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>
42 #include "ext4_jbd2.h"
45 #include "ext4_extents.h"
47 #include <trace/events/ext4.h>
49 #define MPAGE_DA_EXTENT_TAIL 0x01
51 static inline int ext4_begin_ordered_truncate(struct inode
*inode
,
54 return jbd2_journal_begin_ordered_truncate(
55 EXT4_SB(inode
->i_sb
)->s_journal
,
56 &EXT4_I(inode
)->jinode
,
60 static void ext4_invalidatepage(struct page
*page
, unsigned long offset
);
63 * Test whether an inode is a fast symlink.
65 static int ext4_inode_is_fast_symlink(struct inode
*inode
)
67 int ea_blocks
= EXT4_I(inode
)->i_file_acl
?
68 (inode
->i_sb
->s_blocksize
>> 9) : 0;
70 return (S_ISLNK(inode
->i_mode
) && inode
->i_blocks
- ea_blocks
== 0);
74 * Work out how many blocks we need to proceed with the next chunk of a
75 * truncate transaction.
77 static unsigned long blocks_for_truncate(struct inode
*inode
)
81 needed
= inode
->i_blocks
>> (inode
->i_sb
->s_blocksize_bits
- 9);
83 /* Give ourselves just enough room to cope with inodes in which
84 * i_blocks is corrupt: we've seen disk corruptions in the past
85 * which resulted in random data in an inode which looked enough
86 * like a regular file for ext4 to try to delete it. Things
87 * will go a bit crazy if that happens, but at least we should
88 * try not to panic the whole kernel. */
92 /* But we need to bound the transaction so we don't overflow the
94 if (needed
> EXT4_MAX_TRANS_DATA
)
95 needed
= EXT4_MAX_TRANS_DATA
;
97 return EXT4_DATA_TRANS_BLOCKS(inode
->i_sb
) + needed
;
101 * Truncate transactions can be complex and absolutely huge. So we need to
102 * be able to restart the transaction at a conventient checkpoint to make
103 * sure we don't overflow the journal.
105 * start_transaction gets us a new handle for a truncate transaction,
106 * and extend_transaction tries to extend the existing one a bit. If
107 * extend fails, we need to propagate the failure up and restart the
108 * transaction in the top-level truncate loop. --sct
110 static handle_t
*start_transaction(struct inode
*inode
)
114 result
= ext4_journal_start(inode
, blocks_for_truncate(inode
));
118 ext4_std_error(inode
->i_sb
, PTR_ERR(result
));
123 * Try to extend this transaction for the purposes of truncation.
125 * Returns 0 if we managed to create more room. If we can't create more
126 * room, and the transaction must be restarted we return 1.
128 static int try_to_extend_transaction(handle_t
*handle
, struct inode
*inode
)
130 if (!ext4_handle_valid(handle
))
132 if (ext4_handle_has_enough_credits(handle
, EXT4_RESERVE_TRANS_BLOCKS
+1))
134 if (!ext4_journal_extend(handle
, blocks_for_truncate(inode
)))
140 * Restart the transaction associated with *handle. This does a commit,
141 * so before we call here everything must be consistently dirtied against
144 int ext4_truncate_restart_trans(handle_t
*handle
, struct inode
*inode
,
150 * Drop i_data_sem to avoid deadlock with ext4_get_blocks At this
151 * moment, get_block can be called only for blocks inside i_size since
152 * page cache has been already dropped and writes are blocked by
153 * i_mutex. So we can safely drop the i_data_sem here.
155 BUG_ON(EXT4_JOURNAL(inode
) == NULL
);
156 jbd_debug(2, "restarting handle %p\n", handle
);
157 up_write(&EXT4_I(inode
)->i_data_sem
);
158 ret
= ext4_journal_restart(handle
, blocks_for_truncate(inode
));
159 down_write(&EXT4_I(inode
)->i_data_sem
);
160 ext4_discard_preallocations(inode
);
166 * Called at the last iput() if i_nlink is zero.
168 void ext4_delete_inode(struct inode
*inode
)
173 if (ext4_should_order_data(inode
))
174 ext4_begin_ordered_truncate(inode
, 0);
175 truncate_inode_pages(&inode
->i_data
, 0);
177 if (is_bad_inode(inode
))
180 handle
= ext4_journal_start(inode
, blocks_for_truncate(inode
)+3);
181 if (IS_ERR(handle
)) {
182 ext4_std_error(inode
->i_sb
, PTR_ERR(handle
));
184 * If we're going to skip the normal cleanup, we still need to
185 * make sure that the in-core orphan linked list is properly
188 ext4_orphan_del(NULL
, inode
);
193 ext4_handle_sync(handle
);
195 err
= ext4_mark_inode_dirty(handle
, inode
);
197 ext4_warning(inode
->i_sb
, __func__
,
198 "couldn't mark inode dirty (err %d)", err
);
202 ext4_truncate(inode
);
205 * ext4_ext_truncate() doesn't reserve any slop when it
206 * restarts journal transactions; therefore there may not be
207 * enough credits left in the handle to remove the inode from
208 * the orphan list and set the dtime field.
210 if (!ext4_handle_has_enough_credits(handle
, 3)) {
211 err
= ext4_journal_extend(handle
, 3);
213 err
= ext4_journal_restart(handle
, 3);
215 ext4_warning(inode
->i_sb
, __func__
,
216 "couldn't extend journal (err %d)", err
);
218 ext4_journal_stop(handle
);
224 * Kill off the orphan record which ext4_truncate created.
225 * AKPM: I think this can be inside the above `if'.
226 * Note that ext4_orphan_del() has to be able to cope with the
227 * deletion of a non-existent orphan - this is because we don't
228 * know if ext4_truncate() actually created an orphan record.
229 * (Well, we could do this if we need to, but heck - it works)
231 ext4_orphan_del(handle
, inode
);
232 EXT4_I(inode
)->i_dtime
= get_seconds();
235 * One subtle ordering requirement: if anything has gone wrong
236 * (transaction abort, IO errors, whatever), then we can still
237 * do these next steps (the fs will already have been marked as
238 * having errors), but we can't free the inode if the mark_dirty
241 if (ext4_mark_inode_dirty(handle
, inode
))
242 /* If that failed, just do the required in-core inode clear. */
245 ext4_free_inode(handle
, inode
);
246 ext4_journal_stop(handle
);
249 clear_inode(inode
); /* We must guarantee clearing of inode... */
255 struct buffer_head
*bh
;
258 static inline void add_chain(Indirect
*p
, struct buffer_head
*bh
, __le32
*v
)
260 p
->key
= *(p
->p
= v
);
265 * ext4_block_to_path - parse the block number into array of offsets
266 * @inode: inode in question (we are only interested in its superblock)
267 * @i_block: block number to be parsed
268 * @offsets: array to store the offsets in
269 * @boundary: set this non-zero if the referred-to block is likely to be
270 * followed (on disk) by an indirect block.
272 * To store the locations of file's data ext4 uses a data structure common
273 * for UNIX filesystems - tree of pointers anchored in the inode, with
274 * data blocks at leaves and indirect blocks in intermediate nodes.
275 * This function translates the block number into path in that tree -
276 * return value is the path length and @offsets[n] is the offset of
277 * pointer to (n+1)th node in the nth one. If @block is out of range
278 * (negative or too large) warning is printed and zero returned.
280 * Note: function doesn't find node addresses, so no IO is needed. All
281 * we need to know is the capacity of indirect blocks (taken from the
286 * Portability note: the last comparison (check that we fit into triple
287 * indirect block) is spelled differently, because otherwise on an
288 * architecture with 32-bit longs and 8Kb pages we might get into trouble
289 * if our filesystem had 8Kb blocks. We might use long long, but that would
290 * kill us on x86. Oh, well, at least the sign propagation does not matter -
291 * i_block would have to be negative in the very beginning, so we would not
295 static int ext4_block_to_path(struct inode
*inode
,
297 ext4_lblk_t offsets
[4], int *boundary
)
299 int ptrs
= EXT4_ADDR_PER_BLOCK(inode
->i_sb
);
300 int ptrs_bits
= EXT4_ADDR_PER_BLOCK_BITS(inode
->i_sb
);
301 const long direct_blocks
= EXT4_NDIR_BLOCKS
,
302 indirect_blocks
= ptrs
,
303 double_blocks
= (1 << (ptrs_bits
* 2));
307 if (i_block
< direct_blocks
) {
308 offsets
[n
++] = i_block
;
309 final
= direct_blocks
;
310 } else if ((i_block
-= direct_blocks
) < indirect_blocks
) {
311 offsets
[n
++] = EXT4_IND_BLOCK
;
312 offsets
[n
++] = i_block
;
314 } else if ((i_block
-= indirect_blocks
) < double_blocks
) {
315 offsets
[n
++] = EXT4_DIND_BLOCK
;
316 offsets
[n
++] = i_block
>> ptrs_bits
;
317 offsets
[n
++] = i_block
& (ptrs
- 1);
319 } else if (((i_block
-= double_blocks
) >> (ptrs_bits
* 2)) < ptrs
) {
320 offsets
[n
++] = EXT4_TIND_BLOCK
;
321 offsets
[n
++] = i_block
>> (ptrs_bits
* 2);
322 offsets
[n
++] = (i_block
>> ptrs_bits
) & (ptrs
- 1);
323 offsets
[n
++] = i_block
& (ptrs
- 1);
326 ext4_warning(inode
->i_sb
, "ext4_block_to_path",
327 "block %lu > max in inode %lu",
328 i_block
+ direct_blocks
+
329 indirect_blocks
+ double_blocks
, inode
->i_ino
);
332 *boundary
= final
- 1 - (i_block
& (ptrs
- 1));
336 static int __ext4_check_blockref(const char *function
, struct inode
*inode
,
337 __le32
*p
, unsigned int max
)
342 while (bref
< p
+max
) {
343 blk
= le32_to_cpu(*bref
++);
345 unlikely(!ext4_data_block_valid(EXT4_SB(inode
->i_sb
),
347 ext4_error(inode
->i_sb
, function
,
348 "invalid block reference %u "
349 "in inode #%lu", blk
, inode
->i_ino
);
357 #define ext4_check_indirect_blockref(inode, bh) \
358 __ext4_check_blockref(__func__, inode, (__le32 *)(bh)->b_data, \
359 EXT4_ADDR_PER_BLOCK((inode)->i_sb))
361 #define ext4_check_inode_blockref(inode) \
362 __ext4_check_blockref(__func__, inode, EXT4_I(inode)->i_data, \
366 * ext4_get_branch - read the chain of indirect blocks leading to data
367 * @inode: inode in question
368 * @depth: depth of the chain (1 - direct pointer, etc.)
369 * @offsets: offsets of pointers in inode/indirect blocks
370 * @chain: place to store the result
371 * @err: here we store the error value
373 * Function fills the array of triples <key, p, bh> and returns %NULL
374 * if everything went OK or the pointer to the last filled triple
375 * (incomplete one) otherwise. Upon the return chain[i].key contains
376 * the number of (i+1)-th block in the chain (as it is stored in memory,
377 * i.e. little-endian 32-bit), chain[i].p contains the address of that
378 * number (it points into struct inode for i==0 and into the bh->b_data
379 * for i>0) and chain[i].bh points to the buffer_head of i-th indirect
380 * block for i>0 and NULL for i==0. In other words, it holds the block
381 * numbers of the chain, addresses they were taken from (and where we can
382 * verify that chain did not change) and buffer_heads hosting these
385 * Function stops when it stumbles upon zero pointer (absent block)
386 * (pointer to last triple returned, *@err == 0)
387 * or when it gets an IO error reading an indirect block
388 * (ditto, *@err == -EIO)
389 * or when it reads all @depth-1 indirect blocks successfully and finds
390 * the whole chain, all way to the data (returns %NULL, *err == 0).
392 * Need to be called with
393 * down_read(&EXT4_I(inode)->i_data_sem)
395 static Indirect
*ext4_get_branch(struct inode
*inode
, int depth
,
396 ext4_lblk_t
*offsets
,
397 Indirect chain
[4], int *err
)
399 struct super_block
*sb
= inode
->i_sb
;
401 struct buffer_head
*bh
;
404 /* i_data is not going away, no lock needed */
405 add_chain(chain
, NULL
, EXT4_I(inode
)->i_data
+ *offsets
);
409 bh
= sb_getblk(sb
, le32_to_cpu(p
->key
));
413 if (!bh_uptodate_or_lock(bh
)) {
414 if (bh_submit_read(bh
) < 0) {
418 /* validate block references */
419 if (ext4_check_indirect_blockref(inode
, bh
)) {
425 add_chain(++p
, bh
, (__le32
*)bh
->b_data
+ *++offsets
);
439 * ext4_find_near - find a place for allocation with sufficient locality
441 * @ind: descriptor of indirect block.
443 * This function returns the preferred place for block allocation.
444 * It is used when heuristic for sequential allocation fails.
446 * + if there is a block to the left of our position - allocate near it.
447 * + if pointer will live in indirect block - allocate near that block.
448 * + if pointer will live in inode - allocate in the same
451 * In the latter case we colour the starting block by the callers PID to
452 * prevent it from clashing with concurrent allocations for a different inode
453 * in the same block group. The PID is used here so that functionally related
454 * files will be close-by on-disk.
456 * Caller must make sure that @ind is valid and will stay that way.
458 static ext4_fsblk_t
ext4_find_near(struct inode
*inode
, Indirect
*ind
)
460 struct ext4_inode_info
*ei
= EXT4_I(inode
);
461 __le32
*start
= ind
->bh
? (__le32
*) ind
->bh
->b_data
: ei
->i_data
;
463 ext4_fsblk_t bg_start
;
464 ext4_fsblk_t last_block
;
465 ext4_grpblk_t colour
;
466 ext4_group_t block_group
;
467 int flex_size
= ext4_flex_bg_size(EXT4_SB(inode
->i_sb
));
469 /* Try to find previous block */
470 for (p
= ind
->p
- 1; p
>= start
; p
--) {
472 return le32_to_cpu(*p
);
475 /* No such thing, so let's try location of indirect block */
477 return ind
->bh
->b_blocknr
;
480 * It is going to be referred to from the inode itself? OK, just put it
481 * into the same cylinder group then.
483 block_group
= ei
->i_block_group
;
484 if (flex_size
>= EXT4_FLEX_SIZE_DIR_ALLOC_SCHEME
) {
485 block_group
&= ~(flex_size
-1);
486 if (S_ISREG(inode
->i_mode
))
489 bg_start
= ext4_group_first_block_no(inode
->i_sb
, block_group
);
490 last_block
= ext4_blocks_count(EXT4_SB(inode
->i_sb
)->s_es
) - 1;
493 * If we are doing delayed allocation, we don't need take
494 * colour into account.
496 if (test_opt(inode
->i_sb
, DELALLOC
))
499 if (bg_start
+ EXT4_BLOCKS_PER_GROUP(inode
->i_sb
) <= last_block
)
500 colour
= (current
->pid
% 16) *
501 (EXT4_BLOCKS_PER_GROUP(inode
->i_sb
) / 16);
503 colour
= (current
->pid
% 16) * ((last_block
- bg_start
) / 16);
504 return bg_start
+ colour
;
508 * ext4_find_goal - find a preferred place for allocation.
510 * @block: block we want
511 * @partial: pointer to the last triple within a chain
513 * Normally this function find the preferred place for block allocation,
515 * Because this is only used for non-extent files, we limit the block nr
518 static ext4_fsblk_t
ext4_find_goal(struct inode
*inode
, ext4_lblk_t block
,
524 * XXX need to get goal block from mballoc's data structures
527 goal
= ext4_find_near(inode
, partial
);
528 goal
= goal
& EXT4_MAX_BLOCK_FILE_PHYS
;
533 * ext4_blks_to_allocate: Look up the block map and count the number
534 * of direct blocks need to be allocated for the given branch.
536 * @branch: chain of indirect blocks
537 * @k: number of blocks need for indirect blocks
538 * @blks: number of data blocks to be mapped.
539 * @blocks_to_boundary: the offset in the indirect block
541 * return the total number of blocks to be allocate, including the
542 * direct and indirect blocks.
544 static int ext4_blks_to_allocate(Indirect
*branch
, int k
, unsigned int blks
,
545 int blocks_to_boundary
)
547 unsigned int count
= 0;
550 * Simple case, [t,d]Indirect block(s) has not allocated yet
551 * then it's clear blocks on that path have not allocated
554 /* right now we don't handle cross boundary allocation */
555 if (blks
< blocks_to_boundary
+ 1)
558 count
+= blocks_to_boundary
+ 1;
563 while (count
< blks
&& count
<= blocks_to_boundary
&&
564 le32_to_cpu(*(branch
[0].p
+ count
)) == 0) {
571 * ext4_alloc_blocks: multiple allocate blocks needed for a branch
572 * @indirect_blks: the number of blocks need to allocate for indirect
575 * @new_blocks: on return it will store the new block numbers for
576 * the indirect blocks(if needed) and the first direct block,
577 * @blks: on return it will store the total number of allocated
580 static int ext4_alloc_blocks(handle_t
*handle
, struct inode
*inode
,
581 ext4_lblk_t iblock
, ext4_fsblk_t goal
,
582 int indirect_blks
, int blks
,
583 ext4_fsblk_t new_blocks
[4], int *err
)
585 struct ext4_allocation_request ar
;
587 unsigned long count
= 0, blk_allocated
= 0;
589 ext4_fsblk_t current_block
= 0;
593 * Here we try to allocate the requested multiple blocks at once,
594 * on a best-effort basis.
595 * To build a branch, we should allocate blocks for
596 * the indirect blocks(if not allocated yet), and at least
597 * the first direct block of this branch. That's the
598 * minimum number of blocks need to allocate(required)
600 /* first we try to allocate the indirect blocks */
601 target
= indirect_blks
;
604 /* allocating blocks for indirect blocks and direct blocks */
605 current_block
= ext4_new_meta_blocks(handle
, inode
,
610 BUG_ON(current_block
+ count
> EXT4_MAX_BLOCK_FILE_PHYS
);
613 /* allocate blocks for indirect blocks */
614 while (index
< indirect_blks
&& count
) {
615 new_blocks
[index
++] = current_block
++;
620 * save the new block number
621 * for the first direct block
623 new_blocks
[index
] = current_block
;
624 printk(KERN_INFO
"%s returned more blocks than "
625 "requested\n", __func__
);
631 target
= blks
- count
;
632 blk_allocated
= count
;
635 /* Now allocate data blocks */
636 memset(&ar
, 0, sizeof(ar
));
641 if (S_ISREG(inode
->i_mode
))
642 /* enable in-core preallocation only for regular files */
643 ar
.flags
= EXT4_MB_HINT_DATA
;
645 current_block
= ext4_mb_new_blocks(handle
, &ar
, err
);
646 BUG_ON(current_block
+ ar
.len
> EXT4_MAX_BLOCK_FILE_PHYS
);
648 if (*err
&& (target
== blks
)) {
650 * if the allocation failed and we didn't allocate
656 if (target
== blks
) {
658 * save the new block number
659 * for the first direct block
661 new_blocks
[index
] = current_block
;
663 blk_allocated
+= ar
.len
;
666 /* total number of blocks allocated for direct blocks */
671 for (i
= 0; i
< index
; i
++)
672 ext4_free_blocks(handle
, inode
, 0, new_blocks
[i
], 1, 0);
677 * ext4_alloc_branch - allocate and set up a chain of blocks.
679 * @indirect_blks: number of allocated indirect blocks
680 * @blks: number of allocated direct blocks
681 * @offsets: offsets (in the blocks) to store the pointers to next.
682 * @branch: place to store the chain in.
684 * This function allocates blocks, zeroes out all but the last one,
685 * links them into chain and (if we are synchronous) writes them to disk.
686 * In other words, it prepares a branch that can be spliced onto the
687 * inode. It stores the information about that chain in the branch[], in
688 * the same format as ext4_get_branch() would do. We are calling it after
689 * we had read the existing part of chain and partial points to the last
690 * triple of that (one with zero ->key). Upon the exit we have the same
691 * picture as after the successful ext4_get_block(), except that in one
692 * place chain is disconnected - *branch->p is still zero (we did not
693 * set the last link), but branch->key contains the number that should
694 * be placed into *branch->p to fill that gap.
696 * If allocation fails we free all blocks we've allocated (and forget
697 * their buffer_heads) and return the error value the from failed
698 * ext4_alloc_block() (normally -ENOSPC). Otherwise we set the chain
699 * as described above and return 0.
701 static int ext4_alloc_branch(handle_t
*handle
, struct inode
*inode
,
702 ext4_lblk_t iblock
, int indirect_blks
,
703 int *blks
, ext4_fsblk_t goal
,
704 ext4_lblk_t
*offsets
, Indirect
*branch
)
706 int blocksize
= inode
->i_sb
->s_blocksize
;
709 struct buffer_head
*bh
;
711 ext4_fsblk_t new_blocks
[4];
712 ext4_fsblk_t current_block
;
714 num
= ext4_alloc_blocks(handle
, inode
, iblock
, goal
, indirect_blks
,
715 *blks
, new_blocks
, &err
);
719 branch
[0].key
= cpu_to_le32(new_blocks
[0]);
721 * metadata blocks and data blocks are allocated.
723 for (n
= 1; n
<= indirect_blks
; n
++) {
725 * Get buffer_head for parent block, zero it out
726 * and set the pointer to new one, then send
729 bh
= sb_getblk(inode
->i_sb
, new_blocks
[n
-1]);
732 BUFFER_TRACE(bh
, "call get_create_access");
733 err
= ext4_journal_get_create_access(handle
, bh
);
735 /* Don't brelse(bh) here; it's done in
736 * ext4_journal_forget() below */
741 memset(bh
->b_data
, 0, blocksize
);
742 branch
[n
].p
= (__le32
*) bh
->b_data
+ offsets
[n
];
743 branch
[n
].key
= cpu_to_le32(new_blocks
[n
]);
744 *branch
[n
].p
= branch
[n
].key
;
745 if (n
== indirect_blks
) {
746 current_block
= new_blocks
[n
];
748 * End of chain, update the last new metablock of
749 * the chain to point to the new allocated
750 * data blocks numbers
752 for (i
= 1; i
< num
; i
++)
753 *(branch
[n
].p
+ i
) = cpu_to_le32(++current_block
);
755 BUFFER_TRACE(bh
, "marking uptodate");
756 set_buffer_uptodate(bh
);
759 BUFFER_TRACE(bh
, "call ext4_handle_dirty_metadata");
760 err
= ext4_handle_dirty_metadata(handle
, inode
, bh
);
767 /* Allocation failed, free what we already allocated */
768 ext4_free_blocks(handle
, inode
, 0, new_blocks
[0], 1, 0);
769 for (i
= 1; i
<= n
; i
++) {
771 * branch[i].bh is newly allocated, so there is no
772 * need to revoke the block, which is why we don't
773 * need to set EXT4_FREE_BLOCKS_METADATA.
775 ext4_free_blocks(handle
, inode
, 0, new_blocks
[i
], 1,
776 EXT4_FREE_BLOCKS_FORGET
);
778 for (i
= n
+1; i
< indirect_blks
; i
++)
779 ext4_free_blocks(handle
, inode
, 0, new_blocks
[i
], 1, 0);
781 ext4_free_blocks(handle
, inode
, 0, new_blocks
[i
], num
, 0);
787 * ext4_splice_branch - splice the allocated branch onto inode.
789 * @block: (logical) number of block we are adding
790 * @chain: chain of indirect blocks (with a missing link - see
792 * @where: location of missing link
793 * @num: number of indirect blocks we are adding
794 * @blks: number of direct blocks we are adding
796 * This function fills the missing link and does all housekeeping needed in
797 * inode (->i_blocks, etc.). In case of success we end up with the full
798 * chain to new block and return 0.
