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_map_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(function
, inode
,
352 "invalid block reference %u", blk
);
360 #define ext4_check_indirect_blockref(inode, bh) \
361 __ext4_check_blockref(__func__, inode, (__le32 *)(bh)->b_data, \
362 EXT4_ADDR_PER_BLOCK((inode)->i_sb))
364 #define ext4_check_inode_blockref(inode) \
365 __ext4_check_blockref(__func__, inode, EXT4_I(inode)->i_data, \
369 * ext4_get_branch - read the chain of indirect blocks leading to data
370 * @inode: inode in question
371 * @depth: depth of the chain (1 - direct pointer, etc.)
372 * @offsets: offsets of pointers in inode/indirect blocks
373 * @chain: place to store the result
374 * @err: here we store the error value
376 * Function fills the array of triples <key, p, bh> and returns %NULL
377 * if everything went OK or the pointer to the last filled triple
378 * (incomplete one) otherwise. Upon the return chain[i].key contains
379 * the number of (i+1)-th block in the chain (as it is stored in memory,
380 * i.e. little-endian 32-bit), chain[i].p contains the address of that
381 * number (it points into struct inode for i==0 and into the bh->b_data
382 * for i>0) and chain[i].bh points to the buffer_head of i-th indirect
383 * block for i>0 and NULL for i==0. In other words, it holds the block
384 * numbers of the chain, addresses they were taken from (and where we can
385 * verify that chain did not change) and buffer_heads hosting these
388 * Function stops when it stumbles upon zero pointer (absent block)
389 * (pointer to last triple returned, *@err == 0)
390 * or when it gets an IO error reading an indirect block
391 * (ditto, *@err == -EIO)
392 * or when it reads all @depth-1 indirect blocks successfully and finds
393 * the whole chain, all way to the data (returns %NULL, *err == 0).
395 * Need to be called with
396 * down_read(&EXT4_I(inode)->i_data_sem)
398 static Indirect
*ext4_get_branch(struct inode
*inode
, int depth
,
399 ext4_lblk_t
*offsets
,
400 Indirect chain
[4], int *err
)
402 struct super_block
*sb
= inode
->i_sb
;
404 struct buffer_head
*bh
;
407 /* i_data is not going away, no lock needed */
408 add_chain(chain
, NULL
, EXT4_I(inode
)->i_data
+ *offsets
);
412 bh
= sb_getblk(sb
, le32_to_cpu(p
->key
));
416 if (!bh_uptodate_or_lock(bh
)) {
417 if (bh_submit_read(bh
) < 0) {
421 /* validate block references */
422 if (ext4_check_indirect_blockref(inode
, bh
)) {
428 add_chain(++p
, bh
, (__le32
*)bh
->b_data
+ *++offsets
);
442 * ext4_find_near - find a place for allocation with sufficient locality
444 * @ind: descriptor of indirect block.
446 * This function returns the preferred place for block allocation.
447 * It is used when heuristic for sequential allocation fails.
449 * + if there is a block to the left of our position - allocate near it.
450 * + if pointer will live in indirect block - allocate near that block.
451 * + if pointer will live in inode - allocate in the same
454 * In the latter case we colour the starting block by the callers PID to
455 * prevent it from clashing with concurrent allocations for a different inode
456 * in the same block group. The PID is used here so that functionally related
457 * files will be close-by on-disk.
459 * Caller must make sure that @ind is valid and will stay that way.
461 static ext4_fsblk_t
ext4_find_near(struct inode
*inode
, Indirect
*ind
)
463 struct ext4_inode_info
*ei
= EXT4_I(inode
);
464 __le32
*start
= ind
->bh
? (__le32
*) ind
->bh
->b_data
: ei
->i_data
;
466 ext4_fsblk_t bg_start
;
467 ext4_fsblk_t last_block
;
468 ext4_grpblk_t colour
;
469 ext4_group_t block_group
;
470 int flex_size
= ext4_flex_bg_size(EXT4_SB(inode
->i_sb
));
472 /* Try to find previous block */
473 for (p
= ind
->p
- 1; p
>= start
; p
--) {
475 return le32_to_cpu(*p
);
478 /* No such thing, so let's try location of indirect block */
480 return ind
->bh
->b_blocknr
;
483 * It is going to be referred to from the inode itself? OK, just put it
484 * into the same cylinder group then.
486 block_group
= ei
->i_block_group
;
487 if (flex_size
>= EXT4_FLEX_SIZE_DIR_ALLOC_SCHEME
) {
488 block_group
&= ~(flex_size
-1);
489 if (S_ISREG(inode
->i_mode
))
492 bg_start
= ext4_group_first_block_no(inode
->i_sb
, block_group
);
493 last_block
= ext4_blocks_count(EXT4_SB(inode
->i_sb
)->s_es
) - 1;
496 * If we are doing delayed allocation, we don't need take
497 * colour into account.
499 if (test_opt(inode
->i_sb
, DELALLOC
))
502 if (bg_start
+ EXT4_BLOCKS_PER_GROUP(inode
->i_sb
) <= last_block
)
503 colour
= (current
->pid
% 16) *
504 (EXT4_BLOCKS_PER_GROUP(inode
->i_sb
) / 16);
506 colour
= (current
->pid
% 16) * ((last_block
- bg_start
) / 16);
507 return bg_start
+ colour
;
511 * ext4_find_goal - find a preferred place for allocation.
513 * @block: block we want
514 * @partial: pointer to the last triple within a chain
516 * Normally this function find the preferred place for block allocation,
518 * Because this is only used for non-extent files, we limit the block nr
521 static ext4_fsblk_t
ext4_find_goal(struct inode
*inode
, ext4_lblk_t block
,
527 * XXX need to get goal block from mballoc's data structures
530 goal
= ext4_find_near(inode
, partial
);
531 goal
= goal
& EXT4_MAX_BLOCK_FILE_PHYS
;
536 * ext4_blks_to_allocate: Look up the block map and count the number
537 * of direct blocks need to be allocated for the given branch.
539 * @branch: chain of indirect blocks
540 * @k: number of blocks need for indirect blocks
541 * @blks: number of data blocks to be mapped.
542 * @blocks_to_boundary: the offset in the indirect block
544 * return the total number of blocks to be allocate, including the
545 * direct and indirect blocks.
547 static int ext4_blks_to_allocate(Indirect
*branch
, int k
, unsigned int blks
,
548 int blocks_to_boundary
)
550 unsigned int count
= 0;
553 * Simple case, [t,d]Indirect block(s) has not allocated yet
554 * then it's clear blocks on that path have not allocated
557 /* right now we don't handle cross boundary allocation */
558 if (blks
< blocks_to_boundary
+ 1)
561 count
+= blocks_to_boundary
+ 1;
566 while (count
< blks
&& count
<= blocks_to_boundary
&&
567 le32_to_cpu(*(branch
[0].p
+ count
)) == 0) {
574 * ext4_alloc_blocks: multiple allocate blocks needed for a branch
575 * @indirect_blks: the number of blocks need to allocate for indirect
578 * @new_blocks: on return it will store the new block numbers for
579 * the indirect blocks(if needed) and the first direct block,
580 * @blks: on return it will store the total number of allocated
583 static int ext4_alloc_blocks(handle_t
*handle
, struct inode
*inode
,
584 ext4_lblk_t iblock
, ext4_fsblk_t goal
,
585 int indirect_blks
, int blks
,
586 ext4_fsblk_t new_blocks
[4], int *err
)
588 struct ext4_allocation_request ar
;
590 unsigned long count
= 0, blk_allocated
= 0;
592 ext4_fsblk_t current_block
= 0;
596 * Here we try to allocate the requested multiple blocks at once,
597 * on a best-effort basis.
598 * To build a branch, we should allocate blocks for
599 * the indirect blocks(if not allocated yet), and at least
600 * the first direct block of this branch. That's the
601 * minimum number of blocks need to allocate(required)
603 /* first we try to allocate the indirect blocks */
604 target
= indirect_blks
;
607 /* allocating blocks for indirect blocks and direct blocks */
608 current_block
= ext4_new_meta_blocks(handle
, inode
,
613 if (unlikely(current_block
+ count
> EXT4_MAX_BLOCK_FILE_PHYS
)) {
614 EXT4_ERROR_INODE(inode
,
615 "current_block %llu + count %lu > %d!",
616 current_block
, count
,
617 EXT4_MAX_BLOCK_FILE_PHYS
);
623 /* allocate blocks for indirect blocks */
624 while (index
< indirect_blks
&& count
) {
625 new_blocks
[index
++] = current_block
++;
630 * save the new block number
631 * for the first direct block
633 new_blocks
[index
] = current_block
;
634 printk(KERN_INFO
"%s returned more blocks than "
635 "requested\n", __func__
);
641 target
= blks
- count
;
642 blk_allocated
= count
;
645 /* Now allocate data blocks */
646 memset(&ar
, 0, sizeof(ar
));
651 if (S_ISREG(inode
->i_mode
))
652 /* enable in-core preallocation only for regular files */
653 ar
.flags
= EXT4_MB_HINT_DATA
;
655 current_block
= ext4_mb_new_blocks(handle
, &ar
, err
);
656 if (unlikely(current_block
+ ar
.len
> EXT4_MAX_BLOCK_FILE_PHYS
)) {
657 EXT4_ERROR_INODE(inode
,
658 "current_block %llu + ar.len %d > %d!",
659 current_block
, ar
.len
,
660 EXT4_MAX_BLOCK_FILE_PHYS
);
665 if (*err
&& (target
== blks
)) {
667 * if the allocation failed and we didn't allocate
673 if (target
== blks
) {
675 * save the new block number
676 * for the first direct block
678 new_blocks
[index
] = current_block
;
680 blk_allocated
+= ar
.len
;
683 /* total number of blocks allocated for direct blocks */
688 for (i
= 0; i
< index
; i
++)
689 ext4_free_blocks(handle
, inode
, 0, new_blocks
[i
], 1, 0);
694 * ext4_alloc_branch - allocate and set up a chain of blocks.
696 * @indirect_blks: number of allocated indirect blocks
697 * @blks: number of allocated direct blocks
698 * @offsets: offsets (in the blocks) to store the pointers to next.
699 * @branch: place to store the chain in.
701 * This function allocates blocks, zeroes out all but the last one,
702 * links them into chain and (if we are synchronous) writes them to disk.
703 * In other words, it prepares a branch that can be spliced onto the
704 * inode. It stores the information about that chain in the branch[], in
705 * the same format as ext4_get_branch() would do. We are calling it after
706 * we had read the existing part of chain and partial points to the last
707 * triple of that (one with zero ->key). Upon the exit we have the same
708 * picture as after the successful ext4_get_block(), except that in one
709 * place chain is disconnected - *branch->p is still zero (we did not
710 * set the last link), but branch->key contains the number that should
711 * be placed into *branch->p to fill that gap.
713 * If allocation fails we free all blocks we've allocated (and forget
714 * their buffer_heads) and return the error value the from failed
715 * ext4_alloc_block() (normally -ENOSPC). Otherwise we set the chain
716 * as described above and return 0.
718 static int ext4_alloc_branch(handle_t
*handle
, struct inode
*inode
,
719 ext4_lblk_t iblock
, int indirect_blks
,
720 int *blks
, ext4_fsblk_t goal
,
721 ext4_lblk_t
*offsets
, Indirect
*branch
)
723 int blocksize
= inode
->i_sb
->s_blocksize
;
726 struct buffer_head
*bh
;
728 ext4_fsblk_t new_blocks
[4];
729 ext4_fsblk_t current_block
;
731 num
= ext4_alloc_blocks(handle
, inode
, iblock
, goal
, indirect_blks
,
732 *blks
, new_blocks
, &err
);
736 branch
[0].key
= cpu_to_le32(new_blocks
[0]);
738 * metadata blocks and data blocks are allocated.
740 for (n
= 1; n
<= indirect_blks
; n
++) {
742 * Get buffer_head for parent block, zero it out
743 * and set the pointer to new one, then send
746 bh
= sb_getblk(inode
->i_sb
, new_blocks
[n
-1]);
749 BUFFER_TRACE(bh
, "call get_create_access");
750 err
= ext4_journal_get_create_access(handle
, bh
);
752 /* Don't brelse(bh) here; it's done in
753 * ext4_journal_forget() below */
758 memset(bh
->b_data
, 0, blocksize
);
759 branch
[n
].p
= (__le32
*) bh
->b_data
+ offsets
[n
];
760 branch
[n
].key
= cpu_to_le32(new_blocks
[n
]);
761 *branch
[n
].p
= branch
[n
].key
;
762 if (n
== indirect_blks
) {
763 current_block
= new_blocks
[n
];
765 * End of chain, update the last new metablock of
766 * the chain to point to the new allocated
767 * data blocks numbers
769 for (i
= 1; i
< num
; i
++)
770 *(branch
[n
].p
+ i
) = cpu_to_le32(++current_block
);
772 BUFFER_TRACE(bh
, "marking uptodate");
773 set_buffer_uptodate(bh
);
776 BUFFER_TRACE(bh
, "call ext4_handle_dirty_metadata");
777 err
= ext4_handle_dirty_metadata(handle
, inode
, bh
);
784 /* Allocation failed, free what we already allocated */
785 ext4_free_blocks(handle
, inode
, 0, new_blocks
[0], 1, 0);
786 for (i
= 1; i
<= n
; i
++) {
788 * branch[i].bh is newly allocated, so there is no
789 * need to revoke the block, which is why we don't
790 * need to set EXT4_FREE_BLOCKS_METADATA.
792 ext4_free_blocks(handle
, inode
, 0, new_blocks
[i
], 1,
793 EXT4_FREE_BLOCKS_FORGET
);
795 for (i
= n
+1; i
< indirect_blks
; i
++)
796 ext4_free_blocks(handle
, inode
, 0, new_blocks
[i
], 1, 0);
798 ext4_free_blocks(handle
, inode
, 0, new_blocks
[i
], num
, 0);
804 * ext4_splice_branch - splice the allocated branch onto inode.
806 * @block: (logical) number of block we are adding
807 * @chain: chain of indirect blocks (with a missing link - see
809 * @where: location of missing link
810 * @num: number of indirect blocks we are adding
811 * @blks: number of direct blocks we are adding
813 * This function fills the missing link and does all housekeeping needed in
814 * inode (->i_blocks, etc.). In case of success we end up with the full
815 * chain to new block and return 0.
817 static int ext4_splice_branch(handle_t
*handle
, struct inode
*inode
,
818 ext4_lblk_t block
, Indirect
*where
, int num
,
823 ext4_fsblk_t current_block
;
826 * If we're splicing into a [td]indirect block (as opposed to the
827 * inode) then we need to get write access to the [td]indirect block
831 BUFFER_TRACE(where
->bh
, "get_write_access");
832 err
= ext4_journal_get_write_access(handle
, where
->bh
);
838 *where
->p
= where
->key
;
841 * Update the host buffer_head or inode to point to more just allocated
842 * direct blocks blocks
844 if (num
== 0 && blks
> 1) {
845 current_block
= le32_to_cpu(where
->key
) + 1;
846 for (i
= 1; i
< blks
; i
++)
847 *(where
->p
+ i
) = cpu_to_le32(current_block
++);
850 /* We are done with atomic stuff, now do the rest of housekeeping */
851 /* had we spliced it onto indirect block? */
854 * If we spliced it onto an indirect block, we haven't
855 * altered the inode. Note however that if it is being spliced
856 * onto an indirect block at the very end of the file (the
857 * file is growing) then we *will* alter the inode to reflect
858 * the new i_size. But that is not done here - it is done in
859 * generic_commit_write->__mark_inode_dirty->ext4_dirty_inode.
861 jbd_debug(5, "splicing indirect only\n");
862 BUFFER_TRACE(where
->bh
, "call ext4_handle_dirty_metadata");
863 err
= ext4_handle_dirty_metadata(handle
, inode
, where
->bh
);
868 * OK, we spliced it into the inode itself on a direct block.
870 ext4_mark_inode_dirty(handle
, inode
);
871 jbd_debug(5, "splicing direct\n");
876 for (i
= 1; i
<= num
; i
++) {
878 * branch[i].bh is newly allocated, so there is no
879 * need to revoke the block, which is why we don't
880 * need to set EXT4_FREE_BLOCKS_METADATA.
882 ext4_free_blocks(handle
, inode
, where
[i
].bh
, 0, 1,
883 EXT4_FREE_BLOCKS_FORGET
);
885 ext4_free_blocks(handle
, inode
, 0, le32_to_cpu(where
[num
].key
),
892 * The ext4_ind_map_blocks() function handles non-extents inodes
893 * (i.e., using the traditional indirect/double-indirect i_blocks
894 * scheme) for ext4_map_blocks().
896 * Allocation strategy is simple: if we have to allocate something, we will
897 * have to go the whole way to leaf. So let's do it before attaching anything
898 * to tree, set linkage between the newborn blocks, write them if sync is
899 * required, recheck the path, free and repeat if check fails, otherwise
900 * set the last missing link (that will protect us from any truncate-generated
901 * removals - all blocks on the path are immune now) and possibly force the
902 * write on the parent block.
903 * That has a nice additional property: no special recovery from the failed
904 * allocations is needed - we simply release blocks and do not touch anything
905 * reachable from inode.
907 * `handle' can be NULL if create == 0.
909 * return > 0, # of blocks mapped or allocated.
910 * return = 0, if plain lookup failed.
911 * return < 0, error case.
913 * The ext4_ind_get_blocks() function should be called with
914 * down_write(&EXT4_I(inode)->i_data_sem) if allocating filesystem
915 * blocks (i.e., flags has EXT4_GET_BLOCKS_CREATE set) or
916 * down_read(&EXT4_I(inode)->i_data_sem) if not allocating file system
919 static int ext4_ind_map_blocks(handle_t
*handle
, struct inode
*inode
,
920 struct ext4_map_blocks
*map
,
924 ext4_lblk_t offsets
[4];
929 int blocks_to_boundary
= 0;
932 ext4_fsblk_t first_block
= 0;
934 J_ASSERT(!(ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
)));
935 J_ASSERT(handle
!= NULL
|| (flags
& EXT4_GET_BLOCKS_CREATE
) == 0);
936 depth
= ext4_block_to_path(inode
, map
->m_lblk
, offsets
,
937 &blocks_to_boundary
);
942 partial
= ext4_get_branch(inode
, depth
, offsets
, chain
, &err
);
944 /* Simplest case - block found, no allocation needed */
946 first_block
= le32_to_cpu(chain
[depth
- 1].key
);
949 while (count
< map
->m_len
&& count
<= blocks_to_boundary
) {
952 blk
= le32_to_cpu(*(chain
[depth
-1].p
+ count
));
954 if (blk
== first_block
+ count
)
962 /* Next simple case - plain lookup or failed read of indirect block */
963 if ((flags
& EXT4_GET_BLOCKS_CREATE
) == 0 || err
== -EIO
)
967 * Okay, we need to do block allocation.
969 goal
= ext4_find_goal(inode
, map
->m_lblk
, partial
);
971 /* the number of blocks need to allocate for [d,t]indirect blocks */
972 indirect_blks
= (chain
+ depth
) - partial
- 1;
975 * Next look up the indirect map to count the totoal number of
976 * direct blocks to allocate for this branch.
978 count
= ext4_blks_to_allocate(partial
, indirect_blks
,
979 map
->m_len
, blocks_to_boundary
);
981 * Block out ext4_truncate while we alter the tree
983 err
= ext4_alloc_branch(handle
, inode
, map
->m_lblk
, indirect_blks
,
985 offsets
+ (partial
- chain
), partial
);
988 * The ext4_splice_branch call will free and forget any buffers
989 * on the new chain if there is a failure, but that risks using
990 * up transaction credits, especially for bitmaps where the
991 * credits cannot be returned. Can we handle this somehow? We
992 * may need to return -EAGAIN upwards in the worst case. --sct
995 err
= ext4_splice_branch(handle
, inode
, map
->m_lblk
,
996 partial
, indirect_blks
, count
);
1000 map
->m_flags
|= EXT4_MAP_NEW
;
1002 ext4_update_inode_fsync_trans(handle
, inode
, 1);
1004 map
->m_flags
|= EXT4_MAP_MAPPED
;
1005 map
->m_pblk
= le32_to_cpu(chain
[depth
-1].key
);
1007 if (count
> blocks_to_boundary
)
1008 map
->m_flags
|= EXT4_MAP_BOUNDARY
;
1010 /* Clean up and exit */
1011 partial
= chain
+ depth
- 1; /* the whole chain */
1013 while (partial
> chain
) {
1014 BUFFER_TRACE(partial
->bh
, "call brelse");
1015 brelse(partial
->bh
);
1023 qsize_t
*ext4_get_reserved_space(struct inode
*inode
)
1025 return &EXT4_I(inode
)->i_reserved_quota
;
1030 * Calculate the number of metadata blocks need to reserve
1031 * to allocate a new block at @lblocks for non extent file based file
1033 static int ext4_indirect_calc_metadata_amount(struct inode
*inode
,
1036 struct ext4_inode_info
*ei
= EXT4_I(inode
);
1037 sector_t dind_mask
= ~((sector_t
)EXT4_ADDR_PER_BLOCK(inode
->i_sb
) - 1);
1040 if (lblock
< EXT4_NDIR_BLOCKS
)
1043 lblock
-= EXT4_NDIR_BLOCKS
;
1045 if (ei
->i_da_metadata_calc_len
&&
1046 (lblock
& dind_mask
) == ei
->i_da_metadata_calc_last_lblock
) {
1047 ei
->i_da_metadata_calc_len
++;
1050 ei
->i_da_metadata_calc_last_lblock
= lblock
& dind_mask
;
1051 ei
->i_da_metadata_calc_len
= 1;
1052 blk_bits
= order_base_2(lblock
);
1053 return (blk_bits
/ EXT4_ADDR_PER_BLOCK_BITS(inode
->i_sb
)) + 1;
1057 * Calculate the number of metadata blocks need to reserve
1058 * to allocate a block located at @lblock
1060 static int ext4_calc_metadata_amount(struct inode
*inode
, sector_t lblock
)
1062 if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
))
1063 return ext4_ext_calc_metadata_amount(inode
, lblock
);
1065 return ext4_indirect_calc_metadata_amount(inode
, lblock
);
1069 * Called with i_data_sem down, which is important since we can call
1070 * ext4_discard_preallocations() from here.
1072 void ext4_da_update_reserve_space(struct inode
*inode
,
1073 int used
, int quota_claim
)
1075 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
1076 struct ext4_inode_info
*ei
= EXT4_I(inode
);
1078 spin_lock(&ei
->i_block_reservation_lock
);
1079 trace_ext4_da_update_reserve_space(inode
, used
);
1080 if (unlikely(used
> ei
->i_reserved_data_blocks
)) {
1081 ext4_msg(inode
->i_sb
, KERN_NOTICE
, "%s: ino %lu, used %d "
1082 "with only %d reserved data blocks\n",
1083 __func__
, inode
->i_ino
, used
,
1084 ei
->i_reserved_data_blocks
);
1086 used
= ei
->i_reserved_data_blocks
;
1089 /* Update per-inode reservations */
1090 ei
->i_reserved_data_blocks
-= used
;
1091 ei
->i_reserved_meta_blocks
-= ei
->i_allocated_meta_blocks
;
1092 percpu_counter_sub(&sbi
->s_dirtyblocks_counter
,
1093 used
+ ei
->i_allocated_meta_blocks
);
1094 ei
->i_allocated_meta_blocks
= 0;
1096 if (ei
->i_reserved_data_blocks
== 0) {
1098 * We can release all of the reserved metadata blocks
1099 * only when we have written all of the delayed
1100 * allocation blocks.
