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/uio.h>
38 #include <linux/bio.h>
39 #include "ext4_jbd2.h"
42 #include "ext4_extents.h"
44 #define MPAGE_DA_EXTENT_TAIL 0x01
46 static inline int ext4_begin_ordered_truncate(struct inode
*inode
,
49 return jbd2_journal_begin_ordered_truncate(&EXT4_I(inode
)->jinode
,
53 static void ext4_invalidatepage(struct page
*page
, unsigned long offset
);
56 * Test whether an inode is a fast symlink.
58 static int ext4_inode_is_fast_symlink(struct inode
*inode
)
60 int ea_blocks
= EXT4_I(inode
)->i_file_acl
?
61 (inode
->i_sb
->s_blocksize
>> 9) : 0;
63 return (S_ISLNK(inode
->i_mode
) && inode
->i_blocks
- ea_blocks
== 0);
67 * The ext4 forget function must perform a revoke if we are freeing data
68 * which has been journaled. Metadata (eg. indirect blocks) must be
69 * revoked in all cases.
71 * "bh" may be NULL: a metadata block may have been freed from memory
72 * but there may still be a record of it in the journal, and that record
73 * still needs to be revoked.
75 int ext4_forget(handle_t
*handle
, int is_metadata
, struct inode
*inode
,
76 struct buffer_head
*bh
, ext4_fsblk_t blocknr
)
82 BUFFER_TRACE(bh
, "enter");
84 jbd_debug(4, "forgetting bh %p: is_metadata = %d, mode %o, "
86 bh
, is_metadata
, inode
->i_mode
,
87 test_opt(inode
->i_sb
, DATA_FLAGS
));
89 /* Never use the revoke function if we are doing full data
90 * journaling: there is no need to, and a V1 superblock won't
91 * support it. Otherwise, only skip the revoke on un-journaled
94 if (test_opt(inode
->i_sb
, DATA_FLAGS
) == EXT4_MOUNT_JOURNAL_DATA
||
95 (!is_metadata
&& !ext4_should_journal_data(inode
))) {
97 BUFFER_TRACE(bh
, "call jbd2_journal_forget");
98 return ext4_journal_forget(handle
, bh
);
104 * data!=journal && (is_metadata || should_journal_data(inode))
106 BUFFER_TRACE(bh
, "call ext4_journal_revoke");
107 err
= ext4_journal_revoke(handle
, blocknr
, bh
);
109 ext4_abort(inode
->i_sb
, __func__
,
110 "error %d when attempting revoke", err
);
111 BUFFER_TRACE(bh
, "exit");
116 * Work out how many blocks we need to proceed with the next chunk of a
117 * truncate transaction.
119 static unsigned long blocks_for_truncate(struct inode
*inode
)
123 needed
= inode
->i_blocks
>> (inode
->i_sb
->s_blocksize_bits
- 9);
125 /* Give ourselves just enough room to cope with inodes in which
126 * i_blocks is corrupt: we've seen disk corruptions in the past
127 * which resulted in random data in an inode which looked enough
128 * like a regular file for ext4 to try to delete it. Things
129 * will go a bit crazy if that happens, but at least we should
130 * try not to panic the whole kernel. */
134 /* But we need to bound the transaction so we don't overflow the
136 if (needed
> EXT4_MAX_TRANS_DATA
)
137 needed
= EXT4_MAX_TRANS_DATA
;
139 return EXT4_DATA_TRANS_BLOCKS(inode
->i_sb
) + needed
;
143 * Truncate transactions can be complex and absolutely huge. So we need to
144 * be able to restart the transaction at a conventient checkpoint to make
145 * sure we don't overflow the journal.
147 * start_transaction gets us a new handle for a truncate transaction,
148 * and extend_transaction tries to extend the existing one a bit. If
149 * extend fails, we need to propagate the failure up and restart the
150 * transaction in the top-level truncate loop. --sct
152 static handle_t
*start_transaction(struct inode
*inode
)
156 result
= ext4_journal_start(inode
, blocks_for_truncate(inode
));
160 ext4_std_error(inode
->i_sb
, PTR_ERR(result
));
165 * Try to extend this transaction for the purposes of truncation.
