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_evict_inode(struct inode
*inode
)
175 if (inode
->i_nlink
) {
176 truncate_inode_pages(&inode
->i_data
, 0);
180 if (!is_bad_inode(inode
))
181 dquot_initialize(inode
);
183 if (ext4_should_order_data(inode
))
184 ext4_begin_ordered_truncate(inode
, 0);
185 truncate_inode_pages(&inode
->i_data
, 0);
187 if (is_bad_inode(inode
))
190 handle
= ext4_journal_start(inode
, blocks_for_truncate(inode
)+3);
191 if (IS_ERR(handle
)) {
192 ext4_std_error(inode
->i_sb
, PTR_ERR(handle
));
194 * If we're going to skip the normal cleanup, we still need to
195 * make sure that the in-core orphan linked list is properly
198 ext4_orphan_del(NULL
, inode
);
203 ext4_handle_sync(handle
);
205 err
= ext4_mark_inode_dirty(handle
, inode
);
207 ext4_warning(inode
->i_sb
,
208 "couldn't mark inode dirty (err %d)", err
);
212 ext4_truncate(inode
);
215 * ext4_ext_truncate() doesn't reserve any slop when it
216 * restarts journal transactions; therefore there may not be
217 * enough credits left in the handle to remove the inode from
218 * the orphan list and set the dtime field.
220 if (!ext4_handle_has_enough_credits(handle
, 3)) {
221 err
= ext4_journal_extend(handle
, 3);
223 err
= ext4_journal_restart(handle
, 3);
225 ext4_warning(inode
->i_sb
,
226 "couldn't extend journal (err %d)", err
);
228 ext4_journal_stop(handle
);
229 ext4_orphan_del(NULL
, inode
);
235 * Kill off the orphan record which ext4_truncate created.
236 * AKPM: I think this can be inside the above `if'.
237 * Note that ext4_orphan_del() has to be able to cope with the
238 * deletion of a non-existent orphan - this is because we don't
239 * know if ext4_truncate() actually created an orphan record.
240 * (Well, we could do this if we need to, but heck - it works)
242 ext4_orphan_del(handle
, inode
);
243 EXT4_I(inode
)->i_dtime
= get_seconds();
246 * One subtle ordering requirement: if anything has gone wrong
247 * (transaction abort, IO errors, whatever), then we can still
248 * do these next steps (the fs will already have been marked as
249 * having errors), but we can't free the inode if the mark_dirty
252 if (ext4_mark_inode_dirty(handle
, inode
))
253 /* If that failed, just do the required in-core inode clear. */
254 ext4_clear_inode(inode
);
256 ext4_free_inode(handle
, inode
);
257 ext4_journal_stop(handle
);
260 ext4_clear_inode(inode
); /* We must guarantee clearing of inode... */
266 struct buffer_head
*bh
;
269 static inline void add_chain(Indirect
*p
, struct buffer_head
*bh
, __le32
*v
)
271 p
->key
= *(p
->p
= v
);
276 * ext4_block_to_path - parse the block number into array of offsets
277 * @inode: inode in question (we are only interested in its superblock)
278 * @i_block: block number to be parsed
279 * @offsets: array to store the offsets in
280 * @boundary: set this non-zero if the referred-to block is likely to be
281 * followed (on disk) by an indirect block.
283 * To store the locations of file's data ext4 uses a data structure common
284 * for UNIX filesystems - tree of pointers anchored in the inode, with
285 * data blocks at leaves and indirect blocks in intermediate nodes.
286 * This function translates the block number into path in that tree -
287 * return value is the path length and @offsets[n] is the offset of
288 * pointer to (n+1)th node in the nth one. If @block is out of range
289 * (negative or too large) warning is printed and zero returned.
291 * Note: function doesn't find node addresses, so no IO is needed. All
292 * we need to know is the capacity of indirect blocks (taken from the
297 * Portability note: the last comparison (check that we fit into triple
298 * indirect block) is spelled differently, because otherwise on an
299 * architecture with 32-bit longs and 8Kb pages we might get into trouble
300 * if our filesystem had 8Kb blocks. We might use long long, but that would
301 * kill us on x86. Oh, well, at least the sign propagation does not matter -
302 * i_block would have to be negative in the very beginning, so we would not
306 static int ext4_block_to_path(struct inode
*inode
,
308 ext4_lblk_t offsets
[4], int *boundary
)
310 int ptrs
= EXT4_ADDR_PER_BLOCK(inode
->i_sb
);
311 int ptrs_bits
= EXT4_ADDR_PER_BLOCK_BITS(inode
->i_sb
);
312 const long direct_blocks
= EXT4_NDIR_BLOCKS
,
313 indirect_blocks
= ptrs
,
314 double_blocks
= (1 << (ptrs_bits
* 2));
318 if (i_block
< direct_blocks
) {
319 offsets
[n
++] = i_block
;
320 final
= direct_blocks
;
321 } else if ((i_block
-= direct_blocks
) < indirect_blocks
) {
322 offsets
[n
++] = EXT4_IND_BLOCK
;
323 offsets
[n
++] = i_block
;
325 } else if ((i_block
-= indirect_blocks
) < double_blocks
) {
326 offsets
[n
++] = EXT4_DIND_BLOCK
;
327 offsets
[n
++] = i_block
>> ptrs_bits
;
328 offsets
[n
++] = i_block
& (ptrs
- 1);
330 } else if (((i_block
-= double_blocks
) >> (ptrs_bits
* 2)) < ptrs
) {
331 offsets
[n
++] = EXT4_TIND_BLOCK
;
332 offsets
[n
++] = i_block
>> (ptrs_bits
* 2);
333 offsets
[n
++] = (i_block
>> ptrs_bits
) & (ptrs
- 1);
334 offsets
[n
++] = i_block
& (ptrs
- 1);
337 ext4_warning(inode
->i_sb
, "block %lu > max in inode %lu",
338 i_block
+ direct_blocks
+
339 indirect_blocks
+ double_blocks
, inode
->i_ino
);
342 *boundary
= final
- 1 - (i_block
& (ptrs
- 1));
346 static int __ext4_check_blockref(const char *function
, unsigned int line
,
348 __le32
*p
, unsigned int max
)
350 struct ext4_super_block
*es
= EXT4_SB(inode
->i_sb
)->s_es
;
354 while (bref
< p
+max
) {
355 blk
= le32_to_cpu(*bref
++);
357 unlikely(!ext4_data_block_valid(EXT4_SB(inode
->i_sb
),
359 es
->s_last_error_block
= cpu_to_le64(blk
);
360 ext4_error_inode(inode
, function
, line
, blk
,
369 #define ext4_check_indirect_blockref(inode, bh) \
370 __ext4_check_blockref(__func__, __LINE__, inode, \
371 (__le32 *)(bh)->b_data, \
372 EXT4_ADDR_PER_BLOCK((inode)->i_sb))
374 #define ext4_check_inode_blockref(inode) \
375 __ext4_check_blockref(__func__, __LINE__, inode, \
376 EXT4_I(inode)->i_data, \
380 * ext4_get_branch - read the chain of indirect blocks leading to data
381 * @inode: inode in question
382 * @depth: depth of the chain (1 - direct pointer, etc.)
383 * @offsets: offsets of pointers in inode/indirect blocks
384 * @chain: place to store the result
385 * @err: here we store the error value
387 * Function fills the array of triples <key, p, bh> and returns %NULL
388 * if everything went OK or the pointer to the last filled triple
389 * (incomplete one) otherwise. Upon the return chain[i].key contains
390 * the number of (i+1)-th block in the chain (as it is stored in memory,
391 * i.e. little-endian 32-bit), chain[i].p contains the address of that
392 * number (it points into struct inode for i==0 and into the bh->b_data
393 * for i>0) and chain[i].bh points to the buffer_head of i-th indirect
394 * block for i>0 and NULL for i==0. In other words, it holds the block
395 * numbers of the chain, addresses they were taken from (and where we can
396 * verify that chain did not change) and buffer_heads hosting these
399 * Function stops when it stumbles upon zero pointer (absent block)
400 * (pointer to last triple returned, *@err == 0)
401 * or when it gets an IO error reading an indirect block
402 * (ditto, *@err == -EIO)
403 * or when it reads all @depth-1 indirect blocks successfully and finds
404 * the whole chain, all way to the data (returns %NULL, *err == 0).
406 * Need to be called with
407 * down_read(&EXT4_I(inode)->i_data_sem)
409 static Indirect
*ext4_get_branch(struct inode
*inode
, int depth
,
410 ext4_lblk_t
*offsets
,
411 Indirect chain
[4], int *err
)
413 struct super_block
*sb
= inode
->i_sb
;
415 struct buffer_head
*bh
;
418 /* i_data is not going away, no lock needed */
419 add_chain(chain
, NULL
, EXT4_I(inode
)->i_data
+ *offsets
);
423 bh
= sb_getblk(sb
, le32_to_cpu(p
->key
));
427 if (!bh_uptodate_or_lock(bh
)) {
428 if (bh_submit_read(bh
) < 0) {
432 /* validate block references */
433 if (ext4_check_indirect_blockref(inode
, bh
)) {
439 add_chain(++p
, bh
, (__le32
*)bh
->b_data
+ *++offsets
);
453 * ext4_find_near - find a place for allocation with sufficient locality
455 * @ind: descriptor of indirect block.
457 * This function returns the preferred place for block allocation.
458 * It is used when heuristic for sequential allocation fails.
460 * + if there is a block to the left of our position - allocate near it.
461 * + if pointer will live in indirect block - allocate near that block.
462 * + if pointer will live in inode - allocate in the same
465 * In the latter case we colour the starting block by the callers PID to
466 * prevent it from clashing with concurrent allocations for a different inode
467 * in the same block group. The PID is used here so that functionally related
468 * files will be close-by on-disk.
470 * Caller must make sure that @ind is valid and will stay that way.
472 static ext4_fsblk_t
ext4_find_near(struct inode
*inode
, Indirect
*ind
)
474 struct ext4_inode_info
*ei
= EXT4_I(inode
);
475 __le32
*start
= ind
->bh
? (__le32
*) ind
->bh
->b_data
: ei
->i_data
;
477 ext4_fsblk_t bg_start
;
478 ext4_fsblk_t last_block
;
479 ext4_grpblk_t colour
;
480 ext4_group_t block_group
;
481 int flex_size
= ext4_flex_bg_size(EXT4_SB(inode
->i_sb
));
483 /* Try to find previous block */
484 for (p
= ind
->p
- 1; p
>= start
; p
--) {
486 return le32_to_cpu(*p
);
489 /* No such thing, so let's try location of indirect block */
491 return ind
->bh
->b_blocknr
;
494 * It is going to be referred to from the inode itself? OK, just put it
495 * into the same cylinder group then.
497 block_group
= ei
->i_block_group
;
498 if (flex_size
>= EXT4_FLEX_SIZE_DIR_ALLOC_SCHEME
) {
499 block_group
&= ~(flex_size
-1);
500 if (S_ISREG(inode
->i_mode
))
503 bg_start
= ext4_group_first_block_no(inode
->i_sb
, block_group
);
504 last_block
= ext4_blocks_count(EXT4_SB(inode
->i_sb
)->s_es
) - 1;
507 * If we are doing delayed allocation, we don't need take
508 * colour into account.
510 if (test_opt(inode
->i_sb
, DELALLOC
))
513 if (bg_start
+ EXT4_BLOCKS_PER_GROUP(inode
->i_sb
) <= last_block
)
514 colour
= (current
->pid
% 16) *
515 (EXT4_BLOCKS_PER_GROUP(inode
->i_sb
) / 16);
517 colour
= (current
->pid
% 16) * ((last_block
- bg_start
) / 16);
518 return bg_start
+ colour
;
522 * ext4_find_goal - find a preferred place for allocation.
524 * @block: block we want
525 * @partial: pointer to the last triple within a chain
527 * Normally this function find the preferred place for block allocation,
529 * Because this is only used for non-extent files, we limit the block nr
532 static ext4_fsblk_t
ext4_find_goal(struct inode
*inode
, ext4_lblk_t block
,
538 goal
= ext4_find_near(inode
, partial
);
539 goal
= goal
& EXT4_MAX_BLOCK_FILE_PHYS
;
544 * ext4_blks_to_allocate: Look up the block map and count the number
545 * of direct blocks need to be allocated for the given branch.
547 * @branch: chain of indirect blocks
548 * @k: number of blocks need for indirect blocks
549 * @blks: number of data blocks to be mapped.
550 * @blocks_to_boundary: the offset in the indirect block
552 * return the total number of blocks to be allocate, including the
553 * direct and indirect blocks.
555 static int ext4_blks_to_allocate(Indirect
*branch
, int k
, unsigned int blks
,
556 int blocks_to_boundary
)
558 unsigned int count
= 0;
561 * Simple case, [t,d]Indirect block(s) has not allocated yet
562 * then it's clear blocks on that path have not allocated
565 /* right now we don't handle cross boundary allocation */
566 if (blks
< blocks_to_boundary
+ 1)
569 count
+= blocks_to_boundary
+ 1;
574 while (count
< blks
&& count
<= blocks_to_boundary
&&
575 le32_to_cpu(*(branch
[0].p
+ count
)) == 0) {
582 * ext4_alloc_blocks: multiple allocate blocks needed for a branch
583 * @indirect_blks: the number of blocks need to allocate for indirect
586 * @new_blocks: on return it will store the new block numbers for
587 * the indirect blocks(if needed) and the first direct block,
588 * @blks: on return it will store the total number of allocated
591 static int ext4_alloc_blocks(handle_t
*handle
, struct inode
*inode
,
592 ext4_lblk_t iblock
, ext4_fsblk_t goal
,
593 int indirect_blks
, int blks
,
594 ext4_fsblk_t new_blocks
[4], int *err
)
596 struct ext4_allocation_request ar
;
598 unsigned long count
= 0, blk_allocated
= 0;
600 ext4_fsblk_t current_block
= 0;
604 * Here we try to allocate the requested multiple blocks at once,
605 * on a best-effort basis.
606 * To build a branch, we should allocate blocks for
607 * the indirect blocks(if not allocated yet), and at least
608 * the first direct block of this branch. That's the
609 * minimum number of blocks need to allocate(required)
611 /* first we try to allocate the indirect blocks */
612 target
= indirect_blks
;
615 /* allocating blocks for indirect blocks and direct blocks */
616 current_block
= ext4_new_meta_blocks(handle
, inode
,
621 if (unlikely(current_block
+ count
> EXT4_MAX_BLOCK_FILE_PHYS
)) {
622 EXT4_ERROR_INODE(inode
,
623 "current_block %llu + count %lu > %d!",
624 current_block
, count
,
625 EXT4_MAX_BLOCK_FILE_PHYS
);
631 /* allocate blocks for indirect blocks */
632 while (index
< indirect_blks
&& count
) {
633 new_blocks
[index
++] = current_block
++;
638 * save the new block number
639 * for the first direct block
641 new_blocks
[index
] = current_block
;
642 printk(KERN_INFO
"%s returned more blocks than "
643 "requested\n", __func__
);
649 target
= blks
- count
;
650 blk_allocated
= count
;
653 /* Now allocate data blocks */
654 memset(&ar
, 0, sizeof(ar
));
659 if (S_ISREG(inode
->i_mode
))
660 /* enable in-core preallocation only for regular files */
661 ar
.flags
= EXT4_MB_HINT_DATA
;
663 current_block
= ext4_mb_new_blocks(handle
, &ar
, err
);
664 if (unlikely(current_block
+ ar
.len
> EXT4_MAX_BLOCK_FILE_PHYS
)) {
665 EXT4_ERROR_INODE(inode
,
666 "current_block %llu + ar.len %d > %d!",
667 current_block
, ar
.len
,
668 EXT4_MAX_BLOCK_FILE_PHYS
);
673 if (*err
&& (target
== blks
)) {
675 * if the allocation failed and we didn't allocate
681 if (target
== blks
) {
683 * save the new block number
684 * for the first direct block
686 new_blocks
[index
] = current_block
;
688 blk_allocated
+= ar
.len
;
691 /* total number of blocks allocated for direct blocks */
696 for (i
= 0; i
< index
; i
++)
697 ext4_free_blocks(handle
, inode
, 0, new_blocks
[i
], 1, 0);
702 * ext4_alloc_branch - allocate and set up a chain of blocks.
704 * @indirect_blks: number of allocated indirect blocks
705 * @blks: number of allocated direct blocks
706 * @offsets: offsets (in the blocks) to store the pointers to next.
707 * @branch: place to store the chain in.
709 * This function allocates blocks, zeroes out all but the last one,
710 * links them into chain and (if we are synchronous) writes them to disk.
711 * In other words, it prepares a branch that can be spliced onto the
712 * inode. It stores the information about that chain in the branch[], in
713 * the same format as ext4_get_branch() would do. We are calling it after
714 * we had read the existing part of chain and partial points to the last
715 * triple of that (one with zero ->key). Upon the exit we have the same
716 * picture as after the successful ext4_get_block(), except that in one
717 * place chain is disconnected - *branch->p is still zero (we did not
718 * set the last link), but branch->key contains the number that should
719 * be placed into *branch->p to fill that gap.
721 * If allocation fails we free all blocks we've allocated (and forget
722 * their buffer_heads) and return the error value the from failed
723 * ext4_alloc_block() (normally -ENOSPC). Otherwise we set the chain
724 * as described above and return 0.
726 static int ext4_alloc_branch(handle_t
*handle
, struct inode
*inode
,
727 ext4_lblk_t iblock
, int indirect_blks
,
728 int *blks
, ext4_fsblk_t goal
,
729 ext4_lblk_t
*offsets
, Indirect
*branch
)
731 int blocksize
= inode
->i_sb
->s_blocksize
;
734 struct buffer_head
*bh
;
736 ext4_fsblk_t new_blocks
[4];
737 ext4_fsblk_t current_block
;
739 num
= ext4_alloc_blocks(handle
, inode
, iblock
, goal
, indirect_blks
,
740 *blks
, new_blocks
, &err
);
744 branch
[0].key
= cpu_to_le32(new_blocks
[0]);
746 * metadata blocks and data blocks are allocated.
748 for (n
= 1; n
<= indirect_blks
; n
++) {
750 * Get buffer_head for parent block, zero it out
751 * and set the pointer to new one, then send
754 bh
= sb_getblk(inode
->i_sb
, new_blocks
[n
-1]);
757 BUFFER_TRACE(bh
, "call get_create_access");
758 err
= ext4_journal_get_create_access(handle
, bh
);
760 /* Don't brelse(bh) here; it's done in
761 * ext4_journal_forget() below */
766 memset(bh
->b_data
, 0, blocksize
);
767 branch
[n
].p
= (__le32
*) bh
->b_data
+ offsets
[n
];
768 branch
[n
].key
= cpu_to_le32(new_blocks
[n
]);
769 *branch
[n
].p
= branch
[n
].key
;
770 if (n
== indirect_blks
) {
771 current_block
= new_blocks
[n
];
773 * End of chain, update the last new metablock of
774 * the chain to point to the new allocated
775 * data blocks numbers
777 for (i
= 1; i
< num
; i
++)
778 *(branch
[n
].p
+ i
) = cpu_to_le32(++current_block
);
780 BUFFER_TRACE(bh
, "marking uptodate");
781 set_buffer_uptodate(bh
);
784 BUFFER_TRACE(bh
, "call ext4_handle_dirty_metadata");
785 err
= ext4_handle_dirty_metadata(handle
, inode
, bh
);
792 /* Allocation failed, free what we already allocated */
793 ext4_free_blocks(handle
, inode
, 0, new_blocks
[0], 1, 0);
794 for (i
= 1; i
<= n
; i
++) {
796 * branch[i].bh is newly allocated, so there is no
797 * need to revoke the block, which is why we don't
798 * need to set EXT4_FREE_BLOCKS_METADATA.
800 ext4_free_blocks(handle
, inode
, 0, new_blocks
[i
], 1,
801 EXT4_FREE_BLOCKS_FORGET
);
803 for (i
= n
+1; i
< indirect_blks
; i
++)
804 ext4_free_blocks(handle
, inode
, 0, new_blocks
[i
], 1, 0);
806 ext4_free_blocks(handle
, inode
, 0, new_blocks
[i
], num
, 0);
812 * ext4_splice_branch - splice the allocated branch onto inode.
814 * @block: (logical) number of block we are adding
815 * @chain: chain of indirect blocks (with a missing link - see
817 * @where: location of missing link
818 * @num: number of indirect blocks we are adding
819 * @blks: number of direct blocks we are adding
821 * This function fills the missing link and does all housekeeping needed in
822 * inode (->i_blocks, etc.). In case of success we end up with the full
823 * chain to new block and return 0.
825 static int ext4_splice_branch(handle_t
*handle
, struct inode
*inode
,
826 ext4_lblk_t block
, Indirect
*where
, int num
,
831 ext4_fsblk_t current_block
;
834 * If we're splicing into a [td]indirect block (as opposed to the
835 * inode) then we need to get write access to the [td]indirect block
839 BUFFER_TRACE(where
->bh
, "get_write_access");
840 err
= ext4_journal_get_write_access(handle
, where
->bh
);
846 *where
->p
= where
->key
;
849 * Update the host buffer_head or inode to point to more just allocated
850 * direct blocks blocks
852 if (num
== 0 && blks
> 1) {
853 current_block
= le32_to_cpu(where
->key
) + 1;
854 for (i
= 1; i
< blks
; i
++)
855 *(where
->p
+ i
) = cpu_to_le32(current_block
++);
858 /* We are done with atomic stuff, now do the rest of housekeeping */
859 /* had we spliced it onto indirect block? */
862 * If we spliced it onto an indirect block, we haven't
863 * altered the inode. Note however that if it is being spliced
864 * onto an indirect block at the very end of the file (the
865 * file is growing) then we *will* alter the inode to reflect
866 * the new i_size. But that is not done here - it is done in
867 * generic_commit_write->__mark_inode_dirty->ext4_dirty_inode.
869 jbd_debug(5, "splicing indirect only\n");
870 BUFFER_TRACE(where
->bh
, "call ext4_handle_dirty_metadata");
871 err
= ext4_handle_dirty_metadata(handle
, inode
, where
->bh
);
876 * OK, we spliced it into the inode itself on a direct block.
878 ext4_mark_inode_dirty(handle
, inode
);
879 jbd_debug(5, "splicing direct\n");
884 for (i
= 1; i
<= num
; i
++) {
886 * branch[i].bh is newly allocated, so there is no
887 * need to revoke the block, which is why we don't
888 * need to set EXT4_FREE_BLOCKS_METADATA.
890 ext4_free_blocks(handle
, inode
, where
[i
].bh
, 0, 1,
891 EXT4_FREE_BLOCKS_FORGET
);
893 ext4_free_blocks(handle
, inode
, 0, le32_to_cpu(where
[num
].key
),
900 * The ext4_ind_map_blocks() function handles non-extents inodes
901 * (i.e., using the traditional indirect/double-indirect i_blocks
902 * scheme) for ext4_map_blocks().
904 * Allocation strategy is simple: if we have to allocate something, we will
905 * have to go the whole way to leaf. So let's do it before attaching anything
906 * to tree, set linkage between the newborn blocks, write them if sync is
907 * required, recheck the path, free and repeat if check fails, otherwise
908 * set the last missing link (that will protect us from any truncate-generated
909 * removals - all blocks on the path are immune now) and possibly force the
910 * write on the parent block.
911 * That has a nice additional property: no special recovery from the failed
912 * allocations is needed - we simply release blocks and do not touch anything
913 * reachable from inode.
915 * `handle' can be NULL if create == 0.
917 * return > 0, # of blocks mapped or allocated.
918 * return = 0, if plain lookup failed.
919 * return < 0, error case.
921 * The ext4_ind_get_blocks() function should be called with
922 * down_write(&EXT4_I(inode)->i_data_sem) if allocating filesystem
923 * blocks (i.e., flags has EXT4_GET_BLOCKS_CREATE set) or
924 * down_read(&EXT4_I(inode)->i_data_sem) if not allocating file system
927 static int ext4_ind_map_blocks(handle_t
*handle
, struct inode
*inode
,
928 struct ext4_map_blocks
*map
,
932 ext4_lblk_t offsets
[4];
937 int blocks_to_boundary
= 0;
940 ext4_fsblk_t first_block
= 0;
942 J_ASSERT(!(ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
)));
943 J_ASSERT(handle
!= NULL
|| (flags
& EXT4_GET_BLOCKS_CREATE
) == 0);
944 depth
= ext4_block_to_path(inode
, map
->m_lblk
, offsets
,
945 &blocks_to_boundary
);
950 partial
= ext4_get_branch(inode
, depth
, offsets
, chain
, &err
);
952 /* Simplest case - block found, no allocation needed */
954 first_block
= le32_to_cpu(chain
[depth
- 1].key
);
957 while (count
< map
->m_len
&& count
<= blocks_to_boundary
) {
960 blk
= le32_to_cpu(*(chain
[depth
-1].p
+ count
));
962 if (blk
== first_block
+ count
)
970 /* Next simple case - plain lookup or failed read of indirect block */
971 if ((flags
& EXT4_GET_BLOCKS_CREATE
) == 0 || err
== -EIO
)
975 * Okay, we need to do block allocation.
