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 "ext4_jbd2.h"
43 #include "ext4_extents.h"
45 #define MPAGE_DA_EXTENT_TAIL 0x01
47 static inline int ext4_begin_ordered_truncate(struct inode
*inode
,
50 return jbd2_journal_begin_ordered_truncate(
51 EXT4_SB(inode
->i_sb
)->s_journal
,
52 &EXT4_I(inode
)->jinode
,
56 static void ext4_invalidatepage(struct page
*page
, unsigned long offset
);
59 * Test whether an inode is a fast symlink.
61 static int ext4_inode_is_fast_symlink(struct inode
*inode
)
63 int ea_blocks
= EXT4_I(inode
)->i_file_acl
?
64 (inode
->i_sb
->s_blocksize
>> 9) : 0;
66 return (S_ISLNK(inode
->i_mode
) && inode
->i_blocks
- ea_blocks
== 0);
70 * The ext4 forget function must perform a revoke if we are freeing data
71 * which has been journaled. Metadata (eg. indirect blocks) must be
72 * revoked in all cases.
74 * "bh" may be NULL: a metadata block may have been freed from memory
75 * but there may still be a record of it in the journal, and that record
76 * still needs to be revoked.
78 * If the handle isn't valid we're not journaling so there's nothing to do.
80 int ext4_forget(handle_t
*handle
, int is_metadata
, struct inode
*inode
,
81 struct buffer_head
*bh
, ext4_fsblk_t blocknr
)
85 if (!ext4_handle_valid(handle
))
90 BUFFER_TRACE(bh
, "enter");
92 jbd_debug(4, "forgetting bh %p: is_metadata = %d, mode %o, "
94 bh
, is_metadata
, inode
->i_mode
,
95 test_opt(inode
->i_sb
, DATA_FLAGS
));
97 /* Never use the revoke function if we are doing full data
98 * journaling: there is no need to, and a V1 superblock won't
99 * support it. Otherwise, only skip the revoke on un-journaled
102 if (test_opt(inode
->i_sb
, DATA_FLAGS
) == EXT4_MOUNT_JOURNAL_DATA
||
103 (!is_metadata
&& !ext4_should_journal_data(inode
))) {
105 BUFFER_TRACE(bh
, "call jbd2_journal_forget");
106 return ext4_journal_forget(handle
, bh
);
112 * data!=journal && (is_metadata || should_journal_data(inode))
114 BUFFER_TRACE(bh
, "call ext4_journal_revoke");
115 err
= ext4_journal_revoke(handle
, blocknr
, bh
);
117 ext4_abort(inode
->i_sb
, __func__
,
118 "error %d when attempting revoke", err
);
119 BUFFER_TRACE(bh
, "exit");
124 * Work out how many blocks we need to proceed with the next chunk of a
125 * truncate transaction.
127 static unsigned long blocks_for_truncate(struct inode
*inode
)
131 needed
= inode
->i_blocks
>> (inode
->i_sb
->s_blocksize_bits
- 9);
133 /* Give ourselves just enough room to cope with inodes in which
134 * i_blocks is corrupt: we've seen disk corruptions in the past
135 * which resulted in random data in an inode which looked enough
136 * like a regular file for ext4 to try to delete it. Things
137 * will go a bit crazy if that happens, but at least we should
138 * try not to panic the whole kernel. */
142 /* But we need to bound the transaction so we don't overflow the
144 if (needed
> EXT4_MAX_TRANS_DATA
)
145 needed
= EXT4_MAX_TRANS_DATA
;
147 return EXT4_DATA_TRANS_BLOCKS(inode
->i_sb
) + needed
;
151 * Truncate transactions can be complex and absolutely huge. So we need to
152 * be able to restart the transaction at a conventient checkpoint to make
153 * sure we don't overflow the journal.
155 * start_transaction gets us a new handle for a truncate transaction,
156 * and extend_transaction tries to extend the existing one a bit. If
157 * extend fails, we need to propagate the failure up and restart the
158 * transaction in the top-level truncate loop. --sct
160 static handle_t
*start_transaction(struct inode
*inode
)
164 result
= ext4_journal_start(inode
, blocks_for_truncate(inode
));
168 ext4_std_error(inode
->i_sb
, PTR_ERR(result
));
173 * Try to extend this transaction for the purposes of truncation.
175 * Returns 0 if we managed to create more room. If we can't create more
176 * room, and the transaction must be restarted we return 1.
178 static int try_to_extend_transaction(handle_t
*handle
, struct inode
*inode
)
180 if (!ext4_handle_valid(handle
))
182 if (ext4_handle_has_enough_credits(handle
, EXT4_RESERVE_TRANS_BLOCKS
+1))
184 if (!ext4_journal_extend(handle
, blocks_for_truncate(inode
)))
190 * Restart the transaction associated with *handle. This does a commit,
191 * so before we call here everything must be consistently dirtied against
194 static int ext4_journal_test_restart(handle_t
*handle
, struct inode
*inode
)
196 BUG_ON(EXT4_JOURNAL(inode
) == NULL
);
197 jbd_debug(2, "restarting handle %p\n", handle
);
198 return ext4_journal_restart(handle
, blocks_for_truncate(inode
));
202 * Called at the last iput() if i_nlink is zero.
204 void ext4_delete_inode(struct inode
*inode
)
209 if (ext4_should_order_data(inode
))
210 ext4_begin_ordered_truncate(inode
, 0);
211 truncate_inode_pages(&inode
->i_data
, 0);
213 if (is_bad_inode(inode
))
216 handle
= ext4_journal_start(inode
, blocks_for_truncate(inode
)+3);
217 if (IS_ERR(handle
)) {
218 ext4_std_error(inode
->i_sb
, PTR_ERR(handle
));
220 * If we're going to skip the normal cleanup, we still need to
221 * make sure that the in-core orphan linked list is properly
224 ext4_orphan_del(NULL
, inode
);
229 ext4_handle_sync(handle
);
231 err
= ext4_mark_inode_dirty(handle
, inode
);
233 ext4_warning(inode
->i_sb
, __func__
,
234 "couldn't mark inode dirty (err %d)", err
);
238 ext4_truncate(inode
);
241 * ext4_ext_truncate() doesn't reserve any slop when it
242 * restarts journal transactions; therefore there may not be
243 * enough credits left in the handle to remove the inode from
244 * the orphan list and set the dtime field.
246 if (!ext4_handle_has_enough_credits(handle
, 3)) {
247 err
= ext4_journal_extend(handle
, 3);
249 err
= ext4_journal_restart(handle
, 3);
251 ext4_warning(inode
->i_sb
, __func__
,
252 "couldn't extend journal (err %d)", err
);
254 ext4_journal_stop(handle
);
260 * Kill off the orphan record which ext4_truncate created.
261 * AKPM: I think this can be inside the above `if'.
262 * Note that ext4_orphan_del() has to be able to cope with the
263 * deletion of a non-existent orphan - this is because we don't
264 * know if ext4_truncate() actually created an orphan record.
265 * (Well, we could do this if we need to, but heck - it works)
267 ext4_orphan_del(handle
, inode
);
268 EXT4_I(inode
)->i_dtime
= get_seconds();
271 * One subtle ordering requirement: if anything has gone wrong
272 * (transaction abort, IO errors, whatever), then we can still
273 * do these next steps (the fs will already have been marked as
274 * having errors), but we can't free the inode if the mark_dirty
277 if (ext4_mark_inode_dirty(handle
, inode
))
278 /* If that failed, just do the required in-core inode clear. */
281 ext4_free_inode(handle
, inode
);
282 ext4_journal_stop(handle
);
285 clear_inode(inode
); /* We must guarantee clearing of inode... */
291 struct buffer_head
*bh
;
294 static inline void add_chain(Indirect
*p
, struct buffer_head
*bh
, __le32
*v
)
296 p
->key
= *(p
->p
= v
);
301 * ext4_block_to_path - parse the block number into array of offsets
302 * @inode: inode in question (we are only interested in its superblock)
303 * @i_block: block number to be parsed
304 * @offsets: array to store the offsets in
305 * @boundary: set this non-zero if the referred-to block is likely to be
306 * followed (on disk) by an indirect block.
308 * To store the locations of file's data ext4 uses a data structure common
309 * for UNIX filesystems - tree of pointers anchored in the inode, with
310 * data blocks at leaves and indirect blocks in intermediate nodes.
311 * This function translates the block number into path in that tree -
312 * return value is the path length and @offsets[n] is the offset of
313 * pointer to (n+1)th node in the nth one. If @block is out of range
314 * (negative or too large) warning is printed and zero returned.
316 * Note: function doesn't find node addresses, so no IO is needed. All
317 * we need to know is the capacity of indirect blocks (taken from the
322 * Portability note: the last comparison (check that we fit into triple
323 * indirect block) is spelled differently, because otherwise on an
324 * architecture with 32-bit longs and 8Kb pages we might get into trouble
325 * if our filesystem had 8Kb blocks. We might use long long, but that would
326 * kill us on x86. Oh, well, at least the sign propagation does not matter -
327 * i_block would have to be negative in the very beginning, so we would not
331 static int ext4_block_to_path(struct inode
*inode
,
333 ext4_lblk_t offsets
[4], int *boundary
)
335 int ptrs
= EXT4_ADDR_PER_BLOCK(inode
->i_sb
);
336 int ptrs_bits
= EXT4_ADDR_PER_BLOCK_BITS(inode
->i_sb
);
337 const long direct_blocks
= EXT4_NDIR_BLOCKS
,
338 indirect_blocks
= ptrs
,
339 double_blocks
= (1 << (ptrs_bits
* 2));
344 ext4_warning(inode
->i_sb
, "ext4_block_to_path", "block < 0");
345 } else if (i_block
< direct_blocks
) {
346 offsets
[n
++] = i_block
;
347 final
= direct_blocks
;
348 } else if ((i_block
-= direct_blocks
) < indirect_blocks
) {
349 offsets
[n
++] = EXT4_IND_BLOCK
;
350 offsets
[n
++] = i_block
;
352 } else if ((i_block
-= indirect_blocks
) < double_blocks
) {
353 offsets
[n
++] = EXT4_DIND_BLOCK
;
354 offsets
[n
++] = i_block
>> ptrs_bits
;
355 offsets
[n
++] = i_block
& (ptrs
- 1);
357 } else if (((i_block
-= double_blocks
) >> (ptrs_bits
* 2)) < ptrs
) {
358 offsets
[n
++] = EXT4_TIND_BLOCK
;
359 offsets
[n
++] = i_block
>> (ptrs_bits
* 2);
360 offsets
[n
++] = (i_block
>> ptrs_bits
) & (ptrs
- 1);
361 offsets
[n
++] = i_block
& (ptrs
- 1);
364 ext4_warning(inode
->i_sb
, "ext4_block_to_path",
365 "block %lu > max in inode %lu",
366 i_block
+ direct_blocks
+
367 indirect_blocks
+ double_blocks
, inode
->i_ino
);
370 *boundary
= final
- 1 - (i_block
& (ptrs
- 1));
375 * ext4_get_branch - read the chain of indirect blocks leading to data
376 * @inode: inode in question
377 * @depth: depth of the chain (1 - direct pointer, etc.)
378 * @offsets: offsets of pointers in inode/indirect blocks
379 * @chain: place to store the result
380 * @err: here we store the error value
382 * Function fills the array of triples <key, p, bh> and returns %NULL
383 * if everything went OK or the pointer to the last filled triple
384 * (incomplete one) otherwise. Upon the return chain[i].key contains
385 * the number of (i+1)-th block in the chain (as it is stored in memory,
386 * i.e. little-endian 32-bit), chain[i].p contains the address of that
387 * number (it points into struct inode for i==0 and into the bh->b_data
388 * for i>0) and chain[i].bh points to the buffer_head of i-th indirect
389 * block for i>0 and NULL for i==0. In other words, it holds the block
390 * numbers of the chain, addresses they were taken from (and where we can
391 * verify that chain did not change) and buffer_heads hosting these
394 * Function stops when it stumbles upon zero pointer (absent block)
395 * (pointer to last triple returned, *@err == 0)
396 * or when it gets an IO error reading an indirect block
397 * (ditto, *@err == -EIO)
398 * or when it reads all @depth-1 indirect blocks successfully and finds
399 * the whole chain, all way to the data (returns %NULL, *err == 0).
401 * Need to be called with
402 * down_read(&EXT4_I(inode)->i_data_sem)
404 static Indirect
*ext4_get_branch(struct inode
*inode
, int depth
,
405 ext4_lblk_t
*offsets
,
406 Indirect chain
[4], int *err
)
408 struct super_block
*sb
= inode
->i_sb
;
410 struct buffer_head
*bh
;
413 /* i_data is not going away, no lock needed */
414 add_chain(chain
, NULL
, EXT4_I(inode
)->i_data
+ *offsets
);
418 bh
= sb_bread(sb
, le32_to_cpu(p
->key
));
421 add_chain(++p
, bh
, (__le32
*)bh
->b_data
+ *++offsets
);
435 * ext4_find_near - find a place for allocation with sufficient locality
437 * @ind: descriptor of indirect block.
439 * This function returns the preferred place for block allocation.
440 * It is used when heuristic for sequential allocation fails.
442 * + if there is a block to the left of our position - allocate near it.
443 * + if pointer will live in indirect block - allocate near that block.
444 * + if pointer will live in inode - allocate in the same
447 * In the latter case we colour the starting block by the callers PID to
448 * prevent it from clashing with concurrent allocations for a different inode
449 * in the same block group. The PID is used here so that functionally related
450 * files will be close-by on-disk.
452 * Caller must make sure that @ind is valid and will stay that way.
454 static ext4_fsblk_t
ext4_find_near(struct inode
*inode
, Indirect
*ind
)
456 struct ext4_inode_info
*ei
= EXT4_I(inode
);
457 __le32
*start
= ind
->bh
? (__le32
*) ind
->bh
->b_data
: ei
->i_data
;
459 ext4_fsblk_t bg_start
;
460 ext4_fsblk_t last_block
;
461 ext4_grpblk_t colour
;
463 /* Try to find previous block */
464 for (p
= ind
->p
- 1; p
>= start
; p
--) {
466 return le32_to_cpu(*p
);
469 /* No such thing, so let's try location of indirect block */
471 return ind
->bh
->b_blocknr
;
474 * It is going to be referred to from the inode itself? OK, just put it
475 * into the same cylinder group then.
477 bg_start
= ext4_group_first_block_no(inode
->i_sb
, ei
->i_block_group
);
478 last_block
= ext4_blocks_count(EXT4_SB(inode
->i_sb
)->s_es
) - 1;
480 if (bg_start
+ EXT4_BLOCKS_PER_GROUP(inode
->i_sb
) <= last_block
)
481 colour
= (current
->pid
% 16) *
482 (EXT4_BLOCKS_PER_GROUP(inode
->i_sb
) / 16);
484 colour
= (current
->pid
% 16) * ((last_block
- bg_start
) / 16);
485 return bg_start
+ colour
;
489 * ext4_find_goal - find a preferred place for allocation.
491 * @block: block we want
492 * @partial: pointer to the last triple within a chain
494 * Normally this function find the preferred place for block allocation,
497 static ext4_fsblk_t
ext4_find_goal(struct inode
*inode
, ext4_lblk_t block
,
501 * XXX need to get goal block from mballoc's data structures
504 return ext4_find_near(inode
, partial
);
508 * ext4_blks_to_allocate: Look up the block map and count the number
509 * of direct blocks need to be allocated for the given branch.
511 * @branch: chain of indirect blocks
512 * @k: number of blocks need for indirect blocks
513 * @blks: number of data blocks to be mapped.
514 * @blocks_to_boundary: the offset in the indirect block
516 * return the total number of blocks to be allocate, including the
517 * direct and indirect blocks.
519 static int ext4_blks_to_allocate(Indirect
*branch
, int k
, unsigned int blks
,
520 int blocks_to_boundary
)
522 unsigned int count
= 0;
525 * Simple case, [t,d]Indirect block(s) has not allocated yet
526 * then it's clear blocks on that path have not allocated
529 /* right now we don't handle cross boundary allocation */
530 if (blks
< blocks_to_boundary
+ 1)
533 count
+= blocks_to_boundary
+ 1;
538 while (count
< blks
&& count
<= blocks_to_boundary
&&
539 le32_to_cpu(*(branch
[0].p
+ count
)) == 0) {
546 * ext4_alloc_blocks: multiple allocate blocks needed for a branch
547 * @indirect_blks: the number of blocks need to allocate for indirect
550 * @new_blocks: on return it will store the new block numbers for
551 * the indirect blocks(if needed) and the first direct block,
552 * @blks: on return it will store the total number of allocated
555 static int ext4_alloc_blocks(handle_t
*handle
, struct inode
*inode
,
556 ext4_lblk_t iblock
, ext4_fsblk_t goal
,
557 int indirect_blks
, int blks
,
558 ext4_fsblk_t new_blocks
[4], int *err
)
560 struct ext4_allocation_request ar
;
562 unsigned long count
= 0, blk_allocated
= 0;
564 ext4_fsblk_t current_block
= 0;
568 * Here we try to allocate the requested multiple blocks at once,
569 * on a best-effort basis.
570 * To build a branch, we should allocate blocks for
571 * the indirect blocks(if not allocated yet), and at least
572 * the first direct block of this branch. That's the
573 * minimum number of blocks need to allocate(required)
575 /* first we try to allocate the indirect blocks */
576 target
= indirect_blks
;
579 /* allocating blocks for indirect blocks and direct blocks */
580 current_block
= ext4_new_meta_blocks(handle
, inode
,
586 /* allocate blocks for indirect blocks */
587 while (index
< indirect_blks
&& count
) {
588 new_blocks
[index
++] = current_block
++;
593 * save the new block number
594 * for the first direct block
596 new_blocks
[index
] = current_block
;
597 printk(KERN_INFO
"%s returned more blocks than "
598 "requested\n", __func__
);
604 target
= blks
- count
;
605 blk_allocated
= count
;
608 /* Now allocate data blocks */
609 memset(&ar
, 0, sizeof(ar
));
614 if (S_ISREG(inode
->i_mode
))
615 /* enable in-core preallocation only for regular files */
616 ar
.flags
= EXT4_MB_HINT_DATA
;
618 current_block
= ext4_mb_new_blocks(handle
, &ar
, err
);
620 if (*err
&& (target
== blks
)) {
622 * if the allocation failed and we didn't allocate
628 if (target
== blks
) {
630 * save the new block number
631 * for the first direct block
633 new_blocks
[index
] = current_block
;
635 blk_allocated
+= ar
.len
;
638 /* total number of blocks allocated for direct blocks */
643 for (i
= 0; i
< index
; i
++)
644 ext4_free_blocks(handle
, inode
, new_blocks
[i
], 1, 0);
649 * ext4_alloc_branch - allocate and set up a chain of blocks.
651 * @indirect_blks: number of allocated indirect blocks
652 * @blks: number of allocated direct blocks
653 * @offsets: offsets (in the blocks) to store the pointers to next.
654 * @branch: place to store the chain in.
656 * This function allocates blocks, zeroes out all but the last one,
657 * links them into chain and (if we are synchronous) writes them to disk.
658 * In other words, it prepares a branch that can be spliced onto the
659 * inode. It stores the information about that chain in the branch[], in
660 * the same format as ext4_get_branch() would do. We are calling it after
661 * we had read the existing part of chain and partial points to the last
662 * triple of that (one with zero ->key). Upon the exit we have the same
663 * picture as after the successful ext4_get_block(), except that in one
664 * place chain is disconnected - *branch->p is still zero (we did not
665 * set the last link), but branch->key contains the number that should
666 * be placed into *branch->p to fill that gap.
668 * If allocation fails we free all blocks we've allocated (and forget
669 * their buffer_heads) and return the error value the from failed
670 * ext4_alloc_block() (normally -ENOSPC). Otherwise we set the chain
671 * as described above and return 0.
673 static int ext4_alloc_branch(handle_t
*handle
, struct inode
*inode
,
674 ext4_lblk_t iblock
, int indirect_blks
,
675 int *blks
, ext4_fsblk_t goal
,
676 ext4_lblk_t
*offsets
, Indirect
*branch
)
678 int blocksize
= inode
->i_sb
->s_blocksize
;
681 struct buffer_head
*bh
;
683 ext4_fsblk_t new_blocks
[4];
684 ext4_fsblk_t current_block
;
686 num
= ext4_alloc_blocks(handle
, inode
, iblock
, goal
, indirect_blks
,
687 *blks
, new_blocks
, &err
);
691 branch
[0].key
= cpu_to_le32(new_blocks
[0]);
693 * metadata blocks and data blocks are allocated.
695 for (n
= 1; n
<= indirect_blks
; n
++) {
697 * Get buffer_head for parent block, zero it out
698 * and set the pointer to new one, then send
701 bh
= sb_getblk(inode
->i_sb
, new_blocks
[n
-1]);
704 BUFFER_TRACE(bh
, "call get_create_access");
705 err
= ext4_journal_get_create_access(handle
, bh
);
712 memset(bh
->b_data
, 0, blocksize
);
713 branch
[n
].p
= (__le32
*) bh
->b_data
+ offsets
[n
];
714 branch
[n
].key
= cpu_to_le32(new_blocks
[n
]);
715 *branch
[n
].p
= branch
[n
].key
;
716 if (n
== indirect_blks
) {
717 current_block
= new_blocks
[n
];
719 * End of chain, update the last new metablock of
720 * the chain to point to the new allocated
721 * data blocks numbers
723 for (i
=1; i
< num
; i
++)
724 *(branch
[n
].p
+ i
) = cpu_to_le32(++current_block
);
726 BUFFER_TRACE(bh
, "marking uptodate");
727 set_buffer_uptodate(bh
);
730 BUFFER_TRACE(bh
, "call ext4_handle_dirty_metadata");
731 err
= ext4_handle_dirty_metadata(handle
, inode
, bh
);
738 /* Allocation failed, free what we already allocated */
739 for (i
= 1; i
<= n
; i
++) {
740 BUFFER_TRACE(branch
[i
].bh
, "call jbd2_journal_forget");
741 ext4_journal_forget(handle
, branch
[i
].bh
);
743 for (i
= 0; i
< indirect_blks
; i
++)
744 ext4_free_blocks(handle
, inode
, new_blocks
[i
], 1, 0);
746 ext4_free_blocks(handle
, inode
, new_blocks
[i
], num
, 0);
752 * ext4_splice_branch - splice the allocated branch onto inode.