167 * Returns 0 if we managed to create more room. If we can't create more
168 * room, and the transaction must be restarted we return 1.
170 static int try_to_extend_transaction(handle_t
*handle
, struct inode
*inode
)
172 if (handle
->h_buffer_credits
> EXT4_RESERVE_TRANS_BLOCKS
)
174 if (!ext4_journal_extend(handle
, blocks_for_truncate(inode
)))
180 * Restart the transaction associated with *handle. This does a commit,
181 * so before we call here everything must be consistently dirtied against
184 static int ext4_journal_test_restart(handle_t
*handle
, struct inode
*inode
)
186 jbd_debug(2, "restarting handle %p\n", handle
);
187 return ext4_journal_restart(handle
, blocks_for_truncate(inode
));
191 * Called at the last iput() if i_nlink is zero.
193 void ext4_delete_inode(struct inode
*inode
)
198 if (ext4_should_order_data(inode
))
199 ext4_begin_ordered_truncate(inode
, 0);
200 truncate_inode_pages(&inode
->i_data
, 0);
202 if (is_bad_inode(inode
))
205 handle
= ext4_journal_start(inode
, blocks_for_truncate(inode
)+3);
206 if (IS_ERR(handle
)) {
207 ext4_std_error(inode
->i_sb
, PTR_ERR(handle
));
209 * If we're going to skip the normal cleanup, we still need to
210 * make sure that the in-core orphan linked list is properly
213 ext4_orphan_del(NULL
, inode
);
220 err
= ext4_mark_inode_dirty(handle
, inode
);
222 ext4_warning(inode
->i_sb
, __func__
,
223 "couldn't mark inode dirty (err %d)", err
);
227 ext4_truncate(inode
);
230 * ext4_ext_truncate() doesn't reserve any slop when it
231 * restarts journal transactions; therefore there may not be
232 * enough credits left in the handle to remove the inode from
233 * the orphan list and set the dtime field.
235 if (handle
->h_buffer_credits
< 3) {
236 err
= ext4_journal_extend(handle
, 3);
238 err
= ext4_journal_restart(handle
, 3);
240 ext4_warning(inode
->i_sb
, __func__
,
241 "couldn't extend journal (err %d)", err
);
243 ext4_journal_stop(handle
);
249 * Kill off the orphan record which ext4_truncate created.
250 * AKPM: I think this can be inside the above `if'.
251 * Note that ext4_orphan_del() has to be able to cope with the
252 * deletion of a non-existent orphan - this is because we don't
253 * know if ext4_truncate() actually created an orphan record.
254 * (Well, we could do this if we need to, but heck - it works)
256 ext4_orphan_del(handle
, inode
);
257 EXT4_I(inode
)->i_dtime
= get_seconds();
260 * One subtle ordering requirement: if anything has gone wrong
261 * (transaction abort, IO errors, whatever), then we can still
262 * do these next steps (the fs will already have been marked as
263 * having errors), but we can't free the inode if the mark_dirty
266 if (ext4_mark_inode_dirty(handle
, inode
))
267 /* If that failed, just do the required in-core inode clear. */
270 ext4_free_inode(handle
, inode
);
271 ext4_journal_stop(handle
);
274 clear_inode(inode
); /* We must guarantee clearing of inode... */
280 struct buffer_head
*bh
;
283 static inline void add_chain(Indirect
*p
, struct buffer_head
*bh
, __le32
*v
)
285 p
->key
= *(p
->p
= v
);
290 * ext4_block_to_path - parse the block number into array of offsets
291 * @inode: inode in question (we are only interested in its superblock)
292 * @i_block: block number to be parsed
293 * @offsets: array to store the offsets in
294 * @boundary: set this non-zero if the referred-to block is likely to be
295 * followed (on disk) by an indirect block.
