2 * linux/fs/ext4/inode.c
4 * Copyright (C) 1992, 1993, 1994, 1995
5 * Remy Card (card@masi.ibp.fr)
6 * Laboratoire MASI - Institut Blaise Pascal
7 * Universite Pierre et Marie Curie (Paris VI)
11 * linux/fs/minix/inode.c
13 * Copyright (C) 1991, 1992 Linus Torvalds
15 * Goal-directed block allocation by Stephen Tweedie
16 * (sct@redhat.com), 1993, 1998
17 * Big-endian to little-endian byte-swapping/bitmaps by
18 * David S. Miller (davem@caip.rutgers.edu), 1995
19 * 64-bit file support on 64-bit platforms by Jakub Jelinek
20 * (jj@sunsite.ms.mff.cuni.cz)
22 * Assorted race fixes, rewrite of ext4_get_block() by Al Viro, 2000
25 #include <linux/module.h>
27 #include <linux/time.h>
28 #include <linux/jbd2.h>
29 #include <linux/highuid.h>
30 #include <linux/pagemap.h>
31 #include <linux/quotaops.h>
32 #include <linux/string.h>
33 #include <linux/buffer_head.h>
34 #include <linux/writeback.h>
35 #include <linux/pagevec.h>
36 #include <linux/mpage.h>
37 #include <linux/namei.h>
38 #include <linux/uio.h>
39 #include <linux/bio.h>
40 #include <linux/workqueue.h>
41 #include <linux/kernel.h>
42 #include <linux/slab.h>
44 #include "ext4_jbd2.h"
47 #include "ext4_extents.h"
49 #include <trace/events/ext4.h>
51 #define MPAGE_DA_EXTENT_TAIL 0x01
53 static inline int ext4_begin_ordered_truncate(struct inode
*inode
,
56 return jbd2_journal_begin_ordered_truncate(
57 EXT4_SB(inode
->i_sb
)->s_journal
,
58 &EXT4_I(inode
)->jinode
,
62 static void ext4_invalidatepage(struct page
*page
, unsigned long offset
);
63 static int ext4_writepage(struct page
*page
, struct writeback_control
*wbc
);
66 * Test whether an inode is a fast symlink.
68 static int ext4_inode_is_fast_symlink(struct inode
*inode
)
70 int ea_blocks
= EXT4_I(inode
)->i_file_acl
?
71 (inode
->i_sb
->s_blocksize
>> 9) : 0;
73 return (S_ISLNK(inode
->i_mode
) && inode
->i_blocks
- ea_blocks
== 0);
77 * Work out how many blocks we need to proceed with the next chunk of a
78 * truncate transaction.
80 static unsigned long blocks_for_truncate(struct inode
*inode
)
84 needed
= inode
->i_blocks
>> (inode
->i_sb
->s_blocksize_bits
- 9);
86 /* Give ourselves just enough room to cope with inodes in which
87 * i_blocks is corrupt: we've seen disk corruptions in the past
88 * which resulted in random data in an inode which looked enough
89 * like a regular file for ext4 to try to delete it. Things
90 * will go a bit crazy if that happens, but at least we should
91 * try not to panic the whole kernel. */
95 /* But we need to bound the transaction so we don't overflow the
97 if (needed
> EXT4_MAX_TRANS_DATA
)
98 needed
= EXT4_MAX_TRANS_DATA
;
100 return EXT4_DATA_TRANS_BLOCKS(inode
->i_sb
) + needed
;
104 * Truncate transactions can be complex and absolutely huge. So we need to
105 * be able to restart the transaction at a conventient checkpoint to make
106 * sure we don't overflow the journal.
108 * start_transaction gets us a new handle for a truncate transaction,
109 * and extend_transaction tries to extend the existing one a bit. If
110 * extend fails, we need to propagate the failure up and restart the
111 * transaction in the top-level truncate loop. --sct
113 static handle_t
*start_transaction(struct inode
*inode
)
117 result
= ext4_journal_start(inode
, blocks_for_truncate(inode
));
121 ext4_std_error(inode
->i_sb
, PTR_ERR(result
));
126 * Try to extend this transaction for the purposes of truncation.
128 * Returns 0 if we managed to create more room. If we can't create more
129 * room, and the transaction must be restarted we return 1.
131 static int try_to_extend_transaction(handle_t
*handle
, struct inode
*inode
)
133 if (!ext4_handle_valid(handle
))
135 if (ext4_handle_has_enough_credits(handle
, EXT4_RESERVE_TRANS_BLOCKS
+1))
137 if (!ext4_journal_extend(handle
, blocks_for_truncate(inode
)))
143 * Restart the transaction associated with *handle. This does a commit,
144 * so before we call here everything must be consistently dirtied against
147 int ext4_truncate_restart_trans(handle_t
*handle
, struct inode
*inode
,
153 * Drop i_data_sem to avoid deadlock with ext4_map_blocks. At this
154 * moment, get_block can be called only for blocks inside i_size since
155 * page cache has been already dropped and writes are blocked by
156 * i_mutex. So we can safely drop the i_data_sem here.
158 BUG_ON(EXT4_JOURNAL(inode
) == NULL
);
159 jbd_debug(2, "restarting handle %p\n", handle
);
160 up_write(&EXT4_I(inode
)->i_data_sem
);
161 ret
= ext4_journal_restart(handle
, blocks_for_truncate(inode
));
162 down_write(&EXT4_I(inode
)->i_data_sem
);
163 ext4_discard_preallocations(inode
);
169 * Called at the last iput() if i_nlink is zero.
171 void ext4_evict_inode(struct inode
*inode
)
176 if (inode
->i_nlink
) {
177 truncate_inode_pages(&inode
->i_data
, 0);
181 if (!is_bad_inode(inode
))
182 dquot_initialize(inode
);
184 if (ext4_should_order_data(inode
))
185 ext4_begin_ordered_truncate(inode
, 0);
186 truncate_inode_pages(&inode
->i_data
, 0);
188 if (is_bad_inode(inode
))
191 handle
= ext4_journal_start(inode
, blocks_for_truncate(inode
)+3);
192 if (IS_ERR(handle
)) {
193 ext4_std_error(inode
->i_sb
, PTR_ERR(handle
));
195 * If we're going to skip the normal cleanup, we still need to
196 * make sure that the in-core orphan linked list is properly
199 ext4_orphan_del(NULL
, inode
);
204 ext4_handle_sync(handle
);
206 err
= ext4_mark_inode_dirty(handle
, inode
);
208 ext4_warning(inode
->i_sb
,
209 "couldn't mark inode dirty (err %d)", err
);
213 ext4_truncate(inode
);
216 * ext4_ext_truncate() doesn't reserve any slop when it
217 * restarts journal transactions; therefore there may not be
218 * enough credits left in the handle to remove the inode from
219 * the orphan list and set the dtime field.
221 if (!ext4_handle_has_enough_credits(handle
, 3)) {
222 err
= ext4_journal_extend(handle
, 3);
224 err
= ext4_journal_restart(handle
, 3);
226 ext4_warning(inode
->i_sb
,
227 "couldn't extend journal (err %d)", err
);
229 ext4_journal_stop(handle
);
230 ext4_orphan_del(NULL
, inode
);
236 * Kill off the orphan record which ext4_truncate created.
237 * AKPM: I think this can be inside the above `if'.
238 * Note that ext4_orphan_del() has to be able to cope with the
239 * deletion of a non-existent orphan - this is because we don't
240 * know if ext4_truncate() actually created an orphan record.
241 * (Well, we could do this if we need to, but heck - it works)
243 ext4_orphan_del(handle
, inode
);
244 EXT4_I(inode
)->i_dtime
= get_seconds();
247 * One subtle ordering requirement: if anything has gone wrong
248 * (transaction abort, IO errors, whatever), then we can still
249 * do these next steps (the fs will already have been marked as
250 * having errors), but we can't free the inode if the mark_dirty
253 if (ext4_mark_inode_dirty(handle
, inode
))
254 /* If that failed, just do the required in-core inode clear. */
255 ext4_clear_inode(inode
);
257 ext4_free_inode(handle
, inode
);
258 ext4_journal_stop(handle
);
261 ext4_clear_inode(inode
); /* We must guarantee clearing of inode... */
267 struct buffer_head
*bh
;
270 static inline void add_chain(Indirect
*p
, struct buffer_head
*bh
, __le32
*v
)
272 p
->key
= *(p
->p
= v
);
277 * ext4_block_to_path - parse the block number into array of offsets
278 * @inode: inode in question (we are only interested in its superblock)
279 * @i_block: block number to be parsed
280 * @offsets: array to store the offsets in
281 * @boundary: set this non-zero if the referred-to block is likely to be
282 * followed (on disk) by an indirect block.
284 * To store the locations of file's data ext4 uses a data structure common
285 * for UNIX filesystems - tree of pointers anchored in the inode, with
286 * data blocks at leaves and indirect blocks in intermediate nodes.
287 * This function translates the block number into path in that tree -
288 * return value is the path length and @offsets[n] is the offset of
289 * pointer to (n+1)th node in the nth one. If @block is out of range
290 * (negative or too large) warning is printed and zero returned.
292 * Note: function doesn't find node addresses, so no IO is needed. All
293 * we need to know is the capacity of indirect blocks (taken from the
298 * Portability note: the last comparison (check that we fit into triple
299 * indirect block) is spelled differently, because otherwise on an
300 * architecture with 32-bit longs and 8Kb pages we might get into trouble
301 * if our filesystem had 8Kb blocks. We might use long long, but that would
302 * kill us on x86. Oh, well, at least the sign propagation does not matter -
303 * i_block would have to be negative in the very beginning, so we would not
307 static int ext4_block_to_path(struct inode
*inode
,
309 ext4_lblk_t offsets
[4], int *boundary
)
311 int ptrs
= EXT4_ADDR_PER_BLOCK(inode
->i_sb
);
312 int ptrs_bits
= EXT4_ADDR_PER_BLOCK_BITS(inode
->i_sb
);
313 const long direct_blocks
= EXT4_NDIR_BLOCKS
,
314 indirect_blocks
= ptrs
,
315 double_blocks
= (1 << (ptrs_bits
* 2));
319 if (i_block
< direct_blocks
) {
320 offsets
[n
++] = i_block
;
321 final
= direct_blocks
;
322 } else if ((i_block
-= direct_blocks
) < indirect_blocks
) {
323 offsets
[n
++] = EXT4_IND_BLOCK
;
324 offsets
[n
++] = i_block
;
326 } else if ((i_block
-= indirect_blocks
) < double_blocks
) {
327 offsets
[n
++] = EXT4_DIND_BLOCK
;
328 offsets
[n
++] = i_block
>> ptrs_bits
;
329 offsets
[n
++] = i_block
& (ptrs
- 1);
331 } else if (((i_block
-= double_blocks
) >> (ptrs_bits
* 2)) < ptrs
) {
332 offsets
[n
++] = EXT4_TIND_BLOCK
;
333 offsets
[n
++] = i_block
>> (ptrs_bits
* 2);
334 offsets
[n
++] = (i_block
>> ptrs_bits
) & (ptrs
- 1);
335 offsets
[n
++] = i_block
& (ptrs
- 1);
338 ext4_warning(inode
->i_sb
, "block %lu > max in inode %lu",
339 i_block
+ direct_blocks
+
340 indirect_blocks
+ double_blocks
, inode
->i_ino
);
343 *boundary
= final
- 1 - (i_block
& (ptrs
- 1));
347 static int __ext4_check_blockref(const char *function
, unsigned int line
,
349 __le32
*p
, unsigned int max
)
351 struct ext4_super_block
*es
= EXT4_SB(inode
->i_sb
)->s_es
;
355 while (bref
< p
+max
) {
356 blk
= le32_to_cpu(*bref
++);
358 unlikely(!ext4_data_block_valid(EXT4_SB(inode
->i_sb
),
360 es
->s_last_error_block
= cpu_to_le64(blk
);
361 ext4_error_inode(inode
, function
, line
, blk
,
370 #define ext4_check_indirect_blockref(inode, bh) \
371 __ext4_check_blockref(__func__, __LINE__, inode, \
372 (__le32 *)(bh)->b_data, \
373 EXT4_ADDR_PER_BLOCK((inode)->i_sb))
375 #define ext4_check_inode_blockref(inode) \
376 __ext4_check_blockref(__func__, __LINE__, inode, \
377 EXT4_I(inode)->i_data, \
381 * ext4_get_branch - read the chain of indirect blocks leading to data
382 * @inode: inode in question
383 * @depth: depth of the chain (1 - direct pointer, etc.)
384 * @offsets: offsets of pointers in inode/indirect blocks
385 * @chain: place to store the result
386 * @err: here we store the error value
388 * Function fills the array of triples <key, p, bh> and returns %NULL
389 * if everything went OK or the pointer to the last filled triple
390 * (incomplete one) otherwise. Upon the return chain[i].key contains
391 * the number of (i+1)-th block in the chain (as it is stored in memory,
392 * i.e. little-endian 32-bit), chain[i].p contains the address of that
393 * number (it points into struct inode for i==0 and into the bh->b_data
394 * for i>0) and chain[i].bh points to the buffer_head of i-th indirect
395 * block for i>0 and NULL for i==0. In other words, it holds the block
396 * numbers of the chain, addresses they were taken from (and where we can
397 * verify that chain did not change) and buffer_heads hosting these
400 * Function stops when it stumbles upon zero pointer (absent block)
401 * (pointer to last triple returned, *@err == 0)
402 * or when it gets an IO error reading an indirect block
403 * (ditto, *@err == -EIO)
404 * or when it reads all @depth-1 indirect blocks successfully and finds
405 * the whole chain, all way to the data (returns %NULL, *err == 0).
407 * Need to be called with
408 * down_read(&EXT4_I(inode)->i_data_sem)
410 static Indirect
*ext4_get_branch(struct inode
*inode
, int depth
,
411 ext4_lblk_t
*offsets
,
412 Indirect chain
[4], int *err
)
414 struct super_block
*sb
= inode
->i_sb
;
416 struct buffer_head
*bh
;
419 /* i_data is not going away, no lock needed */
420 add_chain(chain
, NULL
, EXT4_I(inode
)->i_data
+ *offsets
);
424 bh
= sb_getblk(sb
, le32_to_cpu(p
->key
));
428 if (!bh_uptodate_or_lock(bh
)) {
429 if (bh_submit_read(bh
) < 0) {
433 /* validate block references */
434 if (ext4_check_indirect_blockref(inode
, bh
)) {
440 add_chain(++p
, bh
, (__le32
*)bh
->b_data
+ *++offsets
);
454 * ext4_find_near - find a place for allocation with sufficient locality
456 * @ind: descriptor of indirect block.
458 * This function returns the preferred place for block allocation.
459 * It is used when heuristic for sequential allocation fails.
461 * + if there is a block to the left of our position - allocate near it.
462 * + if pointer will live in indirect block - allocate near that block.
463 * + if pointer will live in inode - allocate in the same
466 * In the latter case we colour the starting block by the callers PID to
467 * prevent it from clashing with concurrent allocations for a different inode
468 * in the same block group. The PID is used here so that functionally related
469 * files will be close-by on-disk.
471 * Caller must make sure that @ind is valid and will stay that way.
473 static ext4_fsblk_t
ext4_find_near(struct inode
*inode
, Indirect
*ind
)
475 struct ext4_inode_info
*ei
= EXT4_I(inode
);
476 __le32
*start
= ind
->bh
? (__le32
*) ind
->bh
->b_data
: ei
->i_data
;
478 ext4_fsblk_t bg_start
;
479 ext4_fsblk_t last_block
;
480 ext4_grpblk_t colour
;
481 ext4_group_t block_group
;
482 int flex_size
= ext4_flex_bg_size(EXT4_SB(inode
->i_sb
));
484 /* Try to find previous block */
485 for (p
= ind
->p
- 1; p
>= start
; p
--) {
487 return le32_to_cpu(*p
);
490 /* No such thing, so let's try location of indirect block */
492 return ind
->bh
->b_blocknr
;
495 * It is going to be referred to from the inode itself? OK, just put it
496 * into the same cylinder group then.
498 block_group
= ei
->i_block_group
;
499 if (flex_size
>= EXT4_FLEX_SIZE_DIR_ALLOC_SCHEME
) {
500 block_group
&= ~(flex_size
-1);
501 if (S_ISREG(inode
->i_mode
))
504 bg_start
= ext4_group_first_block_no(inode
->i_sb
, block_group
);
505 last_block
= ext4_blocks_count(EXT4_SB(inode
->i_sb
)->s_es
) - 1;
508 * If we are doing delayed allocation, we don't need take
509 * colour into account.
511 if (test_opt(inode
->i_sb
, DELALLOC
))
514 if (bg_start
+ EXT4_BLOCKS_PER_GROUP(inode
->i_sb
) <= last_block
)
515 colour
= (current
->pid
% 16) *
516 (EXT4_BLOCKS_PER_GROUP(inode
->i_sb
) / 16);
518 colour
= (current
->pid
% 16) * ((last_block
- bg_start
) / 16);
519 return bg_start
+ colour
;
523 * ext4_find_goal - find a preferred place for allocation.
525 * @block: block we want
526 * @partial: pointer to the last triple within a chain
528 * Normally this function find the preferred place for block allocation,
530 * Because this is only used for non-extent files, we limit the block nr
533 static ext4_fsblk_t
ext4_find_goal(struct inode
*inode
, ext4_lblk_t block
,
539 * XXX need to get goal block from mballoc's data structures
542 goal
= ext4_find_near(inode
, partial
);
543 goal
= goal
& EXT4_MAX_BLOCK_FILE_PHYS
;
548 * ext4_blks_to_allocate: Look up the block map and count the number
549 * of direct blocks need to be allocated for the given branch.
551 * @branch: chain of indirect blocks
552 * @k: number of blocks need for indirect blocks
553 * @blks: number of data blocks to be mapped.
554 * @blocks_to_boundary: the offset in the indirect block
556 * return the total number of blocks to be allocate, including the
557 * direct and indirect blocks.
559 static int ext4_blks_to_allocate(Indirect
*branch
, int k
, unsigned int blks
,
560 int blocks_to_boundary
)
562 unsigned int count
= 0;
565 * Simple case, [t,d]Indirect block(s) has not allocated yet
566 * then it's clear blocks on that path have not allocated
569 /* right now we don't handle cross boundary allocation */
570 if (blks
< blocks_to_boundary
+ 1)
573 count
+= blocks_to_boundary
+ 1;
578 while (count
< blks
&& count
<= blocks_to_boundary
&&
579 le32_to_cpu(*(branch
[0].p
+ count
)) == 0) {
586 * ext4_alloc_blocks: multiple allocate blocks needed for a branch
587 * @indirect_blks: the number of blocks need to allocate for indirect
590 * @new_blocks: on return it will store the new block numbers for
591 * the indirect blocks(if needed) and the first direct block,
592 * @blks: on return it will store the total number of allocated
595 static int ext4_alloc_blocks(handle_t
*handle
, struct inode
*inode
,
596 ext4_lblk_t iblock
, ext4_fsblk_t goal
,
597 int indirect_blks
, int blks
,
598 ext4_fsblk_t new_blocks
[4], int *err
)
600 struct ext4_allocation_request ar
;
602 unsigned long count
= 0, blk_allocated
= 0;
604 ext4_fsblk_t current_block
= 0;
608 * Here we try to allocate the requested multiple blocks at once,
609 * on a best-effort basis.
610 * To build a branch, we should allocate blocks for
611 * the indirect blocks(if not allocated yet), and at least
612 * the first direct block of this branch. That's the
613 * minimum number of blocks need to allocate(required)
615 /* first we try to allocate the indirect blocks */
616 target
= indirect_blks
;
619 /* allocating blocks for indirect blocks and direct blocks */
620 current_block
= ext4_new_meta_blocks(handle
, inode
,
625 if (unlikely(current_block
+ count
> EXT4_MAX_BLOCK_FILE_PHYS
)) {
626 EXT4_ERROR_INODE(inode
,
627 "current_block %llu + count %lu > %d!",
628 current_block
, count
,
629 EXT4_MAX_BLOCK_FILE_PHYS
);
635 /* allocate blocks for indirect blocks */
636 while (index
< indirect_blks
&& count
) {
637 new_blocks
[index
++] = current_block
++;
642 * save the new block number
643 * for the first direct block
645 new_blocks
[index
] = current_block
;
646 printk(KERN_INFO
"%s returned more blocks than "
647 "requested\n", __func__
);
653 target
= blks
- count
;
654 blk_allocated
= count
;
657 /* Now allocate data blocks */
658 memset(&ar
, 0, sizeof(ar
));
663 if (S_ISREG(inode
->i_mode
))
664 /* enable in-core preallocation only for regular files */
665 ar
.flags
= EXT4_MB_HINT_DATA
;
667 current_block
= ext4_mb_new_blocks(handle
, &ar
, err
);
668 if (unlikely(current_block
+ ar
.len
> EXT4_MAX_BLOCK_FILE_PHYS
)) {
669 EXT4_ERROR_INODE(inode
,
670 "current_block %llu + ar.len %d > %d!",
671 current_block
, ar
.len
,
672 EXT4_MAX_BLOCK_FILE_PHYS
);
677 if (*err
&& (target
== blks
)) {
679 * if the allocation failed and we didn't allocate
685 if (target
== blks
) {
687 * save the new block number
688 * for the first direct block
690 new_blocks
[index
] = current_block
;
692 blk_allocated
+= ar
.len
;
695 /* total number of blocks allocated for direct blocks */
700 for (i
= 0; i
< index
; i
++)
701 ext4_free_blocks(handle
, inode
, 0, new_blocks
[i
], 1, 0);
706 * ext4_alloc_branch - allocate and set up a chain of blocks.
708 * @indirect_blks: number of allocated indirect blocks
709 * @blks: number of allocated direct blocks
710 * @offsets: offsets (in the blocks) to store the pointers to next.
711 * @branch: place to store the chain in.
713 * This function allocates blocks, zeroes out all but the last one,
714 * links them into chain and (if we are synchronous) writes them to disk.
715 * In other words, it prepares a branch that can be spliced onto the
716 * inode. It stores the information about that chain in the branch[], in
717 * the same format as ext4_get_branch() would do. We are calling it after
718 * we had read the existing part of chain and partial points to the last
719 * triple of that (one with zero ->key). Upon the exit we have the same
720 * picture as after the successful ext4_get_block(), except that in one
721 * place chain is disconnected - *branch->p is still zero (we did not
722 * set the last link), but branch->key contains the number that should
723 * be placed into *branch->p to fill that gap.
725 * If allocation fails we free all blocks we've allocated (and forget
726 * their buffer_heads) and return the error value the from failed
727 * ext4_alloc_block() (normally -ENOSPC). Otherwise we set the chain
728 * as described above and return 0.
730 static int ext4_alloc_branch(handle_t
*handle
, struct inode
*inode
,
731 ext4_lblk_t iblock
, int indirect_blks
,
732 int *blks
, ext4_fsblk_t goal
,
733 ext4_lblk_t
*offsets
, Indirect
*branch
)
735 int blocksize
= inode
->i_sb
->s_blocksize
;
738 struct buffer_head
*bh
;
740 ext4_fsblk_t new_blocks
[4];
741 ext4_fsblk_t current_block
;
743 num
= ext4_alloc_blocks(handle
, inode
, iblock
, goal
, indirect_blks
,
744 *blks
, new_blocks
, &err
);
748 branch
[0].key
= cpu_to_le32(new_blocks
[0]);
750 * metadata blocks and data blocks are allocated.
752 for (n
= 1; n
<= indirect_blks
; n
++) {
754 * Get buffer_head for parent block, zero it out
755 * and set the pointer to new one, then send
758 bh
= sb_getblk(inode
->i_sb
, new_blocks
[n
-1]);
761 BUFFER_TRACE(bh
, "call get_create_access");
762 err
= ext4_journal_get_create_access(handle
, bh
);
764 /* Don't brelse(bh) here; it's done in
765 * ext4_journal_forget() below */
770 memset(bh
->b_data
, 0, blocksize
);
771 branch
[n
].p
= (__le32
*) bh
->b_data
+ offsets
[n
];
772 branch
[n
].key
= cpu_to_le32(new_blocks
[n
]);
773 *branch
[n
].p
= branch
[n
].key
;
774 if (n
== indirect_blks
) {
775 current_block
= new_blocks
[n
];
777 * End of chain, update the last new metablock of
778 * the chain to point to the new allocated
779 * data blocks numbers
781 for (i
= 1; i
< num
; i
++)
782 *(branch
[n
].p
+ i
) = cpu_to_le32(++current_block
);
784 BUFFER_TRACE(bh
, "marking uptodate");
785 set_buffer_uptodate(bh
);
788 BUFFER_TRACE(bh
, "call ext4_handle_dirty_metadata");
789 err
= ext4_handle_dirty_metadata(handle
, inode
, bh
);
796 /* Allocation failed, free what we already allocated */
797 ext4_free_blocks(handle
, inode
, 0, new_blocks
[0], 1, 0);
798 for (i
= 1; i
<= n
; i
++) {
800 * branch[i].bh is newly allocated, so there is no
801 * need to revoke the block, which is why we don't
802 * need to set EXT4_FREE_BLOCKS_METADATA.
