
4487 void ext4_truncate(struct inode
*inode
)
4490 struct ext4_inode_info
*ei
= EXT4_I(inode
);
4491 __le32
*i_data
= ei
->i_data
;
4492 int addr_per_block
= EXT4_ADDR_PER_BLOCK(inode
->i_sb
);
4493 struct address_space
*mapping
= inode
->i_mapping
;
4494 ext4_lblk_t offsets
[4];
4499 ext4_lblk_t last_block
;
4500 unsigned blocksize
= inode
->i_sb
->s_blocksize
;
4502 if (!ext4_can_truncate(inode
))
4505 ext4_clear_inode_flag(inode
, EXT4_INODE_EOFBLOCKS
);
4507 if (inode
->i_size
== 0 && !test_opt(inode
->i_sb
, NO_AUTO_DA_ALLOC
))
4508 ext4_set_inode_state(inode
, EXT4_STATE_DA_ALLOC_CLOSE
);
4510 if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
)) {
4511 ext4_ext_truncate(inode
);
4515 handle
= start_transaction(inode
);
4517 return; /* AKPM: return what? */
4519 last_block
= (inode
->i_size
+ blocksize
-1)
4520 >> EXT4_BLOCK_SIZE_BITS(inode
->i_sb
);
4522 if (inode
->i_size
& (blocksize
- 1))
4523 if (ext4_block_truncate_page(handle
, mapping
, inode
->i_size
))
4526 n
= ext4_block_to_path(inode
, last_block
, offsets
, NULL
);
4528 goto out_stop
; /* error */
4531 * OK. This truncate is going to happen. We add the inode to the
4532 * orphan list, so that if this truncate spans multiple transactions,
4533 * and we crash, we will resume the truncate when the filesystem
4534 * recovers. It also marks the inode dirty, to catch the new size.
4536 * Implication: the file must always be in a sane, consistent
4537 * truncatable state while each transaction commits.
4539 if (ext4_orphan_add(handle
, inode
))
4543 * From here we block out all ext4_get_block() callers who want to
4544 * modify the block allocation tree.
4546 down_write(&ei
->i_data_sem
);
4548 ext4_discard_preallocations(inode
);
4551 * The orphan list entry will now protect us from any crash which
4552 * occurs before the truncate completes, so it is now safe to propagate
4553 * the new, shorter inode size (held for now in i_size) into the
4554 * on-disk inode. We do this via i_disksize, which is the value which
4555 * ext4 *really* writes onto the disk inode.
4557 ei
->i_disksize
= inode
->i_size
;
4559 if (n
== 1) { /* direct blocks */
4560 ext4_free_data(handle
, inode
, NULL
, i_data
+offsets
[0],
4561 i_data
+ EXT4_NDIR_BLOCKS
);
4565 partial
= ext4_find_shared(inode
, n
, offsets
, chain
, &nr
);
4566 /* Kill the top of shared branch (not detached) */
4568 if (partial
== chain
) {
4569 /* Shared branch grows from the inode */
4570 ext4_free_branches(handle
, inode
, NULL
,
4571 &nr
, &nr
+1, (chain
+n
-1) - partial
);
4574 * We mark the inode dirty prior to restart,
4575 * and prior to stop. No need for it here.
4578 /* Shared branch grows from an indirect block */
4579 BUFFER_TRACE(partial
->bh
, "get_write_access");
4580 ext4_free_branches(handle
, inode
, partial
->bh
,
4582 partial
->p
+1, (chain
+n
-1) - partial
);
4585 /* Clear the ends of indirect blocks on the shared branch */
4586 while (partial
> chain
) {
4587 ext4_free_branches(handle
, inode
, partial
->bh
, partial
->p
+ 1,
4588 (__le32
*)partial
->bh
->b_data
+addr_per_block
,
4589 (chain
+n
-1) - partial
);
4590 BUFFER_TRACE(partial
->bh
, "call brelse");
4591 brelse(partial
->bh
);
4595 /* Kill the remaining (whole) subtrees */
4596 switch (offsets
[0]) {
4598 nr
= i_data
[EXT4_IND_BLOCK
];
4600 ext4_free_branches(handle
, inode
, NULL
, &nr
, &nr
+1, 1);
4601 i_data
[EXT4_IND_BLOCK
] = 0;
4603 case EXT4_IND_BLOCK
:
4604 nr
= i_data
[EXT4_DIND_BLOCK
];
4606 ext4_free_branches(handle
, inode
, NULL
, &nr
, &nr
+1, 2);
4607 i_data
[EXT4_DIND_BLOCK
] = 0;
4609 case EXT4_DIND_BLOCK
:
4610 nr
= i_data
[EXT4_TIND_BLOCK
];
4612 ext4_free_branches(handle
, inode
, NULL
, &nr
, &nr
+1, 3);
4613 i_data
[EXT4_TIND_BLOCK
] = 0;
4615 case EXT4_TIND_BLOCK
:
4619 up_write(&ei
->i_data_sem
);
4620 inode
->i_mtime
= inode
->i_ctime
= ext4_current_time(inode
);
4621 ext4_mark_inode_dirty(handle
, inode
);
4624 * In a multi-transaction truncate, we only make the final transaction
4628 ext4_handle_sync(handle
);
4631 * If this was a simple ftruncate(), and the file will remain alive
4632 * then we need to clear up the orphan record which we created above.
4633 * However, if this was a real unlink then we were called by
4634 * ext4_delete_inode(), and we allow that function to clean up the
4635 * orphan info for us.
4638 ext4_orphan_del(handle
, inode
);
4640 ext4_journal_stop(handle
);
4644 * ext4_get_inode_loc returns with an extra refcount against the inode's
4645 * underlying buffer_head on success. If 'in_mem' is true, we have all
4646 * data in memory that is needed to recreate the on-disk version of this
4649 static int __ext4_get_inode_loc(struct inode
*inode
,
4650 struct ext4_iloc
*iloc
, int in_mem
)
4652 struct ext4_group_desc
*gdp
;
4653 struct buffer_head
*bh
;
4654 struct super_block
*sb
= inode
->i_sb
;
4656 int inodes_per_block
, inode_offset
;
4659 if (!ext4_valid_inum(sb
, inode
->i_ino
))
4662 iloc
->block_group
= (inode
->i_ino
- 1) / EXT4_INODES_PER_GROUP(sb
);
4663 gdp
= ext4_get_group_desc(sb
, iloc
->block_group
, NULL
);
4668 * Figure out the offset within the block group inode table
4670 inodes_per_block
= (EXT4_BLOCK_SIZE(sb
) / EXT4_INODE_SIZE(sb
));
4671 inode_offset
= ((inode
->i_ino
- 1) %
4672 EXT4_INODES_PER_GROUP(sb
));
4673 block
= ext4_inode_table(sb
, gdp
) + (inode_offset
/ inodes_per_block
);
4674 iloc
->offset
= (inode_offset
% inodes_per_block
) * EXT4_INODE_SIZE(sb
);
4676 bh
= sb_getblk(sb
, block
);
4678 EXT4_ERROR_INODE_BLOCK(inode
, block
,
4679 "unable to read itable block");
4682 if (!buffer_uptodate(bh
)) {
4686 * If the buffer has the write error flag, we have failed
4687 * to write out another inode in the same block. In this
4688 * case, we don't have to read the block because we may
4689 * read the old inode data successfully.