1102 percpu_counter_sub(&sbi
->s_dirtyblocks_counter
,
1103 ei
->i_reserved_meta_blocks
);
1104 ei
->i_reserved_meta_blocks
= 0;
1105 ei
->i_da_metadata_calc_len
= 0;
1107 spin_unlock(&EXT4_I(inode
)->i_block_reservation_lock
);
1109 /* Update quota subsystem for data blocks */
1111 dquot_claim_block(inode
, used
);
1114 * We did fallocate with an offset that is already delayed
1115 * allocated. So on delayed allocated writeback we should
1116 * not re-claim the quota for fallocated blocks.
1118 dquot_release_reservation_block(inode
, used
);
1122 * If we have done all the pending block allocations and if
1123 * there aren't any writers on the inode, we can discard the
1124 * inode's preallocations.
1126 if ((ei
->i_reserved_data_blocks
== 0) &&
1127 (atomic_read(&inode
->i_writecount
) == 0))
1128 ext4_discard_preallocations(inode
);
1131 static int __check_block_validity(struct inode
*inode
, const char *func
,
1132 struct ext4_map_blocks
*map
)
1134 if (!ext4_data_block_valid(EXT4_SB(inode
->i_sb
), map
->m_pblk
,
1136 ext4_error_inode(func
, inode
,
1137 "lblock %lu mapped to illegal pblock %llu "
1138 "(length %d)", (unsigned long) map
->m_lblk
,
1139 map
->m_pblk
, map
->m_len
);
1145 #define check_block_validity(inode, map) \
1146 __check_block_validity((inode), __func__, (map))
1149 * Return the number of contiguous dirty pages in a given inode
1150 * starting at page frame idx.
1152 static pgoff_t
ext4_num_dirty_pages(struct inode
*inode
, pgoff_t idx
,
1153 unsigned int max_pages
)
1155 struct address_space
*mapping
= inode
->i_mapping
;
1157 struct pagevec pvec
;
1159 int i
, nr_pages
, done
= 0;
1163 pagevec_init(&pvec
, 0);
1166 nr_pages
= pagevec_lookup_tag(&pvec
, mapping
, &index
,
1167 PAGECACHE_TAG_DIRTY
,
1168 (pgoff_t
)PAGEVEC_SIZE
);
1171 for (i
= 0; i
< nr_pages
; i
++) {
1172 struct page
*page
= pvec
.pages
[i
];
1173 struct buffer_head
*bh
, *head
;
1176 if (unlikely(page
->mapping
!= mapping
) ||
1178 PageWriteback(page
) ||
1179 page
->index
!= idx
) {
1184 if (page_has_buffers(page
)) {
1185 bh
= head
= page_buffers(page
);
1187 if (!buffer_delay(bh
) &&
1188 !buffer_unwritten(bh
))
1190 bh
= bh
->b_this_page
;
1191 } while (!done
&& (bh
!= head
));
1198 if (num
>= max_pages
)
1201 pagevec_release(&pvec
);
1207 * The ext4_map_blocks() function tries to look up the requested blocks,
1208 * and returns if the blocks are already mapped.
1210 * Otherwise it takes the write lock of the i_data_sem and allocate blocks
1211 * and store the allocated blocks in the result buffer head and mark it
1214 * If file type is extents based, it will call ext4_ext_map_blocks(),
1215 * Otherwise, call with ext4_ind_map_blocks() to handle indirect mapping
1218 * On success, it returns the number of blocks being mapped or allocate.
1219 * if create==0 and the blocks are pre-allocated and uninitialized block,
1220 * the result buffer head is unmapped. If the create ==1, it will make sure
1221 * the buffer head is mapped.
1223 * It returns 0 if plain look up failed (blocks have not been allocated), in
1224 * that casem, buffer head is unmapped
1226 * It returns the error in case of allocation failure.
1228 int ext4_map_blocks(handle_t
*handle
, struct inode
*inode
,
1229 struct ext4_map_blocks
*map
, int flags
)
1234 ext_debug("ext4_map_blocks(): inode %lu, flag %d, max_blocks %u,"
1235 "logical block %lu\n", inode
->i_ino
, flags
, map
->m_len
,
1236 (unsigned long) map
->m_lblk
);
1238 * Try to see if we can get the block without requesting a new
1239 * file system block.
1241 down_read((&EXT4_I(inode
)->i_data_sem
));
1242 if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
)) {
1243 retval
= ext4_ext_map_blocks(handle
, inode
, map
, 0);
1245 retval
= ext4_ind_map_blocks(handle
, inode
, map
, 0);
1247 up_read((&EXT4_I(inode
)->i_data_sem
));
1249 if (retval
> 0 && map
->m_flags
& EXT4_MAP_MAPPED
) {
1250 int ret
= check_block_validity(inode
, map
);
1255 /* If it is only a block(s) look up */
1256 if ((flags
& EXT4_GET_BLOCKS_CREATE
) == 0)
1260 * Returns if the blocks have already allocated
1262 * Note that if blocks have been preallocated
1263 * ext4_ext_get_block() returns th create = 0
1264 * with buffer head unmapped.
1266 if (retval
> 0 && map
->m_flags
& EXT4_MAP_MAPPED
)
1270 * When we call get_blocks without the create flag, the
1271 * BH_Unwritten flag could have gotten set if the blocks
1272 * requested were part of a uninitialized extent. We need to
1273 * clear this flag now that we are committed to convert all or
1274 * part of the uninitialized extent to be an initialized
1275 * extent. This is because we need to avoid the combination
1276 * of BH_Unwritten and BH_Mapped flags being simultaneously
1277 * set on the buffer_head.
1279 map
->m_flags
&= ~EXT4_MAP_UNWRITTEN
;
1282 * New blocks allocate and/or writing to uninitialized extent
1283 * will possibly result in updating i_data, so we take
1284 * the write lock of i_data_sem, and call get_blocks()
1285 * with create == 1 flag.
1287 down_write((&EXT4_I(inode
)->i_data_sem
));
1290 * if the caller is from delayed allocation writeout path
1291 * we have already reserved fs blocks for allocation
1292 * let the underlying get_block() function know to
1293 * avoid double accounting
1295 if (flags
& EXT4_GET_BLOCKS_DELALLOC_RESERVE
)
1296 EXT4_I(inode
)->i_delalloc_reserved_flag
= 1;
1298 * We need to check for EXT4 here because migrate
1299 * could have changed the inode type in between
1301 if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
)) {
1302 retval
= ext4_ext_map_blocks(handle
, inode
, map
, flags
);
1304 retval
= ext4_ind_map_blocks(handle
, inode
, map
, flags
);
1306 if (retval
> 0 && map
->m_flags
& EXT4_MAP_NEW
) {
1308 * We allocated new blocks which will result in
1309 * i_data's format changing. Force the migrate
1310 * to fail by clearing migrate flags
1312 ext4_clear_inode_state(inode
, EXT4_STATE_EXT_MIGRATE
);
1316 * Update reserved blocks/metadata blocks after successful
1317 * block allocation which had been deferred till now. We don't
1318 * support fallocate for non extent files. So we can update
1319 * reserve space here.
1322 (flags
& EXT4_GET_BLOCKS_DELALLOC_RESERVE
))
1323 ext4_da_update_reserve_space(inode
, retval
, 1);
1325 if (flags
& EXT4_GET_BLOCKS_DELALLOC_RESERVE
)
1326 EXT4_I(inode
)->i_delalloc_reserved_flag
= 0;
1328 up_write((&EXT4_I(inode
)->i_data_sem
));
1329 if (retval
> 0 && map
->m_flags
& EXT4_MAP_MAPPED
) {
1330 int ret
= check_block_validity(inode
, map
);
1337 /* Maximum number of blocks we map for direct IO at once. */
1338 #define DIO_MAX_BLOCKS 4096
1340 static int _ext4_get_block(struct inode
*inode
, sector_t iblock
,
1341 struct buffer_head
*bh
, int flags
)
1343 handle_t
*handle
= ext4_journal_current_handle();
1344 struct ext4_map_blocks map
;
1345 int ret
= 0, started
= 0;
1348 map
.m_lblk
= iblock
;
1349 map
.m_len
= bh
->b_size
>> inode
->i_blkbits
;
1351 if (flags
&& !handle
) {
1352 /* Direct IO write... */
1353 if (map
.m_len
> DIO_MAX_BLOCKS
)
1354 map
.m_len
= DIO_MAX_BLOCKS
;
1355 dio_credits
= ext4_chunk_trans_blocks(inode
, map
.m_len
);
1356 handle
= ext4_journal_start(inode
, dio_credits
);
1357 if (IS_ERR(handle
)) {
1358 ret
= PTR_ERR(handle
);
1364 ret
= ext4_map_blocks(handle
, inode
, &map
, flags
);
1366 map_bh(bh
, inode
->i_sb
, map
.m_pblk
);
1367 bh
->b_state
= (bh
->b_state
& ~EXT4_MAP_FLAGS
) | map
.m_flags
;
1368 bh
->b_size
= inode
->i_sb
->s_blocksize
* map
.m_len
;
1372 ext4_journal_stop(handle
);
1376 int ext4_get_block(struct inode
*inode
, sector_t iblock
,
1377 struct buffer_head
*bh
, int create
)
1379 return _ext4_get_block(inode
, iblock
, bh
,
1380 create
? EXT4_GET_BLOCKS_CREATE
: 0);
1384 * `handle' can be NULL if create is zero
1386 struct buffer_head
*ext4_getblk(handle_t
*handle
, struct inode
*inode
,
1387 ext4_lblk_t block
, int create
, int *errp
)
1389 struct ext4_map_blocks map
;
1390 struct buffer_head
*bh
;
1393 J_ASSERT(handle
!= NULL
|| create
== 0);
1397 err
= ext4_map_blocks(handle
, inode
, &map
,
1398 create
? EXT4_GET_BLOCKS_CREATE
: 0);
1406 bh
= sb_getblk(inode
->i_sb
, map
.m_pblk
);
1411 if (map
.m_flags
& EXT4_MAP_NEW
) {
1412 J_ASSERT(create
!= 0);
1413 J_ASSERT(handle
!= NULL
);
1416 * Now that we do not always journal data, we should
1417 * keep in mind whether this should always journal the
1418 * new buffer as metadata. For now, regular file
1419 * writes use ext4_get_block instead, so it's not a
1423 BUFFER_TRACE(bh
, "call get_create_access");
1424 fatal
= ext4_journal_get_create_access(handle
, bh
);
1425 if (!fatal
&& !buffer_uptodate(bh
)) {
1426 memset(bh
->b_data
, 0, inode
->i_sb
->s_blocksize
);
1427 set_buffer_uptodate(bh
);
1430 BUFFER_TRACE(bh
, "call ext4_handle_dirty_metadata");
1431 err
= ext4_handle_dirty_metadata(handle
, inode
, bh
);
1435 BUFFER_TRACE(bh
, "not a new buffer");
1445 struct buffer_head
*ext4_bread(handle_t
*handle
, struct inode
*inode
,
1446 ext4_lblk_t block
, int create
, int *err
)
1448 struct buffer_head
*bh
;
1450 bh
= ext4_getblk(handle
, inode
, block
, create
, err
);
1453 if (buffer_uptodate(bh
))
1455 ll_rw_block(READ_META
, 1, &bh
);
1457 if (buffer_uptodate(bh
))
1464 static int walk_page_buffers(handle_t
*handle
,
1465 struct buffer_head
*head
,
1469 int (*fn
)(handle_t
*handle
,
1470 struct buffer_head
*bh
))
1472 struct buffer_head
*bh
;
1473 unsigned block_start
, block_end
;
1474 unsigned blocksize
= head
->b_size
;
1476 struct buffer_head
*next
;
1478 for (bh
= head
, block_start
= 0;
1479 ret
== 0 && (bh
!= head
|| !block_start
);
1480 block_start
= block_end
, bh
= next
) {
1481 next
= bh
->b_this_page
;
1482 block_end
= block_start
+ blocksize
;
1483 if (block_end
<= from
|| block_start
>= to
) {
1484 if (partial
&& !buffer_uptodate(bh
))
1488 err
= (*fn
)(handle
, bh
);
1496 * To preserve ordering, it is essential that the hole instantiation and
1497 * the data write be encapsulated in a single transaction. We cannot
1498 * close off a transaction and start a new one between the ext4_get_block()
1499 * and the commit_write(). So doing the jbd2_journal_start at the start of
1500 * prepare_write() is the right place.
1502 * Also, this function can nest inside ext4_writepage() ->
1503 * block_write_full_page(). In that case, we *know* that ext4_writepage()
1504 * has generated enough buffer credits to do the whole page. So we won't
1505 * block on the journal in that case, which is good, because the caller may
1508 * By accident, ext4 can be reentered when a transaction is open via
1509 * quota file writes. If we were to commit the transaction while thus
1510 * reentered, there can be a deadlock - we would be holding a quota
1511 * lock, and the commit would never complete if another thread had a
1512 * transaction open and was blocking on the quota lock - a ranking
1515 * So what we do is to rely on the fact that jbd2_journal_stop/journal_start
1516 * will _not_ run commit under these circumstances because handle->h_ref
1517 * is elevated. We'll still have enough credits for the tiny quotafile
1520 static int do_journal_get_write_access(handle_t
*handle
,
1521 struct buffer_head
*bh
)
1523 if (!buffer_mapped(bh
) || buffer_freed(bh
))
1525 return ext4_journal_get_write_access(handle
, bh
);
1529 * Truncate blocks that were not used by write. We have to truncate the
1530 * pagecache as well so that corresponding buffers get properly unmapped.
1532 static void ext4_truncate_failed_write(struct inode
*inode
)
1534 truncate_inode_pages(inode
->i_mapping
, inode
->i_size
);
1535 ext4_truncate(inode
);
1538 static int ext4_get_block_write(struct inode
*inode
, sector_t iblock
,
1539 struct buffer_head
*bh_result
, int create
);
1540 static int ext4_write_begin(struct file
*file
, struct address_space
*mapping
,
1541 loff_t pos
, unsigned len
, unsigned flags
,
1542 struct page
**pagep
, void **fsdata
)
1544 struct inode
*inode
= mapping
->host
;
1545 int ret
, needed_blocks
;
1552 trace_ext4_write_begin(inode
, pos
, len
, flags
);
1554 * Reserve one block more for addition to orphan list in case
1555 * we allocate blocks but write fails for some reason
1557 needed_blocks
= ext4_writepage_trans_blocks(inode
) + 1;
1558 index
= pos
>> PAGE_CACHE_SHIFT
;
1559 from
= pos
& (PAGE_CACHE_SIZE
- 1);
1563 handle
= ext4_journal_start(inode
, needed_blocks
);
1564 if (IS_ERR(handle
)) {
1565 ret
= PTR_ERR(handle
);
1569 /* We cannot recurse into the filesystem as the transaction is already
1571 flags
|= AOP_FLAG_NOFS
;
1573 page
= grab_cache_page_write_begin(mapping
, index
, flags
);
1575 ext4_journal_stop(handle
);
1581 if (ext4_should_dioread_nolock(inode
))
1582 ret
= block_write_begin(file
, mapping
, pos
, len
, flags
, pagep
,
1583 fsdata
, ext4_get_block_write
);
1585 ret
= block_write_begin(file
, mapping
, pos
, len
, flags
, pagep
,
1586 fsdata
, ext4_get_block
);
1588 if (!ret
&& ext4_should_journal_data(inode
)) {
1589 ret
= walk_page_buffers(handle
, page_buffers(page
),
1590 from
, to
, NULL
, do_journal_get_write_access
);
1595 page_cache_release(page
);
1597 * block_write_begin may have instantiated a few blocks
1598 * outside i_size. Trim these off again. Don't need
1599 * i_size_read because we hold i_mutex.
1601 * Add inode to orphan list in case we crash before
1604 if (pos
+ len
> inode
->i_size
&& ext4_can_truncate(inode
))
1605 ext4_orphan_add(handle
, inode
);
1607 ext4_journal_stop(handle
);
1608 if (pos
+ len
> inode
->i_size
) {
1609 ext4_truncate_failed_write(inode
);
1611 * If truncate failed early the inode might
1612 * still be on the orphan list; we need to
1613 * make sure the inode is removed from the
1614 * orphan list in that case.
1617 ext4_orphan_del(NULL
, inode
);
1621 if (ret
== -ENOSPC
&& ext4_should_retry_alloc(inode
->i_sb
, &retries
))
1627 /* For write_end() in data=journal mode */
1628 static int write_end_fn(handle_t
*handle
, struct buffer_head
*bh
)
1630 if (!buffer_mapped(bh
) || buffer_freed(bh
))
1632 set_buffer_uptodate(bh
);
1633 return ext4_handle_dirty_metadata(handle
, NULL
, bh
);
1636 static int ext4_generic_write_end(struct file
*file
,
1637 struct address_space
*mapping
,
1638 loff_t pos
, unsigned len
, unsigned copied
,
1639 struct page
*page
, void *fsdata
)
1641 int i_size_changed
= 0;
1642 struct inode
*inode
= mapping
->host
;
1643 handle_t
*handle
= ext4_journal_current_handle();
1645 copied
= block_write_end(file
, mapping
, pos
, len
, copied
, page
, fsdata
);
1648 * No need to use i_size_read() here, the i_size
1649 * cannot change under us because we hold i_mutex.
1651 * But it's important to update i_size while still holding page lock:
1652 * page writeout could otherwise come in and zero beyond i_size.
1654 if (pos
+ copied
> inode
->i_size
) {
1655 i_size_write(inode
, pos
+ copied
);
1659 if (pos
+ copied
> EXT4_I(inode
)->i_disksize
) {
1660 /* We need to mark inode dirty even if
1661 * new_i_size is less that inode->i_size
1662 * bu greater than i_disksize.(hint delalloc)
1664 ext4_update_i_disksize(inode
, (pos
+ copied
));
1668 page_cache_release(page
);
1671 * Don't mark the inode dirty under page lock. First, it unnecessarily
1672 * makes the holding time of page lock longer. Second, it forces lock
1673 * ordering of page lock and transaction start for journaling
1677 ext4_mark_inode_dirty(handle
, inode
);
1683 * We need to pick up the new inode size which generic_commit_write gave us
1684 * `file' can be NULL - eg, when called from page_symlink().
1686 * ext4 never places buffers on inode->i_mapping->private_list. metadata
1687 * buffers are managed internally.
1689 static int ext4_ordered_write_end(struct file
*file
,
1690 struct address_space
*mapping
,
1691 loff_t pos
, unsigned len
, unsigned copied
,
1692 struct page
*page
, void *fsdata
)
1694 handle_t
*handle
= ext4_journal_current_handle();
1695 struct inode
*inode
= mapping
->host
;
1698 trace_ext4_ordered_write_end(inode
, pos
, len
, copied
);
1699 ret
= ext4_jbd2_file_inode(handle
, inode
);
1702 ret2
= ext4_generic_write_end(file
, mapping
, pos
, len
, copied
,
1705 if (pos
+ len
> inode
->i_size
&& ext4_can_truncate(inode
))
1706 /* if we have allocated more blocks and copied
1707 * less. We will have blocks allocated outside
1708 * inode->i_size. So truncate them
1710 ext4_orphan_add(handle
, inode
);
1714 ret2
= ext4_journal_stop(handle
);
1718 if (pos
+ len
> inode
->i_size
) {
1719 ext4_truncate_failed_write(inode
);
1721 * If truncate failed early the inode might still be
1722 * on the orphan list; we need to make sure the inode
1723 * is removed from the orphan list in that case.
1726 ext4_orphan_del(NULL
, inode
);
1730 return ret
? ret
: copied
;
1733 static int ext4_writeback_write_end(struct file
*file
,
1734 struct address_space
*mapping
,
1735 loff_t pos
, unsigned len
, unsigned copied
,
1736 struct page
*page
, void *fsdata
)
1738 handle_t
*handle
= ext4_journal_current_handle();
1739 struct inode
*inode
= mapping
->host
;
1742 trace_ext4_writeback_write_end(inode
, pos
, len
, copied
);
1743 ret2
= ext4_generic_write_end(file
, mapping
, pos
, len
, copied
,
1746 if (pos
+ len
> inode
->i_size
&& ext4_can_truncate(inode
))
1747 /* if we have allocated more blocks and copied
1748 * less. We will have blocks allocated outside
1749 * inode->i_size. So truncate them
1751 ext4_orphan_add(handle
, inode
);
1756 ret2
= ext4_journal_stop(handle
);
1760 if (pos
+ len
> inode
->i_size
) {
1761 ext4_truncate_failed_write(inode
);
1763 * If truncate failed early the inode might still be
1764 * on the orphan list; we need to make sure the inode
1765 * is removed from the orphan list in that case.
1768 ext4_orphan_del(NULL
, inode
);
1771 return ret
? ret
: copied
;
1774 static int ext4_journalled_write_end(struct file
*file
,
1775 struct address_space
*mapping
,
1776 loff_t pos
, unsigned len
, unsigned copied
,
1777 struct page
*page
, void *fsdata
)
1779 handle_t
*handle
= ext4_journal_current_handle();
1780 struct inode
*inode
= mapping
->host
;
1786 trace_ext4_journalled_write_end(inode
, pos
, len
, copied
);
1787 from
= pos
& (PAGE_CACHE_SIZE
- 1);
1791 if (!PageUptodate(page
))
1793 page_zero_new_buffers(page
, from
+copied
, to
);
1796 ret
= walk_page_buffers(handle
, page_buffers(page
), from
,
1797 to
, &partial
, write_end_fn
);
1799 SetPageUptodate(page
);
1800 new_i_size
= pos
+ copied
;
1801 if (new_i_size
> inode
->i_size
)
1802 i_size_write(inode
, pos
+copied
);
1803 ext4_set_inode_state(inode
, EXT4_STATE_JDATA
);
1804 if (new_i_size
> EXT4_I(inode
)->i_disksize
) {
1805 ext4_update_i_disksize(inode
, new_i_size
);
1806 ret2
= ext4_mark_inode_dirty(handle
, inode
);
1812 page_cache_release(page
);
1813 if (pos
+ len
> inode
->i_size
&& ext4_can_truncate(inode
))
1814 /* if we have allocated more blocks and copied
1815 * less. We will have blocks allocated outside
1816 * inode->i_size. So truncate them
1818 ext4_orphan_add(handle
, inode
);
1820 ret2
= ext4_journal_stop(handle
);
1823 if (pos
+ len
> inode
->i_size
) {
1824 ext4_truncate_failed_write(inode
);
1826 * If truncate failed early the inode might still be
1827 * on the orphan list; we need to make sure the inode
1828 * is removed from the orphan list in that case.
1831 ext4_orphan_del(NULL
, inode
);
1834 return ret
? ret
: copied
;
1838 * Reserve a single block located at lblock
1840 static int ext4_da_reserve_space(struct inode
*inode
, sector_t lblock
)
1843 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
1844 struct ext4_inode_info
*ei
= EXT4_I(inode
);
1845 unsigned long md_needed
;
1849 * recalculate the amount of metadata blocks to reserve
1850 * in order to allocate nrblocks
1851 * worse case is one extent per block
1854 spin_lock(&ei
->i_block_reservation_lock
);
1855 md_needed
= ext4_calc_metadata_amount(inode
, lblock
);
1856 trace_ext4_da_reserve_space(inode
, md_needed
);
1857 spin_unlock(&ei
->i_block_reservation_lock
);
1860 * We will charge metadata quota at writeout time; this saves
1861 * us from metadata over-estimation, though we may go over by
1862 * a small amount in the end. Here we just reserve for data.
1864 ret
= dquot_reserve_block(inode
, 1);
1868 * We do still charge estimated metadata to the sb though;
1869 * we cannot afford to run out of free blocks.