297 * To store the locations of file's data ext4 uses a data structure common
298 * for UNIX filesystems - tree of pointers anchored in the inode, with
299 * data blocks at leaves and indirect blocks in intermediate nodes.
300 * This function translates the block number into path in that tree -
301 * return value is the path length and @offsets[n] is the offset of
302 * pointer to (n+1)th node in the nth one. If @block is out of range
303 * (negative or too large) warning is printed and zero returned.
305 * Note: function doesn't find node addresses, so no IO is needed. All
306 * we need to know is the capacity of indirect blocks (taken from the
311 * Portability note: the last comparison (check that we fit into triple
312 * indirect block) is spelled differently, because otherwise on an
313 * architecture with 32-bit longs and 8Kb pages we might get into trouble
314 * if our filesystem had 8Kb blocks. We might use long long, but that would
315 * kill us on x86. Oh, well, at least the sign propagation does not matter -
316 * i_block would have to be negative in the very beginning, so we would not
320 static int ext4_block_to_path(struct inode
*inode
,
322 ext4_lblk_t offsets
[4], int *boundary
)
324 int ptrs
= EXT4_ADDR_PER_BLOCK(inode
->i_sb
);
325 int ptrs_bits
= EXT4_ADDR_PER_BLOCK_BITS(inode
->i_sb
);
326 const long direct_blocks
= EXT4_NDIR_BLOCKS
,
327 indirect_blocks
= ptrs
,
328 double_blocks
= (1 << (ptrs_bits
* 2));
333 ext4_warning(inode
->i_sb
, "ext4_block_to_path", "block < 0");
334 } else if (i_block
< direct_blocks
) {
335 offsets
[n
++] = i_block
;
336 final
= direct_blocks
;
337 } else if ((i_block
-= direct_blocks
) < indirect_blocks
) {
338 offsets
[n
++] = EXT4_IND_BLOCK
;
339 offsets
[n
++] = i_block
;
341 } else if ((i_block
-= indirect_blocks
) < double_blocks
) {
342 offsets
[n
++] = EXT4_DIND_BLOCK
;
343 offsets
[n
++] = i_block
>> ptrs_bits
;
344 offsets
[n
++] = i_block
& (ptrs
- 1);
346 } else if (((i_block
-= double_blocks
) >> (ptrs_bits
* 2)) < ptrs
) {
347 offsets
[n
++] = EXT4_TIND_BLOCK
;
348 offsets
[n
++] = i_block
>> (ptrs_bits
* 2);
349 offsets
[n
++] = (i_block
>> ptrs_bits
) & (ptrs
- 1);
350 offsets
[n
++] = i_block
& (ptrs
- 1);
353 ext4_warning(inode
->i_sb
, "ext4_block_to_path",
355 i_block
+ direct_blocks
+
356 indirect_blocks
+ double_blocks
);
359 *boundary
= final
- 1 - (i_block
& (ptrs
- 1));
364 * ext4_get_branch - read the chain of indirect blocks leading to data
365 * @inode: inode in question
366 * @depth: depth of the chain (1 - direct pointer, etc.)
367 * @offsets: offsets of pointers in inode/indirect blocks
368 * @chain: place to store the result
369 * @err: here we store the error value
371 * Function fills the array of triples <key, p, bh> and returns %NULL
372 * if everything went OK or the pointer to the last filled triple
373 * (incomplete one) otherwise. Upon the return chain[i].key contains
374 * the number of (i+1)-th block in the chain (as it is stored in memory,
375 * i.e. little-endian 32-bit), chain[i].p contains the address of that
376 * number (it points into struct inode for i==0 and into the bh->b_data
377 * for i>0) and chain[i].bh points to the buffer_head of i-th indirect
378 * block for i>0 and NULL for i==0. In other words, it holds the block
379 * numbers of the chain, addresses they were taken from (and where we can
380 * verify that chain did not change) and buffer_heads hosting these
383 * Function stops when it stumbles upon zero pointer (absent block)
384 * (pointer to last triple returned, *@err == 0)
385 * or when it gets an IO error reading an indirect block
386 * (ditto, *@err == -EIO)
387 * or when it reads all @depth-1 indirect blocks successfully and finds
388 * the whole chain, all way to the data (returns %NULL, *err == 0).