800 static int ext4_splice_branch(handle_t
*handle
, struct inode
*inode
,
801 ext4_lblk_t block
, Indirect
*where
, int num
,
806 ext4_fsblk_t current_block
;
809 * If we're splicing into a [td]indirect block (as opposed to the
810 * inode) then we need to get write access to the [td]indirect block
814 BUFFER_TRACE(where
->bh
, "get_write_access");
815 err
= ext4_journal_get_write_access(handle
, where
->bh
);
821 *where
->p
= where
->key
;
824 * Update the host buffer_head or inode to point to more just allocated
825 * direct blocks blocks
827 if (num
== 0 && blks
> 1) {
828 current_block
= le32_to_cpu(where
->key
) + 1;
829 for (i
= 1; i
< blks
; i
++)
830 *(where
->p
+ i
) = cpu_to_le32(current_block
++);
833 /* We are done with atomic stuff, now do the rest of housekeeping */
834 /* had we spliced it onto indirect block? */
837 * If we spliced it onto an indirect block, we haven't
838 * altered the inode. Note however that if it is being spliced
839 * onto an indirect block at the very end of the file (the
840 * file is growing) then we *will* alter the inode to reflect
841 * the new i_size. But that is not done here - it is done in
842 * generic_commit_write->__mark_inode_dirty->ext4_dirty_inode.
844 jbd_debug(5, "splicing indirect only\n");
845 BUFFER_TRACE(where
->bh
, "call ext4_handle_dirty_metadata");
846 err
= ext4_handle_dirty_metadata(handle
, inode
, where
->bh
);
851 * OK, we spliced it into the inode itself on a direct block.
853 ext4_mark_inode_dirty(handle
, inode
);
854 jbd_debug(5, "splicing direct\n");
859 for (i
= 1; i
<= num
; i
++) {
861 * branch[i].bh is newly allocated, so there is no
862 * need to revoke the block, which is why we don't
863 * need to set EXT4_FREE_BLOCKS_METADATA.
865 ext4_free_blocks(handle
, inode
, where
[i
].bh
, 0, 1,
866 EXT4_FREE_BLOCKS_FORGET
);
868 ext4_free_blocks(handle
, inode
, 0, le32_to_cpu(where
[num
].key
),
875 * The ext4_ind_get_blocks() function handles non-extents inodes
876 * (i.e., using the traditional indirect/double-indirect i_blocks
877 * scheme) for ext4_get_blocks().
879 * Allocation strategy is simple: if we have to allocate something, we will
880 * have to go the whole way to leaf. So let's do it before attaching anything
881 * to tree, set linkage between the newborn blocks, write them if sync is
882 * required, recheck the path, free and repeat if check fails, otherwise
883 * set the last missing link (that will protect us from any truncate-generated
884 * removals - all blocks on the path are immune now) and possibly force the
885 * write on the parent block.
886 * That has a nice additional property: no special recovery from the failed
887 * allocations is needed - we simply release blocks and do not touch anything
888 * reachable from inode.
890 * `handle' can be NULL if create == 0.
892 * return > 0, # of blocks mapped or allocated.
893 * return = 0, if plain lookup failed.
894 * return < 0, error case.
896 * The ext4_ind_get_blocks() function should be called with
897 * down_write(&EXT4_I(inode)->i_data_sem) if allocating filesystem
898 * blocks (i.e., flags has EXT4_GET_BLOCKS_CREATE set) or
899 * down_read(&EXT4_I(inode)->i_data_sem) if not allocating file system
902 static int ext4_ind_get_blocks(handle_t
*handle
, struct inode
*inode
,
903 ext4_lblk_t iblock
, unsigned int maxblocks
,
904 struct buffer_head
*bh_result
,
908 ext4_lblk_t offsets
[4];
913 int blocks_to_boundary
= 0;
916 ext4_fsblk_t first_block
= 0;
918 J_ASSERT(!(EXT4_I(inode
)->i_flags
& EXT4_EXTENTS_FL
));
919 J_ASSERT(handle
!= NULL
|| (flags
& EXT4_GET_BLOCKS_CREATE
) == 0);
920 depth
= ext4_block_to_path(inode
, iblock
, offsets
,
921 &blocks_to_boundary
);
926 partial
= ext4_get_branch(inode
, depth
, offsets
, chain
, &err
);
928 /* Simplest case - block found, no allocation needed */
930 first_block
= le32_to_cpu(chain
[depth
- 1].key
);
931 clear_buffer_new(bh_result
);
934 while (count
< maxblocks
&& count
<= blocks_to_boundary
) {
937 blk
= le32_to_cpu(*(chain
[depth
-1].p
+ count
));
939 if (blk
== first_block
+ count
)
947 /* Next simple case - plain lookup or failed read of indirect block */
948 if ((flags
& EXT4_GET_BLOCKS_CREATE
) == 0 || err
== -EIO
)
952 * Okay, we need to do block allocation.
954 goal
= ext4_find_goal(inode
, iblock
, partial
);
956 /* the number of blocks need to allocate for [d,t]indirect blocks */
957 indirect_blks
= (chain
+ depth
) - partial
- 1;
960 * Next look up the indirect map to count the totoal number of
961 * direct blocks to allocate for this branch.
963 count
= ext4_blks_to_allocate(partial
, indirect_blks
,
964 maxblocks
, blocks_to_boundary
);
966 * Block out ext4_truncate while we alter the tree
968 err
= ext4_alloc_branch(handle
, inode
, iblock
, indirect_blks
,
970 offsets
+ (partial
- chain
), partial
);
973 * The ext4_splice_branch call will free and forget any buffers
974 * on the new chain if there is a failure, but that risks using
975 * up transaction credits, especially for bitmaps where the
976 * credits cannot be returned. Can we handle this somehow? We
977 * may need to return -EAGAIN upwards in the worst case. --sct
980 err
= ext4_splice_branch(handle
, inode
, iblock
,
981 partial
, indirect_blks
, count
);
985 set_buffer_new(bh_result
);
987 map_bh(bh_result
, inode
->i_sb
, le32_to_cpu(chain
[depth
-1].key
));
988 if (count
> blocks_to_boundary
)
989 set_buffer_boundary(bh_result
);
991 /* Clean up and exit */
992 partial
= chain
+ depth
- 1; /* the whole chain */
994 while (partial
> chain
) {
995 BUFFER_TRACE(partial
->bh
, "call brelse");
999 BUFFER_TRACE(bh_result
, "returned");
1004 qsize_t
ext4_get_reserved_space(struct inode
*inode
)
1006 unsigned long long total
;
1008 spin_lock(&EXT4_I(inode
)->i_block_reservation_lock
);
1009 total
= EXT4_I(inode
)->i_reserved_data_blocks
+
1010 EXT4_I(inode
)->i_reserved_meta_blocks
;
1011 spin_unlock(&EXT4_I(inode
)->i_block_reservation_lock
);
1013 return (total
<< inode
->i_blkbits
);
1016 * Calculate the number of metadata blocks need to reserve
1017 * to allocate @blocks for non extent file based file
1019 static int ext4_indirect_calc_metadata_amount(struct inode
*inode
, int blocks
)
1021 int icap
= EXT4_ADDR_PER_BLOCK(inode
->i_sb
);
1022 int ind_blks
, dind_blks
, tind_blks
;
1024 /* number of new indirect blocks needed */
1025 ind_blks
= (blocks
+ icap
- 1) / icap
;
1027 dind_blks
= (ind_blks
+ icap
- 1) / icap
;
1031 return ind_blks
+ dind_blks
+ tind_blks
;
1035 * Calculate the number of metadata blocks need to reserve
1036 * to allocate given number of blocks
1038 static int ext4_calc_metadata_amount(struct inode
*inode
, int blocks
)
1043 if (EXT4_I(inode
)->i_flags
& EXT4_EXTENTS_FL
)
1044 return ext4_ext_calc_metadata_amount(inode
, blocks
);
1046 return ext4_indirect_calc_metadata_amount(inode
, blocks
);
1049 static void ext4_da_update_reserve_space(struct inode
*inode
, int used
)
1051 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
1052 int total
, mdb
, mdb_free
;
1054 spin_lock(&EXT4_I(inode
)->i_block_reservation_lock
);
1055 /* recalculate the number of metablocks still need to be reserved */
1056 total
= EXT4_I(inode
)->i_reserved_data_blocks
- used
;
1057 mdb
= ext4_calc_metadata_amount(inode
, total
);
1059 /* figure out how many metablocks to release */
1060 BUG_ON(mdb
> EXT4_I(inode
)->i_reserved_meta_blocks
);
1061 mdb_free
= EXT4_I(inode
)->i_reserved_meta_blocks
- mdb
;
1064 /* Account for allocated meta_blocks */
1065 mdb_free
-= EXT4_I(inode
)->i_allocated_meta_blocks
;
1067 /* update fs dirty blocks counter */
1068 percpu_counter_sub(&sbi
->s_dirtyblocks_counter
, mdb_free
);
1069 EXT4_I(inode
)->i_allocated_meta_blocks
= 0;
1070 EXT4_I(inode
)->i_reserved_meta_blocks
= mdb
;
1073 /* update per-inode reservations */
1074 BUG_ON(used
> EXT4_I(inode
)->i_reserved_data_blocks
);
1075 EXT4_I(inode
)->i_reserved_data_blocks
-= used
;
1076 spin_unlock(&EXT4_I(inode
)->i_block_reservation_lock
);
1079 * free those over-booking quota for metadata blocks
1082 vfs_dq_release_reservation_block(inode
, mdb_free
);
1085 * If we have done all the pending block allocations and if
1086 * there aren't any writers on the inode, we can discard the
1087 * inode's preallocations.
1089 if (!total
&& (atomic_read(&inode
->i_writecount
) == 0))
1090 ext4_discard_preallocations(inode
);
1093 static int check_block_validity(struct inode
*inode
, const char *msg
,
1094 sector_t logical
, sector_t phys
, int len
)
1096 if (!ext4_data_block_valid(EXT4_SB(inode
->i_sb
), phys
, len
)) {
1097 ext4_error(inode
->i_sb
, msg
,
1098 "inode #%lu logical block %llu mapped to %llu "
1099 "(size %d)", inode
->i_ino
,
1100 (unsigned long long) logical
,
1101 (unsigned long long) phys
, len
);
1108 * Return the number of contiguous dirty pages in a given inode
1109 * starting at page frame idx.
1111 static pgoff_t
ext4_num_dirty_pages(struct inode
*inode
, pgoff_t idx
,
1112 unsigned int max_pages
)
1114 struct address_space
*mapping
= inode
->i_mapping
;
1116 struct pagevec pvec
;
1118 int i
, nr_pages
, done
= 0;
1122 pagevec_init(&pvec
, 0);
1125 nr_pages
= pagevec_lookup_tag(&pvec
, mapping
, &index
,
1126 PAGECACHE_TAG_DIRTY
,
1127 (pgoff_t
)PAGEVEC_SIZE
);
1130 for (i
= 0; i
< nr_pages
; i
++) {
1131 struct page
*page
= pvec
.pages
[i
];
1132 struct buffer_head
*bh
, *head
;
1135 if (unlikely(page
->mapping
!= mapping
) ||
1137 PageWriteback(page
) ||
1138 page
->index
!= idx
) {
1143 if (page_has_buffers(page
)) {
1144 bh
= head
= page_buffers(page
);
1146 if (!buffer_delay(bh
) &&
1147 !buffer_unwritten(bh
))
1149 bh
= bh
->b_this_page
;
1150 } while (!done
&& (bh
!= head
));
1157 if (num
>= max_pages
)
1160 pagevec_release(&pvec
);
1166 * The ext4_get_blocks() function tries to look up the requested blocks,
1167 * and returns if the blocks are already mapped.
1169 * Otherwise it takes the write lock of the i_data_sem and allocate blocks
1170 * and store the allocated blocks in the result buffer head and mark it
1173 * If file type is extents based, it will call ext4_ext_get_blocks(),
1174 * Otherwise, call with ext4_ind_get_blocks() to handle indirect mapping
1177 * On success, it returns the number of blocks being mapped or allocate.
1178 * if create==0 and the blocks are pre-allocated and uninitialized block,
1179 * the result buffer head is unmapped. If the create ==1, it will make sure
1180 * the buffer head is mapped.
1182 * It returns 0 if plain look up failed (blocks have not been allocated), in
1183 * that casem, buffer head is unmapped
1185 * It returns the error in case of allocation failure.
1187 int ext4_get_blocks(handle_t
*handle
, struct inode
*inode
, sector_t block
,
1188 unsigned int max_blocks
, struct buffer_head
*bh
,
1193 clear_buffer_mapped(bh
);
1194 clear_buffer_unwritten(bh
);
1196 ext_debug("ext4_get_blocks(): inode %lu, flag %d, max_blocks %u,"
1197 "logical block %lu\n", inode
->i_ino
, flags
, max_blocks
,
1198 (unsigned long)block
);
1200 * Try to see if we can get the block without requesting a new
1201 * file system block.
1203 down_read((&EXT4_I(inode
)->i_data_sem
));
1204 if (EXT4_I(inode
)->i_flags
& EXT4_EXTENTS_FL
) {
1205 retval
= ext4_ext_get_blocks(handle
, inode
, block
, max_blocks
,
1208 retval
= ext4_ind_get_blocks(handle
, inode
, block
, max_blocks
,
1211 up_read((&EXT4_I(inode
)->i_data_sem
));
1213 if (retval
> 0 && buffer_mapped(bh
)) {
1214 int ret
= check_block_validity(inode
, "file system corruption",
1215 block
, bh
->b_blocknr
, retval
);
1220 /* If it is only a block(s) look up */
1221 if ((flags
& EXT4_GET_BLOCKS_CREATE
) == 0)
1225 * Returns if the blocks have already allocated
1227 * Note that if blocks have been preallocated
1228 * ext4_ext_get_block() returns th create = 0
1229 * with buffer head unmapped.
1231 if (retval
> 0 && buffer_mapped(bh
))
1235 * When we call get_blocks without the create flag, the
1236 * BH_Unwritten flag could have gotten set if the blocks
1237 * requested were part of a uninitialized extent. We need to
1238 * clear this flag now that we are committed to convert all or
1239 * part of the uninitialized extent to be an initialized
1240 * extent. This is because we need to avoid the combination
1241 * of BH_Unwritten and BH_Mapped flags being simultaneously
1242 * set on the buffer_head.
1244 clear_buffer_unwritten(bh
);
1247 * New blocks allocate and/or writing to uninitialized extent
1248 * will possibly result in updating i_data, so we take
1249 * the write lock of i_data_sem, and call get_blocks()
1250 * with create == 1 flag.
1252 down_write((&EXT4_I(inode
)->i_data_sem
));
1255 * if the caller is from delayed allocation writeout path
1256 * we have already reserved fs blocks for allocation
1257 * let the underlying get_block() function know to
1258 * avoid double accounting
1260 if (flags
& EXT4_GET_BLOCKS_DELALLOC_RESERVE
)
1261 EXT4_I(inode
)->i_delalloc_reserved_flag
= 1;
1263 * We need to check for EXT4 here because migrate
1264 * could have changed the inode type in between
1266 if (EXT4_I(inode
)->i_flags
& EXT4_EXTENTS_FL
) {
1267 retval
= ext4_ext_get_blocks(handle
, inode
, block
, max_blocks
,
1270 retval
= ext4_ind_get_blocks(handle
, inode
, block
,
1271 max_blocks
, bh
, flags
);
1273 if (retval
> 0 && buffer_new(bh
)) {
1275 * We allocated new blocks which will result in
1276 * i_data's format changing. Force the migrate
1277 * to fail by clearing migrate flags
1279 EXT4_I(inode
)->i_state
&= ~EXT4_STATE_EXT_MIGRATE
;
1283 if (flags
& EXT4_GET_BLOCKS_DELALLOC_RESERVE
)
1284 EXT4_I(inode
)->i_delalloc_reserved_flag
= 0;
1287 * Update reserved blocks/metadata blocks after successful
1288 * block allocation which had been deferred till now.
1290 if ((retval
> 0) && (flags
& EXT4_GET_BLOCKS_UPDATE_RESERVE_SPACE
))
1291 ext4_da_update_reserve_space(inode
, retval
);
1293 up_write((&EXT4_I(inode
)->i_data_sem
));
1294 if (retval
> 0 && buffer_mapped(bh
)) {
1295 int ret
= check_block_validity(inode
, "file system "
1296 "corruption after allocation",
1297 block
, bh
->b_blocknr
, retval
);
1304 /* Maximum number of blocks we map for direct IO at once. */
1305 #define DIO_MAX_BLOCKS 4096
1307 int ext4_get_block(struct inode
*inode
, sector_t iblock
,
1308 struct buffer_head
*bh_result
, int create
)
1310 handle_t
*handle
= ext4_journal_current_handle();
1311 int ret
= 0, started
= 0;
1312 unsigned max_blocks
= bh_result
->b_size
>> inode
->i_blkbits
;
1315 if (create
&& !handle
) {
1316 /* Direct IO write... */
1317 if (max_blocks
> DIO_MAX_BLOCKS
)
1318 max_blocks
= DIO_MAX_BLOCKS
;
1319 dio_credits
= ext4_chunk_trans_blocks(inode
, max_blocks
);
1320 handle
= ext4_journal_start(inode
, dio_credits
);
1321 if (IS_ERR(handle
)) {
1322 ret
= PTR_ERR(handle
);
1328 ret
= ext4_get_blocks(handle
, inode
, iblock
, max_blocks
, bh_result
,
1329 create
? EXT4_GET_BLOCKS_CREATE
: 0);
1331 bh_result
->b_size
= (ret
<< inode
->i_blkbits
);
1335 ext4_journal_stop(handle
);
1341 * `handle' can be NULL if create is zero
1343 struct buffer_head
*ext4_getblk(handle_t
*handle
, struct inode
*inode
,
1344 ext4_lblk_t block
, int create
, int *errp
)
1346 struct buffer_head dummy
;
1350 J_ASSERT(handle
!= NULL
|| create
== 0);
1353 dummy
.b_blocknr
= -1000;
1354 buffer_trace_init(&dummy
.b_history
);
1356 flags
|= EXT4_GET_BLOCKS_CREATE
;
1357 err
= ext4_get_blocks(handle
, inode
, block
, 1, &dummy
, flags
);
1359 * ext4_get_blocks() returns number of blocks mapped. 0 in
1368 if (!err
&& buffer_mapped(&dummy
)) {
1369 struct buffer_head
*bh
;
1370 bh
= sb_getblk(inode
->i_sb
, dummy
.b_blocknr
);
1375 if (buffer_new(&dummy
)) {
1376 J_ASSERT(create
!= 0);
1377 J_ASSERT(handle
!= NULL
);
1380 * Now that we do not always journal data, we should
1381 * keep in mind whether this should always journal the
1382 * new buffer as metadata. For now, regular file
1383 * writes use ext4_get_block instead, so it's not a
1387 BUFFER_TRACE(bh
, "call get_create_access");
1388 fatal
= ext4_journal_get_create_access(handle
, bh
);
1389 if (!fatal
&& !buffer_uptodate(bh
)) {
1390 memset(bh
->b_data
, 0, inode
->i_sb
->s_blocksize
);
1391 set_buffer_uptodate(bh
);
1394 BUFFER_TRACE(bh
, "call ext4_handle_dirty_metadata");
1395 err
= ext4_handle_dirty_metadata(handle
, inode
, bh
);
1399 BUFFER_TRACE(bh
, "not a new buffer");
1412 struct buffer_head
*ext4_bread(handle_t
*handle
, struct inode
*inode
,
1413 ext4_lblk_t block
, int create
, int *err
)
1415 struct buffer_head
*bh
;
1417 bh
= ext4_getblk(handle
, inode
, block
, create
, err
);
1420 if (buffer_uptodate(bh
))
1422 ll_rw_block(READ_META
, 1, &bh
);
1424 if (buffer_uptodate(bh
))
1431 static int walk_page_buffers(handle_t
*handle
,
1432 struct buffer_head
*head
,
1436 int (*fn
)(handle_t
*handle
,
1437 struct buffer_head
*bh
))
1439 struct buffer_head
*bh
;
1440 unsigned block_start
, block_end
;
1441 unsigned blocksize
= head
->b_size
;
1443 struct buffer_head
*next
;
1445 for (bh
= head
, block_start
= 0;
1446 ret
== 0 && (bh
!= head
|| !block_start
);
1447 block_start
= block_end
, bh
= next
) {
1448 next
= bh
->b_this_page
;
1449 block_end
= block_start
+ blocksize
;
1450 if (block_end
<= from
|| block_start
>= to
) {
1451 if (partial
&& !buffer_uptodate(bh
))
1455 err
= (*fn
)(handle
, bh
);
1463 * To preserve ordering, it is essential that the hole instantiation and
1464 * the data write be encapsulated in a single transaction. We cannot
1465 * close off a transaction and start a new one between the ext4_get_block()
1466 * and the commit_write(). So doing the jbd2_journal_start at the start of
1467 * prepare_write() is the right place.
1469 * Also, this function can nest inside ext4_writepage() ->
1470 * block_write_full_page(). In that case, we *know* that ext4_writepage()
1471 * has generated enough buffer credits to do the whole page. So we won't
1472 * block on the journal in that case, which is good, because the caller may
1475 * By accident, ext4 can be reentered when a transaction is open via
1476 * quota file writes. If we were to commit the transaction while thus
1477 * reentered, there can be a deadlock - we would be holding a quota
1478 * lock, and the commit would never complete if another thread had a
1479 * transaction open and was blocking on the quota lock - a ranking
1482 * So what we do is to rely on the fact that jbd2_journal_stop/journal_start
1483 * will _not_ run commit under these circumstances because handle->h_ref
1484 * is elevated. We'll still have enough credits for the tiny quotafile
1487 static int do_journal_get_write_access(handle_t
*handle
,
1488 struct buffer_head
*bh
)
1490 if (!buffer_mapped(bh
) || buffer_freed(bh
))
1492 return ext4_journal_get_write_access(handle
, bh
);
1496 * Truncate blocks that were not used by write. We have to truncate the
1497 * pagecache as well so that corresponding buffers get properly unmapped.
1499 static void ext4_truncate_failed_write(struct inode
*inode
)
1501 truncate_inode_pages(inode
->i_mapping
, inode
->i_size
);
1502 ext4_truncate(inode
);
1505 static int ext4_write_begin(struct file
*file
, struct address_space
*mapping
,
1506 loff_t pos
, unsigned len
, unsigned flags
,
1507 struct page
**pagep
, void **fsdata
)
1509 struct inode
*inode
= mapping
->host
;
1510 int ret
, needed_blocks
;
1517 trace_ext4_write_begin(inode
, pos
, len
, flags
);
1519 * Reserve one block more for addition to orphan list in case
1520 * we allocate blocks but write fails for some reason
1522 needed_blocks
= ext4_writepage_trans_blocks(inode
) + 1;
1523 index
= pos
>> PAGE_CACHE_SHIFT
;
1524 from
= pos
& (PAGE_CACHE_SIZE
- 1);
1528 handle
= ext4_journal_start(inode
, needed_blocks
);
1529 if (IS_ERR(handle
)) {
1530 ret
= PTR_ERR(handle
);
1534 /* We cannot recurse into the filesystem as the transaction is already
1536 flags
|= AOP_FLAG_NOFS
;
1538 page
= grab_cache_page_write_begin(mapping
, index
, flags
);
1540 ext4_journal_stop(handle
);
1546 ret
= block_write_begin(file
, mapping
, pos
, len
, flags
, pagep
, fsdata
,
1549 if (!ret
&& ext4_should_journal_data(inode
)) {
1550 ret
= walk_page_buffers(handle
, page_buffers(page
),
1551 from
, to
, NULL
, do_journal_get_write_access
);
1556 page_cache_release(page
);
1558 * block_write_begin may have instantiated a few blocks
1559 * outside i_size. Trim these off again. Don't need
1560 * i_size_read because we hold i_mutex.
