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 "ext4_jbd2.h"
43 #include "ext4_extents.h"
45 #define MPAGE_DA_EXTENT_TAIL 0x01
47 static inline int ext4_begin_ordered_truncate(struct inode
*inode
,
50 return jbd2_journal_begin_ordered_truncate(&EXT4_I(inode
)->jinode
,
54 static void ext4_invalidatepage(struct page
*page
, unsigned long offset
);
57 * Test whether an inode is a fast symlink.
59 static int ext4_inode_is_fast_symlink(struct inode
*inode
)
61 int ea_blocks
= EXT4_I(inode
)->i_file_acl
?
62 (inode
->i_sb
->s_blocksize
>> 9) : 0;
64 return (S_ISLNK(inode
->i_mode
) && inode
->i_blocks
- ea_blocks
== 0);
68 * The ext4 forget function must perform a revoke if we are freeing data
69 * which has been journaled. Metadata (eg. indirect blocks) must be
70 * revoked in all cases.
72 * "bh" may be NULL: a metadata block may have been freed from memory
73 * but there may still be a record of it in the journal, and that record
74 * still needs to be revoked.
76 int ext4_forget(handle_t
*handle
, int is_metadata
, struct inode
*inode
,
77 struct buffer_head
*bh
, ext4_fsblk_t blocknr
)
83 BUFFER_TRACE(bh
, "enter");
85 jbd_debug(4, "forgetting bh %p: is_metadata = %d, mode %o, "
87 bh
, is_metadata
, inode
->i_mode
,
88 test_opt(inode
->i_sb
, DATA_FLAGS
));
90 /* Never use the revoke function if we are doing full data
91 * journaling: there is no need to, and a V1 superblock won't
92 * support it. Otherwise, only skip the revoke on un-journaled
95 if (test_opt(inode
->i_sb
, DATA_FLAGS
) == EXT4_MOUNT_JOURNAL_DATA
||
96 (!is_metadata
&& !ext4_should_journal_data(inode
))) {
98 BUFFER_TRACE(bh
, "call jbd2_journal_forget");
99 return ext4_journal_forget(handle
, bh
);
105 * data!=journal && (is_metadata || should_journal_data(inode))
107 BUFFER_TRACE(bh
, "call ext4_journal_revoke");
108 err
= ext4_journal_revoke(handle
, blocknr
, bh
);
110 ext4_abort(inode
->i_sb
, __func__
,
111 "error %d when attempting revoke", err
);
112 BUFFER_TRACE(bh
, "exit");
117 * Work out how many blocks we need to proceed with the next chunk of a
118 * truncate transaction.
120 static unsigned long blocks_for_truncate(struct inode
*inode
)
124 needed
= inode
->i_blocks
>> (inode
->i_sb
->s_blocksize_bits
- 9);
126 /* Give ourselves just enough room to cope with inodes in which
127 * i_blocks is corrupt: we've seen disk corruptions in the past
128 * which resulted in random data in an inode which looked enough
129 * like a regular file for ext4 to try to delete it. Things
130 * will go a bit crazy if that happens, but at least we should
131 * try not to panic the whole kernel. */
135 /* But we need to bound the transaction so we don't overflow the
137 if (needed
> EXT4_MAX_TRANS_DATA
)
138 needed
= EXT4_MAX_TRANS_DATA
;
140 return EXT4_DATA_TRANS_BLOCKS(inode
->i_sb
) + needed
;
144 * Truncate transactions can be complex and absolutely huge. So we need to
145 * be able to restart the transaction at a conventient checkpoint to make
146 * sure we don't overflow the journal.
148 * start_transaction gets us a new handle for a truncate transaction,
149 * and extend_transaction tries to extend the existing one a bit. If
150 * extend fails, we need to propagate the failure up and restart the
151 * transaction in the top-level truncate loop. --sct
153 static handle_t
*start_transaction(struct inode
*inode
)
157 result
= ext4_journal_start(inode
, blocks_for_truncate(inode
));
161 ext4_std_error(inode
->i_sb
, PTR_ERR(result
));
166 * Try to extend this transaction for the purposes of truncation.
168 * Returns 0 if we managed to create more room. If we can't create more
169 * room, and the transaction must be restarted we return 1.
171 static int try_to_extend_transaction(handle_t
*handle
, struct inode
*inode
)
173 if (handle
->h_buffer_credits
> EXT4_RESERVE_TRANS_BLOCKS
)
175 if (!ext4_journal_extend(handle
, blocks_for_truncate(inode
)))
181 * Restart the transaction associated with *handle. This does a commit,
182 * so before we call here everything must be consistently dirtied against
185 static int ext4_journal_test_restart(handle_t
*handle
, struct inode
*inode
)
187 jbd_debug(2, "restarting handle %p\n", handle
);
188 return ext4_journal_restart(handle
, blocks_for_truncate(inode
));
192 * Called at the last iput() if i_nlink is zero.
194 void ext4_delete_inode(struct inode
*inode
)
199 if (ext4_should_order_data(inode
))
200 ext4_begin_ordered_truncate(inode
, 0);
201 truncate_inode_pages(&inode
->i_data
, 0);
203 if (is_bad_inode(inode
))
206 handle
= ext4_journal_start(inode
, blocks_for_truncate(inode
)+3);
207 if (IS_ERR(handle
)) {
208 ext4_std_error(inode
->i_sb
, PTR_ERR(handle
));
210 * If we're going to skip the normal cleanup, we still need to
211 * make sure that the in-core orphan linked list is properly
214 ext4_orphan_del(NULL
, inode
);
221 err
= ext4_mark_inode_dirty(handle
, inode
);
223 ext4_warning(inode
->i_sb
, __func__
,
224 "couldn't mark inode dirty (err %d)", err
);
228 ext4_truncate(inode
);
231 * ext4_ext_truncate() doesn't reserve any slop when it
232 * restarts journal transactions; therefore there may not be
233 * enough credits left in the handle to remove the inode from
234 * the orphan list and set the dtime field.
236 if (handle
->h_buffer_credits
< 3) {
237 err
= ext4_journal_extend(handle
, 3);
239 err
= ext4_journal_restart(handle
, 3);
241 ext4_warning(inode
->i_sb
, __func__
,
242 "couldn't extend journal (err %d)", err
);
244 ext4_journal_stop(handle
);
250 * Kill off the orphan record which ext4_truncate created.
251 * AKPM: I think this can be inside the above `if'.
252 * Note that ext4_orphan_del() has to be able to cope with the
253 * deletion of a non-existent orphan - this is because we don't
254 * know if ext4_truncate() actually created an orphan record.
255 * (Well, we could do this if we need to, but heck - it works)
257 ext4_orphan_del(handle
, inode
);
258 EXT4_I(inode
)->i_dtime
= get_seconds();
261 * One subtle ordering requirement: if anything has gone wrong
262 * (transaction abort, IO errors, whatever), then we can still
263 * do these next steps (the fs will already have been marked as
264 * having errors), but we can't free the inode if the mark_dirty
267 if (ext4_mark_inode_dirty(handle
, inode
))
268 /* If that failed, just do the required in-core inode clear. */
271 ext4_free_inode(handle
, inode
);
272 ext4_journal_stop(handle
);
275 clear_inode(inode
); /* We must guarantee clearing of inode... */
281 struct buffer_head
*bh
;
284 static inline void add_chain(Indirect
*p
, struct buffer_head
*bh
, __le32
*v
)
286 p
->key
= *(p
->p
= v
);
291 * ext4_block_to_path - parse the block number into array of offsets
292 * @inode: inode in question (we are only interested in its superblock)
293 * @i_block: block number to be parsed
294 * @offsets: array to store the offsets in
295 * @boundary: set this non-zero if the referred-to block is likely to be
296 * followed (on disk) by an indirect block.
298 * To store the locations of file's data ext4 uses a data structure common
299 * for UNIX filesystems - tree of pointers anchored in the inode, with
300 * data blocks at leaves and indirect blocks in intermediate nodes.
301 * This function translates the block number into path in that tree -
302 * return value is the path length and @offsets[n] is the offset of
303 * pointer to (n+1)th node in the nth one. If @block is out of range
304 * (negative or too large) warning is printed and zero returned.
306 * Note: function doesn't find node addresses, so no IO is needed. All
307 * we need to know is the capacity of indirect blocks (taken from the
312 * Portability note: the last comparison (check that we fit into triple
313 * indirect block) is spelled differently, because otherwise on an
314 * architecture with 32-bit longs and 8Kb pages we might get into trouble
315 * if our filesystem had 8Kb blocks. We might use long long, but that would
316 * kill us on x86. Oh, well, at least the sign propagation does not matter -
317 * i_block would have to be negative in the very beginning, so we would not
321 static int ext4_block_to_path(struct inode
*inode
,
323 ext4_lblk_t offsets
[4], int *boundary
)
325 int ptrs
= EXT4_ADDR_PER_BLOCK(inode
->i_sb
);
326 int ptrs_bits
= EXT4_ADDR_PER_BLOCK_BITS(inode
->i_sb
);
327 const long direct_blocks
= EXT4_NDIR_BLOCKS
,
328 indirect_blocks
= ptrs
,
329 double_blocks
= (1 << (ptrs_bits
* 2));
334 ext4_warning(inode
->i_sb
, "ext4_block_to_path", "block < 0");
335 } else if (i_block
< direct_blocks
) {
336 offsets
[n
++] = i_block
;
337 final
= direct_blocks
;
338 } else if ((i_block
-= direct_blocks
) < indirect_blocks
) {
339 offsets
[n
++] = EXT4_IND_BLOCK
;
340 offsets
[n
++] = i_block
;
342 } else if ((i_block
-= indirect_blocks
) < double_blocks
) {
343 offsets
[n
++] = EXT4_DIND_BLOCK
;
344 offsets
[n
++] = i_block
>> ptrs_bits
;
345 offsets
[n
++] = i_block
& (ptrs
- 1);
347 } else if (((i_block
-= double_blocks
) >> (ptrs_bits
* 2)) < ptrs
) {
348 offsets
[n
++] = EXT4_TIND_BLOCK
;
349 offsets
[n
++] = i_block
>> (ptrs_bits
* 2);
350 offsets
[n
++] = (i_block
>> ptrs_bits
) & (ptrs
- 1);
351 offsets
[n
++] = i_block
& (ptrs
- 1);
354 ext4_warning(inode
->i_sb
, "ext4_block_to_path",
356 i_block
+ direct_blocks
+
357 indirect_blocks
+ double_blocks
);
360 *boundary
= final
- 1 - (i_block
& (ptrs
- 1));
365 * ext4_get_branch - read the chain of indirect blocks leading to data
366 * @inode: inode in question
367 * @depth: depth of the chain (1 - direct pointer, etc.)
368 * @offsets: offsets of pointers in inode/indirect blocks
369 * @chain: place to store the result
370 * @err: here we store the error value
372 * Function fills the array of triples <key, p, bh> and returns %NULL
373 * if everything went OK or the pointer to the last filled triple
374 * (incomplete one) otherwise. Upon the return chain[i].key contains
375 * the number of (i+1)-th block in the chain (as it is stored in memory,
376 * i.e. little-endian 32-bit), chain[i].p contains the address of that
377 * number (it points into struct inode for i==0 and into the bh->b_data
378 * for i>0) and chain[i].bh points to the buffer_head of i-th indirect
379 * block for i>0 and NULL for i==0. In other words, it holds the block
380 * numbers of the chain, addresses they were taken from (and where we can
381 * verify that chain did not change) and buffer_heads hosting these
384 * Function stops when it stumbles upon zero pointer (absent block)
385 * (pointer to last triple returned, *@err == 0)
386 * or when it gets an IO error reading an indirect block
387 * (ditto, *@err == -EIO)
388 * or when it reads all @depth-1 indirect blocks successfully and finds
389 * the whole chain, all way to the data (returns %NULL, *err == 0).
391 * Need to be called with
392 * down_read(&EXT4_I(inode)->i_data_sem)
394 static Indirect
*ext4_get_branch(struct inode
*inode
, int depth
,
395 ext4_lblk_t
*offsets
,
396 Indirect chain
[4], int *err
)
398 struct super_block
*sb
= inode
->i_sb
;
400 struct buffer_head
*bh
;
403 /* i_data is not going away, no lock needed */
404 add_chain(chain
, NULL
, EXT4_I(inode
)->i_data
+ *offsets
);
408 bh
= sb_bread(sb
, le32_to_cpu(p
->key
));
411 add_chain(++p
, bh
, (__le32
*)bh
->b_data
+ *++offsets
);
425 * ext4_find_near - find a place for allocation with sufficient locality
427 * @ind: descriptor of indirect block.
429 * This function returns the preferred place for block allocation.
430 * It is used when heuristic for sequential allocation fails.
432 * + if there is a block to the left of our position - allocate near it.
433 * + if pointer will live in indirect block - allocate near that block.
434 * + if pointer will live in inode - allocate in the same
437 * In the latter case we colour the starting block by the callers PID to
438 * prevent it from clashing with concurrent allocations for a different inode
439 * in the same block group. The PID is used here so that functionally related
440 * files will be close-by on-disk.
442 * Caller must make sure that @ind is valid and will stay that way.
444 static ext4_fsblk_t
ext4_find_near(struct inode
*inode
, Indirect
*ind
)
446 struct ext4_inode_info
*ei
= EXT4_I(inode
);
447 __le32
*start
= ind
->bh
? (__le32
*) ind
->bh
->b_data
: ei
->i_data
;
449 ext4_fsblk_t bg_start
;
450 ext4_fsblk_t last_block
;
451 ext4_grpblk_t colour
;
453 /* Try to find previous block */
454 for (p
= ind
->p
- 1; p
>= start
; p
--) {
456 return le32_to_cpu(*p
);
459 /* No such thing, so let's try location of indirect block */
461 return ind
->bh
->b_blocknr
;
464 * It is going to be referred to from the inode itself? OK, just put it
465 * into the same cylinder group then.
467 bg_start
= ext4_group_first_block_no(inode
->i_sb
, ei
->i_block_group
);
468 last_block
= ext4_blocks_count(EXT4_SB(inode
->i_sb
)->s_es
) - 1;
470 if (bg_start
+ EXT4_BLOCKS_PER_GROUP(inode
->i_sb
) <= last_block
)
471 colour
= (current
->pid
% 16) *
472 (EXT4_BLOCKS_PER_GROUP(inode
->i_sb
) / 16);
474 colour
= (current
->pid
% 16) * ((last_block
- bg_start
) / 16);
475 return bg_start
+ colour
;
479 * ext4_find_goal - find a preferred place for allocation.
481 * @block: block we want
482 * @partial: pointer to the last triple within a chain
484 * Normally this function find the preferred place for block allocation,
487 static ext4_fsblk_t
ext4_find_goal(struct inode
*inode
, ext4_lblk_t block
,
491 * XXX need to get goal block from mballoc's data structures
494 return ext4_find_near(inode
, partial
);
498 * ext4_blks_to_allocate: Look up the block map and count the number
499 * of direct blocks need to be allocated for the given branch.
501 * @branch: chain of indirect blocks
502 * @k: number of blocks need for indirect blocks
503 * @blks: number of data blocks to be mapped.
504 * @blocks_to_boundary: the offset in the indirect block
506 * return the total number of blocks to be allocate, including the
507 * direct and indirect blocks.
509 static int ext4_blks_to_allocate(Indirect
*branch
, int k
, unsigned long blks
,
510 int blocks_to_boundary
)
512 unsigned long count
= 0;
515 * Simple case, [t,d]Indirect block(s) has not allocated yet
516 * then it's clear blocks on that path have not allocated
519 /* right now we don't handle cross boundary allocation */
520 if (blks
< blocks_to_boundary
+ 1)
523 count
+= blocks_to_boundary
+ 1;
528 while (count
< blks
&& count
<= blocks_to_boundary
&&
529 le32_to_cpu(*(branch
[0].p
+ count
)) == 0) {
536 * ext4_alloc_blocks: multiple allocate blocks needed for a branch
537 * @indirect_blks: the number of blocks need to allocate for indirect
540 * @new_blocks: on return it will store the new block numbers for
541 * the indirect blocks(if needed) and the first direct block,
542 * @blks: on return it will store the total number of allocated
545 static int ext4_alloc_blocks(handle_t
*handle
, struct inode
*inode
,
546 ext4_lblk_t iblock
, ext4_fsblk_t goal
,
547 int indirect_blks
, int blks
,
548 ext4_fsblk_t new_blocks
[4], int *err
)
550 struct ext4_allocation_request ar
;
552 unsigned long count
= 0, blk_allocated
= 0;
554 ext4_fsblk_t current_block
= 0;
558 * Here we try to allocate the requested multiple blocks at once,
559 * on a best-effort basis.
560 * To build a branch, we should allocate blocks for
561 * the indirect blocks(if not allocated yet), and at least
562 * the first direct block of this branch. That's the
563 * minimum number of blocks need to allocate(required)
565 /* first we try to allocate the indirect blocks */
566 target
= indirect_blks
;
569 /* allocating blocks for indirect blocks and direct blocks */
570 current_block
= ext4_new_meta_blocks(handle
, inode
,
576 /* allocate blocks for indirect blocks */
577 while (index
< indirect_blks
&& count
) {
578 new_blocks
[index
++] = current_block
++;
583 * save the new block number
584 * for the first direct block
586 new_blocks
[index
] = current_block
;
587 printk(KERN_INFO
"%s returned more blocks than "
588 "requested\n", __func__
);
594 target
= blks
- count
;
595 blk_allocated
= count
;
598 /* Now allocate data blocks */
599 memset(&ar
, 0, sizeof(ar
));
604 if (S_ISREG(inode
->i_mode
))
605 /* enable in-core preallocation only for regular files */
606 ar
.flags
= EXT4_MB_HINT_DATA
;
608 current_block
= ext4_mb_new_blocks(handle
, &ar
, err
);
610 if (*err
&& (target
== blks
)) {
612 * if the allocation failed and we didn't allocate
618 if (target
== blks
) {
620 * save the new block number
621 * for the first direct block
623 new_blocks
[index
] = current_block
;
625 blk_allocated
+= ar
.len
;
628 /* total number of blocks allocated for direct blocks */
633 for (i
= 0; i
< index
; i
++)
634 ext4_free_blocks(handle
, inode
, new_blocks
[i
], 1, 0);
639 * ext4_alloc_branch - allocate and set up a chain of blocks.
641 * @indirect_blks: number of allocated indirect blocks
642 * @blks: number of allocated direct blocks
643 * @offsets: offsets (in the blocks) to store the pointers to next.
644 * @branch: place to store the chain in.
646 * This function allocates blocks, zeroes out all but the last one,
647 * links them into chain and (if we are synchronous) writes them to disk.
648 * In other words, it prepares a branch that can be spliced onto the
649 * inode. It stores the information about that chain in the branch[], in
650 * the same format as ext4_get_branch() would do. We are calling it after
651 * we had read the existing part of chain and partial points to the last
652 * triple of that (one with zero ->key). Upon the exit we have the same
653 * picture as after the successful ext4_get_block(), except that in one
654 * place chain is disconnected - *branch->p is still zero (we did not
655 * set the last link), but branch->key contains the number that should
656 * be placed into *branch->p to fill that gap.
658 * If allocation fails we free all blocks we've allocated (and forget
659 * their buffer_heads) and return the error value the from failed
660 * ext4_alloc_block() (normally -ENOSPC). Otherwise we set the chain
661 * as described above and return 0.
663 static int ext4_alloc_branch(handle_t
*handle
, struct inode
*inode
,
664 ext4_lblk_t iblock
, int indirect_blks
,
665 int *blks
, ext4_fsblk_t goal
,
666 ext4_lblk_t
*offsets
, Indirect
*branch
)
668 int blocksize
= inode
->i_sb
->s_blocksize
;
671 struct buffer_head
*bh
;
673 ext4_fsblk_t new_blocks
[4];
674 ext4_fsblk_t current_block
;
676 num
= ext4_alloc_blocks(handle
, inode
, iblock
, goal
, indirect_blks
,
677 *blks
, new_blocks
, &err
);
681 branch
[0].key
= cpu_to_le32(new_blocks
[0]);
683 * metadata blocks and data blocks are allocated.
