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 ext4_update_inode_fsync_trans(handle
, inode
, 1);
989 map_bh(bh_result
, inode
->i_sb
, le32_to_cpu(chain
[depth
-1].key
));
990 if (count
> blocks_to_boundary
)
991 set_buffer_boundary(bh_result
);
993 /* Clean up and exit */
994 partial
= chain
+ depth
- 1; /* the whole chain */
996 while (partial
> chain
) {
997 BUFFER_TRACE(partial
->bh
, "call brelse");
1001 BUFFER_TRACE(bh_result
, "returned");
1007 qsize_t
*ext4_get_reserved_space(struct inode
*inode
)
1009 return &EXT4_I(inode
)->i_reserved_quota
;
1014 * Calculate the number of metadata blocks need to reserve
1015 * to allocate a new block at @lblocks for non extent file based file
1017 static int ext4_indirect_calc_metadata_amount(struct inode
*inode
,
1020 struct ext4_inode_info
*ei
= EXT4_I(inode
);
1021 int dind_mask
= EXT4_ADDR_PER_BLOCK(inode
->i_sb
) - 1;
1024 if (lblock
< EXT4_NDIR_BLOCKS
)
1027 lblock
-= EXT4_NDIR_BLOCKS
;
1029 if (ei
->i_da_metadata_calc_len
&&
1030 (lblock
& dind_mask
) == ei
->i_da_metadata_calc_last_lblock
) {
1031 ei
->i_da_metadata_calc_len
++;
1034 ei
->i_da_metadata_calc_last_lblock
= lblock
& dind_mask
;
1035 ei
->i_da_metadata_calc_len
= 1;
1036 blk_bits
= roundup_pow_of_two(lblock
+ 1);
1037 return (blk_bits
/ EXT4_ADDR_PER_BLOCK_BITS(inode
->i_sb
)) + 1;
1041 * Calculate the number of metadata blocks need to reserve
1042 * to allocate a block located at @lblock
1044 static int ext4_calc_metadata_amount(struct inode
*inode
, sector_t lblock
)
1046 if (EXT4_I(inode
)->i_flags
& EXT4_EXTENTS_FL
)
1047 return ext4_ext_calc_metadata_amount(inode
, lblock
);
1049 return ext4_indirect_calc_metadata_amount(inode
, lblock
);
1053 * Called with i_data_sem down, which is important since we can call
1054 * ext4_discard_preallocations() from here.
1056 static void ext4_da_update_reserve_space(struct inode
*inode
, int used
)
1058 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
1059 struct ext4_inode_info
*ei
= EXT4_I(inode
);
1062 spin_lock(&ei
->i_block_reservation_lock
);
1063 if (unlikely(used
> ei
->i_reserved_data_blocks
)) {
1064 ext4_msg(inode
->i_sb
, KERN_NOTICE
, "%s: ino %lu, used %d "
1065 "with only %d reserved data blocks\n",
1066 __func__
, inode
->i_ino
, used
,
1067 ei
->i_reserved_data_blocks
);
1069 used
= ei
->i_reserved_data_blocks
;
1072 /* Update per-inode reservations */
1073 ei
->i_reserved_data_blocks
-= used
;
1074 used
+= ei
->i_allocated_meta_blocks
;
1075 ei
->i_reserved_meta_blocks
-= ei
->i_allocated_meta_blocks
;
1076 ei
->i_allocated_meta_blocks
= 0;
1077 percpu_counter_sub(&sbi
->s_dirtyblocks_counter
, used
);
1079 if (ei
->i_reserved_data_blocks
== 0) {
1081 * We can release all of the reserved metadata blocks
1082 * only when we have written all of the delayed
1083 * allocation blocks.
1085 mdb_free
= ei
->i_reserved_meta_blocks
;
1086 ei
->i_reserved_meta_blocks
= 0;
1087 ei
->i_da_metadata_calc_len
= 0;
1088 percpu_counter_sub(&sbi
->s_dirtyblocks_counter
, mdb_free
);
1090 spin_unlock(&EXT4_I(inode
)->i_block_reservation_lock
);
1092 /* Update quota subsystem */
1093 vfs_dq_claim_block(inode
, used
);
1095 vfs_dq_release_reservation_block(inode
, mdb_free
);
1098 * If we have done all the pending block allocations and if
1099 * there aren't any writers on the inode, we can discard the
1100 * inode's preallocations.
1102 if ((ei
->i_reserved_data_blocks
== 0) &&
1103 (atomic_read(&inode
->i_writecount
) == 0))
1104 ext4_discard_preallocations(inode
);
1107 static int check_block_validity(struct inode
*inode
, const char *msg
,
1108 sector_t logical
, sector_t phys
, int len
)
1110 if (!ext4_data_block_valid(EXT4_SB(inode
->i_sb
), phys
, len
)) {
1111 ext4_error(inode
->i_sb
, msg
,
1112 "inode #%lu logical block %llu mapped to %llu "
1113 "(size %d)", inode
->i_ino
,
1114 (unsigned long long) logical
,
1115 (unsigned long long) phys
, len
);
1122 * Return the number of contiguous dirty pages in a given inode
1123 * starting at page frame idx.
1125 static pgoff_t
ext4_num_dirty_pages(struct inode
*inode
, pgoff_t idx
,
1126 unsigned int max_pages
)
1128 struct address_space
*mapping
= inode
->i_mapping
;
1130 struct pagevec pvec
;
1132 int i
, nr_pages
, done
= 0;
1136 pagevec_init(&pvec
, 0);
1139 nr_pages
= pagevec_lookup_tag(&pvec
, mapping
, &index
,
1140 PAGECACHE_TAG_DIRTY
,
1141 (pgoff_t
)PAGEVEC_SIZE
);
1144 for (i
= 0; i
< nr_pages
; i
++) {
1145 struct page
*page
= pvec
.pages
[i
];
1146 struct buffer_head
*bh
, *head
;
1149 if (unlikely(page
->mapping
!= mapping
) ||
1151 PageWriteback(page
) ||
1152 page
->index
!= idx
) {
1157 if (page_has_buffers(page
)) {
1158 bh
= head
= page_buffers(page
);
1160 if (!buffer_delay(bh
) &&
1161 !buffer_unwritten(bh
))
1163 bh
= bh
->b_this_page
;
1164 } while (!done
&& (bh
!= head
));
1171 if (num
>= max_pages
)
1174 pagevec_release(&pvec
);
1180 * The ext4_get_blocks() function tries to look up the requested blocks,
1181 * and returns if the blocks are already mapped.
1183 * Otherwise it takes the write lock of the i_data_sem and allocate blocks
1184 * and store the allocated blocks in the result buffer head and mark it
1187 * If file type is extents based, it will call ext4_ext_get_blocks(),
1188 * Otherwise, call with ext4_ind_get_blocks() to handle indirect mapping
1191 * On success, it returns the number of blocks being mapped or allocate.
1192 * if create==0 and the blocks are pre-allocated and uninitialized block,
1193 * the result buffer head is unmapped. If the create ==1, it will make sure
1194 * the buffer head is mapped.
1196 * It returns 0 if plain look up failed (blocks have not been allocated), in
1197 * that casem, buffer head is unmapped
1199 * It returns the error in case of allocation failure.
1201 int ext4_get_blocks(handle_t
*handle
, struct inode
*inode
, sector_t block
,
1202 unsigned int max_blocks
, struct buffer_head
*bh
,
1207 clear_buffer_mapped(bh
);
1208 clear_buffer_unwritten(bh
);
1210 ext_debug("ext4_get_blocks(): inode %lu, flag %d, max_blocks %u,"
1211 "logical block %lu\n", inode
->i_ino
, flags
, max_blocks
,
1212 (unsigned long)block
);
1214 * Try to see if we can get the block without requesting a new
1215 * file system block.
1217 down_read((&EXT4_I(inode
)->i_data_sem
));
1218 if (EXT4_I(inode
)->i_flags
& EXT4_EXTENTS_FL
) {
1219 retval
= ext4_ext_get_blocks(handle
, inode
, block
, max_blocks
,
1222 retval
= ext4_ind_get_blocks(handle
, inode
, block
, max_blocks
,
1225 up_read((&EXT4_I(inode
)->i_data_sem
));
1227 if (retval
> 0 && buffer_mapped(bh
)) {
1228 int ret
= check_block_validity(inode
, "file system corruption",
1229 block
, bh
->b_blocknr
, retval
);
1234 /* If it is only a block(s) look up */
1235 if ((flags
& EXT4_GET_BLOCKS_CREATE
) == 0)
1239 * Returns if the blocks have already allocated
1241 * Note that if blocks have been preallocated
1242 * ext4_ext_get_block() returns th create = 0
1243 * with buffer head unmapped.
1245 if (retval
> 0 && buffer_mapped(bh
))
1249 * When we call get_blocks without the create flag, the
1250 * BH_Unwritten flag could have gotten set if the blocks
1251 * requested were part of a uninitialized extent. We need to
1252 * clear this flag now that we are committed to convert all or
1253 * part of the uninitialized extent to be an initialized
1254 * extent. This is because we need to avoid the combination
1255 * of BH_Unwritten and BH_Mapped flags being simultaneously
1256 * set on the buffer_head.
1258 clear_buffer_unwritten(bh
);
1261 * New blocks allocate and/or writing to uninitialized extent
1262 * will possibly result in updating i_data, so we take
1263 * the write lock of i_data_sem, and call get_blocks()
1264 * with create == 1 flag.
1266 down_write((&EXT4_I(inode
)->i_data_sem
));
1269 * if the caller is from delayed allocation writeout path
1270 * we have already reserved fs blocks for allocation
1271 * let the underlying get_block() function know to
1272 * avoid double accounting
1274 if (flags
& EXT4_GET_BLOCKS_DELALLOC_RESERVE
)
1275 EXT4_I(inode
)->i_delalloc_reserved_flag
= 1;
1277 * We need to check for EXT4 here because migrate
1278 * could have changed the inode type in between
1280 if (EXT4_I(inode
)->i_flags
& EXT4_EXTENTS_FL
) {
1281 retval
= ext4_ext_get_blocks(handle
, inode
, block
, max_blocks
,
1284 retval
= ext4_ind_get_blocks(handle
, inode
, block
,
1285 max_blocks
, bh
, flags
);
1287 if (retval
> 0 && buffer_new(bh
)) {
1289 * We allocated new blocks which will result in
1290 * i_data's format changing. Force the migrate
1291 * to fail by clearing migrate flags
1293 EXT4_I(inode
)->i_state
&= ~EXT4_STATE_EXT_MIGRATE
;
1297 if (flags
& EXT4_GET_BLOCKS_DELALLOC_RESERVE
)
1298 EXT4_I(inode
)->i_delalloc_reserved_flag
= 0;
1301 * Update reserved blocks/metadata blocks after successful
1302 * block allocation which had been deferred till now.
1304 if ((retval
> 0) && (flags
& EXT4_GET_BLOCKS_UPDATE_RESERVE_SPACE
))
1305 ext4_da_update_reserve_space(inode
, retval
);
1307 up_write((&EXT4_I(inode
)->i_data_sem
));
1308 if (retval
> 0 && buffer_mapped(bh
)) {
1309 int ret
= check_block_validity(inode
, "file system "
1310 "corruption after allocation",
1311 block
, bh
->b_blocknr
, retval
);
1318 /* Maximum number of blocks we map for direct IO at once. */
1319 #define DIO_MAX_BLOCKS 4096
1321 int ext4_get_block(struct inode
*inode
, sector_t iblock
,
1322 struct buffer_head
*bh_result
, int create
)
1324 handle_t
*handle
= ext4_journal_current_handle();
1325 int ret
= 0, started
= 0;
1326 unsigned max_blocks
= bh_result
->b_size
>> inode
->i_blkbits
;
1329 if (create
&& !handle
) {
1330 /* Direct IO write... */
1331 if (max_blocks
> DIO_MAX_BLOCKS
)
1332 max_blocks
= DIO_MAX_BLOCKS
;
1333 dio_credits
= ext4_chunk_trans_blocks(inode
, max_blocks
);
1334 handle
= ext4_journal_start(inode
, dio_credits
);
1335 if (IS_ERR(handle
)) {
1336 ret
= PTR_ERR(handle
);
1342 ret
= ext4_get_blocks(handle
, inode
, iblock
, max_blocks
, bh_result
,
1343 create
? EXT4_GET_BLOCKS_CREATE
: 0);
1345 bh_result
->b_size
= (ret
<< inode
->i_blkbits
);
1349 ext4_journal_stop(handle
);
1355 * `handle' can be NULL if create is zero
1357 struct buffer_head
*ext4_getblk(handle_t
*handle
, struct inode
*inode
,
1358 ext4_lblk_t block
, int create
, int *errp
)
1360 struct buffer_head dummy
;
1364 J_ASSERT(handle
!= NULL
|| create
== 0);
1367 dummy
.b_blocknr
= -1000;
1368 buffer_trace_init(&dummy
.b_history
);
1370 flags
|= EXT4_GET_BLOCKS_CREATE
;
1371 err
= ext4_get_blocks(handle
, inode
, block
, 1, &dummy
, flags
);
1373 * ext4_get_blocks() returns number of blocks mapped. 0 in
1382 if (!err
&& buffer_mapped(&dummy
)) {
1383 struct buffer_head
*bh
;
1384 bh
= sb_getblk(inode
->i_sb
, dummy
.b_blocknr
);
1389 if (buffer_new(&dummy
)) {
1390 J_ASSERT(create
!= 0);
1391 J_ASSERT(handle
!= NULL
);
1394 * Now that we do not always journal data, we should
1395 * keep in mind whether this should always journal the
1396 * new buffer as metadata. For now, regular file
1397 * writes use ext4_get_block instead, so it's not a
1401 BUFFER_TRACE(bh
, "call get_create_access");
1402 fatal
= ext4_journal_get_create_access(handle
, bh
);
1403 if (!fatal
&& !buffer_uptodate(bh
)) {
1404 memset(bh
->b_data
, 0, inode
->i_sb
->s_blocksize
);
1405 set_buffer_uptodate(bh
);
1408 BUFFER_TRACE(bh
, "call ext4_handle_dirty_metadata");
1409 err
= ext4_handle_dirty_metadata(handle
, inode
, bh
);
1413 BUFFER_TRACE(bh
, "not a new buffer");
1426 struct buffer_head
*ext4_bread(handle_t
*handle
, struct inode
*inode
,
1427 ext4_lblk_t block
, int create
, int *err
)
1429 struct buffer_head
*bh
;
1431 bh
= ext4_getblk(handle
, inode
, block
, create
, err
);
1434 if (buffer_uptodate(bh
))
1436 ll_rw_block(READ_META
, 1, &bh
);
1438 if (buffer_uptodate(bh
))
1445 static int walk_page_buffers(handle_t
*handle
,
1446 struct buffer_head
*head
,
1450 int (*fn
)(handle_t
*handle
,
1451 struct buffer_head
*bh
))
1453 struct buffer_head
*bh
;
1454 unsigned block_start
, block_end
;
1455 unsigned blocksize
= head
->b_size
;
1457 struct buffer_head
*next
;
1459 for (bh
= head
, block_start
= 0;
1460 ret
== 0 && (bh
!= head
|| !block_start
);
1461 block_start
= block_end
, bh
= next
) {
1462 next
= bh
->b_this_page
;
1463 block_end
= block_start
+ blocksize
;
1464 if (block_end
<= from
|| block_start
>= to
) {
1465 if (partial
&& !buffer_uptodate(bh
))
1469 err
= (*fn
)(handle
, bh
);
1477 * To preserve ordering, it is essential that the hole instantiation and
1478 * the data write be encapsulated in a single transaction. We cannot
1479 * close off a transaction and start a new one between the ext4_get_block()
1480 * and the commit_write(). So doing the jbd2_journal_start at the start of
1481 * prepare_write() is the right place.
1483 * Also, this function can nest inside ext4_writepage() ->
1484 * block_write_full_page(). In that case, we *know* that ext4_writepage()
1485 * has generated enough buffer credits to do the whole page. So we won't
1486 * block on the journal in that case, which is good, because the caller may
1489 * By accident, ext4 can be reentered when a transaction is open via
1490 * quota file writes. If we were to commit the transaction while thus
1491 * reentered, there can be a deadlock - we would be holding a quota
1492 * lock, and the commit would never complete if another thread had a
1493 * transaction open and was blocking on the quota lock - a ranking
1496 * So what we do is to rely on the fact that jbd2_journal_stop/journal_start
1497 * will _not_ run commit under these circumstances because handle->h_ref
1498 * is elevated. We'll still have enough credits for the tiny quotafile
1501 static int do_journal_get_write_access(handle_t
*handle
,
1502 struct buffer_head
*bh
)
1504 if (!buffer_mapped(bh
) || buffer_freed(bh
))
1506 return ext4_journal_get_write_access(handle
, bh
);
1510 * Truncate blocks that were not used by write. We have to truncate the
1511 * pagecache as well so that corresponding buffers get properly unmapped.
1513 static void ext4_truncate_failed_write(struct inode
*inode
)
1515 truncate_inode_pages(inode
->i_mapping
, inode
->i_size
);
1516 ext4_truncate(inode
);
1519 static int ext4_write_begin(struct file
*file
, struct address_space
*mapping
,
1520 loff_t pos
, unsigned len
, unsigned flags
,
1521 struct page
**pagep
, void **fsdata
)
1523 struct inode
*inode
= mapping
->host
;
1524 int ret
, needed_blocks
;
1531 trace_ext4_write_begin(inode
, pos
, len
, flags
);
1533 * Reserve one block more for addition to orphan list in case
1534 * we allocate blocks but write fails for some reason
1536 needed_blocks
= ext4_writepage_trans_blocks(inode
) + 1;
1537 index
= pos
>> PAGE_CACHE_SHIFT
;
1538 from
= pos
& (PAGE_CACHE_SIZE
- 1);
1542 handle
= ext4_journal_start(inode
, needed_blocks
);
1543 if (IS_ERR(handle
)) {
1544 ret
= PTR_ERR(handle
);
1548 /* We cannot recurse into the filesystem as the transaction is already
1550 flags
|= AOP_FLAG_NOFS
;
1552 page
= grab_cache_page_write_begin(mapping
, index
, flags
);
1554 ext4_journal_stop(handle
);
1560 ret
= block_write_begin(file
, mapping
, pos
, len
, flags
, pagep
, fsdata
,
1563 if (!ret
&& ext4_should_journal_data(inode
)) {
1564 ret
= walk_page_buffers(handle
, page_buffers(page
),
1565 from
, to
, NULL
, do_journal_get_write_access
);
1570 page_cache_release(page
);
1572 * block_write_begin may have instantiated a few blocks
1573 * outside i_size. Trim these off again. Don't need
1574 * i_size_read because we hold i_mutex.
1576 * Add inode to orphan list in case we crash before
1579 if (pos
+ len
> inode
->i_size
&& ext4_can_truncate(inode
))
1580 ext4_orphan_add(handle
, inode
);
1582 ext4_journal_stop(handle
);
1583 if (pos
+ len
> inode
->i_size
) {
1584 ext4_truncate_failed_write(inode
);
1586 * If truncate failed early the inode might
1587 * still be on the orphan list; we need to
1588 * make sure the inode is removed from the
1589 * orphan list in that case.
1592 ext4_orphan_del(NULL
, inode
);
1596 if (ret
== -ENOSPC
&& ext4_should_retry_alloc(inode
->i_sb
, &retries
))
1602 /* For write_end() in data=journal mode */
1603 static int write_end_fn(handle_t
*handle
, struct buffer_head
*bh
)
1605 if (!buffer_mapped(bh
) || buffer_freed(bh
))
1607 set_buffer_uptodate(bh
);
1608 return ext4_handle_dirty_metadata(handle
, NULL
, bh
);
1611 static int ext4_generic_write_end(struct file
*file
,
1612 struct address_space
*mapping
,
1613 loff_t pos
, unsigned len
, unsigned copied
,
1614 struct page
*page
, void *fsdata
)
1616 int i_size_changed
= 0;
1617 struct inode
*inode
= mapping
->host
;
1618 handle_t
*handle
= ext4_journal_current_handle();
1620 copied
= block_write_end(file
, mapping
, pos
, len
, copied
, page
, fsdata
);
1623 * No need to use i_size_read() here, the i_size
1624 * cannot change under us because we hold i_mutex.
1626 * But it's important to update i_size while still holding page lock:
1627 * page writeout could otherwise come in and zero beyond i_size.
