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
);
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
);
1495 static int ext4_write_begin(struct file
*file
, struct address_space
*mapping
,
1496 loff_t pos
, unsigned len
, unsigned flags
,
1497 struct page
**pagep
, void **fsdata
)
1499 struct inode
*inode
= mapping
->host
;
1500 int ret
, needed_blocks
;
1507 trace_ext4_write_begin(inode
, pos
, len
, flags
);
1509 * Reserve one block more for addition to orphan list in case
1510 * we allocate blocks but write fails for some reason
1512 needed_blocks
= ext4_writepage_trans_blocks(inode
) + 1;
1513 index
= pos
>> PAGE_CACHE_SHIFT
;
1514 from
= pos
& (PAGE_CACHE_SIZE
- 1);
1518 handle
= ext4_journal_start(inode
, needed_blocks
);
1519 if (IS_ERR(handle
)) {
1520 ret
= PTR_ERR(handle
);
1524 /* We cannot recurse into the filesystem as the transaction is already
1526 flags
|= AOP_FLAG_NOFS
;
1528 page
= grab_cache_page_write_begin(mapping
, index
, flags
);
1530 ext4_journal_stop(handle
);
1536 ret
= block_write_begin(file
, mapping
, pos
, len
, flags
, pagep
, fsdata
,
1539 if (!ret
&& ext4_should_journal_data(inode
)) {
1540 ret
= walk_page_buffers(handle
, page_buffers(page
),
1541 from
, to
, NULL
, do_journal_get_write_access
);
1546 page_cache_release(page
);
1548 * block_write_begin may have instantiated a few blocks
1549 * outside i_size. Trim these off again. Don't need
1550 * i_size_read because we hold i_mutex.
1552 * Add inode to orphan list in case we crash before
1555 if (pos
+ len
> inode
->i_size
&& ext4_can_truncate(inode
))
1556 ext4_orphan_add(handle
, inode
);
1558 ext4_journal_stop(handle
);
1559 if (pos
+ len
> inode
->i_size
) {
1560 ext4_truncate(inode
);
1562 * If truncate failed early the inode might
1563 * still be on the orphan list; we need to
1564 * make sure the inode is removed from the
1565 * orphan list in that case.
1568 ext4_orphan_del(NULL
, inode
);
1572 if (ret
== -ENOSPC
&& ext4_should_retry_alloc(inode
->i_sb
, &retries
))
1578 /* For write_end() in data=journal mode */
1579 static int write_end_fn(handle_t
*handle
, struct buffer_head
*bh
)
1581 if (!buffer_mapped(bh
) || buffer_freed(bh
))
1583 set_buffer_uptodate(bh
);
1584 return ext4_handle_dirty_metadata(handle
, NULL
, bh
);
1587 static int ext4_generic_write_end(struct file
*file
,
1588 struct address_space
*mapping
,
1589 loff_t pos
, unsigned len
, unsigned copied
,
1590 struct page
*page
, void *fsdata
)
1592 int i_size_changed
= 0;
1593 struct inode
*inode
= mapping
->host
;
1594 handle_t
*handle
= ext4_journal_current_handle();
1596 copied
= block_write_end(file
, mapping
, pos
, len
, copied
, page
, fsdata
);
1599 * No need to use i_size_read() here, the i_size
1600 * cannot change under us because we hold i_mutex.
1602 * But it's important to update i_size while still holding page lock:
1603 * page writeout could otherwise come in and zero beyond i_size.
1605 if (pos
+ copied
> inode
->i_size
) {
1606 i_size_write(inode
, pos
+ copied
);
1610 if (pos
+ copied
> EXT4_I(inode
)->i_disksize
) {
1611 /* We need to mark inode dirty even if
1612 * new_i_size is less that inode->i_size
1613 * bu greater than i_disksize.(hint delalloc)
1615 ext4_update_i_disksize(inode
, (pos
+ copied
));
1619 page_cache_release(page
);
1622 * Don't mark the inode dirty under page lock. First, it unnecessarily
1623 * makes the holding time of page lock longer. Second, it forces lock
1624 * ordering of page lock and transaction start for journaling
1628 ext4_mark_inode_dirty(handle
, inode
);
1634 * We need to pick up the new inode size which generic_commit_write gave us
1635 * `file' can be NULL - eg, when called from page_symlink().
1637 * ext4 never places buffers on inode->i_mapping->private_list. metadata
1638 * buffers are managed internally.
1640 static int ext4_ordered_write_end(struct file
*file
,
1641 struct address_space
*mapping
,
1642 loff_t pos
, unsigned len
, unsigned copied
,
1643 struct page
*page
, void *fsdata
)
1645 handle_t
*handle
= ext4_journal_current_handle();
1646 struct inode
*inode
= mapping
->host
;
1649 trace_ext4_ordered_write_end(inode
, pos
, len
, copied
);
1650 ret
= ext4_jbd2_file_inode(handle
, inode
);
1653 ret2
= ext4_generic_write_end(file
, mapping
, pos
, len
, copied
,
1656 if (pos
+ len
> inode
->i_size
&& ext4_can_truncate(inode
))
1657 /* if we have allocated more blocks and copied
1658 * less. We will have blocks allocated outside
1659 * inode->i_size. So truncate them
1661 ext4_orphan_add(handle
, inode
);
1665 ret2
= ext4_journal_stop(handle
);
1669 if (pos
+ len
> inode
->i_size
) {
1670 ext4_truncate(inode
);
1672 * If truncate failed early the inode might still be
1673 * on the orphan list; we need to make sure the inode
1674 * is removed from the orphan list in that case.
1677 ext4_orphan_del(NULL
, inode
);
1681 return ret
? ret
: copied
;
1684 static int ext4_writeback_write_end(struct file
*file
,
1685 struct address_space
*mapping
,
1686 loff_t pos
, unsigned len
, unsigned copied
,
1687 struct page
*page
, void *fsdata
)
1689 handle_t
*handle
= ext4_journal_current_handle();
1690 struct inode
*inode
= mapping
->host
;
1693 trace_ext4_writeback_write_end(inode
, pos
, len
, copied
);
1694 ret2
= ext4_generic_write_end(file
, mapping
, pos
, len
, copied
,
1697 if (pos
+ len
> inode
->i_size
&& ext4_can_truncate(inode
))
1698 /* if we have allocated more blocks and copied
1699 * less. We will have blocks allocated outside
1700 * inode->i_size. So truncate them
1702 ext4_orphan_add(handle
, inode
);
1707 ret2
= ext4_journal_stop(handle
);
1711 if (pos
+ len
> inode
->i_size
) {
1712 ext4_truncate(inode
);
1714 * If truncate failed early the inode might still be
1715 * on the orphan list; we need to make sure the inode
1716 * is removed from the orphan list in that case.
1719 ext4_orphan_del(NULL
, inode
);
1722 return ret
? ret
: copied
;
1725 static int ext4_journalled_write_end(struct file
*file
,
1726 struct address_space
*mapping
,
1727 loff_t pos
, unsigned len
, unsigned copied
,
1728 struct page
*page
, void *fsdata
)
1730 handle_t
*handle
= ext4_journal_current_handle();
1731 struct inode
*inode
= mapping
->host
;
1737 trace_ext4_journalled_write_end(inode
, pos
, len
, copied
);
1738 from
= pos
& (PAGE_CACHE_SIZE
- 1);
1742 if (!PageUptodate(page
))
1744 page_zero_new_buffers(page
, from
+copied
, to
);
1747 ret
= walk_page_buffers(handle
, page_buffers(page
), from
,
1748 to
, &partial
, write_end_fn
);
1750 SetPageUptodate(page
);
1751 new_i_size
= pos
+ copied
;
1752 if (new_i_size
> inode
->i_size
)
1753 i_size_write(inode
, pos
+copied
);
1754 EXT4_I(inode
)->i_state
|= EXT4_STATE_JDATA
;
1755 if (new_i_size
> EXT4_I(inode
)->i_disksize
) {
1756 ext4_update_i_disksize(inode
, new_i_size
);
1757 ret2
= ext4_mark_inode_dirty(handle
, inode
);
1763 page_cache_release(page
);
1764 if (pos
+ len
> inode
->i_size
&& ext4_can_truncate(inode
))
1765 /* if we have allocated more blocks and copied
1766 * less. We will have blocks allocated outside
1767 * inode->i_size. So truncate them
1769 ext4_orphan_add(handle
, inode
);
1771 ret2
= ext4_journal_stop(handle
);
1774 if (pos
+ len
> inode
->i_size
) {
1775 ext4_truncate(inode
);
1777 * If truncate failed early the inode might still be
1778 * on the orphan list; we need to make sure the inode
1779 * is removed from the orphan list in that case.
1782 ext4_orphan_del(NULL
, inode
);
1785 return ret
? ret
: copied
;
1788 static int ext4_da_reserve_space(struct inode
*inode
, int nrblocks
)
1791 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
1792 unsigned long md_needed
, mdblocks
, total
= 0;
1795 * recalculate the amount of metadata blocks to reserve
1796 * in order to allocate nrblocks
1797 * worse case is one extent per block
1800 spin_lock(&EXT4_I(inode
)->i_block_reservation_lock
);
1801 total
= EXT4_I(inode
)->i_reserved_data_blocks
+ nrblocks
;
1802 mdblocks
= ext4_calc_metadata_amount(inode
, total
);
1803 BUG_ON(mdblocks
< EXT4_I(inode
)->i_reserved_meta_blocks
);
1805 md_needed
= mdblocks
- EXT4_I(inode
)->i_reserved_meta_blocks
;
1806 total
= md_needed
+ nrblocks
;
1809 * Make quota reservation here to prevent quota overflow
1810 * later. Real quota accounting is done at pages writeout
1813 if (vfs_dq_reserve_block(inode
, total
)) {
1814 spin_unlock(&EXT4_I(inode
)->i_block_reservation_lock
);
1818 if (ext4_claim_free_blocks(sbi
, total
)) {
1819 spin_unlock(&EXT4_I(inode
)->i_block_reservation_lock
);
1820 vfs_dq_release_reservation_block(inode
, total
);
1821 if (ext4_should_retry_alloc(inode
->i_sb
, &retries
)) {
1827 EXT4_I(inode
)->i_reserved_data_blocks
+= nrblocks
;
1828 EXT4_I(inode
)->i_reserved_meta_blocks
= mdblocks
;
1830 spin_unlock(&EXT4_I(inode
)->i_block_reservation_lock
);
1831 return 0; /* success */
1834 static void ext4_da_release_space(struct inode
*inode
, int to_free
)
1836 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
1837 int total
, mdb
, mdb_free
, release
;
1840 return; /* Nothing to release, exit */
1842 spin_lock(&EXT4_I(inode
)->i_block_reservation_lock
);
1844 if (!EXT4_I(inode
)->i_reserved_data_blocks
) {
1846 * if there is no reserved blocks, but we try to free some
1847 * then the counter is messed up somewhere.
1848 * but since this function is called from invalidate
1849 * page, it's harmless to return without any action
1851 printk(KERN_INFO
"ext4 delalloc try to release %d reserved "
1852 "blocks for inode %lu, but there is no reserved "
1853 "data blocks\n", to_free
, inode
->i_ino
);
1854 spin_unlock(&EXT4_I(inode
)->i_block_reservation_lock
);
1858 /* recalculate the number of metablocks still need to be reserved */
1859 total
= EXT4_I(inode
)->i_reserved_data_blocks
- to_free
;
1860 mdb
= ext4_calc_metadata_amount(inode
, total
);
1862 /* figure out how many metablocks to release */
1863 BUG_ON(mdb
> EXT4_I(inode
)->i_reserved_meta_blocks
);
1864 mdb_free
= EXT4_I(inode
)->i_reserved_meta_blocks
- mdb
;
1866 release
= to_free
+ mdb_free
;
1868 /* update fs dirty blocks counter for truncate case */
1869 percpu_counter_sub(&sbi
->s_dirtyblocks_counter
, release
);
1871 /* update per-inode reservations */
1872 BUG_ON(to_free
> EXT4_I(inode
)->i_reserved_data_blocks
);
1873 EXT4_I(inode
)->i_reserved_data_blocks
-= to_free
;
1875 BUG_ON(mdb
> EXT4_I(inode
)->i_reserved_meta_blocks
);
1876 EXT4_I(inode
)->i_reserved_meta_blocks
= mdb
;
1877 spin_unlock(&EXT4_I(inode
)->i_block_reservation_lock
);
1879 vfs_dq_release_reservation_block(inode
, release
);
1882 static void ext4_da_page_release_reservation(struct page
*page
,
1883 unsigned long offset
)
1886 struct buffer_head
*head
, *bh
;
1887 unsigned int curr_off
= 0;
1889 head
= page_buffers(page
);
1892 unsigned int next_off
= curr_off
+ bh
->b_size
;
1894 if ((offset
<= curr_off
) && (buffer_delay(bh
))) {
1896 clear_buffer_delay(bh
);
1898 curr_off
= next_off
;
1899 } while ((bh
= bh
->b_this_page
) != head
);
1900 ext4_da_release_space(page
->mapping
->host
, to_release
);
1904 * Delayed allocation stuff
1908 * mpage_da_submit_io - walks through extent of pages and try to write
1909 * them with writepage() call back
1911 * @mpd->inode: inode
1912 * @mpd->first_page: first page of the extent
1913 * @mpd->next_page: page after the last page of the extent
1915 * By the time mpage_da_submit_io() is called we expect all blocks
1916 * to be allocated. this may be wrong if allocation failed.
1918 * As pages are already locked by write_cache_pages(), we can't use it
1920 static int mpage_da_submit_io(struct mpage_da_data
*mpd
)
1923 struct pagevec pvec
;
1924 unsigned long index
, end
;
1925 int ret
= 0, err
, nr_pages
, i
;
1926 struct inode
*inode
= mpd
->inode
;
1927 struct address_space
*mapping
= inode
->i_mapping
;
1929 BUG_ON(mpd
->next_page
<= mpd
->first_page
);
1931 * We need to start from the first_page to the next_page - 1
1932 * to make sure we also write the mapped dirty buffer_heads.
1933 * If we look at mpd->b_blocknr we would only be looking
1934 * at the currently mapped buffer_heads.
1936 index
= mpd
->first_page
;
1937 end
= mpd
->next_page
- 1;
1939 pagevec_init(&pvec
, 0);
1940 while (index
<= end
) {
1941 nr_pages
= pagevec_lookup(&pvec
, mapping
, index
, PAGEVEC_SIZE
);
1944 for (i
= 0; i
< nr_pages
; i
++) {
1945 struct page
*page
= pvec
.pages
[i
];
1947 index
= page
->index
;
1952 BUG_ON(!PageLocked(page
));
1953 BUG_ON(PageWriteback(page
));
1955 pages_skipped
= mpd
->wbc
->pages_skipped
;
1956 err
= mapping
->a_ops
->writepage(page
, mpd
->wbc
);
1957 if (!err
&& (pages_skipped
== mpd
->wbc
->pages_skipped
))
1959 * have successfully written the page
1960 * without skipping the same
1962 mpd
->pages_written
++;
1964 * In error case, we have to continue because
1965 * remaining pages are still locked
1966 * XXX: unlock and re-dirty them?
1971 pagevec_release(&pvec
);
1977 * mpage_put_bnr_to_bhs - walk blocks and assign them actual numbers
1979 * @mpd->inode - inode to walk through
1980 * @exbh->b_blocknr - first block on a disk
1981 * @exbh->b_size - amount of space in bytes
1982 * @logical - first logical block to start assignment with
1984 * the function goes through all passed space and put actual disk
1985 * block numbers into buffer heads, dropping BH_Delay and BH_Unwritten
1987 static void mpage_put_bnr_to_bhs(struct mpage_da_data
*mpd
, sector_t logical
,
1988 struct buffer_head
*exbh
)
1990 struct inode
*inode
= mpd
->inode
;
1991 struct address_space
*mapping
= inode
->i_mapping
;
1992 int blocks
= exbh
->b_size
>> inode
->i_blkbits
;
1993 sector_t pblock
= exbh
->b_blocknr
, cur_logical
;
1994 struct buffer_head
*head
, *bh
;
1996 struct pagevec pvec
;
1999 index
= logical
>> (PAGE_CACHE_SHIFT
- inode
->i_blkbits
);
2000 end
= (logical
+ blocks
- 1) >> (PAGE_CACHE_SHIFT
- inode
->i_blkbits
);
2001 cur_logical
= index
<< (PAGE_CACHE_SHIFT
- inode
->i_blkbits
);
2003 pagevec_init(&pvec
, 0);
2005 while (index
<= end
) {
2006 /* XXX: optimize tail */
2007 nr_pages
= pagevec_lookup(&pvec
, mapping
, index
, PAGEVEC_SIZE
);
2010 for (i
= 0; i
< nr_pages
; i
++) {
2011 struct page
*page
= pvec
.pages
[i
];
2013 index
= page
->index
;
2018 BUG_ON(!PageLocked(page
));
2019 BUG_ON(PageWriteback(page
));
2020 BUG_ON(!page_has_buffers(page
));
2022 bh
= page_buffers(page
);
2025 /* skip blocks out of the range */
2027 if (cur_logical
>= logical
)
2030 } while ((bh
= bh
->b_this_page
) != head
);
2033 if (cur_logical
>= logical
+ blocks
)
2036 if (buffer_delay(bh
) ||
2037 buffer_unwritten(bh
)) {
2039 BUG_ON(bh
->b_bdev
!= inode
->i_sb
->s_bdev
);
2041 if (buffer_delay(bh
)) {
2042 clear_buffer_delay(bh
);
2043 bh
->b_blocknr
= pblock
;
2046 * unwritten already should have
2047 * blocknr assigned. Verify that
2049 clear_buffer_unwritten(bh
);
2050 BUG_ON(bh
->b_blocknr
!= pblock
);
2053 } else if (buffer_mapped(bh
))
2054 BUG_ON(bh
->b_blocknr
!= pblock
);
2058 } while ((bh
= bh
->b_this_page
) != head
);
2060 pagevec_release(&pvec
);
2066 * __unmap_underlying_blocks - just a helper function to unmap
2067 * set of blocks described by @bh
2069 static inline void __unmap_underlying_blocks(struct inode
*inode
,
2070 struct buffer_head
*bh
)
2072 struct block_device
*bdev
= inode
->i_sb
->s_bdev
;
2075 blocks
= bh
->b_size
>> inode
->i_blkbits
;
2076 for (i
= 0; i
< blocks
; i
++)
2077 unmap_underlying_metadata(bdev
, bh
->b_blocknr
+ i
);
2080 static void ext4_da_block_invalidatepages(struct mpage_da_data
*mpd
,
2081 sector_t logical
, long blk_cnt
)
2085 struct pagevec pvec
;
2086 struct inode
*inode
= mpd
->inode
;
2087 struct address_space
*mapping
= inode
->i_mapping
;
2089 index
= logical
>> (PAGE_CACHE_SHIFT
- inode
->i_blkbits
);
2090 end
= (logical
+ blk_cnt
- 1) >>
2091 (PAGE_CACHE_SHIFT
- inode
->i_blkbits
);
2092 while (index
<= end
) {
2093 nr_pages
= pagevec_lookup(&pvec
, mapping
, index
, PAGEVEC_SIZE
);
2096 for (i
= 0; i
< nr_pages
; i
++) {
2097 struct page
*page
= pvec
.pages
[i
];
2098 index
= page
->index
;
2103 BUG_ON(!PageLocked(page
));
2104 BUG_ON(PageWriteback(page
));
2105 block_invalidatepage(page
, 0);
2106 ClearPageUptodate(page
);
2113 static void ext4_print_free_blocks(struct inode
*inode
)
2115 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
2116 printk(KERN_CRIT
"Total free blocks count %lld\n",
2117 ext4_count_free_blocks(inode
->i_sb
));
2118 printk(KERN_CRIT
"Free/Dirty block details\n");
2119 printk(KERN_CRIT
"free_blocks=%lld\n",
2120 (long long) percpu_counter_sum(&sbi
->s_freeblocks_counter
));
2121 printk(KERN_CRIT
"dirty_blocks=%lld\n",
2122 (long long) percpu_counter_sum(&sbi
->s_dirtyblocks_counter
));
2123 printk(KERN_CRIT
"Block reservation details\n");
2124 printk(KERN_CRIT
"i_reserved_data_blocks=%u\n",
2125 EXT4_I(inode
)->i_reserved_data_blocks
);
2126 printk(KERN_CRIT
"i_reserved_meta_blocks=%u\n",
2127 EXT4_I(inode
)->i_reserved_meta_blocks
);
2132 * mpage_da_map_blocks - go through given space
2134 * @mpd - bh describing space
2136 * The function skips space we know is already mapped to disk blocks.