685 for (n
= 1; n
<= indirect_blks
; n
++) {
687 * Get buffer_head for parent block, zero it out
688 * and set the pointer to new one, then send
691 bh
= sb_getblk(inode
->i_sb
, new_blocks
[n
-1]);
694 BUFFER_TRACE(bh
, "call get_create_access");
695 err
= ext4_journal_get_create_access(handle
, bh
);
702 memset(bh
->b_data
, 0, blocksize
);
703 branch
[n
].p
= (__le32
*) bh
->b_data
+ offsets
[n
];
704 branch
[n
].key
= cpu_to_le32(new_blocks
[n
]);
705 *branch
[n
].p
= branch
[n
].key
;
706 if (n
== indirect_blks
) {
707 current_block
= new_blocks
[n
];
709 * End of chain, update the last new metablock of
710 * the chain to point to the new allocated
711 * data blocks numbers
713 for (i
=1; i
< num
; i
++)
714 *(branch
[n
].p
+ i
) = cpu_to_le32(++current_block
);
716 BUFFER_TRACE(bh
, "marking uptodate");
717 set_buffer_uptodate(bh
);
720 BUFFER_TRACE(bh
, "call ext4_journal_dirty_metadata");
721 err
= ext4_journal_dirty_metadata(handle
, bh
);
728 /* Allocation failed, free what we already allocated */
729 for (i
= 1; i
<= n
; i
++) {
730 BUFFER_TRACE(branch
[i
].bh
, "call jbd2_journal_forget");
731 ext4_journal_forget(handle
, branch
[i
].bh
);
733 for (i
= 0; i
< indirect_blks
; i
++)
734 ext4_free_blocks(handle
, inode
, new_blocks
[i
], 1, 0);
736 ext4_free_blocks(handle
, inode
, new_blocks
[i
], num
, 0);
742 * ext4_splice_branch - splice the allocated branch onto inode.
744 * @block: (logical) number of block we are adding
745 * @chain: chain of indirect blocks (with a missing link - see
747 * @where: location of missing link
748 * @num: number of indirect blocks we are adding
749 * @blks: number of direct blocks we are adding
751 * This function fills the missing link and does all housekeeping needed in
752 * inode (->i_blocks, etc.). In case of success we end up with the full
753 * chain to new block and return 0.
755 static int ext4_splice_branch(handle_t
*handle
, struct inode
*inode
,
756 ext4_lblk_t block
, Indirect
*where
, int num
, int blks
)
760 ext4_fsblk_t current_block
;
763 * If we're splicing into a [td]indirect block (as opposed to the
764 * inode) then we need to get write access to the [td]indirect block
768 BUFFER_TRACE(where
->bh
, "get_write_access");
769 err
= ext4_journal_get_write_access(handle
, where
->bh
);
775 *where
->p
= where
->key
;
778 * Update the host buffer_head or inode to point to more just allocated
779 * direct blocks blocks
781 if (num
== 0 && blks
> 1) {
782 current_block
= le32_to_cpu(where
->key
) + 1;
783 for (i
= 1; i
< blks
; i
++)
784 *(where
->p
+ i
) = cpu_to_le32(current_block
++);
787 /* We are done with atomic stuff, now do the rest of housekeeping */
789 inode
->i_ctime
= ext4_current_time(inode
);
790 ext4_mark_inode_dirty(handle
, inode
);
792 /* had we spliced it onto indirect block? */
795 * If we spliced it onto an indirect block, we haven't
796 * altered the inode. Note however that if it is being spliced
797 * onto an indirect block at the very end of the file (the
798 * file is growing) then we *will* alter the inode to reflect
799 * the new i_size. But that is not done here - it is done in
800 * generic_commit_write->__mark_inode_dirty->ext4_dirty_inode.
802 jbd_debug(5, "splicing indirect only\n");
803 BUFFER_TRACE(where
->bh
, "call ext4_journal_dirty_metadata");
804 err
= ext4_journal_dirty_metadata(handle
, where
->bh
);
809 * OK, we spliced it into the inode itself on a direct block.
810 * Inode was dirtied above.
812 jbd_debug(5, "splicing direct\n");
817 for (i
= 1; i
<= num
; i
++) {
818 BUFFER_TRACE(where
[i
].bh
, "call jbd2_journal_forget");
819 ext4_journal_forget(handle
, where
[i
].bh
);
820 ext4_free_blocks(handle
, inode
,
821 le32_to_cpu(where
[i
-1].key
), 1, 0);
823 ext4_free_blocks(handle
, inode
, le32_to_cpu(where
[num
].key
), blks
, 0);
829 * Allocation strategy is simple: if we have to allocate something, we will
830 * have to go the whole way to leaf. So let's do it before attaching anything
831 * to tree, set linkage between the newborn blocks, write them if sync is
832 * required, recheck the path, free and repeat if check fails, otherwise
833 * set the last missing link (that will protect us from any truncate-generated
834 * removals - all blocks on the path are immune now) and possibly force the
835 * write on the parent block.
836 * That has a nice additional property: no special recovery from the failed
837 * allocations is needed - we simply release blocks and do not touch anything
838 * reachable from inode.
840 * `handle' can be NULL if create == 0.
842 * return > 0, # of blocks mapped or allocated.
843 * return = 0, if plain lookup failed.
844 * return < 0, error case.
847 * Need to be called with
848 * down_read(&EXT4_I(inode)->i_data_sem) if not allocating file system block
849 * (ie, create is zero). Otherwise down_write(&EXT4_I(inode)->i_data_sem)
851 int ext4_get_blocks_handle(handle_t
*handle
, struct inode
*inode
,
852 ext4_lblk_t iblock
, unsigned long maxblocks
,
853 struct buffer_head
*bh_result
,
854 int create
, int extend_disksize
)
857 ext4_lblk_t offsets
[4];
862 int blocks_to_boundary
= 0;
864 struct ext4_inode_info
*ei
= EXT4_I(inode
);
866 ext4_fsblk_t first_block
= 0;
870 J_ASSERT(!(EXT4_I(inode
)->i_flags
& EXT4_EXTENTS_FL
));
871 J_ASSERT(handle
!= NULL
|| create
== 0);
872 depth
= ext4_block_to_path(inode
, iblock
, offsets
,
873 &blocks_to_boundary
);
878 partial
= ext4_get_branch(inode
, depth
, offsets
, chain
, &err
);
880 /* Simplest case - block found, no allocation needed */
882 first_block
= le32_to_cpu(chain
[depth
- 1].key
);
883 clear_buffer_new(bh_result
);
886 while (count
< maxblocks
&& count
<= blocks_to_boundary
) {
889 blk
= le32_to_cpu(*(chain
[depth
-1].p
+ count
));
891 if (blk
== first_block
+ count
)
899 /* Next simple case - plain lookup or failed read of indirect block */
900 if (!create
|| err
== -EIO
)
904 * Okay, we need to do block allocation.
906 goal
= ext4_find_goal(inode
, iblock
, partial
);
908 /* the number of blocks need to allocate for [d,t]indirect blocks */
909 indirect_blks
= (chain
+ depth
) - partial
- 1;
912 * Next look up the indirect map to count the totoal number of
913 * direct blocks to allocate for this branch.
915 count
= ext4_blks_to_allocate(partial
, indirect_blks
,
916 maxblocks
, blocks_to_boundary
);
918 * Block out ext4_truncate while we alter the tree
920 err
= ext4_alloc_branch(handle
, inode
, iblock
, indirect_blks
,
922 offsets
+ (partial
- chain
), partial
);
925 * The ext4_splice_branch call will free and forget any buffers
926 * on the new chain if there is a failure, but that risks using
927 * up transaction credits, especially for bitmaps where the
928 * credits cannot be returned. Can we handle this somehow? We
929 * may need to return -EAGAIN upwards in the worst case. --sct
932 err
= ext4_splice_branch(handle
, inode
, iblock
,
933 partial
, indirect_blks
, count
);
935 * i_disksize growing is protected by i_data_sem. Don't forget to
936 * protect it if you're about to implement concurrent
937 * ext4_get_block() -bzzz
939 if (!err
&& extend_disksize
) {
940 disksize
= ((loff_t
) iblock
+ count
) << inode
->i_blkbits
;
941 if (disksize
> i_size_read(inode
))
942 disksize
= i_size_read(inode
);
943 if (disksize
> ei
->i_disksize
)
944 ei
->i_disksize
= disksize
;
949 set_buffer_new(bh_result
);
951 map_bh(bh_result
, inode
->i_sb
, le32_to_cpu(chain
[depth
-1].key
));
952 if (count
> blocks_to_boundary
)
953 set_buffer_boundary(bh_result
);
955 /* Clean up and exit */
956 partial
= chain
+ depth
- 1; /* the whole chain */
958 while (partial
> chain
) {
959 BUFFER_TRACE(partial
->bh
, "call brelse");
963 BUFFER_TRACE(bh_result
, "returned");
969 * Calculate the number of metadata blocks need to reserve
970 * to allocate @blocks for non extent file based file
972 static int ext4_indirect_calc_metadata_amount(struct inode
*inode
, int blocks
)
974 int icap
= EXT4_ADDR_PER_BLOCK(inode
->i_sb
);
975 int ind_blks
, dind_blks
, tind_blks
;
977 /* number of new indirect blocks needed */
978 ind_blks
= (blocks
+ icap
- 1) / icap
;
980 dind_blks
= (ind_blks
+ icap
- 1) / icap
;
984 return ind_blks
+ dind_blks
+ tind_blks
;
988 * Calculate the number of metadata blocks need to reserve
989 * to allocate given number of blocks
991 static int ext4_calc_metadata_amount(struct inode
*inode
, int blocks
)
996 if (EXT4_I(inode
)->i_flags
& EXT4_EXTENTS_FL
)
997 return ext4_ext_calc_metadata_amount(inode
, blocks
);
999 return ext4_indirect_calc_metadata_amount(inode
, blocks
);
1002 static void ext4_da_update_reserve_space(struct inode
*inode
, int used
)
1004 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
1005 int total
, mdb
, mdb_free
;
1007 spin_lock(&EXT4_I(inode
)->i_block_reservation_lock
);
1008 /* recalculate the number of metablocks still need to be reserved */
1009 total
= EXT4_I(inode
)->i_reserved_data_blocks
- used
;
1010 mdb
= ext4_calc_metadata_amount(inode
, total
);
1012 /* figure out how many metablocks to release */
1013 BUG_ON(mdb
> EXT4_I(inode
)->i_reserved_meta_blocks
);
1014 mdb_free
= EXT4_I(inode
)->i_reserved_meta_blocks
- mdb
;
1017 /* Account for allocated meta_blocks */
1018 mdb_free
-= EXT4_I(inode
)->i_allocated_meta_blocks
;
1020 /* update fs dirty blocks counter */
1021 percpu_counter_sub(&sbi
->s_dirtyblocks_counter
, mdb_free
);
1022 EXT4_I(inode
)->i_allocated_meta_blocks
= 0;
1023 EXT4_I(inode
)->i_reserved_meta_blocks
= mdb
;
1026 /* update per-inode reservations */
1027 BUG_ON(used
> EXT4_I(inode
)->i_reserved_data_blocks
);
1028 EXT4_I(inode
)->i_reserved_data_blocks
-= used
;
1030 spin_unlock(&EXT4_I(inode
)->i_block_reservation_lock
);
1034 * The ext4_get_blocks_wrap() function try to look up the requested blocks,
1035 * and returns if the blocks are already mapped.
1037 * Otherwise it takes the write lock of the i_data_sem and allocate blocks
1038 * and store the allocated blocks in the result buffer head and mark it
1041 * If file type is extents based, it will call ext4_ext_get_blocks(),
1042 * Otherwise, call with ext4_get_blocks_handle() to handle indirect mapping
1045 * On success, it returns the number of blocks being mapped or allocate.
1046 * if create==0 and the blocks are pre-allocated and uninitialized block,
1047 * the result buffer head is unmapped. If the create ==1, it will make sure
1048 * the buffer head is mapped.
1050 * It returns 0 if plain look up failed (blocks have not been allocated), in
1051 * that casem, buffer head is unmapped
1053 * It returns the error in case of allocation failure.
1055 int ext4_get_blocks_wrap(handle_t
*handle
, struct inode
*inode
, sector_t block
,
1056 unsigned long max_blocks
, struct buffer_head
*bh
,
1057 int create
, int extend_disksize
, int flag
)
1061 clear_buffer_mapped(bh
);
1064 * Try to see if we can get the block without requesting
1065 * for new file system block.
1067 down_read((&EXT4_I(inode
)->i_data_sem
));
1068 if (EXT4_I(inode
)->i_flags
& EXT4_EXTENTS_FL
) {
1069 retval
= ext4_ext_get_blocks(handle
, inode
, block
, max_blocks
,
1072 retval
= ext4_get_blocks_handle(handle
,
1073 inode
, block
, max_blocks
, bh
, 0, 0);
1075 up_read((&EXT4_I(inode
)->i_data_sem
));
1077 /* If it is only a block(s) look up */
1082 * Returns if the blocks have already allocated
1084 * Note that if blocks have been preallocated
1085 * ext4_ext_get_block() returns th create = 0
1086 * with buffer head unmapped.
1088 if (retval
> 0 && buffer_mapped(bh
))
1092 * New blocks allocate and/or writing to uninitialized extent
1093 * will possibly result in updating i_data, so we take
1094 * the write lock of i_data_sem, and call get_blocks()
1095 * with create == 1 flag.
1097 down_write((&EXT4_I(inode
)->i_data_sem
));
1100 * if the caller is from delayed allocation writeout path
1101 * we have already reserved fs blocks for allocation
1102 * let the underlying get_block() function know to
1103 * avoid double accounting
1106 EXT4_I(inode
)->i_delalloc_reserved_flag
= 1;
1108 * We need to check for EXT4 here because migrate
1109 * could have changed the inode type in between
1111 if (EXT4_I(inode
)->i_flags
& EXT4_EXTENTS_FL
) {
1112 retval
= ext4_ext_get_blocks(handle
, inode
, block
, max_blocks
,
1113 bh
, create
, extend_disksize
);
1115 retval
= ext4_get_blocks_handle(handle
, inode
, block
,
1116 max_blocks
, bh
, create
, extend_disksize
);
1118 if (retval
> 0 && buffer_new(bh
)) {
1120 * We allocated new blocks which will result in
1121 * i_data's format changing. Force the migrate
1122 * to fail by clearing migrate flags
1124 EXT4_I(inode
)->i_flags
= EXT4_I(inode
)->i_flags
&
1130 EXT4_I(inode
)->i_delalloc_reserved_flag
= 0;
1132 * Update reserved blocks/metadata blocks
1133 * after successful block allocation
1134 * which were deferred till now
1136 if ((retval
> 0) && buffer_delay(bh
))
1137 ext4_da_update_reserve_space(inode
, retval
);
1140 up_write((&EXT4_I(inode
)->i_data_sem
));
1144 /* Maximum number of blocks we map for direct IO at once. */
1145 #define DIO_MAX_BLOCKS 4096
1147 int ext4_get_block(struct inode
*inode
, sector_t iblock
,
1148 struct buffer_head
*bh_result
, int create
)
1150 handle_t
*handle
= ext4_journal_current_handle();
1151 int ret
= 0, started
= 0;
1152 unsigned max_blocks
= bh_result
->b_size
>> inode
->i_blkbits
;
1155 if (create
&& !handle
) {
1156 /* Direct IO write... */
1157 if (max_blocks
> DIO_MAX_BLOCKS
)
1158 max_blocks
= DIO_MAX_BLOCKS
;
1159 dio_credits
= ext4_chunk_trans_blocks(inode
, max_blocks
);
1160 handle
= ext4_journal_start(inode
, dio_credits
);
1161 if (IS_ERR(handle
)) {
1162 ret
= PTR_ERR(handle
);
1168 ret
= ext4_get_blocks_wrap(handle
, inode
, iblock
,
1169 max_blocks
, bh_result
, create
, 0, 0);
1171 bh_result
->b_size
= (ret
<< inode
->i_blkbits
);
1175 ext4_journal_stop(handle
);
1181 * `handle' can be NULL if create is zero
1183 struct buffer_head
*ext4_getblk(handle_t
*handle
, struct inode
*inode
,
1184 ext4_lblk_t block
, int create
, int *errp
)
1186 struct buffer_head dummy
;
1189 J_ASSERT(handle
!= NULL
|| create
== 0);
1192 dummy
.b_blocknr
= -1000;
1193 buffer_trace_init(&dummy
.b_history
);
1194 err
= ext4_get_blocks_wrap(handle
, inode
, block
, 1,
1195 &dummy
, create
, 1, 0);
1197 * ext4_get_blocks_handle() returns number of blocks
1198 * mapped. 0 in case of a HOLE.
1206 if (!err
&& buffer_mapped(&dummy
)) {
1207 struct buffer_head
*bh
;
1208 bh
= sb_getblk(inode
->i_sb
, dummy
.b_blocknr
);
1213 if (buffer_new(&dummy
)) {
1214 J_ASSERT(create
!= 0);
1215 J_ASSERT(handle
!= NULL
);
1218 * Now that we do not always journal data, we should
1219 * keep in mind whether this should always journal the
1220 * new buffer as metadata. For now, regular file
1221 * writes use ext4_get_block instead, so it's not a
1225 BUFFER_TRACE(bh
, "call get_create_access");
1226 fatal
= ext4_journal_get_create_access(handle
, bh
);
1227 if (!fatal
&& !buffer_uptodate(bh
)) {
1228 memset(bh
->b_data
, 0, inode
->i_sb
->s_blocksize
);
1229 set_buffer_uptodate(bh
);
1232 BUFFER_TRACE(bh
, "call ext4_journal_dirty_metadata");
1233 err
= ext4_journal_dirty_metadata(handle
, bh
);
1237 BUFFER_TRACE(bh
, "not a new buffer");
1250 struct buffer_head
*ext4_bread(handle_t
*handle
, struct inode
*inode
,
1251 ext4_lblk_t block
, int create
, int *err
)
1253 struct buffer_head
*bh
;
1255 bh
= ext4_getblk(handle
, inode
, block
, create
, err
);
1258 if (buffer_uptodate(bh
))
1260 ll_rw_block(READ_META
, 1, &bh
);
1262 if (buffer_uptodate(bh
))
1269 static int walk_page_buffers(handle_t
*handle
,
1270 struct buffer_head
*head
,
1274 int (*fn
)(handle_t
*handle
,
1275 struct buffer_head
*bh
))
1277 struct buffer_head
*bh
;
1278 unsigned block_start
, block_end
;
1279 unsigned blocksize
= head
->b_size
;
1281 struct buffer_head
*next
;
1283 for (bh
= head
, block_start
= 0;
1284 ret
== 0 && (bh
!= head
|| !block_start
);
1285 block_start
= block_end
, bh
= next
)
1287 next
= bh
->b_this_page
;
1288 block_end
= block_start
+ blocksize
;
1289 if (block_end
<= from
|| block_start
>= to
) {
1290 if (partial
&& !buffer_uptodate(bh
))
1294 err
= (*fn
)(handle
, bh
);
1302 * To preserve ordering, it is essential that the hole instantiation and
1303 * the data write be encapsulated in a single transaction. We cannot
1304 * close off a transaction and start a new one between the ext4_get_block()
1305 * and the commit_write(). So doing the jbd2_journal_start at the start of
1306 * prepare_write() is the right place.
1308 * Also, this function can nest inside ext4_writepage() ->
1309 * block_write_full_page(). In that case, we *know* that ext4_writepage()
1310 * has generated enough buffer credits to do the whole page. So we won't
1311 * block on the journal in that case, which is good, because the caller may
1314 * By accident, ext4 can be reentered when a transaction is open via
1315 * quota file writes. If we were to commit the transaction while thus
1316 * reentered, there can be a deadlock - we would be holding a quota
1317 * lock, and the commit would never complete if another thread had a
1318 * transaction open and was blocking on the quota lock - a ranking
1321 * So what we do is to rely on the fact that jbd2_journal_stop/journal_start
1322 * will _not_ run commit under these circumstances because handle->h_ref
1323 * is elevated. We'll still have enough credits for the tiny quotafile
1326 static int do_journal_get_write_access(handle_t
*handle
,
1327 struct buffer_head
*bh
)
1329 if (!buffer_mapped(bh
) || buffer_freed(bh
))
1331 return ext4_journal_get_write_access(handle
, bh
);
1334 static int ext4_write_begin(struct file
*file
, struct address_space
*mapping
,
1335 loff_t pos
, unsigned len
, unsigned flags
,
1336 struct page
**pagep
, void **fsdata
)
1338 struct inode
*inode
= mapping
->host
;
1339 int ret
, needed_blocks
= ext4_writepage_trans_blocks(inode
);
1346 index
= pos
>> PAGE_CACHE_SHIFT
;
1347 from
= pos
& (PAGE_CACHE_SIZE
- 1);
1351 handle
= ext4_journal_start(inode
, needed_blocks
);
1352 if (IS_ERR(handle
)) {
1353 ret
= PTR_ERR(handle
);
1357 page
= grab_cache_page_write_begin(mapping
, index
, flags
);
1359 ext4_journal_stop(handle
);
1365 ret
= block_write_begin(file
, mapping
, pos
, len
, flags
, pagep
, fsdata
,
1368 if (!ret
&& ext4_should_journal_data(inode
)) {
1369 ret
= walk_page_buffers(handle
, page_buffers(page
),
1370 from
, to
, NULL
, do_journal_get_write_access
);
1375 ext4_journal_stop(handle
);
1376 page_cache_release(page
);
1378 * block_write_begin may have instantiated a few blocks
1379 * outside i_size. Trim these off again. Don't need
1380 * i_size_read because we hold i_mutex.