390 * Need to be called with
391 * down_read(&EXT4_I(inode)->i_data_sem)
393 static Indirect
*ext4_get_branch(struct inode
*inode
, int depth
,
394 ext4_lblk_t
*offsets
,
395 Indirect chain
[4], int *err
)
397 struct super_block
*sb
= inode
->i_sb
;
399 struct buffer_head
*bh
;
402 /* i_data is not going away, no lock needed */
403 add_chain(chain
, NULL
, EXT4_I(inode
)->i_data
+ *offsets
);
407 bh
= sb_bread(sb
, le32_to_cpu(p
->key
));
410 add_chain(++p
, bh
, (__le32
*)bh
->b_data
+ *++offsets
);
424 * ext4_find_near - find a place for allocation with sufficient locality
426 * @ind: descriptor of indirect block.
428 * This function returns the preferred place for block allocation.
429 * It is used when heuristic for sequential allocation fails.
431 * + if there is a block to the left of our position - allocate near it.
432 * + if pointer will live in indirect block - allocate near that block.
433 * + if pointer will live in inode - allocate in the same
436 * In the latter case we colour the starting block by the callers PID to
437 * prevent it from clashing with concurrent allocations for a different inode
438 * in the same block group. The PID is used here so that functionally related
439 * files will be close-by on-disk.
441 * Caller must make sure that @ind is valid and will stay that way.
443 static ext4_fsblk_t
ext4_find_near(struct inode
*inode
, Indirect
*ind
)
445 struct ext4_inode_info
*ei
= EXT4_I(inode
);
446 __le32
*start
= ind
->bh
? (__le32
*) ind
->bh
->b_data
: ei
->i_data
;
448 ext4_fsblk_t bg_start
;
449 ext4_fsblk_t last_block
;
450 ext4_grpblk_t colour
;
452 /* Try to find previous block */
453 for (p
= ind
->p
- 1; p
>= start
; p
--) {
455 return le32_to_cpu(*p
);
458 /* No such thing, so let's try location of indirect block */
460 return ind
->bh
->b_blocknr
;
463 * It is going to be referred to from the inode itself? OK, just put it
464 * into the same cylinder group then.
466 bg_start
= ext4_group_first_block_no(inode
->i_sb
, ei
->i_block_group
);
467 last_block
= ext4_blocks_count(EXT4_SB(inode
->i_sb
)->s_es
) - 1;
469 if (bg_start
+ EXT4_BLOCKS_PER_GROUP(inode
->i_sb
) <= last_block
)
470 colour
= (current
->pid
% 16) *
471 (EXT4_BLOCKS_PER_GROUP(inode
->i_sb
) / 16);
473 colour
= (current
->pid
% 16) * ((last_block
- bg_start
) / 16);
474 return bg_start
+ colour
;
478 * ext4_find_goal - find a preferred place for allocation.
480 * @block: block we want
481 * @partial: pointer to the last triple within a chain
483 * Normally this function find the preferred place for block allocation,
486 static ext4_fsblk_t
ext4_find_goal(struct inode
*inode
, ext4_lblk_t block
,
489 struct ext4_block_alloc_info
*block_i
;
491 block_i
= EXT4_I(inode
)->i_block_alloc_info
;
494 * try the heuristic for sequential allocation,
495 * failing that at least try to get decent locality.
497 if (block_i
&& (block
== block_i
->last_alloc_logical_block
+ 1)
498 && (block_i
->last_alloc_physical_block
!= 0)) {
499 return block_i
->last_alloc_physical_block
+ 1;
502 return ext4_find_near(inode
, partial
);
506 * ext4_blks_to_allocate: Look up the block map and count the number
507 * of direct blocks need to be allocated for the given branch.
509 * @branch: chain of indirect blocks
510 * @k: number of blocks need for indirect blocks
511 * @blks: number of data blocks to be mapped.
