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>
41 #include "ext4_jbd2.h"
44 #include "ext4_extents.h"
46 #include <trace/events/ext4.h>
48 #define MPAGE_DA_EXTENT_TAIL 0x01
50 static inline int ext4_begin_ordered_truncate(struct inode
*inode
,
53 return jbd2_journal_begin_ordered_truncate(
54 EXT4_SB(inode
->i_sb
)->s_journal
,
55 &EXT4_I(inode
)->jinode
,
59 static void ext4_invalidatepage(struct page
*page
, unsigned long offset
);
62 * Test whether an inode is a fast symlink.
64 static int ext4_inode_is_fast_symlink(struct inode
*inode
)
66 int ea_blocks
= EXT4_I(inode
)->i_file_acl
?
67 (inode
->i_sb
->s_blocksize
>> 9) : 0;
69 return (S_ISLNK(inode
->i_mode
) && inode
->i_blocks
- ea_blocks
== 0);
73 * The ext4 forget function must perform a revoke if we are freeing data
74 * which has been journaled. Metadata (eg. indirect blocks) must be
75 * revoked in all cases.
77 * "bh" may be NULL: a metadata block may have been freed from memory
78 * but there may still be a record of it in the journal, and that record
79 * still needs to be revoked.
81 * If the handle isn't valid we're not journaling, but we still need to
82 * call into ext4_journal_revoke() to put the buffer head.
84 int ext4_forget(handle_t
*handle
, int is_metadata
, struct inode
*inode
,
85 struct buffer_head
*bh
, ext4_fsblk_t blocknr
)
91 BUFFER_TRACE(bh
, "enter");
93 jbd_debug(4, "forgetting bh %p: is_metadata = %d, mode %o, "
95 bh
, is_metadata
, inode
->i_mode
,
96 test_opt(inode
->i_sb
, DATA_FLAGS
));
98 /* Never use the revoke function if we are doing full data
99 * journaling: there is no need to, and a V1 superblock won't
100 * support it. Otherwise, only skip the revoke on un-journaled
103 if (test_opt(inode
->i_sb
, DATA_FLAGS
) == EXT4_MOUNT_JOURNAL_DATA
||
104 (!is_metadata
&& !ext4_should_journal_data(inode
))) {
106 BUFFER_TRACE(bh
, "call jbd2_journal_forget");
107 return ext4_journal_forget(handle
, bh
);
113 * data!=journal && (is_metadata || should_journal_data(inode))
115 BUFFER_TRACE(bh
, "call ext4_journal_revoke");
116 err
= ext4_journal_revoke(handle
, blocknr
, bh
);
118 ext4_abort(inode
->i_sb
, __func__
,
119 "error %d when attempting revoke", err
);
120 BUFFER_TRACE(bh
, "exit");
125 * Work out how many blocks we need to proceed with the next chunk of a
126 * truncate transaction.
128 static unsigned long blocks_for_truncate(struct inode
*inode
)
132 needed
= inode
->i_blocks
>> (inode
->i_sb
->s_blocksize_bits
- 9);
134 /* Give ourselves just enough room to cope with inodes in which
135 * i_blocks is corrupt: we've seen disk corruptions in the past
136 * which resulted in random data in an inode which looked enough
137 * like a regular file for ext4 to try to delete it. Things
138 * will go a bit crazy if that happens, but at least we should
139 * try not to panic the whole kernel. */
143 /* But we need to bound the transaction so we don't overflow the
145 if (needed
> EXT4_MAX_TRANS_DATA
)
146 needed
= EXT4_MAX_TRANS_DATA
;
148 return EXT4_DATA_TRANS_BLOCKS(inode
->i_sb
) + needed
;
152 * Truncate transactions can be complex and absolutely huge. So we need to
153 * be able to restart the transaction at a conventient checkpoint to make
154 * sure we don't overflow the journal.
156 * start_transaction gets us a new handle for a truncate transaction,
157 * and extend_transaction tries to extend the existing one a bit. If
158 * extend fails, we need to propagate the failure up and restart the
159 * transaction in the top-level truncate loop. --sct
161 static handle_t
*start_transaction(struct inode
*inode
)
165 result
= ext4_journal_start(inode
, blocks_for_truncate(inode
));
169 ext4_std_error(inode
->i_sb
, PTR_ERR(result
));
174 * Try to extend this transaction for the purposes of truncation.
176 * Returns 0 if we managed to create more room. If we can't create more
177 * room, and the transaction must be restarted we return 1.
179 static int try_to_extend_transaction(handle_t
*handle
, struct inode
*inode
)
181 if (!ext4_handle_valid(handle
))
183 if (ext4_handle_has_enough_credits(handle
, EXT4_RESERVE_TRANS_BLOCKS
+1))
185 if (!ext4_journal_extend(handle
, blocks_for_truncate(inode
)))
191 * Restart the transaction associated with *handle. This does a commit,
192 * so before we call here everything must be consistently dirtied against
195 int ext4_truncate_restart_trans(handle_t
*handle
, struct inode
*inode
,
201 * Drop i_data_sem to avoid deadlock with ext4_get_blocks At this
202 * moment, get_block can be called only for blocks inside i_size since
203 * page cache has been already dropped and writes are blocked by
204 * i_mutex. So we can safely drop the i_data_sem here.
206 BUG_ON(EXT4_JOURNAL(inode
) == NULL
);
207 jbd_debug(2, "restarting handle %p\n", handle
);
208 up_write(&EXT4_I(inode
)->i_data_sem
);
209 ret
= ext4_journal_restart(handle
, blocks_for_truncate(inode
));
210 down_write(&EXT4_I(inode
)->i_data_sem
);
216 * Called at the last iput() if i_nlink is zero.
218 void ext4_delete_inode(struct inode
*inode
)
223 if (ext4_should_order_data(inode
))
224 ext4_begin_ordered_truncate(inode
, 0);
225 truncate_inode_pages(&inode
->i_data
, 0);
227 if (is_bad_inode(inode
))
230 handle
= ext4_journal_start(inode
, blocks_for_truncate(inode
)+3);
231 if (IS_ERR(handle
)) {
232 ext4_std_error(inode
->i_sb
, PTR_ERR(handle
));
234 * If we're going to skip the normal cleanup, we still need to
235 * make sure that the in-core orphan linked list is properly
238 ext4_orphan_del(NULL
, inode
);
243 ext4_handle_sync(handle
);
245 err
= ext4_mark_inode_dirty(handle
, inode
);
247 ext4_warning(inode
->i_sb
, __func__
,
248 "couldn't mark inode dirty (err %d)", err
);
252 ext4_truncate(inode
);
255 * ext4_ext_truncate() doesn't reserve any slop when it
256 * restarts journal transactions; therefore there may not be
257 * enough credits left in the handle to remove the inode from
258 * the orphan list and set the dtime field.
260 if (!ext4_handle_has_enough_credits(handle
, 3)) {
261 err
= ext4_journal_extend(handle
, 3);
263 err
= ext4_journal_restart(handle
, 3);
265 ext4_warning(inode
->i_sb
, __func__
,
266 "couldn't extend journal (err %d)", err
);
268 ext4_journal_stop(handle
);
274 * Kill off the orphan record which ext4_truncate created.
275 * AKPM: I think this can be inside the above `if'.
276 * Note that ext4_orphan_del() has to be able to cope with the
277 * deletion of a non-existent orphan - this is because we don't
278 * know if ext4_truncate() actually created an orphan record.
279 * (Well, we could do this if we need to, but heck - it works)
281 ext4_orphan_del(handle
, inode
);
282 EXT4_I(inode
)->i_dtime
= get_seconds();
285 * One subtle ordering requirement: if anything has gone wrong
286 * (transaction abort, IO errors, whatever), then we can still
287 * do these next steps (the fs will already have been marked as
288 * having errors), but we can't free the inode if the mark_dirty
291 if (ext4_mark_inode_dirty(handle
, inode
))
292 /* If that failed, just do the required in-core inode clear. */
295 ext4_free_inode(handle
, inode
);
296 ext4_journal_stop(handle
);
299 clear_inode(inode
); /* We must guarantee clearing of inode... */
305 struct buffer_head
*bh
;
308 static inline void add_chain(Indirect
*p
, struct buffer_head
*bh
, __le32
*v
)
310 p
->key
= *(p
->p
= v
);
315 * ext4_block_to_path - parse the block number into array of offsets
316 * @inode: inode in question (we are only interested in its superblock)
317 * @i_block: block number to be parsed
318 * @offsets: array to store the offsets in
319 * @boundary: set this non-zero if the referred-to block is likely to be
320 * followed (on disk) by an indirect block.
322 * To store the locations of file's data ext4 uses a data structure common
323 * for UNIX filesystems - tree of pointers anchored in the inode, with
324 * data blocks at leaves and indirect blocks in intermediate nodes.
325 * This function translates the block number into path in that tree -
326 * return value is the path length and @offsets[n] is the offset of
327 * pointer to (n+1)th node in the nth one. If @block is out of range
328 * (negative or too large) warning is printed and zero returned.
330 * Note: function doesn't find node addresses, so no IO is needed. All
331 * we need to know is the capacity of indirect blocks (taken from the
336 * Portability note: the last comparison (check that we fit into triple
337 * indirect block) is spelled differently, because otherwise on an
338 * architecture with 32-bit longs and 8Kb pages we might get into trouble
339 * if our filesystem had 8Kb blocks. We might use long long, but that would
340 * kill us on x86. Oh, well, at least the sign propagation does not matter -
341 * i_block would have to be negative in the very beginning, so we would not
345 static int ext4_block_to_path(struct inode
*inode
,
347 ext4_lblk_t offsets
[4], int *boundary
)
349 int ptrs
= EXT4_ADDR_PER_BLOCK(inode
->i_sb
);
350 int ptrs_bits
= EXT4_ADDR_PER_BLOCK_BITS(inode
->i_sb
);
351 const long direct_blocks
= EXT4_NDIR_BLOCKS
,
352 indirect_blocks
= ptrs
,
353 double_blocks
= (1 << (ptrs_bits
* 2));
357 if (i_block
< direct_blocks
) {
358 offsets
[n
++] = i_block
;
359 final
= direct_blocks
;
360 } else if ((i_block
-= direct_blocks
) < indirect_blocks
) {
361 offsets
[n
++] = EXT4_IND_BLOCK
;
362 offsets
[n
++] = i_block
;
364 } else if ((i_block
-= indirect_blocks
) < double_blocks
) {
365 offsets
[n
++] = EXT4_DIND_BLOCK
;
366 offsets
[n
++] = i_block
>> ptrs_bits
;
367 offsets
[n
++] = i_block
& (ptrs
- 1);
369 } else if (((i_block
-= double_blocks
) >> (ptrs_bits
* 2)) < ptrs
) {
370 offsets
[n
++] = EXT4_TIND_BLOCK
;
371 offsets
[n
++] = i_block
>> (ptrs_bits
* 2);
372 offsets
[n
++] = (i_block
>> ptrs_bits
) & (ptrs
- 1);
373 offsets
[n
++] = i_block
& (ptrs
- 1);
376 ext4_warning(inode
->i_sb
, "ext4_block_to_path",
377 "block %lu > max in inode %lu",
378 i_block
+ direct_blocks
+
379 indirect_blocks
+ double_blocks
, inode
->i_ino
);
382 *boundary
= final
- 1 - (i_block
& (ptrs
- 1));
386 static int __ext4_check_blockref(const char *function
, struct inode
*inode
,
387 __le32
*p
, unsigned int max
)
392 while (bref
< p
+max
) {
393 blk
= le32_to_cpu(*bref
++);
395 unlikely(!ext4_data_block_valid(EXT4_SB(inode
->i_sb
),
397 ext4_error(inode
->i_sb
, function
,
398 "invalid block reference %u "
399 "in inode #%lu", blk
, inode
->i_ino
);
407 #define ext4_check_indirect_blockref(inode, bh) \
408 __ext4_check_blockref(__func__, inode, (__le32 *)(bh)->b_data, \
409 EXT4_ADDR_PER_BLOCK((inode)->i_sb))
411 #define ext4_check_inode_blockref(inode) \
412 __ext4_check_blockref(__func__, inode, EXT4_I(inode)->i_data, \
416 * ext4_get_branch - read the chain of indirect blocks leading to data
417 * @inode: inode in question
418 * @depth: depth of the chain (1 - direct pointer, etc.)
419 * @offsets: offsets of pointers in inode/indirect blocks
420 * @chain: place to store the result
421 * @err: here we store the error value
423 * Function fills the array of triples <key, p, bh> and returns %NULL
424 * if everything went OK or the pointer to the last filled triple
425 * (incomplete one) otherwise. Upon the return chain[i].key contains
426 * the number of (i+1)-th block in the chain (as it is stored in memory,
427 * i.e. little-endian 32-bit), chain[i].p contains the address of that
428 * number (it points into struct inode for i==0 and into the bh->b_data
429 * for i>0) and chain[i].bh points to the buffer_head of i-th indirect
430 * block for i>0 and NULL for i==0. In other words, it holds the block
431 * numbers of the chain, addresses they were taken from (and where we can
432 * verify that chain did not change) and buffer_heads hosting these
435 * Function stops when it stumbles upon zero pointer (absent block)
436 * (pointer to last triple returned, *@err == 0)
437 * or when it gets an IO error reading an indirect block
438 * (ditto, *@err == -EIO)
439 * or when it reads all @depth-1 indirect blocks successfully and finds
440 * the whole chain, all way to the data (returns %NULL, *err == 0).
442 * Need to be called with
443 * down_read(&EXT4_I(inode)->i_data_sem)
445 static Indirect
*ext4_get_branch(struct inode
*inode
, int depth
,
446 ext4_lblk_t
*offsets
,
447 Indirect chain
[4], int *err
)
449 struct super_block
*sb
= inode
->i_sb
;
451 struct buffer_head
*bh
;
454 /* i_data is not going away, no lock needed */
455 add_chain(chain
, NULL
, EXT4_I(inode
)->i_data
+ *offsets
);
459 bh
= sb_getblk(sb
, le32_to_cpu(p
->key
));
463 if (!bh_uptodate_or_lock(bh
)) {
464 if (bh_submit_read(bh
) < 0) {
468 /* validate block references */
469 if (ext4_check_indirect_blockref(inode
, bh
)) {
475 add_chain(++p
, bh
, (__le32
*)bh
->b_data
+ *++offsets
);
489 * ext4_find_near - find a place for allocation with sufficient locality
491 * @ind: descriptor of indirect block.
493 * This function returns the preferred place for block allocation.
494 * It is used when heuristic for sequential allocation fails.
496 * + if there is a block to the left of our position - allocate near it.
497 * + if pointer will live in indirect block - allocate near that block.
498 * + if pointer will live in inode - allocate in the same
501 * In the latter case we colour the starting block by the callers PID to
502 * prevent it from clashing with concurrent allocations for a different inode
503 * in the same block group. The PID is used here so that functionally related
504 * files will be close-by on-disk.
506 * Caller must make sure that @ind is valid and will stay that way.
508 static ext4_fsblk_t
ext4_find_near(struct inode
*inode
, Indirect
*ind
)
510 struct ext4_inode_info
*ei
= EXT4_I(inode
);
511 __le32
*start
= ind
->bh
? (__le32
*) ind
->bh
->b_data
: ei
->i_data
;
513 ext4_fsblk_t bg_start
;
514 ext4_fsblk_t last_block
;
515 ext4_grpblk_t colour
;
516 ext4_group_t block_group
;
517 int flex_size
= ext4_flex_bg_size(EXT4_SB(inode
->i_sb
));
519 /* Try to find previous block */
520 for (p
= ind
->p
- 1; p
>= start
; p
--) {
522 return le32_to_cpu(*p
);
525 /* No such thing, so let's try location of indirect block */
527 return ind
->bh
->b_blocknr
;
530 * It is going to be referred to from the inode itself? OK, just put it
531 * into the same cylinder group then.
533 block_group
= ei
->i_block_group
;
534 if (flex_size
>= EXT4_FLEX_SIZE_DIR_ALLOC_SCHEME
) {
535 block_group
&= ~(flex_size
-1);
536 if (S_ISREG(inode
->i_mode
))
539 bg_start
= ext4_group_first_block_no(inode
->i_sb
, block_group
);
540 last_block
= ext4_blocks_count(EXT4_SB(inode
->i_sb
)->s_es
) - 1;
543 * If we are doing delayed allocation, we don't need take
544 * colour into account.
546 if (test_opt(inode
->i_sb
, DELALLOC
))
549 if (bg_start
+ EXT4_BLOCKS_PER_GROUP(inode
->i_sb
) <= last_block
)
550 colour
= (current
->pid
% 16) *
551 (EXT4_BLOCKS_PER_GROUP(inode
->i_sb
) / 16);
553 colour
= (current
->pid
% 16) * ((last_block
- bg_start
) / 16);
554 return bg_start
+ colour
;
558 * ext4_find_goal - find a preferred place for allocation.
560 * @block: block we want
561 * @partial: pointer to the last triple within a chain
563 * Normally this function find the preferred place for block allocation,
566 static ext4_fsblk_t
ext4_find_goal(struct inode
*inode
, ext4_lblk_t block
,
570 * XXX need to get goal block from mballoc's data structures
573 return ext4_find_near(inode
, partial
);
577 * ext4_blks_to_allocate: Look up the block map and count the number
578 * of direct blocks need to be allocated for the given branch.
580 * @branch: chain of indirect blocks
581 * @k: number of blocks need for indirect blocks
582 * @blks: number of data blocks to be mapped.
583 * @blocks_to_boundary: the offset in the indirect block
585 * return the total number of blocks to be allocate, including the
586 * direct and indirect blocks.
588 static int ext4_blks_to_allocate(Indirect
*branch
, int k
, unsigned int blks
,
589 int blocks_to_boundary
)
591 unsigned int count
= 0;
594 * Simple case, [t,d]Indirect block(s) has not allocated yet
595 * then it's clear blocks on that path have not allocated
598 /* right now we don't handle cross boundary allocation */
599 if (blks
< blocks_to_boundary
+ 1)
602 count
+= blocks_to_boundary
+ 1;
607 while (count
< blks
&& count
<= blocks_to_boundary
&&
608 le32_to_cpu(*(branch
[0].p
+ count
)) == 0) {
615 * ext4_alloc_blocks: multiple allocate blocks needed for a branch
616 * @indirect_blks: the number of blocks need to allocate for indirect
619 * @new_blocks: on return it will store the new block numbers for
620 * the indirect blocks(if needed) and the first direct block,
621 * @blks: on return it will store the total number of allocated
624 static int ext4_alloc_blocks(handle_t
*handle
, struct inode
*inode
,
625 ext4_lblk_t iblock
, ext4_fsblk_t goal
,
626 int indirect_blks
, int blks
,
627 ext4_fsblk_t new_blocks
[4], int *err
)
629 struct ext4_allocation_request ar
;
631 unsigned long count
= 0, blk_allocated
= 0;
633 ext4_fsblk_t current_block
= 0;
637 * Here we try to allocate the requested multiple blocks at once,
638 * on a best-effort basis.
639 * To build a branch, we should allocate blocks for
640 * the indirect blocks(if not allocated yet), and at least
641 * the first direct block of this branch. That's the
642 * minimum number of blocks need to allocate(required)
644 /* first we try to allocate the indirect blocks */
645 target
= indirect_blks
;
648 /* allocating blocks for indirect blocks and direct blocks */
649 current_block
= ext4_new_meta_blocks(handle
, inode
,
655 /* allocate blocks for indirect blocks */
656 while (index
< indirect_blks
&& count
) {
657 new_blocks
[index
++] = current_block
++;
662 * save the new block number
663 * for the first direct block
665 new_blocks
[index
] = current_block
;
666 printk(KERN_INFO
"%s returned more blocks than "
667 "requested\n", __func__
);
673 target
= blks
- count
;
674 blk_allocated
= count
;
677 /* Now allocate data blocks */
678 memset(&ar
, 0, sizeof(ar
));
683 if (S_ISREG(inode
->i_mode
))
684 /* enable in-core preallocation only for regular files */
685 ar
.flags
= EXT4_MB_HINT_DATA
;
687 current_block
= ext4_mb_new_blocks(handle
, &ar
, err
);
689 if (*err
&& (target
== blks
)) {
691 * if the allocation failed and we didn't allocate
697 if (target
== blks
) {
699 * save the new block number
700 * for the first direct block
702 new_blocks
[index
] = current_block
;
704 blk_allocated
+= ar
.len
;
707 /* total number of blocks allocated for direct blocks */
712 for (i
= 0; i
< index
; i
++)
713 ext4_free_blocks(handle
, inode
, new_blocks
[i
], 1, 0);
718 * ext4_alloc_branch - allocate and set up a chain of blocks.
720 * @indirect_blks: number of allocated indirect blocks
721 * @blks: number of allocated direct blocks
722 * @offsets: offsets (in the blocks) to store the pointers to next.
723 * @branch: place to store the chain in.
725 * This function allocates blocks, zeroes out all but the last one,
726 * links them into chain and (if we are synchronous) writes them to disk.
727 * In other words, it prepares a branch that can be spliced onto the
728 * inode. It stores the information about that chain in the branch[], in
729 * the same format as ext4_get_branch() would do. We are calling it after
730 * we had read the existing part of chain and partial points to the last
731 * triple of that (one with zero ->key). Upon the exit we have the same
732 * picture as after the successful ext4_get_block(), except that in one
733 * place chain is disconnected - *branch->p is still zero (we did not
734 * set the last link), but branch->key contains the number that should
735 * be placed into *branch->p to fill that gap.
737 * If allocation fails we free all blocks we've allocated (and forget
738 * their buffer_heads) and return the error value the from failed
739 * ext4_alloc_block() (normally -ENOSPC). Otherwise we set the chain
740 * as described above and return 0.
742 static int ext4_alloc_branch(handle_t
*handle
, struct inode
*inode
,
743 ext4_lblk_t iblock
, int indirect_blks
,
744 int *blks
, ext4_fsblk_t goal
,
745 ext4_lblk_t
*offsets
, Indirect
*branch
)
747 int blocksize
= inode
->i_sb
->s_blocksize
;
750 struct buffer_head
*bh
;
752 ext4_fsblk_t new_blocks
[4];
753 ext4_fsblk_t current_block
;
755 num
= ext4_alloc_blocks(handle
, inode
, iblock
, goal
, indirect_blks
,
756 *blks
, new_blocks
, &err
);
760 branch
[0].key
= cpu_to_le32(new_blocks
[0]);
762 * metadata blocks and data blocks are allocated.