1562 * Add inode to orphan list in case we crash before
1565 if (pos
+ len
> inode
->i_size
&& ext4_can_truncate(inode
))
1566 ext4_orphan_add(handle
, inode
);
1568 ext4_journal_stop(handle
);
1569 if (pos
+ len
> inode
->i_size
) {
1570 ext4_truncate_failed_write(inode
);
1572 * If truncate failed early the inode might
1573 * still be on the orphan list; we need to
1574 * make sure the inode is removed from the
1575 * orphan list in that case.
1578 ext4_orphan_del(NULL
, inode
);
1582 if (ret
== -ENOSPC
&& ext4_should_retry_alloc(inode
->i_sb
, &retries
))
1588 /* For write_end() in data=journal mode */
1589 static int write_end_fn(handle_t
*handle
, struct buffer_head
*bh
)
1591 if (!buffer_mapped(bh
) || buffer_freed(bh
))
1593 set_buffer_uptodate(bh
);
1594 return ext4_handle_dirty_metadata(handle
, NULL
, bh
);
1597 static int ext4_generic_write_end(struct file
*file
,
1598 struct address_space
*mapping
,
1599 loff_t pos
, unsigned len
, unsigned copied
,
1600 struct page
*page
, void *fsdata
)
1602 int i_size_changed
= 0;
1603 struct inode
*inode
= mapping
->host
;
1604 handle_t
*handle
= ext4_journal_current_handle();
1606 copied
= block_write_end(file
, mapping
, pos
, len
, copied
, page
, fsdata
);
1609 * No need to use i_size_read() here, the i_size
1610 * cannot change under us because we hold i_mutex.
1612 * But it's important to update i_size while still holding page lock:
1613 * page writeout could otherwise come in and zero beyond i_size.
1615 if (pos
+ copied
> inode
->i_size
) {
1616 i_size_write(inode
, pos
+ copied
);
1620 if (pos
+ copied
> EXT4_I(inode
)->i_disksize
) {
1621 /* We need to mark inode dirty even if
1622 * new_i_size is less that inode->i_size
1623 * bu greater than i_disksize.(hint delalloc)
1625 ext4_update_i_disksize(inode
, (pos
+ copied
));
1629 page_cache_release(page
);
1632 * Don't mark the inode dirty under page lock. First, it unnecessarily
1633 * makes the holding time of page lock longer. Second, it forces lock
1634 * ordering of page lock and transaction start for journaling
1638 ext4_mark_inode_dirty(handle
, inode
);
1644 * We need to pick up the new inode size which generic_commit_write gave us
1645 * `file' can be NULL - eg, when called from page_symlink().
1647 * ext4 never places buffers on inode->i_mapping->private_list. metadata
1648 * buffers are managed internally.
1650 static int ext4_ordered_write_end(struct file
*file
,
1651 struct address_space
*mapping
,
1652 loff_t pos
, unsigned len
, unsigned copied
,
1653 struct page
*page
, void *fsdata
)
1655 handle_t
*handle
= ext4_journal_current_handle();
1656 struct inode
*inode
= mapping
->host
;
1659 trace_ext4_ordered_write_end(inode
, pos
, len
, copied
);
1660 ret
= ext4_jbd2_file_inode(handle
, inode
);
1663 ret2
= ext4_generic_write_end(file
, mapping
, pos
, len
, copied
,
1666 if (pos
+ len
> inode
->i_size
&& ext4_can_truncate(inode
))
1667 /* if we have allocated more blocks and copied
1668 * less. We will have blocks allocated outside
1669 * inode->i_size. So truncate them
1671 ext4_orphan_add(handle
, inode
);
1675 ret2
= ext4_journal_stop(handle
);
1679 if (pos
+ len
> inode
->i_size
) {
1680 ext4_truncate_failed_write(inode
);
1682 * If truncate failed early the inode might still be
1683 * on the orphan list; we need to make sure the inode
1684 * is removed from the orphan list in that case.
1687 ext4_orphan_del(NULL
, inode
);
1691 return ret
? ret
: copied
;
1694 static int ext4_writeback_write_end(struct file
*file
,
1695 struct address_space
*mapping
,
1696 loff_t pos
, unsigned len
, unsigned copied
,
1697 struct page
*page
, void *fsdata
)
1699 handle_t
*handle
= ext4_journal_current_handle();
1700 struct inode
*inode
= mapping
->host
;
1703 trace_ext4_writeback_write_end(inode
, pos
, len
, copied
);
1704 ret2
= ext4_generic_write_end(file
, mapping
, pos
, len
, copied
,
1707 if (pos
+ len
> inode
->i_size
&& ext4_can_truncate(inode
))
1708 /* if we have allocated more blocks and copied
1709 * less. We will have blocks allocated outside
1710 * inode->i_size. So truncate them
1712 ext4_orphan_add(handle
, inode
);
1717 ret2
= ext4_journal_stop(handle
);
1721 if (pos
+ len
> inode
->i_size
) {
1722 ext4_truncate_failed_write(inode
);
1724 * If truncate failed early the inode might still be
1725 * on the orphan list; we need to make sure the inode
1726 * is removed from the orphan list in that case.
1729 ext4_orphan_del(NULL
, inode
);
1732 return ret
? ret
: copied
;
1735 static int ext4_journalled_write_end(struct file
*file
,
1736 struct address_space
*mapping
,
1737 loff_t pos
, unsigned len
, unsigned copied
,
1738 struct page
*page
, void *fsdata
)
1740 handle_t
*handle
= ext4_journal_current_handle();
1741 struct inode
*inode
= mapping
->host
;
1747 trace_ext4_journalled_write_end(inode
, pos
, len
, copied
);
1748 from
= pos
& (PAGE_CACHE_SIZE
- 1);
1752 if (!PageUptodate(page
))
1754 page_zero_new_buffers(page
, from
+copied
, to
);
1757 ret
= walk_page_buffers(handle
, page_buffers(page
), from
,
1758 to
, &partial
, write_end_fn
);
1760 SetPageUptodate(page
);
1761 new_i_size
= pos
+ copied
;
1762 if (new_i_size
> inode
->i_size
)
1763 i_size_write(inode
, pos
+copied
);
1764 EXT4_I(inode
)->i_state
|= EXT4_STATE_JDATA
;
1765 if (new_i_size
> EXT4_I(inode
)->i_disksize
) {
1766 ext4_update_i_disksize(inode
, new_i_size
);
1767 ret2
= ext4_mark_inode_dirty(handle
, inode
);
1773 page_cache_release(page
);
1774 if (pos
+ len
> inode
->i_size
&& ext4_can_truncate(inode
))
1775 /* if we have allocated more blocks and copied
1776 * less. We will have blocks allocated outside
1777 * inode->i_size. So truncate them
1779 ext4_orphan_add(handle
, inode
);
1781 ret2
= ext4_journal_stop(handle
);
1784 if (pos
+ len
> inode
->i_size
) {
1785 ext4_truncate_failed_write(inode
);
1787 * If truncate failed early the inode might still be
1788 * on the orphan list; we need to make sure the inode
1789 * is removed from the orphan list in that case.
1792 ext4_orphan_del(NULL
, inode
);
1795 return ret
? ret
: copied
;
1798 static int ext4_da_reserve_space(struct inode
*inode
, int nrblocks
)
1801 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
1802 unsigned long md_needed
, mdblocks
, total
= 0;
1805 * recalculate the amount of metadata blocks to reserve
1806 * in order to allocate nrblocks
1807 * worse case is one extent per block
1810 spin_lock(&EXT4_I(inode
)->i_block_reservation_lock
);
1811 total
= EXT4_I(inode
)->i_reserved_data_blocks
+ nrblocks
;
1812 mdblocks
= ext4_calc_metadata_amount(inode
, total
);
1813 BUG_ON(mdblocks
< EXT4_I(inode
)->i_reserved_meta_blocks
);
1815 md_needed
= mdblocks
- EXT4_I(inode
)->i_reserved_meta_blocks
;
1816 total
= md_needed
+ nrblocks
;
1819 * Make quota reservation here to prevent quota overflow
1820 * later. Real quota accounting is done at pages writeout
1823 if (vfs_dq_reserve_block(inode
, total
)) {
1824 spin_unlock(&EXT4_I(inode
)->i_block_reservation_lock
);
1828 if (ext4_claim_free_blocks(sbi
, total
)) {
1829 spin_unlock(&EXT4_I(inode
)->i_block_reservation_lock
);
1830 vfs_dq_release_reservation_block(inode
, total
);
1831 if (ext4_should_retry_alloc(inode
->i_sb
, &retries
)) {
1837 EXT4_I(inode
)->i_reserved_data_blocks
+= nrblocks
;
1838 EXT4_I(inode
)->i_reserved_meta_blocks
= mdblocks
;
1840 spin_unlock(&EXT4_I(inode
)->i_block_reservation_lock
);
1841 return 0; /* success */
1844 static void ext4_da_release_space(struct inode
*inode
, int to_free
)
1846 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
1847 int total
, mdb
, mdb_free
, release
;
1850 return; /* Nothing to release, exit */
1852 spin_lock(&EXT4_I(inode
)->i_block_reservation_lock
);
1854 if (!EXT4_I(inode
)->i_reserved_data_blocks
) {
1856 * if there is no reserved blocks, but we try to free some
1857 * then the counter is messed up somewhere.
1858 * but since this function is called from invalidate
1859 * page, it's harmless to return without any action
1861 printk(KERN_INFO
"ext4 delalloc try to release %d reserved "
1862 "blocks for inode %lu, but there is no reserved "
1863 "data blocks\n", to_free
, inode
->i_ino
);
1864 spin_unlock(&EXT4_I(inode
)->i_block_reservation_lock
);
1868 /* recalculate the number of metablocks still need to be reserved */
1869 total
= EXT4_I(inode
)->i_reserved_data_blocks
- to_free
;
1870 mdb
= ext4_calc_metadata_amount(inode
, total
);
1872 /* figure out how many metablocks to release */
1873 BUG_ON(mdb
> EXT4_I(inode
)->i_reserved_meta_blocks
);
1874 mdb_free
= EXT4_I(inode
)->i_reserved_meta_blocks
- mdb
;
1876 release
= to_free
+ mdb_free
;
1878 /* update fs dirty blocks counter for truncate case */
1879 percpu_counter_sub(&sbi
->s_dirtyblocks_counter
, release
);
1881 /* update per-inode reservations */
1882 BUG_ON(to_free
> EXT4_I(inode
)->i_reserved_data_blocks
);
1883 EXT4_I(inode
)->i_reserved_data_blocks
-= to_free
;
1885 BUG_ON(mdb
> EXT4_I(inode
)->i_reserved_meta_blocks
);
1886 EXT4_I(inode
)->i_reserved_meta_blocks
= mdb
;
1887 spin_unlock(&EXT4_I(inode
)->i_block_reservation_lock
);
1889 vfs_dq_release_reservation_block(inode
, release
);
1892 static void ext4_da_page_release_reservation(struct page
*page
,
1893 unsigned long offset
)
1896 struct buffer_head
*head
, *bh
;
1897 unsigned int curr_off
= 0;
1899 head
= page_buffers(page
);
1902 unsigned int next_off
= curr_off
+ bh
->b_size
;
1904 if ((offset
<= curr_off
) && (buffer_delay(bh
))) {
1906 clear_buffer_delay(bh
);
1908 curr_off
= next_off
;
1909 } while ((bh
= bh
->b_this_page
) != head
);
1910 ext4_da_release_space(page
->mapping
->host
, to_release
);
1914 * Delayed allocation stuff
1918 * mpage_da_submit_io - walks through extent of pages and try to write
1919 * them with writepage() call back
1921 * @mpd->inode: inode
1922 * @mpd->first_page: first page of the extent
1923 * @mpd->next_page: page after the last page of the extent
1925 * By the time mpage_da_submit_io() is called we expect all blocks
1926 * to be allocated. this may be wrong if allocation failed.
1928 * As pages are already locked by write_cache_pages(), we can't use it
1930 static int mpage_da_submit_io(struct mpage_da_data
*mpd
)
1933 struct pagevec pvec
;
1934 unsigned long index
, end
;
1935 int ret
= 0, err
, nr_pages
, i
;
1936 struct inode
*inode
= mpd
->inode
;
1937 struct address_space
*mapping
= inode
->i_mapping
;
1939 BUG_ON(mpd
->next_page
<= mpd
->first_page
);
1941 * We need to start from the first_page to the next_page - 1
1942 * to make sure we also write the mapped dirty buffer_heads.
1943 * If we look at mpd->b_blocknr we would only be looking
1944 * at the currently mapped buffer_heads.
1946 index
= mpd
->first_page
;
1947 end
= mpd
->next_page
- 1;
1949 pagevec_init(&pvec
, 0);
1950 while (index
<= end
) {
1951 nr_pages
= pagevec_lookup(&pvec
, mapping
, index
, PAGEVEC_SIZE
);
1954 for (i
= 0; i
< nr_pages
; i
++) {
1955 struct page
*page
= pvec
.pages
[i
];
1957 index
= page
->index
;
1962 BUG_ON(!PageLocked(page
));
1963 BUG_ON(PageWriteback(page
));
1965 pages_skipped
= mpd
->wbc
->pages_skipped
;
1966 err
= mapping
->a_ops
->writepage(page
, mpd
->wbc
);
1967 if (!err
&& (pages_skipped
== mpd
->wbc
->pages_skipped
))
1969 * have successfully written the page
1970 * without skipping the same
1972 mpd
->pages_written
++;
1974 * In error case, we have to continue because
1975 * remaining pages are still locked
1976 * XXX: unlock and re-dirty them?
1981 pagevec_release(&pvec
);
1987 * mpage_put_bnr_to_bhs - walk blocks and assign them actual numbers
1989 * @mpd->inode - inode to walk through
1990 * @exbh->b_blocknr - first block on a disk
1991 * @exbh->b_size - amount of space in bytes
1992 * @logical - first logical block to start assignment with
1994 * the function goes through all passed space and put actual disk
1995 * block numbers into buffer heads, dropping BH_Delay and BH_Unwritten
1997 static void mpage_put_bnr_to_bhs(struct mpage_da_data
*mpd
, sector_t logical
,
1998 struct buffer_head
*exbh
)
2000 struct inode
*inode
= mpd
->inode
;
2001 struct address_space
*mapping
= inode
->i_mapping
;
2002 int blocks
= exbh
->b_size
>> inode
->i_blkbits
;
2003 sector_t pblock
= exbh
->b_blocknr
, cur_logical
;
2004 struct buffer_head
*head
, *bh
;
2006 struct pagevec pvec
;
2009 index
= logical
>> (PAGE_CACHE_SHIFT
- inode
->i_blkbits
);
2010 end
= (logical
+ blocks
- 1) >> (PAGE_CACHE_SHIFT
- inode
->i_blkbits
);
2011 cur_logical
= index
<< (PAGE_CACHE_SHIFT
- inode
->i_blkbits
);
2013 pagevec_init(&pvec
, 0);
2015 while (index
<= end
) {
2016 /* XXX: optimize tail */
2017 nr_pages
= pagevec_lookup(&pvec
, mapping
, index
, PAGEVEC_SIZE
);
2020 for (i
= 0; i
< nr_pages
; i
++) {
2021 struct page
*page
= pvec
.pages
[i
];
2023 index
= page
->index
;
2028 BUG_ON(!PageLocked(page
));
2029 BUG_ON(PageWriteback(page
));
2030 BUG_ON(!page_has_buffers(page
));
2032 bh
= page_buffers(page
);
2035 /* skip blocks out of the range */
2037 if (cur_logical
>= logical
)
2040 } while ((bh
= bh
->b_this_page
) != head
);
2043 if (cur_logical
>= logical
+ blocks
)
2046 if (buffer_delay(bh
) ||
2047 buffer_unwritten(bh
)) {
2049 BUG_ON(bh
->b_bdev
!= inode
->i_sb
->s_bdev
);
2051 if (buffer_delay(bh
)) {
2052 clear_buffer_delay(bh
);
2053 bh
->b_blocknr
= pblock
;
2056 * unwritten already should have
2057 * blocknr assigned. Verify that
2059 clear_buffer_unwritten(bh
);
2060 BUG_ON(bh
->b_blocknr
!= pblock
);
2063 } else if (buffer_mapped(bh
))
2064 BUG_ON(bh
->b_blocknr
!= pblock
);
2068 } while ((bh
= bh
->b_this_page
) != head
);
2070 pagevec_release(&pvec
);
2076 * __unmap_underlying_blocks - just a helper function to unmap
2077 * set of blocks described by @bh
2079 static inline void __unmap_underlying_blocks(struct inode
*inode
,
2080 struct buffer_head
*bh
)
2082 struct block_device
*bdev
= inode
->i_sb
->s_bdev
;
2085 blocks
= bh
->b_size
>> inode
->i_blkbits
;
2086 for (i
= 0; i
< blocks
; i
++)
2087 unmap_underlying_metadata(bdev
, bh
->b_blocknr
+ i
);
2090 static void ext4_da_block_invalidatepages(struct mpage_da_data
*mpd
,
2091 sector_t logical
, long blk_cnt
)
2095 struct pagevec pvec
;
2096 struct inode
*inode
= mpd
->inode
;
2097 struct address_space
*mapping
= inode
->i_mapping
;
2099 index
= logical
>> (PAGE_CACHE_SHIFT
- inode
->i_blkbits
);
2100 end
= (logical
+ blk_cnt
- 1) >>
2101 (PAGE_CACHE_SHIFT
- inode
->i_blkbits
);
2102 while (index
<= end
) {
2103 nr_pages
= pagevec_lookup(&pvec
, mapping
, index
, PAGEVEC_SIZE
);
2106 for (i
= 0; i
< nr_pages
; i
++) {
2107 struct page
*page
= pvec
.pages
[i
];
2108 index
= page
->index
;
2113 BUG_ON(!PageLocked(page
));
2114 BUG_ON(PageWriteback(page
));
2115 block_invalidatepage(page
, 0);
2116 ClearPageUptodate(page
);
2123 static void ext4_print_free_blocks(struct inode
*inode
)
2125 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
2126 printk(KERN_CRIT
"Total free blocks count %lld\n",
2127 ext4_count_free_blocks(inode
->i_sb
));
2128 printk(KERN_CRIT
"Free/Dirty block details\n");
2129 printk(KERN_CRIT
"free_blocks=%lld\n",
2130 (long long) percpu_counter_sum(&sbi
->s_freeblocks_counter
));
2131 printk(KERN_CRIT
"dirty_blocks=%lld\n",
2132 (long long) percpu_counter_sum(&sbi
->s_dirtyblocks_counter
));
2133 printk(KERN_CRIT
"Block reservation details\n");
2134 printk(KERN_CRIT
"i_reserved_data_blocks=%u\n",
2135 EXT4_I(inode
)->i_reserved_data_blocks
);
2136 printk(KERN_CRIT
"i_reserved_meta_blocks=%u\n",
2137 EXT4_I(inode
)->i_reserved_meta_blocks
);
2142 * mpage_da_map_blocks - go through given space
2144 * @mpd - bh describing space
2146 * The function skips space we know is already mapped to disk blocks.
2149 static int mpage_da_map_blocks(struct mpage_da_data
*mpd
)
2151 int err
, blks
, get_blocks_flags
;
2152 struct buffer_head
new;
2153 sector_t next
= mpd
->b_blocknr
;
2154 unsigned max_blocks
= mpd
->b_size
>> mpd
->inode
->i_blkbits
;
2155 loff_t disksize
= EXT4_I(mpd
->inode
)->i_disksize
;
2156 handle_t
*handle
= NULL
;
2159 * We consider only non-mapped and non-allocated blocks
2161 if ((mpd
->b_state
& (1 << BH_Mapped
)) &&
2162 !(mpd
->b_state
& (1 << BH_Delay
)) &&
2163 !(mpd
->b_state
& (1 << BH_Unwritten
)))
2167 * If we didn't accumulate anything to write simply return
2172 handle
= ext4_journal_current_handle();
2176 * Call ext4_get_blocks() to allocate any delayed allocation
2177 * blocks, or to convert an uninitialized extent to be
2178 * initialized (in the case where we have written into
2179 * one or more preallocated blocks).
2181 * We pass in the magic EXT4_GET_BLOCKS_DELALLOC_RESERVE to
2182 * indicate that we are on the delayed allocation path. This
2183 * affects functions in many different parts of the allocation
2184 * call path. This flag exists primarily because we don't
2185 * want to change *many* call functions, so ext4_get_blocks()
2186 * will set the magic i_delalloc_reserved_flag once the
2187 * inode's allocation semaphore is taken.
2189 * If the blocks in questions were delalloc blocks, set
2190 * EXT4_GET_BLOCKS_DELALLOC_RESERVE so the delalloc accounting
2191 * variables are updated after the blocks have been allocated.
2194 get_blocks_flags
= (EXT4_GET_BLOCKS_CREATE
|
2195 EXT4_GET_BLOCKS_DELALLOC_RESERVE
);
2196 if (mpd
->b_state
& (1 << BH_Delay
))
2197 get_blocks_flags
|= EXT4_GET_BLOCKS_UPDATE_RESERVE_SPACE
;
2198 blks
= ext4_get_blocks(handle
, mpd
->inode
, next
, max_blocks
,
2199 &new, get_blocks_flags
);
2203 * If get block returns with error we simply
2204 * return. Later writepage will redirty the page and
2205 * writepages will find the dirty page again
2210 if (err
== -ENOSPC
&&
2211 ext4_count_free_blocks(mpd
->inode
->i_sb
)) {
2217 * get block failure will cause us to loop in
2218 * writepages, because a_ops->writepage won't be able
2219 * to make progress. The page will be redirtied by
2220 * writepage and writepages will again try to write
2223 ext4_msg(mpd
->inode
->i_sb
, KERN_CRIT
,
2224 "delayed block allocation failed for inode %lu at "
2225 "logical offset %llu with max blocks %zd with "
2226 "error %d\n", mpd
->inode
->i_ino
,
2227 (unsigned long long) next
,
2228 mpd
->b_size
>> mpd
->inode
->i_blkbits
, err
);
2229 printk(KERN_CRIT
"This should not happen!! "
2230 "Data will be lost\n");
2231 if (err
== -ENOSPC
) {
2232 ext4_print_free_blocks(mpd
->inode
);
2234 /* invalidate all the pages */
2235 ext4_da_block_invalidatepages(mpd
, next
,
2236 mpd
->b_size
>> mpd
->inode
->i_blkbits
);
2241 new.b_size
= (blks
<< mpd
->inode
->i_blkbits
);
2243 if (buffer_new(&new))
2244 __unmap_underlying_blocks(mpd
->inode
, &new);
2247 * If blocks are delayed marked, we need to
2248 * put actual blocknr and drop delayed bit
2250 if ((mpd
->b_state
& (1 << BH_Delay
)) ||
2251 (mpd
->b_state
& (1 << BH_Unwritten
)))
2252 mpage_put_bnr_to_bhs(mpd
, next
, &new);
2254 if (ext4_should_order_data(mpd
->inode
)) {
2255 err
= ext4_jbd2_file_inode(handle
, mpd
->inode
);
2261 * Update on-disk size along with block allocation.
2263 disksize
= ((loff_t
) next
+ blks
) << mpd
->inode
->i_blkbits
;
2264 if (disksize
> i_size_read(mpd
->inode
))
2265 disksize
= i_size_read(mpd
->inode
);
2266 if (disksize
> EXT4_I(mpd
->inode
)->i_disksize
) {
2267 ext4_update_i_disksize(mpd
->inode
, disksize
);
2268 return ext4_mark_inode_dirty(handle
, mpd
->inode
);
2274 #define BH_FLAGS ((1 << BH_Uptodate) | (1 << BH_Mapped) | \
2275 (1 << BH_Delay) | (1 << BH_Unwritten))
2278 * mpage_add_bh_to_extent - try to add one more block to extent of blocks
2280 * @mpd->lbh - extent of blocks
2281 * @logical - logical number of the block in the file
2282 * @bh - bh of the block (used to access block's state)
2284 * the function is used to collect contig. blocks in same state
2286 static void mpage_add_bh_to_extent(struct mpage_da_data
*mpd
,
2287 sector_t logical
, size_t b_size
,
2288 unsigned long b_state
)
2291 int nrblocks
= mpd
->b_size
>> mpd
->inode
->i_blkbits
;
2293 /* check if thereserved journal credits might overflow */
2294 if (!(EXT4_I(mpd
->inode
)->i_flags
& EXT4_EXTENTS_FL
)) {
2295 if (nrblocks
>= EXT4_MAX_TRANS_DATA
) {
2297 * With non-extent format we are limited by the journal
2298 * credit available. Total credit needed to insert
2299 * nrblocks contiguous blocks is dependent on the
2300 * nrblocks. So limit nrblocks.