804 ext4_free_blocks(handle
, inode
, 0, new_blocks
[i
], 1,
805 EXT4_FREE_BLOCKS_FORGET
);
807 for (i
= n
+1; i
< indirect_blks
; i
++)
808 ext4_free_blocks(handle
, inode
, 0, new_blocks
[i
], 1, 0);
810 ext4_free_blocks(handle
, inode
, 0, new_blocks
[i
], num
, 0);
816 * ext4_splice_branch - splice the allocated branch onto inode.
818 * @block: (logical) number of block we are adding
819 * @chain: chain of indirect blocks (with a missing link - see
821 * @where: location of missing link
822 * @num: number of indirect blocks we are adding
823 * @blks: number of direct blocks we are adding
825 * This function fills the missing link and does all housekeeping needed in
826 * inode (->i_blocks, etc.). In case of success we end up with the full
827 * chain to new block and return 0.
829 static int ext4_splice_branch(handle_t
*handle
, struct inode
*inode
,
830 ext4_lblk_t block
, Indirect
*where
, int num
,
835 ext4_fsblk_t current_block
;
838 * If we're splicing into a [td]indirect block (as opposed to the
839 * inode) then we need to get write access to the [td]indirect block
843 BUFFER_TRACE(where
->bh
, "get_write_access");
844 err
= ext4_journal_get_write_access(handle
, where
->bh
);
850 *where
->p
= where
->key
;
853 * Update the host buffer_head or inode to point to more just allocated
854 * direct blocks blocks
856 if (num
== 0 && blks
> 1) {
857 current_block
= le32_to_cpu(where
->key
) + 1;
858 for (i
= 1; i
< blks
; i
++)
859 *(where
->p
+ i
) = cpu_to_le32(current_block
++);
862 /* We are done with atomic stuff, now do the rest of housekeeping */
863 /* had we spliced it onto indirect block? */
866 * If we spliced it onto an indirect block, we haven't
867 * altered the inode. Note however that if it is being spliced
868 * onto an indirect block at the very end of the file (the
869 * file is growing) then we *will* alter the inode to reflect
870 * the new i_size. But that is not done here - it is done in
871 * generic_commit_write->__mark_inode_dirty->ext4_dirty_inode.
873 jbd_debug(5, "splicing indirect only\n");
874 BUFFER_TRACE(where
->bh
, "call ext4_handle_dirty_metadata");
875 err
= ext4_handle_dirty_metadata(handle
, inode
, where
->bh
);
880 * OK, we spliced it into the inode itself on a direct block.
882 ext4_mark_inode_dirty(handle
, inode
);
883 jbd_debug(5, "splicing direct\n");
888 for (i
= 1; i
<= num
; i
++) {
890 * branch[i].bh is newly allocated, so there is no
891 * need to revoke the block, which is why we don't
892 * need to set EXT4_FREE_BLOCKS_METADATA.
894 ext4_free_blocks(handle
, inode
, where
[i
].bh
, 0, 1,
895 EXT4_FREE_BLOCKS_FORGET
);
897 ext4_free_blocks(handle
, inode
, 0, le32_to_cpu(where
[num
].key
),
904 * The ext4_ind_map_blocks() function handles non-extents inodes
905 * (i.e., using the traditional indirect/double-indirect i_blocks
906 * scheme) for ext4_map_blocks().
908 * Allocation strategy is simple: if we have to allocate something, we will
909 * have to go the whole way to leaf. So let's do it before attaching anything
910 * to tree, set linkage between the newborn blocks, write them if sync is
911 * required, recheck the path, free and repeat if check fails, otherwise
912 * set the last missing link (that will protect us from any truncate-generated
913 * removals - all blocks on the path are immune now) and possibly force the
914 * write on the parent block.
915 * That has a nice additional property: no special recovery from the failed
916 * allocations is needed - we simply release blocks and do not touch anything
917 * reachable from inode.
919 * `handle' can be NULL if create == 0.
921 * return > 0, # of blocks mapped or allocated.
922 * return = 0, if plain lookup failed.
923 * return < 0, error case.
925 * The ext4_ind_get_blocks() function should be called with
926 * down_write(&EXT4_I(inode)->i_data_sem) if allocating filesystem
927 * blocks (i.e., flags has EXT4_GET_BLOCKS_CREATE set) or
928 * down_read(&EXT4_I(inode)->i_data_sem) if not allocating file system
931 static int ext4_ind_map_blocks(handle_t
*handle
, struct inode
*inode
,
932 struct ext4_map_blocks
*map
,
936 ext4_lblk_t offsets
[4];
941 int blocks_to_boundary
= 0;
944 ext4_fsblk_t first_block
= 0;
946 J_ASSERT(!(ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
)));
947 J_ASSERT(handle
!= NULL
|| (flags
& EXT4_GET_BLOCKS_CREATE
) == 0);
948 depth
= ext4_block_to_path(inode
, map
->m_lblk
, offsets
,
949 &blocks_to_boundary
);
954 partial
= ext4_get_branch(inode
, depth
, offsets
, chain
, &err
);
956 /* Simplest case - block found, no allocation needed */
958 first_block
= le32_to_cpu(chain
[depth
- 1].key
);
961 while (count
< map
->m_len
&& count
<= blocks_to_boundary
) {
964 blk
= le32_to_cpu(*(chain
[depth
-1].p
+ count
));
966 if (blk
== first_block
+ count
)
974 /* Next simple case - plain lookup or failed read of indirect block */
975 if ((flags
& EXT4_GET_BLOCKS_CREATE
) == 0 || err
== -EIO
)
979 * Okay, we need to do block allocation.
981 goal
= ext4_find_goal(inode
, map
->m_lblk
, partial
);
983 /* the number of blocks need to allocate for [d,t]indirect blocks */
984 indirect_blks
= (chain
+ depth
) - partial
- 1;
987 * Next look up the indirect map to count the totoal number of
988 * direct blocks to allocate for this branch.
990 count
= ext4_blks_to_allocate(partial
, indirect_blks
,
991 map
->m_len
, blocks_to_boundary
);
993 * Block out ext4_truncate while we alter the tree
995 err
= ext4_alloc_branch(handle
, inode
, map
->m_lblk
, indirect_blks
,
997 offsets
+ (partial
- chain
), partial
);
1000 * The ext4_splice_branch call will free and forget any buffers
1001 * on the new chain if there is a failure, but that risks using
1002 * up transaction credits, especially for bitmaps where the
1003 * credits cannot be returned. Can we handle this somehow? We
1004 * may need to return -EAGAIN upwards in the worst case. --sct
1007 err
= ext4_splice_branch(handle
, inode
, map
->m_lblk
,
1008 partial
, indirect_blks
, count
);
1012 map
->m_flags
|= EXT4_MAP_NEW
;
1014 ext4_update_inode_fsync_trans(handle
, inode
, 1);
1016 map
->m_flags
|= EXT4_MAP_MAPPED
;
1017 map
->m_pblk
= le32_to_cpu(chain
[depth
-1].key
);
1019 if (count
> blocks_to_boundary
)
1020 map
->m_flags
|= EXT4_MAP_BOUNDARY
;
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
);
1035 qsize_t
*ext4_get_reserved_space(struct inode
*inode
)
1037 return &EXT4_I(inode
)->i_reserved_quota
;
1042 * Calculate the number of metadata blocks need to reserve
1043 * to allocate a new block at @lblocks for non extent file based file
1045 static int ext4_indirect_calc_metadata_amount(struct inode
*inode
,
1048 struct ext4_inode_info
*ei
= EXT4_I(inode
);
1049 sector_t dind_mask
= ~((sector_t
)EXT4_ADDR_PER_BLOCK(inode
->i_sb
) - 1);
1052 if (lblock
< EXT4_NDIR_BLOCKS
)
1055 lblock
-= EXT4_NDIR_BLOCKS
;
1057 if (ei
->i_da_metadata_calc_len
&&
1058 (lblock
& dind_mask
) == ei
->i_da_metadata_calc_last_lblock
) {
1059 ei
->i_da_metadata_calc_len
++;
1062 ei
->i_da_metadata_calc_last_lblock
= lblock
& dind_mask
;
1063 ei
->i_da_metadata_calc_len
= 1;
1064 blk_bits
= order_base_2(lblock
);
1065 return (blk_bits
/ EXT4_ADDR_PER_BLOCK_BITS(inode
->i_sb
)) + 1;
1069 * Calculate the number of metadata blocks need to reserve
1070 * to allocate a block located at @lblock
1072 static int ext4_calc_metadata_amount(struct inode
*inode
, sector_t lblock
)
1074 if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
))
1075 return ext4_ext_calc_metadata_amount(inode
, lblock
);
1077 return ext4_indirect_calc_metadata_amount(inode
, lblock
);
1081 * Called with i_data_sem down, which is important since we can call
1082 * ext4_discard_preallocations() from here.
1084 void ext4_da_update_reserve_space(struct inode
*inode
,
1085 int used
, int quota_claim
)
1087 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
1088 struct ext4_inode_info
*ei
= EXT4_I(inode
);
1090 spin_lock(&ei
->i_block_reservation_lock
);
1091 trace_ext4_da_update_reserve_space(inode
, used
);
1092 if (unlikely(used
> ei
->i_reserved_data_blocks
)) {
1093 ext4_msg(inode
->i_sb
, KERN_NOTICE
, "%s: ino %lu, used %d "
1094 "with only %d reserved data blocks\n",
1095 __func__
, inode
->i_ino
, used
,
1096 ei
->i_reserved_data_blocks
);
1098 used
= ei
->i_reserved_data_blocks
;
1101 /* Update per-inode reservations */
1102 ei
->i_reserved_data_blocks
-= used
;
1103 ei
->i_reserved_meta_blocks
-= ei
->i_allocated_meta_blocks
;
1104 percpu_counter_sub(&sbi
->s_dirtyblocks_counter
,
1105 used
+ ei
->i_allocated_meta_blocks
);
1106 ei
->i_allocated_meta_blocks
= 0;
1108 if (ei
->i_reserved_data_blocks
== 0) {
1110 * We can release all of the reserved metadata blocks
1111 * only when we have written all of the delayed
1112 * allocation blocks.
1114 percpu_counter_sub(&sbi
->s_dirtyblocks_counter
,
1115 ei
->i_reserved_meta_blocks
);
1116 ei
->i_reserved_meta_blocks
= 0;
1117 ei
->i_da_metadata_calc_len
= 0;
1119 spin_unlock(&EXT4_I(inode
)->i_block_reservation_lock
);
1121 /* Update quota subsystem for data blocks */
1123 dquot_claim_block(inode
, used
);
1126 * We did fallocate with an offset that is already delayed
1127 * allocated. So on delayed allocated writeback we should
1128 * not re-claim the quota for fallocated blocks.
1130 dquot_release_reservation_block(inode
, used
);
1134 * If we have done all the pending block allocations and if
1135 * there aren't any writers on the inode, we can discard the
1136 * inode's preallocations.
1138 if ((ei
->i_reserved_data_blocks
== 0) &&
1139 (atomic_read(&inode
->i_writecount
) == 0))
1140 ext4_discard_preallocations(inode
);
1143 static int __check_block_validity(struct inode
*inode
, const char *func
,
1145 struct ext4_map_blocks
*map
)
1147 if (!ext4_data_block_valid(EXT4_SB(inode
->i_sb
), map
->m_pblk
,
1149 ext4_error_inode(inode
, func
, line
, map
->m_pblk
,
1150 "lblock %lu mapped to illegal pblock "
1151 "(length %d)", (unsigned long) map
->m_lblk
,
1158 #define check_block_validity(inode, map) \
1159 __check_block_validity((inode), __func__, __LINE__, (map))
1162 * Return the number of contiguous dirty pages in a given inode
1163 * starting at page frame idx.
1165 static pgoff_t
ext4_num_dirty_pages(struct inode
*inode
, pgoff_t idx
,
1166 unsigned int max_pages
)
1168 struct address_space
*mapping
= inode
->i_mapping
;
1170 struct pagevec pvec
;
1172 int i
, nr_pages
, done
= 0;
1176 pagevec_init(&pvec
, 0);
1179 nr_pages
= pagevec_lookup_tag(&pvec
, mapping
, &index
,
1180 PAGECACHE_TAG_DIRTY
,
1181 (pgoff_t
)PAGEVEC_SIZE
);
1184 for (i
= 0; i
< nr_pages
; i
++) {
1185 struct page
*page
= pvec
.pages
[i
];
1186 struct buffer_head
*bh
, *head
;
1189 if (unlikely(page
->mapping
!= mapping
) ||
1191 PageWriteback(page
) ||
1192 page
->index
!= idx
) {
1197 if (page_has_buffers(page
)) {
1198 bh
= head
= page_buffers(page
);
1200 if (!buffer_delay(bh
) &&
1201 !buffer_unwritten(bh
))
1203 bh
= bh
->b_this_page
;
1204 } while (!done
&& (bh
!= head
));
1211 if (num
>= max_pages
) {
1216 pagevec_release(&pvec
);
1222 * The ext4_map_blocks() function tries to look up the requested blocks,
1223 * and returns if the blocks are already mapped.
1225 * Otherwise it takes the write lock of the i_data_sem and allocate blocks
1226 * and store the allocated blocks in the result buffer head and mark it
1229 * If file type is extents based, it will call ext4_ext_map_blocks(),
1230 * Otherwise, call with ext4_ind_map_blocks() to handle indirect mapping
1233 * On success, it returns the number of blocks being mapped or allocate.
1234 * if create==0 and the blocks are pre-allocated and uninitialized block,
1235 * the result buffer head is unmapped. If the create ==1, it will make sure
1236 * the buffer head is mapped.
1238 * It returns 0 if plain look up failed (blocks have not been allocated), in
1239 * that casem, buffer head is unmapped
1241 * It returns the error in case of allocation failure.
1243 int ext4_map_blocks(handle_t
*handle
, struct inode
*inode
,
1244 struct ext4_map_blocks
*map
, int flags
)
1249 ext_debug("ext4_map_blocks(): inode %lu, flag %d, max_blocks %u,"
1250 "logical block %lu\n", inode
->i_ino
, flags
, map
->m_len
,
1251 (unsigned long) map
->m_lblk
);
1253 * Try to see if we can get the block without requesting a new
1254 * file system block.
1256 down_read((&EXT4_I(inode
)->i_data_sem
));
1257 if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
)) {
1258 retval
= ext4_ext_map_blocks(handle
, inode
, map
, 0);
1260 retval
= ext4_ind_map_blocks(handle
, inode
, map
, 0);
1262 up_read((&EXT4_I(inode
)->i_data_sem
));
1264 if (retval
> 0 && map
->m_flags
& EXT4_MAP_MAPPED
) {
1265 int ret
= check_block_validity(inode
, map
);
1270 /* If it is only a block(s) look up */
1271 if ((flags
& EXT4_GET_BLOCKS_CREATE
) == 0)
1275 * Returns if the blocks have already allocated
1277 * Note that if blocks have been preallocated
1278 * ext4_ext_get_block() returns th create = 0
1279 * with buffer head unmapped.
1281 if (retval
> 0 && map
->m_flags
& EXT4_MAP_MAPPED
)
1285 * When we call get_blocks without the create flag, the
1286 * BH_Unwritten flag could have gotten set if the blocks
1287 * requested were part of a uninitialized extent. We need to
1288 * clear this flag now that we are committed to convert all or
1289 * part of the uninitialized extent to be an initialized
1290 * extent. This is because we need to avoid the combination
1291 * of BH_Unwritten and BH_Mapped flags being simultaneously
1292 * set on the buffer_head.
1294 map
->m_flags
&= ~EXT4_MAP_UNWRITTEN
;
1297 * New blocks allocate and/or writing to uninitialized extent
1298 * will possibly result in updating i_data, so we take
1299 * the write lock of i_data_sem, and call get_blocks()
1300 * with create == 1 flag.
1302 down_write((&EXT4_I(inode
)->i_data_sem
));
1305 * if the caller is from delayed allocation writeout path
1306 * we have already reserved fs blocks for allocation
1307 * let the underlying get_block() function know to
1308 * avoid double accounting
1310 if (flags
& EXT4_GET_BLOCKS_DELALLOC_RESERVE
)
1311 EXT4_I(inode
)->i_delalloc_reserved_flag
= 1;
1313 * We need to check for EXT4 here because migrate
1314 * could have changed the inode type in between
1316 if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
)) {
1317 retval
= ext4_ext_map_blocks(handle
, inode
, map
, flags
);
1319 retval
= ext4_ind_map_blocks(handle
, inode
, map
, flags
);
1321 if (retval
> 0 && map
->m_flags
& EXT4_MAP_NEW
) {
1323 * We allocated new blocks which will result in
1324 * i_data's format changing. Force the migrate
1325 * to fail by clearing migrate flags
1327 ext4_clear_inode_state(inode
, EXT4_STATE_EXT_MIGRATE
);
1331 * Update reserved blocks/metadata blocks after successful
1332 * block allocation which had been deferred till now. We don't
1333 * support fallocate for non extent files. So we can update
1334 * reserve space here.
1337 (flags
& EXT4_GET_BLOCKS_DELALLOC_RESERVE
))
1338 ext4_da_update_reserve_space(inode
, retval
, 1);
1340 if (flags
& EXT4_GET_BLOCKS_DELALLOC_RESERVE
)
1341 EXT4_I(inode
)->i_delalloc_reserved_flag
= 0;
1343 up_write((&EXT4_I(inode
)->i_data_sem
));
1344 if (retval
> 0 && map
->m_flags
& EXT4_MAP_MAPPED
) {
1345 int ret
= check_block_validity(inode
, map
);
1352 /* Maximum number of blocks we map for direct IO at once. */
1353 #define DIO_MAX_BLOCKS 4096
1355 static int _ext4_get_block(struct inode
*inode
, sector_t iblock
,
1356 struct buffer_head
*bh
, int flags
)
1358 handle_t
*handle
= ext4_journal_current_handle();
1359 struct ext4_map_blocks map
;
1360 int ret
= 0, started
= 0;
1363 map
.m_lblk
= iblock
;
1364 map
.m_len
= bh
->b_size
>> inode
->i_blkbits
;
1366 if (flags
&& !handle
) {
1367 /* Direct IO write... */
1368 if (map
.m_len
> DIO_MAX_BLOCKS
)
1369 map
.m_len
= DIO_MAX_BLOCKS
;
1370 dio_credits
= ext4_chunk_trans_blocks(inode
, map
.m_len
);
1371 handle
= ext4_journal_start(inode
, dio_credits
);
1372 if (IS_ERR(handle
)) {
1373 ret
= PTR_ERR(handle
);
1379 ret
= ext4_map_blocks(handle
, inode
, &map
, flags
);
1381 map_bh(bh
, inode
->i_sb
, map
.m_pblk
);
1382 bh
->b_state
= (bh
->b_state
& ~EXT4_MAP_FLAGS
) | map
.m_flags
;
1383 bh
->b_size
= inode
->i_sb
->s_blocksize
* map
.m_len
;
1387 ext4_journal_stop(handle
);
1391 int ext4_get_block(struct inode
*inode
, sector_t iblock
,
1392 struct buffer_head
*bh
, int create
)
1394 return _ext4_get_block(inode
, iblock
, bh
,
1395 create
? EXT4_GET_BLOCKS_CREATE
: 0);
1399 * `handle' can be NULL if create is zero
1401 struct buffer_head
*ext4_getblk(handle_t
*handle
, struct inode
*inode
,
1402 ext4_lblk_t block
, int create
, int *errp
)
1404 struct ext4_map_blocks map
;
1405 struct buffer_head
*bh
;
1408 J_ASSERT(handle
!= NULL
|| create
== 0);
1412 err
= ext4_map_blocks(handle
, inode
, &map
,
1413 create
? EXT4_GET_BLOCKS_CREATE
: 0);
1421 bh
= sb_getblk(inode
->i_sb
, map
.m_pblk
);
1426 if (map
.m_flags
& EXT4_MAP_NEW
) {
1427 J_ASSERT(create
!= 0);
1428 J_ASSERT(handle
!= NULL
);
1431 * Now that we do not always journal data, we should
1432 * keep in mind whether this should always journal the
1433 * new buffer as metadata. For now, regular file
1434 * writes use ext4_get_block instead, so it's not a
1438 BUFFER_TRACE(bh
, "call get_create_access");
1439 fatal
= ext4_journal_get_create_access(handle
, bh
);
1440 if (!fatal
&& !buffer_uptodate(bh
)) {
1441 memset(bh
->b_data
, 0, inode
->i_sb
->s_blocksize
);
1442 set_buffer_uptodate(bh
);
1445 BUFFER_TRACE(bh
, "call ext4_handle_dirty_metadata");
1446 err
= ext4_handle_dirty_metadata(handle
, inode
, bh
);
1450 BUFFER_TRACE(bh
, "not a new buffer");
1460 struct buffer_head
*ext4_bread(handle_t
*handle
, struct inode
*inode
,
1461 ext4_lblk_t block
, int create
, int *err
)
1463 struct buffer_head
*bh
;
1465 bh
= ext4_getblk(handle
, inode
, block
, create
, err
);
1468 if (buffer_uptodate(bh
))
1470 ll_rw_block(READ_META
, 1, &bh
);
1472 if (buffer_uptodate(bh
))
1479 static int walk_page_buffers(handle_t
*handle
,
1480 struct buffer_head
*head
,
1484 int (*fn
)(handle_t
*handle
,
1485 struct buffer_head
*bh
))
1487 struct buffer_head
*bh
;
1488 unsigned block_start
, block_end
;
1489 unsigned blocksize
= head
->b_size
;
1491 struct buffer_head
*next
;
1493 for (bh
= head
, block_start
= 0;
1494 ret
== 0 && (bh
!= head
|| !block_start
);
1495 block_start
= block_end
, bh
= next
) {
1496 next
= bh
->b_this_page
;
1497 block_end
= block_start
+ blocksize
;
1498 if (block_end
<= from
|| block_start
>= to
) {
1499 if (partial
&& !buffer_uptodate(bh
))
1503 err
= (*fn
)(handle
, bh
);
1511 * To preserve ordering, it is essential that the hole instantiation and
1512 * the data write be encapsulated in a single transaction. We cannot
1513 * close off a transaction and start a new one between the ext4_get_block()
1514 * and the commit_write(). So doing the jbd2_journal_start at the start of
1515 * prepare_write() is the right place.
1517 * Also, this function can nest inside ext4_writepage() ->
1518 * block_write_full_page(). In that case, we *know* that ext4_writepage()
1519 * has generated enough buffer credits to do the whole page. So we won't
1520 * block on the journal in that case, which is good, because the caller may
1523 * By accident, ext4 can be reentered when a transaction is open via
1524 * quota file writes. If we were to commit the transaction while thus
1525 * reentered, there can be a deadlock - we would be holding a quota
1526 * lock, and the commit would never complete if another thread had a
1527 * transaction open and was blocking on the quota lock - a ranking
1530 * So what we do is to rely on the fact that jbd2_journal_stop/journal_start
1531 * will _not_ run commit under these circumstances because handle->h_ref
1532 * is elevated. We'll still have enough credits for the tiny quotafile
1535 static int do_journal_get_write_access(handle_t
*handle
,
1536 struct buffer_head
*bh
)
1538 int dirty
= buffer_dirty(bh
);
1541 if (!buffer_mapped(bh
) || buffer_freed(bh
))
1544 * __block_prepare_write() could have dirtied some buffers. Clean
1545 * the dirty bit as jbd2_journal_get_write_access() could complain
1546 * otherwise about fs integrity issues. Setting of the dirty bit
1547 * by __block_prepare_write() isn't a real problem here as we clear
1548 * the bit before releasing a page lock and thus writeback cannot
1549 * ever write the buffer.
1552 clear_buffer_dirty(bh
);
1553 ret
= ext4_journal_get_write_access(handle
, bh
);
1555 ret
= ext4_handle_dirty_metadata(handle
, NULL
, bh
);
1560 * Truncate blocks that were not used by write. We have to truncate the
1561 * pagecache as well so that corresponding buffers get properly unmapped.
1563 static void ext4_truncate_failed_write(struct inode
*inode
)
1565 truncate_inode_pages(inode
->i_mapping
, inode
->i_size
);
1566 ext4_truncate(inode
);
1569 static int ext4_get_block_write(struct inode
*inode
, sector_t iblock
,
1570 struct buffer_head
*bh_result
, int create
);
1571 static int ext4_write_begin(struct file
*file
, struct address_space
*mapping
,
1572 loff_t pos
, unsigned len
, unsigned flags
,
1573 struct page
**pagep
, void **fsdata
)
1575 struct inode
*inode
= mapping
->host
;
1576 int ret
, needed_blocks
;
1583 trace_ext4_write_begin(inode
, pos
, len
, flags
);
1585 * Reserve one block more for addition to orphan list in case
1586 * we allocate blocks but write fails for some reason
1588 needed_blocks
= ext4_writepage_trans_blocks(inode
) + 1;
1589 index
= pos
>> PAGE_CACHE_SHIFT
;
1590 from
= pos
& (PAGE_CACHE_SIZE
- 1);
1594 handle
= ext4_journal_start(inode
, needed_blocks
);
1595 if (IS_ERR(handle
)) {
1596 ret
= PTR_ERR(handle
);
1600 /* We cannot recurse into the filesystem as the transaction is already
1602 flags
|= AOP_FLAG_NOFS
;
1604 page
= grab_cache_page_write_begin(mapping
, index
, flags
);
1606 ext4_journal_stop(handle
);
1612 if (ext4_should_dioread_nolock(inode
))
1613 ret
= __block_write_begin(page
, pos
, len
, ext4_get_block_write
);
1615 ret
= __block_write_begin(page
, pos
, len
, ext4_get_block
);
1617 if (!ret
&& ext4_should_journal_data(inode
)) {
1618 ret
= walk_page_buffers(handle
, page_buffers(page
),
1619 from
, to
, NULL
, do_journal_get_write_access
);
1624 page_cache_release(page
);
1626 * __block_write_begin may have instantiated a few blocks
1627 * outside i_size. Trim these off again. Don't need
1628 * i_size_read because we hold i_mutex.