764 for (n
= 1; n
<= indirect_blks
; n
++) {
766 * Get buffer_head for parent block, zero it out
767 * and set the pointer to new one, then send
770 bh
= sb_getblk(inode
->i_sb
, new_blocks
[n
-1]);
773 BUFFER_TRACE(bh
, "call get_create_access");
774 err
= ext4_journal_get_create_access(handle
, bh
);
776 /* Don't brelse(bh) here; it's done in
777 * ext4_journal_forget() below */
782 memset(bh
->b_data
, 0, blocksize
);
783 branch
[n
].p
= (__le32
*) bh
->b_data
+ offsets
[n
];
784 branch
[n
].key
= cpu_to_le32(new_blocks
[n
]);
785 *branch
[n
].p
= branch
[n
].key
;
786 if (n
== indirect_blks
) {
787 current_block
= new_blocks
[n
];
789 * End of chain, update the last new metablock of
790 * the chain to point to the new allocated
791 * data blocks numbers
793 for (i
= 1; i
< num
; i
++)
794 *(branch
[n
].p
+ i
) = cpu_to_le32(++current_block
);
796 BUFFER_TRACE(bh
, "marking uptodate");
797 set_buffer_uptodate(bh
);
800 BUFFER_TRACE(bh
, "call ext4_handle_dirty_metadata");
801 err
= ext4_handle_dirty_metadata(handle
, inode
, bh
);
808 /* Allocation failed, free what we already allocated */
809 for (i
= 1; i
<= n
; i
++) {
810 BUFFER_TRACE(branch
[i
].bh
, "call jbd2_journal_forget");
811 ext4_journal_forget(handle
, branch
[i
].bh
);
813 for (i
= 0; i
< indirect_blks
; i
++)
814 ext4_free_blocks(handle
, inode
, new_blocks
[i
], 1, 0);
816 ext4_free_blocks(handle
, inode
, new_blocks
[i
], num
, 0);
822 * ext4_splice_branch - splice the allocated branch onto inode.
824 * @block: (logical) number of block we are adding
825 * @chain: chain of indirect blocks (with a missing link - see
827 * @where: location of missing link
828 * @num: number of indirect blocks we are adding
829 * @blks: number of direct blocks we are adding
831 * This function fills the missing link and does all housekeeping needed in
832 * inode (->i_blocks, etc.). In case of success we end up with the full
833 * chain to new block and return 0.
835 static int ext4_splice_branch(handle_t
*handle
, struct inode
*inode
,
836 ext4_lblk_t block
, Indirect
*where
, int num
,
841 ext4_fsblk_t current_block
;
844 * If we're splicing into a [td]indirect block (as opposed to the
845 * inode) then we need to get write access to the [td]indirect block
849 BUFFER_TRACE(where
->bh
, "get_write_access");
850 err
= ext4_journal_get_write_access(handle
, where
->bh
);
856 *where
->p
= where
->key
;
859 * Update the host buffer_head or inode to point to more just allocated
860 * direct blocks blocks
862 if (num
== 0 && blks
> 1) {
863 current_block
= le32_to_cpu(where
->key
) + 1;
864 for (i
= 1; i
< blks
; i
++)
865 *(where
->p
+ i
) = cpu_to_le32(current_block
++);
868 /* We are done with atomic stuff, now do the rest of housekeeping */
869 /* had we spliced it onto indirect block? */
872 * If we spliced it onto an indirect block, we haven't
873 * altered the inode. Note however that if it is being spliced
874 * onto an indirect block at the very end of the file (the
875 * file is growing) then we *will* alter the inode to reflect
876 * the new i_size. But that is not done here - it is done in
877 * generic_commit_write->__mark_inode_dirty->ext4_dirty_inode.
879 jbd_debug(5, "splicing indirect only\n");
880 BUFFER_TRACE(where
->bh
, "call ext4_handle_dirty_metadata");
881 err
= ext4_handle_dirty_metadata(handle
, inode
, where
->bh
);
886 * OK, we spliced it into the inode itself on a direct block.
888 ext4_mark_inode_dirty(handle
, inode
);
889 jbd_debug(5, "splicing direct\n");
894 for (i
= 1; i
<= num
; i
++) {
895 BUFFER_TRACE(where
[i
].bh
, "call jbd2_journal_forget");
896 ext4_journal_forget(handle
, where
[i
].bh
);
897 ext4_free_blocks(handle
, inode
,
898 le32_to_cpu(where
[i
-1].key
), 1, 0);
900 ext4_free_blocks(handle
, inode
, le32_to_cpu(where
[num
].key
), blks
, 0);
906 * The ext4_ind_get_blocks() function handles non-extents inodes
907 * (i.e., using the traditional indirect/double-indirect i_blocks
908 * scheme) for ext4_get_blocks().
910 * Allocation strategy is simple: if we have to allocate something, we will
911 * have to go the whole way to leaf. So let's do it before attaching anything
912 * to tree, set linkage between the newborn blocks, write them if sync is
913 * required, recheck the path, free and repeat if check fails, otherwise
914 * set the last missing link (that will protect us from any truncate-generated
915 * removals - all blocks on the path are immune now) and possibly force the
916 * write on the parent block.
917 * That has a nice additional property: no special recovery from the failed
918 * allocations is needed - we simply release blocks and do not touch anything
919 * reachable from inode.
921 * `handle' can be NULL if create == 0.
923 * return > 0, # of blocks mapped or allocated.
924 * return = 0, if plain lookup failed.
925 * return < 0, error case.
927 * The ext4_ind_get_blocks() function should be called with
928 * down_write(&EXT4_I(inode)->i_data_sem) if allocating filesystem
929 * blocks (i.e., flags has EXT4_GET_BLOCKS_CREATE set) or
930 * down_read(&EXT4_I(inode)->i_data_sem) if not allocating file system
933 static int ext4_ind_get_blocks(handle_t
*handle
, struct inode
*inode
,
934 ext4_lblk_t iblock
, unsigned int maxblocks
,
935 struct buffer_head
*bh_result
,
939 ext4_lblk_t offsets
[4];
944 int blocks_to_boundary
= 0;
947 ext4_fsblk_t first_block
= 0;
949 J_ASSERT(!(EXT4_I(inode
)->i_flags
& EXT4_EXTENTS_FL
));
950 J_ASSERT(handle
!= NULL
|| (flags
& EXT4_GET_BLOCKS_CREATE
) == 0);
951 depth
= ext4_block_to_path(inode
, iblock
, offsets
,
952 &blocks_to_boundary
);
957 partial
= ext4_get_branch(inode
, depth
, offsets
, chain
, &err
);
959 /* Simplest case - block found, no allocation needed */
961 first_block
= le32_to_cpu(chain
[depth
- 1].key
);
962 clear_buffer_new(bh_result
);
965 while (count
< maxblocks
&& count
<= blocks_to_boundary
) {
968 blk
= le32_to_cpu(*(chain
[depth
-1].p
+ count
));
970 if (blk
== first_block
+ count
)
978 /* Next simple case - plain lookup or failed read of indirect block */
979 if ((flags
& EXT4_GET_BLOCKS_CREATE
) == 0 || err
== -EIO
)
983 * Okay, we need to do block allocation.
985 goal
= ext4_find_goal(inode
, iblock
, partial
);
987 /* the number of blocks need to allocate for [d,t]indirect blocks */
988 indirect_blks
= (chain
+ depth
) - partial
- 1;
991 * Next look up the indirect map to count the totoal number of
992 * direct blocks to allocate for this branch.
994 count
= ext4_blks_to_allocate(partial
, indirect_blks
,
995 maxblocks
, blocks_to_boundary
);
997 * Block out ext4_truncate while we alter the tree
999 err
= ext4_alloc_branch(handle
, inode
, iblock
, indirect_blks
,
1001 offsets
+ (partial
- chain
), partial
);
1004 * The ext4_splice_branch call will free and forget any buffers
1005 * on the new chain if there is a failure, but that risks using
1006 * up transaction credits, especially for bitmaps where the
1007 * credits cannot be returned. Can we handle this somehow? We
1008 * may need to return -EAGAIN upwards in the worst case. --sct
1011 err
= ext4_splice_branch(handle
, inode
, iblock
,
1012 partial
, indirect_blks
, count
);
1016 set_buffer_new(bh_result
);
1018 map_bh(bh_result
, inode
->i_sb
, le32_to_cpu(chain
[depth
-1].key
));
1019 if (count
> blocks_to_boundary
)
1020 set_buffer_boundary(bh_result
);
1022 /* Clean up and exit */
1023 partial
= chain
+ depth
- 1; /* the whole chain */
1025 while (partial
> chain
) {
1026 BUFFER_TRACE(partial
->bh
, "call brelse");
1027 brelse(partial
->bh
);
1030 BUFFER_TRACE(bh_result
, "returned");
1035 qsize_t
ext4_get_reserved_space(struct inode
*inode
)
1037 unsigned long long total
;
1039 spin_lock(&EXT4_I(inode
)->i_block_reservation_lock
);
1040 total
= EXT4_I(inode
)->i_reserved_data_blocks
+
1041 EXT4_I(inode
)->i_reserved_meta_blocks
;
1042 spin_unlock(&EXT4_I(inode
)->i_block_reservation_lock
);
1047 * Calculate the number of metadata blocks need to reserve
1048 * to allocate @blocks for non extent file based file
1050 static int ext4_indirect_calc_metadata_amount(struct inode
*inode
, int blocks
)
1052 int icap
= EXT4_ADDR_PER_BLOCK(inode
->i_sb
);
1053 int ind_blks
, dind_blks
, tind_blks
;
1055 /* number of new indirect blocks needed */
1056 ind_blks
= (blocks
+ icap
- 1) / icap
;
1058 dind_blks
= (ind_blks
+ icap
- 1) / icap
;
1062 return ind_blks
+ dind_blks
+ tind_blks
;
1066 * Calculate the number of metadata blocks need to reserve
1067 * to allocate given number of blocks
1069 static int ext4_calc_metadata_amount(struct inode
*inode
, int blocks
)
1074 if (EXT4_I(inode
)->i_flags
& EXT4_EXTENTS_FL
)
1075 return ext4_ext_calc_metadata_amount(inode
, blocks
);
1077 return ext4_indirect_calc_metadata_amount(inode
, blocks
);
1080 static void ext4_da_update_reserve_space(struct inode
*inode
, int used
)
1082 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
1083 int total
, mdb
, mdb_free
;
1085 spin_lock(&EXT4_I(inode
)->i_block_reservation_lock
);
1086 /* recalculate the number of metablocks still need to be reserved */
1087 total
= EXT4_I(inode
)->i_reserved_data_blocks
- used
;
1088 mdb
= ext4_calc_metadata_amount(inode
, total
);
1090 /* figure out how many metablocks to release */
1091 BUG_ON(mdb
> EXT4_I(inode
)->i_reserved_meta_blocks
);
1092 mdb_free
= EXT4_I(inode
)->i_reserved_meta_blocks
- mdb
;
1095 /* Account for allocated meta_blocks */
1096 mdb_free
-= EXT4_I(inode
)->i_allocated_meta_blocks
;
1098 /* update fs dirty blocks counter */
1099 percpu_counter_sub(&sbi
->s_dirtyblocks_counter
, mdb_free
);
1100 EXT4_I(inode
)->i_allocated_meta_blocks
= 0;
1101 EXT4_I(inode
)->i_reserved_meta_blocks
= mdb
;
1104 /* update per-inode reservations */
1105 BUG_ON(used
> EXT4_I(inode
)->i_reserved_data_blocks
);
1106 EXT4_I(inode
)->i_reserved_data_blocks
-= used
;
1107 spin_unlock(&EXT4_I(inode
)->i_block_reservation_lock
);
1110 * free those over-booking quota for metadata blocks
1113 vfs_dq_release_reservation_block(inode
, mdb_free
);
1116 * If we have done all the pending block allocations and if
1117 * there aren't any writers on the inode, we can discard the
1118 * inode's preallocations.
1120 if (!total
&& (atomic_read(&inode
->i_writecount
) == 0))
1121 ext4_discard_preallocations(inode
);
1124 static int check_block_validity(struct inode
*inode
, sector_t logical
,
1125 sector_t phys
, int len
)
1127 if (!ext4_data_block_valid(EXT4_SB(inode
->i_sb
), phys
, len
)) {
1128 ext4_error(inode
->i_sb
, "check_block_validity",
1129 "inode #%lu logical block %llu mapped to %llu "
1130 "(size %d)", inode
->i_ino
,
1131 (unsigned long long) logical
,
1132 (unsigned long long) phys
, len
);
1140 * The ext4_get_blocks() function tries to look up the requested blocks,
1141 * and returns if the blocks are already mapped.
1143 * Otherwise it takes the write lock of the i_data_sem and allocate blocks
1144 * and store the allocated blocks in the result buffer head and mark it
1147 * If file type is extents based, it will call ext4_ext_get_blocks(),
1148 * Otherwise, call with ext4_ind_get_blocks() to handle indirect mapping
1151 * On success, it returns the number of blocks being mapped or allocate.
1152 * if create==0 and the blocks are pre-allocated and uninitialized block,
1153 * the result buffer head is unmapped. If the create ==1, it will make sure
1154 * the buffer head is mapped.
1156 * It returns 0 if plain look up failed (blocks have not been allocated), in
1157 * that casem, buffer head is unmapped
1159 * It returns the error in case of allocation failure.
1161 int ext4_get_blocks(handle_t
*handle
, struct inode
*inode
, sector_t block
,
1162 unsigned int max_blocks
, struct buffer_head
*bh
,
1167 clear_buffer_mapped(bh
);
1168 clear_buffer_unwritten(bh
);
1171 * Try to see if we can get the block without requesting a new
1172 * file system block.
1174 down_read((&EXT4_I(inode
)->i_data_sem
));
1175 if (EXT4_I(inode
)->i_flags
& EXT4_EXTENTS_FL
) {
1176 retval
= ext4_ext_get_blocks(handle
, inode
, block
, max_blocks
,
1179 retval
= ext4_ind_get_blocks(handle
, inode
, block
, max_blocks
,
1182 up_read((&EXT4_I(inode
)->i_data_sem
));
1184 if (retval
> 0 && buffer_mapped(bh
)) {
1185 int ret
= check_block_validity(inode
, block
,
1186 bh
->b_blocknr
, retval
);
1191 /* If it is only a block(s) look up */
1192 if ((flags
& EXT4_GET_BLOCKS_CREATE
) == 0)
1196 * Returns if the blocks have already allocated
1198 * Note that if blocks have been preallocated
1199 * ext4_ext_get_block() returns th create = 0
1200 * with buffer head unmapped.
1202 if (retval
> 0 && buffer_mapped(bh
))
1206 * When we call get_blocks without the create flag, the
1207 * BH_Unwritten flag could have gotten set if the blocks
1208 * requested were part of a uninitialized extent. We need to
1209 * clear this flag now that we are committed to convert all or
1210 * part of the uninitialized extent to be an initialized
1211 * extent. This is because we need to avoid the combination
1212 * of BH_Unwritten and BH_Mapped flags being simultaneously
1213 * set on the buffer_head.
1215 clear_buffer_unwritten(bh
);
1218 * New blocks allocate and/or writing to uninitialized extent
1219 * will possibly result in updating i_data, so we take
1220 * the write lock of i_data_sem, and call get_blocks()
1221 * with create == 1 flag.
1223 down_write((&EXT4_I(inode
)->i_data_sem
));
1226 * if the caller is from delayed allocation writeout path
1227 * we have already reserved fs blocks for allocation
1228 * let the underlying get_block() function know to
1229 * avoid double accounting
1231 if (flags
& EXT4_GET_BLOCKS_DELALLOC_RESERVE
)
1232 EXT4_I(inode
)->i_delalloc_reserved_flag
= 1;
1234 * We need to check for EXT4 here because migrate
1235 * could have changed the inode type in between
1237 if (EXT4_I(inode
)->i_flags
& EXT4_EXTENTS_FL
) {
1238 retval
= ext4_ext_get_blocks(handle
, inode
, block
, max_blocks
,
1241 retval
= ext4_ind_get_blocks(handle
, inode
, block
,
1242 max_blocks
, bh
, flags
);
1244 if (retval
> 0 && buffer_new(bh
)) {
1246 * We allocated new blocks which will result in
1247 * i_data's format changing. Force the migrate
1248 * to fail by clearing migrate flags
1250 EXT4_I(inode
)->i_flags
= EXT4_I(inode
)->i_flags
&
1255 if (flags
& EXT4_GET_BLOCKS_DELALLOC_RESERVE
)
1256 EXT4_I(inode
)->i_delalloc_reserved_flag
= 0;
1259 * Update reserved blocks/metadata blocks after successful
1260 * block allocation which had been deferred till now.
1262 if ((retval
> 0) && (flags
& EXT4_GET_BLOCKS_UPDATE_RESERVE_SPACE
))
1263 ext4_da_update_reserve_space(inode
, retval
);
1265 up_write((&EXT4_I(inode
)->i_data_sem
));
1266 if (retval
> 0 && buffer_mapped(bh
)) {
1267 int ret
= check_block_validity(inode
, block
,
1268 bh
->b_blocknr
, retval
);
1275 /* Maximum number of blocks we map for direct IO at once. */
1276 #define DIO_MAX_BLOCKS 4096
1278 int ext4_get_block(struct inode
*inode
, sector_t iblock
,
1279 struct buffer_head
*bh_result
, int create
)
1281 handle_t
*handle
= ext4_journal_current_handle();
1282 int ret
= 0, started
= 0;
1283 unsigned max_blocks
= bh_result
->b_size
>> inode
->i_blkbits
;
1286 if (create
&& !handle
) {
1287 /* Direct IO write... */
1288 if (max_blocks
> DIO_MAX_BLOCKS
)
1289 max_blocks
= DIO_MAX_BLOCKS
;
1290 dio_credits
= ext4_chunk_trans_blocks(inode
, max_blocks
);
1291 handle
= ext4_journal_start(inode
, dio_credits
);
1292 if (IS_ERR(handle
)) {
1293 ret
= PTR_ERR(handle
);
1299 ret
= ext4_get_blocks(handle
, inode
, iblock
, max_blocks
, bh_result
,
1300 create
? EXT4_GET_BLOCKS_CREATE
: 0);
1302 bh_result
->b_size
= (ret
<< inode
->i_blkbits
);
1306 ext4_journal_stop(handle
);
1312 * `handle' can be NULL if create is zero
1314 struct buffer_head
*ext4_getblk(handle_t
*handle
, struct inode
*inode
,
1315 ext4_lblk_t block
, int create
, int *errp
)
1317 struct buffer_head dummy
;
1321 J_ASSERT(handle
!= NULL
|| create
== 0);
1324 dummy
.b_blocknr
= -1000;
1325 buffer_trace_init(&dummy
.b_history
);
1327 flags
|= EXT4_GET_BLOCKS_CREATE
;
1328 err
= ext4_get_blocks(handle
, inode
, block
, 1, &dummy
, flags
);
1330 * ext4_get_blocks() returns number of blocks mapped. 0 in
1339 if (!err
&& buffer_mapped(&dummy
)) {
1340 struct buffer_head
*bh
;
1341 bh
= sb_getblk(inode
->i_sb
, dummy
.b_blocknr
);
1346 if (buffer_new(&dummy
)) {
1347 J_ASSERT(create
!= 0);
1348 J_ASSERT(handle
!= NULL
);
1351 * Now that we do not always journal data, we should
1352 * keep in mind whether this should always journal the
1353 * new buffer as metadata. For now, regular file
1354 * writes use ext4_get_block instead, so it's not a
1358 BUFFER_TRACE(bh
, "call get_create_access");
1359 fatal
= ext4_journal_get_create_access(handle
, bh
);
1360 if (!fatal
&& !buffer_uptodate(bh
)) {
1361 memset(bh
->b_data
, 0, inode
->i_sb
->s_blocksize
);
1362 set_buffer_uptodate(bh
);
1365 BUFFER_TRACE(bh
, "call ext4_handle_dirty_metadata");
1366 err
= ext4_handle_dirty_metadata(handle
, inode
, bh
);
1370 BUFFER_TRACE(bh
, "not a new buffer");
1383 struct buffer_head
*ext4_bread(handle_t
*handle
, struct inode
*inode
,
1384 ext4_lblk_t block
, int create
, int *err
)
1386 struct buffer_head
*bh
;
1388 bh
= ext4_getblk(handle
, inode
, block
, create
, err
);
1391 if (buffer_uptodate(bh
))
1393 ll_rw_block(READ_META
, 1, &bh
);
1395 if (buffer_uptodate(bh
))
1402 static int walk_page_buffers(handle_t
*handle
,
1403 struct buffer_head
*head
,
1407 int (*fn
)(handle_t
*handle
,
1408 struct buffer_head
*bh
))
1410 struct buffer_head
*bh
;
1411 unsigned block_start
, block_end
;
1412 unsigned blocksize
= head
->b_size
;
1414 struct buffer_head
*next
;
1416 for (bh
= head
, block_start
= 0;
1417 ret
== 0 && (bh
!= head
|| !block_start
);
1418 block_start
= block_end
, bh
= next
) {
1419 next
= bh
->b_this_page
;
1420 block_end
= block_start
+ blocksize
;
1421 if (block_end
<= from
|| block_start
>= to
) {
1422 if (partial
&& !buffer_uptodate(bh
))
1426 err
= (*fn
)(handle
, bh
);
1434 * To preserve ordering, it is essential that the hole instantiation and
1435 * the data write be encapsulated in a single transaction. We cannot
1436 * close off a transaction and start a new one between the ext4_get_block()
1437 * and the commit_write(). So doing the jbd2_journal_start at the start of
1438 * prepare_write() is the right place.