754 * @block: (logical) number of block we are adding
755 * @chain: chain of indirect blocks (with a missing link - see
757 * @where: location of missing link
758 * @num: number of indirect blocks we are adding
759 * @blks: number of direct blocks we are adding
761 * This function fills the missing link and does all housekeeping needed in
762 * inode (->i_blocks, etc.). In case of success we end up with the full
763 * chain to new block and return 0.
765 static int ext4_splice_branch(handle_t
*handle
, struct inode
*inode
,
766 ext4_lblk_t block
, Indirect
*where
, int num
, int blks
)
770 ext4_fsblk_t current_block
;
773 * If we're splicing into a [td]indirect block (as opposed to the
774 * inode) then we need to get write access to the [td]indirect block
778 BUFFER_TRACE(where
->bh
, "get_write_access");
779 err
= ext4_journal_get_write_access(handle
, where
->bh
);
785 *where
->p
= where
->key
;
788 * Update the host buffer_head or inode to point to more just allocated
789 * direct blocks blocks
791 if (num
== 0 && blks
> 1) {
792 current_block
= le32_to_cpu(where
->key
) + 1;
793 for (i
= 1; i
< blks
; i
++)
794 *(where
->p
+ i
) = cpu_to_le32(current_block
++);
797 /* We are done with atomic stuff, now do the rest of housekeeping */
799 inode
->i_ctime
= ext4_current_time(inode
);
800 ext4_mark_inode_dirty(handle
, inode
);
802 /* had we spliced it onto indirect block? */
805 * If we spliced it onto an indirect block, we haven't
806 * altered the inode. Note however that if it is being spliced
807 * onto an indirect block at the very end of the file (the
808 * file is growing) then we *will* alter the inode to reflect
809 * the new i_size. But that is not done here - it is done in
810 * generic_commit_write->__mark_inode_dirty->ext4_dirty_inode.
812 jbd_debug(5, "splicing indirect only\n");
813 BUFFER_TRACE(where
->bh
, "call ext4_handle_dirty_metadata");
814 err
= ext4_handle_dirty_metadata(handle
, inode
, where
->bh
);
819 * OK, we spliced it into the inode itself on a direct block.
820 * Inode was dirtied above.
822 jbd_debug(5, "splicing direct\n");
827 for (i
= 1; i
<= num
; i
++) {
828 BUFFER_TRACE(where
[i
].bh
, "call jbd2_journal_forget");
829 ext4_journal_forget(handle
, where
[i
].bh
);
830 ext4_free_blocks(handle
, inode
,
831 le32_to_cpu(where
[i
-1].key
), 1, 0);
833 ext4_free_blocks(handle
, inode
, le32_to_cpu(where
[num
].key
), blks
, 0);
839 * Allocation strategy is simple: if we have to allocate something, we will
840 * have to go the whole way to leaf. So let's do it before attaching anything
841 * to tree, set linkage between the newborn blocks, write them if sync is
842 * required, recheck the path, free and repeat if check fails, otherwise
843 * set the last missing link (that will protect us from any truncate-generated
844 * removals - all blocks on the path are immune now) and possibly force the
845 * write on the parent block.
846 * That has a nice additional property: no special recovery from the failed
847 * allocations is needed - we simply release blocks and do not touch anything
848 * reachable from inode.
850 * `handle' can be NULL if create == 0.
852 * return > 0, # of blocks mapped or allocated.
853 * return = 0, if plain lookup failed.
854 * return < 0, error case.
857 * Need to be called with
858 * down_read(&EXT4_I(inode)->i_data_sem) if not allocating file system block
859 * (ie, create is zero). Otherwise down_write(&EXT4_I(inode)->i_data_sem)
861 static int ext4_get_blocks_handle(handle_t
*handle
, struct inode
*inode
,
862 ext4_lblk_t iblock
, unsigned int maxblocks
,
863 struct buffer_head
*bh_result
,
864 int create
, int extend_disksize
)
867 ext4_lblk_t offsets
[4];
872 int blocks_to_boundary
= 0;
874 struct ext4_inode_info
*ei
= EXT4_I(inode
);
876 ext4_fsblk_t first_block
= 0;
880 J_ASSERT(!(EXT4_I(inode
)->i_flags
& EXT4_EXTENTS_FL
));
881 J_ASSERT(handle
!= NULL
|| create
== 0);
882 depth
= ext4_block_to_path(inode
, iblock
, offsets
,
883 &blocks_to_boundary
);
888 partial
= ext4_get_branch(inode
, depth
, offsets
, chain
, &err
);
890 /* Simplest case - block found, no allocation needed */
892 first_block
= le32_to_cpu(chain
[depth
- 1].key
);
893 clear_buffer_new(bh_result
);
896 while (count
< maxblocks
&& count
<= blocks_to_boundary
) {
899 blk
= le32_to_cpu(*(chain
[depth
-1].p
+ count
));
901 if (blk
== first_block
+ count
)
909 /* Next simple case - plain lookup or failed read of indirect block */
910 if (!create
|| err
== -EIO
)
914 * Okay, we need to do block allocation.
916 goal
= ext4_find_goal(inode
, iblock
, partial
);
918 /* the number of blocks need to allocate for [d,t]indirect blocks */
919 indirect_blks
= (chain
+ depth
) - partial
- 1;
922 * Next look up the indirect map to count the totoal number of
923 * direct blocks to allocate for this branch.
925 count
= ext4_blks_to_allocate(partial
, indirect_blks
,
926 maxblocks
, blocks_to_boundary
);
928 * Block out ext4_truncate while we alter the tree
930 err
= ext4_alloc_branch(handle
, inode
, iblock
, indirect_blks
,
932 offsets
+ (partial
- chain
), partial
);
935 * The ext4_splice_branch call will free and forget any buffers
936 * on the new chain if there is a failure, but that risks using
937 * up transaction credits, especially for bitmaps where the
938 * credits cannot be returned. Can we handle this somehow? We
939 * may need to return -EAGAIN upwards in the worst case. --sct
942 err
= ext4_splice_branch(handle
, inode
, iblock
,
943 partial
, indirect_blks
, count
);
945 * i_disksize growing is protected by i_data_sem. Don't forget to
946 * protect it if you're about to implement concurrent
947 * ext4_get_block() -bzzz
949 if (!err
&& extend_disksize
) {
950 disksize
= ((loff_t
) iblock
+ count
) << inode
->i_blkbits
;
951 if (disksize
> i_size_read(inode
))
952 disksize
= i_size_read(inode
);
953 if (disksize
> ei
->i_disksize
)
954 ei
->i_disksize
= disksize
;
959 set_buffer_new(bh_result
);
961 map_bh(bh_result
, inode
->i_sb
, le32_to_cpu(chain
[depth
-1].key
));
962 if (count
> blocks_to_boundary
)
963 set_buffer_boundary(bh_result
);
965 /* Clean up and exit */
966 partial
= chain
+ depth
- 1; /* the whole chain */
968 while (partial
> chain
) {
969 BUFFER_TRACE(partial
->bh
, "call brelse");
973 BUFFER_TRACE(bh_result
, "returned");
979 * Calculate the number of metadata blocks need to reserve
980 * to allocate @blocks for non extent file based file
982 static int ext4_indirect_calc_metadata_amount(struct inode
*inode
, int blocks
)
984 int icap
= EXT4_ADDR_PER_BLOCK(inode
->i_sb
);
985 int ind_blks
, dind_blks
, tind_blks
;
987 /* number of new indirect blocks needed */
988 ind_blks
= (blocks
+ icap
- 1) / icap
;
990 dind_blks
= (ind_blks
+ icap
- 1) / icap
;
994 return ind_blks
+ dind_blks
+ tind_blks
;
998 * Calculate the number of metadata blocks need to reserve
999 * to allocate given number of blocks
1001 static int ext4_calc_metadata_amount(struct inode
*inode
, int blocks
)
1006 if (EXT4_I(inode
)->i_flags
& EXT4_EXTENTS_FL
)
1007 return ext4_ext_calc_metadata_amount(inode
, blocks
);
1009 return ext4_indirect_calc_metadata_amount(inode
, blocks
);
1012 static void ext4_da_update_reserve_space(struct inode
*inode
, int used
)
1014 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
1015 int total
, mdb
, mdb_free
;
1017 spin_lock(&EXT4_I(inode
)->i_block_reservation_lock
);
1018 /* recalculate the number of metablocks still need to be reserved */
1019 total
= EXT4_I(inode
)->i_reserved_data_blocks
- used
;
1020 mdb
= ext4_calc_metadata_amount(inode
, total
);
1022 /* figure out how many metablocks to release */
1023 BUG_ON(mdb
> EXT4_I(inode
)->i_reserved_meta_blocks
);
1024 mdb_free
= EXT4_I(inode
)->i_reserved_meta_blocks
- mdb
;
1027 /* Account for allocated meta_blocks */
1028 mdb_free
-= EXT4_I(inode
)->i_allocated_meta_blocks
;
1030 /* update fs dirty blocks counter */
1031 percpu_counter_sub(&sbi
->s_dirtyblocks_counter
, mdb_free
);
1032 EXT4_I(inode
)->i_allocated_meta_blocks
= 0;
1033 EXT4_I(inode
)->i_reserved_meta_blocks
= mdb
;
1036 /* update per-inode reservations */
1037 BUG_ON(used
> EXT4_I(inode
)->i_reserved_data_blocks
);
1038 EXT4_I(inode
)->i_reserved_data_blocks
-= used
;
1039 spin_unlock(&EXT4_I(inode
)->i_block_reservation_lock
);
1042 * If we have done all the pending block allocations and if
1043 * there aren't any writers on the inode, we can discard the
1044 * inode's preallocations.
1046 if (!total
&& (atomic_read(&inode
->i_writecount
) == 0))
1047 ext4_discard_preallocations(inode
);
1051 * The ext4_get_blocks_wrap() function try to look up the requested blocks,
1052 * and returns if the blocks are already mapped.
1054 * Otherwise it takes the write lock of the i_data_sem and allocate blocks
1055 * and store the allocated blocks in the result buffer head and mark it
1058 * If file type is extents based, it will call ext4_ext_get_blocks(),
1059 * Otherwise, call with ext4_get_blocks_handle() to handle indirect mapping
1062 * On success, it returns the number of blocks being mapped or allocate.
1063 * if create==0 and the blocks are pre-allocated and uninitialized block,
1064 * the result buffer head is unmapped. If the create ==1, it will make sure
1065 * the buffer head is mapped.
1067 * It returns 0 if plain look up failed (blocks have not been allocated), in
1068 * that casem, buffer head is unmapped
1070 * It returns the error in case of allocation failure.
1072 int ext4_get_blocks_wrap(handle_t
*handle
, struct inode
*inode
, sector_t block
,
1073 unsigned int max_blocks
, struct buffer_head
*bh
,
1074 int create
, int extend_disksize
, int flag
)
1078 clear_buffer_mapped(bh
);
1081 * Try to see if we can get the block without requesting
1082 * for new file system block.
1084 down_read((&EXT4_I(inode
)->i_data_sem
));
1085 if (EXT4_I(inode
)->i_flags
& EXT4_EXTENTS_FL
) {
1086 retval
= ext4_ext_get_blocks(handle
, inode
, block
, max_blocks
,
1089 retval
= ext4_get_blocks_handle(handle
,
1090 inode
, block
, max_blocks
, bh
, 0, 0);
1092 up_read((&EXT4_I(inode
)->i_data_sem
));
1094 /* If it is only a block(s) look up */
1099 * Returns if the blocks have already allocated
1101 * Note that if blocks have been preallocated
1102 * ext4_ext_get_block() returns th create = 0
1103 * with buffer head unmapped.
1105 if (retval
> 0 && buffer_mapped(bh
))
1109 * New blocks allocate and/or writing to uninitialized extent
1110 * will possibly result in updating i_data, so we take
1111 * the write lock of i_data_sem, and call get_blocks()
1112 * with create == 1 flag.
1114 down_write((&EXT4_I(inode
)->i_data_sem
));
1117 * if the caller is from delayed allocation writeout path
1118 * we have already reserved fs blocks for allocation
1119 * let the underlying get_block() function know to
1120 * avoid double accounting
1123 EXT4_I(inode
)->i_delalloc_reserved_flag
= 1;
1125 * We need to check for EXT4 here because migrate
1126 * could have changed the inode type in between
1128 if (EXT4_I(inode
)->i_flags
& EXT4_EXTENTS_FL
) {
1129 retval
= ext4_ext_get_blocks(handle
, inode
, block
, max_blocks
,
1130 bh
, create
, extend_disksize
);
1132 retval
= ext4_get_blocks_handle(handle
, inode
, block
,
1133 max_blocks
, bh
, create
, extend_disksize
);
1135 if (retval
> 0 && buffer_new(bh
)) {
1137 * We allocated new blocks which will result in
1138 * i_data's format changing. Force the migrate
1139 * to fail by clearing migrate flags
1141 EXT4_I(inode
)->i_flags
= EXT4_I(inode
)->i_flags
&
1147 EXT4_I(inode
)->i_delalloc_reserved_flag
= 0;
1149 * Update reserved blocks/metadata blocks
1150 * after successful block allocation
1151 * which were deferred till now
1153 if ((retval
> 0) && buffer_delay(bh
))
1154 ext4_da_update_reserve_space(inode
, retval
);
1157 up_write((&EXT4_I(inode
)->i_data_sem
));
1161 /* Maximum number of blocks we map for direct IO at once. */
1162 #define DIO_MAX_BLOCKS 4096
1164 int ext4_get_block(struct inode
*inode
, sector_t iblock
,
1165 struct buffer_head
*bh_result
, int create
)
1167 handle_t
*handle
= ext4_journal_current_handle();
1168 int ret
= 0, started
= 0;
1169 unsigned max_blocks
= bh_result
->b_size
>> inode
->i_blkbits
;
1172 if (create
&& !handle
) {
1173 /* Direct IO write... */
1174 if (max_blocks
> DIO_MAX_BLOCKS
)
1175 max_blocks
= DIO_MAX_BLOCKS
;
1176 dio_credits
= ext4_chunk_trans_blocks(inode
, max_blocks
);
1177 handle
= ext4_journal_start(inode
, dio_credits
);
1178 if (IS_ERR(handle
)) {
1179 ret
= PTR_ERR(handle
);
1185 ret
= ext4_get_blocks_wrap(handle
, inode
, iblock
,
1186 max_blocks
, bh_result
, create
, 0, 0);
1188 bh_result
->b_size
= (ret
<< inode
->i_blkbits
);
1192 ext4_journal_stop(handle
);
1198 * `handle' can be NULL if create is zero
1200 struct buffer_head
*ext4_getblk(handle_t
*handle
, struct inode
*inode
,
1201 ext4_lblk_t block
, int create
, int *errp
)
1203 struct buffer_head dummy
;
1206 J_ASSERT(handle
!= NULL
|| create
== 0);
1209 dummy
.b_blocknr
= -1000;
1210 buffer_trace_init(&dummy
.b_history
);
1211 err
= ext4_get_blocks_wrap(handle
, inode
, block
, 1,
1212 &dummy
, create
, 1, 0);
1214 * ext4_get_blocks_handle() returns number of blocks
1215 * mapped. 0 in case of a HOLE.
1223 if (!err
&& buffer_mapped(&dummy
)) {
1224 struct buffer_head
*bh
;
1225 bh
= sb_getblk(inode
->i_sb
, dummy
.b_blocknr
);
1230 if (buffer_new(&dummy
)) {
1231 J_ASSERT(create
!= 0);
1232 J_ASSERT(handle
!= NULL
);
1235 * Now that we do not always journal data, we should
1236 * keep in mind whether this should always journal the
1237 * new buffer as metadata. For now, regular file
1238 * writes use ext4_get_block instead, so it's not a
1242 BUFFER_TRACE(bh
, "call get_create_access");
1243 fatal
= ext4_journal_get_create_access(handle
, bh
);
1244 if (!fatal
&& !buffer_uptodate(bh
)) {
1245 memset(bh
->b_data
, 0, inode
->i_sb
->s_blocksize
);
1246 set_buffer_uptodate(bh
);
1249 BUFFER_TRACE(bh
, "call ext4_handle_dirty_metadata");
1250 err
= ext4_handle_dirty_metadata(handle
, inode
, bh
);
1254 BUFFER_TRACE(bh
, "not a new buffer");
1267 struct buffer_head
*ext4_bread(handle_t
*handle
, struct inode
*inode
,
1268 ext4_lblk_t block
, int create
, int *err
)
1270 struct buffer_head
*bh
;
1272 bh
= ext4_getblk(handle
, inode
, block
, create
, err
);
1275 if (buffer_uptodate(bh
))
1277 ll_rw_block(READ_META
, 1, &bh
);
1279 if (buffer_uptodate(bh
))
1286 static int walk_page_buffers(handle_t
*handle
,
1287 struct buffer_head
*head
,
1291 int (*fn
)(handle_t
*handle
,
1292 struct buffer_head
*bh
))
1294 struct buffer_head
*bh
;
1295 unsigned block_start
, block_end
;
1296 unsigned blocksize
= head
->b_size
;
1298 struct buffer_head
*next
;
1300 for (bh
= head
, block_start
= 0;
1301 ret
== 0 && (bh
!= head
|| !block_start
);
1302 block_start
= block_end
, bh
= next
)
1304 next
= bh
->b_this_page
;
1305 block_end
= block_start
+ blocksize
;
1306 if (block_end
<= from
|| block_start
>= to
) {
1307 if (partial
&& !buffer_uptodate(bh
))
1311 err
= (*fn
)(handle
, bh
);
1319 * To preserve ordering, it is essential that the hole instantiation and
1320 * the data write be encapsulated in a single transaction. We cannot
1321 * close off a transaction and start a new one between the ext4_get_block()
1322 * and the commit_write(). So doing the jbd2_journal_start at the start of
1323 * prepare_write() is the right place.
1325 * Also, this function can nest inside ext4_writepage() ->
1326 * block_write_full_page(). In that case, we *know* that ext4_writepage()
1327 * has generated enough buffer credits to do the whole page. So we won't
1328 * block on the journal in that case, which is good, because the caller may
1331 * By accident, ext4 can be reentered when a transaction is open via
1332 * quota file writes. If we were to commit the transaction while thus
1333 * reentered, there can be a deadlock - we would be holding a quota
1334 * lock, and the commit would never complete if another thread had a
1335 * transaction open and was blocking on the quota lock - a ranking
1338 * So what we do is to rely on the fact that jbd2_journal_stop/journal_start
1339 * will _not_ run commit under these circumstances because handle->h_ref
1340 * is elevated. We'll still have enough credits for the tiny quotafile
1343 static int do_journal_get_write_access(handle_t
*handle
,
1344 struct buffer_head
*bh
)
1346 if (!buffer_mapped(bh
) || buffer_freed(bh
))
1348 return ext4_journal_get_write_access(handle
, bh
);
1351 static int ext4_write_begin(struct file
*file
, struct address_space
*mapping
,
1352 loff_t pos
, unsigned len
, unsigned flags
,
1353 struct page
**pagep
, void **fsdata
)
1355 struct inode
*inode
= mapping
->host
;
1356 int ret
, needed_blocks
= ext4_writepage_trans_blocks(inode
);
1363 trace_mark(ext4_write_begin
,
1364 "dev %s ino %lu pos %llu len %u flags %u",
1365 inode
->i_sb
->s_id
, inode
->i_ino
,
1366 (unsigned long long) pos
, len
, flags
);
1367 index
= pos
>> PAGE_CACHE_SHIFT
;
1368 from
= pos
& (PAGE_CACHE_SIZE
- 1);
1372 handle
= ext4_journal_start(inode
, needed_blocks
);
1373 if (IS_ERR(handle
)) {
1374 ret
= PTR_ERR(handle
);
1378 /* We cannot recurse into the filesystem as the transaction is already
1380 flags
|= AOP_FLAG_NOFS
;
1382 page
= grab_cache_page_write_begin(mapping
, index
, flags
);
1384 ext4_journal_stop(handle
);
1390 ret
= block_write_begin(file
, mapping
, pos
, len
, flags
, pagep
, fsdata
,
1393 if (!ret
&& ext4_should_journal_data(inode
)) {
1394 ret
= walk_page_buffers(handle
, page_buffers(page
),
1395 from
, to
, NULL
, do_journal_get_write_access
);
1400 ext4_journal_stop(handle
);
1401 page_cache_release(page
);
1403 * block_write_begin may have instantiated a few blocks
1404 * outside i_size. Trim these off again. Don't need
1405 * i_size_read because we hold i_mutex.
1407 if (pos
+ len
> inode
->i_size
)
1408 vmtruncate(inode
, inode
->i_size
);
1411 if (ret
== -ENOSPC
&& ext4_should_retry_alloc(inode
->i_sb
, &retries
))
1417 /* For write_end() in data=journal mode */
1418 static int write_end_fn(handle_t
*handle
, struct buffer_head
*bh
)
1420 if (!buffer_mapped(bh
) || buffer_freed(bh
))
1422 set_buffer_uptodate(bh
);
1423 return ext4_handle_dirty_metadata(handle
, NULL
, bh
);
1427 * We need to pick up the new inode size which generic_commit_write gave us
1428 * `file' can be NULL - eg, when called from page_symlink().
1430 * ext4 never places buffers on inode->i_mapping->private_list. metadata
1431 * buffers are managed internally.