1630 * Add inode to orphan list in case we crash before
1633 if (pos
+ len
> inode
->i_size
&& ext4_can_truncate(inode
))
1634 ext4_orphan_add(handle
, inode
);
1636 ext4_journal_stop(handle
);
1637 if (pos
+ len
> inode
->i_size
) {
1638 ext4_truncate_failed_write(inode
);
1640 * If truncate failed early the inode might
1641 * still be on the orphan list; we need to
1642 * make sure the inode is removed from the
1643 * orphan list in that case.
1646 ext4_orphan_del(NULL
, inode
);
1650 if (ret
== -ENOSPC
&& ext4_should_retry_alloc(inode
->i_sb
, &retries
))
1656 /* For write_end() in data=journal mode */
1657 static int write_end_fn(handle_t
*handle
, struct buffer_head
*bh
)
1659 if (!buffer_mapped(bh
) || buffer_freed(bh
))
1661 set_buffer_uptodate(bh
);
1662 return ext4_handle_dirty_metadata(handle
, NULL
, bh
);
1665 static int ext4_generic_write_end(struct file
*file
,
1666 struct address_space
*mapping
,
1667 loff_t pos
, unsigned len
, unsigned copied
,
1668 struct page
*page
, void *fsdata
)
1670 int i_size_changed
= 0;
1671 struct inode
*inode
= mapping
->host
;
1672 handle_t
*handle
= ext4_journal_current_handle();
1674 copied
= block_write_end(file
, mapping
, pos
, len
, copied
, page
, fsdata
);
1677 * No need to use i_size_read() here, the i_size
1678 * cannot change under us because we hold i_mutex.
1680 * But it's important to update i_size while still holding page lock:
1681 * page writeout could otherwise come in and zero beyond i_size.
1683 if (pos
+ copied
> inode
->i_size
) {
1684 i_size_write(inode
, pos
+ copied
);
1688 if (pos
+ copied
> EXT4_I(inode
)->i_disksize
) {
1689 /* We need to mark inode dirty even if
1690 * new_i_size is less that inode->i_size
1691 * bu greater than i_disksize.(hint delalloc)
1693 ext4_update_i_disksize(inode
, (pos
+ copied
));
1697 page_cache_release(page
);
1700 * Don't mark the inode dirty under page lock. First, it unnecessarily
1701 * makes the holding time of page lock longer. Second, it forces lock
1702 * ordering of page lock and transaction start for journaling
1706 ext4_mark_inode_dirty(handle
, inode
);
1712 * We need to pick up the new inode size which generic_commit_write gave us
1713 * `file' can be NULL - eg, when called from page_symlink().
1715 * ext4 never places buffers on inode->i_mapping->private_list. metadata
1716 * buffers are managed internally.
1718 static int ext4_ordered_write_end(struct file
*file
,
1719 struct address_space
*mapping
,
1720 loff_t pos
, unsigned len
, unsigned copied
,
1721 struct page
*page
, void *fsdata
)
1723 handle_t
*handle
= ext4_journal_current_handle();
1724 struct inode
*inode
= mapping
->host
;
1727 trace_ext4_ordered_write_end(inode
, pos
, len
, copied
);
1728 ret
= ext4_jbd2_file_inode(handle
, inode
);
1731 ret2
= ext4_generic_write_end(file
, mapping
, pos
, len
, copied
,
1734 if (pos
+ len
> inode
->i_size
&& ext4_can_truncate(inode
))
1735 /* if we have allocated more blocks and copied
1736 * less. We will have blocks allocated outside
1737 * inode->i_size. So truncate them
1739 ext4_orphan_add(handle
, inode
);
1743 ret2
= ext4_journal_stop(handle
);
1747 if (pos
+ len
> inode
->i_size
) {
1748 ext4_truncate_failed_write(inode
);
1750 * If truncate failed early the inode might still be
1751 * on the orphan list; we need to make sure the inode
1752 * is removed from the orphan list in that case.
1755 ext4_orphan_del(NULL
, inode
);
1759 return ret
? ret
: copied
;
1762 static int ext4_writeback_write_end(struct file
*file
,
1763 struct address_space
*mapping
,
1764 loff_t pos
, unsigned len
, unsigned copied
,
1765 struct page
*page
, void *fsdata
)
1767 handle_t
*handle
= ext4_journal_current_handle();
1768 struct inode
*inode
= mapping
->host
;
1771 trace_ext4_writeback_write_end(inode
, pos
, len
, copied
);
1772 ret2
= ext4_generic_write_end(file
, mapping
, pos
, len
, copied
,
1775 if (pos
+ len
> inode
->i_size
&& ext4_can_truncate(inode
))
1776 /* if we have allocated more blocks and copied
1777 * less. We will have blocks allocated outside
1778 * inode->i_size. So truncate them
1780 ext4_orphan_add(handle
, inode
);
1785 ret2
= ext4_journal_stop(handle
);
1789 if (pos
+ len
> inode
->i_size
) {
1790 ext4_truncate_failed_write(inode
);
1792 * If truncate failed early the inode might still be
1793 * on the orphan list; we need to make sure the inode
1794 * is removed from the orphan list in that case.
1797 ext4_orphan_del(NULL
, inode
);
1800 return ret
? ret
: copied
;
1803 static int ext4_journalled_write_end(struct file
*file
,
1804 struct address_space
*mapping
,
1805 loff_t pos
, unsigned len
, unsigned copied
,
1806 struct page
*page
, void *fsdata
)
1808 handle_t
*handle
= ext4_journal_current_handle();
1809 struct inode
*inode
= mapping
->host
;
1815 trace_ext4_journalled_write_end(inode
, pos
, len
, copied
);
1816 from
= pos
& (PAGE_CACHE_SIZE
- 1);
1820 if (!PageUptodate(page
))
1822 page_zero_new_buffers(page
, from
+copied
, to
);
1825 ret
= walk_page_buffers(handle
, page_buffers(page
), from
,
1826 to
, &partial
, write_end_fn
);
1828 SetPageUptodate(page
);
1829 new_i_size
= pos
+ copied
;
1830 if (new_i_size
> inode
->i_size
)
1831 i_size_write(inode
, pos
+copied
);
1832 ext4_set_inode_state(inode
, EXT4_STATE_JDATA
);
1833 if (new_i_size
> EXT4_I(inode
)->i_disksize
) {
1834 ext4_update_i_disksize(inode
, new_i_size
);
1835 ret2
= ext4_mark_inode_dirty(handle
, inode
);
1841 page_cache_release(page
);
1842 if (pos
+ len
> inode
->i_size
&& ext4_can_truncate(inode
))
1843 /* if we have allocated more blocks and copied
1844 * less. We will have blocks allocated outside
1845 * inode->i_size. So truncate them
1847 ext4_orphan_add(handle
, inode
);
1849 ret2
= ext4_journal_stop(handle
);
1852 if (pos
+ len
> inode
->i_size
) {
1853 ext4_truncate_failed_write(inode
);
1855 * If truncate failed early the inode might still be
1856 * on the orphan list; we need to make sure the inode
1857 * is removed from the orphan list in that case.
1860 ext4_orphan_del(NULL
, inode
);
1863 return ret
? ret
: copied
;
1867 * Reserve a single block located at lblock
1869 static int ext4_da_reserve_space(struct inode
*inode
, sector_t lblock
)
1872 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
1873 struct ext4_inode_info
*ei
= EXT4_I(inode
);
1874 unsigned long md_needed
;
1878 * recalculate the amount of metadata blocks to reserve
1879 * in order to allocate nrblocks
1880 * worse case is one extent per block
1883 spin_lock(&ei
->i_block_reservation_lock
);
1884 md_needed
= ext4_calc_metadata_amount(inode
, lblock
);
1885 trace_ext4_da_reserve_space(inode
, md_needed
);
1886 spin_unlock(&ei
->i_block_reservation_lock
);
1889 * We will charge metadata quota at writeout time; this saves
1890 * us from metadata over-estimation, though we may go over by
1891 * a small amount in the end. Here we just reserve for data.
1893 ret
= dquot_reserve_block(inode
, 1);
1897 * We do still charge estimated metadata to the sb though;
1898 * we cannot afford to run out of free blocks.
1900 if (ext4_claim_free_blocks(sbi
, md_needed
+ 1)) {
1901 dquot_release_reservation_block(inode
, 1);
1902 if (ext4_should_retry_alloc(inode
->i_sb
, &retries
)) {
1908 spin_lock(&ei
->i_block_reservation_lock
);
1909 ei
->i_reserved_data_blocks
++;
1910 ei
->i_reserved_meta_blocks
+= md_needed
;
1911 spin_unlock(&ei
->i_block_reservation_lock
);
1913 return 0; /* success */
1916 static void ext4_da_release_space(struct inode
*inode
, int to_free
)
1918 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
1919 struct ext4_inode_info
*ei
= EXT4_I(inode
);
1922 return; /* Nothing to release, exit */
1924 spin_lock(&EXT4_I(inode
)->i_block_reservation_lock
);
1926 trace_ext4_da_release_space(inode
, to_free
);
1927 if (unlikely(to_free
> ei
->i_reserved_data_blocks
)) {
1929 * if there aren't enough reserved blocks, then the
1930 * counter is messed up somewhere. Since this
1931 * function is called from invalidate page, it's
1932 * harmless to return without any action.
1934 ext4_msg(inode
->i_sb
, KERN_NOTICE
, "ext4_da_release_space: "
1935 "ino %lu, to_free %d with only %d reserved "
1936 "data blocks\n", inode
->i_ino
, to_free
,
1937 ei
->i_reserved_data_blocks
);
1939 to_free
= ei
->i_reserved_data_blocks
;
1941 ei
->i_reserved_data_blocks
-= to_free
;
1943 if (ei
->i_reserved_data_blocks
== 0) {
1945 * We can release all of the reserved metadata blocks
1946 * only when we have written all of the delayed
1947 * allocation blocks.
1949 percpu_counter_sub(&sbi
->s_dirtyblocks_counter
,
1950 ei
->i_reserved_meta_blocks
);
1951 ei
->i_reserved_meta_blocks
= 0;
1952 ei
->i_da_metadata_calc_len
= 0;
1955 /* update fs dirty data blocks counter */
1956 percpu_counter_sub(&sbi
->s_dirtyblocks_counter
, to_free
);
1958 spin_unlock(&EXT4_I(inode
)->i_block_reservation_lock
);
1960 dquot_release_reservation_block(inode
, to_free
);
1963 static void ext4_da_page_release_reservation(struct page
*page
,
1964 unsigned long offset
)
1967 struct buffer_head
*head
, *bh
;
1968 unsigned int curr_off
= 0;
1970 head
= page_buffers(page
);
1973 unsigned int next_off
= curr_off
+ bh
->b_size
;
1975 if ((offset
<= curr_off
) && (buffer_delay(bh
))) {
1977 clear_buffer_delay(bh
);
1979 curr_off
= next_off
;
1980 } while ((bh
= bh
->b_this_page
) != head
);
1981 ext4_da_release_space(page
->mapping
->host
, to_release
);
1985 * Delayed allocation stuff
1989 * mpage_da_submit_io - walks through extent of pages and try to write
1990 * them with writepage() call back
1992 * @mpd->inode: inode
1993 * @mpd->first_page: first page of the extent
1994 * @mpd->next_page: page after the last page of the extent
1996 * By the time mpage_da_submit_io() is called we expect all blocks
1997 * to be allocated. this may be wrong if allocation failed.
1999 * As pages are already locked by write_cache_pages(), we can't use it
2001 static int mpage_da_submit_io(struct mpage_da_data
*mpd
)
2004 struct pagevec pvec
;
2005 unsigned long index
, end
;
2006 int ret
= 0, err
, nr_pages
, i
;
2007 struct inode
*inode
= mpd
->inode
;
2008 struct address_space
*mapping
= inode
->i_mapping
;
2010 BUG_ON(mpd
->next_page
<= mpd
->first_page
);
2012 * We need to start from the first_page to the next_page - 1
2013 * to make sure we also write the mapped dirty buffer_heads.
2014 * If we look at mpd->b_blocknr we would only be looking
2015 * at the currently mapped buffer_heads.
2017 index
= mpd
->first_page
;
2018 end
= mpd
->next_page
- 1;
2020 pagevec_init(&pvec
, 0);
2021 while (index
<= end
) {
2022 nr_pages
= pagevec_lookup(&pvec
, mapping
, index
, PAGEVEC_SIZE
);
2025 for (i
= 0; i
< nr_pages
; i
++) {
2026 struct page
*page
= pvec
.pages
[i
];
2028 index
= page
->index
;
2033 BUG_ON(!PageLocked(page
));
2034 BUG_ON(PageWriteback(page
));
2036 pages_skipped
= mpd
->wbc
->pages_skipped
;
2037 err
= ext4_writepage(page
, mpd
->wbc
);
2038 if (!err
&& (pages_skipped
== mpd
->wbc
->pages_skipped
))
2040 * have successfully written the page
2041 * without skipping the same
2043 mpd
->pages_written
++;
2045 * In error case, we have to continue because
2046 * remaining pages are still locked
2047 * XXX: unlock and re-dirty them?
2052 pagevec_release(&pvec
);
2058 * mpage_put_bnr_to_bhs - walk blocks and assign them actual numbers
2060 * the function goes through all passed space and put actual disk
2061 * block numbers into buffer heads, dropping BH_Delay and BH_Unwritten
2063 static void mpage_put_bnr_to_bhs(struct mpage_da_data
*mpd
,
2064 struct ext4_map_blocks
*map
)
2066 struct inode
*inode
= mpd
->inode
;
2067 struct address_space
*mapping
= inode
->i_mapping
;
2068 int blocks
= map
->m_len
;
2069 sector_t pblock
= map
->m_pblk
, cur_logical
;
2070 struct buffer_head
*head
, *bh
;
2072 struct pagevec pvec
;
2075 index
= map
->m_lblk
>> (PAGE_CACHE_SHIFT
- inode
->i_blkbits
);
2076 end
= (map
->m_lblk
+ blocks
- 1) >> (PAGE_CACHE_SHIFT
- inode
->i_blkbits
);
2077 cur_logical
= index
<< (PAGE_CACHE_SHIFT
- inode
->i_blkbits
);
2079 pagevec_init(&pvec
, 0);
2081 while (index
<= end
) {
2082 /* XXX: optimize tail */
2083 nr_pages
= pagevec_lookup(&pvec
, mapping
, index
, PAGEVEC_SIZE
);
2086 for (i
= 0; i
< nr_pages
; i
++) {
2087 struct page
*page
= pvec
.pages
[i
];
2089 index
= page
->index
;
2094 BUG_ON(!PageLocked(page
));
2095 BUG_ON(PageWriteback(page
));
2096 BUG_ON(!page_has_buffers(page
));
2098 bh
= page_buffers(page
);
2101 /* skip blocks out of the range */
2103 if (cur_logical
>= map
->m_lblk
)
2106 } while ((bh
= bh
->b_this_page
) != head
);
2109 if (cur_logical
> map
->m_lblk
+ (blocks
- 1))
2112 if (buffer_delay(bh
) || buffer_unwritten(bh
)) {
2114 BUG_ON(bh
->b_bdev
!= inode
->i_sb
->s_bdev
);
2116 if (buffer_delay(bh
)) {
2117 clear_buffer_delay(bh
);
2118 bh
->b_blocknr
= pblock
;
2121 * unwritten already should have
2122 * blocknr assigned. Verify that
2124 clear_buffer_unwritten(bh
);
2125 BUG_ON(bh
->b_blocknr
!= pblock
);
2128 } else if (buffer_mapped(bh
))
2129 BUG_ON(bh
->b_blocknr
!= pblock
);
2131 if (map
->m_flags
& EXT4_MAP_UNINIT
)
2132 set_buffer_uninit(bh
);
2135 } while ((bh
= bh
->b_this_page
) != head
);
2137 pagevec_release(&pvec
);
2142 static void ext4_da_block_invalidatepages(struct mpage_da_data
*mpd
,
2143 sector_t logical
, long blk_cnt
)
2147 struct pagevec pvec
;
2148 struct inode
*inode
= mpd
->inode
;
2149 struct address_space
*mapping
= inode
->i_mapping
;
2151 index
= logical
>> (PAGE_CACHE_SHIFT
- inode
->i_blkbits
);
2152 end
= (logical
+ blk_cnt
- 1) >>
2153 (PAGE_CACHE_SHIFT
- inode
->i_blkbits
);
2154 while (index
<= end
) {
2155 nr_pages
= pagevec_lookup(&pvec
, mapping
, index
, PAGEVEC_SIZE
);
2158 for (i
= 0; i
< nr_pages
; i
++) {
2159 struct page
*page
= pvec
.pages
[i
];
2160 if (page
->index
> end
)
2162 BUG_ON(!PageLocked(page
));
2163 BUG_ON(PageWriteback(page
));
2164 block_invalidatepage(page
, 0);
2165 ClearPageUptodate(page
);
2168 index
= pvec
.pages
[nr_pages
- 1]->index
+ 1;
2169 pagevec_release(&pvec
);
2174 static void ext4_print_free_blocks(struct inode
*inode
)
2176 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
2177 printk(KERN_CRIT
"Total free blocks count %lld\n",
2178 ext4_count_free_blocks(inode
->i_sb
));
2179 printk(KERN_CRIT
"Free/Dirty block details\n");
2180 printk(KERN_CRIT
"free_blocks=%lld\n",
2181 (long long) percpu_counter_sum(&sbi
->s_freeblocks_counter
));
2182 printk(KERN_CRIT
"dirty_blocks=%lld\n",
2183 (long long) percpu_counter_sum(&sbi
->s_dirtyblocks_counter
));
2184 printk(KERN_CRIT
"Block reservation details\n");
2185 printk(KERN_CRIT
"i_reserved_data_blocks=%u\n",
2186 EXT4_I(inode
)->i_reserved_data_blocks
);
2187 printk(KERN_CRIT
"i_reserved_meta_blocks=%u\n",
2188 EXT4_I(inode
)->i_reserved_meta_blocks
);
2193 * mpage_da_map_and_submit - go through given space, map them
2194 * if necessary, and then submit them for I/O
2196 * @mpd - bh describing space
2198 * The function skips space we know is already mapped to disk blocks.
2201 static void mpage_da_map_and_submit(struct mpage_da_data
*mpd
)
2203 int err
, blks
, get_blocks_flags
;
2204 struct ext4_map_blocks map
;
2205 sector_t next
= mpd
->b_blocknr
;
2206 unsigned max_blocks
= mpd
->b_size
>> mpd
->inode
->i_blkbits
;
2207 loff_t disksize
= EXT4_I(mpd
->inode
)->i_disksize
;
2208 handle_t
*handle
= NULL
;
2211 * If the blocks are mapped already, or we couldn't accumulate
2212 * any blocks, then proceed immediately to the submission stage.
2214 if ((mpd
->b_size
== 0) ||
2215 ((mpd
->b_state
& (1 << BH_Mapped
)) &&
2216 !(mpd
->b_state
& (1 << BH_Delay
)) &&
2217 !(mpd
->b_state
& (1 << BH_Unwritten
))))
2220 handle
= ext4_journal_current_handle();
2224 * Call ext4_map_blocks() to allocate any delayed allocation
2225 * blocks, or to convert an uninitialized extent to be
2226 * initialized (in the case where we have written into
2227 * one or more preallocated blocks).
2229 * We pass in the magic EXT4_GET_BLOCKS_DELALLOC_RESERVE to
2230 * indicate that we are on the delayed allocation path. This
2231 * affects functions in many different parts of the allocation
2232 * call path. This flag exists primarily because we don't
2233 * want to change *many* call functions, so ext4_map_blocks()
2234 * will set the magic i_delalloc_reserved_flag once the
2235 * inode's allocation semaphore is taken.
2237 * If the blocks in questions were delalloc blocks, set
2238 * EXT4_GET_BLOCKS_DELALLOC_RESERVE so the delalloc accounting
2239 * variables are updated after the blocks have been allocated.
2242 map
.m_len
= max_blocks
;
2243 get_blocks_flags
= EXT4_GET_BLOCKS_CREATE
;
2244 if (ext4_should_dioread_nolock(mpd
->inode
))
2245 get_blocks_flags
|= EXT4_GET_BLOCKS_IO_CREATE_EXT
;
2246 if (mpd
->b_state
& (1 << BH_Delay
))
2247 get_blocks_flags
|= EXT4_GET_BLOCKS_DELALLOC_RESERVE
;
2249 blks
= ext4_map_blocks(handle
, mpd
->inode
, &map
, get_blocks_flags
);
2251 struct super_block
*sb
= mpd
->inode
->i_sb
;
2255 * If get block returns EAGAIN or ENOSPC and there
2256 * appears to be free blocks we will call
2257 * ext4_writepage() for all of the pages which will
2258 * just redirty the pages.
2263 if (err
== -ENOSPC
&&
2264 ext4_count_free_blocks(sb
)) {
2270 * get block failure will cause us to loop in
2271 * writepages, because a_ops->writepage won't be able
2272 * to make progress. The page will be redirtied by
2273 * writepage and writepages will again try to write
2276 if (!(EXT4_SB(sb
)->s_mount_flags
& EXT4_MF_FS_ABORTED
)) {
2277 ext4_msg(sb
, KERN_CRIT
,
2278 "delayed block allocation failed for inode %lu "
2279 "at logical offset %llu with max blocks %zd "
2280 "with error %d", mpd
->inode
->i_ino
,
2281 (unsigned long long) next
,
2282 mpd
->b_size
>> mpd
->inode
->i_blkbits
, err
);
2283 ext4_msg(sb
, KERN_CRIT
,
2284 "This should not happen!! Data will be lost\n");
2286 ext4_print_free_blocks(mpd
->inode
);
2288 /* invalidate all the pages */
2289 ext4_da_block_invalidatepages(mpd
, next
,
2290 mpd
->b_size
>> mpd
->inode
->i_blkbits
);
2295 if (map
.m_flags
& EXT4_MAP_NEW
) {
2296 struct block_device
*bdev
= mpd
->inode
->i_sb
->s_bdev
;
2299 for (i
= 0; i
< map
.m_len
; i
++)
2300 unmap_underlying_metadata(bdev
, map
.m_pblk
+ i
);
2304 * If blocks are delayed marked, we need to
2305 * put actual blocknr and drop delayed bit
2307 if ((mpd
->b_state
& (1 << BH_Delay
)) ||
2308 (mpd
->b_state
& (1 << BH_Unwritten
)))
2309 mpage_put_bnr_to_bhs(mpd
, &map
);
2311 if (ext4_should_order_data(mpd
->inode
)) {
2312 err
= ext4_jbd2_file_inode(handle
, mpd
->inode
);
2314 /* This only happens if the journal is aborted */
2319 * Update on-disk size along with block allocation.
2321 disksize
= ((loff_t
) next
+ blks
) << mpd
->inode
->i_blkbits
;
2322 if (disksize
> i_size_read(mpd
->inode
))
2323 disksize
= i_size_read(mpd
->inode
);
2324 if (disksize
> EXT4_I(mpd
->inode
)->i_disksize
) {
2325 ext4_update_i_disksize(mpd
->inode
, disksize
);
2326 err
= ext4_mark_inode_dirty(handle
, mpd
->inode
);
2328 ext4_error(mpd
->inode
->i_sb
,
2329 "Failed to mark inode %lu dirty",
2334 mpage_da_submit_io(mpd
);
2338 #define BH_FLAGS ((1 << BH_Uptodate) | (1 << BH_Mapped) | \
2339 (1 << BH_Delay) | (1 << BH_Unwritten))
2342 * mpage_add_bh_to_extent - try to add one more block to extent of blocks
2344 * @mpd->lbh - extent of blocks
2345 * @logical - logical number of the block in the file
2346 * @bh - bh of the block (used to access block's state)
2348 * the function is used to collect contig. blocks in same state
2350 static void mpage_add_bh_to_extent(struct mpage_da_data
*mpd
,
2351 sector_t logical
, size_t b_size
,
2352 unsigned long b_state
)
2355 int nrblocks
= mpd
->b_size
>> mpd
->inode
->i_blkbits
;
2358 * XXX Don't go larger than mballoc is willing to allocate
2359 * This is a stopgap solution. We eventually need to fold
2360 * mpage_da_submit_io() into this function and then call
2361 * ext4_map_blocks() multiple times in a loop
2363 if (nrblocks
>= 8*1024*1024/mpd
->inode
->i_sb
->s_blocksize
)
2366 /* check if thereserved journal credits might overflow */
2367 if (!(ext4_test_inode_flag(mpd
->inode
, EXT4_INODE_EXTENTS
))) {
2368 if (nrblocks
>= EXT4_MAX_TRANS_DATA
) {
2370 * With non-extent format we are limited by the journal
2371 * credit available. Total credit needed to insert
2372 * nrblocks contiguous blocks is dependent on the
2373 * nrblocks. So limit nrblocks.