1440 * Also, this function can nest inside ext4_writepage() ->
1441 * block_write_full_page(). In that case, we *know* that ext4_writepage()
1442 * has generated enough buffer credits to do the whole page. So we won't
1443 * block on the journal in that case, which is good, because the caller may
1446 * By accident, ext4 can be reentered when a transaction is open via
1447 * quota file writes. If we were to commit the transaction while thus
1448 * reentered, there can be a deadlock - we would be holding a quota
1449 * lock, and the commit would never complete if another thread had a
1450 * transaction open and was blocking on the quota lock - a ranking
1453 * So what we do is to rely on the fact that jbd2_journal_stop/journal_start
1454 * will _not_ run commit under these circumstances because handle->h_ref
1455 * is elevated. We'll still have enough credits for the tiny quotafile
1458 static int do_journal_get_write_access(handle_t
*handle
,
1459 struct buffer_head
*bh
)
1461 if (!buffer_mapped(bh
) || buffer_freed(bh
))
1463 return ext4_journal_get_write_access(handle
, bh
);
1466 static int ext4_write_begin(struct file
*file
, struct address_space
*mapping
,
1467 loff_t pos
, unsigned len
, unsigned flags
,
1468 struct page
**pagep
, void **fsdata
)
1470 struct inode
*inode
= mapping
->host
;
1471 int ret
, needed_blocks
;
1478 trace_ext4_write_begin(inode
, pos
, len
, flags
);
1480 * Reserve one block more for addition to orphan list in case
1481 * we allocate blocks but write fails for some reason
1483 needed_blocks
= ext4_writepage_trans_blocks(inode
) + 1;
1484 index
= pos
>> PAGE_CACHE_SHIFT
;
1485 from
= pos
& (PAGE_CACHE_SIZE
- 1);
1489 handle
= ext4_journal_start(inode
, needed_blocks
);
1490 if (IS_ERR(handle
)) {
1491 ret
= PTR_ERR(handle
);
1495 /* We cannot recurse into the filesystem as the transaction is already
1497 flags
|= AOP_FLAG_NOFS
;
1499 page
= grab_cache_page_write_begin(mapping
, index
, flags
);
1501 ext4_journal_stop(handle
);
1507 ret
= block_write_begin(file
, mapping
, pos
, len
, flags
, pagep
, fsdata
,
1510 if (!ret
&& ext4_should_journal_data(inode
)) {
1511 ret
= walk_page_buffers(handle
, page_buffers(page
),
1512 from
, to
, NULL
, do_journal_get_write_access
);
1517 page_cache_release(page
);
1519 * block_write_begin may have instantiated a few blocks
1520 * outside i_size. Trim these off again. Don't need
1521 * i_size_read because we hold i_mutex.
1523 * Add inode to orphan list in case we crash before
1526 if (pos
+ len
> inode
->i_size
&& ext4_can_truncate(inode
))
1527 ext4_orphan_add(handle
, inode
);
1529 ext4_journal_stop(handle
);
1530 if (pos
+ len
> inode
->i_size
) {
1531 ext4_truncate(inode
);
1533 * If truncate failed early the inode might
1534 * still be on the orphan list; we need to
1535 * make sure the inode is removed from the
1536 * orphan list in that case.
1539 ext4_orphan_del(NULL
, inode
);
1543 if (ret
== -ENOSPC
&& ext4_should_retry_alloc(inode
->i_sb
, &retries
))
1549 /* For write_end() in data=journal mode */
1550 static int write_end_fn(handle_t
*handle
, struct buffer_head
*bh
)
1552 if (!buffer_mapped(bh
) || buffer_freed(bh
))
1554 set_buffer_uptodate(bh
);
1555 return ext4_handle_dirty_metadata(handle
, NULL
, bh
);
1558 static int ext4_generic_write_end(struct file
*file
,
1559 struct address_space
*mapping
,
1560 loff_t pos
, unsigned len
, unsigned copied
,
1561 struct page
*page
, void *fsdata
)
1563 int i_size_changed
= 0;
1564 struct inode
*inode
= mapping
->host
;
1565 handle_t
*handle
= ext4_journal_current_handle();
1567 copied
= block_write_end(file
, mapping
, pos
, len
, copied
, page
, fsdata
);
1570 * No need to use i_size_read() here, the i_size
1571 * cannot change under us because we hold i_mutex.
1573 * But it's important to update i_size while still holding page lock:
1574 * page writeout could otherwise come in and zero beyond i_size.
1576 if (pos
+ copied
> inode
->i_size
) {
1577 i_size_write(inode
, pos
+ copied
);
1581 if (pos
+ copied
> EXT4_I(inode
)->i_disksize
) {
1582 /* We need to mark inode dirty even if
1583 * new_i_size is less that inode->i_size
1584 * bu greater than i_disksize.(hint delalloc)
1586 ext4_update_i_disksize(inode
, (pos
+ copied
));
1590 page_cache_release(page
);
1593 * Don't mark the inode dirty under page lock. First, it unnecessarily
1594 * makes the holding time of page lock longer. Second, it forces lock
1595 * ordering of page lock and transaction start for journaling
1599 ext4_mark_inode_dirty(handle
, inode
);
1605 * We need to pick up the new inode size which generic_commit_write gave us
1606 * `file' can be NULL - eg, when called from page_symlink().
1608 * ext4 never places buffers on inode->i_mapping->private_list. metadata
1609 * buffers are managed internally.
1611 static int ext4_ordered_write_end(struct file
*file
,
1612 struct address_space
*mapping
,
1613 loff_t pos
, unsigned len
, unsigned copied
,
1614 struct page
*page
, void *fsdata
)
1616 handle_t
*handle
= ext4_journal_current_handle();
1617 struct inode
*inode
= mapping
->host
;
1620 trace_ext4_ordered_write_end(inode
, pos
, len
, copied
);
1621 ret
= ext4_jbd2_file_inode(handle
, inode
);
1624 ret2
= ext4_generic_write_end(file
, mapping
, pos
, len
, copied
,
1627 if (pos
+ len
> inode
->i_size
&& ext4_can_truncate(inode
))
1628 /* if we have allocated more blocks and copied
1629 * less. We will have blocks allocated outside
1630 * inode->i_size. So truncate them
1632 ext4_orphan_add(handle
, inode
);
1636 ret2
= ext4_journal_stop(handle
);
1640 if (pos
+ len
> inode
->i_size
) {
1641 ext4_truncate(inode
);
1643 * If truncate failed early the inode might still be
1644 * on the orphan list; we need to make sure the inode
1645 * is removed from the orphan list in that case.
1648 ext4_orphan_del(NULL
, inode
);
1652 return ret
? ret
: copied
;
1655 static int ext4_writeback_write_end(struct file
*file
,
1656 struct address_space
*mapping
,
1657 loff_t pos
, unsigned len
, unsigned copied
,
1658 struct page
*page
, void *fsdata
)
1660 handle_t
*handle
= ext4_journal_current_handle();
1661 struct inode
*inode
= mapping
->host
;
1664 trace_ext4_writeback_write_end(inode
, pos
, len
, copied
);
1665 ret2
= ext4_generic_write_end(file
, mapping
, pos
, len
, copied
,
1668 if (pos
+ len
> inode
->i_size
&& ext4_can_truncate(inode
))
1669 /* if we have allocated more blocks and copied
1670 * less. We will have blocks allocated outside
1671 * inode->i_size. So truncate them
1673 ext4_orphan_add(handle
, inode
);
1678 ret2
= ext4_journal_stop(handle
);
1682 if (pos
+ len
> inode
->i_size
) {
1683 ext4_truncate(inode
);
1685 * If truncate failed early the inode might still be
1686 * on the orphan list; we need to make sure the inode
1687 * is removed from the orphan list in that case.
1690 ext4_orphan_del(NULL
, inode
);
1693 return ret
? ret
: copied
;
1696 static int ext4_journalled_write_end(struct file
*file
,
1697 struct address_space
*mapping
,
1698 loff_t pos
, unsigned len
, unsigned copied
,
1699 struct page
*page
, void *fsdata
)
1701 handle_t
*handle
= ext4_journal_current_handle();
1702 struct inode
*inode
= mapping
->host
;
1708 trace_ext4_journalled_write_end(inode
, pos
, len
, copied
);
1709 from
= pos
& (PAGE_CACHE_SIZE
- 1);
1713 if (!PageUptodate(page
))
1715 page_zero_new_buffers(page
, from
+copied
, to
);
1718 ret
= walk_page_buffers(handle
, page_buffers(page
), from
,
1719 to
, &partial
, write_end_fn
);
1721 SetPageUptodate(page
);
1722 new_i_size
= pos
+ copied
;
1723 if (new_i_size
> inode
->i_size
)
1724 i_size_write(inode
, pos
+copied
);
1725 EXT4_I(inode
)->i_state
|= EXT4_STATE_JDATA
;
1726 if (new_i_size
> EXT4_I(inode
)->i_disksize
) {
1727 ext4_update_i_disksize(inode
, new_i_size
);
1728 ret2
= ext4_mark_inode_dirty(handle
, inode
);
1734 page_cache_release(page
);
1735 if (pos
+ len
> inode
->i_size
&& ext4_can_truncate(inode
))
1736 /* if we have allocated more blocks and copied
1737 * less. We will have blocks allocated outside
1738 * inode->i_size. So truncate them
1740 ext4_orphan_add(handle
, inode
);
1742 ret2
= ext4_journal_stop(handle
);
1745 if (pos
+ len
> inode
->i_size
) {
1746 ext4_truncate(inode
);
1748 * If truncate failed early the inode might still be
1749 * on the orphan list; we need to make sure the inode
1750 * is removed from the orphan list in that case.
1753 ext4_orphan_del(NULL
, inode
);
1756 return ret
? ret
: copied
;
1759 static int ext4_da_reserve_space(struct inode
*inode
, int nrblocks
)
1762 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
1763 unsigned long md_needed
, mdblocks
, total
= 0;
1766 * recalculate the amount of metadata blocks to reserve
1767 * in order to allocate nrblocks
1768 * worse case is one extent per block
1771 spin_lock(&EXT4_I(inode
)->i_block_reservation_lock
);
1772 total
= EXT4_I(inode
)->i_reserved_data_blocks
+ nrblocks
;
1773 mdblocks
= ext4_calc_metadata_amount(inode
, total
);
1774 BUG_ON(mdblocks
< EXT4_I(inode
)->i_reserved_meta_blocks
);
1776 md_needed
= mdblocks
- EXT4_I(inode
)->i_reserved_meta_blocks
;
1777 total
= md_needed
+ nrblocks
;
1780 * Make quota reservation here to prevent quota overflow
1781 * later. Real quota accounting is done at pages writeout
1784 if (vfs_dq_reserve_block(inode
, total
)) {
1785 spin_unlock(&EXT4_I(inode
)->i_block_reservation_lock
);
1789 if (ext4_claim_free_blocks(sbi
, total
)) {
1790 spin_unlock(&EXT4_I(inode
)->i_block_reservation_lock
);
1791 if (ext4_should_retry_alloc(inode
->i_sb
, &retries
)) {
1795 vfs_dq_release_reservation_block(inode
, total
);
1798 EXT4_I(inode
)->i_reserved_data_blocks
+= nrblocks
;
1799 EXT4_I(inode
)->i_reserved_meta_blocks
= mdblocks
;
1801 spin_unlock(&EXT4_I(inode
)->i_block_reservation_lock
);
1802 return 0; /* success */
1805 static void ext4_da_release_space(struct inode
*inode
, int to_free
)
1807 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
1808 int total
, mdb
, mdb_free
, release
;
1811 return; /* Nothing to release, exit */
1813 spin_lock(&EXT4_I(inode
)->i_block_reservation_lock
);
1815 if (!EXT4_I(inode
)->i_reserved_data_blocks
) {
1817 * if there is no reserved blocks, but we try to free some
1818 * then the counter is messed up somewhere.
1819 * but since this function is called from invalidate
1820 * page, it's harmless to return without any action
1822 printk(KERN_INFO
"ext4 delalloc try to release %d reserved "
1823 "blocks for inode %lu, but there is no reserved "
1824 "data blocks\n", to_free
, inode
->i_ino
);
1825 spin_unlock(&EXT4_I(inode
)->i_block_reservation_lock
);
1829 /* recalculate the number of metablocks still need to be reserved */
1830 total
= EXT4_I(inode
)->i_reserved_data_blocks
- to_free
;
1831 mdb
= ext4_calc_metadata_amount(inode
, total
);
1833 /* figure out how many metablocks to release */
1834 BUG_ON(mdb
> EXT4_I(inode
)->i_reserved_meta_blocks
);
1835 mdb_free
= EXT4_I(inode
)->i_reserved_meta_blocks
- mdb
;
1837 release
= to_free
+ mdb_free
;
1839 /* update fs dirty blocks counter for truncate case */
1840 percpu_counter_sub(&sbi
->s_dirtyblocks_counter
, release
);
1842 /* update per-inode reservations */
1843 BUG_ON(to_free
> EXT4_I(inode
)->i_reserved_data_blocks
);
1844 EXT4_I(inode
)->i_reserved_data_blocks
-= to_free
;
1846 BUG_ON(mdb
> EXT4_I(inode
)->i_reserved_meta_blocks
);
1847 EXT4_I(inode
)->i_reserved_meta_blocks
= mdb
;
1848 spin_unlock(&EXT4_I(inode
)->i_block_reservation_lock
);
1850 vfs_dq_release_reservation_block(inode
, release
);
1853 static void ext4_da_page_release_reservation(struct page
*page
,
1854 unsigned long offset
)
1857 struct buffer_head
*head
, *bh
;
1858 unsigned int curr_off
= 0;
1860 head
= page_buffers(page
);
1863 unsigned int next_off
= curr_off
+ bh
->b_size
;
1865 if ((offset
<= curr_off
) && (buffer_delay(bh
))) {
1867 clear_buffer_delay(bh
);
1869 curr_off
= next_off
;
1870 } while ((bh
= bh
->b_this_page
) != head
);
1871 ext4_da_release_space(page
->mapping
->host
, to_release
);
1875 * Delayed allocation stuff
1878 struct mpage_da_data
{
1879 struct inode
*inode
;
1880 sector_t b_blocknr
; /* start block number of extent */
1881 size_t b_size
; /* size of extent */
1882 unsigned long b_state
; /* state of the extent */
1883 unsigned long first_page
, next_page
; /* extent of pages */
1884 struct writeback_control
*wbc
;
1891 * mpage_da_submit_io - walks through extent of pages and try to write
1892 * them with writepage() call back
1894 * @mpd->inode: inode
1895 * @mpd->first_page: first page of the extent
1896 * @mpd->next_page: page after the last page of the extent
1898 * By the time mpage_da_submit_io() is called we expect all blocks
1899 * to be allocated. this may be wrong if allocation failed.
1901 * As pages are already locked by write_cache_pages(), we can't use it
1903 static int mpage_da_submit_io(struct mpage_da_data
*mpd
)
1906 struct pagevec pvec
;
1907 unsigned long index
, end
;
1908 int ret
= 0, err
, nr_pages
, i
;
1909 struct inode
*inode
= mpd
->inode
;
1910 struct address_space
*mapping
= inode
->i_mapping
;
1912 BUG_ON(mpd
->next_page
<= mpd
->first_page
);
1914 * We need to start from the first_page to the next_page - 1
1915 * to make sure we also write the mapped dirty buffer_heads.
1916 * If we look at mpd->b_blocknr we would only be looking
1917 * at the currently mapped buffer_heads.
1919 index
= mpd
->first_page
;
1920 end
= mpd
->next_page
- 1;
1922 pagevec_init(&pvec
, 0);
1923 while (index
<= end
) {
1924 nr_pages
= pagevec_lookup(&pvec
, mapping
, index
, PAGEVEC_SIZE
);
1927 for (i
= 0; i
< nr_pages
; i
++) {
1928 struct page
*page
= pvec
.pages
[i
];
1930 index
= page
->index
;
1935 BUG_ON(!PageLocked(page
));
1936 BUG_ON(PageWriteback(page
));
1938 pages_skipped
= mpd
->wbc
->pages_skipped
;
1939 err
= mapping
->a_ops
->writepage(page
, mpd
->wbc
);
1940 if (!err
&& (pages_skipped
== mpd
->wbc
->pages_skipped
))
1942 * have successfully written the page
1943 * without skipping the same
1945 mpd
->pages_written
++;
1947 * In error case, we have to continue because
1948 * remaining pages are still locked
1949 * XXX: unlock and re-dirty them?
1954 pagevec_release(&pvec
);
1960 * mpage_put_bnr_to_bhs - walk blocks and assign them actual numbers
1962 * @mpd->inode - inode to walk through
1963 * @exbh->b_blocknr - first block on a disk
1964 * @exbh->b_size - amount of space in bytes
1965 * @logical - first logical block to start assignment with
1967 * the function goes through all passed space and put actual disk
1968 * block numbers into buffer heads, dropping BH_Delay and BH_Unwritten
1970 static void mpage_put_bnr_to_bhs(struct mpage_da_data
*mpd
, sector_t logical
,
1971 struct buffer_head
*exbh
)
1973 struct inode
*inode
= mpd
->inode
;
1974 struct address_space
*mapping
= inode
->i_mapping
;
1975 int blocks
= exbh
->b_size
>> inode
->i_blkbits
;
1976 sector_t pblock
= exbh
->b_blocknr
, cur_logical
;
1977 struct buffer_head
*head
, *bh
;
1979 struct pagevec pvec
;
1982 index
= logical
>> (PAGE_CACHE_SHIFT
- inode
->i_blkbits
);
1983 end
= (logical
+ blocks
- 1) >> (PAGE_CACHE_SHIFT
- inode
->i_blkbits
);
1984 cur_logical
= index
<< (PAGE_CACHE_SHIFT
- inode
->i_blkbits
);
1986 pagevec_init(&pvec
, 0);
1988 while (index
<= end
) {
1989 /* XXX: optimize tail */
1990 nr_pages
= pagevec_lookup(&pvec
, mapping
, index
, PAGEVEC_SIZE
);
1993 for (i
= 0; i
< nr_pages
; i
++) {
1994 struct page
*page
= pvec
.pages
[i
];
1996 index
= page
->index
;
2001 BUG_ON(!PageLocked(page
));
2002 BUG_ON(PageWriteback(page
));
2003 BUG_ON(!page_has_buffers(page
));
2005 bh
= page_buffers(page
);
2008 /* skip blocks out of the range */
2010 if (cur_logical
>= logical
)
2013 } while ((bh
= bh
->b_this_page
) != head
);
2016 if (cur_logical
>= logical
+ blocks
)
2019 if (buffer_delay(bh
) ||
2020 buffer_unwritten(bh
)) {
2022 BUG_ON(bh
->b_bdev
!= inode
->i_sb
->s_bdev
);
2024 if (buffer_delay(bh
)) {
2025 clear_buffer_delay(bh
);
2026 bh
->b_blocknr
= pblock
;
2029 * unwritten already should have
2030 * blocknr assigned. Verify that
2032 clear_buffer_unwritten(bh
);
2033 BUG_ON(bh
->b_blocknr
!= pblock
);
2036 } else if (buffer_mapped(bh
))
2037 BUG_ON(bh
->b_blocknr
!= pblock
);
2041 } while ((bh
= bh
->b_this_page
) != head
);
2043 pagevec_release(&pvec
);
2049 * __unmap_underlying_blocks - just a helper function to unmap
2050 * set of blocks described by @bh
2052 static inline void __unmap_underlying_blocks(struct inode
*inode
,
2053 struct buffer_head
*bh
)
2055 struct block_device
*bdev
= inode
->i_sb
->s_bdev
;
2058 blocks
= bh
->b_size
>> inode
->i_blkbits
;
2059 for (i
= 0; i
< blocks
; i
++)
2060 unmap_underlying_metadata(bdev
, bh
->b_blocknr
+ i
);
2063 static void ext4_da_block_invalidatepages(struct mpage_da_data
*mpd
,
2064 sector_t logical
, long blk_cnt
)
2068 struct pagevec pvec
;
2069 struct inode
*inode
= mpd
->inode
;
2070 struct address_space
*mapping
= inode
->i_mapping
;
2072 index
= logical
>> (PAGE_CACHE_SHIFT
- inode
->i_blkbits
);
2073 end
= (logical
+ blk_cnt
- 1) >>
2074 (PAGE_CACHE_SHIFT
- inode
->i_blkbits
);
2075 while (index
<= end
) {
2076 nr_pages
= pagevec_lookup(&pvec
, mapping
, index
, PAGEVEC_SIZE
);
2079 for (i
= 0; i
< nr_pages
; i
++) {
2080 struct page
*page
= pvec
.pages
[i
];
2081 index
= page
->index
;
2086 BUG_ON(!PageLocked(page
));
2087 BUG_ON(PageWriteback(page
));
2088 block_invalidatepage(page
, 0);
2089 ClearPageUptodate(page
);
2096 static void ext4_print_free_blocks(struct inode
*inode
)
2098 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
2099 printk(KERN_EMERG
"Total free blocks count %lld\n",
2100 ext4_count_free_blocks(inode
->i_sb
));
2101 printk(KERN_EMERG
"Free/Dirty block details\n");
2102 printk(KERN_EMERG
"free_blocks=%lld\n",
2103 (long long)percpu_counter_sum(&sbi
->s_freeblocks_counter
));
2104 printk(KERN_EMERG
"dirty_blocks=%lld\n",
2105 (long long)percpu_counter_sum(&sbi
->s_dirtyblocks_counter
));
2106 printk(KERN_EMERG
"Block reservation details\n");
2107 printk(KERN_EMERG
"i_reserved_data_blocks=%u\n",
2108 EXT4_I(inode
)->i_reserved_data_blocks
);
2109 printk(KERN_EMERG
"i_reserved_meta_blocks=%u\n",
2110 EXT4_I(inode
)->i_reserved_meta_blocks
);
2115 * mpage_da_map_blocks - go through given space
2117 * @mpd - bh describing space
2119 * The function skips space we know is already mapped to disk blocks.