2139 static int mpage_da_map_blocks(struct mpage_da_data
*mpd
)
2141 int err
, blks
, get_blocks_flags
;
2142 struct buffer_head
new;
2143 sector_t next
= mpd
->b_blocknr
;
2144 unsigned max_blocks
= mpd
->b_size
>> mpd
->inode
->i_blkbits
;
2145 loff_t disksize
= EXT4_I(mpd
->inode
)->i_disksize
;
2146 handle_t
*handle
= NULL
;
2149 * We consider only non-mapped and non-allocated blocks
2151 if ((mpd
->b_state
& (1 << BH_Mapped
)) &&
2152 !(mpd
->b_state
& (1 << BH_Delay
)) &&
2153 !(mpd
->b_state
& (1 << BH_Unwritten
)))
2157 * If we didn't accumulate anything to write simply return
2162 handle
= ext4_journal_current_handle();
2166 * Call ext4_get_blocks() to allocate any delayed allocation
2167 * blocks, or to convert an uninitialized extent to be
2168 * initialized (in the case where we have written into
2169 * one or more preallocated blocks).
2171 * We pass in the magic EXT4_GET_BLOCKS_DELALLOC_RESERVE to
2172 * indicate that we are on the delayed allocation path. This
2173 * affects functions in many different parts of the allocation
2174 * call path. This flag exists primarily because we don't
2175 * want to change *many* call functions, so ext4_get_blocks()
2176 * will set the magic i_delalloc_reserved_flag once the
2177 * inode's allocation semaphore is taken.
2179 * If the blocks in questions were delalloc blocks, set
2180 * EXT4_GET_BLOCKS_DELALLOC_RESERVE so the delalloc accounting
2181 * variables are updated after the blocks have been allocated.
2184 get_blocks_flags
= (EXT4_GET_BLOCKS_CREATE
|
2185 EXT4_GET_BLOCKS_DELALLOC_RESERVE
);
2186 if (mpd
->b_state
& (1 << BH_Delay
))
2187 get_blocks_flags
|= EXT4_GET_BLOCKS_UPDATE_RESERVE_SPACE
;
2188 blks
= ext4_get_blocks(handle
, mpd
->inode
, next
, max_blocks
,
2189 &new, get_blocks_flags
);
2193 * If get block returns with error we simply
2194 * return. Later writepage will redirty the page and
2195 * writepages will find the dirty page again
2200 if (err
== -ENOSPC
&&
2201 ext4_count_free_blocks(mpd
->inode
->i_sb
)) {
2207 * get block failure will cause us to loop in
2208 * writepages, because a_ops->writepage won't be able
2209 * to make progress. The page will be redirtied by
2210 * writepage and writepages will again try to write
2213 ext4_msg(mpd
->inode
->i_sb
, KERN_CRIT
,
2214 "delayed block allocation failed for inode %lu at "
2215 "logical offset %llu with max blocks %zd with "
2216 "error %d\n", mpd
->inode
->i_ino
,
2217 (unsigned long long) next
,
2218 mpd
->b_size
>> mpd
->inode
->i_blkbits
, err
);
2219 printk(KERN_CRIT
"This should not happen!! "
2220 "Data will be lost\n");
2221 if (err
== -ENOSPC
) {
2222 ext4_print_free_blocks(mpd
->inode
);
2224 /* invalidate all the pages */
2225 ext4_da_block_invalidatepages(mpd
, next
,
2226 mpd
->b_size
>> mpd
->inode
->i_blkbits
);
2231 new.b_size
= (blks
<< mpd
->inode
->i_blkbits
);
2233 if (buffer_new(&new))
2234 __unmap_underlying_blocks(mpd
->inode
, &new);
2237 * If blocks are delayed marked, we need to
2238 * put actual blocknr and drop delayed bit
2240 if ((mpd
->b_state
& (1 << BH_Delay
)) ||
2241 (mpd
->b_state
& (1 << BH_Unwritten
)))
2242 mpage_put_bnr_to_bhs(mpd
, next
, &new);
2244 if (ext4_should_order_data(mpd
->inode
)) {
2245 err
= ext4_jbd2_file_inode(handle
, mpd
->inode
);
2251 * Update on-disk size along with block allocation.
2253 disksize
= ((loff_t
) next
+ blks
) << mpd
->inode
->i_blkbits
;
2254 if (disksize
> i_size_read(mpd
->inode
))
2255 disksize
= i_size_read(mpd
->inode
);
2256 if (disksize
> EXT4_I(mpd
->inode
)->i_disksize
) {
2257 ext4_update_i_disksize(mpd
->inode
, disksize
);
2258 return ext4_mark_inode_dirty(handle
, mpd
->inode
);
2264 #define BH_FLAGS ((1 << BH_Uptodate) | (1 << BH_Mapped) | \
2265 (1 << BH_Delay) | (1 << BH_Unwritten))
2268 * mpage_add_bh_to_extent - try to add one more block to extent of blocks
2270 * @mpd->lbh - extent of blocks
2271 * @logical - logical number of the block in the file
2272 * @bh - bh of the block (used to access block's state)
2274 * the function is used to collect contig. blocks in same state
2276 static void mpage_add_bh_to_extent(struct mpage_da_data
*mpd
,
2277 sector_t logical
, size_t b_size
,
2278 unsigned long b_state
)
2281 int nrblocks
= mpd
->b_size
>> mpd
->inode
->i_blkbits
;
2283 /* check if thereserved journal credits might overflow */
2284 if (!(EXT4_I(mpd
->inode
)->i_flags
& EXT4_EXTENTS_FL
)) {
2285 if (nrblocks
>= EXT4_MAX_TRANS_DATA
) {
2287 * With non-extent format we are limited by the journal
2288 * credit available. Total credit needed to insert
2289 * nrblocks contiguous blocks is dependent on the
2290 * nrblocks. So limit nrblocks.
2293 } else if ((nrblocks
+ (b_size
>> mpd
->inode
->i_blkbits
)) >
2294 EXT4_MAX_TRANS_DATA
) {
2296 * Adding the new buffer_head would make it cross the
2297 * allowed limit for which we have journal credit
2298 * reserved. So limit the new bh->b_size
2300 b_size
= (EXT4_MAX_TRANS_DATA
- nrblocks
) <<
2301 mpd
->inode
->i_blkbits
;
2302 /* we will do mpage_da_submit_io in the next loop */
2306 * First block in the extent
2308 if (mpd
->b_size
== 0) {
2309 mpd
->b_blocknr
= logical
;
2310 mpd
->b_size
= b_size
;
2311 mpd
->b_state
= b_state
& BH_FLAGS
;
2315 next
= mpd
->b_blocknr
+ nrblocks
;
2317 * Can we merge the block to our big extent?
2319 if (logical
== next
&& (b_state
& BH_FLAGS
) == mpd
->b_state
) {
2320 mpd
->b_size
+= b_size
;
2326 * We couldn't merge the block to our extent, so we
2327 * need to flush current extent and start new one
2329 if (mpage_da_map_blocks(mpd
) == 0)
2330 mpage_da_submit_io(mpd
);
2335 static int ext4_bh_delay_or_unwritten(handle_t
*handle
, struct buffer_head
*bh
)
2337 return (buffer_delay(bh
) || buffer_unwritten(bh
)) && buffer_dirty(bh
);
2341 * __mpage_da_writepage - finds extent of pages and blocks
2343 * @page: page to consider
2344 * @wbc: not used, we just follow rules
2347 * The function finds extents of pages and scan them for all blocks.
2349 static int __mpage_da_writepage(struct page
*page
,
2350 struct writeback_control
*wbc
, void *data
)
2352 struct mpage_da_data
*mpd
= data
;
2353 struct inode
*inode
= mpd
->inode
;
2354 struct buffer_head
*bh
, *head
;
2359 * Rest of the page in the page_vec
2360 * redirty then and skip then. We will
2361 * try to write them again after
2362 * starting a new transaction
2364 redirty_page_for_writepage(wbc
, page
);
2366 return MPAGE_DA_EXTENT_TAIL
;
2369 * Can we merge this page to current extent?
2371 if (mpd
->next_page
!= page
->index
) {
2373 * Nope, we can't. So, we map non-allocated blocks
2374 * and start IO on them using writepage()
2376 if (mpd
->next_page
!= mpd
->first_page
) {
2377 if (mpage_da_map_blocks(mpd
) == 0)
2378 mpage_da_submit_io(mpd
);
2380 * skip rest of the page in the page_vec
2383 redirty_page_for_writepage(wbc
, page
);
2385 return MPAGE_DA_EXTENT_TAIL
;
2389 * Start next extent of pages ...
2391 mpd
->first_page
= page
->index
;
2401 mpd
->next_page
= page
->index
+ 1;
2402 logical
= (sector_t
) page
->index
<<
2403 (PAGE_CACHE_SHIFT
- inode
->i_blkbits
);
2405 if (!page_has_buffers(page
)) {
2406 mpage_add_bh_to_extent(mpd
, logical
, PAGE_CACHE_SIZE
,
2407 (1 << BH_Dirty
) | (1 << BH_Uptodate
));
2409 return MPAGE_DA_EXTENT_TAIL
;
2412 * Page with regular buffer heads, just add all dirty ones
2414 head
= page_buffers(page
);
2417 BUG_ON(buffer_locked(bh
));
2419 * We need to try to allocate
2420 * unmapped blocks in the same page.
2421 * Otherwise we won't make progress
2422 * with the page in ext4_writepage
2424 if (ext4_bh_delay_or_unwritten(NULL
, bh
)) {
2425 mpage_add_bh_to_extent(mpd
, logical
,
2429 return MPAGE_DA_EXTENT_TAIL
;
2430 } else if (buffer_dirty(bh
) && (buffer_mapped(bh
))) {
2432 * mapped dirty buffer. We need to update
2433 * the b_state because we look at
2434 * b_state in mpage_da_map_blocks. We don't
2435 * update b_size because if we find an
2436 * unmapped buffer_head later we need to
2437 * use the b_state flag of that buffer_head.
2439 if (mpd
->b_size
== 0)
2440 mpd
->b_state
= bh
->b_state
& BH_FLAGS
;
2443 } while ((bh
= bh
->b_this_page
) != head
);
2450 * This is a special get_blocks_t callback which is used by
2451 * ext4_da_write_begin(). It will either return mapped block or
2452 * reserve space for a single block.
2454 * For delayed buffer_head we have BH_Mapped, BH_New, BH_Delay set.
2455 * We also have b_blocknr = -1 and b_bdev initialized properly
2457 * For unwritten buffer_head we have BH_Mapped, BH_New, BH_Unwritten set.
2458 * We also have b_blocknr = physicalblock mapping unwritten extent and b_bdev
2459 * initialized properly.
2461 static int ext4_da_get_block_prep(struct inode
*inode
, sector_t iblock
,
2462 struct buffer_head
*bh_result
, int create
)
2465 sector_t invalid_block
= ~((sector_t
) 0xffff);
2467 if (invalid_block
< ext4_blocks_count(EXT4_SB(inode
->i_sb
)->s_es
))
2470 BUG_ON(create
== 0);
2471 BUG_ON(bh_result
->b_size
!= inode
->i_sb
->s_blocksize
);
2474 * first, we need to know whether the block is allocated already
2475 * preallocated blocks are unmapped but should treated
2476 * the same as allocated blocks.
2478 ret
= ext4_get_blocks(NULL
, inode
, iblock
, 1, bh_result
, 0);
2479 if ((ret
== 0) && !buffer_delay(bh_result
)) {
2480 /* the block isn't (pre)allocated yet, let's reserve space */
2482 * XXX: __block_prepare_write() unmaps passed block,
2485 ret
= ext4_da_reserve_space(inode
, 1);
2487 /* not enough space to reserve */
2490 map_bh(bh_result
, inode
->i_sb
, invalid_block
);
2491 set_buffer_new(bh_result
);
2492 set_buffer_delay(bh_result
);
2493 } else if (ret
> 0) {
2494 bh_result
->b_size
= (ret
<< inode
->i_blkbits
);
2495 if (buffer_unwritten(bh_result
)) {
2496 /* A delayed write to unwritten bh should
2497 * be marked new and mapped. Mapped ensures
2498 * that we don't do get_block multiple times
2499 * when we write to the same offset and new
2500 * ensures that we do proper zero out for
2503 set_buffer_new(bh_result
);
2504 set_buffer_mapped(bh_result
);
2513 * This function is used as a standard get_block_t calback function
2514 * when there is no desire to allocate any blocks. It is used as a
2515 * callback function for block_prepare_write(), nobh_writepage(), and
2516 * block_write_full_page(). These functions should only try to map a
2517 * single block at a time.
2519 * Since this function doesn't do block allocations even if the caller
2520 * requests it by passing in create=1, it is critically important that
2521 * any caller checks to make sure that any buffer heads are returned
2522 * by this function are either all already mapped or marked for
2523 * delayed allocation before calling nobh_writepage() or
2524 * block_write_full_page(). Otherwise, b_blocknr could be left
2525 * unitialized, and the page write functions will be taken by
2528 static int noalloc_get_block_write(struct inode
*inode
, sector_t iblock
,
2529 struct buffer_head
*bh_result
, int create
)
2532 unsigned max_blocks
= bh_result
->b_size
>> inode
->i_blkbits
;
2534 BUG_ON(bh_result
->b_size
!= inode
->i_sb
->s_blocksize
);
2537 * we don't want to do block allocation in writepage
2538 * so call get_block_wrap with create = 0
2540 ret
= ext4_get_blocks(NULL
, inode
, iblock
, max_blocks
, bh_result
, 0);
2542 bh_result
->b_size
= (ret
<< inode
->i_blkbits
);
2548 static int bget_one(handle_t
*handle
, struct buffer_head
*bh
)
2554 static int bput_one(handle_t
*handle
, struct buffer_head
*bh
)
2560 static int __ext4_journalled_writepage(struct page
*page
,
2561 struct writeback_control
*wbc
,
2564 struct address_space
*mapping
= page
->mapping
;
2565 struct inode
*inode
= mapping
->host
;
2566 struct buffer_head
*page_bufs
;
2567 handle_t
*handle
= NULL
;
2571 page_bufs
= page_buffers(page
);
2573 walk_page_buffers(handle
, page_bufs
, 0, len
, NULL
, bget_one
);
2574 /* As soon as we unlock the page, it can go away, but we have
2575 * references to buffers so we are safe */
2578 handle
= ext4_journal_start(inode
, ext4_writepage_trans_blocks(inode
));
2579 if (IS_ERR(handle
)) {
2580 ret
= PTR_ERR(handle
);
2584 ret
= walk_page_buffers(handle
, page_bufs
, 0, len
, NULL
,
2585 do_journal_get_write_access
);
2587 err
= walk_page_buffers(handle
, page_bufs
, 0, len
, NULL
,
2591 err
= ext4_journal_stop(handle
);
2595 walk_page_buffers(handle
, page_bufs
, 0, len
, NULL
, bput_one
);
2596 EXT4_I(inode
)->i_state
|= EXT4_STATE_JDATA
;
2602 * Note that we don't need to start a transaction unless we're journaling data
2603 * because we should have holes filled from ext4_page_mkwrite(). We even don't
2604 * need to file the inode to the transaction's list in ordered mode because if
2605 * we are writing back data added by write(), the inode is already there and if
2606 * we are writing back data modified via mmap(), noone guarantees in which
2607 * transaction the data will hit the disk. In case we are journaling data, we
2608 * cannot start transaction directly because transaction start ranks above page
2609 * lock so we have to do some magic.
2611 * This function can get called via...