1433 static int ext4_ordered_write_end(struct file
*file
,
1434 struct address_space
*mapping
,
1435 loff_t pos
, unsigned len
, unsigned copied
,
1436 struct page
*page
, void *fsdata
)
1438 handle_t
*handle
= ext4_journal_current_handle();
1439 struct inode
*inode
= mapping
->host
;
1442 trace_mark(ext4_ordered_write_end
,
1443 "dev %s ino %lu pos %llu len %u copied %u",
1444 inode
->i_sb
->s_id
, inode
->i_ino
,
1445 (unsigned long long) pos
, len
, copied
);
1446 ret
= ext4_jbd2_file_inode(handle
, inode
);
1451 new_i_size
= pos
+ copied
;
1452 if (new_i_size
> EXT4_I(inode
)->i_disksize
) {
1453 ext4_update_i_disksize(inode
, new_i_size
);
1454 /* We need to mark inode dirty even if
1455 * new_i_size is less that inode->i_size
1456 * bu greater than i_disksize.(hint delalloc)
1458 ext4_mark_inode_dirty(handle
, inode
);
1461 ret2
= generic_write_end(file
, mapping
, pos
, len
, copied
,
1467 ret2
= ext4_journal_stop(handle
);
1471 return ret
? ret
: copied
;
1474 static int ext4_writeback_write_end(struct file
*file
,
1475 struct address_space
*mapping
,
1476 loff_t pos
, unsigned len
, unsigned copied
,
1477 struct page
*page
, void *fsdata
)
1479 handle_t
*handle
= ext4_journal_current_handle();
1480 struct inode
*inode
= mapping
->host
;
1484 trace_mark(ext4_writeback_write_end
,
1485 "dev %s ino %lu pos %llu len %u copied %u",
1486 inode
->i_sb
->s_id
, inode
->i_ino
,
1487 (unsigned long long) pos
, len
, copied
);
1488 new_i_size
= pos
+ copied
;
1489 if (new_i_size
> EXT4_I(inode
)->i_disksize
) {
1490 ext4_update_i_disksize(inode
, new_i_size
);
1491 /* We need to mark inode dirty even if
1492 * new_i_size is less that inode->i_size
1493 * bu greater than i_disksize.(hint delalloc)
1495 ext4_mark_inode_dirty(handle
, inode
);
1498 ret2
= generic_write_end(file
, mapping
, pos
, len
, copied
,
1504 ret2
= ext4_journal_stop(handle
);
1508 return ret
? ret
: copied
;
1511 static int ext4_journalled_write_end(struct file
*file
,
1512 struct address_space
*mapping
,
1513 loff_t pos
, unsigned len
, unsigned copied
,
1514 struct page
*page
, void *fsdata
)
1516 handle_t
*handle
= ext4_journal_current_handle();
1517 struct inode
*inode
= mapping
->host
;
1523 trace_mark(ext4_journalled_write_end
,
1524 "dev %s ino %lu pos %llu len %u copied %u",
1525 inode
->i_sb
->s_id
, inode
->i_ino
,
1526 (unsigned long long) pos
, len
, copied
);
1527 from
= pos
& (PAGE_CACHE_SIZE
- 1);
1531 if (!PageUptodate(page
))
1533 page_zero_new_buffers(page
, from
+copied
, to
);
1536 ret
= walk_page_buffers(handle
, page_buffers(page
), from
,
1537 to
, &partial
, write_end_fn
);
1539 SetPageUptodate(page
);
1540 new_i_size
= pos
+ copied
;
1541 if (new_i_size
> inode
->i_size
)
1542 i_size_write(inode
, pos
+copied
);
1543 EXT4_I(inode
)->i_state
|= EXT4_STATE_JDATA
;
1544 if (new_i_size
> EXT4_I(inode
)->i_disksize
) {
1545 ext4_update_i_disksize(inode
, new_i_size
);
1546 ret2
= ext4_mark_inode_dirty(handle
, inode
);
1552 ret2
= ext4_journal_stop(handle
);
1555 page_cache_release(page
);
1557 return ret
? ret
: copied
;
1560 static int ext4_da_reserve_space(struct inode
*inode
, int nrblocks
)
1563 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
1564 unsigned long md_needed
, mdblocks
, total
= 0;
1567 * recalculate the amount of metadata blocks to reserve
1568 * in order to allocate nrblocks
1569 * worse case is one extent per block
1572 spin_lock(&EXT4_I(inode
)->i_block_reservation_lock
);
1573 total
= EXT4_I(inode
)->i_reserved_data_blocks
+ nrblocks
;
1574 mdblocks
= ext4_calc_metadata_amount(inode
, total
);
1575 BUG_ON(mdblocks
< EXT4_I(inode
)->i_reserved_meta_blocks
);
1577 md_needed
= mdblocks
- EXT4_I(inode
)->i_reserved_meta_blocks
;
1578 total
= md_needed
+ nrblocks
;
1580 if (ext4_claim_free_blocks(sbi
, total
)) {
1581 spin_unlock(&EXT4_I(inode
)->i_block_reservation_lock
);
1582 if (ext4_should_retry_alloc(inode
->i_sb
, &retries
)) {
1588 EXT4_I(inode
)->i_reserved_data_blocks
+= nrblocks
;
1589 EXT4_I(inode
)->i_reserved_meta_blocks
= mdblocks
;
1591 spin_unlock(&EXT4_I(inode
)->i_block_reservation_lock
);
1592 return 0; /* success */
1595 static void ext4_da_release_space(struct inode
*inode
, int to_free
)
1597 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
1598 int total
, mdb
, mdb_free
, release
;
1601 return; /* Nothing to release, exit */
1603 spin_lock(&EXT4_I(inode
)->i_block_reservation_lock
);
1605 if (!EXT4_I(inode
)->i_reserved_data_blocks
) {
1607 * if there is no reserved blocks, but we try to free some
1608 * then the counter is messed up somewhere.
1609 * but since this function is called from invalidate
1610 * page, it's harmless to return without any action
1612 printk(KERN_INFO
"ext4 delalloc try to release %d reserved "
1613 "blocks for inode %lu, but there is no reserved "
1614 "data blocks\n", to_free
, inode
->i_ino
);
1615 spin_unlock(&EXT4_I(inode
)->i_block_reservation_lock
);
1619 /* recalculate the number of metablocks still need to be reserved */
1620 total
= EXT4_I(inode
)->i_reserved_data_blocks
- to_free
;
1621 mdb
= ext4_calc_metadata_amount(inode
, total
);
1623 /* figure out how many metablocks to release */
1624 BUG_ON(mdb
> EXT4_I(inode
)->i_reserved_meta_blocks
);
1625 mdb_free
= EXT4_I(inode
)->i_reserved_meta_blocks
- mdb
;
1627 release
= to_free
+ mdb_free
;
1629 /* update fs dirty blocks counter for truncate case */
1630 percpu_counter_sub(&sbi
->s_dirtyblocks_counter
, release
);
1632 /* update per-inode reservations */
1633 BUG_ON(to_free
> EXT4_I(inode
)->i_reserved_data_blocks
);
1634 EXT4_I(inode
)->i_reserved_data_blocks
-= to_free
;
1636 BUG_ON(mdb
> EXT4_I(inode
)->i_reserved_meta_blocks
);
1637 EXT4_I(inode
)->i_reserved_meta_blocks
= mdb
;
1638 spin_unlock(&EXT4_I(inode
)->i_block_reservation_lock
);
1641 static void ext4_da_page_release_reservation(struct page
*page
,
1642 unsigned long offset
)
1645 struct buffer_head
*head
, *bh
;
1646 unsigned int curr_off
= 0;
1648 head
= page_buffers(page
);
1651 unsigned int next_off
= curr_off
+ bh
->b_size
;
1653 if ((offset
<= curr_off
) && (buffer_delay(bh
))) {
1655 clear_buffer_delay(bh
);
1657 curr_off
= next_off
;
1658 } while ((bh
= bh
->b_this_page
) != head
);
1659 ext4_da_release_space(page
->mapping
->host
, to_release
);
1663 * Delayed allocation stuff
1666 struct mpage_da_data
{
1667 struct inode
*inode
;
1668 struct buffer_head lbh
; /* extent of blocks */
1669 unsigned long first_page
, next_page
; /* extent of pages */
1670 get_block_t
*get_block
;
1671 struct writeback_control
*wbc
;
1678 * mpage_da_submit_io - walks through extent of pages and try to write
1679 * them with writepage() call back
1681 * @mpd->inode: inode
1682 * @mpd->first_page: first page of the extent
1683 * @mpd->next_page: page after the last page of the extent
1684 * @mpd->get_block: the filesystem's block mapper function
1686 * By the time mpage_da_submit_io() is called we expect all blocks
1687 * to be allocated. this may be wrong if allocation failed.
1689 * As pages are already locked by write_cache_pages(), we can't use it
1691 static int mpage_da_submit_io(struct mpage_da_data
*mpd
)
1694 struct pagevec pvec
;
1695 unsigned long index
, end
;
1696 int ret
= 0, err
, nr_pages
, i
;
1697 struct inode
*inode
= mpd
->inode
;
1698 struct address_space
*mapping
= inode
->i_mapping
;
1700 BUG_ON(mpd
->next_page
<= mpd
->first_page
);
1702 * We need to start from the first_page to the next_page - 1
1703 * to make sure we also write the mapped dirty buffer_heads.
1704 * If we look at mpd->lbh.b_blocknr we would only be looking
1705 * at the currently mapped buffer_heads.
1707 index
= mpd
->first_page
;
1708 end
= mpd
->next_page
- 1;
1710 pagevec_init(&pvec
, 0);
1711 while (index
<= end
) {
1712 nr_pages
= pagevec_lookup(&pvec
, mapping
, index
, PAGEVEC_SIZE
);
1715 for (i
= 0; i
< nr_pages
; i
++) {
1716 struct page
*page
= pvec
.pages
[i
];
1718 index
= page
->index
;
1723 BUG_ON(!PageLocked(page
));
1724 BUG_ON(PageWriteback(page
));
1726 pages_skipped
= mpd
->wbc
->pages_skipped
;
1727 err
= mapping
->a_ops
->writepage(page
, mpd
->wbc
);
1728 if (!err
&& (pages_skipped
== mpd
->wbc
->pages_skipped
))
1730 * have successfully written the page
1731 * without skipping the same
1733 mpd
->pages_written
++;
1735 * In error case, we have to continue because
1736 * remaining pages are still locked
1737 * XXX: unlock and re-dirty them?
1742 pagevec_release(&pvec
);
1748 * mpage_put_bnr_to_bhs - walk blocks and assign them actual numbers
1750 * @mpd->inode - inode to walk through
1751 * @exbh->b_blocknr - first block on a disk
1752 * @exbh->b_size - amount of space in bytes
1753 * @logical - first logical block to start assignment with
1755 * the function goes through all passed space and put actual disk
1756 * block numbers into buffer heads, dropping BH_Delay
1758 static void mpage_put_bnr_to_bhs(struct mpage_da_data
*mpd
, sector_t logical
,
1759 struct buffer_head
*exbh
)
1761 struct inode
*inode
= mpd
->inode
;
1762 struct address_space
*mapping
= inode
->i_mapping
;
1763 int blocks
= exbh
->b_size
>> inode
->i_blkbits
;
1764 sector_t pblock
= exbh
->b_blocknr
, cur_logical
;
1765 struct buffer_head
*head
, *bh
;
1767 struct pagevec pvec
;
1770 index
= logical
>> (PAGE_CACHE_SHIFT
- inode
->i_blkbits
);
1771 end
= (logical
+ blocks
- 1) >> (PAGE_CACHE_SHIFT
- inode
->i_blkbits
);
1772 cur_logical
= index
<< (PAGE_CACHE_SHIFT
- inode
->i_blkbits
);
1774 pagevec_init(&pvec
, 0);
1776 while (index
<= end
) {
1777 /* XXX: optimize tail */
1778 nr_pages
= pagevec_lookup(&pvec
, mapping
, index
, PAGEVEC_SIZE
);
1781 for (i
= 0; i
< nr_pages
; i
++) {
1782 struct page
*page
= pvec
.pages
[i
];
1784 index
= page
->index
;
1789 BUG_ON(!PageLocked(page
));
1790 BUG_ON(PageWriteback(page
));
1791 BUG_ON(!page_has_buffers(page
));
1793 bh
= page_buffers(page
);
1796 /* skip blocks out of the range */
1798 if (cur_logical
>= logical
)
1801 } while ((bh
= bh
->b_this_page
) != head
);
1804 if (cur_logical
>= logical
+ blocks
)
1806 if (buffer_delay(bh
)) {
1807 bh
->b_blocknr
= pblock
;
1808 clear_buffer_delay(bh
);
1809 bh
->b_bdev
= inode
->i_sb
->s_bdev
;
1810 } else if (buffer_unwritten(bh
)) {
1811 bh
->b_blocknr
= pblock
;
1812 clear_buffer_unwritten(bh
);
1813 set_buffer_mapped(bh
);
1815 bh
->b_bdev
= inode
->i_sb
->s_bdev
;
1816 } else if (buffer_mapped(bh
))
1817 BUG_ON(bh
->b_blocknr
!= pblock
);
1821 } while ((bh
= bh
->b_this_page
) != head
);
1823 pagevec_release(&pvec
);
1829 * __unmap_underlying_blocks - just a helper function to unmap
1830 * set of blocks described by @bh
1832 static inline void __unmap_underlying_blocks(struct inode
*inode
,
1833 struct buffer_head
*bh
)
1835 struct block_device
*bdev
= inode
->i_sb
->s_bdev
;
1838 blocks
= bh
->b_size
>> inode
->i_blkbits
;
1839 for (i
= 0; i
< blocks
; i
++)
1840 unmap_underlying_metadata(bdev
, bh
->b_blocknr
+ i
);
1843 static void ext4_da_block_invalidatepages(struct mpage_da_data
*mpd
,
1844 sector_t logical
, long blk_cnt
)
1848 struct pagevec pvec
;
1849 struct inode
*inode
= mpd
->inode
;
1850 struct address_space
*mapping
= inode
->i_mapping
;
1852 index
= logical
>> (PAGE_CACHE_SHIFT
- inode
->i_blkbits
);
1853 end
= (logical
+ blk_cnt
- 1) >>
1854 (PAGE_CACHE_SHIFT
- inode
->i_blkbits
);
1855 while (index
<= end
) {
1856 nr_pages
= pagevec_lookup(&pvec
, mapping
, index
, PAGEVEC_SIZE
);
1859 for (i
= 0; i
< nr_pages
; i
++) {
1860 struct page
*page
= pvec
.pages
[i
];
1861 index
= page
->index
;
1866 BUG_ON(!PageLocked(page
));
1867 BUG_ON(PageWriteback(page
));
1868 block_invalidatepage(page
, 0);
1869 ClearPageUptodate(page
);
1876 static void ext4_print_free_blocks(struct inode
*inode
)
1878 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
1879 printk(KERN_EMERG
"Total free blocks count %lld\n",
1880 ext4_count_free_blocks(inode
->i_sb
));
1881 printk(KERN_EMERG
"Free/Dirty block details\n");
1882 printk(KERN_EMERG
"free_blocks=%lld\n",
1883 (long long)percpu_counter_sum(&sbi
->s_freeblocks_counter
));
1884 printk(KERN_EMERG
"dirty_blocks=%lld\n",
1885 (long long)percpu_counter_sum(&sbi
->s_dirtyblocks_counter
));
1886 printk(KERN_EMERG
"Block reservation details\n");
1887 printk(KERN_EMERG
"i_reserved_data_blocks=%u\n",
1888 EXT4_I(inode
)->i_reserved_data_blocks
);
1889 printk(KERN_EMERG
"i_reserved_meta_blocks=%u\n",
1890 EXT4_I(inode
)->i_reserved_meta_blocks
);
1895 * mpage_da_map_blocks - go through given space
1897 * @mpd->lbh - bh describing space
1898 * @mpd->get_block - the filesystem's block mapper function
1900 * The function skips space we know is already mapped to disk blocks.
1903 static int mpage_da_map_blocks(struct mpage_da_data
*mpd
)
1906 struct buffer_head
new;
1907 struct buffer_head
*lbh
= &mpd
->lbh
;
1911 * We consider only non-mapped and non-allocated blocks
1913 if (buffer_mapped(lbh
) && !buffer_delay(lbh
))
1915 new.b_state
= lbh
->b_state
;
1917 new.b_size
= lbh
->b_size
;
1918 next
= lbh
->b_blocknr
;
1920 * If we didn't accumulate anything
1921 * to write simply return
1925 err
= mpd
->get_block(mpd
->inode
, next
, &new, 1);
1928 /* If get block returns with error
1929 * we simply return. Later writepage
1930 * will redirty the page and writepages
1931 * will find the dirty page again
1936 if (err
== -ENOSPC
&&
1937 ext4_count_free_blocks(mpd
->inode
->i_sb
)) {
1943 * get block failure will cause us
1944 * to loop in writepages. Because
1945 * a_ops->writepage won't be able to
1946 * make progress. The page will be redirtied
1947 * by writepage and writepages will again
1948 * try to write the same.
1950 printk(KERN_EMERG
"%s block allocation failed for inode %lu "
1951 "at logical offset %llu with max blocks "
1952 "%zd with error %d\n",
1953 __func__
, mpd
->inode
->i_ino
,
1954 (unsigned long long)next
,
1955 lbh
->b_size
>> mpd
->inode
->i_blkbits
, err
);
1956 printk(KERN_EMERG
"This should not happen.!! "
1957 "Data will be lost\n");
1958 if (err
== -ENOSPC
) {
1959 ext4_print_free_blocks(mpd
->inode
);
1961 /* invlaidate all the pages */
1962 ext4_da_block_invalidatepages(mpd
, next
,
1963 lbh
->b_size
>> mpd
->inode
->i_blkbits
);
1966 BUG_ON(new.b_size
== 0);
1968 if (buffer_new(&new))
1969 __unmap_underlying_blocks(mpd
->inode
, &new);
1972 * If blocks are delayed marked, we need to
1973 * put actual blocknr and drop delayed bit
1975 if (buffer_delay(lbh
) || buffer_unwritten(lbh
))
1976 mpage_put_bnr_to_bhs(mpd
, next
, &new);
1981 #define BH_FLAGS ((1 << BH_Uptodate) | (1 << BH_Mapped) | \
1982 (1 << BH_Delay) | (1 << BH_Unwritten))
1985 * mpage_add_bh_to_extent - try to add one more block to extent of blocks
1987 * @mpd->lbh - extent of blocks
1988 * @logical - logical number of the block in the file
1989 * @bh - bh of the block (used to access block's state)
1991 * the function is used to collect contig. blocks in same state
1993 static void mpage_add_bh_to_extent(struct mpage_da_data
*mpd
,
1994 sector_t logical
, struct buffer_head
*bh
)
1997 size_t b_size
= bh
->b_size
;
1998 struct buffer_head
*lbh
= &mpd
->lbh
;
1999 int nrblocks
= lbh
->b_size
>> mpd
->inode
->i_blkbits
;
2001 /* check if thereserved journal credits might overflow */
2002 if (!(EXT4_I(mpd
->inode
)->i_flags
& EXT4_EXTENTS_FL
)) {
2003 if (nrblocks
>= EXT4_MAX_TRANS_DATA
) {
2005 * With non-extent format we are limited by the journal
2006 * credit available. Total credit needed to insert
2007 * nrblocks contiguous blocks is dependent on the
2008 * nrblocks. So limit nrblocks.
2011 } else if ((nrblocks
+ (b_size
>> mpd
->inode
->i_blkbits
)) >
2012 EXT4_MAX_TRANS_DATA
) {
2014 * Adding the new buffer_head would make it cross the
2015 * allowed limit for which we have journal credit
2016 * reserved. So limit the new bh->b_size
2018 b_size
= (EXT4_MAX_TRANS_DATA
- nrblocks
) <<
2019 mpd
->inode
->i_blkbits
;
2020 /* we will do mpage_da_submit_io in the next loop */
2024 * First block in the extent
2026 if (lbh
->b_size
== 0) {
2027 lbh
->b_blocknr
= logical
;
2028 lbh
->b_size
= b_size
;
2029 lbh
->b_state
= bh
->b_state
& BH_FLAGS
;
2033 next
= lbh
->b_blocknr
+ nrblocks
;
2035 * Can we merge the block to our big extent?
2037 if (logical
== next
&& (bh
->b_state
& BH_FLAGS
) == lbh
->b_state
) {
2038 lbh
->b_size
+= b_size
;
2044 * We couldn't merge the block to our extent, so we
2045 * need to flush current extent and start new one
2047 if (mpage_da_map_blocks(mpd
) == 0)
2048 mpage_da_submit_io(mpd
);
2054 * __mpage_da_writepage - finds extent of pages and blocks
2056 * @page: page to consider
2057 * @wbc: not used, we just follow rules
2060 * The function finds extents of pages and scan them for all blocks.
2062 static int __mpage_da_writepage(struct page
*page
,
2063 struct writeback_control
*wbc
, void *data
)
2065 struct mpage_da_data
*mpd
= data
;
2066 struct inode
*inode
= mpd
->inode
;
2067 struct buffer_head
*bh
, *head
, fake
;
2072 * Rest of the page in the page_vec
2073 * redirty then and skip then. We will
2074 * try to to write them again after
2075 * starting a new transaction
2077 redirty_page_for_writepage(wbc
, page
);
2079 return MPAGE_DA_EXTENT_TAIL
;
2082 * Can we merge this page to current extent?
2084 if (mpd
->next_page
!= page
->index
) {
2086 * Nope, we can't. So, we map non-allocated blocks
2087 * and start IO on them using writepage()
2089 if (mpd
->next_page
!= mpd
->first_page
) {
2090 if (mpage_da_map_blocks(mpd
) == 0)
2091 mpage_da_submit_io(mpd
);
2093 * skip rest of the page in the page_vec
2096 redirty_page_for_writepage(wbc
, page
);
2098 return MPAGE_DA_EXTENT_TAIL
;
2102 * Start next extent of pages ...