2376 } else if ((nrblocks
+ (b_size
>> mpd
->inode
->i_blkbits
)) >
2377 EXT4_MAX_TRANS_DATA
) {
2379 * Adding the new buffer_head would make it cross the
2380 * allowed limit for which we have journal credit
2381 * reserved. So limit the new bh->b_size
2383 b_size
= (EXT4_MAX_TRANS_DATA
- nrblocks
) <<
2384 mpd
->inode
->i_blkbits
;
2385 /* we will do mpage_da_submit_io in the next loop */
2389 * First block in the extent
2391 if (mpd
->b_size
== 0) {
2392 mpd
->b_blocknr
= logical
;
2393 mpd
->b_size
= b_size
;
2394 mpd
->b_state
= b_state
& BH_FLAGS
;
2398 next
= mpd
->b_blocknr
+ nrblocks
;
2400 * Can we merge the block to our big extent?
2402 if (logical
== next
&& (b_state
& BH_FLAGS
) == mpd
->b_state
) {
2403 mpd
->b_size
+= b_size
;
2409 * We couldn't merge the block to our extent, so we
2410 * need to flush current extent and start new one
2412 mpage_da_map_and_submit(mpd
);
2416 static int ext4_bh_delay_or_unwritten(handle_t
*handle
, struct buffer_head
*bh
)
2418 return (buffer_delay(bh
) || buffer_unwritten(bh
)) && buffer_dirty(bh
);
2422 * __mpage_da_writepage - finds extent of pages and blocks
2424 * @page: page to consider
2425 * @wbc: not used, we just follow rules
2428 * The function finds extents of pages and scan them for all blocks.
2430 static int __mpage_da_writepage(struct page
*page
,
2431 struct writeback_control
*wbc
, void *data
)
2433 struct mpage_da_data
*mpd
= data
;
2434 struct inode
*inode
= mpd
->inode
;
2435 struct buffer_head
*bh
, *head
;
2439 * Can we merge this page to current extent?
2441 if (mpd
->next_page
!= page
->index
) {
2443 * Nope, we can't. So, we map non-allocated blocks
2444 * and start IO on them
2446 if (mpd
->next_page
!= mpd
->first_page
) {
2447 mpage_da_map_and_submit(mpd
);
2449 * skip rest of the page in the page_vec
2451 redirty_page_for_writepage(wbc
, page
);
2453 return MPAGE_DA_EXTENT_TAIL
;
2457 * Start next extent of pages ...
2459 mpd
->first_page
= page
->index
;
2469 mpd
->next_page
= page
->index
+ 1;
2470 logical
= (sector_t
) page
->index
<<
2471 (PAGE_CACHE_SHIFT
- inode
->i_blkbits
);
2473 if (!page_has_buffers(page
)) {
2474 mpage_add_bh_to_extent(mpd
, logical
, PAGE_CACHE_SIZE
,
2475 (1 << BH_Dirty
) | (1 << BH_Uptodate
));
2477 return MPAGE_DA_EXTENT_TAIL
;
2480 * Page with regular buffer heads, just add all dirty ones
2482 head
= page_buffers(page
);
2485 BUG_ON(buffer_locked(bh
));
2487 * We need to try to allocate
2488 * unmapped blocks in the same page.
2489 * Otherwise we won't make progress
2490 * with the page in ext4_writepage
2492 if (ext4_bh_delay_or_unwritten(NULL
, bh
)) {
2493 mpage_add_bh_to_extent(mpd
, logical
,
2497 return MPAGE_DA_EXTENT_TAIL
;
2498 } else if (buffer_dirty(bh
) && (buffer_mapped(bh
))) {
2500 * mapped dirty buffer. We need to update
2501 * the b_state because we look at
2502 * b_state in mpage_da_map_blocks. We don't
2503 * update b_size because if we find an
2504 * unmapped buffer_head later we need to
2505 * use the b_state flag of that buffer_head.
2507 if (mpd
->b_size
== 0)
2508 mpd
->b_state
= bh
->b_state
& BH_FLAGS
;
2511 } while ((bh
= bh
->b_this_page
) != head
);
2518 * This is a special get_blocks_t callback which is used by
2519 * ext4_da_write_begin(). It will either return mapped block or
2520 * reserve space for a single block.
2522 * For delayed buffer_head we have BH_Mapped, BH_New, BH_Delay set.
2523 * We also have b_blocknr = -1 and b_bdev initialized properly
2525 * For unwritten buffer_head we have BH_Mapped, BH_New, BH_Unwritten set.
2526 * We also have b_blocknr = physicalblock mapping unwritten extent and b_bdev
2527 * initialized properly.
2529 static int ext4_da_get_block_prep(struct inode
*inode
, sector_t iblock
,
2530 struct buffer_head
*bh
, int create
)
2532 struct ext4_map_blocks map
;
2534 sector_t invalid_block
= ~((sector_t
) 0xffff);
2536 if (invalid_block
< ext4_blocks_count(EXT4_SB(inode
->i_sb
)->s_es
))
2539 BUG_ON(create
== 0);
2540 BUG_ON(bh
->b_size
!= inode
->i_sb
->s_blocksize
);
2542 map
.m_lblk
= iblock
;
2546 * first, we need to know whether the block is allocated already
2547 * preallocated blocks are unmapped but should treated
2548 * the same as allocated blocks.
2550 ret
= ext4_map_blocks(NULL
, inode
, &map
, 0);
2554 if (buffer_delay(bh
))
2555 return 0; /* Not sure this could or should happen */
2557 * XXX: __block_prepare_write() unmaps passed block,
2560 ret
= ext4_da_reserve_space(inode
, iblock
);
2562 /* not enough space to reserve */
2565 map_bh(bh
, inode
->i_sb
, invalid_block
);
2567 set_buffer_delay(bh
);
2571 map_bh(bh
, inode
->i_sb
, map
.m_pblk
);
2572 bh
->b_state
= (bh
->b_state
& ~EXT4_MAP_FLAGS
) | map
.m_flags
;
2574 if (buffer_unwritten(bh
)) {
2575 /* A delayed write to unwritten bh should be marked
2576 * new and mapped. Mapped ensures that we don't do
2577 * get_block multiple times when we write to the same
2578 * offset and new ensures that we do proper zero out
2579 * for partial write.
2582 set_buffer_mapped(bh
);
2588 * This function is used as a standard get_block_t calback function
2589 * when there is no desire to allocate any blocks. It is used as a
2590 * callback function for block_prepare_write() and block_write_full_page().
2591 * These functions should only try to map a single block at a time.
2593 * Since this function doesn't do block allocations even if the caller
2594 * requests it by passing in create=1, it is critically important that
2595 * any caller checks to make sure that any buffer heads are returned
2596 * by this function are either all already mapped or marked for
2597 * delayed allocation before calling block_write_full_page(). Otherwise,
2598 * b_blocknr could be left unitialized, and the page write functions will
2599 * be taken by surprise.
2601 static int noalloc_get_block_write(struct inode
*inode
, sector_t iblock
,
2602 struct buffer_head
*bh_result
, int create
)
2604 BUG_ON(bh_result
->b_size
!= inode
->i_sb
->s_blocksize
);
2605 return _ext4_get_block(inode
, iblock
, bh_result
, 0);
2608 static int bget_one(handle_t
*handle
, struct buffer_head
*bh
)
2614 static int bput_one(handle_t
*handle
, struct buffer_head
*bh
)
2620 static int __ext4_journalled_writepage(struct page
*page
,
2623 struct address_space
*mapping
= page
->mapping
;
2624 struct inode
*inode
= mapping
->host
;
2625 struct buffer_head
*page_bufs
;
2626 handle_t
*handle
= NULL
;
2630 page_bufs
= page_buffers(page
);
2632 walk_page_buffers(handle
, page_bufs
, 0, len
, NULL
, bget_one
);
2633 /* As soon as we unlock the page, it can go away, but we have
2634 * references to buffers so we are safe */
2637 handle
= ext4_journal_start(inode
, ext4_writepage_trans_blocks(inode
));
2638 if (IS_ERR(handle
)) {
2639 ret
= PTR_ERR(handle
);
2643 ret
= walk_page_buffers(handle
, page_bufs
, 0, len
, NULL
,
2644 do_journal_get_write_access
);
2646 err
= walk_page_buffers(handle
, page_bufs
, 0, len
, NULL
,
2650 err
= ext4_journal_stop(handle
);
2654 walk_page_buffers(handle
, page_bufs
, 0, len
, NULL
, bput_one
);
2655 ext4_set_inode_state(inode
, EXT4_STATE_JDATA
);
2660 static int ext4_set_bh_endio(struct buffer_head
*bh
, struct inode
*inode
);
2661 static void ext4_end_io_buffer_write(struct buffer_head
*bh
, int uptodate
);
2664 * Note that we don't need to start a transaction unless we're journaling data
2665 * because we should have holes filled from ext4_page_mkwrite(). We even don't
2666 * need to file the inode to the transaction's list in ordered mode because if
2667 * we are writing back data added by write(), the inode is already there and if
2668 * we are writing back data modified via mmap(), noone guarantees in which
2669 * transaction the data will hit the disk. In case we are journaling data, we
2670 * cannot start transaction directly because transaction start ranks above page
2671 * lock so we have to do some magic.
2673 * This function can get called via...
2674 * - ext4_da_writepages after taking page lock (have journal handle)
2675 * - journal_submit_inode_data_buffers (no journal handle)
2676 * - shrink_page_list via pdflush (no journal handle)
2677 * - grab_page_cache when doing write_begin (have journal handle)
2679 * We don't do any block allocation in this function. If we have page with
2680 * multiple blocks we need to write those buffer_heads that are mapped. This
2681 * is important for mmaped based write. So if we do with blocksize 1K
2682 * truncate(f, 1024);
2683 * a = mmap(f, 0, 4096);
2685 * truncate(f, 4096);
2686 * we have in the page first buffer_head mapped via page_mkwrite call back
2687 * but other bufer_heads would be unmapped but dirty(dirty done via the
2688 * do_wp_page). So writepage should write the first block. If we modify
2689 * the mmap area beyond 1024 we will again get a page_fault and the
2690 * page_mkwrite callback will do the block allocation and mark the
2691 * buffer_heads mapped.
2693 * We redirty the page if we have any buffer_heads that is either delay or
2694 * unwritten in the page.
2696 * We can get recursively called as show below.
2698 * ext4_writepage() -> kmalloc() -> __alloc_pages() -> page_launder() ->
2701 * But since we don't do any block allocation we should not deadlock.
2702 * Page also have the dirty flag cleared so we don't get recurive page_lock.
2704 static int ext4_writepage(struct page
*page
,
2705 struct writeback_control
*wbc
)
2707 int ret
= 0, commit_write
= 0;
2710 struct buffer_head
*page_bufs
= NULL
;
2711 struct inode
*inode
= page
->mapping
->host
;
2713 trace_ext4_writepage(inode
, page
);
2714 size
= i_size_read(inode
);
2715 if (page
->index
== size
>> PAGE_CACHE_SHIFT
)
2716 len
= size
& ~PAGE_CACHE_MASK
;
2718 len
= PAGE_CACHE_SIZE
;
2721 * If the page does not have buffers (for whatever reason),
2722 * try to create them using block_prepare_write. If this
2723 * fails, redirty the page and move on.
2725 if (!page_buffers(page
)) {
2726 if (block_prepare_write(page
, 0, len
,
2727 noalloc_get_block_write
)) {
2729 redirty_page_for_writepage(wbc
, page
);
2735 page_bufs
= page_buffers(page
);
2736 if (walk_page_buffers(NULL
, page_bufs
, 0, len
, NULL
,
2737 ext4_bh_delay_or_unwritten
)) {
2739 * We don't want to do block allocation So redirty the
2740 * page and return We may reach here when we do a
2741 * journal commit via
2742 * journal_submit_inode_data_buffers. If we don't
2743 * have mapping block we just ignore them. We can also
2744 * reach here via shrink_page_list
2749 /* now mark the buffer_heads as dirty and uptodate */
2750 block_commit_write(page
, 0, len
);
2752 if (PageChecked(page
) && ext4_should_journal_data(inode
)) {
2754 * It's mmapped pagecache. Add buffers and journal it. There
2755 * doesn't seem much point in redirtying the page here.
2757 ClearPageChecked(page
);
2758 return __ext4_journalled_writepage(page
, len
);
2761 if (buffer_uninit(page_bufs
)) {
2762 ext4_set_bh_endio(page_bufs
, inode
);
2763 ret
= block_write_full_page_endio(page
, noalloc_get_block_write
,
2764 wbc
, ext4_end_io_buffer_write
);
2766 ret
= block_write_full_page(page
, noalloc_get_block_write
,
2773 * This is called via ext4_da_writepages() to
2774 * calulate the total number of credits to reserve to fit
2775 * a single extent allocation into a single transaction,
2776 * ext4_da_writpeages() will loop calling this before
2777 * the block allocation.
2780 static int ext4_da_writepages_trans_blocks(struct inode
*inode
)
2782 int max_blocks
= EXT4_I(inode
)->i_reserved_data_blocks
;
2785 * With non-extent format the journal credit needed to
2786 * insert nrblocks contiguous block is dependent on
2787 * number of contiguous block. So we will limit
2788 * number of contiguous block to a sane value
2790 if (!(ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
)) &&
2791 (max_blocks
> EXT4_MAX_TRANS_DATA
))
2792 max_blocks
= EXT4_MAX_TRANS_DATA
;
2794 return ext4_chunk_trans_blocks(inode
, max_blocks
);
2798 * write_cache_pages_da - walk the list of dirty pages of the given
2799 * address space and call the callback function (which usually writes
2802 * This is a forked version of write_cache_pages(). Differences:
2803 * Range cyclic is ignored.
2804 * no_nrwrite_index_update is always presumed true
2806 static int write_cache_pages_da(struct address_space
*mapping
,
2807 struct writeback_control
*wbc
,
2808 struct mpage_da_data
*mpd
)
2812 struct pagevec pvec
;
2815 pgoff_t end
; /* Inclusive */
2816 long nr_to_write
= wbc
->nr_to_write
;
2818 pagevec_init(&pvec
, 0);
2819 index
= wbc
->range_start
>> PAGE_CACHE_SHIFT
;
2820 end
= wbc
->range_end
>> PAGE_CACHE_SHIFT
;
2822 while (!done
&& (index
<= end
)) {
2825 nr_pages
= pagevec_lookup_tag(&pvec
, mapping
, &index
,
2826 PAGECACHE_TAG_DIRTY
,
2827 min(end
- index
, (pgoff_t
)PAGEVEC_SIZE
-1) + 1);
2831 for (i
= 0; i
< nr_pages
; i
++) {
2832 struct page
*page
= pvec
.pages
[i
];
2835 * At this point, the page may be truncated or
2836 * invalidated (changing page->mapping to NULL), or
2837 * even swizzled back from swapper_space to tmpfs file
2838 * mapping. However, page->index will not change
2839 * because we have a reference on the page.
2841 if (page
->index
> end
) {
2849 * Page truncated or invalidated. We can freely skip it
2850 * then, even for data integrity operations: the page
2851 * has disappeared concurrently, so there could be no
2852 * real expectation of this data interity operation
2853 * even if there is now a new, dirty page at the same
2854 * pagecache address.
2856 if (unlikely(page
->mapping
!= mapping
)) {
2862 if (!PageDirty(page
)) {
2863 /* someone wrote it for us */
2864 goto continue_unlock
;
2867 if (PageWriteback(page
)) {
2868 if (wbc
->sync_mode
!= WB_SYNC_NONE
)
2869 wait_on_page_writeback(page
);
2871 goto continue_unlock
;
2874 BUG_ON(PageWriteback(page
));
2875 if (!clear_page_dirty_for_io(page
))
2876 goto continue_unlock
;
2878 ret
= __mpage_da_writepage(page
, wbc
, mpd
);
2879 if (unlikely(ret
)) {
2880 if (ret
== AOP_WRITEPAGE_ACTIVATE
) {
2889 if (nr_to_write
> 0) {
2891 if (nr_to_write
== 0 &&
2892 wbc
->sync_mode
== WB_SYNC_NONE
) {
2894 * We stop writing back only if we are
2895 * not doing integrity sync. In case of
2896 * integrity sync we have to keep going
2897 * because someone may be concurrently
2898 * dirtying pages, and we might have
2899 * synced a lot of newly appeared dirty
2900 * pages, but have not synced all of the
2908 pagevec_release(&pvec
);
2915 static int ext4_da_writepages(struct address_space
*mapping
,
2916 struct writeback_control
*wbc
)
2919 int range_whole
= 0;
2920 handle_t
*handle
= NULL
;
2921 struct mpage_da_data mpd
;
2922 struct inode
*inode
= mapping
->host
;
2923 int pages_written
= 0;
2925 unsigned int max_pages
;
2926 int range_cyclic
, cycled
= 1, io_done
= 0;
2927 int needed_blocks
, ret
= 0;
2928 long desired_nr_to_write
, nr_to_writebump
= 0;
2929 loff_t range_start
= wbc
->range_start
;
2930 struct ext4_sb_info
*sbi
= EXT4_SB(mapping
->host
->i_sb
);
2932 trace_ext4_da_writepages(inode
, wbc
);
2935 * No pages to write? This is mainly a kludge to avoid starting
2936 * a transaction for special inodes like journal inode on last iput()
2937 * because that could violate lock ordering on umount
2939 if (!mapping
->nrpages
|| !mapping_tagged(mapping
, PAGECACHE_TAG_DIRTY
))
2943 * If the filesystem has aborted, it is read-only, so return
2944 * right away instead of dumping stack traces later on that
2945 * will obscure the real source of the problem. We test
2946 * EXT4_MF_FS_ABORTED instead of sb->s_flag's MS_RDONLY because
2947 * the latter could be true if the filesystem is mounted
2948 * read-only, and in that case, ext4_da_writepages should
2949 * *never* be called, so if that ever happens, we would want
2952 if (unlikely(sbi
->s_mount_flags
& EXT4_MF_FS_ABORTED
))
2955 if (wbc
->range_start
== 0 && wbc
->range_end
== LLONG_MAX
)
2958 range_cyclic
= wbc
->range_cyclic
;
2959 if (wbc
->range_cyclic
) {
2960 index
= mapping
->writeback_index
;
2963 wbc
->range_start
= index
<< PAGE_CACHE_SHIFT
;
2964 wbc
->range_end
= LLONG_MAX
;
2965 wbc
->range_cyclic
= 0;
2967 index
= wbc
->range_start
>> PAGE_CACHE_SHIFT
;
2970 * This works around two forms of stupidity. The first is in
2971 * the writeback code, which caps the maximum number of pages
2972 * written to be 1024 pages. This is wrong on multiple
2973 * levels; different architectues have a different page size,
2974 * which changes the maximum amount of data which gets
2975 * written. Secondly, 4 megabytes is way too small. XFS
2976 * forces this value to be 16 megabytes by multiplying
2977 * nr_to_write parameter by four, and then relies on its
2978 * allocator to allocate larger extents to make them
2979 * contiguous. Unfortunately this brings us to the second
2980 * stupidity, which is that ext4's mballoc code only allocates
2981 * at most 2048 blocks. So we force contiguous writes up to
2982 * the number of dirty blocks in the inode, or
2983 * sbi->max_writeback_mb_bump whichever is smaller.
2985 max_pages
= sbi
->s_max_writeback_mb_bump
<< (20 - PAGE_CACHE_SHIFT
);
2986 if (!range_cyclic
&& range_whole
) {
2987 if (wbc
->nr_to_write
== LONG_MAX
)
2988 desired_nr_to_write
= wbc
->nr_to_write
;
2990 desired_nr_to_write
= wbc
->nr_to_write
* 8;
2992 desired_nr_to_write
= ext4_num_dirty_pages(inode
, index
,
2994 if (desired_nr_to_write
> max_pages
)
2995 desired_nr_to_write
= max_pages
;
2997 if (wbc
->nr_to_write
< desired_nr_to_write
) {
2998 nr_to_writebump
= desired_nr_to_write
- wbc
->nr_to_write
;
2999 wbc
->nr_to_write
= desired_nr_to_write
;
3003 mpd
.inode
= mapping
->host
;
3005 pages_skipped
= wbc
->pages_skipped
;
3008 while (!ret
&& wbc
->nr_to_write
> 0) {
3011 * we insert one extent at a time. So we need
3012 * credit needed for single extent allocation.
3013 * journalled mode is currently not supported
3016 BUG_ON(ext4_should_journal_data(inode
));
3017 needed_blocks
= ext4_da_writepages_trans_blocks(inode
);
3019 /* start a new transaction*/
3020 handle
= ext4_journal_start(inode
, needed_blocks
);
3021 if (IS_ERR(handle
)) {
3022 ret
= PTR_ERR(handle
);
3023 ext4_msg(inode
->i_sb
, KERN_CRIT
, "%s: jbd2_start: "
3024 "%ld pages, ino %lu; err %d", __func__
,
3025 wbc
->nr_to_write
, inode
->i_ino
, ret
);
3026 goto out_writepages
;
3030 * Now call __mpage_da_writepage to find the next
3031 * contiguous region of logical blocks that need
3032 * blocks to be allocated by ext4. We don't actually
3033 * submit the blocks for I/O here, even though
3034 * write_cache_pages thinks it will, and will set the
3035 * pages as clean for write before calling
3036 * __mpage_da_writepage().
3044 mpd
.pages_written
= 0;
3046 ret
= write_cache_pages_da(mapping
, wbc
, &mpd
);
3048 * If we have a contiguous extent of pages and we
3049 * haven't done the I/O yet, map the blocks and submit
3052 if (!mpd
.io_done
&& mpd
.next_page
!= mpd
.first_page
) {
3053 mpage_da_map_and_submit(&mpd
);
3054 ret
= MPAGE_DA_EXTENT_TAIL
;
3056 trace_ext4_da_write_pages(inode
, &mpd
);
3057 wbc
->nr_to_write
-= mpd
.pages_written
;
3059 ext4_journal_stop(handle
);
3061 if ((mpd
.retval
== -ENOSPC
) && sbi
->s_journal
) {
3062 /* commit the transaction which would
3063 * free blocks released in the transaction
3066 jbd2_journal_force_commit_nested(sbi
->s_journal
);
3067 wbc
->pages_skipped
= pages_skipped
;
3069 } else if (ret
== MPAGE_DA_EXTENT_TAIL
) {
3071 * got one extent now try with
3074 pages_written
+= mpd
.pages_written
;
3075 wbc
->pages_skipped
= pages_skipped
;
3078 } else if (wbc
->nr_to_write
)
3080 * There is no more writeout needed
3081 * or we requested for a noblocking writeout
3082 * and we found the device congested
3086 if (!io_done
&& !cycled
) {
3089 wbc
->range_start
= index
<< PAGE_CACHE_SHIFT
;
3090 wbc
->range_end
= mapping
->writeback_index
- 1;
3093 if (pages_skipped
!= wbc
->pages_skipped
)
3094 ext4_msg(inode
->i_sb
, KERN_CRIT
,
3095 "This should not happen leaving %s "
3096 "with nr_to_write = %ld ret = %d",
3097 __func__
, wbc
->nr_to_write
, ret
);
3100 index
+= pages_written
;
3101 wbc
->range_cyclic
= range_cyclic
;
3102 if (wbc
->range_cyclic
|| (range_whole
&& wbc
->nr_to_write
> 0))
3104 * set the writeback_index so that range_cyclic
3105 * mode will write it back later
3107 mapping
->writeback_index
= index
;
3110 wbc
->nr_to_write
-= nr_to_writebump
;
3111 wbc
->range_start
= range_start
;
3112 trace_ext4_da_writepages_result(inode
, wbc
, ret
, pages_written
);
3116 #define FALL_BACK_TO_NONDELALLOC 1
3117 static int ext4_nonda_switch(struct super_block
*sb
)
3119 s64 free_blocks
, dirty_blocks
;
3120 struct ext4_sb_info
*sbi
= EXT4_SB(sb
);
3123 * switch to non delalloc mode if we are running low
3124 * on free block. The free block accounting via percpu
3125 * counters can get slightly wrong with percpu_counter_batch getting
3126 * accumulated on each CPU without updating global counters
3127 * Delalloc need an accurate free block accounting. So switch
3128 * to non delalloc when we are near to error range.
3130 free_blocks
= percpu_counter_read_positive(&sbi
->s_freeblocks_counter
);
3131 dirty_blocks
= percpu_counter_read_positive(&sbi
->s_dirtyblocks_counter
);
3132 if (2 * free_blocks
< 3 * dirty_blocks
||
3133 free_blocks
< (dirty_blocks
+ EXT4_FREEBLOCKS_WATERMARK
)) {
3135 * free block count is less than 150% of dirty blocks
3136 * or free blocks is less than watermark
3141 * Even if we don't switch but are nearing capacity,
3142 * start pushing delalloc when 1/2 of free blocks are dirty.