1629 if (pos
+ copied
> inode
->i_size
) {
1630 i_size_write(inode
, pos
+ copied
);
1634 if (pos
+ copied
> EXT4_I(inode
)->i_disksize
) {
1635 /* We need to mark inode dirty even if
1636 * new_i_size is less that inode->i_size
1637 * bu greater than i_disksize.(hint delalloc)
1639 ext4_update_i_disksize(inode
, (pos
+ copied
));
1643 page_cache_release(page
);
1646 * Don't mark the inode dirty under page lock. First, it unnecessarily
1647 * makes the holding time of page lock longer. Second, it forces lock
1648 * ordering of page lock and transaction start for journaling
1652 ext4_mark_inode_dirty(handle
, inode
);
1658 * We need to pick up the new inode size which generic_commit_write gave us
1659 * `file' can be NULL - eg, when called from page_symlink().
1661 * ext4 never places buffers on inode->i_mapping->private_list. metadata
1662 * buffers are managed internally.
1664 static int ext4_ordered_write_end(struct file
*file
,
1665 struct address_space
*mapping
,
1666 loff_t pos
, unsigned len
, unsigned copied
,
1667 struct page
*page
, void *fsdata
)
1669 handle_t
*handle
= ext4_journal_current_handle();
1670 struct inode
*inode
= mapping
->host
;
1673 trace_ext4_ordered_write_end(inode
, pos
, len
, copied
);
1674 ret
= ext4_jbd2_file_inode(handle
, inode
);
1677 ret2
= ext4_generic_write_end(file
, mapping
, pos
, len
, copied
,
1680 if (pos
+ len
> inode
->i_size
&& ext4_can_truncate(inode
))
1681 /* if we have allocated more blocks and copied
1682 * less. We will have blocks allocated outside
1683 * inode->i_size. So truncate them
1685 ext4_orphan_add(handle
, inode
);
1689 ret2
= ext4_journal_stop(handle
);
1693 if (pos
+ len
> inode
->i_size
) {
1694 ext4_truncate_failed_write(inode
);
1696 * If truncate failed early the inode might still be
1697 * on the orphan list; we need to make sure the inode
1698 * is removed from the orphan list in that case.
1701 ext4_orphan_del(NULL
, inode
);
1705 return ret
? ret
: copied
;
1708 static int ext4_writeback_write_end(struct file
*file
,
1709 struct address_space
*mapping
,
1710 loff_t pos
, unsigned len
, unsigned copied
,
1711 struct page
*page
, void *fsdata
)
1713 handle_t
*handle
= ext4_journal_current_handle();
1714 struct inode
*inode
= mapping
->host
;
1717 trace_ext4_writeback_write_end(inode
, pos
, len
, copied
);
1718 ret2
= ext4_generic_write_end(file
, mapping
, pos
, len
, copied
,
1721 if (pos
+ len
> inode
->i_size
&& ext4_can_truncate(inode
))
1722 /* if we have allocated more blocks and copied
1723 * less. We will have blocks allocated outside
1724 * inode->i_size. So truncate them
1726 ext4_orphan_add(handle
, inode
);
1731 ret2
= ext4_journal_stop(handle
);
1735 if (pos
+ len
> inode
->i_size
) {
1736 ext4_truncate_failed_write(inode
);
1738 * If truncate failed early the inode might still be
1739 * on the orphan list; we need to make sure the inode
1740 * is removed from the orphan list in that case.
1743 ext4_orphan_del(NULL
, inode
);
1746 return ret
? ret
: copied
;
1749 static int ext4_journalled_write_end(struct file
*file
,
1750 struct address_space
*mapping
,
1751 loff_t pos
, unsigned len
, unsigned copied
,
1752 struct page
*page
, void *fsdata
)
1754 handle_t
*handle
= ext4_journal_current_handle();
1755 struct inode
*inode
= mapping
->host
;
1761 trace_ext4_journalled_write_end(inode
, pos
, len
, copied
);
1762 from
= pos
& (PAGE_CACHE_SIZE
- 1);
1766 if (!PageUptodate(page
))
1768 page_zero_new_buffers(page
, from
+copied
, to
);
1771 ret
= walk_page_buffers(handle
, page_buffers(page
), from
,
1772 to
, &partial
, write_end_fn
);
1774 SetPageUptodate(page
);
1775 new_i_size
= pos
+ copied
;
1776 if (new_i_size
> inode
->i_size
)
1777 i_size_write(inode
, pos
+copied
);
1778 EXT4_I(inode
)->i_state
|= EXT4_STATE_JDATA
;
1779 if (new_i_size
> EXT4_I(inode
)->i_disksize
) {
1780 ext4_update_i_disksize(inode
, new_i_size
);
1781 ret2
= ext4_mark_inode_dirty(handle
, inode
);
1787 page_cache_release(page
);
1788 if (pos
+ len
> inode
->i_size
&& ext4_can_truncate(inode
))
1789 /* if we have allocated more blocks and copied
1790 * less. We will have blocks allocated outside
1791 * inode->i_size. So truncate them
1793 ext4_orphan_add(handle
, inode
);
1795 ret2
= ext4_journal_stop(handle
);
1798 if (pos
+ len
> inode
->i_size
) {
1799 ext4_truncate_failed_write(inode
);
1801 * If truncate failed early the inode might still be
1802 * on the orphan list; we need to make sure the inode
1803 * is removed from the orphan list in that case.
1806 ext4_orphan_del(NULL
, inode
);
1809 return ret
? ret
: copied
;
1813 * Reserve a single block located at lblock
1815 static int ext4_da_reserve_space(struct inode
*inode
, sector_t lblock
)
1818 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
1819 struct ext4_inode_info
*ei
= EXT4_I(inode
);
1820 unsigned long md_needed
, md_reserved
;
1823 * recalculate the amount of metadata blocks to reserve
1824 * in order to allocate nrblocks
1825 * worse case is one extent per block
1828 spin_lock(&ei
->i_block_reservation_lock
);
1829 md_reserved
= ei
->i_reserved_meta_blocks
;
1830 md_needed
= ext4_calc_metadata_amount(inode
, lblock
);
1831 spin_unlock(&ei
->i_block_reservation_lock
);
1834 * Make quota reservation here to prevent quota overflow
1835 * later. Real quota accounting is done at pages writeout
1838 if (vfs_dq_reserve_block(inode
, md_needed
+ 1)) {
1840 * We tend to badly over-estimate the amount of
1841 * metadata blocks which are needed, so if we have
1842 * reserved any metadata blocks, try to force out the
1843 * inode and see if we have any better luck.
1845 if (md_reserved
&& retries
++ <= 3)
1850 if (ext4_claim_free_blocks(sbi
, md_needed
+ 1)) {
1851 vfs_dq_release_reservation_block(inode
, md_needed
+ 1);
1852 if (ext4_should_retry_alloc(inode
->i_sb
, &retries
)) {
1855 write_inode_now(inode
, (retries
== 3));
1861 spin_lock(&ei
->i_block_reservation_lock
);
1862 ei
->i_reserved_data_blocks
++;
1863 ei
->i_reserved_meta_blocks
+= md_needed
;
1864 spin_unlock(&ei
->i_block_reservation_lock
);
1866 return 0; /* success */
1869 static void ext4_da_release_space(struct inode
*inode
, int to_free
)
1871 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
1872 struct ext4_inode_info
*ei
= EXT4_I(inode
);
1875 return; /* Nothing to release, exit */
1877 spin_lock(&EXT4_I(inode
)->i_block_reservation_lock
);
1879 if (unlikely(to_free
> ei
->i_reserved_data_blocks
)) {
1881 * if there aren't enough reserved blocks, then the
1882 * counter is messed up somewhere. Since this
1883 * function is called from invalidate page, it's
1884 * harmless to return without any action.
1886 ext4_msg(inode
->i_sb
, KERN_NOTICE
, "ext4_da_release_space: "
1887 "ino %lu, to_free %d with only %d reserved "
1888 "data blocks\n", inode
->i_ino
, to_free
,
1889 ei
->i_reserved_data_blocks
);
1891 to_free
= ei
->i_reserved_data_blocks
;
1893 ei
->i_reserved_data_blocks
-= to_free
;
1895 if (ei
->i_reserved_data_blocks
== 0) {
1897 * We can release all of the reserved metadata blocks
1898 * only when we have written all of the delayed
1899 * allocation blocks.
1901 to_free
+= ei
->i_reserved_meta_blocks
;
1902 ei
->i_reserved_meta_blocks
= 0;
1903 ei
->i_da_metadata_calc_len
= 0;
1906 /* update fs dirty blocks counter */
1907 percpu_counter_sub(&sbi
->s_dirtyblocks_counter
, to_free
);
1909 spin_unlock(&EXT4_I(inode
)->i_block_reservation_lock
);
1911 vfs_dq_release_reservation_block(inode
, to_free
);
1914 static void ext4_da_page_release_reservation(struct page
*page
,
1915 unsigned long offset
)
1918 struct buffer_head
*head
, *bh
;
1919 unsigned int curr_off
= 0;
1921 head
= page_buffers(page
);
1924 unsigned int next_off
= curr_off
+ bh
->b_size
;
1926 if ((offset
<= curr_off
) && (buffer_delay(bh
))) {
1928 clear_buffer_delay(bh
);
1930 curr_off
= next_off
;
1931 } while ((bh
= bh
->b_this_page
) != head
);
1932 ext4_da_release_space(page
->mapping
->host
, to_release
);
1936 * Delayed allocation stuff
1940 * mpage_da_submit_io - walks through extent of pages and try to write
1941 * them with writepage() call back
1943 * @mpd->inode: inode
1944 * @mpd->first_page: first page of the extent
1945 * @mpd->next_page: page after the last page of the extent
1947 * By the time mpage_da_submit_io() is called we expect all blocks
1948 * to be allocated. this may be wrong if allocation failed.
1950 * As pages are already locked by write_cache_pages(), we can't use it
1952 static int mpage_da_submit_io(struct mpage_da_data
*mpd
)
1955 struct pagevec pvec
;
1956 unsigned long index
, end
;
1957 int ret
= 0, err
, nr_pages
, i
;
1958 struct inode
*inode
= mpd
->inode
;
1959 struct address_space
*mapping
= inode
->i_mapping
;
1961 BUG_ON(mpd
->next_page
<= mpd
->first_page
);
1963 * We need to start from the first_page to the next_page - 1
1964 * to make sure we also write the mapped dirty buffer_heads.
1965 * If we look at mpd->b_blocknr we would only be looking
1966 * at the currently mapped buffer_heads.
1968 index
= mpd
->first_page
;
1969 end
= mpd
->next_page
- 1;
1971 pagevec_init(&pvec
, 0);
1972 while (index
<= end
) {
1973 nr_pages
= pagevec_lookup(&pvec
, mapping
, index
, PAGEVEC_SIZE
);
1976 for (i
= 0; i
< nr_pages
; i
++) {
1977 struct page
*page
= pvec
.pages
[i
];
1979 index
= page
->index
;
1984 BUG_ON(!PageLocked(page
));
1985 BUG_ON(PageWriteback(page
));
1987 pages_skipped
= mpd
->wbc
->pages_skipped
;
1988 err
= mapping
->a_ops
->writepage(page
, mpd
->wbc
);
1989 if (!err
&& (pages_skipped
== mpd
->wbc
->pages_skipped
))
1991 * have successfully written the page
1992 * without skipping the same
1994 mpd
->pages_written
++;
1996 * In error case, we have to continue because
1997 * remaining pages are still locked
1998 * XXX: unlock and re-dirty them?
2003 pagevec_release(&pvec
);
2009 * mpage_put_bnr_to_bhs - walk blocks and assign them actual numbers
2011 * @mpd->inode - inode to walk through
2012 * @exbh->b_blocknr - first block on a disk
2013 * @exbh->b_size - amount of space in bytes
2014 * @logical - first logical block to start assignment with
2016 * the function goes through all passed space and put actual disk
2017 * block numbers into buffer heads, dropping BH_Delay and BH_Unwritten
2019 static void mpage_put_bnr_to_bhs(struct mpage_da_data
*mpd
, sector_t logical
,
2020 struct buffer_head
*exbh
)
2022 struct inode
*inode
= mpd
->inode
;
2023 struct address_space
*mapping
= inode
->i_mapping
;
2024 int blocks
= exbh
->b_size
>> inode
->i_blkbits
;
2025 sector_t pblock
= exbh
->b_blocknr
, cur_logical
;
2026 struct buffer_head
*head
, *bh
;
2028 struct pagevec pvec
;
2031 index
= logical
>> (PAGE_CACHE_SHIFT
- inode
->i_blkbits
);
2032 end
= (logical
+ blocks
- 1) >> (PAGE_CACHE_SHIFT
- inode
->i_blkbits
);
2033 cur_logical
= index
<< (PAGE_CACHE_SHIFT
- inode
->i_blkbits
);
2035 pagevec_init(&pvec
, 0);
2037 while (index
<= end
) {
2038 /* XXX: optimize tail */
2039 nr_pages
= pagevec_lookup(&pvec
, mapping
, index
, PAGEVEC_SIZE
);
2042 for (i
= 0; i
< nr_pages
; i
++) {
2043 struct page
*page
= pvec
.pages
[i
];
2045 index
= page
->index
;
2050 BUG_ON(!PageLocked(page
));
2051 BUG_ON(PageWriteback(page
));
2052 BUG_ON(!page_has_buffers(page
));
2054 bh
= page_buffers(page
);
2057 /* skip blocks out of the range */
2059 if (cur_logical
>= logical
)
2062 } while ((bh
= bh
->b_this_page
) != head
);
2065 if (cur_logical
>= logical
+ blocks
)
2068 if (buffer_delay(bh
) ||
2069 buffer_unwritten(bh
)) {
2071 BUG_ON(bh
->b_bdev
!= inode
->i_sb
->s_bdev
);
2073 if (buffer_delay(bh
)) {
2074 clear_buffer_delay(bh
);
2075 bh
->b_blocknr
= pblock
;
2078 * unwritten already should have
2079 * blocknr assigned. Verify that
2081 clear_buffer_unwritten(bh
);
2082 BUG_ON(bh
->b_blocknr
!= pblock
);
2085 } else if (buffer_mapped(bh
))
2086 BUG_ON(bh
->b_blocknr
!= pblock
);
2090 } while ((bh
= bh
->b_this_page
) != head
);
2092 pagevec_release(&pvec
);
2098 * __unmap_underlying_blocks - just a helper function to unmap
2099 * set of blocks described by @bh
2101 static inline void __unmap_underlying_blocks(struct inode
*inode
,
2102 struct buffer_head
*bh
)
2104 struct block_device
*bdev
= inode
->i_sb
->s_bdev
;
2107 blocks
= bh
->b_size
>> inode
->i_blkbits
;
2108 for (i
= 0; i
< blocks
; i
++)
2109 unmap_underlying_metadata(bdev
, bh
->b_blocknr
+ i
);
2112 static void ext4_da_block_invalidatepages(struct mpage_da_data
*mpd
,
2113 sector_t logical
, long blk_cnt
)
2117 struct pagevec pvec
;
2118 struct inode
*inode
= mpd
->inode
;
2119 struct address_space
*mapping
= inode
->i_mapping
;
2121 index
= logical
>> (PAGE_CACHE_SHIFT
- inode
->i_blkbits
);
2122 end
= (logical
+ blk_cnt
- 1) >>
2123 (PAGE_CACHE_SHIFT
- inode
->i_blkbits
);
2124 while (index
<= end
) {
2125 nr_pages
= pagevec_lookup(&pvec
, mapping
, index
, PAGEVEC_SIZE
);
2128 for (i
= 0; i
< nr_pages
; i
++) {
2129 struct page
*page
= pvec
.pages
[i
];
2130 index
= page
->index
;
2135 BUG_ON(!PageLocked(page
));
2136 BUG_ON(PageWriteback(page
));
2137 block_invalidatepage(page
, 0);
2138 ClearPageUptodate(page
);
2145 static void ext4_print_free_blocks(struct inode
*inode
)
2147 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
2148 printk(KERN_CRIT
"Total free blocks count %lld\n",
2149 ext4_count_free_blocks(inode
->i_sb
));
2150 printk(KERN_CRIT
"Free/Dirty block details\n");
2151 printk(KERN_CRIT
"free_blocks=%lld\n",
2152 (long long) percpu_counter_sum(&sbi
->s_freeblocks_counter
));
2153 printk(KERN_CRIT
"dirty_blocks=%lld\n",
2154 (long long) percpu_counter_sum(&sbi
->s_dirtyblocks_counter
));
2155 printk(KERN_CRIT
"Block reservation details\n");
2156 printk(KERN_CRIT
"i_reserved_data_blocks=%u\n",
2157 EXT4_I(inode
)->i_reserved_data_blocks
);
2158 printk(KERN_CRIT
"i_reserved_meta_blocks=%u\n",
2159 EXT4_I(inode
)->i_reserved_meta_blocks
);
2164 * mpage_da_map_blocks - go through given space
2166 * @mpd - bh describing space
2168 * The function skips space we know is already mapped to disk blocks.
2171 static int mpage_da_map_blocks(struct mpage_da_data
*mpd
)
2173 int err
, blks
, get_blocks_flags
;
2174 struct buffer_head
new;
2175 sector_t next
= mpd
->b_blocknr
;
2176 unsigned max_blocks
= mpd
->b_size
>> mpd
->inode
->i_blkbits
;
2177 loff_t disksize
= EXT4_I(mpd
->inode
)->i_disksize
;
2178 handle_t
*handle
= NULL
;
2181 * We consider only non-mapped and non-allocated blocks
2183 if ((mpd
->b_state
& (1 << BH_Mapped
)) &&
2184 !(mpd
->b_state
& (1 << BH_Delay
)) &&
2185 !(mpd
->b_state
& (1 << BH_Unwritten
)))
2189 * If we didn't accumulate anything to write simply return
2194 handle
= ext4_journal_current_handle();
2198 * Call ext4_get_blocks() to allocate any delayed allocation
2199 * blocks, or to convert an uninitialized extent to be
2200 * initialized (in the case where we have written into
2201 * one or more preallocated blocks).
2203 * We pass in the magic EXT4_GET_BLOCKS_DELALLOC_RESERVE to
2204 * indicate that we are on the delayed allocation path. This
2205 * affects functions in many different parts of the allocation
2206 * call path. This flag exists primarily because we don't
2207 * want to change *many* call functions, so ext4_get_blocks()
2208 * will set the magic i_delalloc_reserved_flag once the
2209 * inode's allocation semaphore is taken.
2211 * If the blocks in questions were delalloc blocks, set
2212 * EXT4_GET_BLOCKS_DELALLOC_RESERVE so the delalloc accounting
2213 * variables are updated after the blocks have been allocated.
2216 get_blocks_flags
= (EXT4_GET_BLOCKS_CREATE
|
2217 EXT4_GET_BLOCKS_DELALLOC_RESERVE
);
2218 if (mpd
->b_state
& (1 << BH_Delay
))
2219 get_blocks_flags
|= EXT4_GET_BLOCKS_UPDATE_RESERVE_SPACE
;
2220 blks
= ext4_get_blocks(handle
, mpd
->inode
, next
, max_blocks
,
2221 &new, get_blocks_flags
);
2225 * If get block returns with error we simply
2226 * return. Later writepage will redirty the page and
2227 * writepages will find the dirty page again
2232 if (err
== -ENOSPC
&&
2233 ext4_count_free_blocks(mpd
->inode
->i_sb
)) {
2239 * get block failure will cause us to loop in
2240 * writepages, because a_ops->writepage won't be able
2241 * to make progress. The page will be redirtied by
2242 * writepage and writepages will again try to write
2245 ext4_msg(mpd
->inode
->i_sb
, KERN_CRIT
,
2246 "delayed block allocation failed for inode %lu at "
2247 "logical offset %llu with max blocks %zd with "
2248 "error %d\n", mpd
->inode
->i_ino
,
2249 (unsigned long long) next
,
2250 mpd
->b_size
>> mpd
->inode
->i_blkbits
, err
);
2251 printk(KERN_CRIT
"This should not happen!! "
2252 "Data will be lost\n");
2253 if (err
== -ENOSPC
) {
2254 ext4_print_free_blocks(mpd
->inode
);
2256 /* invalidate all the pages */
2257 ext4_da_block_invalidatepages(mpd
, next
,
2258 mpd
->b_size
>> mpd
->inode
->i_blkbits
);
2263 new.b_size
= (blks
<< mpd
->inode
->i_blkbits
);
2265 if (buffer_new(&new))
2266 __unmap_underlying_blocks(mpd
->inode
, &new);
2269 * If blocks are delayed marked, we need to
2270 * put actual blocknr and drop delayed bit
2272 if ((mpd
->b_state
& (1 << BH_Delay
)) ||
2273 (mpd
->b_state
& (1 << BH_Unwritten
)))
2274 mpage_put_bnr_to_bhs(mpd
, next
, &new);
2276 if (ext4_should_order_data(mpd
->inode
)) {
2277 err
= ext4_jbd2_file_inode(handle
, mpd
->inode
);
2283 * Update on-disk size along with block allocation.