1871 if (ext4_claim_free_blocks(sbi
, md_needed
+ 1)) {
1872 dquot_release_reservation_block(inode
, 1);
1873 if (ext4_should_retry_alloc(inode
->i_sb
, &retries
)) {
1879 spin_lock(&ei
->i_block_reservation_lock
);
1880 ei
->i_reserved_data_blocks
++;
1881 ei
->i_reserved_meta_blocks
+= md_needed
;
1882 spin_unlock(&ei
->i_block_reservation_lock
);
1884 return 0; /* success */
1887 static void ext4_da_release_space(struct inode
*inode
, int to_free
)
1889 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
1890 struct ext4_inode_info
*ei
= EXT4_I(inode
);
1893 return; /* Nothing to release, exit */
1895 spin_lock(&EXT4_I(inode
)->i_block_reservation_lock
);
1897 trace_ext4_da_release_space(inode
, to_free
);
1898 if (unlikely(to_free
> ei
->i_reserved_data_blocks
)) {
1900 * if there aren't enough reserved blocks, then the
1901 * counter is messed up somewhere. Since this
1902 * function is called from invalidate page, it's
1903 * harmless to return without any action.
1905 ext4_msg(inode
->i_sb
, KERN_NOTICE
, "ext4_da_release_space: "
1906 "ino %lu, to_free %d with only %d reserved "
1907 "data blocks\n", inode
->i_ino
, to_free
,
1908 ei
->i_reserved_data_blocks
);
1910 to_free
= ei
->i_reserved_data_blocks
;
1912 ei
->i_reserved_data_blocks
-= to_free
;
1914 if (ei
->i_reserved_data_blocks
== 0) {
1916 * We can release all of the reserved metadata blocks
1917 * only when we have written all of the delayed
1918 * allocation blocks.
1920 percpu_counter_sub(&sbi
->s_dirtyblocks_counter
,
1921 ei
->i_reserved_meta_blocks
);
1922 ei
->i_reserved_meta_blocks
= 0;
1923 ei
->i_da_metadata_calc_len
= 0;
1926 /* update fs dirty data blocks counter */
1927 percpu_counter_sub(&sbi
->s_dirtyblocks_counter
, to_free
);
1929 spin_unlock(&EXT4_I(inode
)->i_block_reservation_lock
);
1931 dquot_release_reservation_block(inode
, to_free
);
1934 static void ext4_da_page_release_reservation(struct page
*page
,
1935 unsigned long offset
)
1938 struct buffer_head
*head
, *bh
;
1939 unsigned int curr_off
= 0;
1941 head
= page_buffers(page
);
1944 unsigned int next_off
= curr_off
+ bh
->b_size
;
1946 if ((offset
<= curr_off
) && (buffer_delay(bh
))) {
1948 clear_buffer_delay(bh
);
1950 curr_off
= next_off
;
1951 } while ((bh
= bh
->b_this_page
) != head
);
1952 ext4_da_release_space(page
->mapping
->host
, to_release
);
1956 * Delayed allocation stuff
1960 * mpage_da_submit_io - walks through extent of pages and try to write
1961 * them with writepage() call back
1963 * @mpd->inode: inode
1964 * @mpd->first_page: first page of the extent
1965 * @mpd->next_page: page after the last page of the extent
1967 * By the time mpage_da_submit_io() is called we expect all blocks
1968 * to be allocated. this may be wrong if allocation failed.
1970 * As pages are already locked by write_cache_pages(), we can't use it
1972 static int mpage_da_submit_io(struct mpage_da_data
*mpd
)
1975 struct pagevec pvec
;
1976 unsigned long index
, end
;
1977 int ret
= 0, err
, nr_pages
, i
;
1978 struct inode
*inode
= mpd
->inode
;
1979 struct address_space
*mapping
= inode
->i_mapping
;
1981 BUG_ON(mpd
->next_page
<= mpd
->first_page
);
1983 * We need to start from the first_page to the next_page - 1
1984 * to make sure we also write the mapped dirty buffer_heads.
1985 * If we look at mpd->b_blocknr we would only be looking
1986 * at the currently mapped buffer_heads.
1988 index
= mpd
->first_page
;
1989 end
= mpd
->next_page
- 1;
1991 pagevec_init(&pvec
, 0);
1992 while (index
<= end
) {
1993 nr_pages
= pagevec_lookup(&pvec
, mapping
, index
, PAGEVEC_SIZE
);
1996 for (i
= 0; i
< nr_pages
; i
++) {
1997 struct page
*page
= pvec
.pages
[i
];
1999 index
= page
->index
;
2004 BUG_ON(!PageLocked(page
));
2005 BUG_ON(PageWriteback(page
));
2007 pages_skipped
= mpd
->wbc
->pages_skipped
;
2008 err
= mapping
->a_ops
->writepage(page
, mpd
->wbc
);
2009 if (!err
&& (pages_skipped
== mpd
->wbc
->pages_skipped
))
2011 * have successfully written the page
2012 * without skipping the same
2014 mpd
->pages_written
++;
2016 * In error case, we have to continue because
2017 * remaining pages are still locked
2018 * XXX: unlock and re-dirty them?
2023 pagevec_release(&pvec
);
2029 * mpage_put_bnr_to_bhs - walk blocks and assign them actual numbers
2031 * the function goes through all passed space and put actual disk
2032 * block numbers into buffer heads, dropping BH_Delay and BH_Unwritten
2034 static void mpage_put_bnr_to_bhs(struct mpage_da_data
*mpd
,
2035 struct ext4_map_blocks
*map
)
2037 struct inode
*inode
= mpd
->inode
;
2038 struct address_space
*mapping
= inode
->i_mapping
;
2039 int blocks
= map
->m_len
;
2040 sector_t pblock
= map
->m_pblk
, cur_logical
;
2041 struct buffer_head
*head
, *bh
;
2043 struct pagevec pvec
;
2046 index
= map
->m_lblk
>> (PAGE_CACHE_SHIFT
- inode
->i_blkbits
);
2047 end
= (map
->m_lblk
+ blocks
- 1) >> (PAGE_CACHE_SHIFT
- inode
->i_blkbits
);
2048 cur_logical
= index
<< (PAGE_CACHE_SHIFT
- inode
->i_blkbits
);
2050 pagevec_init(&pvec
, 0);
2052 while (index
<= end
) {
2053 /* XXX: optimize tail */
2054 nr_pages
= pagevec_lookup(&pvec
, mapping
, index
, PAGEVEC_SIZE
);
2057 for (i
= 0; i
< nr_pages
; i
++) {
2058 struct page
*page
= pvec
.pages
[i
];
2060 index
= page
->index
;
2065 BUG_ON(!PageLocked(page
));
2066 BUG_ON(PageWriteback(page
));
2067 BUG_ON(!page_has_buffers(page
));
2069 bh
= page_buffers(page
);
2072 /* skip blocks out of the range */
2074 if (cur_logical
>= map
->m_lblk
)
2077 } while ((bh
= bh
->b_this_page
) != head
);
2080 if (cur_logical
>= map
->m_lblk
+ blocks
)
2083 if (buffer_delay(bh
) || buffer_unwritten(bh
)) {
2085 BUG_ON(bh
->b_bdev
!= inode
->i_sb
->s_bdev
);
2087 if (buffer_delay(bh
)) {
2088 clear_buffer_delay(bh
);
2089 bh
->b_blocknr
= pblock
;
2092 * unwritten already should have
2093 * blocknr assigned. Verify that
2095 clear_buffer_unwritten(bh
);
2096 BUG_ON(bh
->b_blocknr
!= pblock
);
2099 } else if (buffer_mapped(bh
))
2100 BUG_ON(bh
->b_blocknr
!= pblock
);
2102 if (map
->m_flags
& EXT4_MAP_UNINIT
)
2103 set_buffer_uninit(bh
);
2106 } while ((bh
= bh
->b_this_page
) != head
);
2108 pagevec_release(&pvec
);
2113 static void ext4_da_block_invalidatepages(struct mpage_da_data
*mpd
,
2114 sector_t logical
, long blk_cnt
)
2118 struct pagevec pvec
;
2119 struct inode
*inode
= mpd
->inode
;
2120 struct address_space
*mapping
= inode
->i_mapping
;
2122 index
= logical
>> (PAGE_CACHE_SHIFT
- inode
->i_blkbits
);
2123 end
= (logical
+ blk_cnt
- 1) >>
2124 (PAGE_CACHE_SHIFT
- inode
->i_blkbits
);
2125 while (index
<= end
) {
2126 nr_pages
= pagevec_lookup(&pvec
, mapping
, index
, PAGEVEC_SIZE
);
2129 for (i
= 0; i
< nr_pages
; i
++) {
2130 struct page
*page
= pvec
.pages
[i
];
2131 if (page
->index
> end
)
2133 BUG_ON(!PageLocked(page
));
2134 BUG_ON(PageWriteback(page
));
2135 block_invalidatepage(page
, 0);
2136 ClearPageUptodate(page
);
2139 index
= pvec
.pages
[nr_pages
- 1]->index
+ 1;
2140 pagevec_release(&pvec
);
2145 static void ext4_print_free_blocks(struct inode
*inode
)
2147 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
2148 printk(KERN_CRIT
"Total free blocks count %lld\n",
2149 ext4_count_free_blocks(inode
->i_sb
));
2150 printk(KERN_CRIT
"Free/Dirty block details\n");
2151 printk(KERN_CRIT
"free_blocks=%lld\n",
2152 (long long) percpu_counter_sum(&sbi
->s_freeblocks_counter
));
2153 printk(KERN_CRIT
"dirty_blocks=%lld\n",
2154 (long long) percpu_counter_sum(&sbi
->s_dirtyblocks_counter
));
2155 printk(KERN_CRIT
"Block reservation details\n");
2156 printk(KERN_CRIT
"i_reserved_data_blocks=%u\n",
2157 EXT4_I(inode
)->i_reserved_data_blocks
);
2158 printk(KERN_CRIT
"i_reserved_meta_blocks=%u\n",
2159 EXT4_I(inode
)->i_reserved_meta_blocks
);
2164 * mpage_da_map_blocks - go through given space
2166 * @mpd - bh describing space
2168 * The function skips space we know is already mapped to disk blocks.
2171 static int mpage_da_map_blocks(struct mpage_da_data
*mpd
)
2173 int err
, blks
, get_blocks_flags
;
2174 struct ext4_map_blocks map
;
2175 sector_t next
= mpd
->b_blocknr
;
2176 unsigned max_blocks
= mpd
->b_size
>> mpd
->inode
->i_blkbits
;
2177 loff_t disksize
= EXT4_I(mpd
->inode
)->i_disksize
;
2178 handle_t
*handle
= NULL
;
2181 * We consider only non-mapped and non-allocated blocks
2183 if ((mpd
->b_state
& (1 << BH_Mapped
)) &&
2184 !(mpd
->b_state
& (1 << BH_Delay
)) &&
2185 !(mpd
->b_state
& (1 << BH_Unwritten
)))
2189 * If we didn't accumulate anything to write simply return
2194 handle
= ext4_journal_current_handle();
2198 * Call ext4_get_blocks() to allocate any delayed allocation
2199 * blocks, or to convert an uninitialized extent to be
2200 * initialized (in the case where we have written into
2201 * one or more preallocated blocks).
2203 * We pass in the magic EXT4_GET_BLOCKS_DELALLOC_RESERVE to
2204 * indicate that we are on the delayed allocation path. This
2205 * affects functions in many different parts of the allocation
2206 * call path. This flag exists primarily because we don't
2207 * want to change *many* call functions, so ext4_get_blocks()
2208 * will set the magic i_delalloc_reserved_flag once the
2209 * inode's allocation semaphore is taken.
2211 * If the blocks in questions were delalloc blocks, set
2212 * EXT4_GET_BLOCKS_DELALLOC_RESERVE so the delalloc accounting
2213 * variables are updated after the blocks have been allocated.
2216 map
.m_len
= max_blocks
;
2217 get_blocks_flags
= EXT4_GET_BLOCKS_CREATE
;
2218 if (ext4_should_dioread_nolock(mpd
->inode
))
2219 get_blocks_flags
|= EXT4_GET_BLOCKS_IO_CREATE_EXT
;
2220 if (mpd
->b_state
& (1 << BH_Delay
))
2221 get_blocks_flags
|= EXT4_GET_BLOCKS_DELALLOC_RESERVE
;
2223 blks
= ext4_map_blocks(handle
, mpd
->inode
, &map
, get_blocks_flags
);
2227 * If get block returns with error we simply
2228 * return. Later writepage will redirty the page and
2229 * writepages will find the dirty page again
2234 if (err
== -ENOSPC
&&
2235 ext4_count_free_blocks(mpd
->inode
->i_sb
)) {
2241 * get block failure will cause us to loop in
2242 * writepages, because a_ops->writepage won't be able
2243 * to make progress. The page will be redirtied by
2244 * writepage and writepages will again try to write
2247 ext4_msg(mpd
->inode
->i_sb
, KERN_CRIT
,
2248 "delayed block allocation failed for inode %lu at "
2249 "logical offset %llu with max blocks %zd with "
2250 "error %d", mpd
->inode
->i_ino
,
2251 (unsigned long long) next
,
2252 mpd
->b_size
>> mpd
->inode
->i_blkbits
, err
);
2253 printk(KERN_CRIT
"This should not happen!! "
2254 "Data will be lost\n");
2255 if (err
== -ENOSPC
) {
2256 ext4_print_free_blocks(mpd
->inode
);
2258 /* invalidate all the pages */
2259 ext4_da_block_invalidatepages(mpd
, next
,
2260 mpd
->b_size
>> mpd
->inode
->i_blkbits
);
2265 if (map
.m_flags
& EXT4_MAP_NEW
) {
2266 struct block_device
*bdev
= mpd
->inode
->i_sb
->s_bdev
;
2269 for (i
= 0; i
< map
.m_len
; i
++)
2270 unmap_underlying_metadata(bdev
, map
.m_pblk
+ i
);
2274 * If blocks are delayed marked, we need to
2275 * put actual blocknr and drop delayed bit
2277 if ((mpd
->b_state
& (1 << BH_Delay
)) ||
2278 (mpd
->b_state
& (1 << BH_Unwritten
)))
2279 mpage_put_bnr_to_bhs(mpd
, &map
);
2281 if (ext4_should_order_data(mpd
->inode
)) {
2282 err
= ext4_jbd2_file_inode(handle
, mpd
->inode
);
2288 * Update on-disk size along with block allocation.
2290 disksize
= ((loff_t
) next
+ blks
) << mpd
->inode
->i_blkbits
;
2291 if (disksize
> i_size_read(mpd
->inode
))
2292 disksize
= i_size_read(mpd
->inode
);
2293 if (disksize
> EXT4_I(mpd
->inode
)->i_disksize
) {
2294 ext4_update_i_disksize(mpd
->inode
, disksize
);
2295 return ext4_mark_inode_dirty(handle
, mpd
->inode
);
2301 #define BH_FLAGS ((1 << BH_Uptodate) | (1 << BH_Mapped) | \
2302 (1 << BH_Delay) | (1 << BH_Unwritten))
2305 * mpage_add_bh_to_extent - try to add one more block to extent of blocks
2307 * @mpd->lbh - extent of blocks
2308 * @logical - logical number of the block in the file
2309 * @bh - bh of the block (used to access block's state)
2311 * the function is used to collect contig. blocks in same state
2313 static void mpage_add_bh_to_extent(struct mpage_da_data
*mpd
,
2314 sector_t logical
, size_t b_size
,
2315 unsigned long b_state
)
2318 int nrblocks
= mpd
->b_size
>> mpd
->inode
->i_blkbits
;
2321 * XXX Don't go larger than mballoc is willing to allocate
2322 * This is a stopgap solution. We eventually need to fold
2323 * mpage_da_submit_io() into this function and then call
2324 * ext4_get_blocks() multiple times in a loop
2326 if (nrblocks
>= 8*1024*1024/mpd
->inode
->i_sb
->s_blocksize
)
2329 /* check if thereserved journal credits might overflow */
2330 if (!(ext4_test_inode_flag(mpd
->inode
, EXT4_INODE_EXTENTS
))) {
2331 if (nrblocks
>= EXT4_MAX_TRANS_DATA
) {
2333 * With non-extent format we are limited by the journal
2334 * credit available. Total credit needed to insert
2335 * nrblocks contiguous blocks is dependent on the
2336 * nrblocks. So limit nrblocks.
2339 } else if ((nrblocks
+ (b_size
>> mpd
->inode
->i_blkbits
)) >
2340 EXT4_MAX_TRANS_DATA
) {
2342 * Adding the new buffer_head would make it cross the
2343 * allowed limit for which we have journal credit
2344 * reserved. So limit the new bh->b_size
2346 b_size
= (EXT4_MAX_TRANS_DATA
- nrblocks
) <<
2347 mpd
->inode
->i_blkbits
;
2348 /* we will do mpage_da_submit_io in the next loop */
2352 * First block in the extent
2354 if (mpd
->b_size
== 0) {
2355 mpd
->b_blocknr
= logical
;
2356 mpd
->b_size
= b_size
;
2357 mpd
->b_state
= b_state
& BH_FLAGS
;
2361 next
= mpd
->b_blocknr
+ nrblocks
;
2363 * Can we merge the block to our big extent?
2365 if (logical
== next
&& (b_state
& BH_FLAGS
) == mpd
->b_state
) {
2366 mpd
->b_size
+= b_size
;
2372 * We couldn't merge the block to our extent, so we
2373 * need to flush current extent and start new one
2375 if (mpage_da_map_blocks(mpd
) == 0)
2376 mpage_da_submit_io(mpd
);
2381 static int ext4_bh_delay_or_unwritten(handle_t
*handle
, struct buffer_head
*bh
)
2383 return (buffer_delay(bh
) || buffer_unwritten(bh
)) && buffer_dirty(bh
);
2387 * __mpage_da_writepage - finds extent of pages and blocks
2389 * @page: page to consider
2390 * @wbc: not used, we just follow rules
2393 * The function finds extents of pages and scan them for all blocks.
2395 static int __mpage_da_writepage(struct page
*page
,
2396 struct writeback_control
*wbc
, void *data
)
2398 struct mpage_da_data
*mpd
= data
;
2399 struct inode
*inode
= mpd
->inode
;
2400 struct buffer_head
*bh
, *head
;
2404 * Can we merge this page to current extent?
2406 if (mpd
->next_page
!= page
->index
) {
2408 * Nope, we can't. So, we map non-allocated blocks
2409 * and start IO on them using writepage()
2411 if (mpd
->next_page
!= mpd
->first_page
) {
2412 if (mpage_da_map_blocks(mpd
) == 0)
2413 mpage_da_submit_io(mpd
);
2415 * skip rest of the page in the page_vec
2418 redirty_page_for_writepage(wbc
, page
);
2420 return MPAGE_DA_EXTENT_TAIL
;
2424 * Start next extent of pages ...
2426 mpd
->first_page
= page
->index
;
2436 mpd
->next_page
= page
->index
+ 1;
2437 logical
= (sector_t
) page
->index
<<
2438 (PAGE_CACHE_SHIFT
- inode
->i_blkbits
);
2440 if (!page_has_buffers(page
)) {
2441 mpage_add_bh_to_extent(mpd
, logical
, PAGE_CACHE_SIZE
,
2442 (1 << BH_Dirty
) | (1 << BH_Uptodate
));
2444 return MPAGE_DA_EXTENT_TAIL
;
2447 * Page with regular buffer heads, just add all dirty ones
2449 head
= page_buffers(page
);
2452 BUG_ON(buffer_locked(bh
));
2454 * We need to try to allocate
2455 * unmapped blocks in the same page.
2456 * Otherwise we won't make progress
2457 * with the page in ext4_writepage
2459 if (ext4_bh_delay_or_unwritten(NULL
, bh
)) {
2460 mpage_add_bh_to_extent(mpd
, logical
,
2464 return MPAGE_DA_EXTENT_TAIL
;
2465 } else if (buffer_dirty(bh
) && (buffer_mapped(bh
))) {
2467 * mapped dirty buffer. We need to update
2468 * the b_state because we look at
2469 * b_state in mpage_da_map_blocks. We don't
2470 * update b_size because if we find an
2471 * unmapped buffer_head later we need to
2472 * use the b_state flag of that buffer_head.
2474 if (mpd
->b_size
== 0)
2475 mpd
->b_state
= bh
->b_state
& BH_FLAGS
;
2478 } while ((bh
= bh
->b_this_page
) != head
);
2485 * This is a special get_blocks_t callback which is used by
2486 * ext4_da_write_begin(). It will either return mapped block or
2487 * reserve space for a single block.
2489 * For delayed buffer_head we have BH_Mapped, BH_New, BH_Delay set.
2490 * We also have b_blocknr = -1 and b_bdev initialized properly
2492 * For unwritten buffer_head we have BH_Mapped, BH_New, BH_Unwritten set.
2493 * We also have b_blocknr = physicalblock mapping unwritten extent and b_bdev
2494 * initialized properly.
2496 static int ext4_da_get_block_prep(struct inode
*inode
, sector_t iblock
,
2497 struct buffer_head
*bh
, int create
)
2499 struct ext4_map_blocks map
;
2501 sector_t invalid_block
= ~((sector_t
) 0xffff);
2503 if (invalid_block
< ext4_blocks_count(EXT4_SB(inode
->i_sb
)->s_es
))
2506 BUG_ON(create
== 0);
2507 BUG_ON(bh
->b_size
!= inode
->i_sb
->s_blocksize
);
2509 map
.m_lblk
= iblock
;
2513 * first, we need to know whether the block is allocated already
2514 * preallocated blocks are unmapped but should treated
2515 * the same as allocated blocks.
2517 ret
= ext4_map_blocks(NULL
, inode
, &map
, 0);
2521 if (buffer_delay(bh
))
2522 return 0; /* Not sure this could or should happen */
2524 * XXX: __block_prepare_write() unmaps passed block,
2527 ret
= ext4_da_reserve_space(inode
, iblock
);
2529 /* not enough space to reserve */
2532 map_bh(bh
, inode
->i_sb
, invalid_block
);
2534 set_buffer_delay(bh
);
2538 map_bh(bh
, inode
->i_sb
, map
.m_pblk
);
2539 bh
->b_state
= (bh
->b_state
& ~EXT4_MAP_FLAGS
) | map
.m_flags
;
2541 if (buffer_unwritten(bh
)) {
2542 /* A delayed write to unwritten bh should be marked
2543 * new and mapped. Mapped ensures that we don't do
2544 * get_block multiple times when we write to the same
2545 * offset and new ensures that we do proper zero out
2546 * for partial write.
2549 set_buffer_mapped(bh
);
2555 * This function is used as a standard get_block_t calback function
2556 * when there is no desire to allocate any blocks. It is used as a
2557 * callback function for block_prepare_write() and block_write_full_page().
2558 * These functions should only try to map a single block at a time.
2560 * Since this function doesn't do block allocations even if the caller
2561 * requests it by passing in create=1, it is critically important that
2562 * any caller checks to make sure that any buffer heads are returned
2563 * by this function are either all already mapped or marked for
2564 * delayed allocation before calling block_write_full_page(). Otherwise,
2565 * b_blocknr could be left unitialized, and the page write functions will
2566 * be taken by surprise.
