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>
44 #include "ext4_jbd2.h"
47 #include "ext4_extents.h"
49 #include <trace/events/ext4.h>
51 #define MPAGE_DA_EXTENT_TAIL 0x01
53 static inline int ext4_begin_ordered_truncate(struct inode
*inode
,
56 return jbd2_journal_begin_ordered_truncate(
57 EXT4_SB(inode
->i_sb
)->s_journal
,
58 &EXT4_I(inode
)->jinode
,
62 static void ext4_invalidatepage(struct page
*page
, unsigned long offset
);
65 * Test whether an inode is a fast symlink.
67 static int ext4_inode_is_fast_symlink(struct inode
*inode
)
69 int ea_blocks
= EXT4_I(inode
)->i_file_acl
?
70 (inode
->i_sb
->s_blocksize
>> 9) : 0;
72 return (S_ISLNK(inode
->i_mode
) && inode
->i_blocks
- ea_blocks
== 0);
76 * Work out how many blocks we need to proceed with the next chunk of a
77 * truncate transaction.
79 static unsigned long blocks_for_truncate(struct inode
*inode
)
83 needed
= inode
->i_blocks
>> (inode
->i_sb
->s_blocksize_bits
- 9);
85 /* Give ourselves just enough room to cope with inodes in which
86 * i_blocks is corrupt: we've seen disk corruptions in the past
87 * which resulted in random data in an inode which looked enough
88 * like a regular file for ext4 to try to delete it. Things
89 * will go a bit crazy if that happens, but at least we should
90 * try not to panic the whole kernel. */
94 /* But we need to bound the transaction so we don't overflow the
96 if (needed
> EXT4_MAX_TRANS_DATA
)
97 needed
= EXT4_MAX_TRANS_DATA
;
99 return EXT4_DATA_TRANS_BLOCKS(inode
->i_sb
) + needed
;
103 * Truncate transactions can be complex and absolutely huge. So we need to
104 * be able to restart the transaction at a conventient checkpoint to make
105 * sure we don't overflow the journal.
107 * start_transaction gets us a new handle for a truncate transaction,
108 * and extend_transaction tries to extend the existing one a bit. If
109 * extend fails, we need to propagate the failure up and restart the
110 * transaction in the top-level truncate loop. --sct
112 static handle_t
*start_transaction(struct inode
*inode
)
116 result
= ext4_journal_start(inode
, blocks_for_truncate(inode
));
120 ext4_std_error(inode
->i_sb
, PTR_ERR(result
));
125 * Try to extend this transaction for the purposes of truncation.
127 * Returns 0 if we managed to create more room. If we can't create more
128 * room, and the transaction must be restarted we return 1.
130 static int try_to_extend_transaction(handle_t
*handle
, struct inode
*inode
)
132 if (!ext4_handle_valid(handle
))
134 if (ext4_handle_has_enough_credits(handle
, EXT4_RESERVE_TRANS_BLOCKS
+1))
136 if (!ext4_journal_extend(handle
, blocks_for_truncate(inode
)))
142 * Restart the transaction associated with *handle. This does a commit,
143 * so before we call here everything must be consistently dirtied against
146 int ext4_truncate_restart_trans(handle_t
*handle
, struct inode
*inode
,
152 * Drop i_data_sem to avoid deadlock with ext4_get_blocks At this
153 * moment, get_block can be called only for blocks inside i_size since
154 * page cache has been already dropped and writes are blocked by
155 * i_mutex. So we can safely drop the i_data_sem here.
157 BUG_ON(EXT4_JOURNAL(inode
) == NULL
);
158 jbd_debug(2, "restarting handle %p\n", handle
);
159 up_write(&EXT4_I(inode
)->i_data_sem
);
160 ret
= ext4_journal_restart(handle
, blocks_for_truncate(inode
));
161 down_write(&EXT4_I(inode
)->i_data_sem
);
162 ext4_discard_preallocations(inode
);
168 * Called at the last iput() if i_nlink is zero.
170 void ext4_delete_inode(struct inode
*inode
)
175 if (!is_bad_inode(inode
))
176 dquot_initialize(inode
);
178 if (ext4_should_order_data(inode
))
179 ext4_begin_ordered_truncate(inode
, 0);
180 truncate_inode_pages(&inode
->i_data
, 0);
182 if (is_bad_inode(inode
))
185 handle
= ext4_journal_start(inode
, blocks_for_truncate(inode
)+3);
186 if (IS_ERR(handle
)) {
187 ext4_std_error(inode
->i_sb
, PTR_ERR(handle
));
189 * If we're going to skip the normal cleanup, we still need to
190 * make sure that the in-core orphan linked list is properly
193 ext4_orphan_del(NULL
, inode
);
198 ext4_handle_sync(handle
);
200 err
= ext4_mark_inode_dirty(handle
, inode
);
202 ext4_warning(inode
->i_sb
,
203 "couldn't mark inode dirty (err %d)", err
);
207 ext4_truncate(inode
);
210 * ext4_ext_truncate() doesn't reserve any slop when it
211 * restarts journal transactions; therefore there may not be
212 * enough credits left in the handle to remove the inode from
213 * the orphan list and set the dtime field.
215 if (!ext4_handle_has_enough_credits(handle
, 3)) {
216 err
= ext4_journal_extend(handle
, 3);
218 err
= ext4_journal_restart(handle
, 3);
220 ext4_warning(inode
->i_sb
,
221 "couldn't extend journal (err %d)", err
);
223 ext4_journal_stop(handle
);
229 * Kill off the orphan record which ext4_truncate created.
230 * AKPM: I think this can be inside the above `if'.
231 * Note that ext4_orphan_del() has to be able to cope with the
232 * deletion of a non-existent orphan - this is because we don't
233 * know if ext4_truncate() actually created an orphan record.
234 * (Well, we could do this if we need to, but heck - it works)
236 ext4_orphan_del(handle
, inode
);
237 EXT4_I(inode
)->i_dtime
= get_seconds();
240 * One subtle ordering requirement: if anything has gone wrong
241 * (transaction abort, IO errors, whatever), then we can still
242 * do these next steps (the fs will already have been marked as
243 * having errors), but we can't free the inode if the mark_dirty
246 if (ext4_mark_inode_dirty(handle
, inode
))
247 /* If that failed, just do the required in-core inode clear. */
250 ext4_free_inode(handle
, inode
);
251 ext4_journal_stop(handle
);
254 clear_inode(inode
); /* We must guarantee clearing of inode... */
260 struct buffer_head
*bh
;
263 static inline void add_chain(Indirect
*p
, struct buffer_head
*bh
, __le32
*v
)
265 p
->key
= *(p
->p
= v
);
270 * ext4_block_to_path - parse the block number into array of offsets
271 * @inode: inode in question (we are only interested in its superblock)
272 * @i_block: block number to be parsed
273 * @offsets: array to store the offsets in
274 * @boundary: set this non-zero if the referred-to block is likely to be
275 * followed (on disk) by an indirect block.
277 * To store the locations of file's data ext4 uses a data structure common
278 * for UNIX filesystems - tree of pointers anchored in the inode, with
279 * data blocks at leaves and indirect blocks in intermediate nodes.
280 * This function translates the block number into path in that tree -
281 * return value is the path length and @offsets[n] is the offset of
282 * pointer to (n+1)th node in the nth one. If @block is out of range
283 * (negative or too large) warning is printed and zero returned.
285 * Note: function doesn't find node addresses, so no IO is needed. All
286 * we need to know is the capacity of indirect blocks (taken from the
291 * Portability note: the last comparison (check that we fit into triple
292 * indirect block) is spelled differently, because otherwise on an
293 * architecture with 32-bit longs and 8Kb pages we might get into trouble
294 * if our filesystem had 8Kb blocks. We might use long long, but that would
295 * kill us on x86. Oh, well, at least the sign propagation does not matter -
296 * i_block would have to be negative in the very beginning, so we would not
300 static int ext4_block_to_path(struct inode
*inode
,
302 ext4_lblk_t offsets
[4], int *boundary
)
304 int ptrs
= EXT4_ADDR_PER_BLOCK(inode
->i_sb
);
305 int ptrs_bits
= EXT4_ADDR_PER_BLOCK_BITS(inode
->i_sb
);
306 const long direct_blocks
= EXT4_NDIR_BLOCKS
,
307 indirect_blocks
= ptrs
,
308 double_blocks
= (1 << (ptrs_bits
* 2));
312 if (i_block
< direct_blocks
) {
313 offsets
[n
++] = i_block
;
314 final
= direct_blocks
;
315 } else if ((i_block
-= direct_blocks
) < indirect_blocks
) {
316 offsets
[n
++] = EXT4_IND_BLOCK
;
317 offsets
[n
++] = i_block
;
319 } else if ((i_block
-= indirect_blocks
) < double_blocks
) {
320 offsets
[n
++] = EXT4_DIND_BLOCK
;
321 offsets
[n
++] = i_block
>> ptrs_bits
;
322 offsets
[n
++] = i_block
& (ptrs
- 1);
324 } else if (((i_block
-= double_blocks
) >> (ptrs_bits
* 2)) < ptrs
) {
325 offsets
[n
++] = EXT4_TIND_BLOCK
;
326 offsets
[n
++] = i_block
>> (ptrs_bits
* 2);
327 offsets
[n
++] = (i_block
>> ptrs_bits
) & (ptrs
- 1);
328 offsets
[n
++] = i_block
& (ptrs
- 1);
331 ext4_warning(inode
->i_sb
, "block %lu > max in inode %lu",
332 i_block
+ direct_blocks
+
333 indirect_blocks
+ double_blocks
, inode
->i_ino
);
336 *boundary
= final
- 1 - (i_block
& (ptrs
- 1));
340 static int __ext4_check_blockref(const char *function
, struct inode
*inode
,
341 __le32
*p
, unsigned int max
)
346 while (bref
< p
+max
) {
347 blk
= le32_to_cpu(*bref
++);
349 unlikely(!ext4_data_block_valid(EXT4_SB(inode
->i_sb
),
351 __ext4_error(inode
->i_sb
, function
,
352 "invalid block reference %u "
353 "in inode #%lu", blk
, inode
->i_ino
);
361 #define ext4_check_indirect_blockref(inode, bh) \
362 __ext4_check_blockref(__func__, inode, (__le32 *)(bh)->b_data, \
363 EXT4_ADDR_PER_BLOCK((inode)->i_sb))
365 #define ext4_check_inode_blockref(inode) \
366 __ext4_check_blockref(__func__, inode, EXT4_I(inode)->i_data, \
370 * ext4_get_branch - read the chain of indirect blocks leading to data
371 * @inode: inode in question
372 * @depth: depth of the chain (1 - direct pointer, etc.)
373 * @offsets: offsets of pointers in inode/indirect blocks
374 * @chain: place to store the result
375 * @err: here we store the error value
377 * Function fills the array of triples <key, p, bh> and returns %NULL
378 * if everything went OK or the pointer to the last filled triple
379 * (incomplete one) otherwise. Upon the return chain[i].key contains
380 * the number of (i+1)-th block in the chain (as it is stored in memory,
381 * i.e. little-endian 32-bit), chain[i].p contains the address of that
382 * number (it points into struct inode for i==0 and into the bh->b_data
383 * for i>0) and chain[i].bh points to the buffer_head of i-th indirect
384 * block for i>0 and NULL for i==0. In other words, it holds the block
385 * numbers of the chain, addresses they were taken from (and where we can
386 * verify that chain did not change) and buffer_heads hosting these
389 * Function stops when it stumbles upon zero pointer (absent block)
390 * (pointer to last triple returned, *@err == 0)
391 * or when it gets an IO error reading an indirect block
392 * (ditto, *@err == -EIO)
393 * or when it reads all @depth-1 indirect blocks successfully and finds
394 * the whole chain, all way to the data (returns %NULL, *err == 0).
396 * Need to be called with
397 * down_read(&EXT4_I(inode)->i_data_sem)
399 static Indirect
*ext4_get_branch(struct inode
*inode
, int depth
,
400 ext4_lblk_t
*offsets
,
401 Indirect chain
[4], int *err
)
403 struct super_block
*sb
= inode
->i_sb
;
405 struct buffer_head
*bh
;
408 /* i_data is not going away, no lock needed */
409 add_chain(chain
, NULL
, EXT4_I(inode
)->i_data
+ *offsets
);
413 bh
= sb_getblk(sb
, le32_to_cpu(p
->key
));
417 if (!bh_uptodate_or_lock(bh
)) {
418 if (bh_submit_read(bh
) < 0) {
422 /* validate block references */
423 if (ext4_check_indirect_blockref(inode
, bh
)) {
429 add_chain(++p
, bh
, (__le32
*)bh
->b_data
+ *++offsets
);
443 * ext4_find_near - find a place for allocation with sufficient locality
445 * @ind: descriptor of indirect block.
447 * This function returns the preferred place for block allocation.
448 * It is used when heuristic for sequential allocation fails.
450 * + if there is a block to the left of our position - allocate near it.
451 * + if pointer will live in indirect block - allocate near that block.
452 * + if pointer will live in inode - allocate in the same
455 * In the latter case we colour the starting block by the callers PID to
456 * prevent it from clashing with concurrent allocations for a different inode
457 * in the same block group. The PID is used here so that functionally related
458 * files will be close-by on-disk.
460 * Caller must make sure that @ind is valid and will stay that way.
462 static ext4_fsblk_t
ext4_find_near(struct inode
*inode
, Indirect
*ind
)
464 struct ext4_inode_info
*ei
= EXT4_I(inode
);
465 __le32
*start
= ind
->bh
? (__le32
*) ind
->bh
->b_data
: ei
->i_data
;
467 ext4_fsblk_t bg_start
;
468 ext4_fsblk_t last_block
;
469 ext4_grpblk_t colour
;
470 ext4_group_t block_group
;
471 int flex_size
= ext4_flex_bg_size(EXT4_SB(inode
->i_sb
));
473 /* Try to find previous block */
474 for (p
= ind
->p
- 1; p
>= start
; p
--) {
476 return le32_to_cpu(*p
);
479 /* No such thing, so let's try location of indirect block */
481 return ind
->bh
->b_blocknr
;
484 * It is going to be referred to from the inode itself? OK, just put it
485 * into the same cylinder group then.
487 block_group
= ei
->i_block_group
;
488 if (flex_size
>= EXT4_FLEX_SIZE_DIR_ALLOC_SCHEME
) {
489 block_group
&= ~(flex_size
-1);
490 if (S_ISREG(inode
->i_mode
))
493 bg_start
= ext4_group_first_block_no(inode
->i_sb
, block_group
);
494 last_block
= ext4_blocks_count(EXT4_SB(inode
->i_sb
)->s_es
) - 1;
497 * If we are doing delayed allocation, we don't need take
498 * colour into account.
500 if (test_opt(inode
->i_sb
, DELALLOC
))
503 if (bg_start
+ EXT4_BLOCKS_PER_GROUP(inode
->i_sb
) <= last_block
)
504 colour
= (current
->pid
% 16) *
505 (EXT4_BLOCKS_PER_GROUP(inode
->i_sb
) / 16);
507 colour
= (current
->pid
% 16) * ((last_block
- bg_start
) / 16);
508 return bg_start
+ colour
;
512 * ext4_find_goal - find a preferred place for allocation.
514 * @block: block we want
515 * @partial: pointer to the last triple within a chain
517 * Normally this function find the preferred place for block allocation,
519 * Because this is only used for non-extent files, we limit the block nr
522 static ext4_fsblk_t
ext4_find_goal(struct inode
*inode
, ext4_lblk_t block
,
528 * XXX need to get goal block from mballoc's data structures
531 goal
= ext4_find_near(inode
, partial
);
532 goal
= goal
& EXT4_MAX_BLOCK_FILE_PHYS
;
537 * ext4_blks_to_allocate: Look up the block map and count the number
538 * of direct blocks need to be allocated for the given branch.
540 * @branch: chain of indirect blocks
541 * @k: number of blocks need for indirect blocks
542 * @blks: number of data blocks to be mapped.
543 * @blocks_to_boundary: the offset in the indirect block
545 * return the total number of blocks to be allocate, including the
546 * direct and indirect blocks.
548 static int ext4_blks_to_allocate(Indirect
*branch
, int k
, unsigned int blks
,
549 int blocks_to_boundary
)
551 unsigned int count
= 0;
554 * Simple case, [t,d]Indirect block(s) has not allocated yet
555 * then it's clear blocks on that path have not allocated
558 /* right now we don't handle cross boundary allocation */
559 if (blks
< blocks_to_boundary
+ 1)
562 count
+= blocks_to_boundary
+ 1;
567 while (count
< blks
&& count
<= blocks_to_boundary
&&
568 le32_to_cpu(*(branch
[0].p
+ count
)) == 0) {
575 * ext4_alloc_blocks: multiple allocate blocks needed for a branch
576 * @indirect_blks: the number of blocks need to allocate for indirect
579 * @new_blocks: on return it will store the new block numbers for
580 * the indirect blocks(if needed) and the first direct block,
581 * @blks: on return it will store the total number of allocated
584 static int ext4_alloc_blocks(handle_t
*handle
, struct inode
*inode
,
585 ext4_lblk_t iblock
, ext4_fsblk_t goal
,
586 int indirect_blks
, int blks
,
587 ext4_fsblk_t new_blocks
[4], int *err
)
589 struct ext4_allocation_request ar
;
591 unsigned long count
= 0, blk_allocated
= 0;
593 ext4_fsblk_t current_block
= 0;
597 * Here we try to allocate the requested multiple blocks at once,
598 * on a best-effort basis.
599 * To build a branch, we should allocate blocks for
600 * the indirect blocks(if not allocated yet), and at least
601 * the first direct block of this branch. That's the
602 * minimum number of blocks need to allocate(required)
604 /* first we try to allocate the indirect blocks */
605 target
= indirect_blks
;
608 /* allocating blocks for indirect blocks and direct blocks */
609 current_block
= ext4_new_meta_blocks(handle
, inode
,
614 if (unlikely(current_block
+ count
> EXT4_MAX_BLOCK_FILE_PHYS
)) {
615 EXT4_ERROR_INODE(inode
,
616 "current_block %llu + count %lu > %d!",
617 current_block
, count
,
618 EXT4_MAX_BLOCK_FILE_PHYS
);
624 /* allocate blocks for indirect blocks */
625 while (index
< indirect_blks
&& count
) {
626 new_blocks
[index
++] = current_block
++;
631 * save the new block number
632 * for the first direct block
634 new_blocks
[index
] = current_block
;
635 printk(KERN_INFO
"%s returned more blocks than "
636 "requested\n", __func__
);
642 target
= blks
- count
;
643 blk_allocated
= count
;
646 /* Now allocate data blocks */
647 memset(&ar
, 0, sizeof(ar
));
652 if (S_ISREG(inode
->i_mode
))
653 /* enable in-core preallocation only for regular files */
654 ar
.flags
= EXT4_MB_HINT_DATA
;
656 current_block
= ext4_mb_new_blocks(handle
, &ar
, err
);
657 if (unlikely(current_block
+ ar
.len
> EXT4_MAX_BLOCK_FILE_PHYS
)) {
658 EXT4_ERROR_INODE(inode
,
659 "current_block %llu + ar.len %d > %d!",
660 current_block
, ar
.len
,
661 EXT4_MAX_BLOCK_FILE_PHYS
);
666 if (*err
&& (target
== blks
)) {
668 * if the allocation failed and we didn't allocate
674 if (target
== blks
) {
676 * save the new block number
677 * for the first direct block
679 new_blocks
[index
] = current_block
;
681 blk_allocated
+= ar
.len
;
684 /* total number of blocks allocated for direct blocks */
689 for (i
= 0; i
< index
; i
++)
690 ext4_free_blocks(handle
, inode
, 0, new_blocks
[i
], 1, 0);
695 * ext4_alloc_branch - allocate and set up a chain of blocks.
697 * @indirect_blks: number of allocated indirect blocks
698 * @blks: number of allocated direct blocks
699 * @offsets: offsets (in the blocks) to store the pointers to next.
700 * @branch: place to store the chain in.
702 * This function allocates blocks, zeroes out all but the last one,
703 * links them into chain and (if we are synchronous) writes them to disk.
704 * In other words, it prepares a branch that can be spliced onto the
705 * inode. It stores the information about that chain in the branch[], in
706 * the same format as ext4_get_branch() would do. We are calling it after
707 * we had read the existing part of chain and partial points to the last
708 * triple of that (one with zero ->key). Upon the exit we have the same
709 * picture as after the successful ext4_get_block(), except that in one
710 * place chain is disconnected - *branch->p is still zero (we did not
711 * set the last link), but branch->key contains the number that should
712 * be placed into *branch->p to fill that gap.
714 * If allocation fails we free all blocks we've allocated (and forget
715 * their buffer_heads) and return the error value the from failed
716 * ext4_alloc_block() (normally -ENOSPC). Otherwise we set the chain
717 * as described above and return 0.
719 static int ext4_alloc_branch(handle_t
*handle
, struct inode
*inode
,
720 ext4_lblk_t iblock
, int indirect_blks
,
721 int *blks
, ext4_fsblk_t goal
,
722 ext4_lblk_t
*offsets
, Indirect
*branch
)
724 int blocksize
= inode
->i_sb
->s_blocksize
;
727 struct buffer_head
*bh
;
729 ext4_fsblk_t new_blocks
[4];
730 ext4_fsblk_t current_block
;
732 num
= ext4_alloc_blocks(handle
, inode
, iblock
, goal
, indirect_blks
,
733 *blks
, new_blocks
, &err
);
737 branch
[0].key
= cpu_to_le32(new_blocks
[0]);
739 * metadata blocks and data blocks are allocated.
741 for (n
= 1; n
<= indirect_blks
; n
++) {
743 * Get buffer_head for parent block, zero it out
744 * and set the pointer to new one, then send
747 bh
= sb_getblk(inode
->i_sb
, new_blocks
[n
-1]);
750 BUFFER_TRACE(bh
, "call get_create_access");
751 err
= ext4_journal_get_create_access(handle
, bh
);
753 /* Don't brelse(bh) here; it's done in
754 * ext4_journal_forget() below */
759 memset(bh
->b_data
, 0, blocksize
);
760 branch
[n
].p
= (__le32
*) bh
->b_data
+ offsets
[n
];
761 branch
[n
].key
= cpu_to_le32(new_blocks
[n
]);
762 *branch
[n
].p
= branch
[n
].key
;
763 if (n
== indirect_blks
) {
764 current_block
= new_blocks
[n
];
766 * End of chain, update the last new metablock of
767 * the chain to point to the new allocated
768 * data blocks numbers
770 for (i
= 1; i
< num
; i
++)
771 *(branch
[n
].p
+ i
) = cpu_to_le32(++current_block
);
773 BUFFER_TRACE(bh
, "marking uptodate");
774 set_buffer_uptodate(bh
);
777 BUFFER_TRACE(bh
, "call ext4_handle_dirty_metadata");
778 err
= ext4_handle_dirty_metadata(handle
, inode
, bh
);
785 /* Allocation failed, free what we already allocated */
786 ext4_free_blocks(handle
, inode
, 0, new_blocks
[0], 1, 0);
787 for (i
= 1; i
<= n
; i
++) {
789 * branch[i].bh is newly allocated, so there is no
790 * need to revoke the block, which is why we don't
791 * need to set EXT4_FREE_BLOCKS_METADATA.
793 ext4_free_blocks(handle
, inode
, 0, new_blocks
[i
], 1,
794 EXT4_FREE_BLOCKS_FORGET
);
796 for (i
= n
+1; i
< indirect_blks
; i
++)
797 ext4_free_blocks(handle
, inode
, 0, new_blocks
[i
], 1, 0);
799 ext4_free_blocks(handle
, inode
, 0, new_blocks
[i
], num
, 0);
805 * ext4_splice_branch - splice the allocated branch onto inode.
807 * @block: (logical) number of block we are adding
808 * @chain: chain of indirect blocks (with a missing link - see
810 * @where: location of missing link
811 * @num: number of indirect blocks we are adding
812 * @blks: number of direct blocks we are adding
814 * This function fills the missing link and does all housekeeping needed in
815 * inode (->i_blocks, etc.). In case of success we end up with the full
816 * chain to new block and return 0.
818 static int ext4_splice_branch(handle_t
*handle
, struct inode
*inode
,
819 ext4_lblk_t block
, Indirect
*where
, int num
,
824 ext4_fsblk_t current_block
;
827 * If we're splicing into a [td]indirect block (as opposed to the
828 * inode) then we need to get write access to the [td]indirect block
832 BUFFER_TRACE(where
->bh
, "get_write_access");
833 err
= ext4_journal_get_write_access(handle
, where
->bh
);
839 *where
->p
= where
->key
;
842 * Update the host buffer_head or inode to point to more just allocated
843 * direct blocks blocks
845 if (num
== 0 && blks
> 1) {
846 current_block
= le32_to_cpu(where
->key
) + 1;
847 for (i
= 1; i
< blks
; i
++)
848 *(where
->p
+ i
) = cpu_to_le32(current_block
++);
851 /* We are done with atomic stuff, now do the rest of housekeeping */
852 /* had we spliced it onto indirect block? */
855 * If we spliced it onto an indirect block, we haven't
856 * altered the inode. Note however that if it is being spliced
857 * onto an indirect block at the very end of the file (the
858 * file is growing) then we *will* alter the inode to reflect
859 * the new i_size. But that is not done here - it is done in
860 * generic_commit_write->__mark_inode_dirty->ext4_dirty_inode.