2104 mpd
->first_page
= page
->index
;
2109 mpd
->lbh
.b_size
= 0;
2110 mpd
->lbh
.b_state
= 0;
2111 mpd
->lbh
.b_blocknr
= 0;
2114 mpd
->next_page
= page
->index
+ 1;
2115 logical
= (sector_t
) page
->index
<<
2116 (PAGE_CACHE_SHIFT
- inode
->i_blkbits
);
2118 if (!page_has_buffers(page
)) {
2120 * There is no attached buffer heads yet (mmap?)
2121 * we treat the page asfull of dirty blocks
2124 bh
->b_size
= PAGE_CACHE_SIZE
;
2126 set_buffer_dirty(bh
);
2127 set_buffer_uptodate(bh
);
2128 mpage_add_bh_to_extent(mpd
, logical
, bh
);
2130 return MPAGE_DA_EXTENT_TAIL
;
2133 * Page with regular buffer heads, just add all dirty ones
2135 head
= page_buffers(page
);
2138 BUG_ON(buffer_locked(bh
));
2140 * We need to try to allocate
2141 * unmapped blocks in the same page.
2142 * Otherwise we won't make progress
2143 * with the page in ext4_da_writepage
2145 if (buffer_dirty(bh
) &&
2146 (!buffer_mapped(bh
) || buffer_delay(bh
))) {
2147 mpage_add_bh_to_extent(mpd
, logical
, bh
);
2149 return MPAGE_DA_EXTENT_TAIL
;
2150 } else if (buffer_dirty(bh
) && (buffer_mapped(bh
))) {
2152 * mapped dirty buffer. We need to update
2153 * the b_state because we look at
2154 * b_state in mpage_da_map_blocks. We don't
2155 * update b_size because if we find an
2156 * unmapped buffer_head later we need to
2157 * use the b_state flag of that buffer_head.
2159 if (mpd
->lbh
.b_size
== 0)
2161 bh
->b_state
& BH_FLAGS
;
2164 } while ((bh
= bh
->b_this_page
) != head
);
2171 * mpage_da_writepages - walk the list of dirty pages of the given
2172 * address space, allocates non-allocated blocks, maps newly-allocated
2173 * blocks to existing bhs and issue IO them
2175 * @mapping: address space structure to write
2176 * @wbc: subtract the number of written pages from *@wbc->nr_to_write
2177 * @get_block: the filesystem's block mapper function.
2179 * This is a library function, which implements the writepages()
2180 * address_space_operation.
2182 static int mpage_da_writepages(struct address_space
*mapping
,
2183 struct writeback_control
*wbc
,
2184 struct mpage_da_data
*mpd
)
2188 if (!mpd
->get_block
)
2189 return generic_writepages(mapping
, wbc
);
2191 mpd
->lbh
.b_size
= 0;
2192 mpd
->lbh
.b_state
= 0;
2193 mpd
->lbh
.b_blocknr
= 0;
2194 mpd
->first_page
= 0;
2197 mpd
->pages_written
= 0;
2200 ret
= write_cache_pages(mapping
, wbc
, __mpage_da_writepage
, mpd
);
2202 * Handle last extent of pages
2204 if (!mpd
->io_done
&& mpd
->next_page
!= mpd
->first_page
) {
2205 if (mpage_da_map_blocks(mpd
) == 0)
2206 mpage_da_submit_io(mpd
);
2209 ret
= MPAGE_DA_EXTENT_TAIL
;
2211 wbc
->nr_to_write
-= mpd
->pages_written
;
2216 * this is a special callback for ->write_begin() only
2217 * it's intention is to return mapped block or reserve space
2219 static int ext4_da_get_block_prep(struct inode
*inode
, sector_t iblock
,
2220 struct buffer_head
*bh_result
, int create
)
2223 sector_t invalid_block
= ~((sector_t
) 0xffff);
2225 if (invalid_block
< ext4_blocks_count(EXT4_SB(inode
->i_sb
)->s_es
))
2228 BUG_ON(create
== 0);
2229 BUG_ON(bh_result
->b_size
!= inode
->i_sb
->s_blocksize
);
2232 * first, we need to know whether the block is allocated already
2233 * preallocated blocks are unmapped but should treated
2234 * the same as allocated blocks.
2236 ret
= ext4_get_blocks_wrap(NULL
, inode
, iblock
, 1, bh_result
, 0, 0, 0);
2237 if ((ret
== 0) && !buffer_delay(bh_result
)) {
2238 /* the block isn't (pre)allocated yet, let's reserve space */
2240 * XXX: __block_prepare_write() unmaps passed block,
2243 ret
= ext4_da_reserve_space(inode
, 1);
2245 /* not enough space to reserve */
2248 map_bh(bh_result
, inode
->i_sb
, invalid_block
);
2249 set_buffer_new(bh_result
);
2250 set_buffer_delay(bh_result
);
2251 } else if (ret
> 0) {
2252 bh_result
->b_size
= (ret
<< inode
->i_blkbits
);
2254 * With sub-block writes into unwritten extents
2255 * we also need to mark the buffer as new so that
2256 * the unwritten parts of the buffer gets correctly zeroed.
2258 if (buffer_unwritten(bh_result
))
2259 set_buffer_new(bh_result
);
2265 #define EXT4_DELALLOC_RSVED 1
2266 static int ext4_da_get_block_write(struct inode
*inode
, sector_t iblock
,
2267 struct buffer_head
*bh_result
, int create
)
2270 unsigned max_blocks
= bh_result
->b_size
>> inode
->i_blkbits
;
2271 loff_t disksize
= EXT4_I(inode
)->i_disksize
;
2272 handle_t
*handle
= NULL
;
2274 handle
= ext4_journal_current_handle();
2276 ret
= ext4_get_blocks_wrap(handle
, inode
, iblock
, max_blocks
,
2277 bh_result
, create
, 0, EXT4_DELALLOC_RSVED
);
2280 bh_result
->b_size
= (ret
<< inode
->i_blkbits
);
2282 if (ext4_should_order_data(inode
)) {
2284 retval
= ext4_jbd2_file_inode(handle
, inode
);
2287 * Failed to add inode for ordered
2288 * mode. Don't update file size
2294 * Update on-disk size along with block allocation
2295 * we don't use 'extend_disksize' as size may change
2296 * within already allocated block -bzzz
2298 disksize
= ((loff_t
) iblock
+ ret
) << inode
->i_blkbits
;
2299 if (disksize
> i_size_read(inode
))
2300 disksize
= i_size_read(inode
);
2301 if (disksize
> EXT4_I(inode
)->i_disksize
) {
2302 ext4_update_i_disksize(inode
, disksize
);
2303 ret
= ext4_mark_inode_dirty(handle
, inode
);
2311 static int ext4_bh_unmapped_or_delay(handle_t
*handle
, struct buffer_head
*bh
)
2314 * unmapped buffer is possible for holes.
2315 * delay buffer is possible with delayed allocation
2317 return ((!buffer_mapped(bh
) || buffer_delay(bh
)) && buffer_dirty(bh
));
2320 static int ext4_normal_get_block_write(struct inode
*inode
, sector_t iblock
,
2321 struct buffer_head
*bh_result
, int create
)
2324 unsigned max_blocks
= bh_result
->b_size
>> inode
->i_blkbits
;
2327 * we don't want to do block allocation in writepage
2328 * so call get_block_wrap with create = 0
2330 ret
= ext4_get_blocks_wrap(NULL
, inode
, iblock
, max_blocks
,
2331 bh_result
, 0, 0, 0);
2333 bh_result
->b_size
= (ret
<< inode
->i_blkbits
);
2340 * get called vi ext4_da_writepages after taking page lock (have journal handle)
2341 * get called via journal_submit_inode_data_buffers (no journal handle)
2342 * get called via shrink_page_list via pdflush (no journal handle)
2343 * or grab_page_cache when doing write_begin (have journal handle)
2345 static int ext4_da_writepage(struct page
*page
,
2346 struct writeback_control
*wbc
)
2351 struct buffer_head
*page_bufs
;
2352 struct inode
*inode
= page
->mapping
->host
;
2354 trace_mark(ext4_da_writepage
,
2355 "dev %s ino %lu page_index %lu",
2356 inode
->i_sb
->s_id
, inode
->i_ino
, page
->index
);
2357 size
= i_size_read(inode
);
2358 if (page
->index
== size
>> PAGE_CACHE_SHIFT
)
2359 len
= size
& ~PAGE_CACHE_MASK
;
2361 len
= PAGE_CACHE_SIZE
;
2363 if (page_has_buffers(page
)) {
2364 page_bufs
= page_buffers(page
);
2365 if (walk_page_buffers(NULL
, page_bufs
, 0, len
, NULL
,
2366 ext4_bh_unmapped_or_delay
)) {
2368 * We don't want to do block allocation
2369 * So redirty the page and return
2370 * We may reach here when we do a journal commit
2371 * via journal_submit_inode_data_buffers.
2372 * If we don't have mapping block we just ignore
2373 * them. We can also reach here via shrink_page_list
2375 redirty_page_for_writepage(wbc
, page
);
2381 * The test for page_has_buffers() is subtle:
2382 * We know the page is dirty but it lost buffers. That means
2383 * that at some moment in time after write_begin()/write_end()
2384 * has been called all buffers have been clean and thus they
2385 * must have been written at least once. So they are all
2386 * mapped and we can happily proceed with mapping them
2387 * and writing the page.
2389 * Try to initialize the buffer_heads and check whether
2390 * all are mapped and non delay. We don't want to
2391 * do block allocation here.
2393 ret
= block_prepare_write(page
, 0, PAGE_CACHE_SIZE
,
2394 ext4_normal_get_block_write
);
2396 page_bufs
= page_buffers(page
);
2397 /* check whether all are mapped and non delay */
2398 if (walk_page_buffers(NULL
, page_bufs
, 0, len
, NULL
,
2399 ext4_bh_unmapped_or_delay
)) {
2400 redirty_page_for_writepage(wbc
, page
);
2406 * We can't do block allocation here
2407 * so just redity the page and unlock
2410 redirty_page_for_writepage(wbc
, page
);
2414 /* now mark the buffer_heads as dirty and uptodate */
2415 block_commit_write(page
, 0, PAGE_CACHE_SIZE
);
2418 if (test_opt(inode
->i_sb
, NOBH
) && ext4_should_writeback_data(inode
))
2419 ret
= nobh_writepage(page
, ext4_normal_get_block_write
, wbc
);
2421 ret
= block_write_full_page(page
,
2422 ext4_normal_get_block_write
,
2429 * This is called via ext4_da_writepages() to
2430 * calulate the total number of credits to reserve to fit
2431 * a single extent allocation into a single transaction,
2432 * ext4_da_writpeages() will loop calling this before
2433 * the block allocation.
2436 static int ext4_da_writepages_trans_blocks(struct inode
*inode
)
2438 int max_blocks
= EXT4_I(inode
)->i_reserved_data_blocks
;
2441 * With non-extent format the journal credit needed to
2442 * insert nrblocks contiguous block is dependent on
2443 * number of contiguous block. So we will limit
2444 * number of contiguous block to a sane value
2446 if (!(inode
->i_flags
& EXT4_EXTENTS_FL
) &&
2447 (max_blocks
> EXT4_MAX_TRANS_DATA
))
2448 max_blocks
= EXT4_MAX_TRANS_DATA
;
2450 return ext4_chunk_trans_blocks(inode
, max_blocks
);
2453 static int ext4_da_writepages(struct address_space
*mapping
,
2454 struct writeback_control
*wbc
)
2457 int range_whole
= 0;
2458 handle_t
*handle
= NULL
;
2459 struct mpage_da_data mpd
;
2460 struct inode
*inode
= mapping
->host
;
2461 int no_nrwrite_index_update
;
2462 int pages_written
= 0;
2464 int range_cyclic
, cycled
= 1, io_done
= 0;
2465 int needed_blocks
, ret
= 0, nr_to_writebump
= 0;
2466 struct ext4_sb_info
*sbi
= EXT4_SB(mapping
->host
->i_sb
);
2468 trace_mark(ext4_da_writepages
,
2469 "dev %s ino %lu nr_t_write %ld "
2470 "pages_skipped %ld range_start %llu "
2471 "range_end %llu nonblocking %d "
2472 "for_kupdate %d for_reclaim %d "
2473 "for_writepages %d range_cyclic %d",
2474 inode
->i_sb
->s_id
, inode
->i_ino
,
2475 wbc
->nr_to_write
, wbc
->pages_skipped
,
2476 (unsigned long long) wbc
->range_start
,
2477 (unsigned long long) wbc
->range_end
,
2478 wbc
->nonblocking
, wbc
->for_kupdate
,
2479 wbc
->for_reclaim
, wbc
->for_writepages
,
2483 * No pages to write? This is mainly a kludge to avoid starting
2484 * a transaction for special inodes like journal inode on last iput()
2485 * because that could violate lock ordering on umount
2487 if (!mapping
->nrpages
|| !mapping_tagged(mapping
, PAGECACHE_TAG_DIRTY
))
2491 * If the filesystem has aborted, it is read-only, so return
2492 * right away instead of dumping stack traces later on that
2493 * will obscure the real source of the problem. We test
2494 * EXT4_MOUNT_ABORT instead of sb->s_flag's MS_RDONLY because
2495 * the latter could be true if the filesystem is mounted
2496 * read-only, and in that case, ext4_da_writepages should
2497 * *never* be called, so if that ever happens, we would want
2500 if (unlikely(sbi
->s_mount_opt
& EXT4_MOUNT_ABORT
))
2504 * Make sure nr_to_write is >= sbi->s_mb_stream_request
2505 * This make sure small files blocks are allocated in
2506 * single attempt. This ensure that small files
2507 * get less fragmented.
2509 if (wbc
->nr_to_write
< sbi
->s_mb_stream_request
) {
2510 nr_to_writebump
= sbi
->s_mb_stream_request
- wbc
->nr_to_write
;
2511 wbc
->nr_to_write
= sbi
->s_mb_stream_request
;
2513 if (wbc
->range_start
== 0 && wbc
->range_end
== LLONG_MAX
)
2516 range_cyclic
= wbc
->range_cyclic
;
2517 if (wbc
->range_cyclic
) {
2518 index
= mapping
->writeback_index
;
2521 wbc
->range_start
= index
<< PAGE_CACHE_SHIFT
;
2522 wbc
->range_end
= LLONG_MAX
;
2523 wbc
->range_cyclic
= 0;
2525 index
= wbc
->range_start
>> PAGE_CACHE_SHIFT
;
2528 mpd
.inode
= mapping
->host
;
2531 * we don't want write_cache_pages to update
2532 * nr_to_write and writeback_index
2534 no_nrwrite_index_update
= wbc
->no_nrwrite_index_update
;
2535 wbc
->no_nrwrite_index_update
= 1;
2536 pages_skipped
= wbc
->pages_skipped
;
2539 while (!ret
&& wbc
->nr_to_write
> 0) {
2542 * we insert one extent at a time. So we need
2543 * credit needed for single extent allocation.
2544 * journalled mode is currently not supported
2547 BUG_ON(ext4_should_journal_data(inode
));
2548 needed_blocks
= ext4_da_writepages_trans_blocks(inode
);
2550 /* start a new transaction*/
2551 handle
= ext4_journal_start(inode
, needed_blocks
);
2552 if (IS_ERR(handle
)) {
2553 ret
= PTR_ERR(handle
);
2554 printk(KERN_CRIT
"%s: jbd2_start: "
2555 "%ld pages, ino %lu; err %d\n", __func__
,
2556 wbc
->nr_to_write
, inode
->i_ino
, ret
);
2558 goto out_writepages
;
2560 mpd
.get_block
= ext4_da_get_block_write
;
2561 ret
= mpage_da_writepages(mapping
, wbc
, &mpd
);
2563 ext4_journal_stop(handle
);
2565 if ((mpd
.retval
== -ENOSPC
) && sbi
->s_journal
) {
2566 /* commit the transaction which would
2567 * free blocks released in the transaction
2570 jbd2_journal_force_commit_nested(sbi
->s_journal
);
2571 wbc
->pages_skipped
= pages_skipped
;
2573 } else if (ret
== MPAGE_DA_EXTENT_TAIL
) {
2575 * got one extent now try with
2578 pages_written
+= mpd
.pages_written
;
2579 wbc
->pages_skipped
= pages_skipped
;
2582 } else if (wbc
->nr_to_write
)
2584 * There is no more writeout needed
2585 * or we requested for a noblocking writeout
2586 * and we found the device congested
2590 if (!io_done
&& !cycled
) {
2593 wbc
->range_start
= index
<< PAGE_CACHE_SHIFT
;
2594 wbc
->range_end
= mapping
->writeback_index
- 1;
2597 if (pages_skipped
!= wbc
->pages_skipped
)
2598 printk(KERN_EMERG
"This should not happen leaving %s "
2599 "with nr_to_write = %ld ret = %d\n",
2600 __func__
, wbc
->nr_to_write
, ret
);
2603 index
+= pages_written
;
2604 wbc
->range_cyclic
= range_cyclic
;
2605 if (wbc
->range_cyclic
|| (range_whole
&& wbc
->nr_to_write
> 0))
2607 * set the writeback_index so that range_cyclic
2608 * mode will write it back later
2610 mapping
->writeback_index
= index
;
2613 if (!no_nrwrite_index_update
)
2614 wbc
->no_nrwrite_index_update
= 0;
2615 wbc
->nr_to_write
-= nr_to_writebump
;
2616 trace_mark(ext4_da_writepage_result
,
2617 "dev %s ino %lu ret %d pages_written %d "
2618 "pages_skipped %ld congestion %d "
2619 "more_io %d no_nrwrite_index_update %d",
2620 inode
->i_sb
->s_id
, inode
->i_ino
, ret
,
2621 pages_written
, wbc
->pages_skipped
,
2622 wbc
->encountered_congestion
, wbc
->more_io
,
2623 wbc
->no_nrwrite_index_update
);
2627 #define FALL_BACK_TO_NONDELALLOC 1
2628 static int ext4_nonda_switch(struct super_block
*sb
)
2630 s64 free_blocks
, dirty_blocks
;
2631 struct ext4_sb_info
*sbi
= EXT4_SB(sb
);
2634 * switch to non delalloc mode if we are running low
2635 * on free block. The free block accounting via percpu
2636 * counters can get slightly wrong with percpu_counter_batch getting
2637 * accumulated on each CPU without updating global counters
2638 * Delalloc need an accurate free block accounting. So switch
2639 * to non delalloc when we are near to error range.
2641 free_blocks
= percpu_counter_read_positive(&sbi
->s_freeblocks_counter
);
2642 dirty_blocks
= percpu_counter_read_positive(&sbi
->s_dirtyblocks_counter
);
2643 if (2 * free_blocks
< 3 * dirty_blocks
||
2644 free_blocks
< (dirty_blocks
+ EXT4_FREEBLOCKS_WATERMARK
)) {
2646 * free block count is less that 150% of dirty blocks
2647 * or free blocks is less that watermark
2654 static int ext4_da_write_begin(struct file
*file
, struct address_space
*mapping
,
2655 loff_t pos
, unsigned len
, unsigned flags
,
2656 struct page
**pagep
, void **fsdata
)
2658 int ret
, retries
= 0;
2662 struct inode
*inode
= mapping
->host
;
2665 index
= pos
>> PAGE_CACHE_SHIFT
;
2666 from
= pos
& (PAGE_CACHE_SIZE
- 1);
2669 if (ext4_nonda_switch(inode
->i_sb
)) {
2670 *fsdata
= (void *)FALL_BACK_TO_NONDELALLOC
;
2671 return ext4_write_begin(file
, mapping
, pos
,
2672 len
, flags
, pagep
, fsdata
);
2674 *fsdata
= (void *)0;
2676 trace_mark(ext4_da_write_begin
,
2677 "dev %s ino %lu pos %llu len %u flags %u",
2678 inode
->i_sb
->s_id
, inode
->i_ino
,
2679 (unsigned long long) pos
, len
, flags
);
2682 * With delayed allocation, we don't log the i_disksize update
2683 * if there is delayed block allocation. But we still need
2684 * to journalling the i_disksize update if writes to the end
2685 * of file which has an already mapped buffer.
2687 handle
= ext4_journal_start(inode
, 1);
2688 if (IS_ERR(handle
)) {
2689 ret
= PTR_ERR(handle
);
2692 /* We cannot recurse into the filesystem as the transaction is already
2694 flags
|= AOP_FLAG_NOFS
;
2696 page
= grab_cache_page_write_begin(mapping
, index
, flags
);
2698 ext4_journal_stop(handle
);
2704 ret
= block_write_begin(file
, mapping
, pos
, len
, flags
, pagep
, fsdata
,
2705 ext4_da_get_block_prep
);
2708 ext4_journal_stop(handle
);
2709 page_cache_release(page
);
2711 * block_write_begin may have instantiated a few blocks
2712 * outside i_size. Trim these off again. Don't need
2713 * i_size_read because we hold i_mutex.