2612 * - ext4_da_writepages after taking page lock (have journal handle)
2613 * - journal_submit_inode_data_buffers (no journal handle)
2614 * - shrink_page_list via pdflush (no journal handle)
2615 * - grab_page_cache when doing write_begin (have journal handle)
2617 * We don't do any block allocation in this function. If we have page with
2618 * multiple blocks we need to write those buffer_heads that are mapped. This
2619 * is important for mmaped based write. So if we do with blocksize 1K
2620 * truncate(f, 1024);
2621 * a = mmap(f, 0, 4096);
2623 * truncate(f, 4096);
2624 * we have in the page first buffer_head mapped via page_mkwrite call back
2625 * but other bufer_heads would be unmapped but dirty(dirty done via the
2626 * do_wp_page). So writepage should write the first block. If we modify
2627 * the mmap area beyond 1024 we will again get a page_fault and the
2628 * page_mkwrite callback will do the block allocation and mark the
2629 * buffer_heads mapped.
2631 * We redirty the page if we have any buffer_heads that is either delay or
2632 * unwritten in the page.
2634 * We can get recursively called as show below.
2636 * ext4_writepage() -> kmalloc() -> __alloc_pages() -> page_launder() ->
2639 * But since we don't do any block allocation we should not deadlock.
2640 * Page also have the dirty flag cleared so we don't get recurive page_lock.
2642 static int ext4_writepage(struct page
*page
,
2643 struct writeback_control
*wbc
)
2648 struct buffer_head
*page_bufs
;
2649 struct inode
*inode
= page
->mapping
->host
;
2651 trace_ext4_writepage(inode
, page
);
2652 size
= i_size_read(inode
);
2653 if (page
->index
== size
>> PAGE_CACHE_SHIFT
)
2654 len
= size
& ~PAGE_CACHE_MASK
;
2656 len
= PAGE_CACHE_SIZE
;
2658 if (page_has_buffers(page
)) {
2659 page_bufs
= page_buffers(page
);
2660 if (walk_page_buffers(NULL
, page_bufs
, 0, len
, NULL
,
2661 ext4_bh_delay_or_unwritten
)) {
2663 * We don't want to do block allocation
2664 * So redirty the page and return
2665 * We may reach here when we do a journal commit
2666 * via journal_submit_inode_data_buffers.
2667 * If we don't have mapping block we just ignore
2668 * them. We can also reach here via shrink_page_list
2670 redirty_page_for_writepage(wbc
, page
);
2676 * The test for page_has_buffers() is subtle:
2677 * We know the page is dirty but it lost buffers. That means
2678 * that at some moment in time after write_begin()/write_end()
2679 * has been called all buffers have been clean and thus they
2680 * must have been written at least once. So they are all
2681 * mapped and we can happily proceed with mapping them
2682 * and writing the page.
2684 * Try to initialize the buffer_heads and check whether
2685 * all are mapped and non delay. We don't want to
2686 * do block allocation here.
2688 ret
= block_prepare_write(page
, 0, len
,
2689 noalloc_get_block_write
);
2691 page_bufs
= page_buffers(page
);
2692 /* check whether all are mapped and non delay */
2693 if (walk_page_buffers(NULL
, page_bufs
, 0, len
, NULL
,
2694 ext4_bh_delay_or_unwritten
)) {
2695 redirty_page_for_writepage(wbc
, page
);
2701 * We can't do block allocation here
2702 * so just redity the page and unlock
2705 redirty_page_for_writepage(wbc
, page
);
2709 /* now mark the buffer_heads as dirty and uptodate */
2710 block_commit_write(page
, 0, len
);
2713 if (PageChecked(page
) && ext4_should_journal_data(inode
)) {
2715 * It's mmapped pagecache. Add buffers and journal it. There
2716 * doesn't seem much point in redirtying the page here.
2718 ClearPageChecked(page
);
2719 return __ext4_journalled_writepage(page
, wbc
, len
);
2722 if (test_opt(inode
->i_sb
, NOBH
) && ext4_should_writeback_data(inode
))
2723 ret
= nobh_writepage(page
, noalloc_get_block_write
, wbc
);
2725 ret
= block_write_full_page(page
, noalloc_get_block_write
,
2732 * This is called via ext4_da_writepages() to
2733 * calulate the total number of credits to reserve to fit
2734 * a single extent allocation into a single transaction,
2735 * ext4_da_writpeages() will loop calling this before
2736 * the block allocation.
2739 static int ext4_da_writepages_trans_blocks(struct inode
*inode
)
2741 int max_blocks
= EXT4_I(inode
)->i_reserved_data_blocks
;
2744 * With non-extent format the journal credit needed to
2745 * insert nrblocks contiguous block is dependent on
2746 * number of contiguous block. So we will limit
2747 * number of contiguous block to a sane value
2749 if (!(EXT4_I(inode
)->i_flags
& EXT4_EXTENTS_FL
) &&
2750 (max_blocks
> EXT4_MAX_TRANS_DATA
))
2751 max_blocks
= EXT4_MAX_TRANS_DATA
;
2753 return ext4_chunk_trans_blocks(inode
, max_blocks
);
2756 static int ext4_da_writepages(struct address_space
*mapping
,
2757 struct writeback_control
*wbc
)
2760 int range_whole
= 0;
2761 handle_t
*handle
= NULL
;
2762 struct mpage_da_data mpd
;
2763 struct inode
*inode
= mapping
->host
;
2764 int no_nrwrite_index_update
;
2765 int pages_written
= 0;
2767 unsigned int max_pages
;
2768 int range_cyclic
, cycled
= 1, io_done
= 0;
2769 int needed_blocks
, ret
= 0;
2770 long desired_nr_to_write
, nr_to_writebump
= 0;
2771 loff_t range_start
= wbc
->range_start
;
2772 struct ext4_sb_info
*sbi
= EXT4_SB(mapping
->host
->i_sb
);
2774 trace_ext4_da_writepages(inode
, wbc
);
2777 * No pages to write? This is mainly a kludge to avoid starting
2778 * a transaction for special inodes like journal inode on last iput()
2779 * because that could violate lock ordering on umount
2781 if (!mapping
->nrpages
|| !mapping_tagged(mapping
, PAGECACHE_TAG_DIRTY
))
2785 * If the filesystem has aborted, it is read-only, so return
2786 * right away instead of dumping stack traces later on that
2787 * will obscure the real source of the problem. We test
2788 * EXT4_MF_FS_ABORTED instead of sb->s_flag's MS_RDONLY because
2789 * the latter could be true if the filesystem is mounted
2790 * read-only, and in that case, ext4_da_writepages should
2791 * *never* be called, so if that ever happens, we would want
2794 if (unlikely(sbi
->s_mount_flags
& EXT4_MF_FS_ABORTED
))
2797 if (wbc
->range_start
== 0 && wbc
->range_end
== LLONG_MAX
)
2800 range_cyclic
= wbc
->range_cyclic
;
2801 if (wbc
->range_cyclic
) {
2802 index
= mapping
->writeback_index
;
2805 wbc
->range_start
= index
<< PAGE_CACHE_SHIFT
;
2806 wbc
->range_end
= LLONG_MAX
;
2807 wbc
->range_cyclic
= 0;
2809 index
= wbc
->range_start
>> PAGE_CACHE_SHIFT
;
2812 * This works around two forms of stupidity. The first is in
2813 * the writeback code, which caps the maximum number of pages
2814 * written to be 1024 pages. This is wrong on multiple
2815 * levels; different architectues have a different page size,
2816 * which changes the maximum amount of data which gets
2817 * written. Secondly, 4 megabytes is way too small. XFS
2818 * forces this value to be 16 megabytes by multiplying
2819 * nr_to_write parameter by four, and then relies on its
2820 * allocator to allocate larger extents to make them
2821 * contiguous. Unfortunately this brings us to the second
2822 * stupidity, which is that ext4's mballoc code only allocates
2823 * at most 2048 blocks. So we force contiguous writes up to
2824 * the number of dirty blocks in the inode, or
2825 * sbi->max_writeback_mb_bump whichever is smaller.
2827 max_pages
= sbi
->s_max_writeback_mb_bump
<< (20 - PAGE_CACHE_SHIFT
);
2828 if (!range_cyclic
&& range_whole
)
2829 desired_nr_to_write
= wbc
->nr_to_write
* 8;
2831 desired_nr_to_write
= ext4_num_dirty_pages(inode
, index
,
2833 if (desired_nr_to_write
> max_pages
)
2834 desired_nr_to_write
= max_pages
;
2836 if (wbc
->nr_to_write
< desired_nr_to_write
) {
2837 nr_to_writebump
= desired_nr_to_write
- wbc
->nr_to_write
;
2838 wbc
->nr_to_write
= desired_nr_to_write
;
2842 mpd
.inode
= mapping
->host
;
2845 * we don't want write_cache_pages to update
2846 * nr_to_write and writeback_index
2848 no_nrwrite_index_update
= wbc
->no_nrwrite_index_update
;
2849 wbc
->no_nrwrite_index_update
= 1;
2850 pages_skipped
= wbc
->pages_skipped
;
2853 while (!ret
&& wbc
->nr_to_write
> 0) {
2856 * we insert one extent at a time. So we need
2857 * credit needed for single extent allocation.
2858 * journalled mode is currently not supported
2861 BUG_ON(ext4_should_journal_data(inode
));
2862 needed_blocks
= ext4_da_writepages_trans_blocks(inode
);
2864 /* start a new transaction*/
2865 handle
= ext4_journal_start(inode
, needed_blocks
);
2866 if (IS_ERR(handle
)) {
2867 ret
= PTR_ERR(handle
);
2868 ext4_msg(inode
->i_sb
, KERN_CRIT
, "%s: jbd2_start: "
2869 "%ld pages, ino %lu; err %d\n", __func__
,
2870 wbc
->nr_to_write
, inode
->i_ino
, ret
);
2871 goto out_writepages
;
2875 * Now call __mpage_da_writepage to find the next
2876 * contiguous region of logical blocks that need
2877 * blocks to be allocated by ext4. We don't actually
2878 * submit the blocks for I/O here, even though
2879 * write_cache_pages thinks it will, and will set the
2880 * pages as clean for write before calling
2881 * __mpage_da_writepage().
2889 mpd
.pages_written
= 0;
2891 ret
= write_cache_pages(mapping
, wbc
, __mpage_da_writepage
,
2894 * If we have a contigous extent of pages and we
2895 * haven't done the I/O yet, map the blocks and submit
2898 if (!mpd
.io_done
&& mpd
.next_page
!= mpd
.first_page
) {
2899 if (mpage_da_map_blocks(&mpd
) == 0)
2900 mpage_da_submit_io(&mpd
);
2902 ret
= MPAGE_DA_EXTENT_TAIL
;
2904 trace_ext4_da_write_pages(inode
, &mpd
);
2905 wbc
->nr_to_write
-= mpd
.pages_written
;
2907 ext4_journal_stop(handle
);
2909 if ((mpd
.retval
== -ENOSPC
) && sbi
->s_journal
) {
2910 /* commit the transaction which would
2911 * free blocks released in the transaction
2914 jbd2_journal_force_commit_nested(sbi
->s_journal
);
2915 wbc
->pages_skipped
= pages_skipped
;
2917 } else if (ret
== MPAGE_DA_EXTENT_TAIL
) {
2919 * got one extent now try with
2922 pages_written
+= mpd
.pages_written
;
2923 wbc
->pages_skipped
= pages_skipped
;
2926 } else if (wbc
->nr_to_write
)
2928 * There is no more writeout needed
2929 * or we requested for a noblocking writeout
2930 * and we found the device congested
2934 if (!io_done
&& !cycled
) {
2937 wbc
->range_start
= index
<< PAGE_CACHE_SHIFT
;
2938 wbc
->range_end
= mapping
->writeback_index
- 1;
2941 if (pages_skipped
!= wbc
->pages_skipped
)
2942 ext4_msg(inode
->i_sb
, KERN_CRIT
,
2943 "This should not happen leaving %s "
2944 "with nr_to_write = %ld ret = %d\n",
2945 __func__
, wbc
->nr_to_write
, ret
);
2948 index
+= pages_written
;
2949 wbc
->range_cyclic
= range_cyclic
;
2950 if (wbc
->range_cyclic
|| (range_whole
&& wbc
->nr_to_write
> 0))
2952 * set the writeback_index so that range_cyclic
2953 * mode will write it back later
2955 mapping
->writeback_index
= index
;
2958 if (!no_nrwrite_index_update
)
2959 wbc
->no_nrwrite_index_update
= 0;
2960 if (wbc
->nr_to_write
> nr_to_writebump
)
2961 wbc
->nr_to_write
-= nr_to_writebump
;
2962 wbc
->range_start
= range_start
;
2963 trace_ext4_da_writepages_result(inode
, wbc
, ret
, pages_written
);
2967 #define FALL_BACK_TO_NONDELALLOC 1
2968 static int ext4_nonda_switch(struct super_block
*sb
)
2970 s64 free_blocks
, dirty_blocks
;
2971 struct ext4_sb_info
*sbi
= EXT4_SB(sb
);
2974 * switch to non delalloc mode if we are running low
2975 * on free block. The free block accounting via percpu
2976 * counters can get slightly wrong with percpu_counter_batch getting
2977 * accumulated on each CPU without updating global counters
2978 * Delalloc need an accurate free block accounting. So switch
2979 * to non delalloc when we are near to error range.
2981 free_blocks
= percpu_counter_read_positive(&sbi
->s_freeblocks_counter
);
2982 dirty_blocks
= percpu_counter_read_positive(&sbi
->s_dirtyblocks_counter
);
2983 if (2 * free_blocks
< 3 * dirty_blocks
||
2984 free_blocks
< (dirty_blocks
+ EXT4_FREEBLOCKS_WATERMARK
)) {
2986 * free block count is less that 150% of dirty blocks
2987 * or free blocks is less that watermark
2994 static int ext4_da_write_begin(struct file
*file
, struct address_space
*mapping
,
2995 loff_t pos
, unsigned len
, unsigned flags
,
2996 struct page
**pagep
, void **fsdata
)
2998 int ret
, retries
= 0;
3002 struct inode
*inode
= mapping
->host
;
3005 index
= pos
>> PAGE_CACHE_SHIFT
;
3006 from
= pos
& (PAGE_CACHE_SIZE
- 1);
3009 if (ext4_nonda_switch(inode
->i_sb
)) {
3010 *fsdata
= (void *)FALL_BACK_TO_NONDELALLOC
;
3011 return ext4_write_begin(file
, mapping
, pos
,
3012 len
, flags
, pagep
, fsdata
);
3014 *fsdata
= (void *)0;
3015 trace_ext4_da_write_begin(inode
, pos
, len
, flags
);
3018 * With delayed allocation, we don't log the i_disksize update
3019 * if there is delayed block allocation. But we still need
3020 * to journalling the i_disksize update if writes to the end
3021 * of file which has an already mapped buffer.
3023 handle
= ext4_journal_start(inode
, 1);
3024 if (IS_ERR(handle
)) {
3025 ret
= PTR_ERR(handle
);
3028 /* We cannot recurse into the filesystem as the transaction is already
3030 flags
|= AOP_FLAG_NOFS
;
3032 page
= grab_cache_page_write_begin(mapping
, index
, flags
);
3034 ext4_journal_stop(handle
);
3040 ret
= block_write_begin(file
, mapping
, pos
, len
, flags
, pagep
, fsdata
,
3041 ext4_da_get_block_prep
);
3044 ext4_journal_stop(handle
);
3045 page_cache_release(page
);
3047 * block_write_begin may have instantiated a few blocks
3048 * outside i_size. Trim these off again. Don't need
3049 * i_size_read because we hold i_mutex.
3051 if (pos
+ len
> inode
->i_size
)
3052 ext4_truncate(inode
);
3055 if (ret
== -ENOSPC
&& ext4_should_retry_alloc(inode
->i_sb
, &retries
))
3062 * Check if we should update i_disksize
3063 * when write to the end of file but not require block allocation
3065 static int ext4_da_should_update_i_disksize(struct page
*page
,
3066 unsigned long offset
)
3068 struct buffer_head
*bh
;
3069 struct inode
*inode
= page
->mapping
->host
;
3073 bh
= page_buffers(page
);
3074 idx
= offset
>> inode
->i_blkbits
;
3076 for (i
= 0; i
< idx
; i
++)
3077 bh
= bh
->b_this_page
;
3079 if (!buffer_mapped(bh
) || (buffer_delay(bh
)) || buffer_unwritten(bh
))
3084 static int ext4_da_write_end(struct file
*file
,
3085 struct address_space
*mapping
,
3086 loff_t pos
, unsigned len
, unsigned copied
,
3087 struct page
*page
, void *fsdata
)
3089 struct inode
*inode
= mapping
->host
;
3091 handle_t
*handle
= ext4_journal_current_handle();
3093 unsigned long start
, end
;
3094 int write_mode
= (int)(unsigned long)fsdata
;
3096 if (write_mode
== FALL_BACK_TO_NONDELALLOC
) {
3097 if (ext4_should_order_data(inode
)) {
3098 return ext4_ordered_write_end(file
, mapping
, pos
,
3099 len
, copied
, page
, fsdata
);
3100 } else if (ext4_should_writeback_data(inode
)) {
3101 return ext4_writeback_write_end(file
, mapping
, pos
,
3102 len
, copied
, page
, fsdata
);
3108 trace_ext4_da_write_end(inode
, pos
, len
, copied
);
3109 start
= pos
& (PAGE_CACHE_SIZE
- 1);
3110 end
= start
+ copied
- 1;
3113 * generic_write_end() will run mark_inode_dirty() if i_size
3114 * changes. So let's piggyback the i_disksize mark_inode_dirty
3118 new_i_size
= pos
+ copied
;
3119 if (new_i_size
> EXT4_I(inode
)->i_disksize
) {
3120 if (ext4_da_should_update_i_disksize(page
, end
)) {
3121 down_write(&EXT4_I(inode
)->i_data_sem
);
3122 if (new_i_size
> EXT4_I(inode
)->i_disksize
) {
3124 * Updating i_disksize when extending file
3125 * without needing block allocation
3127 if (ext4_should_order_data(inode
))
3128 ret
= ext4_jbd2_file_inode(handle
,
3131 EXT4_I(inode
)->i_disksize
= new_i_size
;
3133 up_write(&EXT4_I(inode
)->i_data_sem
);
3134 /* We need to mark inode dirty even if
3135 * new_i_size is less that inode->i_size
3136 * bu greater than i_disksize.(hint delalloc)
3138 ext4_mark_inode_dirty(handle
, inode
);
3141 ret2
= generic_write_end(file
, mapping
, pos
, len
, copied
,
3146 ret2
= ext4_journal_stop(handle
);
3150 return ret
? ret
: copied
;
3153 static void ext4_da_invalidatepage(struct page
*page
, unsigned long offset
)
3156 * Drop reserved blocks
3158 BUG_ON(!PageLocked(page
));
3159 if (!page_has_buffers(page
))
3162 ext4_da_page_release_reservation(page
, offset
);
3165 ext4_invalidatepage(page
, offset
);
3171 * Force all delayed allocation blocks to be allocated for a given inode.