2285 disksize
= ((loff_t
) next
+ blks
) << mpd
->inode
->i_blkbits
;
2286 if (disksize
> i_size_read(mpd
->inode
))
2287 disksize
= i_size_read(mpd
->inode
);
2288 if (disksize
> EXT4_I(mpd
->inode
)->i_disksize
) {
2289 ext4_update_i_disksize(mpd
->inode
, disksize
);
2290 return ext4_mark_inode_dirty(handle
, mpd
->inode
);
2296 #define BH_FLAGS ((1 << BH_Uptodate) | (1 << BH_Mapped) | \
2297 (1 << BH_Delay) | (1 << BH_Unwritten))
2300 * mpage_add_bh_to_extent - try to add one more block to extent of blocks
2302 * @mpd->lbh - extent of blocks
2303 * @logical - logical number of the block in the file
2304 * @bh - bh of the block (used to access block's state)
2306 * the function is used to collect contig. blocks in same state
2308 static void mpage_add_bh_to_extent(struct mpage_da_data
*mpd
,
2309 sector_t logical
, size_t b_size
,
2310 unsigned long b_state
)
2313 int nrblocks
= mpd
->b_size
>> mpd
->inode
->i_blkbits
;
2315 /* check if thereserved journal credits might overflow */
2316 if (!(EXT4_I(mpd
->inode
)->i_flags
& EXT4_EXTENTS_FL
)) {
2317 if (nrblocks
>= EXT4_MAX_TRANS_DATA
) {
2319 * With non-extent format we are limited by the journal
2320 * credit available. Total credit needed to insert
2321 * nrblocks contiguous blocks is dependent on the
2322 * nrblocks. So limit nrblocks.
2325 } else if ((nrblocks
+ (b_size
>> mpd
->inode
->i_blkbits
)) >
2326 EXT4_MAX_TRANS_DATA
) {
2328 * Adding the new buffer_head would make it cross the
2329 * allowed limit for which we have journal credit
2330 * reserved. So limit the new bh->b_size
2332 b_size
= (EXT4_MAX_TRANS_DATA
- nrblocks
) <<
2333 mpd
->inode
->i_blkbits
;
2334 /* we will do mpage_da_submit_io in the next loop */
2338 * First block in the extent
2340 if (mpd
->b_size
== 0) {
2341 mpd
->b_blocknr
= logical
;
2342 mpd
->b_size
= b_size
;
2343 mpd
->b_state
= b_state
& BH_FLAGS
;
2347 next
= mpd
->b_blocknr
+ nrblocks
;
2349 * Can we merge the block to our big extent?
2351 if (logical
== next
&& (b_state
& BH_FLAGS
) == mpd
->b_state
) {
2352 mpd
->b_size
+= b_size
;
2358 * We couldn't merge the block to our extent, so we
2359 * need to flush current extent and start new one
2361 if (mpage_da_map_blocks(mpd
) == 0)
2362 mpage_da_submit_io(mpd
);
2367 static int ext4_bh_delay_or_unwritten(handle_t
*handle
, struct buffer_head
*bh
)
2369 return (buffer_delay(bh
) || buffer_unwritten(bh
)) && buffer_dirty(bh
);
2373 * __mpage_da_writepage - finds extent of pages and blocks
2375 * @page: page to consider
2376 * @wbc: not used, we just follow rules
2379 * The function finds extents of pages and scan them for all blocks.
2381 static int __mpage_da_writepage(struct page
*page
,
2382 struct writeback_control
*wbc
, void *data
)
2384 struct mpage_da_data
*mpd
= data
;
2385 struct inode
*inode
= mpd
->inode
;
2386 struct buffer_head
*bh
, *head
;
2391 * Rest of the page in the page_vec
2392 * redirty then and skip then. We will
2393 * try to write them again after
2394 * starting a new transaction
2396 redirty_page_for_writepage(wbc
, page
);
2398 return MPAGE_DA_EXTENT_TAIL
;
2401 * Can we merge this page to current extent?
2403 if (mpd
->next_page
!= page
->index
) {
2405 * Nope, we can't. So, we map non-allocated blocks
2406 * and start IO on them using writepage()
2408 if (mpd
->next_page
!= mpd
->first_page
) {
2409 if (mpage_da_map_blocks(mpd
) == 0)
2410 mpage_da_submit_io(mpd
);
2412 * skip rest of the page in the page_vec
2415 redirty_page_for_writepage(wbc
, page
);
2417 return MPAGE_DA_EXTENT_TAIL
;
2421 * Start next extent of pages ...
2423 mpd
->first_page
= page
->index
;
2433 mpd
->next_page
= page
->index
+ 1;
2434 logical
= (sector_t
) page
->index
<<
2435 (PAGE_CACHE_SHIFT
- inode
->i_blkbits
);
2437 if (!page_has_buffers(page
)) {
2438 mpage_add_bh_to_extent(mpd
, logical
, PAGE_CACHE_SIZE
,
2439 (1 << BH_Dirty
) | (1 << BH_Uptodate
));
2441 return MPAGE_DA_EXTENT_TAIL
;
2444 * Page with regular buffer heads, just add all dirty ones
2446 head
= page_buffers(page
);
2449 BUG_ON(buffer_locked(bh
));
2451 * We need to try to allocate
2452 * unmapped blocks in the same page.
2453 * Otherwise we won't make progress
2454 * with the page in ext4_writepage
2456 if (ext4_bh_delay_or_unwritten(NULL
, bh
)) {
2457 mpage_add_bh_to_extent(mpd
, logical
,
2461 return MPAGE_DA_EXTENT_TAIL
;
2462 } else if (buffer_dirty(bh
) && (buffer_mapped(bh
))) {
2464 * mapped dirty buffer. We need to update
2465 * the b_state because we look at
2466 * b_state in mpage_da_map_blocks. We don't
2467 * update b_size because if we find an
2468 * unmapped buffer_head later we need to
2469 * use the b_state flag of that buffer_head.
2471 if (mpd
->b_size
== 0)
2472 mpd
->b_state
= bh
->b_state
& BH_FLAGS
;
2475 } while ((bh
= bh
->b_this_page
) != head
);
2482 * This is a special get_blocks_t callback which is used by
2483 * ext4_da_write_begin(). It will either return mapped block or
2484 * reserve space for a single block.
2486 * For delayed buffer_head we have BH_Mapped, BH_New, BH_Delay set.
2487 * We also have b_blocknr = -1 and b_bdev initialized properly
2489 * For unwritten buffer_head we have BH_Mapped, BH_New, BH_Unwritten set.
2490 * We also have b_blocknr = physicalblock mapping unwritten extent and b_bdev
2491 * initialized properly.
2493 static int ext4_da_get_block_prep(struct inode
*inode
, sector_t iblock
,
2494 struct buffer_head
*bh_result
, int create
)
2497 sector_t invalid_block
= ~((sector_t
) 0xffff);
2499 if (invalid_block
< ext4_blocks_count(EXT4_SB(inode
->i_sb
)->s_es
))
2502 BUG_ON(create
== 0);
2503 BUG_ON(bh_result
->b_size
!= inode
->i_sb
->s_blocksize
);
2506 * first, we need to know whether the block is allocated already
2507 * preallocated blocks are unmapped but should treated
2508 * the same as allocated blocks.
2510 ret
= ext4_get_blocks(NULL
, inode
, iblock
, 1, bh_result
, 0);
2511 if ((ret
== 0) && !buffer_delay(bh_result
)) {
2512 /* the block isn't (pre)allocated yet, let's reserve space */
2514 * XXX: __block_prepare_write() unmaps passed block,
2517 ret
= ext4_da_reserve_space(inode
, iblock
);
2519 /* not enough space to reserve */
2522 map_bh(bh_result
, inode
->i_sb
, invalid_block
);
2523 set_buffer_new(bh_result
);
2524 set_buffer_delay(bh_result
);
2525 } else if (ret
> 0) {
2526 bh_result
->b_size
= (ret
<< inode
->i_blkbits
);
2527 if (buffer_unwritten(bh_result
)) {
2528 /* A delayed write to unwritten bh should
2529 * be marked new and mapped. Mapped ensures
2530 * that we don't do get_block multiple times
2531 * when we write to the same offset and new
2532 * ensures that we do proper zero out for
2535 set_buffer_new(bh_result
);
2536 set_buffer_mapped(bh_result
);
2545 * This function is used as a standard get_block_t calback function
2546 * when there is no desire to allocate any blocks. It is used as a
2547 * callback function for block_prepare_write(), nobh_writepage(), and
2548 * block_write_full_page(). These functions should only try to map a
2549 * single block at a time.
2551 * Since this function doesn't do block allocations even if the caller
2552 * requests it by passing in create=1, it is critically important that
2553 * any caller checks to make sure that any buffer heads are returned
2554 * by this function are either all already mapped or marked for
2555 * delayed allocation before calling nobh_writepage() or
2556 * block_write_full_page(). Otherwise, b_blocknr could be left
2557 * unitialized, and the page write functions will be taken by
2560 static int noalloc_get_block_write(struct inode
*inode
, sector_t iblock
,
2561 struct buffer_head
*bh_result
, int create
)
2564 unsigned max_blocks
= bh_result
->b_size
>> inode
->i_blkbits
;
2566 BUG_ON(bh_result
->b_size
!= inode
->i_sb
->s_blocksize
);
2569 * we don't want to do block allocation in writepage
2570 * so call get_block_wrap with create = 0
2572 ret
= ext4_get_blocks(NULL
, inode
, iblock
, max_blocks
, bh_result
, 0);
2574 bh_result
->b_size
= (ret
<< inode
->i_blkbits
);
2580 static int bget_one(handle_t
*handle
, struct buffer_head
*bh
)
2586 static int bput_one(handle_t
*handle
, struct buffer_head
*bh
)
2592 static int __ext4_journalled_writepage(struct page
*page
,
2595 struct address_space
*mapping
= page
->mapping
;
2596 struct inode
*inode
= mapping
->host
;
2597 struct buffer_head
*page_bufs
;
2598 handle_t
*handle
= NULL
;
2602 page_bufs
= page_buffers(page
);
2604 walk_page_buffers(handle
, page_bufs
, 0, len
, NULL
, bget_one
);
2605 /* As soon as we unlock the page, it can go away, but we have
2606 * references to buffers so we are safe */
2609 handle
= ext4_journal_start(inode
, ext4_writepage_trans_blocks(inode
));
2610 if (IS_ERR(handle
)) {
2611 ret
= PTR_ERR(handle
);
2615 ret
= walk_page_buffers(handle
, page_bufs
, 0, len
, NULL
,
2616 do_journal_get_write_access
);
2618 err
= walk_page_buffers(handle
, page_bufs
, 0, len
, NULL
,
2622 err
= ext4_journal_stop(handle
);
2626 walk_page_buffers(handle
, page_bufs
, 0, len
, NULL
, bput_one
);
2627 EXT4_I(inode
)->i_state
|= EXT4_STATE_JDATA
;
2633 * Note that we don't need to start a transaction unless we're journaling data
2634 * because we should have holes filled from ext4_page_mkwrite(). We even don't
2635 * need to file the inode to the transaction's list in ordered mode because if
2636 * we are writing back data added by write(), the inode is already there and if
2637 * we are writing back data modified via mmap(), noone guarantees in which
2638 * transaction the data will hit the disk. In case we are journaling data, we
2639 * cannot start transaction directly because transaction start ranks above page
2640 * lock so we have to do some magic.
2642 * This function can get called via...
2643 * - ext4_da_writepages after taking page lock (have journal handle)
2644 * - journal_submit_inode_data_buffers (no journal handle)
2645 * - shrink_page_list via pdflush (no journal handle)
2646 * - grab_page_cache when doing write_begin (have journal handle)
2648 * We don't do any block allocation in this function. If we have page with
2649 * multiple blocks we need to write those buffer_heads that are mapped. This
2650 * is important for mmaped based write. So if we do with blocksize 1K
2651 * truncate(f, 1024);
2652 * a = mmap(f, 0, 4096);
2654 * truncate(f, 4096);
2655 * we have in the page first buffer_head mapped via page_mkwrite call back
2656 * but other bufer_heads would be unmapped but dirty(dirty done via the
2657 * do_wp_page). So writepage should write the first block. If we modify
2658 * the mmap area beyond 1024 we will again get a page_fault and the
2659 * page_mkwrite callback will do the block allocation and mark the
2660 * buffer_heads mapped.
2662 * We redirty the page if we have any buffer_heads that is either delay or
2663 * unwritten in the page.
2665 * We can get recursively called as show below.
2667 * ext4_writepage() -> kmalloc() -> __alloc_pages() -> page_launder() ->
2670 * But since we don't do any block allocation we should not deadlock.
2671 * Page also have the dirty flag cleared so we don't get recurive page_lock.
2673 static int ext4_writepage(struct page
*page
,
2674 struct writeback_control
*wbc
)
2679 struct buffer_head
*page_bufs
;
2680 struct inode
*inode
= page
->mapping
->host
;
2682 trace_ext4_writepage(inode
, page
);
2683 size
= i_size_read(inode
);
2684 if (page
->index
== size
>> PAGE_CACHE_SHIFT
)
2685 len
= size
& ~PAGE_CACHE_MASK
;
2687 len
= PAGE_CACHE_SIZE
;
2689 if (page_has_buffers(page
)) {
2690 page_bufs
= page_buffers(page
);
2691 if (walk_page_buffers(NULL
, page_bufs
, 0, len
, NULL
,
2692 ext4_bh_delay_or_unwritten
)) {
2694 * We don't want to do block allocation
2695 * So redirty the page and return
2696 * We may reach here when we do a journal commit
2697 * via journal_submit_inode_data_buffers.
2698 * If we don't have mapping block we just ignore
2699 * them. We can also reach here via shrink_page_list
2701 redirty_page_for_writepage(wbc
, page
);
2707 * The test for page_has_buffers() is subtle:
2708 * We know the page is dirty but it lost buffers. That means
2709 * that at some moment in time after write_begin()/write_end()
2710 * has been called all buffers have been clean and thus they
2711 * must have been written at least once. So they are all
2712 * mapped and we can happily proceed with mapping them
2713 * and writing the page.
2715 * Try to initialize the buffer_heads and check whether
2716 * all are mapped and non delay. We don't want to
2717 * do block allocation here.
2719 ret
= block_prepare_write(page
, 0, len
,
2720 noalloc_get_block_write
);
2722 page_bufs
= page_buffers(page
);
2723 /* check whether all are mapped and non delay */
2724 if (walk_page_buffers(NULL
, page_bufs
, 0, len
, NULL
,
2725 ext4_bh_delay_or_unwritten
)) {
2726 redirty_page_for_writepage(wbc
, page
);
2732 * We can't do block allocation here
2733 * so just redity the page and unlock
2736 redirty_page_for_writepage(wbc
, page
);
2740 /* now mark the buffer_heads as dirty and uptodate */
2741 block_commit_write(page
, 0, len
);
2744 if (PageChecked(page
) && ext4_should_journal_data(inode
)) {
2746 * It's mmapped pagecache. Add buffers and journal it. There
2747 * doesn't seem much point in redirtying the page here.
2749 ClearPageChecked(page
);
2750 return __ext4_journalled_writepage(page
, len
);
2753 if (test_opt(inode
->i_sb
, NOBH
) && ext4_should_writeback_data(inode
))
2754 ret
= nobh_writepage(page
, noalloc_get_block_write
, wbc
);
2756 ret
= block_write_full_page(page
, noalloc_get_block_write
,
2763 * This is called via ext4_da_writepages() to
2764 * calulate the total number of credits to reserve to fit
2765 * a single extent allocation into a single transaction,
2766 * ext4_da_writpeages() will loop calling this before
2767 * the block allocation.
2770 static int ext4_da_writepages_trans_blocks(struct inode
*inode
)
2772 int max_blocks
= EXT4_I(inode
)->i_reserved_data_blocks
;
2775 * With non-extent format the journal credit needed to
2776 * insert nrblocks contiguous block is dependent on
2777 * number of contiguous block. So we will limit
2778 * number of contiguous block to a sane value
2780 if (!(EXT4_I(inode
)->i_flags
& EXT4_EXTENTS_FL
) &&
2781 (max_blocks
> EXT4_MAX_TRANS_DATA
))
2782 max_blocks
= EXT4_MAX_TRANS_DATA
;
2784 return ext4_chunk_trans_blocks(inode
, max_blocks
);
2787 static int ext4_da_writepages(struct address_space
*mapping
,
2788 struct writeback_control
*wbc
)
2791 int range_whole
= 0;
2792 handle_t
*handle
= NULL
;
2793 struct mpage_da_data mpd
;
2794 struct inode
*inode
= mapping
->host
;
2795 int no_nrwrite_index_update
;
2796 int pages_written
= 0;
2798 unsigned int max_pages
;
2799 int range_cyclic
, cycled
= 1, io_done
= 0;
2800 int needed_blocks
, ret
= 0;
2801 long desired_nr_to_write
, nr_to_writebump
= 0;
2802 loff_t range_start
= wbc
->range_start
;
2803 struct ext4_sb_info
*sbi
= EXT4_SB(mapping
->host
->i_sb
);
2805 trace_ext4_da_writepages(inode
, wbc
);
2808 * No pages to write? This is mainly a kludge to avoid starting
2809 * a transaction for special inodes like journal inode on last iput()
2810 * because that could violate lock ordering on umount
2812 if (!mapping
->nrpages
|| !mapping_tagged(mapping
, PAGECACHE_TAG_DIRTY
))
2816 * If the filesystem has aborted, it is read-only, so return
2817 * right away instead of dumping stack traces later on that
2818 * will obscure the real source of the problem. We test
2819 * EXT4_MF_FS_ABORTED instead of sb->s_flag's MS_RDONLY because
2820 * the latter could be true if the filesystem is mounted
2821 * read-only, and in that case, ext4_da_writepages should
2822 * *never* be called, so if that ever happens, we would want
2825 if (unlikely(sbi
->s_mount_flags
& EXT4_MF_FS_ABORTED
))
2828 if (wbc
->range_start
== 0 && wbc
->range_end
== LLONG_MAX
)
2831 range_cyclic
= wbc
->range_cyclic
;
2832 if (wbc
->range_cyclic
) {
2833 index
= mapping
->writeback_index
;
2836 wbc
->range_start
= index
<< PAGE_CACHE_SHIFT
;
2837 wbc
->range_end
= LLONG_MAX
;
2838 wbc
->range_cyclic
= 0;
2840 index
= wbc
->range_start
>> PAGE_CACHE_SHIFT
;
2843 * This works around two forms of stupidity. The first is in
2844 * the writeback code, which caps the maximum number of pages
2845 * written to be 1024 pages. This is wrong on multiple
2846 * levels; different architectues have a different page size,
2847 * which changes the maximum amount of data which gets
2848 * written. Secondly, 4 megabytes is way too small. XFS
2849 * forces this value to be 16 megabytes by multiplying
2850 * nr_to_write parameter by four, and then relies on its
2851 * allocator to allocate larger extents to make them
2852 * contiguous. Unfortunately this brings us to the second
2853 * stupidity, which is that ext4's mballoc code only allocates
2854 * at most 2048 blocks. So we force contiguous writes up to
2855 * the number of dirty blocks in the inode, or
2856 * sbi->max_writeback_mb_bump whichever is smaller.