2715 if (pos
+ len
> inode
->i_size
)
2716 vmtruncate(inode
, inode
->i_size
);
2719 if (ret
== -ENOSPC
&& ext4_should_retry_alloc(inode
->i_sb
, &retries
))
2726 * Check if we should update i_disksize
2727 * when write to the end of file but not require block allocation
2729 static int ext4_da_should_update_i_disksize(struct page
*page
,
2730 unsigned long offset
)
2732 struct buffer_head
*bh
;
2733 struct inode
*inode
= page
->mapping
->host
;
2737 bh
= page_buffers(page
);
2738 idx
= offset
>> inode
->i_blkbits
;
2740 for (i
= 0; i
< idx
; i
++)
2741 bh
= bh
->b_this_page
;
2743 if (!buffer_mapped(bh
) || (buffer_delay(bh
)))
2748 static int ext4_da_write_end(struct file
*file
,
2749 struct address_space
*mapping
,
2750 loff_t pos
, unsigned len
, unsigned copied
,
2751 struct page
*page
, void *fsdata
)
2753 struct inode
*inode
= mapping
->host
;
2755 handle_t
*handle
= ext4_journal_current_handle();
2757 unsigned long start
, end
;
2758 int write_mode
= (int)(unsigned long)fsdata
;
2760 if (write_mode
== FALL_BACK_TO_NONDELALLOC
) {
2761 if (ext4_should_order_data(inode
)) {
2762 return ext4_ordered_write_end(file
, mapping
, pos
,
2763 len
, copied
, page
, fsdata
);
2764 } else if (ext4_should_writeback_data(inode
)) {
2765 return ext4_writeback_write_end(file
, mapping
, pos
,
2766 len
, copied
, page
, fsdata
);
2772 trace_mark(ext4_da_write_end
,
2773 "dev %s ino %lu pos %llu len %u copied %u",
2774 inode
->i_sb
->s_id
, inode
->i_ino
,
2775 (unsigned long long) pos
, len
, copied
);
2776 start
= pos
& (PAGE_CACHE_SIZE
- 1);
2777 end
= start
+ copied
- 1;
2780 * generic_write_end() will run mark_inode_dirty() if i_size
2781 * changes. So let's piggyback the i_disksize mark_inode_dirty
2785 new_i_size
= pos
+ copied
;
2786 if (new_i_size
> EXT4_I(inode
)->i_disksize
) {
2787 if (ext4_da_should_update_i_disksize(page
, end
)) {
2788 down_write(&EXT4_I(inode
)->i_data_sem
);
2789 if (new_i_size
> EXT4_I(inode
)->i_disksize
) {
2791 * Updating i_disksize when extending file
2792 * without needing block allocation
2794 if (ext4_should_order_data(inode
))
2795 ret
= ext4_jbd2_file_inode(handle
,
2798 EXT4_I(inode
)->i_disksize
= new_i_size
;
2800 up_write(&EXT4_I(inode
)->i_data_sem
);
2801 /* We need to mark inode dirty even if
2802 * new_i_size is less that inode->i_size
2803 * bu greater than i_disksize.(hint delalloc)
2805 ext4_mark_inode_dirty(handle
, inode
);
2808 ret2
= generic_write_end(file
, mapping
, pos
, len
, copied
,
2813 ret2
= ext4_journal_stop(handle
);
2817 return ret
? ret
: copied
;
2820 static void ext4_da_invalidatepage(struct page
*page
, unsigned long offset
)
2823 * Drop reserved blocks
2825 BUG_ON(!PageLocked(page
));
2826 if (!page_has_buffers(page
))
2829 ext4_da_page_release_reservation(page
, offset
);
2832 ext4_invalidatepage(page
, offset
);
2838 * Force all delayed allocation blocks to be allocated for a given inode.
2840 int ext4_alloc_da_blocks(struct inode
*inode
)
2842 if (!EXT4_I(inode
)->i_reserved_data_blocks
&&
2843 !EXT4_I(inode
)->i_reserved_meta_blocks
)
2847 * We do something simple for now. The filemap_flush() will
2848 * also start triggering a write of the data blocks, which is
2849 * not strictly speaking necessary (and for users of
2850 * laptop_mode, not even desirable). However, to do otherwise
2851 * would require replicating code paths in:
2853 * ext4_da_writepages() ->
2854 * write_cache_pages() ---> (via passed in callback function)
2855 * __mpage_da_writepage() -->
2856 * mpage_add_bh_to_extent()
2857 * mpage_da_map_blocks()
2859 * The problem is that write_cache_pages(), located in
2860 * mm/page-writeback.c, marks pages clean in preparation for
2861 * doing I/O, which is not desirable if we're not planning on
2864 * We could call write_cache_pages(), and then redirty all of
2865 * the pages by calling redirty_page_for_writeback() but that
2866 * would be ugly in the extreme. So instead we would need to
2867 * replicate parts of the code in the above functions,
2868 * simplifying them becuase we wouldn't actually intend to
2869 * write out the pages, but rather only collect contiguous
2870 * logical block extents, call the multi-block allocator, and
2871 * then update the buffer heads with the block allocations.
2873 * For now, though, we'll cheat by calling filemap_flush(),
2874 * which will map the blocks, and start the I/O, but not
2875 * actually wait for the I/O to complete.
2877 return filemap_flush(inode
->i_mapping
);
2881 * bmap() is special. It gets used by applications such as lilo and by
2882 * the swapper to find the on-disk block of a specific piece of data.
2884 * Naturally, this is dangerous if the block concerned is still in the
2885 * journal. If somebody makes a swapfile on an ext4 data-journaling
2886 * filesystem and enables swap, then they may get a nasty shock when the
2887 * data getting swapped to that swapfile suddenly gets overwritten by
2888 * the original zero's written out previously to the journal and
2889 * awaiting writeback in the kernel's buffer cache.
2891 * So, if we see any bmap calls here on a modified, data-journaled file,
2892 * take extra steps to flush any blocks which might be in the cache.
2894 static sector_t
ext4_bmap(struct address_space
*mapping
, sector_t block
)
2896 struct inode
*inode
= mapping
->host
;
2900 if (mapping_tagged(mapping
, PAGECACHE_TAG_DIRTY
) &&
2901 test_opt(inode
->i_sb
, DELALLOC
)) {
2903 * With delalloc we want to sync the file
2904 * so that we can make sure we allocate
2907 filemap_write_and_wait(mapping
);
2910 if (EXT4_JOURNAL(inode
) && EXT4_I(inode
)->i_state
& EXT4_STATE_JDATA
) {
2912 * This is a REALLY heavyweight approach, but the use of
2913 * bmap on dirty files is expected to be extremely rare:
2914 * only if we run lilo or swapon on a freshly made file
2915 * do we expect this to happen.
2917 * (bmap requires CAP_SYS_RAWIO so this does not
2918 * represent an unprivileged user DOS attack --- we'd be
2919 * in trouble if mortal users could trigger this path at
2922 * NB. EXT4_STATE_JDATA is not set on files other than
2923 * regular files. If somebody wants to bmap a directory
2924 * or symlink and gets confused because the buffer
2925 * hasn't yet been flushed to disk, they deserve
2926 * everything they get.
2929 EXT4_I(inode
)->i_state
&= ~EXT4_STATE_JDATA
;
2930 journal
= EXT4_JOURNAL(inode
);
2931 jbd2_journal_lock_updates(journal
);
2932 err
= jbd2_journal_flush(journal
);
2933 jbd2_journal_unlock_updates(journal
);
2939 return generic_block_bmap(mapping
, block
, ext4_get_block
);
2942 static int bget_one(handle_t
*handle
, struct buffer_head
*bh
)
2948 static int bput_one(handle_t
*handle
, struct buffer_head
*bh
)
2955 * Note that we don't need to start a transaction unless we're journaling data
2956 * because we should have holes filled from ext4_page_mkwrite(). We even don't
2957 * need to file the inode to the transaction's list in ordered mode because if
2958 * we are writing back data added by write(), the inode is already there and if
2959 * we are writing back data modified via mmap(), noone guarantees in which
2960 * transaction the data will hit the disk. In case we are journaling data, we
2961 * cannot start transaction directly because transaction start ranks above page
2962 * lock so we have to do some magic.
2964 * In all journaling modes block_write_full_page() will start the I/O.
2968 * ext4_writepage() -> kmalloc() -> __alloc_pages() -> page_launder() ->
2973 * ext4_file_write() -> generic_file_write() -> __alloc_pages() -> ...
2975 * Same applies to ext4_get_block(). We will deadlock on various things like
2976 * lock_journal and i_data_sem
2978 * Setting PF_MEMALLOC here doesn't work - too many internal memory
2981 * 16May01: If we're reentered then journal_current_handle() will be
2982 * non-zero. We simply *return*.
2984 * 1 July 2001: @@@ FIXME:
2985 * In journalled data mode, a data buffer may be metadata against the
2986 * current transaction. But the same file is part of a shared mapping
2987 * and someone does a writepage() on it.
2989 * We will move the buffer onto the async_data list, but *after* it has
2990 * been dirtied. So there's a small window where we have dirty data on
2993 * Note that this only applies to the last partial page in the file. The
2994 * bit which block_write_full_page() uses prepare/commit for. (That's
2995 * broken code anyway: it's wrong for msync()).
2997 * It's a rare case: affects the final partial page, for journalled data
2998 * where the file is subject to bith write() and writepage() in the same
2999 * transction. To fix it we'll need a custom block_write_full_page().
3000 * We'll probably need that anyway for journalling writepage() output.
3002 * We don't honour synchronous mounts for writepage(). That would be
3003 * disastrous. Any write() or metadata operation will sync the fs for
3007 static int __ext4_normal_writepage(struct page
*page
,
3008 struct writeback_control
*wbc
)
3010 struct inode
*inode
= page
->mapping
->host
;
3012 if (test_opt(inode
->i_sb
, NOBH
))
3013 return nobh_writepage(page
,
3014 ext4_normal_get_block_write
, wbc
);
3016 return block_write_full_page(page
,
3017 ext4_normal_get_block_write
,
3021 static int ext4_normal_writepage(struct page
*page
,
3022 struct writeback_control
*wbc
)
3024 struct inode
*inode
= page
->mapping
->host
;
3025 loff_t size
= i_size_read(inode
);
3028 trace_mark(ext4_normal_writepage
,
3029 "dev %s ino %lu page_index %lu",
3030 inode
->i_sb
->s_id
, inode
->i_ino
, page
->index
);
3031 J_ASSERT(PageLocked(page
));
3032 if (page
->index
== size
>> PAGE_CACHE_SHIFT
)
3033 len
= size
& ~PAGE_CACHE_MASK
;
3035 len
= PAGE_CACHE_SIZE
;
3037 if (page_has_buffers(page
)) {
3038 /* if page has buffers it should all be mapped
3039 * and allocated. If there are not buffers attached
3040 * to the page we know the page is dirty but it lost
3041 * buffers. That means that at some moment in time
3042 * after write_begin() / write_end() has been called
3043 * all buffers have been clean and thus they must have been
3044 * written at least once. So they are all mapped and we can
3045 * happily proceed with mapping them and writing the page.
3047 BUG_ON(walk_page_buffers(NULL
, page_buffers(page
), 0, len
, NULL
,
3048 ext4_bh_unmapped_or_delay
));
3051 if (!ext4_journal_current_handle())
3052 return __ext4_normal_writepage(page
, wbc
);
3054 redirty_page_for_writepage(wbc
, page
);
3059 static int __ext4_journalled_writepage(struct page
*page
,
3060 struct writeback_control
*wbc
)
3062 struct address_space
*mapping
= page
->mapping
;
3063 struct inode
*inode
= mapping
->host
;
3064 struct buffer_head
*page_bufs
;
3065 handle_t
*handle
= NULL
;
3069 ret
= block_prepare_write(page
, 0, PAGE_CACHE_SIZE
,
3070 ext4_normal_get_block_write
);
3074 page_bufs
= page_buffers(page
);
3075 walk_page_buffers(handle
, page_bufs
, 0, PAGE_CACHE_SIZE
, NULL
,
3077 /* As soon as we unlock the page, it can go away, but we have
3078 * references to buffers so we are safe */
3081 handle
= ext4_journal_start(inode
, ext4_writepage_trans_blocks(inode
));
3082 if (IS_ERR(handle
)) {
3083 ret
= PTR_ERR(handle
);
3087 ret
= walk_page_buffers(handle
, page_bufs
, 0,
3088 PAGE_CACHE_SIZE
, NULL
, do_journal_get_write_access
);
3090 err
= walk_page_buffers(handle
, page_bufs
, 0,
3091 PAGE_CACHE_SIZE
, NULL
, write_end_fn
);
3094 err
= ext4_journal_stop(handle
);
3098 walk_page_buffers(handle
, page_bufs
, 0,
3099 PAGE_CACHE_SIZE
, NULL
, bput_one
);
3100 EXT4_I(inode
)->i_state
|= EXT4_STATE_JDATA
;
3109 static int ext4_journalled_writepage(struct page
*page
,
3110 struct writeback_control
*wbc
)
3112 struct inode
*inode
= page
->mapping
->host
;
3113 loff_t size
= i_size_read(inode
);
3116 trace_mark(ext4_journalled_writepage
,
3117 "dev %s ino %lu page_index %lu",
3118 inode
->i_sb
->s_id
, inode
->i_ino
, page
->index
);
3119 J_ASSERT(PageLocked(page
));
3120 if (page
->index
== size
>> PAGE_CACHE_SHIFT
)
3121 len
= size
& ~PAGE_CACHE_MASK
;
3123 len
= PAGE_CACHE_SIZE
;
3125 if (page_has_buffers(page
)) {
3126 /* if page has buffers it should all be mapped
3127 * and allocated. If there are not buffers attached
3128 * to the page we know the page is dirty but it lost
3129 * buffers. That means that at some moment in time
3130 * after write_begin() / write_end() has been called
3131 * all buffers have been clean and thus they must have been
3132 * written at least once. So they are all mapped and we can
3133 * happily proceed with mapping them and writing the page.
3135 BUG_ON(walk_page_buffers(NULL
, page_buffers(page
), 0, len
, NULL
,
3136 ext4_bh_unmapped_or_delay
));
3139 if (ext4_journal_current_handle())
3142 if (PageChecked(page
)) {
3144 * It's mmapped pagecache. Add buffers and journal it. There
3145 * doesn't seem much point in redirtying the page here.
3147 ClearPageChecked(page
);
3148 return __ext4_journalled_writepage(page
, wbc
);
3151 * It may be a page full of checkpoint-mode buffers. We don't
3152 * really know unless we go poke around in the buffer_heads.
3153 * But block_write_full_page will do the right thing.
3155 return block_write_full_page(page
,
3156 ext4_normal_get_block_write
,
3160 redirty_page_for_writepage(wbc
, page
);
3165 static int ext4_readpage(struct file
*file
, struct page
*page
)
3167 return mpage_readpage(page
, ext4_get_block
);
3171 ext4_readpages(struct file
*file
, struct address_space
*mapping
,
3172 struct list_head
*pages
, unsigned nr_pages
)
3174 return mpage_readpages(mapping
, pages
, nr_pages
, ext4_get_block
);
3177 static void ext4_invalidatepage(struct page
*page
, unsigned long offset
)
3179 journal_t
*journal
= EXT4_JOURNAL(page
->mapping
->host
);
3182 * If it's a full truncate we just forget about the pending dirtying
3185 ClearPageChecked(page
);
3188 jbd2_journal_invalidatepage(journal
, page
, offset
);
3190 block_invalidatepage(page
, offset
);
3193 static int ext4_releasepage(struct page
*page
, gfp_t wait
)
3195 journal_t
*journal
= EXT4_JOURNAL(page
->mapping
->host
);
3197 WARN_ON(PageChecked(page
));
3198 if (!page_has_buffers(page
))
3201 return jbd2_journal_try_to_free_buffers(journal
, page
, wait
);
3203 return try_to_free_buffers(page
);
3207 * If the O_DIRECT write will extend the file then add this inode to the
3208 * orphan list. So recovery will truncate it back to the original size
3209 * if the machine crashes during the write.
3211 * If the O_DIRECT write is intantiating holes inside i_size and the machine
3212 * crashes then stale disk data _may_ be exposed inside the file. But current
3213 * VFS code falls back into buffered path in that case so we are safe.
3215 static ssize_t
ext4_direct_IO(int rw
, struct kiocb
*iocb
,
3216 const struct iovec
*iov
, loff_t offset
,
3217 unsigned long nr_segs
)
3219 struct file
*file
= iocb
->ki_filp
;
3220 struct inode
*inode
= file
->f_mapping
->host
;
3221 struct ext4_inode_info
*ei
= EXT4_I(inode
);
3225 size_t count
= iov_length(iov
, nr_segs
);
3228 loff_t final_size
= offset
+ count
;
3230 if (final_size
> inode
->i_size
) {
3231 /* Credits for sb + inode write */
3232 handle
= ext4_journal_start(inode
, 2);
3233 if (IS_ERR(handle
)) {
3234 ret
= PTR_ERR(handle
);
3237 ret
= ext4_orphan_add(handle
, inode
);
3239 ext4_journal_stop(handle
);
3243 ei
->i_disksize
= inode
->i_size
;
3244 ext4_journal_stop(handle
);
3248 ret
= blockdev_direct_IO(rw
, iocb
, inode
, inode
->i_sb
->s_bdev
, iov
,
3250 ext4_get_block
, NULL
);
3255 /* Credits for sb + inode write */
3256 handle
= ext4_journal_start(inode
, 2);
3257 if (IS_ERR(handle
)) {
3258 /* This is really bad luck. We've written the data
3259 * but cannot extend i_size. Bail out and pretend
3260 * the write failed... */
3261 ret
= PTR_ERR(handle
);
3265 ext4_orphan_del(handle
, inode
);
3267 loff_t end
= offset
+ ret
;
3268 if (end
> inode
->i_size
) {
3269 ei
->i_disksize
= end
;
3270 i_size_write(inode
, end
);
3272 * We're going to return a positive `ret'
3273 * here due to non-zero-length I/O, so there's
3274 * no way of reporting error returns from
3275 * ext4_mark_inode_dirty() to userspace. So
3278 ext4_mark_inode_dirty(handle
, inode
);
3281 err
= ext4_journal_stop(handle
);
3290 * Pages can be marked dirty completely asynchronously from ext4's journalling
3291 * activity. By filemap_sync_pte(), try_to_unmap_one(), etc. We cannot do
3292 * much here because ->set_page_dirty is called under VFS locks. The page is
3293 * not necessarily locked.
3295 * We cannot just dirty the page and leave attached buffers clean, because the
3296 * buffers' dirty state is "definitive". We cannot just set the buffers dirty
3297 * or jbddirty because all the journalling code will explode.
3299 * So what we do is to mark the page "pending dirty" and next time writepage
3300 * is called, propagate that into the buffers appropriately.
3302 static int ext4_journalled_set_page_dirty(struct page
*page
)
3304 SetPageChecked(page
);
3305 return __set_page_dirty_nobuffers(page
);
3308 static const struct address_space_operations ext4_ordered_aops
= {
3309 .readpage
= ext4_readpage
,
3310 .readpages
= ext4_readpages
,
3311 .writepage
= ext4_normal_writepage
,
3312 .sync_page
= block_sync_page
,
3313 .write_begin
= ext4_write_begin
,
3314 .write_end
= ext4_ordered_write_end
,
3316 .invalidatepage
= ext4_invalidatepage
,
3317 .releasepage
= ext4_releasepage
,
3318 .direct_IO
= ext4_direct_IO
,
3319 .migratepage
= buffer_migrate_page
,
3320 .is_partially_uptodate
= block_is_partially_uptodate
,
3323 static const struct address_space_operations ext4_writeback_aops
= {
3324 .readpage
= ext4_readpage
,
3325 .readpages
= ext4_readpages
,
3326 .writepage
= ext4_normal_writepage
,
3327 .sync_page
= block_sync_page
,
3328 .write_begin
= ext4_write_begin
,
3329 .write_end
= ext4_writeback_write_end
,
3331 .invalidatepage
= ext4_invalidatepage
,
3332 .releasepage
= ext4_releasepage
,
3333 .direct_IO
= ext4_direct_IO
,
3334 .migratepage
= buffer_migrate_page
,
3335 .is_partially_uptodate
= block_is_partially_uptodate
,
3338 static const struct address_space_operations ext4_journalled_aops
= {
3339 .readpage
= ext4_readpage
,
3340 .readpages
= ext4_readpages
,
3341 .writepage
= ext4_journalled_writepage
,
3342 .sync_page
= block_sync_page
,
3343 .write_begin
= ext4_write_begin
,
3344 .write_end
= ext4_journalled_write_end
,
3345 .set_page_dirty
= ext4_journalled_set_page_dirty
,
3347 .invalidatepage
= ext4_invalidatepage
,
3348 .releasepage
= ext4_releasepage
,
3349 .is_partially_uptodate
= block_is_partially_uptodate
,
3352 static const struct address_space_operations ext4_da_aops
= {
3353 .readpage
= ext4_readpage
,
3354 .readpages
= ext4_readpages
,
3355 .writepage
= ext4_da_writepage
,
3356 .writepages
= ext4_da_writepages
,
3357 .sync_page
= block_sync_page
,
3358 .write_begin
= ext4_da_write_begin
,
3359 .write_end
= ext4_da_write_end
,
3361 .invalidatepage
= ext4_da_invalidatepage
,
3362 .releasepage
= ext4_releasepage
,
3363 .direct_IO
= ext4_direct_IO
,
3364 .migratepage
= buffer_migrate_page
,
3365 .is_partially_uptodate
= block_is_partially_uptodate
,
3368 void ext4_set_aops(struct inode
*inode
)
3370 if (ext4_should_order_data(inode
) &&
3371 test_opt(inode
->i_sb
, DELALLOC
))
3372 inode
->i_mapping
->a_ops
= &ext4_da_aops
;
3373 else if (ext4_should_order_data(inode
))
3374 inode
->i_mapping
->a_ops
= &ext4_ordered_aops
;
3375 else if (ext4_should_writeback_data(inode
) &&
3376 test_opt(inode
->i_sb
, DELALLOC
))
3377 inode
->i_mapping
->a_ops
= &ext4_da_aops
;
3378 else if (ext4_should_writeback_data(inode
))
3379 inode
->i_mapping
->a_ops
= &ext4_writeback_aops
;
3381 inode
->i_mapping
->a_ops
= &ext4_journalled_aops
;
3385 * ext4_block_truncate_page() zeroes out a mapping from file offset `from'
3386 * up to the end of the block which corresponds to `from'.
3387 * This required during truncate. We need to physically zero the tail end
3388 * of that block so it doesn't yield old data if the file is later grown.
3390 int ext4_block_truncate_page(handle_t
*handle
,
3391 struct address_space
*mapping
, loff_t from
)
3393 ext4_fsblk_t index
= from
>> PAGE_CACHE_SHIFT
;
3394 unsigned offset
= from
& (PAGE_CACHE_SIZE
-1);
3395 unsigned blocksize
, length
, pos
;
3397 struct inode
*inode
= mapping
->host
;
3398 struct buffer_head
*bh
;
3402 page
= grab_cache_page(mapping
, from
>> PAGE_CACHE_SHIFT
);
3406 blocksize
= inode
->i_sb
->s_blocksize
;
3407 length
= blocksize
- (offset
& (blocksize
- 1));
3408 iblock
= index
<< (PAGE_CACHE_SHIFT
- inode
->i_sb
->s_blocksize_bits
);
3411 * For "nobh" option, we can only work if we don't need to
3412 * read-in the page - otherwise we create buffers to do the IO.