3144 if (free_blocks
< 2 * dirty_blocks
)
3145 writeback_inodes_sb_if_idle(sb
);
3150 static int ext4_da_write_begin(struct file
*file
, struct address_space
*mapping
,
3151 loff_t pos
, unsigned len
, unsigned flags
,
3152 struct page
**pagep
, void **fsdata
)
3154 int ret
, retries
= 0;
3157 struct inode
*inode
= mapping
->host
;
3160 index
= pos
>> PAGE_CACHE_SHIFT
;
3162 if (ext4_nonda_switch(inode
->i_sb
)) {
3163 *fsdata
= (void *)FALL_BACK_TO_NONDELALLOC
;
3164 return ext4_write_begin(file
, mapping
, pos
,
3165 len
, flags
, pagep
, fsdata
);
3167 *fsdata
= (void *)0;
3168 trace_ext4_da_write_begin(inode
, pos
, len
, flags
);
3171 * With delayed allocation, we don't log the i_disksize update
3172 * if there is delayed block allocation. But we still need
3173 * to journalling the i_disksize update if writes to the end
3174 * of file which has an already mapped buffer.
3176 handle
= ext4_journal_start(inode
, 1);
3177 if (IS_ERR(handle
)) {
3178 ret
= PTR_ERR(handle
);
3181 /* We cannot recurse into the filesystem as the transaction is already
3183 flags
|= AOP_FLAG_NOFS
;
3185 page
= grab_cache_page_write_begin(mapping
, index
, flags
);
3187 ext4_journal_stop(handle
);
3193 ret
= __block_write_begin(page
, pos
, len
, ext4_da_get_block_prep
);
3196 ext4_journal_stop(handle
);
3197 page_cache_release(page
);
3199 * block_write_begin may have instantiated a few blocks
3200 * outside i_size. Trim these off again. Don't need
3201 * i_size_read because we hold i_mutex.
3203 if (pos
+ len
> inode
->i_size
)
3204 ext4_truncate_failed_write(inode
);
3207 if (ret
== -ENOSPC
&& ext4_should_retry_alloc(inode
->i_sb
, &retries
))
3214 * Check if we should update i_disksize
3215 * when write to the end of file but not require block allocation
3217 static int ext4_da_should_update_i_disksize(struct page
*page
,
3218 unsigned long offset
)
3220 struct buffer_head
*bh
;
3221 struct inode
*inode
= page
->mapping
->host
;
3225 bh
= page_buffers(page
);
3226 idx
= offset
>> inode
->i_blkbits
;
3228 for (i
= 0; i
< idx
; i
++)
3229 bh
= bh
->b_this_page
;
3231 if (!buffer_mapped(bh
) || (buffer_delay(bh
)) || buffer_unwritten(bh
))
3236 static int ext4_da_write_end(struct file
*file
,
3237 struct address_space
*mapping
,
3238 loff_t pos
, unsigned len
, unsigned copied
,
3239 struct page
*page
, void *fsdata
)
3241 struct inode
*inode
= mapping
->host
;
3243 handle_t
*handle
= ext4_journal_current_handle();
3245 unsigned long start
, end
;
3246 int write_mode
= (int)(unsigned long)fsdata
;
3248 if (write_mode
== FALL_BACK_TO_NONDELALLOC
) {
3249 if (ext4_should_order_data(inode
)) {
3250 return ext4_ordered_write_end(file
, mapping
, pos
,
3251 len
, copied
, page
, fsdata
);
3252 } else if (ext4_should_writeback_data(inode
)) {
3253 return ext4_writeback_write_end(file
, mapping
, pos
,
3254 len
, copied
, page
, fsdata
);
3260 trace_ext4_da_write_end(inode
, pos
, len
, copied
);
3261 start
= pos
& (PAGE_CACHE_SIZE
- 1);
3262 end
= start
+ copied
- 1;
3265 * generic_write_end() will run mark_inode_dirty() if i_size
3266 * changes. So let's piggyback the i_disksize mark_inode_dirty
3270 new_i_size
= pos
+ copied
;
3271 if (new_i_size
> EXT4_I(inode
)->i_disksize
) {
3272 if (ext4_da_should_update_i_disksize(page
, end
)) {
3273 down_write(&EXT4_I(inode
)->i_data_sem
);
3274 if (new_i_size
> EXT4_I(inode
)->i_disksize
) {
3276 * Updating i_disksize when extending file
3277 * without needing block allocation
3279 if (ext4_should_order_data(inode
))
3280 ret
= ext4_jbd2_file_inode(handle
,
3283 EXT4_I(inode
)->i_disksize
= new_i_size
;
3285 up_write(&EXT4_I(inode
)->i_data_sem
);
3286 /* We need to mark inode dirty even if
3287 * new_i_size is less that inode->i_size
3288 * bu greater than i_disksize.(hint delalloc)
3290 ext4_mark_inode_dirty(handle
, inode
);
3293 ret2
= generic_write_end(file
, mapping
, pos
, len
, copied
,
3298 ret2
= ext4_journal_stop(handle
);
3302 return ret
? ret
: copied
;
3305 static void ext4_da_invalidatepage(struct page
*page
, unsigned long offset
)
3308 * Drop reserved blocks
3310 BUG_ON(!PageLocked(page
));
3311 if (!page_has_buffers(page
))
3314 ext4_da_page_release_reservation(page
, offset
);
3317 ext4_invalidatepage(page
, offset
);
3323 * Force all delayed allocation blocks to be allocated for a given inode.
3325 int ext4_alloc_da_blocks(struct inode
*inode
)
3327 trace_ext4_alloc_da_blocks(inode
);
3329 if (!EXT4_I(inode
)->i_reserved_data_blocks
&&
3330 !EXT4_I(inode
)->i_reserved_meta_blocks
)
3334 * We do something simple for now. The filemap_flush() will
3335 * also start triggering a write of the data blocks, which is
3336 * not strictly speaking necessary (and for users of
3337 * laptop_mode, not even desirable). However, to do otherwise
3338 * would require replicating code paths in:
3340 * ext4_da_writepages() ->
3341 * write_cache_pages() ---> (via passed in callback function)
3342 * __mpage_da_writepage() -->
3343 * mpage_add_bh_to_extent()
3344 * mpage_da_map_blocks()
3346 * The problem is that write_cache_pages(), located in
3347 * mm/page-writeback.c, marks pages clean in preparation for
3348 * doing I/O, which is not desirable if we're not planning on
3351 * We could call write_cache_pages(), and then redirty all of
3352 * the pages by calling redirty_page_for_writeback() but that
3353 * would be ugly in the extreme. So instead we would need to
3354 * replicate parts of the code in the above functions,
3355 * simplifying them becuase we wouldn't actually intend to
3356 * write out the pages, but rather only collect contiguous
3357 * logical block extents, call the multi-block allocator, and
3358 * then update the buffer heads with the block allocations.
3360 * For now, though, we'll cheat by calling filemap_flush(),
3361 * which will map the blocks, and start the I/O, but not
3362 * actually wait for the I/O to complete.
3364 return filemap_flush(inode
->i_mapping
);
3368 * bmap() is special. It gets used by applications such as lilo and by
3369 * the swapper to find the on-disk block of a specific piece of data.
3371 * Naturally, this is dangerous if the block concerned is still in the
3372 * journal. If somebody makes a swapfile on an ext4 data-journaling
3373 * filesystem and enables swap, then they may get a nasty shock when the
3374 * data getting swapped to that swapfile suddenly gets overwritten by
3375 * the original zero's written out previously to the journal and
3376 * awaiting writeback in the kernel's buffer cache.
3378 * So, if we see any bmap calls here on a modified, data-journaled file,
3379 * take extra steps to flush any blocks which might be in the cache.
3381 static sector_t
ext4_bmap(struct address_space
*mapping
, sector_t block
)
3383 struct inode
*inode
= mapping
->host
;
3387 if (mapping_tagged(mapping
, PAGECACHE_TAG_DIRTY
) &&
3388 test_opt(inode
->i_sb
, DELALLOC
)) {
3390 * With delalloc we want to sync the file
3391 * so that we can make sure we allocate
3394 filemap_write_and_wait(mapping
);
3397 if (EXT4_JOURNAL(inode
) &&
3398 ext4_test_inode_state(inode
, EXT4_STATE_JDATA
)) {
3400 * This is a REALLY heavyweight approach, but the use of
3401 * bmap on dirty files is expected to be extremely rare:
3402 * only if we run lilo or swapon on a freshly made file
3403 * do we expect this to happen.
3405 * (bmap requires CAP_SYS_RAWIO so this does not
3406 * represent an unprivileged user DOS attack --- we'd be
3407 * in trouble if mortal users could trigger this path at
3410 * NB. EXT4_STATE_JDATA is not set on files other than
3411 * regular files. If somebody wants to bmap a directory
3412 * or symlink and gets confused because the buffer
3413 * hasn't yet been flushed to disk, they deserve
3414 * everything they get.
3417 ext4_clear_inode_state(inode
, EXT4_STATE_JDATA
);
3418 journal
= EXT4_JOURNAL(inode
);
3419 jbd2_journal_lock_updates(journal
);
3420 err
= jbd2_journal_flush(journal
);
3421 jbd2_journal_unlock_updates(journal
);
3427 return generic_block_bmap(mapping
, block
, ext4_get_block
);
3430 static int ext4_readpage(struct file
*file
, struct page
*page
)
3432 return mpage_readpage(page
, ext4_get_block
);
3436 ext4_readpages(struct file
*file
, struct address_space
*mapping
,
3437 struct list_head
*pages
, unsigned nr_pages
)
3439 return mpage_readpages(mapping
, pages
, nr_pages
, ext4_get_block
);
3442 static void ext4_free_io_end(ext4_io_end_t
*io
)
3451 static void ext4_invalidatepage_free_endio(struct page
*page
, unsigned long offset
)
3453 struct buffer_head
*head
, *bh
;
3454 unsigned int curr_off
= 0;
3456 if (!page_has_buffers(page
))
3458 head
= bh
= page_buffers(page
);
3460 if (offset
<= curr_off
&& test_clear_buffer_uninit(bh
)
3462 ext4_free_io_end(bh
->b_private
);
3463 bh
->b_private
= NULL
;
3464 bh
->b_end_io
= NULL
;
3466 curr_off
= curr_off
+ bh
->b_size
;
3467 bh
= bh
->b_this_page
;
3468 } while (bh
!= head
);
3471 static void ext4_invalidatepage(struct page
*page
, unsigned long offset
)
3473 journal_t
*journal
= EXT4_JOURNAL(page
->mapping
->host
);
3476 * free any io_end structure allocated for buffers to be discarded
3478 if (ext4_should_dioread_nolock(page
->mapping
->host
))
3479 ext4_invalidatepage_free_endio(page
, offset
);
3481 * If it's a full truncate we just forget about the pending dirtying
3484 ClearPageChecked(page
);
3487 jbd2_journal_invalidatepage(journal
, page
, offset
);
3489 block_invalidatepage(page
, offset
);
3492 static int ext4_releasepage(struct page
*page
, gfp_t wait
)
3494 journal_t
*journal
= EXT4_JOURNAL(page
->mapping
->host
);
3496 WARN_ON(PageChecked(page
));
3497 if (!page_has_buffers(page
))
3500 return jbd2_journal_try_to_free_buffers(journal
, page
, wait
);
3502 return try_to_free_buffers(page
);
3506 * O_DIRECT for ext3 (or indirect map) based files
3508 * If the O_DIRECT write will extend the file then add this inode to the
3509 * orphan list. So recovery will truncate it back to the original size
3510 * if the machine crashes during the write.
3512 * If the O_DIRECT write is intantiating holes inside i_size and the machine
3513 * crashes then stale disk data _may_ be exposed inside the file. But current
3514 * VFS code falls back into buffered path in that case so we are safe.
3516 static ssize_t
ext4_ind_direct_IO(int rw
, struct kiocb
*iocb
,
3517 const struct iovec
*iov
, loff_t offset
,
3518 unsigned long nr_segs
)
3520 struct file
*file
= iocb
->ki_filp
;
3521 struct inode
*inode
= file
->f_mapping
->host
;
3522 struct ext4_inode_info
*ei
= EXT4_I(inode
);
3526 size_t count
= iov_length(iov
, nr_segs
);
3530 loff_t final_size
= offset
+ count
;
3532 if (final_size
> inode
->i_size
) {
3533 /* Credits for sb + inode write */
3534 handle
= ext4_journal_start(inode
, 2);
3535 if (IS_ERR(handle
)) {
3536 ret
= PTR_ERR(handle
);
3539 ret
= ext4_orphan_add(handle
, inode
);
3541 ext4_journal_stop(handle
);
3545 ei
->i_disksize
= inode
->i_size
;
3546 ext4_journal_stop(handle
);
3551 if (rw
== READ
&& ext4_should_dioread_nolock(inode
))
3552 ret
= __blockdev_direct_IO(rw
, iocb
, inode
,
3553 inode
->i_sb
->s_bdev
, iov
,
3555 ext4_get_block
, NULL
, NULL
, 0);
3557 ret
= blockdev_direct_IO(rw
, iocb
, inode
,
3558 inode
->i_sb
->s_bdev
, iov
,
3560 ext4_get_block
, NULL
);
3562 if (unlikely((rw
& WRITE
) && ret
< 0)) {
3563 loff_t isize
= i_size_read(inode
);
3564 loff_t end
= offset
+ iov_length(iov
, nr_segs
);
3567 vmtruncate(inode
, isize
);
3570 if (ret
== -ENOSPC
&& ext4_should_retry_alloc(inode
->i_sb
, &retries
))
3576 /* Credits for sb + inode write */
3577 handle
= ext4_journal_start(inode
, 2);
3578 if (IS_ERR(handle
)) {
3579 /* This is really bad luck. We've written the data
3580 * but cannot extend i_size. Bail out and pretend
3581 * the write failed... */
3582 ret
= PTR_ERR(handle
);
3584 ext4_orphan_del(NULL
, inode
);
3589 ext4_orphan_del(handle
, inode
);
3591 loff_t end
= offset
+ ret
;
3592 if (end
> inode
->i_size
) {
3593 ei
->i_disksize
= end
;
3594 i_size_write(inode
, end
);
3596 * We're going to return a positive `ret'
3597 * here due to non-zero-length I/O, so there's
3598 * no way of reporting error returns from
3599 * ext4_mark_inode_dirty() to userspace. So
3602 ext4_mark_inode_dirty(handle
, inode
);
3605 err
= ext4_journal_stop(handle
);
3614 * ext4_get_block used when preparing for a DIO write or buffer write.
3615 * We allocate an uinitialized extent if blocks haven't been allocated.
3616 * The extent will be converted to initialized after the IO is complete.
3618 static int ext4_get_block_write(struct inode
*inode
, sector_t iblock
,
3619 struct buffer_head
*bh_result
, int create
)
3621 ext4_debug("ext4_get_block_write: inode %lu, create flag %d\n",
3622 inode
->i_ino
, create
);
3623 return _ext4_get_block(inode
, iblock
, bh_result
,
3624 EXT4_GET_BLOCKS_IO_CREATE_EXT
);
3627 static void dump_completed_IO(struct inode
* inode
)
3630 struct list_head
*cur
, *before
, *after
;
3631 ext4_io_end_t
*io
, *io0
, *io1
;
3632 unsigned long flags
;
3634 if (list_empty(&EXT4_I(inode
)->i_completed_io_list
)){
3635 ext4_debug("inode %lu completed_io list is empty\n", inode
->i_ino
);
3639 ext4_debug("Dump inode %lu completed_io list \n", inode
->i_ino
);
3640 spin_lock_irqsave(&EXT4_I(inode
)->i_completed_io_lock
, flags
);
3641 list_for_each_entry(io
, &EXT4_I(inode
)->i_completed_io_list
, list
){
3644 io0
= container_of(before
, ext4_io_end_t
, list
);
3646 io1
= container_of(after
, ext4_io_end_t
, list
);
3648 ext4_debug("io 0x%p from inode %lu,prev 0x%p,next 0x%p\n",
3649 io
, inode
->i_ino
, io0
, io1
);
3651 spin_unlock_irqrestore(&EXT4_I(inode
)->i_completed_io_lock
, flags
);
3656 * check a range of space and convert unwritten extents to written.
3658 static int ext4_end_io_nolock(ext4_io_end_t
*io
)
3660 struct inode
*inode
= io
->inode
;
3661 loff_t offset
= io
->offset
;
3662 ssize_t size
= io
->size
;
3665 ext4_debug("ext4_end_io_nolock: io 0x%p from inode %lu,list->next 0x%p,"
3666 "list->prev 0x%p\n",
3667 io
, inode
->i_ino
, io
->list
.next
, io
->list
.prev
);
3669 if (list_empty(&io
->list
))
3672 if (io
->flag
!= EXT4_IO_UNWRITTEN
)
3675 ret
= ext4_convert_unwritten_extents(inode
, offset
, size
);
3677 printk(KERN_EMERG
"%s: failed to convert unwritten"
3678 "extents to written extents, error is %d"
3679 " io is still on inode %lu aio dio list\n",
3680 __func__
, ret
, inode
->i_ino
);
3685 aio_complete(io
->iocb
, io
->result
, 0);
3686 /* clear the DIO AIO unwritten flag */
3692 * work on completed aio dio IO, to convert unwritten extents to extents
3694 static void ext4_end_io_work(struct work_struct
*work
)
3696 ext4_io_end_t
*io
= container_of(work
, ext4_io_end_t
, work
);
3697 struct inode
*inode
= io
->inode
;
3698 struct ext4_inode_info
*ei
= EXT4_I(inode
);
3699 unsigned long flags
;
3702 mutex_lock(&inode
->i_mutex
);
3703 ret
= ext4_end_io_nolock(io
);
3705 mutex_unlock(&inode
->i_mutex
);
3709 spin_lock_irqsave(&ei
->i_completed_io_lock
, flags
);
3710 if (!list_empty(&io
->list
))
3711 list_del_init(&io
->list
);
3712 spin_unlock_irqrestore(&ei
->i_completed_io_lock
, flags
);
3713 mutex_unlock(&inode
->i_mutex
);
3714 ext4_free_io_end(io
);
3718 * This function is called from ext4_sync_file().
3720 * When IO is completed, the work to convert unwritten extents to
3721 * written is queued on workqueue but may not get immediately
3722 * scheduled. When fsync is called, we need to ensure the
3723 * conversion is complete before fsync returns.
3724 * The inode keeps track of a list of pending/completed IO that
3725 * might needs to do the conversion. This function walks through
3726 * the list and convert the related unwritten extents for completed IO
3728 * The function return the number of pending IOs on success.
3730 int flush_completed_IO(struct inode
*inode
)
3733 struct ext4_inode_info
*ei
= EXT4_I(inode
);
3734 unsigned long flags
;
3738 if (list_empty(&ei
->i_completed_io_list
))
3741 dump_completed_IO(inode
);
3742 spin_lock_irqsave(&ei
->i_completed_io_lock
, flags
);
3743 while (!list_empty(&ei
->i_completed_io_list
)){
3744 io
= list_entry(ei
->i_completed_io_list
.next
,
3745 ext4_io_end_t
, list
);
3747 * Calling ext4_end_io_nolock() to convert completed
3750 * When ext4_sync_file() is called, run_queue() may already
3751 * about to flush the work corresponding to this io structure.
3752 * It will be upset if it founds the io structure related
3753 * to the work-to-be schedule is freed.
3755 * Thus we need to keep the io structure still valid here after
3756 * convertion finished. The io structure has a flag to
3757 * avoid double converting from both fsync and background work
3760 spin_unlock_irqrestore(&ei
->i_completed_io_lock
, flags
);
3761 ret
= ext4_end_io_nolock(io
);
3762 spin_lock_irqsave(&ei
->i_completed_io_lock
, flags
);
3766 list_del_init(&io
->list
);
3768 spin_unlock_irqrestore(&ei
->i_completed_io_lock
, flags
);
3769 return (ret2
< 0) ? ret2
: 0;
3772 static ext4_io_end_t
*ext4_init_io_end (struct inode
*inode
, gfp_t flags
)
3774 ext4_io_end_t
*io
= NULL
;
3776 io
= kmalloc(sizeof(*io
), flags
);
3787 INIT_WORK(&io
->work
, ext4_end_io_work
);
3788 INIT_LIST_HEAD(&io
->list
);
3794 static void ext4_end_io_dio(struct kiocb
*iocb
, loff_t offset
,
3795 ssize_t size
, void *private, int ret
,
3798 ext4_io_end_t
*io_end
= iocb
->private;
3799 struct workqueue_struct
*wq
;
3800 unsigned long flags
;
3801 struct ext4_inode_info
*ei
;
3803 /* if not async direct IO or dio with 0 bytes write, just return */
3804 if (!io_end
|| !size
)
3807 ext_debug("ext4_end_io_dio(): io_end 0x%p"
3808 "for inode %lu, iocb 0x%p, offset %llu, size %llu\n",
3809 iocb
->private, io_end
->inode
->i_ino
, iocb
, offset
,
3812 /* if not aio dio with unwritten extents, just free io and return */
3813 if (io_end
->flag
!= EXT4_IO_UNWRITTEN
){
3814 ext4_free_io_end(io_end
);
3815 iocb
->private = NULL
;
3818 aio_complete(iocb
, ret
, 0);
3822 io_end
->offset
= offset
;
3823 io_end
->size
= size
;
3825 io_end
->iocb
= iocb
;
3826 io_end
->result
= ret
;
3828 wq
= EXT4_SB(io_end
->inode
->i_sb
)->dio_unwritten_wq
;
3830 /* Add the io_end to per-inode completed aio dio list*/
3831 ei
= EXT4_I(io_end
->inode
);
3832 spin_lock_irqsave(&ei
->i_completed_io_lock
, flags
);
3833 list_add_tail(&io_end
->list
, &ei
->i_completed_io_list
);
3834 spin_unlock_irqrestore(&ei
->i_completed_io_lock
, flags
);
3836 /* queue the work to convert unwritten extents to written */
3837 queue_work(wq
, &io_end
->work
);
3838 iocb
->private = NULL
;
3841 static void ext4_end_io_buffer_write(struct buffer_head
*bh
, int uptodate
)
3843 ext4_io_end_t
*io_end
= bh
->b_private
;
3844 struct workqueue_struct
*wq
;
3845 struct inode
*inode
;
3846 unsigned long flags
;
3848 if (!test_clear_buffer_uninit(bh
) || !io_end
)
3851 if (!(io_end
->inode
->i_sb
->s_flags
& MS_ACTIVE
)) {
3852 printk("sb umounted, discard end_io request for inode %lu\n",
3853 io_end
->inode
->i_ino
);
3854 ext4_free_io_end(io_end
);
3858 io_end
->flag
= EXT4_IO_UNWRITTEN
;
3859 inode
= io_end
->inode
;
3861 /* Add the io_end to per-inode completed io list*/
3862 spin_lock_irqsave(&EXT4_I(inode
)->i_completed_io_lock
, flags
);
3863 list_add_tail(&io_end
->list
, &EXT4_I(inode
)->i_completed_io_list
);
3864 spin_unlock_irqrestore(&EXT4_I(inode
)->i_completed_io_lock
, flags
);
3866 wq
= EXT4_SB(inode
->i_sb
)->dio_unwritten_wq
;
3867 /* queue the work to convert unwritten extents to written */
3868 queue_work(wq
, &io_end
->work
);
3870 bh
->b_private
= NULL
;
3871 bh
->b_end_io
= NULL
;
3872 clear_buffer_uninit(bh
);
3873 end_buffer_async_write(bh
, uptodate
);
3876 static int ext4_set_bh_endio(struct buffer_head
*bh
, struct inode
*inode
)
3878 ext4_io_end_t
*io_end
;
3879 struct page
*page
= bh
->b_page
;
3880 loff_t offset
= (sector_t
)page
->index
<< PAGE_CACHE_SHIFT
;
3881 size_t size
= bh
->b_size
;
3884 io_end
= ext4_init_io_end(inode
, GFP_ATOMIC
);
3886 if (printk_ratelimit())
3887 printk(KERN_WARNING
"%s: allocation fail\n", __func__
);
3891 io_end
->offset
= offset
;
3892 io_end
->size
= size
;
3894 * We need to hold a reference to the page to make sure it
3895 * doesn't get evicted before ext4_end_io_work() has a chance
3896 * to convert the extent from written to unwritten.
3898 io_end
->page
= page
;
3899 get_page(io_end
->page
);
3901 bh
->b_private
= io_end
;
3902 bh
->b_end_io
= ext4_end_io_buffer_write
;
3907 * For ext4 extent files, ext4 will do direct-io write to holes,
3908 * preallocated extents, and those write extend the file, no need to
3909 * fall back to buffered IO.
3911 * For holes, we fallocate those blocks, mark them as unintialized
3912 * If those blocks were preallocated, we mark sure they are splited, but
3913 * still keep the range to write as unintialized.
3915 * The unwrritten extents will be converted to written when DIO is completed.
3916 * For async direct IO, since the IO may still pending when return, we
3917 * set up an end_io call back function, which will do the convertion
3918 * when async direct IO completed.