512 * @blocks_to_boundary: the offset in the indirect block
514 * return the total number of blocks to be allocate, including the
515 * direct and indirect blocks.
517 static int ext4_blks_to_allocate(Indirect
*branch
, int k
, unsigned long blks
,
518 int blocks_to_boundary
)
520 unsigned long count
= 0;
523 * Simple case, [t,d]Indirect block(s) has not allocated yet
524 * then it's clear blocks on that path have not allocated
527 /* right now we don't handle cross boundary allocation */
528 if (blks
< blocks_to_boundary
+ 1)
531 count
+= blocks_to_boundary
+ 1;
536 while (count
< blks
&& count
<= blocks_to_boundary
&&
537 le32_to_cpu(*(branch
[0].p
+ count
)) == 0) {
544 * ext4_alloc_blocks: multiple allocate blocks needed for a branch
545 * @indirect_blks: the number of blocks need to allocate for indirect
548 * @new_blocks: on return it will store the new block numbers for
549 * the indirect blocks(if needed) and the first direct block,
550 * @blks: on return it will store the total number of allocated
553 static int ext4_alloc_blocks(handle_t
*handle
, struct inode
*inode
,
554 ext4_lblk_t iblock
, ext4_fsblk_t goal
,
555 int indirect_blks
, int blks
,
556 ext4_fsblk_t new_blocks
[4], int *err
)
559 unsigned long count
= 0, blk_allocated
= 0;
561 ext4_fsblk_t current_block
= 0;
565 * Here we try to allocate the requested multiple blocks at once,
566 * on a best-effort basis.
567 * To build a branch, we should allocate blocks for
568 * the indirect blocks(if not allocated yet), and at least
569 * the first direct block of this branch. That's the
570 * minimum number of blocks need to allocate(required)
572 /* first we try to allocate the indirect blocks */
573 target
= indirect_blks
;
576 /* allocating blocks for indirect blocks and direct blocks */
577 current_block
= ext4_new_meta_blocks(handle
, inode
,
583 /* allocate blocks for indirect blocks */
584 while (index
< indirect_blks
&& count
) {
585 new_blocks
[index
++] = current_block
++;
590 * save the new block number
591 * for the first direct block
593 new_blocks
[index
] = current_block
;
594 printk(KERN_INFO
"%s returned more blocks than "
595 "requested\n", __func__
);
601 target
= blks
- count
;
602 blk_allocated
= count
;
605 /* Now allocate data blocks */
607 /* allocating blocks for data blocks */
608 current_block
= ext4_new_blocks(handle
, inode
, iblock
,
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
+= count
;
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 struct ext4_block_alloc_info
*block_i
;
761 ext4_fsblk_t current_block
;
763 block_i
= EXT4_I(inode
)->i_block_alloc_info
;
765 * If we're splicing into a [td]indirect block (as opposed to the
766 * inode) then we need to get write access to the [td]indirect block
770 BUFFER_TRACE(where
->bh
, "get_write_access");
771 err
= ext4_journal_get_write_access(handle
, where
->bh
);
777 *where
->p
= where
->key
;
780 * Update the host buffer_head or inode to point to more just allocated
781 * direct blocks blocks
783 if (num
== 0 && blks
> 1) {
784 current_block
= le32_to_cpu(where
->key
) + 1;
785 for (i
= 1; i
< blks
; i
++)
786 *(where
->p
+ i
) = cpu_to_le32(current_block
++);
790 * update the most recently allocated logical & physical block
791 * in i_block_alloc_info, to assist find the proper goal block for next
795 block_i
->last_alloc_logical_block
= block
+ blks
- 1;
796 block_i
->last_alloc_physical_block
=
797 le32_to_cpu(where
[num
].key
) + blks
- 1;
800 /* We are done with atomic stuff, now do the rest of housekeeping */
802 inode
->i_ctime
= ext4_current_time(inode
);
803 ext4_mark_inode_dirty(handle
, inode
);
805 /* had we spliced it onto indirect block? */
808 * If we spliced it onto an indirect block, we haven't
809 * altered the inode. Note however that if it is being spliced
810 * onto an indirect block at the very end of the file (the
811 * file is growing) then we *will* alter the inode to reflect
812 * the new i_size. But that is not done here - it is done in
813 * generic_commit_write->__mark_inode_dirty->ext4_dirty_inode.