862 jbd_debug(5, "splicing indirect only\n");
863 BUFFER_TRACE(where
->bh
, "call ext4_handle_dirty_metadata");
864 err
= ext4_handle_dirty_metadata(handle
, inode
, where
->bh
);
869 * OK, we spliced it into the inode itself on a direct block.
871 ext4_mark_inode_dirty(handle
, inode
);
872 jbd_debug(5, "splicing direct\n");
877 for (i
= 1; i
<= num
; i
++) {
879 * branch[i].bh is newly allocated, so there is no
880 * need to revoke the block, which is why we don't
881 * need to set EXT4_FREE_BLOCKS_METADATA.
883 ext4_free_blocks(handle
, inode
, where
[i
].bh
, 0, 1,
884 EXT4_FREE_BLOCKS_FORGET
);
886 ext4_free_blocks(handle
, inode
, 0, le32_to_cpu(where
[num
].key
),
893 * The ext4_ind_get_blocks() function handles non-extents inodes
894 * (i.e., using the traditional indirect/double-indirect i_blocks
895 * scheme) for ext4_get_blocks().
897 * Allocation strategy is simple: if we have to allocate something, we will
898 * have to go the whole way to leaf. So let's do it before attaching anything
899 * to tree, set linkage between the newborn blocks, write them if sync is
900 * required, recheck the path, free and repeat if check fails, otherwise
901 * set the last missing link (that will protect us from any truncate-generated
902 * removals - all blocks on the path are immune now) and possibly force the
903 * write on the parent block.
904 * That has a nice additional property: no special recovery from the failed
905 * allocations is needed - we simply release blocks and do not touch anything
906 * reachable from inode.
908 * `handle' can be NULL if create == 0.
910 * return > 0, # of blocks mapped or allocated.
911 * return = 0, if plain lookup failed.
912 * return < 0, error case.
914 * The ext4_ind_get_blocks() function should be called with
915 * down_write(&EXT4_I(inode)->i_data_sem) if allocating filesystem
916 * blocks (i.e., flags has EXT4_GET_BLOCKS_CREATE set) or
917 * down_read(&EXT4_I(inode)->i_data_sem) if not allocating file system
920 static int ext4_ind_get_blocks(handle_t
*handle
, struct inode
*inode
,
921 ext4_lblk_t iblock
, unsigned int maxblocks
,
922 struct buffer_head
*bh_result
,
926 ext4_lblk_t offsets
[4];
931 int blocks_to_boundary
= 0;
934 ext4_fsblk_t first_block
= 0;
936 J_ASSERT(!(EXT4_I(inode
)->i_flags
& EXT4_EXTENTS_FL
));
937 J_ASSERT(handle
!= NULL
|| (flags
& EXT4_GET_BLOCKS_CREATE
) == 0);
938 depth
= ext4_block_to_path(inode
, iblock
, offsets
,
939 &blocks_to_boundary
);
944 partial
= ext4_get_branch(inode
, depth
, offsets
, chain
, &err
);
946 /* Simplest case - block found, no allocation needed */
948 first_block
= le32_to_cpu(chain
[depth
- 1].key
);
949 clear_buffer_new(bh_result
);
952 while (count
< maxblocks
&& count
<= blocks_to_boundary
) {
955 blk
= le32_to_cpu(*(chain
[depth
-1].p
+ count
));
957 if (blk
== first_block
+ count
)
965 /* Next simple case - plain lookup or failed read of indirect block */
966 if ((flags
& EXT4_GET_BLOCKS_CREATE
) == 0 || err
== -EIO
)
970 * Okay, we need to do block allocation.
972 goal
= ext4_find_goal(inode
, iblock
, partial
);
974 /* the number of blocks need to allocate for [d,t]indirect blocks */
975 indirect_blks
= (chain
+ depth
) - partial
- 1;
978 * Next look up the indirect map to count the totoal number of
979 * direct blocks to allocate for this branch.
981 count
= ext4_blks_to_allocate(partial
, indirect_blks
,
982 maxblocks
, blocks_to_boundary
);
984 * Block out ext4_truncate while we alter the tree
986 err
= ext4_alloc_branch(handle
, inode
, iblock
, indirect_blks
,
988 offsets
+ (partial
- chain
), partial
);
991 * The ext4_splice_branch call will free and forget any buffers
992 * on the new chain if there is a failure, but that risks using
993 * up transaction credits, especially for bitmaps where the
994 * credits cannot be returned. Can we handle this somehow? We
995 * may need to return -EAGAIN upwards in the worst case. --sct
998 err
= ext4_splice_branch(handle
, inode
, iblock
,
999 partial
, indirect_blks
, count
);
1003 set_buffer_new(bh_result
);
1005 ext4_update_inode_fsync_trans(handle
, inode
, 1);
1007 map_bh(bh_result
, inode
->i_sb
, le32_to_cpu(chain
[depth
-1].key
));
1008 if (count
> blocks_to_boundary
)
1009 set_buffer_boundary(bh_result
);
1011 /* Clean up and exit */
1012 partial
= chain
+ depth
- 1; /* the whole chain */
1014 while (partial
> chain
) {
1015 BUFFER_TRACE(partial
->bh
, "call brelse");
1016 brelse(partial
->bh
);
1019 BUFFER_TRACE(bh_result
, "returned");
1025 qsize_t
*ext4_get_reserved_space(struct inode
*inode
)
1027 return &EXT4_I(inode
)->i_reserved_quota
;
1032 * Calculate the number of metadata blocks need to reserve
1033 * to allocate a new block at @lblocks for non extent file based file
1035 static int ext4_indirect_calc_metadata_amount(struct inode
*inode
,
1038 struct ext4_inode_info
*ei
= EXT4_I(inode
);
1039 sector_t dind_mask
= ~((sector_t
)EXT4_ADDR_PER_BLOCK(inode
->i_sb
) - 1);
1042 if (lblock
< EXT4_NDIR_BLOCKS
)
1045 lblock
-= EXT4_NDIR_BLOCKS
;
1047 if (ei
->i_da_metadata_calc_len
&&
1048 (lblock
& dind_mask
) == ei
->i_da_metadata_calc_last_lblock
) {
1049 ei
->i_da_metadata_calc_len
++;
1052 ei
->i_da_metadata_calc_last_lblock
= lblock
& dind_mask
;
1053 ei
->i_da_metadata_calc_len
= 1;
1054 blk_bits
= order_base_2(lblock
);
1055 return (blk_bits
/ EXT4_ADDR_PER_BLOCK_BITS(inode
->i_sb
)) + 1;
1059 * Calculate the number of metadata blocks need to reserve
1060 * to allocate a block located at @lblock
1062 static int ext4_calc_metadata_amount(struct inode
*inode
, sector_t lblock
)
1064 if (EXT4_I(inode
)->i_flags
& EXT4_EXTENTS_FL
)
1065 return ext4_ext_calc_metadata_amount(inode
, lblock
);
1067 return ext4_indirect_calc_metadata_amount(inode
, lblock
);
1071 * Called with i_data_sem down, which is important since we can call
1072 * ext4_discard_preallocations() from here.
1074 void ext4_da_update_reserve_space(struct inode
*inode
,
1075 int used
, int quota_claim
)
1077 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
1078 struct ext4_inode_info
*ei
= EXT4_I(inode
);
1080 spin_lock(&ei
->i_block_reservation_lock
);
1081 trace_ext4_da_update_reserve_space(inode
, used
);
1082 if (unlikely(used
> ei
->i_reserved_data_blocks
)) {
1083 ext4_msg(inode
->i_sb
, KERN_NOTICE
, "%s: ino %lu, used %d "
1084 "with only %d reserved data blocks\n",
1085 __func__
, inode
->i_ino
, used
,
1086 ei
->i_reserved_data_blocks
);
1088 used
= ei
->i_reserved_data_blocks
;
1091 /* Update per-inode reservations */
1092 ei
->i_reserved_data_blocks
-= used
;
1093 ei
->i_reserved_meta_blocks
-= ei
->i_allocated_meta_blocks
;
1094 percpu_counter_sub(&sbi
->s_dirtyblocks_counter
,
1095 used
+ ei
->i_allocated_meta_blocks
);
1096 ei
->i_allocated_meta_blocks
= 0;
1098 if (ei
->i_reserved_data_blocks
== 0) {
1100 * We can release all of the reserved metadata blocks
1101 * only when we have written all of the delayed
1102 * allocation blocks.
1104 percpu_counter_sub(&sbi
->s_dirtyblocks_counter
,
1105 ei
->i_reserved_meta_blocks
);
1106 ei
->i_reserved_meta_blocks
= 0;
1107 ei
->i_da_metadata_calc_len
= 0;
1109 spin_unlock(&EXT4_I(inode
)->i_block_reservation_lock
);
1111 /* Update quota subsystem for data blocks */
1113 dquot_claim_block(inode
, used
);
1116 * We did fallocate with an offset that is already delayed
1117 * allocated. So on delayed allocated writeback we should
1118 * not re-claim the quota for fallocated blocks.
1120 dquot_release_reservation_block(inode
, used
);
1124 * If we have done all the pending block allocations and if
1125 * there aren't any writers on the inode, we can discard the
1126 * inode's preallocations.
1128 if ((ei
->i_reserved_data_blocks
== 0) &&
1129 (atomic_read(&inode
->i_writecount
) == 0))
1130 ext4_discard_preallocations(inode
);
1133 static int check_block_validity(struct inode
*inode
, const char *msg
,
1134 sector_t logical
, sector_t phys
, int len
)
1136 if (!ext4_data_block_valid(EXT4_SB(inode
->i_sb
), phys
, len
)) {
1137 __ext4_error(inode
->i_sb
, msg
,
1138 "inode #%lu logical block %llu mapped to %llu "
1139 "(size %d)", inode
->i_ino
,
1140 (unsigned long long) logical
,
1141 (unsigned long long) phys
, len
);
1148 * Return the number of contiguous dirty pages in a given inode
1149 * starting at page frame idx.
1151 static pgoff_t
ext4_num_dirty_pages(struct inode
*inode
, pgoff_t idx
,
1152 unsigned int max_pages
)
1154 struct address_space
*mapping
= inode
->i_mapping
;
1156 struct pagevec pvec
;
1158 int i
, nr_pages
, done
= 0;
1162 pagevec_init(&pvec
, 0);
1165 nr_pages
= pagevec_lookup_tag(&pvec
, mapping
, &index
,
1166 PAGECACHE_TAG_DIRTY
,
1167 (pgoff_t
)PAGEVEC_SIZE
);
1170 for (i
= 0; i
< nr_pages
; i
++) {
1171 struct page
*page
= pvec
.pages
[i
];
1172 struct buffer_head
*bh
, *head
;
1175 if (unlikely(page
->mapping
!= mapping
) ||
1177 PageWriteback(page
) ||
1178 page
->index
!= idx
) {
1183 if (page_has_buffers(page
)) {
1184 bh
= head
= page_buffers(page
);
1186 if (!buffer_delay(bh
) &&
1187 !buffer_unwritten(bh
))
1189 bh
= bh
->b_this_page
;
1190 } while (!done
&& (bh
!= head
));
1197 if (num
>= max_pages
)
1200 pagevec_release(&pvec
);
1206 * The ext4_get_blocks() function tries to look up the requested blocks,
1207 * and returns if the blocks are already mapped.
1209 * Otherwise it takes the write lock of the i_data_sem and allocate blocks
1210 * and store the allocated blocks in the result buffer head and mark it
1213 * If file type is extents based, it will call ext4_ext_get_blocks(),
1214 * Otherwise, call with ext4_ind_get_blocks() to handle indirect mapping
1217 * On success, it returns the number of blocks being mapped or allocate.
1218 * if create==0 and the blocks are pre-allocated and uninitialized block,
1219 * the result buffer head is unmapped. If the create ==1, it will make sure
1220 * the buffer head is mapped.
1222 * It returns 0 if plain look up failed (blocks have not been allocated), in
1223 * that casem, buffer head is unmapped
1225 * It returns the error in case of allocation failure.
1227 int ext4_get_blocks(handle_t
*handle
, struct inode
*inode
, sector_t block
,
1228 unsigned int max_blocks
, struct buffer_head
*bh
,
1233 clear_buffer_mapped(bh
);
1234 clear_buffer_unwritten(bh
);
1236 ext_debug("ext4_get_blocks(): inode %lu, flag %d, max_blocks %u,"
1237 "logical block %lu\n", inode
->i_ino
, flags
, max_blocks
,
1238 (unsigned long)block
);
1240 * Try to see if we can get the block without requesting a new
1241 * file system block.
1243 down_read((&EXT4_I(inode
)->i_data_sem
));
1244 if (EXT4_I(inode
)->i_flags
& EXT4_EXTENTS_FL
) {
1245 retval
= ext4_ext_get_blocks(handle
, inode
, block
, max_blocks
,
1248 retval
= ext4_ind_get_blocks(handle
, inode
, block
, max_blocks
,
1251 up_read((&EXT4_I(inode
)->i_data_sem
));
1253 if (retval
> 0 && buffer_mapped(bh
)) {
1254 int ret
= check_block_validity(inode
, "file system corruption",
1255 block
, bh
->b_blocknr
, retval
);
1260 /* If it is only a block(s) look up */
1261 if ((flags
& EXT4_GET_BLOCKS_CREATE
) == 0)
1265 * Returns if the blocks have already allocated
1267 * Note that if blocks have been preallocated
1268 * ext4_ext_get_block() returns th create = 0
1269 * with buffer head unmapped.
1271 if (retval
> 0 && buffer_mapped(bh
))
1275 * When we call get_blocks without the create flag, the
1276 * BH_Unwritten flag could have gotten set if the blocks
1277 * requested were part of a uninitialized extent. We need to
1278 * clear this flag now that we are committed to convert all or
1279 * part of the uninitialized extent to be an initialized
1280 * extent. This is because we need to avoid the combination
1281 * of BH_Unwritten and BH_Mapped flags being simultaneously
1282 * set on the buffer_head.
1284 clear_buffer_unwritten(bh
);
1287 * New blocks allocate and/or writing to uninitialized extent
1288 * will possibly result in updating i_data, so we take
1289 * the write lock of i_data_sem, and call get_blocks()
1290 * with create == 1 flag.
1292 down_write((&EXT4_I(inode
)->i_data_sem
));
1295 * if the caller is from delayed allocation writeout path
1296 * we have already reserved fs blocks for allocation
1297 * let the underlying get_block() function know to
1298 * avoid double accounting
1300 if (flags
& EXT4_GET_BLOCKS_DELALLOC_RESERVE
)
1301 EXT4_I(inode
)->i_delalloc_reserved_flag
= 1;
1303 * We need to check for EXT4 here because migrate
1304 * could have changed the inode type in between
1306 if (EXT4_I(inode
)->i_flags
& EXT4_EXTENTS_FL
) {
1307 retval
= ext4_ext_get_blocks(handle
, inode
, block
, max_blocks
,
1310 retval
= ext4_ind_get_blocks(handle
, inode
, block
,
1311 max_blocks
, bh
, flags
);
1313 if (retval
> 0 && buffer_new(bh
)) {
1315 * We allocated new blocks which will result in
1316 * i_data's format changing. Force the migrate
1317 * to fail by clearing migrate flags
1319 ext4_clear_inode_state(inode
, EXT4_STATE_EXT_MIGRATE
);
1323 * Update reserved blocks/metadata blocks after successful
1324 * block allocation which had been deferred till now. We don't
1325 * support fallocate for non extent files. So we can update
1326 * reserve space here.
1329 (flags
& EXT4_GET_BLOCKS_DELALLOC_RESERVE
))
1330 ext4_da_update_reserve_space(inode
, retval
, 1);
1332 if (flags
& EXT4_GET_BLOCKS_DELALLOC_RESERVE
)
1333 EXT4_I(inode
)->i_delalloc_reserved_flag
= 0;
1335 up_write((&EXT4_I(inode
)->i_data_sem
));
1336 if (retval
> 0 && buffer_mapped(bh
)) {
1337 int ret
= check_block_validity(inode
, "file system "
1338 "corruption after allocation",
1339 block
, bh
->b_blocknr
, retval
);
1346 /* Maximum number of blocks we map for direct IO at once. */
1347 #define DIO_MAX_BLOCKS 4096
1349 int ext4_get_block(struct inode
*inode
, sector_t iblock
,
1350 struct buffer_head
*bh_result
, int create
)
1352 handle_t
*handle
= ext4_journal_current_handle();
1353 int ret
= 0, started
= 0;
1354 unsigned max_blocks
= bh_result
->b_size
>> inode
->i_blkbits
;
1357 if (create
&& !handle
) {
1358 /* Direct IO write... */
1359 if (max_blocks
> DIO_MAX_BLOCKS
)
1360 max_blocks
= DIO_MAX_BLOCKS
;
1361 dio_credits
= ext4_chunk_trans_blocks(inode
, max_blocks
);
1362 handle
= ext4_journal_start(inode
, dio_credits
);
1363 if (IS_ERR(handle
)) {
1364 ret
= PTR_ERR(handle
);
1370 ret
= ext4_get_blocks(handle
, inode
, iblock
, max_blocks
, bh_result
,
1371 create
? EXT4_GET_BLOCKS_CREATE
: 0);
1373 bh_result
->b_size
= (ret
<< inode
->i_blkbits
);
1377 ext4_journal_stop(handle
);
1383 * `handle' can be NULL if create is zero
1385 struct buffer_head
*ext4_getblk(handle_t
*handle
, struct inode
*inode
,
1386 ext4_lblk_t block
, int create
, int *errp
)
1388 struct buffer_head dummy
;
1392 J_ASSERT(handle
!= NULL
|| create
== 0);
1395 dummy
.b_blocknr
= -1000;
1396 buffer_trace_init(&dummy
.b_history
);
1398 flags
|= EXT4_GET_BLOCKS_CREATE
;
1399 err
= ext4_get_blocks(handle
, inode
, block
, 1, &dummy
, flags
);
1401 * ext4_get_blocks() returns number of blocks mapped. 0 in
1410 if (!err
&& buffer_mapped(&dummy
)) {
1411 struct buffer_head
*bh
;
1412 bh
= sb_getblk(inode
->i_sb
, dummy
.b_blocknr
);
1417 if (buffer_new(&dummy
)) {
1418 J_ASSERT(create
!= 0);
1419 J_ASSERT(handle
!= NULL
);
1422 * Now that we do not always journal data, we should
1423 * keep in mind whether this should always journal the
1424 * new buffer as metadata. For now, regular file
1425 * writes use ext4_get_block instead, so it's not a
1429 BUFFER_TRACE(bh
, "call get_create_access");
1430 fatal
= ext4_journal_get_create_access(handle
, bh
);
1431 if (!fatal
&& !buffer_uptodate(bh
)) {
1432 memset(bh
->b_data
, 0, inode
->i_sb
->s_blocksize
);
1433 set_buffer_uptodate(bh
);
1436 BUFFER_TRACE(bh
, "call ext4_handle_dirty_metadata");
1437 err
= ext4_handle_dirty_metadata(handle
, inode
, bh
);
1441 BUFFER_TRACE(bh
, "not a new buffer");
1454 struct buffer_head
*ext4_bread(handle_t
*handle
, struct inode
*inode
,
1455 ext4_lblk_t block
, int create
, int *err
)
1457 struct buffer_head
*bh
;
1459 bh
= ext4_getblk(handle
, inode
, block
, create
, err
);
1462 if (buffer_uptodate(bh
))
1464 ll_rw_block(READ_META
, 1, &bh
);
1466 if (buffer_uptodate(bh
))
1473 static int walk_page_buffers(handle_t
*handle
,
1474 struct buffer_head
*head
,
1478 int (*fn
)(handle_t
*handle
,
1479 struct buffer_head
*bh
))
1481 struct buffer_head
*bh
;
1482 unsigned block_start
, block_end
;
1483 unsigned blocksize
= head
->b_size
;
1485 struct buffer_head
*next
;
1487 for (bh
= head
, block_start
= 0;
1488 ret
== 0 && (bh
!= head
|| !block_start
);
1489 block_start
= block_end
, bh
= next
) {
1490 next
= bh
->b_this_page
;
1491 block_end
= block_start
+ blocksize
;
1492 if (block_end
<= from
|| block_start
>= to
) {
1493 if (partial
&& !buffer_uptodate(bh
))
1497 err
= (*fn
)(handle
, bh
);
1505 * To preserve ordering, it is essential that the hole instantiation and
1506 * the data write be encapsulated in a single transaction. We cannot
1507 * close off a transaction and start a new one between the ext4_get_block()
1508 * and the commit_write(). So doing the jbd2_journal_start at the start of
1509 * prepare_write() is the right place.
1511 * Also, this function can nest inside ext4_writepage() ->
1512 * block_write_full_page(). In that case, we *know* that ext4_writepage()
1513 * has generated enough buffer credits to do the whole page. So we won't
1514 * block on the journal in that case, which is good, because the caller may
1517 * By accident, ext4 can be reentered when a transaction is open via
1518 * quota file writes. If we were to commit the transaction while thus
1519 * reentered, there can be a deadlock - we would be holding a quota
1520 * lock, and the commit would never complete if another thread had a
1521 * transaction open and was blocking on the quota lock - a ranking
1524 * So what we do is to rely on the fact that jbd2_journal_stop/journal_start
1525 * will _not_ run commit under these circumstances because handle->h_ref
1526 * is elevated. We'll still have enough credits for the tiny quotafile
1529 static int do_journal_get_write_access(handle_t
*handle
,
1530 struct buffer_head
*bh
)
1532 if (!buffer_mapped(bh
) || buffer_freed(bh
))
1534 return ext4_journal_get_write_access(handle
, bh
);
1538 * Truncate blocks that were not used by write. We have to truncate the
1539 * pagecache as well so that corresponding buffers get properly unmapped.
1541 static void ext4_truncate_failed_write(struct inode
*inode
)
1543 truncate_inode_pages(inode
->i_mapping
, inode
->i_size
);
1544 ext4_truncate(inode
);
1547 static int ext4_get_block_write(struct inode
*inode
, sector_t iblock
,
1548 struct buffer_head
*bh_result
, int create
);
1549 static int ext4_write_begin(struct file
*file
, struct address_space
*mapping
,
1550 loff_t pos
, unsigned len
, unsigned flags
,
1551 struct page
**pagep
, void **fsdata
)
1553 struct inode
*inode
= mapping
->host
;
1554 int ret
, needed_blocks
;
1561 trace_ext4_write_begin(inode
, pos
, len
, flags
);
1563 * Reserve one block more for addition to orphan list in case
1564 * we allocate blocks but write fails for some reason
1566 needed_blocks
= ext4_writepage_trans_blocks(inode
) + 1;
1567 index
= pos
>> PAGE_CACHE_SHIFT
;
1568 from
= pos
& (PAGE_CACHE_SIZE
- 1);
1572 handle
= ext4_journal_start(inode
, needed_blocks
);
1573 if (IS_ERR(handle
)) {
1574 ret
= PTR_ERR(handle
);
1578 /* We cannot recurse into the filesystem as the transaction is already
1580 flags
|= AOP_FLAG_NOFS
;
1582 page
= grab_cache_page_write_begin(mapping
, index
, flags
);
1584 ext4_journal_stop(handle
);
1590 if (ext4_should_dioread_nolock(inode
))
1591 ret
= block_write_begin(file
, mapping
, pos
, len
, flags
, pagep
,
1592 fsdata
, ext4_get_block_write
);
1594 ret
= block_write_begin(file
, mapping
, pos
, len
, flags
, pagep
,
1595 fsdata
, ext4_get_block
);
1597 if (!ret
&& ext4_should_journal_data(inode
)) {
1598 ret
= walk_page_buffers(handle
, page_buffers(page
),
1599 from
, to
, NULL
, do_journal_get_write_access
);
1604 page_cache_release(page
);
1606 * block_write_begin may have instantiated a few blocks
1607 * outside i_size. Trim these off again. Don't need
1608 * i_size_read because we hold i_mutex.
1610 * Add inode to orphan list in case we crash before
1613 if (pos
+ len
> inode
->i_size
&& ext4_can_truncate(inode
))
1614 ext4_orphan_add(handle
, inode
);
1616 ext4_journal_stop(handle
);
1617 if (pos
+ len
> inode
->i_size
) {
1618 ext4_truncate_failed_write(inode
);
1620 * If truncate failed early the inode might
1621 * still be on the orphan list; we need to
1622 * make sure the inode is removed from the
1623 * orphan list in that case.
1626 ext4_orphan_del(NULL
, inode
);
1630 if (ret
== -ENOSPC
&& ext4_should_retry_alloc(inode
->i_sb
, &retries
))
1636 /* For write_end() in data=journal mode */
1637 static int write_end_fn(handle_t
*handle
, struct buffer_head
*bh
)
1639 if (!buffer_mapped(bh
) || buffer_freed(bh
))
1641 set_buffer_uptodate(bh
);
1642 return ext4_handle_dirty_metadata(handle
, NULL
, bh
);
1645 static int ext4_generic_write_end(struct file
*file
,
1646 struct address_space
*mapping
,
1647 loff_t pos
, unsigned len
, unsigned copied
,
1648 struct page
*page
, void *fsdata
)
1650 int i_size_changed
= 0;
1651 struct inode
*inode
= mapping
->host
;
1652 handle_t
*handle
= ext4_journal_current_handle();
1654 copied
= block_write_end(file
, mapping
, pos
, len
, copied
, page
, fsdata
);
1657 * No need to use i_size_read() here, the i_size
1658 * cannot change under us because we hold i_mutex.