3920 * If the O_DIRECT write will extend the file then add this inode to the
3921 * orphan list. So recovery will truncate it back to the original size
3922 * if the machine crashes during the write.
3925 static ssize_t
ext4_ext_direct_IO(int rw
, struct kiocb
*iocb
,
3926 const struct iovec
*iov
, loff_t offset
,
3927 unsigned long nr_segs
)
3929 struct file
*file
= iocb
->ki_filp
;
3930 struct inode
*inode
= file
->f_mapping
->host
;
3932 size_t count
= iov_length(iov
, nr_segs
);
3934 loff_t final_size
= offset
+ count
;
3935 if (rw
== WRITE
&& final_size
<= inode
->i_size
) {
3937 * We could direct write to holes and fallocate.
3939 * Allocated blocks to fill the hole are marked as uninitialized
3940 * to prevent paralel buffered read to expose the stale data
3941 * before DIO complete the data IO.
3943 * As to previously fallocated extents, ext4 get_block
3944 * will just simply mark the buffer mapped but still
3945 * keep the extents uninitialized.
3947 * for non AIO case, we will convert those unwritten extents
3948 * to written after return back from blockdev_direct_IO.
3950 * for async DIO, the conversion needs to be defered when
3951 * the IO is completed. The ext4 end_io callback function
3952 * will be called to take care of the conversion work.
3953 * Here for async case, we allocate an io_end structure to
3956 iocb
->private = NULL
;
3957 EXT4_I(inode
)->cur_aio_dio
= NULL
;
3958 if (!is_sync_kiocb(iocb
)) {
3959 iocb
->private = ext4_init_io_end(inode
, GFP_NOFS
);
3963 * we save the io structure for current async
3964 * direct IO, so that later ext4_map_blocks()
3965 * could flag the io structure whether there
3966 * is a unwritten extents needs to be converted
3967 * when IO is completed.
3969 EXT4_I(inode
)->cur_aio_dio
= iocb
->private;
3972 ret
= blockdev_direct_IO(rw
, iocb
, inode
,
3973 inode
->i_sb
->s_bdev
, iov
,
3975 ext4_get_block_write
,
3978 EXT4_I(inode
)->cur_aio_dio
= NULL
;
3980 * The io_end structure takes a reference to the inode,
3981 * that structure needs to be destroyed and the
3982 * reference to the inode need to be dropped, when IO is
3983 * complete, even with 0 byte write, or failed.
3985 * In the successful AIO DIO case, the io_end structure will be
3986 * desctroyed and the reference to the inode will be dropped
3987 * after the end_io call back function is called.
3989 * In the case there is 0 byte write, or error case, since
3990 * VFS direct IO won't invoke the end_io call back function,
3991 * we need to free the end_io structure here.
3993 if (ret
!= -EIOCBQUEUED
&& ret
<= 0 && iocb
->private) {
3994 ext4_free_io_end(iocb
->private);
3995 iocb
->private = NULL
;
3996 } else if (ret
> 0 && ext4_test_inode_state(inode
,
3997 EXT4_STATE_DIO_UNWRITTEN
)) {
4000 * for non AIO case, since the IO is already
4001 * completed, we could do the convertion right here
4003 err
= ext4_convert_unwritten_extents(inode
,
4007 ext4_clear_inode_state(inode
, EXT4_STATE_DIO_UNWRITTEN
);
4012 /* for write the the end of file case, we fall back to old way */
4013 return ext4_ind_direct_IO(rw
, iocb
, iov
, offset
, nr_segs
);
4016 static ssize_t
ext4_direct_IO(int rw
, struct kiocb
*iocb
,
4017 const struct iovec
*iov
, loff_t offset
,
4018 unsigned long nr_segs
)
4020 struct file
*file
= iocb
->ki_filp
;
4021 struct inode
*inode
= file
->f_mapping
->host
;
4023 if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
))
4024 return ext4_ext_direct_IO(rw
, iocb
, iov
, offset
, nr_segs
);
4026 return ext4_ind_direct_IO(rw
, iocb
, iov
, offset
, nr_segs
);
4030 * Pages can be marked dirty completely asynchronously from ext4's journalling
4031 * activity. By filemap_sync_pte(), try_to_unmap_one(), etc. We cannot do
4032 * much here because ->set_page_dirty is called under VFS locks. The page is
4033 * not necessarily locked.
4035 * We cannot just dirty the page and leave attached buffers clean, because the
4036 * buffers' dirty state is "definitive". We cannot just set the buffers dirty
4037 * or jbddirty because all the journalling code will explode.
4039 * So what we do is to mark the page "pending dirty" and next time writepage
4040 * is called, propagate that into the buffers appropriately.
4042 static int ext4_journalled_set_page_dirty(struct page
*page
)
4044 SetPageChecked(page
);
4045 return __set_page_dirty_nobuffers(page
);
4048 static const struct address_space_operations ext4_ordered_aops
= {
4049 .readpage
= ext4_readpage
,
4050 .readpages
= ext4_readpages
,
4051 .writepage
= ext4_writepage
,
4052 .sync_page
= block_sync_page
,
4053 .write_begin
= ext4_write_begin
,
4054 .write_end
= ext4_ordered_write_end
,
4056 .invalidatepage
= ext4_invalidatepage
,
4057 .releasepage
= ext4_releasepage
,
4058 .direct_IO
= ext4_direct_IO
,
4059 .migratepage
= buffer_migrate_page
,
4060 .is_partially_uptodate
= block_is_partially_uptodate
,
4061 .error_remove_page
= generic_error_remove_page
,
4064 static const struct address_space_operations ext4_writeback_aops
= {
4065 .readpage
= ext4_readpage
,
4066 .readpages
= ext4_readpages
,
4067 .writepage
= ext4_writepage
,
4068 .sync_page
= block_sync_page
,
4069 .write_begin
= ext4_write_begin
,
4070 .write_end
= ext4_writeback_write_end
,
4072 .invalidatepage
= ext4_invalidatepage
,
4073 .releasepage
= ext4_releasepage
,
4074 .direct_IO
= ext4_direct_IO
,
4075 .migratepage
= buffer_migrate_page
,
4076 .is_partially_uptodate
= block_is_partially_uptodate
,
4077 .error_remove_page
= generic_error_remove_page
,
4080 static const struct address_space_operations ext4_journalled_aops
= {
4081 .readpage
= ext4_readpage
,
4082 .readpages
= ext4_readpages
,
4083 .writepage
= ext4_writepage
,
4084 .sync_page
= block_sync_page
,
4085 .write_begin
= ext4_write_begin
,
4086 .write_end
= ext4_journalled_write_end
,
4087 .set_page_dirty
= ext4_journalled_set_page_dirty
,
4089 .invalidatepage
= ext4_invalidatepage
,
4090 .releasepage
= ext4_releasepage
,
4091 .is_partially_uptodate
= block_is_partially_uptodate
,
4092 .error_remove_page
= generic_error_remove_page
,
4095 static const struct address_space_operations ext4_da_aops
= {
4096 .readpage
= ext4_readpage
,
4097 .readpages
= ext4_readpages
,
4098 .writepage
= ext4_writepage
,
4099 .writepages
= ext4_da_writepages
,
4100 .sync_page
= block_sync_page
,
4101 .write_begin
= ext4_da_write_begin
,
4102 .write_end
= ext4_da_write_end
,
4104 .invalidatepage
= ext4_da_invalidatepage
,
4105 .releasepage
= ext4_releasepage
,
4106 .direct_IO
= ext4_direct_IO
,
4107 .migratepage
= buffer_migrate_page
,
4108 .is_partially_uptodate
= block_is_partially_uptodate
,
4109 .error_remove_page
= generic_error_remove_page
,
4112 void ext4_set_aops(struct inode
*inode
)
4114 if (ext4_should_order_data(inode
) &&
4115 test_opt(inode
->i_sb
, DELALLOC
))
4116 inode
->i_mapping
->a_ops
= &ext4_da_aops
;
4117 else if (ext4_should_order_data(inode
))
4118 inode
->i_mapping
->a_ops
= &ext4_ordered_aops
;
4119 else if (ext4_should_writeback_data(inode
) &&
4120 test_opt(inode
->i_sb
, DELALLOC
))
4121 inode
->i_mapping
->a_ops
= &ext4_da_aops
;
4122 else if (ext4_should_writeback_data(inode
))
4123 inode
->i_mapping
->a_ops
= &ext4_writeback_aops
;
4125 inode
->i_mapping
->a_ops
= &ext4_journalled_aops
;
4129 * ext4_block_truncate_page() zeroes out a mapping from file offset `from'
4130 * up to the end of the block which corresponds to `from'.
4131 * This required during truncate. We need to physically zero the tail end
4132 * of that block so it doesn't yield old data if the file is later grown.
4134 int ext4_block_truncate_page(handle_t
*handle
,
4135 struct address_space
*mapping
, loff_t from
)
4137 ext4_fsblk_t index
= from
>> PAGE_CACHE_SHIFT
;
4138 unsigned offset
= from
& (PAGE_CACHE_SIZE
-1);
4139 unsigned blocksize
, length
, pos
;
4141 struct inode
*inode
= mapping
->host
;
4142 struct buffer_head
*bh
;
4146 page
= find_or_create_page(mapping
, from
>> PAGE_CACHE_SHIFT
,
4147 mapping_gfp_mask(mapping
) & ~__GFP_FS
);
4151 blocksize
= inode
->i_sb
->s_blocksize
;
4152 length
= blocksize
- (offset
& (blocksize
- 1));
4153 iblock
= index
<< (PAGE_CACHE_SHIFT
- inode
->i_sb
->s_blocksize_bits
);
4155 if (!page_has_buffers(page
))
4156 create_empty_buffers(page
, blocksize
, 0);
4158 /* Find the buffer that contains "offset" */
4159 bh
= page_buffers(page
);
4161 while (offset
>= pos
) {
4162 bh
= bh
->b_this_page
;
4168 if (buffer_freed(bh
)) {
4169 BUFFER_TRACE(bh
, "freed: skip");
4173 if (!buffer_mapped(bh
)) {
4174 BUFFER_TRACE(bh
, "unmapped");
4175 ext4_get_block(inode
, iblock
, bh
, 0);
4176 /* unmapped? It's a hole - nothing to do */
4177 if (!buffer_mapped(bh
)) {
4178 BUFFER_TRACE(bh
, "still unmapped");
4183 /* Ok, it's mapped. Make sure it's up-to-date */
4184 if (PageUptodate(page
))
4185 set_buffer_uptodate(bh
);
4187 if (!buffer_uptodate(bh
)) {
4189 ll_rw_block(READ
, 1, &bh
);
4191 /* Uhhuh. Read error. Complain and punt. */
4192 if (!buffer_uptodate(bh
))
4196 if (ext4_should_journal_data(inode
)) {
4197 BUFFER_TRACE(bh
, "get write access");
4198 err
= ext4_journal_get_write_access(handle
, bh
);
4203 zero_user(page
, offset
, length
);
4205 BUFFER_TRACE(bh
, "zeroed end of block");
4208 if (ext4_should_journal_data(inode
)) {
4209 err
= ext4_handle_dirty_metadata(handle
, inode
, bh
);
4211 if (ext4_should_order_data(inode
))
4212 err
= ext4_jbd2_file_inode(handle
, inode
);
4213 mark_buffer_dirty(bh
);
4218 page_cache_release(page
);
4223 * Probably it should be a library function... search for first non-zero word
4224 * or memcmp with zero_page, whatever is better for particular architecture.
4227 static inline int all_zeroes(__le32
*p
, __le32
*q
)
4236 * ext4_find_shared - find the indirect blocks for partial truncation.
4237 * @inode: inode in question
4238 * @depth: depth of the affected branch
4239 * @offsets: offsets of pointers in that branch (see ext4_block_to_path)
4240 * @chain: place to store the pointers to partial indirect blocks
4241 * @top: place to the (detached) top of branch
4243 * This is a helper function used by ext4_truncate().
4245 * When we do truncate() we may have to clean the ends of several
4246 * indirect blocks but leave the blocks themselves alive. Block is
4247 * partially truncated if some data below the new i_size is refered
4248 * from it (and it is on the path to the first completely truncated
4249 * data block, indeed). We have to free the top of that path along
4250 * with everything to the right of the path. Since no allocation
4251 * past the truncation point is possible until ext4_truncate()
4252 * finishes, we may safely do the latter, but top of branch may
4253 * require special attention - pageout below the truncation point
4254 * might try to populate it.
4256 * We atomically detach the top of branch from the tree, store the
4257 * block number of its root in *@top, pointers to buffer_heads of
4258 * partially truncated blocks - in @chain[].bh and pointers to
4259 * their last elements that should not be removed - in
4260 * @chain[].p. Return value is the pointer to last filled element
4263 * The work left to caller to do the actual freeing of subtrees:
4264 * a) free the subtree starting from *@top
4265 * b) free the subtrees whose roots are stored in
4266 * (@chain[i].p+1 .. end of @chain[i].bh->b_data)
4267 * c) free the subtrees growing from the inode past the @chain[0].
4268 * (no partially truncated stuff there). */
4270 static Indirect
*ext4_find_shared(struct inode
*inode
, int depth
,
4271 ext4_lblk_t offsets
[4], Indirect chain
[4],
4274 Indirect
*partial
, *p
;
4278 /* Make k index the deepest non-null offset + 1 */
4279 for (k
= depth
; k
> 1 && !offsets
[k
-1]; k
--)
4281 partial
= ext4_get_branch(inode
, k
, offsets
, chain
, &err
);
4282 /* Writer: pointers */
4284 partial
= chain
+ k
-1;
4286 * If the branch acquired continuation since we've looked at it -
4287 * fine, it should all survive and (new) top doesn't belong to us.
4289 if (!partial
->key
&& *partial
->p
)
4292 for (p
= partial
; (p
> chain
) && all_zeroes((__le32
*) p
->bh
->b_data
, p
->p
); p
--)
4295 * OK, we've found the last block that must survive. The rest of our
4296 * branch should be detached before unlocking. However, if that rest
4297 * of branch is all ours and does not grow immediately from the inode
4298 * it's easier to cheat and just decrement partial->p.
4300 if (p
== chain
+ k
- 1 && p
> chain
) {
4304 /* Nope, don't do this in ext4. Must leave the tree intact */
4311 while (partial
> p
) {
4312 brelse(partial
->bh
);
4320 * Zero a number of block pointers in either an inode or an indirect block.
4321 * If we restart the transaction we must again get write access to the
4322 * indirect block for further modification.
4324 * We release `count' blocks on disk, but (last - first) may be greater
4325 * than `count' because there can be holes in there.
4327 static int ext4_clear_blocks(handle_t
*handle
, struct inode
*inode
,
4328 struct buffer_head
*bh
,
4329 ext4_fsblk_t block_to_free
,
4330 unsigned long count
, __le32
*first
,
4334 int flags
= EXT4_FREE_BLOCKS_FORGET
| EXT4_FREE_BLOCKS_VALIDATED
;
4336 if (S_ISDIR(inode
->i_mode
) || S_ISLNK(inode
->i_mode
))
4337 flags
|= EXT4_FREE_BLOCKS_METADATA
;
4339 if (!ext4_data_block_valid(EXT4_SB(inode
->i_sb
), block_to_free
,
4341 EXT4_ERROR_INODE(inode
, "attempt to clear invalid "
4342 "blocks %llu len %lu",
4343 (unsigned long long) block_to_free
, count
);
4347 if (try_to_extend_transaction(handle
, inode
)) {
4349 BUFFER_TRACE(bh
, "call ext4_handle_dirty_metadata");
4350 ext4_handle_dirty_metadata(handle
, inode
, bh
);
4352 ext4_mark_inode_dirty(handle
, inode
);
4353 ext4_truncate_restart_trans(handle
, inode
,
4354 blocks_for_truncate(inode
));
4356 BUFFER_TRACE(bh
, "retaking write access");
4357 ext4_journal_get_write_access(handle
, bh
);
4361 for (p
= first
; p
< last
; p
++)
4364 ext4_free_blocks(handle
, inode
, 0, block_to_free
, count
, flags
);
4369 * ext4_free_data - free a list of data blocks
4370 * @handle: handle for this transaction
4371 * @inode: inode we are dealing with
4372 * @this_bh: indirect buffer_head which contains *@first and *@last
4373 * @first: array of block numbers
4374 * @last: points immediately past the end of array
4376 * We are freeing all blocks refered from that array (numbers are stored as
4377 * little-endian 32-bit) and updating @inode->i_blocks appropriately.
4379 * We accumulate contiguous runs of blocks to free. Conveniently, if these
4380 * blocks are contiguous then releasing them at one time will only affect one
4381 * or two bitmap blocks (+ group descriptor(s) and superblock) and we won't
4382 * actually use a lot of journal space.
4384 * @this_bh will be %NULL if @first and @last point into the inode's direct
4387 static void ext4_free_data(handle_t
*handle
, struct inode
*inode
,
4388 struct buffer_head
*this_bh
,
4389 __le32
*first
, __le32
*last
)
4391 ext4_fsblk_t block_to_free
= 0; /* Starting block # of a run */
4392 unsigned long count
= 0; /* Number of blocks in the run */
4393 __le32
*block_to_free_p
= NULL
; /* Pointer into inode/ind
4396 ext4_fsblk_t nr
; /* Current block # */
4397 __le32
*p
; /* Pointer into inode/ind
4398 for current block */
4401 if (this_bh
) { /* For indirect block */
4402 BUFFER_TRACE(this_bh
, "get_write_access");
4403 err
= ext4_journal_get_write_access(handle
, this_bh
);
4404 /* Important: if we can't update the indirect pointers
4405 * to the blocks, we can't free them. */
4410 for (p
= first
; p
< last
; p
++) {
4411 nr
= le32_to_cpu(*p
);
4413 /* accumulate blocks to free if they're contiguous */
4416 block_to_free_p
= p
;
4418 } else if (nr
== block_to_free
+ count
) {
4421 if (ext4_clear_blocks(handle
, inode
, this_bh
,
4422 block_to_free
, count
,
4423 block_to_free_p
, p
))
4426 block_to_free_p
= p
;
4433 ext4_clear_blocks(handle
, inode
, this_bh
, block_to_free
,
4434 count
, block_to_free_p
, p
);
4437 BUFFER_TRACE(this_bh
, "call ext4_handle_dirty_metadata");
4440 * The buffer head should have an attached journal head at this
4441 * point. However, if the data is corrupted and an indirect
4442 * block pointed to itself, it would have been detached when
4443 * the block was cleared. Check for this instead of OOPSing.
4445 if ((EXT4_JOURNAL(inode
) == NULL
) || bh2jh(this_bh
))
4446 ext4_handle_dirty_metadata(handle
, inode
, this_bh
);
4448 EXT4_ERROR_INODE(inode
,
4449 "circular indirect block detected at "
4451 (unsigned long long) this_bh
->b_blocknr
);
4456 * ext4_free_branches - free an array of branches
4457 * @handle: JBD handle for this transaction
4458 * @inode: inode we are dealing with
4459 * @parent_bh: the buffer_head which contains *@first and *@last
4460 * @first: array of block numbers
4461 * @last: pointer immediately past the end of array
4462 * @depth: depth of the branches to free
4464 * We are freeing all blocks refered from these branches (numbers are
4465 * stored as little-endian 32-bit) and updating @inode->i_blocks
4468 static void ext4_free_branches(handle_t
*handle
, struct inode
*inode
,
4469 struct buffer_head
*parent_bh
,
4470 __le32
*first
, __le32
*last
, int depth
)
4475 if (ext4_handle_is_aborted(handle
))
4479 struct buffer_head
*bh
;
4480 int addr_per_block
= EXT4_ADDR_PER_BLOCK(inode
->i_sb
);
4482 while (--p
>= first
) {
4483 nr
= le32_to_cpu(*p
);
4485 continue; /* A hole */
4487 if (!ext4_data_block_valid(EXT4_SB(inode
->i_sb
),
4489 EXT4_ERROR_INODE(inode
,
4490 "invalid indirect mapped "
4491 "block %lu (level %d)",
4492 (unsigned long) nr
, depth
);
4496 /* Go read the buffer for the next level down */
4497 bh
= sb_bread(inode
->i_sb
, nr
);
4500 * A read failure? Report error and clear slot
4504 EXT4_ERROR_INODE_BLOCK(inode
, nr
,
4509 /* This zaps the entire block. Bottom up. */
4510 BUFFER_TRACE(bh
, "free child branches");
4511 ext4_free_branches(handle
, inode
, bh
,
4512 (__le32
*) bh
->b_data
,
4513 (__le32
*) bh
->b_data
+ addr_per_block
,
4517 * Everything below this this pointer has been
4518 * released. Now let this top-of-subtree go.
4520 * We want the freeing of this indirect block to be
4521 * atomic in the journal with the updating of the
4522 * bitmap block which owns it. So make some room in
4525 * We zero the parent pointer *after* freeing its
4526 * pointee in the bitmaps, so if extend_transaction()
4527 * for some reason fails to put the bitmap changes and
4528 * the release into the same transaction, recovery
4529 * will merely complain about releasing a free block,
4530 * rather than leaking blocks.
4532 if (ext4_handle_is_aborted(handle
))
4534 if (try_to_extend_transaction(handle
, inode
)) {
4535 ext4_mark_inode_dirty(handle
, inode
);
4536 ext4_truncate_restart_trans(handle
, inode
,
4537 blocks_for_truncate(inode
));
4541 * The forget flag here is critical because if
4542 * we are journaling (and not doing data
4543 * journaling), we have to make sure a revoke
4544 * record is written to prevent the journal
4545 * replay from overwriting the (former)
4546 * indirect block if it gets reallocated as a
4547 * data block. This must happen in the same
4548 * transaction where the data blocks are
4551 ext4_free_blocks(handle
, inode
, 0, nr
, 1,
4552 EXT4_FREE_BLOCKS_METADATA
|
4553 EXT4_FREE_BLOCKS_FORGET
);
4557 * The block which we have just freed is
4558 * pointed to by an indirect block: journal it
4560 BUFFER_TRACE(parent_bh
, "get_write_access");
4561 if (!ext4_journal_get_write_access(handle
,
4564 BUFFER_TRACE(parent_bh
,
4565 "call ext4_handle_dirty_metadata");
4566 ext4_handle_dirty_metadata(handle
,
4573 /* We have reached the bottom of the tree. */
4574 BUFFER_TRACE(parent_bh
, "free data blocks");
4575 ext4_free_data(handle
, inode
, parent_bh
, first
, last
);
4579 int ext4_can_truncate(struct inode
*inode
)
4581 if (IS_APPEND(inode
) || IS_IMMUTABLE(inode
))
4583 if (S_ISREG(inode
->i_mode
))
4585 if (S_ISDIR(inode
->i_mode
))
4587 if (S_ISLNK(inode
->i_mode
))
4588 return !ext4_inode_is_fast_symlink(inode
);
4595 * We block out ext4_get_block() block instantiations across the entire
4596 * transaction, and VFS/VM ensures that ext4_truncate() cannot run
4597 * simultaneously on behalf of the same inode.
4599 * As we work through the truncate and commmit bits of it to the journal there
4600 * is one core, guiding principle: the file's tree must always be consistent on
4601 * disk. We must be able to restart the truncate after a crash.
4603 * The file's tree may be transiently inconsistent in memory (although it
4604 * probably isn't), but whenever we close off and commit a journal transaction,
4605 * the contents of (the filesystem + the journal) must be consistent and
4606 * restartable. It's pretty simple, really: bottom up, right to left (although
4607 * left-to-right works OK too).
4609 * Note that at recovery time, journal replay occurs *before* the restart of
4610 * truncate against the orphan inode list.
4612 * The committed inode has the new, desired i_size (which is the same as
4613 * i_disksize in this case). After a crash, ext4_orphan_cleanup() will see
4614 * that this inode's truncate did not complete and it will again call
4615 * ext4_truncate() to have another go. So there will be instantiated blocks
4616 * to the right of the truncation point in a crashed ext4 filesystem. But
4617 * that's fine - as long as they are linked from the inode, the post-crash
4618 * ext4_truncate() run will find them and release them.
4620 void ext4_truncate(struct inode
*inode
)
4623 struct ext4_inode_info
*ei
= EXT4_I(inode
);
4624 __le32
*i_data
= ei
->i_data
;
4625 int addr_per_block
= EXT4_ADDR_PER_BLOCK(inode
->i_sb
);
4626 struct address_space
*mapping
= inode
->i_mapping
;
4627 ext4_lblk_t offsets
[4];
4632 ext4_lblk_t last_block
;
4633 unsigned blocksize
= inode
->i_sb
->s_blocksize
;
4635 if (!ext4_can_truncate(inode
))
4638 ext4_clear_inode_flag(inode
, EXT4_INODE_EOFBLOCKS
);
4640 if (inode
->i_size
== 0 && !test_opt(inode
->i_sb
, NO_AUTO_DA_ALLOC
))
4641 ext4_set_inode_state(inode
, EXT4_STATE_DA_ALLOC_CLOSE
);
4643 if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
)) {
4644 ext4_ext_truncate(inode
);
4648 handle
= start_transaction(inode
);
4650 return; /* AKPM: return what? */
4652 last_block
= (inode
->i_size
+ blocksize
-1)
4653 >> EXT4_BLOCK_SIZE_BITS(inode
->i_sb
);
4655 if (inode
->i_size
& (blocksize
- 1))
4656 if (ext4_block_truncate_page(handle
, mapping
, inode
->i_size
))
4659 n
= ext4_block_to_path(inode
, last_block
, offsets
, NULL
);
4661 goto out_stop
; /* error */
4664 * OK. This truncate is going to happen. We add the inode to the
4665 * orphan list, so that if this truncate spans multiple transactions,
4666 * and we crash, we will resume the truncate when the filesystem
4667 * recovers. It also marks the inode dirty, to catch the new size.