1382 if (pos
+ len
> inode
->i_size
)
1383 vmtruncate(inode
, inode
->i_size
);
1386 if (ret
== -ENOSPC
&& ext4_should_retry_alloc(inode
->i_sb
, &retries
))
1392 /* For write_end() in data=journal mode */
1393 static int write_end_fn(handle_t
*handle
, struct buffer_head
*bh
)
1395 if (!buffer_mapped(bh
) || buffer_freed(bh
))
1397 set_buffer_uptodate(bh
);
1398 return ext4_journal_dirty_metadata(handle
, bh
);
1402 * We need to pick up the new inode size which generic_commit_write gave us
1403 * `file' can be NULL - eg, when called from page_symlink().
1405 * ext4 never places buffers on inode->i_mapping->private_list. metadata
1406 * buffers are managed internally.
1408 static int ext4_ordered_write_end(struct file
*file
,
1409 struct address_space
*mapping
,
1410 loff_t pos
, unsigned len
, unsigned copied
,
1411 struct page
*page
, void *fsdata
)
1413 handle_t
*handle
= ext4_journal_current_handle();
1414 struct inode
*inode
= mapping
->host
;
1417 ret
= ext4_jbd2_file_inode(handle
, inode
);
1422 new_i_size
= pos
+ copied
;
1423 if (new_i_size
> EXT4_I(inode
)->i_disksize
) {
1424 ext4_update_i_disksize(inode
, new_i_size
);
1425 /* We need to mark inode dirty even if
1426 * new_i_size is less that inode->i_size
1427 * bu greater than i_disksize.(hint delalloc)
1429 ext4_mark_inode_dirty(handle
, inode
);
1432 ret2
= generic_write_end(file
, mapping
, pos
, len
, copied
,
1438 ret2
= ext4_journal_stop(handle
);
1442 return ret
? ret
: copied
;
1445 static int ext4_writeback_write_end(struct file
*file
,
1446 struct address_space
*mapping
,
1447 loff_t pos
, unsigned len
, unsigned copied
,
1448 struct page
*page
, void *fsdata
)
1450 handle_t
*handle
= ext4_journal_current_handle();
1451 struct inode
*inode
= mapping
->host
;
1455 new_i_size
= pos
+ copied
;
1456 if (new_i_size
> EXT4_I(inode
)->i_disksize
) {
1457 ext4_update_i_disksize(inode
, new_i_size
);
1458 /* We need to mark inode dirty even if
1459 * new_i_size is less that inode->i_size
1460 * bu greater than i_disksize.(hint delalloc)
1462 ext4_mark_inode_dirty(handle
, inode
);
1465 ret2
= generic_write_end(file
, mapping
, pos
, len
, copied
,
1471 ret2
= ext4_journal_stop(handle
);
1475 return ret
? ret
: copied
;
1478 static int ext4_journalled_write_end(struct file
*file
,
1479 struct address_space
*mapping
,
1480 loff_t pos
, unsigned len
, unsigned copied
,
1481 struct page
*page
, void *fsdata
)
1483 handle_t
*handle
= ext4_journal_current_handle();
1484 struct inode
*inode
= mapping
->host
;
1490 from
= pos
& (PAGE_CACHE_SIZE
- 1);
1494 if (!PageUptodate(page
))
1496 page_zero_new_buffers(page
, from
+copied
, to
);
1499 ret
= walk_page_buffers(handle
, page_buffers(page
), from
,
1500 to
, &partial
, write_end_fn
);
1502 SetPageUptodate(page
);
1503 new_i_size
= pos
+ copied
;
1504 if (new_i_size
> inode
->i_size
)
1505 i_size_write(inode
, pos
+copied
);
1506 EXT4_I(inode
)->i_state
|= EXT4_STATE_JDATA
;
1507 if (new_i_size
> EXT4_I(inode
)->i_disksize
) {
1508 ext4_update_i_disksize(inode
, new_i_size
);
1509 ret2
= ext4_mark_inode_dirty(handle
, inode
);
1515 ret2
= ext4_journal_stop(handle
);
1518 page_cache_release(page
);
1520 return ret
? ret
: copied
;
1523 static int ext4_da_reserve_space(struct inode
*inode
, int nrblocks
)
1526 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
1527 unsigned long md_needed
, mdblocks
, total
= 0;
1530 * recalculate the amount of metadata blocks to reserve
1531 * in order to allocate nrblocks
1532 * worse case is one extent per block
1535 spin_lock(&EXT4_I(inode
)->i_block_reservation_lock
);
1536 total
= EXT4_I(inode
)->i_reserved_data_blocks
+ nrblocks
;
1537 mdblocks
= ext4_calc_metadata_amount(inode
, total
);
1538 BUG_ON(mdblocks
< EXT4_I(inode
)->i_reserved_meta_blocks
);
1540 md_needed
= mdblocks
- EXT4_I(inode
)->i_reserved_meta_blocks
;
1541 total
= md_needed
+ nrblocks
;
1543 if (ext4_claim_free_blocks(sbi
, total
)) {
1544 spin_unlock(&EXT4_I(inode
)->i_block_reservation_lock
);
1545 if (ext4_should_retry_alloc(inode
->i_sb
, &retries
)) {
1551 EXT4_I(inode
)->i_reserved_data_blocks
+= nrblocks
;
1552 EXT4_I(inode
)->i_reserved_meta_blocks
= mdblocks
;
1554 spin_unlock(&EXT4_I(inode
)->i_block_reservation_lock
);
1555 return 0; /* success */
1558 static void ext4_da_release_space(struct inode
*inode
, int to_free
)
1560 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
1561 int total
, mdb
, mdb_free
, release
;
1564 return; /* Nothing to release, exit */
1566 spin_lock(&EXT4_I(inode
)->i_block_reservation_lock
);
1568 if (!EXT4_I(inode
)->i_reserved_data_blocks
) {
1570 * if there is no reserved blocks, but we try to free some
1571 * then the counter is messed up somewhere.
1572 * but since this function is called from invalidate
1573 * page, it's harmless to return without any action
1575 printk(KERN_INFO
"ext4 delalloc try to release %d reserved "
1576 "blocks for inode %lu, but there is no reserved "
1577 "data blocks\n", to_free
, inode
->i_ino
);
1578 spin_unlock(&EXT4_I(inode
)->i_block_reservation_lock
);
1582 /* recalculate the number of metablocks still need to be reserved */
1583 total
= EXT4_I(inode
)->i_reserved_data_blocks
- to_free
;
1584 mdb
= ext4_calc_metadata_amount(inode
, total
);
1586 /* figure out how many metablocks to release */
1587 BUG_ON(mdb
> EXT4_I(inode
)->i_reserved_meta_blocks
);
1588 mdb_free
= EXT4_I(inode
)->i_reserved_meta_blocks
- mdb
;
1590 release
= to_free
+ mdb_free
;
1592 /* update fs dirty blocks counter for truncate case */
1593 percpu_counter_sub(&sbi
->s_dirtyblocks_counter
, release
);
1595 /* update per-inode reservations */
1596 BUG_ON(to_free
> EXT4_I(inode
)->i_reserved_data_blocks
);
1597 EXT4_I(inode
)->i_reserved_data_blocks
-= to_free
;
1599 BUG_ON(mdb
> EXT4_I(inode
)->i_reserved_meta_blocks
);
1600 EXT4_I(inode
)->i_reserved_meta_blocks
= mdb
;
1601 spin_unlock(&EXT4_I(inode
)->i_block_reservation_lock
);
1604 static void ext4_da_page_release_reservation(struct page
*page
,
1605 unsigned long offset
)
1608 struct buffer_head
*head
, *bh
;
1609 unsigned int curr_off
= 0;
1611 head
= page_buffers(page
);
1614 unsigned int next_off
= curr_off
+ bh
->b_size
;
1616 if ((offset
<= curr_off
) && (buffer_delay(bh
))) {
1618 clear_buffer_delay(bh
);
1620 curr_off
= next_off
;
1621 } while ((bh
= bh
->b_this_page
) != head
);
1622 ext4_da_release_space(page
->mapping
->host
, to_release
);
1626 * Delayed allocation stuff
1629 struct mpage_da_data
{
1630 struct inode
*inode
;
1631 struct buffer_head lbh
; /* extent of blocks */
1632 unsigned long first_page
, next_page
; /* extent of pages */
1633 get_block_t
*get_block
;
1634 struct writeback_control
*wbc
;
1641 * mpage_da_submit_io - walks through extent of pages and try to write
1642 * them with writepage() call back
1644 * @mpd->inode: inode
1645 * @mpd->first_page: first page of the extent
1646 * @mpd->next_page: page after the last page of the extent
1647 * @mpd->get_block: the filesystem's block mapper function
1649 * By the time mpage_da_submit_io() is called we expect all blocks
1650 * to be allocated. this may be wrong if allocation failed.
1652 * As pages are already locked by write_cache_pages(), we can't use it
1654 static int mpage_da_submit_io(struct mpage_da_data
*mpd
)
1657 struct pagevec pvec
;
1658 unsigned long index
, end
;
1659 int ret
= 0, err
, nr_pages
, i
;
1660 struct inode
*inode
= mpd
->inode
;
1661 struct address_space
*mapping
= inode
->i_mapping
;
1663 BUG_ON(mpd
->next_page
<= mpd
->first_page
);
1665 * We need to start from the first_page to the next_page - 1
1666 * to make sure we also write the mapped dirty buffer_heads.
1667 * If we look at mpd->lbh.b_blocknr we would only be looking
1668 * at the currently mapped buffer_heads.
1670 index
= mpd
->first_page
;
1671 end
= mpd
->next_page
- 1;
1673 pagevec_init(&pvec
, 0);
1674 while (index
<= end
) {
1675 nr_pages
= pagevec_lookup(&pvec
, mapping
, index
, PAGEVEC_SIZE
);
1678 for (i
= 0; i
< nr_pages
; i
++) {
1679 struct page
*page
= pvec
.pages
[i
];
1681 index
= page
->index
;
1686 BUG_ON(!PageLocked(page
));
1687 BUG_ON(PageWriteback(page
));
1689 pages_skipped
= mpd
->wbc
->pages_skipped
;
1690 err
= mapping
->a_ops
->writepage(page
, mpd
->wbc
);
1691 if (!err
&& (pages_skipped
== mpd
->wbc
->pages_skipped
))
1693 * have successfully written the page
1694 * without skipping the same
1696 mpd
->pages_written
++;
1698 * In error case, we have to continue because
1699 * remaining pages are still locked
1700 * XXX: unlock and re-dirty them?
1705 pagevec_release(&pvec
);
1711 * mpage_put_bnr_to_bhs - walk blocks and assign them actual numbers
1713 * @mpd->inode - inode to walk through
1714 * @exbh->b_blocknr - first block on a disk
1715 * @exbh->b_size - amount of space in bytes
1716 * @logical - first logical block to start assignment with
1718 * the function goes through all passed space and put actual disk
1719 * block numbers into buffer heads, dropping BH_Delay
1721 static void mpage_put_bnr_to_bhs(struct mpage_da_data
*mpd
, sector_t logical
,
1722 struct buffer_head
*exbh
)
1724 struct inode
*inode
= mpd
->inode
;
1725 struct address_space
*mapping
= inode
->i_mapping
;
1726 int blocks
= exbh
->b_size
>> inode
->i_blkbits
;
1727 sector_t pblock
= exbh
->b_blocknr
, cur_logical
;
1728 struct buffer_head
*head
, *bh
;
1730 struct pagevec pvec
;
1733 index
= logical
>> (PAGE_CACHE_SHIFT
- inode
->i_blkbits
);
1734 end
= (logical
+ blocks
- 1) >> (PAGE_CACHE_SHIFT
- inode
->i_blkbits
);
1735 cur_logical
= index
<< (PAGE_CACHE_SHIFT
- inode
->i_blkbits
);
1737 pagevec_init(&pvec
, 0);
1739 while (index
<= end
) {
1740 /* XXX: optimize tail */
1741 nr_pages
= pagevec_lookup(&pvec
, mapping
, index
, PAGEVEC_SIZE
);
1744 for (i
= 0; i
< nr_pages
; i
++) {
1745 struct page
*page
= pvec
.pages
[i
];
1747 index
= page
->index
;
1752 BUG_ON(!PageLocked(page
));
1753 BUG_ON(PageWriteback(page
));
1754 BUG_ON(!page_has_buffers(page
));
1756 bh
= page_buffers(page
);
1759 /* skip blocks out of the range */
1761 if (cur_logical
>= logical
)
1764 } while ((bh
= bh
->b_this_page
) != head
);
1767 if (cur_logical
>= logical
+ blocks
)
1769 if (buffer_delay(bh
)) {
1770 bh
->b_blocknr
= pblock
;
1771 clear_buffer_delay(bh
);
1772 bh
->b_bdev
= inode
->i_sb
->s_bdev
;
1773 } else if (buffer_unwritten(bh
)) {
1774 bh
->b_blocknr
= pblock
;
1775 clear_buffer_unwritten(bh
);
1776 set_buffer_mapped(bh
);
1778 bh
->b_bdev
= inode
->i_sb
->s_bdev
;
1779 } else if (buffer_mapped(bh
))
1780 BUG_ON(bh
->b_blocknr
!= pblock
);
1784 } while ((bh
= bh
->b_this_page
) != head
);
1786 pagevec_release(&pvec
);
1792 * __unmap_underlying_blocks - just a helper function to unmap
1793 * set of blocks described by @bh
1795 static inline void __unmap_underlying_blocks(struct inode
*inode
,
1796 struct buffer_head
*bh
)
1798 struct block_device
*bdev
= inode
->i_sb
->s_bdev
;
1801 blocks
= bh
->b_size
>> inode
->i_blkbits
;
1802 for (i
= 0; i
< blocks
; i
++)
1803 unmap_underlying_metadata(bdev
, bh
->b_blocknr
+ i
);
1806 static void ext4_da_block_invalidatepages(struct mpage_da_data
*mpd
,
1807 sector_t logical
, long blk_cnt
)
1811 struct pagevec pvec
;
1812 struct inode
*inode
= mpd
->inode
;
1813 struct address_space
*mapping
= inode
->i_mapping
;
1815 index
= logical
>> (PAGE_CACHE_SHIFT
- inode
->i_blkbits
);
1816 end
= (logical
+ blk_cnt
- 1) >>
1817 (PAGE_CACHE_SHIFT
- inode
->i_blkbits
);
1818 while (index
<= end
) {
1819 nr_pages
= pagevec_lookup(&pvec
, mapping
, index
, PAGEVEC_SIZE
);
1822 for (i
= 0; i
< nr_pages
; i
++) {
1823 struct page
*page
= pvec
.pages
[i
];
1824 index
= page
->index
;
1829 BUG_ON(!PageLocked(page
));
1830 BUG_ON(PageWriteback(page
));
1831 block_invalidatepage(page
, 0);
1832 ClearPageUptodate(page
);
1839 static void ext4_print_free_blocks(struct inode
*inode
)
1841 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
1842 printk(KERN_EMERG
"Total free blocks count %lld\n",
1843 ext4_count_free_blocks(inode
->i_sb
));
1844 printk(KERN_EMERG
"Free/Dirty block details\n");
1845 printk(KERN_EMERG
"free_blocks=%lld\n",
1846 (long long)percpu_counter_sum(&sbi
->s_freeblocks_counter
));
1847 printk(KERN_EMERG
"dirty_blocks=%lld\n",
1848 (long long)percpu_counter_sum(&sbi
->s_dirtyblocks_counter
));
1849 printk(KERN_EMERG
"Block reservation details\n");
1850 printk(KERN_EMERG
"i_reserved_data_blocks=%lu\n",
1851 EXT4_I(inode
)->i_reserved_data_blocks
);
1852 printk(KERN_EMERG
"i_reserved_meta_blocks=%lu\n",
1853 EXT4_I(inode
)->i_reserved_meta_blocks
);
1858 * mpage_da_map_blocks - go through given space
1860 * @mpd->lbh - bh describing space
1861 * @mpd->get_block - the filesystem's block mapper function
1863 * The function skips space we know is already mapped to disk blocks.
1866 static int mpage_da_map_blocks(struct mpage_da_data
*mpd
)
1869 struct buffer_head
new;
1870 struct buffer_head
*lbh
= &mpd
->lbh
;
1874 * We consider only non-mapped and non-allocated blocks
1876 if (buffer_mapped(lbh
) && !buffer_delay(lbh
))
1878 new.b_state
= lbh
->b_state
;
1880 new.b_size
= lbh
->b_size
;
1881 next
= lbh
->b_blocknr
;
1883 * If we didn't accumulate anything
1884 * to write simply return
1888 err
= mpd
->get_block(mpd
->inode
, next
, &new, 1);
1891 /* If get block returns with error
1892 * we simply return. Later writepage
1893 * will redirty the page and writepages
1894 * will find the dirty page again
1899 if (err
== -ENOSPC
&&
1900 ext4_count_free_blocks(mpd
->inode
->i_sb
)) {
1906 * get block failure will cause us
1907 * to loop in writepages. Because
1908 * a_ops->writepage won't be able to
1909 * make progress. The page will be redirtied
1910 * by writepage and writepages will again
1911 * try to write the same.
1913 printk(KERN_EMERG
"%s block allocation failed for inode %lu "
1914 "at logical offset %llu with max blocks "
1915 "%zd with error %d\n",
1916 __func__
, mpd
->inode
->i_ino
,
1917 (unsigned long long)next
,
1918 lbh
->b_size
>> mpd
->inode
->i_blkbits
, err
);
1919 printk(KERN_EMERG
"This should not happen.!! "
1920 "Data will be lost\n");
1921 if (err
== -ENOSPC
) {
1922 ext4_print_free_blocks(mpd
->inode
);
1924 /* invlaidate all the pages */
1925 ext4_da_block_invalidatepages(mpd
, next
,
1926 lbh
->b_size
>> mpd
->inode
->i_blkbits
);
1929 BUG_ON(new.b_size
== 0);
1931 if (buffer_new(&new))
1932 __unmap_underlying_blocks(mpd
->inode
, &new);
1935 * If blocks are delayed marked, we need to
1936 * put actual blocknr and drop delayed bit
1938 if (buffer_delay(lbh
) || buffer_unwritten(lbh
))
1939 mpage_put_bnr_to_bhs(mpd
, next
, &new);
1944 #define BH_FLAGS ((1 << BH_Uptodate) | (1 << BH_Mapped) | \
1945 (1 << BH_Delay) | (1 << BH_Unwritten))
1948 * mpage_add_bh_to_extent - try to add one more block to extent of blocks
1950 * @mpd->lbh - extent of blocks
1951 * @logical - logical number of the block in the file
1952 * @bh - bh of the block (used to access block's state)
1954 * the function is used to collect contig. blocks in same state
1956 static void mpage_add_bh_to_extent(struct mpage_da_data
*mpd
,
1957 sector_t logical
, struct buffer_head
*bh
)
1960 size_t b_size
= bh
->b_size
;
1961 struct buffer_head
*lbh
= &mpd
->lbh
;
1962 int nrblocks
= lbh
->b_size
>> mpd
->inode
->i_blkbits
;
1964 /* check if thereserved journal credits might overflow */
1965 if (!(EXT4_I(mpd
->inode
)->i_flags
& EXT4_EXTENTS_FL
)) {
1966 if (nrblocks
>= EXT4_MAX_TRANS_DATA
) {
1968 * With non-extent format we are limited by the journal
1969 * credit available. Total credit needed to insert
1970 * nrblocks contiguous blocks is dependent on the
1971 * nrblocks. So limit nrblocks.
1974 } else if ((nrblocks
+ (b_size
>> mpd
->inode
->i_blkbits
)) >
1975 EXT4_MAX_TRANS_DATA
) {
1977 * Adding the new buffer_head would make it cross the
1978 * allowed limit for which we have journal credit
1979 * reserved. So limit the new bh->b_size
1981 b_size
= (EXT4_MAX_TRANS_DATA
- nrblocks
) <<
1982 mpd
->inode
->i_blkbits
;
1983 /* we will do mpage_da_submit_io in the next loop */
1987 * First block in the extent
1989 if (lbh
->b_size
== 0) {
1990 lbh
->b_blocknr
= logical
;
1991 lbh
->b_size
= b_size
;
1992 lbh
->b_state
= bh
->b_state
& BH_FLAGS
;
1996 next
= lbh
->b_blocknr
+ nrblocks
;
1998 * Can we merge the block to our big extent?