1660 * But it's important to update i_size while still holding page lock:
1661 * page writeout could otherwise come in and zero beyond i_size.
1663 if (pos
+ copied
> inode
->i_size
) {
1664 i_size_write(inode
, pos
+ copied
);
1668 if (pos
+ copied
> EXT4_I(inode
)->i_disksize
) {
1669 /* We need to mark inode dirty even if
1670 * new_i_size is less that inode->i_size
1671 * bu greater than i_disksize.(hint delalloc)
1673 ext4_update_i_disksize(inode
, (pos
+ copied
));
1677 page_cache_release(page
);
1680 * Don't mark the inode dirty under page lock. First, it unnecessarily
1681 * makes the holding time of page lock longer. Second, it forces lock
1682 * ordering of page lock and transaction start for journaling
1686 ext4_mark_inode_dirty(handle
, inode
);
1692 * We need to pick up the new inode size which generic_commit_write gave us
1693 * `file' can be NULL - eg, when called from page_symlink().
1695 * ext4 never places buffers on inode->i_mapping->private_list. metadata
1696 * buffers are managed internally.
1698 static int ext4_ordered_write_end(struct file
*file
,
1699 struct address_space
*mapping
,
1700 loff_t pos
, unsigned len
, unsigned copied
,
1701 struct page
*page
, void *fsdata
)
1703 handle_t
*handle
= ext4_journal_current_handle();
1704 struct inode
*inode
= mapping
->host
;
1707 trace_ext4_ordered_write_end(inode
, pos
, len
, copied
);
1708 ret
= ext4_jbd2_file_inode(handle
, inode
);
1711 ret2
= ext4_generic_write_end(file
, mapping
, pos
, len
, copied
,
1714 if (pos
+ len
> inode
->i_size
&& ext4_can_truncate(inode
))
1715 /* if we have allocated more blocks and copied
1716 * less. We will have blocks allocated outside
1717 * inode->i_size. So truncate them
1719 ext4_orphan_add(handle
, inode
);
1723 ret2
= ext4_journal_stop(handle
);
1727 if (pos
+ len
> inode
->i_size
) {
1728 ext4_truncate_failed_write(inode
);
1730 * If truncate failed early the inode might still be
1731 * on the orphan list; we need to make sure the inode
1732 * is removed from the orphan list in that case.
1735 ext4_orphan_del(NULL
, inode
);
1739 return ret
? ret
: copied
;
1742 static int ext4_writeback_write_end(struct file
*file
,
1743 struct address_space
*mapping
,
1744 loff_t pos
, unsigned len
, unsigned copied
,
1745 struct page
*page
, void *fsdata
)
1747 handle_t
*handle
= ext4_journal_current_handle();
1748 struct inode
*inode
= mapping
->host
;
1751 trace_ext4_writeback_write_end(inode
, pos
, len
, copied
);
1752 ret2
= ext4_generic_write_end(file
, mapping
, pos
, len
, copied
,
1755 if (pos
+ len
> inode
->i_size
&& ext4_can_truncate(inode
))
1756 /* if we have allocated more blocks and copied
1757 * less. We will have blocks allocated outside
1758 * inode->i_size. So truncate them
1760 ext4_orphan_add(handle
, inode
);
1765 ret2
= ext4_journal_stop(handle
);
1769 if (pos
+ len
> inode
->i_size
) {
1770 ext4_truncate_failed_write(inode
);
1772 * If truncate failed early the inode might still be
1773 * on the orphan list; we need to make sure the inode
1774 * is removed from the orphan list in that case.
1777 ext4_orphan_del(NULL
, inode
);
1780 return ret
? ret
: copied
;
1783 static int ext4_journalled_write_end(struct file
*file
,
1784 struct address_space
*mapping
,
1785 loff_t pos
, unsigned len
, unsigned copied
,
1786 struct page
*page
, void *fsdata
)
1788 handle_t
*handle
= ext4_journal_current_handle();
1789 struct inode
*inode
= mapping
->host
;
1795 trace_ext4_journalled_write_end(inode
, pos
, len
, copied
);
1796 from
= pos
& (PAGE_CACHE_SIZE
- 1);
1800 if (!PageUptodate(page
))
1802 page_zero_new_buffers(page
, from
+copied
, to
);
1805 ret
= walk_page_buffers(handle
, page_buffers(page
), from
,
1806 to
, &partial
, write_end_fn
);
1808 SetPageUptodate(page
);
1809 new_i_size
= pos
+ copied
;
1810 if (new_i_size
> inode
->i_size
)
1811 i_size_write(inode
, pos
+copied
);
1812 ext4_set_inode_state(inode
, EXT4_STATE_JDATA
);
1813 if (new_i_size
> EXT4_I(inode
)->i_disksize
) {
1814 ext4_update_i_disksize(inode
, new_i_size
);
1815 ret2
= ext4_mark_inode_dirty(handle
, inode
);
1821 page_cache_release(page
);
1822 if (pos
+ len
> inode
->i_size
&& ext4_can_truncate(inode
))
1823 /* if we have allocated more blocks and copied
1824 * less. We will have blocks allocated outside
1825 * inode->i_size. So truncate them
1827 ext4_orphan_add(handle
, inode
);
1829 ret2
= ext4_journal_stop(handle
);
1832 if (pos
+ len
> inode
->i_size
) {
1833 ext4_truncate_failed_write(inode
);
1835 * If truncate failed early the inode might still be
1836 * on the orphan list; we need to make sure the inode
1837 * is removed from the orphan list in that case.
1840 ext4_orphan_del(NULL
, inode
);
1843 return ret
? ret
: copied
;
1847 * Reserve a single block located at lblock
1849 static int ext4_da_reserve_space(struct inode
*inode
, sector_t lblock
)
1852 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
1853 struct ext4_inode_info
*ei
= EXT4_I(inode
);
1854 unsigned long md_needed
;
1858 * recalculate the amount of metadata blocks to reserve
1859 * in order to allocate nrblocks
1860 * worse case is one extent per block
1863 spin_lock(&ei
->i_block_reservation_lock
);
1864 md_needed
= ext4_calc_metadata_amount(inode
, lblock
);
1865 trace_ext4_da_reserve_space(inode
, md_needed
);
1866 spin_unlock(&ei
->i_block_reservation_lock
);
1869 * We will charge metadata quota at writeout time; this saves
1870 * us from metadata over-estimation, though we may go over by
1871 * a small amount in the end. Here we just reserve for data.
1873 ret
= dquot_reserve_block(inode
, 1);
1877 * We do still charge estimated metadata to the sb though;
1878 * we cannot afford to run out of free blocks.
1880 if (ext4_claim_free_blocks(sbi
, md_needed
+ 1)) {
1881 dquot_release_reservation_block(inode
, 1);
1882 if (ext4_should_retry_alloc(inode
->i_sb
, &retries
)) {
1888 spin_lock(&ei
->i_block_reservation_lock
);
1889 ei
->i_reserved_data_blocks
++;
1890 ei
->i_reserved_meta_blocks
+= md_needed
;
1891 spin_unlock(&ei
->i_block_reservation_lock
);
1893 return 0; /* success */
1896 static void ext4_da_release_space(struct inode
*inode
, int to_free
)
1898 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
1899 struct ext4_inode_info
*ei
= EXT4_I(inode
);
1902 return; /* Nothing to release, exit */
1904 spin_lock(&EXT4_I(inode
)->i_block_reservation_lock
);
1906 if (unlikely(to_free
> ei
->i_reserved_data_blocks
)) {
1908 * if there aren't enough reserved blocks, then the
1909 * counter is messed up somewhere. Since this
1910 * function is called from invalidate page, it's
1911 * harmless to return without any action.
1913 ext4_msg(inode
->i_sb
, KERN_NOTICE
, "ext4_da_release_space: "
1914 "ino %lu, to_free %d with only %d reserved "
1915 "data blocks\n", inode
->i_ino
, to_free
,
1916 ei
->i_reserved_data_blocks
);
1918 to_free
= ei
->i_reserved_data_blocks
;
1920 ei
->i_reserved_data_blocks
-= to_free
;
1922 if (ei
->i_reserved_data_blocks
== 0) {
1924 * We can release all of the reserved metadata blocks
1925 * only when we have written all of the delayed
1926 * allocation blocks.
1928 percpu_counter_sub(&sbi
->s_dirtyblocks_counter
,
1929 ei
->i_reserved_meta_blocks
);
1930 ei
->i_reserved_meta_blocks
= 0;
1931 ei
->i_da_metadata_calc_len
= 0;
1934 /* update fs dirty data blocks counter */
1935 percpu_counter_sub(&sbi
->s_dirtyblocks_counter
, to_free
);
1937 spin_unlock(&EXT4_I(inode
)->i_block_reservation_lock
);
1939 dquot_release_reservation_block(inode
, to_free
);
1942 static void ext4_da_page_release_reservation(struct page
*page
,
1943 unsigned long offset
)
1946 struct buffer_head
*head
, *bh
;
1947 unsigned int curr_off
= 0;
1949 head
= page_buffers(page
);
1952 unsigned int next_off
= curr_off
+ bh
->b_size
;
1954 if ((offset
<= curr_off
) && (buffer_delay(bh
))) {
1956 clear_buffer_delay(bh
);
1958 curr_off
= next_off
;
1959 } while ((bh
= bh
->b_this_page
) != head
);
1960 ext4_da_release_space(page
->mapping
->host
, to_release
);
1964 * Delayed allocation stuff
1968 * mpage_da_submit_io - walks through extent of pages and try to write
1969 * them with writepage() call back
1971 * @mpd->inode: inode
1972 * @mpd->first_page: first page of the extent
1973 * @mpd->next_page: page after the last page of the extent
1975 * By the time mpage_da_submit_io() is called we expect all blocks
1976 * to be allocated. this may be wrong if allocation failed.
1978 * As pages are already locked by write_cache_pages(), we can't use it
1980 static int mpage_da_submit_io(struct mpage_da_data
*mpd
)
1983 struct pagevec pvec
;
1984 unsigned long index
, end
;
1985 int ret
= 0, err
, nr_pages
, i
;
1986 struct inode
*inode
= mpd
->inode
;
1987 struct address_space
*mapping
= inode
->i_mapping
;
1989 BUG_ON(mpd
->next_page
<= mpd
->first_page
);
1991 * We need to start from the first_page to the next_page - 1
1992 * to make sure we also write the mapped dirty buffer_heads.
1993 * If we look at mpd->b_blocknr we would only be looking
1994 * at the currently mapped buffer_heads.
1996 index
= mpd
->first_page
;
1997 end
= mpd
->next_page
- 1;
1999 pagevec_init(&pvec
, 0);
2000 while (index
<= end
) {
2001 nr_pages
= pagevec_lookup(&pvec
, mapping
, index
, PAGEVEC_SIZE
);
2004 for (i
= 0; i
< nr_pages
; i
++) {
2005 struct page
*page
= pvec
.pages
[i
];
2007 index
= page
->index
;
2012 BUG_ON(!PageLocked(page
));
2013 BUG_ON(PageWriteback(page
));
2015 pages_skipped
= mpd
->wbc
->pages_skipped
;
2016 err
= mapping
->a_ops
->writepage(page
, mpd
->wbc
);
2017 if (!err
&& (pages_skipped
== mpd
->wbc
->pages_skipped
))
2019 * have successfully written the page
2020 * without skipping the same
2022 mpd
->pages_written
++;
2024 * In error case, we have to continue because
2025 * remaining pages are still locked
2026 * XXX: unlock and re-dirty them?
2031 pagevec_release(&pvec
);
2037 * mpage_put_bnr_to_bhs - walk blocks and assign them actual numbers
2039 * @mpd->inode - inode to walk through
2040 * @exbh->b_blocknr - first block on a disk
2041 * @exbh->b_size - amount of space in bytes
2042 * @logical - first logical block to start assignment with
2044 * the function goes through all passed space and put actual disk
2045 * block numbers into buffer heads, dropping BH_Delay and BH_Unwritten
2047 static void mpage_put_bnr_to_bhs(struct mpage_da_data
*mpd
, sector_t logical
,
2048 struct buffer_head
*exbh
)
2050 struct inode
*inode
= mpd
->inode
;
2051 struct address_space
*mapping
= inode
->i_mapping
;
2052 int blocks
= exbh
->b_size
>> inode
->i_blkbits
;
2053 sector_t pblock
= exbh
->b_blocknr
, cur_logical
;
2054 struct buffer_head
*head
, *bh
;
2056 struct pagevec pvec
;
2059 index
= logical
>> (PAGE_CACHE_SHIFT
- inode
->i_blkbits
);
2060 end
= (logical
+ blocks
- 1) >> (PAGE_CACHE_SHIFT
- inode
->i_blkbits
);
2061 cur_logical
= index
<< (PAGE_CACHE_SHIFT
- inode
->i_blkbits
);
2063 pagevec_init(&pvec
, 0);
2065 while (index
<= end
) {
2066 /* XXX: optimize tail */
2067 nr_pages
= pagevec_lookup(&pvec
, mapping
, index
, PAGEVEC_SIZE
);
2070 for (i
= 0; i
< nr_pages
; i
++) {
2071 struct page
*page
= pvec
.pages
[i
];
2073 index
= page
->index
;
2078 BUG_ON(!PageLocked(page
));
2079 BUG_ON(PageWriteback(page
));
2080 BUG_ON(!page_has_buffers(page
));
2082 bh
= page_buffers(page
);
2085 /* skip blocks out of the range */
2087 if (cur_logical
>= logical
)
2090 } while ((bh
= bh
->b_this_page
) != head
);
2093 if (cur_logical
>= logical
+ blocks
)
2096 if (buffer_delay(bh
) ||
2097 buffer_unwritten(bh
)) {
2099 BUG_ON(bh
->b_bdev
!= inode
->i_sb
->s_bdev
);
2101 if (buffer_delay(bh
)) {
2102 clear_buffer_delay(bh
);
2103 bh
->b_blocknr
= pblock
;
2106 * unwritten already should have
2107 * blocknr assigned. Verify that
2109 clear_buffer_unwritten(bh
);
2110 BUG_ON(bh
->b_blocknr
!= pblock
);
2113 } else if (buffer_mapped(bh
))
2114 BUG_ON(bh
->b_blocknr
!= pblock
);
2116 if (buffer_uninit(exbh
))
2117 set_buffer_uninit(bh
);
2120 } while ((bh
= bh
->b_this_page
) != head
);
2122 pagevec_release(&pvec
);
2128 * __unmap_underlying_blocks - just a helper function to unmap
2129 * set of blocks described by @bh
2131 static inline void __unmap_underlying_blocks(struct inode
*inode
,
2132 struct buffer_head
*bh
)
2134 struct block_device
*bdev
= inode
->i_sb
->s_bdev
;
2137 blocks
= bh
->b_size
>> inode
->i_blkbits
;
2138 for (i
= 0; i
< blocks
; i
++)
2139 unmap_underlying_metadata(bdev
, bh
->b_blocknr
+ i
);
2142 static void ext4_da_block_invalidatepages(struct mpage_da_data
*mpd
,
2143 sector_t logical
, long blk_cnt
)
2147 struct pagevec pvec
;
2148 struct inode
*inode
= mpd
->inode
;
2149 struct address_space
*mapping
= inode
->i_mapping
;
2151 index
= logical
>> (PAGE_CACHE_SHIFT
- inode
->i_blkbits
);
2152 end
= (logical
+ blk_cnt
- 1) >>
2153 (PAGE_CACHE_SHIFT
- inode
->i_blkbits
);
2154 while (index
<= end
) {
2155 nr_pages
= pagevec_lookup(&pvec
, mapping
, index
, PAGEVEC_SIZE
);
2158 for (i
= 0; i
< nr_pages
; i
++) {
2159 struct page
*page
= pvec
.pages
[i
];
2160 if (page
->index
> end
)
2162 BUG_ON(!PageLocked(page
));
2163 BUG_ON(PageWriteback(page
));
2164 block_invalidatepage(page
, 0);
2165 ClearPageUptodate(page
);
2168 index
= pvec
.pages
[nr_pages
- 1]->index
+ 1;
2169 pagevec_release(&pvec
);
2174 static void ext4_print_free_blocks(struct inode
*inode
)
2176 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
2177 printk(KERN_CRIT
"Total free blocks count %lld\n",
2178 ext4_count_free_blocks(inode
->i_sb
));
2179 printk(KERN_CRIT
"Free/Dirty block details\n");
2180 printk(KERN_CRIT
"free_blocks=%lld\n",
2181 (long long) percpu_counter_sum(&sbi
->s_freeblocks_counter
));
2182 printk(KERN_CRIT
"dirty_blocks=%lld\n",
2183 (long long) percpu_counter_sum(&sbi
->s_dirtyblocks_counter
));
2184 printk(KERN_CRIT
"Block reservation details\n");
2185 printk(KERN_CRIT
"i_reserved_data_blocks=%u\n",
2186 EXT4_I(inode
)->i_reserved_data_blocks
);
2187 printk(KERN_CRIT
"i_reserved_meta_blocks=%u\n",
2188 EXT4_I(inode
)->i_reserved_meta_blocks
);
2193 * mpage_da_map_blocks - go through given space
2195 * @mpd - bh describing space
2197 * The function skips space we know is already mapped to disk blocks.
2200 static int mpage_da_map_blocks(struct mpage_da_data
*mpd
)
2202 int err
, blks
, get_blocks_flags
;
2203 struct buffer_head
new;
2204 sector_t next
= mpd
->b_blocknr
;
2205 unsigned max_blocks
= mpd
->b_size
>> mpd
->inode
->i_blkbits
;
2206 loff_t disksize
= EXT4_I(mpd
->inode
)->i_disksize
;
2207 handle_t
*handle
= NULL
;
2210 * We consider only non-mapped and non-allocated blocks
2212 if ((mpd
->b_state
& (1 << BH_Mapped
)) &&
2213 !(mpd
->b_state
& (1 << BH_Delay
)) &&
2214 !(mpd
->b_state
& (1 << BH_Unwritten
)))
2218 * If we didn't accumulate anything to write simply return
2223 handle
= ext4_journal_current_handle();
2227 * Call ext4_get_blocks() to allocate any delayed allocation
2228 * blocks, or to convert an uninitialized extent to be
2229 * initialized (in the case where we have written into
2230 * one or more preallocated blocks).
2232 * We pass in the magic EXT4_GET_BLOCKS_DELALLOC_RESERVE to
2233 * indicate that we are on the delayed allocation path. This
2234 * affects functions in many different parts of the allocation
2235 * call path. This flag exists primarily because we don't
2236 * want to change *many* call functions, so ext4_get_blocks()
2237 * will set the magic i_delalloc_reserved_flag once the
2238 * inode's allocation semaphore is taken.
2240 * If the blocks in questions were delalloc blocks, set
2241 * EXT4_GET_BLOCKS_DELALLOC_RESERVE so the delalloc accounting
2242 * variables are updated after the blocks have been allocated.
2245 get_blocks_flags
= EXT4_GET_BLOCKS_CREATE
;
2246 if (ext4_should_dioread_nolock(mpd
->inode
))
2247 get_blocks_flags
|= EXT4_GET_BLOCKS_IO_CREATE_EXT
;
2248 if (mpd
->b_state
& (1 << BH_Delay
))
2249 get_blocks_flags
|= EXT4_GET_BLOCKS_DELALLOC_RESERVE
;
2251 blks
= ext4_get_blocks(handle
, mpd
->inode
, next
, max_blocks
,
2252 &new, get_blocks_flags
);
2256 * If get block returns with error we simply
2257 * return. Later writepage will redirty the page and
2258 * writepages will find the dirty page again
2263 if (err
== -ENOSPC
&&
2264 ext4_count_free_blocks(mpd
->inode
->i_sb
)) {
2270 * get block failure will cause us to loop in
2271 * writepages, because a_ops->writepage won't be able
2272 * to make progress. The page will be redirtied by
2273 * writepage and writepages will again try to write
2276 ext4_msg(mpd
->inode
->i_sb
, KERN_CRIT
,
2277 "delayed block allocation failed for inode %lu at "
2278 "logical offset %llu with max blocks %zd with "
2279 "error %d", mpd
->inode
->i_ino
,
2280 (unsigned long long) next
,
2281 mpd
->b_size
>> mpd
->inode
->i_blkbits
, err
);
2282 printk(KERN_CRIT
"This should not happen!! "
2283 "Data will be lost\n");
2284 if (err
== -ENOSPC
) {
2285 ext4_print_free_blocks(mpd
->inode
);
2287 /* invalidate all the pages */
2288 ext4_da_block_invalidatepages(mpd
, next
,
2289 mpd
->b_size
>> mpd
->inode
->i_blkbits
);
2294 new.b_size
= (blks
<< mpd
->inode
->i_blkbits
);
2296 if (buffer_new(&new))
2297 __unmap_underlying_blocks(mpd
->inode
, &new);
2300 * If blocks are delayed marked, we need to
2301 * put actual blocknr and drop delayed bit
2303 if ((mpd
->b_state
& (1 << BH_Delay
)) ||
2304 (mpd
->b_state
& (1 << BH_Unwritten
)))
2305 mpage_put_bnr_to_bhs(mpd
, next
, &new);
2307 if (ext4_should_order_data(mpd
->inode
)) {
2308 err
= ext4_jbd2_file_inode(handle
, mpd
->inode
);
2314 * Update on-disk size along with block allocation.
2316 disksize
= ((loff_t
) next
+ blks
) << mpd
->inode
->i_blkbits
;
2317 if (disksize
> i_size_read(mpd
->inode
))
2318 disksize
= i_size_read(mpd
->inode
);
2319 if (disksize
> EXT4_I(mpd
->inode
)->i_disksize
) {
2320 ext4_update_i_disksize(mpd
->inode
, disksize
);
2321 return ext4_mark_inode_dirty(handle
, mpd
->inode
);
2327 #define BH_FLAGS ((1 << BH_Uptodate) | (1 << BH_Mapped) | \
2328 (1 << BH_Delay) | (1 << BH_Unwritten))
2331 * mpage_add_bh_to_extent - try to add one more block to extent of blocks
2333 * @mpd->lbh - extent of blocks
2334 * @logical - logical number of the block in the file
2335 * @bh - bh of the block (used to access block's state)
2337 * the function is used to collect contig. blocks in same state
2339 static void mpage_add_bh_to_extent(struct mpage_da_data
*mpd
,
2340 sector_t logical
, size_t b_size
,
2341 unsigned long b_state
)
2344 int nrblocks
= mpd
->b_size
>> mpd
->inode
->i_blkbits
;
2347 * XXX Don't go larger than mballoc is willing to allocate
2348 * This is a stopgap solution. We eventually need to fold
2349 * mpage_da_submit_io() into this function and then call
2350 * ext4_get_blocks() multiple times in a loop
2352 if (nrblocks
>= 8*1024*1024/mpd
->inode
->i_sb
->s_blocksize
)
2355 /* check if thereserved journal credits might overflow */
2356 if (!(EXT4_I(mpd
->inode
)->i_flags
& EXT4_EXTENTS_FL
)) {
2357 if (nrblocks
>= EXT4_MAX_TRANS_DATA
) {
2359 * With non-extent format we are limited by the journal
2360 * credit available. Total credit needed to insert
2361 * nrblocks contiguous blocks is dependent on the
2362 * nrblocks. So limit nrblocks.
2365 } else if ((nrblocks
+ (b_size
>> mpd
->inode
->i_blkbits
)) >
2366 EXT4_MAX_TRANS_DATA
) {
2368 * Adding the new buffer_head would make it cross the
2369 * allowed limit for which we have journal credit
2370 * reserved. So limit the new bh->b_size
2372 b_size
= (EXT4_MAX_TRANS_DATA
- nrblocks
) <<
2373 mpd
->inode
->i_blkbits
;
2374 /* we will do mpage_da_submit_io in the next loop */
2378 * First block in the extent
2380 if (mpd
->b_size
== 0) {
2381 mpd
->b_blocknr
= logical
;
2382 mpd
->b_size
= b_size
;
2383 mpd
->b_state
= b_state
& BH_FLAGS
;
2387 next
= mpd
->b_blocknr
+ nrblocks
;
2389 * Can we merge the block to our big extent?
2391 if (logical
== next
&& (b_state
& BH_FLAGS
) == mpd
->b_state
) {
2392 mpd
->b_size
+= b_size
;
2398 * We couldn't merge the block to our extent, so we
2399 * need to flush current extent and start new one
2401 if (mpage_da_map_blocks(mpd
) == 0)
2402 mpage_da_submit_io(mpd
);
2407 static int ext4_bh_delay_or_unwritten(handle_t
*handle
, struct buffer_head
*bh
)
2409 return (buffer_delay(bh
) || buffer_unwritten(bh
)) && buffer_dirty(bh
);
2413 * __mpage_da_writepage - finds extent of pages and blocks
2415 * @page: page to consider
2416 * @wbc: not used, we just follow rules
2419 * The function finds extents of pages and scan them for all blocks.