3414 if (!page_has_buffers(page
) && test_opt(inode
->i_sb
, NOBH
) &&
3415 ext4_should_writeback_data(inode
) && PageUptodate(page
)) {
3416 zero_user(page
, offset
, length
);
3417 set_page_dirty(page
);
3421 if (!page_has_buffers(page
))
3422 create_empty_buffers(page
, blocksize
, 0);
3424 /* Find the buffer that contains "offset" */
3425 bh
= page_buffers(page
);
3427 while (offset
>= pos
) {
3428 bh
= bh
->b_this_page
;
3434 if (buffer_freed(bh
)) {
3435 BUFFER_TRACE(bh
, "freed: skip");
3439 if (!buffer_mapped(bh
)) {
3440 BUFFER_TRACE(bh
, "unmapped");
3441 ext4_get_block(inode
, iblock
, bh
, 0);
3442 /* unmapped? It's a hole - nothing to do */
3443 if (!buffer_mapped(bh
)) {
3444 BUFFER_TRACE(bh
, "still unmapped");
3449 /* Ok, it's mapped. Make sure it's up-to-date */
3450 if (PageUptodate(page
))
3451 set_buffer_uptodate(bh
);
3453 if (!buffer_uptodate(bh
)) {
3455 ll_rw_block(READ
, 1, &bh
);
3457 /* Uhhuh. Read error. Complain and punt. */
3458 if (!buffer_uptodate(bh
))
3462 if (ext4_should_journal_data(inode
)) {
3463 BUFFER_TRACE(bh
, "get write access");
3464 err
= ext4_journal_get_write_access(handle
, bh
);
3469 zero_user(page
, offset
, length
);
3471 BUFFER_TRACE(bh
, "zeroed end of block");
3474 if (ext4_should_journal_data(inode
)) {
3475 err
= ext4_handle_dirty_metadata(handle
, inode
, bh
);
3477 if (ext4_should_order_data(inode
))
3478 err
= ext4_jbd2_file_inode(handle
, inode
);
3479 mark_buffer_dirty(bh
);
3484 page_cache_release(page
);
3489 * Probably it should be a library function... search for first non-zero word
3490 * or memcmp with zero_page, whatever is better for particular architecture.
3493 static inline int all_zeroes(__le32
*p
, __le32
*q
)
3502 * ext4_find_shared - find the indirect blocks for partial truncation.
3503 * @inode: inode in question
3504 * @depth: depth of the affected branch
3505 * @offsets: offsets of pointers in that branch (see ext4_block_to_path)
3506 * @chain: place to store the pointers to partial indirect blocks
3507 * @top: place to the (detached) top of branch
3509 * This is a helper function used by ext4_truncate().
3511 * When we do truncate() we may have to clean the ends of several
3512 * indirect blocks but leave the blocks themselves alive. Block is
3513 * partially truncated if some data below the new i_size is refered
3514 * from it (and it is on the path to the first completely truncated
3515 * data block, indeed). We have to free the top of that path along
3516 * with everything to the right of the path. Since no allocation
3517 * past the truncation point is possible until ext4_truncate()
3518 * finishes, we may safely do the latter, but top of branch may
3519 * require special attention - pageout below the truncation point
3520 * might try to populate it.
3522 * We atomically detach the top of branch from the tree, store the
3523 * block number of its root in *@top, pointers to buffer_heads of
3524 * partially truncated blocks - in @chain[].bh and pointers to
3525 * their last elements that should not be removed - in
3526 * @chain[].p. Return value is the pointer to last filled element
3529 * The work left to caller to do the actual freeing of subtrees:
3530 * a) free the subtree starting from *@top
3531 * b) free the subtrees whose roots are stored in
3532 * (@chain[i].p+1 .. end of @chain[i].bh->b_data)
3533 * c) free the subtrees growing from the inode past the @chain[0].
3534 * (no partially truncated stuff there). */
3536 static Indirect
*ext4_find_shared(struct inode
*inode
, int depth
,
3537 ext4_lblk_t offsets
[4], Indirect chain
[4], __le32
*top
)
3539 Indirect
*partial
, *p
;
3543 /* Make k index the deepest non-null offest + 1 */
3544 for (k
= depth
; k
> 1 && !offsets
[k
-1]; k
--)
3546 partial
= ext4_get_branch(inode
, k
, offsets
, chain
, &err
);
3547 /* Writer: pointers */
3549 partial
= chain
+ k
-1;
3551 * If the branch acquired continuation since we've looked at it -
3552 * fine, it should all survive and (new) top doesn't belong to us.
3554 if (!partial
->key
&& *partial
->p
)
3557 for (p
= partial
; (p
> chain
) && all_zeroes((__le32
*) p
->bh
->b_data
, p
->p
); p
--)
3560 * OK, we've found the last block that must survive. The rest of our
3561 * branch should be detached before unlocking. However, if that rest
3562 * of branch is all ours and does not grow immediately from the inode
3563 * it's easier to cheat and just decrement partial->p.
3565 if (p
== chain
+ k
- 1 && p
> chain
) {
3569 /* Nope, don't do this in ext4. Must leave the tree intact */
3576 while (partial
> p
) {
3577 brelse(partial
->bh
);
3585 * Zero a number of block pointers in either an inode or an indirect block.
3586 * If we restart the transaction we must again get write access to the
3587 * indirect block for further modification.
3589 * We release `count' blocks on disk, but (last - first) may be greater
3590 * than `count' because there can be holes in there.
3592 static void ext4_clear_blocks(handle_t
*handle
, struct inode
*inode
,
3593 struct buffer_head
*bh
, ext4_fsblk_t block_to_free
,
3594 unsigned long count
, __le32
*first
, __le32
*last
)
3597 if (try_to_extend_transaction(handle
, inode
)) {
3599 BUFFER_TRACE(bh
, "call ext4_handle_dirty_metadata");
3600 ext4_handle_dirty_metadata(handle
, inode
, bh
);
3602 ext4_mark_inode_dirty(handle
, inode
);
3603 ext4_journal_test_restart(handle
, inode
);
3605 BUFFER_TRACE(bh
, "retaking write access");
3606 ext4_journal_get_write_access(handle
, bh
);
3611 * Any buffers which are on the journal will be in memory. We find
3612 * them on the hash table so jbd2_journal_revoke() will run jbd2_journal_forget()
3613 * on them. We've already detached each block from the file, so
3614 * bforget() in jbd2_journal_forget() should be safe.
3616 * AKPM: turn on bforget in jbd2_journal_forget()!!!
3618 for (p
= first
; p
< last
; p
++) {
3619 u32 nr
= le32_to_cpu(*p
);
3621 struct buffer_head
*tbh
;
3624 tbh
= sb_find_get_block(inode
->i_sb
, nr
);
3625 ext4_forget(handle
, 0, inode
, tbh
, nr
);
3629 ext4_free_blocks(handle
, inode
, block_to_free
, count
, 0);
3633 * ext4_free_data - free a list of data blocks
3634 * @handle: handle for this transaction
3635 * @inode: inode we are dealing with
3636 * @this_bh: indirect buffer_head which contains *@first and *@last
3637 * @first: array of block numbers
3638 * @last: points immediately past the end of array
3640 * We are freeing all blocks refered from that array (numbers are stored as
3641 * little-endian 32-bit) and updating @inode->i_blocks appropriately.
3643 * We accumulate contiguous runs of blocks to free. Conveniently, if these
3644 * blocks are contiguous then releasing them at one time will only affect one
3645 * or two bitmap blocks (+ group descriptor(s) and superblock) and we won't
3646 * actually use a lot of journal space.
3648 * @this_bh will be %NULL if @first and @last point into the inode's direct
3651 static void ext4_free_data(handle_t
*handle
, struct inode
*inode
,
3652 struct buffer_head
*this_bh
,
3653 __le32
*first
, __le32
*last
)
3655 ext4_fsblk_t block_to_free
= 0; /* Starting block # of a run */
3656 unsigned long count
= 0; /* Number of blocks in the run */
3657 __le32
*block_to_free_p
= NULL
; /* Pointer into inode/ind
3660 ext4_fsblk_t nr
; /* Current block # */
3661 __le32
*p
; /* Pointer into inode/ind
3662 for current block */
3665 if (this_bh
) { /* For indirect block */
3666 BUFFER_TRACE(this_bh
, "get_write_access");
3667 err
= ext4_journal_get_write_access(handle
, this_bh
);
3668 /* Important: if we can't update the indirect pointers
3669 * to the blocks, we can't free them. */
3674 for (p
= first
; p
< last
; p
++) {
3675 nr
= le32_to_cpu(*p
);
3677 /* accumulate blocks to free if they're contiguous */
3680 block_to_free_p
= p
;
3682 } else if (nr
== block_to_free
+ count
) {
3685 ext4_clear_blocks(handle
, inode
, this_bh
,
3687 count
, block_to_free_p
, p
);
3689 block_to_free_p
= p
;
3696 ext4_clear_blocks(handle
, inode
, this_bh
, block_to_free
,
3697 count
, block_to_free_p
, p
);
3700 BUFFER_TRACE(this_bh
, "call ext4_handle_dirty_metadata");
3703 * The buffer head should have an attached journal head at this
3704 * point. However, if the data is corrupted and an indirect
3705 * block pointed to itself, it would have been detached when
3706 * the block was cleared. Check for this instead of OOPSing.
3708 if ((EXT4_JOURNAL(inode
) == NULL
) || bh2jh(this_bh
))
3709 ext4_handle_dirty_metadata(handle
, inode
, this_bh
);
3711 ext4_error(inode
->i_sb
, __func__
,
3712 "circular indirect block detected, "
3713 "inode=%lu, block=%llu",
3715 (unsigned long long) this_bh
->b_blocknr
);
3720 * ext4_free_branches - free an array of branches
3721 * @handle: JBD handle for this transaction
3722 * @inode: inode we are dealing with
3723 * @parent_bh: the buffer_head which contains *@first and *@last
3724 * @first: array of block numbers
3725 * @last: pointer immediately past the end of array
3726 * @depth: depth of the branches to free
3728 * We are freeing all blocks refered from these branches (numbers are
3729 * stored as little-endian 32-bit) and updating @inode->i_blocks
3732 static void ext4_free_branches(handle_t
*handle
, struct inode
*inode
,
3733 struct buffer_head
*parent_bh
,
3734 __le32
*first
, __le32
*last
, int depth
)
3739 if (ext4_handle_is_aborted(handle
))
3743 struct buffer_head
*bh
;
3744 int addr_per_block
= EXT4_ADDR_PER_BLOCK(inode
->i_sb
);
3746 while (--p
>= first
) {
3747 nr
= le32_to_cpu(*p
);
3749 continue; /* A hole */
3751 /* Go read the buffer for the next level down */
3752 bh
= sb_bread(inode
->i_sb
, nr
);
3755 * A read failure? Report error and clear slot
3759 ext4_error(inode
->i_sb
, "ext4_free_branches",
3760 "Read failure, inode=%lu, block=%llu",
3765 /* This zaps the entire block. Bottom up. */
3766 BUFFER_TRACE(bh
, "free child branches");
3767 ext4_free_branches(handle
, inode
, bh
,
3768 (__le32
*) bh
->b_data
,
3769 (__le32
*) bh
->b_data
+ addr_per_block
,
3773 * We've probably journalled the indirect block several
3774 * times during the truncate. But it's no longer
3775 * needed and we now drop it from the transaction via
3776 * jbd2_journal_revoke().
3778 * That's easy if it's exclusively part of this
3779 * transaction. But if it's part of the committing
3780 * transaction then jbd2_journal_forget() will simply
3781 * brelse() it. That means that if the underlying
3782 * block is reallocated in ext4_get_block(),
3783 * unmap_underlying_metadata() will find this block
3784 * and will try to get rid of it. damn, damn.
3786 * If this block has already been committed to the
3787 * journal, a revoke record will be written. And
3788 * revoke records must be emitted *before* clearing
3789 * this block's bit in the bitmaps.
3791 ext4_forget(handle
, 1, inode
, bh
, bh
->b_blocknr
);
3794 * Everything below this this pointer has been
3795 * released. Now let this top-of-subtree go.
3797 * We want the freeing of this indirect block to be
3798 * atomic in the journal with the updating of the
3799 * bitmap block which owns it. So make some room in
3802 * We zero the parent pointer *after* freeing its
3803 * pointee in the bitmaps, so if extend_transaction()
3804 * for some reason fails to put the bitmap changes and
3805 * the release into the same transaction, recovery
3806 * will merely complain about releasing a free block,
3807 * rather than leaking blocks.
3809 if (ext4_handle_is_aborted(handle
))
3811 if (try_to_extend_transaction(handle
, inode
)) {
3812 ext4_mark_inode_dirty(handle
, inode
);
3813 ext4_journal_test_restart(handle
, inode
);
3816 ext4_free_blocks(handle
, inode
, nr
, 1, 1);
3820 * The block which we have just freed is
3821 * pointed to by an indirect block: journal it
3823 BUFFER_TRACE(parent_bh
, "get_write_access");
3824 if (!ext4_journal_get_write_access(handle
,
3827 BUFFER_TRACE(parent_bh
,
3828 "call ext4_handle_dirty_metadata");
3829 ext4_handle_dirty_metadata(handle
,
3836 /* We have reached the bottom of the tree. */
3837 BUFFER_TRACE(parent_bh
, "free data blocks");
3838 ext4_free_data(handle
, inode
, parent_bh
, first
, last
);
3842 int ext4_can_truncate(struct inode
*inode
)
3844 if (IS_APPEND(inode
) || IS_IMMUTABLE(inode
))
3846 if (S_ISREG(inode
->i_mode
))
3848 if (S_ISDIR(inode
->i_mode
))
3850 if (S_ISLNK(inode
->i_mode
))
3851 return !ext4_inode_is_fast_symlink(inode
);
3858 * We block out ext4_get_block() block instantiations across the entire
3859 * transaction, and VFS/VM ensures that ext4_truncate() cannot run
3860 * simultaneously on behalf of the same inode.
3862 * As we work through the truncate and commmit bits of it to the journal there
3863 * is one core, guiding principle: the file's tree must always be consistent on
3864 * disk. We must be able to restart the truncate after a crash.
3866 * The file's tree may be transiently inconsistent in memory (although it
3867 * probably isn't), but whenever we close off and commit a journal transaction,
3868 * the contents of (the filesystem + the journal) must be consistent and
3869 * restartable. It's pretty simple, really: bottom up, right to left (although
3870 * left-to-right works OK too).
3872 * Note that at recovery time, journal replay occurs *before* the restart of
3873 * truncate against the orphan inode list.
3875 * The committed inode has the new, desired i_size (which is the same as
3876 * i_disksize in this case). After a crash, ext4_orphan_cleanup() will see
3877 * that this inode's truncate did not complete and it will again call
3878 * ext4_truncate() to have another go. So there will be instantiated blocks
3879 * to the right of the truncation point in a crashed ext4 filesystem. But
3880 * that's fine - as long as they are linked from the inode, the post-crash
3881 * ext4_truncate() run will find them and release them.
3883 void ext4_truncate(struct inode
*inode
)
3886 struct ext4_inode_info
*ei
= EXT4_I(inode
);
3887 __le32
*i_data
= ei
->i_data
;
3888 int addr_per_block
= EXT4_ADDR_PER_BLOCK(inode
->i_sb
);
3889 struct address_space
*mapping
= inode
->i_mapping
;
3890 ext4_lblk_t offsets
[4];
3895 ext4_lblk_t last_block
;
3896 unsigned blocksize
= inode
->i_sb
->s_blocksize
;
3898 if (!ext4_can_truncate(inode
))
3901 if (inode
->i_size
== 0 && !test_opt(inode
->i_sb
, NO_AUTO_DA_ALLOC
))
3902 ei
->i_state
|= EXT4_STATE_DA_ALLOC_CLOSE
;
3904 if (EXT4_I(inode
)->i_flags
& EXT4_EXTENTS_FL
) {
3905 ext4_ext_truncate(inode
);
3909 handle
= start_transaction(inode
);
3911 return; /* AKPM: return what? */
3913 last_block
= (inode
->i_size
+ blocksize
-1)
3914 >> EXT4_BLOCK_SIZE_BITS(inode
->i_sb
);
3916 if (inode
->i_size
& (blocksize
- 1))
3917 if (ext4_block_truncate_page(handle
, mapping
, inode
->i_size
))
3920 n
= ext4_block_to_path(inode
, last_block
, offsets
, NULL
);
3922 goto out_stop
; /* error */
3925 * OK. This truncate is going to happen. We add the inode to the
3926 * orphan list, so that if this truncate spans multiple transactions,
3927 * and we crash, we will resume the truncate when the filesystem
3928 * recovers. It also marks the inode dirty, to catch the new size.
3930 * Implication: the file must always be in a sane, consistent
3931 * truncatable state while each transaction commits.
3933 if (ext4_orphan_add(handle
, inode
))
3937 * From here we block out all ext4_get_block() callers who want to
3938 * modify the block allocation tree.
3940 down_write(&ei
->i_data_sem
);
3942 ext4_discard_preallocations(inode
);
3945 * The orphan list entry will now protect us from any crash which
3946 * occurs before the truncate completes, so it is now safe to propagate
3947 * the new, shorter inode size (held for now in i_size) into the
3948 * on-disk inode. We do this via i_disksize, which is the value which
3949 * ext4 *really* writes onto the disk inode.
3951 ei
->i_disksize
= inode
->i_size
;
3953 if (n
== 1) { /* direct blocks */
3954 ext4_free_data(handle
, inode
, NULL
, i_data
+offsets
[0],
3955 i_data
+ EXT4_NDIR_BLOCKS
);
3959 partial
= ext4_find_shared(inode
, n
, offsets
, chain
, &nr
);
3960 /* Kill the top of shared branch (not detached) */
3962 if (partial
== chain
) {
3963 /* Shared branch grows from the inode */
3964 ext4_free_branches(handle
, inode
, NULL
,
3965 &nr
, &nr
+1, (chain
+n
-1) - partial
);
3968 * We mark the inode dirty prior to restart,
3969 * and prior to stop. No need for it here.
3972 /* Shared branch grows from an indirect block */
3973 BUFFER_TRACE(partial
->bh
, "get_write_access");
3974 ext4_free_branches(handle
, inode
, partial
->bh
,
3976 partial
->p
+1, (chain
+n
-1) - partial
);
3979 /* Clear the ends of indirect blocks on the shared branch */
3980 while (partial
> chain
) {
3981 ext4_free_branches(handle
, inode
, partial
->bh
, partial
->p
+ 1,
3982 (__le32
*)partial
->bh
->b_data
+addr_per_block
,
3983 (chain
+n
-1) - partial
);
3984 BUFFER_TRACE(partial
->bh
, "call brelse");
3985 brelse (partial
->bh
);
3989 /* Kill the remaining (whole) subtrees */
3990 switch (offsets
[0]) {
3992 nr
= i_data
[EXT4_IND_BLOCK
];
3994 ext4_free_branches(handle
, inode
, NULL
, &nr
, &nr
+1, 1);
3995 i_data
[EXT4_IND_BLOCK
] = 0;
3997 case EXT4_IND_BLOCK
:
3998 nr
= i_data
[EXT4_DIND_BLOCK
];
4000 ext4_free_branches(handle
, inode
, NULL
, &nr
, &nr
+1, 2);
4001 i_data
[EXT4_DIND_BLOCK
] = 0;
4003 case EXT4_DIND_BLOCK
:
4004 nr
= i_data
[EXT4_TIND_BLOCK
];
4006 ext4_free_branches(handle
, inode
, NULL
, &nr
, &nr
+1, 3);
4007 i_data
[EXT4_TIND_BLOCK
] = 0;
4009 case EXT4_TIND_BLOCK
:
4013 up_write(&ei
->i_data_sem
);
4014 inode
->i_mtime
= inode
->i_ctime
= ext4_current_time(inode
);
4015 ext4_mark_inode_dirty(handle
, inode
);
4018 * In a multi-transaction truncate, we only make the final transaction
4022 ext4_handle_sync(handle
);
4025 * If this was a simple ftruncate(), and the file will remain alive
4026 * then we need to clear up the orphan record which we created above.
4027 * However, if this was a real unlink then we were called by
4028 * ext4_delete_inode(), and we allow that function to clean up the
4029 * orphan info for us.
4032 ext4_orphan_del(handle
, inode
);
4034 ext4_journal_stop(handle
);
4038 * ext4_get_inode_loc returns with an extra refcount against the inode's
4039 * underlying buffer_head on success. If 'in_mem' is true, we have all
4040 * data in memory that is needed to recreate the on-disk version of this
4043 static int __ext4_get_inode_loc(struct inode
*inode
,
4044 struct ext4_iloc
*iloc
, int in_mem
)
4046 struct ext4_group_desc
*gdp
;
4047 struct buffer_head
*bh
;
4048 struct super_block
*sb
= inode
->i_sb
;
4050 int inodes_per_block
, inode_offset
;
4053 if (!ext4_valid_inum(sb
, inode
->i_ino
))
4056 iloc
->block_group
= (inode
->i_ino
- 1) / EXT4_INODES_PER_GROUP(sb
);
4057 gdp
= ext4_get_group_desc(sb
, iloc
->block_group
, NULL
);
4062 * Figure out the offset within the block group inode table
4064 inodes_per_block
= (EXT4_BLOCK_SIZE(sb
) / EXT4_INODE_SIZE(sb
));
4065 inode_offset
= ((inode
->i_ino
- 1) %
4066 EXT4_INODES_PER_GROUP(sb
));
4067 block
= ext4_inode_table(sb
, gdp
) + (inode_offset
/ inodes_per_block
);
4068 iloc
->offset
= (inode_offset
% inodes_per_block
) * EXT4_INODE_SIZE(sb
);
4070 bh
= sb_getblk(sb
, block
);
4072 ext4_error(sb
, "ext4_get_inode_loc", "unable to read "
4073 "inode block - inode=%lu, block=%llu",
4074 inode
->i_ino
, block
);
4077 if (!buffer_uptodate(bh
)) {
4081 * If the buffer has the write error flag, we have failed
4082 * to write out another inode in the same block. In this
4083 * case, we don't have to read the block because we may
4084 * read the old inode data successfully.