2000 if (logical
== next
&& (bh
->b_state
& BH_FLAGS
) == lbh
->b_state
) {
2001 lbh
->b_size
+= b_size
;
2007 * We couldn't merge the block to our extent, so we
2008 * need to flush current extent and start new one
2010 if (mpage_da_map_blocks(mpd
) == 0)
2011 mpage_da_submit_io(mpd
);
2017 * __mpage_da_writepage - finds extent of pages and blocks
2019 * @page: page to consider
2020 * @wbc: not used, we just follow rules
2023 * The function finds extents of pages and scan them for all blocks.
2025 static int __mpage_da_writepage(struct page
*page
,
2026 struct writeback_control
*wbc
, void *data
)
2028 struct mpage_da_data
*mpd
= data
;
2029 struct inode
*inode
= mpd
->inode
;
2030 struct buffer_head
*bh
, *head
, fake
;
2035 * Rest of the page in the page_vec
2036 * redirty then and skip then. We will
2037 * try to to write them again after
2038 * starting a new transaction
2040 redirty_page_for_writepage(wbc
, page
);
2042 return MPAGE_DA_EXTENT_TAIL
;
2045 * Can we merge this page to current extent?
2047 if (mpd
->next_page
!= page
->index
) {
2049 * Nope, we can't. So, we map non-allocated blocks
2050 * and start IO on them using writepage()
2052 if (mpd
->next_page
!= mpd
->first_page
) {
2053 if (mpage_da_map_blocks(mpd
) == 0)
2054 mpage_da_submit_io(mpd
);
2056 * skip rest of the page in the page_vec
2059 redirty_page_for_writepage(wbc
, page
);
2061 return MPAGE_DA_EXTENT_TAIL
;
2065 * Start next extent of pages ...
2067 mpd
->first_page
= page
->index
;
2072 mpd
->lbh
.b_size
= 0;
2073 mpd
->lbh
.b_state
= 0;
2074 mpd
->lbh
.b_blocknr
= 0;
2077 mpd
->next_page
= page
->index
+ 1;
2078 logical
= (sector_t
) page
->index
<<
2079 (PAGE_CACHE_SHIFT
- inode
->i_blkbits
);
2081 if (!page_has_buffers(page
)) {
2083 * There is no attached buffer heads yet (mmap?)
2084 * we treat the page asfull of dirty blocks
2087 bh
->b_size
= PAGE_CACHE_SIZE
;
2089 set_buffer_dirty(bh
);
2090 set_buffer_uptodate(bh
);
2091 mpage_add_bh_to_extent(mpd
, logical
, bh
);
2093 return MPAGE_DA_EXTENT_TAIL
;
2096 * Page with regular buffer heads, just add all dirty ones
2098 head
= page_buffers(page
);
2101 BUG_ON(buffer_locked(bh
));
2103 * We need to try to allocate
2104 * unmapped blocks in the same page.
2105 * Otherwise we won't make progress
2106 * with the page in ext4_da_writepage
2108 if (buffer_dirty(bh
) &&
2109 (!buffer_mapped(bh
) || buffer_delay(bh
))) {
2110 mpage_add_bh_to_extent(mpd
, logical
, bh
);
2112 return MPAGE_DA_EXTENT_TAIL
;
2113 } else if (buffer_dirty(bh
) && (buffer_mapped(bh
))) {
2115 * mapped dirty buffer. We need to update
2116 * the b_state because we look at
2117 * b_state in mpage_da_map_blocks. We don't
2118 * update b_size because if we find an
2119 * unmapped buffer_head later we need to
2120 * use the b_state flag of that buffer_head.
2122 if (mpd
->lbh
.b_size
== 0)
2124 bh
->b_state
& BH_FLAGS
;
2127 } while ((bh
= bh
->b_this_page
) != head
);
2134 * mpage_da_writepages - walk the list of dirty pages of the given
2135 * address space, allocates non-allocated blocks, maps newly-allocated
2136 * blocks to existing bhs and issue IO them
2138 * @mapping: address space structure to write
2139 * @wbc: subtract the number of written pages from *@wbc->nr_to_write
2140 * @get_block: the filesystem's block mapper function.
2142 * This is a library function, which implements the writepages()
2143 * address_space_operation.
2145 static int mpage_da_writepages(struct address_space
*mapping
,
2146 struct writeback_control
*wbc
,
2147 struct mpage_da_data
*mpd
)
2151 if (!mpd
->get_block
)
2152 return generic_writepages(mapping
, wbc
);
2154 mpd
->lbh
.b_size
= 0;
2155 mpd
->lbh
.b_state
= 0;
2156 mpd
->lbh
.b_blocknr
= 0;
2157 mpd
->first_page
= 0;
2160 mpd
->pages_written
= 0;
2163 ret
= write_cache_pages(mapping
, wbc
, __mpage_da_writepage
, mpd
);
2165 * Handle last extent of pages
2167 if (!mpd
->io_done
&& mpd
->next_page
!= mpd
->first_page
) {
2168 if (mpage_da_map_blocks(mpd
) == 0)
2169 mpage_da_submit_io(mpd
);
2172 ret
= MPAGE_DA_EXTENT_TAIL
;
2174 wbc
->nr_to_write
-= mpd
->pages_written
;
2179 * this is a special callback for ->write_begin() only
2180 * it's intention is to return mapped block or reserve space
2182 static int ext4_da_get_block_prep(struct inode
*inode
, sector_t iblock
,
2183 struct buffer_head
*bh_result
, int create
)
2187 BUG_ON(create
== 0);
2188 BUG_ON(bh_result
->b_size
!= inode
->i_sb
->s_blocksize
);
2191 * first, we need to know whether the block is allocated already
2192 * preallocated blocks are unmapped but should treated
2193 * the same as allocated blocks.
2195 ret
= ext4_get_blocks_wrap(NULL
, inode
, iblock
, 1, bh_result
, 0, 0, 0);
2196 if ((ret
== 0) && !buffer_delay(bh_result
)) {
2197 /* the block isn't (pre)allocated yet, let's reserve space */
2199 * XXX: __block_prepare_write() unmaps passed block,
2202 ret
= ext4_da_reserve_space(inode
, 1);
2204 /* not enough space to reserve */
2207 map_bh(bh_result
, inode
->i_sb
, 0);
2208 set_buffer_new(bh_result
);
2209 set_buffer_delay(bh_result
);
2210 } else if (ret
> 0) {
2211 bh_result
->b_size
= (ret
<< inode
->i_blkbits
);
2217 #define EXT4_DELALLOC_RSVED 1
2218 static int ext4_da_get_block_write(struct inode
*inode
, sector_t iblock
,
2219 struct buffer_head
*bh_result
, int create
)
2222 unsigned max_blocks
= bh_result
->b_size
>> inode
->i_blkbits
;
2223 loff_t disksize
= EXT4_I(inode
)->i_disksize
;
2224 handle_t
*handle
= NULL
;
2226 handle
= ext4_journal_current_handle();
2228 ret
= ext4_get_blocks_wrap(handle
, inode
, iblock
, max_blocks
,
2229 bh_result
, create
, 0, EXT4_DELALLOC_RSVED
);
2232 bh_result
->b_size
= (ret
<< inode
->i_blkbits
);
2234 if (ext4_should_order_data(inode
)) {
2236 retval
= ext4_jbd2_file_inode(handle
, inode
);
2239 * Failed to add inode for ordered
2240 * mode. Don't update file size
2246 * Update on-disk size along with block allocation
2247 * we don't use 'extend_disksize' as size may change
2248 * within already allocated block -bzzz
2250 disksize
= ((loff_t
) iblock
+ ret
) << inode
->i_blkbits
;
2251 if (disksize
> i_size_read(inode
))
2252 disksize
= i_size_read(inode
);
2253 if (disksize
> EXT4_I(inode
)->i_disksize
) {
2254 ext4_update_i_disksize(inode
, disksize
);
2255 ret
= ext4_mark_inode_dirty(handle
, inode
);
2263 static int ext4_bh_unmapped_or_delay(handle_t
*handle
, struct buffer_head
*bh
)
2266 * unmapped buffer is possible for holes.
2267 * delay buffer is possible with delayed allocation
2269 return ((!buffer_mapped(bh
) || buffer_delay(bh
)) && buffer_dirty(bh
));
2272 static int ext4_normal_get_block_write(struct inode
*inode
, sector_t iblock
,
2273 struct buffer_head
*bh_result
, int create
)
2276 unsigned max_blocks
= bh_result
->b_size
>> inode
->i_blkbits
;
2279 * we don't want to do block allocation in writepage
2280 * so call get_block_wrap with create = 0
2282 ret
= ext4_get_blocks_wrap(NULL
, inode
, iblock
, max_blocks
,
2283 bh_result
, 0, 0, 0);
2285 bh_result
->b_size
= (ret
<< inode
->i_blkbits
);
2292 * get called vi ext4_da_writepages after taking page lock (have journal handle)
2293 * get called via journal_submit_inode_data_buffers (no journal handle)
2294 * get called via shrink_page_list via pdflush (no journal handle)
2295 * or grab_page_cache when doing write_begin (have journal handle)
2297 static int ext4_da_writepage(struct page
*page
,
2298 struct writeback_control
*wbc
)
2303 struct buffer_head
*page_bufs
;
2304 struct inode
*inode
= page
->mapping
->host
;
2306 size
= i_size_read(inode
);
2307 if (page
->index
== size
>> PAGE_CACHE_SHIFT
)
2308 len
= size
& ~PAGE_CACHE_MASK
;
2310 len
= PAGE_CACHE_SIZE
;
2312 if (page_has_buffers(page
)) {
2313 page_bufs
= page_buffers(page
);
2314 if (walk_page_buffers(NULL
, page_bufs
, 0, len
, NULL
,
2315 ext4_bh_unmapped_or_delay
)) {
2317 * We don't want to do block allocation
2318 * So redirty the page and return
2319 * We may reach here when we do a journal commit
2320 * via journal_submit_inode_data_buffers.
2321 * If we don't have mapping block we just ignore
2322 * them. We can also reach here via shrink_page_list
2324 redirty_page_for_writepage(wbc
, page
);
2330 * The test for page_has_buffers() is subtle:
2331 * We know the page is dirty but it lost buffers. That means
2332 * that at some moment in time after write_begin()/write_end()
2333 * has been called all buffers have been clean and thus they
2334 * must have been written at least once. So they are all
2335 * mapped and we can happily proceed with mapping them
2336 * and writing the page.
2338 * Try to initialize the buffer_heads and check whether
2339 * all are mapped and non delay. We don't want to
2340 * do block allocation here.
2342 ret
= block_prepare_write(page
, 0, PAGE_CACHE_SIZE
,
2343 ext4_normal_get_block_write
);
2345 page_bufs
= page_buffers(page
);
2346 /* check whether all are mapped and non delay */
2347 if (walk_page_buffers(NULL
, page_bufs
, 0, len
, NULL
,
2348 ext4_bh_unmapped_or_delay
)) {
2349 redirty_page_for_writepage(wbc
, page
);
2355 * We can't do block allocation here
2356 * so just redity the page and unlock
2359 redirty_page_for_writepage(wbc
, page
);
2363 /* now mark the buffer_heads as dirty and uptodate */
2364 block_commit_write(page
, 0, PAGE_CACHE_SIZE
);
2367 if (test_opt(inode
->i_sb
, NOBH
) && ext4_should_writeback_data(inode
))
2368 ret
= nobh_writepage(page
, ext4_normal_get_block_write
, wbc
);
2370 ret
= block_write_full_page(page
,
2371 ext4_normal_get_block_write
,
2378 * This is called via ext4_da_writepages() to
2379 * calulate the total number of credits to reserve to fit
2380 * a single extent allocation into a single transaction,
2381 * ext4_da_writpeages() will loop calling this before
2382 * the block allocation.
2385 static int ext4_da_writepages_trans_blocks(struct inode
*inode
)
2387 int max_blocks
= EXT4_I(inode
)->i_reserved_data_blocks
;
2390 * With non-extent format the journal credit needed to
2391 * insert nrblocks contiguous block is dependent on
2392 * number of contiguous block. So we will limit
2393 * number of contiguous block to a sane value
2395 if (!(inode
->i_flags
& EXT4_EXTENTS_FL
) &&
2396 (max_blocks
> EXT4_MAX_TRANS_DATA
))
2397 max_blocks
= EXT4_MAX_TRANS_DATA
;
2399 return ext4_chunk_trans_blocks(inode
, max_blocks
);
2402 static int ext4_da_writepages(struct address_space
*mapping
,
2403 struct writeback_control
*wbc
)
2406 int range_whole
= 0;
2407 handle_t
*handle
= NULL
;
2408 struct mpage_da_data mpd
;
2409 struct inode
*inode
= mapping
->host
;
2410 int no_nrwrite_index_update
;
2411 long pages_written
= 0, pages_skipped
;
2412 int needed_blocks
, ret
= 0, nr_to_writebump
= 0;
2413 struct ext4_sb_info
*sbi
= EXT4_SB(mapping
->host
->i_sb
);
2416 * No pages to write? This is mainly a kludge to avoid starting
2417 * a transaction for special inodes like journal inode on last iput()
2418 * because that could violate lock ordering on umount
2420 if (!mapping
->nrpages
|| !mapping_tagged(mapping
, PAGECACHE_TAG_DIRTY
))
2424 * If the filesystem has aborted, it is read-only, so return
2425 * right away instead of dumping stack traces later on that
2426 * will obscure the real source of the problem. We test
2427 * EXT4_MOUNT_ABORT instead of sb->s_flag's MS_RDONLY because
2428 * the latter could be true if the filesystem is mounted
2429 * read-only, and in that case, ext4_da_writepages should
2430 * *never* be called, so if that ever happens, we would want
2433 if (unlikely(sbi
->s_mount_opt
& EXT4_MOUNT_ABORT
))
2437 * Make sure nr_to_write is >= sbi->s_mb_stream_request
2438 * This make sure small files blocks are allocated in
2439 * single attempt. This ensure that small files
2440 * get less fragmented.
2442 if (wbc
->nr_to_write
< sbi
->s_mb_stream_request
) {
2443 nr_to_writebump
= sbi
->s_mb_stream_request
- wbc
->nr_to_write
;
2444 wbc
->nr_to_write
= sbi
->s_mb_stream_request
;
2446 if (wbc
->range_start
== 0 && wbc
->range_end
== LLONG_MAX
)
2449 if (wbc
->range_cyclic
)
2450 index
= mapping
->writeback_index
;
2452 index
= wbc
->range_start
>> PAGE_CACHE_SHIFT
;
2455 mpd
.inode
= mapping
->host
;
2458 * we don't want write_cache_pages to update
2459 * nr_to_write and writeback_index
2461 no_nrwrite_index_update
= wbc
->no_nrwrite_index_update
;
2462 wbc
->no_nrwrite_index_update
= 1;
2463 pages_skipped
= wbc
->pages_skipped
;
2465 while (!ret
&& wbc
->nr_to_write
> 0) {
2468 * we insert one extent at a time. So we need
2469 * credit needed for single extent allocation.
2470 * journalled mode is currently not supported
2473 BUG_ON(ext4_should_journal_data(inode
));
2474 needed_blocks
= ext4_da_writepages_trans_blocks(inode
);
2476 /* start a new transaction*/
2477 handle
= ext4_journal_start(inode
, needed_blocks
);
2478 if (IS_ERR(handle
)) {
2479 ret
= PTR_ERR(handle
);
2480 printk(KERN_CRIT
"%s: jbd2_start: "
2481 "%ld pages, ino %lu; err %d\n", __func__
,
2482 wbc
->nr_to_write
, inode
->i_ino
, ret
);
2484 goto out_writepages
;
2486 mpd
.get_block
= ext4_da_get_block_write
;
2487 ret
= mpage_da_writepages(mapping
, wbc
, &mpd
);
2489 ext4_journal_stop(handle
);
2491 if (mpd
.retval
== -ENOSPC
) {
2492 /* commit the transaction which would
2493 * free blocks released in the transaction
2496 jbd2_journal_force_commit_nested(sbi
->s_journal
);
2497 wbc
->pages_skipped
= pages_skipped
;
2499 } else if (ret
== MPAGE_DA_EXTENT_TAIL
) {
2501 * got one extent now try with
2504 pages_written
+= mpd
.pages_written
;
2505 wbc
->pages_skipped
= pages_skipped
;
2507 } else if (wbc
->nr_to_write
)
2509 * There is no more writeout needed
2510 * or we requested for a noblocking writeout
2511 * and we found the device congested
2515 if (pages_skipped
!= wbc
->pages_skipped
)
2516 printk(KERN_EMERG
"This should not happen leaving %s "
2517 "with nr_to_write = %ld ret = %d\n",
2518 __func__
, wbc
->nr_to_write
, ret
);
2521 index
+= pages_written
;
2522 if (wbc
->range_cyclic
|| (range_whole
&& wbc
->nr_to_write
> 0))
2524 * set the writeback_index so that range_cyclic
2525 * mode will write it back later
2527 mapping
->writeback_index
= index
;
2530 if (!no_nrwrite_index_update
)
2531 wbc
->no_nrwrite_index_update
= 0;
2532 wbc
->nr_to_write
-= nr_to_writebump
;
2536 #define FALL_BACK_TO_NONDELALLOC 1
2537 static int ext4_nonda_switch(struct super_block
*sb
)
2539 s64 free_blocks
, dirty_blocks
;
2540 struct ext4_sb_info
*sbi
= EXT4_SB(sb
);
2543 * switch to non delalloc mode if we are running low
2544 * on free block. The free block accounting via percpu
2545 * counters can get slightly wrong with FBC_BATCH getting
2546 * accumulated on each CPU without updating global counters
2547 * Delalloc need an accurate free block accounting. So switch
2548 * to non delalloc when we are near to error range.
2550 free_blocks
= percpu_counter_read_positive(&sbi
->s_freeblocks_counter
);
2551 dirty_blocks
= percpu_counter_read_positive(&sbi
->s_dirtyblocks_counter
);
2552 if (2 * free_blocks
< 3 * dirty_blocks
||
2553 free_blocks
< (dirty_blocks
+ EXT4_FREEBLOCKS_WATERMARK
)) {
2555 * free block count is less that 150% of dirty blocks
2556 * or free blocks is less that watermark
2563 static int ext4_da_write_begin(struct file
*file
, struct address_space
*mapping
,
2564 loff_t pos
, unsigned len
, unsigned flags
,
2565 struct page
**pagep
, void **fsdata
)
2567 int ret
, retries
= 0;
2571 struct inode
*inode
= mapping
->host
;
2574 index
= pos
>> PAGE_CACHE_SHIFT
;
2575 from
= pos
& (PAGE_CACHE_SIZE
- 1);
2578 if (ext4_nonda_switch(inode
->i_sb
)) {
2579 *fsdata
= (void *)FALL_BACK_TO_NONDELALLOC
;
2580 return ext4_write_begin(file
, mapping
, pos
,
2581 len
, flags
, pagep
, fsdata
);
2583 *fsdata
= (void *)0;
2586 * With delayed allocation, we don't log the i_disksize update
2587 * if there is delayed block allocation. But we still need
2588 * to journalling the i_disksize update if writes to the end
2589 * of file which has an already mapped buffer.
2591 handle
= ext4_journal_start(inode
, 1);
2592 if (IS_ERR(handle
)) {
2593 ret
= PTR_ERR(handle
);
2597 page
= grab_cache_page_write_begin(mapping
, index
, flags
);
2599 ext4_journal_stop(handle
);
2605 ret
= block_write_begin(file
, mapping
, pos
, len
, flags
, pagep
, fsdata
,
2606 ext4_da_get_block_prep
);
2609 ext4_journal_stop(handle
);
2610 page_cache_release(page
);
2612 * block_write_begin may have instantiated a few blocks
2613 * outside i_size. Trim these off again. Don't need
2614 * i_size_read because we hold i_mutex.