2568 static int noalloc_get_block_write(struct inode
*inode
, sector_t iblock
,
2569 struct buffer_head
*bh_result
, int create
)
2571 BUG_ON(bh_result
->b_size
!= inode
->i_sb
->s_blocksize
);
2572 return _ext4_get_block(inode
, iblock
, bh_result
, 0);
2575 static int bget_one(handle_t
*handle
, struct buffer_head
*bh
)
2581 static int bput_one(handle_t
*handle
, struct buffer_head
*bh
)
2587 static int __ext4_journalled_writepage(struct page
*page
,
2590 struct address_space
*mapping
= page
->mapping
;
2591 struct inode
*inode
= mapping
->host
;
2592 struct buffer_head
*page_bufs
;
2593 handle_t
*handle
= NULL
;
2597 page_bufs
= page_buffers(page
);
2599 walk_page_buffers(handle
, page_bufs
, 0, len
, NULL
, bget_one
);
2600 /* As soon as we unlock the page, it can go away, but we have
2601 * references to buffers so we are safe */
2604 handle
= ext4_journal_start(inode
, ext4_writepage_trans_blocks(inode
));
2605 if (IS_ERR(handle
)) {
2606 ret
= PTR_ERR(handle
);
2610 ret
= walk_page_buffers(handle
, page_bufs
, 0, len
, NULL
,
2611 do_journal_get_write_access
);
2613 err
= walk_page_buffers(handle
, page_bufs
, 0, len
, NULL
,
2617 err
= ext4_journal_stop(handle
);
2621 walk_page_buffers(handle
, page_bufs
, 0, len
, NULL
, bput_one
);
2622 ext4_set_inode_state(inode
, EXT4_STATE_JDATA
);
2627 static int ext4_set_bh_endio(struct buffer_head
*bh
, struct inode
*inode
);
2628 static void ext4_end_io_buffer_write(struct buffer_head
*bh
, int uptodate
);
2631 * Note that we don't need to start a transaction unless we're journaling data
2632 * because we should have holes filled from ext4_page_mkwrite(). We even don't
2633 * need to file the inode to the transaction's list in ordered mode because if
2634 * we are writing back data added by write(), the inode is already there and if
2635 * we are writing back data modified via mmap(), noone guarantees in which
2636 * transaction the data will hit the disk. In case we are journaling data, we
2637 * cannot start transaction directly because transaction start ranks above page
2638 * lock so we have to do some magic.
2640 * This function can get called via...
2641 * - ext4_da_writepages after taking page lock (have journal handle)
2642 * - journal_submit_inode_data_buffers (no journal handle)
2643 * - shrink_page_list via pdflush (no journal handle)
2644 * - grab_page_cache when doing write_begin (have journal handle)
2646 * We don't do any block allocation in this function. If we have page with
2647 * multiple blocks we need to write those buffer_heads that are mapped. This
2648 * is important for mmaped based write. So if we do with blocksize 1K
2649 * truncate(f, 1024);
2650 * a = mmap(f, 0, 4096);
2652 * truncate(f, 4096);
2653 * we have in the page first buffer_head mapped via page_mkwrite call back
2654 * but other bufer_heads would be unmapped but dirty(dirty done via the
2655 * do_wp_page). So writepage should write the first block. If we modify
2656 * the mmap area beyond 1024 we will again get a page_fault and the
2657 * page_mkwrite callback will do the block allocation and mark the
2658 * buffer_heads mapped.
2660 * We redirty the page if we have any buffer_heads that is either delay or
2661 * unwritten in the page.
2663 * We can get recursively called as show below.
2665 * ext4_writepage() -> kmalloc() -> __alloc_pages() -> page_launder() ->
2668 * But since we don't do any block allocation we should not deadlock.
2669 * Page also have the dirty flag cleared so we don't get recurive page_lock.
2671 static int ext4_writepage(struct page
*page
,
2672 struct writeback_control
*wbc
)
2677 struct buffer_head
*page_bufs
= NULL
;
2678 struct inode
*inode
= page
->mapping
->host
;
2680 trace_ext4_writepage(inode
, page
);
2681 size
= i_size_read(inode
);
2682 if (page
->index
== size
>> PAGE_CACHE_SHIFT
)
2683 len
= size
& ~PAGE_CACHE_MASK
;
2685 len
= PAGE_CACHE_SIZE
;
2687 if (page_has_buffers(page
)) {
2688 page_bufs
= page_buffers(page
);
2689 if (walk_page_buffers(NULL
, page_bufs
, 0, len
, NULL
,
2690 ext4_bh_delay_or_unwritten
)) {
2692 * We don't want to do block allocation
2693 * So redirty the page and return
2694 * We may reach here when we do a journal commit
2695 * via journal_submit_inode_data_buffers.
2696 * If we don't have mapping block we just ignore
2697 * them. We can also reach here via shrink_page_list
2699 redirty_page_for_writepage(wbc
, page
);
2705 * The test for page_has_buffers() is subtle:
2706 * We know the page is dirty but it lost buffers. That means
2707 * that at some moment in time after write_begin()/write_end()
2708 * has been called all buffers have been clean and thus they
2709 * must have been written at least once. So they are all
2710 * mapped and we can happily proceed with mapping them
2711 * and writing the page.
2713 * Try to initialize the buffer_heads and check whether
2714 * all are mapped and non delay. We don't want to
2715 * do block allocation here.
2717 ret
= block_prepare_write(page
, 0, len
,
2718 noalloc_get_block_write
);
2720 page_bufs
= page_buffers(page
);
2721 /* check whether all are mapped and non delay */
2722 if (walk_page_buffers(NULL
, page_bufs
, 0, len
, NULL
,
2723 ext4_bh_delay_or_unwritten
)) {
2724 redirty_page_for_writepage(wbc
, page
);
2730 * We can't do block allocation here
2731 * so just redity the page and unlock
2734 redirty_page_for_writepage(wbc
, page
);
2738 /* now mark the buffer_heads as dirty and uptodate */
2739 block_commit_write(page
, 0, len
);
2742 if (PageChecked(page
) && ext4_should_journal_data(inode
)) {
2744 * It's mmapped pagecache. Add buffers and journal it. There
2745 * doesn't seem much point in redirtying the page here.
2747 ClearPageChecked(page
);
2748 return __ext4_journalled_writepage(page
, len
);
2751 if (page_bufs
&& buffer_uninit(page_bufs
)) {
2752 ext4_set_bh_endio(page_bufs
, inode
);
2753 ret
= block_write_full_page_endio(page
, noalloc_get_block_write
,
2754 wbc
, ext4_end_io_buffer_write
);
2756 ret
= block_write_full_page(page
, noalloc_get_block_write
,
2763 * This is called via ext4_da_writepages() to
2764 * calulate the total number of credits to reserve to fit
2765 * a single extent allocation into a single transaction,
2766 * ext4_da_writpeages() will loop calling this before
2767 * the block allocation.
2770 static int ext4_da_writepages_trans_blocks(struct inode
*inode
)
2772 int max_blocks
= EXT4_I(inode
)->i_reserved_data_blocks
;
2775 * With non-extent format the journal credit needed to
2776 * insert nrblocks contiguous block is dependent on
2777 * number of contiguous block. So we will limit
2778 * number of contiguous block to a sane value
2780 if (!(ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
)) &&
2781 (max_blocks
> EXT4_MAX_TRANS_DATA
))
2782 max_blocks
= EXT4_MAX_TRANS_DATA
;
2784 return ext4_chunk_trans_blocks(inode
, max_blocks
);
2788 * write_cache_pages_da - walk the list of dirty pages of the given
2789 * address space and call the callback function (which usually writes
2792 * This is a forked version of write_cache_pages(). Differences:
2793 * Range cyclic is ignored.
2794 * no_nrwrite_index_update is always presumed true
2796 static int write_cache_pages_da(struct address_space
*mapping
,
2797 struct writeback_control
*wbc
,
2798 struct mpage_da_data
*mpd
)
2802 struct pagevec pvec
;
2805 pgoff_t end
; /* Inclusive */
2806 long nr_to_write
= wbc
->nr_to_write
;
2808 pagevec_init(&pvec
, 0);
2809 index
= wbc
->range_start
>> PAGE_CACHE_SHIFT
;
2810 end
= wbc
->range_end
>> PAGE_CACHE_SHIFT
;
2812 while (!done
&& (index
<= end
)) {
2815 nr_pages
= pagevec_lookup_tag(&pvec
, mapping
, &index
,
2816 PAGECACHE_TAG_DIRTY
,
2817 min(end
- index
, (pgoff_t
)PAGEVEC_SIZE
-1) + 1);
2821 for (i
= 0; i
< nr_pages
; i
++) {
2822 struct page
*page
= pvec
.pages
[i
];
2825 * At this point, the page may be truncated or
2826 * invalidated (changing page->mapping to NULL), or
2827 * even swizzled back from swapper_space to tmpfs file
2828 * mapping. However, page->index will not change
2829 * because we have a reference on the page.
2831 if (page
->index
> end
) {
2839 * Page truncated or invalidated. We can freely skip it
2840 * then, even for data integrity operations: the page
2841 * has disappeared concurrently, so there could be no
2842 * real expectation of this data interity operation
2843 * even if there is now a new, dirty page at the same
2844 * pagecache address.
2846 if (unlikely(page
->mapping
!= mapping
)) {
2852 if (!PageDirty(page
)) {
2853 /* someone wrote it for us */
2854 goto continue_unlock
;
2857 if (PageWriteback(page
)) {
2858 if (wbc
->sync_mode
!= WB_SYNC_NONE
)
2859 wait_on_page_writeback(page
);
2861 goto continue_unlock
;
2864 BUG_ON(PageWriteback(page
));
2865 if (!clear_page_dirty_for_io(page
))
2866 goto continue_unlock
;
2868 ret
= __mpage_da_writepage(page
, wbc
, mpd
);
2869 if (unlikely(ret
)) {
2870 if (ret
== AOP_WRITEPAGE_ACTIVATE
) {
2879 if (nr_to_write
> 0) {
2881 if (nr_to_write
== 0 &&
2882 wbc
->sync_mode
== WB_SYNC_NONE
) {
2884 * We stop writing back only if we are
2885 * not doing integrity sync. In case of
2886 * integrity sync we have to keep going
2887 * because someone may be concurrently
2888 * dirtying pages, and we might have
2889 * synced a lot of newly appeared dirty
2890 * pages, but have not synced all of the
2898 pagevec_release(&pvec
);
2905 static int ext4_da_writepages(struct address_space
*mapping
,
2906 struct writeback_control
*wbc
)
2909 int range_whole
= 0;
2910 handle_t
*handle
= NULL
;
2911 struct mpage_da_data mpd
;
2912 struct inode
*inode
= mapping
->host
;
2913 int pages_written
= 0;
2915 unsigned int max_pages
;
2916 int range_cyclic
, cycled
= 1, io_done
= 0;
2917 int needed_blocks
, ret
= 0;
2918 long desired_nr_to_write
, nr_to_writebump
= 0;
2919 loff_t range_start
= wbc
->range_start
;
2920 struct ext4_sb_info
*sbi
= EXT4_SB(mapping
->host
->i_sb
);
2922 trace_ext4_da_writepages(inode
, wbc
);
2925 * No pages to write? This is mainly a kludge to avoid starting
2926 * a transaction for special inodes like journal inode on last iput()
2927 * because that could violate lock ordering on umount
2929 if (!mapping
->nrpages
|| !mapping_tagged(mapping
, PAGECACHE_TAG_DIRTY
))
2933 * If the filesystem has aborted, it is read-only, so return
2934 * right away instead of dumping stack traces later on that
2935 * will obscure the real source of the problem. We test
2936 * EXT4_MF_FS_ABORTED instead of sb->s_flag's MS_RDONLY because
2937 * the latter could be true if the filesystem is mounted
2938 * read-only, and in that case, ext4_da_writepages should
2939 * *never* be called, so if that ever happens, we would want
2942 if (unlikely(sbi
->s_mount_flags
& EXT4_MF_FS_ABORTED
))
2945 if (wbc
->range_start
== 0 && wbc
->range_end
== LLONG_MAX
)
2948 range_cyclic
= wbc
->range_cyclic
;
2949 if (wbc
->range_cyclic
) {
2950 index
= mapping
->writeback_index
;
2953 wbc
->range_start
= index
<< PAGE_CACHE_SHIFT
;
2954 wbc
->range_end
= LLONG_MAX
;
2955 wbc
->range_cyclic
= 0;
2957 index
= wbc
->range_start
>> PAGE_CACHE_SHIFT
;
2960 * This works around two forms of stupidity. The first is in
2961 * the writeback code, which caps the maximum number of pages
2962 * written to be 1024 pages. This is wrong on multiple
2963 * levels; different architectues have a different page size,
2964 * which changes the maximum amount of data which gets
2965 * written. Secondly, 4 megabytes is way too small. XFS
2966 * forces this value to be 16 megabytes by multiplying
2967 * nr_to_write parameter by four, and then relies on its
2968 * allocator to allocate larger extents to make them
2969 * contiguous. Unfortunately this brings us to the second
2970 * stupidity, which is that ext4's mballoc code only allocates
2971 * at most 2048 blocks. So we force contiguous writes up to
2972 * the number of dirty blocks in the inode, or
2973 * sbi->max_writeback_mb_bump whichever is smaller.
2975 max_pages
= sbi
->s_max_writeback_mb_bump
<< (20 - PAGE_CACHE_SHIFT
);
2976 if (!range_cyclic
&& range_whole
)
2977 desired_nr_to_write
= wbc
->nr_to_write
* 8;
2979 desired_nr_to_write
= ext4_num_dirty_pages(inode
, index
,
2981 if (desired_nr_to_write
> max_pages
)
2982 desired_nr_to_write
= max_pages
;
2984 if (wbc
->nr_to_write
< desired_nr_to_write
) {
2985 nr_to_writebump
= desired_nr_to_write
- wbc
->nr_to_write
;
2986 wbc
->nr_to_write
= desired_nr_to_write
;
2990 mpd
.inode
= mapping
->host
;
2992 pages_skipped
= wbc
->pages_skipped
;
2995 while (!ret
&& wbc
->nr_to_write
> 0) {
2998 * we insert one extent at a time. So we need
2999 * credit needed for single extent allocation.
3000 * journalled mode is currently not supported
3003 BUG_ON(ext4_should_journal_data(inode
));
3004 needed_blocks
= ext4_da_writepages_trans_blocks(inode
);
3006 /* start a new transaction*/
3007 handle
= ext4_journal_start(inode
, needed_blocks
);
3008 if (IS_ERR(handle
)) {
3009 ret
= PTR_ERR(handle
);
3010 ext4_msg(inode
->i_sb
, KERN_CRIT
, "%s: jbd2_start: "
3011 "%ld pages, ino %lu; err %d", __func__
,
3012 wbc
->nr_to_write
, inode
->i_ino
, ret
);
3013 goto out_writepages
;
3017 * Now call __mpage_da_writepage to find the next
3018 * contiguous region of logical blocks that need
3019 * blocks to be allocated by ext4. We don't actually
3020 * submit the blocks for I/O here, even though
3021 * write_cache_pages thinks it will, and will set the
3022 * pages as clean for write before calling
3023 * __mpage_da_writepage().
3031 mpd
.pages_written
= 0;
3033 ret
= write_cache_pages_da(mapping
, wbc
, &mpd
);
3035 * If we have a contiguous extent of pages and we
3036 * haven't done the I/O yet, map the blocks and submit
3039 if (!mpd
.io_done
&& mpd
.next_page
!= mpd
.first_page
) {
3040 if (mpage_da_map_blocks(&mpd
) == 0)
3041 mpage_da_submit_io(&mpd
);
3043 ret
= MPAGE_DA_EXTENT_TAIL
;
3045 trace_ext4_da_write_pages(inode
, &mpd
);
3046 wbc
->nr_to_write
-= mpd
.pages_written
;
3048 ext4_journal_stop(handle
);
3050 if ((mpd
.retval
== -ENOSPC
) && sbi
->s_journal
) {
3051 /* commit the transaction which would
3052 * free blocks released in the transaction
3055 jbd2_journal_force_commit_nested(sbi
->s_journal
);
3056 wbc
->pages_skipped
= pages_skipped
;
3058 } else if (ret
== MPAGE_DA_EXTENT_TAIL
) {
3060 * got one extent now try with
3063 pages_written
+= mpd
.pages_written
;
3064 wbc
->pages_skipped
= pages_skipped
;
3067 } else if (wbc
->nr_to_write
)
3069 * There is no more writeout needed
3070 * or we requested for a noblocking writeout
3071 * and we found the device congested
3075 if (!io_done
&& !cycled
) {
3078 wbc
->range_start
= index
<< PAGE_CACHE_SHIFT
;
3079 wbc
->range_end
= mapping
->writeback_index
- 1;
3082 if (pages_skipped
!= wbc
->pages_skipped
)
3083 ext4_msg(inode
->i_sb
, KERN_CRIT
,
3084 "This should not happen leaving %s "
3085 "with nr_to_write = %ld ret = %d",
3086 __func__
, wbc
->nr_to_write
, ret
);
3089 index
+= pages_written
;
3090 wbc
->range_cyclic
= range_cyclic
;
3091 if (wbc
->range_cyclic
|| (range_whole
&& wbc
->nr_to_write
> 0))
3093 * set the writeback_index so that range_cyclic
3094 * mode will write it back later
3096 mapping
->writeback_index
= index
;
3099 wbc
->nr_to_write
-= nr_to_writebump
;
3100 wbc
->range_start
= range_start
;
3101 trace_ext4_da_writepages_result(inode
, wbc
, ret
, pages_written
);
3105 #define FALL_BACK_TO_NONDELALLOC 1
3106 static int ext4_nonda_switch(struct super_block
*sb
)
3108 s64 free_blocks
, dirty_blocks
;
3109 struct ext4_sb_info
*sbi
= EXT4_SB(sb
);
3112 * switch to non delalloc mode if we are running low
3113 * on free block. The free block accounting via percpu
3114 * counters can get slightly wrong with percpu_counter_batch getting
3115 * accumulated on each CPU without updating global counters
3116 * Delalloc need an accurate free block accounting. So switch
3117 * to non delalloc when we are near to error range.
3119 free_blocks
= percpu_counter_read_positive(&sbi
->s_freeblocks_counter
);
3120 dirty_blocks
= percpu_counter_read_positive(&sbi
->s_dirtyblocks_counter
);
3121 if (2 * free_blocks
< 3 * dirty_blocks
||
3122 free_blocks
< (dirty_blocks
+ EXT4_FREEBLOCKS_WATERMARK
)) {
3124 * free block count is less than 150% of dirty blocks
3125 * or free blocks is less than watermark
3130 * Even if we don't switch but are nearing capacity,
3131 * start pushing delalloc when 1/2 of free blocks are dirty.
3133 if (free_blocks
< 2 * dirty_blocks
)
3134 writeback_inodes_sb_if_idle(sb
);
3139 static int ext4_da_write_begin(struct file
*file
, struct address_space
*mapping
,
3140 loff_t pos
, unsigned len
, unsigned flags
,
3141 struct page
**pagep
, void **fsdata
)
3143 int ret
, retries
= 0;
3146 struct inode
*inode
= mapping
->host
;
3149 index
= pos
>> PAGE_CACHE_SHIFT
;
3151 if (ext4_nonda_switch(inode
->i_sb
)) {
3152 *fsdata
= (void *)FALL_BACK_TO_NONDELALLOC
;
3153 return ext4_write_begin(file
, mapping
, pos
,
3154 len
, flags
, pagep
, fsdata
);
3156 *fsdata
= (void *)0;
3157 trace_ext4_da_write_begin(inode
, pos
, len
, flags
);
3160 * With delayed allocation, we don't log the i_disksize update
3161 * if there is delayed block allocation. But we still need
3162 * to journalling the i_disksize update if writes to the end
3163 * of file which has an already mapped buffer.
3165 handle
= ext4_journal_start(inode
, 1);
3166 if (IS_ERR(handle
)) {
3167 ret
= PTR_ERR(handle
);
3170 /* We cannot recurse into the filesystem as the transaction is already
3172 flags
|= AOP_FLAG_NOFS
;
3174 page
= grab_cache_page_write_begin(mapping
, index
, flags
);
3176 ext4_journal_stop(handle
);
3182 ret
= block_write_begin(file
, mapping
, pos
, len
, flags
, pagep
, fsdata
,
3183 ext4_da_get_block_prep
);
3186 ext4_journal_stop(handle
);
3187 page_cache_release(page
);
3189 * block_write_begin may have instantiated a few blocks
3190 * outside i_size. Trim these off again. Don't need
3191 * i_size_read because we hold i_mutex.
3193 if (pos
+ len
> inode
->i_size
)
3194 ext4_truncate_failed_write(inode
);
3197 if (ret
== -ENOSPC
&& ext4_should_retry_alloc(inode
->i_sb
, &retries
))
3204 * Check if we should update i_disksize
3205 * when write to the end of file but not require block allocation
3207 static int ext4_da_should_update_i_disksize(struct page
*page
,
3208 unsigned long offset
)
3210 struct buffer_head
*bh
;
3211 struct inode
*inode
= page
->mapping
->host
;
3215 bh
= page_buffers(page
);
3216 idx
= offset
>> inode
->i_blkbits
;
3218 for (i
= 0; i
< idx
; i
++)
3219 bh
= bh
->b_this_page
;
3221 if (!buffer_mapped(bh
) || (buffer_delay(bh
)) || buffer_unwritten(bh
))
3226 static int ext4_da_write_end(struct file
*file
,
3227 struct address_space
*mapping
,
3228 loff_t pos
, unsigned len
, unsigned copied
,
3229 struct page
*page
, void *fsdata
)
3231 struct inode
*inode
= mapping
->host
;
3233 handle_t
*handle
= ext4_journal_current_handle();
3235 unsigned long start
, end
;
3236 int write_mode
= (int)(unsigned long)fsdata
;
3238 if (write_mode
== FALL_BACK_TO_NONDELALLOC
) {
3239 if (ext4_should_order_data(inode
)) {
3240 return ext4_ordered_write_end(file
, mapping
, pos
,
3241 len
, copied
, page
, fsdata
);
3242 } else if (ext4_should_writeback_data(inode
)) {
3243 return ext4_writeback_write_end(file
, mapping
, pos
,
3244 len
, copied
, page
, fsdata
);
3250 trace_ext4_da_write_end(inode
, pos
, len
, copied
);
3251 start
= pos
& (PAGE_CACHE_SIZE
- 1);
3252 end
= start
+ copied
- 1;
3255 * generic_write_end() will run mark_inode_dirty() if i_size
3256 * changes. So let's piggyback the i_disksize mark_inode_dirty
3260 new_i_size
= pos
+ copied
;
3261 if (new_i_size
> EXT4_I(inode
)->i_disksize
) {
3262 if (ext4_da_should_update_i_disksize(page
, end
)) {
3263 down_write(&EXT4_I(inode
)->i_data_sem
);
3264 if (new_i_size
> EXT4_I(inode
)->i_disksize
) {
3266 * Updating i_disksize when extending file
3267 * without needing block allocation
3269 if (ext4_should_order_data(inode
))
3270 ret
= ext4_jbd2_file_inode(handle
,
3273 EXT4_I(inode
)->i_disksize
= new_i_size
;
3275 up_write(&EXT4_I(inode
)->i_data_sem
);
3276 /* We need to mark inode dirty even if
3277 * new_i_size is less that inode->i_size
3278 * bu greater than i_disksize.(hint delalloc)
3280 ext4_mark_inode_dirty(handle
, inode
);
3283 ret2
= generic_write_end(file
, mapping
, pos
, len
, copied
,
3288 ret2
= ext4_journal_stop(handle
);
3292 return ret
? ret
: copied
;
3295 static void ext4_da_invalidatepage(struct page
*page
, unsigned long offset
)
3298 * Drop reserved blocks
3300 BUG_ON(!PageLocked(page
));
3301 if (!page_has_buffers(page
))
3304 ext4_da_page_release_reservation(page
, offset
);
3307 ext4_invalidatepage(page
, offset
);
3313 * Force all delayed allocation blocks to be allocated for a given inode.