2858 max_pages
= sbi
->s_max_writeback_mb_bump
<< (20 - PAGE_CACHE_SHIFT
);
2859 if (!range_cyclic
&& range_whole
)
2860 desired_nr_to_write
= wbc
->nr_to_write
* 8;
2862 desired_nr_to_write
= ext4_num_dirty_pages(inode
, index
,
2864 if (desired_nr_to_write
> max_pages
)
2865 desired_nr_to_write
= max_pages
;
2867 if (wbc
->nr_to_write
< desired_nr_to_write
) {
2868 nr_to_writebump
= desired_nr_to_write
- wbc
->nr_to_write
;
2869 wbc
->nr_to_write
= desired_nr_to_write
;
2873 mpd
.inode
= mapping
->host
;
2876 * we don't want write_cache_pages to update
2877 * nr_to_write and writeback_index
2879 no_nrwrite_index_update
= wbc
->no_nrwrite_index_update
;
2880 wbc
->no_nrwrite_index_update
= 1;
2881 pages_skipped
= wbc
->pages_skipped
;
2884 while (!ret
&& wbc
->nr_to_write
> 0) {
2887 * we insert one extent at a time. So we need
2888 * credit needed for single extent allocation.
2889 * journalled mode is currently not supported
2892 BUG_ON(ext4_should_journal_data(inode
));
2893 needed_blocks
= ext4_da_writepages_trans_blocks(inode
);
2895 /* start a new transaction*/
2896 handle
= ext4_journal_start(inode
, needed_blocks
);
2897 if (IS_ERR(handle
)) {
2898 ret
= PTR_ERR(handle
);
2899 ext4_msg(inode
->i_sb
, KERN_CRIT
, "%s: jbd2_start: "
2900 "%ld pages, ino %lu; err %d\n", __func__
,
2901 wbc
->nr_to_write
, inode
->i_ino
, ret
);
2902 goto out_writepages
;
2906 * Now call __mpage_da_writepage to find the next
2907 * contiguous region of logical blocks that need
2908 * blocks to be allocated by ext4. We don't actually
2909 * submit the blocks for I/O here, even though
2910 * write_cache_pages thinks it will, and will set the
2911 * pages as clean for write before calling
2912 * __mpage_da_writepage().
2920 mpd
.pages_written
= 0;
2922 ret
= write_cache_pages(mapping
, wbc
, __mpage_da_writepage
,
2925 * If we have a contiguous extent of pages and we
2926 * haven't done the I/O yet, map the blocks and submit
2929 if (!mpd
.io_done
&& mpd
.next_page
!= mpd
.first_page
) {
2930 if (mpage_da_map_blocks(&mpd
) == 0)
2931 mpage_da_submit_io(&mpd
);
2933 ret
= MPAGE_DA_EXTENT_TAIL
;
2935 trace_ext4_da_write_pages(inode
, &mpd
);
2936 wbc
->nr_to_write
-= mpd
.pages_written
;
2938 ext4_journal_stop(handle
);
2940 if ((mpd
.retval
== -ENOSPC
) && sbi
->s_journal
) {
2941 /* commit the transaction which would
2942 * free blocks released in the transaction
2945 jbd2_journal_force_commit_nested(sbi
->s_journal
);
2946 wbc
->pages_skipped
= pages_skipped
;
2948 } else if (ret
== MPAGE_DA_EXTENT_TAIL
) {
2950 * got one extent now try with
2953 pages_written
+= mpd
.pages_written
;
2954 wbc
->pages_skipped
= pages_skipped
;
2957 } else if (wbc
->nr_to_write
)
2959 * There is no more writeout needed
2960 * or we requested for a noblocking writeout
2961 * and we found the device congested
2965 if (!io_done
&& !cycled
) {
2968 wbc
->range_start
= index
<< PAGE_CACHE_SHIFT
;
2969 wbc
->range_end
= mapping
->writeback_index
- 1;
2972 if (pages_skipped
!= wbc
->pages_skipped
)
2973 ext4_msg(inode
->i_sb
, KERN_CRIT
,
2974 "This should not happen leaving %s "
2975 "with nr_to_write = %ld ret = %d\n",
2976 __func__
, wbc
->nr_to_write
, ret
);
2979 index
+= pages_written
;
2980 wbc
->range_cyclic
= range_cyclic
;
2981 if (wbc
->range_cyclic
|| (range_whole
&& wbc
->nr_to_write
> 0))
2983 * set the writeback_index so that range_cyclic
2984 * mode will write it back later
2986 mapping
->writeback_index
= index
;
2989 if (!no_nrwrite_index_update
)
2990 wbc
->no_nrwrite_index_update
= 0;
2991 wbc
->nr_to_write
-= nr_to_writebump
;
2992 wbc
->range_start
= range_start
;
2993 trace_ext4_da_writepages_result(inode
, wbc
, ret
, pages_written
);
2997 #define FALL_BACK_TO_NONDELALLOC 1
2998 static int ext4_nonda_switch(struct super_block
*sb
)
3000 s64 free_blocks
, dirty_blocks
;
3001 struct ext4_sb_info
*sbi
= EXT4_SB(sb
);
3004 * switch to non delalloc mode if we are running low
3005 * on free block. The free block accounting via percpu
3006 * counters can get slightly wrong with percpu_counter_batch getting
3007 * accumulated on each CPU without updating global counters
3008 * Delalloc need an accurate free block accounting. So switch
3009 * to non delalloc when we are near to error range.
3011 free_blocks
= percpu_counter_read_positive(&sbi
->s_freeblocks_counter
);
3012 dirty_blocks
= percpu_counter_read_positive(&sbi
->s_dirtyblocks_counter
);
3013 if (2 * free_blocks
< 3 * dirty_blocks
||
3014 free_blocks
< (dirty_blocks
+ EXT4_FREEBLOCKS_WATERMARK
)) {
3016 * free block count is less than 150% of dirty blocks
3017 * or free blocks is less than watermark
3022 * Even if we don't switch but are nearing capacity,
3023 * start pushing delalloc when 1/2 of free blocks are dirty.
3025 if (free_blocks
< 2 * dirty_blocks
)
3026 writeback_inodes_sb_if_idle(sb
);
3031 static int ext4_da_write_begin(struct file
*file
, struct address_space
*mapping
,
3032 loff_t pos
, unsigned len
, unsigned flags
,
3033 struct page
**pagep
, void **fsdata
)
3035 int ret
, retries
= 0;
3039 struct inode
*inode
= mapping
->host
;
3042 index
= pos
>> PAGE_CACHE_SHIFT
;
3043 from
= pos
& (PAGE_CACHE_SIZE
- 1);
3046 if (ext4_nonda_switch(inode
->i_sb
)) {
3047 *fsdata
= (void *)FALL_BACK_TO_NONDELALLOC
;
3048 return ext4_write_begin(file
, mapping
, pos
,
3049 len
, flags
, pagep
, fsdata
);
3051 *fsdata
= (void *)0;
3052 trace_ext4_da_write_begin(inode
, pos
, len
, flags
);
3055 * With delayed allocation, we don't log the i_disksize update
3056 * if there is delayed block allocation. But we still need
3057 * to journalling the i_disksize update if writes to the end
3058 * of file which has an already mapped buffer.
3060 handle
= ext4_journal_start(inode
, 1);
3061 if (IS_ERR(handle
)) {
3062 ret
= PTR_ERR(handle
);
3065 /* We cannot recurse into the filesystem as the transaction is already
3067 flags
|= AOP_FLAG_NOFS
;
3069 page
= grab_cache_page_write_begin(mapping
, index
, flags
);
3071 ext4_journal_stop(handle
);
3077 ret
= block_write_begin(file
, mapping
, pos
, len
, flags
, pagep
, fsdata
,
3078 ext4_da_get_block_prep
);
3081 ext4_journal_stop(handle
);
3082 page_cache_release(page
);
3084 * block_write_begin may have instantiated a few blocks
3085 * outside i_size. Trim these off again. Don't need
3086 * i_size_read because we hold i_mutex.
3088 if (pos
+ len
> inode
->i_size
)
3089 ext4_truncate_failed_write(inode
);
3092 if (ret
== -ENOSPC
&& ext4_should_retry_alloc(inode
->i_sb
, &retries
))
3099 * Check if we should update i_disksize
3100 * when write to the end of file but not require block allocation
3102 static int ext4_da_should_update_i_disksize(struct page
*page
,
3103 unsigned long offset
)
3105 struct buffer_head
*bh
;
3106 struct inode
*inode
= page
->mapping
->host
;
3110 bh
= page_buffers(page
);
3111 idx
= offset
>> inode
->i_blkbits
;
3113 for (i
= 0; i
< idx
; i
++)
3114 bh
= bh
->b_this_page
;
3116 if (!buffer_mapped(bh
) || (buffer_delay(bh
)) || buffer_unwritten(bh
))
3121 static int ext4_da_write_end(struct file
*file
,
3122 struct address_space
*mapping
,
3123 loff_t pos
, unsigned len
, unsigned copied
,
3124 struct page
*page
, void *fsdata
)
3126 struct inode
*inode
= mapping
->host
;
3128 handle_t
*handle
= ext4_journal_current_handle();
3130 unsigned long start
, end
;
3131 int write_mode
= (int)(unsigned long)fsdata
;
3133 if (write_mode
== FALL_BACK_TO_NONDELALLOC
) {
3134 if (ext4_should_order_data(inode
)) {
3135 return ext4_ordered_write_end(file
, mapping
, pos
,
3136 len
, copied
, page
, fsdata
);
3137 } else if (ext4_should_writeback_data(inode
)) {
3138 return ext4_writeback_write_end(file
, mapping
, pos
,
3139 len
, copied
, page
, fsdata
);
3145 trace_ext4_da_write_end(inode
, pos
, len
, copied
);
3146 start
= pos
& (PAGE_CACHE_SIZE
- 1);
3147 end
= start
+ copied
- 1;
3150 * generic_write_end() will run mark_inode_dirty() if i_size
3151 * changes. So let's piggyback the i_disksize mark_inode_dirty
3155 new_i_size
= pos
+ copied
;
3156 if (new_i_size
> EXT4_I(inode
)->i_disksize
) {
3157 if (ext4_da_should_update_i_disksize(page
, end
)) {
3158 down_write(&EXT4_I(inode
)->i_data_sem
);
3159 if (new_i_size
> EXT4_I(inode
)->i_disksize
) {
3161 * Updating i_disksize when extending file
3162 * without needing block allocation
3164 if (ext4_should_order_data(inode
))
3165 ret
= ext4_jbd2_file_inode(handle
,
3168 EXT4_I(inode
)->i_disksize
= new_i_size
;
3170 up_write(&EXT4_I(inode
)->i_data_sem
);
3171 /* We need to mark inode dirty even if
3172 * new_i_size is less that inode->i_size
3173 * bu greater than i_disksize.(hint delalloc)
3175 ext4_mark_inode_dirty(handle
, inode
);
3178 ret2
= generic_write_end(file
, mapping
, pos
, len
, copied
,
3183 ret2
= ext4_journal_stop(handle
);
3187 return ret
? ret
: copied
;
3190 static void ext4_da_invalidatepage(struct page
*page
, unsigned long offset
)
3193 * Drop reserved blocks
3195 BUG_ON(!PageLocked(page
));
3196 if (!page_has_buffers(page
))
3199 ext4_da_page_release_reservation(page
, offset
);
3202 ext4_invalidatepage(page
, offset
);
3208 * Force all delayed allocation blocks to be allocated for a given inode.
3210 int ext4_alloc_da_blocks(struct inode
*inode
)
3212 trace_ext4_alloc_da_blocks(inode
);
3214 if (!EXT4_I(inode
)->i_reserved_data_blocks
&&
3215 !EXT4_I(inode
)->i_reserved_meta_blocks
)
3219 * We do something simple for now. The filemap_flush() will
3220 * also start triggering a write of the data blocks, which is
3221 * not strictly speaking necessary (and for users of
3222 * laptop_mode, not even desirable). However, to do otherwise
3223 * would require replicating code paths in:
3225 * ext4_da_writepages() ->
3226 * write_cache_pages() ---> (via passed in callback function)
3227 * __mpage_da_writepage() -->
3228 * mpage_add_bh_to_extent()
3229 * mpage_da_map_blocks()
3231 * The problem is that write_cache_pages(), located in
3232 * mm/page-writeback.c, marks pages clean in preparation for
3233 * doing I/O, which is not desirable if we're not planning on
3236 * We could call write_cache_pages(), and then redirty all of
3237 * the pages by calling redirty_page_for_writeback() but that
3238 * would be ugly in the extreme. So instead we would need to
3239 * replicate parts of the code in the above functions,
3240 * simplifying them becuase we wouldn't actually intend to
3241 * write out the pages, but rather only collect contiguous
3242 * logical block extents, call the multi-block allocator, and
3243 * then update the buffer heads with the block allocations.
3245 * For now, though, we'll cheat by calling filemap_flush(),
3246 * which will map the blocks, and start the I/O, but not
3247 * actually wait for the I/O to complete.
3249 return filemap_flush(inode
->i_mapping
);
3253 * bmap() is special. It gets used by applications such as lilo and by
3254 * the swapper to find the on-disk block of a specific piece of data.
3256 * Naturally, this is dangerous if the block concerned is still in the
3257 * journal. If somebody makes a swapfile on an ext4 data-journaling
3258 * filesystem and enables swap, then they may get a nasty shock when the
3259 * data getting swapped to that swapfile suddenly gets overwritten by
3260 * the original zero's written out previously to the journal and
3261 * awaiting writeback in the kernel's buffer cache.
3263 * So, if we see any bmap calls here on a modified, data-journaled file,
3264 * take extra steps to flush any blocks which might be in the cache.
3266 static sector_t
ext4_bmap(struct address_space
*mapping
, sector_t block
)
3268 struct inode
*inode
= mapping
->host
;
3272 if (mapping_tagged(mapping
, PAGECACHE_TAG_DIRTY
) &&
3273 test_opt(inode
->i_sb
, DELALLOC
)) {
3275 * With delalloc we want to sync the file
3276 * so that we can make sure we allocate
3279 filemap_write_and_wait(mapping
);
3282 if (EXT4_JOURNAL(inode
) && EXT4_I(inode
)->i_state
& EXT4_STATE_JDATA
) {
3284 * This is a REALLY heavyweight approach, but the use of
3285 * bmap on dirty files is expected to be extremely rare:
3286 * only if we run lilo or swapon on a freshly made file
3287 * do we expect this to happen.
3289 * (bmap requires CAP_SYS_RAWIO so this does not
3290 * represent an unprivileged user DOS attack --- we'd be
3291 * in trouble if mortal users could trigger this path at
3294 * NB. EXT4_STATE_JDATA is not set on files other than
3295 * regular files. If somebody wants to bmap a directory
3296 * or symlink and gets confused because the buffer
3297 * hasn't yet been flushed to disk, they deserve
3298 * everything they get.
3301 EXT4_I(inode
)->i_state
&= ~EXT4_STATE_JDATA
;
3302 journal
= EXT4_JOURNAL(inode
);
3303 jbd2_journal_lock_updates(journal
);
3304 err
= jbd2_journal_flush(journal
);
3305 jbd2_journal_unlock_updates(journal
);
3311 return generic_block_bmap(mapping
, block
, ext4_get_block
);
3314 static int ext4_readpage(struct file
*file
, struct page
*page
)
3316 return mpage_readpage(page
, ext4_get_block
);
3320 ext4_readpages(struct file
*file
, struct address_space
*mapping
,
3321 struct list_head
*pages
, unsigned nr_pages
)
3323 return mpage_readpages(mapping
, pages
, nr_pages
, ext4_get_block
);
3326 static void ext4_invalidatepage(struct page
*page
, unsigned long offset
)
3328 journal_t
*journal
= EXT4_JOURNAL(page
->mapping
->host
);
3331 * If it's a full truncate we just forget about the pending dirtying
3334 ClearPageChecked(page
);
3337 jbd2_journal_invalidatepage(journal
, page
, offset
);
3339 block_invalidatepage(page
, offset
);
3342 static int ext4_releasepage(struct page
*page
, gfp_t wait
)
3344 journal_t
*journal
= EXT4_JOURNAL(page
->mapping
->host
);
3346 WARN_ON(PageChecked(page
));
3347 if (!page_has_buffers(page
))
3350 return jbd2_journal_try_to_free_buffers(journal
, page
, wait
);
3352 return try_to_free_buffers(page
);
3356 * O_DIRECT for ext3 (or indirect map) based files
3358 * If the O_DIRECT write will extend the file then add this inode to the
3359 * orphan list. So recovery will truncate it back to the original size
3360 * if the machine crashes during the write.
3362 * If the O_DIRECT write is intantiating holes inside i_size and the machine
3363 * crashes then stale disk data _may_ be exposed inside the file. But current
3364 * VFS code falls back into buffered path in that case so we are safe.
3366 static ssize_t
ext4_ind_direct_IO(int rw
, struct kiocb
*iocb
,
3367 const struct iovec
*iov
, loff_t offset
,
3368 unsigned long nr_segs
)
3370 struct file
*file
= iocb
->ki_filp
;
3371 struct inode
*inode
= file
->f_mapping
->host
;
3372 struct ext4_inode_info
*ei
= EXT4_I(inode
);
3376 size_t count
= iov_length(iov
, nr_segs
);
3380 loff_t final_size
= offset
+ count
;
3382 if (final_size
> inode
->i_size
) {
3383 /* Credits for sb + inode write */
3384 handle
= ext4_journal_start(inode
, 2);
3385 if (IS_ERR(handle
)) {
3386 ret
= PTR_ERR(handle
);
3389 ret
= ext4_orphan_add(handle
, inode
);
3391 ext4_journal_stop(handle
);
3395 ei
->i_disksize
= inode
->i_size
;
3396 ext4_journal_stop(handle
);
3401 ret
= blockdev_direct_IO(rw
, iocb
, inode
, inode
->i_sb
->s_bdev
, iov
,
3403 ext4_get_block
, NULL
);
3404 if (ret
== -ENOSPC
&& ext4_should_retry_alloc(inode
->i_sb
, &retries
))
3410 /* Credits for sb + inode write */
3411 handle
= ext4_journal_start(inode
, 2);
3412 if (IS_ERR(handle
)) {
3413 /* This is really bad luck. We've written the data
3414 * but cannot extend i_size. Bail out and pretend
3415 * the write failed... */
3416 ret
= PTR_ERR(handle
);
3420 ext4_orphan_del(handle
, inode
);
3422 loff_t end
= offset
+ ret
;
3423 if (end
> inode
->i_size
) {
3424 ei
->i_disksize
= end
;
3425 i_size_write(inode
, end
);
3427 * We're going to return a positive `ret'
3428 * here due to non-zero-length I/O, so there's
3429 * no way of reporting error returns from
3430 * ext4_mark_inode_dirty() to userspace. So
3433 ext4_mark_inode_dirty(handle
, inode
);
3436 err
= ext4_journal_stop(handle
);
3444 static int ext4_get_block_dio_write(struct inode
*inode
, sector_t iblock
,
3445 struct buffer_head
*bh_result
, int create
)
3447 handle_t
*handle
= NULL
;
3449 unsigned max_blocks
= bh_result
->b_size
>> inode
->i_blkbits
;
3452 ext4_debug("ext4_get_block_dio_write: inode %lu, create flag %d\n",
3453 inode
->i_ino
, create
);
3455 * DIO VFS code passes create = 0 flag for write to
3456 * the middle of file. It does this to avoid block
3457 * allocation for holes, to prevent expose stale data
3458 * out when there is parallel buffered read (which does
3459 * not hold the i_mutex lock) while direct IO write has
3460 * not completed. DIO request on holes finally falls back
3461 * to buffered IO for this reason.
3463 * For ext4 extent based file, since we support fallocate,
3464 * new allocated extent as uninitialized, for holes, we
3465 * could fallocate blocks for holes, thus parallel
3466 * buffered IO read will zero out the page when read on
3467 * a hole while parallel DIO write to the hole has not completed.
3469 * when we come here, we know it's a direct IO write to
3470 * to the middle of file (<i_size)
3471 * so it's safe to override the create flag from VFS.
3473 create
= EXT4_GET_BLOCKS_DIO_CREATE_EXT
;
3475 if (max_blocks
> DIO_MAX_BLOCKS
)
3476 max_blocks
= DIO_MAX_BLOCKS
;
3477 dio_credits
= ext4_chunk_trans_blocks(inode
, max_blocks
);
3478 handle
= ext4_journal_start(inode
, dio_credits
);
3479 if (IS_ERR(handle
)) {
3480 ret
= PTR_ERR(handle
);
3483 ret
= ext4_get_blocks(handle
, inode
, iblock
, max_blocks
, bh_result
,
3486 bh_result
->b_size
= (ret
<< inode
->i_blkbits
);
3489 ext4_journal_stop(handle
);
3494 static void ext4_free_io_end(ext4_io_end_t
*io
)
3500 static void dump_aio_dio_list(struct inode
* inode
)
3503 struct list_head
*cur
, *before
, *after
;
3504 ext4_io_end_t
*io
, *io0
, *io1
;
3506 if (list_empty(&EXT4_I(inode
)->i_aio_dio_complete_list
)){
3507 ext4_debug("inode %lu aio dio list is empty\n", inode
->i_ino
);
3511 ext4_debug("Dump inode %lu aio_dio_completed_IO list \n", inode
->i_ino
);
3512 list_for_each_entry(io
, &EXT4_I(inode
)->i_aio_dio_complete_list
, list
){
3515 io0
= container_of(before
, ext4_io_end_t
, list
);
3517 io1
= container_of(after
, ext4_io_end_t
, list
);
3519 ext4_debug("io 0x%p from inode %lu,prev 0x%p,next 0x%p\n",
3520 io
, inode
->i_ino
, io0
, io1
);
3526 * check a range of space and convert unwritten extents to written.