2303 } else if ((nrblocks
+ (b_size
>> mpd
->inode
->i_blkbits
)) >
2304 EXT4_MAX_TRANS_DATA
) {
2306 * Adding the new buffer_head would make it cross the
2307 * allowed limit for which we have journal credit
2308 * reserved. So limit the new bh->b_size
2310 b_size
= (EXT4_MAX_TRANS_DATA
- nrblocks
) <<
2311 mpd
->inode
->i_blkbits
;
2312 /* we will do mpage_da_submit_io in the next loop */
2316 * First block in the extent
2318 if (mpd
->b_size
== 0) {
2319 mpd
->b_blocknr
= logical
;
2320 mpd
->b_size
= b_size
;
2321 mpd
->b_state
= b_state
& BH_FLAGS
;
2325 next
= mpd
->b_blocknr
+ nrblocks
;
2327 * Can we merge the block to our big extent?
2329 if (logical
== next
&& (b_state
& BH_FLAGS
) == mpd
->b_state
) {
2330 mpd
->b_size
+= b_size
;
2336 * We couldn't merge the block to our extent, so we
2337 * need to flush current extent and start new one
2339 if (mpage_da_map_blocks(mpd
) == 0)
2340 mpage_da_submit_io(mpd
);
2345 static int ext4_bh_delay_or_unwritten(handle_t
*handle
, struct buffer_head
*bh
)
2347 return (buffer_delay(bh
) || buffer_unwritten(bh
)) && buffer_dirty(bh
);
2351 * __mpage_da_writepage - finds extent of pages and blocks
2353 * @page: page to consider
2354 * @wbc: not used, we just follow rules
2357 * The function finds extents of pages and scan them for all blocks.
2359 static int __mpage_da_writepage(struct page
*page
,
2360 struct writeback_control
*wbc
, void *data
)
2362 struct mpage_da_data
*mpd
= data
;
2363 struct inode
*inode
= mpd
->inode
;
2364 struct buffer_head
*bh
, *head
;
2369 * Rest of the page in the page_vec
2370 * redirty then and skip then. We will
2371 * try to write them again after
2372 * starting a new transaction
2374 redirty_page_for_writepage(wbc
, page
);
2376 return MPAGE_DA_EXTENT_TAIL
;
2379 * Can we merge this page to current extent?
2381 if (mpd
->next_page
!= page
->index
) {
2383 * Nope, we can't. So, we map non-allocated blocks
2384 * and start IO on them using writepage()
2386 if (mpd
->next_page
!= mpd
->first_page
) {
2387 if (mpage_da_map_blocks(mpd
) == 0)
2388 mpage_da_submit_io(mpd
);
2390 * skip rest of the page in the page_vec
2393 redirty_page_for_writepage(wbc
, page
);
2395 return MPAGE_DA_EXTENT_TAIL
;
2399 * Start next extent of pages ...
2401 mpd
->first_page
= page
->index
;
2411 mpd
->next_page
= page
->index
+ 1;
2412 logical
= (sector_t
) page
->index
<<
2413 (PAGE_CACHE_SHIFT
- inode
->i_blkbits
);
2415 if (!page_has_buffers(page
)) {
2416 mpage_add_bh_to_extent(mpd
, logical
, PAGE_CACHE_SIZE
,
2417 (1 << BH_Dirty
) | (1 << BH_Uptodate
));
2419 return MPAGE_DA_EXTENT_TAIL
;
2422 * Page with regular buffer heads, just add all dirty ones
2424 head
= page_buffers(page
);
2427 BUG_ON(buffer_locked(bh
));
2429 * We need to try to allocate
2430 * unmapped blocks in the same page.
2431 * Otherwise we won't make progress
2432 * with the page in ext4_writepage
2434 if (ext4_bh_delay_or_unwritten(NULL
, bh
)) {
2435 mpage_add_bh_to_extent(mpd
, logical
,
2439 return MPAGE_DA_EXTENT_TAIL
;
2440 } else if (buffer_dirty(bh
) && (buffer_mapped(bh
))) {
2442 * mapped dirty buffer. We need to update
2443 * the b_state because we look at
2444 * b_state in mpage_da_map_blocks. We don't
2445 * update b_size because if we find an
2446 * unmapped buffer_head later we need to
2447 * use the b_state flag of that buffer_head.
2449 if (mpd
->b_size
== 0)
2450 mpd
->b_state
= bh
->b_state
& BH_FLAGS
;
2453 } while ((bh
= bh
->b_this_page
) != head
);
2460 * This is a special get_blocks_t callback which is used by
2461 * ext4_da_write_begin(). It will either return mapped block or
2462 * reserve space for a single block.
2464 * For delayed buffer_head we have BH_Mapped, BH_New, BH_Delay set.
2465 * We also have b_blocknr = -1 and b_bdev initialized properly
2467 * For unwritten buffer_head we have BH_Mapped, BH_New, BH_Unwritten set.
2468 * We also have b_blocknr = physicalblock mapping unwritten extent and b_bdev
2469 * initialized properly.
2471 static int ext4_da_get_block_prep(struct inode
*inode
, sector_t iblock
,
2472 struct buffer_head
*bh_result
, int create
)
2475 sector_t invalid_block
= ~((sector_t
) 0xffff);
2477 if (invalid_block
< ext4_blocks_count(EXT4_SB(inode
->i_sb
)->s_es
))
2480 BUG_ON(create
== 0);
2481 BUG_ON(bh_result
->b_size
!= inode
->i_sb
->s_blocksize
);
2484 * first, we need to know whether the block is allocated already
2485 * preallocated blocks are unmapped but should treated
2486 * the same as allocated blocks.
2488 ret
= ext4_get_blocks(NULL
, inode
, iblock
, 1, bh_result
, 0);
2489 if ((ret
== 0) && !buffer_delay(bh_result
)) {
2490 /* the block isn't (pre)allocated yet, let's reserve space */
2492 * XXX: __block_prepare_write() unmaps passed block,
2495 ret
= ext4_da_reserve_space(inode
, 1);
2497 /* not enough space to reserve */
2500 map_bh(bh_result
, inode
->i_sb
, invalid_block
);
2501 set_buffer_new(bh_result
);
2502 set_buffer_delay(bh_result
);
2503 } else if (ret
> 0) {
2504 bh_result
->b_size
= (ret
<< inode
->i_blkbits
);
2505 if (buffer_unwritten(bh_result
)) {
2506 /* A delayed write to unwritten bh should
2507 * be marked new and mapped. Mapped ensures
2508 * that we don't do get_block multiple times
2509 * when we write to the same offset and new
2510 * ensures that we do proper zero out for
2513 set_buffer_new(bh_result
);
2514 set_buffer_mapped(bh_result
);
2523 * This function is used as a standard get_block_t calback function
2524 * when there is no desire to allocate any blocks. It is used as a
2525 * callback function for block_prepare_write(), nobh_writepage(), and
2526 * block_write_full_page(). These functions should only try to map a
2527 * single block at a time.
2529 * Since this function doesn't do block allocations even if the caller
2530 * requests it by passing in create=1, it is critically important that
2531 * any caller checks to make sure that any buffer heads are returned
2532 * by this function are either all already mapped or marked for
2533 * delayed allocation before calling nobh_writepage() or
2534 * block_write_full_page(). Otherwise, b_blocknr could be left
2535 * unitialized, and the page write functions will be taken by
2538 static int noalloc_get_block_write(struct inode
*inode
, sector_t iblock
,
2539 struct buffer_head
*bh_result
, int create
)
2542 unsigned max_blocks
= bh_result
->b_size
>> inode
->i_blkbits
;
2544 BUG_ON(bh_result
->b_size
!= inode
->i_sb
->s_blocksize
);
2547 * we don't want to do block allocation in writepage
2548 * so call get_block_wrap with create = 0
2550 ret
= ext4_get_blocks(NULL
, inode
, iblock
, max_blocks
, bh_result
, 0);
2552 bh_result
->b_size
= (ret
<< inode
->i_blkbits
);
2558 static int bget_one(handle_t
*handle
, struct buffer_head
*bh
)
2564 static int bput_one(handle_t
*handle
, struct buffer_head
*bh
)
2570 static int __ext4_journalled_writepage(struct page
*page
,
2573 struct address_space
*mapping
= page
->mapping
;
2574 struct inode
*inode
= mapping
->host
;
2575 struct buffer_head
*page_bufs
;
2576 handle_t
*handle
= NULL
;
2580 page_bufs
= page_buffers(page
);
2582 walk_page_buffers(handle
, page_bufs
, 0, len
, NULL
, bget_one
);
2583 /* As soon as we unlock the page, it can go away, but we have
2584 * references to buffers so we are safe */
2587 handle
= ext4_journal_start(inode
, ext4_writepage_trans_blocks(inode
));
2588 if (IS_ERR(handle
)) {
2589 ret
= PTR_ERR(handle
);
2593 ret
= walk_page_buffers(handle
, page_bufs
, 0, len
, NULL
,
2594 do_journal_get_write_access
);
2596 err
= walk_page_buffers(handle
, page_bufs
, 0, len
, NULL
,
2600 err
= ext4_journal_stop(handle
);
2604 walk_page_buffers(handle
, page_bufs
, 0, len
, NULL
, bput_one
);
2605 EXT4_I(inode
)->i_state
|= EXT4_STATE_JDATA
;
2611 * Note that we don't need to start a transaction unless we're journaling data
2612 * because we should have holes filled from ext4_page_mkwrite(). We even don't
2613 * need to file the inode to the transaction's list in ordered mode because if
2614 * we are writing back data added by write(), the inode is already there and if
2615 * we are writing back data modified via mmap(), noone guarantees in which
2616 * transaction the data will hit the disk. In case we are journaling data, we
2617 * cannot start transaction directly because transaction start ranks above page
2618 * lock so we have to do some magic.
2620 * This function can get called via...
2621 * - ext4_da_writepages after taking page lock (have journal handle)
2622 * - journal_submit_inode_data_buffers (no journal handle)
2623 * - shrink_page_list via pdflush (no journal handle)
2624 * - grab_page_cache when doing write_begin (have journal handle)
2626 * We don't do any block allocation in this function. If we have page with
2627 * multiple blocks we need to write those buffer_heads that are mapped. This
2628 * is important for mmaped based write. So if we do with blocksize 1K
2629 * truncate(f, 1024);
2630 * a = mmap(f, 0, 4096);
2632 * truncate(f, 4096);
2633 * we have in the page first buffer_head mapped via page_mkwrite call back
2634 * but other bufer_heads would be unmapped but dirty(dirty done via the
2635 * do_wp_page). So writepage should write the first block. If we modify
2636 * the mmap area beyond 1024 we will again get a page_fault and the
2637 * page_mkwrite callback will do the block allocation and mark the
2638 * buffer_heads mapped.
2640 * We redirty the page if we have any buffer_heads that is either delay or
2641 * unwritten in the page.
2643 * We can get recursively called as show below.
2645 * ext4_writepage() -> kmalloc() -> __alloc_pages() -> page_launder() ->
2648 * But since we don't do any block allocation we should not deadlock.
2649 * Page also have the dirty flag cleared so we don't get recurive page_lock.
2651 static int ext4_writepage(struct page
*page
,
2652 struct writeback_control
*wbc
)
2657 struct buffer_head
*page_bufs
;
2658 struct inode
*inode
= page
->mapping
->host
;
2660 trace_ext4_writepage(inode
, page
);
2661 size
= i_size_read(inode
);
2662 if (page
->index
== size
>> PAGE_CACHE_SHIFT
)
2663 len
= size
& ~PAGE_CACHE_MASK
;
2665 len
= PAGE_CACHE_SIZE
;
2667 if (page_has_buffers(page
)) {
2668 page_bufs
= page_buffers(page
);
2669 if (walk_page_buffers(NULL
, page_bufs
, 0, len
, NULL
,
2670 ext4_bh_delay_or_unwritten
)) {
2672 * We don't want to do block allocation
2673 * So redirty the page and return
2674 * We may reach here when we do a journal commit
2675 * via journal_submit_inode_data_buffers.
2676 * If we don't have mapping block we just ignore
2677 * them. We can also reach here via shrink_page_list
2679 redirty_page_for_writepage(wbc
, page
);
2685 * The test for page_has_buffers() is subtle:
2686 * We know the page is dirty but it lost buffers. That means
2687 * that at some moment in time after write_begin()/write_end()
2688 * has been called all buffers have been clean and thus they
2689 * must have been written at least once. So they are all
2690 * mapped and we can happily proceed with mapping them
2691 * and writing the page.
2693 * Try to initialize the buffer_heads and check whether
2694 * all are mapped and non delay. We don't want to
2695 * do block allocation here.
2697 ret
= block_prepare_write(page
, 0, len
,
2698 noalloc_get_block_write
);
2700 page_bufs
= page_buffers(page
);
2701 /* check whether all are mapped and non delay */
2702 if (walk_page_buffers(NULL
, page_bufs
, 0, len
, NULL
,
2703 ext4_bh_delay_or_unwritten
)) {
2704 redirty_page_for_writepage(wbc
, page
);
2710 * We can't do block allocation here
2711 * so just redity the page and unlock
2714 redirty_page_for_writepage(wbc
, page
);
2718 /* now mark the buffer_heads as dirty and uptodate */
2719 block_commit_write(page
, 0, len
);
2722 if (PageChecked(page
) && ext4_should_journal_data(inode
)) {
2724 * It's mmapped pagecache. Add buffers and journal it. There
2725 * doesn't seem much point in redirtying the page here.
2727 ClearPageChecked(page
);
2728 return __ext4_journalled_writepage(page
, len
);
2731 if (test_opt(inode
->i_sb
, NOBH
) && ext4_should_writeback_data(inode
))
2732 ret
= nobh_writepage(page
, noalloc_get_block_write
, wbc
);
2734 ret
= block_write_full_page(page
, noalloc_get_block_write
,
2741 * This is called via ext4_da_writepages() to
2742 * calulate the total number of credits to reserve to fit
2743 * a single extent allocation into a single transaction,
2744 * ext4_da_writpeages() will loop calling this before
2745 * the block allocation.
2748 static int ext4_da_writepages_trans_blocks(struct inode
*inode
)
2750 int max_blocks
= EXT4_I(inode
)->i_reserved_data_blocks
;
2753 * With non-extent format the journal credit needed to
2754 * insert nrblocks contiguous block is dependent on
2755 * number of contiguous block. So we will limit
2756 * number of contiguous block to a sane value
2758 if (!(EXT4_I(inode
)->i_flags
& EXT4_EXTENTS_FL
) &&
2759 (max_blocks
> EXT4_MAX_TRANS_DATA
))
2760 max_blocks
= EXT4_MAX_TRANS_DATA
;
2762 return ext4_chunk_trans_blocks(inode
, max_blocks
);
2765 static int ext4_da_writepages(struct address_space
*mapping
,
2766 struct writeback_control
*wbc
)
2769 int range_whole
= 0;
2770 handle_t
*handle
= NULL
;
2771 struct mpage_da_data mpd
;
2772 struct inode
*inode
= mapping
->host
;
2773 int no_nrwrite_index_update
;
2774 int pages_written
= 0;
2776 unsigned int max_pages
;
2777 int range_cyclic
, cycled
= 1, io_done
= 0;
2778 int needed_blocks
, ret
= 0;
2779 long desired_nr_to_write
, nr_to_writebump
= 0;
2780 loff_t range_start
= wbc
->range_start
;
2781 struct ext4_sb_info
*sbi
= EXT4_SB(mapping
->host
->i_sb
);
2783 trace_ext4_da_writepages(inode
, wbc
);
2786 * No pages to write? This is mainly a kludge to avoid starting
2787 * a transaction for special inodes like journal inode on last iput()
2788 * because that could violate lock ordering on umount
2790 if (!mapping
->nrpages
|| !mapping_tagged(mapping
, PAGECACHE_TAG_DIRTY
))
2794 * If the filesystem has aborted, it is read-only, so return
2795 * right away instead of dumping stack traces later on that
2796 * will obscure the real source of the problem. We test
2797 * EXT4_MF_FS_ABORTED instead of sb->s_flag's MS_RDONLY because
2798 * the latter could be true if the filesystem is mounted
2799 * read-only, and in that case, ext4_da_writepages should
2800 * *never* be called, so if that ever happens, we would want
2803 if (unlikely(sbi
->s_mount_flags
& EXT4_MF_FS_ABORTED
))
2806 if (wbc
->range_start
== 0 && wbc
->range_end
== LLONG_MAX
)
2809 range_cyclic
= wbc
->range_cyclic
;
2810 if (wbc
->range_cyclic
) {
2811 index
= mapping
->writeback_index
;
2814 wbc
->range_start
= index
<< PAGE_CACHE_SHIFT
;
2815 wbc
->range_end
= LLONG_MAX
;
2816 wbc
->range_cyclic
= 0;
2818 index
= wbc
->range_start
>> PAGE_CACHE_SHIFT
;
2821 * This works around two forms of stupidity. The first is in
2822 * the writeback code, which caps the maximum number of pages
2823 * written to be 1024 pages. This is wrong on multiple
2824 * levels; different architectues have a different page size,
2825 * which changes the maximum amount of data which gets
2826 * written. Secondly, 4 megabytes is way too small. XFS
2827 * forces this value to be 16 megabytes by multiplying
2828 * nr_to_write parameter by four, and then relies on its
2829 * allocator to allocate larger extents to make them
2830 * contiguous. Unfortunately this brings us to the second
2831 * stupidity, which is that ext4's mballoc code only allocates
2832 * at most 2048 blocks. So we force contiguous writes up to
2833 * the number of dirty blocks in the inode, or
2834 * sbi->max_writeback_mb_bump whichever is smaller.
2836 max_pages
= sbi
->s_max_writeback_mb_bump
<< (20 - PAGE_CACHE_SHIFT
);
2837 if (!range_cyclic
&& range_whole
)
2838 desired_nr_to_write
= wbc
->nr_to_write
* 8;
2840 desired_nr_to_write
= ext4_num_dirty_pages(inode
, index
,
2842 if (desired_nr_to_write
> max_pages
)
2843 desired_nr_to_write
= max_pages
;
2845 if (wbc
->nr_to_write
< desired_nr_to_write
) {
2846 nr_to_writebump
= desired_nr_to_write
- wbc
->nr_to_write
;
2847 wbc
->nr_to_write
= desired_nr_to_write
;
2851 mpd
.inode
= mapping
->host
;
2854 * we don't want write_cache_pages to update
2855 * nr_to_write and writeback_index
2857 no_nrwrite_index_update
= wbc
->no_nrwrite_index_update
;
2858 wbc
->no_nrwrite_index_update
= 1;
2859 pages_skipped
= wbc
->pages_skipped
;
2862 while (!ret
&& wbc
->nr_to_write
> 0) {
2865 * we insert one extent at a time. So we need
2866 * credit needed for single extent allocation.
2867 * journalled mode is currently not supported
2870 BUG_ON(ext4_should_journal_data(inode
));
2871 needed_blocks
= ext4_da_writepages_trans_blocks(inode
);
2873 /* start a new transaction*/
2874 handle
= ext4_journal_start(inode
, needed_blocks
);
2875 if (IS_ERR(handle
)) {
2876 ret
= PTR_ERR(handle
);
2877 ext4_msg(inode
->i_sb
, KERN_CRIT
, "%s: jbd2_start: "
2878 "%ld pages, ino %lu; err %d\n", __func__
,
2879 wbc
->nr_to_write
, inode
->i_ino
, ret
);
2880 goto out_writepages
;
2884 * Now call __mpage_da_writepage to find the next
2885 * contiguous region of logical blocks that need
2886 * blocks to be allocated by ext4. We don't actually
2887 * submit the blocks for I/O here, even though
2888 * write_cache_pages thinks it will, and will set the
2889 * pages as clean for write before calling
2890 * __mpage_da_writepage().
2898 mpd
.pages_written
= 0;
2900 ret
= write_cache_pages(mapping
, wbc
, __mpage_da_writepage
,
2903 * If we have a contigous extent of pages and we
2904 * haven't done the I/O yet, map the blocks and submit
2907 if (!mpd
.io_done
&& mpd
.next_page
!= mpd
.first_page
) {
2908 if (mpage_da_map_blocks(&mpd
) == 0)
2909 mpage_da_submit_io(&mpd
);
2911 ret
= MPAGE_DA_EXTENT_TAIL
;
2913 trace_ext4_da_write_pages(inode
, &mpd
);
2914 wbc
->nr_to_write
-= mpd
.pages_written
;
2916 ext4_journal_stop(handle
);
2918 if ((mpd
.retval
== -ENOSPC
) && sbi
->s_journal
) {
2919 /* commit the transaction which would
2920 * free blocks released in the transaction
2923 jbd2_journal_force_commit_nested(sbi
->s_journal
);
2924 wbc
->pages_skipped
= pages_skipped
;
2926 } else if (ret
== MPAGE_DA_EXTENT_TAIL
) {
2928 * got one extent now try with
2931 pages_written
+= mpd
.pages_written
;
2932 wbc
->pages_skipped
= pages_skipped
;
2935 } else if (wbc
->nr_to_write
)
2937 * There is no more writeout needed
2938 * or we requested for a noblocking writeout
2939 * and we found the device congested
2943 if (!io_done
&& !cycled
) {
2946 wbc
->range_start
= index
<< PAGE_CACHE_SHIFT
;
2947 wbc
->range_end
= mapping
->writeback_index
- 1;
2950 if (pages_skipped
!= wbc
->pages_skipped
)
2951 ext4_msg(inode
->i_sb
, KERN_CRIT
,
2952 "This should not happen leaving %s "
2953 "with nr_to_write = %ld ret = %d\n",
2954 __func__
, wbc
->nr_to_write
, ret
);
2957 index
+= pages_written
;
2958 wbc
->range_cyclic
= range_cyclic
;
2959 if (wbc
->range_cyclic
|| (range_whole
&& wbc
->nr_to_write
> 0))
2961 * set the writeback_index so that range_cyclic
2962 * mode will write it back later
2964 mapping
->writeback_index
= index
;
2967 if (!no_nrwrite_index_update
)
2968 wbc
->no_nrwrite_index_update
= 0;
2969 if (wbc
->nr_to_write
> nr_to_writebump
)
2970 wbc
->nr_to_write
-= nr_to_writebump
;
2971 wbc
->range_start
= range_start
;
2972 trace_ext4_da_writepages_result(inode
, wbc
, ret
, pages_written
);
2976 #define FALL_BACK_TO_NONDELALLOC 1
2977 static int ext4_nonda_switch(struct super_block
*sb
)
2979 s64 free_blocks
, dirty_blocks
;
2980 struct ext4_sb_info
*sbi
= EXT4_SB(sb
);
2983 * switch to non delalloc mode if we are running low
2984 * on free block. The free block accounting via percpu
2985 * counters can get slightly wrong with percpu_counter_batch getting
2986 * accumulated on each CPU without updating global counters
2987 * Delalloc need an accurate free block accounting. So switch
2988 * to non delalloc when we are near to error range.