2122 static int mpage_da_map_blocks(struct mpage_da_data
*mpd
)
2124 int err
, blks
, get_blocks_flags
;
2125 struct buffer_head
new;
2126 sector_t next
= mpd
->b_blocknr
;
2127 unsigned max_blocks
= mpd
->b_size
>> mpd
->inode
->i_blkbits
;
2128 loff_t disksize
= EXT4_I(mpd
->inode
)->i_disksize
;
2129 handle_t
*handle
= NULL
;
2132 * We consider only non-mapped and non-allocated blocks
2134 if ((mpd
->b_state
& (1 << BH_Mapped
)) &&
2135 !(mpd
->b_state
& (1 << BH_Delay
)) &&
2136 !(mpd
->b_state
& (1 << BH_Unwritten
)))
2140 * If we didn't accumulate anything to write simply return
2145 handle
= ext4_journal_current_handle();
2149 * Call ext4_get_blocks() to allocate any delayed allocation
2150 * blocks, or to convert an uninitialized extent to be
2151 * initialized (in the case where we have written into
2152 * one or more preallocated blocks).
2154 * We pass in the magic EXT4_GET_BLOCKS_DELALLOC_RESERVE to
2155 * indicate that we are on the delayed allocation path. This
2156 * affects functions in many different parts of the allocation
2157 * call path. This flag exists primarily because we don't
2158 * want to change *many* call functions, so ext4_get_blocks()
2159 * will set the magic i_delalloc_reserved_flag once the
2160 * inode's allocation semaphore is taken.
2162 * If the blocks in questions were delalloc blocks, set
2163 * EXT4_GET_BLOCKS_DELALLOC_RESERVE so the delalloc accounting
2164 * variables are updated after the blocks have been allocated.
2167 get_blocks_flags
= (EXT4_GET_BLOCKS_CREATE
|
2168 EXT4_GET_BLOCKS_DELALLOC_RESERVE
);
2169 if (mpd
->b_state
& (1 << BH_Delay
))
2170 get_blocks_flags
|= EXT4_GET_BLOCKS_UPDATE_RESERVE_SPACE
;
2171 blks
= ext4_get_blocks(handle
, mpd
->inode
, next
, max_blocks
,
2172 &new, get_blocks_flags
);
2176 * If get block returns with error we simply
2177 * return. Later writepage will redirty the page and
2178 * writepages will find the dirty page again
2183 if (err
== -ENOSPC
&&
2184 ext4_count_free_blocks(mpd
->inode
->i_sb
)) {
2190 * get block failure will cause us to loop in
2191 * writepages, because a_ops->writepage won't be able
2192 * to make progress. The page will be redirtied by
2193 * writepage and writepages will again try to write
2196 printk(KERN_EMERG
"%s block allocation failed for inode %lu "
2197 "at logical offset %llu with max blocks "
2198 "%zd with error %d\n",
2199 __func__
, mpd
->inode
->i_ino
,
2200 (unsigned long long)next
,
2201 mpd
->b_size
>> mpd
->inode
->i_blkbits
, err
);
2202 printk(KERN_EMERG
"This should not happen.!! "
2203 "Data will be lost\n");
2204 if (err
== -ENOSPC
) {
2205 ext4_print_free_blocks(mpd
->inode
);
2207 /* invalidate all the pages */
2208 ext4_da_block_invalidatepages(mpd
, next
,
2209 mpd
->b_size
>> mpd
->inode
->i_blkbits
);
2214 new.b_size
= (blks
<< mpd
->inode
->i_blkbits
);
2216 if (buffer_new(&new))
2217 __unmap_underlying_blocks(mpd
->inode
, &new);
2220 * If blocks are delayed marked, we need to
2221 * put actual blocknr and drop delayed bit
2223 if ((mpd
->b_state
& (1 << BH_Delay
)) ||
2224 (mpd
->b_state
& (1 << BH_Unwritten
)))
2225 mpage_put_bnr_to_bhs(mpd
, next
, &new);
2227 if (ext4_should_order_data(mpd
->inode
)) {
2228 err
= ext4_jbd2_file_inode(handle
, mpd
->inode
);
2234 * Update on-disk size along with block allocation.
2236 disksize
= ((loff_t
) next
+ blks
) << mpd
->inode
->i_blkbits
;
2237 if (disksize
> i_size_read(mpd
->inode
))
2238 disksize
= i_size_read(mpd
->inode
);
2239 if (disksize
> EXT4_I(mpd
->inode
)->i_disksize
) {
2240 ext4_update_i_disksize(mpd
->inode
, disksize
);
2241 return ext4_mark_inode_dirty(handle
, mpd
->inode
);
2247 #define BH_FLAGS ((1 << BH_Uptodate) | (1 << BH_Mapped) | \
2248 (1 << BH_Delay) | (1 << BH_Unwritten))
2251 * mpage_add_bh_to_extent - try to add one more block to extent of blocks
2253 * @mpd->lbh - extent of blocks
2254 * @logical - logical number of the block in the file
2255 * @bh - bh of the block (used to access block's state)
2257 * the function is used to collect contig. blocks in same state
2259 static void mpage_add_bh_to_extent(struct mpage_da_data
*mpd
,
2260 sector_t logical
, size_t b_size
,
2261 unsigned long b_state
)
2264 int nrblocks
= mpd
->b_size
>> mpd
->inode
->i_blkbits
;
2266 /* check if thereserved journal credits might overflow */
2267 if (!(EXT4_I(mpd
->inode
)->i_flags
& EXT4_EXTENTS_FL
)) {
2268 if (nrblocks
>= EXT4_MAX_TRANS_DATA
) {
2270 * With non-extent format we are limited by the journal
2271 * credit available. Total credit needed to insert
2272 * nrblocks contiguous blocks is dependent on the
2273 * nrblocks. So limit nrblocks.
2276 } else if ((nrblocks
+ (b_size
>> mpd
->inode
->i_blkbits
)) >
2277 EXT4_MAX_TRANS_DATA
) {
2279 * Adding the new buffer_head would make it cross the
2280 * allowed limit for which we have journal credit
2281 * reserved. So limit the new bh->b_size
2283 b_size
= (EXT4_MAX_TRANS_DATA
- nrblocks
) <<
2284 mpd
->inode
->i_blkbits
;
2285 /* we will do mpage_da_submit_io in the next loop */
2289 * First block in the extent
2291 if (mpd
->b_size
== 0) {
2292 mpd
->b_blocknr
= logical
;
2293 mpd
->b_size
= b_size
;
2294 mpd
->b_state
= b_state
& BH_FLAGS
;
2298 next
= mpd
->b_blocknr
+ nrblocks
;
2300 * Can we merge the block to our big extent?
2302 if (logical
== next
&& (b_state
& BH_FLAGS
) == mpd
->b_state
) {
2303 mpd
->b_size
+= b_size
;
2309 * We couldn't merge the block to our extent, so we
2310 * need to flush current extent and start new one
2312 if (mpage_da_map_blocks(mpd
) == 0)
2313 mpage_da_submit_io(mpd
);
2318 static int ext4_bh_delay_or_unwritten(handle_t
*handle
, struct buffer_head
*bh
)
2320 return (buffer_delay(bh
) || buffer_unwritten(bh
)) && buffer_dirty(bh
);
2324 * __mpage_da_writepage - finds extent of pages and blocks
2326 * @page: page to consider
2327 * @wbc: not used, we just follow rules
2330 * The function finds extents of pages and scan them for all blocks.
2332 static int __mpage_da_writepage(struct page
*page
,
2333 struct writeback_control
*wbc
, void *data
)
2335 struct mpage_da_data
*mpd
= data
;
2336 struct inode
*inode
= mpd
->inode
;
2337 struct buffer_head
*bh
, *head
;
2342 * Rest of the page in the page_vec
2343 * redirty then and skip then. We will
2344 * try to to write them again after
2345 * starting a new transaction
2347 redirty_page_for_writepage(wbc
, page
);
2349 return MPAGE_DA_EXTENT_TAIL
;
2352 * Can we merge this page to current extent?
2354 if (mpd
->next_page
!= page
->index
) {
2356 * Nope, we can't. So, we map non-allocated blocks
2357 * and start IO on them using writepage()
2359 if (mpd
->next_page
!= mpd
->first_page
) {
2360 if (mpage_da_map_blocks(mpd
) == 0)
2361 mpage_da_submit_io(mpd
);
2363 * skip rest of the page in the page_vec
2366 redirty_page_for_writepage(wbc
, page
);
2368 return MPAGE_DA_EXTENT_TAIL
;
2372 * Start next extent of pages ...
2374 mpd
->first_page
= page
->index
;
2384 mpd
->next_page
= page
->index
+ 1;
2385 logical
= (sector_t
) page
->index
<<
2386 (PAGE_CACHE_SHIFT
- inode
->i_blkbits
);
2388 if (!page_has_buffers(page
)) {
2389 mpage_add_bh_to_extent(mpd
, logical
, PAGE_CACHE_SIZE
,
2390 (1 << BH_Dirty
) | (1 << BH_Uptodate
));
2392 return MPAGE_DA_EXTENT_TAIL
;
2395 * Page with regular buffer heads, just add all dirty ones
2397 head
= page_buffers(page
);
2400 BUG_ON(buffer_locked(bh
));
2402 * We need to try to allocate
2403 * unmapped blocks in the same page.
2404 * Otherwise we won't make progress
2405 * with the page in ext4_writepage
2407 if (ext4_bh_delay_or_unwritten(NULL
, bh
)) {
2408 mpage_add_bh_to_extent(mpd
, logical
,
2412 return MPAGE_DA_EXTENT_TAIL
;
2413 } else if (buffer_dirty(bh
) && (buffer_mapped(bh
))) {
2415 * mapped dirty buffer. We need to update
2416 * the b_state because we look at
2417 * b_state in mpage_da_map_blocks. We don't
2418 * update b_size because if we find an
2419 * unmapped buffer_head later we need to
2420 * use the b_state flag of that buffer_head.
2422 if (mpd
->b_size
== 0)
2423 mpd
->b_state
= bh
->b_state
& BH_FLAGS
;
2426 } while ((bh
= bh
->b_this_page
) != head
);
2433 * This is a special get_blocks_t callback which is used by
2434 * ext4_da_write_begin(). It will either return mapped block or
2435 * reserve space for a single block.
2437 * For delayed buffer_head we have BH_Mapped, BH_New, BH_Delay set.
2438 * We also have b_blocknr = -1 and b_bdev initialized properly
2440 * For unwritten buffer_head we have BH_Mapped, BH_New, BH_Unwritten set.
2441 * We also have b_blocknr = physicalblock mapping unwritten extent and b_bdev
2442 * initialized properly.
2444 static int ext4_da_get_block_prep(struct inode
*inode
, sector_t iblock
,
2445 struct buffer_head
*bh_result
, int create
)
2448 sector_t invalid_block
= ~((sector_t
) 0xffff);
2450 if (invalid_block
< ext4_blocks_count(EXT4_SB(inode
->i_sb
)->s_es
))
2453 BUG_ON(create
== 0);
2454 BUG_ON(bh_result
->b_size
!= inode
->i_sb
->s_blocksize
);
2457 * first, we need to know whether the block is allocated already
2458 * preallocated blocks are unmapped but should treated
2459 * the same as allocated blocks.
2461 ret
= ext4_get_blocks(NULL
, inode
, iblock
, 1, bh_result
, 0);
2462 if ((ret
== 0) && !buffer_delay(bh_result
)) {
2463 /* the block isn't (pre)allocated yet, let's reserve space */
2465 * XXX: __block_prepare_write() unmaps passed block,
2468 ret
= ext4_da_reserve_space(inode
, 1);
2470 /* not enough space to reserve */
2473 map_bh(bh_result
, inode
->i_sb
, invalid_block
);
2474 set_buffer_new(bh_result
);
2475 set_buffer_delay(bh_result
);
2476 } else if (ret
> 0) {
2477 bh_result
->b_size
= (ret
<< inode
->i_blkbits
);
2478 if (buffer_unwritten(bh_result
)) {
2479 /* A delayed write to unwritten bh should
2480 * be marked new and mapped. Mapped ensures
2481 * that we don't do get_block multiple times
2482 * when we write to the same offset and new
2483 * ensures that we do proper zero out for
2486 set_buffer_new(bh_result
);
2487 set_buffer_mapped(bh_result
);
2496 * This function is used as a standard get_block_t calback function
2497 * when there is no desire to allocate any blocks. It is used as a
2498 * callback function for block_prepare_write(), nobh_writepage(), and
2499 * block_write_full_page(). These functions should only try to map a
2500 * single block at a time.
2502 * Since this function doesn't do block allocations even if the caller
2503 * requests it by passing in create=1, it is critically important that
2504 * any caller checks to make sure that any buffer heads are returned
2505 * by this function are either all already mapped or marked for
2506 * delayed allocation before calling nobh_writepage() or
2507 * block_write_full_page(). Otherwise, b_blocknr could be left
2508 * unitialized, and the page write functions will be taken by
2511 static int noalloc_get_block_write(struct inode
*inode
, sector_t iblock
,
2512 struct buffer_head
*bh_result
, int create
)
2515 unsigned max_blocks
= bh_result
->b_size
>> inode
->i_blkbits
;
2517 BUG_ON(bh_result
->b_size
!= inode
->i_sb
->s_blocksize
);
2520 * we don't want to do block allocation in writepage
2521 * so call get_block_wrap with create = 0
2523 ret
= ext4_get_blocks(NULL
, inode
, iblock
, max_blocks
, bh_result
, 0);
2525 bh_result
->b_size
= (ret
<< inode
->i_blkbits
);
2531 static int bget_one(handle_t
*handle
, struct buffer_head
*bh
)
2537 static int bput_one(handle_t
*handle
, struct buffer_head
*bh
)
2543 static int __ext4_journalled_writepage(struct page
*page
,
2544 struct writeback_control
*wbc
,
2547 struct address_space
*mapping
= page
->mapping
;
2548 struct inode
*inode
= mapping
->host
;
2549 struct buffer_head
*page_bufs
;
2550 handle_t
*handle
= NULL
;
2554 page_bufs
= page_buffers(page
);
2556 walk_page_buffers(handle
, page_bufs
, 0, len
, NULL
, bget_one
);
2557 /* As soon as we unlock the page, it can go away, but we have
2558 * references to buffers so we are safe */
2561 handle
= ext4_journal_start(inode
, ext4_writepage_trans_blocks(inode
));
2562 if (IS_ERR(handle
)) {
2563 ret
= PTR_ERR(handle
);
2567 ret
= walk_page_buffers(handle
, page_bufs
, 0, len
, NULL
,
2568 do_journal_get_write_access
);
2570 err
= walk_page_buffers(handle
, page_bufs
, 0, len
, NULL
,
2574 err
= ext4_journal_stop(handle
);
2578 walk_page_buffers(handle
, page_bufs
, 0, len
, NULL
, bput_one
);
2579 EXT4_I(inode
)->i_state
|= EXT4_STATE_JDATA
;
2585 * Note that we don't need to start a transaction unless we're journaling data
2586 * because we should have holes filled from ext4_page_mkwrite(). We even don't
2587 * need to file the inode to the transaction's list in ordered mode because if
2588 * we are writing back data added by write(), the inode is already there and if
2589 * we are writing back data modified via mmap(), noone guarantees in which
2590 * transaction the data will hit the disk. In case we are journaling data, we
2591 * cannot start transaction directly because transaction start ranks above page
2592 * lock so we have to do some magic.
2594 * This function can get called via...
2595 * - ext4_da_writepages after taking page lock (have journal handle)
2596 * - journal_submit_inode_data_buffers (no journal handle)
2597 * - shrink_page_list via pdflush (no journal handle)
2598 * - grab_page_cache when doing write_begin (have journal handle)
2600 * We don't do any block allocation in this function. If we have page with
2601 * multiple blocks we need to write those buffer_heads that are mapped. This
2602 * is important for mmaped based write. So if we do with blocksize 1K
2603 * truncate(f, 1024);
2604 * a = mmap(f, 0, 4096);
2606 * truncate(f, 4096);
2607 * we have in the page first buffer_head mapped via page_mkwrite call back
2608 * but other bufer_heads would be unmapped but dirty(dirty done via the
2609 * do_wp_page). So writepage should write the first block. If we modify
2610 * the mmap area beyond 1024 we will again get a page_fault and the
2611 * page_mkwrite callback will do the block allocation and mark the
2612 * buffer_heads mapped.
2614 * We redirty the page if we have any buffer_heads that is either delay or
2615 * unwritten in the page.
2617 * We can get recursively called as show below.
2619 * ext4_writepage() -> kmalloc() -> __alloc_pages() -> page_launder() ->
2622 * But since we don't do any block allocation we should not deadlock.
2623 * Page also have the dirty flag cleared so we don't get recurive page_lock.
2625 static int ext4_writepage(struct page
*page
,
2626 struct writeback_control
*wbc
)
2631 struct buffer_head
*page_bufs
;
2632 struct inode
*inode
= page
->mapping
->host
;
2634 trace_ext4_writepage(inode
, page
);
2635 size
= i_size_read(inode
);
2636 if (page
->index
== size
>> PAGE_CACHE_SHIFT
)
2637 len
= size
& ~PAGE_CACHE_MASK
;
2639 len
= PAGE_CACHE_SIZE
;
2641 if (page_has_buffers(page
)) {
2642 page_bufs
= page_buffers(page
);
2643 if (walk_page_buffers(NULL
, page_bufs
, 0, len
, NULL
,
2644 ext4_bh_delay_or_unwritten
)) {
2646 * We don't want to do block allocation
2647 * So redirty the page and return
2648 * We may reach here when we do a journal commit
2649 * via journal_submit_inode_data_buffers.
2650 * If we don't have mapping block we just ignore
2651 * them. We can also reach here via shrink_page_list
2653 redirty_page_for_writepage(wbc
, page
);
2659 * The test for page_has_buffers() is subtle:
2660 * We know the page is dirty but it lost buffers. That means
2661 * that at some moment in time after write_begin()/write_end()
2662 * has been called all buffers have been clean and thus they
2663 * must have been written at least once. So they are all
2664 * mapped and we can happily proceed with mapping them
2665 * and writing the page.
2667 * Try to initialize the buffer_heads and check whether
2668 * all are mapped and non delay. We don't want to
2669 * do block allocation here.
2671 ret
= block_prepare_write(page
, 0, len
,
2672 noalloc_get_block_write
);
2674 page_bufs
= page_buffers(page
);
2675 /* check whether all are mapped and non delay */
2676 if (walk_page_buffers(NULL
, page_bufs
, 0, len
, NULL
,
2677 ext4_bh_delay_or_unwritten
)) {
2678 redirty_page_for_writepage(wbc
, page
);
2684 * We can't do block allocation here
2685 * so just redity the page and unlock
2688 redirty_page_for_writepage(wbc
, page
);
2692 /* now mark the buffer_heads as dirty and uptodate */
2693 block_commit_write(page
, 0, len
);
2696 if (PageChecked(page
) && ext4_should_journal_data(inode
)) {
2698 * It's mmapped pagecache. Add buffers and journal it. There
2699 * doesn't seem much point in redirtying the page here.
2701 ClearPageChecked(page
);
2702 return __ext4_journalled_writepage(page
, wbc
, len
);
2705 if (test_opt(inode
->i_sb
, NOBH
) && ext4_should_writeback_data(inode
))
2706 ret
= nobh_writepage(page
, noalloc_get_block_write
, wbc
);
2708 ret
= block_write_full_page(page
, noalloc_get_block_write
,
2715 * This is called via ext4_da_writepages() to
2716 * calulate the total number of credits to reserve to fit
2717 * a single extent allocation into a single transaction,
2718 * ext4_da_writpeages() will loop calling this before
2719 * the block allocation.
2722 static int ext4_da_writepages_trans_blocks(struct inode
*inode
)
2724 int max_blocks
= EXT4_I(inode
)->i_reserved_data_blocks
;
2727 * With non-extent format the journal credit needed to
2728 * insert nrblocks contiguous block is dependent on
2729 * number of contiguous block. So we will limit
2730 * number of contiguous block to a sane value
2732 if (!(inode
->i_flags
& EXT4_EXTENTS_FL
) &&
2733 (max_blocks
> EXT4_MAX_TRANS_DATA
))
2734 max_blocks
= EXT4_MAX_TRANS_DATA
;
2736 return ext4_chunk_trans_blocks(inode
, max_blocks
);
2739 static int ext4_da_writepages(struct address_space
*mapping
,
2740 struct writeback_control
*wbc
)
2743 int range_whole
= 0;
2744 handle_t
*handle
= NULL
;
2745 struct mpage_da_data mpd
;
2746 struct inode
*inode
= mapping
->host
;
2747 int no_nrwrite_index_update
;
2748 int pages_written
= 0;
2750 int range_cyclic
, cycled
= 1, io_done
= 0;
2751 int needed_blocks
, ret
= 0, nr_to_writebump
= 0;
2752 loff_t range_start
= wbc
->range_start
;
2753 struct ext4_sb_info
*sbi
= EXT4_SB(mapping
->host
->i_sb
);
2755 trace_ext4_da_writepages(inode
, wbc
);
2758 * No pages to write? This is mainly a kludge to avoid starting
2759 * a transaction for special inodes like journal inode on last iput()
2760 * because that could violate lock ordering on umount
2762 if (!mapping
->nrpages
|| !mapping_tagged(mapping
, PAGECACHE_TAG_DIRTY
))
2766 * If the filesystem has aborted, it is read-only, so return
2767 * right away instead of dumping stack traces later on that
2768 * will obscure the real source of the problem. We test
2769 * EXT4_MF_FS_ABORTED instead of sb->s_flag's MS_RDONLY because
2770 * the latter could be true if the filesystem is mounted
2771 * read-only, and in that case, ext4_da_writepages should
2772 * *never* be called, so if that ever happens, we would want
2775 if (unlikely(sbi
->s_mount_flags
& EXT4_MF_FS_ABORTED
))
2779 * Make sure nr_to_write is >= sbi->s_mb_stream_request
2780 * This make sure small files blocks are allocated in
2781 * single attempt. This ensure that small files
2782 * get less fragmented.