3315 int ext4_alloc_da_blocks(struct inode
*inode
)
3317 trace_ext4_alloc_da_blocks(inode
);
3319 if (!EXT4_I(inode
)->i_reserved_data_blocks
&&
3320 !EXT4_I(inode
)->i_reserved_meta_blocks
)
3324 * We do something simple for now. The filemap_flush() will
3325 * also start triggering a write of the data blocks, which is
3326 * not strictly speaking necessary (and for users of
3327 * laptop_mode, not even desirable). However, to do otherwise
3328 * would require replicating code paths in:
3330 * ext4_da_writepages() ->
3331 * write_cache_pages() ---> (via passed in callback function)
3332 * __mpage_da_writepage() -->
3333 * mpage_add_bh_to_extent()
3334 * mpage_da_map_blocks()
3336 * The problem is that write_cache_pages(), located in
3337 * mm/page-writeback.c, marks pages clean in preparation for
3338 * doing I/O, which is not desirable if we're not planning on
3341 * We could call write_cache_pages(), and then redirty all of
3342 * the pages by calling redirty_page_for_writeback() but that
3343 * would be ugly in the extreme. So instead we would need to
3344 * replicate parts of the code in the above functions,
3345 * simplifying them becuase we wouldn't actually intend to
3346 * write out the pages, but rather only collect contiguous
3347 * logical block extents, call the multi-block allocator, and
3348 * then update the buffer heads with the block allocations.
3350 * For now, though, we'll cheat by calling filemap_flush(),
3351 * which will map the blocks, and start the I/O, but not
3352 * actually wait for the I/O to complete.
3354 return filemap_flush(inode
->i_mapping
);
3358 * bmap() is special. It gets used by applications such as lilo and by
3359 * the swapper to find the on-disk block of a specific piece of data.
3361 * Naturally, this is dangerous if the block concerned is still in the
3362 * journal. If somebody makes a swapfile on an ext4 data-journaling
3363 * filesystem and enables swap, then they may get a nasty shock when the
3364 * data getting swapped to that swapfile suddenly gets overwritten by
3365 * the original zero's written out previously to the journal and
3366 * awaiting writeback in the kernel's buffer cache.
3368 * So, if we see any bmap calls here on a modified, data-journaled file,
3369 * take extra steps to flush any blocks which might be in the cache.
3371 static sector_t
ext4_bmap(struct address_space
*mapping
, sector_t block
)
3373 struct inode
*inode
= mapping
->host
;
3377 if (mapping_tagged(mapping
, PAGECACHE_TAG_DIRTY
) &&
3378 test_opt(inode
->i_sb
, DELALLOC
)) {
3380 * With delalloc we want to sync the file
3381 * so that we can make sure we allocate
3384 filemap_write_and_wait(mapping
);
3387 if (EXT4_JOURNAL(inode
) &&
3388 ext4_test_inode_state(inode
, EXT4_STATE_JDATA
)) {
3390 * This is a REALLY heavyweight approach, but the use of
3391 * bmap on dirty files is expected to be extremely rare:
3392 * only if we run lilo or swapon on a freshly made file
3393 * do we expect this to happen.
3395 * (bmap requires CAP_SYS_RAWIO so this does not
3396 * represent an unprivileged user DOS attack --- we'd be
3397 * in trouble if mortal users could trigger this path at
3400 * NB. EXT4_STATE_JDATA is not set on files other than
3401 * regular files. If somebody wants to bmap a directory
3402 * or symlink and gets confused because the buffer
3403 * hasn't yet been flushed to disk, they deserve
3404 * everything they get.
3407 ext4_clear_inode_state(inode
, EXT4_STATE_JDATA
);
3408 journal
= EXT4_JOURNAL(inode
);
3409 jbd2_journal_lock_updates(journal
);
3410 err
= jbd2_journal_flush(journal
);
3411 jbd2_journal_unlock_updates(journal
);
3417 return generic_block_bmap(mapping
, block
, ext4_get_block
);
3420 static int ext4_readpage(struct file
*file
, struct page
*page
)
3422 return mpage_readpage(page
, ext4_get_block
);
3426 ext4_readpages(struct file
*file
, struct address_space
*mapping
,
3427 struct list_head
*pages
, unsigned nr_pages
)
3429 return mpage_readpages(mapping
, pages
, nr_pages
, ext4_get_block
);
3432 static void ext4_free_io_end(ext4_io_end_t
*io
)
3441 static void ext4_invalidatepage_free_endio(struct page
*page
, unsigned long offset
)
3443 struct buffer_head
*head
, *bh
;
3444 unsigned int curr_off
= 0;
3446 if (!page_has_buffers(page
))
3448 head
= bh
= page_buffers(page
);
3450 if (offset
<= curr_off
&& test_clear_buffer_uninit(bh
)
3452 ext4_free_io_end(bh
->b_private
);
3453 bh
->b_private
= NULL
;
3454 bh
->b_end_io
= NULL
;
3456 curr_off
= curr_off
+ bh
->b_size
;
3457 bh
= bh
->b_this_page
;
3458 } while (bh
!= head
);
3461 static void ext4_invalidatepage(struct page
*page
, unsigned long offset
)
3463 journal_t
*journal
= EXT4_JOURNAL(page
->mapping
->host
);
3466 * free any io_end structure allocated for buffers to be discarded
3468 if (ext4_should_dioread_nolock(page
->mapping
->host
))
3469 ext4_invalidatepage_free_endio(page
, offset
);
3471 * If it's a full truncate we just forget about the pending dirtying
3474 ClearPageChecked(page
);
3477 jbd2_journal_invalidatepage(journal
, page
, offset
);
3479 block_invalidatepage(page
, offset
);
3482 static int ext4_releasepage(struct page
*page
, gfp_t wait
)
3484 journal_t
*journal
= EXT4_JOURNAL(page
->mapping
->host
);
3486 WARN_ON(PageChecked(page
));
3487 if (!page_has_buffers(page
))
3490 return jbd2_journal_try_to_free_buffers(journal
, page
, wait
);
3492 return try_to_free_buffers(page
);
3496 * O_DIRECT for ext3 (or indirect map) based files
3498 * If the O_DIRECT write will extend the file then add this inode to the
3499 * orphan list. So recovery will truncate it back to the original size
3500 * if the machine crashes during the write.
3502 * If the O_DIRECT write is intantiating holes inside i_size and the machine
3503 * crashes then stale disk data _may_ be exposed inside the file. But current
3504 * VFS code falls back into buffered path in that case so we are safe.
3506 static ssize_t
ext4_ind_direct_IO(int rw
, struct kiocb
*iocb
,
3507 const struct iovec
*iov
, loff_t offset
,
3508 unsigned long nr_segs
)
3510 struct file
*file
= iocb
->ki_filp
;
3511 struct inode
*inode
= file
->f_mapping
->host
;
3512 struct ext4_inode_info
*ei
= EXT4_I(inode
);
3516 size_t count
= iov_length(iov
, nr_segs
);
3520 loff_t final_size
= offset
+ count
;
3522 if (final_size
> inode
->i_size
) {
3523 /* Credits for sb + inode write */
3524 handle
= ext4_journal_start(inode
, 2);
3525 if (IS_ERR(handle
)) {
3526 ret
= PTR_ERR(handle
);
3529 ret
= ext4_orphan_add(handle
, inode
);
3531 ext4_journal_stop(handle
);
3535 ei
->i_disksize
= inode
->i_size
;
3536 ext4_journal_stop(handle
);
3541 if (rw
== READ
&& ext4_should_dioread_nolock(inode
))
3542 ret
= blockdev_direct_IO_no_locking(rw
, iocb
, inode
,
3543 inode
->i_sb
->s_bdev
, iov
,
3545 ext4_get_block
, NULL
);
3547 ret
= blockdev_direct_IO(rw
, iocb
, inode
,
3548 inode
->i_sb
->s_bdev
, iov
,
3550 ext4_get_block
, NULL
);
3551 if (ret
== -ENOSPC
&& ext4_should_retry_alloc(inode
->i_sb
, &retries
))
3557 /* Credits for sb + inode write */
3558 handle
= ext4_journal_start(inode
, 2);
3559 if (IS_ERR(handle
)) {
3560 /* This is really bad luck. We've written the data
3561 * but cannot extend i_size. Bail out and pretend
3562 * the write failed... */
3563 ret
= PTR_ERR(handle
);
3565 ext4_orphan_del(NULL
, inode
);
3570 ext4_orphan_del(handle
, inode
);
3572 loff_t end
= offset
+ ret
;
3573 if (end
> inode
->i_size
) {
3574 ei
->i_disksize
= end
;
3575 i_size_write(inode
, end
);
3577 * We're going to return a positive `ret'
3578 * here due to non-zero-length I/O, so there's
3579 * no way of reporting error returns from
3580 * ext4_mark_inode_dirty() to userspace. So
3583 ext4_mark_inode_dirty(handle
, inode
);
3586 err
= ext4_journal_stop(handle
);
3595 * ext4_get_block used when preparing for a DIO write or buffer write.
3596 * We allocate an uinitialized extent if blocks haven't been allocated.
3597 * The extent will be converted to initialized after the IO is complete.
3599 static int ext4_get_block_write(struct inode
*inode
, sector_t iblock
,
3600 struct buffer_head
*bh_result
, int create
)
3602 ext4_debug("ext4_get_block_write: inode %lu, create flag %d\n",
3603 inode
->i_ino
, create
);
3604 return _ext4_get_block(inode
, iblock
, bh_result
,
3605 EXT4_GET_BLOCKS_IO_CREATE_EXT
);
3608 static void dump_completed_IO(struct inode
* inode
)
3611 struct list_head
*cur
, *before
, *after
;
3612 ext4_io_end_t
*io
, *io0
, *io1
;
3613 unsigned long flags
;
3615 if (list_empty(&EXT4_I(inode
)->i_completed_io_list
)){
3616 ext4_debug("inode %lu completed_io list is empty\n", inode
->i_ino
);
3620 ext4_debug("Dump inode %lu completed_io list \n", inode
->i_ino
);
3621 spin_lock_irqsave(&EXT4_I(inode
)->i_completed_io_lock
, flags
);
3622 list_for_each_entry(io
, &EXT4_I(inode
)->i_completed_io_list
, list
){
3625 io0
= container_of(before
, ext4_io_end_t
, list
);
3627 io1
= container_of(after
, ext4_io_end_t
, list
);
3629 ext4_debug("io 0x%p from inode %lu,prev 0x%p,next 0x%p\n",
3630 io
, inode
->i_ino
, io0
, io1
);
3632 spin_unlock_irqrestore(&EXT4_I(inode
)->i_completed_io_lock
, flags
);
3637 * check a range of space and convert unwritten extents to written.
3639 static int ext4_end_io_nolock(ext4_io_end_t
*io
)
3641 struct inode
*inode
= io
->inode
;
3642 loff_t offset
= io
->offset
;
3643 ssize_t size
= io
->size
;
3646 ext4_debug("ext4_end_io_nolock: io 0x%p from inode %lu,list->next 0x%p,"
3647 "list->prev 0x%p\n",
3648 io
, inode
->i_ino
, io
->list
.next
, io
->list
.prev
);
3650 if (list_empty(&io
->list
))
3653 if (io
->flag
!= EXT4_IO_UNWRITTEN
)
3656 ret
= ext4_convert_unwritten_extents(inode
, offset
, size
);
3658 printk(KERN_EMERG
"%s: failed to convert unwritten"
3659 "extents to written extents, error is %d"
3660 " io is still on inode %lu aio dio list\n",
3661 __func__
, ret
, inode
->i_ino
);
3665 /* clear the DIO AIO unwritten flag */
3671 * work on completed aio dio IO, to convert unwritten extents to extents
3673 static void ext4_end_io_work(struct work_struct
*work
)
3675 ext4_io_end_t
*io
= container_of(work
, ext4_io_end_t
, work
);
3676 struct inode
*inode
= io
->inode
;
3677 struct ext4_inode_info
*ei
= EXT4_I(inode
);
3678 unsigned long flags
;
3681 mutex_lock(&inode
->i_mutex
);
3682 ret
= ext4_end_io_nolock(io
);
3684 mutex_unlock(&inode
->i_mutex
);
3688 spin_lock_irqsave(&ei
->i_completed_io_lock
, flags
);
3689 if (!list_empty(&io
->list
))
3690 list_del_init(&io
->list
);
3691 spin_unlock_irqrestore(&ei
->i_completed_io_lock
, flags
);
3692 mutex_unlock(&inode
->i_mutex
);
3693 ext4_free_io_end(io
);
3697 * This function is called from ext4_sync_file().
3699 * When IO is completed, the work to convert unwritten extents to
3700 * written is queued on workqueue but may not get immediately
3701 * scheduled. When fsync is called, we need to ensure the
3702 * conversion is complete before fsync returns.
3703 * The inode keeps track of a list of pending/completed IO that
3704 * might needs to do the conversion. This function walks through
3705 * the list and convert the related unwritten extents for completed IO
3707 * The function return the number of pending IOs on success.
3709 int flush_completed_IO(struct inode
*inode
)
3712 struct ext4_inode_info
*ei
= EXT4_I(inode
);
3713 unsigned long flags
;
3717 if (list_empty(&ei
->i_completed_io_list
))
3720 dump_completed_IO(inode
);
3721 spin_lock_irqsave(&ei
->i_completed_io_lock
, flags
);
3722 while (!list_empty(&ei
->i_completed_io_list
)){
3723 io
= list_entry(ei
->i_completed_io_list
.next
,
3724 ext4_io_end_t
, list
);
3726 * Calling ext4_end_io_nolock() to convert completed
3729 * When ext4_sync_file() is called, run_queue() may already
3730 * about to flush the work corresponding to this io structure.
3731 * It will be upset if it founds the io structure related
3732 * to the work-to-be schedule is freed.
3734 * Thus we need to keep the io structure still valid here after
3735 * convertion finished. The io structure has a flag to
3736 * avoid double converting from both fsync and background work
3739 spin_unlock_irqrestore(&ei
->i_completed_io_lock
, flags
);
3740 ret
= ext4_end_io_nolock(io
);
3741 spin_lock_irqsave(&ei
->i_completed_io_lock
, flags
);
3745 list_del_init(&io
->list
);
3747 spin_unlock_irqrestore(&ei
->i_completed_io_lock
, flags
);
3748 return (ret2
< 0) ? ret2
: 0;
3751 static ext4_io_end_t
*ext4_init_io_end (struct inode
*inode
, gfp_t flags
)
3753 ext4_io_end_t
*io
= NULL
;
3755 io
= kmalloc(sizeof(*io
), flags
);
3764 INIT_WORK(&io
->work
, ext4_end_io_work
);
3765 INIT_LIST_HEAD(&io
->list
);
3771 static void ext4_end_io_dio(struct kiocb
*iocb
, loff_t offset
,
3772 ssize_t size
, void *private)
3774 ext4_io_end_t
*io_end
= iocb
->private;
3775 struct workqueue_struct
*wq
;
3776 unsigned long flags
;
3777 struct ext4_inode_info
*ei
;
3779 /* if not async direct IO or dio with 0 bytes write, just return */
3780 if (!io_end
|| !size
)
3783 ext_debug("ext4_end_io_dio(): io_end 0x%p"
3784 "for inode %lu, iocb 0x%p, offset %llu, size %llu\n",
3785 iocb
->private, io_end
->inode
->i_ino
, iocb
, offset
,
3788 /* if not aio dio with unwritten extents, just free io and return */
3789 if (io_end
->flag
!= EXT4_IO_UNWRITTEN
){
3790 ext4_free_io_end(io_end
);
3791 iocb
->private = NULL
;
3795 io_end
->offset
= offset
;
3796 io_end
->size
= size
;
3797 io_end
->flag
= EXT4_IO_UNWRITTEN
;
3798 wq
= EXT4_SB(io_end
->inode
->i_sb
)->dio_unwritten_wq
;
3800 /* queue the work to convert unwritten extents to written */
3801 queue_work(wq
, &io_end
->work
);
3803 /* Add the io_end to per-inode completed aio dio list*/
3804 ei
= EXT4_I(io_end
->inode
);
3805 spin_lock_irqsave(&ei
->i_completed_io_lock
, flags
);
3806 list_add_tail(&io_end
->list
, &ei
->i_completed_io_list
);
3807 spin_unlock_irqrestore(&ei
->i_completed_io_lock
, flags
);
3808 iocb
->private = NULL
;
3811 static void ext4_end_io_buffer_write(struct buffer_head
*bh
, int uptodate
)
3813 ext4_io_end_t
*io_end
= bh
->b_private
;
3814 struct workqueue_struct
*wq
;
3815 struct inode
*inode
;
3816 unsigned long flags
;
3818 if (!test_clear_buffer_uninit(bh
) || !io_end
)
3821 if (!(io_end
->inode
->i_sb
->s_flags
& MS_ACTIVE
)) {
3822 printk("sb umounted, discard end_io request for inode %lu\n",
3823 io_end
->inode
->i_ino
);
3824 ext4_free_io_end(io_end
);
3828 io_end
->flag
= EXT4_IO_UNWRITTEN
;
3829 inode
= io_end
->inode
;
3831 /* Add the io_end to per-inode completed io list*/
3832 spin_lock_irqsave(&EXT4_I(inode
)->i_completed_io_lock
, flags
);
3833 list_add_tail(&io_end
->list
, &EXT4_I(inode
)->i_completed_io_list
);
3834 spin_unlock_irqrestore(&EXT4_I(inode
)->i_completed_io_lock
, flags
);
3836 wq
= EXT4_SB(inode
->i_sb
)->dio_unwritten_wq
;
3837 /* queue the work to convert unwritten extents to written */
3838 queue_work(wq
, &io_end
->work
);
3840 bh
->b_private
= NULL
;
3841 bh
->b_end_io
= NULL
;
3842 clear_buffer_uninit(bh
);
3843 end_buffer_async_write(bh
, uptodate
);
3846 static int ext4_set_bh_endio(struct buffer_head
*bh
, struct inode
*inode
)
3848 ext4_io_end_t
*io_end
;
3849 struct page
*page
= bh
->b_page
;
3850 loff_t offset
= (sector_t
)page
->index
<< PAGE_CACHE_SHIFT
;
3851 size_t size
= bh
->b_size
;
3854 io_end
= ext4_init_io_end(inode
, GFP_ATOMIC
);
3856 if (printk_ratelimit())
3857 printk(KERN_WARNING
"%s: allocation fail\n", __func__
);
3861 io_end
->offset
= offset
;
3862 io_end
->size
= size
;
3864 * We need to hold a reference to the page to make sure it
3865 * doesn't get evicted before ext4_end_io_work() has a chance
3866 * to convert the extent from written to unwritten.
3868 io_end
->page
= page
;
3869 get_page(io_end
->page
);
3871 bh
->b_private
= io_end
;
3872 bh
->b_end_io
= ext4_end_io_buffer_write
;
3877 * For ext4 extent files, ext4 will do direct-io write to holes,
3878 * preallocated extents, and those write extend the file, no need to
3879 * fall back to buffered IO.
3881 * For holes, we fallocate those blocks, mark them as unintialized
3882 * If those blocks were preallocated, we mark sure they are splited, but
3883 * still keep the range to write as unintialized.
3885 * The unwrritten extents will be converted to written when DIO is completed.
3886 * For async direct IO, since the IO may still pending when return, we
3887 * set up an end_io call back function, which will do the convertion
3888 * when async direct IO completed.
3890 * If the O_DIRECT write will extend the file then add this inode to the
3891 * orphan list. So recovery will truncate it back to the original size
3892 * if the machine crashes during the write.
3895 static ssize_t
ext4_ext_direct_IO(int rw
, struct kiocb
*iocb
,
3896 const struct iovec
*iov
, loff_t offset
,
3897 unsigned long nr_segs
)
3899 struct file
*file
= iocb
->ki_filp
;
3900 struct inode
*inode
= file
->f_mapping
->host
;
3902 size_t count
= iov_length(iov
, nr_segs
);
3904 loff_t final_size
= offset
+ count
;
3905 if (rw
== WRITE
&& final_size
<= inode
->i_size
) {
3907 * We could direct write to holes and fallocate.
3909 * Allocated blocks to fill the hole are marked as uninitialized
3910 * to prevent paralel buffered read to expose the stale data
3911 * before DIO complete the data IO.
3913 * As to previously fallocated extents, ext4 get_block
3914 * will just simply mark the buffer mapped but still
3915 * keep the extents uninitialized.
3917 * for non AIO case, we will convert those unwritten extents
3918 * to written after return back from blockdev_direct_IO.
3920 * for async DIO, the conversion needs to be defered when
3921 * the IO is completed. The ext4 end_io callback function
3922 * will be called to take care of the conversion work.
3923 * Here for async case, we allocate an io_end structure to
3926 iocb
->private = NULL
;
3927 EXT4_I(inode
)->cur_aio_dio
= NULL
;
3928 if (!is_sync_kiocb(iocb
)) {
3929 iocb
->private = ext4_init_io_end(inode
, GFP_NOFS
);
3933 * we save the io structure for current async
3934 * direct IO, so that later ext4_get_blocks()
3935 * could flag the io structure whether there
3936 * is a unwritten extents needs to be converted
3937 * when IO is completed.
3939 EXT4_I(inode
)->cur_aio_dio
= iocb
->private;
3942 ret
= blockdev_direct_IO(rw
, iocb
, inode
,
3943 inode
->i_sb
->s_bdev
, iov
,
3945 ext4_get_block_write
,
3948 EXT4_I(inode
)->cur_aio_dio
= NULL
;
3950 * The io_end structure takes a reference to the inode,
3951 * that structure needs to be destroyed and the
3952 * reference to the inode need to be dropped, when IO is
3953 * complete, even with 0 byte write, or failed.
3955 * In the successful AIO DIO case, the io_end structure will be
3956 * desctroyed and the reference to the inode will be dropped
3957 * after the end_io call back function is called.
3959 * In the case there is 0 byte write, or error case, since
3960 * VFS direct IO won't invoke the end_io call back function,
3961 * we need to free the end_io structure here.
3963 if (ret
!= -EIOCBQUEUED
&& ret
<= 0 && iocb
->private) {
3964 ext4_free_io_end(iocb
->private);
3965 iocb
->private = NULL
;
3966 } else if (ret
> 0 && ext4_test_inode_state(inode
,
3967 EXT4_STATE_DIO_UNWRITTEN
)) {
3970 * for non AIO case, since the IO is already
3971 * completed, we could do the convertion right here
3973 err
= ext4_convert_unwritten_extents(inode
,
3977 ext4_clear_inode_state(inode
, EXT4_STATE_DIO_UNWRITTEN
);
3982 /* for write the the end of file case, we fall back to old way */
3983 return ext4_ind_direct_IO(rw
, iocb
, iov
, offset
, nr_segs
);
3986 static ssize_t
ext4_direct_IO(int rw
, struct kiocb
*iocb
,
3987 const struct iovec
*iov
, loff_t offset
,
3988 unsigned long nr_segs
)
3990 struct file
*file
= iocb
->ki_filp
;
3991 struct inode
*inode
= file
->f_mapping
->host
;
3993 if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
))
3994 return ext4_ext_direct_IO(rw
, iocb
, iov
, offset
, nr_segs
);
3996 return ext4_ind_direct_IO(rw
, iocb
, iov
, offset
, nr_segs
);
4000 * Pages can be marked dirty completely asynchronously from ext4's journalling
4001 * activity. By filemap_sync_pte(), try_to_unmap_one(), etc. We cannot do
4002 * much here because ->set_page_dirty is called under VFS locks. The page is
4003 * not necessarily locked.
4005 * We cannot just dirty the page and leave attached buffers clean, because the
4006 * buffers' dirty state is "definitive". We cannot just set the buffers dirty
4007 * or jbddirty because all the journalling code will explode.
4009 * So what we do is to mark the page "pending dirty" and next time writepage
4010 * is called, propagate that into the buffers appropriately.