977 goal
= ext4_find_goal(inode
, map
->m_lblk
, partial
);
979 /* the number of blocks need to allocate for [d,t]indirect blocks */
980 indirect_blks
= (chain
+ depth
) - partial
- 1;
983 * Next look up the indirect map to count the totoal number of
984 * direct blocks to allocate for this branch.
986 count
= ext4_blks_to_allocate(partial
, indirect_blks
,
987 map
->m_len
, blocks_to_boundary
);
989 * Block out ext4_truncate while we alter the tree
991 err
= ext4_alloc_branch(handle
, inode
, map
->m_lblk
, indirect_blks
,
993 offsets
+ (partial
- chain
), partial
);
996 * The ext4_splice_branch call will free and forget any buffers
997 * on the new chain if there is a failure, but that risks using
998 * up transaction credits, especially for bitmaps where the
999 * credits cannot be returned. Can we handle this somehow? We
1000 * may need to return -EAGAIN upwards in the worst case. --sct
1003 err
= ext4_splice_branch(handle
, inode
, map
->m_lblk
,
1004 partial
, indirect_blks
, count
);
1008 map
->m_flags
|= EXT4_MAP_NEW
;
1010 ext4_update_inode_fsync_trans(handle
, inode
, 1);
1012 map
->m_flags
|= EXT4_MAP_MAPPED
;
1013 map
->m_pblk
= le32_to_cpu(chain
[depth
-1].key
);
1015 if (count
> blocks_to_boundary
)
1016 map
->m_flags
|= EXT4_MAP_BOUNDARY
;
1018 /* Clean up and exit */
1019 partial
= chain
+ depth
- 1; /* the whole chain */
1021 while (partial
> chain
) {
1022 BUFFER_TRACE(partial
->bh
, "call brelse");
1023 brelse(partial
->bh
);
1031 qsize_t
*ext4_get_reserved_space(struct inode
*inode
)
1033 return &EXT4_I(inode
)->i_reserved_quota
;
1038 * Calculate the number of metadata blocks need to reserve
1039 * to allocate a new block at @lblocks for non extent file based file
1041 static int ext4_indirect_calc_metadata_amount(struct inode
*inode
,
1044 struct ext4_inode_info
*ei
= EXT4_I(inode
);
1045 sector_t dind_mask
= ~((sector_t
)EXT4_ADDR_PER_BLOCK(inode
->i_sb
) - 1);
1048 if (lblock
< EXT4_NDIR_BLOCKS
)
1051 lblock
-= EXT4_NDIR_BLOCKS
;
1053 if (ei
->i_da_metadata_calc_len
&&
1054 (lblock
& dind_mask
) == ei
->i_da_metadata_calc_last_lblock
) {
1055 ei
->i_da_metadata_calc_len
++;
1058 ei
->i_da_metadata_calc_last_lblock
= lblock
& dind_mask
;
1059 ei
->i_da_metadata_calc_len
= 1;
1060 blk_bits
= order_base_2(lblock
);
1061 return (blk_bits
/ EXT4_ADDR_PER_BLOCK_BITS(inode
->i_sb
)) + 1;
1065 * Calculate the number of metadata blocks need to reserve
1066 * to allocate a block located at @lblock
1068 static int ext4_calc_metadata_amount(struct inode
*inode
, sector_t lblock
)
1070 if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
))
1071 return ext4_ext_calc_metadata_amount(inode
, lblock
);
1073 return ext4_indirect_calc_metadata_amount(inode
, lblock
);
1077 * Called with i_data_sem down, which is important since we can call
1078 * ext4_discard_preallocations() from here.
1080 void ext4_da_update_reserve_space(struct inode
*inode
,
1081 int used
, int quota_claim
)
1083 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
1084 struct ext4_inode_info
*ei
= EXT4_I(inode
);
1086 spin_lock(&ei
->i_block_reservation_lock
);
1087 trace_ext4_da_update_reserve_space(inode
, used
);
1088 if (unlikely(used
> ei
->i_reserved_data_blocks
)) {
1089 ext4_msg(inode
->i_sb
, KERN_NOTICE
, "%s: ino %lu, used %d "
1090 "with only %d reserved data blocks\n",
1091 __func__
, inode
->i_ino
, used
,
1092 ei
->i_reserved_data_blocks
);
1094 used
= ei
->i_reserved_data_blocks
;
1097 /* Update per-inode reservations */
1098 ei
->i_reserved_data_blocks
-= used
;
1099 ei
->i_reserved_meta_blocks
-= ei
->i_allocated_meta_blocks
;
1100 percpu_counter_sub(&sbi
->s_dirtyblocks_counter
,
1101 used
+ ei
->i_allocated_meta_blocks
);
1102 ei
->i_allocated_meta_blocks
= 0;
1104 if (ei
->i_reserved_data_blocks
== 0) {
1106 * We can release all of the reserved metadata blocks
1107 * only when we have written all of the delayed
1108 * allocation blocks.
1110 percpu_counter_sub(&sbi
->s_dirtyblocks_counter
,
1111 ei
->i_reserved_meta_blocks
);
1112 ei
->i_reserved_meta_blocks
= 0;
1113 ei
->i_da_metadata_calc_len
= 0;
1115 spin_unlock(&EXT4_I(inode
)->i_block_reservation_lock
);
1117 /* Update quota subsystem for data blocks */
1119 dquot_claim_block(inode
, used
);
1122 * We did fallocate with an offset that is already delayed
1123 * allocated. So on delayed allocated writeback we should
1124 * not re-claim the quota for fallocated blocks.
1126 dquot_release_reservation_block(inode
, used
);
1130 * If we have done all the pending block allocations and if
1131 * there aren't any writers on the inode, we can discard the
1132 * inode's preallocations.
1134 if ((ei
->i_reserved_data_blocks
== 0) &&
1135 (atomic_read(&inode
->i_writecount
) == 0))
1136 ext4_discard_preallocations(inode
);
1139 static int __check_block_validity(struct inode
*inode
, const char *func
,
1141 struct ext4_map_blocks
*map
)
1143 if (!ext4_data_block_valid(EXT4_SB(inode
->i_sb
), map
->m_pblk
,
1145 ext4_error_inode(inode
, func
, line
, map
->m_pblk
,
1146 "lblock %lu mapped to illegal pblock "
1147 "(length %d)", (unsigned long) map
->m_lblk
,
1154 #define check_block_validity(inode, map) \
1155 __check_block_validity((inode), __func__, __LINE__, (map))
1158 * Return the number of contiguous dirty pages in a given inode
1159 * starting at page frame idx.
1161 static pgoff_t
ext4_num_dirty_pages(struct inode
*inode
, pgoff_t idx
,
1162 unsigned int max_pages
)
1164 struct address_space
*mapping
= inode
->i_mapping
;
1166 struct pagevec pvec
;
1168 int i
, nr_pages
, done
= 0;
1172 pagevec_init(&pvec
, 0);
1175 nr_pages
= pagevec_lookup_tag(&pvec
, mapping
, &index
,
1176 PAGECACHE_TAG_DIRTY
,
1177 (pgoff_t
)PAGEVEC_SIZE
);
1180 for (i
= 0; i
< nr_pages
; i
++) {
1181 struct page
*page
= pvec
.pages
[i
];
1182 struct buffer_head
*bh
, *head
;
1185 if (unlikely(page
->mapping
!= mapping
) ||
1187 PageWriteback(page
) ||
1188 page
->index
!= idx
) {
1193 if (page_has_buffers(page
)) {
1194 bh
= head
= page_buffers(page
);
1196 if (!buffer_delay(bh
) &&
1197 !buffer_unwritten(bh
))
1199 bh
= bh
->b_this_page
;
1200 } while (!done
&& (bh
!= head
));
1207 if (num
>= max_pages
)
1210 pagevec_release(&pvec
);
1216 * The ext4_map_blocks() function tries to look up the requested blocks,
1217 * and returns if the blocks are already mapped.
1219 * Otherwise it takes the write lock of the i_data_sem and allocate blocks
1220 * and store the allocated blocks in the result buffer head and mark it
1223 * If file type is extents based, it will call ext4_ext_map_blocks(),
1224 * Otherwise, call with ext4_ind_map_blocks() to handle indirect mapping
1227 * On success, it returns the number of blocks being mapped or allocate.
1228 * if create==0 and the blocks are pre-allocated and uninitialized block,
1229 * the result buffer head is unmapped. If the create ==1, it will make sure
1230 * the buffer head is mapped.
1232 * It returns 0 if plain look up failed (blocks have not been allocated), in
1233 * that casem, buffer head is unmapped
1235 * It returns the error in case of allocation failure.
1237 int ext4_map_blocks(handle_t
*handle
, struct inode
*inode
,
1238 struct ext4_map_blocks
*map
, int flags
)
1243 ext_debug("ext4_map_blocks(): inode %lu, flag %d, max_blocks %u,"
1244 "logical block %lu\n", inode
->i_ino
, flags
, map
->m_len
,
1245 (unsigned long) map
->m_lblk
);
1247 * Try to see if we can get the block without requesting a new
1248 * file system block.
1250 down_read((&EXT4_I(inode
)->i_data_sem
));
1251 if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
)) {
1252 retval
= ext4_ext_map_blocks(handle
, inode
, map
, 0);
1254 retval
= ext4_ind_map_blocks(handle
, inode
, map
, 0);
1256 up_read((&EXT4_I(inode
)->i_data_sem
));
1258 if (retval
> 0 && map
->m_flags
& EXT4_MAP_MAPPED
) {
1259 int ret
= check_block_validity(inode
, map
);
1264 /* If it is only a block(s) look up */
1265 if ((flags
& EXT4_GET_BLOCKS_CREATE
) == 0)
1269 * Returns if the blocks have already allocated
1271 * Note that if blocks have been preallocated
1272 * ext4_ext_get_block() returns th create = 0
1273 * with buffer head unmapped.
1275 if (retval
> 0 && map
->m_flags
& EXT4_MAP_MAPPED
)
1279 * When we call get_blocks without the create flag, the
1280 * BH_Unwritten flag could have gotten set if the blocks
1281 * requested were part of a uninitialized extent. We need to
1282 * clear this flag now that we are committed to convert all or
1283 * part of the uninitialized extent to be an initialized
1284 * extent. This is because we need to avoid the combination
1285 * of BH_Unwritten and BH_Mapped flags being simultaneously
1286 * set on the buffer_head.
1288 map
->m_flags
&= ~EXT4_MAP_UNWRITTEN
;
1291 * New blocks allocate and/or writing to uninitialized extent
1292 * will possibly result in updating i_data, so we take
1293 * the write lock of i_data_sem, and call get_blocks()
1294 * with create == 1 flag.
1296 down_write((&EXT4_I(inode
)->i_data_sem
));
1299 * if the caller is from delayed allocation writeout path
1300 * we have already reserved fs blocks for allocation
1301 * let the underlying get_block() function know to
1302 * avoid double accounting
1304 if (flags
& EXT4_GET_BLOCKS_DELALLOC_RESERVE
)
1305 EXT4_I(inode
)->i_delalloc_reserved_flag
= 1;
1307 * We need to check for EXT4 here because migrate
1308 * could have changed the inode type in between
1310 if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
)) {
1311 retval
= ext4_ext_map_blocks(handle
, inode
, map
, flags
);
1313 retval
= ext4_ind_map_blocks(handle
, inode
, map
, flags
);
1315 if (retval
> 0 && map
->m_flags
& EXT4_MAP_NEW
) {
1317 * We allocated new blocks which will result in
1318 * i_data's format changing. Force the migrate
1319 * to fail by clearing migrate flags
1321 ext4_clear_inode_state(inode
, EXT4_STATE_EXT_MIGRATE
);
1325 * Update reserved blocks/metadata blocks after successful
1326 * block allocation which had been deferred till now. We don't
1327 * support fallocate for non extent files. So we can update
1328 * reserve space here.
1331 (flags
& EXT4_GET_BLOCKS_DELALLOC_RESERVE
))
1332 ext4_da_update_reserve_space(inode
, retval
, 1);
1334 if (flags
& EXT4_GET_BLOCKS_DELALLOC_RESERVE
)
1335 EXT4_I(inode
)->i_delalloc_reserved_flag
= 0;
1337 up_write((&EXT4_I(inode
)->i_data_sem
));
1338 if (retval
> 0 && map
->m_flags
& EXT4_MAP_MAPPED
) {
1339 int ret
= check_block_validity(inode
, map
);
1346 /* Maximum number of blocks we map for direct IO at once. */
1347 #define DIO_MAX_BLOCKS 4096
1349 static int _ext4_get_block(struct inode
*inode
, sector_t iblock
,
1350 struct buffer_head
*bh
, int flags
)
1352 handle_t
*handle
= ext4_journal_current_handle();
1353 struct ext4_map_blocks map
;
1354 int ret
= 0, started
= 0;
1357 map
.m_lblk
= iblock
;
1358 map
.m_len
= bh
->b_size
>> inode
->i_blkbits
;
1360 if (flags
&& !handle
) {
1361 /* Direct IO write... */
1362 if (map
.m_len
> DIO_MAX_BLOCKS
)
1363 map
.m_len
= DIO_MAX_BLOCKS
;
1364 dio_credits
= ext4_chunk_trans_blocks(inode
, map
.m_len
);
1365 handle
= ext4_journal_start(inode
, dio_credits
);
1366 if (IS_ERR(handle
)) {
1367 ret
= PTR_ERR(handle
);
1373 ret
= ext4_map_blocks(handle
, inode
, &map
, flags
);
1375 map_bh(bh
, inode
->i_sb
, map
.m_pblk
);
1376 bh
->b_state
= (bh
->b_state
& ~EXT4_MAP_FLAGS
) | map
.m_flags
;
1377 bh
->b_size
= inode
->i_sb
->s_blocksize
* map
.m_len
;
1381 ext4_journal_stop(handle
);
1385 int ext4_get_block(struct inode
*inode
, sector_t iblock
,
1386 struct buffer_head
*bh
, int create
)
1388 return _ext4_get_block(inode
, iblock
, bh
,
1389 create
? EXT4_GET_BLOCKS_CREATE
: 0);
1393 * `handle' can be NULL if create is zero
1395 struct buffer_head
*ext4_getblk(handle_t
*handle
, struct inode
*inode
,
1396 ext4_lblk_t block
, int create
, int *errp
)
1398 struct ext4_map_blocks map
;
1399 struct buffer_head
*bh
;
1402 J_ASSERT(handle
!= NULL
|| create
== 0);
1406 err
= ext4_map_blocks(handle
, inode
, &map
,
1407 create
? EXT4_GET_BLOCKS_CREATE
: 0);
1415 bh
= sb_getblk(inode
->i_sb
, map
.m_pblk
);
1420 if (map
.m_flags
& EXT4_MAP_NEW
) {
1421 J_ASSERT(create
!= 0);
1422 J_ASSERT(handle
!= NULL
);
1425 * Now that we do not always journal data, we should
1426 * keep in mind whether this should always journal the
1427 * new buffer as metadata. For now, regular file
1428 * writes use ext4_get_block instead, so it's not a
1432 BUFFER_TRACE(bh
, "call get_create_access");
1433 fatal
= ext4_journal_get_create_access(handle
, bh
);
1434 if (!fatal
&& !buffer_uptodate(bh
)) {
1435 memset(bh
->b_data
, 0, inode
->i_sb
->s_blocksize
);
1436 set_buffer_uptodate(bh
);
1439 BUFFER_TRACE(bh
, "call ext4_handle_dirty_metadata");
1440 err
= ext4_handle_dirty_metadata(handle
, inode
, bh
);
1444 BUFFER_TRACE(bh
, "not a new buffer");
1454 struct buffer_head
*ext4_bread(handle_t
*handle
, struct inode
*inode
,
1455 ext4_lblk_t block
, int create
, int *err
)
1457 struct buffer_head
*bh
;
1459 bh
= ext4_getblk(handle
, inode
, block
, create
, err
);
1462 if (buffer_uptodate(bh
))
1464 ll_rw_block(READ_META
, 1, &bh
);
1466 if (buffer_uptodate(bh
))
1473 static int walk_page_buffers(handle_t
*handle
,
1474 struct buffer_head
*head
,
1478 int (*fn
)(handle_t
*handle
,
1479 struct buffer_head
*bh
))
1481 struct buffer_head
*bh
;
1482 unsigned block_start
, block_end
;
1483 unsigned blocksize
= head
->b_size
;
1485 struct buffer_head
*next
;
1487 for (bh
= head
, block_start
= 0;
1488 ret
== 0 && (bh
!= head
|| !block_start
);
1489 block_start
= block_end
, bh
= next
) {
1490 next
= bh
->b_this_page
;
1491 block_end
= block_start
+ blocksize
;
1492 if (block_end
<= from
|| block_start
>= to
) {
1493 if (partial
&& !buffer_uptodate(bh
))
1497 err
= (*fn
)(handle
, bh
);
1505 * To preserve ordering, it is essential that the hole instantiation and
1506 * the data write be encapsulated in a single transaction. We cannot
1507 * close off a transaction and start a new one between the ext4_get_block()
1508 * and the commit_write(). So doing the jbd2_journal_start at the start of
1509 * prepare_write() is the right place.
1511 * Also, this function can nest inside ext4_writepage() ->
1512 * block_write_full_page(). In that case, we *know* that ext4_writepage()
1513 * has generated enough buffer credits to do the whole page. So we won't
1514 * block on the journal in that case, which is good, because the caller may
1517 * By accident, ext4 can be reentered when a transaction is open via
1518 * quota file writes. If we were to commit the transaction while thus
1519 * reentered, there can be a deadlock - we would be holding a quota
1520 * lock, and the commit would never complete if another thread had a
1521 * transaction open and was blocking on the quota lock - a ranking
1524 * So what we do is to rely on the fact that jbd2_journal_stop/journal_start
1525 * will _not_ run commit under these circumstances because handle->h_ref
1526 * is elevated. We'll still have enough credits for the tiny quotafile
1529 static int do_journal_get_write_access(handle_t
*handle
,
1530 struct buffer_head
*bh
)
1532 int dirty
= buffer_dirty(bh
);
1535 if (!buffer_mapped(bh
) || buffer_freed(bh
))
1538 * __block_prepare_write() could have dirtied some buffers. Clean
1539 * the dirty bit as jbd2_journal_get_write_access() could complain
1540 * otherwise about fs integrity issues. Setting of the dirty bit
1541 * by __block_prepare_write() isn't a real problem here as we clear
1542 * the bit before releasing a page lock and thus writeback cannot
1543 * ever write the buffer.
1546 clear_buffer_dirty(bh
);
1547 ret
= ext4_journal_get_write_access(handle
, bh
);
1549 ret
= ext4_handle_dirty_metadata(handle
, NULL
, bh
);
1554 * Truncate blocks that were not used by write. We have to truncate the
1555 * pagecache as well so that corresponding buffers get properly unmapped.
1557 static void ext4_truncate_failed_write(struct inode
*inode
)
1559 truncate_inode_pages(inode
->i_mapping
, inode
->i_size
);
1560 ext4_truncate(inode
);
1563 static int ext4_get_block_write(struct inode
*inode
, sector_t iblock
,
1564 struct buffer_head
*bh_result
, int create
);
1565 static int ext4_write_begin(struct file
*file
, struct address_space
*mapping
,
1566 loff_t pos
, unsigned len
, unsigned flags
,
1567 struct page
**pagep
, void **fsdata
)
1569 struct inode
*inode
= mapping
->host
;
1570 int ret
, needed_blocks
;
1577 trace_ext4_write_begin(inode
, pos
, len
, flags
);
1579 * Reserve one block more for addition to orphan list in case
1580 * we allocate blocks but write fails for some reason
1582 needed_blocks
= ext4_writepage_trans_blocks(inode
) + 1;
1583 index
= pos
>> PAGE_CACHE_SHIFT
;
1584 from
= pos
& (PAGE_CACHE_SIZE
- 1);
1588 handle
= ext4_journal_start(inode
, needed_blocks
);
1589 if (IS_ERR(handle
)) {
1590 ret
= PTR_ERR(handle
);
1594 /* We cannot recurse into the filesystem as the transaction is already
1596 flags
|= AOP_FLAG_NOFS
;
1598 page
= grab_cache_page_write_begin(mapping
, index
, flags
);
1600 ext4_journal_stop(handle
);
1606 if (ext4_should_dioread_nolock(inode
))
1607 ret
= __block_write_begin(page
, pos
, len
, ext4_get_block_write
);
1609 ret
= __block_write_begin(page
, pos
, len
, ext4_get_block
);
1611 if (!ret
&& ext4_should_journal_data(inode
)) {
1612 ret
= walk_page_buffers(handle
, page_buffers(page
),
1613 from
, to
, NULL
, do_journal_get_write_access
);
1618 page_cache_release(page
);
1620 * __block_write_begin may have instantiated a few blocks
1621 * outside i_size. Trim these off again. Don't need
1622 * i_size_read because we hold i_mutex.
1624 * Add inode to orphan list in case we crash before
1627 if (pos
+ len
> inode
->i_size
&& ext4_can_truncate(inode
))
1628 ext4_orphan_add(handle
, inode
);
1630 ext4_journal_stop(handle
);
1631 if (pos
+ len
> inode
->i_size
) {
1632 ext4_truncate_failed_write(inode
);
1634 * If truncate failed early the inode might
1635 * still be on the orphan list; we need to
1636 * make sure the inode is removed from the
1637 * orphan list in that case.
1640 ext4_orphan_del(NULL
, inode
);
1644 if (ret
== -ENOSPC
&& ext4_should_retry_alloc(inode
->i_sb
, &retries
))
1650 /* For write_end() in data=journal mode */
1651 static int write_end_fn(handle_t
*handle
, struct buffer_head
*bh
)
1653 if (!buffer_mapped(bh
) || buffer_freed(bh
))
1655 set_buffer_uptodate(bh
);
1656 return ext4_handle_dirty_metadata(handle
, NULL
, bh
);
1659 static int ext4_generic_write_end(struct file
*file
,
1660 struct address_space
*mapping
,
1661 loff_t pos
, unsigned len
, unsigned copied
,
1662 struct page
*page
, void *fsdata
)
1664 int i_size_changed
= 0;
1665 struct inode
*inode
= mapping
->host
;
1666 handle_t
*handle
= ext4_journal_current_handle();
1668 copied
= block_write_end(file
, mapping
, pos
, len
, copied
, page
, fsdata
);
1671 * No need to use i_size_read() here, the i_size
1672 * cannot change under us because we hold i_mutex.
1674 * But it's important to update i_size while still holding page lock:
1675 * page writeout could otherwise come in and zero beyond i_size.
1677 if (pos
+ copied
> inode
->i_size
) {
1678 i_size_write(inode
, pos
+ copied
);
1682 if (pos
+ copied
> EXT4_I(inode
)->i_disksize
) {
1683 /* We need to mark inode dirty even if
1684 * new_i_size is less that inode->i_size
1685 * bu greater than i_disksize.(hint delalloc)
1687 ext4_update_i_disksize(inode
, (pos
+ copied
));
1691 page_cache_release(page
);
1694 * Don't mark the inode dirty under page lock. First, it unnecessarily
1695 * makes the holding time of page lock longer. Second, it forces lock
1696 * ordering of page lock and transaction start for journaling
1700 ext4_mark_inode_dirty(handle
, inode
);
1706 * We need to pick up the new inode size which generic_commit_write gave us
1707 * `file' can be NULL - eg, when called from page_symlink().
1709 * ext4 never places buffers on inode->i_mapping->private_list. metadata
1710 * buffers are managed internally.