2616 if (pos
+ len
> inode
->i_size
)
2617 vmtruncate(inode
, inode
->i_size
);
2620 if (ret
== -ENOSPC
&& ext4_should_retry_alloc(inode
->i_sb
, &retries
))
2627 * Check if we should update i_disksize
2628 * when write to the end of file but not require block allocation
2630 static int ext4_da_should_update_i_disksize(struct page
*page
,
2631 unsigned long offset
)
2633 struct buffer_head
*bh
;
2634 struct inode
*inode
= page
->mapping
->host
;
2638 bh
= page_buffers(page
);
2639 idx
= offset
>> inode
->i_blkbits
;
2641 for (i
= 0; i
< idx
; i
++)
2642 bh
= bh
->b_this_page
;
2644 if (!buffer_mapped(bh
) || (buffer_delay(bh
)))
2649 static int ext4_da_write_end(struct file
*file
,
2650 struct address_space
*mapping
,
2651 loff_t pos
, unsigned len
, unsigned copied
,
2652 struct page
*page
, void *fsdata
)
2654 struct inode
*inode
= mapping
->host
;
2656 handle_t
*handle
= ext4_journal_current_handle();
2658 unsigned long start
, end
;
2659 int write_mode
= (int)(unsigned long)fsdata
;
2661 if (write_mode
== FALL_BACK_TO_NONDELALLOC
) {
2662 if (ext4_should_order_data(inode
)) {
2663 return ext4_ordered_write_end(file
, mapping
, pos
,
2664 len
, copied
, page
, fsdata
);
2665 } else if (ext4_should_writeback_data(inode
)) {
2666 return ext4_writeback_write_end(file
, mapping
, pos
,
2667 len
, copied
, page
, fsdata
);
2673 start
= pos
& (PAGE_CACHE_SIZE
- 1);
2674 end
= start
+ copied
- 1;
2677 * generic_write_end() will run mark_inode_dirty() if i_size
2678 * changes. So let's piggyback the i_disksize mark_inode_dirty
2682 new_i_size
= pos
+ copied
;
2683 if (new_i_size
> EXT4_I(inode
)->i_disksize
) {
2684 if (ext4_da_should_update_i_disksize(page
, end
)) {
2685 down_write(&EXT4_I(inode
)->i_data_sem
);
2686 if (new_i_size
> EXT4_I(inode
)->i_disksize
) {
2688 * Updating i_disksize when extending file
2689 * without needing block allocation
2691 if (ext4_should_order_data(inode
))
2692 ret
= ext4_jbd2_file_inode(handle
,
2695 EXT4_I(inode
)->i_disksize
= new_i_size
;
2697 up_write(&EXT4_I(inode
)->i_data_sem
);
2698 /* We need to mark inode dirty even if
2699 * new_i_size is less that inode->i_size
2700 * bu greater than i_disksize.(hint delalloc)
2702 ext4_mark_inode_dirty(handle
, inode
);
2705 ret2
= generic_write_end(file
, mapping
, pos
, len
, copied
,
2710 ret2
= ext4_journal_stop(handle
);
2714 return ret
? ret
: copied
;
2717 static void ext4_da_invalidatepage(struct page
*page
, unsigned long offset
)
2720 * Drop reserved blocks
2722 BUG_ON(!PageLocked(page
));
2723 if (!page_has_buffers(page
))
2726 ext4_da_page_release_reservation(page
, offset
);
2729 ext4_invalidatepage(page
, offset
);
2736 * bmap() is special. It gets used by applications such as lilo and by
2737 * the swapper to find the on-disk block of a specific piece of data.
2739 * Naturally, this is dangerous if the block concerned is still in the
2740 * journal. If somebody makes a swapfile on an ext4 data-journaling
2741 * filesystem and enables swap, then they may get a nasty shock when the
2742 * data getting swapped to that swapfile suddenly gets overwritten by
2743 * the original zero's written out previously to the journal and
2744 * awaiting writeback in the kernel's buffer cache.
2746 * So, if we see any bmap calls here on a modified, data-journaled file,
2747 * take extra steps to flush any blocks which might be in the cache.
2749 static sector_t
ext4_bmap(struct address_space
*mapping
, sector_t block
)
2751 struct inode
*inode
= mapping
->host
;
2755 if (mapping_tagged(mapping
, PAGECACHE_TAG_DIRTY
) &&
2756 test_opt(inode
->i_sb
, DELALLOC
)) {
2758 * With delalloc we want to sync the file
2759 * so that we can make sure we allocate
2762 filemap_write_and_wait(mapping
);
2765 if (EXT4_I(inode
)->i_state
& EXT4_STATE_JDATA
) {
2767 * This is a REALLY heavyweight approach, but the use of
2768 * bmap on dirty files is expected to be extremely rare:
2769 * only if we run lilo or swapon on a freshly made file
2770 * do we expect this to happen.
2772 * (bmap requires CAP_SYS_RAWIO so this does not
2773 * represent an unprivileged user DOS attack --- we'd be
2774 * in trouble if mortal users could trigger this path at
2777 * NB. EXT4_STATE_JDATA is not set on files other than
2778 * regular files. If somebody wants to bmap a directory
2779 * or symlink and gets confused because the buffer
2780 * hasn't yet been flushed to disk, they deserve
2781 * everything they get.
2784 EXT4_I(inode
)->i_state
&= ~EXT4_STATE_JDATA
;
2785 journal
= EXT4_JOURNAL(inode
);
2786 jbd2_journal_lock_updates(journal
);
2787 err
= jbd2_journal_flush(journal
);
2788 jbd2_journal_unlock_updates(journal
);
2794 return generic_block_bmap(mapping
, block
, ext4_get_block
);
2797 static int bget_one(handle_t
*handle
, struct buffer_head
*bh
)
2803 static int bput_one(handle_t
*handle
, struct buffer_head
*bh
)
2810 * Note that we don't need to start a transaction unless we're journaling data
2811 * because we should have holes filled from ext4_page_mkwrite(). We even don't
2812 * need to file the inode to the transaction's list in ordered mode because if
2813 * we are writing back data added by write(), the inode is already there and if
2814 * we are writing back data modified via mmap(), noone guarantees in which
2815 * transaction the data will hit the disk. In case we are journaling data, we
2816 * cannot start transaction directly because transaction start ranks above page
2817 * lock so we have to do some magic.
2819 * In all journaling modes block_write_full_page() will start the I/O.
2823 * ext4_writepage() -> kmalloc() -> __alloc_pages() -> page_launder() ->
2828 * ext4_file_write() -> generic_file_write() -> __alloc_pages() -> ...
2830 * Same applies to ext4_get_block(). We will deadlock on various things like
2831 * lock_journal and i_data_sem
2833 * Setting PF_MEMALLOC here doesn't work - too many internal memory
2836 * 16May01: If we're reentered then journal_current_handle() will be
2837 * non-zero. We simply *return*.
2839 * 1 July 2001: @@@ FIXME:
2840 * In journalled data mode, a data buffer may be metadata against the
2841 * current transaction. But the same file is part of a shared mapping
2842 * and someone does a writepage() on it.
2844 * We will move the buffer onto the async_data list, but *after* it has
2845 * been dirtied. So there's a small window where we have dirty data on
2848 * Note that this only applies to the last partial page in the file. The
2849 * bit which block_write_full_page() uses prepare/commit for. (That's
2850 * broken code anyway: it's wrong for msync()).
2852 * It's a rare case: affects the final partial page, for journalled data
2853 * where the file is subject to bith write() and writepage() in the same
2854 * transction. To fix it we'll need a custom block_write_full_page().
2855 * We'll probably need that anyway for journalling writepage() output.
2857 * We don't honour synchronous mounts for writepage(). That would be
2858 * disastrous. Any write() or metadata operation will sync the fs for
2862 static int __ext4_normal_writepage(struct page
*page
,
2863 struct writeback_control
*wbc
)
2865 struct inode
*inode
= page
->mapping
->host
;
2867 if (test_opt(inode
->i_sb
, NOBH
))
2868 return nobh_writepage(page
,
2869 ext4_normal_get_block_write
, wbc
);
2871 return block_write_full_page(page
,
2872 ext4_normal_get_block_write
,
2876 static int ext4_normal_writepage(struct page
*page
,
2877 struct writeback_control
*wbc
)
2879 struct inode
*inode
= page
->mapping
->host
;
2880 loff_t size
= i_size_read(inode
);
2883 J_ASSERT(PageLocked(page
));
2884 if (page
->index
== size
>> PAGE_CACHE_SHIFT
)
2885 len
= size
& ~PAGE_CACHE_MASK
;
2887 len
= PAGE_CACHE_SIZE
;
2889 if (page_has_buffers(page
)) {
2890 /* if page has buffers it should all be mapped
2891 * and allocated. If there are not buffers attached
2892 * to the page we know the page is dirty but it lost
2893 * buffers. That means that at some moment in time
2894 * after write_begin() / write_end() has been called
2895 * all buffers have been clean and thus they must have been
2896 * written at least once. So they are all mapped and we can
2897 * happily proceed with mapping them and writing the page.
2899 BUG_ON(walk_page_buffers(NULL
, page_buffers(page
), 0, len
, NULL
,
2900 ext4_bh_unmapped_or_delay
));
2903 if (!ext4_journal_current_handle())
2904 return __ext4_normal_writepage(page
, wbc
);
2906 redirty_page_for_writepage(wbc
, page
);
2911 static int __ext4_journalled_writepage(struct page
*page
,
2912 struct writeback_control
*wbc
)
2914 struct address_space
*mapping
= page
->mapping
;
2915 struct inode
*inode
= mapping
->host
;
2916 struct buffer_head
*page_bufs
;
2917 handle_t
*handle
= NULL
;
2921 ret
= block_prepare_write(page
, 0, PAGE_CACHE_SIZE
,
2922 ext4_normal_get_block_write
);
2926 page_bufs
= page_buffers(page
);
2927 walk_page_buffers(handle
, page_bufs
, 0, PAGE_CACHE_SIZE
, NULL
,
2929 /* As soon as we unlock the page, it can go away, but we have
2930 * references to buffers so we are safe */
2933 handle
= ext4_journal_start(inode
, ext4_writepage_trans_blocks(inode
));
2934 if (IS_ERR(handle
)) {
2935 ret
= PTR_ERR(handle
);
2939 ret
= walk_page_buffers(handle
, page_bufs
, 0,
2940 PAGE_CACHE_SIZE
, NULL
, do_journal_get_write_access
);
2942 err
= walk_page_buffers(handle
, page_bufs
, 0,
2943 PAGE_CACHE_SIZE
, NULL
, write_end_fn
);
2946 err
= ext4_journal_stop(handle
);
2950 walk_page_buffers(handle
, page_bufs
, 0,
2951 PAGE_CACHE_SIZE
, NULL
, bput_one
);
2952 EXT4_I(inode
)->i_state
|= EXT4_STATE_JDATA
;
2961 static int ext4_journalled_writepage(struct page
*page
,
2962 struct writeback_control
*wbc
)
2964 struct inode
*inode
= page
->mapping
->host
;
2965 loff_t size
= i_size_read(inode
);
2968 J_ASSERT(PageLocked(page
));
2969 if (page
->index
== size
>> PAGE_CACHE_SHIFT
)
2970 len
= size
& ~PAGE_CACHE_MASK
;
2972 len
= PAGE_CACHE_SIZE
;
2974 if (page_has_buffers(page
)) {
2975 /* if page has buffers it should all be mapped
2976 * and allocated. If there are not buffers attached
2977 * to the page we know the page is dirty but it lost
2978 * buffers. That means that at some moment in time
2979 * after write_begin() / write_end() has been called
2980 * all buffers have been clean and thus they must have been
2981 * written at least once. So they are all mapped and we can
2982 * happily proceed with mapping them and writing the page.
2984 BUG_ON(walk_page_buffers(NULL
, page_buffers(page
), 0, len
, NULL
,
2985 ext4_bh_unmapped_or_delay
));
2988 if (ext4_journal_current_handle())
2991 if (PageChecked(page
)) {
2993 * It's mmapped pagecache. Add buffers and journal it. There
2994 * doesn't seem much point in redirtying the page here.
2996 ClearPageChecked(page
);
2997 return __ext4_journalled_writepage(page
, wbc
);
3000 * It may be a page full of checkpoint-mode buffers. We don't
3001 * really know unless we go poke around in the buffer_heads.
3002 * But block_write_full_page will do the right thing.
3004 return block_write_full_page(page
,
3005 ext4_normal_get_block_write
,
3009 redirty_page_for_writepage(wbc
, page
);
3014 static int ext4_readpage(struct file
*file
, struct page
*page
)
3016 return mpage_readpage(page
, ext4_get_block
);
3020 ext4_readpages(struct file
*file
, struct address_space
*mapping
,
3021 struct list_head
*pages
, unsigned nr_pages
)
3023 return mpage_readpages(mapping
, pages
, nr_pages
, ext4_get_block
);
3026 static void ext4_invalidatepage(struct page
*page
, unsigned long offset
)
3028 journal_t
*journal
= EXT4_JOURNAL(page
->mapping
->host
);
3031 * If it's a full truncate we just forget about the pending dirtying
3034 ClearPageChecked(page
);
3036 jbd2_journal_invalidatepage(journal
, page
, offset
);
3039 static int ext4_releasepage(struct page
*page
, gfp_t wait
)
3041 journal_t
*journal
= EXT4_JOURNAL(page
->mapping
->host
);
3043 WARN_ON(PageChecked(page
));
3044 if (!page_has_buffers(page
))
3046 return jbd2_journal_try_to_free_buffers(journal
, page
, wait
);
3050 * If the O_DIRECT write will extend the file then add this inode to the
3051 * orphan list. So recovery will truncate it back to the original size
3052 * if the machine crashes during the write.
3054 * If the O_DIRECT write is intantiating holes inside i_size and the machine
3055 * crashes then stale disk data _may_ be exposed inside the file. But current
3056 * VFS code falls back into buffered path in that case so we are safe.
3058 static ssize_t
ext4_direct_IO(int rw
, struct kiocb
*iocb
,
3059 const struct iovec
*iov
, loff_t offset
,
3060 unsigned long nr_segs
)
3062 struct file
*file
= iocb
->ki_filp
;
3063 struct inode
*inode
= file
->f_mapping
->host
;
3064 struct ext4_inode_info
*ei
= EXT4_I(inode
);
3068 size_t count
= iov_length(iov
, nr_segs
);
3071 loff_t final_size
= offset
+ count
;
3073 if (final_size
> inode
->i_size
) {
3074 /* Credits for sb + inode write */
3075 handle
= ext4_journal_start(inode
, 2);
3076 if (IS_ERR(handle
)) {
3077 ret
= PTR_ERR(handle
);
3080 ret
= ext4_orphan_add(handle
, inode
);
3082 ext4_journal_stop(handle
);
3086 ei
->i_disksize
= inode
->i_size
;
3087 ext4_journal_stop(handle
);
3091 ret
= blockdev_direct_IO(rw
, iocb
, inode
, inode
->i_sb
->s_bdev
, iov
,
3093 ext4_get_block
, NULL
);
3098 /* Credits for sb + inode write */
3099 handle
= ext4_journal_start(inode
, 2);
3100 if (IS_ERR(handle
)) {
3101 /* This is really bad luck. We've written the data
3102 * but cannot extend i_size. Bail out and pretend
3103 * the write failed... */
3104 ret
= PTR_ERR(handle
);
3108 ext4_orphan_del(handle
, inode
);
3110 loff_t end
= offset
+ ret
;
3111 if (end
> inode
->i_size
) {
3112 ei
->i_disksize
= end
;
3113 i_size_write(inode
, end
);
3115 * We're going to return a positive `ret'
3116 * here due to non-zero-length I/O, so there's
3117 * no way of reporting error returns from
3118 * ext4_mark_inode_dirty() to userspace. So
3121 ext4_mark_inode_dirty(handle
, inode
);
3124 err
= ext4_journal_stop(handle
);
3133 * Pages can be marked dirty completely asynchronously from ext4's journalling
3134 * activity. By filemap_sync_pte(), try_to_unmap_one(), etc. We cannot do
3135 * much here because ->set_page_dirty is called under VFS locks. The page is
3136 * not necessarily locked.
3138 * We cannot just dirty the page and leave attached buffers clean, because the
3139 * buffers' dirty state is "definitive". We cannot just set the buffers dirty
3140 * or jbddirty because all the journalling code will explode.
3142 * So what we do is to mark the page "pending dirty" and next time writepage
3143 * is called, propagate that into the buffers appropriately.
3145 static int ext4_journalled_set_page_dirty(struct page
*page
)
3147 SetPageChecked(page
);
3148 return __set_page_dirty_nobuffers(page
);
3151 static const struct address_space_operations ext4_ordered_aops
= {
3152 .readpage
= ext4_readpage
,
3153 .readpages
= ext4_readpages
,
3154 .writepage
= ext4_normal_writepage
,
3155 .sync_page
= block_sync_page
,
3156 .write_begin
= ext4_write_begin
,
3157 .write_end
= ext4_ordered_write_end
,
3159 .invalidatepage
= ext4_invalidatepage
,
3160 .releasepage
= ext4_releasepage
,
3161 .direct_IO
= ext4_direct_IO
,
3162 .migratepage
= buffer_migrate_page
,
3163 .is_partially_uptodate
= block_is_partially_uptodate
,
3166 static const struct address_space_operations ext4_writeback_aops
= {
3167 .readpage
= ext4_readpage
,
3168 .readpages
= ext4_readpages
,
3169 .writepage
= ext4_normal_writepage
,
3170 .sync_page
= block_sync_page
,
3171 .write_begin
= ext4_write_begin
,
3172 .write_end
= ext4_writeback_write_end
,
3174 .invalidatepage
= ext4_invalidatepage
,
3175 .releasepage
= ext4_releasepage
,
3176 .direct_IO
= ext4_direct_IO
,
3177 .migratepage
= buffer_migrate_page
,
3178 .is_partially_uptodate
= block_is_partially_uptodate
,
3181 static const struct address_space_operations ext4_journalled_aops
= {
3182 .readpage
= ext4_readpage
,
3183 .readpages
= ext4_readpages
,
3184 .writepage
= ext4_journalled_writepage
,
3185 .sync_page
= block_sync_page
,
3186 .write_begin
= ext4_write_begin
,
3187 .write_end
= ext4_journalled_write_end
,
3188 .set_page_dirty
= ext4_journalled_set_page_dirty
,
3190 .invalidatepage
= ext4_invalidatepage
,
3191 .releasepage
= ext4_releasepage
,
3192 .is_partially_uptodate
= block_is_partially_uptodate
,
3195 static const struct address_space_operations ext4_da_aops
= {
3196 .readpage
= ext4_readpage
,
3197 .readpages
= ext4_readpages
,
3198 .writepage
= ext4_da_writepage
,
3199 .writepages
= ext4_da_writepages
,
3200 .sync_page
= block_sync_page
,
3201 .write_begin
= ext4_da_write_begin
,
3202 .write_end
= ext4_da_write_end
,
3204 .invalidatepage
= ext4_da_invalidatepage
,
3205 .releasepage
= ext4_releasepage
,
3206 .direct_IO
= ext4_direct_IO
,
3207 .migratepage
= buffer_migrate_page
,
3208 .is_partially_uptodate
= block_is_partially_uptodate
,
3211 void ext4_set_aops(struct inode
*inode
)
3213 if (ext4_should_order_data(inode
) &&
3214 test_opt(inode
->i_sb
, DELALLOC
))
3215 inode
->i_mapping
->a_ops
= &ext4_da_aops
;
3216 else if (ext4_should_order_data(inode
))
3217 inode
->i_mapping
->a_ops
= &ext4_ordered_aops
;
3218 else if (ext4_should_writeback_data(inode
) &&
3219 test_opt(inode
->i_sb
, DELALLOC
))
3220 inode
->i_mapping
->a_ops
= &ext4_da_aops
;
3221 else if (ext4_should_writeback_data(inode
))
3222 inode
->i_mapping
->a_ops
= &ext4_writeback_aops
;
3224 inode
->i_mapping
->a_ops
= &ext4_journalled_aops
;
3228 * ext4_block_truncate_page() zeroes out a mapping from file offset `from'
3229 * up to the end of the block which corresponds to `from'.
3230 * This required during truncate. We need to physically zero the tail end
3231 * of that block so it doesn't yield old data if the file is later grown.
3233 int ext4_block_truncate_page(handle_t
*handle
,
3234 struct address_space
*mapping
, loff_t from
)
3236 ext4_fsblk_t index
= from
>> PAGE_CACHE_SHIFT
;
3237 unsigned offset
= from
& (PAGE_CACHE_SIZE
-1);
3238 unsigned blocksize
, length
, pos
;
3240 struct inode
*inode
= mapping
->host
;
3241 struct buffer_head
*bh
;
3245 page
= grab_cache_page(mapping
, from
>> PAGE_CACHE_SHIFT
);
3249 blocksize
= inode
->i_sb
->s_blocksize
;
3250 length
= blocksize
- (offset
& (blocksize
- 1));
3251 iblock
= index
<< (PAGE_CACHE_SHIFT
- inode
->i_sb
->s_blocksize_bits
);
3254 * For "nobh" option, we can only work if we don't need to
3255 * read-in the page - otherwise we create buffers to do the IO.