3173 int ext4_alloc_da_blocks(struct inode
*inode
)
3175 trace_ext4_alloc_da_blocks(inode
);
3177 if (!EXT4_I(inode
)->i_reserved_data_blocks
&&
3178 !EXT4_I(inode
)->i_reserved_meta_blocks
)
3182 * We do something simple for now. The filemap_flush() will
3183 * also start triggering a write of the data blocks, which is
3184 * not strictly speaking necessary (and for users of
3185 * laptop_mode, not even desirable). However, to do otherwise
3186 * would require replicating code paths in:
3188 * ext4_da_writepages() ->
3189 * write_cache_pages() ---> (via passed in callback function)
3190 * __mpage_da_writepage() -->
3191 * mpage_add_bh_to_extent()
3192 * mpage_da_map_blocks()
3194 * The problem is that write_cache_pages(), located in
3195 * mm/page-writeback.c, marks pages clean in preparation for
3196 * doing I/O, which is not desirable if we're not planning on
3199 * We could call write_cache_pages(), and then redirty all of
3200 * the pages by calling redirty_page_for_writeback() but that
3201 * would be ugly in the extreme. So instead we would need to
3202 * replicate parts of the code in the above functions,
3203 * simplifying them becuase we wouldn't actually intend to
3204 * write out the pages, but rather only collect contiguous
3205 * logical block extents, call the multi-block allocator, and
3206 * then update the buffer heads with the block allocations.
3208 * For now, though, we'll cheat by calling filemap_flush(),
3209 * which will map the blocks, and start the I/O, but not
3210 * actually wait for the I/O to complete.
3212 return filemap_flush(inode
->i_mapping
);
3216 * bmap() is special. It gets used by applications such as lilo and by
3217 * the swapper to find the on-disk block of a specific piece of data.
3219 * Naturally, this is dangerous if the block concerned is still in the
3220 * journal. If somebody makes a swapfile on an ext4 data-journaling
3221 * filesystem and enables swap, then they may get a nasty shock when the
3222 * data getting swapped to that swapfile suddenly gets overwritten by
3223 * the original zero's written out previously to the journal and
3224 * awaiting writeback in the kernel's buffer cache.
3226 * So, if we see any bmap calls here on a modified, data-journaled file,
3227 * take extra steps to flush any blocks which might be in the cache.
3229 static sector_t
ext4_bmap(struct address_space
*mapping
, sector_t block
)
3231 struct inode
*inode
= mapping
->host
;
3235 if (mapping_tagged(mapping
, PAGECACHE_TAG_DIRTY
) &&
3236 test_opt(inode
->i_sb
, DELALLOC
)) {
3238 * With delalloc we want to sync the file
3239 * so that we can make sure we allocate
3242 filemap_write_and_wait(mapping
);
3245 if (EXT4_JOURNAL(inode
) && EXT4_I(inode
)->i_state
& EXT4_STATE_JDATA
) {
3247 * This is a REALLY heavyweight approach, but the use of
3248 * bmap on dirty files is expected to be extremely rare:
3249 * only if we run lilo or swapon on a freshly made file
3250 * do we expect this to happen.
3252 * (bmap requires CAP_SYS_RAWIO so this does not
3253 * represent an unprivileged user DOS attack --- we'd be
3254 * in trouble if mortal users could trigger this path at
3257 * NB. EXT4_STATE_JDATA is not set on files other than
3258 * regular files. If somebody wants to bmap a directory
3259 * or symlink and gets confused because the buffer
3260 * hasn't yet been flushed to disk, they deserve
3261 * everything they get.
3264 EXT4_I(inode
)->i_state
&= ~EXT4_STATE_JDATA
;
3265 journal
= EXT4_JOURNAL(inode
);
3266 jbd2_journal_lock_updates(journal
);
3267 err
= jbd2_journal_flush(journal
);
3268 jbd2_journal_unlock_updates(journal
);
3274 return generic_block_bmap(mapping
, block
, ext4_get_block
);
3277 static int ext4_readpage(struct file
*file
, struct page
*page
)
3279 return mpage_readpage(page
, ext4_get_block
);
3283 ext4_readpages(struct file
*file
, struct address_space
*mapping
,
3284 struct list_head
*pages
, unsigned nr_pages
)
3286 return mpage_readpages(mapping
, pages
, nr_pages
, ext4_get_block
);
3289 static void ext4_invalidatepage(struct page
*page
, unsigned long offset
)
3291 journal_t
*journal
= EXT4_JOURNAL(page
->mapping
->host
);
3294 * If it's a full truncate we just forget about the pending dirtying
3297 ClearPageChecked(page
);
3300 jbd2_journal_invalidatepage(journal
, page
, offset
);
3302 block_invalidatepage(page
, offset
);
3305 static int ext4_releasepage(struct page
*page
, gfp_t wait
)
3307 journal_t
*journal
= EXT4_JOURNAL(page
->mapping
->host
);
3309 WARN_ON(PageChecked(page
));
3310 if (!page_has_buffers(page
))
3313 return jbd2_journal_try_to_free_buffers(journal
, page
, wait
);
3315 return try_to_free_buffers(page
);
3319 * O_DIRECT for ext3 (or indirect map) based files
3321 * If the O_DIRECT write will extend the file then add this inode to the
3322 * orphan list. So recovery will truncate it back to the original size
3323 * if the machine crashes during the write.
3325 * If the O_DIRECT write is intantiating holes inside i_size and the machine
3326 * crashes then stale disk data _may_ be exposed inside the file. But current
3327 * VFS code falls back into buffered path in that case so we are safe.
3329 static ssize_t
ext4_ind_direct_IO(int rw
, struct kiocb
*iocb
,
3330 const struct iovec
*iov
, loff_t offset
,
3331 unsigned long nr_segs
)
3333 struct file
*file
= iocb
->ki_filp
;
3334 struct inode
*inode
= file
->f_mapping
->host
;
3335 struct ext4_inode_info
*ei
= EXT4_I(inode
);
3339 size_t count
= iov_length(iov
, nr_segs
);
3343 loff_t final_size
= offset
+ count
;
3345 if (final_size
> inode
->i_size
) {
3346 /* Credits for sb + inode write */
3347 handle
= ext4_journal_start(inode
, 2);
3348 if (IS_ERR(handle
)) {
3349 ret
= PTR_ERR(handle
);
3352 ret
= ext4_orphan_add(handle
, inode
);
3354 ext4_journal_stop(handle
);
3358 ei
->i_disksize
= inode
->i_size
;
3359 ext4_journal_stop(handle
);
3364 ret
= blockdev_direct_IO(rw
, iocb
, inode
, inode
->i_sb
->s_bdev
, iov
,
3366 ext4_get_block
, NULL
);
3367 if (ret
== -ENOSPC
&& ext4_should_retry_alloc(inode
->i_sb
, &retries
))
3373 /* Credits for sb + inode write */
3374 handle
= ext4_journal_start(inode
, 2);
3375 if (IS_ERR(handle
)) {
3376 /* This is really bad luck. We've written the data
3377 * but cannot extend i_size. Bail out and pretend
3378 * the write failed... */
3379 ret
= PTR_ERR(handle
);
3383 ext4_orphan_del(handle
, inode
);
3385 loff_t end
= offset
+ ret
;
3386 if (end
> inode
->i_size
) {
3387 ei
->i_disksize
= end
;
3388 i_size_write(inode
, end
);
3390 * We're going to return a positive `ret'
3391 * here due to non-zero-length I/O, so there's
3392 * no way of reporting error returns from
3393 * ext4_mark_inode_dirty() to userspace. So
3396 ext4_mark_inode_dirty(handle
, inode
);
3399 err
= ext4_journal_stop(handle
);
3407 static int ext4_get_block_dio_write(struct inode
*inode
, sector_t iblock
,
3408 struct buffer_head
*bh_result
, int create
)
3410 handle_t
*handle
= NULL
;
3412 unsigned max_blocks
= bh_result
->b_size
>> inode
->i_blkbits
;
3415 ext4_debug("ext4_get_block_dio_write: inode %lu, create flag %d\n",
3416 inode
->i_ino
, create
);
3418 * DIO VFS code passes create = 0 flag for write to
3419 * the middle of file. It does this to avoid block
3420 * allocation for holes, to prevent expose stale data
3421 * out when there is parallel buffered read (which does
3422 * not hold the i_mutex lock) while direct IO write has
3423 * not completed. DIO request on holes finally falls back
3424 * to buffered IO for this reason.
3426 * For ext4 extent based file, since we support fallocate,
3427 * new allocated extent as uninitialized, for holes, we
3428 * could fallocate blocks for holes, thus parallel
3429 * buffered IO read will zero out the page when read on
3430 * a hole while parallel DIO write to the hole has not completed.
3432 * when we come here, we know it's a direct IO write to
3433 * to the middle of file (<i_size)
3434 * so it's safe to override the create flag from VFS.
3436 create
= EXT4_GET_BLOCKS_DIO_CREATE_EXT
;
3438 if (max_blocks
> DIO_MAX_BLOCKS
)
3439 max_blocks
= DIO_MAX_BLOCKS
;
3440 dio_credits
= ext4_chunk_trans_blocks(inode
, max_blocks
);
3441 handle
= ext4_journal_start(inode
, dio_credits
);
3442 if (IS_ERR(handle
)) {
3443 ret
= PTR_ERR(handle
);
3446 ret
= ext4_get_blocks(handle
, inode
, iblock
, max_blocks
, bh_result
,
3449 bh_result
->b_size
= (ret
<< inode
->i_blkbits
);
3452 ext4_journal_stop(handle
);
3457 static void ext4_free_io_end(ext4_io_end_t
*io
)
3463 static void dump_aio_dio_list(struct inode
* inode
)
3466 struct list_head
*cur
, *before
, *after
;
3467 ext4_io_end_t
*io
, *io0
, *io1
;
3469 if (list_empty(&EXT4_I(inode
)->i_aio_dio_complete_list
)){
3470 ext4_debug("inode %lu aio dio list is empty\n", inode
->i_ino
);
3474 ext4_debug("Dump inode %lu aio_dio_completed_IO list \n", inode
->i_ino
);
3475 list_for_each_entry(io
, &EXT4_I(inode
)->i_aio_dio_complete_list
, list
){
3478 io0
= container_of(before
, ext4_io_end_t
, list
);
3480 io1
= container_of(after
, ext4_io_end_t
, list
);
3482 ext4_debug("io 0x%p from inode %lu,prev 0x%p,next 0x%p\n",
3483 io
, inode
->i_ino
, io0
, io1
);
3489 * check a range of space and convert unwritten extents to written.
3491 static int ext4_end_aio_dio_nolock(ext4_io_end_t
*io
)
3493 struct inode
*inode
= io
->inode
;
3494 loff_t offset
= io
->offset
;
3495 size_t size
= io
->size
;
3498 ext4_debug("end_aio_dio_onlock: io 0x%p from inode %lu,list->next 0x%p,"
3499 "list->prev 0x%p\n",
3500 io
, inode
->i_ino
, io
->list
.next
, io
->list
.prev
);
3502 if (list_empty(&io
->list
))
3505 if (io
->flag
!= DIO_AIO_UNWRITTEN
)
3508 if (offset
+ size
<= i_size_read(inode
))
3509 ret
= ext4_convert_unwritten_extents(inode
, offset
, size
);
3512 printk(KERN_EMERG
"%s: failed to convert unwritten"
3513 "extents to written extents, error is %d"
3514 " io is still on inode %lu aio dio list\n",
3515 __func__
, ret
, inode
->i_ino
);
3519 /* clear the DIO AIO unwritten flag */
3524 * work on completed aio dio IO, to convert unwritten extents to extents
3526 static void ext4_end_aio_dio_work(struct work_struct
*work
)
3528 ext4_io_end_t
*io
= container_of(work
, ext4_io_end_t
, work
);
3529 struct inode
*inode
= io
->inode
;
3532 mutex_lock(&inode
->i_mutex
);
3533 ret
= ext4_end_aio_dio_nolock(io
);
3535 if (!list_empty(&io
->list
))
3536 list_del_init(&io
->list
);
3537 ext4_free_io_end(io
);
3539 mutex_unlock(&inode
->i_mutex
);
3542 * This function is called from ext4_sync_file().
3544 * When AIO DIO IO is completed, the work to convert unwritten
3545 * extents to written is queued on workqueue but may not get immediately
3546 * scheduled. When fsync is called, we need to ensure the
3547 * conversion is complete before fsync returns.
3548 * The inode keeps track of a list of completed AIO from DIO path
3549 * that might needs to do the conversion. This function walks through
3550 * the list and convert the related unwritten extents to written.
3552 int flush_aio_dio_completed_IO(struct inode
*inode
)
3558 if (list_empty(&EXT4_I(inode
)->i_aio_dio_complete_list
))
3561 dump_aio_dio_list(inode
);
3562 while (!list_empty(&EXT4_I(inode
)->i_aio_dio_complete_list
)){
3563 io
= list_entry(EXT4_I(inode
)->i_aio_dio_complete_list
.next
,
3564 ext4_io_end_t
, list
);
3566 * Calling ext4_end_aio_dio_nolock() to convert completed
3569 * When ext4_sync_file() is called, run_queue() may already
3570 * about to flush the work corresponding to this io structure.
3571 * It will be upset if it founds the io structure related
3572 * to the work-to-be schedule is freed.
3574 * Thus we need to keep the io structure still valid here after
3575 * convertion finished. The io structure has a flag to
3576 * avoid double converting from both fsync and background work
3579 ret
= ext4_end_aio_dio_nolock(io
);
3583 list_del_init(&io
->list
);
3585 return (ret2
< 0) ? ret2
: 0;
3588 static ext4_io_end_t
*ext4_init_io_end (struct inode
*inode
)
3590 ext4_io_end_t
*io
= NULL
;
3592 io
= kmalloc(sizeof(*io
), GFP_NOFS
);
3601 INIT_WORK(&io
->work
, ext4_end_aio_dio_work
);
3602 INIT_LIST_HEAD(&io
->list
);
3608 static void ext4_end_io_dio(struct kiocb
*iocb
, loff_t offset
,
3609 ssize_t size
, void *private)
3611 ext4_io_end_t
*io_end
= iocb
->private;
3612 struct workqueue_struct
*wq
;
3614 /* if not async direct IO or dio with 0 bytes write, just return */
3615 if (!io_end
|| !size
)
3618 ext_debug("ext4_end_io_dio(): io_end 0x%p"
3619 "for inode %lu, iocb 0x%p, offset %llu, size %llu\n",
3620 iocb
->private, io_end
->inode
->i_ino
, iocb
, offset
,
3623 /* if not aio dio with unwritten extents, just free io and return */
3624 if (io_end
->flag
!= DIO_AIO_UNWRITTEN
){
3625 ext4_free_io_end(io_end
);
3626 iocb
->private = NULL
;
3630 io_end
->offset
= offset
;
3631 io_end
->size
= size
;
3632 wq
= EXT4_SB(io_end
->inode
->i_sb
)->dio_unwritten_wq
;
3634 /* queue the work to convert unwritten extents to written */
3635 queue_work(wq
, &io_end
->work
);
3637 /* Add the io_end to per-inode completed aio dio list*/
3638 list_add_tail(&io_end
->list
,
3639 &EXT4_I(io_end
->inode
)->i_aio_dio_complete_list
);
3640 iocb
->private = NULL
;
3643 * For ext4 extent files, ext4 will do direct-io write to holes,
3644 * preallocated extents, and those write extend the file, no need to
3645 * fall back to buffered IO.
3647 * For holes, we fallocate those blocks, mark them as unintialized
3648 * If those blocks were preallocated, we mark sure they are splited, but
3649 * still keep the range to write as unintialized.
3651 * The unwrritten extents will be converted to written when DIO is completed.
3652 * For async direct IO, since the IO may still pending when return, we
3653 * set up an end_io call back function, which will do the convertion
3654 * when async direct IO completed.
3656 * If the O_DIRECT write will extend the file then add this inode to the
3657 * orphan list. So recovery will truncate it back to the original size
3658 * if the machine crashes during the write.
3661 static ssize_t
ext4_ext_direct_IO(int rw
, struct kiocb
*iocb
,
3662 const struct iovec
*iov
, loff_t offset
,
3663 unsigned long nr_segs
)
3665 struct file
*file
= iocb
->ki_filp
;
3666 struct inode
*inode
= file
->f_mapping
->host
;
3668 size_t count
= iov_length(iov
, nr_segs
);
3670 loff_t final_size
= offset
+ count
;
3671 if (rw
== WRITE
&& final_size
<= inode
->i_size
) {
3673 * We could direct write to holes and fallocate.
3675 * Allocated blocks to fill the hole are marked as uninitialized
3676 * to prevent paralel buffered read to expose the stale data
3677 * before DIO complete the data IO.
3679 * As to previously fallocated extents, ext4 get_block
3680 * will just simply mark the buffer mapped but still
3681 * keep the extents uninitialized.
3683 * for non AIO case, we will convert those unwritten extents
3684 * to written after return back from blockdev_direct_IO.
3686 * for async DIO, the conversion needs to be defered when
3687 * the IO is completed. The ext4 end_io callback function
3688 * will be called to take care of the conversion work.
3689 * Here for async case, we allocate an io_end structure to
3692 iocb
->private = NULL
;
3693 EXT4_I(inode
)->cur_aio_dio
= NULL
;
3694 if (!is_sync_kiocb(iocb
)) {
3695 iocb
->private = ext4_init_io_end(inode
);
3699 * we save the io structure for current async
3700 * direct IO, so that later ext4_get_blocks()
3701 * could flag the io structure whether there
3702 * is a unwritten extents needs to be converted
3703 * when IO is completed.