1712 static int ext4_ordered_write_end(struct file
*file
,
1713 struct address_space
*mapping
,
1714 loff_t pos
, unsigned len
, unsigned copied
,
1715 struct page
*page
, void *fsdata
)
1717 handle_t
*handle
= ext4_journal_current_handle();
1718 struct inode
*inode
= mapping
->host
;
1721 trace_ext4_ordered_write_end(inode
, pos
, len
, copied
);
1722 ret
= ext4_jbd2_file_inode(handle
, inode
);
1725 ret2
= ext4_generic_write_end(file
, mapping
, pos
, len
, copied
,
1728 if (pos
+ len
> inode
->i_size
&& ext4_can_truncate(inode
))
1729 /* if we have allocated more blocks and copied
1730 * less. We will have blocks allocated outside
1731 * inode->i_size. So truncate them
1733 ext4_orphan_add(handle
, inode
);
1737 ret2
= ext4_journal_stop(handle
);
1741 if (pos
+ len
> inode
->i_size
) {
1742 ext4_truncate_failed_write(inode
);
1744 * If truncate failed early the inode might still be
1745 * on the orphan list; we need to make sure the inode
1746 * is removed from the orphan list in that case.
1749 ext4_orphan_del(NULL
, inode
);
1753 return ret
? ret
: copied
;
1756 static int ext4_writeback_write_end(struct file
*file
,
1757 struct address_space
*mapping
,
1758 loff_t pos
, unsigned len
, unsigned copied
,
1759 struct page
*page
, void *fsdata
)
1761 handle_t
*handle
= ext4_journal_current_handle();
1762 struct inode
*inode
= mapping
->host
;
1765 trace_ext4_writeback_write_end(inode
, pos
, len
, copied
);
1766 ret2
= ext4_generic_write_end(file
, mapping
, pos
, len
, copied
,
1769 if (pos
+ len
> inode
->i_size
&& ext4_can_truncate(inode
))
1770 /* if we have allocated more blocks and copied
1771 * less. We will have blocks allocated outside
1772 * inode->i_size. So truncate them
1774 ext4_orphan_add(handle
, inode
);
1779 ret2
= ext4_journal_stop(handle
);
1783 if (pos
+ len
> inode
->i_size
) {
1784 ext4_truncate_failed_write(inode
);
1786 * If truncate failed early the inode might still be
1787 * on the orphan list; we need to make sure the inode
1788 * is removed from the orphan list in that case.
1791 ext4_orphan_del(NULL
, inode
);
1794 return ret
? ret
: copied
;
1797 static int ext4_journalled_write_end(struct file
*file
,
1798 struct address_space
*mapping
,
1799 loff_t pos
, unsigned len
, unsigned copied
,
1800 struct page
*page
, void *fsdata
)
1802 handle_t
*handle
= ext4_journal_current_handle();
1803 struct inode
*inode
= mapping
->host
;
1809 trace_ext4_journalled_write_end(inode
, pos
, len
, copied
);
1810 from
= pos
& (PAGE_CACHE_SIZE
- 1);
1814 if (!PageUptodate(page
))
1816 page_zero_new_buffers(page
, from
+copied
, to
);
1819 ret
= walk_page_buffers(handle
, page_buffers(page
), from
,
1820 to
, &partial
, write_end_fn
);
1822 SetPageUptodate(page
);
1823 new_i_size
= pos
+ copied
;
1824 if (new_i_size
> inode
->i_size
)
1825 i_size_write(inode
, pos
+copied
);
1826 ext4_set_inode_state(inode
, EXT4_STATE_JDATA
);
1827 if (new_i_size
> EXT4_I(inode
)->i_disksize
) {
1828 ext4_update_i_disksize(inode
, new_i_size
);
1829 ret2
= ext4_mark_inode_dirty(handle
, inode
);
1835 page_cache_release(page
);
1836 if (pos
+ len
> inode
->i_size
&& ext4_can_truncate(inode
))
1837 /* if we have allocated more blocks and copied
1838 * less. We will have blocks allocated outside
1839 * inode->i_size. So truncate them
1841 ext4_orphan_add(handle
, inode
);
1843 ret2
= ext4_journal_stop(handle
);
1846 if (pos
+ len
> inode
->i_size
) {
1847 ext4_truncate_failed_write(inode
);
1849 * If truncate failed early the inode might still be
1850 * on the orphan list; we need to make sure the inode
1851 * is removed from the orphan list in that case.
1854 ext4_orphan_del(NULL
, inode
);
1857 return ret
? ret
: copied
;
1861 * Reserve a single block located at lblock
1863 static int ext4_da_reserve_space(struct inode
*inode
, sector_t lblock
)
1866 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
1867 struct ext4_inode_info
*ei
= EXT4_I(inode
);
1868 unsigned long md_needed
;
1872 * recalculate the amount of metadata blocks to reserve
1873 * in order to allocate nrblocks
1874 * worse case is one extent per block
1877 spin_lock(&ei
->i_block_reservation_lock
);
1878 md_needed
= ext4_calc_metadata_amount(inode
, lblock
);
1879 trace_ext4_da_reserve_space(inode
, md_needed
);
1880 spin_unlock(&ei
->i_block_reservation_lock
);
1883 * We will charge metadata quota at writeout time; this saves
1884 * us from metadata over-estimation, though we may go over by
1885 * a small amount in the end. Here we just reserve for data.
1887 ret
= dquot_reserve_block(inode
, 1);
1891 * We do still charge estimated metadata to the sb though;
1892 * we cannot afford to run out of free blocks.
1894 if (ext4_claim_free_blocks(sbi
, md_needed
+ 1)) {
1895 dquot_release_reservation_block(inode
, 1);
1896 if (ext4_should_retry_alloc(inode
->i_sb
, &retries
)) {
1902 spin_lock(&ei
->i_block_reservation_lock
);
1903 ei
->i_reserved_data_blocks
++;
1904 ei
->i_reserved_meta_blocks
+= md_needed
;
1905 spin_unlock(&ei
->i_block_reservation_lock
);
1907 return 0; /* success */
1910 static void ext4_da_release_space(struct inode
*inode
, int to_free
)
1912 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
1913 struct ext4_inode_info
*ei
= EXT4_I(inode
);
1916 return; /* Nothing to release, exit */
1918 spin_lock(&EXT4_I(inode
)->i_block_reservation_lock
);
1920 trace_ext4_da_release_space(inode
, to_free
);
1921 if (unlikely(to_free
> ei
->i_reserved_data_blocks
)) {
1923 * if there aren't enough reserved blocks, then the
1924 * counter is messed up somewhere. Since this
1925 * function is called from invalidate page, it's
1926 * harmless to return without any action.
1928 ext4_msg(inode
->i_sb
, KERN_NOTICE
, "ext4_da_release_space: "
1929 "ino %lu, to_free %d with only %d reserved "
1930 "data blocks\n", inode
->i_ino
, to_free
,
1931 ei
->i_reserved_data_blocks
);
1933 to_free
= ei
->i_reserved_data_blocks
;
1935 ei
->i_reserved_data_blocks
-= to_free
;
1937 if (ei
->i_reserved_data_blocks
== 0) {
1939 * We can release all of the reserved metadata blocks
1940 * only when we have written all of the delayed
1941 * allocation blocks.
1943 percpu_counter_sub(&sbi
->s_dirtyblocks_counter
,
1944 ei
->i_reserved_meta_blocks
);
1945 ei
->i_reserved_meta_blocks
= 0;
1946 ei
->i_da_metadata_calc_len
= 0;
1949 /* update fs dirty data blocks counter */
1950 percpu_counter_sub(&sbi
->s_dirtyblocks_counter
, to_free
);
1952 spin_unlock(&EXT4_I(inode
)->i_block_reservation_lock
);
1954 dquot_release_reservation_block(inode
, to_free
);
1957 static void ext4_da_page_release_reservation(struct page
*page
,
1958 unsigned long offset
)
1961 struct buffer_head
*head
, *bh
;
1962 unsigned int curr_off
= 0;
1964 head
= page_buffers(page
);
1967 unsigned int next_off
= curr_off
+ bh
->b_size
;
1969 if ((offset
<= curr_off
) && (buffer_delay(bh
))) {
1971 clear_buffer_delay(bh
);
1973 curr_off
= next_off
;
1974 } while ((bh
= bh
->b_this_page
) != head
);
1975 ext4_da_release_space(page
->mapping
->host
, to_release
);
1979 * Delayed allocation stuff
1983 * mpage_da_submit_io - walks through extent of pages and try to write
1984 * them with writepage() call back
1986 * @mpd->inode: inode
1987 * @mpd->first_page: first page of the extent
1988 * @mpd->next_page: page after the last page of the extent
1990 * By the time mpage_da_submit_io() is called we expect all blocks
1991 * to be allocated. this may be wrong if allocation failed.
1993 * As pages are already locked by write_cache_pages(), we can't use it
1995 static int mpage_da_submit_io(struct mpage_da_data
*mpd
)
1998 struct pagevec pvec
;
1999 unsigned long index
, end
;
2000 int ret
= 0, err
, nr_pages
, i
;
2001 struct inode
*inode
= mpd
->inode
;
2002 struct address_space
*mapping
= inode
->i_mapping
;
2004 BUG_ON(mpd
->next_page
<= mpd
->first_page
);
2006 * We need to start from the first_page to the next_page - 1
2007 * to make sure we also write the mapped dirty buffer_heads.
2008 * If we look at mpd->b_blocknr we would only be looking
2009 * at the currently mapped buffer_heads.
2011 index
= mpd
->first_page
;
2012 end
= mpd
->next_page
- 1;
2014 pagevec_init(&pvec
, 0);
2015 while (index
<= end
) {
2016 nr_pages
= pagevec_lookup(&pvec
, mapping
, index
, PAGEVEC_SIZE
);
2019 for (i
= 0; i
< nr_pages
; i
++) {
2020 struct page
*page
= pvec
.pages
[i
];
2022 index
= page
->index
;
2027 BUG_ON(!PageLocked(page
));
2028 BUG_ON(PageWriteback(page
));
2030 pages_skipped
= mpd
->wbc
->pages_skipped
;
2031 err
= mapping
->a_ops
->writepage(page
, mpd
->wbc
);
2032 if (!err
&& (pages_skipped
== mpd
->wbc
->pages_skipped
))
2034 * have successfully written the page
2035 * without skipping the same
2037 mpd
->pages_written
++;
2041 pagevec_release(&pvec
);
2047 * mpage_put_bnr_to_bhs - walk blocks and assign them actual numbers
2049 * the function goes through all passed space and put actual disk
2050 * block numbers into buffer heads, dropping BH_Delay and BH_Unwritten
2052 static void mpage_put_bnr_to_bhs(struct mpage_da_data
*mpd
,
2053 struct ext4_map_blocks
*map
)
2055 struct inode
*inode
= mpd
->inode
;
2056 struct address_space
*mapping
= inode
->i_mapping
;
2057 int blocks
= map
->m_len
;
2058 sector_t pblock
= map
->m_pblk
, cur_logical
;
2059 struct buffer_head
*head
, *bh
;
2061 struct pagevec pvec
;
2064 index
= map
->m_lblk
>> (PAGE_CACHE_SHIFT
- inode
->i_blkbits
);
2065 end
= (map
->m_lblk
+ blocks
- 1) >> (PAGE_CACHE_SHIFT
- inode
->i_blkbits
);
2066 cur_logical
= index
<< (PAGE_CACHE_SHIFT
- inode
->i_blkbits
);
2068 pagevec_init(&pvec
, 0);
2070 while (index
<= end
) {
2071 nr_pages
= pagevec_lookup(&pvec
, mapping
, index
, PAGEVEC_SIZE
);
2074 for (i
= 0; i
< nr_pages
; i
++) {
2075 struct page
*page
= pvec
.pages
[i
];
2077 index
= page
->index
;
2082 BUG_ON(!PageLocked(page
));
2083 BUG_ON(PageWriteback(page
));
2084 BUG_ON(!page_has_buffers(page
));
2086 bh
= page_buffers(page
);
2089 /* skip blocks out of the range */
2091 if (cur_logical
>= map
->m_lblk
)
2094 } while ((bh
= bh
->b_this_page
) != head
);
2097 if (cur_logical
>= map
->m_lblk
+ blocks
)
2100 if (buffer_delay(bh
) || buffer_unwritten(bh
)) {
2102 BUG_ON(bh
->b_bdev
!= inode
->i_sb
->s_bdev
);
2104 if (buffer_delay(bh
)) {
2105 clear_buffer_delay(bh
);
2106 bh
->b_blocknr
= pblock
;
2109 * unwritten already should have
2110 * blocknr assigned. Verify that
2112 clear_buffer_unwritten(bh
);
2113 BUG_ON(bh
->b_blocknr
!= pblock
);
2116 } else if (buffer_mapped(bh
))
2117 BUG_ON(bh
->b_blocknr
!= pblock
);
2119 if (map
->m_flags
& EXT4_MAP_UNINIT
)
2120 set_buffer_uninit(bh
);
2123 } while ((bh
= bh
->b_this_page
) != head
);
2125 pagevec_release(&pvec
);
2130 static void ext4_da_block_invalidatepages(struct mpage_da_data
*mpd
,
2131 sector_t logical
, long blk_cnt
)
2135 struct pagevec pvec
;
2136 struct inode
*inode
= mpd
->inode
;
2137 struct address_space
*mapping
= inode
->i_mapping
;
2139 index
= logical
>> (PAGE_CACHE_SHIFT
- inode
->i_blkbits
);
2140 end
= (logical
+ blk_cnt
- 1) >>
2141 (PAGE_CACHE_SHIFT
- inode
->i_blkbits
);
2142 while (index
<= end
) {
2143 nr_pages
= pagevec_lookup(&pvec
, mapping
, index
, PAGEVEC_SIZE
);
2146 for (i
= 0; i
< nr_pages
; i
++) {
2147 struct page
*page
= pvec
.pages
[i
];
2148 if (page
->index
> end
)
2150 BUG_ON(!PageLocked(page
));
2151 BUG_ON(PageWriteback(page
));
2152 block_invalidatepage(page
, 0);
2153 ClearPageUptodate(page
);
2156 index
= pvec
.pages
[nr_pages
- 1]->index
+ 1;
2157 pagevec_release(&pvec
);
2162 static void ext4_print_free_blocks(struct inode
*inode
)
2164 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
2165 printk(KERN_CRIT
"Total free blocks count %lld\n",
2166 ext4_count_free_blocks(inode
->i_sb
));
2167 printk(KERN_CRIT
"Free/Dirty block details\n");
2168 printk(KERN_CRIT
"free_blocks=%lld\n",
2169 (long long) percpu_counter_sum(&sbi
->s_freeblocks_counter
));
2170 printk(KERN_CRIT
"dirty_blocks=%lld\n",
2171 (long long) percpu_counter_sum(&sbi
->s_dirtyblocks_counter
));
2172 printk(KERN_CRIT
"Block reservation details\n");
2173 printk(KERN_CRIT
"i_reserved_data_blocks=%u\n",
2174 EXT4_I(inode
)->i_reserved_data_blocks
);
2175 printk(KERN_CRIT
"i_reserved_meta_blocks=%u\n",
2176 EXT4_I(inode
)->i_reserved_meta_blocks
);
2181 * mpage_da_map_blocks - go through given space
2183 * @mpd - bh describing space
2185 * The function skips space we know is already mapped to disk blocks.
2188 static int mpage_da_map_blocks(struct mpage_da_data
*mpd
)
2190 int err
, blks
, get_blocks_flags
;
2191 struct ext4_map_blocks map
;
2192 sector_t next
= mpd
->b_blocknr
;
2193 unsigned max_blocks
= mpd
->b_size
>> mpd
->inode
->i_blkbits
;
2194 loff_t disksize
= EXT4_I(mpd
->inode
)->i_disksize
;
2195 handle_t
*handle
= NULL
;
2198 * We consider only non-mapped and non-allocated blocks
2200 if ((mpd
->b_state
& (1 << BH_Mapped
)) &&
2201 !(mpd
->b_state
& (1 << BH_Delay
)) &&
2202 !(mpd
->b_state
& (1 << BH_Unwritten
)))
2206 * If we didn't accumulate anything to write simply return
2211 handle
= ext4_journal_current_handle();
2215 * Call ext4_map_blocks() to allocate any delayed allocation
2216 * blocks, or to convert an uninitialized extent to be
2217 * initialized (in the case where we have written into
2218 * one or more preallocated blocks).
2220 * We pass in the magic EXT4_GET_BLOCKS_DELALLOC_RESERVE to
2221 * indicate that we are on the delayed allocation path. This
2222 * affects functions in many different parts of the allocation
2223 * call path. This flag exists primarily because we don't
2224 * want to change *many* call functions, so ext4_map_blocks()
2225 * will set the magic i_delalloc_reserved_flag once the
2226 * inode's allocation semaphore is taken.
2228 * If the blocks in questions were delalloc blocks, set
2229 * EXT4_GET_BLOCKS_DELALLOC_RESERVE so the delalloc accounting
2230 * variables are updated after the blocks have been allocated.
2233 map
.m_len
= max_blocks
;
2234 get_blocks_flags
= EXT4_GET_BLOCKS_CREATE
;
2235 if (ext4_should_dioread_nolock(mpd
->inode
))
2236 get_blocks_flags
|= EXT4_GET_BLOCKS_IO_CREATE_EXT
;
2237 if (mpd
->b_state
& (1 << BH_Delay
))
2238 get_blocks_flags
|= EXT4_GET_BLOCKS_DELALLOC_RESERVE
;
2240 blks
= ext4_map_blocks(handle
, mpd
->inode
, &map
, get_blocks_flags
);
2242 struct super_block
*sb
= mpd
->inode
->i_sb
;
2246 * If get block returns with error we simply
2247 * return. Later writepage will redirty the page and
2248 * writepages will find the dirty page again
2253 if (err
== -ENOSPC
&&
2254 ext4_count_free_blocks(sb
)) {
2260 * get block failure will cause us to loop in
2261 * writepages, because a_ops->writepage won't be able
2262 * to make progress. The page will be redirtied by
2263 * writepage and writepages will again try to write
2266 if (!(EXT4_SB(sb
)->s_mount_flags
& EXT4_MF_FS_ABORTED
)) {
2267 ext4_msg(sb
, KERN_CRIT
,
2268 "delayed block allocation failed for inode %lu "
2269 "at logical offset %llu with max blocks %zd "
2270 "with error %d", mpd
->inode
->i_ino
,
2271 (unsigned long long) next
,
2272 mpd
->b_size
>> mpd
->inode
->i_blkbits
, err
);
2273 ext4_msg(sb
, KERN_CRIT
,
2274 "This should not happen!! Data will be lost\n");
2276 ext4_print_free_blocks(mpd
->inode
);
2278 /* invalidate all the pages */
2279 ext4_da_block_invalidatepages(mpd
, next
,
2280 mpd
->b_size
>> mpd
->inode
->i_blkbits
);
2285 if (map
.m_flags
& EXT4_MAP_NEW
) {
2286 struct block_device
*bdev
= mpd
->inode
->i_sb
->s_bdev
;
2289 for (i
= 0; i
< map
.m_len
; i
++)
2290 unmap_underlying_metadata(bdev
, map
.m_pblk
+ i
);
2294 * If blocks are delayed marked, we need to
2295 * put actual blocknr and drop delayed bit
2297 if ((mpd
->b_state
& (1 << BH_Delay
)) ||
2298 (mpd
->b_state
& (1 << BH_Unwritten
)))
2299 mpage_put_bnr_to_bhs(mpd
, &map
);
2301 if (ext4_should_order_data(mpd
->inode
)) {
2302 err
= ext4_jbd2_file_inode(handle
, mpd
->inode
);
2308 * Update on-disk size along with block allocation.
2310 disksize
= ((loff_t
) next
+ blks
) << mpd
->inode
->i_blkbits
;
2311 if (disksize
> i_size_read(mpd
->inode
))
2312 disksize
= i_size_read(mpd
->inode
);
2313 if (disksize
> EXT4_I(mpd
->inode
)->i_disksize
) {
2314 ext4_update_i_disksize(mpd
->inode
, disksize
);
2315 return ext4_mark_inode_dirty(handle
, mpd
->inode
);
2321 #define BH_FLAGS ((1 << BH_Uptodate) | (1 << BH_Mapped) | \
2322 (1 << BH_Delay) | (1 << BH_Unwritten))
2325 * mpage_add_bh_to_extent - try to add one more block to extent of blocks
2327 * @mpd->lbh - extent of blocks
2328 * @logical - logical number of the block in the file
2329 * @bh - bh of the block (used to access block's state)
2331 * the function is used to collect contig. blocks in same state
2333 static void mpage_add_bh_to_extent(struct mpage_da_data
*mpd
,
2334 sector_t logical
, size_t b_size
,
2335 unsigned long b_state
)
2338 int nrblocks
= mpd
->b_size
>> mpd
->inode
->i_blkbits
;
2340 if (nrblocks
>= 8*1024*1024/mpd
->inode
->i_sb
->s_blocksize
)
2343 /* check if thereserved journal credits might overflow */
2344 if (!(ext4_test_inode_flag(mpd
->inode
, EXT4_INODE_EXTENTS
))) {
2345 if (nrblocks
>= EXT4_MAX_TRANS_DATA
) {
2347 * With non-extent format we are limited by the journal
2348 * credit available. Total credit needed to insert
2349 * nrblocks contiguous blocks is dependent on the
2350 * nrblocks. So limit nrblocks.
2353 } else if ((nrblocks
+ (b_size
>> mpd
->inode
->i_blkbits
)) >
2354 EXT4_MAX_TRANS_DATA
) {
2356 * Adding the new buffer_head would make it cross the
2357 * allowed limit for which we have journal credit
2358 * reserved. So limit the new bh->b_size
2360 b_size
= (EXT4_MAX_TRANS_DATA
- nrblocks
) <<
2361 mpd
->inode
->i_blkbits
;
2362 /* we will do mpage_da_submit_io in the next loop */
2366 * First block in the extent
2368 if (mpd
->b_size
== 0) {
2369 mpd
->b_blocknr
= logical
;
2370 mpd
->b_size
= b_size
;
2371 mpd
->b_state
= b_state
& BH_FLAGS
;
2375 next
= mpd
->b_blocknr
+ nrblocks
;
2377 * Can we merge the block to our big extent?
2379 if (logical
== next
&& (b_state
& BH_FLAGS
) == mpd
->b_state
) {
2380 mpd
->b_size
+= b_size
;
2386 * We couldn't merge the block to our extent, so we
2387 * need to flush current extent and start new one
2389 if (mpage_da_map_blocks(mpd
) == 0)
2390 mpage_da_submit_io(mpd
);
2395 static int ext4_bh_delay_or_unwritten(handle_t
*handle
, struct buffer_head
*bh
)
2397 return (buffer_delay(bh
) || buffer_unwritten(bh
)) && buffer_dirty(bh
);
2401 * __mpage_da_writepage - finds extent of pages and blocks
2403 * @page: page to consider
2404 * @wbc: not used, we just follow rules
2407 * The function finds extents of pages and scan them for all blocks.
2409 static int __mpage_da_writepage(struct page
*page
,
2410 struct writeback_control
*wbc
, void *data
)
2412 struct mpage_da_data
*mpd
= data
;
2413 struct inode
*inode
= mpd
->inode
;
2414 struct buffer_head
*bh
, *head
;
2418 * Can we merge this page to current extent?
2420 if (mpd
->next_page
!= page
->index
) {
2422 * Nope, we can't. So, we map non-allocated blocks
2423 * and start IO on them using writepage()
2425 if (mpd
->next_page
!= mpd
->first_page
) {
2426 if (mpage_da_map_blocks(mpd
) == 0)
2427 mpage_da_submit_io(mpd
);
2429 * skip rest of the page in the page_vec
2432 redirty_page_for_writepage(wbc
, page
);
2434 return MPAGE_DA_EXTENT_TAIL
;
2438 * Start next extent of pages ...
2440 mpd
->first_page
= page
->index
;
2450 mpd
->next_page
= page
->index
+ 1;
2451 logical
= (sector_t
) page
->index
<<
2452 (PAGE_CACHE_SHIFT
- inode
->i_blkbits
);
2454 if (!page_has_buffers(page
)) {
2455 mpage_add_bh_to_extent(mpd
, logical
, PAGE_CACHE_SIZE
,
2456 (1 << BH_Dirty
) | (1 << BH_Uptodate
));
2458 return MPAGE_DA_EXTENT_TAIL
;
2461 * Page with regular buffer heads, just add all dirty ones
2463 head
= page_buffers(page
);
2466 BUG_ON(buffer_locked(bh
));
2468 * We need to try to allocate
2469 * unmapped blocks in the same page.