4086 if (buffer_write_io_error(bh
) && !buffer_uptodate(bh
))
4087 set_buffer_uptodate(bh
);
4089 if (buffer_uptodate(bh
)) {
4090 /* someone brought it uptodate while we waited */
4096 * If we have all information of the inode in memory and this
4097 * is the only valid inode in the block, we need not read the
4101 struct buffer_head
*bitmap_bh
;
4104 start
= inode_offset
& ~(inodes_per_block
- 1);
4106 /* Is the inode bitmap in cache? */
4107 bitmap_bh
= sb_getblk(sb
, ext4_inode_bitmap(sb
, gdp
));
4112 * If the inode bitmap isn't in cache then the
4113 * optimisation may end up performing two reads instead
4114 * of one, so skip it.
4116 if (!buffer_uptodate(bitmap_bh
)) {
4120 for (i
= start
; i
< start
+ inodes_per_block
; i
++) {
4121 if (i
== inode_offset
)
4123 if (ext4_test_bit(i
, bitmap_bh
->b_data
))
4127 if (i
== start
+ inodes_per_block
) {
4128 /* all other inodes are free, so skip I/O */
4129 memset(bh
->b_data
, 0, bh
->b_size
);
4130 set_buffer_uptodate(bh
);
4138 * If we need to do any I/O, try to pre-readahead extra
4139 * blocks from the inode table.
4141 if (EXT4_SB(sb
)->s_inode_readahead_blks
) {
4142 ext4_fsblk_t b
, end
, table
;
4145 table
= ext4_inode_table(sb
, gdp
);
4146 /* Make sure s_inode_readahead_blks is a power of 2 */
4147 while (EXT4_SB(sb
)->s_inode_readahead_blks
&
4148 (EXT4_SB(sb
)->s_inode_readahead_blks
-1))
4149 EXT4_SB(sb
)->s_inode_readahead_blks
=
4150 (EXT4_SB(sb
)->s_inode_readahead_blks
&
4151 (EXT4_SB(sb
)->s_inode_readahead_blks
-1));
4152 b
= block
& ~(EXT4_SB(sb
)->s_inode_readahead_blks
-1);
4155 end
= b
+ EXT4_SB(sb
)->s_inode_readahead_blks
;
4156 num
= EXT4_INODES_PER_GROUP(sb
);
4157 if (EXT4_HAS_RO_COMPAT_FEATURE(sb
,
4158 EXT4_FEATURE_RO_COMPAT_GDT_CSUM
))
4159 num
-= ext4_itable_unused_count(sb
, gdp
);
4160 table
+= num
/ inodes_per_block
;
4164 sb_breadahead(sb
, b
++);
4168 * There are other valid inodes in the buffer, this inode
4169 * has in-inode xattrs, or we don't have this inode in memory.
4170 * Read the block from disk.
4173 bh
->b_end_io
= end_buffer_read_sync
;
4174 submit_bh(READ_META
, bh
);
4176 if (!buffer_uptodate(bh
)) {
4177 ext4_error(sb
, __func__
,
4178 "unable to read inode block - inode=%lu, "
4179 "block=%llu", inode
->i_ino
, block
);
4189 int ext4_get_inode_loc(struct inode
*inode
, struct ext4_iloc
*iloc
)
4191 /* We have all inode data except xattrs in memory here. */
4192 return __ext4_get_inode_loc(inode
, iloc
,
4193 !(EXT4_I(inode
)->i_state
& EXT4_STATE_XATTR
));
4196 void ext4_set_inode_flags(struct inode
*inode
)
4198 unsigned int flags
= EXT4_I(inode
)->i_flags
;
4200 inode
->i_flags
&= ~(S_SYNC
|S_APPEND
|S_IMMUTABLE
|S_NOATIME
|S_DIRSYNC
);
4201 if (flags
& EXT4_SYNC_FL
)
4202 inode
->i_flags
|= S_SYNC
;
4203 if (flags
& EXT4_APPEND_FL
)
4204 inode
->i_flags
|= S_APPEND
;
4205 if (flags
& EXT4_IMMUTABLE_FL
)
4206 inode
->i_flags
|= S_IMMUTABLE
;
4207 if (flags
& EXT4_NOATIME_FL
)
4208 inode
->i_flags
|= S_NOATIME
;
4209 if (flags
& EXT4_DIRSYNC_FL
)
4210 inode
->i_flags
|= S_DIRSYNC
;
4213 /* Propagate flags from i_flags to EXT4_I(inode)->i_flags */
4214 void ext4_get_inode_flags(struct ext4_inode_info
*ei
)
4216 unsigned int flags
= ei
->vfs_inode
.i_flags
;
4218 ei
->i_flags
&= ~(EXT4_SYNC_FL
|EXT4_APPEND_FL
|
4219 EXT4_IMMUTABLE_FL
|EXT4_NOATIME_FL
|EXT4_DIRSYNC_FL
);
4221 ei
->i_flags
|= EXT4_SYNC_FL
;
4222 if (flags
& S_APPEND
)
4223 ei
->i_flags
|= EXT4_APPEND_FL
;
4224 if (flags
& S_IMMUTABLE
)
4225 ei
->i_flags
|= EXT4_IMMUTABLE_FL
;
4226 if (flags
& S_NOATIME
)
4227 ei
->i_flags
|= EXT4_NOATIME_FL
;
4228 if (flags
& S_DIRSYNC
)
4229 ei
->i_flags
|= EXT4_DIRSYNC_FL
;
4231 static blkcnt_t
ext4_inode_blocks(struct ext4_inode
*raw_inode
,
4232 struct ext4_inode_info
*ei
)
4235 struct inode
*inode
= &(ei
->vfs_inode
);
4236 struct super_block
*sb
= inode
->i_sb
;
4238 if (EXT4_HAS_RO_COMPAT_FEATURE(sb
,
4239 EXT4_FEATURE_RO_COMPAT_HUGE_FILE
)) {
4240 /* we are using combined 48 bit field */
4241 i_blocks
= ((u64
)le16_to_cpu(raw_inode
->i_blocks_high
)) << 32 |
4242 le32_to_cpu(raw_inode
->i_blocks_lo
);
4243 if (ei
->i_flags
& EXT4_HUGE_FILE_FL
) {
4244 /* i_blocks represent file system block size */
4245 return i_blocks
<< (inode
->i_blkbits
- 9);
4250 return le32_to_cpu(raw_inode
->i_blocks_lo
);
4254 struct inode
*ext4_iget(struct super_block
*sb
, unsigned long ino
)
4256 struct ext4_iloc iloc
;
4257 struct ext4_inode
*raw_inode
;
4258 struct ext4_inode_info
*ei
;
4259 struct buffer_head
*bh
;
4260 struct inode
*inode
;
4264 inode
= iget_locked(sb
, ino
);
4266 return ERR_PTR(-ENOMEM
);
4267 if (!(inode
->i_state
& I_NEW
))
4271 #ifdef CONFIG_EXT4_FS_POSIX_ACL
4272 ei
->i_acl
= EXT4_ACL_NOT_CACHED
;
4273 ei
->i_default_acl
= EXT4_ACL_NOT_CACHED
;
4276 ret
= __ext4_get_inode_loc(inode
, &iloc
, 0);
4280 raw_inode
= ext4_raw_inode(&iloc
);
4281 inode
->i_mode
= le16_to_cpu(raw_inode
->i_mode
);
4282 inode
->i_uid
= (uid_t
)le16_to_cpu(raw_inode
->i_uid_low
);
4283 inode
->i_gid
= (gid_t
)le16_to_cpu(raw_inode
->i_gid_low
);
4284 if (!(test_opt(inode
->i_sb
, NO_UID32
))) {
4285 inode
->i_uid
|= le16_to_cpu(raw_inode
->i_uid_high
) << 16;
4286 inode
->i_gid
|= le16_to_cpu(raw_inode
->i_gid_high
) << 16;
4288 inode
->i_nlink
= le16_to_cpu(raw_inode
->i_links_count
);
4291 ei
->i_dir_start_lookup
= 0;
4292 ei
->i_dtime
= le32_to_cpu(raw_inode
->i_dtime
);
4293 /* We now have enough fields to check if the inode was active or not.
4294 * This is needed because nfsd might try to access dead inodes
4295 * the test is that same one that e2fsck uses
4296 * NeilBrown 1999oct15
4298 if (inode
->i_nlink
== 0) {
4299 if (inode
->i_mode
== 0 ||
4300 !(EXT4_SB(inode
->i_sb
)->s_mount_state
& EXT4_ORPHAN_FS
)) {
4301 /* this inode is deleted */
4306 /* The only unlinked inodes we let through here have
4307 * valid i_mode and are being read by the orphan
4308 * recovery code: that's fine, we're about to complete
4309 * the process of deleting those. */
4311 ei
->i_flags
= le32_to_cpu(raw_inode
->i_flags
);
4312 inode
->i_blocks
= ext4_inode_blocks(raw_inode
, ei
);
4313 ei
->i_file_acl
= le32_to_cpu(raw_inode
->i_file_acl_lo
);
4314 if (EXT4_HAS_INCOMPAT_FEATURE(sb
, EXT4_FEATURE_INCOMPAT_64BIT
))
4316 ((__u64
)le16_to_cpu(raw_inode
->i_file_acl_high
)) << 32;
4317 inode
->i_size
= ext4_isize(raw_inode
);
4318 ei
->i_disksize
= inode
->i_size
;
4319 inode
->i_generation
= le32_to_cpu(raw_inode
->i_generation
);
4320 ei
->i_block_group
= iloc
.block_group
;
4322 * NOTE! The in-memory inode i_data array is in little-endian order
4323 * even on big-endian machines: we do NOT byteswap the block numbers!
4325 for (block
= 0; block
< EXT4_N_BLOCKS
; block
++)
4326 ei
->i_data
[block
] = raw_inode
->i_block
[block
];
4327 INIT_LIST_HEAD(&ei
->i_orphan
);
4329 if (EXT4_INODE_SIZE(inode
->i_sb
) > EXT4_GOOD_OLD_INODE_SIZE
) {
4330 ei
->i_extra_isize
= le16_to_cpu(raw_inode
->i_extra_isize
);
4331 if (EXT4_GOOD_OLD_INODE_SIZE
+ ei
->i_extra_isize
>
4332 EXT4_INODE_SIZE(inode
->i_sb
)) {
4337 if (ei
->i_extra_isize
== 0) {
4338 /* The extra space is currently unused. Use it. */
4339 ei
->i_extra_isize
= sizeof(struct ext4_inode
) -
4340 EXT4_GOOD_OLD_INODE_SIZE
;
4342 __le32
*magic
= (void *)raw_inode
+
4343 EXT4_GOOD_OLD_INODE_SIZE
+
4345 if (*magic
== cpu_to_le32(EXT4_XATTR_MAGIC
))
4346 ei
->i_state
|= EXT4_STATE_XATTR
;
4349 ei
->i_extra_isize
= 0;
4351 EXT4_INODE_GET_XTIME(i_ctime
, inode
, raw_inode
);
4352 EXT4_INODE_GET_XTIME(i_mtime
, inode
, raw_inode
);
4353 EXT4_INODE_GET_XTIME(i_atime
, inode
, raw_inode
);
4354 EXT4_EINODE_GET_XTIME(i_crtime
, ei
, raw_inode
);
4356 inode
->i_version
= le32_to_cpu(raw_inode
->i_disk_version
);
4357 if (EXT4_INODE_SIZE(inode
->i_sb
) > EXT4_GOOD_OLD_INODE_SIZE
) {
4358 if (EXT4_FITS_IN_INODE(raw_inode
, ei
, i_version_hi
))
4360 (__u64
)(le32_to_cpu(raw_inode
->i_version_hi
)) << 32;
4363 if (ei
->i_file_acl
&&
4365 (le32_to_cpu(EXT4_SB(sb
)->s_es
->s_first_data_block
) +
4366 EXT4_SB(sb
)->s_gdb_count
)) ||
4367 (ei
->i_file_acl
>= ext4_blocks_count(EXT4_SB(sb
)->s_es
)))) {
4368 ext4_error(sb
, __func__
,
4369 "bad extended attribute block %llu in inode #%lu",
4370 ei
->i_file_acl
, inode
->i_ino
);
4375 if (S_ISREG(inode
->i_mode
)) {
4376 inode
->i_op
= &ext4_file_inode_operations
;
4377 inode
->i_fop
= &ext4_file_operations
;
4378 ext4_set_aops(inode
);
4379 } else if (S_ISDIR(inode
->i_mode
)) {
4380 inode
->i_op
= &ext4_dir_inode_operations
;
4381 inode
->i_fop
= &ext4_dir_operations
;
4382 } else if (S_ISLNK(inode
->i_mode
)) {
4383 if (ext4_inode_is_fast_symlink(inode
)) {
4384 inode
->i_op
= &ext4_fast_symlink_inode_operations
;
4385 nd_terminate_link(ei
->i_data
, inode
->i_size
,
4386 sizeof(ei
->i_data
) - 1);
4388 inode
->i_op
= &ext4_symlink_inode_operations
;
4389 ext4_set_aops(inode
);
4391 } else if (S_ISCHR(inode
->i_mode
) || S_ISBLK(inode
->i_mode
) ||
4392 S_ISFIFO(inode
->i_mode
) || S_ISSOCK(inode
->i_mode
)) {
4393 inode
->i_op
= &ext4_special_inode_operations
;
4394 if (raw_inode
->i_block
[0])
4395 init_special_inode(inode
, inode
->i_mode
,
4396 old_decode_dev(le32_to_cpu(raw_inode
->i_block
[0])));
4398 init_special_inode(inode
, inode
->i_mode
,
4399 new_decode_dev(le32_to_cpu(raw_inode
->i_block
[1])));
4403 ext4_error(inode
->i_sb
, __func__
,
4404 "bogus i_mode (%o) for inode=%lu",
4405 inode
->i_mode
, inode
->i_ino
);
4409 ext4_set_inode_flags(inode
);
4410 unlock_new_inode(inode
);
4415 return ERR_PTR(ret
);
4418 static int ext4_inode_blocks_set(handle_t
*handle
,
4419 struct ext4_inode
*raw_inode
,
4420 struct ext4_inode_info
*ei
)
4422 struct inode
*inode
= &(ei
->vfs_inode
);
4423 u64 i_blocks
= inode
->i_blocks
;
4424 struct super_block
*sb
= inode
->i_sb
;
4426 if (i_blocks
<= ~0U) {
4428 * i_blocks can be represnted in a 32 bit variable
4429 * as multiple of 512 bytes
4431 raw_inode
->i_blocks_lo
= cpu_to_le32(i_blocks
);
4432 raw_inode
->i_blocks_high
= 0;
4433 ei
->i_flags
&= ~EXT4_HUGE_FILE_FL
;
4436 if (!EXT4_HAS_RO_COMPAT_FEATURE(sb
, EXT4_FEATURE_RO_COMPAT_HUGE_FILE
))
4439 if (i_blocks
<= 0xffffffffffffULL
) {
4441 * i_blocks can be represented in a 48 bit variable
4442 * as multiple of 512 bytes
4444 raw_inode
->i_blocks_lo
= cpu_to_le32(i_blocks
);
4445 raw_inode
->i_blocks_high
= cpu_to_le16(i_blocks
>> 32);
4446 ei
->i_flags
&= ~EXT4_HUGE_FILE_FL
;
4448 ei
->i_flags
|= EXT4_HUGE_FILE_FL
;
4449 /* i_block is stored in file system block size */
4450 i_blocks
= i_blocks
>> (inode
->i_blkbits
- 9);
4451 raw_inode
->i_blocks_lo
= cpu_to_le32(i_blocks
);
4452 raw_inode
->i_blocks_high
= cpu_to_le16(i_blocks
>> 32);
4458 * Post the struct inode info into an on-disk inode location in the
4459 * buffer-cache. This gobbles the caller's reference to the
4460 * buffer_head in the inode location struct.
4462 * The caller must have write access to iloc->bh.
4464 static int ext4_do_update_inode(handle_t
*handle
,
4465 struct inode
*inode
,
4466 struct ext4_iloc
*iloc
)
4468 struct ext4_inode
*raw_inode
= ext4_raw_inode(iloc
);
4469 struct ext4_inode_info
*ei
= EXT4_I(inode
);
4470 struct buffer_head
*bh
= iloc
->bh
;
4471 int err
= 0, rc
, block
;
4473 /* For fields not not tracking in the in-memory inode,
4474 * initialise them to zero for new inodes. */
4475 if (ei
->i_state
& EXT4_STATE_NEW
)
4476 memset(raw_inode
, 0, EXT4_SB(inode
->i_sb
)->s_inode_size
);
4478 ext4_get_inode_flags(ei
);
4479 raw_inode
->i_mode
= cpu_to_le16(inode
->i_mode
);
4480 if (!(test_opt(inode
->i_sb
, NO_UID32
))) {
4481 raw_inode
->i_uid_low
= cpu_to_le16(low_16_bits(inode
->i_uid
));
4482 raw_inode
->i_gid_low
= cpu_to_le16(low_16_bits(inode
->i_gid
));
4484 * Fix up interoperability with old kernels. Otherwise, old inodes get
4485 * re-used with the upper 16 bits of the uid/gid intact
4488 raw_inode
->i_uid_high
=
4489 cpu_to_le16(high_16_bits(inode
->i_uid
));
4490 raw_inode
->i_gid_high
=
4491 cpu_to_le16(high_16_bits(inode
->i_gid
));
4493 raw_inode
->i_uid_high
= 0;
4494 raw_inode
->i_gid_high
= 0;
4497 raw_inode
->i_uid_low
=
4498 cpu_to_le16(fs_high2lowuid(inode
->i_uid
));
4499 raw_inode
->i_gid_low
=
4500 cpu_to_le16(fs_high2lowgid(inode
->i_gid
));
4501 raw_inode
->i_uid_high
= 0;
4502 raw_inode
->i_gid_high
= 0;
4504 raw_inode
->i_links_count
= cpu_to_le16(inode
->i_nlink
);
4506 EXT4_INODE_SET_XTIME(i_ctime
, inode
, raw_inode
);
4507 EXT4_INODE_SET_XTIME(i_mtime
, inode
, raw_inode
);
4508 EXT4_INODE_SET_XTIME(i_atime
, inode
, raw_inode
);
4509 EXT4_EINODE_SET_XTIME(i_crtime
, ei
, raw_inode
);
4511 if (ext4_inode_blocks_set(handle
, raw_inode
, ei
))
4513 raw_inode
->i_dtime
= cpu_to_le32(ei
->i_dtime
);
4514 /* clear the migrate flag in the raw_inode */
4515 raw_inode
->i_flags
= cpu_to_le32(ei
->i_flags
& ~EXT4_EXT_MIGRATE
);
4516 if (EXT4_SB(inode
->i_sb
)->s_es
->s_creator_os
!=
4517 cpu_to_le32(EXT4_OS_HURD
))
4518 raw_inode
->i_file_acl_high
=
4519 cpu_to_le16(ei
->i_file_acl
>> 32);
4520 raw_inode
->i_file_acl_lo
= cpu_to_le32(ei
->i_file_acl
);
4521 ext4_isize_set(raw_inode
, ei
->i_disksize
);
4522 if (ei
->i_disksize
> 0x7fffffffULL
) {
4523 struct super_block
*sb
= inode
->i_sb
;
4524 if (!EXT4_HAS_RO_COMPAT_FEATURE(sb
,
4525 EXT4_FEATURE_RO_COMPAT_LARGE_FILE
) ||
4526 EXT4_SB(sb
)->s_es
->s_rev_level
==
4527 cpu_to_le32(EXT4_GOOD_OLD_REV
)) {
4528 /* If this is the first large file
4529 * created, add a flag to the superblock.
4531 err
= ext4_journal_get_write_access(handle
,
4532 EXT4_SB(sb
)->s_sbh
);
4535 ext4_update_dynamic_rev(sb
);
4536 EXT4_SET_RO_COMPAT_FEATURE(sb
,
4537 EXT4_FEATURE_RO_COMPAT_LARGE_FILE
);
4539 ext4_handle_sync(handle
);
4540 err
= ext4_handle_dirty_metadata(handle
, inode
,
4541 EXT4_SB(sb
)->s_sbh
);
4544 raw_inode
->i_generation
= cpu_to_le32(inode
->i_generation
);
4545 if (S_ISCHR(inode
->i_mode
) || S_ISBLK(inode
->i_mode
)) {
4546 if (old_valid_dev(inode
->i_rdev
)) {
4547 raw_inode
->i_block
[0] =
4548 cpu_to_le32(old_encode_dev(inode
->i_rdev
));
4549 raw_inode
->i_block
[1] = 0;
4551 raw_inode
->i_block
[0] = 0;
4552 raw_inode
->i_block
[1] =
4553 cpu_to_le32(new_encode_dev(inode
->i_rdev
));
4554 raw_inode
->i_block
[2] = 0;
4556 } else for (block
= 0; block
< EXT4_N_BLOCKS
; block
++)
4557 raw_inode
->i_block
[block
] = ei
->i_data
[block
];
4559 raw_inode
->i_disk_version
= cpu_to_le32(inode
->i_version
);
4560 if (ei
->i_extra_isize
) {
4561 if (EXT4_FITS_IN_INODE(raw_inode
, ei
, i_version_hi
))
4562 raw_inode
->i_version_hi
=
4563 cpu_to_le32(inode
->i_version
>> 32);
4564 raw_inode
->i_extra_isize
= cpu_to_le16(ei
->i_extra_isize
);
4567 BUFFER_TRACE(bh
, "call ext4_handle_dirty_metadata");
4568 rc
= ext4_handle_dirty_metadata(handle
, inode
, bh
);
4571 ei
->i_state
&= ~EXT4_STATE_NEW
;
4575 ext4_std_error(inode
->i_sb
, err
);
4580 * ext4_write_inode()
4582 * We are called from a few places:
4584 * - Within generic_file_write() for O_SYNC files.