3257 if (!page_has_buffers(page
) && test_opt(inode
->i_sb
, NOBH
) &&
3258 ext4_should_writeback_data(inode
) && PageUptodate(page
)) {
3259 zero_user(page
, offset
, length
);
3260 set_page_dirty(page
);
3264 if (!page_has_buffers(page
))
3265 create_empty_buffers(page
, blocksize
, 0);
3267 /* Find the buffer that contains "offset" */
3268 bh
= page_buffers(page
);
3270 while (offset
>= pos
) {
3271 bh
= bh
->b_this_page
;
3277 if (buffer_freed(bh
)) {
3278 BUFFER_TRACE(bh
, "freed: skip");
3282 if (!buffer_mapped(bh
)) {
3283 BUFFER_TRACE(bh
, "unmapped");
3284 ext4_get_block(inode
, iblock
, bh
, 0);
3285 /* unmapped? It's a hole - nothing to do */
3286 if (!buffer_mapped(bh
)) {
3287 BUFFER_TRACE(bh
, "still unmapped");
3292 /* Ok, it's mapped. Make sure it's up-to-date */
3293 if (PageUptodate(page
))
3294 set_buffer_uptodate(bh
);
3296 if (!buffer_uptodate(bh
)) {
3298 ll_rw_block(READ
, 1, &bh
);
3300 /* Uhhuh. Read error. Complain and punt. */
3301 if (!buffer_uptodate(bh
))
3305 if (ext4_should_journal_data(inode
)) {
3306 BUFFER_TRACE(bh
, "get write access");
3307 err
= ext4_journal_get_write_access(handle
, bh
);
3312 zero_user(page
, offset
, length
);
3314 BUFFER_TRACE(bh
, "zeroed end of block");
3317 if (ext4_should_journal_data(inode
)) {
3318 err
= ext4_journal_dirty_metadata(handle
, bh
);
3320 if (ext4_should_order_data(inode
))
3321 err
= ext4_jbd2_file_inode(handle
, inode
);
3322 mark_buffer_dirty(bh
);
3327 page_cache_release(page
);
3332 * Probably it should be a library function... search for first non-zero word
3333 * or memcmp with zero_page, whatever is better for particular architecture.
3336 static inline int all_zeroes(__le32
*p
, __le32
*q
)
3345 * ext4_find_shared - find the indirect blocks for partial truncation.
3346 * @inode: inode in question
3347 * @depth: depth of the affected branch
3348 * @offsets: offsets of pointers in that branch (see ext4_block_to_path)
3349 * @chain: place to store the pointers to partial indirect blocks
3350 * @top: place to the (detached) top of branch
3352 * This is a helper function used by ext4_truncate().
3354 * When we do truncate() we may have to clean the ends of several
3355 * indirect blocks but leave the blocks themselves alive. Block is
3356 * partially truncated if some data below the new i_size is refered
3357 * from it (and it is on the path to the first completely truncated
3358 * data block, indeed). We have to free the top of that path along
3359 * with everything to the right of the path. Since no allocation
3360 * past the truncation point is possible until ext4_truncate()
3361 * finishes, we may safely do the latter, but top of branch may
3362 * require special attention - pageout below the truncation point
3363 * might try to populate it.
3365 * We atomically detach the top of branch from the tree, store the
3366 * block number of its root in *@top, pointers to buffer_heads of
3367 * partially truncated blocks - in @chain[].bh and pointers to
3368 * their last elements that should not be removed - in
3369 * @chain[].p. Return value is the pointer to last filled element
3372 * The work left to caller to do the actual freeing of subtrees:
3373 * a) free the subtree starting from *@top
3374 * b) free the subtrees whose roots are stored in
3375 * (@chain[i].p+1 .. end of @chain[i].bh->b_data)
3376 * c) free the subtrees growing from the inode past the @chain[0].
3377 * (no partially truncated stuff there). */
3379 static Indirect
*ext4_find_shared(struct inode
*inode
, int depth
,
3380 ext4_lblk_t offsets
[4], Indirect chain
[4], __le32
*top
)
3382 Indirect
*partial
, *p
;
3386 /* Make k index the deepest non-null offest + 1 */
3387 for (k
= depth
; k
> 1 && !offsets
[k
-1]; k
--)
3389 partial
= ext4_get_branch(inode
, k
, offsets
, chain
, &err
);
3390 /* Writer: pointers */
3392 partial
= chain
+ k
-1;
3394 * If the branch acquired continuation since we've looked at it -
3395 * fine, it should all survive and (new) top doesn't belong to us.
3397 if (!partial
->key
&& *partial
->p
)
3400 for (p
= partial
; (p
> chain
) && all_zeroes((__le32
*) p
->bh
->b_data
, p
->p
); p
--)
3403 * OK, we've found the last block that must survive. The rest of our
3404 * branch should be detached before unlocking. However, if that rest
3405 * of branch is all ours and does not grow immediately from the inode
3406 * it's easier to cheat and just decrement partial->p.
3408 if (p
== chain
+ k
- 1 && p
> chain
) {
3412 /* Nope, don't do this in ext4. Must leave the tree intact */
3419 while (partial
> p
) {
3420 brelse(partial
->bh
);
3428 * Zero a number of block pointers in either an inode or an indirect block.
3429 * If we restart the transaction we must again get write access to the
3430 * indirect block for further modification.
3432 * We release `count' blocks on disk, but (last - first) may be greater
3433 * than `count' because there can be holes in there.
3435 static void ext4_clear_blocks(handle_t
*handle
, struct inode
*inode
,
3436 struct buffer_head
*bh
, ext4_fsblk_t block_to_free
,
3437 unsigned long count
, __le32
*first
, __le32
*last
)
3440 if (try_to_extend_transaction(handle
, inode
)) {
3442 BUFFER_TRACE(bh
, "call ext4_journal_dirty_metadata");
3443 ext4_journal_dirty_metadata(handle
, bh
);
3445 ext4_mark_inode_dirty(handle
, inode
);
3446 ext4_journal_test_restart(handle
, inode
);
3448 BUFFER_TRACE(bh
, "retaking write access");
3449 ext4_journal_get_write_access(handle
, bh
);
3454 * Any buffers which are on the journal will be in memory. We find
3455 * them on the hash table so jbd2_journal_revoke() will run jbd2_journal_forget()
3456 * on them. We've already detached each block from the file, so
3457 * bforget() in jbd2_journal_forget() should be safe.
3459 * AKPM: turn on bforget in jbd2_journal_forget()!!!
3461 for (p
= first
; p
< last
; p
++) {
3462 u32 nr
= le32_to_cpu(*p
);
3464 struct buffer_head
*tbh
;
3467 tbh
= sb_find_get_block(inode
->i_sb
, nr
);
3468 ext4_forget(handle
, 0, inode
, tbh
, nr
);
3472 ext4_free_blocks(handle
, inode
, block_to_free
, count
, 0);
3476 * ext4_free_data - free a list of data blocks
3477 * @handle: handle for this transaction
3478 * @inode: inode we are dealing with
3479 * @this_bh: indirect buffer_head which contains *@first and *@last
3480 * @first: array of block numbers
3481 * @last: points immediately past the end of array
3483 * We are freeing all blocks refered from that array (numbers are stored as
3484 * little-endian 32-bit) and updating @inode->i_blocks appropriately.
3486 * We accumulate contiguous runs of blocks to free. Conveniently, if these
3487 * blocks are contiguous then releasing them at one time will only affect one
3488 * or two bitmap blocks (+ group descriptor(s) and superblock) and we won't
3489 * actually use a lot of journal space.
3491 * @this_bh will be %NULL if @first and @last point into the inode's direct
3494 static void ext4_free_data(handle_t
*handle
, struct inode
*inode
,
3495 struct buffer_head
*this_bh
,
3496 __le32
*first
, __le32
*last
)
3498 ext4_fsblk_t block_to_free
= 0; /* Starting block # of a run */
3499 unsigned long count
= 0; /* Number of blocks in the run */
3500 __le32
*block_to_free_p
= NULL
; /* Pointer into inode/ind
3503 ext4_fsblk_t nr
; /* Current block # */
3504 __le32
*p
; /* Pointer into inode/ind
3505 for current block */
3508 if (this_bh
) { /* For indirect block */
3509 BUFFER_TRACE(this_bh
, "get_write_access");
3510 err
= ext4_journal_get_write_access(handle
, this_bh
);
3511 /* Important: if we can't update the indirect pointers
3512 * to the blocks, we can't free them. */
3517 for (p
= first
; p
< last
; p
++) {
3518 nr
= le32_to_cpu(*p
);
3520 /* accumulate blocks to free if they're contiguous */
3523 block_to_free_p
= p
;
3525 } else if (nr
== block_to_free
+ count
) {
3528 ext4_clear_blocks(handle
, inode
, this_bh
,
3530 count
, block_to_free_p
, p
);
3532 block_to_free_p
= p
;
3539 ext4_clear_blocks(handle
, inode
, this_bh
, block_to_free
,
3540 count
, block_to_free_p
, p
);
3543 BUFFER_TRACE(this_bh
, "call ext4_journal_dirty_metadata");
3546 * The buffer head should have an attached journal head at this
3547 * point. However, if the data is corrupted and an indirect
3548 * block pointed to itself, it would have been detached when
3549 * the block was cleared. Check for this instead of OOPSing.
3552 ext4_journal_dirty_metadata(handle
, this_bh
);
3554 ext4_error(inode
->i_sb
, __func__
,
3555 "circular indirect block detected, "
3556 "inode=%lu, block=%llu",
3558 (unsigned long long) this_bh
->b_blocknr
);
3563 * ext4_free_branches - free an array of branches
3564 * @handle: JBD handle for this transaction
3565 * @inode: inode we are dealing with
3566 * @parent_bh: the buffer_head which contains *@first and *@last
3567 * @first: array of block numbers
3568 * @last: pointer immediately past the end of array
3569 * @depth: depth of the branches to free
3571 * We are freeing all blocks refered from these branches (numbers are
3572 * stored as little-endian 32-bit) and updating @inode->i_blocks
3575 static void ext4_free_branches(handle_t
*handle
, struct inode
*inode
,
3576 struct buffer_head
*parent_bh
,
3577 __le32
*first
, __le32
*last
, int depth
)
3582 if (is_handle_aborted(handle
))
3586 struct buffer_head
*bh
;
3587 int addr_per_block
= EXT4_ADDR_PER_BLOCK(inode
->i_sb
);
3589 while (--p
>= first
) {
3590 nr
= le32_to_cpu(*p
);
3592 continue; /* A hole */
3594 /* Go read the buffer for the next level down */
3595 bh
= sb_bread(inode
->i_sb
, nr
);
3598 * A read failure? Report error and clear slot
3602 ext4_error(inode
->i_sb
, "ext4_free_branches",
3603 "Read failure, inode=%lu, block=%llu",
3608 /* This zaps the entire block. Bottom up. */
3609 BUFFER_TRACE(bh
, "free child branches");
3610 ext4_free_branches(handle
, inode
, bh
,
3611 (__le32
*) bh
->b_data
,
3612 (__le32
*) bh
->b_data
+ addr_per_block
,
3616 * We've probably journalled the indirect block several
3617 * times during the truncate. But it's no longer
3618 * needed and we now drop it from the transaction via
3619 * jbd2_journal_revoke().
3621 * That's easy if it's exclusively part of this
3622 * transaction. But if it's part of the committing
3623 * transaction then jbd2_journal_forget() will simply
3624 * brelse() it. That means that if the underlying
3625 * block is reallocated in ext4_get_block(),
3626 * unmap_underlying_metadata() will find this block
3627 * and will try to get rid of it. damn, damn.
3629 * If this block has already been committed to the
3630 * journal, a revoke record will be written. And
3631 * revoke records must be emitted *before* clearing
3632 * this block's bit in the bitmaps.
3634 ext4_forget(handle
, 1, inode
, bh
, bh
->b_blocknr
);
3637 * Everything below this this pointer has been
3638 * released. Now let this top-of-subtree go.
3640 * We want the freeing of this indirect block to be
3641 * atomic in the journal with the updating of the
3642 * bitmap block which owns it. So make some room in
3645 * We zero the parent pointer *after* freeing its
3646 * pointee in the bitmaps, so if extend_transaction()
3647 * for some reason fails to put the bitmap changes and
3648 * the release into the same transaction, recovery
3649 * will merely complain about releasing a free block,
3650 * rather than leaking blocks.
3652 if (is_handle_aborted(handle
))
3654 if (try_to_extend_transaction(handle
, inode
)) {
3655 ext4_mark_inode_dirty(handle
, inode
);
3656 ext4_journal_test_restart(handle
, inode
);
3659 ext4_free_blocks(handle
, inode
, nr
, 1, 1);
3663 * The block which we have just freed is
3664 * pointed to by an indirect block: journal it
3666 BUFFER_TRACE(parent_bh
, "get_write_access");
3667 if (!ext4_journal_get_write_access(handle
,
3670 BUFFER_TRACE(parent_bh
,
3671 "call ext4_journal_dirty_metadata");
3672 ext4_journal_dirty_metadata(handle
,
3678 /* We have reached the bottom of the tree. */
3679 BUFFER_TRACE(parent_bh
, "free data blocks");
3680 ext4_free_data(handle
, inode
, parent_bh
, first
, last
);
3684 int ext4_can_truncate(struct inode
*inode
)
3686 if (IS_APPEND(inode
) || IS_IMMUTABLE(inode
))
3688 if (S_ISREG(inode
->i_mode
))
3690 if (S_ISDIR(inode
->i_mode
))
3692 if (S_ISLNK(inode
->i_mode
))
3693 return !ext4_inode_is_fast_symlink(inode
);
3700 * We block out ext4_get_block() block instantiations across the entire
3701 * transaction, and VFS/VM ensures that ext4_truncate() cannot run
3702 * simultaneously on behalf of the same inode.
3704 * As we work through the truncate and commmit bits of it to the journal there
3705 * is one core, guiding principle: the file's tree must always be consistent on
3706 * disk. We must be able to restart the truncate after a crash.
3708 * The file's tree may be transiently inconsistent in memory (although it
3709 * probably isn't), but whenever we close off and commit a journal transaction,
3710 * the contents of (the filesystem + the journal) must be consistent and
3711 * restartable. It's pretty simple, really: bottom up, right to left (although
3712 * left-to-right works OK too).
3714 * Note that at recovery time, journal replay occurs *before* the restart of
3715 * truncate against the orphan inode list.
3717 * The committed inode has the new, desired i_size (which is the same as
3718 * i_disksize in this case). After a crash, ext4_orphan_cleanup() will see
3719 * that this inode's truncate did not complete and it will again call
3720 * ext4_truncate() to have another go. So there will be instantiated blocks
3721 * to the right of the truncation point in a crashed ext4 filesystem. But
3722 * that's fine - as long as they are linked from the inode, the post-crash
3723 * ext4_truncate() run will find them and release them.
3725 void ext4_truncate(struct inode
*inode
)
3728 struct ext4_inode_info
*ei
= EXT4_I(inode
);
3729 __le32
*i_data
= ei
->i_data
;
3730 int addr_per_block
= EXT4_ADDR_PER_BLOCK(inode
->i_sb
);
3731 struct address_space
*mapping
= inode
->i_mapping
;
3732 ext4_lblk_t offsets
[4];
3737 ext4_lblk_t last_block
;
3738 unsigned blocksize
= inode
->i_sb
->s_blocksize
;
3740 if (!ext4_can_truncate(inode
))
3743 if (EXT4_I(inode
)->i_flags
& EXT4_EXTENTS_FL
) {
3744 ext4_ext_truncate(inode
);
3748 handle
= start_transaction(inode
);
3750 return; /* AKPM: return what? */
3752 last_block
= (inode
->i_size
+ blocksize
-1)
3753 >> EXT4_BLOCK_SIZE_BITS(inode
->i_sb
);
3755 if (inode
->i_size
& (blocksize
- 1))
3756 if (ext4_block_truncate_page(handle
, mapping
, inode
->i_size
))
3759 n
= ext4_block_to_path(inode
, last_block
, offsets
, NULL
);
3761 goto out_stop
; /* error */
3764 * OK. This truncate is going to happen. We add the inode to the
3765 * orphan list, so that if this truncate spans multiple transactions,
3766 * and we crash, we will resume the truncate when the filesystem
3767 * recovers. It also marks the inode dirty, to catch the new size.
3769 * Implication: the file must always be in a sane, consistent
3770 * truncatable state while each transaction commits.
3772 if (ext4_orphan_add(handle
, inode
))
3776 * From here we block out all ext4_get_block() callers who want to
3777 * modify the block allocation tree.
3779 down_write(&ei
->i_data_sem
);
3781 ext4_discard_preallocations(inode
);
3784 * The orphan list entry will now protect us from any crash which
3785 * occurs before the truncate completes, so it is now safe to propagate
3786 * the new, shorter inode size (held for now in i_size) into the
3787 * on-disk inode. We do this via i_disksize, which is the value which
3788 * ext4 *really* writes onto the disk inode.
3790 ei
->i_disksize
= inode
->i_size
;
3792 if (n
== 1) { /* direct blocks */
3793 ext4_free_data(handle
, inode
, NULL
, i_data
+offsets
[0],
3794 i_data
+ EXT4_NDIR_BLOCKS
);
3798 partial
= ext4_find_shared(inode
, n
, offsets
, chain
, &nr
);
3799 /* Kill the top of shared branch (not detached) */
3801 if (partial
== chain
) {
3802 /* Shared branch grows from the inode */
3803 ext4_free_branches(handle
, inode
, NULL
,
3804 &nr
, &nr
+1, (chain
+n
-1) - partial
);
3807 * We mark the inode dirty prior to restart,
3808 * and prior to stop. No need for it here.
3811 /* Shared branch grows from an indirect block */
3812 BUFFER_TRACE(partial
->bh
, "get_write_access");
3813 ext4_free_branches(handle
, inode
, partial
->bh
,
3815 partial
->p
+1, (chain
+n
-1) - partial
);
3818 /* Clear the ends of indirect blocks on the shared branch */
3819 while (partial
> chain
) {
3820 ext4_free_branches(handle
, inode
, partial
->bh
, partial
->p
+ 1,
3821 (__le32
*)partial
->bh
->b_data
+addr_per_block
,
3822 (chain
+n
-1) - partial
);
3823 BUFFER_TRACE(partial
->bh
, "call brelse");
3824 brelse (partial
->bh
);
3828 /* Kill the remaining (whole) subtrees */
3829 switch (offsets
[0]) {
3831 nr
= i_data
[EXT4_IND_BLOCK
];
3833 ext4_free_branches(handle
, inode
, NULL
, &nr
, &nr
+1, 1);
3834 i_data
[EXT4_IND_BLOCK
] = 0;
3836 case EXT4_IND_BLOCK
:
3837 nr
= i_data
[EXT4_DIND_BLOCK
];
3839 ext4_free_branches(handle
, inode
, NULL
, &nr
, &nr
+1, 2);
3840 i_data
[EXT4_DIND_BLOCK
] = 0;
3842 case EXT4_DIND_BLOCK
:
3843 nr
= i_data
[EXT4_TIND_BLOCK
];
3845 ext4_free_branches(handle
, inode
, NULL
, &nr
, &nr
+1, 3);
3846 i_data
[EXT4_TIND_BLOCK
] = 0;
3848 case EXT4_TIND_BLOCK
:
3852 up_write(&ei
->i_data_sem
);
3853 inode
->i_mtime
= inode
->i_ctime
= ext4_current_time(inode
);
3854 ext4_mark_inode_dirty(handle
, inode
);
3857 * In a multi-transaction truncate, we only make the final transaction
3864 * If this was a simple ftruncate(), and the file will remain alive
3865 * then we need to clear up the orphan record which we created above.
3866 * However, if this was a real unlink then we were called by
3867 * ext4_delete_inode(), and we allow that function to clean up the
3868 * orphan info for us.
3871 ext4_orphan_del(handle
, inode
);
3873 ext4_journal_stop(handle
);
3877 * ext4_get_inode_loc returns with an extra refcount against the inode's
3878 * underlying buffer_head on success. If 'in_mem' is true, we have all
3879 * data in memory that is needed to recreate the on-disk version of this
3882 static int __ext4_get_inode_loc(struct inode
*inode
,
3883 struct ext4_iloc
*iloc
, int in_mem
)
3885 struct ext4_group_desc
*gdp
;
3886 struct buffer_head
*bh
;
3887 struct super_block
*sb
= inode
->i_sb
;
3889 int inodes_per_block
, inode_offset
;
3892 if (!ext4_valid_inum(sb
, inode
->i_ino
))
3895 iloc
->block_group
= (inode
->i_ino
- 1) / EXT4_INODES_PER_GROUP(sb
);
3896 gdp
= ext4_get_group_desc(sb
, iloc
->block_group
, NULL
);
3901 * Figure out the offset within the block group inode table
3903 inodes_per_block
= (EXT4_BLOCK_SIZE(sb
) / EXT4_INODE_SIZE(sb
));
3904 inode_offset
= ((inode
->i_ino
- 1) %
3905 EXT4_INODES_PER_GROUP(sb
));
3906 block
= ext4_inode_table(sb
, gdp
) + (inode_offset
/ inodes_per_block
);
3907 iloc
->offset
= (inode_offset
% inodes_per_block
) * EXT4_INODE_SIZE(sb
);
3909 bh
= sb_getblk(sb
, block
);
3911 ext4_error(sb
, "ext4_get_inode_loc", "unable to read "
3912 "inode block - inode=%lu, block=%llu",
3913 inode
->i_ino
, block
);
3916 if (!buffer_uptodate(bh
)) {
3920 * If the buffer has the write error flag, we have failed
3921 * to write out another inode in the same block. In this
3922 * case, we don't have to read the block because we may
3923 * read the old inode data successfully.