4669 * Implication: the file must always be in a sane, consistent
4670 * truncatable state while each transaction commits.
4672 if (ext4_orphan_add(handle
, inode
))
4676 * From here we block out all ext4_get_block() callers who want to
4677 * modify the block allocation tree.
4679 down_write(&ei
->i_data_sem
);
4681 ext4_discard_preallocations(inode
);
4684 * The orphan list entry will now protect us from any crash which
4685 * occurs before the truncate completes, so it is now safe to propagate
4686 * the new, shorter inode size (held for now in i_size) into the
4687 * on-disk inode. We do this via i_disksize, which is the value which
4688 * ext4 *really* writes onto the disk inode.
4690 ei
->i_disksize
= inode
->i_size
;
4692 if (n
== 1) { /* direct blocks */
4693 ext4_free_data(handle
, inode
, NULL
, i_data
+offsets
[0],
4694 i_data
+ EXT4_NDIR_BLOCKS
);
4698 partial
= ext4_find_shared(inode
, n
, offsets
, chain
, &nr
);
4699 /* Kill the top of shared branch (not detached) */
4701 if (partial
== chain
) {
4702 /* Shared branch grows from the inode */
4703 ext4_free_branches(handle
, inode
, NULL
,
4704 &nr
, &nr
+1, (chain
+n
-1) - partial
);
4707 * We mark the inode dirty prior to restart,
4708 * and prior to stop. No need for it here.
4711 /* Shared branch grows from an indirect block */
4712 BUFFER_TRACE(partial
->bh
, "get_write_access");
4713 ext4_free_branches(handle
, inode
, partial
->bh
,
4715 partial
->p
+1, (chain
+n
-1) - partial
);
4718 /* Clear the ends of indirect blocks on the shared branch */
4719 while (partial
> chain
) {
4720 ext4_free_branches(handle
, inode
, partial
->bh
, partial
->p
+ 1,
4721 (__le32
*)partial
->bh
->b_data
+addr_per_block
,
4722 (chain
+n
-1) - partial
);
4723 BUFFER_TRACE(partial
->bh
, "call brelse");
4724 brelse(partial
->bh
);
4728 /* Kill the remaining (whole) subtrees */
4729 switch (offsets
[0]) {
4731 nr
= i_data
[EXT4_IND_BLOCK
];
4733 ext4_free_branches(handle
, inode
, NULL
, &nr
, &nr
+1, 1);
4734 i_data
[EXT4_IND_BLOCK
] = 0;
4736 case EXT4_IND_BLOCK
:
4737 nr
= i_data
[EXT4_DIND_BLOCK
];
4739 ext4_free_branches(handle
, inode
, NULL
, &nr
, &nr
+1, 2);
4740 i_data
[EXT4_DIND_BLOCK
] = 0;
4742 case EXT4_DIND_BLOCK
:
4743 nr
= i_data
[EXT4_TIND_BLOCK
];
4745 ext4_free_branches(handle
, inode
, NULL
, &nr
, &nr
+1, 3);
4746 i_data
[EXT4_TIND_BLOCK
] = 0;
4748 case EXT4_TIND_BLOCK
:
4752 up_write(&ei
->i_data_sem
);
4753 inode
->i_mtime
= inode
->i_ctime
= ext4_current_time(inode
);
4754 ext4_mark_inode_dirty(handle
, inode
);
4757 * In a multi-transaction truncate, we only make the final transaction
4761 ext4_handle_sync(handle
);
4764 * If this was a simple ftruncate(), and the file will remain alive
4765 * then we need to clear up the orphan record which we created above.
4766 * However, if this was a real unlink then we were called by
4767 * ext4_delete_inode(), and we allow that function to clean up the
4768 * orphan info for us.
4771 ext4_orphan_del(handle
, inode
);
4773 ext4_journal_stop(handle
);
4777 * ext4_get_inode_loc returns with an extra refcount against the inode's
4778 * underlying buffer_head on success. If 'in_mem' is true, we have all
4779 * data in memory that is needed to recreate the on-disk version of this
4782 static int __ext4_get_inode_loc(struct inode
*inode
,
4783 struct ext4_iloc
*iloc
, int in_mem
)
4785 struct ext4_group_desc
*gdp
;
4786 struct buffer_head
*bh
;
4787 struct super_block
*sb
= inode
->i_sb
;
4789 int inodes_per_block
, inode_offset
;
4792 if (!ext4_valid_inum(sb
, inode
->i_ino
))
4795 iloc
->block_group
= (inode
->i_ino
- 1) / EXT4_INODES_PER_GROUP(sb
);
4796 gdp
= ext4_get_group_desc(sb
, iloc
->block_group
, NULL
);
4801 * Figure out the offset within the block group inode table
4803 inodes_per_block
= (EXT4_BLOCK_SIZE(sb
) / EXT4_INODE_SIZE(sb
));
4804 inode_offset
= ((inode
->i_ino
- 1) %
4805 EXT4_INODES_PER_GROUP(sb
));
4806 block
= ext4_inode_table(sb
, gdp
) + (inode_offset
/ inodes_per_block
);
4807 iloc
->offset
= (inode_offset
% inodes_per_block
) * EXT4_INODE_SIZE(sb
);
4809 bh
= sb_getblk(sb
, block
);
4811 EXT4_ERROR_INODE_BLOCK(inode
, block
,
4812 "unable to read itable block");
4815 if (!buffer_uptodate(bh
)) {
4819 * If the buffer has the write error flag, we have failed
4820 * to write out another inode in the same block. In this
4821 * case, we don't have to read the block because we may
4822 * read the old inode data successfully.
4824 if (buffer_write_io_error(bh
) && !buffer_uptodate(bh
))
4825 set_buffer_uptodate(bh
);
4827 if (buffer_uptodate(bh
)) {
4828 /* someone brought it uptodate while we waited */
4834 * If we have all information of the inode in memory and this
4835 * is the only valid inode in the block, we need not read the
4839 struct buffer_head
*bitmap_bh
;
4842 start
= inode_offset
& ~(inodes_per_block
- 1);
4844 /* Is the inode bitmap in cache? */
4845 bitmap_bh
= sb_getblk(sb
, ext4_inode_bitmap(sb
, gdp
));
4850 * If the inode bitmap isn't in cache then the
4851 * optimisation may end up performing two reads instead
4852 * of one, so skip it.
4854 if (!buffer_uptodate(bitmap_bh
)) {
4858 for (i
= start
; i
< start
+ inodes_per_block
; i
++) {
4859 if (i
== inode_offset
)
4861 if (ext4_test_bit(i
, bitmap_bh
->b_data
))
4865 if (i
== start
+ inodes_per_block
) {
4866 /* all other inodes are free, so skip I/O */
4867 memset(bh
->b_data
, 0, bh
->b_size
);
4868 set_buffer_uptodate(bh
);
4876 * If we need to do any I/O, try to pre-readahead extra
4877 * blocks from the inode table.
4879 if (EXT4_SB(sb
)->s_inode_readahead_blks
) {
4880 ext4_fsblk_t b
, end
, table
;
4883 table
= ext4_inode_table(sb
, gdp
);
4884 /* s_inode_readahead_blks is always a power of 2 */
4885 b
= block
& ~(EXT4_SB(sb
)->s_inode_readahead_blks
-1);
4888 end
= b
+ EXT4_SB(sb
)->s_inode_readahead_blks
;
4889 num
= EXT4_INODES_PER_GROUP(sb
);
4890 if (EXT4_HAS_RO_COMPAT_FEATURE(sb
,
4891 EXT4_FEATURE_RO_COMPAT_GDT_CSUM
))
4892 num
-= ext4_itable_unused_count(sb
, gdp
);
4893 table
+= num
/ inodes_per_block
;
4897 sb_breadahead(sb
, b
++);
4901 * There are other valid inodes in the buffer, this inode
4902 * has in-inode xattrs, or we don't have this inode in memory.
4903 * Read the block from disk.
4906 bh
->b_end_io
= end_buffer_read_sync
;
4907 submit_bh(READ_META
, bh
);
4909 if (!buffer_uptodate(bh
)) {
4910 EXT4_ERROR_INODE_BLOCK(inode
, block
,
4911 "unable to read itable block");
4921 int ext4_get_inode_loc(struct inode
*inode
, struct ext4_iloc
*iloc
)
4923 /* We have all inode data except xattrs in memory here. */
4924 return __ext4_get_inode_loc(inode
, iloc
,
4925 !ext4_test_inode_state(inode
, EXT4_STATE_XATTR
));
4928 void ext4_set_inode_flags(struct inode
*inode
)
4930 unsigned int flags
= EXT4_I(inode
)->i_flags
;
4932 inode
->i_flags
&= ~(S_SYNC
|S_APPEND
|S_IMMUTABLE
|S_NOATIME
|S_DIRSYNC
);
4933 if (flags
& EXT4_SYNC_FL
)
4934 inode
->i_flags
|= S_SYNC
;
4935 if (flags
& EXT4_APPEND_FL
)
4936 inode
->i_flags
|= S_APPEND
;
4937 if (flags
& EXT4_IMMUTABLE_FL
)
4938 inode
->i_flags
|= S_IMMUTABLE
;
4939 if (flags
& EXT4_NOATIME_FL
)
4940 inode
->i_flags
|= S_NOATIME
;
4941 if (flags
& EXT4_DIRSYNC_FL
)
4942 inode
->i_flags
|= S_DIRSYNC
;
4945 /* Propagate flags from i_flags to EXT4_I(inode)->i_flags */
4946 void ext4_get_inode_flags(struct ext4_inode_info
*ei
)
4948 unsigned int vfs_fl
;
4949 unsigned long old_fl
, new_fl
;
4952 vfs_fl
= ei
->vfs_inode
.i_flags
;
4953 old_fl
= ei
->i_flags
;
4954 new_fl
= old_fl
& ~(EXT4_SYNC_FL
|EXT4_APPEND_FL
|
4955 EXT4_IMMUTABLE_FL
|EXT4_NOATIME_FL
|
4957 if (vfs_fl
& S_SYNC
)
4958 new_fl
|= EXT4_SYNC_FL
;
4959 if (vfs_fl
& S_APPEND
)
4960 new_fl
|= EXT4_APPEND_FL
;
4961 if (vfs_fl
& S_IMMUTABLE
)
4962 new_fl
|= EXT4_IMMUTABLE_FL
;
4963 if (vfs_fl
& S_NOATIME
)
4964 new_fl
|= EXT4_NOATIME_FL
;
4965 if (vfs_fl
& S_DIRSYNC
)
4966 new_fl
|= EXT4_DIRSYNC_FL
;
4967 } while (cmpxchg(&ei
->i_flags
, old_fl
, new_fl
) != old_fl
);
4970 static blkcnt_t
ext4_inode_blocks(struct ext4_inode
*raw_inode
,
4971 struct ext4_inode_info
*ei
)
4974 struct inode
*inode
= &(ei
->vfs_inode
);
4975 struct super_block
*sb
= inode
->i_sb
;
4977 if (EXT4_HAS_RO_COMPAT_FEATURE(sb
,
4978 EXT4_FEATURE_RO_COMPAT_HUGE_FILE
)) {
4979 /* we are using combined 48 bit field */
4980 i_blocks
= ((u64
)le16_to_cpu(raw_inode
->i_blocks_high
)) << 32 |
4981 le32_to_cpu(raw_inode
->i_blocks_lo
);
4982 if (ext4_test_inode_flag(inode
, EXT4_INODE_HUGE_FILE
)) {
4983 /* i_blocks represent file system block size */
4984 return i_blocks
<< (inode
->i_blkbits
- 9);
4989 return le32_to_cpu(raw_inode
->i_blocks_lo
);
4993 struct inode
*ext4_iget(struct super_block
*sb
, unsigned long ino
)
4995 struct ext4_iloc iloc
;
4996 struct ext4_inode
*raw_inode
;
4997 struct ext4_inode_info
*ei
;
4998 struct inode
*inode
;
4999 journal_t
*journal
= EXT4_SB(sb
)->s_journal
;
5003 inode
= iget_locked(sb
, ino
);
5005 return ERR_PTR(-ENOMEM
);
5006 if (!(inode
->i_state
& I_NEW
))
5012 ret
= __ext4_get_inode_loc(inode
, &iloc
, 0);
5015 raw_inode
= ext4_raw_inode(&iloc
);
5016 inode
->i_mode
= le16_to_cpu(raw_inode
->i_mode
);
5017 inode
->i_uid
= (uid_t
)le16_to_cpu(raw_inode
->i_uid_low
);
5018 inode
->i_gid
= (gid_t
)le16_to_cpu(raw_inode
->i_gid_low
);
5019 if (!(test_opt(inode
->i_sb
, NO_UID32
))) {
5020 inode
->i_uid
|= le16_to_cpu(raw_inode
->i_uid_high
) << 16;
5021 inode
->i_gid
|= le16_to_cpu(raw_inode
->i_gid_high
) << 16;
5023 inode
->i_nlink
= le16_to_cpu(raw_inode
->i_links_count
);
5025 ei
->i_state_flags
= 0;
5026 ei
->i_dir_start_lookup
= 0;
5027 ei
->i_dtime
= le32_to_cpu(raw_inode
->i_dtime
);
5028 /* We now have enough fields to check if the inode was active or not.
5029 * This is needed because nfsd might try to access dead inodes
5030 * the test is that same one that e2fsck uses
5031 * NeilBrown 1999oct15
5033 if (inode
->i_nlink
== 0) {
5034 if (inode
->i_mode
== 0 ||
5035 !(EXT4_SB(inode
->i_sb
)->s_mount_state
& EXT4_ORPHAN_FS
)) {
5036 /* this inode is deleted */
5040 /* The only unlinked inodes we let through here have
5041 * valid i_mode and are being read by the orphan
5042 * recovery code: that's fine, we're about to complete
5043 * the process of deleting those. */
5045 ei
->i_flags
= le32_to_cpu(raw_inode
->i_flags
);
5046 inode
->i_blocks
= ext4_inode_blocks(raw_inode
, ei
);
5047 ei
->i_file_acl
= le32_to_cpu(raw_inode
->i_file_acl_lo
);
5048 if (EXT4_HAS_INCOMPAT_FEATURE(sb
, EXT4_FEATURE_INCOMPAT_64BIT
))
5050 ((__u64
)le16_to_cpu(raw_inode
->i_file_acl_high
)) << 32;
5051 inode
->i_size
= ext4_isize(raw_inode
);
5052 ei
->i_disksize
= inode
->i_size
;
5054 ei
->i_reserved_quota
= 0;
5056 inode
->i_generation
= le32_to_cpu(raw_inode
->i_generation
);
5057 ei
->i_block_group
= iloc
.block_group
;
5058 ei
->i_last_alloc_group
= ~0;
5060 * NOTE! The in-memory inode i_data array is in little-endian order
5061 * even on big-endian machines: we do NOT byteswap the block numbers!
5063 for (block
= 0; block
< EXT4_N_BLOCKS
; block
++)
5064 ei
->i_data
[block
] = raw_inode
->i_block
[block
];
5065 INIT_LIST_HEAD(&ei
->i_orphan
);
5068 * Set transaction id's of transactions that have to be committed
5069 * to finish f[data]sync. We set them to currently running transaction
5070 * as we cannot be sure that the inode or some of its metadata isn't
5071 * part of the transaction - the inode could have been reclaimed and
5072 * now it is reread from disk.
5075 transaction_t
*transaction
;
5078 read_lock(&journal
->j_state_lock
);
5079 if (journal
->j_running_transaction
)
5080 transaction
= journal
->j_running_transaction
;
5082 transaction
= journal
->j_committing_transaction
;
5084 tid
= transaction
->t_tid
;
5086 tid
= journal
->j_commit_sequence
;
5087 read_unlock(&journal
->j_state_lock
);
5088 ei
->i_sync_tid
= tid
;
5089 ei
->i_datasync_tid
= tid
;
5092 if (EXT4_INODE_SIZE(inode
->i_sb
) > EXT4_GOOD_OLD_INODE_SIZE
) {
5093 ei
->i_extra_isize
= le16_to_cpu(raw_inode
->i_extra_isize
);
5094 if (EXT4_GOOD_OLD_INODE_SIZE
+ ei
->i_extra_isize
>
5095 EXT4_INODE_SIZE(inode
->i_sb
)) {
5099 if (ei
->i_extra_isize
== 0) {
5100 /* The extra space is currently unused. Use it. */
5101 ei
->i_extra_isize
= sizeof(struct ext4_inode
) -
5102 EXT4_GOOD_OLD_INODE_SIZE
;
5104 __le32
*magic
= (void *)raw_inode
+
5105 EXT4_GOOD_OLD_INODE_SIZE
+
5107 if (*magic
== cpu_to_le32(EXT4_XATTR_MAGIC
))
5108 ext4_set_inode_state(inode
, EXT4_STATE_XATTR
);
5111 ei
->i_extra_isize
= 0;
5113 EXT4_INODE_GET_XTIME(i_ctime
, inode
, raw_inode
);
5114 EXT4_INODE_GET_XTIME(i_mtime
, inode
, raw_inode
);
5115 EXT4_INODE_GET_XTIME(i_atime
, inode
, raw_inode
);
5116 EXT4_EINODE_GET_XTIME(i_crtime
, ei
, raw_inode
);
5118 inode
->i_version
= le32_to_cpu(raw_inode
->i_disk_version
);
5119 if (EXT4_INODE_SIZE(inode
->i_sb
) > EXT4_GOOD_OLD_INODE_SIZE
) {
5120 if (EXT4_FITS_IN_INODE(raw_inode
, ei
, i_version_hi
))
5122 (__u64
)(le32_to_cpu(raw_inode
->i_version_hi
)) << 32;
5126 if (ei
->i_file_acl
&&
5127 !ext4_data_block_valid(EXT4_SB(sb
), ei
->i_file_acl
, 1)) {
5128 EXT4_ERROR_INODE(inode
, "bad extended attribute block %llu",
5132 } else if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
)) {
5133 if (S_ISREG(inode
->i_mode
) || S_ISDIR(inode
->i_mode
) ||
5134 (S_ISLNK(inode
->i_mode
) &&
5135 !ext4_inode_is_fast_symlink(inode
)))
5136 /* Validate extent which is part of inode */
5137 ret
= ext4_ext_check_inode(inode
);
5138 } else if (S_ISREG(inode
->i_mode
) || S_ISDIR(inode
->i_mode
) ||
5139 (S_ISLNK(inode
->i_mode
) &&
5140 !ext4_inode_is_fast_symlink(inode
))) {
5141 /* Validate block references which are part of inode */
5142 ret
= ext4_check_inode_blockref(inode
);
5147 if (S_ISREG(inode
->i_mode
)) {
5148 inode
->i_op
= &ext4_file_inode_operations
;
5149 inode
->i_fop
= &ext4_file_operations
;
5150 ext4_set_aops(inode
);
5151 } else if (S_ISDIR(inode
->i_mode
)) {
5152 inode
->i_op
= &ext4_dir_inode_operations
;
5153 inode
->i_fop
= &ext4_dir_operations
;
5154 } else if (S_ISLNK(inode
->i_mode
)) {
5155 if (ext4_inode_is_fast_symlink(inode
)) {
5156 inode
->i_op
= &ext4_fast_symlink_inode_operations
;
5157 nd_terminate_link(ei
->i_data
, inode
->i_size
,
5158 sizeof(ei
->i_data
) - 1);
5160 inode
->i_op
= &ext4_symlink_inode_operations
;
5161 ext4_set_aops(inode
);
5163 } else if (S_ISCHR(inode
->i_mode
) || S_ISBLK(inode
->i_mode
) ||
5164 S_ISFIFO(inode
->i_mode
) || S_ISSOCK(inode
->i_mode
)) {
5165 inode
->i_op
= &ext4_special_inode_operations
;
5166 if (raw_inode
->i_block
[0])
5167 init_special_inode(inode
, inode
->i_mode
,
5168 old_decode_dev(le32_to_cpu(raw_inode
->i_block
[0])));
5170 init_special_inode(inode
, inode
->i_mode
,
5171 new_decode_dev(le32_to_cpu(raw_inode
->i_block
[1])));
5174 EXT4_ERROR_INODE(inode
, "bogus i_mode (%o)", inode
->i_mode
);
5178 ext4_set_inode_flags(inode
);
5179 unlock_new_inode(inode
);
5185 return ERR_PTR(ret
);
5188 static int ext4_inode_blocks_set(handle_t
*handle
,
5189 struct ext4_inode
*raw_inode
,
5190 struct ext4_inode_info
*ei
)
5192 struct inode
*inode
= &(ei
->vfs_inode
);
5193 u64 i_blocks
= inode
->i_blocks
;
5194 struct super_block
*sb
= inode
->i_sb
;
5196 if (i_blocks
<= ~0U) {
5198 * i_blocks can be represnted in a 32 bit variable
5199 * as multiple of 512 bytes
5201 raw_inode
->i_blocks_lo
= cpu_to_le32(i_blocks
);
5202 raw_inode
->i_blocks_high
= 0;
5203 ext4_clear_inode_flag(inode
, EXT4_INODE_HUGE_FILE
);
5206 if (!EXT4_HAS_RO_COMPAT_FEATURE(sb
, EXT4_FEATURE_RO_COMPAT_HUGE_FILE
))
5209 if (i_blocks
<= 0xffffffffffffULL
) {
5211 * i_blocks can be represented in a 48 bit variable
5212 * as multiple of 512 bytes
5214 raw_inode
->i_blocks_lo
= cpu_to_le32(i_blocks
);
5215 raw_inode
->i_blocks_high
= cpu_to_le16(i_blocks
>> 32);
5216 ext4_clear_inode_flag(inode
, EXT4_INODE_HUGE_FILE
);
5218 ext4_set_inode_flag(inode
, EXT4_INODE_HUGE_FILE
);
5219 /* i_block is stored in file system block size */
5220 i_blocks
= i_blocks
>> (inode
->i_blkbits
- 9);
5221 raw_inode
->i_blocks_lo
= cpu_to_le32(i_blocks
);
5222 raw_inode
->i_blocks_high
= cpu_to_le16(i_blocks
>> 32);
5228 * Post the struct inode info into an on-disk inode location in the
5229 * buffer-cache. This gobbles the caller's reference to the
5230 * buffer_head in the inode location struct.
5232 * The caller must have write access to iloc->bh.
5234 static int ext4_do_update_inode(handle_t
*handle
,
5235 struct inode
*inode
,
5236 struct ext4_iloc
*iloc
)
5238 struct ext4_inode
*raw_inode
= ext4_raw_inode(iloc
);
5239 struct ext4_inode_info
*ei
= EXT4_I(inode
);
5240 struct buffer_head
*bh
= iloc
->bh
;
5241 int err
= 0, rc
, block
;
5243 /* For fields not not tracking in the in-memory inode,
5244 * initialise them to zero for new inodes. */
5245 if (ext4_test_inode_state(inode
, EXT4_STATE_NEW
))
5246 memset(raw_inode
, 0, EXT4_SB(inode
->i_sb
)->s_inode_size
);
5248 ext4_get_inode_flags(ei
);
5249 raw_inode
->i_mode
= cpu_to_le16(inode
->i_mode
);
5250 if (!(test_opt(inode
->i_sb
, NO_UID32
))) {
5251 raw_inode
->i_uid_low
= cpu_to_le16(low_16_bits(inode
->i_uid
));
5252 raw_inode
->i_gid_low
= cpu_to_le16(low_16_bits(inode
->i_gid
));
5254 * Fix up interoperability with old kernels. Otherwise, old inodes get
5255 * re-used with the upper 16 bits of the uid/gid intact
5258 raw_inode
->i_uid_high
=
5259 cpu_to_le16(high_16_bits(inode
->i_uid
));
5260 raw_inode
->i_gid_high
=
5261 cpu_to_le16(high_16_bits(inode
->i_gid
));
5263 raw_inode
->i_uid_high
= 0;
5264 raw_inode
->i_gid_high
= 0;
5267 raw_inode
->i_uid_low
=
5268 cpu_to_le16(fs_high2lowuid(inode
->i_uid
));
5269 raw_inode
->i_gid_low
=
5270 cpu_to_le16(fs_high2lowgid(inode
->i_gid
));
5271 raw_inode
->i_uid_high
= 0;
5272 raw_inode
->i_gid_high
= 0;
5274 raw_inode
->i_links_count
= cpu_to_le16(inode
->i_nlink
);
5276 EXT4_INODE_SET_XTIME(i_ctime
, inode
, raw_inode
);
5277 EXT4_INODE_SET_XTIME(i_mtime
, inode
, raw_inode
);
5278 EXT4_INODE_SET_XTIME(i_atime
, inode
, raw_inode
);
5279 EXT4_EINODE_SET_XTIME(i_crtime
, ei
, raw_inode
);
5281 if (ext4_inode_blocks_set(handle
, raw_inode
, ei
))
5283 raw_inode
->i_dtime
= cpu_to_le32(ei
->i_dtime
);
5284 raw_inode
->i_flags
= cpu_to_le32(ei
->i_flags
);
5285 if (EXT4_SB(inode
->i_sb
)->s_es
->s_creator_os
!=
5286 cpu_to_le32(EXT4_OS_HURD
))
5287 raw_inode
->i_file_acl_high
=
5288 cpu_to_le16(ei
->i_file_acl
>> 32);
5289 raw_inode
->i_file_acl_lo
= cpu_to_le32(ei
->i_file_acl
);
5290 ext4_isize_set(raw_inode
, ei
->i_disksize
);
5291 if (ei
->i_disksize
> 0x7fffffffULL
) {
5292 struct super_block
*sb
= inode
->i_sb
;
5293 if (!EXT4_HAS_RO_COMPAT_FEATURE(sb
,
5294 EXT4_FEATURE_RO_COMPAT_LARGE_FILE
) ||
5295 EXT4_SB(sb
)->s_es
->s_rev_level
==
5296 cpu_to_le32(EXT4_GOOD_OLD_REV
)) {
5297 /* If this is the first large file
5298 * created, add a flag to the superblock.