2784 if (wbc
->nr_to_write
< sbi
->s_mb_stream_request
) {
2785 nr_to_writebump
= sbi
->s_mb_stream_request
- wbc
->nr_to_write
;
2786 wbc
->nr_to_write
= sbi
->s_mb_stream_request
;
2788 if (wbc
->range_start
== 0 && wbc
->range_end
== LLONG_MAX
)
2791 range_cyclic
= wbc
->range_cyclic
;
2792 if (wbc
->range_cyclic
) {
2793 index
= mapping
->writeback_index
;
2796 wbc
->range_start
= index
<< PAGE_CACHE_SHIFT
;
2797 wbc
->range_end
= LLONG_MAX
;
2798 wbc
->range_cyclic
= 0;
2800 index
= wbc
->range_start
>> PAGE_CACHE_SHIFT
;
2803 mpd
.inode
= mapping
->host
;
2806 * we don't want write_cache_pages to update
2807 * nr_to_write and writeback_index
2809 no_nrwrite_index_update
= wbc
->no_nrwrite_index_update
;
2810 wbc
->no_nrwrite_index_update
= 1;
2811 pages_skipped
= wbc
->pages_skipped
;
2814 while (!ret
&& wbc
->nr_to_write
> 0) {
2817 * we insert one extent at a time. So we need
2818 * credit needed for single extent allocation.
2819 * journalled mode is currently not supported
2822 BUG_ON(ext4_should_journal_data(inode
));
2823 needed_blocks
= ext4_da_writepages_trans_blocks(inode
);
2825 /* start a new transaction*/
2826 handle
= ext4_journal_start(inode
, needed_blocks
);
2827 if (IS_ERR(handle
)) {
2828 ret
= PTR_ERR(handle
);
2829 printk(KERN_CRIT
"%s: jbd2_start: "
2830 "%ld pages, ino %lu; err %d\n", __func__
,
2831 wbc
->nr_to_write
, inode
->i_ino
, ret
);
2833 goto out_writepages
;
2837 * Now call __mpage_da_writepage to find the next
2838 * contiguous region of logical blocks that need
2839 * blocks to be allocated by ext4. We don't actually
2840 * submit the blocks for I/O here, even though
2841 * write_cache_pages thinks it will, and will set the
2842 * pages as clean for write before calling
2843 * __mpage_da_writepage().
2851 mpd
.pages_written
= 0;
2853 ret
= write_cache_pages(mapping
, wbc
, __mpage_da_writepage
,
2856 * If we have a contigous extent of pages and we
2857 * haven't done the I/O yet, map the blocks and submit
2860 if (!mpd
.io_done
&& mpd
.next_page
!= mpd
.first_page
) {
2861 if (mpage_da_map_blocks(&mpd
) == 0)
2862 mpage_da_submit_io(&mpd
);
2864 ret
= MPAGE_DA_EXTENT_TAIL
;
2866 wbc
->nr_to_write
-= mpd
.pages_written
;
2868 ext4_journal_stop(handle
);
2870 if ((mpd
.retval
== -ENOSPC
) && sbi
->s_journal
) {
2871 /* commit the transaction which would
2872 * free blocks released in the transaction
2875 jbd2_journal_force_commit_nested(sbi
->s_journal
);
2876 wbc
->pages_skipped
= pages_skipped
;
2878 } else if (ret
== MPAGE_DA_EXTENT_TAIL
) {
2880 * got one extent now try with
2883 pages_written
+= mpd
.pages_written
;
2884 wbc
->pages_skipped
= pages_skipped
;
2887 } else if (wbc
->nr_to_write
)
2889 * There is no more writeout needed
2890 * or we requested for a noblocking writeout
2891 * and we found the device congested
2895 if (!io_done
&& !cycled
) {
2898 wbc
->range_start
= index
<< PAGE_CACHE_SHIFT
;
2899 wbc
->range_end
= mapping
->writeback_index
- 1;
2902 if (pages_skipped
!= wbc
->pages_skipped
)
2903 printk(KERN_EMERG
"This should not happen leaving %s "
2904 "with nr_to_write = %ld ret = %d\n",
2905 __func__
, wbc
->nr_to_write
, ret
);
2908 index
+= pages_written
;
2909 wbc
->range_cyclic
= range_cyclic
;
2910 if (wbc
->range_cyclic
|| (range_whole
&& wbc
->nr_to_write
> 0))
2912 * set the writeback_index so that range_cyclic
2913 * mode will write it back later
2915 mapping
->writeback_index
= index
;
2918 if (!no_nrwrite_index_update
)
2919 wbc
->no_nrwrite_index_update
= 0;
2920 wbc
->nr_to_write
-= nr_to_writebump
;
2921 wbc
->range_start
= range_start
;
2922 trace_ext4_da_writepages_result(inode
, wbc
, ret
, pages_written
);
2926 #define FALL_BACK_TO_NONDELALLOC 1
2927 static int ext4_nonda_switch(struct super_block
*sb
)
2929 s64 free_blocks
, dirty_blocks
;
2930 struct ext4_sb_info
*sbi
= EXT4_SB(sb
);
2933 * switch to non delalloc mode if we are running low
2934 * on free block. The free block accounting via percpu
2935 * counters can get slightly wrong with percpu_counter_batch getting
2936 * accumulated on each CPU without updating global counters
2937 * Delalloc need an accurate free block accounting. So switch
2938 * to non delalloc when we are near to error range.
2940 free_blocks
= percpu_counter_read_positive(&sbi
->s_freeblocks_counter
);
2941 dirty_blocks
= percpu_counter_read_positive(&sbi
->s_dirtyblocks_counter
);
2942 if (2 * free_blocks
< 3 * dirty_blocks
||
2943 free_blocks
< (dirty_blocks
+ EXT4_FREEBLOCKS_WATERMARK
)) {
2945 * free block count is less that 150% of dirty blocks
2946 * or free blocks is less that watermark
2953 static int ext4_da_write_begin(struct file
*file
, struct address_space
*mapping
,
2954 loff_t pos
, unsigned len
, unsigned flags
,
2955 struct page
**pagep
, void **fsdata
)
2957 int ret
, retries
= 0;
2961 struct inode
*inode
= mapping
->host
;
2964 index
= pos
>> PAGE_CACHE_SHIFT
;
2965 from
= pos
& (PAGE_CACHE_SIZE
- 1);
2968 if (ext4_nonda_switch(inode
->i_sb
)) {
2969 *fsdata
= (void *)FALL_BACK_TO_NONDELALLOC
;
2970 return ext4_write_begin(file
, mapping
, pos
,
2971 len
, flags
, pagep
, fsdata
);
2973 *fsdata
= (void *)0;
2974 trace_ext4_da_write_begin(inode
, pos
, len
, flags
);
2977 * With delayed allocation, we don't log the i_disksize update
2978 * if there is delayed block allocation. But we still need
2979 * to journalling the i_disksize update if writes to the end
2980 * of file which has an already mapped buffer.
2982 handle
= ext4_journal_start(inode
, 1);
2983 if (IS_ERR(handle
)) {
2984 ret
= PTR_ERR(handle
);
2987 /* We cannot recurse into the filesystem as the transaction is already
2989 flags
|= AOP_FLAG_NOFS
;
2991 page
= grab_cache_page_write_begin(mapping
, index
, flags
);
2993 ext4_journal_stop(handle
);
2999 ret
= block_write_begin(file
, mapping
, pos
, len
, flags
, pagep
, fsdata
,
3000 ext4_da_get_block_prep
);
3003 ext4_journal_stop(handle
);
3004 page_cache_release(page
);
3006 * block_write_begin may have instantiated a few blocks
3007 * outside i_size. Trim these off again. Don't need
3008 * i_size_read because we hold i_mutex.
3010 if (pos
+ len
> inode
->i_size
)
3011 ext4_truncate(inode
);
3014 if (ret
== -ENOSPC
&& ext4_should_retry_alloc(inode
->i_sb
, &retries
))
3021 * Check if we should update i_disksize
3022 * when write to the end of file but not require block allocation
3024 static int ext4_da_should_update_i_disksize(struct page
*page
,
3025 unsigned long offset
)
3027 struct buffer_head
*bh
;
3028 struct inode
*inode
= page
->mapping
->host
;
3032 bh
= page_buffers(page
);
3033 idx
= offset
>> inode
->i_blkbits
;
3035 for (i
= 0; i
< idx
; i
++)
3036 bh
= bh
->b_this_page
;
3038 if (!buffer_mapped(bh
) || (buffer_delay(bh
)) || buffer_unwritten(bh
))
3043 static int ext4_da_write_end(struct file
*file
,
3044 struct address_space
*mapping
,
3045 loff_t pos
, unsigned len
, unsigned copied
,
3046 struct page
*page
, void *fsdata
)
3048 struct inode
*inode
= mapping
->host
;
3050 handle_t
*handle
= ext4_journal_current_handle();
3052 unsigned long start
, end
;
3053 int write_mode
= (int)(unsigned long)fsdata
;
3055 if (write_mode
== FALL_BACK_TO_NONDELALLOC
) {
3056 if (ext4_should_order_data(inode
)) {
3057 return ext4_ordered_write_end(file
, mapping
, pos
,
3058 len
, copied
, page
, fsdata
);
3059 } else if (ext4_should_writeback_data(inode
)) {
3060 return ext4_writeback_write_end(file
, mapping
, pos
,
3061 len
, copied
, page
, fsdata
);
3067 trace_ext4_da_write_end(inode
, pos
, len
, copied
);
3068 start
= pos
& (PAGE_CACHE_SIZE
- 1);
3069 end
= start
+ copied
- 1;
3072 * generic_write_end() will run mark_inode_dirty() if i_size
3073 * changes. So let's piggyback the i_disksize mark_inode_dirty
3077 new_i_size
= pos
+ copied
;
3078 if (new_i_size
> EXT4_I(inode
)->i_disksize
) {
3079 if (ext4_da_should_update_i_disksize(page
, end
)) {
3080 down_write(&EXT4_I(inode
)->i_data_sem
);
3081 if (new_i_size
> EXT4_I(inode
)->i_disksize
) {
3083 * Updating i_disksize when extending file
3084 * without needing block allocation
3086 if (ext4_should_order_data(inode
))
3087 ret
= ext4_jbd2_file_inode(handle
,
3090 EXT4_I(inode
)->i_disksize
= new_i_size
;
3092 up_write(&EXT4_I(inode
)->i_data_sem
);
3093 /* We need to mark inode dirty even if
3094 * new_i_size is less that inode->i_size
3095 * bu greater than i_disksize.(hint delalloc)
3097 ext4_mark_inode_dirty(handle
, inode
);
3100 ret2
= generic_write_end(file
, mapping
, pos
, len
, copied
,
3105 ret2
= ext4_journal_stop(handle
);
3109 return ret
? ret
: copied
;
3112 static void ext4_da_invalidatepage(struct page
*page
, unsigned long offset
)
3115 * Drop reserved blocks
3117 BUG_ON(!PageLocked(page
));
3118 if (!page_has_buffers(page
))
3121 ext4_da_page_release_reservation(page
, offset
);
3124 ext4_invalidatepage(page
, offset
);
3130 * Force all delayed allocation blocks to be allocated for a given inode.
3132 int ext4_alloc_da_blocks(struct inode
*inode
)
3134 if (!EXT4_I(inode
)->i_reserved_data_blocks
&&
3135 !EXT4_I(inode
)->i_reserved_meta_blocks
)
3139 * We do something simple for now. The filemap_flush() will
3140 * also start triggering a write of the data blocks, which is
3141 * not strictly speaking necessary (and for users of
3142 * laptop_mode, not even desirable). However, to do otherwise
3143 * would require replicating code paths in:
3145 * ext4_da_writepages() ->
3146 * write_cache_pages() ---> (via passed in callback function)
3147 * __mpage_da_writepage() -->
3148 * mpage_add_bh_to_extent()
3149 * mpage_da_map_blocks()
3151 * The problem is that write_cache_pages(), located in
3152 * mm/page-writeback.c, marks pages clean in preparation for
3153 * doing I/O, which is not desirable if we're not planning on
3156 * We could call write_cache_pages(), and then redirty all of
3157 * the pages by calling redirty_page_for_writeback() but that
3158 * would be ugly in the extreme. So instead we would need to
3159 * replicate parts of the code in the above functions,
3160 * simplifying them becuase we wouldn't actually intend to
3161 * write out the pages, but rather only collect contiguous
3162 * logical block extents, call the multi-block allocator, and
3163 * then update the buffer heads with the block allocations.
3165 * For now, though, we'll cheat by calling filemap_flush(),
3166 * which will map the blocks, and start the I/O, but not
3167 * actually wait for the I/O to complete.
3169 return filemap_flush(inode
->i_mapping
);
3173 * bmap() is special. It gets used by applications such as lilo and by
3174 * the swapper to find the on-disk block of a specific piece of data.
3176 * Naturally, this is dangerous if the block concerned is still in the
3177 * journal. If somebody makes a swapfile on an ext4 data-journaling
3178 * filesystem and enables swap, then they may get a nasty shock when the
3179 * data getting swapped to that swapfile suddenly gets overwritten by
3180 * the original zero's written out previously to the journal and
3181 * awaiting writeback in the kernel's buffer cache.
3183 * So, if we see any bmap calls here on a modified, data-journaled file,
3184 * take extra steps to flush any blocks which might be in the cache.
3186 static sector_t
ext4_bmap(struct address_space
*mapping
, sector_t block
)
3188 struct inode
*inode
= mapping
->host
;
3192 if (mapping_tagged(mapping
, PAGECACHE_TAG_DIRTY
) &&
3193 test_opt(inode
->i_sb
, DELALLOC
)) {
3195 * With delalloc we want to sync the file
3196 * so that we can make sure we allocate
3199 filemap_write_and_wait(mapping
);
3202 if (EXT4_JOURNAL(inode
) && EXT4_I(inode
)->i_state
& EXT4_STATE_JDATA
) {
3204 * This is a REALLY heavyweight approach, but the use of
3205 * bmap on dirty files is expected to be extremely rare:
3206 * only if we run lilo or swapon on a freshly made file
3207 * do we expect this to happen.
3209 * (bmap requires CAP_SYS_RAWIO so this does not
3210 * represent an unprivileged user DOS attack --- we'd be
3211 * in trouble if mortal users could trigger this path at
3214 * NB. EXT4_STATE_JDATA is not set on files other than
3215 * regular files. If somebody wants to bmap a directory
3216 * or symlink and gets confused because the buffer
3217 * hasn't yet been flushed to disk, they deserve
3218 * everything they get.
3221 EXT4_I(inode
)->i_state
&= ~EXT4_STATE_JDATA
;
3222 journal
= EXT4_JOURNAL(inode
);
3223 jbd2_journal_lock_updates(journal
);
3224 err
= jbd2_journal_flush(journal
);
3225 jbd2_journal_unlock_updates(journal
);
3231 return generic_block_bmap(mapping
, block
, ext4_get_block
);
3234 static int ext4_readpage(struct file
*file
, struct page
*page
)
3236 return mpage_readpage(page
, ext4_get_block
);
3240 ext4_readpages(struct file
*file
, struct address_space
*mapping
,
3241 struct list_head
*pages
, unsigned nr_pages
)
3243 return mpage_readpages(mapping
, pages
, nr_pages
, ext4_get_block
);
3246 static void ext4_invalidatepage(struct page
*page
, unsigned long offset
)
3248 journal_t
*journal
= EXT4_JOURNAL(page
->mapping
->host
);
3251 * If it's a full truncate we just forget about the pending dirtying
3254 ClearPageChecked(page
);
3257 jbd2_journal_invalidatepage(journal
, page
, offset
);
3259 block_invalidatepage(page
, offset
);
3262 static int ext4_releasepage(struct page
*page
, gfp_t wait
)
3264 journal_t
*journal
= EXT4_JOURNAL(page
->mapping
->host
);
3266 WARN_ON(PageChecked(page
));
3267 if (!page_has_buffers(page
))
3270 return jbd2_journal_try_to_free_buffers(journal
, page
, wait
);
3272 return try_to_free_buffers(page
);
3276 * If the O_DIRECT write will extend the file then add this inode to the
3277 * orphan list. So recovery will truncate it back to the original size
3278 * if the machine crashes during the write.
3280 * If the O_DIRECT write is intantiating holes inside i_size and the machine
3281 * crashes then stale disk data _may_ be exposed inside the file. But current
3282 * VFS code falls back into buffered path in that case so we are safe.
3284 static ssize_t
ext4_direct_IO(int rw
, struct kiocb
*iocb
,
3285 const struct iovec
*iov
, loff_t offset
,
3286 unsigned long nr_segs
)
3288 struct file
*file
= iocb
->ki_filp
;
3289 struct inode
*inode
= file
->f_mapping
->host
;
3290 struct ext4_inode_info
*ei
= EXT4_I(inode
);
3294 size_t count
= iov_length(iov
, nr_segs
);
3297 loff_t final_size
= offset
+ count
;
3299 if (final_size
> inode
->i_size
) {
3300 /* Credits for sb + inode write */
3301 handle
= ext4_journal_start(inode
, 2);
3302 if (IS_ERR(handle
)) {
3303 ret
= PTR_ERR(handle
);
3306 ret
= ext4_orphan_add(handle
, inode
);
3308 ext4_journal_stop(handle
);
3312 ei
->i_disksize
= inode
->i_size
;
3313 ext4_journal_stop(handle
);
3317 ret
= blockdev_direct_IO(rw
, iocb
, inode
, inode
->i_sb
->s_bdev
, iov
,
3319 ext4_get_block
, NULL
);
3324 /* Credits for sb + inode write */
3325 handle
= ext4_journal_start(inode
, 2);
3326 if (IS_ERR(handle
)) {
3327 /* This is really bad luck. We've written the data
3328 * but cannot extend i_size. Bail out and pretend
3329 * the write failed... */
3330 ret
= PTR_ERR(handle
);
3334 ext4_orphan_del(handle
, inode
);
3336 loff_t end
= offset
+ ret
;
3337 if (end
> inode
->i_size
) {
3338 ei
->i_disksize
= end
;
3339 i_size_write(inode
, end
);
3341 * We're going to return a positive `ret'
3342 * here due to non-zero-length I/O, so there's
3343 * no way of reporting error returns from
3344 * ext4_mark_inode_dirty() to userspace. So
3347 ext4_mark_inode_dirty(handle
, inode
);
3350 err
= ext4_journal_stop(handle
);
3359 * Pages can be marked dirty completely asynchronously from ext4's journalling
3360 * activity. By filemap_sync_pte(), try_to_unmap_one(), etc. We cannot do
3361 * much here because ->set_page_dirty is called under VFS locks. The page is
3362 * not necessarily locked.
3364 * We cannot just dirty the page and leave attached buffers clean, because the
3365 * buffers' dirty state is "definitive". We cannot just set the buffers dirty
3366 * or jbddirty because all the journalling code will explode.
3368 * So what we do is to mark the page "pending dirty" and next time writepage
3369 * is called, propagate that into the buffers appropriately.
3371 static int ext4_journalled_set_page_dirty(struct page
*page
)
3373 SetPageChecked(page
);
3374 return __set_page_dirty_nobuffers(page
);
3377 static const struct address_space_operations ext4_ordered_aops
= {
3378 .readpage
= ext4_readpage
,
3379 .readpages
= ext4_readpages
,
3380 .writepage
= ext4_writepage
,
3381 .sync_page
= block_sync_page
,
3382 .write_begin
= ext4_write_begin
,
3383 .write_end
= ext4_ordered_write_end
,
3385 .invalidatepage
= ext4_invalidatepage
,
3386 .releasepage
= ext4_releasepage
,
3387 .direct_IO
= ext4_direct_IO
,
3388 .migratepage
= buffer_migrate_page
,
3389 .is_partially_uptodate
= block_is_partially_uptodate
,
3392 static const struct address_space_operations ext4_writeback_aops
= {
3393 .readpage
= ext4_readpage
,
3394 .readpages
= ext4_readpages
,
3395 .writepage
= ext4_writepage
,
3396 .sync_page
= block_sync_page
,
3397 .write_begin
= ext4_write_begin
,
3398 .write_end
= ext4_writeback_write_end
,
3400 .invalidatepage
= ext4_invalidatepage
,
3401 .releasepage
= ext4_releasepage
,
3402 .direct_IO
= ext4_direct_IO
,
3403 .migratepage
= buffer_migrate_page
,
3404 .is_partially_uptodate
= block_is_partially_uptodate
,
3407 static const struct address_space_operations ext4_journalled_aops
= {
3408 .readpage
= ext4_readpage
,
3409 .readpages
= ext4_readpages
,
3410 .writepage
= ext4_writepage
,
3411 .sync_page
= block_sync_page
,
3412 .write_begin
= ext4_write_begin
,
3413 .write_end
= ext4_journalled_write_end
,
3414 .set_page_dirty
= ext4_journalled_set_page_dirty
,
3416 .invalidatepage
= ext4_invalidatepage
,
3417 .releasepage
= ext4_releasepage
,
3418 .is_partially_uptodate
= block_is_partially_uptodate
,
3421 static const struct address_space_operations ext4_da_aops
= {
3422 .readpage
= ext4_readpage
,
3423 .readpages
= ext4_readpages
,
3424 .writepage
= ext4_writepage
,
3425 .writepages
= ext4_da_writepages
,
3426 .sync_page
= block_sync_page
,
3427 .write_begin
= ext4_da_write_begin
,
3428 .write_end
= ext4_da_write_end
,
3430 .invalidatepage
= ext4_da_invalidatepage
,
3431 .releasepage
= ext4_releasepage
,
3432 .direct_IO
= ext4_direct_IO
,
3433 .migratepage
= buffer_migrate_page
,
3434 .is_partially_uptodate
= block_is_partially_uptodate
,
3437 void ext4_set_aops(struct inode
*inode
)
3439 if (ext4_should_order_data(inode
) &&
3440 test_opt(inode
->i_sb
, DELALLOC
))
3441 inode
->i_mapping
->a_ops
= &ext4_da_aops
;
3442 else if (ext4_should_order_data(inode
))
3443 inode
->i_mapping
->a_ops
= &ext4_ordered_aops
;
3444 else if (ext4_should_writeback_data(inode
) &&
3445 test_opt(inode
->i_sb
, DELALLOC
))
3446 inode
->i_mapping
->a_ops
= &ext4_da_aops
;
3447 else if (ext4_should_writeback_data(inode
))
3448 inode
->i_mapping
->a_ops
= &ext4_writeback_aops
;
3450 inode
->i_mapping
->a_ops
= &ext4_journalled_aops
;
3454 * ext4_block_truncate_page() zeroes out a mapping from file offset `from'
3455 * up to the end of the block which corresponds to `from'.