3705 EXT4_I(inode
)->cur_aio_dio
= iocb
->private;
3708 ret
= blockdev_direct_IO(rw
, iocb
, inode
,
3709 inode
->i_sb
->s_bdev
, iov
,
3711 ext4_get_block_dio_write
,
3714 EXT4_I(inode
)->cur_aio_dio
= NULL
;
3716 * The io_end structure takes a reference to the inode,
3717 * that structure needs to be destroyed and the
3718 * reference to the inode need to be dropped, when IO is
3719 * complete, even with 0 byte write, or failed.
3721 * In the successful AIO DIO case, the io_end structure will be
3722 * desctroyed and the reference to the inode will be dropped
3723 * after the end_io call back function is called.
3725 * In the case there is 0 byte write, or error case, since
3726 * VFS direct IO won't invoke the end_io call back function,
3727 * we need to free the end_io structure here.
3729 if (ret
!= -EIOCBQUEUED
&& ret
<= 0 && iocb
->private) {
3730 ext4_free_io_end(iocb
->private);
3731 iocb
->private = NULL
;
3732 } else if (ret
> 0 && (EXT4_I(inode
)->i_state
&
3733 EXT4_STATE_DIO_UNWRITTEN
)) {
3736 * for non AIO case, since the IO is already
3737 * completed, we could do the convertion right here
3739 err
= ext4_convert_unwritten_extents(inode
,
3743 EXT4_I(inode
)->i_state
&= ~EXT4_STATE_DIO_UNWRITTEN
;
3748 /* for write the the end of file case, we fall back to old way */
3749 return ext4_ind_direct_IO(rw
, iocb
, iov
, offset
, nr_segs
);
3752 static ssize_t
ext4_direct_IO(int rw
, struct kiocb
*iocb
,
3753 const struct iovec
*iov
, loff_t offset
,
3754 unsigned long nr_segs
)
3756 struct file
*file
= iocb
->ki_filp
;
3757 struct inode
*inode
= file
->f_mapping
->host
;
3759 if (EXT4_I(inode
)->i_flags
& EXT4_EXTENTS_FL
)
3760 return ext4_ext_direct_IO(rw
, iocb
, iov
, offset
, nr_segs
);
3762 return ext4_ind_direct_IO(rw
, iocb
, iov
, offset
, nr_segs
);
3766 * Pages can be marked dirty completely asynchronously from ext4's journalling
3767 * activity. By filemap_sync_pte(), try_to_unmap_one(), etc. We cannot do
3768 * much here because ->set_page_dirty is called under VFS locks. The page is
3769 * not necessarily locked.
3771 * We cannot just dirty the page and leave attached buffers clean, because the
3772 * buffers' dirty state is "definitive". We cannot just set the buffers dirty
3773 * or jbddirty because all the journalling code will explode.
3775 * So what we do is to mark the page "pending dirty" and next time writepage
3776 * is called, propagate that into the buffers appropriately.
3778 static int ext4_journalled_set_page_dirty(struct page
*page
)
3780 SetPageChecked(page
);
3781 return __set_page_dirty_nobuffers(page
);
3784 static const struct address_space_operations ext4_ordered_aops
= {
3785 .readpage
= ext4_readpage
,
3786 .readpages
= ext4_readpages
,
3787 .writepage
= ext4_writepage
,
3788 .sync_page
= block_sync_page
,
3789 .write_begin
= ext4_write_begin
,
3790 .write_end
= ext4_ordered_write_end
,
3792 .invalidatepage
= ext4_invalidatepage
,
3793 .releasepage
= ext4_releasepage
,
3794 .direct_IO
= ext4_direct_IO
,
3795 .migratepage
= buffer_migrate_page
,
3796 .is_partially_uptodate
= block_is_partially_uptodate
,
3797 .error_remove_page
= generic_error_remove_page
,
3800 static const struct address_space_operations ext4_writeback_aops
= {
3801 .readpage
= ext4_readpage
,
3802 .readpages
= ext4_readpages
,
3803 .writepage
= ext4_writepage
,
3804 .sync_page
= block_sync_page
,
3805 .write_begin
= ext4_write_begin
,
3806 .write_end
= ext4_writeback_write_end
,
3808 .invalidatepage
= ext4_invalidatepage
,
3809 .releasepage
= ext4_releasepage
,
3810 .direct_IO
= ext4_direct_IO
,
3811 .migratepage
= buffer_migrate_page
,
3812 .is_partially_uptodate
= block_is_partially_uptodate
,
3813 .error_remove_page
= generic_error_remove_page
,
3816 static const struct address_space_operations ext4_journalled_aops
= {
3817 .readpage
= ext4_readpage
,
3818 .readpages
= ext4_readpages
,
3819 .writepage
= ext4_writepage
,
3820 .sync_page
= block_sync_page
,
3821 .write_begin
= ext4_write_begin
,
3822 .write_end
= ext4_journalled_write_end
,
3823 .set_page_dirty
= ext4_journalled_set_page_dirty
,
3825 .invalidatepage
= ext4_invalidatepage
,
3826 .releasepage
= ext4_releasepage
,
3827 .is_partially_uptodate
= block_is_partially_uptodate
,
3828 .error_remove_page
= generic_error_remove_page
,
3831 static const struct address_space_operations ext4_da_aops
= {
3832 .readpage
= ext4_readpage
,
3833 .readpages
= ext4_readpages
,
3834 .writepage
= ext4_writepage
,
3835 .writepages
= ext4_da_writepages
,
3836 .sync_page
= block_sync_page
,
3837 .write_begin
= ext4_da_write_begin
,
3838 .write_end
= ext4_da_write_end
,
3840 .invalidatepage
= ext4_da_invalidatepage
,
3841 .releasepage
= ext4_releasepage
,
3842 .direct_IO
= ext4_direct_IO
,
3843 .migratepage
= buffer_migrate_page
,
3844 .is_partially_uptodate
= block_is_partially_uptodate
,
3845 .error_remove_page
= generic_error_remove_page
,
3848 void ext4_set_aops(struct inode
*inode
)
3850 if (ext4_should_order_data(inode
) &&
3851 test_opt(inode
->i_sb
, DELALLOC
))
3852 inode
->i_mapping
->a_ops
= &ext4_da_aops
;
3853 else if (ext4_should_order_data(inode
))
3854 inode
->i_mapping
->a_ops
= &ext4_ordered_aops
;
3855 else if (ext4_should_writeback_data(inode
) &&
3856 test_opt(inode
->i_sb
, DELALLOC
))
3857 inode
->i_mapping
->a_ops
= &ext4_da_aops
;
3858 else if (ext4_should_writeback_data(inode
))
3859 inode
->i_mapping
->a_ops
= &ext4_writeback_aops
;
3861 inode
->i_mapping
->a_ops
= &ext4_journalled_aops
;
3865 * ext4_block_truncate_page() zeroes out a mapping from file offset `from'
3866 * up to the end of the block which corresponds to `from'.
3867 * This required during truncate. We need to physically zero the tail end
3868 * of that block so it doesn't yield old data if the file is later grown.
3870 int ext4_block_truncate_page(handle_t
*handle
,
3871 struct address_space
*mapping
, loff_t from
)
3873 ext4_fsblk_t index
= from
>> PAGE_CACHE_SHIFT
;
3874 unsigned offset
= from
& (PAGE_CACHE_SIZE
-1);
3875 unsigned blocksize
, length
, pos
;
3877 struct inode
*inode
= mapping
->host
;
3878 struct buffer_head
*bh
;
3882 page
= find_or_create_page(mapping
, from
>> PAGE_CACHE_SHIFT
,
3883 mapping_gfp_mask(mapping
) & ~__GFP_FS
);
3887 blocksize
= inode
->i_sb
->s_blocksize
;
3888 length
= blocksize
- (offset
& (blocksize
- 1));
3889 iblock
= index
<< (PAGE_CACHE_SHIFT
- inode
->i_sb
->s_blocksize_bits
);
3892 * For "nobh" option, we can only work if we don't need to
3893 * read-in the page - otherwise we create buffers to do the IO.
3895 if (!page_has_buffers(page
) && test_opt(inode
->i_sb
, NOBH
) &&
3896 ext4_should_writeback_data(inode
) && PageUptodate(page
)) {
3897 zero_user(page
, offset
, length
);
3898 set_page_dirty(page
);
3902 if (!page_has_buffers(page
))
3903 create_empty_buffers(page
, blocksize
, 0);
3905 /* Find the buffer that contains "offset" */
3906 bh
= page_buffers(page
);
3908 while (offset
>= pos
) {
3909 bh
= bh
->b_this_page
;
3915 if (buffer_freed(bh
)) {
3916 BUFFER_TRACE(bh
, "freed: skip");
3920 if (!buffer_mapped(bh
)) {
3921 BUFFER_TRACE(bh
, "unmapped");
3922 ext4_get_block(inode
, iblock
, bh
, 0);
3923 /* unmapped? It's a hole - nothing to do */
3924 if (!buffer_mapped(bh
)) {
3925 BUFFER_TRACE(bh
, "still unmapped");
3930 /* Ok, it's mapped. Make sure it's up-to-date */
3931 if (PageUptodate(page
))
3932 set_buffer_uptodate(bh
);
3934 if (!buffer_uptodate(bh
)) {
3936 ll_rw_block(READ
, 1, &bh
);
3938 /* Uhhuh. Read error. Complain and punt. */
3939 if (!buffer_uptodate(bh
))
3943 if (ext4_should_journal_data(inode
)) {
3944 BUFFER_TRACE(bh
, "get write access");
3945 err
= ext4_journal_get_write_access(handle
, bh
);
3950 zero_user(page
, offset
, length
);
3952 BUFFER_TRACE(bh
, "zeroed end of block");
3955 if (ext4_should_journal_data(inode
)) {
3956 err
= ext4_handle_dirty_metadata(handle
, inode
, bh
);
3958 if (ext4_should_order_data(inode
))
3959 err
= ext4_jbd2_file_inode(handle
, inode
);
3960 mark_buffer_dirty(bh
);
3965 page_cache_release(page
);
3970 * Probably it should be a library function... search for first non-zero word
3971 * or memcmp with zero_page, whatever is better for particular architecture.
3974 static inline int all_zeroes(__le32
*p
, __le32
*q
)
3983 * ext4_find_shared - find the indirect blocks for partial truncation.
3984 * @inode: inode in question
3985 * @depth: depth of the affected branch
3986 * @offsets: offsets of pointers in that branch (see ext4_block_to_path)
3987 * @chain: place to store the pointers to partial indirect blocks
3988 * @top: place to the (detached) top of branch
3990 * This is a helper function used by ext4_truncate().
3992 * When we do truncate() we may have to clean the ends of several
3993 * indirect blocks but leave the blocks themselves alive. Block is
3994 * partially truncated if some data below the new i_size is refered
3995 * from it (and it is on the path to the first completely truncated
3996 * data block, indeed). We have to free the top of that path along
3997 * with everything to the right of the path. Since no allocation
3998 * past the truncation point is possible until ext4_truncate()
3999 * finishes, we may safely do the latter, but top of branch may
4000 * require special attention - pageout below the truncation point
4001 * might try to populate it.
4003 * We atomically detach the top of branch from the tree, store the
4004 * block number of its root in *@top, pointers to buffer_heads of
4005 * partially truncated blocks - in @chain[].bh and pointers to
4006 * their last elements that should not be removed - in
4007 * @chain[].p. Return value is the pointer to last filled element
4010 * The work left to caller to do the actual freeing of subtrees:
4011 * a) free the subtree starting from *@top
4012 * b) free the subtrees whose roots are stored in
4013 * (@chain[i].p+1 .. end of @chain[i].bh->b_data)
4014 * c) free the subtrees growing from the inode past the @chain[0].
4015 * (no partially truncated stuff there). */
4017 static Indirect
*ext4_find_shared(struct inode
*inode
, int depth
,
4018 ext4_lblk_t offsets
[4], Indirect chain
[4],
4021 Indirect
*partial
, *p
;
4025 /* Make k index the deepest non-null offest + 1 */
4026 for (k
= depth
; k
> 1 && !offsets
[k
-1]; k
--)
4028 partial
= ext4_get_branch(inode
, k
, offsets
, chain
, &err
);
4029 /* Writer: pointers */
4031 partial
= chain
+ k
-1;
4033 * If the branch acquired continuation since we've looked at it -
4034 * fine, it should all survive and (new) top doesn't belong to us.
4036 if (!partial
->key
&& *partial
->p
)
4039 for (p
= partial
; (p
> chain
) && all_zeroes((__le32
*) p
->bh
->b_data
, p
->p
); p
--)
4042 * OK, we've found the last block that must survive. The rest of our
4043 * branch should be detached before unlocking. However, if that rest
4044 * of branch is all ours and does not grow immediately from the inode
4045 * it's easier to cheat and just decrement partial->p.
4047 if (p
== chain
+ k
- 1 && p
> chain
) {
4051 /* Nope, don't do this in ext4. Must leave the tree intact */
4058 while (partial
> p
) {
4059 brelse(partial
->bh
);
4067 * Zero a number of block pointers in either an inode or an indirect block.
4068 * If we restart the transaction we must again get write access to the
4069 * indirect block for further modification.
4071 * We release `count' blocks on disk, but (last - first) may be greater
4072 * than `count' because there can be holes in there.
4074 static void ext4_clear_blocks(handle_t
*handle
, struct inode
*inode
,
4075 struct buffer_head
*bh
,
4076 ext4_fsblk_t block_to_free
,
4077 unsigned long count
, __le32
*first
,
4081 int flags
= EXT4_FREE_BLOCKS_FORGET
;
4083 if (S_ISDIR(inode
->i_mode
) || S_ISLNK(inode
->i_mode
))
4084 flags
|= EXT4_FREE_BLOCKS_METADATA
;
4086 if (try_to_extend_transaction(handle
, inode
)) {
4088 BUFFER_TRACE(bh
, "call ext4_handle_dirty_metadata");
4089 ext4_handle_dirty_metadata(handle
, inode
, bh
);
4091 ext4_mark_inode_dirty(handle
, inode
);
4092 ext4_truncate_restart_trans(handle
, inode
,
4093 blocks_for_truncate(inode
));
4095 BUFFER_TRACE(bh
, "retaking write access");
4096 ext4_journal_get_write_access(handle
, bh
);
4100 for (p
= first
; p
< last
; p
++)
4103 ext4_free_blocks(handle
, inode
, 0, block_to_free
, count
, flags
);
4107 * ext4_free_data - free a list of data blocks
4108 * @handle: handle for this transaction
4109 * @inode: inode we are dealing with
4110 * @this_bh: indirect buffer_head which contains *@first and *@last
4111 * @first: array of block numbers
4112 * @last: points immediately past the end of array
4114 * We are freeing all blocks refered from that array (numbers are stored as
4115 * little-endian 32-bit) and updating @inode->i_blocks appropriately.
4117 * We accumulate contiguous runs of blocks to free. Conveniently, if these
4118 * blocks are contiguous then releasing them at one time will only affect one
4119 * or two bitmap blocks (+ group descriptor(s) and superblock) and we won't
4120 * actually use a lot of journal space.
4122 * @this_bh will be %NULL if @first and @last point into the inode's direct
4125 static void ext4_free_data(handle_t
*handle
, struct inode
*inode
,
4126 struct buffer_head
*this_bh
,
4127 __le32
*first
, __le32
*last
)
4129 ext4_fsblk_t block_to_free
= 0; /* Starting block # of a run */
4130 unsigned long count
= 0; /* Number of blocks in the run */
4131 __le32
*block_to_free_p
= NULL
; /* Pointer into inode/ind
4134 ext4_fsblk_t nr
; /* Current block # */
4135 __le32
*p
; /* Pointer into inode/ind
4136 for current block */
4139 if (this_bh
) { /* For indirect block */
4140 BUFFER_TRACE(this_bh
, "get_write_access");
4141 err
= ext4_journal_get_write_access(handle
, this_bh
);
4142 /* Important: if we can't update the indirect pointers
4143 * to the blocks, we can't free them. */
4148 for (p
= first
; p
< last
; p
++) {
4149 nr
= le32_to_cpu(*p
);
4151 /* accumulate blocks to free if they're contiguous */
4154 block_to_free_p
= p
;
4156 } else if (nr
== block_to_free
+ count
) {
4159 ext4_clear_blocks(handle
, inode
, this_bh
,
4161 count
, block_to_free_p
, p
);
4163 block_to_free_p
= p
;
4170 ext4_clear_blocks(handle
, inode
, this_bh
, block_to_free
,
4171 count
, block_to_free_p
, p
);
4174 BUFFER_TRACE(this_bh
, "call ext4_handle_dirty_metadata");
4177 * The buffer head should have an attached journal head at this
4178 * point. However, if the data is corrupted and an indirect
4179 * block pointed to itself, it would have been detached when
4180 * the block was cleared. Check for this instead of OOPSing.
4182 if ((EXT4_JOURNAL(inode
) == NULL
) || bh2jh(this_bh
))
4183 ext4_handle_dirty_metadata(handle
, inode
, this_bh
);
4185 ext4_error(inode
->i_sb
, __func__
,
4186 "circular indirect block detected, "
4187 "inode=%lu, block=%llu",
4189 (unsigned long long) this_bh
->b_blocknr
);
4194 * ext4_free_branches - free an array of branches
4195 * @handle: JBD handle for this transaction
4196 * @inode: inode we are dealing with
4197 * @parent_bh: the buffer_head which contains *@first and *@last
4198 * @first: array of block numbers
4199 * @last: pointer immediately past the end of array
4200 * @depth: depth of the branches to free
4202 * We are freeing all blocks refered from these branches (numbers are
4203 * stored as little-endian 32-bit) and updating @inode->i_blocks
4206 static void ext4_free_branches(handle_t
*handle
, struct inode
*inode
,
4207 struct buffer_head
*parent_bh
,
4208 __le32
*first
, __le32
*last
, int depth
)
4213 if (ext4_handle_is_aborted(handle
))
4217 struct buffer_head
*bh
;
4218 int addr_per_block
= EXT4_ADDR_PER_BLOCK(inode
->i_sb
);
4220 while (--p
>= first
) {
4221 nr
= le32_to_cpu(*p
);
4223 continue; /* A hole */
4225 /* Go read the buffer for the next level down */
4226 bh
= sb_bread(inode
->i_sb
, nr
);
4229 * A read failure? Report error and clear slot
4233 ext4_error(inode
->i_sb
, "ext4_free_branches",
4234 "Read failure, inode=%lu, block=%llu",
4239 /* This zaps the entire block. Bottom up. */
4240 BUFFER_TRACE(bh
, "free child branches");
4241 ext4_free_branches(handle
, inode
, bh
,
4242 (__le32
*) bh
->b_data
,
4243 (__le32
*) bh
->b_data
+ addr_per_block
,
4247 * We've probably journalled the indirect block several
4248 * times during the truncate. But it's no longer
4249 * needed and we now drop it from the transaction via
4250 * jbd2_journal_revoke().