2421 static int __mpage_da_writepage(struct page
*page
,
2422 struct writeback_control
*wbc
, void *data
)
2424 struct mpage_da_data
*mpd
= data
;
2425 struct inode
*inode
= mpd
->inode
;
2426 struct buffer_head
*bh
, *head
;
2430 * Can we merge this page to current extent?
2432 if (mpd
->next_page
!= page
->index
) {
2434 * Nope, we can't. So, we map non-allocated blocks
2435 * and start IO on them using writepage()
2437 if (mpd
->next_page
!= mpd
->first_page
) {
2438 if (mpage_da_map_blocks(mpd
) == 0)
2439 mpage_da_submit_io(mpd
);
2441 * skip rest of the page in the page_vec
2444 redirty_page_for_writepage(wbc
, page
);
2446 return MPAGE_DA_EXTENT_TAIL
;
2450 * Start next extent of pages ...
2452 mpd
->first_page
= page
->index
;
2462 mpd
->next_page
= page
->index
+ 1;
2463 logical
= (sector_t
) page
->index
<<
2464 (PAGE_CACHE_SHIFT
- inode
->i_blkbits
);
2466 if (!page_has_buffers(page
)) {
2467 mpage_add_bh_to_extent(mpd
, logical
, PAGE_CACHE_SIZE
,
2468 (1 << BH_Dirty
) | (1 << BH_Uptodate
));
2470 return MPAGE_DA_EXTENT_TAIL
;
2473 * Page with regular buffer heads, just add all dirty ones
2475 head
= page_buffers(page
);
2478 BUG_ON(buffer_locked(bh
));
2480 * We need to try to allocate
2481 * unmapped blocks in the same page.
2482 * Otherwise we won't make progress
2483 * with the page in ext4_writepage
2485 if (ext4_bh_delay_or_unwritten(NULL
, bh
)) {
2486 mpage_add_bh_to_extent(mpd
, logical
,
2490 return MPAGE_DA_EXTENT_TAIL
;
2491 } else if (buffer_dirty(bh
) && (buffer_mapped(bh
))) {
2493 * mapped dirty buffer. We need to update
2494 * the b_state because we look at
2495 * b_state in mpage_da_map_blocks. We don't
2496 * update b_size because if we find an
2497 * unmapped buffer_head later we need to
2498 * use the b_state flag of that buffer_head.
2500 if (mpd
->b_size
== 0)
2501 mpd
->b_state
= bh
->b_state
& BH_FLAGS
;
2504 } while ((bh
= bh
->b_this_page
) != head
);
2511 * This is a special get_blocks_t callback which is used by
2512 * ext4_da_write_begin(). It will either return mapped block or
2513 * reserve space for a single block.
2515 * For delayed buffer_head we have BH_Mapped, BH_New, BH_Delay set.
2516 * We also have b_blocknr = -1 and b_bdev initialized properly
2518 * For unwritten buffer_head we have BH_Mapped, BH_New, BH_Unwritten set.
2519 * We also have b_blocknr = physicalblock mapping unwritten extent and b_bdev
2520 * initialized properly.
2522 static int ext4_da_get_block_prep(struct inode
*inode
, sector_t iblock
,
2523 struct buffer_head
*bh_result
, int create
)
2526 sector_t invalid_block
= ~((sector_t
) 0xffff);
2528 if (invalid_block
< ext4_blocks_count(EXT4_SB(inode
->i_sb
)->s_es
))
2531 BUG_ON(create
== 0);
2532 BUG_ON(bh_result
->b_size
!= inode
->i_sb
->s_blocksize
);
2535 * first, we need to know whether the block is allocated already
2536 * preallocated blocks are unmapped but should treated
2537 * the same as allocated blocks.
2539 ret
= ext4_get_blocks(NULL
, inode
, iblock
, 1, bh_result
, 0);
2540 if ((ret
== 0) && !buffer_delay(bh_result
)) {
2541 /* the block isn't (pre)allocated yet, let's reserve space */
2543 * XXX: __block_prepare_write() unmaps passed block,
2546 ret
= ext4_da_reserve_space(inode
, iblock
);
2548 /* not enough space to reserve */
2551 map_bh(bh_result
, inode
->i_sb
, invalid_block
);
2552 set_buffer_new(bh_result
);
2553 set_buffer_delay(bh_result
);
2554 } else if (ret
> 0) {
2555 bh_result
->b_size
= (ret
<< inode
->i_blkbits
);
2556 if (buffer_unwritten(bh_result
)) {
2557 /* A delayed write to unwritten bh should
2558 * be marked new and mapped. Mapped ensures
2559 * that we don't do get_block multiple times
2560 * when we write to the same offset and new
2561 * ensures that we do proper zero out for
2564 set_buffer_new(bh_result
);
2565 set_buffer_mapped(bh_result
);
2574 * This function is used as a standard get_block_t calback function
2575 * when there is no desire to allocate any blocks. It is used as a
2576 * callback function for block_prepare_write(), nobh_writepage(), and
2577 * block_write_full_page(). These functions should only try to map a
2578 * single block at a time.
2580 * Since this function doesn't do block allocations even if the caller
2581 * requests it by passing in create=1, it is critically important that
2582 * any caller checks to make sure that any buffer heads are returned
2583 * by this function are either all already mapped or marked for
2584 * delayed allocation before calling nobh_writepage() or
2585 * block_write_full_page(). Otherwise, b_blocknr could be left
2586 * unitialized, and the page write functions will be taken by
2589 static int noalloc_get_block_write(struct inode
*inode
, sector_t iblock
,
2590 struct buffer_head
*bh_result
, int create
)
2593 unsigned max_blocks
= bh_result
->b_size
>> inode
->i_blkbits
;
2595 BUG_ON(bh_result
->b_size
!= inode
->i_sb
->s_blocksize
);
2598 * we don't want to do block allocation in writepage
2599 * so call get_block_wrap with create = 0
2601 ret
= ext4_get_blocks(NULL
, inode
, iblock
, max_blocks
, bh_result
, 0);
2603 bh_result
->b_size
= (ret
<< inode
->i_blkbits
);
2609 static int bget_one(handle_t
*handle
, struct buffer_head
*bh
)
2615 static int bput_one(handle_t
*handle
, struct buffer_head
*bh
)
2621 static int __ext4_journalled_writepage(struct page
*page
,
2624 struct address_space
*mapping
= page
->mapping
;
2625 struct inode
*inode
= mapping
->host
;
2626 struct buffer_head
*page_bufs
;
2627 handle_t
*handle
= NULL
;
2631 page_bufs
= page_buffers(page
);
2633 walk_page_buffers(handle
, page_bufs
, 0, len
, NULL
, bget_one
);
2634 /* As soon as we unlock the page, it can go away, but we have
2635 * references to buffers so we are safe */
2638 handle
= ext4_journal_start(inode
, ext4_writepage_trans_blocks(inode
));
2639 if (IS_ERR(handle
)) {
2640 ret
= PTR_ERR(handle
);
2644 ret
= walk_page_buffers(handle
, page_bufs
, 0, len
, NULL
,
2645 do_journal_get_write_access
);
2647 err
= walk_page_buffers(handle
, page_bufs
, 0, len
, NULL
,
2651 err
= ext4_journal_stop(handle
);
2655 walk_page_buffers(handle
, page_bufs
, 0, len
, NULL
, bput_one
);
2656 ext4_set_inode_state(inode
, EXT4_STATE_JDATA
);
2661 static int ext4_set_bh_endio(struct buffer_head
*bh
, struct inode
*inode
);
2662 static void ext4_end_io_buffer_write(struct buffer_head
*bh
, int uptodate
);
2665 * Note that we don't need to start a transaction unless we're journaling data
2666 * because we should have holes filled from ext4_page_mkwrite(). We even don't
2667 * need to file the inode to the transaction's list in ordered mode because if
2668 * we are writing back data added by write(), the inode is already there and if
2669 * we are writing back data modified via mmap(), noone guarantees in which
2670 * transaction the data will hit the disk. In case we are journaling data, we
2671 * cannot start transaction directly because transaction start ranks above page
2672 * lock so we have to do some magic.
2674 * This function can get called via...
2675 * - ext4_da_writepages after taking page lock (have journal handle)
2676 * - journal_submit_inode_data_buffers (no journal handle)
2677 * - shrink_page_list via pdflush (no journal handle)
2678 * - grab_page_cache when doing write_begin (have journal handle)
2680 * We don't do any block allocation in this function. If we have page with
2681 * multiple blocks we need to write those buffer_heads that are mapped. This
2682 * is important for mmaped based write. So if we do with blocksize 1K
2683 * truncate(f, 1024);
2684 * a = mmap(f, 0, 4096);
2686 * truncate(f, 4096);
2687 * we have in the page first buffer_head mapped via page_mkwrite call back
2688 * but other bufer_heads would be unmapped but dirty(dirty done via the
2689 * do_wp_page). So writepage should write the first block. If we modify
2690 * the mmap area beyond 1024 we will again get a page_fault and the
2691 * page_mkwrite callback will do the block allocation and mark the
2692 * buffer_heads mapped.
2694 * We redirty the page if we have any buffer_heads that is either delay or
2695 * unwritten in the page.
2697 * We can get recursively called as show below.
2699 * ext4_writepage() -> kmalloc() -> __alloc_pages() -> page_launder() ->
2702 * But since we don't do any block allocation we should not deadlock.
2703 * Page also have the dirty flag cleared so we don't get recurive page_lock.
2705 static int ext4_writepage(struct page
*page
,
2706 struct writeback_control
*wbc
)
2711 struct buffer_head
*page_bufs
= NULL
;
2712 struct inode
*inode
= page
->mapping
->host
;
2714 trace_ext4_writepage(inode
, page
);
2715 size
= i_size_read(inode
);
2716 if (page
->index
== size
>> PAGE_CACHE_SHIFT
)
2717 len
= size
& ~PAGE_CACHE_MASK
;
2719 len
= PAGE_CACHE_SIZE
;
2721 if (page_has_buffers(page
)) {
2722 page_bufs
= page_buffers(page
);
2723 if (walk_page_buffers(NULL
, page_bufs
, 0, len
, NULL
,
2724 ext4_bh_delay_or_unwritten
)) {
2726 * We don't want to do block allocation
2727 * So redirty the page and return
2728 * We may reach here when we do a journal commit
2729 * via journal_submit_inode_data_buffers.
2730 * If we don't have mapping block we just ignore
2731 * them. We can also reach here via shrink_page_list
2733 redirty_page_for_writepage(wbc
, page
);
2739 * The test for page_has_buffers() is subtle:
2740 * We know the page is dirty but it lost buffers. That means
2741 * that at some moment in time after write_begin()/write_end()
2742 * has been called all buffers have been clean and thus they
2743 * must have been written at least once. So they are all
2744 * mapped and we can happily proceed with mapping them
2745 * and writing the page.
2747 * Try to initialize the buffer_heads and check whether
2748 * all are mapped and non delay. We don't want to
2749 * do block allocation here.
2751 ret
= block_prepare_write(page
, 0, len
,
2752 noalloc_get_block_write
);
2754 page_bufs
= page_buffers(page
);
2755 /* check whether all are mapped and non delay */
2756 if (walk_page_buffers(NULL
, page_bufs
, 0, len
, NULL
,
2757 ext4_bh_delay_or_unwritten
)) {
2758 redirty_page_for_writepage(wbc
, page
);
2764 * We can't do block allocation here
2765 * so just redity the page and unlock
2768 redirty_page_for_writepage(wbc
, page
);
2772 /* now mark the buffer_heads as dirty and uptodate */
2773 block_commit_write(page
, 0, len
);
2776 if (PageChecked(page
) && ext4_should_journal_data(inode
)) {
2778 * It's mmapped pagecache. Add buffers and journal it. There
2779 * doesn't seem much point in redirtying the page here.
2781 ClearPageChecked(page
);
2782 return __ext4_journalled_writepage(page
, len
);
2785 if (test_opt(inode
->i_sb
, NOBH
) && ext4_should_writeback_data(inode
))
2786 ret
= nobh_writepage(page
, noalloc_get_block_write
, wbc
);
2787 else if (page_bufs
&& buffer_uninit(page_bufs
)) {
2788 ext4_set_bh_endio(page_bufs
, inode
);
2789 ret
= block_write_full_page_endio(page
, noalloc_get_block_write
,
2790 wbc
, ext4_end_io_buffer_write
);
2792 ret
= block_write_full_page(page
, noalloc_get_block_write
,
2799 * This is called via ext4_da_writepages() to
2800 * calulate the total number of credits to reserve to fit
2801 * a single extent allocation into a single transaction,
2802 * ext4_da_writpeages() will loop calling this before
2803 * the block allocation.
2806 static int ext4_da_writepages_trans_blocks(struct inode
*inode
)
2808 int max_blocks
= EXT4_I(inode
)->i_reserved_data_blocks
;
2811 * With non-extent format the journal credit needed to
2812 * insert nrblocks contiguous block is dependent on
2813 * number of contiguous block. So we will limit
2814 * number of contiguous block to a sane value
2816 if (!(EXT4_I(inode
)->i_flags
& EXT4_EXTENTS_FL
) &&
2817 (max_blocks
> EXT4_MAX_TRANS_DATA
))
2818 max_blocks
= EXT4_MAX_TRANS_DATA
;
2820 return ext4_chunk_trans_blocks(inode
, max_blocks
);
2824 * write_cache_pages_da - walk the list of dirty pages of the given
2825 * address space and call the callback function (which usually writes
2828 * This is a forked version of write_cache_pages(). Differences:
2829 * Range cyclic is ignored.
2830 * no_nrwrite_index_update is always presumed true
2832 static int write_cache_pages_da(struct address_space
*mapping
,
2833 struct writeback_control
*wbc
,
2834 struct mpage_da_data
*mpd
)
2838 struct pagevec pvec
;
2841 pgoff_t end
; /* Inclusive */
2842 long nr_to_write
= wbc
->nr_to_write
;
2844 pagevec_init(&pvec
, 0);
2845 index
= wbc
->range_start
>> PAGE_CACHE_SHIFT
;
2846 end
= wbc
->range_end
>> PAGE_CACHE_SHIFT
;
2848 while (!done
&& (index
<= end
)) {
2851 nr_pages
= pagevec_lookup_tag(&pvec
, mapping
, &index
,
2852 PAGECACHE_TAG_DIRTY
,
2853 min(end
- index
, (pgoff_t
)PAGEVEC_SIZE
-1) + 1);
2857 for (i
= 0; i
< nr_pages
; i
++) {
2858 struct page
*page
= pvec
.pages
[i
];
2861 * At this point, the page may be truncated or
2862 * invalidated (changing page->mapping to NULL), or
2863 * even swizzled back from swapper_space to tmpfs file
2864 * mapping. However, page->index will not change
2865 * because we have a reference on the page.
2867 if (page
->index
> end
) {
2875 * Page truncated or invalidated. We can freely skip it
2876 * then, even for data integrity operations: the page
2877 * has disappeared concurrently, so there could be no
2878 * real expectation of this data interity operation
2879 * even if there is now a new, dirty page at the same
2880 * pagecache address.
2882 if (unlikely(page
->mapping
!= mapping
)) {
2888 if (!PageDirty(page
)) {
2889 /* someone wrote it for us */
2890 goto continue_unlock
;
2893 if (PageWriteback(page
)) {
2894 if (wbc
->sync_mode
!= WB_SYNC_NONE
)
2895 wait_on_page_writeback(page
);
2897 goto continue_unlock
;
2900 BUG_ON(PageWriteback(page
));
2901 if (!clear_page_dirty_for_io(page
))
2902 goto continue_unlock
;
2904 ret
= __mpage_da_writepage(page
, wbc
, mpd
);
2905 if (unlikely(ret
)) {
2906 if (ret
== AOP_WRITEPAGE_ACTIVATE
) {
2915 if (nr_to_write
> 0) {
2917 if (nr_to_write
== 0 &&
2918 wbc
->sync_mode
== WB_SYNC_NONE
) {
2920 * We stop writing back only if we are
2921 * not doing integrity sync. In case of
2922 * integrity sync we have to keep going
2923 * because someone may be concurrently
2924 * dirtying pages, and we might have
2925 * synced a lot of newly appeared dirty
2926 * pages, but have not synced all of the
2934 pagevec_release(&pvec
);
2941 static int ext4_da_writepages(struct address_space
*mapping
,
2942 struct writeback_control
*wbc
)
2945 int range_whole
= 0;
2946 handle_t
*handle
= NULL
;
2947 struct mpage_da_data mpd
;
2948 struct inode
*inode
= mapping
->host
;
2949 int pages_written
= 0;
2951 unsigned int max_pages
;
2952 int range_cyclic
, cycled
= 1, io_done
= 0;
2953 int needed_blocks
, ret
= 0;
2954 long desired_nr_to_write
, nr_to_writebump
= 0;
2955 loff_t range_start
= wbc
->range_start
;
2956 struct ext4_sb_info
*sbi
= EXT4_SB(mapping
->host
->i_sb
);
2958 trace_ext4_da_writepages(inode
, wbc
);
2961 * No pages to write? This is mainly a kludge to avoid starting
2962 * a transaction for special inodes like journal inode on last iput()
2963 * because that could violate lock ordering on umount
2965 if (!mapping
->nrpages
|| !mapping_tagged(mapping
, PAGECACHE_TAG_DIRTY
))
2969 * If the filesystem has aborted, it is read-only, so return
2970 * right away instead of dumping stack traces later on that
2971 * will obscure the real source of the problem. We test
2972 * EXT4_MF_FS_ABORTED instead of sb->s_flag's MS_RDONLY because
2973 * the latter could be true if the filesystem is mounted
2974 * read-only, and in that case, ext4_da_writepages should
2975 * *never* be called, so if that ever happens, we would want
2978 if (unlikely(sbi
->s_mount_flags
& EXT4_MF_FS_ABORTED
))
2981 if (wbc
->range_start
== 0 && wbc
->range_end
== LLONG_MAX
)
2984 range_cyclic
= wbc
->range_cyclic
;
2985 if (wbc
->range_cyclic
) {
2986 index
= mapping
->writeback_index
;
2989 wbc
->range_start
= index
<< PAGE_CACHE_SHIFT
;
2990 wbc
->range_end
= LLONG_MAX
;
2991 wbc
->range_cyclic
= 0;
2993 index
= wbc
->range_start
>> PAGE_CACHE_SHIFT
;
2996 * This works around two forms of stupidity. The first is in
2997 * the writeback code, which caps the maximum number of pages
2998 * written to be 1024 pages. This is wrong on multiple
2999 * levels; different architectues have a different page size,
3000 * which changes the maximum amount of data which gets
3001 * written. Secondly, 4 megabytes is way too small. XFS
3002 * forces this value to be 16 megabytes by multiplying
3003 * nr_to_write parameter by four, and then relies on its
3004 * allocator to allocate larger extents to make them
3005 * contiguous. Unfortunately this brings us to the second
3006 * stupidity, which is that ext4's mballoc code only allocates
3007 * at most 2048 blocks. So we force contiguous writes up to
3008 * the number of dirty blocks in the inode, or
3009 * sbi->max_writeback_mb_bump whichever is smaller.
3011 max_pages
= sbi
->s_max_writeback_mb_bump
<< (20 - PAGE_CACHE_SHIFT
);
3012 if (!range_cyclic
&& range_whole
)
3013 desired_nr_to_write
= wbc
->nr_to_write
* 8;
3015 desired_nr_to_write
= ext4_num_dirty_pages(inode
, index
,
3017 if (desired_nr_to_write
> max_pages
)
3018 desired_nr_to_write
= max_pages
;
3020 if (wbc
->nr_to_write
< desired_nr_to_write
) {
3021 nr_to_writebump
= desired_nr_to_write
- wbc
->nr_to_write
;
3022 wbc
->nr_to_write
= desired_nr_to_write
;
3026 mpd
.inode
= mapping
->host
;
3028 pages_skipped
= wbc
->pages_skipped
;
3031 while (!ret
&& wbc
->nr_to_write
> 0) {
3034 * we insert one extent at a time. So we need
3035 * credit needed for single extent allocation.
3036 * journalled mode is currently not supported
3039 BUG_ON(ext4_should_journal_data(inode
));
3040 needed_blocks
= ext4_da_writepages_trans_blocks(inode
);
3042 /* start a new transaction*/
3043 handle
= ext4_journal_start(inode
, needed_blocks
);
3044 if (IS_ERR(handle
)) {
3045 ret
= PTR_ERR(handle
);
3046 ext4_msg(inode
->i_sb
, KERN_CRIT
, "%s: jbd2_start: "
3047 "%ld pages, ino %lu; err %d", __func__
,
3048 wbc
->nr_to_write
, inode
->i_ino
, ret
);
3049 goto out_writepages
;
3053 * Now call __mpage_da_writepage to find the next
3054 * contiguous region of logical blocks that need
3055 * blocks to be allocated by ext4. We don't actually
3056 * submit the blocks for I/O here, even though
3057 * write_cache_pages thinks it will, and will set the
3058 * pages as clean for write before calling
3059 * __mpage_da_writepage().
3067 mpd
.pages_written
= 0;
3069 ret
= write_cache_pages_da(mapping
, wbc
, &mpd
);
3071 * If we have a contiguous extent of pages and we
3072 * haven't done the I/O yet, map the blocks and submit
3075 if (!mpd
.io_done
&& mpd
.next_page
!= mpd
.first_page
) {
3076 if (mpage_da_map_blocks(&mpd
) == 0)
3077 mpage_da_submit_io(&mpd
);
3079 ret
= MPAGE_DA_EXTENT_TAIL
;
3081 trace_ext4_da_write_pages(inode
, &mpd
);
3082 wbc
->nr_to_write
-= mpd
.pages_written
;
3084 ext4_journal_stop(handle
);
3086 if ((mpd
.retval
== -ENOSPC
) && sbi
->s_journal
) {
3087 /* commit the transaction which would
3088 * free blocks released in the transaction
3091 jbd2_journal_force_commit_nested(sbi
->s_journal
);
3092 wbc
->pages_skipped
= pages_skipped
;
3094 } else if (ret
== MPAGE_DA_EXTENT_TAIL
) {
3096 * got one extent now try with
3099 pages_written
+= mpd
.pages_written
;
3100 wbc
->pages_skipped
= pages_skipped
;
3103 } else if (wbc
->nr_to_write
)
3105 * There is no more writeout needed
3106 * or we requested for a noblocking writeout
3107 * and we found the device congested
3111 if (!io_done
&& !cycled
) {
3114 wbc
->range_start
= index
<< PAGE_CACHE_SHIFT
;
3115 wbc
->range_end
= mapping
->writeback_index
- 1;
3118 if (pages_skipped
!= wbc
->pages_skipped
)
3119 ext4_msg(inode
->i_sb
, KERN_CRIT
,
3120 "This should not happen leaving %s "
3121 "with nr_to_write = %ld ret = %d",
3122 __func__
, wbc
->nr_to_write
, ret
);
3125 index
+= pages_written
;
3126 wbc
->range_cyclic
= range_cyclic
;
3127 if (wbc
->range_cyclic
|| (range_whole
&& wbc
->nr_to_write
> 0))
3129 * set the writeback_index so that range_cyclic
3130 * mode will write it back later
3132 mapping
->writeback_index
= index
;
3135 wbc
->nr_to_write
-= nr_to_writebump
;
3136 wbc
->range_start
= range_start
;
3137 trace_ext4_da_writepages_result(inode
, wbc
, ret
, pages_written
);
3141 #define FALL_BACK_TO_NONDELALLOC 1
3142 static int ext4_nonda_switch(struct super_block
*sb
)
3144 s64 free_blocks
, dirty_blocks
;
3145 struct ext4_sb_info
*sbi
= EXT4_SB(sb
);
3148 * switch to non delalloc mode if we are running low
3149 * on free block. The free block accounting via percpu
3150 * counters can get slightly wrong with percpu_counter_batch getting
3151 * accumulated on each CPU without updating global counters
3152 * Delalloc need an accurate free block accounting. So switch
3153 * to non delalloc when we are near to error range.
3155 free_blocks
= percpu_counter_read_positive(&sbi
->s_freeblocks_counter
);
3156 dirty_blocks
= percpu_counter_read_positive(&sbi
->s_dirtyblocks_counter
);
3157 if (2 * free_blocks
< 3 * dirty_blocks
||
3158 free_blocks
< (dirty_blocks
+ EXT4_FREEBLOCKS_WATERMARK
)) {
3160 * free block count is less than 150% of dirty blocks
3161 * or free blocks is less than watermark
3166 * Even if we don't switch but are nearing capacity,
3167 * start pushing delalloc when 1/2 of free blocks are dirty.
3169 if (free_blocks
< 2 * dirty_blocks
)
3170 writeback_inodes_sb_if_idle(sb
);
3175 static int ext4_da_write_begin(struct file
*file
, struct address_space
*mapping
,
3176 loff_t pos
, unsigned len
, unsigned flags
,
3177 struct page
**pagep
, void **fsdata
)
3179 int ret
, retries
= 0;
3183 struct inode
*inode
= mapping
->host
;
3186 index
= pos
>> PAGE_CACHE_SHIFT
;
3187 from
= pos
& (PAGE_CACHE_SIZE
- 1);
3190 if (ext4_nonda_switch(inode
->i_sb
)) {
3191 *fsdata
= (void *)FALL_BACK_TO_NONDELALLOC
;
3192 return ext4_write_begin(file
, mapping
, pos
,
3193 len
, flags
, pagep
, fsdata
);
3195 *fsdata
= (void *)0;
3196 trace_ext4_da_write_begin(inode
, pos
, len
, flags
);
3199 * With delayed allocation, we don't log the i_disksize update
3200 * if there is delayed block allocation. But we still need
3201 * to journalling the i_disksize update if writes to the end
3202 * of file which has an already mapped buffer.