3456 * This required during truncate. We need to physically zero the tail end
3457 * of that block so it doesn't yield old data if the file is later grown.
3459 int ext4_block_truncate_page(handle_t
*handle
,
3460 struct address_space
*mapping
, loff_t from
)
3462 ext4_fsblk_t index
= from
>> PAGE_CACHE_SHIFT
;
3463 unsigned offset
= from
& (PAGE_CACHE_SIZE
-1);
3464 unsigned blocksize
, length
, pos
;
3466 struct inode
*inode
= mapping
->host
;
3467 struct buffer_head
*bh
;
3471 page
= find_or_create_page(mapping
, from
>> PAGE_CACHE_SHIFT
,
3472 mapping_gfp_mask(mapping
) & ~__GFP_FS
);
3476 blocksize
= inode
->i_sb
->s_blocksize
;
3477 length
= blocksize
- (offset
& (blocksize
- 1));
3478 iblock
= index
<< (PAGE_CACHE_SHIFT
- inode
->i_sb
->s_blocksize_bits
);
3481 * For "nobh" option, we can only work if we don't need to
3482 * read-in the page - otherwise we create buffers to do the IO.
3484 if (!page_has_buffers(page
) && test_opt(inode
->i_sb
, NOBH
) &&
3485 ext4_should_writeback_data(inode
) && PageUptodate(page
)) {
3486 zero_user(page
, offset
, length
);
3487 set_page_dirty(page
);
3491 if (!page_has_buffers(page
))
3492 create_empty_buffers(page
, blocksize
, 0);
3494 /* Find the buffer that contains "offset" */
3495 bh
= page_buffers(page
);
3497 while (offset
>= pos
) {
3498 bh
= bh
->b_this_page
;
3504 if (buffer_freed(bh
)) {
3505 BUFFER_TRACE(bh
, "freed: skip");
3509 if (!buffer_mapped(bh
)) {
3510 BUFFER_TRACE(bh
, "unmapped");
3511 ext4_get_block(inode
, iblock
, bh
, 0);
3512 /* unmapped? It's a hole - nothing to do */
3513 if (!buffer_mapped(bh
)) {
3514 BUFFER_TRACE(bh
, "still unmapped");
3519 /* Ok, it's mapped. Make sure it's up-to-date */
3520 if (PageUptodate(page
))
3521 set_buffer_uptodate(bh
);
3523 if (!buffer_uptodate(bh
)) {
3525 ll_rw_block(READ
, 1, &bh
);
3527 /* Uhhuh. Read error. Complain and punt. */
3528 if (!buffer_uptodate(bh
))
3532 if (ext4_should_journal_data(inode
)) {
3533 BUFFER_TRACE(bh
, "get write access");
3534 err
= ext4_journal_get_write_access(handle
, bh
);
3539 zero_user(page
, offset
, length
);
3541 BUFFER_TRACE(bh
, "zeroed end of block");
3544 if (ext4_should_journal_data(inode
)) {
3545 err
= ext4_handle_dirty_metadata(handle
, inode
, bh
);
3547 if (ext4_should_order_data(inode
))
3548 err
= ext4_jbd2_file_inode(handle
, inode
);
3549 mark_buffer_dirty(bh
);
3554 page_cache_release(page
);
3559 * Probably it should be a library function... search for first non-zero word
3560 * or memcmp with zero_page, whatever is better for particular architecture.
3563 static inline int all_zeroes(__le32
*p
, __le32
*q
)
3572 * ext4_find_shared - find the indirect blocks for partial truncation.
3573 * @inode: inode in question
3574 * @depth: depth of the affected branch
3575 * @offsets: offsets of pointers in that branch (see ext4_block_to_path)
3576 * @chain: place to store the pointers to partial indirect blocks
3577 * @top: place to the (detached) top of branch
3579 * This is a helper function used by ext4_truncate().
3581 * When we do truncate() we may have to clean the ends of several
3582 * indirect blocks but leave the blocks themselves alive. Block is
3583 * partially truncated if some data below the new i_size is refered
3584 * from it (and it is on the path to the first completely truncated
3585 * data block, indeed). We have to free the top of that path along
3586 * with everything to the right of the path. Since no allocation
3587 * past the truncation point is possible until ext4_truncate()
3588 * finishes, we may safely do the latter, but top of branch may
3589 * require special attention - pageout below the truncation point
3590 * might try to populate it.
3592 * We atomically detach the top of branch from the tree, store the
3593 * block number of its root in *@top, pointers to buffer_heads of
3594 * partially truncated blocks - in @chain[].bh and pointers to
3595 * their last elements that should not be removed - in
3596 * @chain[].p. Return value is the pointer to last filled element
3599 * The work left to caller to do the actual freeing of subtrees:
3600 * a) free the subtree starting from *@top
3601 * b) free the subtrees whose roots are stored in
3602 * (@chain[i].p+1 .. end of @chain[i].bh->b_data)
3603 * c) free the subtrees growing from the inode past the @chain[0].
3604 * (no partially truncated stuff there). */
3606 static Indirect
*ext4_find_shared(struct inode
*inode
, int depth
,
3607 ext4_lblk_t offsets
[4], Indirect chain
[4],
3610 Indirect
*partial
, *p
;
3614 /* Make k index the deepest non-null offest + 1 */
3615 for (k
= depth
; k
> 1 && !offsets
[k
-1]; k
--)
3617 partial
= ext4_get_branch(inode
, k
, offsets
, chain
, &err
);
3618 /* Writer: pointers */
3620 partial
= chain
+ k
-1;
3622 * If the branch acquired continuation since we've looked at it -
3623 * fine, it should all survive and (new) top doesn't belong to us.
3625 if (!partial
->key
&& *partial
->p
)
3628 for (p
= partial
; (p
> chain
) && all_zeroes((__le32
*) p
->bh
->b_data
, p
->p
); p
--)
3631 * OK, we've found the last block that must survive. The rest of our
3632 * branch should be detached before unlocking. However, if that rest
3633 * of branch is all ours and does not grow immediately from the inode
3634 * it's easier to cheat and just decrement partial->p.
3636 if (p
== chain
+ k
- 1 && p
> chain
) {
3640 /* Nope, don't do this in ext4. Must leave the tree intact */
3647 while (partial
> p
) {
3648 brelse(partial
->bh
);
3656 * Zero a number of block pointers in either an inode or an indirect block.
3657 * If we restart the transaction we must again get write access to the
3658 * indirect block for further modification.
3660 * We release `count' blocks on disk, but (last - first) may be greater
3661 * than `count' because there can be holes in there.
3663 static void ext4_clear_blocks(handle_t
*handle
, struct inode
*inode
,
3664 struct buffer_head
*bh
,
3665 ext4_fsblk_t block_to_free
,
3666 unsigned long count
, __le32
*first
,
3670 if (try_to_extend_transaction(handle
, inode
)) {
3672 BUFFER_TRACE(bh
, "call ext4_handle_dirty_metadata");
3673 ext4_handle_dirty_metadata(handle
, inode
, bh
);
3675 ext4_mark_inode_dirty(handle
, inode
);
3676 ext4_truncate_restart_trans(handle
, inode
,
3677 blocks_for_truncate(inode
));
3679 BUFFER_TRACE(bh
, "retaking write access");
3680 ext4_journal_get_write_access(handle
, bh
);
3685 * Any buffers which are on the journal will be in memory. We
3686 * find them on the hash table so jbd2_journal_revoke() will
3687 * run jbd2_journal_forget() on them. We've already detached
3688 * each block from the file, so bforget() in
3689 * jbd2_journal_forget() should be safe.
3691 * AKPM: turn on bforget in jbd2_journal_forget()!!!
3693 for (p
= first
; p
< last
; p
++) {
3694 u32 nr
= le32_to_cpu(*p
);
3696 struct buffer_head
*tbh
;
3699 tbh
= sb_find_get_block(inode
->i_sb
, nr
);
3700 ext4_forget(handle
, 0, inode
, tbh
, nr
);
3704 ext4_free_blocks(handle
, inode
, block_to_free
, count
, 0);
3708 * ext4_free_data - free a list of data blocks
3709 * @handle: handle for this transaction
3710 * @inode: inode we are dealing with
3711 * @this_bh: indirect buffer_head which contains *@first and *@last
3712 * @first: array of block numbers
3713 * @last: points immediately past the end of array
3715 * We are freeing all blocks refered from that array (numbers are stored as
3716 * little-endian 32-bit) and updating @inode->i_blocks appropriately.
3718 * We accumulate contiguous runs of blocks to free. Conveniently, if these
3719 * blocks are contiguous then releasing them at one time will only affect one
3720 * or two bitmap blocks (+ group descriptor(s) and superblock) and we won't
3721 * actually use a lot of journal space.
3723 * @this_bh will be %NULL if @first and @last point into the inode's direct
3726 static void ext4_free_data(handle_t
*handle
, struct inode
*inode
,
3727 struct buffer_head
*this_bh
,
3728 __le32
*first
, __le32
*last
)
3730 ext4_fsblk_t block_to_free
= 0; /* Starting block # of a run */
3731 unsigned long count
= 0; /* Number of blocks in the run */
3732 __le32
*block_to_free_p
= NULL
; /* Pointer into inode/ind
3735 ext4_fsblk_t nr
; /* Current block # */
3736 __le32
*p
; /* Pointer into inode/ind
3737 for current block */
3740 if (this_bh
) { /* For indirect block */
3741 BUFFER_TRACE(this_bh
, "get_write_access");
3742 err
= ext4_journal_get_write_access(handle
, this_bh
);
3743 /* Important: if we can't update the indirect pointers
3744 * to the blocks, we can't free them. */
3749 for (p
= first
; p
< last
; p
++) {
3750 nr
= le32_to_cpu(*p
);
3752 /* accumulate blocks to free if they're contiguous */
3755 block_to_free_p
= p
;
3757 } else if (nr
== block_to_free
+ count
) {
3760 ext4_clear_blocks(handle
, inode
, this_bh
,
3762 count
, block_to_free_p
, p
);
3764 block_to_free_p
= p
;
3771 ext4_clear_blocks(handle
, inode
, this_bh
, block_to_free
,
3772 count
, block_to_free_p
, p
);
3775 BUFFER_TRACE(this_bh
, "call ext4_handle_dirty_metadata");
3778 * The buffer head should have an attached journal head at this
3779 * point. However, if the data is corrupted and an indirect
3780 * block pointed to itself, it would have been detached when
3781 * the block was cleared. Check for this instead of OOPSing.
3783 if ((EXT4_JOURNAL(inode
) == NULL
) || bh2jh(this_bh
))
3784 ext4_handle_dirty_metadata(handle
, inode
, this_bh
);
3786 ext4_error(inode
->i_sb
, __func__
,
3787 "circular indirect block detected, "
3788 "inode=%lu, block=%llu",
3790 (unsigned long long) this_bh
->b_blocknr
);
3795 * ext4_free_branches - free an array of branches
3796 * @handle: JBD handle for this transaction
3797 * @inode: inode we are dealing with
3798 * @parent_bh: the buffer_head which contains *@first and *@last
3799 * @first: array of block numbers
3800 * @last: pointer immediately past the end of array
3801 * @depth: depth of the branches to free
3803 * We are freeing all blocks refered from these branches (numbers are
3804 * stored as little-endian 32-bit) and updating @inode->i_blocks
3807 static void ext4_free_branches(handle_t
*handle
, struct inode
*inode
,
3808 struct buffer_head
*parent_bh
,
3809 __le32
*first
, __le32
*last
, int depth
)
3814 if (ext4_handle_is_aborted(handle
))
3818 struct buffer_head
*bh
;
3819 int addr_per_block
= EXT4_ADDR_PER_BLOCK(inode
->i_sb
);
3821 while (--p
>= first
) {
3822 nr
= le32_to_cpu(*p
);
3824 continue; /* A hole */
3826 /* Go read the buffer for the next level down */
3827 bh
= sb_bread(inode
->i_sb
, nr
);
3830 * A read failure? Report error and clear slot
3834 ext4_error(inode
->i_sb
, "ext4_free_branches",
3835 "Read failure, inode=%lu, block=%llu",
3840 /* This zaps the entire block. Bottom up. */
3841 BUFFER_TRACE(bh
, "free child branches");
3842 ext4_free_branches(handle
, inode
, bh
,
3843 (__le32
*) bh
->b_data
,
3844 (__le32
*) bh
->b_data
+ addr_per_block
,
3848 * We've probably journalled the indirect block several
3849 * times during the truncate. But it's no longer
3850 * needed and we now drop it from the transaction via
3851 * jbd2_journal_revoke().
3853 * That's easy if it's exclusively part of this
3854 * transaction. But if it's part of the committing
3855 * transaction then jbd2_journal_forget() will simply
3856 * brelse() it. That means that if the underlying
3857 * block is reallocated in ext4_get_block(),
3858 * unmap_underlying_metadata() will find this block
3859 * and will try to get rid of it. damn, damn.
3861 * If this block has already been committed to the
3862 * journal, a revoke record will be written. And
3863 * revoke records must be emitted *before* clearing
3864 * this block's bit in the bitmaps.
3866 ext4_forget(handle
, 1, inode
, bh
, bh
->b_blocknr
);
3869 * Everything below this this pointer has been
3870 * released. Now let this top-of-subtree go.
3872 * We want the freeing of this indirect block to be
3873 * atomic in the journal with the updating of the
3874 * bitmap block which owns it. So make some room in
3877 * We zero the parent pointer *after* freeing its
3878 * pointee in the bitmaps, so if extend_transaction()
3879 * for some reason fails to put the bitmap changes and
3880 * the release into the same transaction, recovery
3881 * will merely complain about releasing a free block,
3882 * rather than leaking blocks.
3884 if (ext4_handle_is_aborted(handle
))
3886 if (try_to_extend_transaction(handle
, inode
)) {
3887 ext4_mark_inode_dirty(handle
, inode
);
3888 ext4_truncate_restart_trans(handle
, inode
,
3889 blocks_for_truncate(inode
));
3892 ext4_free_blocks(handle
, inode
, nr
, 1, 1);
3896 * The block which we have just freed is
3897 * pointed to by an indirect block: journal it
3899 BUFFER_TRACE(parent_bh
, "get_write_access");
3900 if (!ext4_journal_get_write_access(handle
,
3903 BUFFER_TRACE(parent_bh
,
3904 "call ext4_handle_dirty_metadata");
3905 ext4_handle_dirty_metadata(handle
,
3912 /* We have reached the bottom of the tree. */
3913 BUFFER_TRACE(parent_bh
, "free data blocks");
3914 ext4_free_data(handle
, inode
, parent_bh
, first
, last
);
3918 int ext4_can_truncate(struct inode
*inode
)
3920 if (IS_APPEND(inode
) || IS_IMMUTABLE(inode
))
3922 if (S_ISREG(inode
->i_mode
))
3924 if (S_ISDIR(inode
->i_mode
))
3926 if (S_ISLNK(inode
->i_mode
))
3927 return !ext4_inode_is_fast_symlink(inode
);
3934 * We block out ext4_get_block() block instantiations across the entire
3935 * transaction, and VFS/VM ensures that ext4_truncate() cannot run
3936 * simultaneously on behalf of the same inode.
3938 * As we work through the truncate and commmit bits of it to the journal there
3939 * is one core, guiding principle: the file's tree must always be consistent on
3940 * disk. We must be able to restart the truncate after a crash.
3942 * The file's tree may be transiently inconsistent in memory (although it
3943 * probably isn't), but whenever we close off and commit a journal transaction,
3944 * the contents of (the filesystem + the journal) must be consistent and
3945 * restartable. It's pretty simple, really: bottom up, right to left (although
3946 * left-to-right works OK too).
3948 * Note that at recovery time, journal replay occurs *before* the restart of
3949 * truncate against the orphan inode list.
3951 * The committed inode has the new, desired i_size (which is the same as
3952 * i_disksize in this case). After a crash, ext4_orphan_cleanup() will see
3953 * that this inode's truncate did not complete and it will again call
3954 * ext4_truncate() to have another go. So there will be instantiated blocks
3955 * to the right of the truncation point in a crashed ext4 filesystem. But
3956 * that's fine - as long as they are linked from the inode, the post-crash
3957 * ext4_truncate() run will find them and release them.
3959 void ext4_truncate(struct inode
*inode
)
3962 struct ext4_inode_info
*ei
= EXT4_I(inode
);
3963 __le32
*i_data
= ei
->i_data
;
3964 int addr_per_block
= EXT4_ADDR_PER_BLOCK(inode
->i_sb
);
3965 struct address_space
*mapping
= inode
->i_mapping
;
3966 ext4_lblk_t offsets
[4];
3971 ext4_lblk_t last_block
;
3972 unsigned blocksize
= inode
->i_sb
->s_blocksize
;
3974 if (!ext4_can_truncate(inode
))
3977 if (ei
->i_disksize
&& inode
->i_size
== 0 &&
3978 !test_opt(inode
->i_sb
, NO_AUTO_DA_ALLOC
))
3979 ei
->i_state
|= EXT4_STATE_DA_ALLOC_CLOSE
;
3981 if (EXT4_I(inode
)->i_flags
& EXT4_EXTENTS_FL
) {
3982 ext4_ext_truncate(inode
);
3986 handle
= start_transaction(inode
);
3988 return; /* AKPM: return what? */
3990 last_block
= (inode
->i_size
+ blocksize
-1)
3991 >> EXT4_BLOCK_SIZE_BITS(inode
->i_sb
);
3993 if (inode
->i_size
& (blocksize
- 1))
3994 if (ext4_block_truncate_page(handle
, mapping
, inode
->i_size
))
3997 n
= ext4_block_to_path(inode
, last_block
, offsets
, NULL
);
3999 goto out_stop
; /* error */
4002 * OK. This truncate is going to happen. We add the inode to the
4003 * orphan list, so that if this truncate spans multiple transactions,
4004 * and we crash, we will resume the truncate when the filesystem
4005 * recovers. It also marks the inode dirty, to catch the new size.
4007 * Implication: the file must always be in a sane, consistent
4008 * truncatable state while each transaction commits.
4010 if (ext4_orphan_add(handle
, inode
))
4014 * From here we block out all ext4_get_block() callers who want to
4015 * modify the block allocation tree.
4017 down_write(&ei
->i_data_sem
);
4019 ext4_discard_preallocations(inode
);
4022 * The orphan list entry will now protect us from any crash which
4023 * occurs before the truncate completes, so it is now safe to propagate
4024 * the new, shorter inode size (held for now in i_size) into the
4025 * on-disk inode. We do this via i_disksize, which is the value which
4026 * ext4 *really* writes onto the disk inode.
4028 ei
->i_disksize
= inode
->i_size
;
4030 if (n
== 1) { /* direct blocks */
4031 ext4_free_data(handle
, inode
, NULL
, i_data
+offsets
[0],
4032 i_data
+ EXT4_NDIR_BLOCKS
);
4036 partial
= ext4_find_shared(inode
, n
, offsets
, chain
, &nr
);
4037 /* Kill the top of shared branch (not detached) */
4039 if (partial
== chain
) {
4040 /* Shared branch grows from the inode */
4041 ext4_free_branches(handle
, inode
, NULL
,
4042 &nr
, &nr
+1, (chain
+n
-1) - partial
);
4045 * We mark the inode dirty prior to restart,
4046 * and prior to stop. No need for it here.
4049 /* Shared branch grows from an indirect block */
4050 BUFFER_TRACE(partial
->bh
, "get_write_access");
4051 ext4_free_branches(handle
, inode
, partial
->bh
,
4053 partial
->p
+1, (chain
+n
-1) - partial
);
4056 /* Clear the ends of indirect blocks on the shared branch */
4057 while (partial
> chain
) {
4058 ext4_free_branches(handle
, inode
, partial
->bh
, partial
->p
+ 1,
4059 (__le32
*)partial
->bh
->b_data
+addr_per_block
,
4060 (chain
+n
-1) - partial
);
4061 BUFFER_TRACE(partial
->bh
, "call brelse");
4062 brelse(partial
->bh
);
4066 /* Kill the remaining (whole) subtrees */
4067 switch (offsets
[0]) {
4069 nr
= i_data
[EXT4_IND_BLOCK
];
4071 ext4_free_branches(handle
, inode
, NULL
, &nr
, &nr
+1, 1);
4072 i_data
[EXT4_IND_BLOCK
] = 0;
4074 case EXT4_IND_BLOCK
:
4075 nr
= i_data
[EXT4_DIND_BLOCK
];
4077 ext4_free_branches(handle
, inode
, NULL
, &nr
, &nr
+1, 2);
4078 i_data
[EXT4_DIND_BLOCK
] = 0;
4080 case EXT4_DIND_BLOCK
:
4081 nr
= i_data
[EXT4_TIND_BLOCK
];
4083 ext4_free_branches(handle
, inode
, NULL
, &nr
, &nr
+1, 3);
4084 i_data
[EXT4_TIND_BLOCK
] = 0;
4086 case EXT4_TIND_BLOCK
:
4090 up_write(&ei
->i_data_sem
);
4091 inode
->i_mtime
= inode
->i_ctime
= ext4_current_time(inode
);
4092 ext4_mark_inode_dirty(handle
, inode
);
4095 * In a multi-transaction truncate, we only make the final transaction
4099 ext4_handle_sync(handle
);
4102 * If this was a simple ftruncate(), and the file will remain alive
4103 * then we need to clear up the orphan record which we created above.
4104 * However, if this was a real unlink then we were called by
4105 * ext4_delete_inode(), and we allow that function to clean up the
4106 * orphan info for us.