4012 static int ext4_journalled_set_page_dirty(struct page
*page
)
4014 SetPageChecked(page
);
4015 return __set_page_dirty_nobuffers(page
);
4018 static const struct address_space_operations ext4_ordered_aops
= {
4019 .readpage
= ext4_readpage
,
4020 .readpages
= ext4_readpages
,
4021 .writepage
= ext4_writepage
,
4022 .sync_page
= block_sync_page
,
4023 .write_begin
= ext4_write_begin
,
4024 .write_end
= ext4_ordered_write_end
,
4026 .invalidatepage
= ext4_invalidatepage
,
4027 .releasepage
= ext4_releasepage
,
4028 .direct_IO
= ext4_direct_IO
,
4029 .migratepage
= buffer_migrate_page
,
4030 .is_partially_uptodate
= block_is_partially_uptodate
,
4031 .error_remove_page
= generic_error_remove_page
,
4034 static const struct address_space_operations ext4_writeback_aops
= {
4035 .readpage
= ext4_readpage
,
4036 .readpages
= ext4_readpages
,
4037 .writepage
= ext4_writepage
,
4038 .sync_page
= block_sync_page
,
4039 .write_begin
= ext4_write_begin
,
4040 .write_end
= ext4_writeback_write_end
,
4042 .invalidatepage
= ext4_invalidatepage
,
4043 .releasepage
= ext4_releasepage
,
4044 .direct_IO
= ext4_direct_IO
,
4045 .migratepage
= buffer_migrate_page
,
4046 .is_partially_uptodate
= block_is_partially_uptodate
,
4047 .error_remove_page
= generic_error_remove_page
,
4050 static const struct address_space_operations ext4_journalled_aops
= {
4051 .readpage
= ext4_readpage
,
4052 .readpages
= ext4_readpages
,
4053 .writepage
= ext4_writepage
,
4054 .sync_page
= block_sync_page
,
4055 .write_begin
= ext4_write_begin
,
4056 .write_end
= ext4_journalled_write_end
,
4057 .set_page_dirty
= ext4_journalled_set_page_dirty
,
4059 .invalidatepage
= ext4_invalidatepage
,
4060 .releasepage
= ext4_releasepage
,
4061 .is_partially_uptodate
= block_is_partially_uptodate
,
4062 .error_remove_page
= generic_error_remove_page
,
4065 static const struct address_space_operations ext4_da_aops
= {
4066 .readpage
= ext4_readpage
,
4067 .readpages
= ext4_readpages
,
4068 .writepage
= ext4_writepage
,
4069 .writepages
= ext4_da_writepages
,
4070 .sync_page
= block_sync_page
,
4071 .write_begin
= ext4_da_write_begin
,
4072 .write_end
= ext4_da_write_end
,
4074 .invalidatepage
= ext4_da_invalidatepage
,
4075 .releasepage
= ext4_releasepage
,
4076 .direct_IO
= ext4_direct_IO
,
4077 .migratepage
= buffer_migrate_page
,
4078 .is_partially_uptodate
= block_is_partially_uptodate
,
4079 .error_remove_page
= generic_error_remove_page
,
4082 void ext4_set_aops(struct inode
*inode
)
4084 if (ext4_should_order_data(inode
) &&
4085 test_opt(inode
->i_sb
, DELALLOC
))
4086 inode
->i_mapping
->a_ops
= &ext4_da_aops
;
4087 else if (ext4_should_order_data(inode
))
4088 inode
->i_mapping
->a_ops
= &ext4_ordered_aops
;
4089 else if (ext4_should_writeback_data(inode
) &&
4090 test_opt(inode
->i_sb
, DELALLOC
))
4091 inode
->i_mapping
->a_ops
= &ext4_da_aops
;
4092 else if (ext4_should_writeback_data(inode
))
4093 inode
->i_mapping
->a_ops
= &ext4_writeback_aops
;
4095 inode
->i_mapping
->a_ops
= &ext4_journalled_aops
;
4099 * ext4_block_truncate_page() zeroes out a mapping from file offset `from'
4100 * up to the end of the block which corresponds to `from'.
4101 * This required during truncate. We need to physically zero the tail end
4102 * of that block so it doesn't yield old data if the file is later grown.
4104 int ext4_block_truncate_page(handle_t
*handle
,
4105 struct address_space
*mapping
, loff_t from
)
4107 ext4_fsblk_t index
= from
>> PAGE_CACHE_SHIFT
;
4108 unsigned offset
= from
& (PAGE_CACHE_SIZE
-1);
4109 unsigned blocksize
, length
, pos
;
4111 struct inode
*inode
= mapping
->host
;
4112 struct buffer_head
*bh
;
4116 page
= find_or_create_page(mapping
, from
>> PAGE_CACHE_SHIFT
,
4117 mapping_gfp_mask(mapping
) & ~__GFP_FS
);
4121 blocksize
= inode
->i_sb
->s_blocksize
;
4122 length
= blocksize
- (offset
& (blocksize
- 1));
4123 iblock
= index
<< (PAGE_CACHE_SHIFT
- inode
->i_sb
->s_blocksize_bits
);
4125 if (!page_has_buffers(page
))
4126 create_empty_buffers(page
, blocksize
, 0);
4128 /* Find the buffer that contains "offset" */
4129 bh
= page_buffers(page
);
4131 while (offset
>= pos
) {
4132 bh
= bh
->b_this_page
;
4138 if (buffer_freed(bh
)) {
4139 BUFFER_TRACE(bh
, "freed: skip");
4143 if (!buffer_mapped(bh
)) {
4144 BUFFER_TRACE(bh
, "unmapped");
4145 ext4_get_block(inode
, iblock
, bh
, 0);
4146 /* unmapped? It's a hole - nothing to do */
4147 if (!buffer_mapped(bh
)) {
4148 BUFFER_TRACE(bh
, "still unmapped");
4153 /* Ok, it's mapped. Make sure it's up-to-date */
4154 if (PageUptodate(page
))
4155 set_buffer_uptodate(bh
);
4157 if (!buffer_uptodate(bh
)) {
4159 ll_rw_block(READ
, 1, &bh
);
4161 /* Uhhuh. Read error. Complain and punt. */
4162 if (!buffer_uptodate(bh
))
4166 if (ext4_should_journal_data(inode
)) {
4167 BUFFER_TRACE(bh
, "get write access");
4168 err
= ext4_journal_get_write_access(handle
, bh
);
4173 zero_user(page
, offset
, length
);
4175 BUFFER_TRACE(bh
, "zeroed end of block");
4178 if (ext4_should_journal_data(inode
)) {
4179 err
= ext4_handle_dirty_metadata(handle
, inode
, bh
);
4181 if (ext4_should_order_data(inode
))
4182 err
= ext4_jbd2_file_inode(handle
, inode
);
4183 mark_buffer_dirty(bh
);
4188 page_cache_release(page
);
4193 * Probably it should be a library function... search for first non-zero word
4194 * or memcmp with zero_page, whatever is better for particular architecture.
4197 static inline int all_zeroes(__le32
*p
, __le32
*q
)
4206 * ext4_find_shared - find the indirect blocks for partial truncation.
4207 * @inode: inode in question
4208 * @depth: depth of the affected branch
4209 * @offsets: offsets of pointers in that branch (see ext4_block_to_path)
4210 * @chain: place to store the pointers to partial indirect blocks
4211 * @top: place to the (detached) top of branch
4213 * This is a helper function used by ext4_truncate().
4215 * When we do truncate() we may have to clean the ends of several
4216 * indirect blocks but leave the blocks themselves alive. Block is
4217 * partially truncated if some data below the new i_size is refered
4218 * from it (and it is on the path to the first completely truncated
4219 * data block, indeed). We have to free the top of that path along
4220 * with everything to the right of the path. Since no allocation
4221 * past the truncation point is possible until ext4_truncate()
4222 * finishes, we may safely do the latter, but top of branch may
4223 * require special attention - pageout below the truncation point
4224 * might try to populate it.
4226 * We atomically detach the top of branch from the tree, store the
4227 * block number of its root in *@top, pointers to buffer_heads of
4228 * partially truncated blocks - in @chain[].bh and pointers to
4229 * their last elements that should not be removed - in
4230 * @chain[].p. Return value is the pointer to last filled element
4233 * The work left to caller to do the actual freeing of subtrees:
4234 * a) free the subtree starting from *@top
4235 * b) free the subtrees whose roots are stored in
4236 * (@chain[i].p+1 .. end of @chain[i].bh->b_data)
4237 * c) free the subtrees growing from the inode past the @chain[0].
4238 * (no partially truncated stuff there). */
4240 static Indirect
*ext4_find_shared(struct inode
*inode
, int depth
,
4241 ext4_lblk_t offsets
[4], Indirect chain
[4],
4244 Indirect
*partial
, *p
;
4248 /* Make k index the deepest non-null offset + 1 */
4249 for (k
= depth
; k
> 1 && !offsets
[k
-1]; k
--)
4251 partial
= ext4_get_branch(inode
, k
, offsets
, chain
, &err
);
4252 /* Writer: pointers */
4254 partial
= chain
+ k
-1;
4256 * If the branch acquired continuation since we've looked at it -
4257 * fine, it should all survive and (new) top doesn't belong to us.
4259 if (!partial
->key
&& *partial
->p
)
4262 for (p
= partial
; (p
> chain
) && all_zeroes((__le32
*) p
->bh
->b_data
, p
->p
); p
--)
4265 * OK, we've found the last block that must survive. The rest of our
4266 * branch should be detached before unlocking. However, if that rest
4267 * of branch is all ours and does not grow immediately from the inode
4268 * it's easier to cheat and just decrement partial->p.
4270 if (p
== chain
+ k
- 1 && p
> chain
) {
4274 /* Nope, don't do this in ext4. Must leave the tree intact */
4281 while (partial
> p
) {
4282 brelse(partial
->bh
);
4290 * Zero a number of block pointers in either an inode or an indirect block.
4291 * If we restart the transaction we must again get write access to the
4292 * indirect block for further modification.
4294 * We release `count' blocks on disk, but (last - first) may be greater
4295 * than `count' because there can be holes in there.
4297 static int ext4_clear_blocks(handle_t
*handle
, struct inode
*inode
,
4298 struct buffer_head
*bh
,
4299 ext4_fsblk_t block_to_free
,
4300 unsigned long count
, __le32
*first
,
4304 int flags
= EXT4_FREE_BLOCKS_FORGET
| EXT4_FREE_BLOCKS_VALIDATED
;
4306 if (S_ISDIR(inode
->i_mode
) || S_ISLNK(inode
->i_mode
))
4307 flags
|= EXT4_FREE_BLOCKS_METADATA
;
4309 if (!ext4_data_block_valid(EXT4_SB(inode
->i_sb
), block_to_free
,
4311 EXT4_ERROR_INODE(inode
, "attempt to clear invalid "
4312 "blocks %llu len %lu",
4313 (unsigned long long) block_to_free
, count
);
4317 if (try_to_extend_transaction(handle
, inode
)) {
4319 BUFFER_TRACE(bh
, "call ext4_handle_dirty_metadata");
4320 ext4_handle_dirty_metadata(handle
, inode
, bh
);
4322 ext4_mark_inode_dirty(handle
, inode
);
4323 ext4_truncate_restart_trans(handle
, inode
,
4324 blocks_for_truncate(inode
));
4326 BUFFER_TRACE(bh
, "retaking write access");
4327 ext4_journal_get_write_access(handle
, bh
);
4331 for (p
= first
; p
< last
; p
++)
4334 ext4_free_blocks(handle
, inode
, 0, block_to_free
, count
, flags
);
4339 * ext4_free_data - free a list of data blocks
4340 * @handle: handle for this transaction
4341 * @inode: inode we are dealing with
4342 * @this_bh: indirect buffer_head which contains *@first and *@last
4343 * @first: array of block numbers
4344 * @last: points immediately past the end of array
4346 * We are freeing all blocks refered from that array (numbers are stored as
4347 * little-endian 32-bit) and updating @inode->i_blocks appropriately.
4349 * We accumulate contiguous runs of blocks to free. Conveniently, if these
4350 * blocks are contiguous then releasing them at one time will only affect one
4351 * or two bitmap blocks (+ group descriptor(s) and superblock) and we won't
4352 * actually use a lot of journal space.
4354 * @this_bh will be %NULL if @first and @last point into the inode's direct
4357 static void ext4_free_data(handle_t
*handle
, struct inode
*inode
,
4358 struct buffer_head
*this_bh
,
4359 __le32
*first
, __le32
*last
)
4361 ext4_fsblk_t block_to_free
= 0; /* Starting block # of a run */
4362 unsigned long count
= 0; /* Number of blocks in the run */
4363 __le32
*block_to_free_p
= NULL
; /* Pointer into inode/ind
4366 ext4_fsblk_t nr
; /* Current block # */
4367 __le32
*p
; /* Pointer into inode/ind
4368 for current block */
4371 if (this_bh
) { /* For indirect block */
4372 BUFFER_TRACE(this_bh
, "get_write_access");
4373 err
= ext4_journal_get_write_access(handle
, this_bh
);
4374 /* Important: if we can't update the indirect pointers
4375 * to the blocks, we can't free them. */
4380 for (p
= first
; p
< last
; p
++) {
4381 nr
= le32_to_cpu(*p
);
4383 /* accumulate blocks to free if they're contiguous */
4386 block_to_free_p
= p
;
4388 } else if (nr
== block_to_free
+ count
) {
4391 if (ext4_clear_blocks(handle
, inode
, this_bh
,
4392 block_to_free
, count
,
4393 block_to_free_p
, p
))
4396 block_to_free_p
= p
;
4403 ext4_clear_blocks(handle
, inode
, this_bh
, block_to_free
,
4404 count
, block_to_free_p
, p
);
4407 BUFFER_TRACE(this_bh
, "call ext4_handle_dirty_metadata");
4410 * The buffer head should have an attached journal head at this
4411 * point. However, if the data is corrupted and an indirect
4412 * block pointed to itself, it would have been detached when
4413 * the block was cleared. Check for this instead of OOPSing.
4415 if ((EXT4_JOURNAL(inode
) == NULL
) || bh2jh(this_bh
))
4416 ext4_handle_dirty_metadata(handle
, inode
, this_bh
);
4418 EXT4_ERROR_INODE(inode
,
4419 "circular indirect block detected at "
4421 (unsigned long long) this_bh
->b_blocknr
);
4426 * ext4_free_branches - free an array of branches
4427 * @handle: JBD handle for this transaction
4428 * @inode: inode we are dealing with
4429 * @parent_bh: the buffer_head which contains *@first and *@last
4430 * @first: array of block numbers
4431 * @last: pointer immediately past the end of array
4432 * @depth: depth of the branches to free
4434 * We are freeing all blocks refered from these branches (numbers are
4435 * stored as little-endian 32-bit) and updating @inode->i_blocks
4438 static void ext4_free_branches(handle_t
*handle
, struct inode
*inode
,
4439 struct buffer_head
*parent_bh
,
4440 __le32
*first
, __le32
*last
, int depth
)
4445 if (ext4_handle_is_aborted(handle
))
4449 struct buffer_head
*bh
;
4450 int addr_per_block
= EXT4_ADDR_PER_BLOCK(inode
->i_sb
);
4452 while (--p
>= first
) {
4453 nr
= le32_to_cpu(*p
);
4455 continue; /* A hole */
4457 if (!ext4_data_block_valid(EXT4_SB(inode
->i_sb
),
4459 EXT4_ERROR_INODE(inode
,
4460 "invalid indirect mapped "
4461 "block %lu (level %d)",
4462 (unsigned long) nr
, depth
);
4466 /* Go read the buffer for the next level down */
4467 bh
= sb_bread(inode
->i_sb
, nr
);
4470 * A read failure? Report error and clear slot
4474 EXT4_ERROR_INODE(inode
,
4475 "Read failure block=%llu",
4476 (unsigned long long) nr
);
4480 /* This zaps the entire block. Bottom up. */
4481 BUFFER_TRACE(bh
, "free child branches");
4482 ext4_free_branches(handle
, inode
, bh
,
4483 (__le32
*) bh
->b_data
,
4484 (__le32
*) bh
->b_data
+ addr_per_block
,
4488 * We've probably journalled the indirect block several
4489 * times during the truncate. But it's no longer
4490 * needed and we now drop it from the transaction via
4491 * jbd2_journal_revoke().
4493 * That's easy if it's exclusively part of this
4494 * transaction. But if it's part of the committing
4495 * transaction then jbd2_journal_forget() will simply
4496 * brelse() it. That means that if the underlying
4497 * block is reallocated in ext4_get_block(),
4498 * unmap_underlying_metadata() will find this block
4499 * and will try to get rid of it. damn, damn.
4501 * If this block has already been committed to the
4502 * journal, a revoke record will be written. And
4503 * revoke records must be emitted *before* clearing
4504 * this block's bit in the bitmaps.
4506 ext4_forget(handle
, 1, inode
, bh
, bh
->b_blocknr
);
4509 * Everything below this this pointer has been
4510 * released. Now let this top-of-subtree go.
4512 * We want the freeing of this indirect block to be
4513 * atomic in the journal with the updating of the
4514 * bitmap block which owns it. So make some room in
4517 * We zero the parent pointer *after* freeing its
4518 * pointee in the bitmaps, so if extend_transaction()
4519 * for some reason fails to put the bitmap changes and
4520 * the release into the same transaction, recovery
4521 * will merely complain about releasing a free block,
4522 * rather than leaking blocks.
4524 if (ext4_handle_is_aborted(handle
))
4526 if (try_to_extend_transaction(handle
, inode
)) {
4527 ext4_mark_inode_dirty(handle
, inode
);
4528 ext4_truncate_restart_trans(handle
, inode
,
4529 blocks_for_truncate(inode
));
4532 ext4_free_blocks(handle
, inode
, 0, nr
, 1,
4533 EXT4_FREE_BLOCKS_METADATA
);
4537 * The block which we have just freed is
4538 * pointed to by an indirect block: journal it
4540 BUFFER_TRACE(parent_bh
, "get_write_access");
4541 if (!ext4_journal_get_write_access(handle
,
4544 BUFFER_TRACE(parent_bh
,
4545 "call ext4_handle_dirty_metadata");
4546 ext4_handle_dirty_metadata(handle
,
4553 /* We have reached the bottom of the tree. */
4554 BUFFER_TRACE(parent_bh
, "free data blocks");
4555 ext4_free_data(handle
, inode
, parent_bh
, first
, last
);
4559 int ext4_can_truncate(struct inode
*inode
)
4561 if (IS_APPEND(inode
) || IS_IMMUTABLE(inode
))
4563 if (S_ISREG(inode
->i_mode
))
4565 if (S_ISDIR(inode
->i_mode
))
4567 if (S_ISLNK(inode
->i_mode
))
4568 return !ext4_inode_is_fast_symlink(inode
);
4575 * We block out ext4_get_block() block instantiations across the entire
4576 * transaction, and VFS/VM ensures that ext4_truncate() cannot run
4577 * simultaneously on behalf of the same inode.
4579 * As we work through the truncate and commmit bits of it to the journal there
4580 * is one core, guiding principle: the file's tree must always be consistent on
4581 * disk. We must be able to restart the truncate after a crash.
4583 * The file's tree may be transiently inconsistent in memory (although it
4584 * probably isn't), but whenever we close off and commit a journal transaction,
4585 * the contents of (the filesystem + the journal) must be consistent and
4586 * restartable. It's pretty simple, really: bottom up, right to left (although
4587 * left-to-right works OK too).
4589 * Note that at recovery time, journal replay occurs *before* the restart of
4590 * truncate against the orphan inode list.
4592 * The committed inode has the new, desired i_size (which is the same as
4593 * i_disksize in this case). After a crash, ext4_orphan_cleanup() will see
4594 * that this inode's truncate did not complete and it will again call
4595 * ext4_truncate() to have another go. So there will be instantiated blocks
4596 * to the right of the truncation point in a crashed ext4 filesystem. But
4597 * that's fine - as long as they are linked from the inode, the post-crash
4598 * ext4_truncate() run will find them and release them.
4600 void ext4_truncate(struct inode
*inode
)
4603 struct ext4_inode_info
*ei
= EXT4_I(inode
);
4604 __le32
*i_data
= ei
->i_data
;
4605 int addr_per_block
= EXT4_ADDR_PER_BLOCK(inode
->i_sb
);
4606 struct address_space
*mapping
= inode
->i_mapping
;
4607 ext4_lblk_t offsets
[4];
4612 ext4_lblk_t last_block
;
4613 unsigned blocksize
= inode
->i_sb
->s_blocksize
;
4615 if (!ext4_can_truncate(inode
))
4618 ext4_clear_inode_flag(inode
, EXT4_INODE_EOFBLOCKS
);
4620 if (inode
->i_size
== 0 && !test_opt(inode
->i_sb
, NO_AUTO_DA_ALLOC
))
4621 ext4_set_inode_state(inode
, EXT4_STATE_DA_ALLOC_CLOSE
);
4623 if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
)) {
4624 ext4_ext_truncate(inode
);
4628 handle
= start_transaction(inode
);
4630 return; /* AKPM: return what? */
4632 last_block
= (inode
->i_size
+ blocksize
-1)
4633 >> EXT4_BLOCK_SIZE_BITS(inode
->i_sb
);
4635 if (inode
->i_size
& (blocksize
- 1))
4636 if (ext4_block_truncate_page(handle
, mapping
, inode
->i_size
))
4639 n
= ext4_block_to_path(inode
, last_block
, offsets
, NULL
);
4641 goto out_stop
; /* error */
4644 * OK. This truncate is going to happen. We add the inode to the
4645 * orphan list, so that if this truncate spans multiple transactions,
4646 * and we crash, we will resume the truncate when the filesystem
4647 * recovers. It also marks the inode dirty, to catch the new size.
4649 * Implication: the file must always be in a sane, consistent
4650 * truncatable state while each transaction commits.
4652 if (ext4_orphan_add(handle
, inode
))
4656 * From here we block out all ext4_get_block() callers who want to
4657 * modify the block allocation tree.
4659 down_write(&ei
->i_data_sem
);
4661 ext4_discard_preallocations(inode
);
4664 * The orphan list entry will now protect us from any crash which
4665 * occurs before the truncate completes, so it is now safe to propagate
4666 * the new, shorter inode size (held for now in i_size) into the
4667 * on-disk inode. We do this via i_disksize, which is the value which
4668 * ext4 *really* writes onto the disk inode.
4670 ei
->i_disksize
= inode
->i_size
;
4672 if (n
== 1) { /* direct blocks */
4673 ext4_free_data(handle
, inode
, NULL
, i_data
+offsets
[0],
4674 i_data
+ EXT4_NDIR_BLOCKS
);
4678 partial
= ext4_find_shared(inode
, n
, offsets
, chain
, &nr
);
4679 /* Kill the top of shared branch (not detached) */
4681 if (partial
== chain
) {
4682 /* Shared branch grows from the inode */
4683 ext4_free_branches(handle
, inode
, NULL
,
4684 &nr
, &nr
+1, (chain
+n
-1) - partial
);
4687 * We mark the inode dirty prior to restart,
4688 * and prior to stop. No need for it here.