4585 * Here, there will be no transaction running. We wait for any running
4586 * trasnaction to commit.
4588 * - Within sys_sync(), kupdate and such.
4589 * We wait on commit, if tol to.
4591 * - Within prune_icache() (PF_MEMALLOC == true)
4592 * Here we simply return. We can't afford to block kswapd on the
4595 * In all cases it is actually safe for us to return without doing anything,
4596 * because the inode has been copied into a raw inode buffer in
4597 * ext4_mark_inode_dirty(). This is a correctness thing for O_SYNC and for
4600 * Note that we are absolutely dependent upon all inode dirtiers doing the
4601 * right thing: they *must* call mark_inode_dirty() after dirtying info in
4602 * which we are interested.
4604 * It would be a bug for them to not do this. The code:
4606 * mark_inode_dirty(inode)
4608 * inode->i_size = expr;
4610 * is in error because a kswapd-driven write_inode() could occur while
4611 * `stuff()' is running, and the new i_size will be lost. Plus the inode
4612 * will no longer be on the superblock's dirty inode list.
4614 int ext4_write_inode(struct inode
*inode
, int wait
)
4616 if (current
->flags
& PF_MEMALLOC
)
4619 if (ext4_journal_current_handle()) {
4620 jbd_debug(1, "called recursively, non-PF_MEMALLOC!\n");
4628 return ext4_force_commit(inode
->i_sb
);
4631 int __ext4_write_dirty_metadata(struct inode
*inode
, struct buffer_head
*bh
)
4635 mark_buffer_dirty(bh
);
4636 if (inode
&& inode_needs_sync(inode
)) {
4637 sync_dirty_buffer(bh
);
4638 if (buffer_req(bh
) && !buffer_uptodate(bh
)) {
4639 ext4_error(inode
->i_sb
, __func__
,
4640 "IO error syncing inode, "
4641 "inode=%lu, block=%llu",
4643 (unsigned long long)bh
->b_blocknr
);
4653 * Called from notify_change.
4655 * We want to trap VFS attempts to truncate the file as soon as
4656 * possible. In particular, we want to make sure that when the VFS
4657 * shrinks i_size, we put the inode on the orphan list and modify
4658 * i_disksize immediately, so that during the subsequent flushing of
4659 * dirty pages and freeing of disk blocks, we can guarantee that any
4660 * commit will leave the blocks being flushed in an unused state on
4661 * disk. (On recovery, the inode will get truncated and the blocks will
4662 * be freed, so we have a strong guarantee that no future commit will
4663 * leave these blocks visible to the user.)
4665 * Another thing we have to assure is that if we are in ordered mode
4666 * and inode is still attached to the committing transaction, we must
4667 * we start writeout of all the dirty pages which are being truncated.
4668 * This way we are sure that all the data written in the previous
4669 * transaction are already on disk (truncate waits for pages under
4672 * Called with inode->i_mutex down.
4674 int ext4_setattr(struct dentry
*dentry
, struct iattr
*attr
)
4676 struct inode
*inode
= dentry
->d_inode
;
4678 const unsigned int ia_valid
= attr
->ia_valid
;
4680 error
= inode_change_ok(inode
, attr
);
4684 if ((ia_valid
& ATTR_UID
&& attr
->ia_uid
!= inode
->i_uid
) ||
4685 (ia_valid
& ATTR_GID
&& attr
->ia_gid
!= inode
->i_gid
)) {
4688 /* (user+group)*(old+new) structure, inode write (sb,
4689 * inode block, ? - but truncate inode update has it) */
4690 handle
= ext4_journal_start(inode
, 2*(EXT4_QUOTA_INIT_BLOCKS(inode
->i_sb
)+
4691 EXT4_QUOTA_DEL_BLOCKS(inode
->i_sb
))+3);
4692 if (IS_ERR(handle
)) {
4693 error
= PTR_ERR(handle
);
4696 error
= DQUOT_TRANSFER(inode
, attr
) ? -EDQUOT
: 0;
4698 ext4_journal_stop(handle
);
4701 /* Update corresponding info in inode so that everything is in
4702 * one transaction */
4703 if (attr
->ia_valid
& ATTR_UID
)
4704 inode
->i_uid
= attr
->ia_uid
;
4705 if (attr
->ia_valid
& ATTR_GID
)
4706 inode
->i_gid
= attr
->ia_gid
;
4707 error
= ext4_mark_inode_dirty(handle
, inode
);
4708 ext4_journal_stop(handle
);
4711 if (attr
->ia_valid
& ATTR_SIZE
) {
4712 if (!(EXT4_I(inode
)->i_flags
& EXT4_EXTENTS_FL
)) {
4713 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
4715 if (attr
->ia_size
> sbi
->s_bitmap_maxbytes
) {
4722 if (S_ISREG(inode
->i_mode
) &&
4723 attr
->ia_valid
& ATTR_SIZE
&& attr
->ia_size
< inode
->i_size
) {
4726 handle
= ext4_journal_start(inode
, 3);
4727 if (IS_ERR(handle
)) {
4728 error
= PTR_ERR(handle
);
4732 error
= ext4_orphan_add(handle
, inode
);
4733 EXT4_I(inode
)->i_disksize
= attr
->ia_size
;
4734 rc
= ext4_mark_inode_dirty(handle
, inode
);
4737 ext4_journal_stop(handle
);
4739 if (ext4_should_order_data(inode
)) {
4740 error
= ext4_begin_ordered_truncate(inode
,
4743 /* Do as much error cleanup as possible */
4744 handle
= ext4_journal_start(inode
, 3);
4745 if (IS_ERR(handle
)) {
4746 ext4_orphan_del(NULL
, inode
);
4749 ext4_orphan_del(handle
, inode
);
4750 ext4_journal_stop(handle
);
4756 rc
= inode_setattr(inode
, attr
);
4758 /* If inode_setattr's call to ext4_truncate failed to get a
4759 * transaction handle at all, we need to clean up the in-core
4760 * orphan list manually. */
4762 ext4_orphan_del(NULL
, inode
);
4764 if (!rc
&& (ia_valid
& ATTR_MODE
))
4765 rc
= ext4_acl_chmod(inode
);
4768 ext4_std_error(inode
->i_sb
, error
);
4774 int ext4_getattr(struct vfsmount
*mnt
, struct dentry
*dentry
,
4777 struct inode
*inode
;
4778 unsigned long delalloc_blocks
;
4780 inode
= dentry
->d_inode
;
4781 generic_fillattr(inode
, stat
);
4784 * We can't update i_blocks if the block allocation is delayed
4785 * otherwise in the case of system crash before the real block
4786 * allocation is done, we will have i_blocks inconsistent with
4787 * on-disk file blocks.
4788 * We always keep i_blocks updated together with real
4789 * allocation. But to not confuse with user, stat
4790 * will return the blocks that include the delayed allocation
4791 * blocks for this file.
4793 spin_lock(&EXT4_I(inode
)->i_block_reservation_lock
);
4794 delalloc_blocks
= EXT4_I(inode
)->i_reserved_data_blocks
;
4795 spin_unlock(&EXT4_I(inode
)->i_block_reservation_lock
);
4797 stat
->blocks
+= (delalloc_blocks
<< inode
->i_sb
->s_blocksize_bits
)>>9;
4801 static int ext4_indirect_trans_blocks(struct inode
*inode
, int nrblocks
,
4806 /* if nrblocks are contiguous */
4809 * With N contiguous data blocks, it need at most
4810 * N/EXT4_ADDR_PER_BLOCK(inode->i_sb) indirect blocks
4811 * 2 dindirect blocks
4814 indirects
= nrblocks
/ EXT4_ADDR_PER_BLOCK(inode
->i_sb
);
4815 return indirects
+ 3;
4818 * if nrblocks are not contiguous, worse case, each block touch
4819 * a indirect block, and each indirect block touch a double indirect
4820 * block, plus a triple indirect block
4822 indirects
= nrblocks
* 2 + 1;
4826 static int ext4_index_trans_blocks(struct inode
*inode
, int nrblocks
, int chunk
)
4828 if (!(EXT4_I(inode
)->i_flags
& EXT4_EXTENTS_FL
))
4829 return ext4_indirect_trans_blocks(inode
, nrblocks
, chunk
);
4830 return ext4_ext_index_trans_blocks(inode
, nrblocks
, chunk
);
4834 * Account for index blocks, block groups bitmaps and block group
4835 * descriptor blocks if modify datablocks and index blocks
4836 * worse case, the indexs blocks spread over different block groups
4838 * If datablocks are discontiguous, they are possible to spread over
4839 * different block groups too. If they are contiugous, with flexbg,
4840 * they could still across block group boundary.
4842 * Also account for superblock, inode, quota and xattr blocks
4844 int ext4_meta_trans_blocks(struct inode
*inode
, int nrblocks
, int chunk
)
4846 int groups
, gdpblocks
;
4851 * How many index blocks need to touch to modify nrblocks?
4852 * The "Chunk" flag indicating whether the nrblocks is
4853 * physically contiguous on disk
4855 * For Direct IO and fallocate, they calls get_block to allocate
4856 * one single extent at a time, so they could set the "Chunk" flag
4858 idxblocks
= ext4_index_trans_blocks(inode
, nrblocks
, chunk
);
4863 * Now let's see how many group bitmaps and group descriptors need
4873 if (groups
> EXT4_SB(inode
->i_sb
)->s_groups_count
)
4874 groups
= EXT4_SB(inode
->i_sb
)->s_groups_count
;
4875 if (groups
> EXT4_SB(inode
->i_sb
)->s_gdb_count
)
4876 gdpblocks
= EXT4_SB(inode
->i_sb
)->s_gdb_count
;
4878 /* bitmaps and block group descriptor blocks */
4879 ret
+= groups
+ gdpblocks
;
4881 /* Blocks for super block, inode, quota and xattr blocks */
4882 ret
+= EXT4_META_TRANS_BLOCKS(inode
->i_sb
);
4888 * Calulate the total number of credits to reserve to fit
4889 * the modification of a single pages into a single transaction,
4890 * which may include multiple chunks of block allocations.
4892 * This could be called via ext4_write_begin()
4894 * We need to consider the worse case, when
4895 * one new block per extent.
4897 int ext4_writepage_trans_blocks(struct inode
*inode
)
4899 int bpp
= ext4_journal_blocks_per_page(inode
);
4902 ret
= ext4_meta_trans_blocks(inode
, bpp
, 0);
4904 /* Account for data blocks for journalled mode */
4905 if (ext4_should_journal_data(inode
))
4911 * Calculate the journal credits for a chunk of data modification.
4913 * This is called from DIO, fallocate or whoever calling
4914 * ext4_get_blocks_wrap() to map/allocate a chunk of contigous disk blocks.
4916 * journal buffers for data blocks are not included here, as DIO
4917 * and fallocate do no need to journal data buffers.
4919 int ext4_chunk_trans_blocks(struct inode
*inode
, int nrblocks
)
4921 return ext4_meta_trans_blocks(inode
, nrblocks
, 1);
4925 * The caller must have previously called ext4_reserve_inode_write().
4926 * Give this, we know that the caller already has write access to iloc->bh.
4928 int ext4_mark_iloc_dirty(handle_t
*handle
,
4929 struct inode
*inode
, struct ext4_iloc
*iloc
)
4933 if (test_opt(inode
->i_sb
, I_VERSION
))
4934 inode_inc_iversion(inode
);
4936 /* the do_update_inode consumes one bh->b_count */
4939 /* ext4_do_update_inode() does jbd2_journal_dirty_metadata */
4940 err
= ext4_do_update_inode(handle
, inode
, iloc
);
4946 * On success, We end up with an outstanding reference count against
4947 * iloc->bh. This _must_ be cleaned up later.
4951 ext4_reserve_inode_write(handle_t
*handle
, struct inode
*inode
,
4952 struct ext4_iloc
*iloc
)
4956 err
= ext4_get_inode_loc(inode
, iloc
);
4958 BUFFER_TRACE(iloc
->bh
, "get_write_access");
4959 err
= ext4_journal_get_write_access(handle
, iloc
->bh
);
4965 ext4_std_error(inode
->i_sb
, err
);
4970 * Expand an inode by new_extra_isize bytes.
4971 * Returns 0 on success or negative error number on failure.
4973 static int ext4_expand_extra_isize(struct inode
*inode
,
4974 unsigned int new_extra_isize
,
4975 struct ext4_iloc iloc
,
4978 struct ext4_inode
*raw_inode
;
4979 struct ext4_xattr_ibody_header
*header
;
4980 struct ext4_xattr_entry
*entry
;
4982 if (EXT4_I(inode
)->i_extra_isize
>= new_extra_isize
)
4985 raw_inode
= ext4_raw_inode(&iloc
);
4987 header
= IHDR(inode
, raw_inode
);
4988 entry
= IFIRST(header
);
4990 /* No extended attributes present */
4991 if (!(EXT4_I(inode
)->i_state
& EXT4_STATE_XATTR
) ||
4992 header
->h_magic
!= cpu_to_le32(EXT4_XATTR_MAGIC
)) {
4993 memset((void *)raw_inode
+ EXT4_GOOD_OLD_INODE_SIZE
, 0,
4995 EXT4_I(inode
)->i_extra_isize
= new_extra_isize
;
4999 /* try to expand with EAs present */
5000 return ext4_expand_extra_isize_ea(inode
, new_extra_isize
,
5005 * What we do here is to mark the in-core inode as clean with respect to inode
5006 * dirtiness (it may still be data-dirty).
5007 * This means that the in-core inode may be reaped by prune_icache
5008 * without having to perform any I/O. This is a very good thing,
5009 * because *any* task may call prune_icache - even ones which
5010 * have a transaction open against a different journal.
5012 * Is this cheating? Not really. Sure, we haven't written the
5013 * inode out, but prune_icache isn't a user-visible syncing function.
5014 * Whenever the user wants stuff synced (sys_sync, sys_msync, sys_fsync)
5015 * we start and wait on commits.
5017 * Is this efficient/effective? Well, we're being nice to the system
5018 * by cleaning up our inodes proactively so they can be reaped
5019 * without I/O. But we are potentially leaving up to five seconds'
5020 * worth of inodes floating about which prune_icache wants us to
5021 * write out. One way to fix that would be to get prune_icache()
5022 * to do a write_super() to free up some memory. It has the desired
5025 int ext4_mark_inode_dirty(handle_t
*handle
, struct inode
*inode
)
5027 struct ext4_iloc iloc
;
5028 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
5029 static unsigned int mnt_count
;
5033 err
= ext4_reserve_inode_write(handle
, inode
, &iloc
);
5034 if (ext4_handle_valid(handle
) &&
5035 EXT4_I(inode
)->i_extra_isize
< sbi
->s_want_extra_isize
&&
5036 !(EXT4_I(inode
)->i_state
& EXT4_STATE_NO_EXPAND
)) {
5038 * We need extra buffer credits since we may write into EA block
5039 * with this same handle. If journal_extend fails, then it will
5040 * only result in a minor loss of functionality for that inode.
5041 * If this is felt to be critical, then e2fsck should be run to
5042 * force a large enough s_min_extra_isize.
5044 if ((jbd2_journal_extend(handle
,
5045 EXT4_DATA_TRANS_BLOCKS(inode
->i_sb
))) == 0) {
5046 ret
= ext4_expand_extra_isize(inode
,
5047 sbi
->s_want_extra_isize
,
5050 EXT4_I(inode
)->i_state
|= EXT4_STATE_NO_EXPAND
;
5052 le16_to_cpu(sbi
->s_es
->s_mnt_count
)) {
5053 ext4_warning(inode
->i_sb
, __func__
,
5054 "Unable to expand inode %lu. Delete"
5055 " some EAs or run e2fsck.",
5058 le16_to_cpu(sbi
->s_es
->s_mnt_count
);
5064 err
= ext4_mark_iloc_dirty(handle
, inode
, &iloc
);
5069 * ext4_dirty_inode() is called from __mark_inode_dirty()
5071 * We're really interested in the case where a file is being extended.
5072 * i_size has been changed by generic_commit_write() and we thus need
5073 * to include the updated inode in the current transaction.
5075 * Also, DQUOT_ALLOC_SPACE() will always dirty the inode when blocks
5076 * are allocated to the file.
5078 * If the inode is marked synchronous, we don't honour that here - doing
5079 * so would cause a commit on atime updates, which we don't bother doing.
5080 * We handle synchronous inodes at the highest possible level.
5082 void ext4_dirty_inode(struct inode
*inode
)
5084 handle_t
*current_handle
= ext4_journal_current_handle();
5087 if (!ext4_handle_valid(current_handle
)) {
5088 ext4_mark_inode_dirty(current_handle
, inode
);
5092 handle
= ext4_journal_start(inode
, 2);
5095 if (current_handle
&&
5096 current_handle
->h_transaction
!= handle
->h_transaction
) {
5097 /* This task has a transaction open against a different fs */
5098 printk(KERN_EMERG
"%s: transactions do not match!\n",
5101 jbd_debug(5, "marking dirty. outer handle=%p\n",
5103 ext4_mark_inode_dirty(handle
, inode
);
5105 ext4_journal_stop(handle
);
5112 * Bind an inode's backing buffer_head into this transaction, to prevent
5113 * it from being flushed to disk early. Unlike
5114 * ext4_reserve_inode_write, this leaves behind no bh reference and
5115 * returns no iloc structure, so the caller needs to repeat the iloc
5116 * lookup to mark the inode dirty later.
5118 static int ext4_pin_inode(handle_t
*handle
, struct inode
*inode
)
5120 struct ext4_iloc iloc
;
5124 err
= ext4_get_inode_loc(inode
, &iloc
);
5126 BUFFER_TRACE(iloc
.bh
, "get_write_access");
5127 err
= jbd2_journal_get_write_access(handle
, iloc
.bh
);
5129 err
= ext4_handle_dirty_metadata(handle
,
5135 ext4_std_error(inode
->i_sb
, err
);
5140 int ext4_change_inode_journal_flag(struct inode
*inode
, int val
)
5147 * We have to be very careful here: changing a data block's
5148 * journaling status dynamically is dangerous. If we write a
5149 * data block to the journal, change the status and then delete
5150 * that block, we risk forgetting to revoke the old log record
5151 * from the journal and so a subsequent replay can corrupt data.
5152 * So, first we make sure that the journal is empty and that
5153 * nobody is changing anything.
5156 journal
= EXT4_JOURNAL(inode
);
5159 if (is_journal_aborted(journal
))
5162 jbd2_journal_lock_updates(journal
);
5163 jbd2_journal_flush(journal
);
5166 * OK, there are no updates running now, and all cached data is
5167 * synced to disk. We are now in a completely consistent state
5168 * which doesn't have anything in the journal, and we know that
5169 * no filesystem updates are running, so it is safe to modify
5170 * the inode's in-core data-journaling state flag now.
5174 EXT4_I(inode
)->i_flags
|= EXT4_JOURNAL_DATA_FL
;
5176 EXT4_I(inode
)->i_flags
&= ~EXT4_JOURNAL_DATA_FL
;
5177 ext4_set_aops(inode
);
5179 jbd2_journal_unlock_updates(journal
);
5181 /* Finally we can mark the inode as dirty. */
5183 handle
= ext4_journal_start(inode
, 1);
5185 return PTR_ERR(handle
);
5187 err
= ext4_mark_inode_dirty(handle
, inode
);
5188 ext4_handle_sync(handle
);
5189 ext4_journal_stop(handle
);
5190 ext4_std_error(inode
->i_sb
, err
);
5195 static int ext4_bh_unmapped(handle_t
*handle
, struct buffer_head
*bh
)
5197 return !buffer_mapped(bh
);
5200 int ext4_page_mkwrite(struct vm_area_struct
*vma
, struct vm_fault
*vmf
)
5202 struct page
*page
= vmf
->page
;
5207 struct file
*file
= vma
->vm_file
;
5208 struct inode
*inode
= file
->f_path
.dentry
->d_inode
;
5209 struct address_space
*mapping
= inode
->i_mapping
;
5212 * Get i_alloc_sem to stop truncates messing with the inode. We cannot
5213 * get i_mutex because we are already holding mmap_sem.
5215 down_read(&inode
->i_alloc_sem
);
5216 size
= i_size_read(inode
);
5217 if (page
->mapping
!= mapping
|| size
<= page_offset(page
)
5218 || !PageUptodate(page
)) {
5219 /* page got truncated from under us? */
5223 if (PageMappedToDisk(page
))
5226 if (page
->index
== size
>> PAGE_CACHE_SHIFT
)
5227 len
= size
& ~PAGE_CACHE_MASK
;
5229 len
= PAGE_CACHE_SIZE
;
5231 if (page_has_buffers(page
)) {
5232 /* return if we have all the buffers mapped */
5233 if (!walk_page_buffers(NULL
, page_buffers(page
), 0, len
, NULL
,
5238 * OK, we need to fill the hole... Do write_begin write_end
5239 * to do block allocation/reservation.We are not holding
5240 * inode.i__mutex here. That allow * parallel write_begin,
5241 * write_end call. lock_page prevent this from happening
5242 * on the same page though
5244 ret
= mapping
->a_ops
->write_begin(file
, mapping
, page_offset(page
),
5245 len
, AOP_FLAG_UNINTERRUPTIBLE
, &page
, &fsdata
);
5248 ret
= mapping
->a_ops
->write_end(file
, mapping
, page_offset(page
),
5249 len
, len
, page
, fsdata
);
5255 ret
= VM_FAULT_SIGBUS
;
5256 up_read(&inode
->i_alloc_sem
);