2990 free_blocks
= percpu_counter_read_positive(&sbi
->s_freeblocks_counter
);
2991 dirty_blocks
= percpu_counter_read_positive(&sbi
->s_dirtyblocks_counter
);
2992 if (2 * free_blocks
< 3 * dirty_blocks
||
2993 free_blocks
< (dirty_blocks
+ EXT4_FREEBLOCKS_WATERMARK
)) {
2995 * free block count is less that 150% of dirty blocks
2996 * or free blocks is less that watermark
3003 static int ext4_da_write_begin(struct file
*file
, struct address_space
*mapping
,
3004 loff_t pos
, unsigned len
, unsigned flags
,
3005 struct page
**pagep
, void **fsdata
)
3007 int ret
, retries
= 0;
3011 struct inode
*inode
= mapping
->host
;
3014 index
= pos
>> PAGE_CACHE_SHIFT
;
3015 from
= pos
& (PAGE_CACHE_SIZE
- 1);
3018 if (ext4_nonda_switch(inode
->i_sb
)) {
3019 *fsdata
= (void *)FALL_BACK_TO_NONDELALLOC
;
3020 return ext4_write_begin(file
, mapping
, pos
,
3021 len
, flags
, pagep
, fsdata
);
3023 *fsdata
= (void *)0;
3024 trace_ext4_da_write_begin(inode
, pos
, len
, flags
);
3027 * With delayed allocation, we don't log the i_disksize update
3028 * if there is delayed block allocation. But we still need
3029 * to journalling the i_disksize update if writes to the end
3030 * of file which has an already mapped buffer.
3032 handle
= ext4_journal_start(inode
, 1);
3033 if (IS_ERR(handle
)) {
3034 ret
= PTR_ERR(handle
);
3037 /* We cannot recurse into the filesystem as the transaction is already
3039 flags
|= AOP_FLAG_NOFS
;
3041 page
= grab_cache_page_write_begin(mapping
, index
, flags
);
3043 ext4_journal_stop(handle
);
3049 ret
= block_write_begin(file
, mapping
, pos
, len
, flags
, pagep
, fsdata
,
3050 ext4_da_get_block_prep
);
3053 ext4_journal_stop(handle
);
3054 page_cache_release(page
);
3056 * block_write_begin may have instantiated a few blocks
3057 * outside i_size. Trim these off again. Don't need
3058 * i_size_read because we hold i_mutex.
3060 if (pos
+ len
> inode
->i_size
)
3061 ext4_truncate_failed_write(inode
);
3064 if (ret
== -ENOSPC
&& ext4_should_retry_alloc(inode
->i_sb
, &retries
))
3071 * Check if we should update i_disksize
3072 * when write to the end of file but not require block allocation
3074 static int ext4_da_should_update_i_disksize(struct page
*page
,
3075 unsigned long offset
)
3077 struct buffer_head
*bh
;
3078 struct inode
*inode
= page
->mapping
->host
;
3082 bh
= page_buffers(page
);
3083 idx
= offset
>> inode
->i_blkbits
;
3085 for (i
= 0; i
< idx
; i
++)
3086 bh
= bh
->b_this_page
;
3088 if (!buffer_mapped(bh
) || (buffer_delay(bh
)) || buffer_unwritten(bh
))
3093 static int ext4_da_write_end(struct file
*file
,
3094 struct address_space
*mapping
,
3095 loff_t pos
, unsigned len
, unsigned copied
,
3096 struct page
*page
, void *fsdata
)
3098 struct inode
*inode
= mapping
->host
;
3100 handle_t
*handle
= ext4_journal_current_handle();
3102 unsigned long start
, end
;
3103 int write_mode
= (int)(unsigned long)fsdata
;
3105 if (write_mode
== FALL_BACK_TO_NONDELALLOC
) {
3106 if (ext4_should_order_data(inode
)) {
3107 return ext4_ordered_write_end(file
, mapping
, pos
,
3108 len
, copied
, page
, fsdata
);
3109 } else if (ext4_should_writeback_data(inode
)) {
3110 return ext4_writeback_write_end(file
, mapping
, pos
,
3111 len
, copied
, page
, fsdata
);
3117 trace_ext4_da_write_end(inode
, pos
, len
, copied
);
3118 start
= pos
& (PAGE_CACHE_SIZE
- 1);
3119 end
= start
+ copied
- 1;
3122 * generic_write_end() will run mark_inode_dirty() if i_size
3123 * changes. So let's piggyback the i_disksize mark_inode_dirty
3127 new_i_size
= pos
+ copied
;
3128 if (new_i_size
> EXT4_I(inode
)->i_disksize
) {
3129 if (ext4_da_should_update_i_disksize(page
, end
)) {
3130 down_write(&EXT4_I(inode
)->i_data_sem
);
3131 if (new_i_size
> EXT4_I(inode
)->i_disksize
) {
3133 * Updating i_disksize when extending file
3134 * without needing block allocation
3136 if (ext4_should_order_data(inode
))
3137 ret
= ext4_jbd2_file_inode(handle
,
3140 EXT4_I(inode
)->i_disksize
= new_i_size
;
3142 up_write(&EXT4_I(inode
)->i_data_sem
);
3143 /* We need to mark inode dirty even if
3144 * new_i_size is less that inode->i_size
3145 * bu greater than i_disksize.(hint delalloc)
3147 ext4_mark_inode_dirty(handle
, inode
);
3150 ret2
= generic_write_end(file
, mapping
, pos
, len
, copied
,
3155 ret2
= ext4_journal_stop(handle
);
3159 return ret
? ret
: copied
;
3162 static void ext4_da_invalidatepage(struct page
*page
, unsigned long offset
)
3165 * Drop reserved blocks
3167 BUG_ON(!PageLocked(page
));
3168 if (!page_has_buffers(page
))
3171 ext4_da_page_release_reservation(page
, offset
);
3174 ext4_invalidatepage(page
, offset
);
3180 * Force all delayed allocation blocks to be allocated for a given inode.
3182 int ext4_alloc_da_blocks(struct inode
*inode
)
3184 trace_ext4_alloc_da_blocks(inode
);
3186 if (!EXT4_I(inode
)->i_reserved_data_blocks
&&
3187 !EXT4_I(inode
)->i_reserved_meta_blocks
)
3191 * We do something simple for now. The filemap_flush() will
3192 * also start triggering a write of the data blocks, which is
3193 * not strictly speaking necessary (and for users of
3194 * laptop_mode, not even desirable). However, to do otherwise
3195 * would require replicating code paths in:
3197 * ext4_da_writepages() ->
3198 * write_cache_pages() ---> (via passed in callback function)
3199 * __mpage_da_writepage() -->
3200 * mpage_add_bh_to_extent()
3201 * mpage_da_map_blocks()
3203 * The problem is that write_cache_pages(), located in
3204 * mm/page-writeback.c, marks pages clean in preparation for
3205 * doing I/O, which is not desirable if we're not planning on
3208 * We could call write_cache_pages(), and then redirty all of
3209 * the pages by calling redirty_page_for_writeback() but that
3210 * would be ugly in the extreme. So instead we would need to
3211 * replicate parts of the code in the above functions,
3212 * simplifying them becuase we wouldn't actually intend to
3213 * write out the pages, but rather only collect contiguous
3214 * logical block extents, call the multi-block allocator, and
3215 * then update the buffer heads with the block allocations.
3217 * For now, though, we'll cheat by calling filemap_flush(),
3218 * which will map the blocks, and start the I/O, but not
3219 * actually wait for the I/O to complete.
3221 return filemap_flush(inode
->i_mapping
);
3225 * bmap() is special. It gets used by applications such as lilo and by
3226 * the swapper to find the on-disk block of a specific piece of data.
3228 * Naturally, this is dangerous if the block concerned is still in the
3229 * journal. If somebody makes a swapfile on an ext4 data-journaling
3230 * filesystem and enables swap, then they may get a nasty shock when the
3231 * data getting swapped to that swapfile suddenly gets overwritten by
3232 * the original zero's written out previously to the journal and
3233 * awaiting writeback in the kernel's buffer cache.
3235 * So, if we see any bmap calls here on a modified, data-journaled file,
3236 * take extra steps to flush any blocks which might be in the cache.
3238 static sector_t
ext4_bmap(struct address_space
*mapping
, sector_t block
)
3240 struct inode
*inode
= mapping
->host
;
3244 if (mapping_tagged(mapping
, PAGECACHE_TAG_DIRTY
) &&
3245 test_opt(inode
->i_sb
, DELALLOC
)) {
3247 * With delalloc we want to sync the file
3248 * so that we can make sure we allocate
3251 filemap_write_and_wait(mapping
);
3254 if (EXT4_JOURNAL(inode
) && EXT4_I(inode
)->i_state
& EXT4_STATE_JDATA
) {
3256 * This is a REALLY heavyweight approach, but the use of
3257 * bmap on dirty files is expected to be extremely rare:
3258 * only if we run lilo or swapon on a freshly made file
3259 * do we expect this to happen.
3261 * (bmap requires CAP_SYS_RAWIO so this does not
3262 * represent an unprivileged user DOS attack --- we'd be
3263 * in trouble if mortal users could trigger this path at
3266 * NB. EXT4_STATE_JDATA is not set on files other than
3267 * regular files. If somebody wants to bmap a directory
3268 * or symlink and gets confused because the buffer
3269 * hasn't yet been flushed to disk, they deserve
3270 * everything they get.
3273 EXT4_I(inode
)->i_state
&= ~EXT4_STATE_JDATA
;
3274 journal
= EXT4_JOURNAL(inode
);
3275 jbd2_journal_lock_updates(journal
);
3276 err
= jbd2_journal_flush(journal
);
3277 jbd2_journal_unlock_updates(journal
);
3283 return generic_block_bmap(mapping
, block
, ext4_get_block
);
3286 static int ext4_readpage(struct file
*file
, struct page
*page
)
3288 return mpage_readpage(page
, ext4_get_block
);
3292 ext4_readpages(struct file
*file
, struct address_space
*mapping
,
3293 struct list_head
*pages
, unsigned nr_pages
)
3295 return mpage_readpages(mapping
, pages
, nr_pages
, ext4_get_block
);
3298 static void ext4_invalidatepage(struct page
*page
, unsigned long offset
)
3300 journal_t
*journal
= EXT4_JOURNAL(page
->mapping
->host
);
3303 * If it's a full truncate we just forget about the pending dirtying
3306 ClearPageChecked(page
);
3309 jbd2_journal_invalidatepage(journal
, page
, offset
);
3311 block_invalidatepage(page
, offset
);
3314 static int ext4_releasepage(struct page
*page
, gfp_t wait
)
3316 journal_t
*journal
= EXT4_JOURNAL(page
->mapping
->host
);
3318 WARN_ON(PageChecked(page
));
3319 if (!page_has_buffers(page
))
3322 return jbd2_journal_try_to_free_buffers(journal
, page
, wait
);
3324 return try_to_free_buffers(page
);
3328 * O_DIRECT for ext3 (or indirect map) based files
3330 * If the O_DIRECT write will extend the file then add this inode to the
3331 * orphan list. So recovery will truncate it back to the original size
3332 * if the machine crashes during the write.
3334 * If the O_DIRECT write is intantiating holes inside i_size and the machine
3335 * crashes then stale disk data _may_ be exposed inside the file. But current
3336 * VFS code falls back into buffered path in that case so we are safe.
3338 static ssize_t
ext4_ind_direct_IO(int rw
, struct kiocb
*iocb
,
3339 const struct iovec
*iov
, loff_t offset
,
3340 unsigned long nr_segs
)
3342 struct file
*file
= iocb
->ki_filp
;
3343 struct inode
*inode
= file
->f_mapping
->host
;
3344 struct ext4_inode_info
*ei
= EXT4_I(inode
);
3348 size_t count
= iov_length(iov
, nr_segs
);
3352 loff_t final_size
= offset
+ count
;
3354 if (final_size
> inode
->i_size
) {
3355 /* Credits for sb + inode write */
3356 handle
= ext4_journal_start(inode
, 2);
3357 if (IS_ERR(handle
)) {
3358 ret
= PTR_ERR(handle
);
3361 ret
= ext4_orphan_add(handle
, inode
);
3363 ext4_journal_stop(handle
);
3367 ei
->i_disksize
= inode
->i_size
;
3368 ext4_journal_stop(handle
);
3373 ret
= blockdev_direct_IO(rw
, iocb
, inode
, inode
->i_sb
->s_bdev
, iov
,
3375 ext4_get_block
, NULL
);
3376 if (ret
== -ENOSPC
&& ext4_should_retry_alloc(inode
->i_sb
, &retries
))
3382 /* Credits for sb + inode write */
3383 handle
= ext4_journal_start(inode
, 2);
3384 if (IS_ERR(handle
)) {
3385 /* This is really bad luck. We've written the data
3386 * but cannot extend i_size. Bail out and pretend
3387 * the write failed... */
3388 ret
= PTR_ERR(handle
);
3392 ext4_orphan_del(handle
, inode
);
3394 loff_t end
= offset
+ ret
;
3395 if (end
> inode
->i_size
) {
3396 ei
->i_disksize
= end
;
3397 i_size_write(inode
, end
);
3399 * We're going to return a positive `ret'
3400 * here due to non-zero-length I/O, so there's
3401 * no way of reporting error returns from
3402 * ext4_mark_inode_dirty() to userspace. So
3405 ext4_mark_inode_dirty(handle
, inode
);
3408 err
= ext4_journal_stop(handle
);
3416 static int ext4_get_block_dio_write(struct inode
*inode
, sector_t iblock
,
3417 struct buffer_head
*bh_result
, int create
)
3419 handle_t
*handle
= NULL
;
3421 unsigned max_blocks
= bh_result
->b_size
>> inode
->i_blkbits
;
3424 ext4_debug("ext4_get_block_dio_write: inode %lu, create flag %d\n",
3425 inode
->i_ino
, create
);
3427 * DIO VFS code passes create = 0 flag for write to
3428 * the middle of file. It does this to avoid block
3429 * allocation for holes, to prevent expose stale data
3430 * out when there is parallel buffered read (which does
3431 * not hold the i_mutex lock) while direct IO write has
3432 * not completed. DIO request on holes finally falls back
3433 * to buffered IO for this reason.
3435 * For ext4 extent based file, since we support fallocate,
3436 * new allocated extent as uninitialized, for holes, we
3437 * could fallocate blocks for holes, thus parallel
3438 * buffered IO read will zero out the page when read on
3439 * a hole while parallel DIO write to the hole has not completed.
3441 * when we come here, we know it's a direct IO write to
3442 * to the middle of file (<i_size)
3443 * so it's safe to override the create flag from VFS.
3445 create
= EXT4_GET_BLOCKS_DIO_CREATE_EXT
;
3447 if (max_blocks
> DIO_MAX_BLOCKS
)
3448 max_blocks
= DIO_MAX_BLOCKS
;
3449 dio_credits
= ext4_chunk_trans_blocks(inode
, max_blocks
);
3450 handle
= ext4_journal_start(inode
, dio_credits
);
3451 if (IS_ERR(handle
)) {
3452 ret
= PTR_ERR(handle
);
3455 ret
= ext4_get_blocks(handle
, inode
, iblock
, max_blocks
, bh_result
,
3458 bh_result
->b_size
= (ret
<< inode
->i_blkbits
);
3461 ext4_journal_stop(handle
);
3466 static void ext4_free_io_end(ext4_io_end_t
*io
)
3472 static void dump_aio_dio_list(struct inode
* inode
)
3475 struct list_head
*cur
, *before
, *after
;
3476 ext4_io_end_t
*io
, *io0
, *io1
;
3478 if (list_empty(&EXT4_I(inode
)->i_aio_dio_complete_list
)){
3479 ext4_debug("inode %lu aio dio list is empty\n", inode
->i_ino
);
3483 ext4_debug("Dump inode %lu aio_dio_completed_IO list \n", inode
->i_ino
);
3484 list_for_each_entry(io
, &EXT4_I(inode
)->i_aio_dio_complete_list
, list
){
3487 io0
= container_of(before
, ext4_io_end_t
, list
);
3489 io1
= container_of(after
, ext4_io_end_t
, list
);
3491 ext4_debug("io 0x%p from inode %lu,prev 0x%p,next 0x%p\n",
3492 io
, inode
->i_ino
, io0
, io1
);
3498 * check a range of space and convert unwritten extents to written.
3500 static int ext4_end_aio_dio_nolock(ext4_io_end_t
*io
)
3502 struct inode
*inode
= io
->inode
;
3503 loff_t offset
= io
->offset
;
3504 size_t size
= io
->size
;
3507 ext4_debug("end_aio_dio_onlock: io 0x%p from inode %lu,list->next 0x%p,"
3508 "list->prev 0x%p\n",
3509 io
, inode
->i_ino
, io
->list
.next
, io
->list
.prev
);
3511 if (list_empty(&io
->list
))
3514 if (io
->flag
!= DIO_AIO_UNWRITTEN
)
3517 if (offset
+ size
<= i_size_read(inode
))
3518 ret
= ext4_convert_unwritten_extents(inode
, offset
, size
);
3521 printk(KERN_EMERG
"%s: failed to convert unwritten"
3522 "extents to written extents, error is %d"
3523 " io is still on inode %lu aio dio list\n",
3524 __func__
, ret
, inode
->i_ino
);
3528 /* clear the DIO AIO unwritten flag */
3533 * work on completed aio dio IO, to convert unwritten extents to extents
3535 static void ext4_end_aio_dio_work(struct work_struct
*work
)
3537 ext4_io_end_t
*io
= container_of(work
, ext4_io_end_t
, work
);
3538 struct inode
*inode
= io
->inode
;
3541 mutex_lock(&inode
->i_mutex
);
3542 ret
= ext4_end_aio_dio_nolock(io
);
3544 if (!list_empty(&io
->list
))
3545 list_del_init(&io
->list
);
3546 ext4_free_io_end(io
);
3548 mutex_unlock(&inode
->i_mutex
);
3551 * This function is called from ext4_sync_file().
3553 * When AIO DIO IO is completed, the work to convert unwritten
3554 * extents to written is queued on workqueue but may not get immediately
3555 * scheduled. When fsync is called, we need to ensure the
3556 * conversion is complete before fsync returns.
3557 * The inode keeps track of a list of completed AIO from DIO path
3558 * that might needs to do the conversion. This function walks through
3559 * the list and convert the related unwritten extents to written.
3561 int flush_aio_dio_completed_IO(struct inode
*inode
)
3567 if (list_empty(&EXT4_I(inode
)->i_aio_dio_complete_list
))
3570 dump_aio_dio_list(inode
);
3571 while (!list_empty(&EXT4_I(inode
)->i_aio_dio_complete_list
)){
3572 io
= list_entry(EXT4_I(inode
)->i_aio_dio_complete_list
.next
,
3573 ext4_io_end_t
, list
);
3575 * Calling ext4_end_aio_dio_nolock() to convert completed
3578 * When ext4_sync_file() is called, run_queue() may already
3579 * about to flush the work corresponding to this io structure.
3580 * It will be upset if it founds the io structure related
3581 * to the work-to-be schedule is freed.
3583 * Thus we need to keep the io structure still valid here after
3584 * convertion finished. The io structure has a flag to
3585 * avoid double converting from both fsync and background work
3588 ret
= ext4_end_aio_dio_nolock(io
);
3592 list_del_init(&io
->list
);
3594 return (ret2
< 0) ? ret2
: 0;
3597 static ext4_io_end_t
*ext4_init_io_end (struct inode
*inode
)
3599 ext4_io_end_t
*io
= NULL
;
3601 io
= kmalloc(sizeof(*io
), GFP_NOFS
);
3610 INIT_WORK(&io
->work
, ext4_end_aio_dio_work
);
3611 INIT_LIST_HEAD(&io
->list
);
3617 static void ext4_end_io_dio(struct kiocb
*iocb
, loff_t offset
,
3618 ssize_t size
, void *private)
3620 ext4_io_end_t
*io_end
= iocb
->private;
3621 struct workqueue_struct
*wq
;
3623 /* if not async direct IO or dio with 0 bytes write, just return */
3624 if (!io_end
|| !size
)
3627 ext_debug("ext4_end_io_dio(): io_end 0x%p"
3628 "for inode %lu, iocb 0x%p, offset %llu, size %llu\n",
3629 iocb
->private, io_end
->inode
->i_ino
, iocb
, offset
,
3632 /* if not aio dio with unwritten extents, just free io and return */
3633 if (io_end
->flag
!= DIO_AIO_UNWRITTEN
){
3634 ext4_free_io_end(io_end
);
3635 iocb
->private = NULL
;
3639 io_end
->offset
= offset
;
3640 io_end
->size
= size
;
3641 wq
= EXT4_SB(io_end
->inode
->i_sb
)->dio_unwritten_wq
;
3643 /* queue the work to convert unwritten extents to written */
3644 queue_work(wq
, &io_end
->work
);
3646 /* Add the io_end to per-inode completed aio dio list*/
3647 list_add_tail(&io_end
->list
,
3648 &EXT4_I(io_end
->inode
)->i_aio_dio_complete_list
);
3649 iocb
->private = NULL
;
3652 * For ext4 extent files, ext4 will do direct-io write to holes,
3653 * preallocated extents, and those write extend the file, no need to
3654 * fall back to buffered IO.
3656 * For holes, we fallocate those blocks, mark them as unintialized
3657 * If those blocks were preallocated, we mark sure they are splited, but
3658 * still keep the range to write as unintialized.
3660 * The unwrritten extents will be converted to written when DIO is completed.
3661 * For async direct IO, since the IO may still pending when return, we
3662 * set up an end_io call back function, which will do the convertion
3663 * when async direct IO completed.
3665 * If the O_DIRECT write will extend the file then add this inode to the
3666 * orphan list. So recovery will truncate it back to the original size
3667 * if the machine crashes during the write.
3670 static ssize_t
ext4_ext_direct_IO(int rw
, struct kiocb
*iocb
,
3671 const struct iovec
*iov
, loff_t offset
,
3672 unsigned long nr_segs
)
3674 struct file
*file
= iocb
->ki_filp
;
3675 struct inode
*inode
= file
->f_mapping
->host
;
3677 size_t count
= iov_length(iov
, nr_segs
);
3679 loff_t final_size
= offset
+ count
;
3680 if (rw
== WRITE
&& final_size
<= inode
->i_size
) {
3682 * We could direct write to holes and fallocate.
3684 * Allocated blocks to fill the hole are marked as uninitialized
3685 * to prevent paralel buffered read to expose the stale data
3686 * before DIO complete the data IO.
3688 * As to previously fallocated extents, ext4 get_block
3689 * will just simply mark the buffer mapped but still
3690 * keep the extents uninitialized.
3692 * for non AIO case, we will convert those unwritten extents
3693 * to written after return back from blockdev_direct_IO.
3695 * for async DIO, the conversion needs to be defered when
3696 * the IO is completed. The ext4 end_io callback function
3697 * will be called to take care of the conversion work.
3698 * Here for async case, we allocate an io_end structure to
3701 iocb
->private = NULL
;
3702 EXT4_I(inode
)->cur_aio_dio
= NULL
;
3703 if (!is_sync_kiocb(iocb
)) {
3704 iocb
->private = ext4_init_io_end(inode
);
3708 * we save the io structure for current async
3709 * direct IO, so that later ext4_get_blocks()
3710 * could flag the io structure whether there
3711 * is a unwritten extents needs to be converted
3712 * when IO is completed.
3714 EXT4_I(inode
)->cur_aio_dio
= iocb
->private;
3717 ret
= blockdev_direct_IO(rw
, iocb
, inode
,
3718 inode
->i_sb
->s_bdev
, iov
,
3720 ext4_get_block_dio_write
,
3723 EXT4_I(inode
)->cur_aio_dio
= NULL
;
3725 * The io_end structure takes a reference to the inode,
3726 * that structure needs to be destroyed and the
3727 * reference to the inode need to be dropped, when IO is
3728 * complete, even with 0 byte write, or failed.
3730 * In the successful AIO DIO case, the io_end structure will be
3731 * desctroyed and the reference to the inode will be dropped
3732 * after the end_io call back function is called.
3734 * In the case there is 0 byte write, or error case, since
3735 * VFS direct IO won't invoke the end_io call back function,
3736 * we need to free the end_io structure here.