3204 handle
= ext4_journal_start(inode
, 1);
3205 if (IS_ERR(handle
)) {
3206 ret
= PTR_ERR(handle
);
3209 /* We cannot recurse into the filesystem as the transaction is already
3211 flags
|= AOP_FLAG_NOFS
;
3213 page
= grab_cache_page_write_begin(mapping
, index
, flags
);
3215 ext4_journal_stop(handle
);
3221 ret
= block_write_begin(file
, mapping
, pos
, len
, flags
, pagep
, fsdata
,
3222 ext4_da_get_block_prep
);
3225 ext4_journal_stop(handle
);
3226 page_cache_release(page
);
3228 * block_write_begin may have instantiated a few blocks
3229 * outside i_size. Trim these off again. Don't need
3230 * i_size_read because we hold i_mutex.
3232 if (pos
+ len
> inode
->i_size
)
3233 ext4_truncate_failed_write(inode
);
3236 if (ret
== -ENOSPC
&& ext4_should_retry_alloc(inode
->i_sb
, &retries
))
3243 * Check if we should update i_disksize
3244 * when write to the end of file but not require block allocation
3246 static int ext4_da_should_update_i_disksize(struct page
*page
,
3247 unsigned long offset
)
3249 struct buffer_head
*bh
;
3250 struct inode
*inode
= page
->mapping
->host
;
3254 bh
= page_buffers(page
);
3255 idx
= offset
>> inode
->i_blkbits
;
3257 for (i
= 0; i
< idx
; i
++)
3258 bh
= bh
->b_this_page
;
3260 if (!buffer_mapped(bh
) || (buffer_delay(bh
)) || buffer_unwritten(bh
))
3265 static int ext4_da_write_end(struct file
*file
,
3266 struct address_space
*mapping
,
3267 loff_t pos
, unsigned len
, unsigned copied
,
3268 struct page
*page
, void *fsdata
)
3270 struct inode
*inode
= mapping
->host
;
3272 handle_t
*handle
= ext4_journal_current_handle();
3274 unsigned long start
, end
;
3275 int write_mode
= (int)(unsigned long)fsdata
;
3277 if (write_mode
== FALL_BACK_TO_NONDELALLOC
) {
3278 if (ext4_should_order_data(inode
)) {
3279 return ext4_ordered_write_end(file
, mapping
, pos
,
3280 len
, copied
, page
, fsdata
);
3281 } else if (ext4_should_writeback_data(inode
)) {
3282 return ext4_writeback_write_end(file
, mapping
, pos
,
3283 len
, copied
, page
, fsdata
);
3289 trace_ext4_da_write_end(inode
, pos
, len
, copied
);
3290 start
= pos
& (PAGE_CACHE_SIZE
- 1);
3291 end
= start
+ copied
- 1;
3294 * generic_write_end() will run mark_inode_dirty() if i_size
3295 * changes. So let's piggyback the i_disksize mark_inode_dirty
3299 new_i_size
= pos
+ copied
;
3300 if (new_i_size
> EXT4_I(inode
)->i_disksize
) {
3301 if (ext4_da_should_update_i_disksize(page
, end
)) {
3302 down_write(&EXT4_I(inode
)->i_data_sem
);
3303 if (new_i_size
> EXT4_I(inode
)->i_disksize
) {
3305 * Updating i_disksize when extending file
3306 * without needing block allocation
3308 if (ext4_should_order_data(inode
))
3309 ret
= ext4_jbd2_file_inode(handle
,
3312 EXT4_I(inode
)->i_disksize
= new_i_size
;
3314 up_write(&EXT4_I(inode
)->i_data_sem
);
3315 /* We need to mark inode dirty even if
3316 * new_i_size is less that inode->i_size
3317 * bu greater than i_disksize.(hint delalloc)
3319 ext4_mark_inode_dirty(handle
, inode
);
3322 ret2
= generic_write_end(file
, mapping
, pos
, len
, copied
,
3327 ret2
= ext4_journal_stop(handle
);
3331 return ret
? ret
: copied
;
3334 static void ext4_da_invalidatepage(struct page
*page
, unsigned long offset
)
3337 * Drop reserved blocks
3339 BUG_ON(!PageLocked(page
));
3340 if (!page_has_buffers(page
))
3343 ext4_da_page_release_reservation(page
, offset
);
3346 ext4_invalidatepage(page
, offset
);
3352 * Force all delayed allocation blocks to be allocated for a given inode.
3354 int ext4_alloc_da_blocks(struct inode
*inode
)
3356 trace_ext4_alloc_da_blocks(inode
);
3358 if (!EXT4_I(inode
)->i_reserved_data_blocks
&&
3359 !EXT4_I(inode
)->i_reserved_meta_blocks
)
3363 * We do something simple for now. The filemap_flush() will
3364 * also start triggering a write of the data blocks, which is
3365 * not strictly speaking necessary (and for users of
3366 * laptop_mode, not even desirable). However, to do otherwise
3367 * would require replicating code paths in:
3369 * ext4_da_writepages() ->
3370 * write_cache_pages() ---> (via passed in callback function)
3371 * __mpage_da_writepage() -->
3372 * mpage_add_bh_to_extent()
3373 * mpage_da_map_blocks()
3375 * The problem is that write_cache_pages(), located in
3376 * mm/page-writeback.c, marks pages clean in preparation for
3377 * doing I/O, which is not desirable if we're not planning on
3380 * We could call write_cache_pages(), and then redirty all of
3381 * the pages by calling redirty_page_for_writeback() but that
3382 * would be ugly in the extreme. So instead we would need to
3383 * replicate parts of the code in the above functions,
3384 * simplifying them becuase we wouldn't actually intend to
3385 * write out the pages, but rather only collect contiguous
3386 * logical block extents, call the multi-block allocator, and
3387 * then update the buffer heads with the block allocations.
3389 * For now, though, we'll cheat by calling filemap_flush(),
3390 * which will map the blocks, and start the I/O, but not
3391 * actually wait for the I/O to complete.
3393 return filemap_flush(inode
->i_mapping
);
3397 * bmap() is special. It gets used by applications such as lilo and by
3398 * the swapper to find the on-disk block of a specific piece of data.
3400 * Naturally, this is dangerous if the block concerned is still in the
3401 * journal. If somebody makes a swapfile on an ext4 data-journaling
3402 * filesystem and enables swap, then they may get a nasty shock when the
3403 * data getting swapped to that swapfile suddenly gets overwritten by
3404 * the original zero's written out previously to the journal and
3405 * awaiting writeback in the kernel's buffer cache.
3407 * So, if we see any bmap calls here on a modified, data-journaled file,
3408 * take extra steps to flush any blocks which might be in the cache.
3410 static sector_t
ext4_bmap(struct address_space
*mapping
, sector_t block
)
3412 struct inode
*inode
= mapping
->host
;
3416 if (mapping_tagged(mapping
, PAGECACHE_TAG_DIRTY
) &&
3417 test_opt(inode
->i_sb
, DELALLOC
)) {
3419 * With delalloc we want to sync the file
3420 * so that we can make sure we allocate
3423 filemap_write_and_wait(mapping
);
3426 if (EXT4_JOURNAL(inode
) &&
3427 ext4_test_inode_state(inode
, EXT4_STATE_JDATA
)) {
3429 * This is a REALLY heavyweight approach, but the use of
3430 * bmap on dirty files is expected to be extremely rare:
3431 * only if we run lilo or swapon on a freshly made file
3432 * do we expect this to happen.
3434 * (bmap requires CAP_SYS_RAWIO so this does not
3435 * represent an unprivileged user DOS attack --- we'd be
3436 * in trouble if mortal users could trigger this path at
3439 * NB. EXT4_STATE_JDATA is not set on files other than
3440 * regular files. If somebody wants to bmap a directory
3441 * or symlink and gets confused because the buffer
3442 * hasn't yet been flushed to disk, they deserve
3443 * everything they get.
3446 ext4_clear_inode_state(inode
, EXT4_STATE_JDATA
);
3447 journal
= EXT4_JOURNAL(inode
);
3448 jbd2_journal_lock_updates(journal
);
3449 err
= jbd2_journal_flush(journal
);
3450 jbd2_journal_unlock_updates(journal
);
3456 return generic_block_bmap(mapping
, block
, ext4_get_block
);
3459 static int ext4_readpage(struct file
*file
, struct page
*page
)
3461 return mpage_readpage(page
, ext4_get_block
);
3465 ext4_readpages(struct file
*file
, struct address_space
*mapping
,
3466 struct list_head
*pages
, unsigned nr_pages
)
3468 return mpage_readpages(mapping
, pages
, nr_pages
, ext4_get_block
);
3471 static void ext4_free_io_end(ext4_io_end_t
*io
)
3480 static void ext4_invalidatepage_free_endio(struct page
*page
, unsigned long offset
)
3482 struct buffer_head
*head
, *bh
;
3483 unsigned int curr_off
= 0;
3485 if (!page_has_buffers(page
))
3487 head
= bh
= page_buffers(page
);
3489 if (offset
<= curr_off
&& test_clear_buffer_uninit(bh
)
3491 ext4_free_io_end(bh
->b_private
);
3492 bh
->b_private
= NULL
;
3493 bh
->b_end_io
= NULL
;
3495 curr_off
= curr_off
+ bh
->b_size
;
3496 bh
= bh
->b_this_page
;
3497 } while (bh
!= head
);
3500 static void ext4_invalidatepage(struct page
*page
, unsigned long offset
)
3502 journal_t
*journal
= EXT4_JOURNAL(page
->mapping
->host
);
3505 * free any io_end structure allocated for buffers to be discarded
3507 if (ext4_should_dioread_nolock(page
->mapping
->host
))
3508 ext4_invalidatepage_free_endio(page
, offset
);
3510 * If it's a full truncate we just forget about the pending dirtying
3513 ClearPageChecked(page
);
3516 jbd2_journal_invalidatepage(journal
, page
, offset
);
3518 block_invalidatepage(page
, offset
);
3521 static int ext4_releasepage(struct page
*page
, gfp_t wait
)
3523 journal_t
*journal
= EXT4_JOURNAL(page
->mapping
->host
);
3525 WARN_ON(PageChecked(page
));
3526 if (!page_has_buffers(page
))
3529 return jbd2_journal_try_to_free_buffers(journal
, page
, wait
);
3531 return try_to_free_buffers(page
);
3535 * O_DIRECT for ext3 (or indirect map) based files
3537 * If the O_DIRECT write will extend the file then add this inode to the
3538 * orphan list. So recovery will truncate it back to the original size
3539 * if the machine crashes during the write.
3541 * If the O_DIRECT write is intantiating holes inside i_size and the machine
3542 * crashes then stale disk data _may_ be exposed inside the file. But current
3543 * VFS code falls back into buffered path in that case so we are safe.
3545 static ssize_t
ext4_ind_direct_IO(int rw
, struct kiocb
*iocb
,
3546 const struct iovec
*iov
, loff_t offset
,
3547 unsigned long nr_segs
)
3549 struct file
*file
= iocb
->ki_filp
;
3550 struct inode
*inode
= file
->f_mapping
->host
;
3551 struct ext4_inode_info
*ei
= EXT4_I(inode
);
3555 size_t count
= iov_length(iov
, nr_segs
);
3559 loff_t final_size
= offset
+ count
;
3561 if (final_size
> inode
->i_size
) {
3562 /* Credits for sb + inode write */
3563 handle
= ext4_journal_start(inode
, 2);
3564 if (IS_ERR(handle
)) {
3565 ret
= PTR_ERR(handle
);
3568 ret
= ext4_orphan_add(handle
, inode
);
3570 ext4_journal_stop(handle
);
3574 ei
->i_disksize
= inode
->i_size
;
3575 ext4_journal_stop(handle
);
3580 if (rw
== READ
&& ext4_should_dioread_nolock(inode
))
3581 ret
= blockdev_direct_IO_no_locking(rw
, iocb
, inode
,
3582 inode
->i_sb
->s_bdev
, iov
,
3584 ext4_get_block
, NULL
);
3586 ret
= blockdev_direct_IO(rw
, iocb
, inode
,
3587 inode
->i_sb
->s_bdev
, iov
,
3589 ext4_get_block
, NULL
);
3590 if (ret
== -ENOSPC
&& ext4_should_retry_alloc(inode
->i_sb
, &retries
))
3596 /* Credits for sb + inode write */
3597 handle
= ext4_journal_start(inode
, 2);
3598 if (IS_ERR(handle
)) {
3599 /* This is really bad luck. We've written the data
3600 * but cannot extend i_size. Bail out and pretend
3601 * the write failed... */
3602 ret
= PTR_ERR(handle
);
3604 ext4_orphan_del(NULL
, inode
);
3609 ext4_orphan_del(handle
, inode
);
3611 loff_t end
= offset
+ ret
;
3612 if (end
> inode
->i_size
) {
3613 ei
->i_disksize
= end
;
3614 i_size_write(inode
, end
);
3616 * We're going to return a positive `ret'
3617 * here due to non-zero-length I/O, so there's
3618 * no way of reporting error returns from
3619 * ext4_mark_inode_dirty() to userspace. So
3622 ext4_mark_inode_dirty(handle
, inode
);
3625 err
= ext4_journal_stop(handle
);
3633 static int ext4_get_block_write(struct inode
*inode
, sector_t iblock
,
3634 struct buffer_head
*bh_result
, int create
)
3636 handle_t
*handle
= ext4_journal_current_handle();
3638 unsigned max_blocks
= bh_result
->b_size
>> inode
->i_blkbits
;
3642 ext4_debug("ext4_get_block_write: inode %lu, create flag %d\n",
3643 inode
->i_ino
, create
);
3645 * ext4_get_block in prepare for a DIO write or buffer write.
3646 * We allocate an uinitialized extent if blocks haven't been allocated.
3647 * The extent will be converted to initialized after IO complete.
3649 create
= EXT4_GET_BLOCKS_IO_CREATE_EXT
;
3652 if (max_blocks
> DIO_MAX_BLOCKS
)
3653 max_blocks
= DIO_MAX_BLOCKS
;
3654 dio_credits
= ext4_chunk_trans_blocks(inode
, max_blocks
);
3655 handle
= ext4_journal_start(inode
, dio_credits
);
3656 if (IS_ERR(handle
)) {
3657 ret
= PTR_ERR(handle
);
3663 ret
= ext4_get_blocks(handle
, inode
, iblock
, max_blocks
, bh_result
,
3666 bh_result
->b_size
= (ret
<< inode
->i_blkbits
);
3670 ext4_journal_stop(handle
);
3675 static void dump_completed_IO(struct inode
* inode
)
3678 struct list_head
*cur
, *before
, *after
;
3679 ext4_io_end_t
*io
, *io0
, *io1
;
3680 unsigned long flags
;
3682 if (list_empty(&EXT4_I(inode
)->i_completed_io_list
)){
3683 ext4_debug("inode %lu completed_io list is empty\n", inode
->i_ino
);
3687 ext4_debug("Dump inode %lu completed_io list \n", inode
->i_ino
);
3688 spin_lock_irqsave(&EXT4_I(inode
)->i_completed_io_lock
, flags
);
3689 list_for_each_entry(io
, &EXT4_I(inode
)->i_completed_io_list
, list
){
3692 io0
= container_of(before
, ext4_io_end_t
, list
);
3694 io1
= container_of(after
, ext4_io_end_t
, list
);
3696 ext4_debug("io 0x%p from inode %lu,prev 0x%p,next 0x%p\n",
3697 io
, inode
->i_ino
, io0
, io1
);
3699 spin_unlock_irqrestore(&EXT4_I(inode
)->i_completed_io_lock
, flags
);
3704 * check a range of space and convert unwritten extents to written.
3706 static int ext4_end_io_nolock(ext4_io_end_t
*io
)
3708 struct inode
*inode
= io
->inode
;
3709 loff_t offset
= io
->offset
;
3710 ssize_t size
= io
->size
;
3713 ext4_debug("ext4_end_io_nolock: io 0x%p from inode %lu,list->next 0x%p,"
3714 "list->prev 0x%p\n",
3715 io
, inode
->i_ino
, io
->list
.next
, io
->list
.prev
);
3717 if (list_empty(&io
->list
))
3720 if (io
->flag
!= EXT4_IO_UNWRITTEN
)
3723 ret
= ext4_convert_unwritten_extents(inode
, offset
, size
);
3725 printk(KERN_EMERG
"%s: failed to convert unwritten"
3726 "extents to written extents, error is %d"
3727 " io is still on inode %lu aio dio list\n",
3728 __func__
, ret
, inode
->i_ino
);
3732 /* clear the DIO AIO unwritten flag */
3738 * work on completed aio dio IO, to convert unwritten extents to extents
3740 static void ext4_end_io_work(struct work_struct
*work
)
3742 ext4_io_end_t
*io
= container_of(work
, ext4_io_end_t
, work
);
3743 struct inode
*inode
= io
->inode
;
3744 struct ext4_inode_info
*ei
= EXT4_I(inode
);
3745 unsigned long flags
;
3748 mutex_lock(&inode
->i_mutex
);
3749 ret
= ext4_end_io_nolock(io
);
3751 mutex_unlock(&inode
->i_mutex
);
3755 spin_lock_irqsave(&ei
->i_completed_io_lock
, flags
);
3756 if (!list_empty(&io
->list
))
3757 list_del_init(&io
->list
);
3758 spin_unlock_irqrestore(&ei
->i_completed_io_lock
, flags
);
3759 mutex_unlock(&inode
->i_mutex
);
3760 ext4_free_io_end(io
);
3764 * This function is called from ext4_sync_file().
3766 * When IO is completed, the work to convert unwritten extents to
3767 * written is queued on workqueue but may not get immediately
3768 * scheduled. When fsync is called, we need to ensure the
3769 * conversion is complete before fsync returns.
3770 * The inode keeps track of a list of pending/completed IO that
3771 * might needs to do the conversion. This function walks through
3772 * the list and convert the related unwritten extents for completed IO
3774 * The function return the number of pending IOs on success.
3776 int flush_completed_IO(struct inode
*inode
)
3779 struct ext4_inode_info
*ei
= EXT4_I(inode
);
3780 unsigned long flags
;
3784 if (list_empty(&ei
->i_completed_io_list
))
3787 dump_completed_IO(inode
);
3788 spin_lock_irqsave(&ei
->i_completed_io_lock
, flags
);
3789 while (!list_empty(&ei
->i_completed_io_list
)){
3790 io
= list_entry(ei
->i_completed_io_list
.next
,
3791 ext4_io_end_t
, list
);
3793 * Calling ext4_end_io_nolock() to convert completed
3796 * When ext4_sync_file() is called, run_queue() may already
3797 * about to flush the work corresponding to this io structure.
3798 * It will be upset if it founds the io structure related
3799 * to the work-to-be schedule is freed.
3801 * Thus we need to keep the io structure still valid here after
3802 * convertion finished. The io structure has a flag to
3803 * avoid double converting from both fsync and background work
3806 spin_unlock_irqrestore(&ei
->i_completed_io_lock
, flags
);
3807 ret
= ext4_end_io_nolock(io
);
3808 spin_lock_irqsave(&ei
->i_completed_io_lock
, flags
);
3812 list_del_init(&io
->list
);
3814 spin_unlock_irqrestore(&ei
->i_completed_io_lock
, flags
);
3815 return (ret2
< 0) ? ret2
: 0;
3818 static ext4_io_end_t
*ext4_init_io_end (struct inode
*inode
, gfp_t flags
)
3820 ext4_io_end_t
*io
= NULL
;
3822 io
= kmalloc(sizeof(*io
), flags
);
3831 INIT_WORK(&io
->work
, ext4_end_io_work
);
3832 INIT_LIST_HEAD(&io
->list
);
3838 static void ext4_end_io_dio(struct kiocb
*iocb
, loff_t offset
,
3839 ssize_t size
, void *private)
3841 ext4_io_end_t
*io_end
= iocb
->private;
3842 struct workqueue_struct
*wq
;
3843 unsigned long flags
;
3844 struct ext4_inode_info
*ei
;
3846 /* if not async direct IO or dio with 0 bytes write, just return */
3847 if (!io_end
|| !size
)
3850 ext_debug("ext4_end_io_dio(): io_end 0x%p"
3851 "for inode %lu, iocb 0x%p, offset %llu, size %llu\n",
3852 iocb
->private, io_end
->inode
->i_ino
, iocb
, offset
,
3855 /* if not aio dio with unwritten extents, just free io and return */
3856 if (io_end
->flag
!= EXT4_IO_UNWRITTEN
){
3857 ext4_free_io_end(io_end
);
3858 iocb
->private = NULL
;
3862 io_end
->offset
= offset
;
3863 io_end
->size
= size
;
3864 io_end
->flag
= EXT4_IO_UNWRITTEN
;
3865 wq
= EXT4_SB(io_end
->inode
->i_sb
)->dio_unwritten_wq
;
3867 /* queue the work to convert unwritten extents to written */
3868 queue_work(wq
, &io_end
->work
);
3870 /* Add the io_end to per-inode completed aio dio list*/
3871 ei
= EXT4_I(io_end
->inode
);
3872 spin_lock_irqsave(&ei
->i_completed_io_lock
, flags
);
3873 list_add_tail(&io_end
->list
, &ei
->i_completed_io_list
);
3874 spin_unlock_irqrestore(&ei
->i_completed_io_lock
, flags
);
3875 iocb
->private = NULL
;
3878 static void ext4_end_io_buffer_write(struct buffer_head
*bh
, int uptodate
)
3880 ext4_io_end_t
*io_end
= bh
->b_private
;
3881 struct workqueue_struct
*wq
;
3882 struct inode
*inode
;
3883 unsigned long flags
;
3885 if (!test_clear_buffer_uninit(bh
) || !io_end
)
3888 if (!(io_end
->inode
->i_sb
->s_flags
& MS_ACTIVE
)) {
3889 printk("sb umounted, discard end_io request for inode %lu\n",
3890 io_end
->inode
->i_ino
);
3891 ext4_free_io_end(io_end
);
3895 io_end
->flag
= EXT4_IO_UNWRITTEN
;
3896 inode
= io_end
->inode
;
3898 /* Add the io_end to per-inode completed io list*/
3899 spin_lock_irqsave(&EXT4_I(inode
)->i_completed_io_lock
, flags
);
3900 list_add_tail(&io_end
->list
, &EXT4_I(inode
)->i_completed_io_list
);
3901 spin_unlock_irqrestore(&EXT4_I(inode
)->i_completed_io_lock
, flags
);
3903 wq
= EXT4_SB(inode
->i_sb
)->dio_unwritten_wq
;
3904 /* queue the work to convert unwritten extents to written */
3905 queue_work(wq
, &io_end
->work
);
3907 bh
->b_private
= NULL
;
3908 bh
->b_end_io
= NULL
;
3909 clear_buffer_uninit(bh
);
3910 end_buffer_async_write(bh
, uptodate
);
3913 static int ext4_set_bh_endio(struct buffer_head
*bh
, struct inode
*inode
)
3915 ext4_io_end_t
*io_end
;
3916 struct page
*page
= bh
->b_page
;
3917 loff_t offset
= (sector_t
)page
->index
<< PAGE_CACHE_SHIFT
;
3918 size_t size
= bh
->b_size
;
3921 io_end
= ext4_init_io_end(inode
, GFP_ATOMIC
);
3923 if (printk_ratelimit())
3924 printk(KERN_WARNING
"%s: allocation fail\n", __func__
);
3928 io_end
->offset
= offset
;
3929 io_end
->size
= size
;
3931 * We need to hold a reference to the page to make sure it
3932 * doesn't get evicted before ext4_end_io_work() has a chance
3933 * to convert the extent from written to unwritten.
3935 io_end
->page
= page
;
3936 get_page(io_end
->page
);
3938 bh
->b_private
= io_end
;
3939 bh
->b_end_io
= ext4_end_io_buffer_write
;
3944 * For ext4 extent files, ext4 will do direct-io write to holes,
3945 * preallocated extents, and those write extend the file, no need to
3946 * fall back to buffered IO.
3948 * For holes, we fallocate those blocks, mark them as unintialized
3949 * If those blocks were preallocated, we mark sure they are splited, but
3950 * still keep the range to write as unintialized.
3952 * The unwrritten extents will be converted to written when DIO is completed.
3953 * For async direct IO, since the IO may still pending when return, we
3954 * set up an end_io call back function, which will do the convertion
3955 * when async direct IO completed.
3957 * If the O_DIRECT write will extend the file then add this inode to the
3958 * orphan list. So recovery will truncate it back to the original size
3959 * if the machine crashes during the write.
3962 static ssize_t
ext4_ext_direct_IO(int rw
, struct kiocb
*iocb
,
3963 const struct iovec
*iov
, loff_t offset
,
3964 unsigned long nr_segs
)
3966 struct file
*file
= iocb
->ki_filp
;
3967 struct inode
*inode
= file
->f_mapping
->host
;
3969 size_t count
= iov_length(iov
, nr_segs
);
3971 loff_t final_size
= offset
+ count
;
3972 if (rw
== WRITE
&& final_size
<= inode
->i_size
) {
3974 * We could direct write to holes and fallocate.