4691 if (buffer_write_io_error(bh
) && !buffer_uptodate(bh
))
4692 set_buffer_uptodate(bh
);
4694 if (buffer_uptodate(bh
)) {
4695 /* someone brought it uptodate while we waited */
4701 * If we have all information of the inode in memory and this
4702 * is the only valid inode in the block, we need not read the
4706 struct buffer_head
*bitmap_bh
;
4709 start
= inode_offset
& ~(inodes_per_block
- 1);
4711 /* Is the inode bitmap in cache? */
4712 bitmap_bh
= sb_getblk(sb
, ext4_inode_bitmap(sb
, gdp
));
4717 * If the inode bitmap isn't in cache then the
4718 * optimisation may end up performing two reads instead
4719 * of one, so skip it.
4721 if (!buffer_uptodate(bitmap_bh
)) {
4725 for (i
= start
; i
< start
+ inodes_per_block
; i
++) {
4726 if (i
== inode_offset
)
4728 if (ext4_test_bit(i
, bitmap_bh
->b_data
))
4732 if (i
== start
+ inodes_per_block
) {
4733 /* all other inodes are free, so skip I/O */
4734 memset(bh
->b_data
, 0, bh
->b_size
);
4735 set_buffer_uptodate(bh
);
4743 * If we need to do any I/O, try to pre-readahead extra
4744 * blocks from the inode table.
4746 if (EXT4_SB(sb
)->s_inode_readahead_blks
) {
4747 ext4_fsblk_t b
, end
, table
;
4750 table
= ext4_inode_table(sb
, gdp
);
4751 /* s_inode_readahead_blks is always a power of 2 */
4752 b
= block
& ~(EXT4_SB(sb
)->s_inode_readahead_blks
-1);
4755 end
= b
+ EXT4_SB(sb
)->s_inode_readahead_blks
;
4756 num
= EXT4_INODES_PER_GROUP(sb
);
4757 if (EXT4_HAS_RO_COMPAT_FEATURE(sb
,
4758 EXT4_FEATURE_RO_COMPAT_GDT_CSUM
))
4759 num
-= ext4_itable_unused_count(sb
, gdp
);
4760 table
+= num
/ inodes_per_block
;
4764 sb_breadahead(sb
, b
++);
4768 * There are other valid inodes in the buffer, this inode
4769 * has in-inode xattrs, or we don't have this inode in memory.
4770 * Read the block from disk.
4773 bh
->b_end_io
= end_buffer_read_sync
;
4774 submit_bh(READ_META
, bh
);
4776 if (!buffer_uptodate(bh
)) {
4777 EXT4_ERROR_INODE_BLOCK(inode
, block
,
4778 "unable to read itable block");
4788 int ext4_get_inode_loc(struct inode
*inode
, struct ext4_iloc
*iloc
)
4790 /* We have all inode data except xattrs in memory here. */
4791 return __ext4_get_inode_loc(inode
, iloc
,
4792 !ext4_test_inode_state(inode
, EXT4_STATE_XATTR
));
4795 void ext4_set_inode_flags(struct inode
*inode
)
4797 unsigned int flags
= EXT4_I(inode
)->i_flags
;
4799 inode
->i_flags
&= ~(S_SYNC
|S_APPEND
|S_IMMUTABLE
|S_NOATIME
|S_DIRSYNC
);
4800 if (flags
& EXT4_SYNC_FL
)
4801 inode
->i_flags
|= S_SYNC
;
4802 if (flags
& EXT4_APPEND_FL
)
4803 inode
->i_flags
|= S_APPEND
;
4804 if (flags
& EXT4_IMMUTABLE_FL
)
4805 inode
->i_flags
|= S_IMMUTABLE
;
4806 if (flags
& EXT4_NOATIME_FL
)
4807 inode
->i_flags
|= S_NOATIME
;
4808 if (flags
& EXT4_DIRSYNC_FL
)
4809 inode
->i_flags
|= S_DIRSYNC
;
4812 /* Propagate flags from i_flags to EXT4_I(inode)->i_flags */
4813 void ext4_get_inode_flags(struct ext4_inode_info
*ei
)
4815 unsigned int vfs_fl
;
4816 unsigned long old_fl
, new_fl
;
4819 vfs_fl
= ei
->vfs_inode
.i_flags
;
4820 old_fl
= ei
->i_flags
;
4821 new_fl
= old_fl
& ~(EXT4_SYNC_FL
|EXT4_APPEND_FL
|
4822 EXT4_IMMUTABLE_FL
|EXT4_NOATIME_FL
|
4824 if (vfs_fl
& S_SYNC
)
4825 new_fl
|= EXT4_SYNC_FL
;
4826 if (vfs_fl
& S_APPEND
)
4827 new_fl
|= EXT4_APPEND_FL
;
4828 if (vfs_fl
& S_IMMUTABLE
)
4829 new_fl
|= EXT4_IMMUTABLE_FL
;
4830 if (vfs_fl
& S_NOATIME
)
4831 new_fl
|= EXT4_NOATIME_FL
;
4832 if (vfs_fl
& S_DIRSYNC
)
4833 new_fl
|= EXT4_DIRSYNC_FL
;
4834 } while (cmpxchg(&ei
->i_flags
, old_fl
, new_fl
) != old_fl
);
4837 static blkcnt_t
ext4_inode_blocks(struct ext4_inode
*raw_inode
,
4838 struct ext4_inode_info
*ei
)
4841 struct inode
*inode
= &(ei
->vfs_inode
);
4842 struct super_block
*sb
= inode
->i_sb
;
4844 if (EXT4_HAS_RO_COMPAT_FEATURE(sb
,
4845 EXT4_FEATURE_RO_COMPAT_HUGE_FILE
)) {
4846 /* we are using combined 48 bit field */
4847 i_blocks
= ((u64
)le16_to_cpu(raw_inode
->i_blocks_high
)) << 32 |
4848 le32_to_cpu(raw_inode
->i_blocks_lo
);
4849 if (ext4_test_inode_flag(inode