3738 if (ret
!= -EIOCBQUEUED
&& ret
<= 0 && iocb
->private) {
3739 ext4_free_io_end(iocb
->private);
3740 iocb
->private = NULL
;
3741 } else if (ret
> 0 && (EXT4_I(inode
)->i_state
&
3742 EXT4_STATE_DIO_UNWRITTEN
)) {
3745 * for non AIO case, since the IO is already
3746 * completed, we could do the convertion right here
3748 err
= ext4_convert_unwritten_extents(inode
,
3752 EXT4_I(inode
)->i_state
&= ~EXT4_STATE_DIO_UNWRITTEN
;
3757 /* for write the the end of file case, we fall back to old way */
3758 return ext4_ind_direct_IO(rw
, iocb
, iov
, offset
, nr_segs
);
3761 static ssize_t
ext4_direct_IO(int rw
, struct kiocb
*iocb
,
3762 const struct iovec
*iov
, loff_t offset
,
3763 unsigned long nr_segs
)
3765 struct file
*file
= iocb
->ki_filp
;
3766 struct inode
*inode
= file
->f_mapping
->host
;
3768 if (EXT4_I(inode
)->i_flags
& EXT4_EXTENTS_FL
)
3769 return ext4_ext_direct_IO(rw
, iocb
, iov
, offset
, nr_segs
);
3771 return ext4_ind_direct_IO(rw
, iocb
, iov
, offset
, nr_segs
);
3775 * Pages can be marked dirty completely asynchronously from ext4's journalling
3776 * activity. By filemap_sync_pte(), try_to_unmap_one(), etc. We cannot do
3777 * much here because ->set_page_dirty is called under VFS locks. The page is
3778 * not necessarily locked.
3780 * We cannot just dirty the page and leave attached buffers clean, because the
3781 * buffers' dirty state is "definitive". We cannot just set the buffers dirty
3782 * or jbddirty because all the journalling code will explode.
3784 * So what we do is to mark the page "pending dirty" and next time writepage
3785 * is called, propagate that into the buffers appropriately.
3787 static int ext4_journalled_set_page_dirty(struct page
*page
)
3789 SetPageChecked(page
);
3790 return __set_page_dirty_nobuffers(page
);
3793 static const struct address_space_operations ext4_ordered_aops
= {
3794 .readpage
= ext4_readpage
,
3795 .readpages
= ext4_readpages
,
3796 .writepage
= ext4_writepage
,
3797 .sync_page
= block_sync_page
,
3798 .write_begin
= ext4_write_begin
,
3799 .write_end
= ext4_ordered_write_end
,
3801 .invalidatepage
= ext4_invalidatepage
,
3802 .releasepage
= ext4_releasepage
,
3803 .direct_IO
= ext4_direct_IO
,
3804 .migratepage
= buffer_migrate_page
,
3805 .is_partially_uptodate
= block_is_partially_uptodate
,
3806 .error_remove_page
= generic_error_remove_page
,
3809 static const struct address_space_operations ext4_writeback_aops
= {
3810 .readpage
= ext4_readpage
,
3811 .readpages
= ext4_readpages
,
3812 .writepage
= ext4_writepage
,
3813 .sync_page
= block_sync_page
,
3814 .write_begin
= ext4_write_begin
,
3815 .write_end
= ext4_writeback_write_end
,
3817 .invalidatepage
= ext4_invalidatepage
,
3818 .releasepage
= ext4_releasepage
,
3819 .direct_IO
= ext4_direct_IO
,
3820 .migratepage
= buffer_migrate_page
,
3821 .is_partially_uptodate
= block_is_partially_uptodate
,
3822 .error_remove_page
= generic_error_remove_page
,
3825 static const struct address_space_operations ext4_journalled_aops
= {
3826 .readpage
= ext4_readpage
,
3827 .readpages
= ext4_readpages
,
3828 .writepage
= ext4_writepage
,
3829 .sync_page
= block_sync_page
,
3830 .write_begin
= ext4_write_begin
,
3831 .write_end
= ext4_journalled_write_end
,
3832 .set_page_dirty
= ext4_journalled_set_page_dirty
,
3834 .invalidatepage
= ext4_invalidatepage
,
3835 .releasepage
= ext4_releasepage
,
3836 .is_partially_uptodate
= block_is_partially_uptodate
,
3837 .error_remove_page
= generic_error_remove_page
,
3840 static const struct address_space_operations ext4_da_aops
= {
3841 .readpage
= ext4_readpage
,
3842 .readpages
= ext4_readpages
,
3843 .writepage
= ext4_writepage
,
3844 .writepages
= ext4_da_writepages
,
3845 .sync_page
= block_sync_page
,
3846 .write_begin
= ext4_da_write_begin
,
3847 .write_end
= ext4_da_write_end
,
3849 .invalidatepage
= ext4_da_invalidatepage
,
3850 .releasepage
= ext4_releasepage
,
3851 .direct_IO
= ext4_direct_IO
,
3852 .migratepage
= buffer_migrate_page
,
3853 .is_partially_uptodate
= block_is_partially_uptodate
,
3854 .error_remove_page
= generic_error_remove_page
,
3857 void ext4_set_aops(struct inode
*inode
)
3859 if (ext4_should_order_data(inode
) &&
3860 test_opt(inode
->i_sb
, DELALLOC
))
3861 inode
->i_mapping
->a_ops
= &ext4_da_aops
;
3862 else if (ext4_should_order_data(inode
))
3863 inode
->i_mapping
->a_ops
= &ext4_ordered_aops
;
3864 else if (ext4_should_writeback_data(inode
) &&
3865 test_opt(inode
->i_sb
, DELALLOC
))
3866 inode
->i_mapping
->a_ops
= &ext4_da_aops
;
3867 else if (ext4_should_writeback_data(inode
))
3868 inode
->i_mapping
->a_ops
= &ext4_writeback_aops
;
3870 inode
->i_mapping
->a_ops
= &ext4_journalled_aops
;
3874 * ext4_block_truncate_page() zeroes out a mapping from file offset `from'
3875 * up to the end of the block which corresponds to `from'.
3876 * This required during truncate. We need to physically zero the tail end
3877 * of that block so it doesn't yield old data if the file is later grown.
3879 int ext4_block_truncate_page(handle_t
*handle
,
3880 struct address_space
*mapping
, loff_t from
)
3882 ext4_fsblk_t index
= from
>> PAGE_CACHE_SHIFT
;
3883 unsigned offset
= from
& (PAGE_CACHE_SIZE
-1);
3884 unsigned blocksize
, length
, pos
;
3886 struct inode
*inode
= mapping
->host
;
3887 struct buffer_head
*bh
;
3891 page
= find_or_create_page(mapping
, from
>> PAGE_CACHE_SHIFT
,
3892 mapping_gfp_mask(mapping
) & ~__GFP_FS
);
3896 blocksize
= inode
->i_sb
->s_blocksize
;
3897 length
= blocksize
- (offset
& (blocksize
- 1));
3898 iblock
= index
<< (PAGE_CACHE_SHIFT
- inode
->i_sb
->s_blocksize_bits
);
3901 * For "nobh" option, we can only work if we don't need to
3902 * read-in the page - otherwise we create buffers to do the IO.
3904 if (!page_has_buffers(page
) && test_opt(inode
->i_sb
, NOBH
) &&
3905 ext4_should_writeback_data(inode
) && PageUptodate(page
)) {
3906 zero_user(page
, offset
, length
);
3907 set_page_dirty(page
);
3911 if (!page_has_buffers(page
))
3912 create_empty_buffers(page
, blocksize
, 0);
3914 /* Find the buffer that contains "offset" */
3915 bh
= page_buffers(page
);
3917 while (offset
>= pos
) {
3918 bh
= bh
->b_this_page
;
3924 if (buffer_freed(bh
)) {
3925 BUFFER_TRACE(bh
, "freed: skip");
3929 if (!buffer_mapped(bh
)) {
3930 BUFFER_TRACE(bh
, "unmapped");
3931 ext4_get_block(inode
, iblock
, bh
, 0);
3932 /* unmapped? It's a hole - nothing to do */
3933 if (!buffer_mapped(bh
)) {
3934 BUFFER_TRACE(bh
, "still unmapped");
3939 /* Ok, it's mapped. Make sure it's up-to-date */
3940 if (PageUptodate(page
))
3941 set_buffer_uptodate(bh
);
3943 if (!buffer_uptodate(bh
)) {
3945 ll_rw_block(READ
, 1, &bh
);
3947 /* Uhhuh. Read error. Complain and punt. */
3948 if (!buffer_uptodate(bh
))
3952 if (ext4_should_journal_data(inode
)) {
3953 BUFFER_TRACE(bh
, "get write access");
3954 err
= ext4_journal_get_write_access(handle
, bh
);
3959 zero_user(page
, offset
, length
);
3961 BUFFER_TRACE(bh
, "zeroed end of block");
3964 if (ext4_should_journal_data(inode
)) {
3965 err
= ext4_handle_dirty_metadata(handle
, inode
, bh
);
3967 if (ext4_should_order_data(inode
))
3968 err
= ext4_jbd2_file_inode(handle
, inode
);
3969 mark_buffer_dirty(bh
);
3974 page_cache_release(page
);
3979 * Probably it should be a library function... search for first non-zero word
3980 * or memcmp with zero_page, whatever is better for particular architecture.
3983 static inline int all_zeroes(__le32
*p
, __le32
*q
)
3992 * ext4_find_shared - find the indirect blocks for partial truncation.
3993 * @inode: inode in question
3994 * @depth: depth of the affected branch
3995 * @offsets: offsets of pointers in that branch (see ext4_block_to_path)
3996 * @chain: place to store the pointers to partial indirect blocks
3997 * @top: place to the (detached) top of branch
3999 * This is a helper function used by ext4_truncate().
4001 * When we do truncate() we may have to clean the ends of several
4002 * indirect blocks but leave the blocks themselves alive. Block is
4003 * partially truncated if some data below the new i_size is refered
4004 * from it (and it is on the path to the first completely truncated
4005 * data block, indeed). We have to free the top of that path along
4006 * with everything to the right of the path. Since no allocation
4007 * past the truncation point is possible until ext4_truncate()
4008 * finishes, we may safely do the latter, but top of branch may
4009 * require special attention - pageout below the truncation point
4010 * might try to populate it.
4012 * We atomically detach the top of branch from the tree, store the
4013 * block number of its root in *@top, pointers to buffer_heads of
4014 * partially truncated blocks - in @chain[].bh and pointers to
4015 * their last elements that should not be removed - in
4016 * @chain[].p. Return value is the pointer to last filled element
4019 * The work left to caller to do the actual freeing of subtrees:
4020 * a) free the subtree starting from *@top
4021 * b) free the subtrees whose roots are stored in
4022 * (@chain[i].p+1 .. end of @chain[i].bh->b_data)
4023 * c) free the subtrees growing from the inode past the @chain[0].
4024 * (no partially truncated stuff there). */
4026 static Indirect
*ext4_find_shared(struct inode
*inode
, int depth
,
4027 ext4_lblk_t offsets
[4], Indirect chain
[4],
4030 Indirect
*partial
, *p
;
4034 /* Make k index the deepest non-null offest + 1 */
4035 for (k
= depth
; k
> 1 && !offsets
[k
-1]; k
--)
4037 partial
= ext4_get_branch(inode
, k
, offsets
, chain
, &err
);
4038 /* Writer: pointers */
4040 partial
= chain
+ k
-1;
4042 * If the branch acquired continuation since we've looked at it -
4043 * fine, it should all survive and (new) top doesn't belong to us.
4045 if (!partial
->key
&& *partial
->p
)
4048 for (p
= partial
; (p
> chain
) && all_zeroes((__le32
*) p
->bh
->b_data
, p
->p
); p
--)
4051 * OK, we've found the last block that must survive. The rest of our
4052 * branch should be detached before unlocking. However, if that rest
4053 * of branch is all ours and does not grow immediately from the inode
4054 * it's easier to cheat and just decrement partial->p.
4056 if (p
== chain
+ k
- 1 && p
> chain
) {
4060 /* Nope, don't do this in ext4. Must leave the tree intact */
4067 while (partial
> p
) {
4068 brelse(partial
->bh
);
4076 * Zero a number of block pointers in either an inode or an indirect block.
4077 * If we restart the transaction we must again get write access to the
4078 * indirect block for further modification.
4080 * We release `count' blocks on disk, but (last - first) may be greater
4081 * than `count' because there can be holes in there.
4083 static void ext4_clear_blocks(handle_t
*handle
, struct inode
*inode
,
4084 struct buffer_head
*bh
,
4085 ext4_fsblk_t block_to_free
,
4086 unsigned long count
, __le32
*first
,
4090 int flags
= EXT4_FREE_BLOCKS_FORGET
;
4092 if (S_ISDIR(inode
->i_mode
) || S_ISLNK(inode
->i_mode
))
4093 flags
|= EXT4_FREE_BLOCKS_METADATA
;
4095 if (try_to_extend_transaction(handle
, inode
)) {
4097 BUFFER_TRACE(bh
, "call ext4_handle_dirty_metadata");
4098 ext4_handle_dirty_metadata(handle
, inode
, bh
);
4100 ext4_mark_inode_dirty(handle
, inode
);
4101 ext4_truncate_restart_trans(handle
, inode
,
4102 blocks_for_truncate(inode
));
4104 BUFFER_TRACE(bh
, "retaking write access");
4105 ext4_journal_get_write_access(handle
, bh
);
4109 for (p
= first
; p
< last
; p
++)
4112 ext4_free_blocks(handle
, inode
, 0, block_to_free
, count
, flags
);
4116 * ext4_free_data - free a list of data blocks
4117 * @handle: handle for this transaction
4118 * @inode: inode we are dealing with
4119 * @this_bh: indirect buffer_head which contains *@first and *@last
4120 * @first: array of block numbers
4121 * @last: points immediately past the end of array
4123 * We are freeing all blocks refered from that array (numbers are stored as
4124 * little-endian 32-bit) and updating @inode->i_blocks appropriately.
4126 * We accumulate contiguous runs of blocks to free. Conveniently, if these
4127 * blocks are contiguous then releasing them at one time will only affect one
4128 * or two bitmap blocks (+ group descriptor(s) and superblock) and we won't
4129 * actually use a lot of journal space.
4131 * @this_bh will be %NULL if @first and @last point into the inode's direct
4134 static void ext4_free_data(handle_t
*handle
, struct inode
*inode
,
4135 struct buffer_head
*this_bh
,
4136 __le32
*first
, __le32
*last
)
4138 ext4_fsblk_t block_to_free
= 0; /* Starting block # of a run */
4139 unsigned long count
= 0; /* Number of blocks in the run */
4140 __le32
*block_to_free_p
= NULL
; /* Pointer into inode/ind
4143 ext4_fsblk_t nr
; /* Current block # */
4144 __le32
*p
; /* Pointer into inode/ind
4145 for current block */
4148 if (this_bh
) { /* For indirect block */
4149 BUFFER_TRACE(this_bh
, "get_write_access");
4150 err
= ext4_journal_get_write_access(handle
, this_bh
);
4151 /* Important: if we can't update the indirect pointers
4152 * to the blocks, we can't free them. */
4157 for (p
= first
; p
< last
; p
++) {
4158 nr
= le32_to_cpu(*p
);
4160 /* accumulate blocks to free if they're contiguous */
4163 block_to_free_p
= p
;
4165 } else if (nr
== block_to_free
+ count
) {
4168 ext4_clear_blocks(handle
, inode
, this_bh
,
4170 count
, block_to_free_p
, p
);
4172 block_to_free_p
= p
;
4179 ext4_clear_blocks(handle
, inode
, this_bh
, block_to_free
,
4180 count
, block_to_free_p
, p
);
4183 BUFFER_TRACE(this_bh
, "call ext4_handle_dirty_metadata");
4186 * The buffer head should have an attached journal head at this
4187 * point. However, if the data is corrupted and an indirect
4188 * block pointed to itself, it would have been detached when
4189 * the block was cleared. Check for this instead of OOPSing.
4191 if ((EXT4_JOURNAL(inode
) == NULL
) || bh2jh(this_bh
))
4192 ext4_handle_dirty_metadata(handle
, inode
, this_bh
);
4194 ext4_error(inode
->i_sb
, __func__
,
4195 "circular indirect block detected, "
4196 "inode=%lu, block=%llu",
4198 (unsigned long long) this_bh
->b_blocknr
);
4203 * ext4_free_branches - free an array of branches
4204 * @handle: JBD handle for this transaction
4205 * @inode: inode we are dealing with
4206 * @parent_bh: the buffer_head which contains *@first and *@last
4207 * @first: array of block numbers
4208 * @last: pointer immediately past the end of array
4209 * @depth: depth of the branches to free
4211 * We are freeing all blocks refered from these branches (numbers are
4212 * stored as little-endian 32-bit) and updating @inode->i_blocks
4215 static void ext4_free_branches(handle_t
*handle
, struct inode
*inode
,
4216 struct buffer_head
*parent_bh
,
4217 __le32
*first
, __le32
*last
, int depth
)
4222 if (ext4_handle_is_aborted(handle
))
4226 struct buffer_head
*bh
;
4227 int addr_per_block
= EXT4_ADDR_PER_BLOCK(inode
->i_sb
);
4229 while (--p
>= first
) {
4230 nr
= le32_to_cpu(*p
);
4232 continue; /* A hole */
4234 /* Go read the buffer for the next level down */
4235 bh
= sb_bread(inode
->i_sb
, nr
);
4238 * A read failure? Report error and clear slot
4242 ext4_error(inode
->i_sb
, "ext4_free_branches",
4243 "Read failure, inode=%lu, block=%llu",
4248 /* This zaps the entire block. Bottom up. */
4249 BUFFER_TRACE(bh
, "free child branches");
4250 ext4_free_branches(handle
, inode
, bh
,
4251 (__le32
*) bh
->b_data
,
4252 (__le32
*) bh
->b_data
+ addr_per_block
,
4256 * We've probably journalled the indirect block several
4257 * times during the truncate. But it's no longer
4258 * needed and we now drop it from the transaction via
4259 * jbd2_journal_revoke().
4261 * That's easy if it's exclusively part of this
4262 * transaction. But if it's part of the committing
4263 * transaction then jbd2_journal_forget() will simply
4264 * brelse() it. That means that if the underlying
4265 * block is reallocated in ext4_get_block(),
4266 * unmap_underlying_metadata() will find this block
4267 * and will try to get rid of it. damn, damn.
4269 * If this block has already been committed to the
4270 * journal, a revoke record will be written. And
4271 * revoke records must be emitted *before* clearing
4272 * this block's bit in the bitmaps.
4274 ext4_forget(handle
, 1, inode
, bh
, bh
->b_blocknr
);
4277 * Everything below this this pointer has been
4278 * released. Now let this top-of-subtree go.
4280 * We want the freeing of this indirect block to be
4281 * atomic in the journal with the updating of the
4282 * bitmap block which owns it. So make some room in
4285 * We zero the parent pointer *after* freeing its
4286 * pointee in the bitmaps, so if extend_transaction()
4287 * for some reason fails to put the bitmap changes and
4288 * the release into the same transaction, recovery
4289 * will merely complain about releasing a free block,
4290 * rather than leaking blocks.
4292 if (ext4_handle_is_aborted(handle
))
4294 if (try_to_extend_transaction(handle
, inode
)) {
4295 ext4_mark_inode_dirty(handle
, inode
);
4296 ext4_truncate_restart_trans(handle
, inode
,
4297 blocks_for_truncate(inode
));
4300 ext4_free_blocks(handle
, inode
, 0, nr
, 1,
4301 EXT4_FREE_BLOCKS_METADATA
);
4305 * The block which we have just freed is
4306 * pointed to by an indirect block: journal it
4308 BUFFER_TRACE(parent_bh
, "get_write_access");
4309 if (!ext4_journal_get_write_access(handle
,
4312 BUFFER_TRACE(parent_bh
,
4313 "call ext4_handle_dirty_metadata");
4314 ext4_handle_dirty_metadata(handle
,
4321 /* We have reached the bottom of the tree. */
4322 BUFFER_TRACE(parent_bh
, "free data blocks");
4323 ext4_free_data(handle
, inode
, parent_bh
, first
, last
);
4327 int ext4_can_truncate(struct inode
*inode
)
4329 if (IS_APPEND(inode
) || IS_IMMUTABLE(inode
))
4331 if (S_ISREG(inode
->i_mode
))
4333 if (S_ISDIR(inode
->i_mode
))
4335 if (S_ISLNK(inode
->i_mode
))
4336 return !ext4_inode_is_fast_symlink(inode
);
4343 * We block out ext4_get_block() block instantiations across the entire
4344 * transaction, and VFS/VM ensures that ext4_truncate() cannot run
4345 * simultaneously on behalf of the same inode.
4347 * As we work through the truncate and commmit bits of it to the journal there
4348 * is one core, guiding principle: the file's tree must always be consistent on
4349 * disk. We must be able to restart the truncate after a crash.
4351 * The file's tree may be transiently inconsistent in memory (although it
4352 * probably isn't), but whenever we close off and commit a journal transaction,
4353 * the contents of (the filesystem + the journal) must be consistent and
4354 * restartable. It's pretty simple, really: bottom up, right to left (although
4355 * left-to-right works OK too).
4357 * Note that at recovery time, journal replay occurs *before* the restart of
4358 * truncate against the orphan inode list.
4360 * The committed inode has the new, desired i_size (which is the same as
4361 * i_disksize in this case). After a crash, ext4_orphan_cleanup() will see
4362 * that this inode's truncate did not complete and it will again call
4363 * ext4_truncate() to have another go. So there will be instantiated blocks
4364 * to the right of the truncation point in a crashed ext4 filesystem. But
4365 * that's fine - as long as they are linked from the inode, the post-crash
4366 * ext4_truncate() run will find them and release them.
4368 void ext4_truncate(struct inode
*inode
)
4371 struct ext4_inode_info
*ei
= EXT4_I(inode
);
4372 __le32
*i_data
= ei
->i_data
;
4373 int addr_per_block
= EXT4_ADDR_PER_BLOCK(inode
->i_sb
);
4374 struct address_space
*mapping
= inode
->i_mapping
;
4375 ext4_lblk_t offsets
[4];
4380 ext4_lblk_t last_block
;
4381 unsigned blocksize
= inode
->i_sb
->s_blocksize
;
4383 if (!ext4_can_truncate(inode
))
4386 if (inode
->i_size
== 0 && !test_opt(inode
->i_sb
, NO_AUTO_DA_ALLOC
))
4387 ei
->i_state
|= EXT4_STATE_DA_ALLOC_CLOSE
;
4389 if (EXT4_I(inode
)->i_flags
& EXT4_EXTENTS_FL
) {
4390 ext4_ext_truncate(inode
);
4394 handle
= start_transaction(inode
);
4396 return; /* AKPM: return what? */
4398 last_block
= (inode
->i_size
+ blocksize
-1)
4399 >> EXT4_BLOCK_SIZE_BITS(inode
->i_sb
);
4401 if (inode
->i_size
& (blocksize
- 1))
4402 if (ext4_block_truncate_page(handle
, mapping
, inode
->i_size
))
4405 n
= ext4_block_to_path(inode
, last_block
, offsets
, NULL
);
4407 goto out_stop
; /* error */
4410 * OK. This truncate is going to happen. We add the inode to the
4411 * orphan list, so that if this truncate spans multiple transactions,
4412 * and we crash, we will resume the truncate when the filesystem
4413 * recovers. It also marks the inode dirty, to catch the new size.
4415 * Implication: the file must always be in a sane, consistent
4416 * truncatable state while each transaction commits.
4418 if (ext4_orphan_add(handle
, inode
))
4422 * From here we block out all ext4_get_block() callers who want to
4423 * modify the block allocation tree.