2470 * Otherwise we won't make progress
2471 * with the page in ext4_writepage
2473 if (ext4_bh_delay_or_unwritten(NULL
, bh
)) {
2474 mpage_add_bh_to_extent(mpd
, logical
,
2478 return MPAGE_DA_EXTENT_TAIL
;
2479 } else if (buffer_dirty(bh
) && (buffer_mapped(bh
))) {
2481 * mapped dirty buffer. We need to update
2482 * the b_state because we look at
2483 * b_state in mpage_da_map_blocks. We don't
2484 * update b_size because if we find an
2485 * unmapped buffer_head later we need to
2486 * use the b_state flag of that buffer_head.
2488 if (mpd
->b_size
== 0)
2489 mpd
->b_state
= bh
->b_state
& BH_FLAGS
;
2492 } while ((bh
= bh
->b_this_page
) != head
);
2499 * This is a special get_blocks_t callback which is used by
2500 * ext4_da_write_begin(). It will either return mapped block or
2501 * reserve space for a single block.
2503 * For delayed buffer_head we have BH_Mapped, BH_New, BH_Delay set.
2504 * We also have b_blocknr = -1 and b_bdev initialized properly
2506 * For unwritten buffer_head we have BH_Mapped, BH_New, BH_Unwritten set.
2507 * We also have b_blocknr = physicalblock mapping unwritten extent and b_bdev
2508 * initialized properly.
2510 static int ext4_da_get_block_prep(struct inode
*inode
, sector_t iblock
,
2511 struct buffer_head
*bh
, int create
)
2513 struct ext4_map_blocks map
;
2515 sector_t invalid_block
= ~((sector_t
) 0xffff);
2517 if (invalid_block
< ext4_blocks_count(EXT4_SB(inode
->i_sb
)->s_es
))
2520 BUG_ON(create
== 0);
2521 BUG_ON(bh
->b_size
!= inode
->i_sb
->s_blocksize
);
2523 map
.m_lblk
= iblock
;
2527 * first, we need to know whether the block is allocated already
2528 * preallocated blocks are unmapped but should treated
2529 * the same as allocated blocks.
2531 ret
= ext4_map_blocks(NULL
, inode
, &map
, 0);
2535 if (buffer_delay(bh
))
2536 return 0; /* Not sure this could or should happen */
2537 ret
= ext4_da_reserve_space(inode
, iblock
);
2539 /* not enough space to reserve */
2542 map_bh(bh
, inode
->i_sb
, invalid_block
);
2544 set_buffer_delay(bh
);
2548 map_bh(bh
, inode
->i_sb
, map
.m_pblk
);
2549 bh
->b_state
= (bh
->b_state
& ~EXT4_MAP_FLAGS
) | map
.m_flags
;
2551 if (buffer_unwritten(bh
)) {
2552 /* A delayed write to unwritten bh should be marked
2553 * new and mapped. Mapped ensures that we don't do
2554 * get_block multiple times when we write to the same
2555 * offset and new ensures that we do proper zero out
2556 * for partial write.
2559 set_buffer_mapped(bh
);
2565 * This function is used as a standard get_block_t calback function
2566 * when there is no desire to allocate any blocks. It is used as a
2567 * callback function for block_prepare_write() and block_write_full_page().
2568 * These functions should only try to map a single block at a time.
2570 * Since this function doesn't do block allocations even if the caller
2571 * requests it by passing in create=1, it is critically important that
2572 * any caller checks to make sure that any buffer heads are returned
2573 * by this function are either all already mapped or marked for
2574 * delayed allocation before calling block_write_full_page(). Otherwise,
2575 * b_blocknr could be left unitialized, and the page write functions will
2576 * be taken by surprise.
2578 static int noalloc_get_block_write(struct inode
*inode
, sector_t iblock
,
2579 struct buffer_head
*bh_result
, int create
)
2581 BUG_ON(bh_result
->b_size
!= inode
->i_sb
->s_blocksize
);
2582 return _ext4_get_block(inode
, iblock
, bh_result
, 0);
2585 static int bget_one(handle_t
*handle
, struct buffer_head
*bh
)
2591 static int bput_one(handle_t
*handle
, struct buffer_head
*bh
)
2597 static int __ext4_journalled_writepage(struct page
*page
,
2600 struct address_space
*mapping
= page
->mapping
;
2601 struct inode
*inode
= mapping
->host
;
2602 struct buffer_head
*page_bufs
;
2603 handle_t
*handle
= NULL
;
2607 page_bufs
= page_buffers(page
);
2609 walk_page_buffers(handle
, page_bufs
, 0, len
, NULL
, bget_one
);
2610 /* As soon as we unlock the page, it can go away, but we have
2611 * references to buffers so we are safe */
2614 handle
= ext4_journal_start(inode
, ext4_writepage_trans_blocks(inode
));
2615 if (IS_ERR(handle
)) {
2616 ret
= PTR_ERR(handle
);
2620 ret
= walk_page_buffers(handle
, page_bufs
, 0, len
, NULL
,
2621 do_journal_get_write_access
);
2623 err
= walk_page_buffers(handle
, page_bufs
, 0, len
, NULL
,
2627 err
= ext4_journal_stop(handle
);
2631 walk_page_buffers(handle
, page_bufs
, 0, len
, NULL
, bput_one
);
2632 ext4_set_inode_state(inode
, EXT4_STATE_JDATA
);
2637 static int ext4_set_bh_endio(struct buffer_head
*bh
, struct inode
*inode
);
2638 static void ext4_end_io_buffer_write(struct buffer_head
*bh
, int uptodate
);
2641 * Note that we don't need to start a transaction unless we're journaling data
2642 * because we should have holes filled from ext4_page_mkwrite(). We even don't
2643 * need to file the inode to the transaction's list in ordered mode because if
2644 * we are writing back data added by write(), the inode is already there and if
2645 * we are writing back data modified via mmap(), noone guarantees in which
2646 * transaction the data will hit the disk. In case we are journaling data, we
2647 * cannot start transaction directly because transaction start ranks above page
2648 * lock so we have to do some magic.
2650 * This function can get called via...
2651 * - ext4_da_writepages after taking page lock (have journal handle)
2652 * - journal_submit_inode_data_buffers (no journal handle)
2653 * - shrink_page_list via pdflush (no journal handle)
2654 * - grab_page_cache when doing write_begin (have journal handle)
2656 * We don't do any block allocation in this function. If we have page with
2657 * multiple blocks we need to write those buffer_heads that are mapped. This
2658 * is important for mmaped based write. So if we do with blocksize 1K
2659 * truncate(f, 1024);
2660 * a = mmap(f, 0, 4096);
2662 * truncate(f, 4096);
2663 * we have in the page first buffer_head mapped via page_mkwrite call back
2664 * but other bufer_heads would be unmapped but dirty(dirty done via the
2665 * do_wp_page). So writepage should write the first block. If we modify
2666 * the mmap area beyond 1024 we will again get a page_fault and the
2667 * page_mkwrite callback will do the block allocation and mark the
2668 * buffer_heads mapped.
2670 * We redirty the page if we have any buffer_heads that is either delay or
2671 * unwritten in the page.
2673 * We can get recursively called as show below.
2675 * ext4_writepage() -> kmalloc() -> __alloc_pages() -> page_launder() ->
2678 * But since we don't do any block allocation we should not deadlock.
2679 * Page also have the dirty flag cleared so we don't get recurive page_lock.
2681 static int ext4_writepage(struct page
*page
,
2682 struct writeback_control
*wbc
)
2687 struct buffer_head
*page_bufs
= NULL
;
2688 struct inode
*inode
= page
->mapping
->host
;
2690 trace_ext4_writepage(inode
, page
);
2691 size
= i_size_read(inode
);
2692 if (page
->index
== size
>> PAGE_CACHE_SHIFT
)
2693 len
= size
& ~PAGE_CACHE_MASK
;
2695 len
= PAGE_CACHE_SIZE
;
2697 if (page_has_buffers(page
)) {
2698 page_bufs
= page_buffers(page
);
2699 if (walk_page_buffers(NULL
, page_bufs
, 0, len
, NULL
,
2700 ext4_bh_delay_or_unwritten
)) {
2702 * We don't want to do block allocation
2703 * So redirty the page and return
2704 * We may reach here when we do a journal commit
2705 * via journal_submit_inode_data_buffers.
2706 * If we don't have mapping block we just ignore
2707 * them. We can also reach here via shrink_page_list
2709 redirty_page_for_writepage(wbc
, page
);
2715 * The test for page_has_buffers() is subtle:
2716 * We know the page is dirty but it lost buffers. That means
2717 * that at some moment in time after write_begin()/write_end()
2718 * has been called all buffers have been clean and thus they
2719 * must have been written at least once. So they are all
2720 * mapped and we can happily proceed with mapping them
2721 * and writing the page.
2723 * Try to initialize the buffer_heads and check whether
2724 * all are mapped and non delay. We don't want to
2725 * do block allocation here.
2727 ret
= block_prepare_write(page
, 0, len
,
2728 noalloc_get_block_write
);
2730 page_bufs
= page_buffers(page
);
2731 /* check whether all are mapped and non delay */
2732 if (walk_page_buffers(NULL
, page_bufs
, 0, len
, NULL
,
2733 ext4_bh_delay_or_unwritten
)) {
2734 redirty_page_for_writepage(wbc
, page
);
2740 * We can't do block allocation here
2741 * so just redity the page and unlock
2744 redirty_page_for_writepage(wbc
, page
);
2748 /* now mark the buffer_heads as dirty and uptodate */
2749 block_commit_write(page
, 0, len
);
2752 if (PageChecked(page
) && ext4_should_journal_data(inode
)) {
2754 * It's mmapped pagecache. Add buffers and journal it. There
2755 * doesn't seem much point in redirtying the page here.
2757 ClearPageChecked(page
);
2758 return __ext4_journalled_writepage(page
, len
);
2761 if (page_bufs
&& buffer_uninit(page_bufs
)) {
2762 ext4_set_bh_endio(page_bufs
, inode
);
2763 ret
= block_write_full_page_endio(page
, noalloc_get_block_write
,
2764 wbc
, ext4_end_io_buffer_write
);
2766 ret
= block_write_full_page(page
, noalloc_get_block_write
,
2773 * This is called via ext4_da_writepages() to
2774 * calulate the total number of credits to reserve to fit
2775 * a single extent allocation into a single transaction,
2776 * ext4_da_writpeages() will loop calling this before
2777 * the block allocation.
2780 static int ext4_da_writepages_trans_blocks(struct inode
*inode
)
2782 int max_blocks
= EXT4_I(inode
)->i_reserved_data_blocks
;
2785 * With non-extent format the journal credit needed to
2786 * insert nrblocks contiguous block is dependent on
2787 * number of contiguous block. So we will limit
2788 * number of contiguous block to a sane value
2790 if (!(ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
)) &&
2791 (max_blocks
> EXT4_MAX_TRANS_DATA
))
2792 max_blocks
= EXT4_MAX_TRANS_DATA
;
2794 return ext4_chunk_trans_blocks(inode
, max_blocks
);
2798 * write_cache_pages_da - walk the list of dirty pages of the given
2799 * address space and call the callback function (which usually writes
2802 * This is a forked version of write_cache_pages(). Differences:
2803 * Range cyclic is ignored.
2804 * no_nrwrite_index_update is always presumed true
2806 static int write_cache_pages_da(struct address_space
*mapping
,
2807 struct writeback_control
*wbc
,
2808 struct mpage_da_data
*mpd
)
2812 struct pagevec pvec
;
2815 pgoff_t end
; /* Inclusive */
2816 long nr_to_write
= wbc
->nr_to_write
;
2818 pagevec_init(&pvec
, 0);
2819 index
= wbc
->range_start
>> PAGE_CACHE_SHIFT
;
2820 end
= wbc
->range_end
>> PAGE_CACHE_SHIFT
;
2822 while (!done
&& (index
<= end
)) {
2825 nr_pages
= pagevec_lookup_tag(&pvec
, mapping
, &index
,
2826 PAGECACHE_TAG_DIRTY
,
2827 min(end
- index
, (pgoff_t
)PAGEVEC_SIZE
-1) + 1);
2831 for (i
= 0; i
< nr_pages
; i
++) {
2832 struct page
*page
= pvec
.pages
[i
];
2835 * At this point, the page may be truncated or
2836 * invalidated (changing page->mapping to NULL), or
2837 * even swizzled back from swapper_space to tmpfs file
2838 * mapping. However, page->index will not change
2839 * because we have a reference on the page.
2841 if (page
->index
> end
) {
2849 * Page truncated or invalidated. We can freely skip it
2850 * then, even for data integrity operations: the page
2851 * has disappeared concurrently, so there could be no
2852 * real expectation of this data interity operation
2853 * even if there is now a new, dirty page at the same
2854 * pagecache address.
2856 if (unlikely(page
->mapping
!= mapping
)) {
2862 if (!PageDirty(page
)) {
2863 /* someone wrote it for us */
2864 goto continue_unlock
;
2867 if (PageWriteback(page
)) {
2868 if (wbc
->sync_mode
!= WB_SYNC_NONE
)
2869 wait_on_page_writeback(page
);
2871 goto continue_unlock
;
2874 BUG_ON(PageWriteback(page
));
2875 if (!clear_page_dirty_for_io(page
))
2876 goto continue_unlock
;
2878 ret
= __mpage_da_writepage(page
, wbc
, mpd
);
2879 if (unlikely(ret
)) {
2880 if (ret
== AOP_WRITEPAGE_ACTIVATE
) {
2889 if (nr_to_write
> 0) {
2891 if (nr_to_write
== 0 &&
2892 wbc
->sync_mode
== WB_SYNC_NONE
) {
2894 * We stop writing back only if we are
2895 * not doing integrity sync. In case of
2896 * integrity sync we have to keep going
2897 * because someone may be concurrently
2898 * dirtying pages, and we might have
2899 * synced a lot of newly appeared dirty
2900 * pages, but have not synced all of the
2908 pagevec_release(&pvec
);
2915 static int ext4_da_writepages(struct address_space
*mapping
,
2916 struct writeback_control
*wbc
)
2919 int range_whole
= 0;
2920 handle_t
*handle
= NULL
;
2921 struct mpage_da_data mpd
;
2922 struct inode
*inode
= mapping
->host
;
2923 int pages_written
= 0;
2925 unsigned int max_pages
;
2926 int range_cyclic
, cycled
= 1, io_done
= 0;
2927 int needed_blocks
, ret
= 0;
2928 long desired_nr_to_write
, nr_to_writebump
= 0;
2929 loff_t range_start
= wbc
->range_start
;
2930 struct ext4_sb_info
*sbi
= EXT4_SB(mapping
->host
->i_sb
);
2932 trace_ext4_da_writepages(inode
, wbc
);
2935 * No pages to write? This is mainly a kludge to avoid starting
2936 * a transaction for special inodes like journal inode on last iput()
2937 * because that could violate lock ordering on umount
2939 if (!mapping
->nrpages
|| !mapping_tagged(mapping
, PAGECACHE_TAG_DIRTY
))
2943 * If the filesystem has aborted, it is read-only, so return
2944 * right away instead of dumping stack traces later on that
2945 * will obscure the real source of the problem. We test
2946 * EXT4_MF_FS_ABORTED instead of sb->s_flag's MS_RDONLY because
2947 * the latter could be true if the filesystem is mounted
2948 * read-only, and in that case, ext4_da_writepages should
2949 * *never* be called, so if that ever happens, we would want
2952 if (unlikely(sbi
->s_mount_flags
& EXT4_MF_FS_ABORTED
))
2955 if (wbc
->range_start
== 0 && wbc
->range_end
== LLONG_MAX
)
2958 range_cyclic
= wbc
->range_cyclic
;
2959 if (wbc
->range_cyclic
) {
2960 index
= mapping
->writeback_index
;
2963 wbc
->range_start
= index
<< PAGE_CACHE_SHIFT
;
2964 wbc
->range_end
= LLONG_MAX
;
2965 wbc
->range_cyclic
= 0;
2967 index
= wbc
->range_start
>> PAGE_CACHE_SHIFT
;
2970 * This works around two forms of stupidity. The first is in
2971 * the writeback code, which caps the maximum number of pages
2972 * written to be 1024 pages. This is wrong on multiple
2973 * levels; different architectues have a different page size,
2974 * which changes the maximum amount of data which gets
2975 * written. Secondly, 4 megabytes is way too small. XFS
2976 * forces this value to be 16 megabytes by multiplying
2977 * nr_to_write parameter by four, and then relies on its
2978 * allocator to allocate larger extents to make them
2979 * contiguous. Unfortunately this brings us to the second
2980 * stupidity, which is that ext4's mballoc code only allocates
2981 * at most 2048 blocks. So we force contiguous writes up to
2982 * the number of dirty blocks in the inode, or
2983 * sbi->max_writeback_mb_bump whichever is smaller.
2985 max_pages
= sbi
->s_max_writeback_mb_bump
<< (20 - PAGE_CACHE_SHIFT
);
2986 if (!range_cyclic
&& range_whole
)
2987 desired_nr_to_write
= wbc
->nr_to_write
* 8;
2989 desired_nr_to_write
= ext4_num_dirty_pages(inode
, index
,
2991 if (desired_nr_to_write
> max_pages
)
2992 desired_nr_to_write
= max_pages
;
2994 if (wbc
->nr_to_write
< desired_nr_to_write
) {
2995 nr_to_writebump
= desired_nr_to_write
- wbc
->nr_to_write
;
2996 wbc
->nr_to_write
= desired_nr_to_write
;
3000 mpd
.inode
= mapping
->host
;
3002 pages_skipped
= wbc
->pages_skipped
;
3005 while (!ret
&& wbc
->nr_to_write
> 0) {
3008 * we insert one extent at a time. So we need
3009 * credit needed for single extent allocation.
3010 * journalled mode is currently not supported
3013 BUG_ON(ext4_should_journal_data(inode
));
3014 needed_blocks
= ext4_da_writepages_trans_blocks(inode
);
3016 /* start a new transaction*/
3017 handle
= ext4_journal_start(inode
, needed_blocks
);
3018 if (IS_ERR(handle
)) {
3019 ret
= PTR_ERR(handle
);
3020 ext4_msg(inode
->i_sb
, KERN_CRIT
, "%s: jbd2_start: "
3021 "%ld pages, ino %lu; err %d", __func__
,
3022 wbc
->nr_to_write
, inode
->i_ino
, ret
);
3023 goto out_writepages
;
3027 * Now call __mpage_da_writepage to find the next
3028 * contiguous region of logical blocks that need
3029 * blocks to be allocated by ext4. We don't actually
3030 * submit the blocks for I/O here, even though
3031 * write_cache_pages thinks it will, and will set the
3032 * pages as clean for write before calling
3033 * __mpage_da_writepage().
3041 mpd
.pages_written
= 0;
3043 ret
= write_cache_pages_da(mapping
, wbc
, &mpd
);
3045 * If we have a contiguous extent of pages and we
3046 * haven't done the I/O yet, map the blocks and submit
3049 if (!mpd
.io_done
&& mpd
.next_page
!= mpd
.first_page
) {
3050 if (mpage_da_map_blocks(&mpd
) == 0)
3051 mpage_da_submit_io(&mpd
);
3053 ret
= MPAGE_DA_EXTENT_TAIL
;
3055 trace_ext4_da_write_pages(inode
, &mpd
);
3056 wbc
->nr_to_write
-= mpd
.pages_written
;
3058 ext4_journal_stop(handle
);
3060 if ((mpd
.retval
== -ENOSPC
) && sbi
->s_journal
) {
3061 /* commit the transaction which would
3062 * free blocks released in the transaction
3065 jbd2_journal_force_commit_nested(sbi
->s_journal
);
3066 wbc
->pages_skipped
= pages_skipped
;
3068 } else if (ret
== MPAGE_DA_EXTENT_TAIL
) {
3070 * got one extent now try with
3073 pages_written
+= mpd
.pages_written
;
3074 wbc
->pages_skipped
= pages_skipped
;
3077 } else if (wbc
->nr_to_write
)
3079 * There is no more writeout needed
3080 * or we requested for a noblocking writeout
3081 * and we found the device congested
3085 if (!io_done
&& !cycled
) {
3088 wbc
->range_start
= index
<< PAGE_CACHE_SHIFT
;
3089 wbc
->range_end
= mapping
->writeback_index
- 1;
3092 if (pages_skipped
!= wbc
->pages_skipped
)
3093 ext4_msg(inode
->i_sb
, KERN_CRIT
,
3094 "This should not happen leaving %s "
3095 "with nr_to_write = %ld ret = %d",
3096 __func__
, wbc
->nr_to_write
, ret
);
3099 index
+= pages_written
;
3100 wbc
->range_cyclic
= range_cyclic
;
3101 if (wbc
->range_cyclic
|| (range_whole
&& wbc
->nr_to_write
> 0))
3103 * set the writeback_index so that range_cyclic
3104 * mode will write it back later
3106 mapping
->writeback_index
= index
;
3109 wbc
->nr_to_write
-= nr_to_writebump
;
3110 wbc
->range_start
= range_start
;
3111 trace_ext4_da_writepages_result(inode
, wbc
, ret
, pages_written
);
3115 #define FALL_BACK_TO_NONDELALLOC 1
3116 static int ext4_nonda_switch(struct super_block
*sb
)
3118 s64 free_blocks
, dirty_blocks
;
3119 struct ext4_sb_info
*sbi
= EXT4_SB(sb
);
3122 * switch to non delalloc mode if we are running low
3123 * on free block. The free block accounting via percpu
3124 * counters can get slightly wrong with percpu_counter_batch getting
3125 * accumulated on each CPU without updating global counters
3126 * Delalloc need an accurate free block accounting. So switch
3127 * to non delalloc when we are near to error range.
3129 free_blocks
= percpu_counter_read_positive(&sbi
->s_freeblocks_counter
);
3130 dirty_blocks
= percpu_counter_read_positive(&sbi
->s_dirtyblocks_counter
);
3131 if (2 * free_blocks
< 3 * dirty_blocks
||
3132 free_blocks
< (dirty_blocks
+ EXT4_FREEBLOCKS_WATERMARK
)) {
3134 * free block count is less than 150% of dirty blocks
3135 * or free blocks is less than watermark
3140 * Even if we don't switch but are nearing capacity,
3141 * start pushing delalloc when 1/2 of free blocks are dirty.
3143 if (free_blocks
< 2 * dirty_blocks
)
3144 writeback_inodes_sb_if_idle(sb
);
3149 static int ext4_da_write_begin(struct file
*file
, struct address_space
*mapping
,
3150 loff_t pos
, unsigned len
, unsigned flags
,
3151 struct page
**pagep
, void **fsdata
)
3153 int ret
, retries
= 0;
3156 struct inode
*inode
= mapping
->host
;
3159 index
= pos
>> PAGE_CACHE_SHIFT
;
3161 if (ext4_nonda_switch(inode
->i_sb
)) {
3162 *fsdata
= (void *)FALL_BACK_TO_NONDELALLOC
;
3163 return ext4_write_begin(file
, mapping
, pos
,
3164 len
, flags
, pagep
, fsdata
);
3166 *fsdata
= (void *)0;
3167 trace_ext4_da_write_begin(inode
, pos
, len
, flags
);
3170 * With delayed allocation, we don't log the i_disksize update
3171 * if there is delayed block allocation. But we still need
3172 * to journalling the i_disksize update if writes to the end
3173 * of file which has an already mapped buffer.
3175 handle
= ext4_journal_start(inode
, 1);
3176 if (IS_ERR(handle
)) {
3177 ret
= PTR_ERR(handle
);
3180 /* We cannot recurse into the filesystem as the transaction is already
3182 flags
|= AOP_FLAG_NOFS
;
3184 page
= grab_cache_page_write_begin(mapping
, index
, flags
);
3186 ext4_journal_stop(handle
);
3192 ret
= __block_write_begin(page
, pos
, len
, ext4_da_get_block_prep
);
3195 ext4_journal_stop(handle
);
3196 page_cache_release(page
);
3198 * block_write_begin may have instantiated a few blocks
3199 * outside i_size. Trim these off again. Don't need
3200 * i_size_read because we hold i_mutex.