3528 static int ext4_end_aio_dio_nolock(ext4_io_end_t
*io
)
3530 struct inode
*inode
= io
->inode
;
3531 loff_t offset
= io
->offset
;
3532 size_t size
= io
->size
;
3535 ext4_debug("end_aio_dio_onlock: io 0x%p from inode %lu,list->next 0x%p,"
3536 "list->prev 0x%p\n",
3537 io
, inode
->i_ino
, io
->list
.next
, io
->list
.prev
);
3539 if (list_empty(&io
->list
))
3542 if (io
->flag
!= DIO_AIO_UNWRITTEN
)
3545 if (offset
+ size
<= i_size_read(inode
))
3546 ret
= ext4_convert_unwritten_extents(inode
, offset
, size
);
3549 printk(KERN_EMERG
"%s: failed to convert unwritten"
3550 "extents to written extents, error is %d"
3551 " io is still on inode %lu aio dio list\n",
3552 __func__
, ret
, inode
->i_ino
);
3556 /* clear the DIO AIO unwritten flag */
3561 * work on completed aio dio IO, to convert unwritten extents to extents
3563 static void ext4_end_aio_dio_work(struct work_struct
*work
)
3565 ext4_io_end_t
*io
= container_of(work
, ext4_io_end_t
, work
);
3566 struct inode
*inode
= io
->inode
;
3569 mutex_lock(&inode
->i_mutex
);
3570 ret
= ext4_end_aio_dio_nolock(io
);
3572 if (!list_empty(&io
->list
))
3573 list_del_init(&io
->list
);
3574 ext4_free_io_end(io
);
3576 mutex_unlock(&inode
->i_mutex
);
3579 * This function is called from ext4_sync_file().
3581 * When AIO DIO IO is completed, the work to convert unwritten
3582 * extents to written is queued on workqueue but may not get immediately
3583 * scheduled. When fsync is called, we need to ensure the
3584 * conversion is complete before fsync returns.
3585 * The inode keeps track of a list of completed AIO from DIO path
3586 * that might needs to do the conversion. This function walks through
3587 * the list and convert the related unwritten extents to written.
3589 int flush_aio_dio_completed_IO(struct inode
*inode
)
3595 if (list_empty(&EXT4_I(inode
)->i_aio_dio_complete_list
))
3598 dump_aio_dio_list(inode
);
3599 while (!list_empty(&EXT4_I(inode
)->i_aio_dio_complete_list
)){
3600 io
= list_entry(EXT4_I(inode
)->i_aio_dio_complete_list
.next
,
3601 ext4_io_end_t
, list
);
3603 * Calling ext4_end_aio_dio_nolock() to convert completed
3606 * When ext4_sync_file() is called, run_queue() may already
3607 * about to flush the work corresponding to this io structure.
3608 * It will be upset if it founds the io structure related
3609 * to the work-to-be schedule is freed.
3611 * Thus we need to keep the io structure still valid here after
3612 * convertion finished. The io structure has a flag to
3613 * avoid double converting from both fsync and background work
3616 ret
= ext4_end_aio_dio_nolock(io
);
3620 list_del_init(&io
->list
);
3622 return (ret2
< 0) ? ret2
: 0;
3625 static ext4_io_end_t
*ext4_init_io_end (struct inode
*inode
)
3627 ext4_io_end_t
*io
= NULL
;
3629 io
= kmalloc(sizeof(*io
), GFP_NOFS
);
3638 INIT_WORK(&io
->work
, ext4_end_aio_dio_work
);
3639 INIT_LIST_HEAD(&io
->list
);
3645 static void ext4_end_io_dio(struct kiocb
*iocb
, loff_t offset
,
3646 ssize_t size
, void *private)
3648 ext4_io_end_t
*io_end
= iocb
->private;
3649 struct workqueue_struct
*wq
;
3651 /* if not async direct IO or dio with 0 bytes write, just return */
3652 if (!io_end
|| !size
)
3655 ext_debug("ext4_end_io_dio(): io_end 0x%p"
3656 "for inode %lu, iocb 0x%p, offset %llu, size %llu\n",
3657 iocb
->private, io_end
->inode
->i_ino
, iocb
, offset
,
3660 /* if not aio dio with unwritten extents, just free io and return */
3661 if (io_end
->flag
!= DIO_AIO_UNWRITTEN
){
3662 ext4_free_io_end(io_end
);
3663 iocb
->private = NULL
;
3667 io_end
->offset
= offset
;
3668 io_end
->size
= size
;
3669 wq
= EXT4_SB(io_end
->inode
->i_sb
)->dio_unwritten_wq
;
3671 /* queue the work to convert unwritten extents to written */
3672 queue_work(wq
, &io_end
->work
);
3674 /* Add the io_end to per-inode completed aio dio list*/
3675 list_add_tail(&io_end
->list
,
3676 &EXT4_I(io_end
->inode
)->i_aio_dio_complete_list
);
3677 iocb
->private = NULL
;
3680 * For ext4 extent files, ext4 will do direct-io write to holes,
3681 * preallocated extents, and those write extend the file, no need to
3682 * fall back to buffered IO.
3684 * For holes, we fallocate those blocks, mark them as unintialized
3685 * If those blocks were preallocated, we mark sure they are splited, but
3686 * still keep the range to write as unintialized.
3688 * The unwrritten extents will be converted to written when DIO is completed.
3689 * For async direct IO, since the IO may still pending when return, we
3690 * set up an end_io call back function, which will do the convertion
3691 * when async direct IO completed.
3693 * If the O_DIRECT write will extend the file then add this inode to the
3694 * orphan list. So recovery will truncate it back to the original size
3695 * if the machine crashes during the write.
3698 static ssize_t
ext4_ext_direct_IO(int rw
, struct kiocb
*iocb
,
3699 const struct iovec
*iov
, loff_t offset
,
3700 unsigned long nr_segs
)
3702 struct file
*file
= iocb
->ki_filp
;
3703 struct inode
*inode
= file
->f_mapping
->host
;
3705 size_t count
= iov_length(iov
, nr_segs
);
3707 loff_t final_size
= offset
+ count
;
3708 if (rw
== WRITE
&& final_size
<= inode
->i_size
) {
3710 * We could direct write to holes and fallocate.
3712 * Allocated blocks to fill the hole are marked as uninitialized
3713 * to prevent paralel buffered read to expose the stale data
3714 * before DIO complete the data IO.
3716 * As to previously fallocated extents, ext4 get_block
3717 * will just simply mark the buffer mapped but still
3718 * keep the extents uninitialized.
3720 * for non AIO case, we will convert those unwritten extents
3721 * to written after return back from blockdev_direct_IO.
3723 * for async DIO, the conversion needs to be defered when
3724 * the IO is completed. The ext4 end_io callback function
3725 * will be called to take care of the conversion work.
3726 * Here for async case, we allocate an io_end structure to
3729 iocb
->private = NULL
;
3730 EXT4_I(inode
)->cur_aio_dio
= NULL
;
3731 if (!is_sync_kiocb(iocb
)) {
3732 iocb
->private = ext4_init_io_end(inode
);
3736 * we save the io structure for current async
3737 * direct IO, so that later ext4_get_blocks()
3738 * could flag the io structure whether there
3739 * is a unwritten extents needs to be converted
3740 * when IO is completed.
3742 EXT4_I(inode
)->cur_aio_dio
= iocb
->private;
3745 ret
= blockdev_direct_IO(rw
, iocb
, inode
,
3746 inode
->i_sb
->s_bdev
, iov
,
3748 ext4_get_block_dio_write
,
3751 EXT4_I(inode
)->cur_aio_dio
= NULL
;
3753 * The io_end structure takes a reference to the inode,
3754 * that structure needs to be destroyed and the
3755 * reference to the inode need to be dropped, when IO is
3756 * complete, even with 0 byte write, or failed.
3758 * In the successful AIO DIO case, the io_end structure will be
3759 * desctroyed and the reference to the inode will be dropped
3760 * after the end_io call back function is called.
3762 * In the case there is 0 byte write, or error case, since
3763 * VFS direct IO won't invoke the end_io call back function,
3764 * we need to free the end_io structure here.
3766 if (ret
!= -EIOCBQUEUED
&& ret
<= 0 && iocb
->private) {
3767 ext4_free_io_end(iocb
->private);
3768 iocb
->private = NULL
;
3769 } else if (ret
> 0 && (EXT4_I(inode
)->i_state
&
3770 EXT4_STATE_DIO_UNWRITTEN
)) {
3773 * for non AIO case, since the IO is already
3774 * completed, we could do the convertion right here
3776 err
= ext4_convert_unwritten_extents(inode
,
3780 EXT4_I(inode
)->i_state
&= ~EXT4_STATE_DIO_UNWRITTEN
;
3785 /* for write the the end of file case, we fall back to old way */
3786 return ext4_ind_direct_IO(rw
, iocb
, iov
, offset
, nr_segs
);
3789 static ssize_t
ext4_direct_IO(int rw
, struct kiocb
*iocb
,
3790 const struct iovec
*iov
, loff_t offset
,
3791 unsigned long nr_segs
)
3793 struct file
*file
= iocb
->ki_filp
;
3794 struct inode
*inode
= file
->f_mapping
->host
;
3796 if (EXT4_I(inode
)->i_flags
& EXT4_EXTENTS_FL
)
3797 return ext4_ext_direct_IO(rw
, iocb
, iov
, offset
, nr_segs
);
3799 return ext4_ind_direct_IO(rw
, iocb
, iov
, offset
, nr_segs
);
3803 * Pages can be marked dirty completely asynchronously from ext4's journalling
3804 * activity. By filemap_sync_pte(), try_to_unmap_one(), etc. We cannot do
3805 * much here because ->set_page_dirty is called under VFS locks. The page is
3806 * not necessarily locked.
3808 * We cannot just dirty the page and leave attached buffers clean, because the
3809 * buffers' dirty state is "definitive". We cannot just set the buffers dirty
3810 * or jbddirty because all the journalling code will explode.
3812 * So what we do is to mark the page "pending dirty" and next time writepage
3813 * is called, propagate that into the buffers appropriately.
3815 static int ext4_journalled_set_page_dirty(struct page
*page
)
3817 SetPageChecked(page
);
3818 return __set_page_dirty_nobuffers(page
);
3821 static const struct address_space_operations ext4_ordered_aops
= {
3822 .readpage
= ext4_readpage
,
3823 .readpages
= ext4_readpages
,
3824 .writepage
= ext4_writepage
,
3825 .sync_page
= block_sync_page
,
3826 .write_begin
= ext4_write_begin
,
3827 .write_end
= ext4_ordered_write_end
,
3829 .invalidatepage
= ext4_invalidatepage
,
3830 .releasepage
= ext4_releasepage
,
3831 .direct_IO
= ext4_direct_IO
,
3832 .migratepage
= buffer_migrate_page
,
3833 .is_partially_uptodate
= block_is_partially_uptodate
,
3834 .error_remove_page
= generic_error_remove_page
,
3837 static const struct address_space_operations ext4_writeback_aops
= {
3838 .readpage
= ext4_readpage
,
3839 .readpages
= ext4_readpages
,
3840 .writepage
= ext4_writepage
,
3841 .sync_page
= block_sync_page
,
3842 .write_begin
= ext4_write_begin
,
3843 .write_end
= ext4_writeback_write_end
,
3845 .invalidatepage
= ext4_invalidatepage
,
3846 .releasepage
= ext4_releasepage
,
3847 .direct_IO
= ext4_direct_IO
,
3848 .migratepage
= buffer_migrate_page
,
3849 .is_partially_uptodate
= block_is_partially_uptodate
,
3850 .error_remove_page
= generic_error_remove_page
,
3853 static const struct address_space_operations ext4_journalled_aops
= {
3854 .readpage
= ext4_readpage
,
3855 .readpages
= ext4_readpages
,
3856 .writepage
= ext4_writepage
,
3857 .sync_page
= block_sync_page
,
3858 .write_begin
= ext4_write_begin
,
3859 .write_end
= ext4_journalled_write_end
,
3860 .set_page_dirty
= ext4_journalled_set_page_dirty
,
3862 .invalidatepage
= ext4_invalidatepage
,
3863 .releasepage
= ext4_releasepage
,
3864 .is_partially_uptodate
= block_is_partially_uptodate
,
3865 .error_remove_page
= generic_error_remove_page
,
3868 static const struct address_space_operations ext4_da_aops
= {
3869 .readpage
= ext4_readpage
,
3870 .readpages
= ext4_readpages
,
3871 .writepage
= ext4_writepage
,
3872 .writepages
= ext4_da_writepages
,
3873 .sync_page
= block_sync_page
,
3874 .write_begin
= ext4_da_write_begin
,
3875 .write_end
= ext4_da_write_end
,
3877 .invalidatepage
= ext4_da_invalidatepage
,
3878 .releasepage
= ext4_releasepage
,
3879 .direct_IO
= ext4_direct_IO
,
3880 .migratepage
= buffer_migrate_page
,
3881 .is_partially_uptodate
= block_is_partially_uptodate
,
3882 .error_remove_page
= generic_error_remove_page
,
3885 void ext4_set_aops(struct inode
*inode
)
3887 if (ext4_should_order_data(inode
) &&
3888 test_opt(inode
->i_sb
, DELALLOC
))
3889 inode
->i_mapping
->a_ops
= &ext4_da_aops
;
3890 else if (ext4_should_order_data(inode
))
3891 inode
->i_mapping
->a_ops
= &ext4_ordered_aops
;
3892 else if (ext4_should_writeback_data(inode
) &&
3893 test_opt(inode
->i_sb
, DELALLOC
))
3894 inode
->i_mapping
->a_ops
= &ext4_da_aops
;
3895 else if (ext4_should_writeback_data(inode
))
3896 inode
->i_mapping
->a_ops
= &ext4_writeback_aops
;
3898 inode
->i_mapping
->a_ops
= &ext4_journalled_aops
;
3902 * ext4_block_truncate_page() zeroes out a mapping from file offset `from'
3903 * up to the end of the block which corresponds to `from'.
3904 * This required during truncate. We need to physically zero the tail end
3905 * of that block so it doesn't yield old data if the file is later grown.
3907 int ext4_block_truncate_page(handle_t
*handle
,
3908 struct address_space
*mapping
, loff_t from
)
3910 ext4_fsblk_t index
= from
>> PAGE_CACHE_SHIFT
;
3911 unsigned offset
= from
& (PAGE_CACHE_SIZE
-1);
3912 unsigned blocksize
, length
, pos
;
3914 struct inode
*inode
= mapping
->host
;
3915 struct buffer_head
*bh
;
3919 page
= find_or_create_page(mapping
, from
>> PAGE_CACHE_SHIFT
,
3920 mapping_gfp_mask(mapping
) & ~__GFP_FS
);
3924 blocksize
= inode
->i_sb
->s_blocksize
;
3925 length
= blocksize
- (offset
& (blocksize
- 1));
3926 iblock
= index
<< (PAGE_CACHE_SHIFT
- inode
->i_sb
->s_blocksize_bits
);
3929 * For "nobh" option, we can only work if we don't need to
3930 * read-in the page - otherwise we create buffers to do the IO.
3932 if (!page_has_buffers(page
) && test_opt(inode
->i_sb
, NOBH
) &&
3933 ext4_should_writeback_data(inode
) && PageUptodate(page
)) {
3934 zero_user(page
, offset
, length
);
3935 set_page_dirty(page
);
3939 if (!page_has_buffers(page
))
3940 create_empty_buffers(page
, blocksize
, 0);
3942 /* Find the buffer that contains "offset" */
3943 bh
= page_buffers(page
);
3945 while (offset
>= pos
) {
3946 bh
= bh
->b_this_page
;
3952 if (buffer_freed(bh
)) {
3953 BUFFER_TRACE(bh
, "freed: skip");
3957 if (!buffer_mapped(bh
)) {
3958 BUFFER_TRACE(bh
, "unmapped");
3959 ext4_get_block(inode
, iblock
, bh
, 0);
3960 /* unmapped? It's a hole - nothing to do */
3961 if (!buffer_mapped(bh
)) {
3962 BUFFER_TRACE(bh
, "still unmapped");
3967 /* Ok, it's mapped. Make sure it's up-to-date */
3968 if (PageUptodate(page
))
3969 set_buffer_uptodate(bh
);
3971 if (!buffer_uptodate(bh
)) {
3973 ll_rw_block(READ
, 1, &bh
);
3975 /* Uhhuh. Read error. Complain and punt. */
3976 if (!buffer_uptodate(bh
))
3980 if (ext4_should_journal_data(inode
)) {
3981 BUFFER_TRACE(bh
, "get write access");
3982 err
= ext4_journal_get_write_access(handle
, bh
);
3987 zero_user(page
, offset
, length
);
3989 BUFFER_TRACE(bh
, "zeroed end of block");
3992 if (ext4_should_journal_data(inode
)) {
3993 err
= ext4_handle_dirty_metadata(handle
, inode
, bh
);
3995 if (ext4_should_order_data(inode
))
3996 err
= ext4_jbd2_file_inode(handle
, inode
);
3997 mark_buffer_dirty(bh
);
4002 page_cache_release(page
);
4007 * Probably it should be a library function... search for first non-zero word
4008 * or memcmp with zero_page, whatever is better for particular architecture.
4011 static inline int all_zeroes(__le32
*p
, __le32
*q
)
4020 * ext4_find_shared - find the indirect blocks for partial truncation.
4021 * @inode: inode in question
4022 * @depth: depth of the affected branch
4023 * @offsets: offsets of pointers in that branch (see ext4_block_to_path)
4024 * @chain: place to store the pointers to partial indirect blocks
4025 * @top: place to the (detached) top of branch
4027 * This is a helper function used by ext4_truncate().
4029 * When we do truncate() we may have to clean the ends of several
4030 * indirect blocks but leave the blocks themselves alive. Block is
4031 * partially truncated if some data below the new i_size is refered
4032 * from it (and it is on the path to the first completely truncated
4033 * data block, indeed). We have to free the top of that path along
4034 * with everything to the right of the path. Since no allocation
4035 * past the truncation point is possible until ext4_truncate()
4036 * finishes, we may safely do the latter, but top of branch may
4037 * require special attention - pageout below the truncation point
4038 * might try to populate it.
4040 * We atomically detach the top of branch from the tree, store the
4041 * block number of its root in *@top, pointers to buffer_heads of
4042 * partially truncated blocks - in @chain[].bh and pointers to
4043 * their last elements that should not be removed - in
4044 * @chain[].p. Return value is the pointer to last filled element
4047 * The work left to caller to do the actual freeing of subtrees:
4048 * a) free the subtree starting from *@top
4049 * b) free the subtrees whose roots are stored in
4050 * (@chain[i].p+1 .. end of @chain[i].bh->b_data)
4051 * c) free the subtrees growing from the inode past the @chain[0].
4052 * (no partially truncated stuff there). */
4054 static Indirect
*ext4_find_shared(struct inode
*inode
, int depth
,
4055 ext4_lblk_t offsets
[4], Indirect chain
[4],
4058 Indirect
*partial
, *p
;
4062 /* Make k index the deepest non-null offset + 1 */
4063 for (k
= depth
; k
> 1 && !offsets
[k
-1]; k
--)
4065 partial
= ext4_get_branch(inode
, k
, offsets
, chain
, &err
);
4066 /* Writer: pointers */
4068 partial
= chain
+ k
-1;
4070 * If the branch acquired continuation since we've looked at it -
4071 * fine, it should all survive and (new) top doesn't belong to us.