3976 * Allocated blocks to fill the hole are marked as uninitialized
3977 * to prevent paralel buffered read to expose the stale data
3978 * before DIO complete the data IO.
3980 * As to previously fallocated extents, ext4 get_block
3981 * will just simply mark the buffer mapped but still
3982 * keep the extents uninitialized.
3984 * for non AIO case, we will convert those unwritten extents
3985 * to written after return back from blockdev_direct_IO.
3987 * for async DIO, the conversion needs to be defered when
3988 * the IO is completed. The ext4 end_io callback function
3989 * will be called to take care of the conversion work.
3990 * Here for async case, we allocate an io_end structure to
3993 iocb
->private = NULL
;
3994 EXT4_I(inode
)->cur_aio_dio
= NULL
;
3995 if (!is_sync_kiocb(iocb
)) {
3996 iocb
->private = ext4_init_io_end(inode
, GFP_NOFS
);
4000 * we save the io structure for current async
4001 * direct IO, so that later ext4_get_blocks()
4002 * could flag the io structure whether there
4003 * is a unwritten extents needs to be converted
4004 * when IO is completed.
4006 EXT4_I(inode
)->cur_aio_dio
= iocb
->private;
4009 ret
= blockdev_direct_IO(rw
, iocb
, inode
,
4010 inode
->i_sb
->s_bdev
, iov
,
4012 ext4_get_block_write
,
4015 EXT4_I(inode
)->cur_aio_dio
= NULL
;
4017 * The io_end structure takes a reference to the inode,
4018 * that structure needs to be destroyed and the
4019 * reference to the inode need to be dropped, when IO is
4020 * complete, even with 0 byte write, or failed.
4022 * In the successful AIO DIO case, the io_end structure will be
4023 * desctroyed and the reference to the inode will be dropped
4024 * after the end_io call back function is called.
4026 * In the case there is 0 byte write, or error case, since
4027 * VFS direct IO won't invoke the end_io call back function,
4028 * we need to free the end_io structure here.
4030 if (ret
!= -EIOCBQUEUED
&& ret
<= 0 && iocb
->private) {
4031 ext4_free_io_end(iocb
->private);
4032 iocb
->private = NULL
;
4033 } else if (ret
> 0 && ext4_test_inode_state(inode
,
4034 EXT4_STATE_DIO_UNWRITTEN
)) {
4037 * for non AIO case, since the IO is already
4038 * completed, we could do the convertion right here
4040 err
= ext4_convert_unwritten_extents(inode
,
4044 ext4_clear_inode_state(inode
, EXT4_STATE_DIO_UNWRITTEN
);
4049 /* for write the the end of file case, we fall back to old way */
4050 return ext4_ind_direct_IO(rw
, iocb
, iov
, offset
, nr_segs
);
4053 static ssize_t
ext4_direct_IO(int rw
, struct kiocb
*iocb
,
4054 const struct iovec
*iov
, loff_t offset
,
4055 unsigned long nr_segs
)
4057 struct file
*file
= iocb
->ki_filp
;
4058 struct inode
*inode
= file
->f_mapping
->host
;
4060 if (EXT4_I(inode
)->i_flags
& EXT4_EXTENTS_FL
)
4061 return ext4_ext_direct_IO(rw
, iocb
, iov
, offset
, nr_segs
);
4063 return ext4_ind_direct_IO(rw
, iocb
, iov
, offset
, nr_segs
);
4067 * Pages can be marked dirty completely asynchronously from ext4's journalling
4068 * activity. By filemap_sync_pte(), try_to_unmap_one(), etc. We cannot do
4069 * much here because ->set_page_dirty is called under VFS locks. The page is
4070 * not necessarily locked.
4072 * We cannot just dirty the page and leave attached buffers clean, because the
4073 * buffers' dirty state is "definitive". We cannot just set the buffers dirty
4074 * or jbddirty because all the journalling code will explode.
4076 * So what we do is to mark the page "pending dirty" and next time writepage
4077 * is called, propagate that into the buffers appropriately.
4079 static int ext4_journalled_set_page_dirty(struct page
*page
)
4081 SetPageChecked(page
);
4082 return __set_page_dirty_nobuffers(page
);
4085 static const struct address_space_operations ext4_ordered_aops
= {
4086 .readpage
= ext4_readpage
,
4087 .readpages
= ext4_readpages
,
4088 .writepage
= ext4_writepage
,
4089 .sync_page
= block_sync_page
,
4090 .write_begin
= ext4_write_begin
,
4091 .write_end
= ext4_ordered_write_end
,
4093 .invalidatepage
= ext4_invalidatepage
,
4094 .releasepage
= ext4_releasepage
,
4095 .direct_IO
= ext4_direct_IO
,
4096 .migratepage
= buffer_migrate_page
,
4097 .is_partially_uptodate
= block_is_partially_uptodate
,
4098 .error_remove_page
= generic_error_remove_page
,
4101 static const struct address_space_operations ext4_writeback_aops
= {
4102 .readpage
= ext4_readpage
,
4103 .readpages
= ext4_readpages
,
4104 .writepage
= ext4_writepage
,
4105 .sync_page
= block_sync_page
,
4106 .write_begin
= ext4_write_begin
,
4107 .write_end
= ext4_writeback_write_end
,
4109 .invalidatepage
= ext4_invalidatepage
,
4110 .releasepage
= ext4_releasepage
,
4111 .direct_IO
= ext4_direct_IO
,
4112 .migratepage
= buffer_migrate_page
,
4113 .is_partially_uptodate
= block_is_partially_uptodate
,
4114 .error_remove_page
= generic_error_remove_page
,
4117 static const struct address_space_operations ext4_journalled_aops
= {
4118 .readpage
= ext4_readpage
,
4119 .readpages
= ext4_readpages
,
4120 .writepage
= ext4_writepage
,
4121 .sync_page
= block_sync_page
,
4122 .write_begin
= ext4_write_begin
,
4123 .write_end
= ext4_journalled_write_end
,
4124 .set_page_dirty
= ext4_journalled_set_page_dirty
,
4126 .invalidatepage
= ext4_invalidatepage
,
4127 .releasepage
= ext4_releasepage
,
4128 .is_partially_uptodate
= block_is_partially_uptodate
,
4129 .error_remove_page
= generic_error_remove_page
,
4132 static const struct address_space_operations ext4_da_aops
= {
4133 .readpage
= ext4_readpage
,
4134 .readpages
= ext4_readpages
,
4135 .writepage
= ext4_writepage
,
4136 .writepages
= ext4_da_writepages
,
4137 .sync_page
= block_sync_page
,
4138 .write_begin
= ext4_da_write_begin
,
4139 .write_end
= ext4_da_write_end
,
4141 .invalidatepage
= ext4_da_invalidatepage
,
4142 .releasepage
= ext4_releasepage
,
4143 .direct_IO
= ext4_direct_IO
,
4144 .migratepage
= buffer_migrate_page
,
4145 .is_partially_uptodate
= block_is_partially_uptodate
,
4146 .error_remove_page
= generic_error_remove_page
,
4149 void ext4_set_aops(struct inode
*inode
)
4151 if (ext4_should_order_data(inode
) &&
4152 test_opt(inode
->i_sb
, DELALLOC
))
4153 inode
->i_mapping
->a_ops
= &ext4_da_aops
;
4154 else if (ext4_should_order_data(inode
))
4155 inode
->i_mapping
->a_ops
= &ext4_ordered_aops
;
4156 else if (ext4_should_writeback_data(inode
) &&
4157 test_opt(inode
->i_sb
, DELALLOC
))
4158 inode
->i_mapping
->a_ops
= &ext4_da_aops
;
4159 else if (ext4_should_writeback_data(inode
))
4160 inode
->i_mapping
->a_ops
= &ext4_writeback_aops
;
4162 inode
->i_mapping
->a_ops
= &ext4_journalled_aops
;
4166 * ext4_block_truncate_page() zeroes out a mapping from file offset `from'
4167 * up to the end of the block which corresponds to `from'.
4168 * This required during truncate. We need to physically zero the tail end
4169 * of that block so it doesn't yield old data if the file is later grown.
4171 int ext4_block_truncate_page(handle_t
*handle
,
4172 struct address_space
*mapping
, loff_t from
)
4174 ext4_fsblk_t index
= from
>> PAGE_CACHE_SHIFT
;
4175 unsigned offset
= from
& (PAGE_CACHE_SIZE
-1);
4176 unsigned blocksize
, length
, pos
;
4178 struct inode
*inode
= mapping
->host
;
4179 struct buffer_head
*bh
;
4183 page
= find_or_create_page(mapping
, from
>> PAGE_CACHE_SHIFT
,
4184 mapping_gfp_mask(mapping
) & ~__GFP_FS
);
4188 blocksize
= inode
->i_sb
->s_blocksize
;
4189 length
= blocksize
- (offset
& (blocksize
- 1));
4190 iblock
= index
<< (PAGE_CACHE_SHIFT
- inode
->i_sb
->s_blocksize_bits
);
4193 * For "nobh" option, we can only work if we don't need to
4194 * read-in the page - otherwise we create buffers to do the IO.
4196 if (!page_has_buffers(page
) && test_opt(inode
->i_sb
, NOBH
) &&
4197 ext4_should_writeback_data(inode
) && PageUptodate(page
)) {
4198 zero_user(page
, offset
, length
);
4199 set_page_dirty(page
);
4203 if (!page_has_buffers(page
))
4204 create_empty_buffers(page
, blocksize
, 0);
4206 /* Find the buffer that contains "offset" */
4207 bh
= page_buffers(page
);
4209 while (offset
>= pos
) {
4210 bh
= bh
->b_this_page
;
4216 if (buffer_freed(bh
)) {
4217 BUFFER_TRACE(bh
, "freed: skip");
4221 if (!buffer_mapped(bh
)) {
4222 BUFFER_TRACE(bh
, "unmapped");
4223 ext4_get_block(inode
, iblock
, bh
, 0);
4224 /* unmapped? It's a hole - nothing to do */
4225 if (!buffer_mapped(bh
)) {
4226 BUFFER_TRACE(bh
, "still unmapped");
4231 /* Ok, it's mapped. Make sure it's up-to-date */
4232 if (PageUptodate(page
))
4233 set_buffer_uptodate(bh
);
4235 if (!buffer_uptodate(bh
)) {
4237 ll_rw_block(READ
, 1, &bh
);
4239 /* Uhhuh. Read error. Complain and punt. */
4240 if (!buffer_uptodate(bh
))
4244 if (ext4_should_journal_data(inode
)) {
4245 BUFFER_TRACE(bh
, "get write access");
4246 err
= ext4_journal_get_write_access(handle
, bh
);
4251 zero_user(page
, offset
, length
);
4253 BUFFER_TRACE(bh
, "zeroed end of block");
4256 if (ext4_should_journal_data(inode
)) {
4257 err
= ext4_handle_dirty_metadata(handle
, inode
, bh
);
4259 if (ext4_should_order_data(inode
))
4260 err
= ext4_jbd2_file_inode(handle
, inode
);
4261 mark_buffer_dirty(bh
);
4266 page_cache_release(page
);
4271 * Probably it should be a library function... search for first non-zero word
4272 * or memcmp with zero_page, whatever is better for particular architecture.
4275 static inline int all_zeroes(__le32
*p
, __le32
*q
)
4284 * ext4_find_shared - find the indirect blocks for partial truncation.
4285 * @inode: inode in question
4286 * @depth: depth of the affected branch
4287 * @offsets: offsets of pointers in that branch (see ext4_block_to_path)
4288 * @chain: place to store the pointers to partial indirect blocks
4289 * @top: place to the (detached) top of branch
4291 * This is a helper function used by ext4_truncate().
4293 * When we do truncate() we may have to clean the ends of several
4294 * indirect blocks but leave the blocks themselves alive. Block is
4295 * partially truncated if some data below the new i_size is refered
4296 * from it (and it is on the path to the first completely truncated
4297 * data block, indeed). We have to free the top of that path along
4298 * with everything to the right of the path. Since no allocation
4299 * past the truncation point is possible until ext4_truncate()
4300 * finishes, we may safely do the latter, but top of branch may
4301 * require special attention - pageout below the truncation point
4302 * might try to populate it.
4304 * We atomically detach the top of branch from the tree, store the
4305 * block number of its root in *@top, pointers to buffer_heads of
4306 * partially truncated blocks - in @chain[].bh and pointers to
4307 * their last elements that should not be removed - in
4308 * @chain[].p. Return value is the pointer to last filled element
4311 * The work left to caller to do the actual freeing of subtrees:
4312 * a) free the subtree starting from *@top
4313 * b) free the subtrees whose roots are stored in
4314 * (@chain[i].p+1 .. end of @chain[i].bh->b_data)
4315 * c) free the subtrees growing from the inode past the @chain[0].
4316 * (no partially truncated stuff there). */
4318 static Indirect
*ext4_find_shared(struct inode
*inode
, int depth
,
4319 ext4_lblk_t offsets
[4], Indirect chain
[4],
4322 Indirect
*partial
, *p
;
4326 /* Make k index the deepest non-null offset + 1 */
4327 for (k
= depth
; k
> 1 && !offsets
[k
-1]; k
--)
4329 partial
= ext4_get_branch(inode
, k
, offsets
, chain
, &err
);
4330 /* Writer: pointers */
4332 partial
= chain
+ k
-1;
4334 * If the branch acquired continuation since we've looked at it -
4335 * fine, it should all survive and (new) top doesn't belong to us.
4337 if (!partial
->key
&& *partial
->p
)
4340 for (p
= partial
; (p
> chain
) && all_zeroes((__le32
*) p
->bh
->b_data
, p
->p
); p
--)
4343 * OK, we've found the last block that must survive. The rest of our
4344 * branch should be detached before unlocking. However, if that rest
4345 * of branch is all ours and does not grow immediately from the inode
4346 * it's easier to cheat and just decrement partial->p.
4348 if (p
== chain
+ k
- 1 && p
> chain
) {
4352 /* Nope, don't do this in ext4. Must leave the tree intact */
4359 while (partial
> p
) {
4360 brelse(partial
->bh
);
4368 * Zero a number of block pointers in either an inode or an indirect block.
4369 * If we restart the transaction we must again get write access to the
4370 * indirect block for further modification.
4372 * We release `count' blocks on disk, but (last - first) may be greater
4373 * than `count' because there can be holes in there.
4375 static int ext4_clear_blocks(handle_t
*handle
, struct inode
*inode
,
4376 struct buffer_head
*bh
,
4377 ext4_fsblk_t block_to_free
,
4378 unsigned long count
, __le32
*first
,
4382 int flags
= EXT4_FREE_BLOCKS_FORGET
| EXT4_FREE_BLOCKS_VALIDATED
;
4384 if (S_ISDIR(inode
->i_mode
) || S_ISLNK(inode
->i_mode
))
4385 flags
|= EXT4_FREE_BLOCKS_METADATA
;
4387 if (!ext4_data_block_valid(EXT4_SB(inode
->i_sb
), block_to_free
,
4389 ext4_error(inode
->i_sb
, "inode #%lu: "
4390 "attempt to clear blocks %llu len %lu, invalid",
4391 inode
->i_ino
, (unsigned long long) block_to_free
,
4396 if (try_to_extend_transaction(handle
, inode
)) {
4398 BUFFER_TRACE(bh
, "call ext4_handle_dirty_metadata");
4399 ext4_handle_dirty_metadata(handle
, inode
, bh
);
4401 ext4_mark_inode_dirty(handle
, inode
);
4402 ext4_truncate_restart_trans(handle
, inode
,
4403 blocks_for_truncate(inode
));
4405 BUFFER_TRACE(bh
, "retaking write access");
4406 ext4_journal_get_write_access(handle
, bh
);
4410 for (p
= first
; p
< last
; p
++)
4413 ext4_free_blocks(handle
, inode
, 0, block_to_free
, count
, flags
);
4418 * ext4_free_data - free a list of data blocks
4419 * @handle: handle for this transaction
4420 * @inode: inode we are dealing with
4421 * @this_bh: indirect buffer_head which contains *@first and *@last
4422 * @first: array of block numbers
4423 * @last: points immediately past the end of array
4425 * We are freeing all blocks refered from that array (numbers are stored as
4426 * little-endian 32-bit) and updating @inode->i_blocks appropriately.
4428 * We accumulate contiguous runs of blocks to free. Conveniently, if these
4429 * blocks are contiguous then releasing them at one time will only affect one
4430 * or two bitmap blocks (+ group descriptor(s) and superblock) and we won't
4431 * actually use a lot of journal space.
4433 * @this_bh will be %NULL if @first and @last point into the inode's direct
4436 static void ext4_free_data(handle_t
*handle
, struct inode
*inode
,
4437 struct buffer_head
*this_bh
,
4438 __le32
*first
, __le32
*last
)
4440 ext4_fsblk_t block_to_free
= 0; /* Starting block # of a run */
4441 unsigned long count
= 0; /* Number of blocks in the run */
4442 __le32
*block_to_free_p
= NULL
; /* Pointer into inode/ind
4445 ext4_fsblk_t nr
; /* Current block # */
4446 __le32
*p
; /* Pointer into inode/ind
4447 for current block */
4450 if (this_bh
) { /* For indirect block */
4451 BUFFER_TRACE(this_bh
, "get_write_access");
4452 err
= ext4_journal_get_write_access(handle
, this_bh
);
4453 /* Important: if we can't update the indirect pointers
4454 * to the blocks, we can't free them. */
4459 for (p
= first
; p
< last
; p
++) {
4460 nr
= le32_to_cpu(*p
);
4462 /* accumulate blocks to free if they're contiguous */
4465 block_to_free_p
= p
;
4467 } else if (nr
== block_to_free
+ count
) {
4470 if (ext4_clear_blocks(handle
, inode
, this_bh
,
4471 block_to_free
, count
,
4472 block_to_free_p
, p
))
4475 block_to_free_p
= p
;
4482 ext4_clear_blocks(handle
, inode
, this_bh
, block_to_free
,
4483 count
, block_to_free_p
, p
);
4486 BUFFER_TRACE(this_bh
, "call ext4_handle_dirty_metadata");
4489 * The buffer head should have an attached journal head at this
4490 * point. However, if the data is corrupted and an indirect
4491 * block pointed to itself, it would have been detached when
4492 * the block was cleared. Check for this instead of OOPSing.
4494 if ((EXT4_JOURNAL(inode
) == NULL
) || bh2jh(this_bh
))
4495 ext4_handle_dirty_metadata(handle
, inode
, this_bh
);
4497 ext4_error(inode
->i_sb
,
4498 "circular indirect block detected, "
4499 "inode=%lu, block=%llu",
4501 (unsigned long long) this_bh
->b_blocknr
);
4506 * ext4_free_branches - free an array of branches
4507 * @handle: JBD handle for this transaction
4508 * @inode: inode we are dealing with
4509 * @parent_bh: the buffer_head which contains *@first and *@last
4510 * @first: array of block numbers
4511 * @last: pointer immediately past the end of array
4512 * @depth: depth of the branches to free
4514 * We are freeing all blocks refered from these branches (numbers are
4515 * stored as little-endian 32-bit) and updating @inode->i_blocks
4518 static void ext4_free_branches(handle_t
*handle
, struct inode
*inode
,
4519 struct buffer_head
*parent_bh
,
4520 __le32
*first
, __le32
*last
, int depth
)
4525 if (ext4_handle_is_aborted(handle
))
4529 struct buffer_head
*bh
;
4530 int addr_per_block
= EXT4_ADDR_PER_BLOCK(inode
->i_sb
);
4532 while (--p
>= first
) {
4533 nr
= le32_to_cpu(*p
);
4535 continue; /* A hole */
4537 if (!ext4_data_block_valid(EXT4_SB(inode
->i_sb
),
4539 ext4_error(inode
->i_sb
,
4540 "indirect mapped block in inode "
4541 "#%lu invalid (level %d, blk #%lu)",
4542 inode
->i_ino
, depth
,
4543 (unsigned long) nr
);
4547 /* Go read the buffer for the next level down */
4548 bh
= sb_bread(inode
->i_sb
, nr
);
4551 * A read failure? Report error and clear slot
4555 ext4_error(inode
->i_sb
,
4556 "Read failure, inode=%lu, block=%llu",
4561 /* This zaps the entire block. Bottom up. */
4562 BUFFER_TRACE(bh
, "free child branches");
4563 ext4_free_branches(handle
, inode
, bh
,
4564 (__le32
*) bh
->b_data
,
4565 (__le32
*) bh
->b_data
+ addr_per_block
,
4569 * We've probably journalled the indirect block several
4570 * times during the truncate. But it's no longer
4571 * needed and we now drop it from the transaction via
4572 * jbd2_journal_revoke().
4574 * That's easy if it's exclusively part of this
4575 * transaction. But if it's part of the committing
4576 * transaction then jbd2_journal_forget() will simply
4577 * brelse() it. That means that if the underlying
4578 * block is reallocated in ext4_get_block(),
4579 * unmap_underlying_metadata() will find this block
4580 * and will try to get rid of it. damn, damn.
4582 * If this block has already been committed to the
4583 * journal, a revoke record will be written. And
4584 * revoke records must be emitted *before* clearing
4585 * this block's bit in the bitmaps.
4587 ext4_forget(handle
, 1, inode
, bh
, bh
->b_blocknr
);
4590 * Everything below this this pointer has been
4591 * released. Now let this top-of-subtree go.
4593 * We want the freeing of this indirect block to be
4594 * atomic in the journal with the updating of the
4595 * bitmap block which owns it. So make some room in
4598 * We zero the parent pointer *after* freeing its
4599 * pointee in the bitmaps, so if extend_transaction()
4600 * for some reason fails to put the bitmap changes and
4601 * the release into the same transaction, recovery
4602 * will merely complain about releasing a free block,
4603 * rather than leaking blocks.
4605 if (ext4_handle_is_aborted(handle
))
4607 if (try_to_extend_transaction(handle
, inode
)) {
4608 ext4_mark_inode_dirty(handle
, inode
);
4609 ext4_truncate_restart_trans(handle
, inode
,
4610 blocks_for_truncate(inode
));
4613 ext4_free_blocks(handle
, inode
, 0, nr
, 1,
4614 EXT4_FREE_BLOCKS_METADATA
);
4618 * The block which we have just freed is
4619 * pointed to by an indirect block: journal it
4621 BUFFER_TRACE(parent_bh
, "get_write_access");
4622 if (!ext4_journal_get_write_access(handle
,
4625 BUFFER_TRACE(parent_bh
,
4626 "call ext4_handle_dirty_metadata");
4627 ext4_handle_dirty_metadata(handle
,
4634 /* We have reached the bottom of the tree. */
4635 BUFFER_TRACE(parent_bh
, "free data blocks");
4636 ext4_free_data(handle
, inode
, parent_bh
, first
, last
);
4640 int ext4_can_truncate(struct inode
*inode
)
4642 if (IS_APPEND(inode
) || IS_IMMUTABLE(inode
))
4644 if (S_ISREG(inode
->i_mode
))
4646 if (S_ISDIR(inode
->i_mode
))
4648 if (S_ISLNK(inode
->i_mode
))
4649 return !ext4_inode_is_fast_symlink(inode
);
4656 * We block out ext4_get_block() block instantiations across the entire
4657 * transaction, and VFS/VM ensures that ext4_truncate() cannot run
4658 * simultaneously on behalf of the same inode.
4660 * As we work through the truncate and commmit bits of it to the journal there
4661 * is one core, guiding principle: the file's tree must always be consistent on
4662 * disk. We must be able to restart the truncate after a crash.
4664 * The file's tree may be transiently inconsistent in memory (although it
4665 * probably isn't), but whenever we close off and commit a journal transaction,
4666 * the contents of (the filesystem + the journal) must be consistent and
4667 * restartable. It's pretty simple, really: bottom up, right to left (although
4668 * left-to-right works OK too).
4670 * Note that at recovery time, journal replay occurs *before* the restart of
4671 * truncate against the orphan inode list.
4673 * The committed inode has the new, desired i_size (which is the same as
4674 * i_disksize in this case). After a crash, ext4_orphan_cleanup() will see
4675 * that this inode's truncate did not complete and it will again call
4676 * ext4_truncate() to have another go. So there will be instantiated blocks
4677 * to the right of the truncation point in a crashed ext4 filesystem. But
4678 * that's fine - as long as they are linked from the inode, the post-crash
4679 * ext4_truncate() run will find them and release them.