3202 if (pos
+ len
> inode
->i_size
)
3203 ext4_truncate_failed_write(inode
);
3206 if (ret
== -ENOSPC
&& ext4_should_retry_alloc(inode
->i_sb
, &retries
))
3213 * Check if we should update i_disksize
3214 * when write to the end of file but not require block allocation
3216 static int ext4_da_should_update_i_disksize(struct page
*page
,
3217 unsigned long offset
)
3219 struct buffer_head
*bh
;
3220 struct inode
*inode
= page
->mapping
->host
;
3224 bh
= page_buffers(page
);
3225 idx
= offset
>> inode
->i_blkbits
;
3227 for (i
= 0; i
< idx
; i
++)
3228 bh
= bh
->b_this_page
;
3230 if (!buffer_mapped(bh
) || (buffer_delay(bh
)) || buffer_unwritten(bh
))
3235 static int ext4_da_write_end(struct file
*file
,
3236 struct address_space
*mapping
,
3237 loff_t pos
, unsigned len
, unsigned copied
,
3238 struct page
*page
, void *fsdata
)
3240 struct inode
*inode
= mapping
->host
;
3242 handle_t
*handle
= ext4_journal_current_handle();
3244 unsigned long start
, end
;
3245 int write_mode
= (int)(unsigned long)fsdata
;
3247 if (write_mode
== FALL_BACK_TO_NONDELALLOC
) {
3248 if (ext4_should_order_data(inode
)) {
3249 return ext4_ordered_write_end(file
, mapping
, pos
,
3250 len
, copied
, page
, fsdata
);
3251 } else if (ext4_should_writeback_data(inode
)) {
3252 return ext4_writeback_write_end(file
, mapping
, pos
,
3253 len
, copied
, page
, fsdata
);
3259 trace_ext4_da_write_end(inode
, pos
, len
, copied
);
3260 start
= pos
& (PAGE_CACHE_SIZE
- 1);
3261 end
= start
+ copied
- 1;
3264 * generic_write_end() will run mark_inode_dirty() if i_size
3265 * changes. So let's piggyback the i_disksize mark_inode_dirty
3269 new_i_size
= pos
+ copied
;
3270 if (new_i_size
> EXT4_I(inode
)->i_disksize
) {
3271 if (ext4_da_should_update_i_disksize(page
, end
)) {
3272 down_write(&EXT4_I(inode
)->i_data_sem
);
3273 if (new_i_size
> EXT4_I(inode
)->i_disksize
) {
3275 * Updating i_disksize when extending file
3276 * without needing block allocation
3278 if (ext4_should_order_data(inode
))
3279 ret
= ext4_jbd2_file_inode(handle
,
3282 EXT4_I(inode
)->i_disksize
= new_i_size
;
3284 up_write(&EXT4_I(inode
)->i_data_sem
);
3285 /* We need to mark inode dirty even if
3286 * new_i_size is less that inode->i_size
3287 * bu greater than i_disksize.(hint delalloc)
3289 ext4_mark_inode_dirty(handle
, inode
);
3292 ret2
= generic_write_end(file
, mapping
, pos
, len
, copied
,
3297 ret2
= ext4_journal_stop(handle
);
3301 return ret
? ret
: copied
;
3304 static void ext4_da_invalidatepage(struct page
*page
, unsigned long offset
)
3307 * Drop reserved blocks
3309 BUG_ON(!PageLocked(page
));
3310 if (!page_has_buffers(page
))
3313 ext4_da_page_release_reservation(page
, offset
);
3316 ext4_invalidatepage(page
, offset
);
3322 * Force all delayed allocation blocks to be allocated for a given inode.
3324 int ext4_alloc_da_blocks(struct inode
*inode
)
3326 trace_ext4_alloc_da_blocks(inode
);
3328 if (!EXT4_I(inode
)->i_reserved_data_blocks
&&
3329 !EXT4_I(inode
)->i_reserved_meta_blocks
)
3333 * We do something simple for now. The filemap_flush() will
3334 * also start triggering a write of the data blocks, which is
3335 * not strictly speaking necessary (and for users of
3336 * laptop_mode, not even desirable). However, to do otherwise
3337 * would require replicating code paths in:
3339 * ext4_da_writepages() ->
3340 * write_cache_pages() ---> (via passed in callback function)
3341 * __mpage_da_writepage() -->
3342 * mpage_add_bh_to_extent()
3343 * mpage_da_map_blocks()
3345 * The problem is that write_cache_pages(), located in
3346 * mm/page-writeback.c, marks pages clean in preparation for
3347 * doing I/O, which is not desirable if we're not planning on
3350 * We could call write_cache_pages(), and then redirty all of
3351 * the pages by calling redirty_page_for_writeback() but that
3352 * would be ugly in the extreme. So instead we would need to
3353 * replicate parts of the code in the above functions,
3354 * simplifying them becuase we wouldn't actually intend to
3355 * write out the pages, but rather only collect contiguous
3356 * logical block extents, call the multi-block allocator, and
3357 * then update the buffer heads with the block allocations.
3359 * For now, though, we'll cheat by calling filemap_flush(),
3360 * which will map the blocks, and start the I/O, but not
3361 * actually wait for the I/O to complete.
3363 return filemap_flush(inode
->i_mapping
);
3367 * bmap() is special. It gets used by applications such as lilo and by
3368 * the swapper to find the on-disk block of a specific piece of data.
3370 * Naturally, this is dangerous if the block concerned is still in the
3371 * journal. If somebody makes a swapfile on an ext4 data-journaling
3372 * filesystem and enables swap, then they may get a nasty shock when the
3373 * data getting swapped to that swapfile suddenly gets overwritten by
3374 * the original zero's written out previously to the journal and
3375 * awaiting writeback in the kernel's buffer cache.
3377 * So, if we see any bmap calls here on a modified, data-journaled file,
3378 * take extra steps to flush any blocks which might be in the cache.
3380 static sector_t
ext4_bmap(struct address_space
*mapping
, sector_t block
)
3382 struct inode
*inode
= mapping
->host
;
3386 if (mapping_tagged(mapping
, PAGECACHE_TAG_DIRTY
) &&
3387 test_opt(inode
->i_sb
, DELALLOC
)) {
3389 * With delalloc we want to sync the file
3390 * so that we can make sure we allocate
3393 filemap_write_and_wait(mapping
);
3396 if (EXT4_JOURNAL(inode
) &&
3397 ext4_test_inode_state(inode
, EXT4_STATE_JDATA
)) {
3399 * This is a REALLY heavyweight approach, but the use of
3400 * bmap on dirty files is expected to be extremely rare:
3401 * only if we run lilo or swapon on a freshly made file
3402 * do we expect this to happen.
3404 * (bmap requires CAP_SYS_RAWIO so this does not
3405 * represent an unprivileged user DOS attack --- we'd be
3406 * in trouble if mortal users could trigger this path at
3409 * NB. EXT4_STATE_JDATA is not set on files other than
3410 * regular files. If somebody wants to bmap a directory
3411 * or symlink and gets confused because the buffer
3412 * hasn't yet been flushed to disk, they deserve
3413 * everything they get.
3416 ext4_clear_inode_state(inode
, EXT4_STATE_JDATA
);
3417 journal
= EXT4_JOURNAL(inode
);
3418 jbd2_journal_lock_updates(journal
);
3419 err
= jbd2_journal_flush(journal
);
3420 jbd2_journal_unlock_updates(journal
);
3426 return generic_block_bmap(mapping
, block
, ext4_get_block
);
3429 static int ext4_readpage(struct file
*file
, struct page
*page
)
3431 return mpage_readpage(page
, ext4_get_block
);
3435 ext4_readpages(struct file
*file
, struct address_space
*mapping
,
3436 struct list_head
*pages
, unsigned nr_pages
)
3438 return mpage_readpages(mapping
, pages
, nr_pages
, ext4_get_block
);
3441 static void ext4_free_io_end(ext4_io_end_t
*io
)
3450 static void ext4_invalidatepage_free_endio(struct page
*page
, unsigned long offset
)
3452 struct buffer_head
*head
, *bh
;
3453 unsigned int curr_off
= 0;
3455 if (!page_has_buffers(page
))
3457 head
= bh
= page_buffers(page
);
3459 if (offset
<= curr_off
&& test_clear_buffer_uninit(bh
)
3461 ext4_free_io_end(bh
->b_private
);
3462 bh
->b_private
= NULL
;
3463 bh
->b_end_io
= NULL
;
3465 curr_off
= curr_off
+ bh
->b_size
;
3466 bh
= bh
->b_this_page
;
3467 } while (bh
!= head
);
3470 static void ext4_invalidatepage(struct page
*page
, unsigned long offset
)
3472 journal_t
*journal
= EXT4_JOURNAL(page
->mapping
->host
);
3475 * free any io_end structure allocated for buffers to be discarded
3477 if (ext4_should_dioread_nolock(page
->mapping
->host
))
3478 ext4_invalidatepage_free_endio(page
, offset
);
3480 * If it's a full truncate we just forget about the pending dirtying
3483 ClearPageChecked(page
);
3486 jbd2_journal_invalidatepage(journal
, page
, offset
);
3488 block_invalidatepage(page
, offset
);
3491 static int ext4_releasepage(struct page
*page
, gfp_t wait
)
3493 journal_t
*journal
= EXT4_JOURNAL(page
->mapping
->host
);
3495 WARN_ON(PageChecked(page
));
3496 if (!page_has_buffers(page
))
3499 return jbd2_journal_try_to_free_buffers(journal
, page
, wait
);
3501 return try_to_free_buffers(page
);
3505 * O_DIRECT for ext3 (or indirect map) based files
3507 * If the O_DIRECT write will extend the file then add this inode to the
3508 * orphan list. So recovery will truncate it back to the original size
3509 * if the machine crashes during the write.
3511 * If the O_DIRECT write is intantiating holes inside i_size and the machine
3512 * crashes then stale disk data _may_ be exposed inside the file. But current
3513 * VFS code falls back into buffered path in that case so we are safe.
3515 static ssize_t
ext4_ind_direct_IO(int rw
, struct kiocb
*iocb
,
3516 const struct iovec
*iov
, loff_t offset
,
3517 unsigned long nr_segs
)
3519 struct file
*file
= iocb
->ki_filp
;
3520 struct inode
*inode
= file
->f_mapping
->host
;
3521 struct ext4_inode_info
*ei
= EXT4_I(inode
);
3525 size_t count
= iov_length(iov
, nr_segs
);
3529 loff_t final_size
= offset
+ count
;
3531 if (final_size
> inode
->i_size
) {
3532 /* Credits for sb + inode write */
3533 handle
= ext4_journal_start(inode
, 2);
3534 if (IS_ERR(handle
)) {
3535 ret
= PTR_ERR(handle
);
3538 ret
= ext4_orphan_add(handle
, inode
);
3540 ext4_journal_stop(handle
);
3544 ei
->i_disksize
= inode
->i_size
;
3545 ext4_journal_stop(handle
);
3550 if (rw
== READ
&& ext4_should_dioread_nolock(inode
))
3551 ret
= __blockdev_direct_IO(rw
, iocb
, inode
,
3552 inode
->i_sb
->s_bdev
, iov
,
3554 ext4_get_block
, NULL
, NULL
, 0);
3556 ret
= blockdev_direct_IO(rw
, iocb
, inode
,
3557 inode
->i_sb
->s_bdev
, iov
,
3559 ext4_get_block
, NULL
);
3561 if (unlikely((rw
& WRITE
) && ret
< 0)) {
3562 loff_t isize
= i_size_read(inode
);
3563 loff_t end
= offset
+ iov_length(iov
, nr_segs
);
3566 vmtruncate(inode
, isize
);
3569 if (ret
== -ENOSPC
&& ext4_should_retry_alloc(inode
->i_sb
, &retries
))
3575 /* Credits for sb + inode write */
3576 handle
= ext4_journal_start(inode
, 2);
3577 if (IS_ERR(handle
)) {
3578 /* This is really bad luck. We've written the data
3579 * but cannot extend i_size. Bail out and pretend
3580 * the write failed... */
3581 ret
= PTR_ERR(handle
);
3583 ext4_orphan_del(NULL
, inode
);
3588 ext4_orphan_del(handle
, inode
);
3590 loff_t end
= offset
+ ret
;
3591 if (end
> inode
->i_size
) {
3592 ei
->i_disksize
= end
;
3593 i_size_write(inode
, end
);
3595 * We're going to return a positive `ret'
3596 * here due to non-zero-length I/O, so there's
3597 * no way of reporting error returns from
3598 * ext4_mark_inode_dirty() to userspace. So
3601 ext4_mark_inode_dirty(handle
, inode
);
3604 err
= ext4_journal_stop(handle
);
3613 * ext4_get_block used when preparing for a DIO write or buffer write.
3614 * We allocate an uinitialized extent if blocks haven't been allocated.
3615 * The extent will be converted to initialized after the IO is complete.
3617 static int ext4_get_block_write(struct inode
*inode
, sector_t iblock
,
3618 struct buffer_head
*bh_result
, int create
)
3620 ext4_debug("ext4_get_block_write: inode %lu, create flag %d\n",
3621 inode
->i_ino
, create
);
3622 return _ext4_get_block(inode
, iblock
, bh_result
,
3623 EXT4_GET_BLOCKS_IO_CREATE_EXT
);
3626 static void dump_completed_IO(struct inode
* inode
)
3629 struct list_head
*cur
, *before
, *after
;
3630 ext4_io_end_t
*io
, *io0
, *io1
;
3631 unsigned long flags
;
3633 if (list_empty(&EXT4_I(inode
)->i_completed_io_list
)){
3634 ext4_debug("inode %lu completed_io list is empty\n", inode
->i_ino
);
3638 ext4_debug("Dump inode %lu completed_io list \n", inode
->i_ino
);
3639 spin_lock_irqsave(&EXT4_I(inode
)->i_completed_io_lock
, flags
);
3640 list_for_each_entry(io
, &EXT4_I(inode
)->i_completed_io_list
, list
){
3643 io0
= container_of(before
, ext4_io_end_t
, list
);
3645 io1
= container_of(after
, ext4_io_end_t
, list
);
3647 ext4_debug("io 0x%p from inode %lu,prev 0x%p,next 0x%p\n",
3648 io
, inode
->i_ino
, io0
, io1
);
3650 spin_unlock_irqrestore(&EXT4_I(inode
)->i_completed_io_lock
, flags
);
3655 * check a range of space and convert unwritten extents to written.
3657 static int ext4_end_io_nolock(ext4_io_end_t
*io
)
3659 struct inode
*inode
= io
->inode
;
3660 loff_t offset
= io
->offset
;
3661 ssize_t size
= io
->size
;
3664 ext4_debug("ext4_end_io_nolock: io 0x%p from inode %lu,list->next 0x%p,"
3665 "list->prev 0x%p\n",
3666 io
, inode
->i_ino
, io
->list
.next
, io
->list
.prev
);
3668 if (list_empty(&io
->list
))
3671 if (io
->flag
!= EXT4_IO_UNWRITTEN
)
3674 ret
= ext4_convert_unwritten_extents(inode
, offset
, size
);
3676 printk(KERN_EMERG
"%s: failed to convert unwritten"
3677 "extents to written extents, error is %d"
3678 " io is still on inode %lu aio dio list\n",
3679 __func__
, ret
, inode
->i_ino
);
3684 aio_complete(io
->iocb
, io
->result
, 0);
3685 /* clear the DIO AIO unwritten flag */
3691 * work on completed aio dio IO, to convert unwritten extents to extents
3693 static void ext4_end_io_work(struct work_struct
*work
)
3695 ext4_io_end_t
*io
= container_of(work
, ext4_io_end_t
, work
);
3696 struct inode
*inode
= io
->inode
;
3697 struct ext4_inode_info
*ei
= EXT4_I(inode
);
3698 unsigned long flags
;
3701 mutex_lock(&inode
->i_mutex
);
3702 ret
= ext4_end_io_nolock(io
);
3704 mutex_unlock(&inode
->i_mutex
);
3708 spin_lock_irqsave(&ei
->i_completed_io_lock
, flags
);
3709 if (!list_empty(&io
->list
))
3710 list_del_init(&io
->list
);
3711 spin_unlock_irqrestore(&ei
->i_completed_io_lock
, flags
);
3712 mutex_unlock(&inode
->i_mutex
);
3713 ext4_free_io_end(io
);
3717 * This function is called from ext4_sync_file().
3719 * When IO is completed, the work to convert unwritten extents to
3720 * written is queued on workqueue but may not get immediately
3721 * scheduled. When fsync is called, we need to ensure the
3722 * conversion is complete before fsync returns.
3723 * The inode keeps track of a list of pending/completed IO that
3724 * might needs to do the conversion. This function walks through
3725 * the list and convert the related unwritten extents for completed IO
3727 * The function return the number of pending IOs on success.
3729 int flush_completed_IO(struct inode
*inode
)
3732 struct ext4_inode_info
*ei
= EXT4_I(inode
);
3733 unsigned long flags
;
3737 if (list_empty(&ei
->i_completed_io_list
))
3740 dump_completed_IO(inode
);
3741 spin_lock_irqsave(&ei
->i_completed_io_lock
, flags
);
3742 while (!list_empty(&ei
->i_completed_io_list
)){
3743 io
= list_entry(ei
->i_completed_io_list
.next
,
3744 ext4_io_end_t
, list
);
3746 * Calling ext4_end_io_nolock() to convert completed
3749 * When ext4_sync_file() is called, run_queue() may already
3750 * about to flush the work corresponding to this io structure.
3751 * It will be upset if it founds the io structure related
3752 * to the work-to-be schedule is freed.
3754 * Thus we need to keep the io structure still valid here after
3755 * convertion finished. The io structure has a flag to
3756 * avoid double converting from both fsync and background work
3759 spin_unlock_irqrestore(&ei
->i_completed_io_lock
, flags
);
3760 ret
= ext4_end_io_nolock(io
);
3761 spin_lock_irqsave(&ei
->i_completed_io_lock
, flags
);
3765 list_del_init(&io
->list
);
3767 spin_unlock_irqrestore(&ei
->i_completed_io_lock
, flags
);
3768 return (ret2
< 0) ? ret2
: 0;
3771 static ext4_io_end_t
*ext4_init_io_end (struct inode
*inode
, gfp_t flags
)
3773 ext4_io_end_t
*io
= NULL
;
3775 io
= kmalloc(sizeof(*io
), flags
);
3786 INIT_WORK(&io
->work
, ext4_end_io_work
);
3787 INIT_LIST_HEAD(&io
->list
);
3793 static void ext4_end_io_dio(struct kiocb
*iocb
, loff_t offset
,
3794 ssize_t size
, void *private, int ret
,
3797 ext4_io_end_t
*io_end
= iocb
->private;
3798 struct workqueue_struct
*wq
;
3799 unsigned long flags
;
3800 struct ext4_inode_info
*ei
;
3802 /* if not async direct IO or dio with 0 bytes write, just return */
3803 if (!io_end
|| !size
)
3806 ext_debug("ext4_end_io_dio(): io_end 0x%p"
3807 "for inode %lu, iocb 0x%p, offset %llu, size %llu\n",
3808 iocb
->private, io_end
->inode
->i_ino
, iocb
, offset
,
3811 /* if not aio dio with unwritten extents, just free io and return */
3812 if (io_end
->flag
!= EXT4_IO_UNWRITTEN
){
3813 ext4_free_io_end(io_end
);
3814 iocb
->private = NULL
;
3817 aio_complete(iocb
, ret
, 0);
3821 io_end
->offset
= offset
;
3822 io_end
->size
= size
;
3824 io_end
->iocb
= iocb
;
3825 io_end
->result
= ret
;
3827 wq
= EXT4_SB(io_end
->inode
->i_sb
)->dio_unwritten_wq
;
3829 /* queue the work to convert unwritten extents to written */
3830 queue_work(wq
, &io_end
->work
);
3832 /* Add the io_end to per-inode completed aio dio list*/
3833 ei
= EXT4_I(io_end
->inode
);
3834 spin_lock_irqsave(&ei
->i_completed_io_lock
, flags
);
3835 list_add_tail(&io_end
->list
, &ei
->i_completed_io_list
);
3836 spin_unlock_irqrestore(&ei
->i_completed_io_lock
, flags
);
3837 iocb
->private = NULL
;
3840 static void ext4_end_io_buffer_write(struct buffer_head
*bh
, int uptodate
)
3842 ext4_io_end_t
*io_end
= bh
->b_private
;
3843 struct workqueue_struct
*wq
;
3844 struct inode
*inode
;
3845 unsigned long flags
;
3847 if (!test_clear_buffer_uninit(bh
) || !io_end
)
3850 if (!(io_end
->inode
->i_sb
->s_flags
& MS_ACTIVE
)) {
3851 printk("sb umounted, discard end_io request for inode %lu\n",
3852 io_end
->inode
->i_ino
);
3853 ext4_free_io_end(io_end
);
3857 io_end
->flag
= EXT4_IO_UNWRITTEN
;
3858 inode
= io_end
->inode
;
3860 /* Add the io_end to per-inode completed io list*/
3861 spin_lock_irqsave(&EXT4_I(inode
)->i_completed_io_lock
, flags
);
3862 list_add_tail(&io_end
->list
, &EXT4_I(inode
)->i_completed_io_list
);
3863 spin_unlock_irqrestore(&EXT4_I(inode
)->i_completed_io_lock
, flags
);
3865 wq
= EXT4_SB(inode
->i_sb
)->dio_unwritten_wq
;
3866 /* queue the work to convert unwritten extents to written */
3867 queue_work(wq
, &io_end
->work
);
3869 bh
->b_private
= NULL
;
3870 bh
->b_end_io
= NULL
;
3871 clear_buffer_uninit(bh
);
3872 end_buffer_async_write(bh
, uptodate
);
3875 static int ext4_set_bh_endio(struct buffer_head
*bh
, struct inode
*inode
)
3877 ext4_io_end_t
*io_end
;
3878 struct page
*page
= bh
->b_page
;
3879 loff_t offset
= (sector_t
)page
->index
<< PAGE_CACHE_SHIFT
;
3880 size_t size
= bh
->b_size
;
3883 io_end
= ext4_init_io_end(inode
, GFP_ATOMIC
);
3885 if (printk_ratelimit())
3886 printk(KERN_WARNING
"%s: allocation fail\n", __func__
);
3890 io_end
->offset
= offset
;
3891 io_end
->size
= size
;
3893 * We need to hold a reference to the page to make sure it
3894 * doesn't get evicted before ext4_end_io_work() has a chance
3895 * to convert the extent from written to unwritten.
3897 io_end
->page
= page
;
3898 get_page(io_end
->page
);
3900 bh
->b_private
= io_end
;
3901 bh
->b_end_io
= ext4_end_io_buffer_write
;
3906 * For ext4 extent files, ext4 will do direct-io write to holes,
3907 * preallocated extents, and those write extend the file, no need to
3908 * fall back to buffered IO.
3910 * For holes, we fallocate those blocks, mark them as unintialized
3911 * If those blocks were preallocated, we mark sure they are splited, but
3912 * still keep the range to write as unintialized.
3914 * The unwrritten extents will be converted to written when DIO is completed.
3915 * For async direct IO, since the IO may still pending when return, we
3916 * set up an end_io call back function, which will do the convertion
3917 * when async direct IO completed.
3919 * If the O_DIRECT write will extend the file then add this inode to the
3920 * orphan list. So recovery will truncate it back to the original size
3921 * if the machine crashes during the write.
3924 static ssize_t
ext4_ext_direct_IO(int rw
, struct kiocb
*iocb
,
3925 const struct iovec
*iov
, loff_t offset
,
3926 unsigned long nr_segs
)
3928 struct file
*file
= iocb
->ki_filp
;
3929 struct inode
*inode
= file
->f_mapping
->host
;
3931 size_t count
= iov_length(iov
, nr_segs
);
3933 loff_t final_size
= offset
+ count
;
3934 if (rw
== WRITE
&& final_size
<= inode
->i_size
) {
3936 * We could direct write to holes and fallocate.
3938 * Allocated blocks to fill the hole are marked as uninitialized
3939 * to prevent paralel buffered read to expose the stale data
3940 * before DIO complete the data IO.
3942 * As to previously fallocated extents, ext4 get_block
3943 * will just simply mark the buffer mapped but still
3944 * keep the extents uninitialized.
3946 * for non AIO case, we will convert those unwritten extents
3947 * to written after return back from blockdev_direct_IO.
3949 * for async DIO, the conversion needs to be defered when
3950 * the IO is completed. The ext4 end_io callback function
3951 * will be called to take care of the conversion work.