815 jbd_debug(5, "splicing indirect only\n");
816 BUFFER_TRACE(where
->bh
, "call ext4_journal_dirty_metadata");
817 err
= ext4_journal_dirty_metadata(handle
, where
->bh
);
822 * OK, we spliced it into the inode itself on a direct block.
823 * Inode was dirtied above.
825 jbd_debug(5, "splicing direct\n");
830 for (i
= 1; i
<= num
; i
++) {
831 BUFFER_TRACE(where
[i
].bh
, "call jbd2_journal_forget");
832 ext4_journal_forget(handle
, where
[i
].bh
);
833 ext4_free_blocks(handle
, inode
,
834 le32_to_cpu(where
[i
-1].key
), 1, 0);
836 ext4_free_blocks(handle
, inode
, le32_to_cpu(where
[num
].key
), blks
, 0);
842 * Allocation strategy is simple: if we have to allocate something, we will
843 * have to go the whole way to leaf. So let's do it before attaching anything
844 * to tree, set linkage between the newborn blocks, write them if sync is
845 * required, recheck the path, free and repeat if check fails, otherwise
846 * set the last missing link (that will protect us from any truncate-generated
847 * removals - all blocks on the path are immune now) and possibly force the
848 * write on the parent block.
849 * That has a nice additional property: no special recovery from the failed
850 * allocations is needed - we simply release blocks and do not touch anything
851 * reachable from inode.
853 * `handle' can be NULL if create == 0.
855 * return > 0, # of blocks mapped or allocated.
856 * return = 0, if plain lookup failed.
857 * return < 0, error case.
860 * Need to be called with
861 * down_read(&EXT4_I(inode)->i_data_sem) if not allocating file system block
862 * (ie, create is zero). Otherwise down_write(&EXT4_I(inode)->i_data_sem)
864 int ext4_get_blocks_handle(handle_t
*handle
, struct inode
*inode
,
865 ext4_lblk_t iblock
, unsigned long maxblocks
,
866 struct buffer_head
*bh_result
,
867 int create
, int extend_disksize
)
870 ext4_lblk_t offsets
[4];
875 int blocks_to_boundary
= 0;
877 struct ext4_inode_info
*ei
= EXT4_I(inode
);
879 ext4_fsblk_t first_block
= 0;
883 J_ASSERT(!(EXT4_I(inode
)->i_flags
& EXT4_EXTENTS_FL
));
884 J_ASSERT(handle
!= NULL
|| create
== 0);
885 depth
= ext4_block_to_path(inode
, iblock
, offsets
,
886 &blocks_to_boundary
);
891 partial
= ext4_get_branch(inode
, depth
, offsets
, chain
, &err
);
893 /* Simplest case - block found, no allocation needed */
895 first_block
= le32_to_cpu(chain
[depth
- 1].key
);
896 clear_buffer_new(bh_result
);
899 while (count
< maxblocks
&& count
<= blocks_to_boundary
) {
902 blk
= le32_to_cpu(*(chain
[depth
-1].p
+ count
));
904 if (blk
== first_block
+ count
)
912 /* Next simple case - plain lookup or failed read of indirect block */
913 if (!create
|| err
== -EIO
)
917 * Okay, we need to do block allocation. Lazily initialize the block
918 * allocation info here if necessary
920 if (S_ISREG(inode
->i_mode
) && (!ei
->i_block_alloc_info
))
921 ext4_init_block_alloc_info(inode
);
923 goal
= ext4_find_goal(inode
, iblock
, partial
);
925 /* the number of blocks need to allocate for [d,t]indirect blocks */
926 indirect_blks
= (chain
+ depth
) - partial
- 1;
929 * Next look up the indirect map to count the totoal number of
930 * direct blocks to allocate for this branch.