4073 if (!partial
->key
&& *partial
->p
)
4076 for (p
= partial
; (p
> chain
) && all_zeroes((__le32
*) p
->bh
->b_data
, p
->p
); p
--)
4079 * OK, we've found the last block that must survive. The rest of our
4080 * branch should be detached before unlocking. However, if that rest
4081 * of branch is all ours and does not grow immediately from the inode
4082 * it's easier to cheat and just decrement partial->p.
4084 if (p
== chain
+ k
- 1 && p
> chain
) {
4088 /* Nope, don't do this in ext4. Must leave the tree intact */
4095 while (partial
> p
) {
4096 brelse(partial
->bh
);
4104 * Zero a number of block pointers in either an inode or an indirect block.
4105 * If we restart the transaction we must again get write access to the
4106 * indirect block for further modification.
4108 * We release `count' blocks on disk, but (last - first) may be greater
4109 * than `count' because there can be holes in there.
4111 static void ext4_clear_blocks(handle_t
*handle
, struct inode
*inode
,
4112 struct buffer_head
*bh
,
4113 ext4_fsblk_t block_to_free
,
4114 unsigned long count
, __le32
*first
,
4118 int flags
= EXT4_FREE_BLOCKS_FORGET
;
4120 if (S_ISDIR(inode
->i_mode
) || S_ISLNK(inode
->i_mode
))
4121 flags
|= EXT4_FREE_BLOCKS_METADATA
;
4123 if (try_to_extend_transaction(handle
, inode
)) {
4125 BUFFER_TRACE(bh
, "call ext4_handle_dirty_metadata");
4126 ext4_handle_dirty_metadata(handle
, inode
, bh
);
4128 ext4_mark_inode_dirty(handle
, inode
);
4129 ext4_truncate_restart_trans(handle
, inode
,
4130 blocks_for_truncate(inode
));
4132 BUFFER_TRACE(bh
, "retaking write access");
4133 ext4_journal_get_write_access(handle
, bh
);
4137 for (p
= first
; p
< last
; p
++)
4140 ext4_free_blocks(handle
, inode
, 0, block_to_free
, count
, flags
);
4144 * ext4_free_data - free a list of data blocks
4145 * @handle: handle for this transaction
4146 * @inode: inode we are dealing with
4147 * @this_bh: indirect buffer_head which contains *@first and *@last
4148 * @first: array of block numbers
4149 * @last: points immediately past the end of array
4151 * We are freeing all blocks refered from that array (numbers are stored as
4152 * little-endian 32-bit) and updating @inode->i_blocks appropriately.
4154 * We accumulate contiguous runs of blocks to free. Conveniently, if these
4155 * blocks are contiguous then releasing them at one time will only affect one
4156 * or two bitmap blocks (+ group descriptor(s) and superblock) and we won't
4157 * actually use a lot of journal space.
4159 * @this_bh will be %NULL if @first and @last point into the inode's direct
4162 static void ext4_free_data(handle_t
*handle
, struct inode
*inode
,
4163 struct buffer_head
*this_bh
,
4164 __le32
*first
, __le32
*last
)
4166 ext4_fsblk_t block_to_free
= 0; /* Starting block # of a run */
4167 unsigned long count
= 0; /* Number of blocks in the run */
4168 __le32
*block_to_free_p
= NULL
; /* Pointer into inode/ind
4171 ext4_fsblk_t nr
; /* Current block # */
4172 __le32
*p
; /* Pointer into inode/ind
4173 for current block */
4176 if (this_bh
) { /* For indirect block */
4177 BUFFER_TRACE(this_bh
, "get_write_access");
4178 err
= ext4_journal_get_write_access(handle
, this_bh
);
4179 /* Important: if we can't update the indirect pointers
4180 * to the blocks, we can't free them. */
4185 for (p
= first
; p
< last
; p
++) {
4186 nr
= le32_to_cpu(*p
);
4188 /* accumulate blocks to free if they're contiguous */
4191 block_to_free_p
= p
;
4193 } else if (nr
== block_to_free
+ count
) {
4196 ext4_clear_blocks(handle
, inode
, this_bh
,
4198 count
, block_to_free_p
, p
);
4200 block_to_free_p
= p
;
4207 ext4_clear_blocks(handle
, inode
, this_bh
, block_to_free
,
4208 count
, block_to_free_p
, p
);
4211 BUFFER_TRACE(this_bh
, "call ext4_handle_dirty_metadata");
4214 * The buffer head should have an attached journal head at this
4215 * point. However, if the data is corrupted and an indirect
4216 * block pointed to itself, it would have been detached when
4217 * the block was cleared. Check for this instead of OOPSing.
4219 if ((EXT4_JOURNAL(inode
) == NULL
) || bh2jh(this_bh
))
4220 ext4_handle_dirty_metadata(handle
, inode
, this_bh
);
4222 ext4_error(inode
->i_sb
, __func__
,
4223 "circular indirect block detected, "
4224 "inode=%lu, block=%llu",
4226 (unsigned long long) this_bh
->b_blocknr
);
4231 * ext4_free_branches - free an array of branches
4232 * @handle: JBD handle for this transaction
4233 * @inode: inode we are dealing with
4234 * @parent_bh: the buffer_head which contains *@first and *@last
4235 * @first: array of block numbers
4236 * @last: pointer immediately past the end of array
4237 * @depth: depth of the branches to free
4239 * We are freeing all blocks refered from these branches (numbers are
4240 * stored as little-endian 32-bit) and updating @inode->i_blocks
4243 static void ext4_free_branches(handle_t
*handle
, struct inode
*inode
,
4244 struct buffer_head
*parent_bh
,
4245 __le32
*first
, __le32
*last
, int depth
)
4250 if (ext4_handle_is_aborted(handle
))
4254 struct buffer_head
*bh
;
4255 int addr_per_block
= EXT4_ADDR_PER_BLOCK(inode
->i_sb
);
4257 while (--p
>= first
) {
4258 nr
= le32_to_cpu(*p
);
4260 continue; /* A hole */
4262 /* Go read the buffer for the next level down */
4263 bh
= sb_bread(inode
->i_sb
, nr
);
4266 * A read failure? Report error and clear slot
4270 ext4_error(inode
->i_sb
, "ext4_free_branches",
4271 "Read failure, inode=%lu, block=%llu",
4276 /* This zaps the entire block. Bottom up. */
4277 BUFFER_TRACE(bh
, "free child branches");
4278 ext4_free_branches(handle
, inode
, bh
,
4279 (__le32
*) bh
->b_data
,
4280 (__le32
*) bh
->b_data
+ addr_per_block
,
4284 * We've probably journalled the indirect block several
4285 * times during the truncate. But it's no longer
4286 * needed and we now drop it from the transaction via
4287 * jbd2_journal_revoke().
4289 * That's easy if it's exclusively part of this
4290 * transaction. But if it's part of the committing
4291 * transaction then jbd2_journal_forget() will simply
4292 * brelse() it. That means that if the underlying
4293 * block is reallocated in ext4_get_block(),
4294 * unmap_underlying_metadata() will find this block
4295 * and will try to get rid of it. damn, damn.
4297 * If this block has already been committed to the
4298 * journal, a revoke record will be written. And
4299 * revoke records must be emitted *before* clearing
4300 * this block's bit in the bitmaps.
4302 ext4_forget(handle
, 1, inode
, bh
, bh
->b_blocknr
);
4305 * Everything below this this pointer has been
4306 * released. Now let this top-of-subtree go.
4308 * We want the freeing of this indirect block to be
4309 * atomic in the journal with the updating of the
4310 * bitmap block which owns it. So make some room in
4313 * We zero the parent pointer *after* freeing its
4314 * pointee in the bitmaps, so if extend_transaction()
4315 * for some reason fails to put the bitmap changes and
4316 * the release into the same transaction, recovery
4317 * will merely complain about releasing a free block,
4318 * rather than leaking blocks.
4320 if (ext4_handle_is_aborted(handle
))
4322 if (try_to_extend_transaction(handle
, inode
)) {
4323 ext4_mark_inode_dirty(handle
, inode
);
4324 ext4_truncate_restart_trans(handle
, inode
,
4325 blocks_for_truncate(inode
));
4328 ext4_free_blocks(handle
, inode
, 0, nr
, 1,
4329 EXT4_FREE_BLOCKS_METADATA
);
4333 * The block which we have just freed is
4334 * pointed to by an indirect block: journal it
4336 BUFFER_TRACE(parent_bh
, "get_write_access");
4337 if (!ext4_journal_get_write_access(handle
,
4340 BUFFER_TRACE(parent_bh
,
4341 "call ext4_handle_dirty_metadata");
4342 ext4_handle_dirty_metadata(handle
,
4349 /* We have reached the bottom of the tree. */
4350 BUFFER_TRACE(parent_bh
, "free data blocks");
4351 ext4_free_data(handle
, inode
, parent_bh
, first
, last
);
4355 int ext4_can_truncate(struct inode
*inode
)
4357 if (IS_APPEND(inode
) || IS_IMMUTABLE(inode
))
4359 if (S_ISREG(inode
->i_mode
))
4361 if (S_ISDIR(inode
->i_mode
))
4363 if (S_ISLNK(inode
->i_mode
))
4364 return !ext4_inode_is_fast_symlink(inode
);
4371 * We block out ext4_get_block() block instantiations across the entire
4372 * transaction, and VFS/VM ensures that ext4_truncate() cannot run
4373 * simultaneously on behalf of the same inode.
4375 * As we work through the truncate and commmit bits of it to the journal there
4376 * is one core, guiding principle: the file's tree must always be consistent on
4377 * disk. We must be able to restart the truncate after a crash.
4379 * The file's tree may be transiently inconsistent in memory (although it
4380 * probably isn't), but whenever we close off and commit a journal transaction,
4381 * the contents of (the filesystem + the journal) must be consistent and
4382 * restartable. It's pretty simple, really: bottom up, right to left (although
4383 * left-to-right works OK too).
4385 * Note that at recovery time, journal replay occurs *before* the restart of
4386 * truncate against the orphan inode list.
4388 * The committed inode has the new, desired i_size (which is the same as
4389 * i_disksize in this case). After a crash, ext4_orphan_cleanup() will see
4390 * that this inode's truncate did not complete and it will again call
4391 * ext4_truncate() to have another go. So there will be instantiated blocks
4392 * to the right of the truncation point in a crashed ext4 filesystem. But
4393 * that's fine - as long as they are linked from the inode, the post-crash
4394 * ext4_truncate() run will find them and release them.
4396 void ext4_truncate(struct inode
*inode
)
4399 struct ext4_inode_info
*ei
= EXT4_I(inode
);
4400 __le32
*i_data
= ei
->i_data
;
4401 int addr_per_block
= EXT4_ADDR_PER_BLOCK(inode
->i_sb
);
4402 struct address_space
*mapping
= inode
->i_mapping
;
4403 ext4_lblk_t offsets
[4];
4408 ext4_lblk_t last_block
;
4409 unsigned blocksize
= inode
->i_sb
->s_blocksize
;
4411 if (!ext4_can_truncate(inode
))
4414 if (inode
->i_size
== 0 && !test_opt(inode
->i_sb
, NO_AUTO_DA_ALLOC
))
4415 ei
->i_state
|= EXT4_STATE_DA_ALLOC_CLOSE
;
4417 if (EXT4_I(inode
)->i_flags
& EXT4_EXTENTS_FL
) {
4418 ext4_ext_truncate(inode
);
4422 handle
= start_transaction(inode
);
4424 return; /* AKPM: return what? */
4426 last_block
= (inode
->i_size
+ blocksize
-1)
4427 >> EXT4_BLOCK_SIZE_BITS(inode
->i_sb
);
4429 if (inode
->i_size
& (blocksize
- 1))
4430 if (ext4_block_truncate_page(handle
, mapping
, inode
->i_size
))
4433 n
= ext4_block_to_path(inode
, last_block
, offsets
, NULL
);
4435 goto out_stop
; /* error */
4438 * OK. This truncate is going to happen. We add the inode to the
4439 * orphan list, so that if this truncate spans multiple transactions,
4440 * and we crash, we will resume the truncate when the filesystem
4441 * recovers. It also marks the inode dirty, to catch the new size.
4443 * Implication: the file must always be in a sane, consistent
4444 * truncatable state while each transaction commits.
4446 if (ext4_orphan_add(handle
, inode
))
4450 * From here we block out all ext4_get_block() callers who want to
4451 * modify the block allocation tree.
4453 down_write(&ei
->i_data_sem
);
4455 ext4_discard_preallocations(inode
);
4458 * The orphan list entry will now protect us from any crash which
4459 * occurs before the truncate completes, so it is now safe to propagate
4460 * the new, shorter inode size (held for now in i_size) into the
4461 * on-disk inode. We do this via i_disksize, which is the value which
4462 * ext4 *really* writes onto the disk inode.
4464 ei
->i_disksize
= inode
->i_size
;
4466 if (n
== 1) { /* direct blocks */
4467 ext4_free_data(handle
, inode
, NULL
, i_data
+offsets
[0],
4468 i_data
+ EXT4_NDIR_BLOCKS
);
4472 partial
= ext4_find_shared(inode
, n
, offsets
, chain
, &nr
);
4473 /* Kill the top of shared branch (not detached) */
4475 if (partial
== chain
) {
4476 /* Shared branch grows from the inode */
4477 ext4_free_branches(handle
, inode
, NULL
,
4478 &nr
, &nr
+1, (chain
+n
-1) - partial
);
4481 * We mark the inode dirty prior to restart,
4482 * and prior to stop. No need for it here.
4485 /* Shared branch grows from an indirect block */
4486 BUFFER_TRACE(partial
->bh
, "get_write_access");
4487 ext4_free_branches(handle
, inode
, partial
->bh
,
4489 partial
->p
+1, (chain
+n
-1) - partial
);
4492 /* Clear the ends of indirect blocks on the shared branch */
4493 while (partial
> chain
) {
4494 ext4_free_branches(handle
, inode
, partial
->bh
, partial
->p
+ 1,
4495 (__le32
*)partial
->bh
->b_data
+addr_per_block
,
4496 (chain
+n
-1) - partial
);
4497 BUFFER_TRACE(partial
->bh
, "call brelse");
4498 brelse(partial
->bh
);
4502 /* Kill the remaining (whole) subtrees */
4503 switch (offsets
[0]) {
4505 nr
= i_data
[EXT4_IND_BLOCK
];
4507 ext4_free_branches(handle
, inode
, NULL
, &nr
, &nr
+1, 1);
4508 i_data
[EXT4_IND_BLOCK
] = 0;
4510 case EXT4_IND_BLOCK
:
4511 nr
= i_data
[EXT4_DIND_BLOCK
];
4513 ext4_free_branches(handle
, inode
, NULL
, &nr
, &nr
+1, 2);
4514 i_data
[EXT4_DIND_BLOCK
] = 0;
4516 case EXT4_DIND_BLOCK
:
4517 nr
= i_data
[EXT4_TIND_BLOCK
];
4519 ext4_free_branches(handle
, inode
, NULL
, &nr
, &nr
+1, 3);
4520 i_data
[EXT4_TIND_BLOCK
] = 0;
4522 case EXT4_TIND_BLOCK
:
4526 up_write(&ei
->i_data_sem
);
4527 inode
->i_mtime
= inode
->i_ctime
= ext4_current_time(inode
);
4528 ext4_mark_inode_dirty(handle
, inode
);
4531 * In a multi-transaction truncate, we only make the final transaction
4535 ext4_handle_sync(handle
);
4538 * If this was a simple ftruncate(), and the file will remain alive
4539 * then we need to clear up the orphan record which we created above.
4540 * However, if this was a real unlink then we were called by
4541 * ext4_delete_inode(), and we allow that function to clean up the
4542 * orphan info for us.
4545 ext4_orphan_del(handle
, inode
);
4547 ext4_journal_stop(handle
);
4551 * ext4_get_inode_loc returns with an extra refcount against the inode's
4552 * underlying buffer_head on success. If 'in_mem' is true, we have all
4553 * data in memory that is needed to recreate the on-disk version of this
4556 static int __ext4_get_inode_loc(struct inode
*inode
,
4557 struct ext4_iloc
*iloc
, int in_mem
)
4559 struct ext4_group_desc
*gdp
;
4560 struct buffer_head
*bh
;
4561 struct super_block
*sb
= inode
->i_sb
;
4563 int inodes_per_block
, inode_offset
;
4566 if (!ext4_valid_inum(sb
, inode
->i_ino
))
4569 iloc
->block_group
= (inode
->i_ino
- 1) / EXT4_INODES_PER_GROUP(sb
);
4570 gdp
= ext4_get_group_desc(sb
, iloc
->block_group
, NULL
);
4575 * Figure out the offset within the block group inode table
4577 inodes_per_block
= (EXT4_BLOCK_SIZE(sb
) / EXT4_INODE_SIZE(sb
));
4578 inode_offset
= ((inode
->i_ino
- 1) %
4579 EXT4_INODES_PER_GROUP(sb
));
4580 block
= ext4_inode_table(sb
, gdp
) + (inode_offset
/ inodes_per_block
);
4581 iloc
->offset
= (inode_offset
% inodes_per_block
) * EXT4_INODE_SIZE(sb
);
4583 bh
= sb_getblk(sb
, block
);
4585 ext4_error(sb
, "ext4_get_inode_loc", "unable to read "
4586 "inode block - inode=%lu, block=%llu",
4587 inode
->i_ino
, block
);
4590 if (!buffer_uptodate(bh
)) {
4594 * If the buffer has the write error flag, we have failed
4595 * to write out another inode in the same block. In this
4596 * case, we don't have to read the block because we may
4597 * read the old inode data successfully.
4599 if (buffer_write_io_error(bh
) && !buffer_uptodate(bh
))
4600 set_buffer_uptodate(bh
);
4602 if (buffer_uptodate(bh
)) {
4603 /* someone brought it uptodate while we waited */
4609 * If we have all information of the inode in memory and this
4610 * is the only valid inode in the block, we need not read the
4614 struct buffer_head
*bitmap_bh
;
4617 start
= inode_offset
& ~(inodes_per_block
- 1);
4619 /* Is the inode bitmap in cache? */
4620 bitmap_bh
= sb_getblk(sb
, ext4_inode_bitmap(sb
, gdp
));
4625 * If the inode bitmap isn't in cache then the
4626 * optimisation may end up performing two reads instead
4627 * of one, so skip it.
4629 if (!buffer_uptodate(bitmap_bh
)) {
4633 for (i
= start
; i
< start
+ inodes_per_block
; i
++) {
4634 if (i
== inode_offset
)
4636 if (ext4_test_bit(i
, bitmap_bh
->b_data
))
4640 if (i
== start
+ inodes_per_block
) {
4641 /* all other inodes are free, so skip I/O */
4642 memset(bh
->b_data
, 0, bh
->b_size
);
4643 set_buffer_uptodate(bh
);
4651 * If we need to do any I/O, try to pre-readahead extra
4652 * blocks from the inode table.
4654 if (EXT4_SB(sb
)->s_inode_readahead_blks
) {
4655 ext4_fsblk_t b
, end
, table
;
4658 table
= ext4_inode_table(sb
, gdp
);
4659 /* s_inode_readahead_blks is always a power of 2 */
4660 b
= block
& ~(EXT4_SB(sb
)->s_inode_readahead_blks
-1);
4663 end
= b
+ EXT4_SB(sb
)->s_inode_readahead_blks
;
4664 num
= EXT4_INODES_PER_GROUP(sb
);
4665 if (EXT4_HAS_RO_COMPAT_FEATURE(sb
,
4666 EXT4_FEATURE_RO_COMPAT_GDT_CSUM
))
4667 num
-= ext4_itable_unused_count(sb
, gdp
);
4668 table
+= num
/ inodes_per_block
;
4672 sb_breadahead(sb
, b
++);
4676 * There are other valid inodes in the buffer, this inode
4677 * has in-inode xattrs, or we don't have this inode in memory.
4678 * Read the block from disk.