5300 err
= ext4_journal_get_write_access(handle
,
5301 EXT4_SB(sb
)->s_sbh
);
5304 ext4_update_dynamic_rev(sb
);
5305 EXT4_SET_RO_COMPAT_FEATURE(sb
,
5306 EXT4_FEATURE_RO_COMPAT_LARGE_FILE
);
5308 ext4_handle_sync(handle
);
5309 err
= ext4_handle_dirty_metadata(handle
, NULL
,
5310 EXT4_SB(sb
)->s_sbh
);
5313 raw_inode
->i_generation
= cpu_to_le32(inode
->i_generation
);
5314 if (S_ISCHR(inode
->i_mode
) || S_ISBLK(inode
->i_mode
)) {
5315 if (old_valid_dev(inode
->i_rdev
)) {
5316 raw_inode
->i_block
[0] =
5317 cpu_to_le32(old_encode_dev(inode
->i_rdev
));
5318 raw_inode
->i_block
[1] = 0;
5320 raw_inode
->i_block
[0] = 0;
5321 raw_inode
->i_block
[1] =
5322 cpu_to_le32(new_encode_dev(inode
->i_rdev
));
5323 raw_inode
->i_block
[2] = 0;
5326 for (block
= 0; block
< EXT4_N_BLOCKS
; block
++)
5327 raw_inode
->i_block
[block
] = ei
->i_data
[block
];
5329 raw_inode
->i_disk_version
= cpu_to_le32(inode
->i_version
);
5330 if (ei
->i_extra_isize
) {
5331 if (EXT4_FITS_IN_INODE(raw_inode
, ei
, i_version_hi
))
5332 raw_inode
->i_version_hi
=
5333 cpu_to_le32(inode
->i_version
>> 32);
5334 raw_inode
->i_extra_isize
= cpu_to_le16(ei
->i_extra_isize
);
5337 BUFFER_TRACE(bh
, "call ext4_handle_dirty_metadata");
5338 rc
= ext4_handle_dirty_metadata(handle
, NULL
, bh
);
5341 ext4_clear_inode_state(inode
, EXT4_STATE_NEW
);
5343 ext4_update_inode_fsync_trans(handle
, inode
, 0);
5346 ext4_std_error(inode
->i_sb
, err
);
5351 * ext4_write_inode()
5353 * We are called from a few places:
5355 * - Within generic_file_write() for O_SYNC files.
5356 * Here, there will be no transaction running. We wait for any running
5357 * trasnaction to commit.
5359 * - Within sys_sync(), kupdate and such.
5360 * We wait on commit, if tol to.
5362 * - Within prune_icache() (PF_MEMALLOC == true)
5363 * Here we simply return. We can't afford to block kswapd on the
5366 * In all cases it is actually safe for us to return without doing anything,
5367 * because the inode has been copied into a raw inode buffer in
5368 * ext4_mark_inode_dirty(). This is a correctness thing for O_SYNC and for
5371 * Note that we are absolutely dependent upon all inode dirtiers doing the
5372 * right thing: they *must* call mark_inode_dirty() after dirtying info in
5373 * which we are interested.
5375 * It would be a bug for them to not do this. The code:
5377 * mark_inode_dirty(inode)
5379 * inode->i_size = expr;
5381 * is in error because a kswapd-driven write_inode() could occur while
5382 * `stuff()' is running, and the new i_size will be lost. Plus the inode
5383 * will no longer be on the superblock's dirty inode list.
5385 int ext4_write_inode(struct inode
*inode
, struct writeback_control
*wbc
)
5389 if (current
->flags
& PF_MEMALLOC
)
5392 if (EXT4_SB(inode
->i_sb
)->s_journal
) {
5393 if (ext4_journal_current_handle()) {
5394 jbd_debug(1, "called recursively, non-PF_MEMALLOC!\n");
5399 if (wbc
->sync_mode
!= WB_SYNC_ALL
)
5402 err
= ext4_force_commit(inode
->i_sb
);
5404 struct ext4_iloc iloc
;
5406 err
= __ext4_get_inode_loc(inode
, &iloc
, 0);
5409 if (wbc
->sync_mode
== WB_SYNC_ALL
)
5410 sync_dirty_buffer(iloc
.bh
);
5411 if (buffer_req(iloc
.bh
) && !buffer_uptodate(iloc
.bh
)) {
5412 EXT4_ERROR_INODE_BLOCK(inode
, iloc
.bh
->b_blocknr
,
5413 "IO error syncing inode");
5424 * Called from notify_change.
5426 * We want to trap VFS attempts to truncate the file as soon as
5427 * possible. In particular, we want to make sure that when the VFS
5428 * shrinks i_size, we put the inode on the orphan list and modify
5429 * i_disksize immediately, so that during the subsequent flushing of
5430 * dirty pages and freeing of disk blocks, we can guarantee that any
5431 * commit will leave the blocks being flushed in an unused state on
5432 * disk. (On recovery, the inode will get truncated and the blocks will
5433 * be freed, so we have a strong guarantee that no future commit will
5434 * leave these blocks visible to the user.)
5436 * Another thing we have to assure is that if we are in ordered mode
5437 * and inode is still attached to the committing transaction, we must
5438 * we start writeout of all the dirty pages which are being truncated.
5439 * This way we are sure that all the data written in the previous
5440 * transaction are already on disk (truncate waits for pages under
5443 * Called with inode->i_mutex down.
5445 int ext4_setattr(struct dentry
*dentry
, struct iattr
*attr
)
5447 struct inode
*inode
= dentry
->d_inode
;
5449 const unsigned int ia_valid
= attr
->ia_valid
;
5451 error
= inode_change_ok(inode
, attr
);
5455 if (is_quota_modification(inode
, attr
))
5456 dquot_initialize(inode
);
5457 if ((ia_valid
& ATTR_UID
&& attr
->ia_uid
!= inode
->i_uid
) ||
5458 (ia_valid
& ATTR_GID
&& attr
->ia_gid
!= inode
->i_gid
)) {
5461 /* (user+group)*(old+new) structure, inode write (sb,
5462 * inode block, ? - but truncate inode update has it) */
5463 handle
= ext4_journal_start(inode
, (EXT4_MAXQUOTAS_INIT_BLOCKS(inode
->i_sb
)+
5464 EXT4_MAXQUOTAS_DEL_BLOCKS(inode
->i_sb
))+3);
5465 if (IS_ERR(handle
)) {
5466 error
= PTR_ERR(handle
);
5469 error
= dquot_transfer(inode
, attr
);
5471 ext4_journal_stop(handle
);
5474 /* Update corresponding info in inode so that everything is in
5475 * one transaction */
5476 if (attr
->ia_valid
& ATTR_UID
)
5477 inode
->i_uid
= attr
->ia_uid
;
5478 if (attr
->ia_valid
& ATTR_GID
)
5479 inode
->i_gid
= attr
->ia_gid
;
5480 error
= ext4_mark_inode_dirty(handle
, inode
);
5481 ext4_journal_stop(handle
);
5484 if (attr
->ia_valid
& ATTR_SIZE
) {
5485 if (!(ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
))) {
5486 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
5488 if (attr
->ia_size
> sbi
->s_bitmap_maxbytes
)
5493 if (S_ISREG(inode
->i_mode
) &&
5494 attr
->ia_valid
& ATTR_SIZE
&&
5495 (attr
->ia_size
< inode
->i_size
||
5496 (ext4_test_inode_flag(inode
, EXT4_INODE_EOFBLOCKS
)))) {
5499 handle
= ext4_journal_start(inode
, 3);
5500 if (IS_ERR(handle
)) {
5501 error
= PTR_ERR(handle
);
5505 error
= ext4_orphan_add(handle
, inode
);
5506 EXT4_I(inode
)->i_disksize
= attr
->ia_size
;
5507 rc
= ext4_mark_inode_dirty(handle
, inode
);
5510 ext4_journal_stop(handle
);
5512 if (ext4_should_order_data(inode
)) {
5513 error
= ext4_begin_ordered_truncate(inode
,
5516 /* Do as much error cleanup as possible */
5517 handle
= ext4_journal_start(inode
, 3);
5518 if (IS_ERR(handle
)) {
5519 ext4_orphan_del(NULL
, inode
);
5522 ext4_orphan_del(handle
, inode
);
5523 ext4_journal_stop(handle
);
5527 /* ext4_truncate will clear the flag */
5528 if ((ext4_test_inode_flag(inode
, EXT4_INODE_EOFBLOCKS
)))
5529 ext4_truncate(inode
);
5532 if ((attr
->ia_valid
& ATTR_SIZE
) &&
5533 attr
->ia_size
!= i_size_read(inode
))
5534 rc
= vmtruncate(inode
, attr
->ia_size
);
5537 setattr_copy(inode
, attr
);
5538 mark_inode_dirty(inode
);
5542 * If the call to ext4_truncate failed to get a transaction handle at
5543 * all, we need to clean up the in-core orphan list manually.
5546 ext4_orphan_del(NULL
, inode
);
5548 if (!rc
&& (ia_valid
& ATTR_MODE
))
5549 rc
= ext4_acl_chmod(inode
);
5552 ext4_std_error(inode
->i_sb
, error
);
5558 int ext4_getattr(struct vfsmount
*mnt
, struct dentry
*dentry
,
5561 struct inode
*inode
;
5562 unsigned long delalloc_blocks
;
5564 inode
= dentry
->d_inode
;
5565 generic_fillattr(inode
, stat
);
5568 * We can't update i_blocks if the block allocation is delayed
5569 * otherwise in the case of system crash before the real block
5570 * allocation is done, we will have i_blocks inconsistent with
5571 * on-disk file blocks.
5572 * We always keep i_blocks updated together with real
5573 * allocation. But to not confuse with user, stat
5574 * will return the blocks that include the delayed allocation
5575 * blocks for this file.
5577 spin_lock(&EXT4_I(inode
)->i_block_reservation_lock
);
5578 delalloc_blocks
= EXT4_I(inode
)->i_reserved_data_blocks
;
5579 spin_unlock(&EXT4_I(inode
)->i_block_reservation_lock
);
5581 stat
->blocks
+= (delalloc_blocks
<< inode
->i_sb
->s_blocksize_bits
)>>9;
5585 static int ext4_indirect_trans_blocks(struct inode
*inode
, int nrblocks
,
5590 /* if nrblocks are contiguous */
5593 * With N contiguous data blocks, it need at most
5594 * N/EXT4_ADDR_PER_BLOCK(inode->i_sb) indirect blocks
5595 * 2 dindirect blocks
5598 indirects
= nrblocks
/ EXT4_ADDR_PER_BLOCK(inode
->i_sb
);
5599 return indirects
+ 3;
5602 * if nrblocks are not contiguous, worse case, each block touch
5603 * a indirect block, and each indirect block touch a double indirect
5604 * block, plus a triple indirect block
5606 indirects
= nrblocks
* 2 + 1;
5610 static int ext4_index_trans_blocks(struct inode
*inode
, int nrblocks
, int chunk
)
5612 if (!(ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
)))
5613 return ext4_indirect_trans_blocks(inode
, nrblocks
, chunk
);
5614 return ext4_ext_index_trans_blocks(inode
, nrblocks
, chunk
);
5618 * Account for index blocks, block groups bitmaps and block group
5619 * descriptor blocks if modify datablocks and index blocks
5620 * worse case, the indexs blocks spread over different block groups
5622 * If datablocks are discontiguous, they are possible to spread over
5623 * different block groups too. If they are contiuguous, with flexbg,
5624 * they could still across block group boundary.
5626 * Also account for superblock, inode, quota and xattr blocks
5628 int ext4_meta_trans_blocks(struct inode
*inode
, int nrblocks
, int chunk
)
5630 ext4_group_t groups
, ngroups
= ext4_get_groups_count(inode
->i_sb
);
5636 * How many index blocks need to touch to modify nrblocks?
5637 * The "Chunk" flag indicating whether the nrblocks is
5638 * physically contiguous on disk
5640 * For Direct IO and fallocate, they calls get_block to allocate
5641 * one single extent at a time, so they could set the "Chunk" flag
5643 idxblocks
= ext4_index_trans_blocks(inode
, nrblocks
, chunk
);
5648 * Now let's see how many group bitmaps and group descriptors need
5658 if (groups
> ngroups
)
5660 if (groups
> EXT4_SB(inode
->i_sb
)->s_gdb_count
)
5661 gdpblocks
= EXT4_SB(inode
->i_sb
)->s_gdb_count
;
5663 /* bitmaps and block group descriptor blocks */
5664 ret
+= groups
+ gdpblocks
;
5666 /* Blocks for super block, inode, quota and xattr blocks */
5667 ret
+= EXT4_META_TRANS_BLOCKS(inode
->i_sb
);
5673 * Calulate the total number of credits to reserve to fit
5674 * the modification of a single pages into a single transaction,
5675 * which may include multiple chunks of block allocations.
5677 * This could be called via ext4_write_begin()
5679 * We need to consider the worse case, when
5680 * one new block per extent.
5682 int ext4_writepage_trans_blocks(struct inode
*inode
)
5684 int bpp
= ext4_journal_blocks_per_page(inode
);
5687 ret
= ext4_meta_trans_blocks(inode
, bpp
, 0);
5689 /* Account for data blocks for journalled mode */
5690 if (ext4_should_journal_data(inode
))
5696 * Calculate the journal credits for a chunk of data modification.
5698 * This is called from DIO, fallocate or whoever calling
5699 * ext4_map_blocks() to map/allocate a chunk of contiguous disk blocks.
5701 * journal buffers for data blocks are not included here, as DIO
5702 * and fallocate do no need to journal data buffers.
5704 int ext4_chunk_trans_blocks(struct inode
*inode
, int nrblocks
)
5706 return ext4_meta_trans_blocks(inode
, nrblocks
, 1);
5710 * The caller must have previously called ext4_reserve_inode_write().
5711 * Give this, we know that the caller already has write access to iloc->bh.
5713 int ext4_mark_iloc_dirty(handle_t
*handle
,
5714 struct inode
*inode
, struct ext4_iloc
*iloc
)
5718 if (test_opt(inode
->i_sb
, I_VERSION
))
5719 inode_inc_iversion(inode
);
5721 /* the do_update_inode consumes one bh->b_count */
5724 /* ext4_do_update_inode() does jbd2_journal_dirty_metadata */
5725 err
= ext4_do_update_inode(handle
, inode
, iloc
);
5731 * On success, We end up with an outstanding reference count against
5732 * iloc->bh. This _must_ be cleaned up later.
5736 ext4_reserve_inode_write(handle_t
*handle
, struct inode
*inode
,
5737 struct ext4_iloc
*iloc
)
5741 err
= ext4_get_inode_loc(inode
, iloc
);
5743 BUFFER_TRACE(iloc
->bh
, "get_write_access");
5744 err
= ext4_journal_get_write_access(handle
, iloc
->bh
);
5750 ext4_std_error(inode
->i_sb
, err
);
5755 * Expand an inode by new_extra_isize bytes.
5756 * Returns 0 on success or negative error number on failure.
5758 static int ext4_expand_extra_isize(struct inode
*inode
,
5759 unsigned int new_extra_isize
,
5760 struct ext4_iloc iloc
,
5763 struct ext4_inode
*raw_inode
;
5764 struct ext4_xattr_ibody_header
*header
;
5766 if (EXT4_I(inode
)->i_extra_isize
>= new_extra_isize
)
5769 raw_inode
= ext4_raw_inode(&iloc
);
5771 header
= IHDR(inode
, raw_inode
);
5773 /* No extended attributes present */
5774 if (!ext4_test_inode_state(inode
, EXT4_STATE_XATTR
) ||
5775 header
->h_magic
!= cpu_to_le32(EXT4_XATTR_MAGIC
)) {
5776 memset((void *)raw_inode
+ EXT4_GOOD_OLD_INODE_SIZE
, 0,
5778 EXT4_I(inode
)->i_extra_isize
= new_extra_isize
;
5782 /* try to expand with EAs present */
5783 return ext4_expand_extra_isize_ea(inode
, new_extra_isize
,
5788 * What we do here is to mark the in-core inode as clean with respect to inode
5789 * dirtiness (it may still be data-dirty).
5790 * This means that the in-core inode may be reaped by prune_icache
5791 * without having to perform any I/O. This is a very good thing,
5792 * because *any* task may call prune_icache - even ones which
5793 * have a transaction open against a different journal.
5795 * Is this cheating? Not really. Sure, we haven't written the
5796 * inode out, but prune_icache isn't a user-visible syncing function.
5797 * Whenever the user wants stuff synced (sys_sync, sys_msync, sys_fsync)
5798 * we start and wait on commits.
5800 * Is this efficient/effective? Well, we're being nice to the system
5801 * by cleaning up our inodes proactively so they can be reaped
5802 * without I/O. But we are potentially leaving up to five seconds'
5803 * worth of inodes floating about which prune_icache wants us to
5804 * write out. One way to fix that would be to get prune_icache()
5805 * to do a write_super() to free up some memory. It has the desired
5808 int ext4_mark_inode_dirty(handle_t
*handle
, struct inode
*inode
)
5810 struct ext4_iloc iloc
;
5811 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
5812 static unsigned int mnt_count
;
5816 err
= ext4_reserve_inode_write(handle
, inode
, &iloc
);
5817 if (ext4_handle_valid(handle
) &&
5818 EXT4_I(inode
)->i_extra_isize
< sbi
->s_want_extra_isize
&&
5819 !ext4_test_inode_state(inode
, EXT4_STATE_NO_EXPAND
)) {
5821 * We need extra buffer credits since we may write into EA block
5822 * with this same handle. If journal_extend fails, then it will
5823 * only result in a minor loss of functionality for that inode.
5824 * If this is felt to be critical, then e2fsck should be run to
5825 * force a large enough s_min_extra_isize.
5827 if ((jbd2_journal_extend(handle
,
5828 EXT4_DATA_TRANS_BLOCKS(inode
->i_sb
))) == 0) {
5829 ret
= ext4_expand_extra_isize(inode
,
5830 sbi
->s_want_extra_isize
,
5833 ext4_set_inode_state(inode
,
5834 EXT4_STATE_NO_EXPAND
);
5836 le16_to_cpu(sbi
->s_es
->s_mnt_count
)) {
5837 ext4_warning(inode
->i_sb
,
5838 "Unable to expand inode %lu. Delete"
5839 " some EAs or run e2fsck.",
5842 le16_to_cpu(sbi
->s_es
->s_mnt_count
);
5848 err
= ext4_mark_iloc_dirty(handle
, inode
, &iloc
);
5853 * ext4_dirty_inode() is called from __mark_inode_dirty()
5855 * We're really interested in the case where a file is being extended.
5856 * i_size has been changed by generic_commit_write() and we thus need
5857 * to include the updated inode in the current transaction.
5859 * Also, dquot_alloc_block() will always dirty the inode when blocks
5860 * are allocated to the file.
5862 * If the inode is marked synchronous, we don't honour that here - doing
5863 * so would cause a commit on atime updates, which we don't bother doing.
5864 * We handle synchronous inodes at the highest possible level.
5866 void ext4_dirty_inode(struct inode
*inode
)
5870 handle
= ext4_journal_start(inode
, 2);
5874 ext4_mark_inode_dirty(handle
, inode
);
5876 ext4_journal_stop(handle
);
5883 * Bind an inode's backing buffer_head into this transaction, to prevent
5884 * it from being flushed to disk early. Unlike
5885 * ext4_reserve_inode_write, this leaves behind no bh reference and
5886 * returns no iloc structure, so the caller needs to repeat the iloc
5887 * lookup to mark the inode dirty later.
5889 static int ext4_pin_inode(handle_t
*handle
, struct inode
*inode
)
5891 struct ext4_iloc iloc
;
5895 err
= ext4_get_inode_loc(inode
, &iloc
);
5897 BUFFER_TRACE(iloc
.bh
, "get_write_access");
5898 err
= jbd2_journal_get_write_access(handle
, iloc
.bh
);
5900 err
= ext4_handle_dirty_metadata(handle
,
5906 ext4_std_error(inode
->i_sb
, err
);
5911 int ext4_change_inode_journal_flag(struct inode
*inode
, int val
)
5918 * We have to be very careful here: changing a data block's
5919 * journaling status dynamically is dangerous. If we write a
5920 * data block to the journal, change the status and then delete
5921 * that block, we risk forgetting to revoke the old log record
5922 * from the journal and so a subsequent replay can corrupt data.
5923 * So, first we make sure that the journal is empty and that
5924 * nobody is changing anything.
5927 journal
= EXT4_JOURNAL(inode
);
5930 if (is_journal_aborted(journal
))
5933 jbd2_journal_lock_updates(journal
);
5934 jbd2_journal_flush(journal
);
5937 * OK, there are no updates running now, and all cached data is
5938 * synced to disk. We are now in a completely consistent state
5939 * which doesn't have anything in the journal, and we know that
5940 * no filesystem updates are running, so it is safe to modify
5941 * the inode's in-core data-journaling state flag now.
5945 ext4_set_inode_flag(inode
, EXT4_INODE_JOURNAL_DATA
);
5947 ext4_clear_inode_flag(inode
, EXT4_INODE_JOURNAL_DATA
);
5948 ext4_set_aops(inode
);
5950 jbd2_journal_unlock_updates(journal
);
5952 /* Finally we can mark the inode as dirty. */
5954 handle
= ext4_journal_start(inode
, 1);
5956 return PTR_ERR(handle
);
5958 err
= ext4_mark_inode_dirty(handle
, inode
);
5959 ext4_handle_sync(handle
);
5960 ext4_journal_stop(handle
);
5961 ext4_std_error(inode
->i_sb
, err
);
5966 static int ext4_bh_unmapped(handle_t
*handle
, struct buffer_head
*bh
)
5968 return !buffer_mapped(bh
);
5971 int ext4_page_mkwrite(struct vm_area_struct
*vma
, struct vm_fault
*vmf
)
5973 struct page
*page
= vmf
->page
;
5978 struct file
*file
= vma
->vm_file
;
5979 struct inode
*inode
= file
->f_path
.dentry
->d_inode
;
5980 struct address_space
*mapping
= inode
->i_mapping
;
5983 * Get i_alloc_sem to stop truncates messing with the inode. We cannot
5984 * get i_mutex because we are already holding mmap_sem.
5986 down_read(&inode
->i_alloc_sem
);
5987 size
= i_size_read(inode
);
5988 if (page
->mapping
!= mapping
|| size
<= page_offset(page
)
5989 || !PageUptodate(page
)) {
5990 /* page got truncated from under us? */
5994 if (PageMappedToDisk(page
))
5997 if (page
->index
== size
>> PAGE_CACHE_SHIFT
)
5998 len
= size
& ~PAGE_CACHE_MASK
;
6000 len
= PAGE_CACHE_SIZE
;
6004 * return if we have all the buffers mapped. This avoid
6005 * the need to call write_begin/write_end which does a
6006 * journal_start/journal_stop which can block and take
6009 if (page_has_buffers(page
)) {
6010 if (!walk_page_buffers(NULL
, page_buffers(page
), 0, len
, NULL
,
6011 ext4_bh_unmapped
)) {
6018 * OK, we need to fill the hole... Do write_begin write_end
6019 * to do block allocation/reservation.We are not holding
6020 * inode.i__mutex here. That allow * parallel write_begin,
6021 * write_end call. lock_page prevent this from happening
6022 * on the same page though
6024 ret
= mapping
->a_ops
->write_begin(file
, mapping
, page_offset(page
),
6025 len
, AOP_FLAG_UNINTERRUPTIBLE
, &page
, &fsdata
);
6028 ret
= mapping
->a_ops
->write_end(file
, mapping
, page_offset(page
),
6029 len
, len
, page
, fsdata
);
6035 ret
= VM_FAULT_SIGBUS
;
6036 up_read(&inode
->i_alloc_sem
);