3952 * Here for async case, we allocate an io_end structure to
3955 iocb
->private = NULL
;
3956 EXT4_I(inode
)->cur_aio_dio
= NULL
;
3957 if (!is_sync_kiocb(iocb
)) {
3958 iocb
->private = ext4_init_io_end(inode
, GFP_NOFS
);
3962 * we save the io structure for current async
3963 * direct IO, so that later ext4_map_blocks()
3964 * could flag the io structure whether there
3965 * is a unwritten extents needs to be converted
3966 * when IO is completed.
3968 EXT4_I(inode
)->cur_aio_dio
= iocb
->private;
3971 ret
= blockdev_direct_IO(rw
, iocb
, inode
,
3972 inode
->i_sb
->s_bdev
, iov
,
3974 ext4_get_block_write
,
3977 EXT4_I(inode
)->cur_aio_dio
= NULL
;
3979 * The io_end structure takes a reference to the inode,
3980 * that structure needs to be destroyed and the
3981 * reference to the inode need to be dropped, when IO is
3982 * complete, even with 0 byte write, or failed.
3984 * In the successful AIO DIO case, the io_end structure will be
3985 * desctroyed and the reference to the inode will be dropped
3986 * after the end_io call back function is called.
3988 * In the case there is 0 byte write, or error case, since
3989 * VFS direct IO won't invoke the end_io call back function,
3990 * we need to free the end_io structure here.
3992 if (ret
!= -EIOCBQUEUED
&& ret
<= 0 && iocb
->private) {
3993 ext4_free_io_end(iocb
->private);
3994 iocb
->private = NULL
;
3995 } else if (ret
> 0 && ext4_test_inode_state(inode
,
3996 EXT4_STATE_DIO_UNWRITTEN
)) {
3999 * for non AIO case, since the IO is already
4000 * completed, we could do the convertion right here
4002 err
= ext4_convert_unwritten_extents(inode
,
4006 ext4_clear_inode_state(inode
, EXT4_STATE_DIO_UNWRITTEN
);
4011 /* for write the the end of file case, we fall back to old way */
4012 return ext4_ind_direct_IO(rw
, iocb
, iov
, offset
, nr_segs
);
4015 static ssize_t
ext4_direct_IO(int rw
, struct kiocb
*iocb
,
4016 const struct iovec
*iov
, loff_t offset
,
4017 unsigned long nr_segs
)
4019 struct file
*file
= iocb
->ki_filp
;
4020 struct inode
*inode
= file
->f_mapping
->host
;
4022 if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
))
4023 return ext4_ext_direct_IO(rw
, iocb
, iov
, offset
, nr_segs
);
4025 return ext4_ind_direct_IO(rw
, iocb
, iov
, offset
, nr_segs
);
4029 * Pages can be marked dirty completely asynchronously from ext4's journalling
4030 * activity. By filemap_sync_pte(), try_to_unmap_one(), etc. We cannot do
4031 * much here because ->set_page_dirty is called under VFS locks. The page is
4032 * not necessarily locked.
4034 * We cannot just dirty the page and leave attached buffers clean, because the
4035 * buffers' dirty state is "definitive". We cannot just set the buffers dirty
4036 * or jbddirty because all the journalling code will explode.
4038 * So what we do is to mark the page "pending dirty" and next time writepage
4039 * is called, propagate that into the buffers appropriately.
4041 static int ext4_journalled_set_page_dirty(struct page
*page
)
4043 SetPageChecked(page
);
4044 return __set_page_dirty_nobuffers(page
);
4047 static const struct address_space_operations ext4_ordered_aops
= {
4048 .readpage
= ext4_readpage
,
4049 .readpages
= ext4_readpages
,
4050 .writepage
= ext4_writepage
,
4051 .sync_page
= block_sync_page
,
4052 .write_begin
= ext4_write_begin
,
4053 .write_end
= ext4_ordered_write_end
,
4055 .invalidatepage
= ext4_invalidatepage
,
4056 .releasepage
= ext4_releasepage
,
4057 .direct_IO
= ext4_direct_IO
,
4058 .migratepage
= buffer_migrate_page
,
4059 .is_partially_uptodate
= block_is_partially_uptodate
,
4060 .error_remove_page
= generic_error_remove_page
,
4063 static const struct address_space_operations ext4_writeback_aops
= {
4064 .readpage
= ext4_readpage
,
4065 .readpages
= ext4_readpages
,
4066 .writepage
= ext4_writepage
,
4067 .sync_page
= block_sync_page
,
4068 .write_begin
= ext4_write_begin
,
4069 .write_end
= ext4_writeback_write_end
,
4071 .invalidatepage
= ext4_invalidatepage
,
4072 .releasepage
= ext4_releasepage
,
4073 .direct_IO
= ext4_direct_IO
,
4074 .migratepage
= buffer_migrate_page
,
4075 .is_partially_uptodate
= block_is_partially_uptodate
,
4076 .error_remove_page
= generic_error_remove_page
,
4079 static const struct address_space_operations ext4_journalled_aops
= {
4080 .readpage
= ext4_readpage
,
4081 .readpages
= ext4_readpages
,
4082 .writepage
= ext4_writepage
,
4083 .sync_page
= block_sync_page
,
4084 .write_begin
= ext4_write_begin
,
4085 .write_end
= ext4_journalled_write_end
,
4086 .set_page_dirty
= ext4_journalled_set_page_dirty
,
4088 .invalidatepage
= ext4_invalidatepage
,
4089 .releasepage
= ext4_releasepage
,
4090 .is_partially_uptodate
= block_is_partially_uptodate
,
4091 .error_remove_page
= generic_error_remove_page
,
4094 static const struct address_space_operations ext4_da_aops
= {
4095 .readpage
= ext4_readpage
,
4096 .readpages
= ext4_readpages
,
4097 .writepage
= ext4_writepage
,
4098 .writepages
= ext4_da_writepages
,
4099 .sync_page
= block_sync_page
,
4100 .write_begin
= ext4_da_write_begin
,
4101 .write_end
= ext4_da_write_end
,
4103 .invalidatepage
= ext4_da_invalidatepage
,
4104 .releasepage
= ext4_releasepage
,
4105 .direct_IO
= ext4_direct_IO
,
4106 .migratepage
= buffer_migrate_page
,
4107 .is_partially_uptodate
= block_is_partially_uptodate
,
4108 .error_remove_page
= generic_error_remove_page
,
4111 void ext4_set_aops(struct inode
*inode
)
4113 if (ext4_should_order_data(inode
) &&
4114 test_opt(inode
->i_sb
, DELALLOC
))
4115 inode
->i_mapping
->a_ops
= &ext4_da_aops
;
4116 else if (ext4_should_order_data(inode
))
4117 inode
->i_mapping
->a_ops
= &ext4_ordered_aops
;
4118 else if (ext4_should_writeback_data(inode
) &&
4119 test_opt(inode
->i_sb
, DELALLOC
))
4120 inode
->i_mapping
->a_ops
= &ext4_da_aops
;
4121 else if (ext4_should_writeback_data(inode
))
4122 inode
->i_mapping
->a_ops
= &ext4_writeback_aops
;
4124 inode
->i_mapping
->a_ops
= &ext4_journalled_aops
;
4128 * ext4_block_truncate_page() zeroes out a mapping from file offset `from'
4129 * up to the end of the block which corresponds to `from'.
4130 * This required during truncate. We need to physically zero the tail end
4131 * of that block so it doesn't yield old data if the file is later grown.
4133 int ext4_block_truncate_page(handle_t
*handle
,
4134 struct address_space
*mapping
, loff_t from
)
4136 ext4_fsblk_t index
= from
>> PAGE_CACHE_SHIFT
;
4137 unsigned offset
= from
& (PAGE_CACHE_SIZE
-1);
4138 unsigned blocksize
, length
, pos
;
4140 struct inode
*inode
= mapping
->host
;
4141 struct buffer_head
*bh
;
4145 page
= find_or_create_page(mapping
, from
>> PAGE_CACHE_SHIFT
,
4146 mapping_gfp_mask(mapping
) & ~__GFP_FS
);
4150 blocksize
= inode
->i_sb
->s_blocksize
;
4151 length
= blocksize
- (offset
& (blocksize
- 1));
4152 iblock
= index
<< (PAGE_CACHE_SHIFT
- inode
->i_sb
->s_blocksize_bits
);
4154 if (!page_has_buffers(page
))
4155 create_empty_buffers(page
, blocksize
, 0);
4157 /* Find the buffer that contains "offset" */
4158 bh
= page_buffers(page
);
4160 while (offset
>= pos
) {
4161 bh
= bh
->b_this_page
;
4167 if (buffer_freed(bh
)) {
4168 BUFFER_TRACE(bh
, "freed: skip");
4172 if (!buffer_mapped(bh
)) {
4173 BUFFER_TRACE(bh
, "unmapped");
4174 ext4_get_block(inode
, iblock
, bh
, 0);
4175 /* unmapped? It's a hole - nothing to do */
4176 if (!buffer_mapped(bh
)) {
4177 BUFFER_TRACE(bh
, "still unmapped");
4182 /* Ok, it's mapped. Make sure it's up-to-date */
4183 if (PageUptodate(page
))
4184 set_buffer_uptodate(bh
);
4186 if (!buffer_uptodate(bh
)) {
4188 ll_rw_block(READ
, 1, &bh
);
4190 /* Uhhuh. Read error. Complain and punt. */
4191 if (!buffer_uptodate(bh
))
4195 if (ext4_should_journal_data(inode
)) {
4196 BUFFER_TRACE(bh
, "get write access");
4197 err
= ext4_journal_get_write_access(handle
, bh
);
4202 zero_user(page
, offset
, length
);
4204 BUFFER_TRACE(bh
, "zeroed end of block");
4207 if (ext4_should_journal_data(inode
)) {
4208 err
= ext4_handle_dirty_metadata(handle
, inode
, bh
);
4210 if (ext4_should_order_data(inode
))
4211 err
= ext4_jbd2_file_inode(handle
, inode
);
4212 mark_buffer_dirty(bh
);
4217 page_cache_release(page
);
4222 * Probably it should be a library function... search for first non-zero word
4223 * or memcmp with zero_page, whatever is better for particular architecture.
4226 static inline int all_zeroes(__le32
*p
, __le32
*q
)
4235 * ext4_find_shared - find the indirect blocks for partial truncation.
4236 * @inode: inode in question
4237 * @depth: depth of the affected branch
4238 * @offsets: offsets of pointers in that branch (see ext4_block_to_path)
4239 * @chain: place to store the pointers to partial indirect blocks
4240 * @top: place to the (detached) top of branch
4242 * This is a helper function used by ext4_truncate().
4244 * When we do truncate() we may have to clean the ends of several
4245 * indirect blocks but leave the blocks themselves alive. Block is
4246 * partially truncated if some data below the new i_size is refered
4247 * from it (and it is on the path to the first completely truncated
4248 * data block, indeed). We have to free the top of that path along
4249 * with everything to the right of the path. Since no allocation
4250 * past the truncation point is possible until ext4_truncate()
4251 * finishes, we may safely do the latter, but top of branch may
4252 * require special attention - pageout below the truncation point
4253 * might try to populate it.
4255 * We atomically detach the top of branch from the tree, store the
4256 * block number of its root in *@top, pointers to buffer_heads of
4257 * partially truncated blocks - in @chain[].bh and pointers to
4258 * their last elements that should not be removed - in
4259 * @chain[].p. Return value is the pointer to last filled element
4262 * The work left to caller to do the actual freeing of subtrees:
4263 * a) free the subtree starting from *@top
4264 * b) free the subtrees whose roots are stored in
4265 * (@chain[i].p+1 .. end of @chain[i].bh->b_data)
4266 * c) free the subtrees growing from the inode past the @chain[0].
4267 * (no partially truncated stuff there). */
4269 static Indirect
*ext4_find_shared(struct inode
*inode
, int depth
,
4270 ext4_lblk_t offsets
[4], Indirect chain
[4],
4273 Indirect
*partial
, *p
;
4277 /* Make k index the deepest non-null offset + 1 */
4278 for (k
= depth
; k
> 1 && !offsets
[k
-1]; k
--)
4280 partial
= ext4_get_branch(inode
, k
, offsets
, chain
, &err
);
4281 /* Writer: pointers */
4283 partial
= chain
+ k
-1;
4285 * If the branch acquired continuation since we've looked at it -
4286 * fine, it should all survive and (new) top doesn't belong to us.
4288 if (!partial
->key
&& *partial
->p
)
4291 for (p
= partial
; (p
> chain
) && all_zeroes((__le32
*) p
->bh
->b_data
, p
->p
); p
--)
4294 * OK, we've found the last block that must survive. The rest of our
4295 * branch should be detached before unlocking. However, if that rest
4296 * of branch is all ours and does not grow immediately from the inode
4297 * it's easier to cheat and just decrement partial->p.
4299 if (p
== chain
+ k
- 1 && p
> chain
) {
4303 /* Nope, don't do this in ext4. Must leave the tree intact */
4307 while (partial
> p
) {
4308 brelse(partial
->bh
);
4316 * Zero a number of block pointers in either an inode or an indirect block.
4317 * If we restart the transaction we must again get write access to the
4318 * indirect block for further modification.
4320 * We release `count' blocks on disk, but (last - first) may be greater
4321 * than `count' because there can be holes in there.
4323 static int ext4_clear_blocks(handle_t
*handle
, struct inode
*inode
,
4324 struct buffer_head
*bh
,
4325 ext4_fsblk_t block_to_free
,
4326 unsigned long count
, __le32
*first
,
4330 int flags
= EXT4_FREE_BLOCKS_FORGET
| EXT4_FREE_BLOCKS_VALIDATED
;
4332 if (S_ISDIR(inode
->i_mode
) || S_ISLNK(inode
->i_mode
))
4333 flags
|= EXT4_FREE_BLOCKS_METADATA
;
4335 if (!ext4_data_block_valid(EXT4_SB(inode
->i_sb
), block_to_free
,
4337 EXT4_ERROR_INODE(inode
, "attempt to clear invalid "
4338 "blocks %llu len %lu",
4339 (unsigned long long) block_to_free
, count
);
4343 if (try_to_extend_transaction(handle
, inode
)) {
4345 BUFFER_TRACE(bh
, "call ext4_handle_dirty_metadata");
4346 ext4_handle_dirty_metadata(handle
, inode
, bh
);
4348 ext4_mark_inode_dirty(handle
, inode
);
4349 ext4_truncate_restart_trans(handle
, inode
,
4350 blocks_for_truncate(inode
));
4352 BUFFER_TRACE(bh
, "retaking write access");
4353 ext4_journal_get_write_access(handle
, bh
);
4357 for (p
= first
; p
< last
; p
++)
4360 ext4_free_blocks(handle
, inode
, 0, block_to_free
, count
, flags
);
4365 * ext4_free_data - free a list of data blocks
4366 * @handle: handle for this transaction
4367 * @inode: inode we are dealing with
4368 * @this_bh: indirect buffer_head which contains *@first and *@last
4369 * @first: array of block numbers
4370 * @last: points immediately past the end of array
4372 * We are freeing all blocks refered from that array (numbers are stored as
4373 * little-endian 32-bit) and updating @inode->i_blocks appropriately.
4375 * We accumulate contiguous runs of blocks to free. Conveniently, if these
4376 * blocks are contiguous then releasing them at one time will only affect one
4377 * or two bitmap blocks (+ group descriptor(s) and superblock) and we won't
4378 * actually use a lot of journal space.
4380 * @this_bh will be %NULL if @first and @last point into the inode's direct
4383 static void ext4_free_data(handle_t
*handle
, struct inode
*inode
,
4384 struct buffer_head
*this_bh
,
4385 __le32
*first
, __le32
*last
)
4387 ext4_fsblk_t block_to_free
= 0; /* Starting block # of a run */
4388 unsigned long count
= 0; /* Number of blocks in the run */
4389 __le32
*block_to_free_p
= NULL
; /* Pointer into inode/ind
4392 ext4_fsblk_t nr
; /* Current block # */
4393 __le32
*p
; /* Pointer into inode/ind
4394 for current block */
4397 if (this_bh
) { /* For indirect block */
4398 BUFFER_TRACE(this_bh
, "get_write_access");
4399 err
= ext4_journal_get_write_access(handle
, this_bh
);
4400 /* Important: if we can't update the indirect pointers
4401 * to the blocks, we can't free them. */
4406 for (p
= first
; p
< last
; p
++) {
4407 nr
= le32_to_cpu(*p
);
4409 /* accumulate blocks to free if they're contiguous */
4412 block_to_free_p
= p
;
4414 } else if (nr
== block_to_free
+ count
) {
4417 if (ext4_clear_blocks(handle
, inode
, this_bh
,
4418 block_to_free
, count
,
4419 block_to_free_p
, p
))
4422 block_to_free_p
= p
;
4429 ext4_clear_blocks(handle
, inode
, this_bh
, block_to_free
,
4430 count
, block_to_free_p
, p
);
4433 BUFFER_TRACE(this_bh
, "call ext4_handle_dirty_metadata");
4436 * The buffer head should have an attached journal head at this
4437 * point. However, if the data is corrupted and an indirect
4438 * block pointed to itself, it would have been detached when
4439 * the block was cleared. Check for this instead of OOPSing.
4441 if ((EXT4_JOURNAL(inode
) == NULL
) || bh2jh(this_bh
))
4442 ext4_handle_dirty_metadata(handle
, inode
, this_bh
);
4444 EXT4_ERROR_INODE(inode
,
4445 "circular indirect block detected at "
4447 (unsigned long long) this_bh
->b_blocknr
);
4452 * ext4_free_branches - free an array of branches
4453 * @handle: JBD handle for this transaction
4454 * @inode: inode we are dealing with
4455 * @parent_bh: the buffer_head which contains *@first and *@last
4456 * @first: array of block numbers
4457 * @last: pointer immediately past the end of array
4458 * @depth: depth of the branches to free
4460 * We are freeing all blocks refered from these branches (numbers are
4461 * stored as little-endian 32-bit) and updating @inode->i_blocks
4464 static void ext4_free_branches(handle_t
*handle
, struct inode
*inode
,
4465 struct buffer_head
*parent_bh
,
4466 __le32
*first
, __le32
*last
, int depth
)
4471 if (ext4_handle_is_aborted(handle
))
4475 struct buffer_head
*bh
;
4476 int addr_per_block
= EXT4_ADDR_PER_BLOCK(inode
->i_sb
);
4478 while (--p
>= first
) {
4479 nr
= le32_to_cpu(*p
);
4481 continue; /* A hole */
4483 if (!ext4_data_block_valid(EXT4_SB(inode
->i_sb
),
4485 EXT4_ERROR_INODE(inode
,
4486 "invalid indirect mapped "
4487 "block %lu (level %d)",
4488 (unsigned long) nr
, depth
);
4492 /* Go read the buffer for the next level down */
4493 bh
= sb_bread(inode
->i_sb
, nr
);
4496 * A read failure? Report error and clear slot
4500 EXT4_ERROR_INODE_BLOCK(inode
, nr
,
4505 /* This zaps the entire block. Bottom up. */
4506 BUFFER_TRACE(bh
, "free child branches");
4507 ext4_free_branches(handle
, inode
, bh
,
4508 (__le32
*) bh
->b_data
,
4509 (__le32
*) bh
->b_data
+ addr_per_block
,
4514 * Everything below this this pointer has been
4515 * released. Now let this top-of-subtree go.
4517 * We want the freeing of this indirect block to be
4518 * atomic in the journal with the updating of the
4519 * bitmap block which owns it. So make some room in
4522 * We zero the parent pointer *after* freeing its
4523 * pointee in the bitmaps, so if extend_transaction()
4524 * for some reason fails to put the bitmap changes and
4525 * the release into the same transaction, recovery
4526 * will merely complain about releasing a free block,
4527 * rather than leaking blocks.
4529 if (ext4_handle_is_aborted(handle
))
4531 if (try_to_extend_transaction(handle
, inode
)) {
4532 ext4_mark_inode_dirty(handle
, inode
);
4533 ext4_truncate_restart_trans(handle
, inode
,
4534 blocks_for_truncate(inode
));
4538 * The forget flag here is critical because if
4539 * we are journaling (and not doing data
4540 * journaling), we have to make sure a revoke
4541 * record is written to prevent the journal
4542 * replay from overwriting the (former)
4543 * indirect block if it gets reallocated as a
4544 * data block. This must happen in the same
4545 * transaction where the data blocks are
4548 ext4_free_blocks(handle
, inode
, 0, nr
, 1,
4549 EXT4_FREE_BLOCKS_METADATA
|
4550 EXT4_FREE_BLOCKS_FORGET
);
4554 * The block which we have just freed is
4555 * pointed to by an indirect block: journal it
4557 BUFFER_TRACE(parent_bh
, "get_write_access");
4558 if (!ext4_journal_get_write_access(handle
,
4561 BUFFER_TRACE(parent_bh
,
4562 "call ext4_handle_dirty_metadata");
4563 ext4_handle_dirty_metadata(handle
,
4570 /* We have reached the bottom of the tree. */
4571 BUFFER_TRACE(parent_bh
, "free data blocks");
4572 ext4_free_data(handle
, inode
, parent_bh
, first
, last
);
4576 int ext4_can_truncate(struct inode
*inode
)
4578 if (IS_APPEND(inode
) || IS_IMMUTABLE(inode
))
4580 if (S_ISREG(inode
->i_mode
))
4582 if (S_ISDIR(inode
->i_mode
))
4584 if (S_ISLNK(inode
->i_mode
))
4585 return !ext4_inode_is_fast_symlink(inode
);
4592 * We block out ext4_get_block() block instantiations across the entire
4593 * transaction, and VFS/VM ensures that ext4_truncate() cannot run
4594 * simultaneously on behalf of the same inode.
4596 * As we work through the truncate and commmit bits of it to the journal there
4597 * is one core, guiding principle: the file's tree must always be consistent on
4598 * disk. We must be able to restart the truncate after a crash.
4600 * The file's tree may be transiently inconsistent in memory (although it
4601 * probably isn't), but whenever we close off and commit a journal transaction,
4602 * the contents of (the filesystem + the journal) must be consistent and
4603 * restartable. It's pretty simple, really: bottom up, right to left (although
4604 * left-to-right works OK too).
4606 * Note that at recovery time, journal replay occurs *before* the restart of
4607 * truncate against the orphan inode list.
4609 * The committed inode has the new, desired i_size (which is the same as
4610 * i_disksize in this case). After a crash, ext4_orphan_cleanup() will see
4611 * that this inode's truncate did not complete and it will again call
4612 * ext4_truncate() to have another go. So there will be instantiated blocks
4613 * to the right of the truncation point in a crashed ext4 filesystem. But
4614 * that's fine - as long as they are linked from the inode, the post-crash
4615 * ext4_truncate() run will find them and release them.
4617 void ext4_truncate(struct inode
*inode
)
4620 struct ext4_inode_info
*ei
= EXT4_I(inode
);
4621 __le32
*i_data
= ei
->i_data
;
4622 int addr_per_block
= EXT4_ADDR_PER_BLOCK(inode
->i_sb
);
4623 struct address_space
*mapping
= inode
->i_mapping
;
4624 ext4_lblk_t offsets
[4];
4629 ext4_lblk_t last_block
;
4630 unsigned blocksize
= inode
->i_sb
->s_blocksize
;
4632 if (!ext4_can_truncate(inode
))
4635 ext4_clear_inode_flag(inode
, EXT4_INODE_EOFBLOCKS
);
4637 if (inode
->i_size
== 0 && !test_opt(inode
->i_sb
, NO_AUTO_DA_ALLOC
))
4638 ext4_set_inode_state(inode
, EXT4_STATE_DA_ALLOC_CLOSE
);
4640 if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
)) {
4641 ext4_ext_truncate(inode
);
4645 handle
= start_transaction(inode
);
4647 return; /* AKPM: return what? */
4649 last_block
= (inode
->i_size
+ blocksize
-1)
4650 >> EXT4_BLOCK_SIZE_BITS(inode
->i_sb
);
4652 if (inode
->i_size
& (blocksize
- 1))
4653 if (ext4_block_truncate_page(handle
, mapping
, inode
->i_size
))
4656 n
= ext4_block_to_path(inode
, last_block
, offsets
, NULL
);
4658 goto out_stop
; /* error */
4661 * OK. This truncate is going to happen. We add the inode to the
4662 * orphan list, so that if this truncate spans multiple transactions,
4663 * and we crash, we will resume the truncate when the filesystem
4664 * recovers. It also marks the inode dirty, to catch the new size.