4681 bh
->b_end_io
= end_buffer_read_sync
;
4682 submit_bh(READ_META
, bh
);
4684 if (!buffer_uptodate(bh
)) {
4685 ext4_error(sb
, __func__
,
4686 "unable to read inode block - inode=%lu, "
4687 "block=%llu", inode
->i_ino
, block
);
4697 int ext4_get_inode_loc(struct inode
*inode
, struct ext4_iloc
*iloc
)
4699 /* We have all inode data except xattrs in memory here. */
4700 return __ext4_get_inode_loc(inode
, iloc
,
4701 !(EXT4_I(inode
)->i_state
& EXT4_STATE_XATTR
));
4704 void ext4_set_inode_flags(struct inode
*inode
)
4706 unsigned int flags
= EXT4_I(inode
)->i_flags
;
4708 inode
->i_flags
&= ~(S_SYNC
|S_APPEND
|S_IMMUTABLE
|S_NOATIME
|S_DIRSYNC
);
4709 if (flags
& EXT4_SYNC_FL
)
4710 inode
->i_flags
|= S_SYNC
;
4711 if (flags
& EXT4_APPEND_FL
)
4712 inode
->i_flags
|= S_APPEND
;
4713 if (flags
& EXT4_IMMUTABLE_FL
)
4714 inode
->i_flags
|= S_IMMUTABLE
;
4715 if (flags
& EXT4_NOATIME_FL
)
4716 inode
->i_flags
|= S_NOATIME
;
4717 if (flags
& EXT4_DIRSYNC_FL
)
4718 inode
->i_flags
|= S_DIRSYNC
;
4721 /* Propagate flags from i_flags to EXT4_I(inode)->i_flags */
4722 void ext4_get_inode_flags(struct ext4_inode_info
*ei
)
4724 unsigned int flags
= ei
->vfs_inode
.i_flags
;
4726 ei
->i_flags
&= ~(EXT4_SYNC_FL
|EXT4_APPEND_FL
|
4727 EXT4_IMMUTABLE_FL
|EXT4_NOATIME_FL
|EXT4_DIRSYNC_FL
);
4729 ei
->i_flags
|= EXT4_SYNC_FL
;
4730 if (flags
& S_APPEND
)
4731 ei
->i_flags
|= EXT4_APPEND_FL
;
4732 if (flags
& S_IMMUTABLE
)
4733 ei
->i_flags
|= EXT4_IMMUTABLE_FL
;
4734 if (flags
& S_NOATIME
)
4735 ei
->i_flags
|= EXT4_NOATIME_FL
;
4736 if (flags
& S_DIRSYNC
)
4737 ei
->i_flags
|= EXT4_DIRSYNC_FL
;
4740 static blkcnt_t
ext4_inode_blocks(struct ext4_inode
*raw_inode
,
4741 struct ext4_inode_info
*ei
)
4744 struct inode
*inode
= &(ei
->vfs_inode
);
4745 struct super_block
*sb
= inode
->i_sb
;
4747 if (EXT4_HAS_RO_COMPAT_FEATURE(sb
,
4748 EXT4_FEATURE_RO_COMPAT_HUGE_FILE
)) {
4749 /* we are using combined 48 bit field */
4750 i_blocks
= ((u64
)le16_to_cpu(raw_inode
->i_blocks_high
)) << 32 |
4751 le32_to_cpu(raw_inode
->i_blocks_lo
);
4752 if (ei
->i_flags
& EXT4_HUGE_FILE_FL
) {
4753 /* i_blocks represent file system block size */
4754 return i_blocks
<< (inode
->i_blkbits
- 9);
4759 return le32_to_cpu(raw_inode
->i_blocks_lo
);
4763 struct inode
*ext4_iget(struct super_block
*sb
, unsigned long ino
)
4765 struct ext4_iloc iloc
;
4766 struct ext4_inode
*raw_inode
;
4767 struct ext4_inode_info
*ei
;
4768 struct inode
*inode
;
4769 journal_t
*journal
= EXT4_SB(sb
)->s_journal
;
4773 inode
= iget_locked(sb
, ino
);
4775 return ERR_PTR(-ENOMEM
);
4776 if (!(inode
->i_state
& I_NEW
))
4782 ret
= __ext4_get_inode_loc(inode
, &iloc
, 0);
4785 raw_inode
= ext4_raw_inode(&iloc
);
4786 inode
->i_mode
= le16_to_cpu(raw_inode
->i_mode
);
4787 inode
->i_uid
= (uid_t
)le16_to_cpu(raw_inode
->i_uid_low
);
4788 inode
->i_gid
= (gid_t
)le16_to_cpu(raw_inode
->i_gid_low
);
4789 if (!(test_opt(inode
->i_sb
, NO_UID32
))) {
4790 inode
->i_uid
|= le16_to_cpu(raw_inode
->i_uid_high
) << 16;
4791 inode
->i_gid
|= le16_to_cpu(raw_inode
->i_gid_high
) << 16;
4793 inode
->i_nlink
= le16_to_cpu(raw_inode
->i_links_count
);
4796 ei
->i_dir_start_lookup
= 0;
4797 ei
->i_dtime
= le32_to_cpu(raw_inode
->i_dtime
);
4798 /* We now have enough fields to check if the inode was active or not.
4799 * This is needed because nfsd might try to access dead inodes
4800 * the test is that same one that e2fsck uses
4801 * NeilBrown 1999oct15
4803 if (inode
->i_nlink
== 0) {
4804 if (inode
->i_mode
== 0 ||
4805 !(EXT4_SB(inode
->i_sb
)->s_mount_state
& EXT4_ORPHAN_FS
)) {
4806 /* this inode is deleted */
4810 /* The only unlinked inodes we let through here have
4811 * valid i_mode and are being read by the orphan
4812 * recovery code: that's fine, we're about to complete
4813 * the process of deleting those. */
4815 ei
->i_flags
= le32_to_cpu(raw_inode
->i_flags
);
4816 inode
->i_blocks
= ext4_inode_blocks(raw_inode
, ei
);
4817 ei
->i_file_acl
= le32_to_cpu(raw_inode
->i_file_acl_lo
);
4818 if (EXT4_HAS_INCOMPAT_FEATURE(sb
, EXT4_FEATURE_INCOMPAT_64BIT
))
4820 ((__u64
)le16_to_cpu(raw_inode
->i_file_acl_high
)) << 32;
4821 inode
->i_size
= ext4_isize(raw_inode
);
4822 ei
->i_disksize
= inode
->i_size
;
4824 ei
->i_reserved_quota
= 0;
4826 inode
->i_generation
= le32_to_cpu(raw_inode
->i_generation
);
4827 ei
->i_block_group
= iloc
.block_group
;
4828 ei
->i_last_alloc_group
= ~0;
4830 * NOTE! The in-memory inode i_data array is in little-endian order
4831 * even on big-endian machines: we do NOT byteswap the block numbers!
4833 for (block
= 0; block
< EXT4_N_BLOCKS
; block
++)
4834 ei
->i_data
[block
] = raw_inode
->i_block
[block
];
4835 INIT_LIST_HEAD(&ei
->i_orphan
);
4838 * Set transaction id's of transactions that have to be committed
4839 * to finish f[data]sync. We set them to currently running transaction
4840 * as we cannot be sure that the inode or some of its metadata isn't
4841 * part of the transaction - the inode could have been reclaimed and
4842 * now it is reread from disk.
4845 transaction_t
*transaction
;
4848 spin_lock(&journal
->j_state_lock
);
4849 if (journal
->j_running_transaction
)
4850 transaction
= journal
->j_running_transaction
;
4852 transaction
= journal
->j_committing_transaction
;
4854 tid
= transaction
->t_tid
;
4856 tid
= journal
->j_commit_sequence
;
4857 spin_unlock(&journal
->j_state_lock
);
4858 ei
->i_sync_tid
= tid
;
4859 ei
->i_datasync_tid
= tid
;
4862 if (EXT4_INODE_SIZE(inode
->i_sb
) > EXT4_GOOD_OLD_INODE_SIZE
) {
4863 ei
->i_extra_isize
= le16_to_cpu(raw_inode
->i_extra_isize
);
4864 if (EXT4_GOOD_OLD_INODE_SIZE
+ ei
->i_extra_isize
>
4865 EXT4_INODE_SIZE(inode
->i_sb
)) {
4869 if (ei
->i_extra_isize
== 0) {
4870 /* The extra space is currently unused. Use it. */
4871 ei
->i_extra_isize
= sizeof(struct ext4_inode
) -
4872 EXT4_GOOD_OLD_INODE_SIZE
;
4874 __le32
*magic
= (void *)raw_inode
+
4875 EXT4_GOOD_OLD_INODE_SIZE
+
4877 if (*magic
== cpu_to_le32(EXT4_XATTR_MAGIC
))
4878 ei
->i_state
|= EXT4_STATE_XATTR
;
4881 ei
->i_extra_isize
= 0;
4883 EXT4_INODE_GET_XTIME(i_ctime
, inode
, raw_inode
);
4884 EXT4_INODE_GET_XTIME(i_mtime
, inode
, raw_inode
);
4885 EXT4_INODE_GET_XTIME(i_atime
, inode
, raw_inode
);
4886 EXT4_EINODE_GET_XTIME(i_crtime
, ei
, raw_inode
);
4888 inode
->i_version
= le32_to_cpu(raw_inode
->i_disk_version
);
4889 if (EXT4_INODE_SIZE(inode
->i_sb
) > EXT4_GOOD_OLD_INODE_SIZE
) {
4890 if (EXT4_FITS_IN_INODE(raw_inode
, ei
, i_version_hi
))
4892 (__u64
)(le32_to_cpu(raw_inode
->i_version_hi
)) << 32;
4896 if (ei
->i_file_acl
&&
4897 !ext4_data_block_valid(EXT4_SB(sb
), ei
->i_file_acl
, 1)) {
4898 ext4_error(sb
, __func__
,
4899 "bad extended attribute block %llu in inode #%lu",
4900 ei
->i_file_acl
, inode
->i_ino
);
4903 } else if (ei
->i_flags
& EXT4_EXTENTS_FL
) {
4904 if (S_ISREG(inode
->i_mode
) || S_ISDIR(inode
->i_mode
) ||
4905 (S_ISLNK(inode
->i_mode
) &&
4906 !ext4_inode_is_fast_symlink(inode
)))
4907 /* Validate extent which is part of inode */
4908 ret
= ext4_ext_check_inode(inode
);
4909 } else if (S_ISREG(inode
->i_mode
) || S_ISDIR(inode
->i_mode
) ||
4910 (S_ISLNK(inode
->i_mode
) &&
4911 !ext4_inode_is_fast_symlink(inode
))) {
4912 /* Validate block references which are part of inode */
4913 ret
= ext4_check_inode_blockref(inode
);
4918 if (S_ISREG(inode
->i_mode
)) {
4919 inode
->i_op
= &ext4_file_inode_operations
;
4920 inode
->i_fop
= &ext4_file_operations
;
4921 ext4_set_aops(inode
);
4922 } else if (S_ISDIR(inode
->i_mode
)) {
4923 inode
->i_op
= &ext4_dir_inode_operations
;
4924 inode
->i_fop
= &ext4_dir_operations
;
4925 } else if (S_ISLNK(inode
->i_mode
)) {
4926 if (ext4_inode_is_fast_symlink(inode
)) {
4927 inode
->i_op
= &ext4_fast_symlink_inode_operations
;
4928 nd_terminate_link(ei
->i_data
, inode
->i_size
,
4929 sizeof(ei
->i_data
) - 1);
4931 inode
->i_op
= &ext4_symlink_inode_operations
;
4932 ext4_set_aops(inode
);
4934 } else if (S_ISCHR(inode
->i_mode
) || S_ISBLK(inode
->i_mode
) ||
4935 S_ISFIFO(inode
->i_mode
) || S_ISSOCK(inode
->i_mode
)) {
4936 inode
->i_op
= &ext4_special_inode_operations
;
4937 if (raw_inode
->i_block
[0])
4938 init_special_inode(inode
, inode
->i_mode
,
4939 old_decode_dev(le32_to_cpu(raw_inode
->i_block
[0])));
4941 init_special_inode(inode
, inode
->i_mode
,
4942 new_decode_dev(le32_to_cpu(raw_inode
->i_block
[1])));
4945 ext4_error(inode
->i_sb
, __func__
,
4946 "bogus i_mode (%o) for inode=%lu",
4947 inode
->i_mode
, inode
->i_ino
);
4951 ext4_set_inode_flags(inode
);
4952 unlock_new_inode(inode
);
4958 return ERR_PTR(ret
);
4961 static int ext4_inode_blocks_set(handle_t
*handle
,
4962 struct ext4_inode
*raw_inode
,
4963 struct ext4_inode_info
*ei
)
4965 struct inode
*inode
= &(ei
->vfs_inode
);
4966 u64 i_blocks
= inode
->i_blocks
;
4967 struct super_block
*sb
= inode
->i_sb
;
4969 if (i_blocks
<= ~0U) {
4971 * i_blocks can be represnted in a 32 bit variable
4972 * as multiple of 512 bytes
4974 raw_inode
->i_blocks_lo
= cpu_to_le32(i_blocks
);
4975 raw_inode
->i_blocks_high
= 0;
4976 ei
->i_flags
&= ~EXT4_HUGE_FILE_FL
;
4979 if (!EXT4_HAS_RO_COMPAT_FEATURE(sb
, EXT4_FEATURE_RO_COMPAT_HUGE_FILE
))
4982 if (i_blocks
<= 0xffffffffffffULL
) {
4984 * i_blocks can be represented in a 48 bit variable
4985 * as multiple of 512 bytes
4987 raw_inode
->i_blocks_lo
= cpu_to_le32(i_blocks
);
4988 raw_inode
->i_blocks_high
= cpu_to_le16(i_blocks
>> 32);
4989 ei
->i_flags
&= ~EXT4_HUGE_FILE_FL
;
4991 ei
->i_flags
|= EXT4_HUGE_FILE_FL
;
4992 /* i_block is stored in file system block size */
4993 i_blocks
= i_blocks
>> (inode
->i_blkbits
- 9);
4994 raw_inode
->i_blocks_lo
= cpu_to_le32(i_blocks
);
4995 raw_inode
->i_blocks_high
= cpu_to_le16(i_blocks
>> 32);
5001 * Post the struct inode info into an on-disk inode location in the
5002 * buffer-cache. This gobbles the caller's reference to the
5003 * buffer_head in the inode location struct.
5005 * The caller must have write access to iloc->bh.
5007 static int ext4_do_update_inode(handle_t
*handle
,
5008 struct inode
*inode
,
5009 struct ext4_iloc
*iloc
)
5011 struct ext4_inode
*raw_inode
= ext4_raw_inode(iloc
);
5012 struct ext4_inode_info
*ei
= EXT4_I(inode
);
5013 struct buffer_head
*bh
= iloc
->bh
;
5014 int err
= 0, rc
, block
;
5016 /* For fields not not tracking in the in-memory inode,
5017 * initialise them to zero for new inodes. */
5018 if (ei
->i_state
& EXT4_STATE_NEW
)
5019 memset(raw_inode
, 0, EXT4_SB(inode
->i_sb
)->s_inode_size
);
5021 ext4_get_inode_flags(ei
);
5022 raw_inode
->i_mode
= cpu_to_le16(inode
->i_mode
);
5023 if (!(test_opt(inode
->i_sb
, NO_UID32
))) {
5024 raw_inode
->i_uid_low
= cpu_to_le16(low_16_bits(inode
->i_uid
));
5025 raw_inode
->i_gid_low
= cpu_to_le16(low_16_bits(inode
->i_gid
));
5027 * Fix up interoperability with old kernels. Otherwise, old inodes get
5028 * re-used with the upper 16 bits of the uid/gid intact
5031 raw_inode
->i_uid_high
=
5032 cpu_to_le16(high_16_bits(inode
->i_uid
));
5033 raw_inode
->i_gid_high
=
5034 cpu_to_le16(high_16_bits(inode
->i_gid
));
5036 raw_inode
->i_uid_high
= 0;
5037 raw_inode
->i_gid_high
= 0;
5040 raw_inode
->i_uid_low
=
5041 cpu_to_le16(fs_high2lowuid(inode
->i_uid
));
5042 raw_inode
->i_gid_low
=
5043 cpu_to_le16(fs_high2lowgid(inode
->i_gid
));
5044 raw_inode
->i_uid_high
= 0;
5045 raw_inode
->i_gid_high
= 0;
5047 raw_inode
->i_links_count
= cpu_to_le16(inode
->i_nlink
);
5049 EXT4_INODE_SET_XTIME(i_ctime
, inode
, raw_inode
);
5050 EXT4_INODE_SET_XTIME(i_mtime
, inode
, raw_inode
);
5051 EXT4_INODE_SET_XTIME(i_atime
, inode
, raw_inode
);
5052 EXT4_EINODE_SET_XTIME(i_crtime
, ei
, raw_inode
);
5054 if (ext4_inode_blocks_set(handle
, raw_inode
, ei
))
5056 raw_inode
->i_dtime
= cpu_to_le32(ei
->i_dtime
);
5057 raw_inode
->i_flags
= cpu_to_le32(ei
->i_flags
);
5058 if (EXT4_SB(inode
->i_sb
)->s_es
->s_creator_os
!=
5059 cpu_to_le32(EXT4_OS_HURD
))
5060 raw_inode
->i_file_acl_high
=
5061 cpu_to_le16(ei
->i_file_acl
>> 32);
5062 raw_inode
->i_file_acl_lo
= cpu_to_le32(ei
->i_file_acl
);
5063 ext4_isize_set(raw_inode
, ei
->i_disksize
);
5064 if (ei
->i_disksize
> 0x7fffffffULL
) {
5065 struct super_block
*sb
= inode
->i_sb
;
5066 if (!EXT4_HAS_RO_COMPAT_FEATURE(sb
,
5067 EXT4_FEATURE_RO_COMPAT_LARGE_FILE
) ||
5068 EXT4_SB(sb
)->s_es
->s_rev_level
==
5069 cpu_to_le32(EXT4_GOOD_OLD_REV
)) {
5070 /* If this is the first large file
5071 * created, add a flag to the superblock.
5073 err
= ext4_journal_get_write_access(handle
,
5074 EXT4_SB(sb
)->s_sbh
);
5077 ext4_update_dynamic_rev(sb
);
5078 EXT4_SET_RO_COMPAT_FEATURE(sb
,
5079 EXT4_FEATURE_RO_COMPAT_LARGE_FILE
);
5081 ext4_handle_sync(handle
);
5082 err
= ext4_handle_dirty_metadata(handle
, inode
,
5083 EXT4_SB(sb
)->s_sbh
);
5086 raw_inode
->i_generation
= cpu_to_le32(inode
->i_generation
);
5087 if (S_ISCHR(inode
->i_mode
) || S_ISBLK(inode
->i_mode
)) {
5088 if (old_valid_dev(inode
->i_rdev
)) {
5089 raw_inode
->i_block
[0] =
5090 cpu_to_le32(old_encode_dev(inode
->i_rdev
));
5091 raw_inode
->i_block
[1] = 0;
5093 raw_inode
->i_block
[0] = 0;
5094 raw_inode
->i_block
[1] =
5095 cpu_to_le32(new_encode_dev(inode
->i_rdev
));
5096 raw_inode
->i_block
[2] = 0;
5099 for (block
= 0; block
< EXT4_N_BLOCKS
; block
++)
5100 raw_inode
->i_block
[block
] = ei
->i_data
[block
];
5102 raw_inode
->i_disk_version
= cpu_to_le32(inode
->i_version
);
5103 if (ei
->i_extra_isize
) {
5104 if (EXT4_FITS_IN_INODE(raw_inode
, ei
, i_version_hi
))
5105 raw_inode
->i_version_hi
=
5106 cpu_to_le32(inode
->i_version
>> 32);
5107 raw_inode
->i_extra_isize
= cpu_to_le16(ei
->i_extra_isize
);
5110 BUFFER_TRACE(bh
, "call ext4_handle_dirty_metadata");
5111 rc
= ext4_handle_dirty_metadata(handle
, inode
, bh
);
5114 ei
->i_state
&= ~EXT4_STATE_NEW
;
5116 ext4_update_inode_fsync_trans(handle
, inode
, 0);
5119 ext4_std_error(inode
->i_sb
, err
);
5124 * ext4_write_inode()
5126 * We are called from a few places:
5128 * - Within generic_file_write() for O_SYNC files.
5129 * Here, there will be no transaction running. We wait for any running
5130 * trasnaction to commit.
5132 * - Within sys_sync(), kupdate and such.
5133 * We wait on commit, if tol to.
5135 * - Within prune_icache() (PF_MEMALLOC == true)
5136 * Here we simply return. We can't afford to block kswapd on the
5139 * In all cases it is actually safe for us to return without doing anything,
5140 * because the inode has been copied into a raw inode buffer in
5141 * ext4_mark_inode_dirty(). This is a correctness thing for O_SYNC and for
5144 * Note that we are absolutely dependent upon all inode dirtiers doing the
5145 * right thing: they *must* call mark_inode_dirty() after dirtying info in
5146 * which we are interested.