, EXT4_INODE_HUGE_FILE
)) {
4850 /* i_blocks represent file system block size */
4851 return i_blocks
<< (inode
->i_blkbits
- 9);
4856 return le32_to_cpu(raw_inode
->i_blocks_lo
);
4860 struct inode
*ext4_iget(struct super_block
*sb
, unsigned long ino
)
4862 struct ext4_iloc iloc
;
4863 struct ext4_inode
*raw_inode
;
4864 struct ext4_inode_info
*ei
;
4865 struct inode
*inode
;
4866 journal_t
*journal
= EXT4_SB(sb
)->s_journal
;
4870 inode
= iget_locked(sb
, ino
);
4872 return ERR_PTR(-ENOMEM
);
4873 if (!(inode
->i_state
& I_NEW
))
4879 ret
= __ext4_get_inode_loc(inode
, &iloc
, 0);
4882 raw_inode
= ext4_raw_inode(&iloc
);
4883 inode
->i_mode
= le16_to_cpu(raw_inode
->i_mode
);
4884 inode
->i_uid
= (uid_t
)le16_to_cpu(raw_inode
->i_uid_low
);
4885 inode
->i_gid
= (gid_t
)le16_to_cpu(raw_inode
->i_gid_low
);
4886 if (!(test_opt(inode
->i_sb
, NO_UID32
))) {
4887 inode
->i_uid
|= le16_to_cpu(raw_inode
->i_uid_high
) << 16;
4888 inode
->i_gid
|= le16_to_cpu(raw_inode
->i_gid_high
) << 16;
4890 inode
->i_nlink
= le16_to_cpu(raw_inode
->i_links_count
);
4892 ext4_clear_state_flags(ei
); /* Only relevant on 32-bit archs */
4893 ei
->i_dir_start_lookup
= 0;
4894 ei
->i_dtime
= le32_to_cpu(raw_inode
->i_dtime
);
4895 /* We now have enough fields to check if the inode was active or not.
4896 * This is needed because nfsd might try to access dead inodes
4897 * the test is that same one that e2fsck uses
4898 * NeilBrown 1999oct15
4900 if (inode
->i_nlink
== 0) {
4901 if (inode
->i_mode
== 0 ||
4902 !(EXT4_SB(inode
->i_sb
)->s_mount_state
& EXT4_ORPHAN_FS
)) {
4903 /* this inode is deleted */
4907 /* The only unlinked inodes we let through here have
4908 * valid i_mode and are being read by the orphan
4909 * recovery code: that's fine, we're about to complete
4910 * the process of deleting those. */
4912 ei
->i_flags
= le32_to_cpu(raw_inode
->i_flags
);
4913 inode
->i_blocks
= ext4_inode_blocks(raw_inode
, ei
);
4914 ei
->i_file_acl
= le32_to_cpu(raw_inode
->i_file_acl_lo
);
4915 if (EXT4_HAS_INCOMPAT_FEATURE(sb
, EXT4_FEATURE_INCOMPAT_64BIT
))
4917 ((__u64
)le16_to_cpu(raw_inode
->i_file_acl_high
)) << 32;
4918 inode
->i_size
= ext4_isize(raw_inode
);
4919 ei
->i_disksize
= inode
->i_size
;
4921 ei
->i_reserved_quota
= 0;
4923 inode
->i_generation
= le32_to_cpu(raw_inode
->i_generation
);
4924 ei
->i_block_group
= iloc
.block_group
;
4925 ei
->i_last_alloc_group
= ~0;
4927 * NOTE! The in-memory inode i_data array is in little-endian order
4928 * even on big-endian machines: we do NOT byteswap the block numbers!
4930 for (block
= 0; block
< EXT4_N_BLOCKS
; block
++)
4931 ei
->i_data
[block
] = raw_inode
->i_block
[block
];
4932 INIT_LIST_HEAD(&ei
->i_orphan
);
4935 * Set transaction id's of transactions that have to be committed
4936 * to finish f[data]sync. We set them to currently running transaction
4937 * as we cannot be sure that the inode or some of its metadata isn't
4938 * part of the transaction - the inode could have been reclaimed and
4939 * now it is reread from disk.
4942 transaction_t
*transaction
;
4945 read_lock(&journal
->j_state_lock
);
4946 if (journal
->j_running_transaction
)
4947 transaction
= journal
->j_running_transaction
;
4949 transaction
= journal
->j_committing_transaction
;
4951 tid
= transaction
->t_tid
;
4953 tid
= journal
->j_commit_sequence
;
4954 read_unlock(&journal
->j_state_lock
);
4955 ei
->i_sync_tid
= tid
;
4956 ei
->i_datasync_tid
= tid
;
4959 if (EXT4_INODE_SIZE(inode
->i_sb
) > EXT4_GOOD_OLD_INODE_SIZE
) {
4960 ei
->i_extra_isize
= le16_to_cpu(raw_inode
->i_extra_isize
);
4961 if (EXT4_GOOD_OLD_INODE_SIZE
+ ei
->i_extra_isize
>
4962 EXT4_INODE_SIZE(inode
->i_sb
)) {
4966 if (ei
->i_extra_isize
== 0) {
4967 /* The extra space is currently unused. Use it. */
4968 ei
->i_extra_isize
= sizeof(struct ext4_inode
) -
4969 EXT4_GOOD_OLD_INODE_SIZE
;
4971 __le32
*magic
= (void *)raw_inode
+
4972 EXT4_GOOD_OLD_INODE_SIZE
+
4974 if (*magic