4691 /* Shared branch grows from an indirect block */
4692 BUFFER_TRACE(partial
->bh
, "get_write_access");
4693 ext4_free_branches(handle
, inode
, partial
->bh
,
4695 partial
->p
+1, (chain
+n
-1) - partial
);
4698 /* Clear the ends of indirect blocks on the shared branch */
4699 while (partial
> chain
) {
4700 ext4_free_branches(handle
, inode
, partial
->bh
, partial
->p
+ 1,
4701 (__le32
*)partial
->bh
->b_data
+addr_per_block
,
4702 (chain
+n
-1) - partial
);
4703 BUFFER_TRACE(partial
->bh
, "call brelse");
4704 brelse(partial
->bh
);
4708 /* Kill the remaining (whole) subtrees */
4709 switch (offsets
[0]) {
4711 nr
= i_data
[EXT4_IND_BLOCK
];
4713 ext4_free_branches(handle
, inode
, NULL
, &nr
, &nr
+1, 1);
4714 i_data
[EXT4_IND_BLOCK
] = 0;
4716 case EXT4_IND_BLOCK
:
4717 nr
= i_data
[EXT4_DIND_BLOCK
];
4719 ext4_free_branches(handle
, inode
, NULL
, &nr
, &nr
+1, 2);
4720 i_data
[EXT4_DIND_BLOCK
] = 0;
4722 case EXT4_DIND_BLOCK
:
4723 nr
= i_data
[EXT4_TIND_BLOCK
];
4725 ext4_free_branches(handle
, inode
, NULL
, &nr
, &nr
+1, 3);
4726 i_data
[EXT4_TIND_BLOCK
] = 0;
4728 case EXT4_TIND_BLOCK
:
4732 up_write(&ei
->i_data_sem
);
4733 inode
->i_mtime
= inode
->i_ctime
= ext4_current_time(inode
);
4734 ext4_mark_inode_dirty(handle
, inode
);
4737 * In a multi-transaction truncate, we only make the final transaction
4741 ext4_handle_sync(handle
);
4744 * If this was a simple ftruncate(), and the file will remain alive
4745 * then we need to clear up the orphan record which we created above.
4746 * However, if this was a real unlink then we were called by
4747 * ext4_delete_inode(), and we allow that function to clean up the
4748 * orphan info for us.
4751 ext4_orphan_del(handle
, inode
);
4753 ext4_journal_stop(handle
);
4757 * ext4_get_inode_loc returns with an extra refcount against the inode's
4758 * underlying buffer_head on success. If 'in_mem' is true, we have all
4759 * data in memory that is needed to recreate the on-disk version of this
4762 static int __ext4_get_inode_loc(struct inode
*inode
,
4763 struct ext4_iloc
*iloc
, int in_mem
)
4765 struct ext4_group_desc
*gdp
;
4766 struct buffer_head
*bh
;
4767 struct super_block
*sb
= inode
->i_sb
;
4769 int inodes_per_block
, inode_offset
;
4772 if (!ext4_valid_inum(sb
, inode
->i_ino
))
4775 iloc
->block_group
= (inode
->i_ino
- 1) / EXT4_INODES_PER_GROUP(sb
);
4776 gdp
= ext4_get_group_desc(sb
, iloc
->block_group
, NULL
);
4781 * Figure out the offset within the block group inode table
4783 inodes_per_block
= (EXT4_BLOCK_SIZE(sb
) / EXT4_INODE_SIZE(sb
));
4784 inode_offset
= ((inode
->i_ino
- 1) %
4785 EXT4_INODES_PER_GROUP(sb
));
4786 block
= ext4_inode_table(sb
, gdp
) + (inode_offset
/ inodes_per_block
);
4787 iloc
->offset
= (inode_offset
% inodes_per_block
) * EXT4_INODE_SIZE(sb
);
4789 bh
= sb_getblk(sb
, block
);
4791 EXT4_ERROR_INODE(inode
, "unable to read inode block - "
4792 "block %llu", block
);
4795 if (!buffer_uptodate(bh
)) {
4799 * If the buffer has the write error flag, we have failed
4800 * to write out another inode in the same block. In this
4801 * case, we don't have to read the block because we may
4802 * read the old inode data successfully.
4804 if (buffer_write_io_error(bh
) && !buffer_uptodate(bh
))
4805 set_buffer_uptodate(bh
);
4807 if (buffer_uptodate(bh
)) {
4808 /* someone brought it uptodate while we waited */
4814 * If we have all information of the inode in memory and this
4815 * is the only valid inode in the block, we need not read the
4819 struct buffer_head
*bitmap_bh
;
4822 start
= inode_offset
& ~(inodes_per_block
- 1);
4824 /* Is the inode bitmap in cache? */
4825 bitmap_bh
= sb_getblk(sb
, ext4_inode_bitmap(sb
, gdp
));
4830 * If the inode bitmap isn't in cache then the
4831 * optimisation may end up performing two reads instead
4832 * of one, so skip it.
4834 if (!buffer_uptodate(bitmap_bh
)) {
4838 for (i
= start
; i
< start
+ inodes_per_block
; i
++) {
4839 if (i
== inode_offset
)
4841 if (ext4_test_bit(i
, bitmap_bh
->b_data
))
4845 if (i
== start
+ inodes_per_block
) {
4846 /* all other inodes are free, so skip I/O */
4847 memset(bh
->b_data
, 0, bh
->b_size
);
4848 set_buffer_uptodate(bh
);
4856 * If we need to do any I/O, try to pre-readahead extra
4857 * blocks from the inode table.
4859 if (EXT4_SB(sb
)->s_inode_readahead_blks
) {
4860 ext4_fsblk_t b
, end
, table
;
4863 table
= ext4_inode_table(sb
, gdp
);
4864 /* s_inode_readahead_blks is always a power of 2 */
4865 b
= block
& ~(EXT4_SB(sb
)->s_inode_readahead_blks
-1);
4868 end
= b
+ EXT4_SB(sb
)->s_inode_readahead_blks
;
4869 num
= EXT4_INODES_PER_GROUP(sb
);
4870 if (EXT4_HAS_RO_COMPAT_FEATURE(sb
,
4871 EXT4_FEATURE_RO_COMPAT_GDT_CSUM
))
4872 num
-= ext4_itable_unused_count(sb
, gdp
);
4873 table
+= num
/ inodes_per_block
;
4877 sb_breadahead(sb
, b
++);
4881 * There are other valid inodes in the buffer, this inode
4882 * has in-inode xattrs, or we don't have this inode in memory.
4883 * Read the block from disk.
4886 bh
->b_end_io
= end_buffer_read_sync
;
4887 submit_bh(READ_META
, bh
);
4889 if (!buffer_uptodate(bh
)) {
4890 EXT4_ERROR_INODE(inode
, "unable to read inode "
4891 "block %llu", block
);
4901 int ext4_get_inode_loc(struct inode
*inode
, struct ext4_iloc
*iloc
)
4903 /* We have all inode data except xattrs in memory here. */
4904 return __ext4_get_inode_loc(inode
, iloc
,
4905 !ext4_test_inode_state(inode
, EXT4_STATE_XATTR
));
4908 void ext4_set_inode_flags(struct inode
*inode
)
4910 unsigned int flags
= EXT4_I(inode
)->i_flags
;
4912 inode
->i_flags
&= ~(S_SYNC
|S_APPEND
|S_IMMUTABLE
|S_NOATIME
|S_DIRSYNC
);
4913 if (flags
& EXT4_SYNC_FL
)
4914 inode
->i_flags
|= S_SYNC
;
4915 if (flags
& EXT4_APPEND_FL
)
4916 inode
->i_flags
|= S_APPEND
;
4917 if (flags
& EXT4_IMMUTABLE_FL
)
4918 inode
->i_flags
|= S_IMMUTABLE
;
4919 if (flags
& EXT4_NOATIME_FL
)
4920 inode
->i_flags
|= S_NOATIME
;
4921 if (flags
& EXT4_DIRSYNC_FL
)
4922 inode
->i_flags
|= S_DIRSYNC
;
4925 /* Propagate flags from i_flags to EXT4_I(inode)->i_flags */
4926 void ext4_get_inode_flags(struct ext4_inode_info
*ei
)
4928 unsigned int vfs_fl
;
4929 unsigned long old_fl
, new_fl
;
4932 vfs_fl
= ei
->vfs_inode
.i_flags
;
4933 old_fl
= ei
->i_flags
;
4934 new_fl
= old_fl
& ~(EXT4_SYNC_FL
|EXT4_APPEND_FL
|
4935 EXT4_IMMUTABLE_FL
|EXT4_NOATIME_FL
|
4937 if (vfs_fl
& S_SYNC
)
4938 new_fl
|= EXT4_SYNC_FL
;
4939 if (vfs_fl
& S_APPEND
)
4940 new_fl
|= EXT4_APPEND_FL
;
4941 if (vfs_fl
& S_IMMUTABLE
)
4942 new_fl
|= EXT4_IMMUTABLE_FL
;
4943 if (vfs_fl
& S_NOATIME
)
4944 new_fl
|= EXT4_NOATIME_FL
;
4945 if (vfs_fl
& S_DIRSYNC
)
4946 new_fl
|= EXT4_DIRSYNC_FL
;
4947 } while (cmpxchg(&ei
->i_flags
, old_fl
, new_fl
) != old_fl
);
4950 static blkcnt_t
ext4_inode_blocks(struct ext4_inode
*raw_inode
,
4951 struct ext4_inode_info
*ei
)
4954 struct inode
*inode
= &(ei
->vfs_inode
);
4955 struct super_block
*sb
= inode
->i_sb
;
4957 if (EXT4_HAS_RO_COMPAT_FEATURE(sb
,
4958 EXT4_FEATURE_RO_COMPAT_HUGE_FILE
)) {
4959 /* we are using combined 48 bit field */
4960 i_blocks
= ((u64
)le16_to_cpu(raw_inode
->i_blocks_high
)) << 32 |
4961 le32_to_cpu(raw_inode
->i_blocks_lo
);
4962 if (ext4_test_inode_flag(inode
, EXT4_INODE_HUGE_FILE
)) {
4963 /* i_blocks represent file system block size */
4964 return i_blocks
<< (inode
->i_blkbits
- 9);
4969 return le32_to_cpu(raw_inode
->i_blocks_lo
);
4973 struct inode
*ext4_iget(struct super_block
*sb
, unsigned long ino
)
4975 struct ext4_iloc iloc
;
4976 struct ext4_inode
*raw_inode
;
4977 struct ext4_inode_info
*ei
;
4978 struct inode
*inode
;
4979 journal_t
*journal
= EXT4_SB(sb
)->s_journal
;
4983 inode
= iget_locked(sb
, ino
);
4985 return ERR_PTR(-ENOMEM
);
4986 if (!(inode
->i_state
& I_NEW
))
4992 ret
= __ext4_get_inode_loc(inode
, &iloc
, 0);
4995 raw_inode
= ext4_raw_inode(&iloc
);
4996 inode
->i_mode
= le16_to_cpu(raw_inode
->i_mode
);
4997 inode
->i_uid
= (uid_t
)le16_to_cpu(raw_inode
->i_uid_low
);
4998 inode
->i_gid
= (gid_t
)le16_to_cpu(raw_inode
->i_gid_low
);
4999 if (!(test_opt(inode
->i_sb
, NO_UID32
))) {
5000 inode
->i_uid
|= le16_to_cpu(raw_inode
->i_uid_high
) << 16;
5001 inode
->i_gid
|= le16_to_cpu(raw_inode
->i_gid_high
) << 16;
5003 inode
->i_nlink
= le16_to_cpu(raw_inode
->i_links_count
);
5005 ei
->i_state_flags
= 0;
5006 ei
->i_dir_start_lookup
= 0;
5007 ei
->i_dtime
= le32_to_cpu(raw_inode
->i_dtime
);
5008 /* We now have enough fields to check if the inode was active or not.
5009 * This is needed because nfsd might try to access dead inodes
5010 * the test is that same one that e2fsck uses
5011 * NeilBrown 1999oct15
5013 if (inode
->i_nlink
== 0) {
5014 if (inode
->i_mode
== 0 ||
5015 !(EXT4_SB(inode
->i_sb
)->s_mount_state
& EXT4_ORPHAN_FS
)) {
5016 /* this inode is deleted */
5020 /* The only unlinked inodes we let through here have
5021 * valid i_mode and are being read by the orphan
5022 * recovery code: that's fine, we're about to complete
5023 * the process of deleting those. */
5025 ei
->i_flags
= le32_to_cpu(raw_inode
->i_flags
);
5026 inode
->i_blocks
= ext4_inode_blocks(raw_inode
, ei
);
5027 ei
->i_file_acl
= le32_to_cpu(raw_inode
->i_file_acl_lo
);
5028 if (EXT4_HAS_INCOMPAT_FEATURE(sb
, EXT4_FEATURE_INCOMPAT_64BIT
))
5030 ((__u64
)le16_to_cpu(raw_inode
->i_file_acl_high
)) << 32;
5031 inode
->i_size
= ext4_isize(raw_inode
);
5032 ei
->i_disksize
= inode
->i_size
;
5034 ei
->i_reserved_quota
= 0;
5036 inode
->i_generation
= le32_to_cpu(raw_inode
->i_generation
);
5037 ei
->i_block_group
= iloc
.block_group
;
5038 ei
->i_last_alloc_group
= ~0;
5040 * NOTE! The in-memory inode i_data array is in little-endian order
5041 * even on big-endian machines: we do NOT byteswap the block numbers!
5043 for (block
= 0; block
< EXT4_N_BLOCKS
; block
++)
5044 ei
->i_data
[block
] = raw_inode
->i_block
[block
];
5045 INIT_LIST_HEAD(&ei
->i_orphan
);
5048 * Set transaction id's of transactions that have to be committed
5049 * to finish f[data]sync. We set them to currently running transaction
5050 * as we cannot be sure that the inode or some of its metadata isn't
5051 * part of the transaction - the inode could have been reclaimed and
5052 * now it is reread from disk.
5055 transaction_t
*transaction
;
5058 spin_lock(&journal
->j_state_lock
);
5059 if (journal
->j_running_transaction
)
5060 transaction
= journal
->j_running_transaction
;
5062 transaction
= journal
->j_committing_transaction
;
5064 tid
= transaction
->t_tid
;
5066 tid
= journal
->j_commit_sequence
;
5067 spin_unlock(&journal
->j_state_lock
);
5068 ei
->i_sync_tid
= tid
;
5069 ei
->i_datasync_tid
= tid
;
5072 if (EXT4_INODE_SIZE(inode
->i_sb
) > EXT4_GOOD_OLD_INODE_SIZE
) {
5073 ei
->i_extra_isize
= le16_to_cpu(raw_inode
->i_extra_isize
);
5074 if (EXT4_GOOD_OLD_INODE_SIZE
+ ei
->i_extra_isize
>
5075 EXT4_INODE_SIZE(inode
->i_sb
)) {
5079 if (ei
->i_extra_isize
== 0) {
5080 /* The extra space is currently unused. Use it. */
5081 ei
->i_extra_isize
= sizeof(struct ext4_inode
) -
5082 EXT4_GOOD_OLD_INODE_SIZE
;
5084 __le32
*magic
= (void *)raw_inode
+
5085 EXT4_GOOD_OLD_INODE_SIZE
+
5087 if (*magic
== cpu_to_le32(EXT4_XATTR_MAGIC
))
5088 ext4_set_inode_state(inode
, EXT4_STATE_XATTR
);
5091 ei
->i_extra_isize
= 0;
5093 EXT4_INODE_GET_XTIME(i_ctime
, inode
, raw_inode
);
5094 EXT4_INODE_GET_XTIME(i_mtime
, inode
, raw_inode
);
5095 EXT4_INODE_GET_XTIME(i_atime
, inode
, raw_inode
);
5096 EXT4_EINODE_GET_XTIME(i_crtime
, ei
, raw_inode
);
5098 inode
->i_version
= le32_to_cpu(raw_inode
->i_disk_version
);
5099 if (EXT4_INODE_SIZE(inode
->i_sb
) > EXT4_GOOD_OLD_INODE_SIZE
) {
5100 if (EXT4_FITS_IN_INODE(raw_inode
, ei
, i_version_hi
))
5102 (__u64
)(le32_to_cpu(raw_inode
->i_version_hi
)) << 32;
5106 if (ei
->i_file_acl
&&
5107 !ext4_data_block_valid(EXT4_SB(sb
), ei
->i_file_acl
, 1)) {
5108 EXT4_ERROR_INODE(inode
, "bad extended attribute block %llu",
5112 } else if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
)) {
5113 if (S_ISREG(inode
->i_mode
) || S_ISDIR(inode
->i_mode
) ||
5114 (S_ISLNK(inode
->i_mode
) &&
5115 !ext4_inode_is_fast_symlink(inode
)))
5116 /* Validate extent which is part of inode */
5117 ret
= ext4_ext_check_inode(inode
);
5118 } else if (S_ISREG(inode
->i_mode
) || S_ISDIR(inode
->i_mode
) ||
5119 (S_ISLNK(inode
->i_mode
) &&
5120 !ext4_inode_is_fast_symlink(inode
))) {
5121 /* Validate block references which are part of inode */
5122 ret
= ext4_check_inode_blockref(inode
);
5127 if (S_ISREG(inode
->i_mode
)) {
5128 inode
->i_op
= &ext4_file_inode_operations
;
5129 inode
->i_fop
= &ext4_file_operations
;
5130 ext4_set_aops(inode
);
5131 } else if (S_ISDIR(inode
->i_mode
)) {
5132 inode
->i_op
= &ext4_dir_inode_operations
;
5133 inode
->i_fop
= &ext4_dir_operations
;
5134 } else if (S_ISLNK(inode
->i_mode
)) {
5135 if (ext4_inode_is_fast_symlink(inode
)) {
5136 inode
->i_op
= &ext4_fast_symlink_inode_operations
;
5137 nd_terminate_link(ei
->i_data
, inode
->i_size
,
5138 sizeof(ei
->i_data
) - 1);
5140 inode
->i_op
= &ext4_symlink_inode_operations
;
5141 ext4_set_aops(inode
);
5143 } else if (S_ISCHR(inode
->i_mode
) || S_ISBLK(inode
->i_mode
) ||
5144 S_ISFIFO(inode
->i_mode
) || S_ISSOCK(inode
->i_mode
)) {
5145 inode
->i_op
= &ext4_special_inode_operations
;
5146 if (raw_inode
->i_block
[0])
5147 init_special_inode(inode
, inode
->i_mode
,
5148 old_decode_dev(le32_to_cpu(raw_inode
->i_block
[0])));
5150 init_special_inode(inode
, inode
->i_mode
,
5151 new_decode_dev(le32_to_cpu(raw_inode
->i_block
[1])));
5154 EXT4_ERROR_INODE(inode
, "bogus i_mode (%o)", inode
->i_mode
);
5158 ext4_set_inode_flags(inode
);
5159 unlock_new_inode(inode
);
5165 return ERR_PTR(ret
);
5168 static int ext4_inode_blocks_set(handle_t
*handle
,
5169 struct ext4_inode
*raw_inode
,
5170 struct ext4_inode_info
*ei
)
5172 struct inode
*inode
= &(ei
->vfs_inode
);
5173 u64 i_blocks
= inode
->i_blocks
;
5174 struct super_block
*sb
= inode
->i_sb
;
5176 if (i_blocks
<= ~0U) {
5178 * i_blocks can be represnted in a 32 bit variable
5179 * as multiple of 512 bytes
5181 raw_inode
->i_blocks_lo
= cpu_to_le32(i_blocks
);
5182 raw_inode
->i_blocks_high
= 0;
5183 ext4_clear_inode_flag(inode
, EXT4_INODE_HUGE_FILE
);
5186 if (!EXT4_HAS_RO_COMPAT_FEATURE(sb
, EXT4_FEATURE_RO_COMPAT_HUGE_FILE
))
5189 if (i_blocks
<= 0xffffffffffffULL
) {
5191 * i_blocks can be represented in a 48 bit variable
5192 * as multiple of 512 bytes
5194 raw_inode
->i_blocks_lo
= cpu_to_le32(i_blocks
);
5195 raw_inode
->i_blocks_high
= cpu_to_le16(i_blocks
>> 32);
5196 ext4_clear_inode_flag(inode
, EXT4_INODE_HUGE_FILE
);
5198 ext4_set_inode_flag(inode
, EXT4_INODE_HUGE_FILE
);
5199 /* i_block is stored in file system block size */
5200 i_blocks
= i_blocks
>> (inode
->i_blkbits
- 9);
5201 raw_inode
->i_blocks_lo
= cpu_to_le32(i_blocks
);
5202 raw_inode
->i_blocks_high
= cpu_to_le16(i_blocks
>> 32);
5208 * Post the struct inode info into an on-disk inode location in the
5209 * buffer-cache. This gobbles the caller's reference to the
5210 * buffer_head in the inode location struct.
5212 * The caller must have write access to iloc->bh.
5214 static int ext4_do_update_inode(handle_t
*handle
,
5215 struct inode
*inode
,
5216 struct ext4_iloc
*iloc
)
5218 struct ext4_inode
*raw_inode
= ext4_raw_inode(iloc
);
5219 struct ext4_inode_info
*ei
= EXT4_I(inode
);
5220 struct buffer_head
*bh
= iloc
->bh
;
5221 int err
= 0, rc
, block
;
5223 /* For fields not not tracking in the in-memory inode,
5224 * initialise them to zero for new inodes. */
5225 if (ext4_test_inode_state(inode
, EXT4_STATE_NEW
))
5226 memset(raw_inode
, 0, EXT4_SB(inode
->i_sb
)->s_inode_size
);
5228 ext4_get_inode_flags(ei
);
5229 raw_inode
->i_mode
= cpu_to_le16(inode
->i_mode
);
5230 if (!(test_opt(inode
->i_sb
, NO_UID32
))) {
5231 raw_inode
->i_uid_low
= cpu_to_le16(low_16_bits(inode
->i_uid
));
5232 raw_inode
->i_gid_low
= cpu_to_le16(low_16_bits(inode
->i_gid
));
5234 * Fix up interoperability with old kernels. Otherwise, old inodes get
5235 * re-used with the upper 16 bits of the uid/gid intact
5238 raw_inode
->i_uid_high
=
5239 cpu_to_le16(high_16_bits(inode
->i_uid
));
5240 raw_inode
->i_gid_high
=
5241 cpu_to_le16(high_16_bits(inode
->i_gid
));
5243 raw_inode
->i_uid_high
= 0;
5244 raw_inode
->i_gid_high
= 0;
5247 raw_inode
->i_uid_low
=
5248 cpu_to_le16(fs_high2lowuid(inode
->i_uid
));
5249 raw_inode
->i_gid_low
=
5250 cpu_to_le16(fs_high2lowgid(inode
->i_gid
));
5251 raw_inode
->i_uid_high
= 0;
5252 raw_inode
->i_gid_high
= 0;
5254 raw_inode
->i_links_count
= cpu_to_le16(inode
->i_nlink
);
5256 EXT4_INODE_SET_XTIME(i_ctime
, inode
, raw_inode
);
5257 EXT4_INODE_SET_XTIME(i_mtime
, inode
, raw_inode
);
5258 EXT4_INODE_SET_XTIME(i_atime
, inode
, raw_inode
);
5259 EXT4_EINODE_SET_XTIME(i_crtime
, ei
, raw_inode
);
5261 if (ext4_inode_blocks_set(handle
, raw_inode
, ei
))
5263 raw_inode
->i_dtime
= cpu_to_le32(ei
->i_dtime
);
5264 raw_inode
->i_flags
= cpu_to_le32(ei
->i_flags
);
5265 if (EXT4_SB(inode
->i_sb
)->s_es
->s_creator_os
!=
5266 cpu_to_le32(EXT4_OS_HURD
))
5267 raw_inode
->i_file_acl_high
=
5268 cpu_to_le16(ei
->i_file_acl
>> 32);
5269 raw_inode
->i_file_acl_lo
= cpu_to_le32(ei
->i_file_acl
);
5270 ext4_isize_set(raw_inode
, ei
->i_disksize
);
5271 if (ei
->i_disksize
> 0x7fffffffULL
) {
5272 struct super_block
*sb
= inode
->i_sb
;
5273 if (!EXT4_HAS_RO_COMPAT_FEATURE(sb
,
5274 EXT4_FEATURE_RO_COMPAT_LARGE_FILE
) ||
5275 EXT4_SB(sb
)->s_es
->s_rev_level
==
5276 cpu_to_le32(EXT4_GOOD_OLD_REV
)) {
5277 /* If this is the first large file
5278 * created, add a flag to the superblock.