4425 down_write(&ei
->i_data_sem
);
4427 ext4_discard_preallocations(inode
);
4430 * The orphan list entry will now protect us from any crash which
4431 * occurs before the truncate completes, so it is now safe to propagate
4432 * the new, shorter inode size (held for now in i_size) into the
4433 * on-disk inode. We do this via i_disksize, which is the value which
4434 * ext4 *really* writes onto the disk inode.
4436 ei
->i_disksize
= inode
->i_size
;
4438 if (n
== 1) { /* direct blocks */
4439 ext4_free_data(handle
, inode
, NULL
, i_data
+offsets
[0],
4440 i_data
+ EXT4_NDIR_BLOCKS
);
4444 partial
= ext4_find_shared(inode
, n
, offsets
, chain
, &nr
);
4445 /* Kill the top of shared branch (not detached) */
4447 if (partial
== chain
) {
4448 /* Shared branch grows from the inode */
4449 ext4_free_branches(handle
, inode
, NULL
,
4450 &nr
, &nr
+1, (chain
+n
-1) - partial
);
4453 * We mark the inode dirty prior to restart,
4454 * and prior to stop. No need for it here.
4457 /* Shared branch grows from an indirect block */
4458 BUFFER_TRACE(partial
->bh
, "get_write_access");
4459 ext4_free_branches(handle
, inode
, partial
->bh
,
4461 partial
->p
+1, (chain
+n
-1) - partial
);
4464 /* Clear the ends of indirect blocks on the shared branch */
4465 while (partial
> chain
) {
4466 ext4_free_branches(handle
, inode
, partial
->bh
, partial
->p
+ 1,
4467 (__le32
*)partial
->bh
->b_data
+addr_per_block
,
4468 (chain
+n
-1) - partial
);
4469 BUFFER_TRACE(partial
->bh
, "call brelse");
4470 brelse(partial
->bh
);
4474 /* Kill the remaining (whole) subtrees */
4475 switch (offsets
[0]) {
4477 nr
= i_data
[EXT4_IND_BLOCK
];
4479 ext4_free_branches(handle
, inode
, NULL
, &nr
, &nr
+1, 1);
4480 i_data
[EXT4_IND_BLOCK
] = 0;
4482 case EXT4_IND_BLOCK
:
4483 nr
= i_data
[EXT4_DIND_BLOCK
];
4485 ext4_free_branches(handle
, inode
, NULL
, &nr
, &nr
+1, 2);
4486 i_data
[EXT4_DIND_BLOCK
] = 0;
4488 case EXT4_DIND_BLOCK
:
4489 nr
= i_data
[EXT4_TIND_BLOCK
];
4491 ext4_free_branches(handle
, inode
, NULL
, &nr
, &nr
+1, 3);
4492 i_data
[EXT4_TIND_BLOCK
] = 0;
4494 case EXT4_TIND_BLOCK
:
4498 up_write(&ei
->i_data_sem
);
4499 inode
->i_mtime
= inode
->i_ctime
= ext4_current_time(inode
);
4500 ext4_mark_inode_dirty(handle
, inode
);
4503 * In a multi-transaction truncate, we only make the final transaction
4507 ext4_handle_sync(handle
);
4510 * If this was a simple ftruncate(), and the file will remain alive
4511 * then we need to clear up the orphan record which we created above.
4512 * However, if this was a real unlink then we were called by
4513 * ext4_delete_inode(), and we allow that function to clean up the
4514 * orphan info for us.
4517 ext4_orphan_del(handle
, inode
);
4519 ext4_journal_stop(handle
);
4523 * ext4_get_inode_loc returns with an extra refcount against the inode's
4524 * underlying buffer_head on success. If 'in_mem' is true, we have all
4525 * data in memory that is needed to recreate the on-disk version of this
4528 static int __ext4_get_inode_loc(struct inode
*inode
,
4529 struct ext4_iloc
*iloc
, int in_mem
)
4531 struct ext4_group_desc
*gdp
;
4532 struct buffer_head
*bh
;
4533 struct super_block
*sb
= inode
->i_sb
;
4535 int inodes_per_block
, inode_offset
;
4538 if (!ext4_valid_inum(sb
, inode
->i_ino
))
4541 iloc
->block_group
= (inode
->i_ino
- 1) / EXT4_INODES_PER_GROUP(sb
);
4542 gdp
= ext4_get_group_desc(sb
, iloc
->block_group
, NULL
);
4547 * Figure out the offset within the block group inode table
4549 inodes_per_block
= (EXT4_BLOCK_SIZE(sb
) / EXT4_INODE_SIZE(sb
));
4550 inode_offset
= ((inode
->i_ino
- 1) %
4551 EXT4_INODES_PER_GROUP(sb
));
4552 block
= ext4_inode_table(sb
, gdp
) + (inode_offset
/ inodes_per_block
);
4553 iloc
->offset
= (inode_offset
% inodes_per_block
) * EXT4_INODE_SIZE(sb
);
4555 bh
= sb_getblk(sb
, block
);
4557 ext4_error(sb
, "ext4_get_inode_loc", "unable to read "
4558 "inode block - inode=%lu, block=%llu",
4559 inode
->i_ino
, block
);
4562 if (!buffer_uptodate(bh
)) {
4566 * If the buffer has the write error flag, we have failed
4567 * to write out another inode in the same block. In this
4568 * case, we don't have to read the block because we may
4569 * read the old inode data successfully.
4571 if (buffer_write_io_error(bh
) && !buffer_uptodate(bh
))
4572 set_buffer_uptodate(bh
);
4574 if (buffer_uptodate(bh
)) {
4575 /* someone brought it uptodate while we waited */
4581 * If we have all information of the inode in memory and this
4582 * is the only valid inode in the block, we need not read the
4586 struct buffer_head
*bitmap_bh
;
4589 start
= inode_offset
& ~(inodes_per_block
- 1);
4591 /* Is the inode bitmap in cache? */
4592 bitmap_bh
= sb_getblk(sb
, ext4_inode_bitmap(sb
, gdp
));
4597 * If the inode bitmap isn't in cache then the
4598 * optimisation may end up performing two reads instead
4599 * of one, so skip it.
4601 if (!buffer_uptodate(bitmap_bh
)) {
4605 for (i
= start
; i
< start
+ inodes_per_block
; i
++) {
4606 if (i
== inode_offset
)
4608 if (ext4_test_bit(i
, bitmap_bh
->b_data
))
4612 if (i
== start
+ inodes_per_block
) {
4613 /* all other inodes are free, so skip I/O */
4614 memset(bh
->b_data
, 0, bh
->b_size
);
4615 set_buffer_uptodate(bh
);
4623 * If we need to do any I/O, try to pre-readahead extra
4624 * blocks from the inode table.
4626 if (EXT4_SB(sb
)->s_inode_readahead_blks
) {
4627 ext4_fsblk_t b
, end
, table
;
4630 table
= ext4_inode_table(sb
, gdp
);
4631 /* s_inode_readahead_blks is always a power of 2 */
4632 b
= block
& ~(EXT4_SB(sb
)->s_inode_readahead_blks
-1);
4635 end
= b
+ EXT4_SB(sb
)->s_inode_readahead_blks
;
4636 num
= EXT4_INODES_PER_GROUP(sb
);
4637 if (EXT4_HAS_RO_COMPAT_FEATURE(sb
,
4638 EXT4_FEATURE_RO_COMPAT_GDT_CSUM
))
4639 num
-= ext4_itable_unused_count(sb
, gdp
);
4640 table
+= num
/ inodes_per_block
;
4644 sb_breadahead(sb
, b
++);
4648 * There are other valid inodes in the buffer, this inode
4649 * has in-inode xattrs, or we don't have this inode in memory.
4650 * Read the block from disk.
4653 bh
->b_end_io
= end_buffer_read_sync
;
4654 submit_bh(READ_META
, bh
);
4656 if (!buffer_uptodate(bh
)) {
4657 ext4_error(sb
, __func__
,
4658 "unable to read inode block - inode=%lu, "
4659 "block=%llu", inode
->i_ino
, block
);
4669 int ext4_get_inode_loc(struct inode
*inode
, struct ext4_iloc
*iloc
)
4671 /* We have all inode data except xattrs in memory here. */
4672 return __ext4_get_inode_loc(inode
, iloc
,
4673 !(EXT4_I(inode
)->i_state
& EXT4_STATE_XATTR
));
4676 void ext4_set_inode_flags(struct inode
*inode
)
4678 unsigned int flags
= EXT4_I(inode
)->i_flags
;
4680 inode
->i_flags
&= ~(S_SYNC
|S_APPEND
|S_IMMUTABLE
|S_NOATIME
|S_DIRSYNC
);
4681 if (flags
& EXT4_SYNC_FL
)
4682 inode
->i_flags
|= S_SYNC
;
4683 if (flags
& EXT4_APPEND_FL
)
4684 inode
->i_flags
|= S_APPEND
;
4685 if (flags
& EXT4_IMMUTABLE_FL
)
4686 inode
->i_flags
|= S_IMMUTABLE
;
4687 if (flags
& EXT4_NOATIME_FL
)
4688 inode
->i_flags
|= S_NOATIME
;
4689 if (flags
& EXT4_DIRSYNC_FL
)
4690 inode
->i_flags
|= S_DIRSYNC
;
4693 /* Propagate flags from i_flags to EXT4_I(inode)->i_flags */
4694 void ext4_get_inode_flags(struct ext4_inode_info
*ei
)
4696 unsigned int flags
= ei
->vfs_inode
.i_flags
;
4698 ei
->i_flags
&= ~(EXT4_SYNC_FL
|EXT4_APPEND_FL
|
4699 EXT4_IMMUTABLE_FL
|EXT4_NOATIME_FL
|EXT4_DIRSYNC_FL
);
4701 ei
->i_flags
|= EXT4_SYNC_FL
;
4702 if (flags
& S_APPEND
)
4703 ei
->i_flags
|= EXT4_APPEND_FL
;
4704 if (flags
& S_IMMUTABLE
)
4705 ei
->i_flags
|= EXT4_IMMUTABLE_FL
;
4706 if (flags
& S_NOATIME
)
4707 ei
->i_flags
|= EXT4_NOATIME_FL
;
4708 if (flags
& S_DIRSYNC
)
4709 ei
->i_flags
|= EXT4_DIRSYNC_FL
;
4712 static blkcnt_t
ext4_inode_blocks(struct ext4_inode
*raw_inode
,
4713 struct ext4_inode_info
*ei
)
4716 struct inode
*inode
= &(ei
->vfs_inode
);
4717 struct super_block
*sb
= inode
->i_sb
;
4719 if (EXT4_HAS_RO_COMPAT_FEATURE(sb
,
4720 EXT4_FEATURE_RO_COMPAT_HUGE_FILE
)) {
4721 /* we are using combined 48 bit field */
4722 i_blocks
= ((u64
)le16_to_cpu(raw_inode
->i_blocks_high
)) << 32 |
4723 le32_to_cpu(raw_inode
->i_blocks_lo
);
4724 if (ei
->i_flags
& EXT4_HUGE_FILE_FL
) {
4725 /* i_blocks represent file system block size */
4726 return i_blocks
<< (inode
->i_blkbits
- 9);
4731 return le32_to_cpu(raw_inode
->i_blocks_lo
);
4735 struct inode
*ext4_iget(struct super_block
*sb
, unsigned long ino
)
4737 struct ext4_iloc iloc
;
4738 struct ext4_inode
*raw_inode
;
4739 struct ext4_inode_info
*ei
;
4740 struct inode
*inode
;
4744 inode
= iget_locked(sb
, ino
);
4746 return ERR_PTR(-ENOMEM
);
4747 if (!(inode
->i_state
& I_NEW
))
4753 ret
= __ext4_get_inode_loc(inode
, &iloc
, 0);
4756 raw_inode
= ext4_raw_inode(&iloc
);
4757 inode
->i_mode
= le16_to_cpu(raw_inode
->i_mode
);
4758 inode
->i_uid
= (uid_t
)le16_to_cpu(raw_inode
->i_uid_low
);
4759 inode
->i_gid
= (gid_t
)le16_to_cpu(raw_inode
->i_gid_low
);
4760 if (!(test_opt(inode
->i_sb
, NO_UID32
))) {
4761 inode
->i_uid
|= le16_to_cpu(raw_inode
->i_uid_high
) << 16;
4762 inode
->i_gid
|= le16_to_cpu(raw_inode
->i_gid_high
) << 16;
4764 inode
->i_nlink
= le16_to_cpu(raw_inode
->i_links_count
);
4767 ei
->i_dir_start_lookup
= 0;
4768 ei
->i_dtime
= le32_to_cpu(raw_inode
->i_dtime
);
4769 /* We now have enough fields to check if the inode was active or not.
4770 * This is needed because nfsd might try to access dead inodes
4771 * the test is that same one that e2fsck uses
4772 * NeilBrown 1999oct15
4774 if (inode
->i_nlink
== 0) {
4775 if (inode
->i_mode
== 0 ||
4776 !(EXT4_SB(inode
->i_sb
)->s_mount_state
& EXT4_ORPHAN_FS
)) {
4777 /* this inode is deleted */
4781 /* The only unlinked inodes we let through here have
4782 * valid i_mode and are being read by the orphan
4783 * recovery code: that's fine, we're about to complete
4784 * the process of deleting those. */
4786 ei
->i_flags
= le32_to_cpu(raw_inode
->i_flags
);
4787 inode
->i_blocks
= ext4_inode_blocks(raw_inode
, ei
);
4788 ei
->i_file_acl
= le32_to_cpu(raw_inode
->i_file_acl_lo
);
4789 if (EXT4_HAS_INCOMPAT_FEATURE(sb
, EXT4_FEATURE_INCOMPAT_64BIT
))
4791 ((__u64
)le16_to_cpu(raw_inode
->i_file_acl_high
)) << 32;
4792 inode
->i_size
= ext4_isize(raw_inode
);
4793 ei
->i_disksize
= inode
->i_size
;
4794 inode
->i_generation
= le32_to_cpu(raw_inode
->i_generation
);
4795 ei
->i_block_group
= iloc
.block_group
;
4796 ei
->i_last_alloc_group
= ~0;
4798 * NOTE! The in-memory inode i_data array is in little-endian order
4799 * even on big-endian machines: we do NOT byteswap the block numbers!
4801 for (block
= 0; block
< EXT4_N_BLOCKS
; block
++)
4802 ei
->i_data
[block
] = raw_inode
->i_block
[block
];
4803 INIT_LIST_HEAD(&ei
->i_orphan
);
4805 if (EXT4_INODE_SIZE(inode
->i_sb
) > EXT4_GOOD_OLD_INODE_SIZE
) {
4806 ei
->i_extra_isize
= le16_to_cpu(raw_inode
->i_extra_isize
);
4807 if (EXT4_GOOD_OLD_INODE_SIZE
+ ei
->i_extra_isize
>
4808 EXT4_INODE_SIZE(inode
->i_sb
)) {
4812 if (ei
->i_extra_isize
== 0) {
4813 /* The extra space is currently unused. Use it. */
4814 ei
->i_extra_isize
= sizeof(struct ext4_inode
) -
4815 EXT4_GOOD_OLD_INODE_SIZE
;
4817 __le32
*magic
= (void *)raw_inode
+
4818 EXT4_GOOD_OLD_INODE_SIZE
+
4820 if (*magic
== cpu_to_le32(EXT4_XATTR_MAGIC
))
4821 ei
->i_state
|= EXT4_STATE_XATTR
;
4824 ei
->i_extra_isize
= 0;
4826 EXT4_INODE_GET_XTIME(i_ctime
, inode
, raw_inode
);
4827 EXT4_INODE_GET_XTIME(i_mtime
, inode
, raw_inode
);
4828 EXT4_INODE_GET_XTIME(i_atime
, inode
, raw_inode
);
4829 EXT4_EINODE_GET_XTIME(i_crtime
, ei
, raw_inode
);
4831 inode
->i_version
= le32_to_cpu(raw_inode
->i_disk_version
);
4832 if (EXT4_INODE_SIZE(inode
->i_sb
) > EXT4_GOOD_OLD_INODE_SIZE
) {
4833 if (EXT4_FITS_IN_INODE(raw_inode
, ei
, i_version_hi
))
4835 (__u64
)(le32_to_cpu(raw_inode
->i_version_hi
)) << 32;
4839 if (ei
->i_file_acl
&&
4840 !ext4_data_block_valid(EXT4_SB(sb
), ei
->i_file_acl
, 1)) {
4841 ext4_error(sb
, __func__
,
4842 "bad extended attribute block %llu in inode #%lu",
4843 ei
->i_file_acl
, inode
->i_ino
);
4846 } else if (ei
->i_flags
& EXT4_EXTENTS_FL
) {
4847 if (S_ISREG(inode
->i_mode
) || S_ISDIR(inode
->i_mode
) ||
4848 (S_ISLNK(inode
->i_mode
) &&
4849 !ext4_inode_is_fast_symlink(inode
)))
4850 /* Validate extent which is part of inode */
4851 ret
= ext4_ext_check_inode(inode
);
4852 } else if (S_ISREG(inode
->i_mode
) || S_ISDIR(inode
->i_mode
) ||
4853 (S_ISLNK(inode
->i_mode
) &&
4854 !ext4_inode_is_fast_symlink(inode
))) {
4855 /* Validate block references which are part of inode */
4856 ret
= ext4_check_inode_blockref(inode
);
4861 if (S_ISREG(inode
->i_mode
)) {
4862 inode
->i_op
= &ext4_file_inode_operations
;
4863 inode
->i_fop
= &ext4_file_operations
;
4864 ext4_set_aops(inode
);
4865 } else if (S_ISDIR(inode
->i_mode
)) {
4866 inode
->i_op
= &ext4_dir_inode_operations
;
4867 inode
->i_fop
= &ext4_dir_operations
;
4868 } else if (S_ISLNK(inode
->i_mode
)) {
4869 if (ext4_inode_is_fast_symlink(inode
)) {
4870 inode
->i_op
= &ext4_fast_symlink_inode_operations
;
4871 nd_terminate_link(ei
->i_data
, inode
->i_size
,
4872 sizeof(ei
->i_data
) - 1);
4874 inode
->i_op
= &ext4_symlink_inode_operations
;
4875 ext4_set_aops(inode
);
4877 } else if (S_ISCHR(inode
->i_mode
) || S_ISBLK(inode
->i_mode
) ||
4878 S_ISFIFO(inode
->i_mode
) || S_ISSOCK(inode
->i_mode
)) {
4879 inode
->i_op
= &ext4_special_inode_operations
;
4880 if (raw_inode
->i_block
[0])
4881 init_special_inode(inode
, inode
->i_mode
,
4882 old_decode_dev(le32_to_cpu(raw_inode
->i_block
[0])));
4884 init_special_inode(inode
, inode
->i_mode
,
4885 new_decode_dev(le32_to_cpu(raw_inode
->i_block
[1])));
4888 ext4_error(inode
->i_sb
, __func__
,
4889 "bogus i_mode (%o) for inode=%lu",
4890 inode
->i_mode
, inode
->i_ino
);
4894 ext4_set_inode_flags(inode
);
4895 unlock_new_inode(inode
);
4901 return ERR_PTR(ret
);
4904 static int ext4_inode_blocks_set(handle_t
*handle
,
4905 struct ext4_inode
*raw_inode
,
4906 struct ext4_inode_info
*ei
)
4908 struct inode
*inode
= &(ei
->vfs_inode
);
4909 u64 i_blocks
= inode
->i_blocks
;
4910 struct super_block
*sb
= inode
->i_sb
;
4912 if (i_blocks
<= ~0U) {
4914 * i_blocks can be represnted in a 32 bit variable
4915 * as multiple of 512 bytes
4917 raw_inode
->i_blocks_lo
= cpu_to_le32(i_blocks
);
4918 raw_inode
->i_blocks_high
= 0;
4919 ei
->i_flags
&= ~EXT4_HUGE_FILE_FL
;
4922 if (!EXT4_HAS_RO_COMPAT_FEATURE(sb
, EXT4_FEATURE_RO_COMPAT_HUGE_FILE
))
4925 if (i_blocks
<= 0xffffffffffffULL
) {
4927 * i_blocks can be represented in a 48 bit variable
4928 * as multiple of 512 bytes
4930 raw_inode
->i_blocks_lo
= cpu_to_le32(i_blocks
);
4931 raw_inode
->i_blocks_high
= cpu_to_le16(i_blocks
>> 32);
4932 ei
->i_flags
&= ~EXT4_HUGE_FILE_FL
;
4934 ei
->i_flags
|= EXT4_HUGE_FILE_FL
;
4935 /* i_block is stored in file system block size */
4936 i_blocks
= i_blocks
>> (inode
->i_blkbits
- 9);
4937 raw_inode
->i_blocks_lo
= cpu_to_le32(i_blocks
);
4938 raw_inode
->i_blocks_high
= cpu_to_le16(i_blocks
>> 32);
4944 * Post the struct inode info into an on-disk inode location in the
4945 * buffer-cache. This gobbles the caller's reference to the
4946 * buffer_head in the inode location struct.
4948 * The caller must have write access to iloc->bh.
4950 static int ext4_do_update_inode(handle_t
*handle
,
4951 struct inode
*inode
,
4952 struct ext4_iloc
*iloc
)
4954 struct ext4_inode
*raw_inode
= ext4_raw_inode(iloc
);
4955 struct ext4_inode_info
*ei
= EXT4_I(inode
);
4956 struct buffer_head
*bh
= iloc
->bh
;
4957 int err
= 0, rc
, block
;
4959 /* For fields not not tracking in the in-memory inode,
4960 * initialise them to zero for new inodes. */
4961 if (ei
->i_state
& EXT4_STATE_NEW
)
4962 memset(raw_inode
, 0, EXT4_SB(inode
->i_sb
)->s_inode_size
);
4964 ext4_get_inode_flags(ei
);
4965 raw_inode
->i_mode
= cpu_to_le16(inode
->i_mode
);
4966 if (!(test_opt(inode
->i_sb
, NO_UID32
))) {
4967 raw_inode
->i_uid_low
= cpu_to_le16(low_16_bits(inode
->i_uid
));
4968 raw_inode
->i_gid_low
= cpu_to_le16(low_16_bits(inode
->i_gid
));
4970 * Fix up interoperability with old kernels. Otherwise, old inodes get
4971 * re-used with the upper 16 bits of the uid/gid intact
4974 raw_inode
->i_uid_high
=
4975 cpu_to_le16(high_16_bits(inode
->i_uid
));
4976 raw_inode
->i_gid_high
=
4977 cpu_to_le16(high_16_bits(inode
->i_gid
));
4979 raw_inode
->i_uid_high
= 0;
4980 raw_inode
->i_gid_high
= 0;
4983 raw_inode
->i_uid_low
=
4984 cpu_to_le16(fs_high2lowuid(inode
->i_uid
));
4985 raw_inode
->i_gid_low
=
4986 cpu_to_le16(fs_high2lowgid(inode
->i_gid
));
4987 raw_inode
->i_uid_high
= 0;
4988 raw_inode
->i_gid_high
= 0;
4990 raw_inode
->i_links_count
= cpu_to_le16(inode
->i_nlink
);
4992 EXT4_INODE_SET_XTIME(i_ctime
, inode
, raw_inode
);
4993 EXT4_INODE_SET_XTIME(i_mtime
, inode
, raw_inode
);
4994 EXT4_INODE_SET_XTIME(i_atime
, inode
, raw_inode
);
4995 EXT4_EINODE_SET_XTIME(i_crtime
, ei
, raw_inode
);
4997 if (ext4_inode_blocks_set(handle
, raw_inode
, ei
))
4999 raw_inode
->i_dtime
= cpu_to_le32(ei
->i_dtime
);
5000 raw_inode
->i_flags
= cpu_to_le32(ei
->i_flags
);
5001 if (EXT4_SB(inode
->i_sb
)->s_es
->s_creator_os
!=
5002 cpu_to_le32(EXT4_OS_HURD
))
5003 raw_inode
->i_file_acl_high
=
5004 cpu_to_le16(ei
->i_file_acl
>> 32);
5005 raw_inode
->i_file_acl_lo
= cpu_to_le32(ei
->i_file_acl
);
5006 ext4_isize_set(raw_inode
, ei
->i_disksize
);
5007 if (ei
->i_disksize
> 0x7fffffffULL
) {
5008 struct super_block
*sb
= inode
->i_sb
;
5009 if (!EXT4_HAS_RO_COMPAT_FEATURE(sb
,
5010 EXT4_FEATURE_RO_COMPAT_LARGE_FILE
) ||
5011 EXT4_SB(sb
)->s_es
->s_rev_level
==
5012 cpu_to_le32(EXT4_GOOD_OLD_REV
)) {
5013 /* If this is the first large file
5014 * created, add a flag to the superblock.