== cpu_to_le32(EXT4_XATTR_MAGIC
))
4975 ext4_set_inode_state(inode
, EXT4_STATE_XATTR
);
4978 ei
->i_extra_isize
= 0;
4980 EXT4_INODE_GET_XTIME(i_ctime
, inode
, raw_inode
);
4981 EXT4_INODE_GET_XTIME(i_mtime
, inode
, raw_inode
);
4982 EXT4_INODE_GET_XTIME(i_atime
, inode
, raw_inode
);
4983 EXT4_EINODE_GET_XTIME(i_crtime
, ei
, raw_inode
);
4985 inode
->i_version
= le32_to_cpu(raw_inode
->i_disk_version
);
4986 if (EXT4_INODE_SIZE(inode
->i_sb
) > EXT4_GOOD_OLD_INODE_SIZE
) {
4987 if (EXT4_FITS_IN_INODE(raw_inode
, ei
, i_version_hi
))
4989 (__u64
)(le32_to_cpu(raw_inode
->i_version_hi
)) << 32;
4993 if (ei
->i_file_acl
&&
4994 !ext4_data_block_valid(EXT4_SB(sb
), ei
->i_file_acl
, 1)) {
4995 EXT4_ERROR_INODE(inode
, "bad extended attribute block %llu",
4999 } else if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
)) {
5000 if (S_ISREG(inode
->i_mode
) || S_ISDIR(inode
->i_mode
) ||
5001 (S_ISLNK(inode
->i_mode
) &&
5002 !ext4_inode_is_fast_symlink(inode
)))
5003 /* Validate extent which is part of inode */
5004 ret
= ext4_ext_check_inode(inode
);
5005 } else if (S_ISREG(inode
->i_mode
) || S_ISDIR(inode
->i_mode
) ||
5006 (S_ISLNK(inode
->i_mode
) &&
5007 !ext4_inode_is_fast_symlink(inode
))) {
5008 /* Validate block references which are part of inode */
5009 ret
= ext4_check_inode_blockref(inode
);
5014 if (S_ISREG(inode
->i_mode
)) {
5015 inode
->i_op
= &ext4_file_inode_operations
;
5016 inode
->i_fop
= &ext4_file_operations
;
5017 ext4_set_aops(inode
);
5018 } else if (S_ISDIR(inode
->i_mode
)) {
5019 inode
->i_op
= &ext4_dir_inode_operations
;
5020 inode
->i_fop
= &ext4_dir_operations
;
5021 } else if (S_ISLNK(inode
->i_mode
)) {
5022 if (ext4_inode_is_fast_symlink(inode
)) {
5023 inode
->i_op
= &ext4_fast_symlink_inode_operations
;
5024 nd_terminate_link(ei
->i_data
, inode
->i_size
,
5025 sizeof(ei
->i_data
) - 1);
5027 inode
->i_op
= &ext4_symlink_inode_operations
;
5028 ext4_set_aops(inode
);
5030 } else if (S_ISCHR(inode
->i_mode
) || S_ISBLK(inode
->i_mode
) ||
5031 S_ISFIFO(inode
->i_mode
) || S_ISSOCK(inode
->i_mode
)) {
5032 inode
->i_op
= &ext4_special_inode_operations
;
5033 if (raw_inode
->i_block
[0])
5034 init_special_inode(inode
, inode
->i_mode
,
5035 old_decode_dev(le32_to_cpu(raw_inode
->i_block
[0])));
5037 init_special_inode(inode
, inode
->i_mode
,
5038 new_decode_dev(le32_to_cpu(raw_inode
->i_block
[1])));
5041 EXT4_ERROR_INODE(inode
, "bogus i_mode (%o)", inode
->i_mode
);
5045 ext4_set_inode_flags(inode
);
5046 unlock_new_inode(inode
);
5052 return ERR_PTR(ret
);
5055 static int ext4_inode_blocks_set(handle_t
*handle
,
5056 struct ext4_inode
*raw_inode
,
5057 struct ext4_inode_info
*ei
)
5059 struct inode
*inode
= &(ei
->vfs_inode
);
5060 u64 i_blocks
= inode
->i_blocks
;
5061 struct super_block
*sb
= inode
->i_sb
;
5063 if (i_blocks
<= ~0U) {
5065 * i_blocks can be represnted in a 32 bit variable
5066 * as multiple of 512 bytes
5068 raw_inode
->i_blocks_lo
= cpu_to_le32(i_blocks
);
5069 raw_inode
->i_blocks_high
= 0;
5070 ext4_clear_inode_flag(inode
, EXT4_INODE_HUGE_FILE
);
5073 if (!EXT4_HAS_RO_COMPAT_FEATURE(sb
, EXT4_FEATURE_RO_COMPAT_HUGE_FILE
))
5076 if (i_blocks
<= 0xffffffffffffULL
) {
5078 * i_blocks can be represented in a 48 bit variable
5079 * as multiple of 512 bytes
5081 raw_inode
->i_blocks_lo
= cpu_to_le32(i_blocks
);
5082 raw_inode
->i_blocks_high
= cpu_to_le16(i_blocks
>> 32);
5083 ext4_clear_inode_flag(inode
, EXT4_INODE_HUGE_FILE
);
5085 ext4_set_inode_flag(inode
, EXT4_INODE_HUGE_FILE
);
5086 /* i_block is stored in file system block size */
5087 i_blocks
= i_blocks
>> (inode
->i_blkbits
- 9);
5088 raw_inode
->i_blocks_lo
= cpu_to_le32(i_blocks
);
5089 raw_inode
->i_blocks_high
= cpu_to_le16(i_blocks
>> 32);
5095 * Post the struct inode info into an on-disk inode location in the
5096 * buffer-cache. This gobbles the caller's reference to the
5097 * buffer_head in the inode location struct.
5099 * The caller must have write access to iloc->bh.