4666 * Implication: the file must always be in a sane, consistent
4667 * truncatable state while each transaction commits.
4669 if (ext4_orphan_add(handle
, inode
))
4673 * From here we block out all ext4_get_block() callers who want to
4674 * modify the block allocation tree.
4676 down_write(&ei
->i_data_sem
);
4678 ext4_discard_preallocations(inode
);
4681 * The orphan list entry will now protect us from any crash which
4682 * occurs before the truncate completes, so it is now safe to propagate
4683 * the new, shorter inode size (held for now in i_size) into the
4684 * on-disk inode. We do this via i_disksize, which is the value which
4685 * ext4 *really* writes onto the disk inode.
4687 ei
->i_disksize
= inode
->i_size
;
4689 if (n
== 1) { /* direct blocks */
4690 ext4_free_data(handle
, inode
, NULL
, i_data
+offsets
[0],
4691 i_data
+ EXT4_NDIR_BLOCKS
);
4695 partial
= ext4_find_shared(inode
, n
, offsets
, chain
, &nr
);
4696 /* Kill the top of shared branch (not detached) */
4698 if (partial
== chain
) {
4699 /* Shared branch grows from the inode */
4700 ext4_free_branches(handle
, inode
, NULL
,
4701 &nr
, &nr
+1, (chain
+n
-1) - partial
);
4704 * We mark the inode dirty prior to restart,
4705 * and prior to stop. No need for it here.
4708 /* Shared branch grows from an indirect block */
4709 BUFFER_TRACE(partial
->bh
, "get_write_access");
4710 ext4_free_branches(handle
, inode
, partial
->bh
,
4712 partial
->p
+1, (chain
+n
-1) - partial
);
4715 /* Clear the ends of indirect blocks on the shared branch */
4716 while (partial
> chain
) {
4717 ext4_free_branches(handle
, inode
, partial
->bh
, partial
->p
+ 1,
4718 (__le32
*)partial
->bh
->b_data
+addr_per_block
,
4719 (chain
+n
-1) - partial
);
4720 BUFFER_TRACE(partial
->bh
, "call brelse");
4721 brelse(partial
->bh
);
4725 /* Kill the remaining (whole) subtrees */
4726 switch (offsets
[0]) {
4728 nr
= i_data
[EXT4_IND_BLOCK
];
4730 ext4_free_branches(handle
, inode
, NULL
, &nr
, &nr
+1, 1);
4731 i_data
[EXT4_IND_BLOCK
] = 0;
4733 case EXT4_IND_BLOCK
:
4734 nr
= i_data
[EXT4_DIND_BLOCK
];
4736 ext4_free_branches(handle
, inode
, NULL
, &nr
, &nr
+1, 2);
4737 i_data
[EXT4_DIND_BLOCK
] = 0;
4739 case EXT4_DIND_BLOCK
:
4740 nr
= i_data
[EXT4_TIND_BLOCK
];
4742 ext4_free_branches(handle
, inode
, NULL
, &nr
, &nr
+1, 3);
4743 i_data
[EXT4_TIND_BLOCK
] = 0;
4745 case EXT4_TIND_BLOCK
:
4749 up_write(&ei
->i_data_sem
);
4750 inode
->i_mtime
= inode
->i_ctime
= ext4_current_time(inode
);
4751 ext4_mark_inode_dirty(handle
, inode
);
4754 * In a multi-transaction truncate, we only make the final transaction
4758 ext4_handle_sync(handle
);
4761 * If this was a simple ftruncate(), and the file will remain alive
4762 * then we need to clear up the orphan record which we created above.
4763 * However, if this was a real unlink then we were called by
4764 * ext4_delete_inode(), and we allow that function to clean up the
4765 * orphan info for us.
4768 ext4_orphan_del(handle
, inode
);
4770 ext4_journal_stop(handle
);
4774 * ext4_get_inode_loc returns with an extra refcount against the inode's
4775 * underlying buffer_head on success. If 'in_mem' is true, we have all
4776 * data in memory that is needed to recreate the on-disk version of this
4779 static int __ext4_get_inode_loc(struct inode
*inode
,
4780 struct ext4_iloc
*iloc
, int in_mem
)
4782 struct ext4_group_desc
*gdp
;
4783 struct buffer_head
*bh
;
4784 struct super_block
*sb
= inode
->i_sb
;
4786 int inodes_per_block
, inode_offset
;
4789 if (!ext4_valid_inum(sb
, inode
->i_ino
))
4792 iloc
->block_group
= (inode
->i_ino
- 1) / EXT4_INODES_PER_GROUP(sb
);
4793 gdp
= ext4_get_group_desc(sb
, iloc
->block_group
, NULL
);
4798 * Figure out the offset within the block group inode table
4800 inodes_per_block
= (EXT4_BLOCK_SIZE(sb
) / EXT4_INODE_SIZE(sb
));
4801 inode_offset
= ((inode
->i_ino
- 1) %
4802 EXT4_INODES_PER_GROUP(sb
));
4803 block
= ext4_inode_table(sb
, gdp
) + (inode_offset
/ inodes_per_block
);
4804 iloc
->offset
= (inode_offset
% inodes_per_block
) * EXT4_INODE_SIZE(sb
);
4806 bh
= sb_getblk(sb
, block
);
4808 EXT4_ERROR_INODE_BLOCK(inode
, block
,
4809 "unable to read itable block");
4812 if (!buffer_uptodate(bh
)) {
4816 * If the buffer has the write error flag, we have failed
4817 * to write out another inode in the same block. In this
4818 * case, we don't have to read the block because we may
4819 * read the old inode data successfully.
4821 if (buffer_write_io_error(bh
) && !buffer_uptodate(bh
))
4822 set_buffer_uptodate(bh
);
4824 if (buffer_uptodate(bh
)) {
4825 /* someone brought it uptodate while we waited */
4831 * If we have all information of the inode in memory and this
4832 * is the only valid inode in the block, we need not read the
4836 struct buffer_head
*bitmap_bh
;
4839 start
= inode_offset
& ~(inodes_per_block
- 1);
4841 /* Is the inode bitmap in cache? */
4842 bitmap_bh
= sb_getblk(sb
, ext4_inode_bitmap(sb
, gdp
));
4847 * If the inode bitmap isn't in cache then the
4848 * optimisation may end up performing two reads instead
4849 * of one, so skip it.
4851 if (!buffer_uptodate(bitmap_bh
)) {
4855 for (i
= start
; i
< start
+ inodes_per_block
; i
++) {
4856 if (i
== inode_offset
)
4858 if (ext4_test_bit(i
, bitmap_bh
->b_data
))
4862 if (i
== start
+ inodes_per_block
) {
4863 /* all other inodes are free, so skip I/O */
4864 memset(bh
->b_data
, 0, bh
->b_size
);
4865 set_buffer_uptodate(bh
);
4873 * If we need to do any I/O, try to pre-readahead extra
4874 * blocks from the inode table.
4876 if (EXT4_SB(sb
)->s_inode_readahead_blks
) {
4877 ext4_fsblk_t b
, end
, table
;
4880 table
= ext4_inode_table(sb
, gdp
);
4881 /* s_inode_readahead_blks is always a power of 2 */
4882 b
= block
& ~(EXT4_SB(sb
)->s_inode_readahead_blks
-1);
4885 end
= b
+ EXT4_SB(sb
)->s_inode_readahead_blks
;
4886 num
= EXT4_INODES_PER_GROUP(sb
);
4887 if (EXT4_HAS_RO_COMPAT_FEATURE(sb
,
4888 EXT4_FEATURE_RO_COMPAT_GDT_CSUM
))
4889 num
-= ext4_itable_unused_count(sb
, gdp
);
4890 table
+= num
/ inodes_per_block
;
4894 sb_breadahead(sb
, b
++);
4898 * There are other valid inodes in the buffer, this inode
4899 * has in-inode xattrs, or we don't have this inode in memory.
4900 * Read the block from disk.
4903 bh
->b_end_io
= end_buffer_read_sync
;
4904 submit_bh(READ_META
, bh
);
4906 if (!buffer_uptodate(bh
)) {
4907 EXT4_ERROR_INODE_BLOCK(inode
, block
,
4908 "unable to read itable block");
4918 int ext4_get_inode_loc(struct inode
*inode
, struct ext4_iloc
*iloc
)
4920 /* We have all inode data except xattrs in memory here. */
4921 return __ext4_get_inode_loc(inode
, iloc
,
4922 !ext4_test_inode_state(inode
, EXT4_STATE_XATTR
));
4925 void ext4_set_inode_flags(struct inode
*inode
)
4927 unsigned int flags
= EXT4_I(inode
)->i_flags
;
4929 inode
->i_flags
&= ~(S_SYNC
|S_APPEND
|S_IMMUTABLE
|S_NOATIME
|S_DIRSYNC
);
4930 if (flags
& EXT4_SYNC_FL
)
4931 inode
->i_flags
|= S_SYNC
;
4932 if (flags
& EXT4_APPEND_FL
)
4933 inode
->i_flags
|= S_APPEND
;
4934 if (flags
& EXT4_IMMUTABLE_FL
)
4935 inode
->i_flags
|= S_IMMUTABLE
;
4936 if (flags
& EXT4_NOATIME_FL
)
4937 inode
->i_flags
|= S_NOATIME
;
4938 if (flags
& EXT4_DIRSYNC_FL
)
4939 inode
->i_flags
|= S_DIRSYNC
;
4942 /* Propagate flags from i_flags to EXT4_I(inode)->i_flags */
4943 void ext4_get_inode_flags(struct ext4_inode_info
*ei
)
4945 unsigned int vfs_fl
;
4946 unsigned long old_fl
, new_fl
;
4949 vfs_fl
= ei
->vfs_inode
.i_flags
;
4950 old_fl
= ei
->i_flags
;
4951 new_fl
= old_fl
& ~(EXT4_SYNC_FL
|EXT4_APPEND_FL
|
4952 EXT4_IMMUTABLE_FL
|EXT4_NOATIME_FL
|
4954 if (vfs_fl
& S_SYNC
)
4955 new_fl
|= EXT4_SYNC_FL
;
4956 if (vfs_fl
& S_APPEND
)
4957 new_fl
|= EXT4_APPEND_FL
;
4958 if (vfs_fl
& S_IMMUTABLE
)
4959 new_fl
|= EXT4_IMMUTABLE_FL
;
4960 if (vfs_fl
& S_NOATIME
)
4961 new_fl
|= EXT4_NOATIME_FL
;
4962 if (vfs_fl
& S_DIRSYNC
)
4963 new_fl
|= EXT4_DIRSYNC_FL
;
4964 } while (cmpxchg(&ei
->i_flags
, old_fl
, new_fl
) != old_fl
);
4967 static blkcnt_t
ext4_inode_blocks(struct ext4_inode
*raw_inode
,
4968 struct ext4_inode_info
*ei
)
4971 struct inode
*inode
= &(ei
->vfs_inode
);
4972 struct super_block
*sb
= inode
->i_sb
;
4974 if (EXT4_HAS_RO_COMPAT_FEATURE(sb
,
4975 EXT4_FEATURE_RO_COMPAT_HUGE_FILE
)) {
4976 /* we are using combined 48 bit field */
4977 i_blocks
= ((u64
)le16_to_cpu(raw_inode
->i_blocks_high
)) << 32 |
4978 le32_to_cpu(raw_inode
->i_blocks_lo
);
4979 if (ext4_test_inode_flag(inode
, EXT4_INODE_HUGE_FILE
)) {
4980 /* i_blocks represent file system block size */
4981 return i_blocks
<< (inode
->i_blkbits
- 9);
4986 return le32_to_cpu(raw_inode
->i_blocks_lo
);
4990 struct inode
*ext4_iget(struct super_block
*sb
, unsigned long ino
)
4992 struct ext4_iloc iloc
;
4993 struct ext4_inode
*raw_inode
;
4994 struct ext4_inode_info
*ei
;
4995 struct inode
*inode
;
4996 journal_t
*journal
= EXT4_SB(sb
)->s_journal
;
5000 inode
= iget_locked(sb
, ino
);
5002 return ERR_PTR(-ENOMEM
);
5003 if (!(inode
->i_state
& I_NEW
))
5009 ret
= __ext4_get_inode_loc(inode
, &iloc
, 0);
5012 raw_inode
= ext4_raw_inode(&iloc
);
5013 inode
->i_mode
= le16_to_cpu(raw_inode
->i_mode
);
5014 inode
->i_uid
= (uid_t
)le16_to_cpu(raw_inode
->i_uid_low
);
5015 inode
->i_gid
= (gid_t
)le16_to_cpu(raw_inode
->i_gid_low
);
5016 if (!(test_opt(inode
->i_sb
, NO_UID32
))) {
5017 inode
->i_uid
|= le16_to_cpu(raw_inode
->i_uid_high
) << 16;
5018 inode
->i_gid
|= le16_to_cpu(raw_inode
->i_gid_high
) << 16;
5020 inode
->i_nlink
= le16_to_cpu(raw_inode
->i_links_count
);
5022 ei
->i_state_flags
= 0;
5023 ei
->i_dir_start_lookup
= 0;
5024 ei
->i_dtime
= le32_to_cpu(raw_inode
->i_dtime
);
5025 /* We now have enough fields to check if the inode was active or not.
5026 * This is needed because nfsd might try to access dead inodes
5027 * the test is that same one that e2fsck uses
5028 * NeilBrown 1999oct15
5030 if (inode
->i_nlink
== 0) {
5031 if (inode
->i_mode
== 0 ||
5032 !(EXT4_SB(inode
->i_sb
)->s_mount_state
& EXT4_ORPHAN_FS
)) {
5033 /* this inode is deleted */
5037 /* The only unlinked inodes we let through here have
5038 * valid i_mode and are being read by the orphan
5039 * recovery code: that's fine, we're about to complete
5040 * the process of deleting those. */
5042 ei
->i_flags
= le32_to_cpu(raw_inode
->i_flags
);
5043 inode
->i_blocks
= ext4_inode_blocks(raw_inode
, ei
);
5044 ei
->i_file_acl
= le32_to_cpu(raw_inode
->i_file_acl_lo
);
5045 if (EXT4_HAS_INCOMPAT_FEATURE(sb
, EXT4_FEATURE_INCOMPAT_64BIT
))
5047 ((__u64
)le16_to_cpu(raw_inode
->i_file_acl_high
)) << 32;
5048 inode
->i_size
= ext4_isize(raw_inode
);
5049 ei
->i_disksize
= inode
->i_size
;
5051 ei
->i_reserved_quota
= 0;
5053 inode
->i_generation
= le32_to_cpu(raw_inode
->i_generation
);
5054 ei
->i_block_group
= iloc
.block_group
;
5055 ei
->i_last_alloc_group
= ~0;
5057 * NOTE! The in-memory inode i_data array is in little-endian order
5058 * even on big-endian machines: we do NOT byteswap the block numbers!
5060 for (block
= 0; block
< EXT4_N_BLOCKS
; block
++)
5061 ei
->i_data
[block
] = raw_inode
->i_block
[block
];
5062 INIT_LIST_HEAD(&ei
->i_orphan
);
5065 * Set transaction id's of transactions that have to be committed
5066 * to finish f[data]sync. We set them to currently running transaction
5067 * as we cannot be sure that the inode or some of its metadata isn't
5068 * part of the transaction - the inode could have been reclaimed and
5069 * now it is reread from disk.
5072 transaction_t
*transaction
;
5075 read_lock(&journal
->j_state_lock
);
5076 if (journal
->j_running_transaction
)
5077 transaction
= journal
->j_running_transaction
;
5079 transaction
= journal
->j_committing_transaction
;
5081 tid
= transaction
->t_tid
;
5083 tid
= journal
->j_commit_sequence
;
5084 read_unlock(&journal
->j_state_lock
);
5085 ei
->i_sync_tid
= tid
;
5086 ei
->i_datasync_tid
= tid
;
5089 if (EXT4_INODE_SIZE(inode
->i_sb
) > EXT4_GOOD_OLD_INODE_SIZE
) {
5090 ei
->i_extra_isize
= le16_to_cpu(raw_inode
->i_extra_isize
);
5091 if (EXT4_GOOD_OLD_INODE_SIZE
+ ei
->i_extra_isize
>
5092 EXT4_INODE_SIZE(inode
->i_sb
)) {
5096 if (ei
->i_extra_isize
== 0) {
5097 /* The extra space is currently unused. Use it. */
5098 ei
->i_extra_isize
= sizeof(struct ext4_inode
) -
5099 EXT4_GOOD_OLD_INODE_SIZE
;
5101 __le32
*magic
= (void *)raw_inode
+
5102 EXT4_GOOD_OLD_INODE_SIZE
+
5104 if (*magic
== cpu_to_le32(EXT4_XATTR_MAGIC
))
5105 ext4_set_inode_state(inode
, EXT4_STATE_XATTR
);
5108 ei
->i_extra_isize
= 0;
5110 EXT4_INODE_GET_XTIME(i_ctime
, inode
, raw_inode
);
5111 EXT4_INODE_GET_XTIME(i_mtime
, inode
, raw_inode
);
5112 EXT4_INODE_GET_XTIME(i_atime
, inode
, raw_inode
);
5113 EXT4_EINODE_GET_XTIME(i_crtime
, ei
, raw_inode
);
5115 inode
->i_version
= le32_to_cpu(raw_inode
->i_disk_version
);
5116 if (EXT4_INODE_SIZE(inode
->i_sb
) > EXT4_GOOD_OLD_INODE_SIZE
) {
5117 if (EXT4_FITS_IN_INODE(raw_inode
, ei
, i_version_hi
))
5119 (__u64
)(le32_to_cpu(raw_inode
->i_version_hi
)) << 32;
5123 if (ei
->i_file_acl
&&
5124 !ext4_data_block_valid(EXT4_SB(sb
), ei
->i_file_acl
, 1)) {
5125 EXT4_ERROR_INODE(inode
, "bad extended attribute block %llu",
5129 } else if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
)) {
5130 if (S_ISREG(inode
->i_mode
) || S_ISDIR(inode
->i_mode
) ||
5131 (S_ISLNK(inode
->i_mode
) &&
5132 !ext4_inode_is_fast_symlink(inode
)))
5133 /* Validate extent which is part of inode */
5134 ret
= ext4_ext_check_inode(inode
);
5135 } else if (S_ISREG(inode
->i_mode
) || S_ISDIR(inode
->i_mode
) ||
5136 (S_ISLNK(inode
->i_mode
) &&
5137 !ext4_inode_is_fast_symlink(inode
))) {
5138 /* Validate block references which are part of inode */
5139 ret
= ext4_check_inode_blockref(inode
);
5144 if (S_ISREG(inode
->i_mode
)) {
5145 inode
->i_op
= &ext4_file_inode_operations
;
5146 inode
->i_fop
= &ext4_file_operations
;
5147 ext4_set_aops(inode
);
5148 } else if (S_ISDIR(inode
->i_mode
)) {
5149 inode
->i_op
= &ext4_dir_inode_operations
;
5150 inode
->i_fop
= &ext4_dir_operations
;
5151 } else if (S_ISLNK(inode
->i_mode
)) {
5152 if (ext4_inode_is_fast_symlink(inode
)) {
5153 inode
->i_op
= &ext4_fast_symlink_inode_operations
;
5154 nd_terminate_link(ei
->i_data
, inode
->i_size
,
5155 sizeof(ei
->i_data
) - 1);
5157 inode
->i_op
= &ext4_symlink_inode_operations
;
5158 ext4_set_aops(inode
);
5160 } else if (S_ISCHR(inode
->i_mode
) || S_ISBLK(inode
->i_mode
) ||
5161 S_ISFIFO(inode
->i_mode
) || S_ISSOCK(inode
->i_mode
)) {
5162 inode
->i_op
= &ext4_special_inode_operations
;
5163 if (raw_inode
->i_block
[0])
5164 init_special_inode(inode
, inode
->i_mode
,
5165 old_decode_dev(le32_to_cpu(raw_inode
->i_block
[0])));
5167 init_special_inode(inode
, inode
->i_mode
,
5168 new_decode_dev(le32_to_cpu(raw_inode
->i_block
[1])));
5171 EXT4_ERROR_INODE(inode
, "bogus i_mode (%o)", inode
->i_mode
);
5175 ext4_set_inode_flags(inode
);
5176 unlock_new_inode(inode
);
5182 return ERR_PTR(ret
);
5185 static int ext4_inode_blocks_set(handle_t
*handle
,
5186 struct ext4_inode
*raw_inode
,
5187 struct ext4_inode_info
*ei
)
5189 struct inode
*inode
= &(ei
->vfs_inode
);
5190 u64 i_blocks
= inode
->i_blocks
;
5191 struct super_block
*sb
= inode
->i_sb
;
5193 if (i_blocks
<= ~0U) {
5195 * i_blocks can be represnted in a 32 bit variable
5196 * as multiple of 512 bytes
5198 raw_inode
->i_blocks_lo
= cpu_to_le32(i_blocks
);
5199 raw_inode
->i_blocks_high
= 0;
5200 ext4_clear_inode_flag(inode
, EXT4_INODE_HUGE_FILE
);
5203 if (!EXT4_HAS_RO_COMPAT_FEATURE(sb
, EXT4_FEATURE_RO_COMPAT_HUGE_FILE
))
5206 if (i_blocks
<= 0xffffffffffffULL
) {
5208 * i_blocks can be represented in a 48 bit variable
5209 * as multiple of 512 bytes
5211 raw_inode
->i_blocks_lo
= cpu_to_le32(i_blocks
);
5212 raw_inode
->i_blocks_high
= cpu_to_le16(i_blocks
>> 32);
5213 ext4_clear_inode_flag(inode
, EXT4_INODE_HUGE_FILE
);
5215 ext4_set_inode_flag(inode
, EXT4_INODE_HUGE_FILE
);
5216 /* i_block is stored in file system block size */
5217 i_blocks
= i_blocks
>> (inode
->i_blkbits
- 9);
5218 raw_inode
->i_blocks_lo
= cpu_to_le32(i_blocks
);
5219 raw_inode
->i_blocks_high
= cpu_to_le16(i_blocks
>> 32);
5225 * Post the struct inode info into an on-disk inode location in the
5226 * buffer-cache. This gobbles the caller's reference to the
5227 * buffer_head in the inode location struct.
5229 * The caller must have write access to iloc->bh.
5231 static int ext4_do_update_inode(handle_t
*handle
,
5232 struct inode
*inode
,
5233 struct ext4_iloc
*iloc
)
5235 struct ext4_inode
*raw_inode
= ext4_raw_inode(iloc
);
5236 struct ext4_inode_info
*ei
= EXT4_I(inode
);
5237 struct buffer_head
*bh
= iloc
->bh
;
5238 int err
= 0, rc
, block
;
5240 /* For fields not not tracking in the in-memory inode,
5241 * initialise them to zero for new inodes. */
5242 if (ext4_test_inode_state(inode
, EXT4_STATE_NEW
))
5243 memset(raw_inode
, 0, EXT4_SB(inode
->i_sb
)->s_inode_size
);
5245 ext4_get_inode_flags(ei
);
5246 raw_inode
->i_mode
= cpu_to_le16(inode
->i_mode
);
5247 if (!(test_opt(inode
->i_sb
, NO_UID32
))) {
5248 raw_inode
->i_uid_low
= cpu_to_le16(low_16_bits(inode
->i_uid
));
5249 raw_inode
->i_gid_low
= cpu_to_le16(low_16_bits(inode
->i_gid
));
5251 * Fix up interoperability with old kernels. Otherwise, old inodes get
5252 * re-used with the upper 16 bits of the uid/gid intact
5255 raw_inode
->i_uid_high
=
5256 cpu_to_le16(high_16_bits(inode
->i_uid
));
5257 raw_inode
->i_gid_high
=
5258 cpu_to_le16(high_16_bits(inode
->i_gid
));
5260 raw_inode
->i_uid_high
= 0;
5261 raw_inode
->i_gid_high
= 0;
5264 raw_inode
->i_uid_low
=
5265 cpu_to_le16(fs_high2lowuid(inode
->i_uid
));
5266 raw_inode
->i_gid_low
=
5267 cpu_to_le16(fs_high2lowgid(inode
->i_gid
));
5268 raw_inode
->i_uid_high
= 0;
5269 raw_inode
->i_gid_high
= 0;
5271 raw_inode
->i_links_count
= cpu_to_le16(inode
->i_nlink
);
5273 EXT4_INODE_SET_XTIME(i_ctime
, inode
, raw_inode
);
5274 EXT4_INODE_SET_XTIME(i_mtime
, inode
, raw_inode
);
5275 EXT4_INODE_SET_XTIME(i_atime
, inode
, raw_inode
);
5276 EXT4_EINODE_SET_XTIME(i_crtime
, ei
, raw_inode
);
5278 if (ext4_inode_blocks_set(handle
, raw_inode
, ei
))
5280 raw_inode
->i_dtime
= cpu_to_le32(ei
->i_dtime
);
5281 raw_inode
->i_flags
= cpu_to_le32(ei
->i_flags
);
5282 if (EXT4_SB(inode
->i_sb
)->s_es
->s_creator_os
!=
5283 cpu_to_le32(EXT4_OS_HURD
))
5284 raw_inode
->i_file_acl_high
=
5285 cpu_to_le16(ei
->i_file_acl
>> 32);
5286 raw_inode
->i_file_acl_lo
= cpu_to_le32(ei
->i_file_acl
);
5287 ext4_isize_set(raw_inode
, ei
->i_disksize
);
5288 if (ei
->i_disksize
> 0x7fffffffULL
) {
5289 struct super_block
*sb
= inode
->i_sb
;
5290 if (!EXT4_HAS_RO_COMPAT_FEATURE(sb
,
5291 EXT4_FEATURE_RO_COMPAT_LARGE_FILE
) ||
5292 EXT4_SB(sb
)->s_es
->s_rev_level
==
5293 cpu_to_le32(EXT4_GOOD_OLD_REV
)) {
5294 /* If this is the first large file
5295 * created, add a flag to the superblock.