4109 ext4_orphan_del(handle
, inode
);
4111 ext4_journal_stop(handle
);
4115 * ext4_get_inode_loc returns with an extra refcount against the inode's
4116 * underlying buffer_head on success. If 'in_mem' is true, we have all
4117 * data in memory that is needed to recreate the on-disk version of this
4120 static int __ext4_get_inode_loc(struct inode
*inode
,
4121 struct ext4_iloc
*iloc
, int in_mem
)
4123 struct ext4_group_desc
*gdp
;
4124 struct buffer_head
*bh
;
4125 struct super_block
*sb
= inode
->i_sb
;
4127 int inodes_per_block
, inode_offset
;
4130 if (!ext4_valid_inum(sb
, inode
->i_ino
))
4133 iloc
->block_group
= (inode
->i_ino
- 1) / EXT4_INODES_PER_GROUP(sb
);
4134 gdp
= ext4_get_group_desc(sb
, iloc
->block_group
, NULL
);
4139 * Figure out the offset within the block group inode table
4141 inodes_per_block
= (EXT4_BLOCK_SIZE(sb
) / EXT4_INODE_SIZE(sb
));
4142 inode_offset
= ((inode
->i_ino
- 1) %
4143 EXT4_INODES_PER_GROUP(sb
));
4144 block
= ext4_inode_table(sb
, gdp
) + (inode_offset
/ inodes_per_block
);
4145 iloc
->offset
= (inode_offset
% inodes_per_block
) * EXT4_INODE_SIZE(sb
);
4147 bh
= sb_getblk(sb
, block
);
4149 ext4_error(sb
, "ext4_get_inode_loc", "unable to read "
4150 "inode block - inode=%lu, block=%llu",
4151 inode
->i_ino
, block
);
4154 if (!buffer_uptodate(bh
)) {
4158 * If the buffer has the write error flag, we have failed
4159 * to write out another inode in the same block. In this
4160 * case, we don't have to read the block because we may
4161 * read the old inode data successfully.
4163 if (buffer_write_io_error(bh
) && !buffer_uptodate(bh
))
4164 set_buffer_uptodate(bh
);
4166 if (buffer_uptodate(bh
)) {
4167 /* someone brought it uptodate while we waited */
4173 * If we have all information of the inode in memory and this
4174 * is the only valid inode in the block, we need not read the
4178 struct buffer_head
*bitmap_bh
;
4181 start
= inode_offset
& ~(inodes_per_block
- 1);
4183 /* Is the inode bitmap in cache? */
4184 bitmap_bh
= sb_getblk(sb
, ext4_inode_bitmap(sb
, gdp
));
4189 * If the inode bitmap isn't in cache then the
4190 * optimisation may end up performing two reads instead
4191 * of one, so skip it.
4193 if (!buffer_uptodate(bitmap_bh
)) {
4197 for (i
= start
; i
< start
+ inodes_per_block
; i
++) {
4198 if (i
== inode_offset
)
4200 if (ext4_test_bit(i
, bitmap_bh
->b_data
))
4204 if (i
== start
+ inodes_per_block
) {
4205 /* all other inodes are free, so skip I/O */
4206 memset(bh
->b_data
, 0, bh
->b_size
);
4207 set_buffer_uptodate(bh
);
4215 * If we need to do any I/O, try to pre-readahead extra
4216 * blocks from the inode table.
4218 if (EXT4_SB(sb
)->s_inode_readahead_blks
) {
4219 ext4_fsblk_t b
, end
, table
;
4222 table
= ext4_inode_table(sb
, gdp
);
4223 /* s_inode_readahead_blks is always a power of 2 */
4224 b
= block
& ~(EXT4_SB(sb
)->s_inode_readahead_blks
-1);
4227 end
= b
+ EXT4_SB(sb
)->s_inode_readahead_blks
;
4228 num
= EXT4_INODES_PER_GROUP(sb
);
4229 if (EXT4_HAS_RO_COMPAT_FEATURE(sb
,
4230 EXT4_FEATURE_RO_COMPAT_GDT_CSUM
))
4231 num
-= ext4_itable_unused_count(sb
, gdp
);
4232 table
+= num
/ inodes_per_block
;
4236 sb_breadahead(sb
, b
++);
4240 * There are other valid inodes in the buffer, this inode
4241 * has in-inode xattrs, or we don't have this inode in memory.
4242 * Read the block from disk.
4245 bh
->b_end_io
= end_buffer_read_sync
;
4246 submit_bh(READ_META
, bh
);
4248 if (!buffer_uptodate(bh
)) {
4249 ext4_error(sb
, __func__
,
4250 "unable to read inode block - inode=%lu, "
4251 "block=%llu", inode
->i_ino
, block
);
4261 int ext4_get_inode_loc(struct inode
*inode
, struct ext4_iloc
*iloc
)
4263 /* We have all inode data except xattrs in memory here. */
4264 return __ext4_get_inode_loc(inode
, iloc
,
4265 !(EXT4_I(inode
)->i_state
& EXT4_STATE_XATTR
));
4268 void ext4_set_inode_flags(struct inode
*inode
)
4270 unsigned int flags
= EXT4_I(inode
)->i_flags
;
4272 inode
->i_flags
&= ~(S_SYNC
|S_APPEND
|S_IMMUTABLE
|S_NOATIME
|S_DIRSYNC
);
4273 if (flags
& EXT4_SYNC_FL
)
4274 inode
->i_flags
|= S_SYNC
;
4275 if (flags
& EXT4_APPEND_FL
)
4276 inode
->i_flags
|= S_APPEND
;
4277 if (flags
& EXT4_IMMUTABLE_FL
)
4278 inode
->i_flags
|= S_IMMUTABLE
;
4279 if (flags
& EXT4_NOATIME_FL
)
4280 inode
->i_flags
|= S_NOATIME
;
4281 if (flags
& EXT4_DIRSYNC_FL
)
4282 inode
->i_flags
|= S_DIRSYNC
;
4285 /* Propagate flags from i_flags to EXT4_I(inode)->i_flags */
4286 void ext4_get_inode_flags(struct ext4_inode_info
*ei
)
4288 unsigned int flags
= ei
->vfs_inode
.i_flags
;
4290 ei
->i_flags
&= ~(EXT4_SYNC_FL
|EXT4_APPEND_FL
|
4291 EXT4_IMMUTABLE_FL
|EXT4_NOATIME_FL
|EXT4_DIRSYNC_FL
);
4293 ei
->i_flags
|= EXT4_SYNC_FL
;
4294 if (flags
& S_APPEND
)
4295 ei
->i_flags
|= EXT4_APPEND_FL
;
4296 if (flags
& S_IMMUTABLE
)
4297 ei
->i_flags
|= EXT4_IMMUTABLE_FL
;
4298 if (flags
& S_NOATIME
)
4299 ei
->i_flags
|= EXT4_NOATIME_FL
;
4300 if (flags
& S_DIRSYNC
)
4301 ei
->i_flags
|= EXT4_DIRSYNC_FL
;
4304 static blkcnt_t
ext4_inode_blocks(struct ext4_inode
*raw_inode
,
4305 struct ext4_inode_info
*ei
)
4308 struct inode
*inode
= &(ei
->vfs_inode
);
4309 struct super_block
*sb
= inode
->i_sb
;
4311 if (EXT4_HAS_RO_COMPAT_FEATURE(sb
,
4312 EXT4_FEATURE_RO_COMPAT_HUGE_FILE
)) {
4313 /* we are using combined 48 bit field */
4314 i_blocks
= ((u64
)le16_to_cpu(raw_inode
->i_blocks_high
)) << 32 |
4315 le32_to_cpu(raw_inode
->i_blocks_lo
);
4316 if (ei
->i_flags
& EXT4_HUGE_FILE_FL
) {
4317 /* i_blocks represent file system block size */
4318 return i_blocks
<< (inode
->i_blkbits
- 9);
4323 return le32_to_cpu(raw_inode
->i_blocks_lo
);
4327 struct inode
*ext4_iget(struct super_block
*sb
, unsigned long ino
)
4329 struct ext4_iloc iloc
;
4330 struct ext4_inode
*raw_inode
;
4331 struct ext4_inode_info
*ei
;
4332 struct buffer_head
*bh
;
4333 struct inode
*inode
;
4337 inode
= iget_locked(sb
, ino
);
4339 return ERR_PTR(-ENOMEM
);
4340 if (!(inode
->i_state
& I_NEW
))
4345 ret
= __ext4_get_inode_loc(inode
, &iloc
, 0);
4349 raw_inode
= ext4_raw_inode(&iloc
);
4350 inode
->i_mode
= le16_to_cpu(raw_inode
->i_mode
);
4351 inode
->i_uid
= (uid_t
)le16_to_cpu(raw_inode
->i_uid_low
);
4352 inode
->i_gid
= (gid_t
)le16_to_cpu(raw_inode
->i_gid_low
);
4353 if (!(test_opt(inode
->i_sb
, NO_UID32
))) {
4354 inode
->i_uid
|= le16_to_cpu(raw_inode
->i_uid_high
) << 16;
4355 inode
->i_gid
|= le16_to_cpu(raw_inode
->i_gid_high
) << 16;
4357 inode
->i_nlink
= le16_to_cpu(raw_inode
->i_links_count
);
4360 ei
->i_dir_start_lookup
= 0;
4361 ei
->i_dtime
= le32_to_cpu(raw_inode
->i_dtime
);
4362 /* We now have enough fields to check if the inode was active or not.
4363 * This is needed because nfsd might try to access dead inodes
4364 * the test is that same one that e2fsck uses
4365 * NeilBrown 1999oct15
4367 if (inode
->i_nlink
== 0) {
4368 if (inode
->i_mode
== 0 ||
4369 !(EXT4_SB(inode
->i_sb
)->s_mount_state
& EXT4_ORPHAN_FS
)) {
4370 /* this inode is deleted */
4375 /* The only unlinked inodes we let through here have
4376 * valid i_mode and are being read by the orphan
4377 * recovery code: that's fine, we're about to complete
4378 * the process of deleting those. */
4380 ei
->i_flags
= le32_to_cpu(raw_inode
->i_flags
);
4381 inode
->i_blocks
= ext4_inode_blocks(raw_inode
, ei
);
4382 ei
->i_file_acl
= le32_to_cpu(raw_inode
->i_file_acl_lo
);
4383 if (EXT4_HAS_INCOMPAT_FEATURE(sb
, EXT4_FEATURE_INCOMPAT_64BIT
))
4385 ((__u64
)le16_to_cpu(raw_inode
->i_file_acl_high
)) << 32;
4386 inode
->i_size
= ext4_isize(raw_inode
);
4387 ei
->i_disksize
= inode
->i_size
;
4388 inode
->i_generation
= le32_to_cpu(raw_inode
->i_generation
);
4389 ei
->i_block_group
= iloc
.block_group
;
4390 ei
->i_last_alloc_group
= ~0;
4392 * NOTE! The in-memory inode i_data array is in little-endian order
4393 * even on big-endian machines: we do NOT byteswap the block numbers!
4395 for (block
= 0; block
< EXT4_N_BLOCKS
; block
++)
4396 ei
->i_data
[block
] = raw_inode
->i_block
[block
];
4397 INIT_LIST_HEAD(&ei
->i_orphan
);
4399 if (EXT4_INODE_SIZE(inode
->i_sb
) > EXT4_GOOD_OLD_INODE_SIZE
) {
4400 ei
->i_extra_isize
= le16_to_cpu(raw_inode
->i_extra_isize
);
4401 if (EXT4_GOOD_OLD_INODE_SIZE
+ ei
->i_extra_isize
>
4402 EXT4_INODE_SIZE(inode
->i_sb
)) {
4407 if (ei
->i_extra_isize
== 0) {
4408 /* The extra space is currently unused. Use it. */
4409 ei
->i_extra_isize
= sizeof(struct ext4_inode
) -
4410 EXT4_GOOD_OLD_INODE_SIZE
;
4412 __le32
*magic
= (void *)raw_inode
+
4413 EXT4_GOOD_OLD_INODE_SIZE
+
4415 if (*magic
== cpu_to_le32(EXT4_XATTR_MAGIC
))
4416 ei
->i_state
|= EXT4_STATE_XATTR
;
4419 ei
->i_extra_isize
= 0;
4421 EXT4_INODE_GET_XTIME(i_ctime
, inode
, raw_inode
);
4422 EXT4_INODE_GET_XTIME(i_mtime
, inode
, raw_inode
);
4423 EXT4_INODE_GET_XTIME(i_atime
, inode
, raw_inode
);
4424 EXT4_EINODE_GET_XTIME(i_crtime
, ei
, raw_inode
);
4426 inode
->i_version
= le32_to_cpu(raw_inode
->i_disk_version
);
4427 if (EXT4_INODE_SIZE(inode
->i_sb
) > EXT4_GOOD_OLD_INODE_SIZE
) {
4428 if (EXT4_FITS_IN_INODE(raw_inode
, ei
, i_version_hi
))
4430 (__u64
)(le32_to_cpu(raw_inode
->i_version_hi
)) << 32;
4434 if (ei
->i_file_acl
&&
4436 (le32_to_cpu(EXT4_SB(sb
)->s_es
->s_first_data_block
) +
4437 EXT4_SB(sb
)->s_gdb_count
)) ||
4438 (ei
->i_file_acl
>= ext4_blocks_count(EXT4_SB(sb
)->s_es
)))) {
4439 ext4_error(sb
, __func__
,
4440 "bad extended attribute block %llu in inode #%lu",
4441 ei
->i_file_acl
, inode
->i_ino
);
4444 } else if (ei
->i_flags
& EXT4_EXTENTS_FL
) {
4445 if (S_ISREG(inode
->i_mode
) || S_ISDIR(inode
->i_mode
) ||
4446 (S_ISLNK(inode
->i_mode
) &&
4447 !ext4_inode_is_fast_symlink(inode
)))
4448 /* Validate extent which is part of inode */
4449 ret
= ext4_ext_check_inode(inode
);
4450 } else if (S_ISREG(inode
->i_mode
) || S_ISDIR(inode
->i_mode
) ||
4451 (S_ISLNK(inode
->i_mode
) &&
4452 !ext4_inode_is_fast_symlink(inode
))) {
4453 /* Validate block references which are part of inode */
4454 ret
= ext4_check_inode_blockref(inode
);
4461 if (S_ISREG(inode
->i_mode
)) {
4462 inode
->i_op
= &ext4_file_inode_operations
;
4463 inode
->i_fop
= &ext4_file_operations
;
4464 ext4_set_aops(inode
);
4465 } else if (S_ISDIR(inode
->i_mode
)) {
4466 inode
->i_op
= &ext4_dir_inode_operations
;
4467 inode
->i_fop
= &ext4_dir_operations
;
4468 } else if (S_ISLNK(inode
->i_mode
)) {
4469 if (ext4_inode_is_fast_symlink(inode
)) {
4470 inode
->i_op
= &ext4_fast_symlink_inode_operations
;
4471 nd_terminate_link(ei
->i_data
, inode
->i_size
,
4472 sizeof(ei
->i_data
) - 1);
4474 inode
->i_op
= &ext4_symlink_inode_operations
;
4475 ext4_set_aops(inode
);
4477 } else if (S_ISCHR(inode
->i_mode
) || S_ISBLK(inode
->i_mode
) ||
4478 S_ISFIFO(inode
->i_mode
) || S_ISSOCK(inode
->i_mode
)) {
4479 inode
->i_op
= &ext4_special_inode_operations
;
4480 if (raw_inode
->i_block
[0])
4481 init_special_inode(inode
, inode
->i_mode
,
4482 old_decode_dev(le32_to_cpu(raw_inode
->i_block
[0])));
4484 init_special_inode(inode
, inode
->i_mode
,
4485 new_decode_dev(le32_to_cpu(raw_inode
->i_block
[1])));
4489 ext4_error(inode
->i_sb
, __func__
,
4490 "bogus i_mode (%o) for inode=%lu",
4491 inode
->i_mode
, inode
->i_ino
);
4495 ext4_set_inode_flags(inode
);
4496 unlock_new_inode(inode
);
4501 return ERR_PTR(ret
);
4504 static int ext4_inode_blocks_set(handle_t
*handle
,
4505 struct ext4_inode
*raw_inode
,
4506 struct ext4_inode_info
*ei
)
4508 struct inode
*inode
= &(ei
->vfs_inode
);
4509 u64 i_blocks
= inode
->i_blocks
;
4510 struct super_block
*sb
= inode
->i_sb
;
4512 if (i_blocks
<= ~0U) {
4514 * i_blocks can be represnted in a 32 bit variable
4515 * as multiple of 512 bytes
4517 raw_inode
->i_blocks_lo
= cpu_to_le32(i_blocks
);
4518 raw_inode
->i_blocks_high
= 0;
4519 ei
->i_flags
&= ~EXT4_HUGE_FILE_FL
;
4522 if (!EXT4_HAS_RO_COMPAT_FEATURE(sb
, EXT4_FEATURE_RO_COMPAT_HUGE_FILE
))
4525 if (i_blocks
<= 0xffffffffffffULL
) {
4527 * i_blocks can be represented in a 48 bit variable
4528 * as multiple of 512 bytes
4530 raw_inode
->i_blocks_lo
= cpu_to_le32(i_blocks
);
4531 raw_inode
->i_blocks_high
= cpu_to_le16(i_blocks
>> 32);
4532 ei
->i_flags
&= ~EXT4_HUGE_FILE_FL
;
4534 ei
->i_flags
|= EXT4_HUGE_FILE_FL
;
4535 /* i_block is stored in file system block size */
4536 i_blocks
= i_blocks
>> (inode
->i_blkbits
- 9);
4537 raw_inode
->i_blocks_lo
= cpu_to_le32(i_blocks
);
4538 raw_inode
->i_blocks_high
= cpu_to_le16(i_blocks
>> 32);
4544 * Post the struct inode info into an on-disk inode location in the
4545 * buffer-cache. This gobbles the caller's reference to the
4546 * buffer_head in the inode location struct.
4548 * The caller must have write access to iloc->bh.
4550 static int ext4_do_update_inode(handle_t
*handle
,
4551 struct inode
*inode
,
4552 struct ext4_iloc
*iloc
)
4554 struct ext4_inode
*raw_inode
= ext4_raw_inode(iloc
);
4555 struct ext4_inode_info
*ei
= EXT4_I(inode
);
4556 struct buffer_head
*bh
= iloc
->bh
;
4557 int err
= 0, rc
, block
;
4559 /* For fields not not tracking in the in-memory inode,
4560 * initialise them to zero for new inodes. */
4561 if (ei
->i_state
& EXT4_STATE_NEW
)
4562 memset(raw_inode
, 0, EXT4_SB(inode
->i_sb
)->s_inode_size
);
4564 ext4_get_inode_flags(ei
);
4565 raw_inode
->i_mode
= cpu_to_le16(inode
->i_mode
);
4566 if (!(test_opt(inode
->i_sb
, NO_UID32
))) {
4567 raw_inode
->i_uid_low
= cpu_to_le16(low_16_bits(inode
->i_uid
));
4568 raw_inode
->i_gid_low
= cpu_to_le16(low_16_bits(inode
->i_gid
));
4570 * Fix up interoperability with old kernels. Otherwise, old inodes get
4571 * re-used with the upper 16 bits of the uid/gid intact
4574 raw_inode
->i_uid_high
=
4575 cpu_to_le16(high_16_bits(inode
->i_uid
));
4576 raw_inode
->i_gid_high
=
4577 cpu_to_le16(high_16_bits(inode
->i_gid
));
4579 raw_inode
->i_uid_high
= 0;
4580 raw_inode
->i_gid_high
= 0;
4583 raw_inode
->i_uid_low
=
4584 cpu_to_le16(fs_high2lowuid(inode
->i_uid
));
4585 raw_inode
->i_gid_low
=
4586 cpu_to_le16(fs_high2lowgid(inode
->i_gid
));
4587 raw_inode
->i_uid_high
= 0;
4588 raw_inode
->i_gid_high
= 0;
4590 raw_inode
->i_links_count
= cpu_to_le16(inode
->i_nlink
);
4592 EXT4_INODE_SET_XTIME(i_ctime
, inode
, raw_inode
);
4593 EXT4_INODE_SET_XTIME(i_mtime
, inode
, raw_inode
);
4594 EXT4_INODE_SET_XTIME(i_atime
, inode
, raw_inode
);
4595 EXT4_EINODE_SET_XTIME(i_crtime
, ei
, raw_inode
);
4597 if (ext4_inode_blocks_set(handle
, raw_inode
, ei
))
4599 raw_inode
->i_dtime
= cpu_to_le32(ei
->i_dtime
);
4600 /* clear the migrate flag in the raw_inode */
4601 raw_inode
->i_flags
= cpu_to_le32(ei
->i_flags
& ~EXT4_EXT_MIGRATE
);
4602 if (EXT4_SB(inode
->i_sb
)->s_es
->s_creator_os
!=
4603 cpu_to_le32(EXT4_OS_HURD
))
4604 raw_inode
->i_file_acl_high
=
4605 cpu_to_le16(ei
->i_file_acl
>> 32);
4606 raw_inode
->i_file_acl_lo
= cpu_to_le32(ei
->i_file_acl
);
4607 ext4_isize_set(raw_inode
, ei
->i_disksize
);
4608 if (ei
->i_disksize
> 0x7fffffffULL
) {
4609 struct super_block
*sb
= inode
->i_sb
;
4610 if (!EXT4_HAS_RO_COMPAT_FEATURE(sb
,
4611 EXT4_FEATURE_RO_COMPAT_LARGE_FILE
) ||
4612 EXT4_SB(sb
)->s_es
->s_rev_level
==
4613 cpu_to_le32(EXT4_GOOD_OLD_REV
)) {
4614 /* If this is the first large file
4615 * created, add a flag to the superblock.