5101 static int ext4_do_update_inode(handle_t
*handle
,
5102 struct inode
*inode
,
5103 struct ext4_iloc
*iloc
)
5105 struct ext4_inode
*raw_inode
= ext4_raw_inode(iloc
);
5106 struct ext4_inode_info
*ei
= EXT4_I(inode
);
5107 struct buffer_head
*bh
= iloc
->bh
;
5108 int err
= 0, rc
, block
;
5110 /* For fields not not tracking in the in-memory inode,
5111 * initialise them to zero for new inodes. */
5112 if (ext4_test_inode_state(inode
, EXT4_STATE_NEW
))
5113 memset(raw_inode
, 0, EXT4_SB(inode
->i_sb
)->s_inode_size
);
5115 ext4_get_inode_flags(ei
);
5116 raw_inode
->i_mode
= cpu_to_le16(inode
->i_mode
);
5117 if (!(test_opt(inode
->i_sb
, NO_UID32
))) {
5118 raw_inode
->i_uid_low
= cpu_to_le16(low_16_bits(inode
->i_uid
));
5119 raw_inode
->i_gid_low
= cpu_to_le16(low_16_bits(inode
->i_gid
));
5121 * Fix up interoperability with old kernels. Otherwise, old inodes get
5122 * re-used with the upper 16 bits of the uid/gid intact
5125 raw_inode
->i_uid_high
=
5126 cpu_to_le16(high_16_bits(inode
->i_uid
));
5127 raw_inode
->i_gid_high
=
5128 cpu_to_le16(high_16_bits(inode
->i_gid
));
5130 raw_inode
->i_uid_high
= 0;
5131 raw_inode
->i_gid_high
= 0;
5134 raw_inode
->i_uid_low
=
5135 cpu_to_le16(fs_high2lowuid(inode
->i_uid
));
5136 raw_inode
->i_gid_low
=
5137 cpu_to_le16(fs_high2lowgid(inode
->i_gid
));
5138 raw_inode
->i_uid_high
= 0;
5139 raw_inode
->i_gid_high
= 0;
5141 raw_inode
->i_links_count
= cpu_to_le16(inode
->i_nlink
);
5143 EXT4_INODE_SET_XTIME(i_ctime
, inode
, raw_inode
);
5144 EXT4_INODE_SET_XTIME(i_mtime
, inode
, raw_inode
);
5145 EXT4_INODE_SET_XTIME(i_atime
, inode
, raw_inode
);
5146 EXT4_EINODE_SET_XTIME(i_crtime
, ei
, raw_inode
);
5148 if (ext4_inode_blocks_set(handle
, raw_inode
, ei
))
5150 raw_inode
->i_dtime
= cpu_to_le32(ei
->i_dtime
);
5151 raw_inode
->i_flags
= cpu_to_le32(ei
->i_flags
& 0xFFFFFFFF);
5152 if (EXT4_SB(inode
->i_sb
)->s_es
->s_creator_os
!=
5153 cpu_to_le32(EXT4_OS_HURD
))
5154 raw_inode
->i_file_acl_high
=
5155 cpu_to_le16(ei
->i_file_acl
>> 32);
5156 raw_inode
->i_file_acl_lo
= cpu_to_le32(ei
->i_file_acl
);
5157 ext4_isize_set(raw_inode
, ei
->i_disksize
);
5158 if (ei
->i_disksize
> 0x7fffffffULL
) {
5159 struct super_block
*sb
= inode
->i_sb
;
5160 if (!EXT4_HAS_RO_COMPAT_FEATURE(sb
,
5161 EXT4_FEATURE_RO_COMPAT_LARGE_FILE
) ||
5162 EXT4_SB(sb
)->s_es
->s_rev_level
==
5163 cpu_to_le32(EXT4_GOOD_OLD_REV
)) {
5164 /* If this is the first large file
5165 * created, add a flag to the superblock.
5167 err
= ext4_journal_get_write_access(handle
,
5168 EXT4_SB(sb
)->s_sbh
);
5171 ext4_update_dynamic_rev(sb
);
5172 EXT4_SET_RO_COMPAT_FEATURE(sb
,
5173 EXT4_FEATURE_RO_COMPAT_LARGE_FILE
);
5175 ext4_handle_sync(handle
);
5176 err
= ext4_handle_dirty_metadata(handle
, NULL
,
5177 EXT4_SB(sb
)->s_sbh
);
5180 raw_inode
->i_generation
= cpu_to_le32(inode
->i_generation
);
5181 if (S_ISCHR(inode
->i_mode
) || S_ISBLK(inode
->i_mode
)) {
5182 if (old_valid_dev(inode
->i_rdev
)) {
5183 raw_inode
->i_block
[0] =
5184 cpu_to_le32(old_encode_dev(inode
->i_rdev
));
5185 raw_inode
->i_block
[1] = 0;
5187 raw_inode
->i_block
[0] = 0;
5188 raw_inode
->i_block
[1] =
5189 cpu_to_le32(new_encode_dev(inode
->i_rdev
));
5190 raw_inode
->i_block
[2] = 0;
5193 for (block
= 0; block
< EXT4_N_BLOCKS
; block
++)
5194 raw_inode
->i_block
[block
] = ei
->i_data
[block
];
5196 raw_inode
->i_disk_version
= cpu_to_le32(inode
->i_version
);
5197 if (ei
->i_extra_isize
) {
5198 if (EXT4_FITS_IN_INODE(raw_inode
, ei
, i_version_hi
))
5199 raw_inode
->i_version_hi
=
5200 cpu_to_le32(inode
->i_version
>> 32);
5201 raw_inode
->i_extra_isize
= cpu_to_le16(ei
->i_extra_isize
);
5204 BUFFER_TRACE(bh
, "call ext4_handle_dirty_metadata");
5205 rc
= ext4_handle_dirty_metadata(handle
, NULL
, bh
);
5208 ext4_clear_inode_state(inode
, EXT4_STATE_NEW
);
5210 ext4_update_inode_fsync_trans(handle
, inode
, 0);
5213 ext4_std_error(inode
->i_sb
, err
);
5218 * ext4_write_inode()
5220 * We are called from a few places:
5222 * - Within generic_file_write() for O_SYNC files.
5223 * Here, there will be no transaction running. We wait for any running
5224 * trasnaction to commit.
5226 * - Within sys_sync(), kupdate and such.
5227 * We wait on commit, if tol to.
5229 * - Within prune_icache() (PF_MEMALLOC == true)
5230 * Here we simply return. We can't afford to block kswapd on the
5233 * In all cases it is actually safe for us to return without doing anything,
5234 * because the inode has been copied into a raw inode buffer in
5235 * ext4_mark_inode_dirty(). This is a correctness thing for O_SYNC and for
5238 * Note that we are absolutely dependent upon all inode dirtiers doing the
5239 * right thing: they *must* call mark_inode_dirty() after dirtying info in
5240 * which we are interested.
5242 * It would be a bug for them to not do this. The code:
5244 * mark_inode_dirty(inode)
5246 * inode->i_size = expr;
5248 * is in error because a kswapd-driven write_inode() could occur while
5249 * `stuff()' is running, and the new i_size will be lost. Plus the inode
5250 * will no longer be on the superblock's dirty inode list.
5252 int ext4_write_inode(struct inode
*inode
, struct writeback_control
*wbc
)
5256 if (current
->flags
& PF_MEMALLOC
)
5259 if (EXT4_SB(inode
->i_sb
)->s_journal
) {
5260 if (ext4_journal_current_handle()) {
5261 jbd_debug(1, "called recursively, non-PF_MEMALLOC!\n");
5266 if (wbc
->sync_mode
!= WB_SYNC_ALL
)
5269 err
= ext4_force_commit(inode
->i_sb
);
5271 struct ext4_iloc iloc
;
5273 err
= __ext4_get_inode_loc(inode
, &iloc
, 0);
5276 if (wbc
->sync_mode
== WB_SYNC_ALL
)
5277 sync_dirty_buffer(iloc
.bh
);
5278 if (buffer_req(iloc
.bh
) && !buffer_uptodate(iloc
.bh
)) {
5279 EXT4_ERROR_INODE_BLOCK(inode
, iloc
.bh
->b_blocknr
,
5280 "IO error syncing inode");
5291 * Called from notify_change.