5280 err
= ext4_journal_get_write_access(handle
,
5281 EXT4_SB(sb
)->s_sbh
);
5284 ext4_update_dynamic_rev(sb
);
5285 EXT4_SET_RO_COMPAT_FEATURE(sb
,
5286 EXT4_FEATURE_RO_COMPAT_LARGE_FILE
);
5288 ext4_handle_sync(handle
);
5289 err
= ext4_handle_dirty_metadata(handle
, NULL
,
5290 EXT4_SB(sb
)->s_sbh
);
5293 raw_inode
->i_generation
= cpu_to_le32(inode
->i_generation
);
5294 if (S_ISCHR(inode
->i_mode
) || S_ISBLK(inode
->i_mode
)) {
5295 if (old_valid_dev(inode
->i_rdev
)) {
5296 raw_inode
->i_block
[0] =
5297 cpu_to_le32(old_encode_dev(inode
->i_rdev
));
5298 raw_inode
->i_block
[1] = 0;
5300 raw_inode
->i_block
[0] = 0;
5301 raw_inode
->i_block
[1] =
5302 cpu_to_le32(new_encode_dev(inode
->i_rdev
));
5303 raw_inode
->i_block
[2] = 0;
5306 for (block
= 0; block
< EXT4_N_BLOCKS
; block
++)
5307 raw_inode
->i_block
[block
] = ei
->i_data
[block
];
5309 raw_inode
->i_disk_version
= cpu_to_le32(inode
->i_version
);
5310 if (ei
->i_extra_isize
) {
5311 if (EXT4_FITS_IN_INODE(raw_inode
, ei
, i_version_hi
))
5312 raw_inode
->i_version_hi
=
5313 cpu_to_le32(inode
->i_version
>> 32);
5314 raw_inode
->i_extra_isize
= cpu_to_le16(ei
->i_extra_isize
);
5317 BUFFER_TRACE(bh
, "call ext4_handle_dirty_metadata");
5318 rc
= ext4_handle_dirty_metadata(handle
, NULL
, bh
);
5321 ext4_clear_inode_state(inode
, EXT4_STATE_NEW
);
5323 ext4_update_inode_fsync_trans(handle
, inode
, 0);
5326 ext4_std_error(inode
->i_sb
, err
);
5331 * ext4_write_inode()
5333 * We are called from a few places:
5335 * - Within generic_file_write() for O_SYNC files.
5336 * Here, there will be no transaction running. We wait for any running
5337 * trasnaction to commit.
5339 * - Within sys_sync(), kupdate and such.
5340 * We wait on commit, if tol to.
5342 * - Within prune_icache() (PF_MEMALLOC == true)
5343 * Here we simply return. We can't afford to block kswapd on the
5346 * In all cases it is actually safe for us to return without doing anything,
5347 * because the inode has been copied into a raw inode buffer in
5348 * ext4_mark_inode_dirty(). This is a correctness thing for O_SYNC and for
5351 * Note that we are absolutely dependent upon all inode dirtiers doing the
5352 * right thing: they *must* call mark_inode_dirty() after dirtying info in
5353 * which we are interested.
5355 * It would be a bug for them to not do this. The code:
5357 * mark_inode_dirty(inode)
5359 * inode->i_size = expr;
5361 * is in error because a kswapd-driven write_inode() could occur while
5362 * `stuff()' is running, and the new i_size will be lost. Plus the inode
5363 * will no longer be on the superblock's dirty inode list.
5365 int ext4_write_inode(struct inode
*inode
, struct writeback_control
*wbc
)
5369 if (current
->flags
& PF_MEMALLOC
)
5372 if (EXT4_SB(inode
->i_sb
)->s_journal
) {
5373 if (ext4_journal_current_handle()) {
5374 jbd_debug(1, "called recursively, non-PF_MEMALLOC!\n");
5379 if (wbc
->sync_mode
!= WB_SYNC_ALL
)
5382 err
= ext4_force_commit(inode
->i_sb
);
5384 struct ext4_iloc iloc
;
5386 err
= __ext4_get_inode_loc(inode
, &iloc
, 0);
5389 if (wbc
->sync_mode
== WB_SYNC_ALL
)
5390 sync_dirty_buffer(iloc
.bh
);
5391 if (buffer_req(iloc
.bh
) && !buffer_uptodate(iloc
.bh
)) {
5392 EXT4_ERROR_INODE(inode
,
5393 "IO error syncing inode (block=%llu)",
5394 (unsigned long long) iloc
.bh
->b_blocknr
);
5405 * Called from notify_change.
5407 * We want to trap VFS attempts to truncate the file as soon as
5408 * possible. In particular, we want to make sure that when the VFS
5409 * shrinks i_size, we put the inode on the orphan list and modify
5410 * i_disksize immediately, so that during the subsequent flushing of
5411 * dirty pages and freeing of disk blocks, we can guarantee that any
5412 * commit will leave the blocks being flushed in an unused state on
5413 * disk. (On recovery, the inode will get truncated and the blocks will
5414 * be freed, so we have a strong guarantee that no future commit will
5415 * leave these blocks visible to the user.)
5417 * Another thing we have to assure is that if we are in ordered mode
5418 * and inode is still attached to the committing transaction, we must
5419 * we start writeout of all the dirty pages which are being truncated.
5420 * This way we are sure that all the data written in the previous
5421 * transaction are already on disk (truncate waits for pages under
5424 * Called with inode->i_mutex down.
5426 int ext4_setattr(struct dentry
*dentry
, struct iattr
*attr
)
5428 struct inode
*inode
= dentry
->d_inode
;
5430 const unsigned int ia_valid
= attr
->ia_valid
;
5432 error
= inode_change_ok(inode
, attr
);
5436 if (is_quota_modification(inode
, attr
))
5437 dquot_initialize(inode
);
5438 if ((ia_valid
& ATTR_UID
&& attr
->ia_uid
!= inode
->i_uid
) ||
5439 (ia_valid
& ATTR_GID
&& attr
->ia_gid
!= inode
->i_gid
)) {
5442 /* (user+group)*(old+new) structure, inode write (sb,
5443 * inode block, ? - but truncate inode update has it) */
5444 handle
= ext4_journal_start(inode
, (EXT4_MAXQUOTAS_INIT_BLOCKS(inode
->i_sb
)+
5445 EXT4_MAXQUOTAS_DEL_BLOCKS(inode
->i_sb
))+3);
5446 if (IS_ERR(handle
)) {
5447 error
= PTR_ERR(handle
);
5450 error
= dquot_transfer(inode
, attr
);
5452 ext4_journal_stop(handle
);
5455 /* Update corresponding info in inode so that everything is in
5456 * one transaction */
5457 if (attr
->ia_valid
& ATTR_UID
)
5458 inode
->i_uid
= attr
->ia_uid
;
5459 if (attr
->ia_valid
& ATTR_GID
)
5460 inode
->i_gid
= attr
->ia_gid
;
5461 error
= ext4_mark_inode_dirty(handle
, inode
);
5462 ext4_journal_stop(handle
);
5465 if (attr
->ia_valid
& ATTR_SIZE
) {
5466 if (!(ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
))) {
5467 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
5469 if (attr
->ia_size
> sbi
->s_bitmap_maxbytes
) {
5476 if (S_ISREG(inode
->i_mode
) &&
5477 attr
->ia_valid
& ATTR_SIZE
&&
5478 (attr
->ia_size
< inode
->i_size
||
5479 (ext4_test_inode_flag(inode
, EXT4_INODE_EOFBLOCKS
)))) {
5482 handle
= ext4_journal_start(inode
, 3);
5483 if (IS_ERR(handle
)) {
5484 error
= PTR_ERR(handle
);
5488 error
= ext4_orphan_add(handle
, inode
);
5489 EXT4_I(inode
)->i_disksize
= attr
->ia_size
;
5490 rc
= ext4_mark_inode_dirty(handle
, inode
);
5493 ext4_journal_stop(handle
);
5495 if (ext4_should_order_data(inode
)) {
5496 error
= ext4_begin_ordered_truncate(inode
,
5499 /* Do as much error cleanup as possible */
5500 handle
= ext4_journal_start(inode
, 3);
5501 if (IS_ERR(handle
)) {
5502 ext4_orphan_del(NULL
, inode
);
5505 ext4_orphan_del(handle
, inode
);
5506 ext4_journal_stop(handle
);
5510 /* ext4_truncate will clear the flag */
5511 if ((ext4_test_inode_flag(inode
, EXT4_INODE_EOFBLOCKS
)))
5512 ext4_truncate(inode
);
5515 rc
= inode_setattr(inode
, attr
);
5517 /* If inode_setattr's call to ext4_truncate failed to get a
5518 * transaction handle at all, we need to clean up the in-core
5519 * orphan list manually. */
5521 ext4_orphan_del(NULL
, inode
);
5523 if (!rc
&& (ia_valid
& ATTR_MODE
))
5524 rc
= ext4_acl_chmod(inode
);
5527 ext4_std_error(inode
->i_sb
, error
);
5533 int ext4_getattr(struct vfsmount
*mnt
, struct dentry
*dentry
,
5536 struct inode
*inode
;
5537 unsigned long delalloc_blocks
;
5539 inode
= dentry
->d_inode
;
5540 generic_fillattr(inode
, stat
);
5543 * We can't update i_blocks if the block allocation is delayed
5544 * otherwise in the case of system crash before the real block
5545 * allocation is done, we will have i_blocks inconsistent with
5546 * on-disk file blocks.
5547 * We always keep i_blocks updated together with real
5548 * allocation. But to not confuse with user, stat
5549 * will return the blocks that include the delayed allocation
5550 * blocks for this file.
5552 spin_lock(&EXT4_I(inode
)->i_block_reservation_lock
);
5553 delalloc_blocks
= EXT4_I(inode
)->i_reserved_data_blocks
;
5554 spin_unlock(&EXT4_I(inode
)->i_block_reservation_lock
);
5556 stat
->blocks
+= (delalloc_blocks
<< inode
->i_sb
->s_blocksize_bits
)>>9;
5560 static int ext4_indirect_trans_blocks(struct inode
*inode
, int nrblocks
,
5565 /* if nrblocks are contiguous */
5568 * With N contiguous data blocks, it need at most
5569 * N/EXT4_ADDR_PER_BLOCK(inode->i_sb) indirect blocks
5570 * 2 dindirect blocks
5573 indirects
= nrblocks
/ EXT4_ADDR_PER_BLOCK(inode
->i_sb
);
5574 return indirects
+ 3;
5577 * if nrblocks are not contiguous, worse case, each block touch
5578 * a indirect block, and each indirect block touch a double indirect
5579 * block, plus a triple indirect block
5581 indirects
= nrblocks
* 2 + 1;
5585 static int ext4_index_trans_blocks(struct inode
*inode
, int nrblocks
, int chunk
)
5587 if (!(ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
)))
5588 return ext4_indirect_trans_blocks(inode
, nrblocks
, chunk
);
5589 return ext4_ext_index_trans_blocks(inode
, nrblocks
, chunk
);
5593 * Account for index blocks, block groups bitmaps and block group
5594 * descriptor blocks if modify datablocks and index blocks
5595 * worse case, the indexs blocks spread over different block groups
5597 * If datablocks are discontiguous, they are possible to spread over
5598 * different block groups too. If they are contiuguous, with flexbg,
5599 * they could still across block group boundary.
5601 * Also account for superblock, inode, quota and xattr blocks
5603 int ext4_meta_trans_blocks(struct inode
*inode
, int nrblocks
, int chunk
)
5605 ext4_group_t groups
, ngroups
= ext4_get_groups_count(inode
->i_sb
);
5611 * How many index blocks need to touch to modify nrblocks?
5612 * The "Chunk" flag indicating whether the nrblocks is
5613 * physically contiguous on disk
5615 * For Direct IO and fallocate, they calls get_block to allocate
5616 * one single extent at a time, so they could set the "Chunk" flag
5618 idxblocks
= ext4_index_trans_blocks(inode
, nrblocks
, chunk
);
5623 * Now let's see how many group bitmaps and group descriptors need
5633 if (groups
> ngroups
)
5635 if (groups
> EXT4_SB(inode
->i_sb
)->s_gdb_count
)
5636 gdpblocks
= EXT4_SB(inode
->i_sb
)->s_gdb_count
;
5638 /* bitmaps and block group descriptor blocks */
5639 ret
+= groups
+ gdpblocks
;
5641 /* Blocks for super block, inode, quota and xattr blocks */
5642 ret
+= EXT4_META_TRANS_BLOCKS(inode
->i_sb
);
5648 * Calulate the total number of credits to reserve to fit
5649 * the modification of a single pages into a single transaction,
5650 * which may include multiple chunks of block allocations.
5652 * This could be called via ext4_write_begin()
5654 * We need to consider the worse case, when
5655 * one new block per extent.
5657 int ext4_writepage_trans_blocks(struct inode
*inode
)
5659 int bpp
= ext4_journal_blocks_per_page(inode
);
5662 ret
= ext4_meta_trans_blocks(inode
, bpp
, 0);
5664 /* Account for data blocks for journalled mode */
5665 if (ext4_should_journal_data(inode
))
5671 * Calculate the journal credits for a chunk of data modification.
5673 * This is called from DIO, fallocate or whoever calling
5674 * ext4_get_blocks() to map/allocate a chunk of contiguous disk blocks.
5676 * journal buffers for data blocks are not included here, as DIO
5677 * and fallocate do no need to journal data buffers.
5679 int ext4_chunk_trans_blocks(struct inode
*inode
, int nrblocks
)
5681 return ext4_meta_trans_blocks(inode
, nrblocks
, 1);
5685 * The caller must have previously called ext4_reserve_inode_write().
5686 * Give this, we know that the caller already has write access to iloc->bh.
5688 int ext4_mark_iloc_dirty(handle_t
*handle
,
5689 struct inode
*inode
, struct ext4_iloc
*iloc
)
5693 if (test_opt(inode
->i_sb
, I_VERSION
))
5694 inode_inc_iversion(inode
);
5696 /* the do_update_inode consumes one bh->b_count */
5699 /* ext4_do_update_inode() does jbd2_journal_dirty_metadata */
5700 err
= ext4_do_update_inode(handle
, inode
, iloc
);
5706 * On success, We end up with an outstanding reference count against
5707 * iloc->bh. This _must_ be cleaned up later.
5711 ext4_reserve_inode_write(handle_t
*handle
, struct inode
*inode
,
5712 struct ext4_iloc
*iloc
)
5716 err
= ext4_get_inode_loc(inode
, iloc
);
5718 BUFFER_TRACE(iloc
->bh
, "get_write_access");
5719 err
= ext4_journal_get_write_access(handle
, iloc
->bh
);
5725 ext4_std_error(inode
->i_sb
, err
);
5730 * Expand an inode by new_extra_isize bytes.
5731 * Returns 0 on success or negative error number on failure.
5733 static int ext4_expand_extra_isize(struct inode
*inode
,
5734 unsigned int new_extra_isize
,
5735 struct ext4_iloc iloc
,
5738 struct ext4_inode
*raw_inode
;
5739 struct ext4_xattr_ibody_header
*header
;
5741 if (EXT4_I(inode
)->i_extra_isize
>= new_extra_isize
)
5744 raw_inode
= ext4_raw_inode(&iloc
);
5746 header
= IHDR(inode
, raw_inode
);
5748 /* No extended attributes present */
5749 if (!ext4_test_inode_state(inode
, EXT4_STATE_XATTR
) ||
5750 header
->h_magic
!= cpu_to_le32(EXT4_XATTR_MAGIC
)) {
5751 memset((void *)raw_inode
+ EXT4_GOOD_OLD_INODE_SIZE
, 0,
5753 EXT4_I(inode
)->i_extra_isize
= new_extra_isize
;
5757 /* try to expand with EAs present */
5758 return ext4_expand_extra_isize_ea(inode
, new_extra_isize
,
5763 * What we do here is to mark the in-core inode as clean with respect to inode
5764 * dirtiness (it may still be data-dirty).
5765 * This means that the in-core inode may be reaped by prune_icache
5766 * without having to perform any I/O. This is a very good thing,
5767 * because *any* task may call prune_icache - even ones which
5768 * have a transaction open against a different journal.
5770 * Is this cheating? Not really. Sure, we haven't written the
5771 * inode out, but prune_icache isn't a user-visible syncing function.
5772 * Whenever the user wants stuff synced (sys_sync, sys_msync, sys_fsync)
5773 * we start and wait on commits.
5775 * Is this efficient/effective? Well, we're being nice to the system
5776 * by cleaning up our inodes proactively so they can be reaped
5777 * without I/O. But we are potentially leaving up to five seconds'
5778 * worth of inodes floating about which prune_icache wants us to
5779 * write out. One way to fix that would be to get prune_icache()
5780 * to do a write_super() to free up some memory. It has the desired
5783 int ext4_mark_inode_dirty(handle_t
*handle
, struct inode
*inode
)
5785 struct ext4_iloc iloc
;
5786 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
5787 static unsigned int mnt_count
;
5791 err
= ext4_reserve_inode_write(handle
, inode
, &iloc
);
5792 if (ext4_handle_valid(handle
) &&
5793 EXT4_I(inode
)->i_extra_isize
< sbi
->s_want_extra_isize
&&
5794 !ext4_test_inode_state(inode
, EXT4_STATE_NO_EXPAND
)) {
5796 * We need extra buffer credits since we may write into EA block
5797 * with this same handle. If journal_extend fails, then it will
5798 * only result in a minor loss of functionality for that inode.
5799 * If this is felt to be critical, then e2fsck should be run to
5800 * force a large enough s_min_extra_isize.
5802 if ((jbd2_journal_extend(handle
,
5803 EXT4_DATA_TRANS_BLOCKS(inode
->i_sb
))) == 0) {
5804 ret
= ext4_expand_extra_isize(inode
,
5805 sbi
->s_want_extra_isize
,
5808 ext4_set_inode_state(inode
,
5809 EXT4_STATE_NO_EXPAND
);
5811 le16_to_cpu(sbi
->s_es
->s_mnt_count
)) {
5812 ext4_warning(inode
->i_sb
,
5813 "Unable to expand inode %lu. Delete"
5814 " some EAs or run e2fsck.",
5817 le16_to_cpu(sbi
->s_es
->s_mnt_count
);
5823 err
= ext4_mark_iloc_dirty(handle
, inode
, &iloc
);
5828 * ext4_dirty_inode() is called from __mark_inode_dirty()
5830 * We're really interested in the case where a file is being extended.
5831 * i_size has been changed by generic_commit_write() and we thus need
5832 * to include the updated inode in the current transaction.
5834 * Also, dquot_alloc_block() will always dirty the inode when blocks
5835 * are allocated to the file.
5837 * If the inode is marked synchronous, we don't honour that here - doing
5838 * so would cause a commit on atime updates, which we don't bother doing.
5839 * We handle synchronous inodes at the highest possible level.
5841 void ext4_dirty_inode(struct inode
*inode
)
5845 handle
= ext4_journal_start(inode
, 2);
5849 ext4_mark_inode_dirty(handle
, inode
);
5851 ext4_journal_stop(handle
);
5858 * Bind an inode's backing buffer_head into this transaction, to prevent
5859 * it from being flushed to disk early. Unlike
5860 * ext4_reserve_inode_write, this leaves behind no bh reference and
5861 * returns no iloc structure, so the caller needs to repeat the iloc
5862 * lookup to mark the inode dirty later.
5864 static int ext4_pin_inode(handle_t
*handle
, struct inode
*inode
)
5866 struct ext4_iloc iloc
;
5870 err
= ext4_get_inode_loc(inode
, &iloc
);
5872 BUFFER_TRACE(iloc
.bh
, "get_write_access");
5873 err
= jbd2_journal_get_write_access(handle
, iloc
.bh
);
5875 err
= ext4_handle_dirty_metadata(handle
,
5881 ext4_std_error(inode
->i_sb
, err
);
5886 int ext4_change_inode_journal_flag(struct inode
*inode
, int val
)
5893 * We have to be very careful here: changing a data block's
5894 * journaling status dynamically is dangerous. If we write a
5895 * data block to the journal, change the status and then delete
5896 * that block, we risk forgetting to revoke the old log record
5897 * from the journal and so a subsequent replay can corrupt data.
5898 * So, first we make sure that the journal is empty and that
5899 * nobody is changing anything.
5902 journal
= EXT4_JOURNAL(inode
);
5905 if (is_journal_aborted(journal
))
5908 jbd2_journal_lock_updates(journal
);
5909 jbd2_journal_flush(journal
);
5912 * OK, there are no updates running now, and all cached data is
5913 * synced to disk. We are now in a completely consistent state
5914 * which doesn't have anything in the journal, and we know that
5915 * no filesystem updates are running, so it is safe to modify
5916 * the inode's in-core data-journaling state flag now.
5920 ext4_set_inode_flag(inode
, EXT4_INODE_JOURNAL_DATA
);
5922 ext4_clear_inode_flag(inode
, EXT4_INODE_JOURNAL_DATA
);
5923 ext4_set_aops(inode
);
5925 jbd2_journal_unlock_updates(journal
);
5927 /* Finally we can mark the inode as dirty. */
5929 handle
= ext4_journal_start(inode
, 1);
5931 return PTR_ERR(handle
);
5933 err
= ext4_mark_inode_dirty(handle
, inode
);
5934 ext4_handle_sync(handle
);
5935 ext4_journal_stop(handle
);
5936 ext4_std_error(inode
->i_sb
, err
);
5941 static int ext4_bh_unmapped(handle_t
*handle
, struct buffer_head
*bh
)
5943 return !buffer_mapped(bh
);
5946 int ext4_page_mkwrite(struct vm_area_struct
*vma
, struct vm_fault
*vmf
)
5948 struct page
*page
= vmf
->page
;
5953 struct file
*file
= vma
->vm_file
;
5954 struct inode
*inode
= file
->f_path
.dentry
->d_inode
;
5955 struct address_space
*mapping
= inode
->i_mapping
;
5958 * Get i_alloc_sem to stop truncates messing with the inode. We cannot
5959 * get i_mutex because we are already holding mmap_sem.
5961 down_read(&inode
->i_alloc_sem
);
5962 size
= i_size_read(inode
);
5963 if (page
->mapping
!= mapping
|| size
<= page_offset(page
)
5964 || !PageUptodate(page
)) {
5965 /* page got truncated from under us? */
5969 if (PageMappedToDisk(page
))
5972 if (page
->index
== size
>> PAGE_CACHE_SHIFT
)
5973 len
= size
& ~PAGE_CACHE_MASK
;
5975 len
= PAGE_CACHE_SIZE
;
5979 * return if we have all the buffers mapped. This avoid
5980 * the need to call write_begin/write_end which does a
5981 * journal_start/journal_stop which can block and take
5984 if (page_has_buffers(page
)) {
5985 if (!walk_page_buffers(NULL
, page_buffers(page
), 0, len
, NULL
,
5986 ext4_bh_unmapped
)) {
5993 * OK, we need to fill the hole... Do write_begin write_end
5994 * to do block allocation/reservation.We are not holding
5995 * inode.i__mutex here. That allow * parallel write_begin,
5996 * write_end call. lock_page prevent this from happening
5997 * on the same page though
5999 ret
= mapping
->a_ops
->write_begin(file
, mapping
, page_offset(page
),
6000 len
, AOP_FLAG_UNINTERRUPTIBLE
, &page
, &fsdata
);
6003 ret
= mapping
->a_ops
->write_end(file
, mapping
, page_offset(page
),
6004 len
, len
, page
, fsdata
);
6010 ret
= VM_FAULT_SIGBUS
;
6011 up_read(&inode
->i_alloc_sem
);