5148 * It would be a bug for them to not do this. The code:
5150 * mark_inode_dirty(inode)
5152 * inode->i_size = expr;
5154 * is in error because a kswapd-driven write_inode() could occur while
5155 * `stuff()' is running, and the new i_size will be lost. Plus the inode
5156 * will no longer be on the superblock's dirty inode list.
5158 int ext4_write_inode(struct inode
*inode
, int wait
)
5162 if (current
->flags
& PF_MEMALLOC
)
5165 if (EXT4_SB(inode
->i_sb
)->s_journal
) {
5166 if (ext4_journal_current_handle()) {
5167 jbd_debug(1, "called recursively, non-PF_MEMALLOC!\n");
5175 err
= ext4_force_commit(inode
->i_sb
);
5177 struct ext4_iloc iloc
;
5179 err
= ext4_get_inode_loc(inode
, &iloc
);
5183 sync_dirty_buffer(iloc
.bh
);
5184 if (buffer_req(iloc
.bh
) && !buffer_uptodate(iloc
.bh
)) {
5185 ext4_error(inode
->i_sb
, __func__
,
5186 "IO error syncing inode, "
5187 "inode=%lu, block=%llu",
5189 (unsigned long long)iloc
.bh
->b_blocknr
);
5199 * Called from notify_change.
5201 * We want to trap VFS attempts to truncate the file as soon as
5202 * possible. In particular, we want to make sure that when the VFS
5203 * shrinks i_size, we put the inode on the orphan list and modify
5204 * i_disksize immediately, so that during the subsequent flushing of
5205 * dirty pages and freeing of disk blocks, we can guarantee that any
5206 * commit will leave the blocks being flushed in an unused state on
5207 * disk. (On recovery, the inode will get truncated and the blocks will
5208 * be freed, so we have a strong guarantee that no future commit will
5209 * leave these blocks visible to the user.)
5211 * Another thing we have to assure is that if we are in ordered mode
5212 * and inode is still attached to the committing transaction, we must
5213 * we start writeout of all the dirty pages which are being truncated.
5214 * This way we are sure that all the data written in the previous
5215 * transaction are already on disk (truncate waits for pages under
5218 * Called with inode->i_mutex down.
5220 int ext4_setattr(struct dentry
*dentry
, struct iattr
*attr
)
5222 struct inode
*inode
= dentry
->d_inode
;
5224 const unsigned int ia_valid
= attr
->ia_valid
;
5226 error
= inode_change_ok(inode
, attr
);
5230 if ((ia_valid
& ATTR_UID
&& attr
->ia_uid
!= inode
->i_uid
) ||
5231 (ia_valid
& ATTR_GID
&& attr
->ia_gid
!= inode
->i_gid
)) {
5234 /* (user+group)*(old+new) structure, inode write (sb,
5235 * inode block, ? - but truncate inode update has it) */
5236 handle
= ext4_journal_start(inode
, (EXT4_MAXQUOTAS_INIT_BLOCKS(inode
->i_sb
)+
5237 EXT4_MAXQUOTAS_DEL_BLOCKS(inode
->i_sb
))+3);
5238 if (IS_ERR(handle
)) {
5239 error
= PTR_ERR(handle
);
5242 error
= vfs_dq_transfer(inode
, attr
) ? -EDQUOT
: 0;
5244 ext4_journal_stop(handle
);
5247 /* Update corresponding info in inode so that everything is in
5248 * one transaction */
5249 if (attr
->ia_valid
& ATTR_UID
)
5250 inode
->i_uid
= attr
->ia_uid
;
5251 if (attr
->ia_valid
& ATTR_GID
)
5252 inode
->i_gid
= attr
->ia_gid
;
5253 error
= ext4_mark_inode_dirty(handle
, inode
);
5254 ext4_journal_stop(handle
);
5257 if (attr
->ia_valid
& ATTR_SIZE
) {
5258 if (!(EXT4_I(inode
)->i_flags
& EXT4_EXTENTS_FL
)) {
5259 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
5261 if (attr
->ia_size
> sbi
->s_bitmap_maxbytes
) {
5268 if (S_ISREG(inode
->i_mode
) &&
5269 attr
->ia_valid
& ATTR_SIZE
&& attr
->ia_size
< inode
->i_size
) {
5272 handle
= ext4_journal_start(inode
, 3);
5273 if (IS_ERR(handle
)) {
5274 error
= PTR_ERR(handle
);
5278 error
= ext4_orphan_add(handle
, inode
);
5279 EXT4_I(inode
)->i_disksize
= attr
->ia_size
;
5280 rc
= ext4_mark_inode_dirty(handle
, inode
);
5283 ext4_journal_stop(handle
);
5285 if (ext4_should_order_data(inode
)) {
5286 error
= ext4_begin_ordered_truncate(inode
,
5289 /* Do as much error cleanup as possible */
5290 handle
= ext4_journal_start(inode
, 3);
5291 if (IS_ERR(handle
)) {
5292 ext4_orphan_del(NULL
, inode
);
5295 ext4_orphan_del(handle
, inode
);
5296 ext4_journal_stop(handle
);
5302 rc
= inode_setattr(inode
, attr
);
5304 /* If inode_setattr's call to ext4_truncate failed to get a
5305 * transaction handle at all, we need to clean up the in-core
5306 * orphan list manually. */
5308 ext4_orphan_del(NULL
, inode
);
5310 if (!rc
&& (ia_valid
& ATTR_MODE
))
5311 rc
= ext4_acl_chmod(inode
);
5314 ext4_std_error(inode
->i_sb
, error
);
5320 int ext4_getattr(struct vfsmount
*mnt
, struct dentry
*dentry
,
5323 struct inode
*inode
;
5324 unsigned long delalloc_blocks
;
5326 inode
= dentry
->d_inode
;
5327 generic_fillattr(inode
, stat
);
5330 * We can't update i_blocks if the block allocation is delayed
5331 * otherwise in the case of system crash before the real block
5332 * allocation is done, we will have i_blocks inconsistent with
5333 * on-disk file blocks.
5334 * We always keep i_blocks updated together with real
5335 * allocation. But to not confuse with user, stat
5336 * will return the blocks that include the delayed allocation
5337 * blocks for this file.
5339 spin_lock(&EXT4_I(inode
)->i_block_reservation_lock
);
5340 delalloc_blocks
= EXT4_I(inode
)->i_reserved_data_blocks
;
5341 spin_unlock(&EXT4_I(inode
)->i_block_reservation_lock
);
5343 stat
->blocks
+= (delalloc_blocks
<< inode
->i_sb
->s_blocksize_bits
)>>9;
5347 static int ext4_indirect_trans_blocks(struct inode
*inode
, int nrblocks
,
5352 /* if nrblocks are contiguous */
5355 * With N contiguous data blocks, it need at most
5356 * N/EXT4_ADDR_PER_BLOCK(inode->i_sb) indirect blocks
5357 * 2 dindirect blocks
5360 indirects
= nrblocks
/ EXT4_ADDR_PER_BLOCK(inode
->i_sb
);
5361 return indirects
+ 3;
5364 * if nrblocks are not contiguous, worse case, each block touch
5365 * a indirect block, and each indirect block touch a double indirect
5366 * block, plus a triple indirect block
5368 indirects
= nrblocks
* 2 + 1;
5372 static int ext4_index_trans_blocks(struct inode
*inode
, int nrblocks
, int chunk
)
5374 if (!(EXT4_I(inode
)->i_flags
& EXT4_EXTENTS_FL
))
5375 return ext4_indirect_trans_blocks(inode
, nrblocks
, chunk
);
5376 return ext4_ext_index_trans_blocks(inode
, nrblocks
, chunk
);
5380 * Account for index blocks, block groups bitmaps and block group
5381 * descriptor blocks if modify datablocks and index blocks
5382 * worse case, the indexs blocks spread over different block groups
5384 * If datablocks are discontiguous, they are possible to spread over
5385 * different block groups too. If they are contiuguous, with flexbg,
5386 * they could still across block group boundary.
5388 * Also account for superblock, inode, quota and xattr blocks
5390 int ext4_meta_trans_blocks(struct inode
*inode
, int nrblocks
, int chunk
)
5392 ext4_group_t groups
, ngroups
= ext4_get_groups_count(inode
->i_sb
);
5398 * How many index blocks need to touch to modify nrblocks?
5399 * The "Chunk" flag indicating whether the nrblocks is
5400 * physically contiguous on disk
5402 * For Direct IO and fallocate, they calls get_block to allocate
5403 * one single extent at a time, so they could set the "Chunk" flag
5405 idxblocks
= ext4_index_trans_blocks(inode
, nrblocks
, chunk
);
5410 * Now let's see how many group bitmaps and group descriptors need
5420 if (groups
> ngroups
)
5422 if (groups
> EXT4_SB(inode
->i_sb
)->s_gdb_count
)
5423 gdpblocks
= EXT4_SB(inode
->i_sb
)->s_gdb_count
;
5425 /* bitmaps and block group descriptor blocks */
5426 ret
+= groups
+ gdpblocks
;
5428 /* Blocks for super block, inode, quota and xattr blocks */
5429 ret
+= EXT4_META_TRANS_BLOCKS(inode
->i_sb
);
5435 * Calulate the total number of credits to reserve to fit
5436 * the modification of a single pages into a single transaction,
5437 * which may include multiple chunks of block allocations.
5439 * This could be called via ext4_write_begin()
5441 * We need to consider the worse case, when
5442 * one new block per extent.
5444 int ext4_writepage_trans_blocks(struct inode
*inode
)
5446 int bpp
= ext4_journal_blocks_per_page(inode
);
5449 ret
= ext4_meta_trans_blocks(inode
, bpp
, 0);
5451 /* Account for data blocks for journalled mode */
5452 if (ext4_should_journal_data(inode
))
5458 * Calculate the journal credits for a chunk of data modification.
5460 * This is called from DIO, fallocate or whoever calling
5461 * ext4_get_blocks() to map/allocate a chunk of contiguous disk blocks.
5463 * journal buffers for data blocks are not included here, as DIO
5464 * and fallocate do no need to journal data buffers.
5466 int ext4_chunk_trans_blocks(struct inode
*inode
, int nrblocks
)
5468 return ext4_meta_trans_blocks(inode
, nrblocks
, 1);
5472 * The caller must have previously called ext4_reserve_inode_write().
5473 * Give this, we know that the caller already has write access to iloc->bh.
5475 int ext4_mark_iloc_dirty(handle_t
*handle
,
5476 struct inode
*inode
, struct ext4_iloc
*iloc
)
5480 if (test_opt(inode
->i_sb
, I_VERSION
))
5481 inode_inc_iversion(inode
);
5483 /* the do_update_inode consumes one bh->b_count */
5486 /* ext4_do_update_inode() does jbd2_journal_dirty_metadata */
5487 err
= ext4_do_update_inode(handle
, inode
, iloc
);
5493 * On success, We end up with an outstanding reference count against
5494 * iloc->bh. This _must_ be cleaned up later.
5498 ext4_reserve_inode_write(handle_t
*handle
, struct inode
*inode
,
5499 struct ext4_iloc
*iloc
)
5503 err
= ext4_get_inode_loc(inode
, iloc
);
5505 BUFFER_TRACE(iloc
->bh
, "get_write_access");
5506 err
= ext4_journal_get_write_access(handle
, iloc
->bh
);
5512 ext4_std_error(inode
->i_sb
, err
);
5517 * Expand an inode by new_extra_isize bytes.
5518 * Returns 0 on success or negative error number on failure.
5520 static int ext4_expand_extra_isize(struct inode
*inode
,
5521 unsigned int new_extra_isize
,
5522 struct ext4_iloc iloc
,
5525 struct ext4_inode
*raw_inode
;
5526 struct ext4_xattr_ibody_header
*header
;
5527 struct ext4_xattr_entry
*entry
;
5529 if (EXT4_I(inode
)->i_extra_isize
>= new_extra_isize
)
5532 raw_inode
= ext4_raw_inode(&iloc
);
5534 header
= IHDR(inode
, raw_inode
);
5535 entry
= IFIRST(header
);
5537 /* No extended attributes present */
5538 if (!(EXT4_I(inode
)->i_state
& EXT4_STATE_XATTR
) ||
5539 header
->h_magic
!= cpu_to_le32(EXT4_XATTR_MAGIC
)) {
5540 memset((void *)raw_inode
+ EXT4_GOOD_OLD_INODE_SIZE
, 0,
5542 EXT4_I(inode
)->i_extra_isize
= new_extra_isize
;
5546 /* try to expand with EAs present */
5547 return ext4_expand_extra_isize_ea(inode
, new_extra_isize
,
5552 * What we do here is to mark the in-core inode as clean with respect to inode
5553 * dirtiness (it may still be data-dirty).
5554 * This means that the in-core inode may be reaped by prune_icache
5555 * without having to perform any I/O. This is a very good thing,
5556 * because *any* task may call prune_icache - even ones which
5557 * have a transaction open against a different journal.
5559 * Is this cheating? Not really. Sure, we haven't written the
5560 * inode out, but prune_icache isn't a user-visible syncing function.
5561 * Whenever the user wants stuff synced (sys_sync, sys_msync, sys_fsync)
5562 * we start and wait on commits.
5564 * Is this efficient/effective? Well, we're being nice to the system
5565 * by cleaning up our inodes proactively so they can be reaped
5566 * without I/O. But we are potentially leaving up to five seconds'
5567 * worth of inodes floating about which prune_icache wants us to
5568 * write out. One way to fix that would be to get prune_icache()
5569 * to do a write_super() to free up some memory. It has the desired
5572 int ext4_mark_inode_dirty(handle_t
*handle
, struct inode
*inode
)
5574 struct ext4_iloc iloc
;
5575 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
5576 static unsigned int mnt_count
;
5580 err
= ext4_reserve_inode_write(handle
, inode
, &iloc
);
5581 if (ext4_handle_valid(handle
) &&
5582 EXT4_I(inode
)->i_extra_isize
< sbi
->s_want_extra_isize
&&
5583 !(EXT4_I(inode
)->i_state
& EXT4_STATE_NO_EXPAND
)) {
5585 * We need extra buffer credits since we may write into EA block
5586 * with this same handle. If journal_extend fails, then it will
5587 * only result in a minor loss of functionality for that inode.
5588 * If this is felt to be critical, then e2fsck should be run to
5589 * force a large enough s_min_extra_isize.
5591 if ((jbd2_journal_extend(handle
,
5592 EXT4_DATA_TRANS_BLOCKS(inode
->i_sb
))) == 0) {
5593 ret
= ext4_expand_extra_isize(inode
,
5594 sbi
->s_want_extra_isize
,
5597 EXT4_I(inode
)->i_state
|= EXT4_STATE_NO_EXPAND
;
5599 le16_to_cpu(sbi
->s_es
->s_mnt_count
)) {
5600 ext4_warning(inode
->i_sb
, __func__
,
5601 "Unable to expand inode %lu. Delete"
5602 " some EAs or run e2fsck.",
5605 le16_to_cpu(sbi
->s_es
->s_mnt_count
);
5611 err
= ext4_mark_iloc_dirty(handle
, inode
, &iloc
);
5616 * ext4_dirty_inode() is called from __mark_inode_dirty()
5618 * We're really interested in the case where a file is being extended.
5619 * i_size has been changed by generic_commit_write() and we thus need
5620 * to include the updated inode in the current transaction.
5622 * Also, vfs_dq_alloc_block() will always dirty the inode when blocks
5623 * are allocated to the file.
5625 * If the inode is marked synchronous, we don't honour that here - doing
5626 * so would cause a commit on atime updates, which we don't bother doing.
5627 * We handle synchronous inodes at the highest possible level.
5629 void ext4_dirty_inode(struct inode
*inode
)
5633 handle
= ext4_journal_start(inode
, 2);
5637 ext4_mark_inode_dirty(handle
, inode
);
5639 ext4_journal_stop(handle
);
5646 * Bind an inode's backing buffer_head into this transaction, to prevent
5647 * it from being flushed to disk early. Unlike
5648 * ext4_reserve_inode_write, this leaves behind no bh reference and
5649 * returns no iloc structure, so the caller needs to repeat the iloc
5650 * lookup to mark the inode dirty later.
5652 static int ext4_pin_inode(handle_t
*handle
, struct inode
*inode
)
5654 struct ext4_iloc iloc
;
5658 err
= ext4_get_inode_loc(inode
, &iloc
);
5660 BUFFER_TRACE(iloc
.bh
, "get_write_access");
5661 err
= jbd2_journal_get_write_access(handle
, iloc
.bh
);
5663 err
= ext4_handle_dirty_metadata(handle
,
5669 ext4_std_error(inode
->i_sb
, err
);
5674 int ext4_change_inode_journal_flag(struct inode
*inode
, int val
)
5681 * We have to be very careful here: changing a data block's
5682 * journaling status dynamically is dangerous. If we write a
5683 * data block to the journal, change the status and then delete
5684 * that block, we risk forgetting to revoke the old log record
5685 * from the journal and so a subsequent replay can corrupt data.
5686 * So, first we make sure that the journal is empty and that
5687 * nobody is changing anything.
5690 journal
= EXT4_JOURNAL(inode
);
5693 if (is_journal_aborted(journal
))
5696 jbd2_journal_lock_updates(journal
);
5697 jbd2_journal_flush(journal
);
5700 * OK, there are no updates running now, and all cached data is
5701 * synced to disk. We are now in a completely consistent state
5702 * which doesn't have anything in the journal, and we know that
5703 * no filesystem updates are running, so it is safe to modify
5704 * the inode's in-core data-journaling state flag now.
5708 EXT4_I(inode
)->i_flags
|= EXT4_JOURNAL_DATA_FL
;
5710 EXT4_I(inode
)->i_flags
&= ~EXT4_JOURNAL_DATA_FL
;
5711 ext4_set_aops(inode
);
5713 jbd2_journal_unlock_updates(journal
);
5715 /* Finally we can mark the inode as dirty. */
5717 handle
= ext4_journal_start(inode
, 1);
5719 return PTR_ERR(handle
);
5721 err
= ext4_mark_inode_dirty(handle
, inode
);
5722 ext4_handle_sync(handle
);
5723 ext4_journal_stop(handle
);
5724 ext4_std_error(inode
->i_sb
, err
);
5729 static int ext4_bh_unmapped(handle_t
*handle
, struct buffer_head
*bh
)
5731 return !buffer_mapped(bh
);
5734 int ext4_page_mkwrite(struct vm_area_struct
*vma
, struct vm_fault
*vmf
)
5736 struct page
*page
= vmf
->page
;
5741 struct file
*file
= vma
->vm_file
;
5742 struct inode
*inode
= file
->f_path
.dentry
->d_inode
;
5743 struct address_space
*mapping
= inode
->i_mapping
;
5746 * Get i_alloc_sem to stop truncates messing with the inode. We cannot
5747 * get i_mutex because we are already holding mmap_sem.
5749 down_read(&inode
->i_alloc_sem
);
5750 size
= i_size_read(inode
);
5751 if (page
->mapping
!= mapping
|| size
<= page_offset(page
)
5752 || !PageUptodate(page
)) {
5753 /* page got truncated from under us? */
5757 if (PageMappedToDisk(page
))
5760 if (page
->index
== size
>> PAGE_CACHE_SHIFT
)
5761 len
= size
& ~PAGE_CACHE_MASK
;
5763 len
= PAGE_CACHE_SIZE
;
5767 * return if we have all the buffers mapped. This avoid
5768 * the need to call write_begin/write_end which does a
5769 * journal_start/journal_stop which can block and take
5772 if (page_has_buffers(page
)) {
5773 if (!walk_page_buffers(NULL
, page_buffers(page
), 0, len
, NULL
,
5774 ext4_bh_unmapped
)) {
5781 * OK, we need to fill the hole... Do write_begin write_end
5782 * to do block allocation/reservation.We are not holding
5783 * inode.i__mutex here. That allow * parallel write_begin,
5784 * write_end call. lock_page prevent this from happening
5785 * on the same page though
5787 ret
= mapping
->a_ops
->write_begin(file
, mapping
, page_offset(page
),
5788 len
, AOP_FLAG_UNINTERRUPTIBLE
, &page
, &fsdata
);
5791 ret
= mapping
->a_ops
->write_end(file
, mapping
, page_offset(page
),
5792 len
, len
, page
, fsdata
);
5798 ret
= VM_FAULT_SIGBUS
;
5799 up_read(&inode
->i_alloc_sem
);