5293 * We want to trap VFS attempts to truncate the file as soon as
5294 * possible. In particular, we want to make sure that when the VFS
5295 * shrinks i_size, we put the inode on the orphan list and modify
5296 * i_disksize immediately, so that during the subsequent flushing of
5297 * dirty pages and freeing of disk blocks, we can guarantee that any
5298 * commit will leave the blocks being flushed in an unused state on
5299 * disk. (On recovery, the inode will get truncated and the blocks will
5300 * be freed, so we have a strong guarantee that no future commit will
5301 * leave these blocks visible to the user.)
5303 * Another thing we have to assure is that if we are in ordered mode
5304 * and inode is still attached to the committing transaction, we must
5305 * we start writeout of all the dirty pages which are being truncated.
5306 * This way we are sure that all the data written in the previous
5307 * transaction are already on disk (truncate waits for pages under
5310 * Called with inode->i_mutex down.
5312 int ext4_setattr(struct dentry
*dentry
, struct iattr
*attr
)
5314 struct inode
*inode
= dentry
->d_inode
;
5317 const unsigned int ia_valid
= attr
->ia_valid
;
5319 error
= inode_change_ok(inode
, attr
);
5323 if (is_quota_modification(inode
, attr
))
5324 dquot_initialize(inode
);
5325 if ((ia_valid
& ATTR_UID
&& attr
->ia_uid
!= inode
->i_uid
) ||
5326 (ia_valid
& ATTR_GID
&& attr
->ia_gid
!= inode
->i_gid
)) {
5329 /* (user+group)*(old+new) structure, inode write (sb,
5330 * inode block, ? - but truncate inode update has it) */
5331 handle
= ext4_journal_start(inode
, (EXT4_MAXQUOTAS_INIT_BLOCKS(inode
->i_sb
)+
5332 EXT4_MAXQUOTAS_DEL_BLOCKS(inode
->i_sb
))+3);
5333 if (IS_ERR(handle
)) {
5334 error
= PTR_ERR(handle
);
5337 error
= dquot_transfer(inode
, attr
);
5339 ext4_journal_stop(handle
);
5342 /* Update corresponding info in inode so that everything is in
5343 * one transaction */
5344 if (attr
->ia_valid
& ATTR_UID
)
5345 inode
->i_uid
= attr
->ia_uid
;
5346 if (attr
->ia_valid
& ATTR_GID
)
5347 inode
->i_gid
= attr
->ia_gid
;
5348 error
= ext4_mark_inode_dirty(handle
, inode
);
5349 ext4_journal_stop(handle
);
5352 if (attr
->ia_valid
& ATTR_SIZE
) {
5353 if (!(ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
))) {
5354 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
5356 if (attr
->ia_size
> sbi
->s_bitmap_maxbytes
)
5361 if (S_ISREG(inode
->i_mode
) &&
5362 attr
->ia_valid
& ATTR_SIZE
&&
5363 (attr
->ia_size
< inode
->i_size
||
5364 (ext4_test_inode_flag(inode
, EXT4_INODE_EOFBLOCKS
)))) {
5367 handle
= ext4_journal_start(inode
, 3);
5368 if (IS_ERR(handle
)) {
5369 error
= PTR_ERR(handle
);
5372 if (ext4_handle_valid(handle
)) {
5373 error
= ext4_orphan_add(handle
, inode
);
5376 EXT4_I(inode
)->i_disksize
= attr
->ia_size
;
5377 rc
= ext4_mark_inode_dirty(handle
, inode
);
5380 ext4_journal_stop(handle
);
5382 if (ext4_should_order_data(inode
)) {
5383 error
= ext4_begin_ordered_truncate(inode
,
5386 /* Do as much error cleanup as possible */
5387 handle
= ext4_journal_start(inode
, 3);
5388 if (IS_ERR(handle
)) {
5389 ext4_orphan_del(NULL
, inode
);
5392 ext4_orphan_del(handle
, inode
);
5394 ext4_journal_stop(handle
);
5398 /* ext4_truncate will clear the flag */
5399 if ((ext4_test_inode_flag(inode
, EXT4_INODE_EOFBLOCKS
)))
5400 ext4_truncate(inode
);
5403 if ((attr
->ia_valid
& ATTR_SIZE
) &&
5404 attr
->ia_size
!= i_size_read(inode
))
5405 rc
= vmtruncate(inode
, attr
->ia_size
);
5408 setattr_copy(inode
, attr
);
5409 mark_inode_dirty(inode
);
5413 * If the call to ext4_truncate failed to get a transaction handle at
5414 * all, we need to clean up the in-core orphan list manually.
5416 if (orphan
&& inode
->i_nlink
)
5417 ext4_orphan_del(NULL
, inode
);
5419 if (!rc
&& (ia_valid
& ATTR_MODE
))
5420 rc
= ext4_acl_chmod(inode
);
5423 ext4_std_error(inode
->i_sb
, error
);
5429 int ext4_getattr(struct vfsmount
*mnt
, struct dentry
*dentry
,
5432 struct inode
*inode
;
5433 unsigned long delalloc_blocks
;
5435 inode
= dentry
->d_inode
;
5436 generic_fillattr(inode
, stat
);
5439 * We can't update i_blocks if the block allocation is delayed
5440 * otherwise in the case of system crash before the real block
5441 * allocation is done, we will have i_blocks inconsistent with
5442 * on-disk file blocks.
5443 * We always keep i_blocks updated together with real
5444 * allocation. But to not confuse with user, stat
5445 * will return the blocks that include the delayed allocation
5446 * blocks for this file.
5448 delalloc_blocks
= EXT4_I(inode
)->i_reserved_data_blocks
;
5450 stat
->blocks
+= (delalloc_blocks
<< inode
->i_sb
->s_blocksize_bits
)>>9;
5454 static int ext4_indirect_trans_blocks(struct inode
*inode
, int nrblocks
,
5459 /* if nrblocks are contiguous */
5462 * With N contiguous data blocks, it need at most
5463 * N/EXT4_ADDR_PER_BLOCK(inode->i_sb) indirect blocks
5464 * 2 dindirect blocks
5467 indirects
= nrblocks
/ EXT4_ADDR_PER_BLOCK(inode
->i_sb
);
5468 return indirects
+ 3;
5471 * if nrblocks are not contiguous, worse case, each block touch
5472 * a indirect block, and each indirect block touch a double indirect
5473 * block, plus a triple indirect block
5475 indirects
= nrblocks
* 2 + 1;
5479 static int ext4_index_trans_blocks(struct inode
*inode
, int nrblocks
, int chunk
)
5481 if (!(ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
)))
5482 return ext4_indirect_trans_blocks(inode
, nrblocks
, chunk
);
5483 return ext4_ext_index_trans_blocks(inode
, nrblocks
, chunk
);
5487 * Account for index blocks, block groups bitmaps and block group
5488 * descriptor blocks if modify datablocks and index blocks
5489 * worse case, the indexs blocks spread over different block groups
5491 * If datablocks are discontiguous, they are possible to spread over
5492 * different block groups too. If they are contiuguous, with flexbg,
5493 * they could still across block group boundary.