4252 * That's easy if it's exclusively part of this
4253 * transaction. But if it's part of the committing
4254 * transaction then jbd2_journal_forget() will simply
4255 * brelse() it. That means that if the underlying
4256 * block is reallocated in ext4_get_block(),
4257 * unmap_underlying_metadata() will find this block
4258 * and will try to get rid of it. damn, damn.
4260 * If this block has already been committed to the
4261 * journal, a revoke record will be written. And
4262 * revoke records must be emitted *before* clearing
4263 * this block's bit in the bitmaps.
4265 ext4_forget(handle
, 1, inode
, bh
, bh
->b_blocknr
);
4268 * Everything below this this pointer has been
4269 * released. Now let this top-of-subtree go.
4271 * We want the freeing of this indirect block to be
4272 * atomic in the journal with the updating of the
4273 * bitmap block which owns it. So make some room in
4276 * We zero the parent pointer *after* freeing its
4277 * pointee in the bitmaps, so if extend_transaction()
4278 * for some reason fails to put the bitmap changes and
4279 * the release into the same transaction, recovery
4280 * will merely complain about releasing a free block,
4281 * rather than leaking blocks.
4283 if (ext4_handle_is_aborted(handle
))
4285 if (try_to_extend_transaction(handle
, inode
)) {
4286 ext4_mark_inode_dirty(handle
, inode
);
4287 ext4_truncate_restart_trans(handle
, inode
,
4288 blocks_for_truncate(inode
));
4291 ext4_free_blocks(handle
, inode
, 0, nr
, 1,
4292 EXT4_FREE_BLOCKS_METADATA
);
4296 * The block which we have just freed is
4297 * pointed to by an indirect block: journal it
4299 BUFFER_TRACE(parent_bh
, "get_write_access");
4300 if (!ext4_journal_get_write_access(handle
,
4303 BUFFER_TRACE(parent_bh
,
4304 "call ext4_handle_dirty_metadata");
4305 ext4_handle_dirty_metadata(handle
,
4312 /* We have reached the bottom of the tree. */
4313 BUFFER_TRACE(parent_bh
, "free data blocks");
4314 ext4_free_data(handle
, inode
, parent_bh
, first
, last
);
4318 int ext4_can_truncate(struct inode
*inode
)
4320 if (IS_APPEND(inode
) || IS_IMMUTABLE(inode
))
4322 if (S_ISREG(inode
->i_mode
))
4324 if (S_ISDIR(inode
->i_mode
))
4326 if (S_ISLNK(inode
->i_mode
))
4327 return !ext4_inode_is_fast_symlink(inode
);
4334 * We block out ext4_get_block() block instantiations across the entire
4335 * transaction, and VFS/VM ensures that ext4_truncate() cannot run
4336 * simultaneously on behalf of the same inode.
4338 * As we work through the truncate and commmit bits of it to the journal there
4339 * is one core, guiding principle: the file's tree must always be consistent on
4340 * disk. We must be able to restart the truncate after a crash.
4342 * The file's tree may be transiently inconsistent in memory (although it
4343 * probably isn't), but whenever we close off and commit a journal transaction,
4344 * the contents of (the filesystem + the journal) must be consistent and
4345 * restartable. It's pretty simple, really: bottom up, right to left (although
4346 * left-to-right works OK too).
4348 * Note that at recovery time, journal replay occurs *before* the restart of
4349 * truncate against the orphan inode list.
4351 * The committed inode has the new, desired i_size (which is the same as
4352 * i_disksize in this case). After a crash, ext4_orphan_cleanup() will see
4353 * that this inode's truncate did not complete and it will again call
4354 * ext4_truncate() to have another go. So there will be instantiated blocks
4355 * to the right of the truncation point in a crashed ext4 filesystem. But
4356 * that's fine - as long as they are linked from the inode, the post-crash
4357 * ext4_truncate() run will find them and release them.
4359 void ext4_truncate(struct inode
*inode
)
4362 struct ext4_inode_info
*ei
= EXT4_I(inode
);
4363 __le32
*i_data
= ei
->i_data
;
4364 int addr_per_block
= EXT4_ADDR_PER_BLOCK(inode
->i_sb
);
4365 struct address_space
*mapping
= inode
->i_mapping
;
4366 ext4_lblk_t offsets
[4];
4371 ext4_lblk_t last_block
;
4372 unsigned blocksize
= inode
->i_sb
->s_blocksize
;
4374 if (!ext4_can_truncate(inode
))
4377 if (inode
->i_size
== 0 && !test_opt(inode
->i_sb
, NO_AUTO_DA_ALLOC
))
4378 ei
->i_state
|= EXT4_STATE_DA_ALLOC_CLOSE
;
4380 if (EXT4_I(inode
)->i_flags
& EXT4_EXTENTS_FL
) {
4381 ext4_ext_truncate(inode
);
4385 handle
= start_transaction(inode
);
4387 return; /* AKPM: return what? */
4389 last_block
= (inode
->i_size
+ blocksize
-1)
4390 >> EXT4_BLOCK_SIZE_BITS(inode
->i_sb
);
4392 if (inode
->i_size
& (blocksize
- 1))
4393 if (ext4_block_truncate_page(handle
, mapping
, inode
->i_size
))
4396 n
= ext4_block_to_path(inode
, last_block
, offsets
, NULL
);
4398 goto out_stop
; /* error */
4401 * OK. This truncate is going to happen. We add the inode to the
4402 * orphan list, so that if this truncate spans multiple transactions,
4403 * and we crash, we will resume the truncate when the filesystem
4404 * recovers. It also marks the inode dirty, to catch the new size.
4406 * Implication: the file must always be in a sane, consistent
4407 * truncatable state while each transaction commits.
4409 if (ext4_orphan_add(handle
, inode
))
4413 * From here we block out all ext4_get_block() callers who want to
4414 * modify the block allocation tree.
4416 down_write(&ei
->i_data_sem
);
4418 ext4_discard_preallocations(inode
);
4421 * The orphan list entry will now protect us from any crash which
4422 * occurs before the truncate completes, so it is now safe to propagate
4423 * the new, shorter inode size (held for now in i_size) into the
4424 * on-disk inode. We do this via i_disksize, which is the value which
4425 * ext4 *really* writes onto the disk inode.
4427 ei
->i_disksize
= inode
->i_size
;
4429 if (n
== 1) { /* direct blocks */
4430 ext4_free_data(handle
, inode
, NULL
, i_data
+offsets
[0],
4431 i_data
+ EXT4_NDIR_BLOCKS
);
4435 partial
= ext4_find_shared(inode
, n
, offsets
, chain
, &nr
);
4436 /* Kill the top of shared branch (not detached) */
4438 if (partial
== chain
) {
4439 /* Shared branch grows from the inode */
4440 ext4_free_branches(handle
, inode
, NULL
,
4441 &nr
, &nr
+1, (chain
+n
-1) - partial
);
4444 * We mark the inode dirty prior to restart,
4445 * and prior to stop. No need for it here.
4448 /* Shared branch grows from an indirect block */
4449 BUFFER_TRACE(partial
->bh
, "get_write_access");
4450 ext4_free_branches(handle
, inode
, partial
->bh
,
4452 partial
->p
+1, (chain
+n
-1) - partial
);
4455 /* Clear the ends of indirect blocks on the shared branch */
4456 while (partial
> chain
) {
4457 ext4_free_branches(handle
, inode
, partial
->bh
, partial
->p
+ 1,
4458 (__le32
*)partial
->bh
->b_data
+addr_per_block
,
4459 (chain
+n
-1) - partial
);
4460 BUFFER_TRACE(partial
->bh
, "call brelse");
4461 brelse(partial
->bh
);
4465 /* Kill the remaining (whole) subtrees */
4466 switch (offsets
[0]) {
4468 nr
= i_data
[EXT4_IND_BLOCK
];
4470 ext4_free_branches(handle
, inode
, NULL
, &nr
, &nr
+1, 1);
4471 i_data
[EXT4_IND_BLOCK
] = 0;
4473 case EXT4_IND_BLOCK
:
4474 nr
= i_data
[EXT4_DIND_BLOCK
];
4476 ext4_free_branches(handle
, inode
, NULL
, &nr
, &nr
+1, 2);
4477 i_data
[EXT4_DIND_BLOCK
] = 0;
4479 case EXT4_DIND_BLOCK
:
4480 nr
= i_data
[EXT4_TIND_BLOCK
];
4482 ext4_free_branches(handle
, inode
, NULL
, &nr
, &nr
+1, 3);
4483 i_data
[EXT4_TIND_BLOCK
] = 0;
4485 case EXT4_TIND_BLOCK
:
4489 up_write(&ei
->i_data_sem
);
4490 inode
->i_mtime
= inode
->i_ctime
= ext4_current_time(inode
);
4491 ext4_mark_inode_dirty(handle
, inode
);
4494 * In a multi-transaction truncate, we only make the final transaction
4498 ext4_handle_sync(handle
);
4501 * If this was a simple ftruncate(), and the file will remain alive
4502 * then we need to clear up the orphan record which we created above.
4503 * However, if this was a real unlink then we were called by
4504 * ext4_delete_inode(), and we allow that function to clean up the
4505 * orphan info for us.
4508 ext4_orphan_del(handle
, inode
);
4510 ext4_journal_stop(handle
);
4514 * ext4_get_inode_loc returns with an extra refcount against the inode's
4515 * underlying buffer_head on success. If 'in_mem' is true, we have all
4516 * data in memory that is needed to recreate the on-disk version of this
4519 static int __ext4_get_inode_loc(struct inode
*inode
,
4520 struct ext4_iloc
*iloc
, int in_mem
)
4522 struct ext4_group_desc
*gdp
;
4523 struct buffer_head
*bh
;
4524 struct super_block
*sb
= inode
->i_sb
;
4526 int inodes_per_block
, inode_offset
;
4529 if (!ext4_valid_inum(sb
, inode
->i_ino
))
4532 iloc
->block_group
= (inode
->i_ino
- 1) / EXT4_INODES_PER_GROUP(sb
);
4533 gdp
= ext4_get_group_desc(sb
, iloc
->block_group
, NULL
);
4538 * Figure out the offset within the block group inode table
4540 inodes_per_block
= (EXT4_BLOCK_SIZE(sb
) / EXT4_INODE_SIZE(sb
));
4541 inode_offset
= ((inode
->i_ino
- 1) %
4542 EXT4_INODES_PER_GROUP(sb
));
4543 block
= ext4_inode_table(sb
, gdp
) + (inode_offset
/ inodes_per_block
);
4544 iloc
->offset
= (inode_offset
% inodes_per_block
) * EXT4_INODE_SIZE(sb
);
4546 bh
= sb_getblk(sb
, block
);
4548 ext4_error(sb
, "ext4_get_inode_loc", "unable to read "
4549 "inode block - inode=%lu, block=%llu",
4550 inode
->i_ino
, block
);
4553 if (!buffer_uptodate(bh
)) {
4557 * If the buffer has the write error flag, we have failed
4558 * to write out another inode in the same block. In this
4559 * case, we don't have to read the block because we may
4560 * read the old inode data successfully.
4562 if (buffer_write_io_error(bh
) && !buffer_uptodate(bh
))
4563 set_buffer_uptodate(bh
);
4565 if (buffer_uptodate(bh
)) {
4566 /* someone brought it uptodate while we waited */
4572 * If we have all information of the inode in memory and this
4573 * is the only valid inode in the block, we need not read the
4577 struct buffer_head
*bitmap_bh
;
4580 start
= inode_offset
& ~(inodes_per_block
- 1);
4582 /* Is the inode bitmap in cache? */
4583 bitmap_bh
= sb_getblk(sb
, ext4_inode_bitmap(sb
, gdp
));
4588 * If the inode bitmap isn't in cache then the
4589 * optimisation may end up performing two reads instead
4590 * of one, so skip it.
4592 if (!buffer_uptodate(bitmap_bh
)) {
4596 for (i
= start
; i
< start
+ inodes_per_block
; i
++) {
4597 if (i
== inode_offset
)
4599 if (ext4_test_bit(i
, bitmap_bh
->b_data
))
4603 if (i
== start
+ inodes_per_block
) {
4604 /* all other inodes are free, so skip I/O */
4605 memset(bh
->b_data
, 0, bh
->b_size
);
4606 set_buffer_uptodate(bh
);
4614 * If we need to do any I/O, try to pre-readahead extra
4615 * blocks from the inode table.
4617 if (EXT4_SB(sb
)->s_inode_readahead_blks
) {
4618 ext4_fsblk_t b
, end
, table
;
4621 table
= ext4_inode_table(sb
, gdp
);
4622 /* s_inode_readahead_blks is always a power of 2 */
4623 b
= block
& ~(EXT4_SB(sb
)->s_inode_readahead_blks
-1);
4626 end
= b
+ EXT4_SB(sb
)->s_inode_readahead_blks
;
4627 num
= EXT4_INODES_PER_GROUP(sb
);
4628 if (EXT4_HAS_RO_COMPAT_FEATURE(sb
,
4629 EXT4_FEATURE_RO_COMPAT_GDT_CSUM
))
4630 num
-= ext4_itable_unused_count(sb
, gdp
);
4631 table
+= num
/ inodes_per_block
;
4635 sb_breadahead(sb
, b
++);
4639 * There are other valid inodes in the buffer, this inode
4640 * has in-inode xattrs, or we don't have this inode in memory.
4641 * Read the block from disk.
4644 bh
->b_end_io
= end_buffer_read_sync
;
4645 submit_bh(READ_META
, bh
);
4647 if (!buffer_uptodate(bh
)) {
4648 ext4_error(sb
, __func__
,
4649 "unable to read inode block - inode=%lu, "
4650 "block=%llu", inode
->i_ino
, block
);
4660 int ext4_get_inode_loc(struct inode
*inode
, struct ext4_iloc
*iloc
)
4662 /* We have all inode data except xattrs in memory here. */
4663 return __ext4_get_inode_loc(inode
, iloc
,
4664 !(EXT4_I(inode
)->i_state
& EXT4_STATE_XATTR
));
4667 void ext4_set_inode_flags(struct inode
*inode
)
4669 unsigned int flags
= EXT4_I(inode
)->i_flags
;
4671 inode
->i_flags
&= ~(S_SYNC
|S_APPEND
|S_IMMUTABLE
|S_NOATIME
|S_DIRSYNC
);
4672 if (flags
& EXT4_SYNC_FL
)
4673 inode
->i_flags
|= S_SYNC
;
4674 if (flags
& EXT4_APPEND_FL
)
4675 inode
->i_flags
|= S_APPEND
;
4676 if (flags
& EXT4_IMMUTABLE_FL
)
4677 inode
->i_flags
|= S_IMMUTABLE
;
4678 if (flags
& EXT4_NOATIME_FL
)
4679 inode
->i_flags
|= S_NOATIME
;
4680 if (flags
& EXT4_DIRSYNC_FL
)
4681 inode
->i_flags
|= S_DIRSYNC
;
4684 /* Propagate flags from i_flags to EXT4_I(inode)->i_flags */
4685 void ext4_get_inode_flags(struct ext4_inode_info
*ei
)
4687 unsigned int flags
= ei
->vfs_inode
.i_flags
;
4689 ei
->i_flags
&= ~(EXT4_SYNC_FL
|EXT4_APPEND_FL
|
4690 EXT4_IMMUTABLE_FL
|EXT4_NOATIME_FL
|EXT4_DIRSYNC_FL
);
4692 ei
->i_flags
|= EXT4_SYNC_FL
;
4693 if (flags
& S_APPEND
)
4694 ei
->i_flags
|= EXT4_APPEND_FL
;
4695 if (flags
& S_IMMUTABLE
)
4696 ei
->i_flags
|= EXT4_IMMUTABLE_FL
;
4697 if (flags
& S_NOATIME
)
4698 ei
->i_flags
|= EXT4_NOATIME_FL
;
4699 if (flags
& S_DIRSYNC
)
4700 ei
->i_flags
|= EXT4_DIRSYNC_FL
;
4703 static blkcnt_t
ext4_inode_blocks(struct ext4_inode
*raw_inode
,
4704 struct ext4_inode_info
*ei
)
4707 struct inode
*inode
= &(ei
->vfs_inode
);
4708 struct super_block
*sb
= inode
->i_sb
;
4710 if (EXT4_HAS_RO_COMPAT_FEATURE(sb
,
4711 EXT4_FEATURE_RO_COMPAT_HUGE_FILE
)) {
4712 /* we are using combined 48 bit field */
4713 i_blocks
= ((u64
)le16_to_cpu(raw_inode
->i_blocks_high
)) << 32 |
4714 le32_to_cpu(raw_inode
->i_blocks_lo
);
4715 if (ei
->i_flags
& EXT4_HUGE_FILE_FL
) {
4716 /* i_blocks represent file system block size */
4717 return i_blocks
<< (inode
->i_blkbits
- 9);
4722 return le32_to_cpu(raw_inode
->i_blocks_lo
);
4726 struct inode
*ext4_iget(struct super_block
*sb
, unsigned long ino
)
4728 struct ext4_iloc iloc
;
4729 struct ext4_inode
*raw_inode
;
4730 struct ext4_inode_info
*ei
;
4731 struct inode
*inode
;
4735 inode
= iget_locked(sb
, ino
);
4737 return ERR_PTR(-ENOMEM
);
4738 if (!(inode
->i_state
& I_NEW
))
4744 ret
= __ext4_get_inode_loc(inode
, &iloc
, 0);
4747 raw_inode
= ext4_raw_inode(&iloc
);
4748 inode
->i_mode
= le16_to_cpu(raw_inode
->i_mode
);
4749 inode
->i_uid
= (uid_t
)le16_to_cpu(raw_inode
->i_uid_low
);
4750 inode
->i_gid
= (gid_t
)le16_to_cpu(raw_inode
->i_gid_low
);
4751 if (!(test_opt(inode
->i_sb
, NO_UID32
))) {
4752 inode
->i_uid
|= le16_to_cpu(raw_inode
->i_uid_high
) << 16;
4753 inode
->i_gid
|= le16_to_cpu(raw_inode
->i_gid_high
) << 16;
4755 inode
->i_nlink
= le16_to_cpu(raw_inode
->i_links_count
);
4758 ei
->i_dir_start_lookup
= 0;
4759 ei
->i_dtime
= le32_to_cpu(raw_inode
->i_dtime
);
4760 /* We now have enough fields to check if the inode was active or not.