3925 if (buffer_write_io_error(bh
) && !buffer_uptodate(bh
))
3926 set_buffer_uptodate(bh
);
3928 if (buffer_uptodate(bh
)) {
3929 /* someone brought it uptodate while we waited */
3935 * If we have all information of the inode in memory and this
3936 * is the only valid inode in the block, we need not read the
3940 struct buffer_head
*bitmap_bh
;
3943 start
= inode_offset
& ~(inodes_per_block
- 1);
3945 /* Is the inode bitmap in cache? */
3946 bitmap_bh
= sb_getblk(sb
, ext4_inode_bitmap(sb
, gdp
));
3951 * If the inode bitmap isn't in cache then the
3952 * optimisation may end up performing two reads instead
3953 * of one, so skip it.
3955 if (!buffer_uptodate(bitmap_bh
)) {
3959 for (i
= start
; i
< start
+ inodes_per_block
; i
++) {
3960 if (i
== inode_offset
)
3962 if (ext4_test_bit(i
, bitmap_bh
->b_data
))
3966 if (i
== start
+ inodes_per_block
) {
3967 /* all other inodes are free, so skip I/O */
3968 memset(bh
->b_data
, 0, bh
->b_size
);
3969 set_buffer_uptodate(bh
);
3977 * If we need to do any I/O, try to pre-readahead extra
3978 * blocks from the inode table.
3980 if (EXT4_SB(sb
)->s_inode_readahead_blks
) {
3981 ext4_fsblk_t b
, end
, table
;
3984 table
= ext4_inode_table(sb
, gdp
);
3985 /* Make sure s_inode_readahead_blks is a power of 2 */
3986 while (EXT4_SB(sb
)->s_inode_readahead_blks
&
3987 (EXT4_SB(sb
)->s_inode_readahead_blks
-1))
3988 EXT4_SB(sb
)->s_inode_readahead_blks
=
3989 (EXT4_SB(sb
)->s_inode_readahead_blks
&
3990 (EXT4_SB(sb
)->s_inode_readahead_blks
-1));
3991 b
= block
& ~(EXT4_SB(sb
)->s_inode_readahead_blks
-1);
3994 end
= b
+ EXT4_SB(sb
)->s_inode_readahead_blks
;
3995 num
= EXT4_INODES_PER_GROUP(sb
);
3996 if (EXT4_HAS_RO_COMPAT_FEATURE(sb
,
3997 EXT4_FEATURE_RO_COMPAT_GDT_CSUM
))
3998 num
-= le16_to_cpu(gdp
->bg_itable_unused
);
3999 table
+= num
/ inodes_per_block
;
4003 sb_breadahead(sb
, b
++);
4007 * There are other valid inodes in the buffer, this inode
4008 * has in-inode xattrs, or we don't have this inode in memory.
4009 * Read the block from disk.
4012 bh
->b_end_io
= end_buffer_read_sync
;
4013 submit_bh(READ_META
, bh
);
4015 if (!buffer_uptodate(bh
)) {
4016 ext4_error(sb
, __func__
,
4017 "unable to read inode block - inode=%lu, "
4018 "block=%llu", inode
->i_ino
, block
);
4028 int ext4_get_inode_loc(struct inode
*inode
, struct ext4_iloc
*iloc
)
4030 /* We have all inode data except xattrs in memory here. */
4031 return __ext4_get_inode_loc(inode
, iloc
,
4032 !(EXT4_I(inode
)->i_state
& EXT4_STATE_XATTR
));
4035 void ext4_set_inode_flags(struct inode
*inode
)
4037 unsigned int flags
= EXT4_I(inode
)->i_flags
;
4039 inode
->i_flags
&= ~(S_SYNC
|S_APPEND
|S_IMMUTABLE
|S_NOATIME
|S_DIRSYNC
);
4040 if (flags
& EXT4_SYNC_FL
)
4041 inode
->i_flags
|= S_SYNC
;
4042 if (flags
& EXT4_APPEND_FL
)
4043 inode
->i_flags
|= S_APPEND
;
4044 if (flags
& EXT4_IMMUTABLE_FL
)
4045 inode
->i_flags
|= S_IMMUTABLE
;
4046 if (flags
& EXT4_NOATIME_FL
)
4047 inode
->i_flags
|= S_NOATIME
;
4048 if (flags
& EXT4_DIRSYNC_FL
)
4049 inode
->i_flags
|= S_DIRSYNC
;
4052 /* Propagate flags from i_flags to EXT4_I(inode)->i_flags */
4053 void ext4_get_inode_flags(struct ext4_inode_info
*ei
)
4055 unsigned int flags
= ei
->vfs_inode
.i_flags
;
4057 ei
->i_flags
&= ~(EXT4_SYNC_FL
|EXT4_APPEND_FL
|
4058 EXT4_IMMUTABLE_FL
|EXT4_NOATIME_FL
|EXT4_DIRSYNC_FL
);
4060 ei
->i_flags
|= EXT4_SYNC_FL
;
4061 if (flags
& S_APPEND
)
4062 ei
->i_flags
|= EXT4_APPEND_FL
;
4063 if (flags
& S_IMMUTABLE
)
4064 ei
->i_flags
|= EXT4_IMMUTABLE_FL
;
4065 if (flags
& S_NOATIME
)
4066 ei
->i_flags
|= EXT4_NOATIME_FL
;
4067 if (flags
& S_DIRSYNC
)
4068 ei
->i_flags
|= EXT4_DIRSYNC_FL
;
4070 static blkcnt_t
ext4_inode_blocks(struct ext4_inode
*raw_inode
,
4071 struct ext4_inode_info
*ei
)
4074 struct inode
*inode
= &(ei
->vfs_inode
);
4075 struct super_block
*sb
= inode
->i_sb
;
4077 if (EXT4_HAS_RO_COMPAT_FEATURE(sb
,
4078 EXT4_FEATURE_RO_COMPAT_HUGE_FILE
)) {
4079 /* we are using combined 48 bit field */
4080 i_blocks
= ((u64
)le16_to_cpu(raw_inode
->i_blocks_high
)) << 32 |
4081 le32_to_cpu(raw_inode
->i_blocks_lo
);
4082 if (ei
->i_flags
& EXT4_HUGE_FILE_FL
) {
4083 /* i_blocks represent file system block size */
4084 return i_blocks
<< (inode
->i_blkbits
- 9);
4089 return le32_to_cpu(raw_inode
->i_blocks_lo
);
4093 struct inode
*ext4_iget(struct super_block
*sb
, unsigned long ino
)
4095 struct ext4_iloc iloc
;
4096 struct ext4_inode
*raw_inode
;
4097 struct ext4_inode_info
*ei
;
4098 struct buffer_head
*bh
;
4099 struct inode
*inode
;
4103 inode
= iget_locked(sb
, ino
);
4105 return ERR_PTR(-ENOMEM
);
4106 if (!(inode
->i_state
& I_NEW
))
4110 #ifdef CONFIG_EXT4_FS_POSIX_ACL
4111 ei
->i_acl
= EXT4_ACL_NOT_CACHED
;
4112 ei
->i_default_acl
= EXT4_ACL_NOT_CACHED
;
4115 ret
= __ext4_get_inode_loc(inode
, &iloc
, 0);
4119 raw_inode
= ext4_raw_inode(&iloc
);
4120 inode
->i_mode
= le16_to_cpu(raw_inode
->i_mode
);
4121 inode
->i_uid
= (uid_t
)le16_to_cpu(raw_inode
->i_uid_low
);
4122 inode
->i_gid
= (gid_t
)le16_to_cpu(raw_inode
->i_gid_low
);
4123 if (!(test_opt(inode
->i_sb
, NO_UID32
))) {
4124 inode
->i_uid
|= le16_to_cpu(raw_inode
->i_uid_high
) << 16;
4125 inode
->i_gid
|= le16_to_cpu(raw_inode
->i_gid_high
) << 16;
4127 inode
->i_nlink
= le16_to_cpu(raw_inode
->i_links_count
);
4130 ei
->i_dir_start_lookup
= 0;
4131 ei
->i_dtime
= le32_to_cpu(raw_inode
->i_dtime
);
4132 /* We now have enough fields to check if the inode was active or not.
4133 * This is needed because nfsd might try to access dead inodes
4134 * the test is that same one that e2fsck uses
4135 * NeilBrown 1999oct15
4137 if (inode
->i_nlink
== 0) {
4138 if (inode
->i_mode
== 0 ||
4139 !(EXT4_SB(inode
->i_sb
)->s_mount_state
& EXT4_ORPHAN_FS
)) {
4140 /* this inode is deleted */
4145 /* The only unlinked inodes we let through here have
4146 * valid i_mode and are being read by the orphan
4147 * recovery code: that's fine, we're about to complete
4148 * the process of deleting those. */
4150 ei
->i_flags
= le32_to_cpu(raw_inode
->i_flags
);
4151 inode
->i_blocks
= ext4_inode_blocks(raw_inode
, ei
);
4152 ei
->i_file_acl
= le32_to_cpu(raw_inode
->i_file_acl_lo
);
4153 if (EXT4_SB(inode
->i_sb
)->s_es
->s_creator_os
!=
4154 cpu_to_le32(EXT4_OS_HURD
)) {
4156 ((__u64
)le16_to_cpu(raw_inode
->i_file_acl_high
)) << 32;
4158 inode
->i_size
= ext4_isize(raw_inode
);
4159 ei
->i_disksize
= inode
->i_size
;
4160 inode
->i_generation
= le32_to_cpu(raw_inode
->i_generation
);
4161 ei
->i_block_group
= iloc
.block_group
;
4163 * NOTE! The in-memory inode i_data array is in little-endian order
4164 * even on big-endian machines: we do NOT byteswap the block numbers!
4166 for (block
= 0; block
< EXT4_N_BLOCKS
; block
++)
4167 ei
->i_data
[block
] = raw_inode
->i_block
[block
];
4168 INIT_LIST_HEAD(&ei
->i_orphan
);
4170 if (EXT4_INODE_SIZE(inode
->i_sb
) > EXT4_GOOD_OLD_INODE_SIZE
) {
4171 ei
->i_extra_isize
= le16_to_cpu(raw_inode
->i_extra_isize
);
4172 if (EXT4_GOOD_OLD_INODE_SIZE
+ ei
->i_extra_isize
>
4173 EXT4_INODE_SIZE(inode
->i_sb
)) {
4178 if (ei
->i_extra_isize
== 0) {
4179 /* The extra space is currently unused. Use it. */
4180 ei
->i_extra_isize
= sizeof(struct ext4_inode
) -
4181 EXT4_GOOD_OLD_INODE_SIZE
;
4183 __le32
*magic
= (void *)raw_inode
+
4184 EXT4_GOOD_OLD_INODE_SIZE
+
4186 if (*magic
== cpu_to_le32(EXT4_XATTR_MAGIC
))
4187 ei
->i_state
|= EXT4_STATE_XATTR
;
4190 ei
->i_extra_isize
= 0;
4192 EXT4_INODE_GET_XTIME(i_ctime
, inode
, raw_inode
);
4193 EXT4_INODE_GET_XTIME(i_mtime
, inode
, raw_inode
);
4194 EXT4_INODE_GET_XTIME(i_atime
, inode
, raw_inode
);
4195 EXT4_EINODE_GET_XTIME(i_crtime
, ei
, raw_inode
);
4197 inode
->i_version
= le32_to_cpu(raw_inode
->i_disk_version
);
4198 if (EXT4_INODE_SIZE(inode
->i_sb
) > EXT4_GOOD_OLD_INODE_SIZE
) {
4199 if (EXT4_FITS_IN_INODE(raw_inode
, ei
, i_version_hi
))
4201 (__u64
)(le32_to_cpu(raw_inode
->i_version_hi
)) << 32;
4204 if (S_ISREG(inode
->i_mode
)) {
4205 inode
->i_op
= &ext4_file_inode_operations
;
4206 inode
->i_fop
= &ext4_file_operations
;
4207 ext4_set_aops(inode
);
4208 } else if (S_ISDIR(inode
->i_mode
)) {
4209 inode
->i_op
= &ext4_dir_inode_operations
;
4210 inode
->i_fop
= &ext4_dir_operations
;
4211 } else if (S_ISLNK(inode
->i_mode
)) {
4212 if (ext4_inode_is_fast_symlink(inode
)) {
4213 inode
->i_op
= &ext4_fast_symlink_inode_operations
;
4214 nd_terminate_link(ei
->i_data
, inode
->i_size
,
4215 sizeof(ei
->i_data
) - 1);
4217 inode
->i_op
= &ext4_symlink_inode_operations
;
4218 ext4_set_aops(inode
);
4221 inode
->i_op
= &ext4_special_inode_operations
;
4222 if (raw_inode
->i_block
[0])
4223 init_special_inode(inode
, inode
->i_mode
,
4224 old_decode_dev(le32_to_cpu(raw_inode
->i_block
[0])));
4226 init_special_inode(inode
, inode
->i_mode
,
4227 new_decode_dev(le32_to_cpu(raw_inode
->i_block
[1])));
4230 ext4_set_inode_flags(inode
);
4231 unlock_new_inode(inode
);
4236 return ERR_PTR(ret
);
4239 static int ext4_inode_blocks_set(handle_t
*handle
,
4240 struct ext4_inode
*raw_inode
,
4241 struct ext4_inode_info
*ei
)
4243 struct inode
*inode
= &(ei
->vfs_inode
);
4244 u64 i_blocks
= inode
->i_blocks
;
4245 struct super_block
*sb
= inode
->i_sb
;
4247 if (i_blocks
<= ~0U) {
4249 * i_blocks can be represnted in a 32 bit variable
4250 * as multiple of 512 bytes
4252 raw_inode
->i_blocks_lo
= cpu_to_le32(i_blocks
);
4253 raw_inode
->i_blocks_high
= 0;
4254 ei
->i_flags
&= ~EXT4_HUGE_FILE_FL
;
4257 if (!EXT4_HAS_RO_COMPAT_FEATURE(sb
, EXT4_FEATURE_RO_COMPAT_HUGE_FILE
))
4260 if (i_blocks
<= 0xffffffffffffULL
) {
4262 * i_blocks can be represented in a 48 bit variable
4263 * as multiple of 512 bytes
4265 raw_inode
->i_blocks_lo
= cpu_to_le32(i_blocks
);
4266 raw_inode
->i_blocks_high
= cpu_to_le16(i_blocks
>> 32);
4267 ei
->i_flags
&= ~EXT4_HUGE_FILE_FL
;
4269 ei
->i_flags
|= EXT4_HUGE_FILE_FL
;
4270 /* i_block is stored in file system block size */
4271 i_blocks
= i_blocks
>> (inode
->i_blkbits
- 9);
4272 raw_inode
->i_blocks_lo
= cpu_to_le32(i_blocks
);
4273 raw_inode
->i_blocks_high
= cpu_to_le16(i_blocks
>> 32);
4279 * Post the struct inode info into an on-disk inode location in the
4280 * buffer-cache. This gobbles the caller's reference to the
4281 * buffer_head in the inode location struct.
4283 * The caller must have write access to iloc->bh.
4285 static int ext4_do_update_inode(handle_t
*handle
,
4286 struct inode
*inode
,
4287 struct ext4_iloc
*iloc
)
4289 struct ext4_inode
*raw_inode
= ext4_raw_inode(iloc
);
4290 struct ext4_inode_info
*ei
= EXT4_I(inode
);
4291 struct buffer_head
*bh
= iloc
->bh
;
4292 int err
= 0, rc
, block
;
4294 /* For fields not not tracking in the in-memory inode,
4295 * initialise them to zero for new inodes. */
4296 if (ei
->i_state
& EXT4_STATE_NEW
)
4297 memset(raw_inode
, 0, EXT4_SB(inode
->i_sb
)->s_inode_size
);
4299 ext4_get_inode_flags(ei
);
4300 raw_inode
->i_mode
= cpu_to_le16(inode
->i_mode
);
4301 if (!(test_opt(inode
->i_sb
, NO_UID32
))) {
4302 raw_inode
->i_uid_low
= cpu_to_le16(low_16_bits(inode
->i_uid
));
4303 raw_inode
->i_gid_low
= cpu_to_le16(low_16_bits(inode
->i_gid
));
4305 * Fix up interoperability with old kernels. Otherwise, old inodes get
4306 * re-used with the upper 16 bits of the uid/gid intact
4309 raw_inode
->i_uid_high
=
4310 cpu_to_le16(high_16_bits(inode
->i_uid
));
4311 raw_inode
->i_gid_high
=
4312 cpu_to_le16(high_16_bits(inode
->i_gid
));
4314 raw_inode
->i_uid_high
= 0;
4315 raw_inode
->i_gid_high
= 0;
4318 raw_inode
->i_uid_low
=
4319 cpu_to_le16(fs_high2lowuid(inode
->i_uid
));
4320 raw_inode
->i_gid_low
=
4321 cpu_to_le16(fs_high2lowgid(inode
->i_gid
));
4322 raw_inode
->i_uid_high
= 0;
4323 raw_inode
->i_gid_high
= 0;
4325 raw_inode
->i_links_count
= cpu_to_le16(inode
->i_nlink
);
4327 EXT4_INODE_SET_XTIME(i_ctime
, inode
, raw_inode
);
4328 EXT4_INODE_SET_XTIME(i_mtime
, inode
, raw_inode
);
4329 EXT4_INODE_SET_XTIME(i_atime
, inode
, raw_inode
);
4330 EXT4_EINODE_SET_XTIME(i_crtime
, ei
, raw_inode
);
4332 if (ext4_inode_blocks_set(handle
, raw_inode
, ei
))
4334 raw_inode
->i_dtime
= cpu_to_le32(ei
->i_dtime
);
4335 /* clear the migrate flag in the raw_inode */
4336 raw_inode
->i_flags
= cpu_to_le32(ei
->i_flags
& ~EXT4_EXT_MIGRATE
);
4337 if (EXT4_SB(inode
->i_sb
)->s_es
->s_creator_os
!=
4338 cpu_to_le32(EXT4_OS_HURD
))
4339 raw_inode
->i_file_acl_high
=
4340 cpu_to_le16(ei
->i_file_acl
>> 32);
4341 raw_inode
->i_file_acl_lo
= cpu_to_le32(ei
->i_file_acl
);
4342 ext4_isize_set(raw_inode
, ei
->i_disksize
);
4343 if (ei
->i_disksize
> 0x7fffffffULL
) {
4344 struct super_block
*sb
= inode
->i_sb
;
4345 if (!EXT4_HAS_RO_COMPAT_FEATURE(sb
,
4346 EXT4_FEATURE_RO_COMPAT_LARGE_FILE
) ||
4347 EXT4_SB(sb
)->s_es
->s_rev_level
==
4348 cpu_to_le32(EXT4_GOOD_OLD_REV
)) {
4349 /* If this is the first large file
4350 * created, add a flag to the superblock.
4352 err
= ext4_journal_get_write_access(handle
,
4353 EXT4_SB(sb
)->s_sbh
);
4356 ext4_update_dynamic_rev(sb
);
4357 EXT4_SET_RO_COMPAT_FEATURE(sb
,
4358 EXT4_FEATURE_RO_COMPAT_LARGE_FILE
);
4361 err
= ext4_journal_dirty_metadata(handle
,
4362 EXT4_SB(sb
)->s_sbh
);
4365 raw_inode
->i_generation
= cpu_to_le32(inode
->i_generation
);
4366 if (S_ISCHR(inode
->i_mode
) || S_ISBLK(inode
->i_mode
)) {
4367 if (old_valid_dev(inode
->i_rdev
)) {
4368 raw_inode
->i_block
[0] =
4369 cpu_to_le32(old_encode_dev(inode
->i_rdev
));
4370 raw_inode
->i_block
[1] = 0;
4372 raw_inode
->i_block
[0] = 0;
4373 raw_inode
->i_block
[1] =
4374 cpu_to_le32(new_encode_dev(inode
->i_rdev
));
4375 raw_inode
->i_block
[2] = 0;
4377 } else for (block
= 0; block
< EXT4_N_BLOCKS
; block
++)
4378 raw_inode
->i_block
[block
] = ei
->i_data
[block
];
4380 raw_inode
->i_disk_version
= cpu_to_le32(inode
->i_version
);
4381 if (ei
->i_extra_isize
) {
4382 if (EXT4_FITS_IN_INODE(raw_inode
, ei
, i_version_hi
))
4383 raw_inode
->i_version_hi
=
4384 cpu_to_le32(inode
->i_version
>> 32);
4385 raw_inode
->i_extra_isize
= cpu_to_le16(ei
->i_extra_isize
);
4389 BUFFER_TRACE(bh
, "call ext4_journal_dirty_metadata");
4390 rc
= ext4_journal_dirty_metadata(handle
, bh
);
4393 ei
->i_state
&= ~EXT4_STATE_NEW
;
4397 ext4_std_error(inode
->i_sb
, err
);
4402 * ext4_write_inode()
4404 * We are called from a few places:
4406 * - Within generic_file_write() for O_SYNC files.