5495 * Also account for superblock, inode, quota and xattr blocks
5497 static int ext4_meta_trans_blocks(struct inode
*inode
, int nrblocks
, int chunk
)
5499 ext4_group_t groups
, ngroups
= ext4_get_groups_count(inode
->i_sb
);
5505 * How many index blocks need to touch to modify nrblocks?
5506 * The "Chunk" flag indicating whether the nrblocks is
5507 * physically contiguous on disk
5509 * For Direct IO and fallocate, they calls get_block to allocate
5510 * one single extent at a time, so they could set the "Chunk" flag
5512 idxblocks
= ext4_index_trans_blocks(inode
, nrblocks
, chunk
);
5517 * Now let's see how many group bitmaps and group descriptors need
5527 if (groups
> ngroups
)
5529 if (groups
> EXT4_SB(inode
->i_sb
)->s_gdb_count
)
5530 gdpblocks
= EXT4_SB(inode
->i_sb
)->s_gdb_count
;
5532 /* bitmaps and block group descriptor blocks */
5533 ret
+= groups
+ gdpblocks
;
5535 /* Blocks for super block, inode, quota and xattr blocks */
5536 ret
+= EXT4_META_TRANS_BLOCKS(inode
->i_sb
);
5542 * Calulate the total number of credits to reserve to fit
5543 * the modification of a single pages into a single transaction,
5544 * which may include multiple chunks of block allocations.
5546 * This could be called via ext4_write_begin()
5548 * We need to consider the worse case, when
5549 * one new block per extent.
5551 int ext4_writepage_trans_blocks(struct inode
*inode
)
5553 int bpp
= ext4_journal_blocks_per_page(inode
);
5556 ret
= ext4_meta_trans_blocks(inode
, bpp
, 0);
5558 /* Account for data blocks for journalled mode */
5559 if (ext4_should_journal_data(inode
))
5565 * Calculate the journal credits for a chunk of data modification.
5567 * This is called from DIO, fallocate or whoever calling
5568 * ext4_map_blocks() to map/allocate a chunk of contiguous disk blocks.
5570 * journal buffers for data blocks are not included here, as DIO
5571 * and fallocate do no need to journal data buffers.
5573 int ext4_chunk_trans_blocks(struct inode
*inode
, int nrblocks
)
5575 return ext4_meta_trans_blocks(inode
, nrblocks
, 1);
5579 * The caller must have previously called ext4_reserve_inode_write().
5580 * Give this, we know that the caller already has write access to iloc->bh.
5582 int ext4_mark_iloc_dirty(handle_t
*handle
,
5583 struct inode
*inode
, struct ext4_iloc
*iloc
)
5587 if (test_opt(inode
->i_sb
, I_VERSION
))
5588 inode_inc_iversion(inode
);
5590 /* the do_update_inode consumes one bh->b_count */
5593 /* ext4_do_update_inode() does jbd2_journal_dirty_metadata */
5594 err
= ext4_do_update_inode(handle
, inode
, iloc
);
5600 * On success, We end up with an outstanding reference count against
5601 * iloc->bh. This _must_ be cleaned up later.
5605 ext4_reserve_inode_write(handle_t
*handle
, struct inode
*inode
,
5606 struct ext4_iloc
*iloc
)
5610 err
= ext4_get_inode_loc(inode
, iloc
);
5612 BUFFER_TRACE(iloc
->bh
, "get_write_access");
5613 err
= ext4_journal_get_write_access(handle
, iloc
->bh
);
5619 ext4_std_error(inode
->i_sb
, err
);
5624 * Expand an inode by new_extra_isize bytes.
5625 * Returns 0 on success or negative error number on failure.
5627 static int ext4_expand_extra_isize(struct inode
*inode
,
5628 unsigned int new_extra_isize
,
5629 struct ext4_iloc iloc
,
5632 struct ext4_inode
*raw_inode
;
5633 struct ext4_xattr_ibody_header
*header
;
5635 if (EXT4_I(inode
)->i_extra_isize
>= new_extra_isize
)
5638 raw_inode
= ext4_raw_inode(&iloc
);
5640 header
= IHDR(inode
, raw_inode
);
5642 /* No extended attributes present */
5643 if (!ext4_test_inode_state(inode
, EXT4_STATE_XATTR
) ||
5644 header
->h_magic
!= cpu_to_le32(EXT4_XATTR_MAGIC
)) {
5645 memset((void *)raw_inode
+ EXT4_GOOD_OLD_INODE_SIZE
, 0,
5647 EXT4_I(inode
)->i_extra_isize
= new_extra_isize
;
5651 /* try to expand with EAs present */
5652 return ext4_expand_extra_isize_ea(inode
, new_extra_isize
,
5657 * What we do here is to mark the in-core inode as clean with respect to inode
5658 * dirtiness (it may still be data-dirty).
5659 * This means that the in-core inode may be reaped by prune_icache
5660 * without having to perform any I/O. This is a very good thing,
5661 * because *any* task may call prune_icache - even ones which
5662 * have a transaction open against a different journal.
5664 * Is this cheating? Not really. Sure, we haven't written the
5665 * inode out, but prune_icache isn't a user-visible syncing function.
5666 * Whenever the user wants stuff synced (sys_sync, sys_msync, sys_fsync)
5667 * we start and wait on commits.
5669 * Is this efficient/effective? Well, we're being nice to the system
5670 * by cleaning up our inodes proactively so they can be reaped
5671 * without I/O. But we are potentially leaving up to five seconds'
5672 * worth of inodes floating about which prune_icache wants us to
5673 * write out. One way to fix that would be to get prune_icache()
5674 * to do a write_super() to free up some memory. It has the desired
5677 int ext4_mark_inode_dirty(handle_t
*handle
, struct inode
*inode
)
5679 struct ext4_iloc iloc
;
5680 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
5681 static unsigned int mnt_count
;
5685 trace_ext4_mark_inode_dirty(inode
, _RET_IP_
);
5686 err
= ext4_reserve_inode_write(handle
, inode
, &iloc
);
5687 if (ext4_handle_valid(handle
) &&
5688 EXT4_I(inode
)->i_extra_isize
< sbi
->s_want_extra_isize
&&
5689 !ext4_test_inode_state(inode
, EXT4_STATE_NO_EXPAND
)) {
5691 * We need extra buffer credits since we may write into EA block
5692 * with this same handle. If journal_extend fails, then it will
5693 * only result in a minor loss of functionality for that inode.