5016 err
= ext4_journal_get_write_access(handle
,
5017 EXT4_SB(sb
)->s_sbh
);
5020 ext4_update_dynamic_rev(sb
);
5021 EXT4_SET_RO_COMPAT_FEATURE(sb
,
5022 EXT4_FEATURE_RO_COMPAT_LARGE_FILE
);
5024 ext4_handle_sync(handle
);
5025 err
= ext4_handle_dirty_metadata(handle
, inode
,
5026 EXT4_SB(sb
)->s_sbh
);
5029 raw_inode
->i_generation
= cpu_to_le32(inode
->i_generation
);
5030 if (S_ISCHR(inode
->i_mode
) || S_ISBLK(inode
->i_mode
)) {
5031 if (old_valid_dev(inode
->i_rdev
)) {
5032 raw_inode
->i_block
[0] =
5033 cpu_to_le32(old_encode_dev(inode
->i_rdev
));
5034 raw_inode
->i_block
[1] = 0;
5036 raw_inode
->i_block
[0] = 0;
5037 raw_inode
->i_block
[1] =
5038 cpu_to_le32(new_encode_dev(inode
->i_rdev
));
5039 raw_inode
->i_block
[2] = 0;
5042 for (block
= 0; block
< EXT4_N_BLOCKS
; block
++)
5043 raw_inode
->i_block
[block
] = ei
->i_data
[block
];
5045 raw_inode
->i_disk_version
= cpu_to_le32(inode
->i_version
);
5046 if (ei
->i_extra_isize
) {
5047 if (EXT4_FITS_IN_INODE(raw_inode
, ei
, i_version_hi
))
5048 raw_inode
->i_version_hi
=
5049 cpu_to_le32(inode
->i_version
>> 32);
5050 raw_inode
->i_extra_isize
= cpu_to_le16(ei
->i_extra_isize
);
5053 BUFFER_TRACE(bh
, "call ext4_handle_dirty_metadata");
5054 rc
= ext4_handle_dirty_metadata(handle
, inode
, bh
);
5057 ei
->i_state
&= ~EXT4_STATE_NEW
;
5061 ext4_std_error(inode
->i_sb
, err
);
5066 * ext4_write_inode()
5068 * We are called from a few places:
5070 * - Within generic_file_write() for O_SYNC files.
5071 * Here, there will be no transaction running. We wait for any running
5072 * trasnaction to commit.
5074 * - Within sys_sync(), kupdate and such.
5075 * We wait on commit, if tol to.
5077 * - Within prune_icache() (PF_MEMALLOC == true)
5078 * Here we simply return. We can't afford to block kswapd on the
5081 * In all cases it is actually safe for us to return without doing anything,
5082 * because the inode has been copied into a raw inode buffer in
5083 * ext4_mark_inode_dirty(). This is a correctness thing for O_SYNC and for
5086 * Note that we are absolutely dependent upon all inode dirtiers doing the
5087 * right thing: they *must* call mark_inode_dirty() after dirtying info in
5088 * which we are interested.
5090 * It would be a bug for them to not do this. The code:
5092 * mark_inode_dirty(inode)
5094 * inode->i_size = expr;
5096 * is in error because a kswapd-driven write_inode() could occur while
5097 * `stuff()' is running, and the new i_size will be lost. Plus the inode
5098 * will no longer be on the superblock's dirty inode list.
5100 int ext4_write_inode(struct inode
*inode
, int wait
)
5104 if (current
->flags
& PF_MEMALLOC
)
5107 if (EXT4_SB(inode
->i_sb
)->s_journal
) {
5108 if (ext4_journal_current_handle()) {
5109 jbd_debug(1, "called recursively, non-PF_MEMALLOC!\n");
5117 err
= ext4_force_commit(inode
->i_sb
);
5119 struct ext4_iloc iloc
;
5121 err
= ext4_get_inode_loc(inode
, &iloc
);
5125 sync_dirty_buffer(iloc
.bh
);
5126 if (buffer_req(iloc
.bh
) && !buffer_uptodate(iloc
.bh
)) {
5127 ext4_error(inode
->i_sb
, __func__
,
5128 "IO error syncing inode, "
5129 "inode=%lu, block=%llu",
5131 (unsigned long long)iloc
.bh
->b_blocknr
);
5141 * Called from notify_change.
5143 * We want to trap VFS attempts to truncate the file as soon as
5144 * possible. In particular, we want to make sure that when the VFS
5145 * shrinks i_size, we put the inode on the orphan list and modify
5146 * i_disksize immediately, so that during the subsequent flushing of
5147 * dirty pages and freeing of disk blocks, we can guarantee that any
5148 * commit will leave the blocks being flushed in an unused state on
5149 * disk. (On recovery, the inode will get truncated and the blocks will
5150 * be freed, so we have a strong guarantee that no future commit will
5151 * leave these blocks visible to the user.)
5153 * Another thing we have to assure is that if we are in ordered mode
5154 * and inode is still attached to the committing transaction, we must
5155 * we start writeout of all the dirty pages which are being truncated.
5156 * This way we are sure that all the data written in the previous
5157 * transaction are already on disk (truncate waits for pages under
5160 * Called with inode->i_mutex down.
5162 int ext4_setattr(struct dentry
*dentry
, struct iattr
*attr
)
5164 struct inode
*inode
= dentry
->d_inode
;
5166 const unsigned int ia_valid
= attr
->ia_valid
;
5168 error
= inode_change_ok(inode
, attr
);
5172 if ((ia_valid
& ATTR_UID
&& attr
->ia_uid
!= inode
->i_uid
) ||
5173 (ia_valid
& ATTR_GID
&& attr
->ia_gid
!= inode
->i_gid
)) {
5176 /* (user+group)*(old+new) structure, inode write (sb,
5177 * inode block, ? - but truncate inode update has it) */
5178 handle
= ext4_journal_start(inode
, (EXT4_MAXQUOTAS_INIT_BLOCKS(inode
->i_sb
)+
5179 EXT4_MAXQUOTAS_DEL_BLOCKS(inode
->i_sb
))+3);
5180 if (IS_ERR(handle
)) {
5181 error
= PTR_ERR(handle
);
5184 error
= vfs_dq_transfer(inode
, attr
) ? -EDQUOT
: 0;
5186 ext4_journal_stop(handle
);
5189 /* Update corresponding info in inode so that everything is in
5190 * one transaction */
5191 if (attr
->ia_valid
& ATTR_UID
)
5192 inode
->i_uid
= attr
->ia_uid
;
5193 if (attr
->ia_valid
& ATTR_GID
)
5194 inode
->i_gid
= attr
->ia_gid
;
5195 error
= ext4_mark_inode_dirty(handle
, inode
);
5196 ext4_journal_stop(handle
);
5199 if (attr
->ia_valid
& ATTR_SIZE
) {
5200 if (!(EXT4_I(inode
)->i_flags
& EXT4_EXTENTS_FL
)) {
5201 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
5203 if (attr
->ia_size
> sbi
->s_bitmap_maxbytes
) {
5210 if (S_ISREG(inode
->i_mode
) &&
5211 attr
->ia_valid
& ATTR_SIZE
&& attr
->ia_size
< inode
->i_size
) {
5214 handle
= ext4_journal_start(inode
, 3);
5215 if (IS_ERR(handle
)) {
5216 error
= PTR_ERR(handle
);
5220 error
= ext4_orphan_add(handle
, inode
);
5221 EXT4_I(inode
)->i_disksize
= attr
->ia_size
;
5222 rc
= ext4_mark_inode_dirty(handle
, inode
);
5225 ext4_journal_stop(handle
);
5227 if (ext4_should_order_data(inode
)) {
5228 error
= ext4_begin_ordered_truncate(inode
,
5231 /* Do as much error cleanup as possible */
5232 handle
= ext4_journal_start(inode
, 3);
5233 if (IS_ERR(handle
)) {
5234 ext4_orphan_del(NULL
, inode
);
5237 ext4_orphan_del(handle
, inode
);
5238 ext4_journal_stop(handle
);
5244 rc
= inode_setattr(inode
, attr
);
5246 /* If inode_setattr's call to ext4_truncate failed to get a
5247 * transaction handle at all, we need to clean up the in-core
5248 * orphan list manually. */
5250 ext4_orphan_del(NULL
, inode
);
5252 if (!rc
&& (ia_valid
& ATTR_MODE
))
5253 rc
= ext4_acl_chmod(inode
);
5256 ext4_std_error(inode
->i_sb
, error
);
5262 int ext4_getattr(struct vfsmount
*mnt
, struct dentry
*dentry
,
5265 struct inode
*inode
;
5266 unsigned long delalloc_blocks
;
5268 inode
= dentry
->d_inode
;
5269 generic_fillattr(inode
, stat
);
5272 * We can't update i_blocks if the block allocation is delayed
5273 * otherwise in the case of system crash before the real block
5274 * allocation is done, we will have i_blocks inconsistent with
5275 * on-disk file blocks.
5276 * We always keep i_blocks updated together with real
5277 * allocation. But to not confuse with user, stat
5278 * will return the blocks that include the delayed allocation
5279 * blocks for this file.
5281 spin_lock(&EXT4_I(inode
)->i_block_reservation_lock
);
5282 delalloc_blocks
= EXT4_I(inode
)->i_reserved_data_blocks
;
5283 spin_unlock(&EXT4_I(inode
)->i_block_reservation_lock
);
5285 stat
->blocks
+= (delalloc_blocks
<< inode
->i_sb
->s_blocksize_bits
)>>9;
5289 static int ext4_indirect_trans_blocks(struct inode
*inode
, int nrblocks
,
5294 /* if nrblocks are contiguous */
5297 * With N contiguous data blocks, it need at most
5298 * N/EXT4_ADDR_PER_BLOCK(inode->i_sb) indirect blocks
5299 * 2 dindirect blocks
5302 indirects
= nrblocks
/ EXT4_ADDR_PER_BLOCK(inode
->i_sb
);
5303 return indirects
+ 3;
5306 * if nrblocks are not contiguous, worse case, each block touch
5307 * a indirect block, and each indirect block touch a double indirect
5308 * block, plus a triple indirect block
5310 indirects
= nrblocks
* 2 + 1;
5314 static int ext4_index_trans_blocks(struct inode
*inode
, int nrblocks
, int chunk
)
5316 if (!(EXT4_I(inode
)->i_flags
& EXT4_EXTENTS_FL
))
5317 return ext4_indirect_trans_blocks(inode
, nrblocks
, chunk
);
5318 return ext4_ext_index_trans_blocks(inode
, nrblocks
, chunk
);
5322 * Account for index blocks, block groups bitmaps and block group
5323 * descriptor blocks if modify datablocks and index blocks
5324 * worse case, the indexs blocks spread over different block groups
5326 * If datablocks are discontiguous, they are possible to spread over
5327 * different block groups too. If they are contiugous, with flexbg,
5328 * they could still across block group boundary.
5330 * Also account for superblock, inode, quota and xattr blocks
5332 int ext4_meta_trans_blocks(struct inode
*inode
, int nrblocks
, int chunk
)
5334 ext4_group_t groups
, ngroups
= ext4_get_groups_count(inode
->i_sb
);
5340 * How many index blocks need to touch to modify nrblocks?
5341 * The "Chunk" flag indicating whether the nrblocks is
5342 * physically contiguous on disk
5344 * For Direct IO and fallocate, they calls get_block to allocate
5345 * one single extent at a time, so they could set the "Chunk" flag
5347 idxblocks
= ext4_index_trans_blocks(inode
, nrblocks
, chunk
);
5352 * Now let's see how many group bitmaps and group descriptors need
5362 if (groups
> ngroups
)
5364 if (groups
> EXT4_SB(inode
->i_sb
)->s_gdb_count
)
5365 gdpblocks
= EXT4_SB(inode
->i_sb
)->s_gdb_count
;
5367 /* bitmaps and block group descriptor blocks */
5368 ret
+= groups
+ gdpblocks
;
5370 /* Blocks for super block, inode, quota and xattr blocks */
5371 ret
+= EXT4_META_TRANS_BLOCKS(inode
->i_sb
);
5377 * Calulate the total number of credits to reserve to fit
5378 * the modification of a single pages into a single transaction,
5379 * which may include multiple chunks of block allocations.
5381 * This could be called via ext4_write_begin()
5383 * We need to consider the worse case, when
5384 * one new block per extent.
5386 int ext4_writepage_trans_blocks(struct inode
*inode
)
5388 int bpp
= ext4_journal_blocks_per_page(inode
);
5391 ret
= ext4_meta_trans_blocks(inode
, bpp
, 0);
5393 /* Account for data blocks for journalled mode */
5394 if (ext4_should_journal_data(inode
))
5400 * Calculate the journal credits for a chunk of data modification.
5402 * This is called from DIO, fallocate or whoever calling
5403 * ext4_get_blocks() to map/allocate a chunk of contigous disk blocks.
5405 * journal buffers for data blocks are not included here, as DIO
5406 * and fallocate do no need to journal data buffers.
5408 int ext4_chunk_trans_blocks(struct inode
*inode
, int nrblocks
)
5410 return ext4_meta_trans_blocks(inode
, nrblocks
, 1);
5414 * The caller must have previously called ext4_reserve_inode_write().
5415 * Give this, we know that the caller already has write access to iloc->bh.
5417 int ext4_mark_iloc_dirty(handle_t
*handle
,
5418 struct inode
*inode
, struct ext4_iloc
*iloc
)
5422 if (test_opt(inode
->i_sb
, I_VERSION
))
5423 inode_inc_iversion(inode
);
5425 /* the do_update_inode consumes one bh->b_count */
5428 /* ext4_do_update_inode() does jbd2_journal_dirty_metadata */
5429 err
= ext4_do_update_inode(handle
, inode
, iloc
);
5435 * On success, We end up with an outstanding reference count against
5436 * iloc->bh. This _must_ be cleaned up later.
5440 ext4_reserve_inode_write(handle_t
*handle
, struct inode
*inode
,
5441 struct ext4_iloc
*iloc
)
5445 err
= ext4_get_inode_loc(inode
, iloc
);
5447 BUFFER_TRACE(iloc
->bh
, "get_write_access");
5448 err
= ext4_journal_get_write_access(handle
, iloc
->bh
);
5454 ext4_std_error(inode
->i_sb
, err
);
5459 * Expand an inode by new_extra_isize bytes.
5460 * Returns 0 on success or negative error number on failure.
5462 static int ext4_expand_extra_isize(struct inode
*inode
,
5463 unsigned int new_extra_isize
,
5464 struct ext4_iloc iloc
,
5467 struct ext4_inode
*raw_inode
;
5468 struct ext4_xattr_ibody_header
*header
;
5469 struct ext4_xattr_entry
*entry
;
5471 if (EXT4_I(inode
)->i_extra_isize
>= new_extra_isize
)
5474 raw_inode
= ext4_raw_inode(&iloc
);
5476 header
= IHDR(inode
, raw_inode
);
5477 entry
= IFIRST(header
);
5479 /* No extended attributes present */
5480 if (!(EXT4_I(inode
)->i_state
& EXT4_STATE_XATTR
) ||
5481 header
->h_magic
!= cpu_to_le32(EXT4_XATTR_MAGIC
)) {
5482 memset((void *)raw_inode
+ EXT4_GOOD_OLD_INODE_SIZE
, 0,
5484 EXT4_I(inode
)->i_extra_isize
= new_extra_isize
;
5488 /* try to expand with EAs present */
5489 return ext4_expand_extra_isize_ea(inode
, new_extra_isize
,
5494 * What we do here is to mark the in-core inode as clean with respect to inode
5495 * dirtiness (it may still be data-dirty).
5496 * This means that the in-core inode may be reaped by prune_icache
5497 * without having to perform any I/O. This is a very good thing,
5498 * because *any* task may call prune_icache - even ones which
5499 * have a transaction open against a different journal.
5501 * Is this cheating? Not really. Sure, we haven't written the
5502 * inode out, but prune_icache isn't a user-visible syncing function.
5503 * Whenever the user wants stuff synced (sys_sync, sys_msync, sys_fsync)
5504 * we start and wait on commits.
5506 * Is this efficient/effective? Well, we're being nice to the system
5507 * by cleaning up our inodes proactively so they can be reaped
5508 * without I/O. But we are potentially leaving up to five seconds'
5509 * worth of inodes floating about which prune_icache wants us to
5510 * write out. One way to fix that would be to get prune_icache()
5511 * to do a write_super() to free up some memory. It has the desired
5514 int ext4_mark_inode_dirty(handle_t
*handle
, struct inode
*inode
)
5516 struct ext4_iloc iloc
;
5517 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
5518 static unsigned int mnt_count
;
5522 err
= ext4_reserve_inode_write(handle
, inode
, &iloc
);
5523 if (ext4_handle_valid(handle
) &&
5524 EXT4_I(inode
)->i_extra_isize
< sbi
->s_want_extra_isize
&&
5525 !(EXT4_I(inode
)->i_state
& EXT4_STATE_NO_EXPAND
)) {
5527 * We need extra buffer credits since we may write into EA block
5528 * with this same handle. If journal_extend fails, then it will
5529 * only result in a minor loss of functionality for that inode.
5530 * If this is felt to be critical, then e2fsck should be run to
5531 * force a large enough s_min_extra_isize.
5533 if ((jbd2_journal_extend(handle
,
5534 EXT4_DATA_TRANS_BLOCKS(inode
->i_sb
))) == 0) {
5535 ret
= ext4_expand_extra_isize(inode
,
5536 sbi
->s_want_extra_isize
,
5539 EXT4_I(inode
)->i_state
|= EXT4_STATE_NO_EXPAND
;
5541 le16_to_cpu(sbi
->s_es
->s_mnt_count
)) {
5542 ext4_warning(inode
->i_sb
, __func__
,
5543 "Unable to expand inode %lu. Delete"
5544 " some EAs or run e2fsck.",
5547 le16_to_cpu(sbi
->s_es
->s_mnt_count
);
5553 err
= ext4_mark_iloc_dirty(handle
, inode
, &iloc
);
5558 * ext4_dirty_inode() is called from __mark_inode_dirty()
5560 * We're really interested in the case where a file is being extended.
5561 * i_size has been changed by generic_commit_write() and we thus need
5562 * to include the updated inode in the current transaction.
5564 * Also, vfs_dq_alloc_block() will always dirty the inode when blocks
5565 * are allocated to the file.
5567 * If the inode is marked synchronous, we don't honour that here - doing
5568 * so would cause a commit on atime updates, which we don't bother doing.
5569 * We handle synchronous inodes at the highest possible level.
5571 void ext4_dirty_inode(struct inode
*inode
)
5575 handle
= ext4_journal_start(inode
, 2);
5579 ext4_mark_inode_dirty(handle
, inode
);
5581 ext4_journal_stop(handle
);
5588 * Bind an inode's backing buffer_head into this transaction, to prevent
5589 * it from being flushed to disk early. Unlike
5590 * ext4_reserve_inode_write, this leaves behind no bh reference and
5591 * returns no iloc structure, so the caller needs to repeat the iloc
5592 * lookup to mark the inode dirty later.
5594 static int ext4_pin_inode(handle_t
*handle
, struct inode
*inode
)
5596 struct ext4_iloc iloc
;
5600 err
= ext4_get_inode_loc(inode
, &iloc
);
5602 BUFFER_TRACE(iloc
.bh
, "get_write_access");
5603 err
= jbd2_journal_get_write_access(handle
, iloc
.bh
);
5605 err
= ext4_handle_dirty_metadata(handle
,
5611 ext4_std_error(inode
->i_sb
, err
);
5616 int ext4_change_inode_journal_flag(struct inode
*inode
, int val
)
5623 * We have to be very careful here: changing a data block's
5624 * journaling status dynamically is dangerous. If we write a
5625 * data block to the journal, change the status and then delete
5626 * that block, we risk forgetting to revoke the old log record
5627 * from the journal and so a subsequent replay can corrupt data.
5628 * So, first we make sure that the journal is empty and that
5629 * nobody is changing anything.
5632 journal
= EXT4_JOURNAL(inode
);
5635 if (is_journal_aborted(journal
))
5638 jbd2_journal_lock_updates(journal
);
5639 jbd2_journal_flush(journal
);
5642 * OK, there are no updates running now, and all cached data is
5643 * synced to disk. We are now in a completely consistent state
5644 * which doesn't have anything in the journal, and we know that
5645 * no filesystem updates are running, so it is safe to modify
5646 * the inode's in-core data-journaling state flag now.
5650 EXT4_I(inode
)->i_flags
|= EXT4_JOURNAL_DATA_FL
;
5652 EXT4_I(inode
)->i_flags
&= ~EXT4_JOURNAL_DATA_FL
;
5653 ext4_set_aops(inode
);
5655 jbd2_journal_unlock_updates(journal
);
5657 /* Finally we can mark the inode as dirty. */
5659 handle
= ext4_journal_start(inode
, 1);
5661 return PTR_ERR(handle
);
5663 err
= ext4_mark_inode_dirty(handle
, inode
);
5664 ext4_handle_sync(handle
);
5665 ext4_journal_stop(handle
);
5666 ext4_std_error(inode
->i_sb
, err
);
5671 static int ext4_bh_unmapped(handle_t
*handle
, struct buffer_head
*bh
)
5673 return !buffer_mapped(bh
);
5676 int ext4_page_mkwrite(struct vm_area_struct
*vma
, struct vm_fault
*vmf
)
5678 struct page
*page
= vmf
->page
;
5683 struct file
*file
= vma
->vm_file
;
5684 struct inode
*inode
= file
->f_path
.dentry
->d_inode
;
5685 struct address_space
*mapping
= inode
->i_mapping
;
5688 * Get i_alloc_sem to stop truncates messing with the inode. We cannot
5689 * get i_mutex because we are already holding mmap_sem.
5691 down_read(&inode
->i_alloc_sem
);
5692 size
= i_size_read(inode
);
5693 if (page
->mapping
!= mapping
|| size
<= page_offset(page
)
5694 || !PageUptodate(page
)) {
5695 /* page got truncated from under us? */
5699 if (PageMappedToDisk(page
))
5702 if (page
->index
== size
>> PAGE_CACHE_SHIFT
)
5703 len
= size
& ~PAGE_CACHE_MASK
;
5705 len
= PAGE_CACHE_SIZE
;
5709 * return if we have all the buffers mapped. This avoid
5710 * the need to call write_begin/write_end which does a
5711 * journal_start/journal_stop which can block and take
5714 if (page_has_buffers(page
)) {
5715 if (!walk_page_buffers(NULL
, page_buffers(page
), 0, len
, NULL
,
5716 ext4_bh_unmapped
)) {
5723 * OK, we need to fill the hole... Do write_begin write_end
5724 * to do block allocation/reservation.We are not holding
5725 * inode.i__mutex here. That allow * parallel write_begin,
5726 * write_end call. lock_page prevent this from happening
5727 * on the same page though
5729 ret
= mapping
->a_ops
->write_begin(file
, mapping
, page_offset(page
),
5730 len
, AOP_FLAG_UNINTERRUPTIBLE
, &page
, &fsdata
);
5733 ret
= mapping
->a_ops
->write_end(file
, mapping
, page_offset(page
),
5734 len
, len
, page
, fsdata
);
5740 ret
= VM_FAULT_SIGBUS
;
5741 up_read(&inode
->i_alloc_sem
);