4761 * This is needed because nfsd might try to access dead inodes
4762 * the test is that same one that e2fsck uses
4763 * NeilBrown 1999oct15
4765 if (inode
->i_nlink
== 0) {
4766 if (inode
->i_mode
== 0 ||
4767 !(EXT4_SB(inode
->i_sb
)->s_mount_state
& EXT4_ORPHAN_FS
)) {
4768 /* this inode is deleted */
4772 /* The only unlinked inodes we let through here have
4773 * valid i_mode and are being read by the orphan
4774 * recovery code: that's fine, we're about to complete
4775 * the process of deleting those. */
4777 ei
->i_flags
= le32_to_cpu(raw_inode
->i_flags
);
4778 inode
->i_blocks
= ext4_inode_blocks(raw_inode
, ei
);
4779 ei
->i_file_acl
= le32_to_cpu(raw_inode
->i_file_acl_lo
);
4780 if (EXT4_HAS_INCOMPAT_FEATURE(sb
, EXT4_FEATURE_INCOMPAT_64BIT
))
4782 ((__u64
)le16_to_cpu(raw_inode
->i_file_acl_high
)) << 32;
4783 inode
->i_size
= ext4_isize(raw_inode
);
4784 ei
->i_disksize
= inode
->i_size
;
4785 inode
->i_generation
= le32_to_cpu(raw_inode
->i_generation
);
4786 ei
->i_block_group
= iloc
.block_group
;
4787 ei
->i_last_alloc_group
= ~0;
4789 * NOTE! The in-memory inode i_data array is in little-endian order
4790 * even on big-endian machines: we do NOT byteswap the block numbers!
4792 for (block
= 0; block
< EXT4_N_BLOCKS
; block
++)
4793 ei
->i_data
[block
] = raw_inode
->i_block
[block
];
4794 INIT_LIST_HEAD(&ei
->i_orphan
);
4796 if (EXT4_INODE_SIZE(inode
->i_sb
) > EXT4_GOOD_OLD_INODE_SIZE
) {
4797 ei
->i_extra_isize
= le16_to_cpu(raw_inode
->i_extra_isize
);
4798 if (EXT4_GOOD_OLD_INODE_SIZE
+ ei
->i_extra_isize
>
4799 EXT4_INODE_SIZE(inode
->i_sb
)) {
4803 if (ei
->i_extra_isize
== 0) {
4804 /* The extra space is currently unused. Use it. */
4805 ei
->i_extra_isize
= sizeof(struct ext4_inode
) -
4806 EXT4_GOOD_OLD_INODE_SIZE
;
4808 __le32
*magic
= (void *)raw_inode
+
4809 EXT4_GOOD_OLD_INODE_SIZE
+
4811 if (*magic
== cpu_to_le32(EXT4_XATTR_MAGIC
))
4812 ei
->i_state
|= EXT4_STATE_XATTR
;
4815 ei
->i_extra_isize
= 0;
4817 EXT4_INODE_GET_XTIME(i_ctime
, inode
, raw_inode
);
4818 EXT4_INODE_GET_XTIME(i_mtime
, inode
, raw_inode
);
4819 EXT4_INODE_GET_XTIME(i_atime
, inode
, raw_inode
);
4820 EXT4_EINODE_GET_XTIME(i_crtime
, ei
, raw_inode
);
4822 inode
->i_version
= le32_to_cpu(raw_inode
->i_disk_version
);
4823 if (EXT4_INODE_SIZE(inode
->i_sb
) > EXT4_GOOD_OLD_INODE_SIZE
) {
4824 if (EXT4_FITS_IN_INODE(raw_inode
, ei
, i_version_hi
))
4826 (__u64
)(le32_to_cpu(raw_inode
->i_version_hi
)) << 32;
4830 if (ei
->i_file_acl
&&
4831 !ext4_data_block_valid(EXT4_SB(sb
), ei
->i_file_acl
, 1)) {
4832 ext4_error(sb
, __func__
,
4833 "bad extended attribute block %llu in inode #%lu",
4834 ei
->i_file_acl
, inode
->i_ino
);
4837 } else if (ei
->i_flags
& EXT4_EXTENTS_FL
) {
4838 if (S_ISREG(inode
->i_mode
) || S_ISDIR(inode
->i_mode
) ||
4839 (S_ISLNK(inode
->i_mode
) &&
4840 !ext4_inode_is_fast_symlink(inode
)))
4841 /* Validate extent which is part of inode */
4842 ret
= ext4_ext_check_inode(inode
);
4843 } else if (S_ISREG(inode
->i_mode
) || S_ISDIR(inode
->i_mode
) ||
4844 (S_ISLNK(inode
->i_mode
) &&
4845 !ext4_inode_is_fast_symlink(inode
))) {
4846 /* Validate block references which are part of inode */
4847 ret
= ext4_check_inode_blockref(inode
);
4852 if (S_ISREG(inode
->i_mode
)) {
4853 inode
->i_op
= &ext4_file_inode_operations
;
4854 inode
->i_fop
= &ext4_file_operations
;
4855 ext4_set_aops(inode
);
4856 } else if (S_ISDIR(inode
->i_mode
)) {
4857 inode
->i_op
= &ext4_dir_inode_operations
;
4858 inode
->i_fop
= &ext4_dir_operations
;
4859 } else if (S_ISLNK(inode
->i_mode
)) {
4860 if (ext4_inode_is_fast_symlink(inode
)) {
4861 inode
->i_op
= &ext4_fast_symlink_inode_operations
;
4862 nd_terminate_link(ei
->i_data
, inode
->i_size
,
4863 sizeof(ei
->i_data
) - 1);
4865 inode
->i_op
= &ext4_symlink_inode_operations
;
4866 ext4_set_aops(inode
);
4868 } else if (S_ISCHR(inode
->i_mode
) || S_ISBLK(inode
->i_mode
) ||
4869 S_ISFIFO(inode
->i_mode
) || S_ISSOCK(inode
->i_mode
)) {
4870 inode
->i_op
= &ext4_special_inode_operations
;
4871 if (raw_inode
->i_block
[0])
4872 init_special_inode(inode
, inode
->i_mode
,
4873 old_decode_dev(le32_to_cpu(raw_inode
->i_block
[0])));
4875 init_special_inode(inode
, inode
->i_mode
,
4876 new_decode_dev(le32_to_cpu(raw_inode
->i_block
[1])));
4879 ext4_error(inode
->i_sb
, __func__
,
4880 "bogus i_mode (%o) for inode=%lu",
4881 inode
->i_mode
, inode
->i_ino
);
4885 ext4_set_inode_flags(inode
);
4886 unlock_new_inode(inode
);
4892 return ERR_PTR(ret
);
4895 static int ext4_inode_blocks_set(handle_t
*handle
,
4896 struct ext4_inode
*raw_inode
,
4897 struct ext4_inode_info
*ei
)
4899 struct inode
*inode
= &(ei
->vfs_inode
);
4900 u64 i_blocks
= inode
->i_blocks
;
4901 struct super_block
*sb
= inode
->i_sb
;
4903 if (i_blocks
<= ~0U) {
4905 * i_blocks can be represnted in a 32 bit variable
4906 * as multiple of 512 bytes
4908 raw_inode
->i_blocks_lo
= cpu_to_le32(i_blocks
);
4909 raw_inode
->i_blocks_high
= 0;
4910 ei
->i_flags
&= ~EXT4_HUGE_FILE_FL
;
4913 if (!EXT4_HAS_RO_COMPAT_FEATURE(sb
, EXT4_FEATURE_RO_COMPAT_HUGE_FILE
))
4916 if (i_blocks
<= 0xffffffffffffULL
) {
4918 * i_blocks can be represented in a 48 bit variable
4919 * as multiple of 512 bytes
4921 raw_inode
->i_blocks_lo
= cpu_to_le32(i_blocks
);
4922 raw_inode
->i_blocks_high
= cpu_to_le16(i_blocks
>> 32);
4923 ei
->i_flags
&= ~EXT4_HUGE_FILE_FL
;
4925 ei
->i_flags
|= EXT4_HUGE_FILE_FL
;
4926 /* i_block is stored in file system block size */
4927 i_blocks
= i_blocks
>> (inode
->i_blkbits
- 9);
4928 raw_inode
->i_blocks_lo
= cpu_to_le32(i_blocks
);
4929 raw_inode
->i_blocks_high
= cpu_to_le16(i_blocks
>> 32);
4935 * Post the struct inode info into an on-disk inode location in the
4936 * buffer-cache. This gobbles the caller's reference to the
4937 * buffer_head in the inode location struct.
4939 * The caller must have write access to iloc->bh.
4941 static int ext4_do_update_inode(handle_t
*handle
,
4942 struct inode
*inode
,
4943 struct ext4_iloc
*iloc
)
4945 struct ext4_inode
*raw_inode
= ext4_raw_inode(iloc
);
4946 struct ext4_inode_info
*ei
= EXT4_I(inode
);
4947 struct buffer_head
*bh
= iloc
->bh
;
4948 int err
= 0, rc
, block
;
4950 /* For fields not not tracking in the in-memory inode,
4951 * initialise them to zero for new inodes. */
4952 if (ei
->i_state
& EXT4_STATE_NEW
)
4953 memset(raw_inode
, 0, EXT4_SB(inode
->i_sb
)->s_inode_size
);
4955 ext4_get_inode_flags(ei
);
4956 raw_inode
->i_mode
= cpu_to_le16(inode
->i_mode
);
4957 if (!(test_opt(inode
->i_sb
, NO_UID32
))) {
4958 raw_inode
->i_uid_low
= cpu_to_le16(low_16_bits(inode
->i_uid
));
4959 raw_inode
->i_gid_low
= cpu_to_le16(low_16_bits(inode
->i_gid
));
4961 * Fix up interoperability with old kernels. Otherwise, old inodes get
4962 * re-used with the upper 16 bits of the uid/gid intact
4965 raw_inode
->i_uid_high
=
4966 cpu_to_le16(high_16_bits(inode
->i_uid
));
4967 raw_inode
->i_gid_high
=
4968 cpu_to_le16(high_16_bits(inode
->i_gid
));
4970 raw_inode
->i_uid_high
= 0;
4971 raw_inode
->i_gid_high
= 0;
4974 raw_inode
->i_uid_low
=
4975 cpu_to_le16(fs_high2lowuid(inode
->i_uid
));
4976 raw_inode
->i_gid_low
=
4977 cpu_to_le16(fs_high2lowgid(inode
->i_gid
));
4978 raw_inode
->i_uid_high
= 0;
4979 raw_inode
->i_gid_high
= 0;
4981 raw_inode
->i_links_count
= cpu_to_le16(inode
->i_nlink
);
4983 EXT4_INODE_SET_XTIME(i_ctime
, inode
, raw_inode
);
4984 EXT4_INODE_SET_XTIME(i_mtime
, inode
, raw_inode
);
4985 EXT4_INODE_SET_XTIME(i_atime
, inode
, raw_inode
);
4986 EXT4_EINODE_SET_XTIME(i_crtime
, ei
, raw_inode
);
4988 if (ext4_inode_blocks_set(handle
, raw_inode
, ei
))
4990 raw_inode
->i_dtime
= cpu_to_le32(ei
->i_dtime
);
4991 raw_inode
->i_flags
= cpu_to_le32(ei
->i_flags
);
4992 if (EXT4_SB(inode
->i_sb
)->s_es
->s_creator_os
!=
4993 cpu_to_le32(EXT4_OS_HURD
))
4994 raw_inode
->i_file_acl_high
=
4995 cpu_to_le16(ei
->i_file_acl
>> 32);
4996 raw_inode
->i_file_acl_lo
= cpu_to_le32(ei
->i_file_acl
);
4997 ext4_isize_set(raw_inode
, ei
->i_disksize
);
4998 if (ei
->i_disksize
> 0x7fffffffULL
) {
4999 struct super_block
*sb
= inode
->i_sb
;
5000 if (!EXT4_HAS_RO_COMPAT_FEATURE(sb
,
5001 EXT4_FEATURE_RO_COMPAT_LARGE_FILE
) ||
5002 EXT4_SB(sb
)->s_es
->s_rev_level
==
5003 cpu_to_le32(EXT4_GOOD_OLD_REV
)) {
5004 /* If this is the first large file
5005 * created, add a flag to the superblock.
5007 err
= ext4_journal_get_write_access(handle
,
5008 EXT4_SB(sb
)->s_sbh
);
5011 ext4_update_dynamic_rev(sb
);
5012 EXT4_SET_RO_COMPAT_FEATURE(sb
,
5013 EXT4_FEATURE_RO_COMPAT_LARGE_FILE
);
5015 ext4_handle_sync(handle
);
5016 err
= ext4_handle_dirty_metadata(handle
, inode
,
5017 EXT4_SB(sb
)->s_sbh
);
5020 raw_inode
->i_generation
= cpu_to_le32(inode
->i_generation
);
5021 if (S_ISCHR(inode
->i_mode
) || S_ISBLK(inode
->i_mode
)) {
5022 if (old_valid_dev(inode
->i_rdev
)) {
5023 raw_inode
->i_block
[0] =
5024 cpu_to_le32(old_encode_dev(inode
->i_rdev
));
5025 raw_inode
->i_block
[1] = 0;
5027 raw_inode
->i_block
[0] = 0;
5028 raw_inode
->i_block
[1] =
5029 cpu_to_le32(new_encode_dev(inode
->i_rdev
));
5030 raw_inode
->i_block
[2] = 0;
5033 for (block
= 0; block
< EXT4_N_BLOCKS
; block
++)
5034 raw_inode
->i_block
[block
] = ei
->i_data
[block
];
5036 raw_inode
->i_disk_version
= cpu_to_le32(inode
->i_version
);
5037 if (ei
->i_extra_isize
) {
5038 if (EXT4_FITS_IN_INODE(raw_inode
, ei
, i_version_hi
))
5039 raw_inode
->i_version_hi
=
5040 cpu_to_le32(inode
->i_version
>> 32);
5041 raw_inode
->i_extra_isize
= cpu_to_le16(ei
->i_extra_isize
);
5044 BUFFER_TRACE(bh
, "call ext4_handle_dirty_metadata");
5045 rc
= ext4_handle_dirty_metadata(handle
, inode
, bh
);
5048 ei
->i_state
&= ~EXT4_STATE_NEW
;
5052 ext4_std_error(inode
->i_sb
, err
);
5057 * ext4_write_inode()
5059 * We are called from a few places:
5061 * - Within generic_file_write() for O_SYNC files.
5062 * Here, there will be no transaction running. We wait for any running
5063 * trasnaction to commit.
5065 * - Within sys_sync(), kupdate and such.
5066 * We wait on commit, if tol to.
5068 * - Within prune_icache() (PF_MEMALLOC == true)
5069 * Here we simply return. We can't afford to block kswapd on the
5072 * In all cases it is actually safe for us to return without doing anything,
5073 * because the inode has been copied into a raw inode buffer in
5074 * ext4_mark_inode_dirty(). This is a correctness thing for O_SYNC and for
5077 * Note that we are absolutely dependent upon all inode dirtiers doing the
5078 * right thing: they *must* call mark_inode_dirty() after dirtying info in
5079 * which we are interested.
5081 * It would be a bug for them to not do this. The code:
5083 * mark_inode_dirty(inode)
5085 * inode->i_size = expr;
5087 * is in error because a kswapd-driven write_inode() could occur while
5088 * `stuff()' is running, and the new i_size will be lost. Plus the inode
5089 * will no longer be on the superblock's dirty inode list.
5091 int ext4_write_inode(struct inode
*inode
, int wait
)
5095 if (current
->flags
& PF_MEMALLOC
)
5098 if (EXT4_SB(inode
->i_sb
)->s_journal
) {
5099 if (ext4_journal_current_handle()) {
5100 jbd_debug(1, "called recursively, non-PF_MEMALLOC!\n");
5108 err
= ext4_force_commit(inode
->i_sb
);
5110 struct ext4_iloc iloc
;
5112 err
= ext4_get_inode_loc(inode
, &iloc
);
5116 sync_dirty_buffer(iloc
.bh
);
5117 if (buffer_req(iloc
.bh
) && !buffer_uptodate(iloc
.bh
)) {
5118 ext4_error(inode
->i_sb
, __func__
,
5119 "IO error syncing inode, "
5120 "inode=%lu, block=%llu",
5122 (unsigned long long)iloc
.bh
->b_blocknr
);
5132 * Called from notify_change.
5134 * We want to trap VFS attempts to truncate the file as soon as
5135 * possible. In particular, we want to make sure that when the VFS
5136 * shrinks i_size, we put the inode on the orphan list and modify
5137 * i_disksize immediately, so that during the subsequent flushing of
5138 * dirty pages and freeing of disk blocks, we can guarantee that any
5139 * commit will leave the blocks being flushed in an unused state on
5140 * disk. (On recovery, the inode will get truncated and the blocks will
5141 * be freed, so we have a strong guarantee that no future commit will
5142 * leave these blocks visible to the user.)
5144 * Another thing we have to assure is that if we are in ordered mode
5145 * and inode is still attached to the committing transaction, we must
5146 * we start writeout of all the dirty pages which are being truncated.
5147 * This way we are sure that all the data written in the previous
5148 * transaction are already on disk (truncate waits for pages under
5151 * Called with inode->i_mutex down.