5694 * If this is felt to be critical, then e2fsck should be run to
5695 * force a large enough s_min_extra_isize.
5697 if ((jbd2_journal_extend(handle
,
5698 EXT4_DATA_TRANS_BLOCKS(inode
->i_sb
))) == 0) {
5699 ret
= ext4_expand_extra_isize(inode
,
5700 sbi
->s_want_extra_isize
,
5703 ext4_set_inode_state(inode
,
5704 EXT4_STATE_NO_EXPAND
);
5706 le16_to_cpu(sbi
->s_es
->s_mnt_count
)) {
5707 ext4_warning(inode
->i_sb
,
5708 "Unable to expand inode %lu. Delete"
5709 " some EAs or run e2fsck.",
5712 le16_to_cpu(sbi
->s_es
->s_mnt_count
);
5718 err
= ext4_mark_iloc_dirty(handle
, inode
, &iloc
);
5723 * ext4_dirty_inode() is called from __mark_inode_dirty()
5725 * We're really interested in the case where a file is being extended.
5726 * i_size has been changed by generic_commit_write() and we thus need
5727 * to include the updated inode in the current transaction.
5729 * Also, dquot_alloc_block() will always dirty the inode when blocks
5730 * are allocated to the file.
5732 * If the inode is marked synchronous, we don't honour that here - doing
5733 * so would cause a commit on atime updates, which we don't bother doing.
5734 * We handle synchronous inodes at the highest possible level.
5736 void ext4_dirty_inode(struct inode
*inode
)
5740 handle
= ext4_journal_start(inode
, 2);
5744 ext4_mark_inode_dirty(handle
, inode
);
5746 ext4_journal_stop(handle
);
5753 * Bind an inode's backing buffer_head into this transaction, to prevent
5754 * it from being flushed to disk early. Unlike
5755 * ext4_reserve_inode_write, this leaves behind no bh reference and
5756 * returns no iloc structure, so the caller needs to repeat the iloc
5757 * lookup to mark the inode dirty later.
5759 static int ext4_pin_inode(handle_t
*handle
, struct inode
*inode
)
5761 struct ext4_iloc iloc
;
5765 err
= ext4_get_inode_loc(inode
, &iloc
);
5767 BUFFER_TRACE(iloc
.bh
, "get_write_access");
5768 err
= jbd2_journal_get_write_access(handle
, iloc
.bh
);
5770 err
= ext4_handle_dirty_metadata(handle
,
5776 ext4_std_error(inode
->i_sb
, err
);
5781 int ext4_change_inode_journal_flag(struct inode
*inode
, int val
)
5788 * We have to be very careful here: changing a data block's
5789 * journaling status dynamically is dangerous. If we write a
5790 * data block to the journal, change the status and then delete
5791 * that block, we risk forgetting to revoke the old log record
5792 * from the journal and so a subsequent replay can corrupt data.
5793 * So, first we make sure that the journal is empty and that
5794 * nobody is changing anything.
5797 journal
= EXT4_JOURNAL(inode
);
5800 if (is_journal_aborted(journal
))
5803 jbd2_journal_lock_updates(journal
);
5804 jbd2_journal_flush(journal
);
5807 * OK, there are no updates running now, and all cached data is
5808 * synced to disk. We are now in a completely consistent state
5809 * which doesn't have anything in the journal, and we know that
5810 * no filesystem updates are running, so it is safe to modify
5811 * the inode's in-core data-journaling state flag now.
5815 ext4_set_inode_flag(inode
, EXT4_INODE_JOURNAL_DATA
);
5817 ext4_clear_inode_flag(inode
, EXT4_INODE_JOURNAL_DATA
);
5818 ext4_set_aops(inode
);
5820 jbd2_journal_unlock_updates(journal
);
5822 /* Finally we can mark the inode as dirty. */
5824 handle
= ext4_journal_start(inode
, 1);
5826 return PTR_ERR(handle
);
5828 err
= ext4_mark_inode_dirty(handle
, inode
);
5829 ext4_handle_sync(handle
);
5830 ext4_journal_stop(handle
);
5831 ext4_std_error(inode
->i_sb
, err
);
5836 static int ext4_bh_unmapped(handle_t
*handle
, struct buffer_head
*bh
)
5838 return !buffer_mapped(bh
);
5841 int ext4_page_mkwrite(struct vm_area_struct
*vma
, struct vm_fault
*vmf
)
5843 struct page
*page
= vmf
->page
;
5848 struct file
*file
= vma
->vm_file
;
5849 struct inode
*inode
= file
->f_path
.dentry
->d_inode
;
5850 struct address_space
*mapping
= inode
->i_mapping
;
5853 * Get i_alloc_sem to stop truncates messing with the inode. We cannot
5854 * get i_mutex because we are already holding mmap_sem.
5856 down_read(&inode
->i_alloc_sem
);
5857 size
= i_size_read(inode
);
5858 if (page
->mapping
!= mapping
|| size
<= page_offset(page
)
5859 || !PageUptodate(page
)) {
5860 /* page got truncated from under us? */
5864 if (PageMappedToDisk(page
))
5867 if (page
->index
== size
>> PAGE_CACHE_SHIFT
)
5868 len
= size
& ~PAGE_CACHE_MASK
;
5870 len
= PAGE_CACHE_SIZE
;
5874 * return if we have all the buffers mapped. This avoid
5875 * the need to call write_begin/write_end which does a
5876 * journal_start/journal_stop which can block and take
5879 if (page_has_buffers(page
)) {
5880 if (!walk_page_buffers(NULL
, page_buffers(page
), 0, len
, NULL
,
5881 ext4_bh_unmapped
)) {
5888 * OK, we need to fill the hole... Do write_begin write_end
5889 * to do block allocation/reservation.We are not holding
5890 * inode.i__mutex here. That allow * parallel write_begin,
5891 * write_end call. lock_page prevent this from happening
5892 * on the same page though
5894 ret
= mapping
->a_ops
->write_begin(file
, mapping
, page_offset(page
),
5895 len
, AOP_FLAG_UNINTERRUPTIBLE
, &page
, &fsdata
);
5898 ret
= mapping
->a_ops
->write_end(file
, mapping
, page_offset(page
),
5899 len
, len
, page
, fsdata
);
5905 ret
= VM_FAULT_SIGBUS
;
5906 up_read(&inode
->i_alloc_sem
);