4681 void ext4_truncate(struct inode
*inode
)
4684 struct ext4_inode_info
*ei
= EXT4_I(inode
);
4685 __le32
*i_data
= ei
->i_data
;
4686 int addr_per_block
= EXT4_ADDR_PER_BLOCK(inode
->i_sb
);
4687 struct address_space
*mapping
= inode
->i_mapping
;
4688 ext4_lblk_t offsets
[4];
4693 ext4_lblk_t last_block
;
4694 unsigned blocksize
= inode
->i_sb
->s_blocksize
;
4696 if (!ext4_can_truncate(inode
))
4699 EXT4_I(inode
)->i_flags
&= ~EXT4_EOFBLOCKS_FL
;
4701 if (inode
->i_size
== 0 && !test_opt(inode
->i_sb
, NO_AUTO_DA_ALLOC
))
4702 ext4_set_inode_state(inode
, EXT4_STATE_DA_ALLOC_CLOSE
);
4704 if (EXT4_I(inode
)->i_flags
& EXT4_EXTENTS_FL
) {
4705 ext4_ext_truncate(inode
);
4709 handle
= start_transaction(inode
);
4711 return; /* AKPM: return what? */
4713 last_block
= (inode
->i_size
+ blocksize
-1)
4714 >> EXT4_BLOCK_SIZE_BITS(inode
->i_sb
);
4716 if (inode
->i_size
& (blocksize
- 1))
4717 if (ext4_block_truncate_page(handle
, mapping
, inode
->i_size
))
4720 n
= ext4_block_to_path(inode
, last_block
, offsets
, NULL
);
4722 goto out_stop
; /* error */
4725 * OK. This truncate is going to happen. We add the inode to the
4726 * orphan list, so that if this truncate spans multiple transactions,
4727 * and we crash, we will resume the truncate when the filesystem
4728 * recovers. It also marks the inode dirty, to catch the new size.
4730 * Implication: the file must always be in a sane, consistent
4731 * truncatable state while each transaction commits.
4733 if (ext4_orphan_add(handle
, inode
))
4737 * From here we block out all ext4_get_block() callers who want to
4738 * modify the block allocation tree.
4740 down_write(&ei
->i_data_sem
);
4742 ext4_discard_preallocations(inode
);
4745 * The orphan list entry will now protect us from any crash which
4746 * occurs before the truncate completes, so it is now safe to propagate
4747 * the new, shorter inode size (held for now in i_size) into the
4748 * on-disk inode. We do this via i_disksize, which is the value which
4749 * ext4 *really* writes onto the disk inode.
4751 ei
->i_disksize
= inode
->i_size
;
4753 if (n
== 1) { /* direct blocks */
4754 ext4_free_data(handle
, inode
, NULL
, i_data
+offsets
[0],
4755 i_data
+ EXT4_NDIR_BLOCKS
);
4759 partial
= ext4_find_shared(inode
, n
, offsets
, chain
, &nr
);
4760 /* Kill the top of shared branch (not detached) */
4762 if (partial
== chain
) {
4763 /* Shared branch grows from the inode */
4764 ext4_free_branches(handle
, inode
, NULL
,
4765 &nr
, &nr
+1, (chain
+n
-1) - partial
);
4768 * We mark the inode dirty prior to restart,
4769 * and prior to stop. No need for it here.
4772 /* Shared branch grows from an indirect block */
4773 BUFFER_TRACE(partial
->bh
, "get_write_access");
4774 ext4_free_branches(handle
, inode
, partial
->bh
,
4776 partial
->p
+1, (chain
+n
-1) - partial
);
4779 /* Clear the ends of indirect blocks on the shared branch */
4780 while (partial
> chain
) {
4781 ext4_free_branches(handle
, inode
, partial
->bh
, partial
->p
+ 1,
4782 (__le32
*)partial
->bh
->b_data
+addr_per_block
,
4783 (chain
+n
-1) - partial
);
4784 BUFFER_TRACE(partial
->bh
, "call brelse");
4785 brelse(partial
->bh
);
4789 /* Kill the remaining (whole) subtrees */
4790 switch (offsets
[0]) {
4792 nr
= i_data
[EXT4_IND_BLOCK
];
4794 ext4_free_branches(handle
, inode
, NULL
, &nr
, &nr
+1, 1);
4795 i_data
[EXT4_IND_BLOCK
] = 0;
4797 case EXT4_IND_BLOCK
:
4798 nr
= i_data
[EXT4_DIND_BLOCK
];
4800 ext4_free_branches(handle
, inode
, NULL
, &nr
, &nr
+1, 2);
4801 i_data
[EXT4_DIND_BLOCK
] = 0;
4803 case EXT4_DIND_BLOCK
:
4804 nr
= i_data
[EXT4_TIND_BLOCK
];
4806 ext4_free_branches(handle
, inode
, NULL
, &nr
, &nr
+1, 3);
4807 i_data
[EXT4_TIND_BLOCK
] = 0;
4809 case EXT4_TIND_BLOCK
:
4813 up_write(&ei
->i_data_sem
);
4814 inode
->i_mtime
= inode
->i_ctime
= ext4_current_time(inode
);
4815 ext4_mark_inode_dirty(handle
, inode
);
4818 * In a multi-transaction truncate, we only make the final transaction
4822 ext4_handle_sync(handle
);
4825 * If this was a simple ftruncate(), and the file will remain alive
4826 * then we need to clear up the orphan record which we created above.
4827 * However, if this was a real unlink then we were called by
4828 * ext4_delete_inode(), and we allow that function to clean up the
4829 * orphan info for us.
4832 ext4_orphan_del(handle
, inode
);
4834 ext4_journal_stop(handle
);
4838 * ext4_get_inode_loc returns with an extra refcount against the inode's
4839 * underlying buffer_head on success. If 'in_mem' is true, we have all
4840 * data in memory that is needed to recreate the on-disk version of this
4843 static int __ext4_get_inode_loc(struct inode
*inode
,
4844 struct ext4_iloc
*iloc
, int in_mem
)
4846 struct ext4_group_desc
*gdp
;
4847 struct buffer_head
*bh
;
4848 struct super_block
*sb
= inode
->i_sb
;
4850 int inodes_per_block
, inode_offset
;
4853 if (!ext4_valid_inum(sb
, inode
->i_ino
))
4856 iloc
->block_group
= (inode
->i_ino
- 1) / EXT4_INODES_PER_GROUP(sb
);
4857 gdp
= ext4_get_group_desc(sb
, iloc
->block_group
, NULL
);
4862 * Figure out the offset within the block group inode table
4864 inodes_per_block
= (EXT4_BLOCK_SIZE(sb
) / EXT4_INODE_SIZE(sb
));
4865 inode_offset
= ((inode
->i_ino
- 1) %
4866 EXT4_INODES_PER_GROUP(sb
));
4867 block
= ext4_inode_table(sb
, gdp
) + (inode_offset
/ inodes_per_block
);
4868 iloc
->offset
= (inode_offset
% inodes_per_block
) * EXT4_INODE_SIZE(sb
);
4870 bh
= sb_getblk(sb
, block
);
4872 ext4_error(sb
, "unable to read inode block - "
4873 "inode=%lu, block=%llu", inode
->i_ino
, block
);
4876 if (!buffer_uptodate(bh
)) {
4880 * If the buffer has the write error flag, we have failed
4881 * to write out another inode in the same block. In this
4882 * case, we don't have to read the block because we may
4883 * read the old inode data successfully.
4885 if (buffer_write_io_error(bh
) && !buffer_uptodate(bh
))
4886 set_buffer_uptodate(bh
);
4888 if (buffer_uptodate(bh
)) {
4889 /* someone brought it uptodate while we waited */
4895 * If we have all information of the inode in memory and this
4896 * is the only valid inode in the block, we need not read the
4900 struct buffer_head
*bitmap_bh
;
4903 start
= inode_offset
& ~(inodes_per_block
- 1);
4905 /* Is the inode bitmap in cache? */
4906 bitmap_bh
= sb_getblk(sb
, ext4_inode_bitmap(sb
, gdp
));
4911 * If the inode bitmap isn't in cache then the
4912 * optimisation may end up performing two reads instead
4913 * of one, so skip it.
4915 if (!buffer_uptodate(bitmap_bh
)) {
4919 for (i
= start
; i
< start
+ inodes_per_block
; i
++) {
4920 if (i
== inode_offset
)
4922 if (ext4_test_bit(i
, bitmap_bh
->b_data
))
4926 if (i
== start
+ inodes_per_block
) {
4927 /* all other inodes are free, so skip I/O */
4928 memset(bh
->b_data
, 0, bh
->b_size
);
4929 set_buffer_uptodate(bh
);
4937 * If we need to do any I/O, try to pre-readahead extra
4938 * blocks from the inode table.
4940 if (EXT4_SB(sb
)->s_inode_readahead_blks
) {
4941 ext4_fsblk_t b
, end
, table
;
4944 table
= ext4_inode_table(sb
, gdp
);
4945 /* s_inode_readahead_blks is always a power of 2 */
4946 b
= block
& ~(EXT4_SB(sb
)->s_inode_readahead_blks
-1);
4949 end
= b
+ EXT4_SB(sb
)->s_inode_readahead_blks
;
4950 num
= EXT4_INODES_PER_GROUP(sb
);
4951 if (EXT4_HAS_RO_COMPAT_FEATURE(sb
,
4952 EXT4_FEATURE_RO_COMPAT_GDT_CSUM
))
4953 num
-= ext4_itable_unused_count(sb
, gdp
);
4954 table
+= num
/ inodes_per_block
;
4958 sb_breadahead(sb
, b
++);
4962 * There are other valid inodes in the buffer, this inode
4963 * has in-inode xattrs, or we don't have this inode in memory.
4964 * Read the block from disk.
4967 bh
->b_end_io
= end_buffer_read_sync
;
4968 submit_bh(READ_META
, bh
);
4970 if (!buffer_uptodate(bh
)) {
4971 ext4_error(sb
, "unable to read inode block - inode=%lu,"
4972 " block=%llu", inode
->i_ino
, block
);
4982 int ext4_get_inode_loc(struct inode
*inode
, struct ext4_iloc
*iloc
)
4984 /* We have all inode data except xattrs in memory here. */
4985 return __ext4_get_inode_loc(inode
, iloc
,
4986 !ext4_test_inode_state(inode
, EXT4_STATE_XATTR
));
4989 void ext4_set_inode_flags(struct inode
*inode
)
4991 unsigned int flags
= EXT4_I(inode
)->i_flags
;
4993 inode
->i_flags
&= ~(S_SYNC
|S_APPEND
|S_IMMUTABLE
|S_NOATIME
|S_DIRSYNC
);
4994 if (flags
& EXT4_SYNC_FL
)
4995 inode
->i_flags
|= S_SYNC
;
4996 if (flags
& EXT4_APPEND_FL
)
4997 inode
->i_flags
|= S_APPEND
;
4998 if (flags
& EXT4_IMMUTABLE_FL
)
4999 inode
->i_flags
|= S_IMMUTABLE
;
5000 if (flags
& EXT4_NOATIME_FL
)
5001 inode
->i_flags
|= S_NOATIME
;
5002 if (flags
& EXT4_DIRSYNC_FL
)
5003 inode
->i_flags
|= S_DIRSYNC
;
5006 /* Propagate flags from i_flags to EXT4_I(inode)->i_flags */
5007 void ext4_get_inode_flags(struct ext4_inode_info
*ei
)
5009 unsigned int flags
= ei
->vfs_inode
.i_flags
;
5011 ei
->i_flags
&= ~(EXT4_SYNC_FL
|EXT4_APPEND_FL
|
5012 EXT4_IMMUTABLE_FL
|EXT4_NOATIME_FL
|EXT4_DIRSYNC_FL
);
5014 ei
->i_flags
|= EXT4_SYNC_FL
;
5015 if (flags
& S_APPEND
)
5016 ei
->i_flags
|= EXT4_APPEND_FL
;
5017 if (flags
& S_IMMUTABLE
)
5018 ei
->i_flags
|= EXT4_IMMUTABLE_FL
;
5019 if (flags
& S_NOATIME
)
5020 ei
->i_flags
|= EXT4_NOATIME_FL
;
5021 if (flags
& S_DIRSYNC
)
5022 ei
->i_flags
|= EXT4_DIRSYNC_FL
;
5025 static blkcnt_t
ext4_inode_blocks(struct ext4_inode
*raw_inode
,
5026 struct ext4_inode_info
*ei
)
5029 struct inode
*inode
= &(ei
->vfs_inode
);
5030 struct super_block
*sb
= inode
->i_sb
;
5032 if (EXT4_HAS_RO_COMPAT_FEATURE(sb
,
5033 EXT4_FEATURE_RO_COMPAT_HUGE_FILE
)) {
5034 /* we are using combined 48 bit field */
5035 i_blocks
= ((u64
)le16_to_cpu(raw_inode
->i_blocks_high
)) << 32 |
5036 le32_to_cpu(raw_inode
->i_blocks_lo
);
5037 if (ei
->i_flags
& EXT4_HUGE_FILE_FL
) {
5038 /* i_blocks represent file system block size */
5039 return i_blocks
<< (inode
->i_blkbits
- 9);
5044 return le32_to_cpu(raw_inode
->i_blocks_lo
);
5048 struct inode
*ext4_iget(struct super_block
*sb
, unsigned long ino
)
5050 struct ext4_iloc iloc
;
5051 struct ext4_inode
*raw_inode
;
5052 struct ext4_inode_info
*ei
;
5053 struct inode
*inode
;
5054 journal_t
*journal
= EXT4_SB(sb
)->s_journal
;
5058 inode
= iget_locked(sb
, ino
);
5060 return ERR_PTR(-ENOMEM
);
5061 if (!(inode
->i_state
& I_NEW
))
5067 ret
= __ext4_get_inode_loc(inode
, &iloc
, 0);
5070 raw_inode
= ext4_raw_inode(&iloc
);
5071 inode
->i_mode
= le16_to_cpu(raw_inode
->i_mode
);
5072 inode
->i_uid
= (uid_t
)le16_to_cpu(raw_inode
->i_uid_low
);
5073 inode
->i_gid
= (gid_t
)le16_to_cpu(raw_inode
->i_gid_low
);
5074 if (!(test_opt(inode
->i_sb
, NO_UID32
))) {
5075 inode
->i_uid
|= le16_to_cpu(raw_inode
->i_uid_high
) << 16;
5076 inode
->i_gid
|= le16_to_cpu(raw_inode
->i_gid_high
) << 16;
5078 inode
->i_nlink
= le16_to_cpu(raw_inode
->i_links_count
);
5080 ei
->i_state_flags
= 0;
5081 ei
->i_dir_start_lookup
= 0;
5082 ei
->i_dtime
= le32_to_cpu(raw_inode
->i_dtime
);
5083 /* We now have enough fields to check if the inode was active or not.
5084 * This is needed because nfsd might try to access dead inodes
5085 * the test is that same one that e2fsck uses
5086 * NeilBrown 1999oct15
5088 if (inode
->i_nlink
== 0) {
5089 if (inode
->i_mode
== 0 ||
5090 !(EXT4_SB(inode
->i_sb
)->s_mount_state
& EXT4_ORPHAN_FS
)) {
5091 /* this inode is deleted */
5095 /* The only unlinked inodes we let through here have
5096 * valid i_mode and are being read by the orphan
5097 * recovery code: that's fine, we're about to complete
5098 * the process of deleting those. */
5100 ei
->i_flags
= le32_to_cpu(raw_inode
->i_flags
);
5101 inode
->i_blocks
= ext4_inode_blocks(raw_inode
, ei
);
5102 ei
->i_file_acl
= le32_to_cpu(raw_inode
->i_file_acl_lo
);
5103 if (EXT4_HAS_INCOMPAT_FEATURE(sb
, EXT4_FEATURE_INCOMPAT_64BIT
))
5105 ((__u64
)le16_to_cpu(raw_inode
->i_file_acl_high
)) << 32;
5106 inode
->i_size
= ext4_isize(raw_inode
);
5107 ei
->i_disksize
= inode
->i_size
;
5109 ei
->i_reserved_quota
= 0;
5111 inode
->i_generation
= le32_to_cpu(raw_inode
->i_generation
);
5112 ei
->i_block_group
= iloc
.block_group
;
5113 ei
->i_last_alloc_group
= ~0;
5115 * NOTE! The in-memory inode i_data array is in little-endian order
5116 * even on big-endian machines: we do NOT byteswap the block numbers!
5118 for (block
= 0; block
< EXT4_N_BLOCKS
; block
++)
5119 ei
->i_data
[block
] = raw_inode
->i_block
[block
];
5120 INIT_LIST_HEAD(&ei
->i_orphan
);
5123 * Set transaction id's of transactions that have to be committed
5124 * to finish f[data]sync. We set them to currently running transaction
5125 * as we cannot be sure that the inode or some of its metadata isn't
5126 * part of the transaction - the inode could have been reclaimed and
5127 * now it is reread from disk.
5130 transaction_t
*transaction
;
5133 spin_lock(&journal
->j_state_lock
);
5134 if (journal
->j_running_transaction
)
5135 transaction
= journal
->j_running_transaction
;
5137 transaction
= journal
->j_committing_transaction
;
5139 tid
= transaction
->t_tid
;
5141 tid
= journal
->j_commit_sequence
;
5142 spin_unlock(&journal
->j_state_lock
);
5143 ei
->i_sync_tid
= tid
;
5144 ei
->i_datasync_tid
= tid
;
5147 if (EXT4_INODE_SIZE(inode
->i_sb
) > EXT4_GOOD_OLD_INODE_SIZE
) {
5148 ei
->i_extra_isize
= le16_to_cpu(raw_inode
->i_extra_isize
);
5149 if (EXT4_GOOD_OLD_INODE_SIZE
+ ei
->i_extra_isize
>
5150 EXT4_INODE_SIZE(inode
->i_sb
)) {
5154 if (ei
->i_extra_isize
== 0) {
5155 /* The extra space is currently unused. Use it. */
5156 ei
->i_extra_isize
= sizeof(struct ext4_inode
) -
5157 EXT4_GOOD_OLD_INODE_SIZE
;
5159 __le32
*magic
= (void *)raw_inode
+
5160 EXT4_GOOD_OLD_INODE_SIZE
+
5162 if (*magic
== cpu_to_le32(EXT4_XATTR_MAGIC
))
5163 ext4_set_inode_state(inode
, EXT4_STATE_XATTR
);
5166 ei
->i_extra_isize
= 0;
5168 EXT4_INODE_GET_XTIME(i_ctime
, inode
, raw_inode
);
5169 EXT4_INODE_GET_XTIME(i_mtime
, inode
, raw_inode
);
5170 EXT4_INODE_GET_XTIME(i_atime
, inode
, raw_inode
);
5171 EXT4_EINODE_GET_XTIME(i_crtime
, ei
, raw_inode
);
5173 inode
->i_version
= le32_to_cpu(raw_inode
->i_disk_version
);
5174 if (EXT4_INODE_SIZE(inode
->i_sb
) > EXT4_GOOD_OLD_INODE_SIZE
) {
5175 if (EXT4_FITS_IN_INODE(raw_inode
, ei
, i_version_hi
))
5177 (__u64
)(le32_to_cpu(raw_inode
->i_version_hi
)) << 32;
5181 if (ei
->i_file_acl
&&
5182 !ext4_data_block_valid(EXT4_SB(sb
), ei
->i_file_acl
, 1)) {
5183 ext4_error(sb
, "bad extended attribute block %llu inode #%lu",
5184 ei
->i_file_acl
, inode
->i_ino
);
5187 } else if (ei
->i_flags
& EXT4_EXTENTS_FL
) {
5188 if (S_ISREG(inode
->i_mode
) || S_ISDIR(inode
->i_mode
) ||
5189 (S_ISLNK(inode
->i_mode
) &&
5190 !ext4_inode_is_fast_symlink(inode
)))
5191 /* Validate extent which is part of inode */
5192 ret
= ext4_ext_check_inode(inode
);
5193 } else if (S_ISREG(inode
->i_mode
) || S_ISDIR(inode
->i_mode
) ||
5194 (S_ISLNK(inode
->i_mode
) &&
5195 !ext4_inode_is_fast_symlink(inode
))) {
5196 /* Validate block references which are part of inode */
5197 ret
= ext4_check_inode_blockref(inode
);
5202 if (S_ISREG(inode
->i_mode
)) {
5203 inode
->i_op
= &ext4_file_inode_operations
;
5204 inode
->i_fop
= &ext4_file_operations
;
5205 ext4_set_aops(inode
);
5206 } else if (S_ISDIR(inode
->i_mode
)) {
5207 inode
->i_op
= &ext4_dir_inode_operations
;
5208 inode
->i_fop
= &ext4_dir_operations
;
5209 } else if (S_ISLNK(inode
->i_mode
)) {
5210 if (ext4_inode_is_fast_symlink(inode
)) {
5211 inode
->i_op
= &ext4_fast_symlink_inode_operations
;
5212 nd_terminate_link(ei
->i_data
, inode
->i_size
,
5213 sizeof(ei
->i_data
) - 1);
5215 inode
->i_op
= &ext4_symlink_inode_operations
;
5216 ext4_set_aops(inode
);
5218 } else if (S_ISCHR(inode
->i_mode
) || S_ISBLK(inode
->i_mode
) ||
5219 S_ISFIFO(inode
->i_mode
) || S_ISSOCK(inode
->i_mode
)) {
5220 inode
->i_op
= &ext4_special_inode_operations
;
5221 if (raw_inode
->i_block
[0])
5222 init_special_inode(inode
, inode
->i_mode
,
5223 old_decode_dev(le32_to_cpu(raw_inode
->i_block
[0])));
5225 init_special_inode(inode
, inode
->i_mode
,
5226 new_decode_dev(le32_to_cpu(raw_inode
->i_block
[1])));
5229 ext4_error(inode
->i_sb
, "bogus i_mode (%o) for inode=%lu",
5230 inode
->i_mode
, inode
->i_ino
);
5234 ext4_set_inode_flags(inode
);
5235 unlock_new_inode(inode
);
5241 return ERR_PTR(ret
);
5244 static int ext4_inode_blocks_set(handle_t
*handle
,
5245 struct ext4_inode
*raw_inode
,
5246 struct ext4_inode_info
*ei
)
5248 struct inode
*inode
= &(ei
->vfs_inode
);
5249 u64 i_blocks
= inode
->i_blocks
;
5250 struct super_block
*sb
= inode
->i_sb
;
5252 if (i_blocks
<= ~0U) {
5254 * i_blocks can be represnted in a 32 bit variable
5255 * as multiple of 512 bytes
5257 raw_inode
->i_blocks_lo
= cpu_to_le32(i_blocks
);
5258 raw_inode
->i_blocks_high
= 0;
5259 ei
->i_flags
&= ~EXT4_HUGE_FILE_FL
;
5262 if (!EXT4_HAS_RO_COMPAT_FEATURE(sb
, EXT4_FEATURE_RO_COMPAT_HUGE_FILE
))
5265 if (i_blocks
<= 0xffffffffffffULL
) {
5267 * i_blocks can be represented in a 48 bit variable
5268 * as multiple of 512 bytes
5270 raw_inode
->i_blocks_lo
= cpu_to_le32(i_blocks
);
5271 raw_inode
->i_blocks_high
= cpu_to_le16(i_blocks
>> 32);
5272 ei
->i_flags
&= ~EXT4_HUGE_FILE_FL
;
5274 ei
->i_flags
|= EXT4_HUGE_FILE_FL
;
5275 /* i_block is stored in file system block size */
5276 i_blocks
= i_blocks
>> (inode
->i_blkbits
- 9);
5277 raw_inode
->i_blocks_lo
= cpu_to_le32(i_blocks
);
5278 raw_inode
->i_blocks_high
= cpu_to_le16(i_blocks
>> 32);
5284 * Post the struct inode info into an on-disk inode location in the
5285 * buffer-cache. This gobbles the caller's reference to the
5286 * buffer_head in the inode location struct.
5288 * The caller must have write access to iloc->bh.
5290 static int ext4_do_update_inode(handle_t
*handle
,
5291 struct inode
*inode
,
5292 struct ext4_iloc
*iloc
)
5294 struct ext4_inode
*raw_inode
= ext4_raw_inode(iloc
);
5295 struct ext4_inode_info
*ei
= EXT4_I(inode
);
5296 struct buffer_head
*bh
= iloc
->bh
;
5297 int err
= 0, rc
, block
;
5299 /* For fields not not tracking in the in-memory inode,
5300 * initialise them to zero for new inodes. */
5301 if (ext4_test_inode_state(inode
, EXT4_STATE_NEW
))
5302 memset(raw_inode
, 0, EXT4_SB(inode
->i_sb
)->s_inode_size
);
5304 ext4_get_inode_flags(ei
);
5305 raw_inode
->i_mode
= cpu_to_le16(inode
->i_mode
);
5306 if (!(test_opt(inode
->i_sb
, NO_UID32
))) {
5307 raw_inode
->i_uid_low
= cpu_to_le16(low_16_bits(inode
->i_uid
));
5308 raw_inode
->i_gid_low
= cpu_to_le16(low_16_bits(inode
->i_gid
));
5310 * Fix up interoperability with old kernels. Otherwise, old inodes get
5311 * re-used with the upper 16 bits of the uid/gid intact
5314 raw_inode
->i_uid_high
=
5315 cpu_to_le16(high_16_bits(inode
->i_uid
));
5316 raw_inode
->i_gid_high
=
5317 cpu_to_le16(high_16_bits(inode
->i_gid
));
5319 raw_inode
->i_uid_high
= 0;
5320 raw_inode
->i_gid_high
= 0;
5323 raw_inode
->i_uid_low
=
5324 cpu_to_le16(fs_high2lowuid(inode
->i_uid
));
5325 raw_inode
->i_gid_low
=
5326 cpu_to_le16(fs_high2lowgid(inode
->i_gid
));
5327 raw_inode
->i_uid_high
= 0;
5328 raw_inode
->i_gid_high
= 0;
5330 raw_inode
->i_links_count
= cpu_to_le16(inode
->i_nlink
);
5332 EXT4_INODE_SET_XTIME(i_ctime
, inode
, raw_inode
);
5333 EXT4_INODE_SET_XTIME(i_mtime
, inode
, raw_inode
);
5334 EXT4_INODE_SET_XTIME(i_atime
, inode
, raw_inode
);
5335 EXT4_EINODE_SET_XTIME(i_crtime
, ei
, raw_inode
);
5337 if (ext4_inode_blocks_set(handle
, raw_inode
, ei
))
5339 raw_inode
->i_dtime
= cpu_to_le32(ei
->i_dtime
);
5340 raw_inode
->i_flags
= cpu_to_le32(ei
->i_flags
);
5341 if (EXT4_SB(inode
->i_sb
)->s_es
->s_creator_os
!=
5342 cpu_to_le32(EXT4_OS_HURD
))
5343 raw_inode
->i_file_acl_high
=
5344 cpu_to_le16(ei
->i_file_acl
>> 32);
5345 raw_inode
->i_file_acl_lo
= cpu_to_le32(ei
->i_file_acl
);
5346 ext4_isize_set(raw_inode
, ei
->i_disksize
);
5347 if (ei
->i_disksize
> 0x7fffffffULL
) {
5348 struct super_block
*sb
= inode
->i_sb
;
5349 if (!EXT4_HAS_RO_COMPAT_FEATURE(sb
,
5350 EXT4_FEATURE_RO_COMPAT_LARGE_FILE
) ||
5351 EXT4_SB(sb
)->s_es
->s_rev_level
==
5352 cpu_to_le32(EXT4_GOOD_OLD_REV
)) {
5353 /* If this is the first large file
5354 * created, add a flag to the superblock.