5153 int ext4_setattr(struct dentry
*dentry
, struct iattr
*attr
)
5155 struct inode
*inode
= dentry
->d_inode
;
5157 const unsigned int ia_valid
= attr
->ia_valid
;
5159 error
= inode_change_ok(inode
, attr
);
5163 if ((ia_valid
& ATTR_UID
&& attr
->ia_uid
!= inode
->i_uid
) ||
5164 (ia_valid
& ATTR_GID
&& attr
->ia_gid
!= inode
->i_gid
)) {
5167 /* (user+group)*(old+new) structure, inode write (sb,
5168 * inode block, ? - but truncate inode update has it) */
5169 handle
= ext4_journal_start(inode
, 2*(EXT4_QUOTA_INIT_BLOCKS(inode
->i_sb
)+
5170 EXT4_QUOTA_DEL_BLOCKS(inode
->i_sb
))+3);
5171 if (IS_ERR(handle
)) {
5172 error
= PTR_ERR(handle
);
5175 error
= vfs_dq_transfer(inode
, attr
) ? -EDQUOT
: 0;
5177 ext4_journal_stop(handle
);
5180 /* Update corresponding info in inode so that everything is in
5181 * one transaction */
5182 if (attr
->ia_valid
& ATTR_UID
)
5183 inode
->i_uid
= attr
->ia_uid
;
5184 if (attr
->ia_valid
& ATTR_GID
)
5185 inode
->i_gid
= attr
->ia_gid
;
5186 error
= ext4_mark_inode_dirty(handle
, inode
);
5187 ext4_journal_stop(handle
);
5190 if (attr
->ia_valid
& ATTR_SIZE
) {
5191 if (!(EXT4_I(inode
)->i_flags
& EXT4_EXTENTS_FL
)) {
5192 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
5194 if (attr
->ia_size
> sbi
->s_bitmap_maxbytes
) {
5201 if (S_ISREG(inode
->i_mode
) &&
5202 attr
->ia_valid
& ATTR_SIZE
&& attr
->ia_size
< inode
->i_size
) {
5205 handle
= ext4_journal_start(inode
, 3);
5206 if (IS_ERR(handle
)) {
5207 error
= PTR_ERR(handle
);
5211 error
= ext4_orphan_add(handle
, inode
);
5212 EXT4_I(inode
)->i_disksize
= attr
->ia_size
;
5213 rc
= ext4_mark_inode_dirty(handle
, inode
);
5216 ext4_journal_stop(handle
);
5218 if (ext4_should_order_data(inode
)) {
5219 error
= ext4_begin_ordered_truncate(inode
,
5222 /* Do as much error cleanup as possible */
5223 handle
= ext4_journal_start(inode
, 3);
5224 if (IS_ERR(handle
)) {
5225 ext4_orphan_del(NULL
, inode
);
5228 ext4_orphan_del(handle
, inode
);
5229 ext4_journal_stop(handle
);
5235 rc
= inode_setattr(inode
, attr
);
5237 /* If inode_setattr's call to ext4_truncate failed to get a
5238 * transaction handle at all, we need to clean up the in-core
5239 * orphan list manually. */
5241 ext4_orphan_del(NULL
, inode
);
5243 if (!rc
&& (ia_valid
& ATTR_MODE
))
5244 rc
= ext4_acl_chmod(inode
);
5247 ext4_std_error(inode
->i_sb
, error
);
5253 int ext4_getattr(struct vfsmount
*mnt
, struct dentry
*dentry
,
5256 struct inode
*inode
;
5257 unsigned long delalloc_blocks
;
5259 inode
= dentry
->d_inode
;
5260 generic_fillattr(inode
, stat
);
5263 * We can't update i_blocks if the block allocation is delayed
5264 * otherwise in the case of system crash before the real block
5265 * allocation is done, we will have i_blocks inconsistent with
5266 * on-disk file blocks.
5267 * We always keep i_blocks updated together with real
5268 * allocation. But to not confuse with user, stat
5269 * will return the blocks that include the delayed allocation
5270 * blocks for this file.
5272 spin_lock(&EXT4_I(inode
)->i_block_reservation_lock
);
5273 delalloc_blocks
= EXT4_I(inode
)->i_reserved_data_blocks
;
5274 spin_unlock(&EXT4_I(inode
)->i_block_reservation_lock
);
5276 stat
->blocks
+= (delalloc_blocks
<< inode
->i_sb
->s_blocksize_bits
)>>9;
5280 static int ext4_indirect_trans_blocks(struct inode
*inode
, int nrblocks
,
5285 /* if nrblocks are contiguous */
5288 * With N contiguous data blocks, it need at most
5289 * N/EXT4_ADDR_PER_BLOCK(inode->i_sb) indirect blocks
5290 * 2 dindirect blocks
5293 indirects
= nrblocks
/ EXT4_ADDR_PER_BLOCK(inode
->i_sb
);
5294 return indirects
+ 3;
5297 * if nrblocks are not contiguous, worse case, each block touch
5298 * a indirect block, and each indirect block touch a double indirect
5299 * block, plus a triple indirect block
5301 indirects
= nrblocks
* 2 + 1;
5305 static int ext4_index_trans_blocks(struct inode
*inode
, int nrblocks
, int chunk
)
5307 if (!(EXT4_I(inode
)->i_flags
& EXT4_EXTENTS_FL
))
5308 return ext4_indirect_trans_blocks(inode
, nrblocks
, chunk
);
5309 return ext4_ext_index_trans_blocks(inode
, nrblocks
, chunk
);
5313 * Account for index blocks, block groups bitmaps and block group
5314 * descriptor blocks if modify datablocks and index blocks
5315 * worse case, the indexs blocks spread over different block groups
5317 * If datablocks are discontiguous, they are possible to spread over
5318 * different block groups too. If they are contiugous, with flexbg,
5319 * they could still across block group boundary.
5321 * Also account for superblock, inode, quota and xattr blocks
5323 int ext4_meta_trans_blocks(struct inode
*inode
, int nrblocks
, int chunk
)
5325 ext4_group_t groups
, ngroups
= ext4_get_groups_count(inode
->i_sb
);
5331 * How many index blocks need to touch to modify nrblocks?
5332 * The "Chunk" flag indicating whether the nrblocks is
5333 * physically contiguous on disk
5335 * For Direct IO and fallocate, they calls get_block to allocate
5336 * one single extent at a time, so they could set the "Chunk" flag
5338 idxblocks
= ext4_index_trans_blocks(inode
, nrblocks
, chunk
);
5343 * Now let's see how many group bitmaps and group descriptors need
5353 if (groups
> ngroups
)
5355 if (groups
> EXT4_SB(inode
->i_sb
)->s_gdb_count
)
5356 gdpblocks
= EXT4_SB(inode
->i_sb
)->s_gdb_count
;
5358 /* bitmaps and block group descriptor blocks */
5359 ret
+= groups
+ gdpblocks
;
5361 /* Blocks for super block, inode, quota and xattr blocks */
5362 ret
+= EXT4_META_TRANS_BLOCKS(inode
->i_sb
);
5368 * Calulate the total number of credits to reserve to fit
5369 * the modification of a single pages into a single transaction,
5370 * which may include multiple chunks of block allocations.
5372 * This could be called via ext4_write_begin()
5374 * We need to consider the worse case, when
5375 * one new block per extent.
5377 int ext4_writepage_trans_blocks(struct inode
*inode
)
5379 int bpp
= ext4_journal_blocks_per_page(inode
);
5382 ret
= ext4_meta_trans_blocks(inode
, bpp
, 0);
5384 /* Account for data blocks for journalled mode */
5385 if (ext4_should_journal_data(inode
))
5391 * Calculate the journal credits for a chunk of data modification.
5393 * This is called from DIO, fallocate or whoever calling
5394 * ext4_get_blocks() to map/allocate a chunk of contigous disk blocks.
5396 * journal buffers for data blocks are not included here, as DIO
5397 * and fallocate do no need to journal data buffers.
5399 int ext4_chunk_trans_blocks(struct inode
*inode
, int nrblocks
)
5401 return ext4_meta_trans_blocks(inode
, nrblocks
, 1);
5405 * The caller must have previously called ext4_reserve_inode_write().
5406 * Give this, we know that the caller already has write access to iloc->bh.
5408 int ext4_mark_iloc_dirty(handle_t
*handle
,
5409 struct inode
*inode
, struct ext4_iloc
*iloc
)
5413 if (test_opt(inode
->i_sb
, I_VERSION
))
5414 inode_inc_iversion(inode
);
5416 /* the do_update_inode consumes one bh->b_count */
5419 /* ext4_do_update_inode() does jbd2_journal_dirty_metadata */
5420 err
= ext4_do_update_inode(handle
, inode
, iloc
);
5426 * On success, We end up with an outstanding reference count against
5427 * iloc->bh. This _must_ be cleaned up later.
5431 ext4_reserve_inode_write(handle_t
*handle
, struct inode
*inode
,
5432 struct ext4_iloc
*iloc
)
5436 err
= ext4_get_inode_loc(inode
, iloc
);
5438 BUFFER_TRACE(iloc
->bh
, "get_write_access");
5439 err
= ext4_journal_get_write_access(handle
, iloc
->bh
);
5445 ext4_std_error(inode
->i_sb
, err
);
5450 * Expand an inode by new_extra_isize bytes.
5451 * Returns 0 on success or negative error number on failure.
5453 static int ext4_expand_extra_isize(struct inode
*inode
,
5454 unsigned int new_extra_isize
,
5455 struct ext4_iloc iloc
,
5458 struct ext4_inode
*raw_inode
;
5459 struct ext4_xattr_ibody_header
*header
;
5460 struct ext4_xattr_entry
*entry
;
5462 if (EXT4_I(inode
)->i_extra_isize
>= new_extra_isize
)
5465 raw_inode
= ext4_raw_inode(&iloc
);
5467 header
= IHDR(inode
, raw_inode
);
5468 entry
= IFIRST(header
);
5470 /* No extended attributes present */
5471 if (!(EXT4_I(inode
)->i_state
& EXT4_STATE_XATTR
) ||
5472 header
->h_magic
!= cpu_to_le32(EXT4_XATTR_MAGIC
)) {
5473 memset((void *)raw_inode
+ EXT4_GOOD_OLD_INODE_SIZE
, 0,
5475 EXT4_I(inode
)->i_extra_isize
= new_extra_isize
;
5479 /* try to expand with EAs present */
5480 return ext4_expand_extra_isize_ea(inode
, new_extra_isize
,
5485 * What we do here is to mark the in-core inode as clean with respect to inode
5486 * dirtiness (it may still be data-dirty).
5487 * This means that the in-core inode may be reaped by prune_icache
5488 * without having to perform any I/O. This is a very good thing,
5489 * because *any* task may call prune_icache - even ones which
5490 * have a transaction open against a different journal.
5492 * Is this cheating? Not really. Sure, we haven't written the
5493 * inode out, but prune_icache isn't a user-visible syncing function.
5494 * Whenever the user wants stuff synced (sys_sync, sys_msync, sys_fsync)
5495 * we start and wait on commits.
5497 * Is this efficient/effective? Well, we're being nice to the system
5498 * by cleaning up our inodes proactively so they can be reaped
5499 * without I/O. But we are potentially leaving up to five seconds'
5500 * worth of inodes floating about which prune_icache wants us to
5501 * write out. One way to fix that would be to get prune_icache()
5502 * to do a write_super() to free up some memory. It has the desired
5505 int ext4_mark_inode_dirty(handle_t
*handle
, struct inode
*inode
)
5507 struct ext4_iloc iloc
;
5508 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
5509 static unsigned int mnt_count
;
5513 err
= ext4_reserve_inode_write(handle
, inode
, &iloc
);
5514 if (ext4_handle_valid(handle
) &&
5515 EXT4_I(inode
)->i_extra_isize
< sbi
->s_want_extra_isize
&&
5516 !(EXT4_I(inode
)->i_state
& EXT4_STATE_NO_EXPAND
)) {
5518 * We need extra buffer credits since we may write into EA block
5519 * with this same handle. If journal_extend fails, then it will
5520 * only result in a minor loss of functionality for that inode.
5521 * If this is felt to be critical, then e2fsck should be run to
5522 * force a large enough s_min_extra_isize.
5524 if ((jbd2_journal_extend(handle
,
5525 EXT4_DATA_TRANS_BLOCKS(inode
->i_sb
))) == 0) {
5526 ret
= ext4_expand_extra_isize(inode
,
5527 sbi
->s_want_extra_isize
,
5530 EXT4_I(inode
)->i_state
|= EXT4_STATE_NO_EXPAND
;
5532 le16_to_cpu(sbi
->s_es
->s_mnt_count
)) {
5533 ext4_warning(inode
->i_sb
, __func__
,
5534 "Unable to expand inode %lu. Delete"
5535 " some EAs or run e2fsck.",
5538 le16_to_cpu(sbi
->s_es
->s_mnt_count
);
5544 err
= ext4_mark_iloc_dirty(handle
, inode
, &iloc
);
5549 * ext4_dirty_inode() is called from __mark_inode_dirty()
5551 * We're really interested in the case where a file is being extended.
5552 * i_size has been changed by generic_commit_write() and we thus need
5553 * to include the updated inode in the current transaction.
5555 * Also, vfs_dq_alloc_block() will always dirty the inode when blocks
5556 * are allocated to the file.
5558 * If the inode is marked synchronous, we don't honour that here - doing
5559 * so would cause a commit on atime updates, which we don't bother doing.
5560 * We handle synchronous inodes at the highest possible level.
5562 void ext4_dirty_inode(struct inode
*inode
)
5566 handle
= ext4_journal_start(inode
, 2);
5570 ext4_mark_inode_dirty(handle
, inode
);
5572 ext4_journal_stop(handle
);
5579 * Bind an inode's backing buffer_head into this transaction, to prevent
5580 * it from being flushed to disk early. Unlike
5581 * ext4_reserve_inode_write, this leaves behind no bh reference and
5582 * returns no iloc structure, so the caller needs to repeat the iloc
5583 * lookup to mark the inode dirty later.
5585 static int ext4_pin_inode(handle_t
*handle
, struct inode
*inode
)
5587 struct ext4_iloc iloc
;
5591 err
= ext4_get_inode_loc(inode
, &iloc
);
5593 BUFFER_TRACE(iloc
.bh
, "get_write_access");
5594 err
= jbd2_journal_get_write_access(handle
, iloc
.bh
);
5596 err
= ext4_handle_dirty_metadata(handle
,
5602 ext4_std_error(inode
->i_sb
, err
);
5607 int ext4_change_inode_journal_flag(struct inode
*inode
, int val
)
5614 * We have to be very careful here: changing a data block's
5615 * journaling status dynamically is dangerous. If we write a
5616 * data block to the journal, change the status and then delete
5617 * that block, we risk forgetting to revoke the old log record
5618 * from the journal and so a subsequent replay can corrupt data.
5619 * So, first we make sure that the journal is empty and that
5620 * nobody is changing anything.
5623 journal
= EXT4_JOURNAL(inode
);
5626 if (is_journal_aborted(journal
))
5629 jbd2_journal_lock_updates(journal
);
5630 jbd2_journal_flush(journal
);
5633 * OK, there are no updates running now, and all cached data is
5634 * synced to disk. We are now in a completely consistent state
5635 * which doesn't have anything in the journal, and we know that
5636 * no filesystem updates are running, so it is safe to modify
5637 * the inode's in-core data-journaling state flag now.
5641 EXT4_I(inode
)->i_flags
|= EXT4_JOURNAL_DATA_FL
;
5643 EXT4_I(inode
)->i_flags
&= ~EXT4_JOURNAL_DATA_FL
;
5644 ext4_set_aops(inode
);
5646 jbd2_journal_unlock_updates(journal
);
5648 /* Finally we can mark the inode as dirty. */
5650 handle
= ext4_journal_start(inode
, 1);
5652 return PTR_ERR(handle
);
5654 err
= ext4_mark_inode_dirty(handle
, inode
);
5655 ext4_handle_sync(handle
);
5656 ext4_journal_stop(handle
);
5657 ext4_std_error(inode
->i_sb
, err
);
5662 static int ext4_bh_unmapped(handle_t
*handle
, struct buffer_head
*bh
)
5664 return !buffer_mapped(bh
);
5667 int ext4_page_mkwrite(struct vm_area_struct
*vma
, struct vm_fault
*vmf
)
5669 struct page
*page
= vmf
->page
;
5674 struct file
*file
= vma
->vm_file
;
5675 struct inode
*inode
= file
->f_path
.dentry
->d_inode
;
5676 struct address_space
*mapping
= inode
->i_mapping
;
5679 * Get i_alloc_sem to stop truncates messing with the inode. We cannot
5680 * get i_mutex because we are already holding mmap_sem.
5682 down_read(&inode
->i_alloc_sem
);
5683 size
= i_size_read(inode
);
5684 if (page
->mapping
!= mapping
|| size
<= page_offset(page
)
5685 || !PageUptodate(page
)) {
5686 /* page got truncated from under us? */
5690 if (PageMappedToDisk(page
))
5693 if (page
->index
== size
>> PAGE_CACHE_SHIFT
)
5694 len
= size
& ~PAGE_CACHE_MASK
;
5696 len
= PAGE_CACHE_SIZE
;
5700 * return if we have all the buffers mapped. This avoid
5701 * the need to call write_begin/write_end which does a
5702 * journal_start/journal_stop which can block and take
5705 if (page_has_buffers(page
)) {
5706 if (!walk_page_buffers(NULL
, page_buffers(page
), 0, len
, NULL
,
5707 ext4_bh_unmapped
)) {
5714 * OK, we need to fill the hole... Do write_begin write_end
5715 * to do block allocation/reservation.We are not holding
5716 * inode.i__mutex here. That allow * parallel write_begin,
5717 * write_end call. lock_page prevent this from happening
5718 * on the same page though
5720 ret
= mapping
->a_ops
->write_begin(file
, mapping
, page_offset(page
),
5721 len
, AOP_FLAG_UNINTERRUPTIBLE
, &page
, &fsdata
);
5724 ret
= mapping
->a_ops
->write_end(file
, mapping
, page_offset(page
),
5725 len
, len
, page
, fsdata
);
5731 ret
= VM_FAULT_SIGBUS
;
5732 up_read(&inode
->i_alloc_sem
);