4617 err
= ext4_journal_get_write_access(handle
,
4618 EXT4_SB(sb
)->s_sbh
);
4621 ext4_update_dynamic_rev(sb
);
4622 EXT4_SET_RO_COMPAT_FEATURE(sb
,
4623 EXT4_FEATURE_RO_COMPAT_LARGE_FILE
);
4625 ext4_handle_sync(handle
);
4626 err
= ext4_handle_dirty_metadata(handle
, inode
,
4627 EXT4_SB(sb
)->s_sbh
);
4630 raw_inode
->i_generation
= cpu_to_le32(inode
->i_generation
);
4631 if (S_ISCHR(inode
->i_mode
) || S_ISBLK(inode
->i_mode
)) {
4632 if (old_valid_dev(inode
->i_rdev
)) {
4633 raw_inode
->i_block
[0] =
4634 cpu_to_le32(old_encode_dev(inode
->i_rdev
));
4635 raw_inode
->i_block
[1] = 0;
4637 raw_inode
->i_block
[0] = 0;
4638 raw_inode
->i_block
[1] =
4639 cpu_to_le32(new_encode_dev(inode
->i_rdev
));
4640 raw_inode
->i_block
[2] = 0;
4643 for (block
= 0; block
< EXT4_N_BLOCKS
; block
++)
4644 raw_inode
->i_block
[block
] = ei
->i_data
[block
];
4646 raw_inode
->i_disk_version
= cpu_to_le32(inode
->i_version
);
4647 if (ei
->i_extra_isize
) {
4648 if (EXT4_FITS_IN_INODE(raw_inode
, ei
, i_version_hi
))
4649 raw_inode
->i_version_hi
=
4650 cpu_to_le32(inode
->i_version
>> 32);
4651 raw_inode
->i_extra_isize
= cpu_to_le16(ei
->i_extra_isize
);
4654 BUFFER_TRACE(bh
, "call ext4_handle_dirty_metadata");
4655 rc
= ext4_handle_dirty_metadata(handle
, inode
, bh
);
4658 ei
->i_state
&= ~EXT4_STATE_NEW
;
4662 ext4_std_error(inode
->i_sb
, err
);
4667 * ext4_write_inode()
4669 * We are called from a few places:
4671 * - Within generic_file_write() for O_SYNC files.
4672 * Here, there will be no transaction running. We wait for any running
4673 * trasnaction to commit.
4675 * - Within sys_sync(), kupdate and such.
4676 * We wait on commit, if tol to.
4678 * - Within prune_icache() (PF_MEMALLOC == true)
4679 * Here we simply return. We can't afford to block kswapd on the
4682 * In all cases it is actually safe for us to return without doing anything,
4683 * because the inode has been copied into a raw inode buffer in
4684 * ext4_mark_inode_dirty(). This is a correctness thing for O_SYNC and for
4687 * Note that we are absolutely dependent upon all inode dirtiers doing the
4688 * right thing: they *must* call mark_inode_dirty() after dirtying info in
4689 * which we are interested.
4691 * It would be a bug for them to not do this. The code:
4693 * mark_inode_dirty(inode)
4695 * inode->i_size = expr;
4697 * is in error because a kswapd-driven write_inode() could occur while
4698 * `stuff()' is running, and the new i_size will be lost. Plus the inode
4699 * will no longer be on the superblock's dirty inode list.
4701 int ext4_write_inode(struct inode
*inode
, int wait
)
4703 if (current
->flags
& PF_MEMALLOC
)
4706 if (ext4_journal_current_handle()) {
4707 jbd_debug(1, "called recursively, non-PF_MEMALLOC!\n");
4715 return ext4_force_commit(inode
->i_sb
);
4721 * Called from notify_change.
4723 * We want to trap VFS attempts to truncate the file as soon as
4724 * possible. In particular, we want to make sure that when the VFS
4725 * shrinks i_size, we put the inode on the orphan list and modify
4726 * i_disksize immediately, so that during the subsequent flushing of
4727 * dirty pages and freeing of disk blocks, we can guarantee that any
4728 * commit will leave the blocks being flushed in an unused state on
4729 * disk. (On recovery, the inode will get truncated and the blocks will
4730 * be freed, so we have a strong guarantee that no future commit will
4731 * leave these blocks visible to the user.)
4733 * Another thing we have to assure is that if we are in ordered mode
4734 * and inode is still attached to the committing transaction, we must
4735 * we start writeout of all the dirty pages which are being truncated.
4736 * This way we are sure that all the data written in the previous
4737 * transaction are already on disk (truncate waits for pages under
4740 * Called with inode->i_mutex down.
4742 int ext4_setattr(struct dentry
*dentry
, struct iattr
*attr
)
4744 struct inode
*inode
= dentry
->d_inode
;
4746 const unsigned int ia_valid
= attr
->ia_valid
;
4748 error
= inode_change_ok(inode
, attr
);
4752 if ((ia_valid
& ATTR_UID
&& attr
->ia_uid
!= inode
->i_uid
) ||
4753 (ia_valid
& ATTR_GID
&& attr
->ia_gid
!= inode
->i_gid
)) {
4756 /* (user+group)*(old+new) structure, inode write (sb,
4757 * inode block, ? - but truncate inode update has it) */
4758 handle
= ext4_journal_start(inode
, 2*(EXT4_QUOTA_INIT_BLOCKS(inode
->i_sb
)+
4759 EXT4_QUOTA_DEL_BLOCKS(inode
->i_sb
))+3);
4760 if (IS_ERR(handle
)) {
4761 error
= PTR_ERR(handle
);
4764 error
= vfs_dq_transfer(inode
, attr
) ? -EDQUOT
: 0;
4766 ext4_journal_stop(handle
);
4769 /* Update corresponding info in inode so that everything is in
4770 * one transaction */
4771 if (attr
->ia_valid
& ATTR_UID
)
4772 inode
->i_uid
= attr
->ia_uid
;
4773 if (attr
->ia_valid
& ATTR_GID
)
4774 inode
->i_gid
= attr
->ia_gid
;
4775 error
= ext4_mark_inode_dirty(handle
, inode
);
4776 ext4_journal_stop(handle
);
4779 if (attr
->ia_valid
& ATTR_SIZE
) {
4780 if (!(EXT4_I(inode
)->i_flags
& EXT4_EXTENTS_FL
)) {
4781 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
4783 if (attr
->ia_size
> sbi
->s_bitmap_maxbytes
) {
4790 if (S_ISREG(inode
->i_mode
) &&
4791 attr
->ia_valid
& ATTR_SIZE
&& attr
->ia_size
< inode
->i_size
) {
4794 handle
= ext4_journal_start(inode
, 3);
4795 if (IS_ERR(handle
)) {
4796 error
= PTR_ERR(handle
);
4800 error
= ext4_orphan_add(handle
, inode
);
4801 EXT4_I(inode
)->i_disksize
= attr
->ia_size
;
4802 rc
= ext4_mark_inode_dirty(handle
, inode
);
4805 ext4_journal_stop(handle
);
4807 if (ext4_should_order_data(inode
)) {
4808 error
= ext4_begin_ordered_truncate(inode
,
4811 /* Do as much error cleanup as possible */
4812 handle
= ext4_journal_start(inode
, 3);
4813 if (IS_ERR(handle
)) {
4814 ext4_orphan_del(NULL
, inode
);
4817 ext4_orphan_del(handle
, inode
);
4818 ext4_journal_stop(handle
);
4824 rc
= inode_setattr(inode
, attr
);
4826 /* If inode_setattr's call to ext4_truncate failed to get a
4827 * transaction handle at all, we need to clean up the in-core
4828 * orphan list manually. */
4830 ext4_orphan_del(NULL
, inode
);
4832 if (!rc
&& (ia_valid
& ATTR_MODE
))
4833 rc
= ext4_acl_chmod(inode
);
4836 ext4_std_error(inode
->i_sb
, error
);
4842 int ext4_getattr(struct vfsmount
*mnt
, struct dentry
*dentry
,
4845 struct inode
*inode
;
4846 unsigned long delalloc_blocks
;
4848 inode
= dentry
->d_inode
;
4849 generic_fillattr(inode
, stat
);
4852 * We can't update i_blocks if the block allocation is delayed
4853 * otherwise in the case of system crash before the real block
4854 * allocation is done, we will have i_blocks inconsistent with
4855 * on-disk file blocks.
4856 * We always keep i_blocks updated together with real
4857 * allocation. But to not confuse with user, stat
4858 * will return the blocks that include the delayed allocation
4859 * blocks for this file.
4861 spin_lock(&EXT4_I(inode
)->i_block_reservation_lock
);
4862 delalloc_blocks
= EXT4_I(inode
)->i_reserved_data_blocks
;
4863 spin_unlock(&EXT4_I(inode
)->i_block_reservation_lock
);
4865 stat
->blocks
+= (delalloc_blocks
<< inode
->i_sb
->s_blocksize_bits
)>>9;
4869 static int ext4_indirect_trans_blocks(struct inode
*inode
, int nrblocks
,
4874 /* if nrblocks are contiguous */
4877 * With N contiguous data blocks, it need at most
4878 * N/EXT4_ADDR_PER_BLOCK(inode->i_sb) indirect blocks
4879 * 2 dindirect blocks
4882 indirects
= nrblocks
/ EXT4_ADDR_PER_BLOCK(inode
->i_sb
);
4883 return indirects
+ 3;
4886 * if nrblocks are not contiguous, worse case, each block touch
4887 * a indirect block, and each indirect block touch a double indirect
4888 * block, plus a triple indirect block
4890 indirects
= nrblocks
* 2 + 1;
4894 static int ext4_index_trans_blocks(struct inode
*inode
, int nrblocks
, int chunk
)
4896 if (!(EXT4_I(inode
)->i_flags
& EXT4_EXTENTS_FL
))
4897 return ext4_indirect_trans_blocks(inode
, nrblocks
, chunk
);
4898 return ext4_ext_index_trans_blocks(inode
, nrblocks
, chunk
);
4902 * Account for index blocks, block groups bitmaps and block group
4903 * descriptor blocks if modify datablocks and index blocks
4904 * worse case, the indexs blocks spread over different block groups
4906 * If datablocks are discontiguous, they are possible to spread over
4907 * different block groups too. If they are contiugous, with flexbg,
4908 * they could still across block group boundary.
4910 * Also account for superblock, inode, quota and xattr blocks
4912 int ext4_meta_trans_blocks(struct inode
*inode
, int nrblocks
, int chunk
)
4914 ext4_group_t groups
, ngroups
= ext4_get_groups_count(inode
->i_sb
);
4920 * How many index blocks need to touch to modify nrblocks?
4921 * The "Chunk" flag indicating whether the nrblocks is
4922 * physically contiguous on disk
4924 * For Direct IO and fallocate, they calls get_block to allocate
4925 * one single extent at a time, so they could set the "Chunk" flag
4927 idxblocks
= ext4_index_trans_blocks(inode
, nrblocks
, chunk
);
4932 * Now let's see how many group bitmaps and group descriptors need
4942 if (groups
> ngroups
)
4944 if (groups
> EXT4_SB(inode
->i_sb
)->s_gdb_count
)
4945 gdpblocks
= EXT4_SB(inode
->i_sb
)->s_gdb_count
;
4947 /* bitmaps and block group descriptor blocks */
4948 ret
+= groups
+ gdpblocks
;
4950 /* Blocks for super block, inode, quota and xattr blocks */
4951 ret
+= EXT4_META_TRANS_BLOCKS(inode
->i_sb
);
4957 * Calulate the total number of credits to reserve to fit
4958 * the modification of a single pages into a single transaction,
4959 * which may include multiple chunks of block allocations.
4961 * This could be called via ext4_write_begin()
4963 * We need to consider the worse case, when
4964 * one new block per extent.
4966 int ext4_writepage_trans_blocks(struct inode
*inode
)
4968 int bpp
= ext4_journal_blocks_per_page(inode
);
4971 ret
= ext4_meta_trans_blocks(inode
, bpp
, 0);
4973 /* Account for data blocks for journalled mode */
4974 if (ext4_should_journal_data(inode
))
4980 * Calculate the journal credits for a chunk of data modification.
4982 * This is called from DIO, fallocate or whoever calling
4983 * ext4_get_blocks() to map/allocate a chunk of contigous disk blocks.
4985 * journal buffers for data blocks are not included here, as DIO
4986 * and fallocate do no need to journal data buffers.
4988 int ext4_chunk_trans_blocks(struct inode
*inode
, int nrblocks
)
4990 return ext4_meta_trans_blocks(inode
, nrblocks
, 1);
4994 * The caller must have previously called ext4_reserve_inode_write().
4995 * Give this, we know that the caller already has write access to iloc->bh.
4997 int ext4_mark_iloc_dirty(handle_t
*handle
,
4998 struct inode
*inode
, struct ext4_iloc
*iloc
)
5002 if (test_opt(inode
->i_sb
, I_VERSION
))
5003 inode_inc_iversion(inode
);
5005 /* the do_update_inode consumes one bh->b_count */
5008 /* ext4_do_update_inode() does jbd2_journal_dirty_metadata */
5009 err
= ext4_do_update_inode(handle
, inode
, iloc
);
5015 * On success, We end up with an outstanding reference count against
5016 * iloc->bh. This _must_ be cleaned up later.
5020 ext4_reserve_inode_write(handle_t
*handle
, struct inode
*inode
,
5021 struct ext4_iloc
*iloc
)
5025 err
= ext4_get_inode_loc(inode
, iloc
);
5027 BUFFER_TRACE(iloc
->bh
, "get_write_access");
5028 err
= ext4_journal_get_write_access(handle
, iloc
->bh
);
5034 ext4_std_error(inode
->i_sb
, err
);
5039 * Expand an inode by new_extra_isize bytes.
5040 * Returns 0 on success or negative error number on failure.
5042 static int ext4_expand_extra_isize(struct inode
*inode
,
5043 unsigned int new_extra_isize
,
5044 struct ext4_iloc iloc
,
5047 struct ext4_inode
*raw_inode
;
5048 struct ext4_xattr_ibody_header
*header
;
5049 struct ext4_xattr_entry
*entry
;
5051 if (EXT4_I(inode
)->i_extra_isize
>= new_extra_isize
)
5054 raw_inode
= ext4_raw_inode(&iloc
);
5056 header
= IHDR(inode
, raw_inode
);
5057 entry
= IFIRST(header
);
5059 /* No extended attributes present */
5060 if (!(EXT4_I(inode
)->i_state
& EXT4_STATE_XATTR
) ||
5061 header
->h_magic
!= cpu_to_le32(EXT4_XATTR_MAGIC
)) {
5062 memset((void *)raw_inode
+ EXT4_GOOD_OLD_INODE_SIZE
, 0,
5064 EXT4_I(inode
)->i_extra_isize
= new_extra_isize
;
5068 /* try to expand with EAs present */
5069 return ext4_expand_extra_isize_ea(inode
, new_extra_isize
,
5074 * What we do here is to mark the in-core inode as clean with respect to inode
5075 * dirtiness (it may still be data-dirty).
5076 * This means that the in-core inode may be reaped by prune_icache
5077 * without having to perform any I/O. This is a very good thing,
5078 * because *any* task may call prune_icache - even ones which
5079 * have a transaction open against a different journal.
5081 * Is this cheating? Not really. Sure, we haven't written the
5082 * inode out, but prune_icache isn't a user-visible syncing function.
5083 * Whenever the user wants stuff synced (sys_sync, sys_msync, sys_fsync)
5084 * we start and wait on commits.
5086 * Is this efficient/effective? Well, we're being nice to the system
5087 * by cleaning up our inodes proactively so they can be reaped
5088 * without I/O. But we are potentially leaving up to five seconds'
5089 * worth of inodes floating about which prune_icache wants us to
5090 * write out. One way to fix that would be to get prune_icache()
5091 * to do a write_super() to free up some memory. It has the desired
5094 int ext4_mark_inode_dirty(handle_t
*handle
, struct inode
*inode
)
5096 struct ext4_iloc iloc
;
5097 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
5098 static unsigned int mnt_count
;
5102 err
= ext4_reserve_inode_write(handle
, inode
, &iloc
);
5103 if (ext4_handle_valid(handle
) &&
5104 EXT4_I(inode
)->i_extra_isize
< sbi
->s_want_extra_isize
&&
5105 !(EXT4_I(inode
)->i_state
& EXT4_STATE_NO_EXPAND
)) {
5107 * We need extra buffer credits since we may write into EA block
5108 * with this same handle. If journal_extend fails, then it will
5109 * only result in a minor loss of functionality for that inode.
5110 * If this is felt to be critical, then e2fsck should be run to
5111 * force a large enough s_min_extra_isize.
5113 if ((jbd2_journal_extend(handle
,
5114 EXT4_DATA_TRANS_BLOCKS(inode
->i_sb
))) == 0) {
5115 ret
= ext4_expand_extra_isize(inode
,
5116 sbi
->s_want_extra_isize
,
5119 EXT4_I(inode
)->i_state
|= EXT4_STATE_NO_EXPAND
;
5121 le16_to_cpu(sbi
->s_es
->s_mnt_count
)) {
5122 ext4_warning(inode
->i_sb
, __func__
,
5123 "Unable to expand inode %lu. Delete"
5124 " some EAs or run e2fsck.",
5127 le16_to_cpu(sbi
->s_es
->s_mnt_count
);
5133 err
= ext4_mark_iloc_dirty(handle
, inode
, &iloc
);
5138 * ext4_dirty_inode() is called from __mark_inode_dirty()
5140 * We're really interested in the case where a file is being extended.
5141 * i_size has been changed by generic_commit_write() and we thus need
5142 * to include the updated inode in the current transaction.
5144 * Also, vfs_dq_alloc_block() will always dirty the inode when blocks
5145 * are allocated to the file.
5147 * If the inode is marked synchronous, we don't honour that here - doing
5148 * so would cause a commit on atime updates, which we don't bother doing.
5149 * We handle synchronous inodes at the highest possible level.
5151 void ext4_dirty_inode(struct inode
*inode
)
5153 handle_t
*current_handle
= ext4_journal_current_handle();
5156 if (!ext4_handle_valid(current_handle
)) {
5157 ext4_mark_inode_dirty(current_handle
, inode
);
5161 handle
= ext4_journal_start(inode
, 2);
5164 if (current_handle
&&
5165 current_handle
->h_transaction
!= handle
->h_transaction
) {
5166 /* This task has a transaction open against a different fs */
5167 printk(KERN_EMERG
"%s: transactions do not match!\n",
5170 jbd_debug(5, "marking dirty. outer handle=%p\n",
5172 ext4_mark_inode_dirty(handle
, inode
);
5174 ext4_journal_stop(handle
);
5181 * Bind an inode's backing buffer_head into this transaction, to prevent
5182 * it from being flushed to disk early. Unlike
5183 * ext4_reserve_inode_write, this leaves behind no bh reference and
5184 * returns no iloc structure, so the caller needs to repeat the iloc
5185 * lookup to mark the inode dirty later.
5187 static int ext4_pin_inode(handle_t
*handle
, struct inode
*inode
)
5189 struct ext4_iloc iloc
;
5193 err
= ext4_get_inode_loc(inode
, &iloc
);
5195 BUFFER_TRACE(iloc
.bh
, "get_write_access");
5196 err
= jbd2_journal_get_write_access(handle
, iloc
.bh
);
5198 err
= ext4_handle_dirty_metadata(handle
,
5204 ext4_std_error(inode
->i_sb
, err
);
5209 int ext4_change_inode_journal_flag(struct inode
*inode
, int val
)
5216 * We have to be very careful here: changing a data block's
5217 * journaling status dynamically is dangerous. If we write a
5218 * data block to the journal, change the status and then delete
5219 * that block, we risk forgetting to revoke the old log record
5220 * from the journal and so a subsequent replay can corrupt data.
5221 * So, first we make sure that the journal is empty and that
5222 * nobody is changing anything.
5225 journal
= EXT4_JOURNAL(inode
);
5228 if (is_journal_aborted(journal
))
5231 jbd2_journal_lock_updates(journal
);
5232 jbd2_journal_flush(journal
);
5235 * OK, there are no updates running now, and all cached data is
5236 * synced to disk. We are now in a completely consistent state
5237 * which doesn't have anything in the journal, and we know that
5238 * no filesystem updates are running, so it is safe to modify
5239 * the inode's in-core data-journaling state flag now.
5243 EXT4_I(inode
)->i_flags
|= EXT4_JOURNAL_DATA_FL
;
5245 EXT4_I(inode
)->i_flags
&= ~EXT4_JOURNAL_DATA_FL
;
5246 ext4_set_aops(inode
);
5248 jbd2_journal_unlock_updates(journal
);
5250 /* Finally we can mark the inode as dirty. */
5252 handle
= ext4_journal_start(inode
, 1);
5254 return PTR_ERR(handle
);
5256 err
= ext4_mark_inode_dirty(handle
, inode
);
5257 ext4_handle_sync(handle
);
5258 ext4_journal_stop(handle
);
5259 ext4_std_error(inode
->i_sb
, err
);
5264 static int ext4_bh_unmapped(handle_t
*handle
, struct buffer_head
*bh
)
5266 return !buffer_mapped(bh
);
5269 int ext4_page_mkwrite(struct vm_area_struct
*vma
, struct vm_fault
*vmf
)
5271 struct page
*page
= vmf
->page
;
5276 struct file
*file
= vma
->vm_file
;
5277 struct inode
*inode
= file
->f_path
.dentry
->d_inode
;
5278 struct address_space
*mapping
= inode
->i_mapping
;
5281 * Get i_alloc_sem to stop truncates messing with the inode. We cannot
5282 * get i_mutex because we are already holding mmap_sem.
5284 down_read(&inode
->i_alloc_sem
);
5285 size
= i_size_read(inode
);
5286 if (page
->mapping
!= mapping
|| size
<= page_offset(page
)
5287 || !PageUptodate(page
)) {
5288 /* page got truncated from under us? */
5292 if (PageMappedToDisk(page
))
5295 if (page
->index
== size
>> PAGE_CACHE_SHIFT
)
5296 len
= size
& ~PAGE_CACHE_MASK
;
5298 len
= PAGE_CACHE_SIZE
;
5300 if (page_has_buffers(page
)) {
5301 /* return if we have all the buffers mapped */
5302 if (!walk_page_buffers(NULL
, page_buffers(page
), 0, len
, NULL
,
5307 * OK, we need to fill the hole... Do write_begin write_end
5308 * to do block allocation/reservation.We are not holding
5309 * inode.i__mutex here. That allow * parallel write_begin,
5310 * write_end call. lock_page prevent this from happening
5311 * on the same page though
5313 ret
= mapping
->a_ops
->write_begin(file
, mapping
, page_offset(page
),
5314 len
, AOP_FLAG_UNINTERRUPTIBLE
, &page
, &fsdata
);
5317 ret
= mapping
->a_ops
->write_end(file
, mapping
, page_offset(page
),
5318 len
, len
, page
, fsdata
);
5324 ret
= VM_FAULT_SIGBUS
;
5325 up_read(&inode
->i_alloc_sem
);