5297 err
= ext4_journal_get_write_access(handle
,
5298 EXT4_SB(sb
)->s_sbh
);
5301 ext4_update_dynamic_rev(sb
);
5302 EXT4_SET_RO_COMPAT_FEATURE(sb
,
5303 EXT4_FEATURE_RO_COMPAT_LARGE_FILE
);
5305 ext4_handle_sync(handle
);
5306 err
= ext4_handle_dirty_metadata(handle
, NULL
,
5307 EXT4_SB(sb
)->s_sbh
);
5310 raw_inode
->i_generation
= cpu_to_le32(inode
->i_generation
);
5311 if (S_ISCHR(inode
->i_mode
) || S_ISBLK(inode
->i_mode
)) {
5312 if (old_valid_dev(inode
->i_rdev
)) {
5313 raw_inode
->i_block
[0] =
5314 cpu_to_le32(old_encode_dev(inode
->i_rdev
));
5315 raw_inode
->i_block
[1] = 0;
5317 raw_inode
->i_block
[0] = 0;
5318 raw_inode
->i_block
[1] =
5319 cpu_to_le32(new_encode_dev(inode
->i_rdev
));
5320 raw_inode
->i_block
[2] = 0;
5323 for (block
= 0; block
< EXT4_N_BLOCKS
; block
++)
5324 raw_inode
->i_block
[block
] = ei
->i_data
[block
];
5326 raw_inode
->i_disk_version
= cpu_to_le32(inode
->i_version
);
5327 if (ei
->i_extra_isize
) {
5328 if (EXT4_FITS_IN_INODE(raw_inode
, ei
, i_version_hi
))
5329 raw_inode
->i_version_hi
=
5330 cpu_to_le32(inode
->i_version
>> 32);
5331 raw_inode
->i_extra_isize
= cpu_to_le16(ei
->i_extra_isize
);
5334 BUFFER_TRACE(bh
, "call ext4_handle_dirty_metadata");
5335 rc
= ext4_handle_dirty_metadata(handle
, NULL
, bh
);
5338 ext4_clear_inode_state(inode
, EXT4_STATE_NEW
);
5340 ext4_update_inode_fsync_trans(handle
, inode
, 0);
5343 ext4_std_error(inode
->i_sb
, err
);
5348 * ext4_write_inode()
5350 * We are called from a few places:
5352 * - Within generic_file_write() for O_SYNC files.
5353 * Here, there will be no transaction running. We wait for any running
5354 * trasnaction to commit.
5356 * - Within sys_sync(), kupdate and such.
5357 * We wait on commit, if tol to.
5359 * - Within prune_icache() (PF_MEMALLOC == true)
5360 * Here we simply return. We can't afford to block kswapd on the
5363 * In all cases it is actually safe for us to return without doing anything,
5364 * because the inode has been copied into a raw inode buffer in
5365 * ext4_mark_inode_dirty(). This is a correctness thing for O_SYNC and for
5368 * Note that we are absolutely dependent upon all inode dirtiers doing the
5369 * right thing: they *must* call mark_inode_dirty() after dirtying info in
5370 * which we are interested.
5372 * It would be a bug for them to not do this. The code:
5374 * mark_inode_dirty(inode)
5376 * inode->i_size = expr;
5378 * is in error because a kswapd-driven write_inode() could occur while
5379 * `stuff()' is running, and the new i_size will be lost. Plus the inode
5380 * will no longer be on the superblock's dirty inode list.
5382 int ext4_write_inode(struct inode
*inode
, struct writeback_control
*wbc
)
5386 if (current
->flags
& PF_MEMALLOC
)
5389 if (EXT4_SB(inode
->i_sb
)->s_journal
) {
5390 if (ext4_journal_current_handle()) {
5391 jbd_debug(1, "called recursively, non-PF_MEMALLOC!\n");
5396 if (wbc
->sync_mode
!= WB_SYNC_ALL
)
5399 err
= ext4_force_commit(inode
->i_sb
);
5401 struct ext4_iloc iloc
;
5403 err
= __ext4_get_inode_loc(inode
, &iloc
, 0);
5406 if (wbc
->sync_mode
== WB_SYNC_ALL
)
5407 sync_dirty_buffer(iloc
.bh
);
5408 if (buffer_req(iloc
.bh
) && !buffer_uptodate(iloc
.bh
)) {
5409 EXT4_ERROR_INODE_BLOCK(inode
, iloc
.bh
->b_blocknr
,
5410 "IO error syncing inode");
5421 * Called from notify_change.
5423 * We want to trap VFS attempts to truncate the file as soon as
5424 * possible. In particular, we want to make sure that when the VFS
5425 * shrinks i_size, we put the inode on the orphan list and modify
5426 * i_disksize immediately, so that during the subsequent flushing of
5427 * dirty pages and freeing of disk blocks, we can guarantee that any
5428 * commit will leave the blocks being flushed in an unused state on
5429 * disk. (On recovery, the inode will get truncated and the blocks will
5430 * be freed, so we have a strong guarantee that no future commit will
5431 * leave these blocks visible to the user.)
5433 * Another thing we have to assure is that if we are in ordered mode
5434 * and inode is still attached to the committing transaction, we must
5435 * we start writeout of all the dirty pages which are being truncated.
5436 * This way we are sure that all the data written in the previous
5437 * transaction are already on disk (truncate waits for pages under
5440 * Called with inode->i_mutex down.
5442 int ext4_setattr(struct dentry
*dentry
, struct iattr
*attr
)
5444 struct inode
*inode
= dentry
->d_inode
;
5446 const unsigned int ia_valid
= attr
->ia_valid
;
5448 error
= inode_change_ok(inode
, attr
);
5452 if (is_quota_modification(inode
, attr
))
5453 dquot_initialize(inode
);
5454 if ((ia_valid
& ATTR_UID
&& attr
->ia_uid
!= inode
->i_uid
) ||
5455 (ia_valid
& ATTR_GID
&& attr
->ia_gid
!= inode
->i_gid
)) {
5458 /* (user+group)*(old+new) structure, inode write (sb,
5459 * inode block, ? - but truncate inode update has it) */
5460 handle
= ext4_journal_start(inode
, (EXT4_MAXQUOTAS_INIT_BLOCKS(inode
->i_sb
)+
5461 EXT4_MAXQUOTAS_DEL_BLOCKS(inode
->i_sb
))+3);
5462 if (IS_ERR(handle
)) {
5463 error
= PTR_ERR(handle
);
5466 error
= dquot_transfer(inode
, attr
);
5468 ext4_journal_stop(handle
);
5471 /* Update corresponding info in inode so that everything is in
5472 * one transaction */
5473 if (attr
->ia_valid
& ATTR_UID
)
5474 inode
->i_uid
= attr
->ia_uid
;
5475 if (attr
->ia_valid
& ATTR_GID
)
5476 inode
->i_gid
= attr
->ia_gid
;
5477 error
= ext4_mark_inode_dirty(handle
, inode
);
5478 ext4_journal_stop(handle
);
5481 if (attr
->ia_valid
& ATTR_SIZE
) {
5482 if (!(ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
))) {
5483 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
5485 if (attr
->ia_size
> sbi
->s_bitmap_maxbytes
)
5490 if (S_ISREG(inode
->i_mode
) &&
5491 attr
->ia_valid
& ATTR_SIZE
&&
5492 (attr
->ia_size
< inode
->i_size
||
5493 (ext4_test_inode_flag(inode
, EXT4_INODE_EOFBLOCKS
)))) {
5496 handle
= ext4_journal_start(inode
, 3);
5497 if (IS_ERR(handle
)) {
5498 error
= PTR_ERR(handle
);
5502 error
= ext4_orphan_add(handle
, inode
);
5503 EXT4_I(inode
)->i_disksize
= attr
->ia_size
;
5504 rc
= ext4_mark_inode_dirty(handle
, inode
);
5507 ext4_journal_stop(handle
);
5509 if (ext4_should_order_data(inode
)) {
5510 error
= ext4_begin_ordered_truncate(inode
,
5513 /* Do as much error cleanup as possible */
5514 handle
= ext4_journal_start(inode
, 3);
5515 if (IS_ERR(handle
)) {
5516 ext4_orphan_del(NULL
, inode
);
5519 ext4_orphan_del(handle
, inode
);
5520 ext4_journal_stop(handle
);
5524 /* ext4_truncate will clear the flag */
5525 if ((ext4_test_inode_flag(inode
, EXT4_INODE_EOFBLOCKS
)))
5526 ext4_truncate(inode
);
5529 if ((attr
->ia_valid
& ATTR_SIZE
) &&
5530 attr
->ia_size
!= i_size_read(inode
))
5531 rc
= vmtruncate(inode
, attr
->ia_size
);
5534 setattr_copy(inode
, attr
);
5535 mark_inode_dirty(inode
);
5539 * If the call to ext4_truncate failed to get a transaction handle at
5540 * all, we need to clean up the in-core orphan list manually.
5543 ext4_orphan_del(NULL
, inode
);
5545 if (!rc
&& (ia_valid
& ATTR_MODE
))
5546 rc
= ext4_acl_chmod(inode
);
5549 ext4_std_error(inode
->i_sb
, error
);
5555 int ext4_getattr(struct vfsmount
*mnt
, struct dentry
*dentry
,
5558 struct inode
*inode
;
5559 unsigned long delalloc_blocks
;
5561 inode
= dentry
->d_inode
;
5562 generic_fillattr(inode
, stat
);
5565 * We can't update i_blocks if the block allocation is delayed
5566 * otherwise in the case of system crash before the real block
5567 * allocation is done, we will have i_blocks inconsistent with
5568 * on-disk file blocks.
5569 * We always keep i_blocks updated together with real
5570 * allocation. But to not confuse with user, stat
5571 * will return the blocks that include the delayed allocation
5572 * blocks for this file.
5574 spin_lock(&EXT4_I(inode
)->i_block_reservation_lock
);
5575 delalloc_blocks
= EXT4_I(inode
)->i_reserved_data_blocks
;
5576 spin_unlock(&EXT4_I(inode
)->i_block_reservation_lock
);
5578 stat
->blocks
+= (delalloc_blocks
<< inode
->i_sb
->s_blocksize_bits
)>>9;
5582 static int ext4_indirect_trans_blocks(struct inode
*inode
, int nrblocks
,
5587 /* if nrblocks are contiguous */
5590 * With N contiguous data blocks, it need at most
5591 * N/EXT4_ADDR_PER_BLOCK(inode->i_sb) indirect blocks
5592 * 2 dindirect blocks
5595 indirects
= nrblocks
/ EXT4_ADDR_PER_BLOCK(inode
->i_sb
);
5596 return indirects
+ 3;
5599 * if nrblocks are not contiguous, worse case, each block touch
5600 * a indirect block, and each indirect block touch a double indirect
5601 * block, plus a triple indirect block
5603 indirects
= nrblocks
* 2 + 1;
5607 static int ext4_index_trans_blocks(struct inode
*inode
, int nrblocks
, int chunk
)
5609 if (!(ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
)))
5610 return ext4_indirect_trans_blocks(inode
, nrblocks
, chunk
);
5611 return ext4_ext_index_trans_blocks(inode
, nrblocks
, chunk
);
5615 * Account for index blocks, block groups bitmaps and block group
5616 * descriptor blocks if modify datablocks and index blocks
5617 * worse case, the indexs blocks spread over different block groups
5619 * If datablocks are discontiguous, they are possible to spread over
5620 * different block groups too. If they are contiuguous, with flexbg,
5621 * they could still across block group boundary.
5623 * Also account for superblock, inode, quota and xattr blocks
5625 int ext4_meta_trans_blocks(struct inode
*inode
, int nrblocks
, int chunk
)
5627 ext4_group_t groups
, ngroups
= ext4_get_groups_count(inode
->i_sb
);
5633 * How many index blocks need to touch to modify nrblocks?
5634 * The "Chunk" flag indicating whether the nrblocks is
5635 * physically contiguous on disk
5637 * For Direct IO and fallocate, they calls get_block to allocate
5638 * one single extent at a time, so they could set the "Chunk" flag
5640 idxblocks
= ext4_index_trans_blocks(inode
, nrblocks
, chunk
);
5645 * Now let's see how many group bitmaps and group descriptors need
5655 if (groups
> ngroups
)
5657 if (groups
> EXT4_SB(inode
->i_sb
)->s_gdb_count
)
5658 gdpblocks
= EXT4_SB(inode
->i_sb
)->s_gdb_count
;
5660 /* bitmaps and block group descriptor blocks */
5661 ret
+= groups
+ gdpblocks
;
5663 /* Blocks for super block, inode, quota and xattr blocks */
5664 ret
+= EXT4_META_TRANS_BLOCKS(inode
->i_sb
);
5670 * Calulate the total number of credits to reserve to fit
5671 * the modification of a single pages into a single transaction,
5672 * which may include multiple chunks of block allocations.
5674 * This could be called via ext4_write_begin()
5676 * We need to consider the worse case, when
5677 * one new block per extent.
5679 int ext4_writepage_trans_blocks(struct inode
*inode
)
5681 int bpp
= ext4_journal_blocks_per_page(inode
);
5684 ret
= ext4_meta_trans_blocks(inode
, bpp
, 0);
5686 /* Account for data blocks for journalled mode */
5687 if (ext4_should_journal_data(inode
))
5693 * Calculate the journal credits for a chunk of data modification.
5695 * This is called from DIO, fallocate or whoever calling
5696 * ext4_map_blocks() to map/allocate a chunk of contiguous disk blocks.
5698 * journal buffers for data blocks are not included here, as DIO
5699 * and fallocate do no need to journal data buffers.
5701 int ext4_chunk_trans_blocks(struct inode
*inode
, int nrblocks
)
5703 return ext4_meta_trans_blocks(inode
, nrblocks
, 1);
5707 * The caller must have previously called ext4_reserve_inode_write().
5708 * Give this, we know that the caller already has write access to iloc->bh.
5710 int ext4_mark_iloc_dirty(handle_t
*handle
,
5711 struct inode
*inode
, struct ext4_iloc
*iloc
)
5715 if (test_opt(inode
->i_sb
, I_VERSION
))
5716 inode_inc_iversion(inode
);
5718 /* the do_update_inode consumes one bh->b_count */
5721 /* ext4_do_update_inode() does jbd2_journal_dirty_metadata */
5722 err
= ext4_do_update_inode(handle
, inode
, iloc
);
5728 * On success, We end up with an outstanding reference count against
5729 * iloc->bh. This _must_ be cleaned up later.
5733 ext4_reserve_inode_write(handle_t
*handle
, struct inode
*inode
,
5734 struct ext4_iloc
*iloc
)
5738 err
= ext4_get_inode_loc(inode
, iloc
);
5740 BUFFER_TRACE(iloc
->bh
, "get_write_access");
5741 err
= ext4_journal_get_write_access(handle
, iloc
->bh
);
5747 ext4_std_error(inode
->i_sb
, err
);
5752 * Expand an inode by new_extra_isize bytes.
5753 * Returns 0 on success or negative error number on failure.
5755 static int ext4_expand_extra_isize(struct inode
*inode
,
5756 unsigned int new_extra_isize
,
5757 struct ext4_iloc iloc
,
5760 struct ext4_inode
*raw_inode
;
5761 struct ext4_xattr_ibody_header
*header
;
5763 if (EXT4_I(inode
)->i_extra_isize
>= new_extra_isize
)
5766 raw_inode
= ext4_raw_inode(&iloc
);
5768 header
= IHDR(inode
, raw_inode
);
5770 /* No extended attributes present */
5771 if (!ext4_test_inode_state(inode
, EXT4_STATE_XATTR
) ||
5772 header
->h_magic
!= cpu_to_le32(EXT4_XATTR_MAGIC
)) {
5773 memset((void *)raw_inode
+ EXT4_GOOD_OLD_INODE_SIZE
, 0,
5775 EXT4_I(inode
)->i_extra_isize
= new_extra_isize
;
5779 /* try to expand with EAs present */
5780 return ext4_expand_extra_isize_ea(inode
, new_extra_isize
,
5785 * What we do here is to mark the in-core inode as clean with respect to inode
5786 * dirtiness (it may still be data-dirty).
5787 * This means that the in-core inode may be reaped by prune_icache
5788 * without having to perform any I/O. This is a very good thing,
5789 * because *any* task may call prune_icache - even ones which
5790 * have a transaction open against a different journal.
5792 * Is this cheating? Not really. Sure, we haven't written the
5793 * inode out, but prune_icache isn't a user-visible syncing function.
5794 * Whenever the user wants stuff synced (sys_sync, sys_msync, sys_fsync)
5795 * we start and wait on commits.
5797 * Is this efficient/effective? Well, we're being nice to the system
5798 * by cleaning up our inodes proactively so they can be reaped
5799 * without I/O. But we are potentially leaving up to five seconds'
5800 * worth of inodes floating about which prune_icache wants us to
5801 * write out. One way to fix that would be to get prune_icache()
5802 * to do a write_super() to free up some memory. It has the desired
5805 int ext4_mark_inode_dirty(handle_t
*handle
, struct inode
*inode
)
5807 struct ext4_iloc iloc
;
5808 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
5809 static unsigned int mnt_count
;
5813 err
= ext4_reserve_inode_write(handle
, inode
, &iloc
);
5814 if (ext4_handle_valid(handle
) &&
5815 EXT4_I(inode
)->i_extra_isize
< sbi
->s_want_extra_isize
&&
5816 !ext4_test_inode_state(inode
, EXT4_STATE_NO_EXPAND
)) {
5818 * We need extra buffer credits since we may write into EA block
5819 * with this same handle. If journal_extend fails, then it will
5820 * only result in a minor loss of functionality for that inode.
5821 * If this is felt to be critical, then e2fsck should be run to
5822 * force a large enough s_min_extra_isize.
5824 if ((jbd2_journal_extend(handle
,
5825 EXT4_DATA_TRANS_BLOCKS(inode
->i_sb
))) == 0) {
5826 ret
= ext4_expand_extra_isize(inode
,
5827 sbi
->s_want_extra_isize
,
5830 ext4_set_inode_state(inode
,
5831 EXT4_STATE_NO_EXPAND
);
5833 le16_to_cpu(sbi
->s_es
->s_mnt_count
)) {
5834 ext4_warning(inode
->i_sb
,
5835 "Unable to expand inode %lu. Delete"
5836 " some EAs or run e2fsck.",
5839 le16_to_cpu(sbi
->s_es
->s_mnt_count
);
5845 err
= ext4_mark_iloc_dirty(handle
, inode
, &iloc
);
5850 * ext4_dirty_inode() is called from __mark_inode_dirty()
5852 * We're really interested in the case where a file is being extended.
5853 * i_size has been changed by generic_commit_write() and we thus need
5854 * to include the updated inode in the current transaction.
5856 * Also, dquot_alloc_block() will always dirty the inode when blocks
5857 * are allocated to the file.
5859 * If the inode is marked synchronous, we don't honour that here - doing
5860 * so would cause a commit on atime updates, which we don't bother doing.
5861 * We handle synchronous inodes at the highest possible level.
5863 void ext4_dirty_inode(struct inode
*inode
)
5867 handle
= ext4_journal_start(inode
, 2);
5871 ext4_mark_inode_dirty(handle
, inode
);
5873 ext4_journal_stop(handle
);
5879 int ext4_change_inode_journal_flag(struct inode
*inode
, int val
)
5886 * We have to be very careful here: changing a data block's
5887 * journaling status dynamically is dangerous. If we write a
5888 * data block to the journal, change the status and then delete
5889 * that block, we risk forgetting to revoke the old log record
5890 * from the journal and so a subsequent replay can corrupt data.
5891 * So, first we make sure that the journal is empty and that
5892 * nobody is changing anything.
5895 journal
= EXT4_JOURNAL(inode
);
5898 if (is_journal_aborted(journal
))
5901 jbd2_journal_lock_updates(journal
);
5902 jbd2_journal_flush(journal
);
5905 * OK, there are no updates running now, and all cached data is
5906 * synced to disk. We are now in a completely consistent state
5907 * which doesn't have anything in the journal, and we know that
5908 * no filesystem updates are running, so it is safe to modify
5909 * the inode's in-core data-journaling state flag now.
5913 ext4_set_inode_flag(inode
, EXT4_INODE_JOURNAL_DATA
);
5915 ext4_clear_inode_flag(inode
, EXT4_INODE_JOURNAL_DATA
);
5916 ext4_set_aops(inode
);
5918 jbd2_journal_unlock_updates(journal
);
5920 /* Finally we can mark the inode as dirty. */
5922 handle
= ext4_journal_start(inode
, 1);
5924 return PTR_ERR(handle
);
5926 err
= ext4_mark_inode_dirty(handle
, inode
);
5927 ext4_handle_sync(handle
);
5928 ext4_journal_stop(handle
);
5929 ext4_std_error(inode
->i_sb
, err
);
5934 static int ext4_bh_unmapped(handle_t
*handle
, struct buffer_head
*bh
)
5936 return !buffer_mapped(bh
);
5939 int ext4_page_mkwrite(struct vm_area_struct
*vma
, struct vm_fault
*vmf
)
5941 struct page
*page
= vmf
->page
;
5946 struct file
*file
= vma
->vm_file
;
5947 struct inode
*inode
= file
->f_path
.dentry
->d_inode
;
5948 struct address_space
*mapping
= inode
->i_mapping
;
5951 * Get i_alloc_sem to stop truncates messing with the inode. We cannot
5952 * get i_mutex because we are already holding mmap_sem.
5954 down_read(&inode
->i_alloc_sem
);
5955 size
= i_size_read(inode
);
5956 if (page
->mapping
!= mapping
|| size
<= page_offset(page
)
5957 || !PageUptodate(page
)) {
5958 /* page got truncated from under us? */
5962 if (PageMappedToDisk(page
))
5965 if (page
->index
== size
>> PAGE_CACHE_SHIFT
)
5966 len
= size
& ~PAGE_CACHE_MASK
;
5968 len
= PAGE_CACHE_SIZE
;
5972 * return if we have all the buffers mapped. This avoid
5973 * the need to call write_begin/write_end which does a
5974 * journal_start/journal_stop which can block and take
5977 if (page_has_buffers(page
)) {
5978 if (!walk_page_buffers(NULL
, page_buffers(page
), 0, len
, NULL
,
5979 ext4_bh_unmapped
)) {
5986 * OK, we need to fill the hole... Do write_begin write_end
5987 * to do block allocation/reservation.We are not holding
5988 * inode.i__mutex here. That allow * parallel write_begin,
5989 * write_end call. lock_page prevent this from happening
5990 * on the same page though
5992 ret
= mapping
->a_ops
->write_begin(file
, mapping
, page_offset(page
),
5993 len
, AOP_FLAG_UNINTERRUPTIBLE
, &page
, &fsdata
);
5996 ret
= mapping
->a_ops
->write_end(file
, mapping
, page_offset(page
),
5997 len
, len
, page
, fsdata
);
6003 ret
= VM_FAULT_SIGBUS
;
6004 up_read(&inode
->i_alloc_sem
);