4407 * Here, there will be no transaction running. We wait for any running
4408 * trasnaction to commit.
4410 * - Within sys_sync(), kupdate and such.
4411 * We wait on commit, if tol to.
4413 * - Within prune_icache() (PF_MEMALLOC == true)
4414 * Here we simply return. We can't afford to block kswapd on the
4417 * In all cases it is actually safe for us to return without doing anything,
4418 * because the inode has been copied into a raw inode buffer in
4419 * ext4_mark_inode_dirty(). This is a correctness thing for O_SYNC and for
4422 * Note that we are absolutely dependent upon all inode dirtiers doing the
4423 * right thing: they *must* call mark_inode_dirty() after dirtying info in
4424 * which we are interested.
4426 * It would be a bug for them to not do this. The code:
4428 * mark_inode_dirty(inode)
4430 * inode->i_size = expr;
4432 * is in error because a kswapd-driven write_inode() could occur while
4433 * `stuff()' is running, and the new i_size will be lost. Plus the inode
4434 * will no longer be on the superblock's dirty inode list.
4436 int ext4_write_inode(struct inode
*inode
, int wait
)
4438 if (current
->flags
& PF_MEMALLOC
)
4441 if (ext4_journal_current_handle()) {
4442 jbd_debug(1, "called recursively, non-PF_MEMALLOC!\n");
4450 return ext4_force_commit(inode
->i_sb
);
4456 * Called from notify_change.
4458 * We want to trap VFS attempts to truncate the file as soon as
4459 * possible. In particular, we want to make sure that when the VFS
4460 * shrinks i_size, we put the inode on the orphan list and modify
4461 * i_disksize immediately, so that during the subsequent flushing of
4462 * dirty pages and freeing of disk blocks, we can guarantee that any
4463 * commit will leave the blocks being flushed in an unused state on
4464 * disk. (On recovery, the inode will get truncated and the blocks will
4465 * be freed, so we have a strong guarantee that no future commit will
4466 * leave these blocks visible to the user.)
4468 * Another thing we have to assure is that if we are in ordered mode
4469 * and inode is still attached to the committing transaction, we must
4470 * we start writeout of all the dirty pages which are being truncated.
4471 * This way we are sure that all the data written in the previous
4472 * transaction are already on disk (truncate waits for pages under
4475 * Called with inode->i_mutex down.
4477 int ext4_setattr(struct dentry
*dentry
, struct iattr
*attr
)
4479 struct inode
*inode
= dentry
->d_inode
;
4481 const unsigned int ia_valid
= attr
->ia_valid
;
4483 error
= inode_change_ok(inode
, attr
);
4487 if ((ia_valid
& ATTR_UID
&& attr
->ia_uid
!= inode
->i_uid
) ||
4488 (ia_valid
& ATTR_GID
&& attr
->ia_gid
!= inode
->i_gid
)) {
4491 /* (user+group)*(old+new) structure, inode write (sb,
4492 * inode block, ? - but truncate inode update has it) */
4493 handle
= ext4_journal_start(inode
, 2*(EXT4_QUOTA_INIT_BLOCKS(inode
->i_sb
)+
4494 EXT4_QUOTA_DEL_BLOCKS(inode
->i_sb
))+3);
4495 if (IS_ERR(handle
)) {
4496 error
= PTR_ERR(handle
);
4499 error
= DQUOT_TRANSFER(inode
, attr
) ? -EDQUOT
: 0;
4501 ext4_journal_stop(handle
);
4504 /* Update corresponding info in inode so that everything is in
4505 * one transaction */
4506 if (attr
->ia_valid
& ATTR_UID
)
4507 inode
->i_uid
= attr
->ia_uid
;
4508 if (attr
->ia_valid
& ATTR_GID
)
4509 inode
->i_gid
= attr
->ia_gid
;
4510 error
= ext4_mark_inode_dirty(handle
, inode
);
4511 ext4_journal_stop(handle
);
4514 if (attr
->ia_valid
& ATTR_SIZE
) {
4515 if (!(EXT4_I(inode
)->i_flags
& EXT4_EXTENTS_FL
)) {
4516 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
4518 if (attr
->ia_size
> sbi
->s_bitmap_maxbytes
) {
4525 if (S_ISREG(inode
->i_mode
) &&
4526 attr
->ia_valid
& ATTR_SIZE
&& attr
->ia_size
< inode
->i_size
) {
4529 handle
= ext4_journal_start(inode
, 3);
4530 if (IS_ERR(handle
)) {
4531 error
= PTR_ERR(handle
);
4535 error
= ext4_orphan_add(handle
, inode
);
4536 EXT4_I(inode
)->i_disksize
= attr
->ia_size
;
4537 rc
= ext4_mark_inode_dirty(handle
, inode
);
4540 ext4_journal_stop(handle
);
4542 if (ext4_should_order_data(inode
)) {
4543 error
= ext4_begin_ordered_truncate(inode
,
4546 /* Do as much error cleanup as possible */
4547 handle
= ext4_journal_start(inode
, 3);
4548 if (IS_ERR(handle
)) {
4549 ext4_orphan_del(NULL
, inode
);
4552 ext4_orphan_del(handle
, inode
);
4553 ext4_journal_stop(handle
);
4559 rc
= inode_setattr(inode
, attr
);
4561 /* If inode_setattr's call to ext4_truncate failed to get a
4562 * transaction handle at all, we need to clean up the in-core
4563 * orphan list manually. */
4565 ext4_orphan_del(NULL
, inode
);
4567 if (!rc
&& (ia_valid
& ATTR_MODE
))
4568 rc
= ext4_acl_chmod(inode
);
4571 ext4_std_error(inode
->i_sb
, error
);
4577 int ext4_getattr(struct vfsmount
*mnt
, struct dentry
*dentry
,
4580 struct inode
*inode
;
4581 unsigned long delalloc_blocks
;
4583 inode
= dentry
->d_inode
;
4584 generic_fillattr(inode
, stat
);
4587 * We can't update i_blocks if the block allocation is delayed
4588 * otherwise in the case of system crash before the real block
4589 * allocation is done, we will have i_blocks inconsistent with
4590 * on-disk file blocks.
4591 * We always keep i_blocks updated together with real
4592 * allocation. But to not confuse with user, stat
4593 * will return the blocks that include the delayed allocation
4594 * blocks for this file.
4596 spin_lock(&EXT4_I(inode
)->i_block_reservation_lock
);
4597 delalloc_blocks
= EXT4_I(inode
)->i_reserved_data_blocks
;
4598 spin_unlock(&EXT4_I(inode
)->i_block_reservation_lock
);
4600 stat
->blocks
+= (delalloc_blocks
<< inode
->i_sb
->s_blocksize_bits
)>>9;
4604 static int ext4_indirect_trans_blocks(struct inode
*inode
, int nrblocks
,
4609 /* if nrblocks are contiguous */
4612 * With N contiguous data blocks, it need at most
4613 * N/EXT4_ADDR_PER_BLOCK(inode->i_sb) indirect blocks
4614 * 2 dindirect blocks
4617 indirects
= nrblocks
/ EXT4_ADDR_PER_BLOCK(inode
->i_sb
);
4618 return indirects
+ 3;
4621 * if nrblocks are not contiguous, worse case, each block touch
4622 * a indirect block, and each indirect block touch a double indirect
4623 * block, plus a triple indirect block
4625 indirects
= nrblocks
* 2 + 1;
4629 static int ext4_index_trans_blocks(struct inode
*inode
, int nrblocks
, int chunk
)
4631 if (!(EXT4_I(inode
)->i_flags
& EXT4_EXTENTS_FL
))
4632 return ext4_indirect_trans_blocks(inode
, nrblocks
, chunk
);
4633 return ext4_ext_index_trans_blocks(inode
, nrblocks
, chunk
);
4637 * Account for index blocks, block groups bitmaps and block group
4638 * descriptor blocks if modify datablocks and index blocks
4639 * worse case, the indexs blocks spread over different block groups
4641 * If datablocks are discontiguous, they are possible to spread over
4642 * different block groups too. If they are contiugous, with flexbg,
4643 * they could still across block group boundary.
4645 * Also account for superblock, inode, quota and xattr blocks
4647 int ext4_meta_trans_blocks(struct inode
*inode
, int nrblocks
, int chunk
)
4649 int groups
, gdpblocks
;
4654 * How many index blocks need to touch to modify nrblocks?
4655 * The "Chunk" flag indicating whether the nrblocks is
4656 * physically contiguous on disk
4658 * For Direct IO and fallocate, they calls get_block to allocate
4659 * one single extent at a time, so they could set the "Chunk" flag
4661 idxblocks
= ext4_index_trans_blocks(inode
, nrblocks
, chunk
);
4666 * Now let's see how many group bitmaps and group descriptors need
4676 if (groups
> EXT4_SB(inode
->i_sb
)->s_groups_count
)
4677 groups
= EXT4_SB(inode
->i_sb
)->s_groups_count
;
4678 if (groups
> EXT4_SB(inode
->i_sb
)->s_gdb_count
)
4679 gdpblocks
= EXT4_SB(inode
->i_sb
)->s_gdb_count
;
4681 /* bitmaps and block group descriptor blocks */
4682 ret
+= groups
+ gdpblocks
;
4684 /* Blocks for super block, inode, quota and xattr blocks */
4685 ret
+= EXT4_META_TRANS_BLOCKS(inode
->i_sb
);
4691 * Calulate the total number of credits to reserve to fit
4692 * the modification of a single pages into a single transaction,
4693 * which may include multiple chunks of block allocations.
4695 * This could be called via ext4_write_begin()
4697 * We need to consider the worse case, when
4698 * one new block per extent.
4700 int ext4_writepage_trans_blocks(struct inode
*inode
)
4702 int bpp
= ext4_journal_blocks_per_page(inode
);
4705 ret
= ext4_meta_trans_blocks(inode
, bpp
, 0);
4707 /* Account for data blocks for journalled mode */
4708 if (ext4_should_journal_data(inode
))
4714 * Calculate the journal credits for a chunk of data modification.
4716 * This is called from DIO, fallocate or whoever calling
4717 * ext4_get_blocks_wrap() to map/allocate a chunk of contigous disk blocks.
4719 * journal buffers for data blocks are not included here, as DIO
4720 * and fallocate do no need to journal data buffers.
4722 int ext4_chunk_trans_blocks(struct inode
*inode
, int nrblocks
)
4724 return ext4_meta_trans_blocks(inode
, nrblocks
, 1);
4728 * The caller must have previously called ext4_reserve_inode_write().
4729 * Give this, we know that the caller already has write access to iloc->bh.
4731 int ext4_mark_iloc_dirty(handle_t
*handle
,
4732 struct inode
*inode
, struct ext4_iloc
*iloc
)
4736 if (test_opt(inode
->i_sb
, I_VERSION
))
4737 inode_inc_iversion(inode
);
4739 /* the do_update_inode consumes one bh->b_count */
4742 /* ext4_do_update_inode() does jbd2_journal_dirty_metadata */
4743 err
= ext4_do_update_inode(handle
, inode
, iloc
);
4749 * On success, We end up with an outstanding reference count against
4750 * iloc->bh. This _must_ be cleaned up later.
4754 ext4_reserve_inode_write(handle_t
*handle
, struct inode
*inode
,
4755 struct ext4_iloc
*iloc
)
4759 err
= ext4_get_inode_loc(inode
, iloc
);
4761 BUFFER_TRACE(iloc
->bh
, "get_write_access");
4762 err
= ext4_journal_get_write_access(handle
, iloc
->bh
);
4769 ext4_std_error(inode
->i_sb
, err
);
4774 * Expand an inode by new_extra_isize bytes.
4775 * Returns 0 on success or negative error number on failure.
4777 static int ext4_expand_extra_isize(struct inode
*inode
,
4778 unsigned int new_extra_isize
,
4779 struct ext4_iloc iloc
,
4782 struct ext4_inode
*raw_inode
;
4783 struct ext4_xattr_ibody_header
*header
;
4784 struct ext4_xattr_entry
*entry
;
4786 if (EXT4_I(inode
)->i_extra_isize
>= new_extra_isize
)
4789 raw_inode
= ext4_raw_inode(&iloc
);
4791 header
= IHDR(inode
, raw_inode
);
4792 entry
= IFIRST(header
);
4794 /* No extended attributes present */
4795 if (!(EXT4_I(inode
)->i_state
& EXT4_STATE_XATTR
) ||
4796 header
->h_magic
!= cpu_to_le32(EXT4_XATTR_MAGIC
)) {
4797 memset((void *)raw_inode
+ EXT4_GOOD_OLD_INODE_SIZE
, 0,
4799 EXT4_I(inode
)->i_extra_isize
= new_extra_isize
;
4803 /* try to expand with EAs present */
4804 return ext4_expand_extra_isize_ea(inode
, new_extra_isize
,
4809 * What we do here is to mark the in-core inode as clean with respect to inode
4810 * dirtiness (it may still be data-dirty).
4811 * This means that the in-core inode may be reaped by prune_icache
4812 * without having to perform any I/O. This is a very good thing,
4813 * because *any* task may call prune_icache - even ones which
4814 * have a transaction open against a different journal.
4816 * Is this cheating? Not really. Sure, we haven't written the
4817 * inode out, but prune_icache isn't a user-visible syncing function.
4818 * Whenever the user wants stuff synced (sys_sync, sys_msync, sys_fsync)
4819 * we start and wait on commits.
4821 * Is this efficient/effective? Well, we're being nice to the system
4822 * by cleaning up our inodes proactively so they can be reaped
4823 * without I/O. But we are potentially leaving up to five seconds'
4824 * worth of inodes floating about which prune_icache wants us to
4825 * write out. One way to fix that would be to get prune_icache()
4826 * to do a write_super() to free up some memory. It has the desired
4829 int ext4_mark_inode_dirty(handle_t
*handle
, struct inode
*inode
)
4831 struct ext4_iloc iloc
;
4832 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
4833 static unsigned int mnt_count
;
4837 err
= ext4_reserve_inode_write(handle
, inode
, &iloc
);
4838 if (EXT4_I(inode
)->i_extra_isize
< sbi
->s_want_extra_isize
&&
4839 !(EXT4_I(inode
)->i_state
& EXT4_STATE_NO_EXPAND
)) {
4841 * We need extra buffer credits since we may write into EA block
4842 * with this same handle. If journal_extend fails, then it will
4843 * only result in a minor loss of functionality for that inode.
4844 * If this is felt to be critical, then e2fsck should be run to
4845 * force a large enough s_min_extra_isize.
4847 if ((jbd2_journal_extend(handle
,
4848 EXT4_DATA_TRANS_BLOCKS(inode
->i_sb
))) == 0) {
4849 ret
= ext4_expand_extra_isize(inode
,
4850 sbi
->s_want_extra_isize
,
4853 EXT4_I(inode
)->i_state
|= EXT4_STATE_NO_EXPAND
;
4855 le16_to_cpu(sbi
->s_es
->s_mnt_count
)) {
4856 ext4_warning(inode
->i_sb
, __func__
,
4857 "Unable to expand inode %lu. Delete"
4858 " some EAs or run e2fsck.",
4861 le16_to_cpu(sbi
->s_es
->s_mnt_count
);
4867 err
= ext4_mark_iloc_dirty(handle
, inode
, &iloc
);
4872 * ext4_dirty_inode() is called from __mark_inode_dirty()
4874 * We're really interested in the case where a file is being extended.
4875 * i_size has been changed by generic_commit_write() and we thus need
4876 * to include the updated inode in the current transaction.
4878 * Also, DQUOT_ALLOC_SPACE() will always dirty the inode when blocks
4879 * are allocated to the file.
4881 * If the inode is marked synchronous, we don't honour that here - doing
4882 * so would cause a commit on atime updates, which we don't bother doing.
4883 * We handle synchronous inodes at the highest possible level.
4885 void ext4_dirty_inode(struct inode
*inode
)
4887 handle_t
*current_handle
= ext4_journal_current_handle();
4890 handle
= ext4_journal_start(inode
, 2);
4893 if (current_handle
&&
4894 current_handle
->h_transaction
!= handle
->h_transaction
) {
4895 /* This task has a transaction open against a different fs */
4896 printk(KERN_EMERG
"%s: transactions do not match!\n",
4899 jbd_debug(5, "marking dirty. outer handle=%p\n",
4901 ext4_mark_inode_dirty(handle
, inode
);
4903 ext4_journal_stop(handle
);
4910 * Bind an inode's backing buffer_head into this transaction, to prevent
4911 * it from being flushed to disk early. Unlike
4912 * ext4_reserve_inode_write, this leaves behind no bh reference and
4913 * returns no iloc structure, so the caller needs to repeat the iloc
4914 * lookup to mark the inode dirty later.
4916 static int ext4_pin_inode(handle_t
*handle
, struct inode
*inode
)
4918 struct ext4_iloc iloc
;
4922 err
= ext4_get_inode_loc(inode
, &iloc
);
4924 BUFFER_TRACE(iloc
.bh
, "get_write_access");
4925 err
= jbd2_journal_get_write_access(handle
, iloc
.bh
);
4927 err
= ext4_journal_dirty_metadata(handle
,
4932 ext4_std_error(inode
->i_sb
, err
);
4937 int ext4_change_inode_journal_flag(struct inode
*inode
, int val
)
4944 * We have to be very careful here: changing a data block's
4945 * journaling status dynamically is dangerous. If we write a
4946 * data block to the journal, change the status and then delete
4947 * that block, we risk forgetting to revoke the old log record
4948 * from the journal and so a subsequent replay can corrupt data.
4949 * So, first we make sure that the journal is empty and that
4950 * nobody is changing anything.
4953 journal
= EXT4_JOURNAL(inode
);
4954 if (is_journal_aborted(journal
))
4957 jbd2_journal_lock_updates(journal
);
4958 jbd2_journal_flush(journal
);
4961 * OK, there are no updates running now, and all cached data is
4962 * synced to disk. We are now in a completely consistent state
4963 * which doesn't have anything in the journal, and we know that
4964 * no filesystem updates are running, so it is safe to modify
4965 * the inode's in-core data-journaling state flag now.
4969 EXT4_I(inode
)->i_flags
|= EXT4_JOURNAL_DATA_FL
;
4971 EXT4_I(inode
)->i_flags
&= ~EXT4_JOURNAL_DATA_FL
;
4972 ext4_set_aops(inode
);
4974 jbd2_journal_unlock_updates(journal
);
4976 /* Finally we can mark the inode as dirty. */
4978 handle
= ext4_journal_start(inode
, 1);
4980 return PTR_ERR(handle
);
4982 err
= ext4_mark_inode_dirty(handle
, inode
);
4984 ext4_journal_stop(handle
);
4985 ext4_std_error(inode
->i_sb
, err
);
4990 static int ext4_bh_unmapped(handle_t
*handle
, struct buffer_head
*bh
)
4992 return !buffer_mapped(bh
);
4995 int ext4_page_mkwrite(struct vm_area_struct
*vma
, struct page
*page
)
5001 struct file
*file
= vma
->vm_file
;
5002 struct inode
*inode
= file
->f_path
.dentry
->d_inode
;
5003 struct address_space
*mapping
= inode
->i_mapping
;
5006 * Get i_alloc_sem to stop truncates messing with the inode. We cannot
5007 * get i_mutex because we are already holding mmap_sem.
5009 down_read(&inode
->i_alloc_sem
);
5010 size
= i_size_read(inode
);
5011 if (page
->mapping
!= mapping
|| size
<= page_offset(page
)
5012 || !PageUptodate(page
)) {
5013 /* page got truncated from under us? */
5017 if (PageMappedToDisk(page
))
5020 if (page
->index
== size
>> PAGE_CACHE_SHIFT
)
5021 len
= size
& ~PAGE_CACHE_MASK
;
5023 len
= PAGE_CACHE_SIZE
;
5025 if (page_has_buffers(page
)) {
5026 /* return if we have all the buffers mapped */
5027 if (!walk_page_buffers(NULL
, page_buffers(page
), 0, len
, NULL
,
5032 * OK, we need to fill the hole... Do write_begin write_end
5033 * to do block allocation/reservation.We are not holding
5034 * inode.i__mutex here. That allow * parallel write_begin,
5035 * write_end call. lock_page prevent this from happening
5036 * on the same page though
5038 ret
= mapping
->a_ops
->write_begin(file
, mapping
, page_offset(page
),
5039 len
, AOP_FLAG_UNINTERRUPTIBLE
, &page
, &fsdata
);
5042 ret
= mapping
->a_ops
->write_end(file
, mapping
, page_offset(page
),
5043 len
, len
, page
, fsdata
);
5048 up_read(&inode
->i_alloc_sem
);