5356 err
= ext4_journal_get_write_access(handle
,
5357 EXT4_SB(sb
)->s_sbh
);
5360 ext4_update_dynamic_rev(sb
);
5361 EXT4_SET_RO_COMPAT_FEATURE(sb
,
5362 EXT4_FEATURE_RO_COMPAT_LARGE_FILE
);
5364 ext4_handle_sync(handle
);
5365 err
= ext4_handle_dirty_metadata(handle
, NULL
,
5366 EXT4_SB(sb
)->s_sbh
);
5369 raw_inode
->i_generation
= cpu_to_le32(inode
->i_generation
);
5370 if (S_ISCHR(inode
->i_mode
) || S_ISBLK(inode
->i_mode
)) {
5371 if (old_valid_dev(inode
->i_rdev
)) {
5372 raw_inode
->i_block
[0] =
5373 cpu_to_le32(old_encode_dev(inode
->i_rdev
));
5374 raw_inode
->i_block
[1] = 0;
5376 raw_inode
->i_block
[0] = 0;
5377 raw_inode
->i_block
[1] =
5378 cpu_to_le32(new_encode_dev(inode
->i_rdev
));
5379 raw_inode
->i_block
[2] = 0;
5382 for (block
= 0; block
< EXT4_N_BLOCKS
; block
++)
5383 raw_inode
->i_block
[block
] = ei
->i_data
[block
];
5385 raw_inode
->i_disk_version
= cpu_to_le32(inode
->i_version
);
5386 if (ei
->i_extra_isize
) {
5387 if (EXT4_FITS_IN_INODE(raw_inode
, ei
, i_version_hi
))
5388 raw_inode
->i_version_hi
=
5389 cpu_to_le32(inode
->i_version
>> 32);
5390 raw_inode
->i_extra_isize
= cpu_to_le16(ei
->i_extra_isize
);
5393 BUFFER_TRACE(bh
, "call ext4_handle_dirty_metadata");
5394 rc
= ext4_handle_dirty_metadata(handle
, NULL
, bh
);
5397 ext4_clear_inode_state(inode
, EXT4_STATE_NEW
);
5399 ext4_update_inode_fsync_trans(handle
, inode
, 0);
5402 ext4_std_error(inode
->i_sb
, err
);
5407 * ext4_write_inode()
5409 * We are called from a few places:
5411 * - Within generic_file_write() for O_SYNC files.
5412 * Here, there will be no transaction running. We wait for any running
5413 * trasnaction to commit.
5415 * - Within sys_sync(), kupdate and such.
5416 * We wait on commit, if tol to.
5418 * - Within prune_icache() (PF_MEMALLOC == true)
5419 * Here we simply return. We can't afford to block kswapd on the
5422 * In all cases it is actually safe for us to return without doing anything,
5423 * because the inode has been copied into a raw inode buffer in
5424 * ext4_mark_inode_dirty(). This is a correctness thing for O_SYNC and for
5427 * Note that we are absolutely dependent upon all inode dirtiers doing the
5428 * right thing: they *must* call mark_inode_dirty() after dirtying info in
5429 * which we are interested.
5431 * It would be a bug for them to not do this. The code:
5433 * mark_inode_dirty(inode)
5435 * inode->i_size = expr;
5437 * is in error because a kswapd-driven write_inode() could occur while
5438 * `stuff()' is running, and the new i_size will be lost. Plus the inode
5439 * will no longer be on the superblock's dirty inode list.
5441 int ext4_write_inode(struct inode
*inode
, struct writeback_control
*wbc
)
5445 if (current
->flags
& PF_MEMALLOC
)
5448 if (EXT4_SB(inode
->i_sb
)->s_journal
) {
5449 if (ext4_journal_current_handle()) {
5450 jbd_debug(1, "called recursively, non-PF_MEMALLOC!\n");
5455 if (wbc
->sync_mode
!= WB_SYNC_ALL
)
5458 err
= ext4_force_commit(inode
->i_sb
);
5460 struct ext4_iloc iloc
;
5462 err
= __ext4_get_inode_loc(inode
, &iloc
, 0);
5465 if (wbc
->sync_mode
== WB_SYNC_ALL
)
5466 sync_dirty_buffer(iloc
.bh
);
5467 if (buffer_req(iloc
.bh
) && !buffer_uptodate(iloc
.bh
)) {
5468 ext4_error(inode
->i_sb
, "IO error syncing inode, "
5469 "inode=%lu, block=%llu", inode
->i_ino
,
5470 (unsigned long long)iloc
.bh
->b_blocknr
);
5481 * Called from notify_change.
5483 * We want to trap VFS attempts to truncate the file as soon as
5484 * possible. In particular, we want to make sure that when the VFS
5485 * shrinks i_size, we put the inode on the orphan list and modify
5486 * i_disksize immediately, so that during the subsequent flushing of
5487 * dirty pages and freeing of disk blocks, we can guarantee that any
5488 * commit will leave the blocks being flushed in an unused state on
5489 * disk. (On recovery, the inode will get truncated and the blocks will
5490 * be freed, so we have a strong guarantee that no future commit will
5491 * leave these blocks visible to the user.)
5493 * Another thing we have to assure is that if we are in ordered mode
5494 * and inode is still attached to the committing transaction, we must
5495 * we start writeout of all the dirty pages which are being truncated.
5496 * This way we are sure that all the data written in the previous
5497 * transaction are already on disk (truncate waits for pages under
5500 * Called with inode->i_mutex down.
5502 int ext4_setattr(struct dentry
*dentry
, struct iattr
*attr
)
5504 struct inode
*inode
= dentry
->d_inode
;
5506 const unsigned int ia_valid
= attr
->ia_valid
;
5508 error
= inode_change_ok(inode
, attr
);
5512 if (ia_valid
& ATTR_SIZE
)
5513 dquot_initialize(inode
);
5514 if ((ia_valid
& ATTR_UID
&& attr
->ia_uid
!= inode
->i_uid
) ||
5515 (ia_valid
& ATTR_GID
&& attr
->ia_gid
!= inode
->i_gid
)) {
5518 /* (user+group)*(old+new) structure, inode write (sb,
5519 * inode block, ? - but truncate inode update has it) */
5520 handle
= ext4_journal_start(inode
, (EXT4_MAXQUOTAS_INIT_BLOCKS(inode
->i_sb
)+
5521 EXT4_MAXQUOTAS_DEL_BLOCKS(inode
->i_sb
))+3);
5522 if (IS_ERR(handle
)) {
5523 error
= PTR_ERR(handle
);
5526 error
= dquot_transfer(inode
, attr
);
5528 ext4_journal_stop(handle
);
5531 /* Update corresponding info in inode so that everything is in
5532 * one transaction */
5533 if (attr
->ia_valid
& ATTR_UID
)
5534 inode
->i_uid
= attr
->ia_uid
;
5535 if (attr
->ia_valid
& ATTR_GID
)
5536 inode
->i_gid
= attr
->ia_gid
;
5537 error
= ext4_mark_inode_dirty(handle
, inode
);
5538 ext4_journal_stop(handle
);
5541 if (attr
->ia_valid
& ATTR_SIZE
) {
5542 if (!(EXT4_I(inode
)->i_flags
& EXT4_EXTENTS_FL
)) {
5543 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
5545 if (attr
->ia_size
> sbi
->s_bitmap_maxbytes
) {
5552 if (S_ISREG(inode
->i_mode
) &&
5553 attr
->ia_valid
& ATTR_SIZE
&&
5554 (attr
->ia_size
< inode
->i_size
||
5555 (EXT4_I(inode
)->i_flags
& EXT4_EOFBLOCKS_FL
))) {
5558 handle
= ext4_journal_start(inode
, 3);
5559 if (IS_ERR(handle
)) {
5560 error
= PTR_ERR(handle
);
5564 error
= ext4_orphan_add(handle
, inode
);
5565 EXT4_I(inode
)->i_disksize
= attr
->ia_size
;
5566 rc
= ext4_mark_inode_dirty(handle
, inode
);
5569 ext4_journal_stop(handle
);
5571 if (ext4_should_order_data(inode
)) {
5572 error
= ext4_begin_ordered_truncate(inode
,
5575 /* Do as much error cleanup as possible */
5576 handle
= ext4_journal_start(inode
, 3);
5577 if (IS_ERR(handle
)) {
5578 ext4_orphan_del(NULL
, inode
);
5581 ext4_orphan_del(handle
, inode
);
5582 ext4_journal_stop(handle
);
5586 /* ext4_truncate will clear the flag */
5587 if ((EXT4_I(inode
)->i_flags
& EXT4_EOFBLOCKS_FL
))
5588 ext4_truncate(inode
);
5591 rc
= inode_setattr(inode
, attr
);
5593 /* If inode_setattr's call to ext4_truncate failed to get a
5594 * transaction handle at all, we need to clean up the in-core
5595 * orphan list manually. */
5597 ext4_orphan_del(NULL
, inode
);
5599 if (!rc
&& (ia_valid
& ATTR_MODE
))
5600 rc
= ext4_acl_chmod(inode
);
5603 ext4_std_error(inode
->i_sb
, error
);
5609 int ext4_getattr(struct vfsmount
*mnt
, struct dentry
*dentry
,
5612 struct inode
*inode
;
5613 unsigned long delalloc_blocks
;
5615 inode
= dentry
->d_inode
;
5616 generic_fillattr(inode
, stat
);
5619 * We can't update i_blocks if the block allocation is delayed
5620 * otherwise in the case of system crash before the real block
5621 * allocation is done, we will have i_blocks inconsistent with
5622 * on-disk file blocks.
5623 * We always keep i_blocks updated together with real
5624 * allocation. But to not confuse with user, stat
5625 * will return the blocks that include the delayed allocation
5626 * blocks for this file.
5628 spin_lock(&EXT4_I(inode
)->i_block_reservation_lock
);
5629 delalloc_blocks
= EXT4_I(inode
)->i_reserved_data_blocks
;
5630 spin_unlock(&EXT4_I(inode
)->i_block_reservation_lock
);
5632 stat
->blocks
+= (delalloc_blocks
<< inode
->i_sb
->s_blocksize_bits
)>>9;
5636 static int ext4_indirect_trans_blocks(struct inode
*inode
, int nrblocks
,
5641 /* if nrblocks are contiguous */
5644 * With N contiguous data blocks, it need at most
5645 * N/EXT4_ADDR_PER_BLOCK(inode->i_sb) indirect blocks
5646 * 2 dindirect blocks
5649 indirects
= nrblocks
/ EXT4_ADDR_PER_BLOCK(inode
->i_sb
);
5650 return indirects
+ 3;
5653 * if nrblocks are not contiguous, worse case, each block touch
5654 * a indirect block, and each indirect block touch a double indirect
5655 * block, plus a triple indirect block
5657 indirects
= nrblocks
* 2 + 1;
5661 static int ext4_index_trans_blocks(struct inode
*inode
, int nrblocks
, int chunk
)
5663 if (!(EXT4_I(inode
)->i_flags
& EXT4_EXTENTS_FL
))
5664 return ext4_indirect_trans_blocks(inode
, nrblocks
, chunk
);
5665 return ext4_ext_index_trans_blocks(inode
, nrblocks
, chunk
);
5669 * Account for index blocks, block groups bitmaps and block group
5670 * descriptor blocks if modify datablocks and index blocks
5671 * worse case, the indexs blocks spread over different block groups
5673 * If datablocks are discontiguous, they are possible to spread over
5674 * different block groups too. If they are contiuguous, with flexbg,
5675 * they could still across block group boundary.
5677 * Also account for superblock, inode, quota and xattr blocks
5679 int ext4_meta_trans_blocks(struct inode
*inode
, int nrblocks
, int chunk
)
5681 ext4_group_t groups
, ngroups
= ext4_get_groups_count(inode
->i_sb
);
5687 * How many index blocks need to touch to modify nrblocks?
5688 * The "Chunk" flag indicating whether the nrblocks is
5689 * physically contiguous on disk
5691 * For Direct IO and fallocate, they calls get_block to allocate
5692 * one single extent at a time, so they could set the "Chunk" flag
5694 idxblocks
= ext4_index_trans_blocks(inode
, nrblocks
, chunk
);
5699 * Now let's see how many group bitmaps and group descriptors need
5709 if (groups
> ngroups
)
5711 if (groups
> EXT4_SB(inode
->i_sb
)->s_gdb_count
)
5712 gdpblocks
= EXT4_SB(inode
->i_sb
)->s_gdb_count
;
5714 /* bitmaps and block group descriptor blocks */
5715 ret
+= groups
+ gdpblocks
;
5717 /* Blocks for super block, inode, quota and xattr blocks */
5718 ret
+= EXT4_META_TRANS_BLOCKS(inode
->i_sb
);
5724 * Calulate the total number of credits to reserve to fit
5725 * the modification of a single pages into a single transaction,
5726 * which may include multiple chunks of block allocations.
5728 * This could be called via ext4_write_begin()
5730 * We need to consider the worse case, when
5731 * one new block per extent.
5733 int ext4_writepage_trans_blocks(struct inode
*inode
)
5735 int bpp
= ext4_journal_blocks_per_page(inode
);
5738 ret
= ext4_meta_trans_blocks(inode
, bpp
, 0);
5740 /* Account for data blocks for journalled mode */
5741 if (ext4_should_journal_data(inode
))
5747 * Calculate the journal credits for a chunk of data modification.
5749 * This is called from DIO, fallocate or whoever calling
5750 * ext4_get_blocks() to map/allocate a chunk of contiguous disk blocks.
5752 * journal buffers for data blocks are not included here, as DIO
5753 * and fallocate do no need to journal data buffers.
5755 int ext4_chunk_trans_blocks(struct inode
*inode
, int nrblocks
)
5757 return ext4_meta_trans_blocks(inode
, nrblocks
, 1);
5761 * The caller must have previously called ext4_reserve_inode_write().
5762 * Give this, we know that the caller already has write access to iloc->bh.
5764 int ext4_mark_iloc_dirty(handle_t
*handle
,
5765 struct inode
*inode
, struct ext4_iloc
*iloc
)
5769 if (test_opt(inode
->i_sb
, I_VERSION
))
5770 inode_inc_iversion(inode
);
5772 /* the do_update_inode consumes one bh->b_count */
5775 /* ext4_do_update_inode() does jbd2_journal_dirty_metadata */
5776 err
= ext4_do_update_inode(handle
, inode
, iloc
);
5782 * On success, We end up with an outstanding reference count against
5783 * iloc->bh. This _must_ be cleaned up later.
5787 ext4_reserve_inode_write(handle_t
*handle
, struct inode
*inode
,
5788 struct ext4_iloc
*iloc
)
5792 err
= ext4_get_inode_loc(inode
, iloc
);
5794 BUFFER_TRACE(iloc
->bh
, "get_write_access");
5795 err
= ext4_journal_get_write_access(handle
, iloc
->bh
);
5801 ext4_std_error(inode
->i_sb
, err
);
5806 * Expand an inode by new_extra_isize bytes.
5807 * Returns 0 on success or negative error number on failure.
5809 static int ext4_expand_extra_isize(struct inode
*inode
,
5810 unsigned int new_extra_isize
,
5811 struct ext4_iloc iloc
,
5814 struct ext4_inode
*raw_inode
;
5815 struct ext4_xattr_ibody_header
*header
;
5816 struct ext4_xattr_entry
*entry
;
5818 if (EXT4_I(inode
)->i_extra_isize
>= new_extra_isize
)
5821 raw_inode
= ext4_raw_inode(&iloc
);
5823 header
= IHDR(inode
, raw_inode
);
5824 entry
= IFIRST(header
);
5826 /* No extended attributes present */
5827 if (!ext4_test_inode_state(inode
, EXT4_STATE_XATTR
) ||
5828 header
->h_magic
!= cpu_to_le32(EXT4_XATTR_MAGIC
)) {
5829 memset((void *)raw_inode
+ EXT4_GOOD_OLD_INODE_SIZE
, 0,
5831 EXT4_I(inode
)->i_extra_isize
= new_extra_isize
;
5835 /* try to expand with EAs present */
5836 return ext4_expand_extra_isize_ea(inode
, new_extra_isize
,
5841 * What we do here is to mark the in-core inode as clean with respect to inode
5842 * dirtiness (it may still be data-dirty).
5843 * This means that the in-core inode may be reaped by prune_icache
5844 * without having to perform any I/O. This is a very good thing,
5845 * because *any* task may call prune_icache - even ones which
5846 * have a transaction open against a different journal.
5848 * Is this cheating? Not really. Sure, we haven't written the
5849 * inode out, but prune_icache isn't a user-visible syncing function.
5850 * Whenever the user wants stuff synced (sys_sync, sys_msync, sys_fsync)
5851 * we start and wait on commits.
5853 * Is this efficient/effective? Well, we're being nice to the system
5854 * by cleaning up our inodes proactively so they can be reaped
5855 * without I/O. But we are potentially leaving up to five seconds'
5856 * worth of inodes floating about which prune_icache wants us to
5857 * write out. One way to fix that would be to get prune_icache()
5858 * to do a write_super() to free up some memory. It has the desired
5861 int ext4_mark_inode_dirty(handle_t
*handle
, struct inode
*inode
)
5863 struct ext4_iloc iloc
;
5864 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
5865 static unsigned int mnt_count
;
5869 err
= ext4_reserve_inode_write(handle
, inode
, &iloc
);
5870 if (ext4_handle_valid(handle
) &&
5871 EXT4_I(inode
)->i_extra_isize
< sbi
->s_want_extra_isize
&&
5872 !ext4_test_inode_state(inode
, EXT4_STATE_NO_EXPAND
)) {
5874 * We need extra buffer credits since we may write into EA block
5875 * with this same handle. If journal_extend fails, then it will
5876 * only result in a minor loss of functionality for that inode.
5877 * If this is felt to be critical, then e2fsck should be run to
5878 * force a large enough s_min_extra_isize.
5880 if ((jbd2_journal_extend(handle
,
5881 EXT4_DATA_TRANS_BLOCKS(inode
->i_sb
))) == 0) {
5882 ret
= ext4_expand_extra_isize(inode
,
5883 sbi
->s_want_extra_isize
,
5886 ext4_set_inode_state(inode
,
5887 EXT4_STATE_NO_EXPAND
);
5889 le16_to_cpu(sbi
->s_es
->s_mnt_count
)) {
5890 ext4_warning(inode
->i_sb
,
5891 "Unable to expand inode %lu. Delete"
5892 " some EAs or run e2fsck.",
5895 le16_to_cpu(sbi
->s_es
->s_mnt_count
);
5901 err
= ext4_mark_iloc_dirty(handle
, inode
, &iloc
);
5906 * ext4_dirty_inode() is called from __mark_inode_dirty()
5908 * We're really interested in the case where a file is being extended.
5909 * i_size has been changed by generic_commit_write() and we thus need
5910 * to include the updated inode in the current transaction.
5912 * Also, dquot_alloc_block() will always dirty the inode when blocks
5913 * are allocated to the file.
5915 * If the inode is marked synchronous, we don't honour that here - doing
5916 * so would cause a commit on atime updates, which we don't bother doing.
5917 * We handle synchronous inodes at the highest possible level.
5919 void ext4_dirty_inode(struct inode
*inode
)
5923 handle
= ext4_journal_start(inode
, 2);
5927 ext4_mark_inode_dirty(handle
, inode
);
5929 ext4_journal_stop(handle
);
5936 * Bind an inode's backing buffer_head into this transaction, to prevent
5937 * it from being flushed to disk early. Unlike
5938 * ext4_reserve_inode_write, this leaves behind no bh reference and
5939 * returns no iloc structure, so the caller needs to repeat the iloc
5940 * lookup to mark the inode dirty later.
5942 static int ext4_pin_inode(handle_t
*handle
, struct inode
*inode
)
5944 struct ext4_iloc iloc
;
5948 err
= ext4_get_inode_loc(inode
, &iloc
);
5950 BUFFER_TRACE(iloc
.bh
, "get_write_access");
5951 err
= jbd2_journal_get_write_access(handle
, iloc
.bh
);
5953 err
= ext4_handle_dirty_metadata(handle
,
5959 ext4_std_error(inode
->i_sb
, err
);
5964 int ext4_change_inode_journal_flag(struct inode
*inode
, int val
)
5971 * We have to be very careful here: changing a data block's
5972 * journaling status dynamically is dangerous. If we write a
5973 * data block to the journal, change the status and then delete
5974 * that block, we risk forgetting to revoke the old log record
5975 * from the journal and so a subsequent replay can corrupt data.
5976 * So, first we make sure that the journal is empty and that
5977 * nobody is changing anything.
5980 journal
= EXT4_JOURNAL(inode
);
5983 if (is_journal_aborted(journal
))
5986 jbd2_journal_lock_updates(journal
);
5987 jbd2_journal_flush(journal
);
5990 * OK, there are no updates running now, and all cached data is
5991 * synced to disk. We are now in a completely consistent state
5992 * which doesn't have anything in the journal, and we know that
5993 * no filesystem updates are running, so it is safe to modify
5994 * the inode's in-core data-journaling state flag now.
5998 EXT4_I(inode
)->i_flags
|= EXT4_JOURNAL_DATA_FL
;
6000 EXT4_I(inode
)->i_flags
&= ~EXT4_JOURNAL_DATA_FL
;
6001 ext4_set_aops(inode
);
6003 jbd2_journal_unlock_updates(journal
);
6005 /* Finally we can mark the inode as dirty. */
6007 handle
= ext4_journal_start(inode
, 1);
6009 return PTR_ERR(handle
);
6011 err
= ext4_mark_inode_dirty(handle
, inode
);
6012 ext4_handle_sync(handle
);
6013 ext4_journal_stop(handle
);
6014 ext4_std_error(inode
->i_sb
, err
);
6019 static int ext4_bh_unmapped(handle_t
*handle
, struct buffer_head
*bh
)
6021 return !buffer_mapped(bh
);
6024 int ext4_page_mkwrite(struct vm_area_struct
*vma
, struct vm_fault
*vmf
)
6026 struct page
*page
= vmf
->page
;
6031 struct file
*file
= vma
->vm_file
;
6032 struct inode
*inode
= file
->f_path
.dentry
->d_inode
;
6033 struct address_space
*mapping
= inode
->i_mapping
;
6036 * Get i_alloc_sem to stop truncates messing with the inode. We cannot
6037 * get i_mutex because we are already holding mmap_sem.
6039 down_read(&inode
->i_alloc_sem
);
6040 size
= i_size_read(inode
);
6041 if (page
->mapping
!= mapping
|| size
<= page_offset(page
)
6042 || !PageUptodate(page
)) {
6043 /* page got truncated from under us? */
6047 if (PageMappedToDisk(page
))
6050 if (page
->index
== size
>> PAGE_CACHE_SHIFT
)
6051 len
= size
& ~PAGE_CACHE_MASK
;
6053 len
= PAGE_CACHE_SIZE
;
6057 * return if we have all the buffers mapped. This avoid
6058 * the need to call write_begin/write_end which does a
6059 * journal_start/journal_stop which can block and take
6062 if (page_has_buffers(page
)) {
6063 if (!walk_page_buffers(NULL
, page_buffers(page
), 0, len
, NULL
,
6064 ext4_bh_unmapped
)) {
6071 * OK, we need to fill the hole... Do write_begin write_end
6072 * to do block allocation/reservation.We are not holding
6073 * inode.i__mutex here. That allow * parallel write_begin,
6074 * write_end call. lock_page prevent this from happening
6075 * on the same page though
6077 ret
= mapping
->a_ops
->write_begin(file
, mapping
, page_offset(page
),
6078 len
, AOP_FLAG_UNINTERRUPTIBLE
, &page
, &fsdata
);
6081 ret
= mapping
->a_ops
->write_end(file
, mapping
, page_offset(page
),
6082 len
, len
, page
, fsdata
);
6088 ret
= VM_FAULT_SIGBUS
;
6089 up_read(&inode
->i_alloc_sem
);