932 count
= ext4_blks_to_allocate(partial
, indirect_blks
,
933 maxblocks
, blocks_to_boundary
);
935 * Block out ext4_truncate while we alter the tree
937 err
= ext4_alloc_branch(handle
, inode
, iblock
, indirect_blks
,
939 offsets
+ (partial
- chain
), partial
);
942 * The ext4_splice_branch call will free and forget any buffers
943 * on the new chain if there is a failure, but that risks using
944 * up transaction credits, especially for bitmaps where the
945 * credits cannot be returned. Can we handle this somehow? We
946 * may need to return -EAGAIN upwards in the worst case. --sct
949 err
= ext4_splice_branch(handle
, inode
, iblock
,
950 partial
, indirect_blks
, count
);
952 * i_disksize growing is protected by i_data_sem. Don't forget to
953 * protect it if you're about to implement concurrent
954 * ext4_get_block() -bzzz
956 if (!err
&& extend_disksize
) {
957 disksize
= ((loff_t
) iblock
+ count
) << inode
->i_blkbits
;
958 if (disksize
> i_size_read(inode
))
959 disksize
= i_size_read(inode
);
960 if (disksize
> ei
->i_disksize
)
961 ei
->i_disksize
= disksize
;
966 set_buffer_new(bh_result
);
968 map_bh(bh_result
, inode
->i_sb
, le32_to_cpu(chain
[depth
-1].key
));
969 if (count
> blocks_to_boundary
)
970 set_buffer_boundary(bh_result
);
972 /* Clean up and exit */
973 partial
= chain
+ depth
- 1; /* the whole chain */
975 while (partial
> chain
) {
976 BUFFER_TRACE(partial
->bh
, "call brelse");
980 BUFFER_TRACE(bh_result
, "returned");
986 * Calculate the number of metadata blocks need to reserve
987 * to allocate @blocks for non extent file based file
989 static int ext4_indirect_calc_metadata_amount(struct inode
*inode
, int blocks
)
991 int icap
= EXT4_ADDR_PER_BLOCK(inode
->i_sb
);
992 int ind_blks
, dind_blks
, tind_blks
;
994 /* number of new indirect blocks needed */
995 ind_blks
= (blocks
+ icap
- 1) / icap
;
997 dind_blks
= (ind_blks
+ icap
- 1) / icap
;
1001 return ind_blks
+ dind_blks
+ tind_blks
;
1005 * Calculate the number of metadata blocks need to reserve
1006 * to allocate given number of blocks
1008 static int ext4_calc_metadata_amount(struct inode
*inode
, int blocks
)
1013 if (EXT4_I(inode
)->i_flags
& EXT4_EXTENTS_FL
)
1014 return ext4_ext_calc_metadata_amount(inode
, blocks
);
1016 return ext4_indirect_calc_metadata_amount(inode
, blocks
);
1019 static void ext4_da_update_reserve_space(struct inode
*inode
, int used
)
1021 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
1022 int total
, mdb
, mdb_free
;
1024 spin_lock(&EXT4_I(inode
)->i_block_reservation_lock
);
1025 /* recalculate the number of metablocks still need to be reserved */
1026 total
= EXT4_I(inode
)->i_reserved_data_blocks
- used
;
1027 mdb
= ext4_calc_metadata_amount(inode
, total
);
1029 /* figure out how many metablocks to release */
1030 BUG_ON(mdb
> EXT4_I(inode
)->i_reserved_meta_blocks
);
1031 mdb_free
= EXT4_I(inode
)->i_reserved_meta_blocks
- mdb
;
1034 /* Account for allocated meta_blocks */
1035 mdb_free
-= EXT4_I(inode
)->i_allocated_meta_blocks
;
1037 /* update fs dirty blocks counter */
1038 percpu_counter_sub(&sbi
->s_dirtyblocks_counter
, mdb_free
);
1039 EXT4_I(inode
)->i_allocated_meta_blocks
= 0;
1040 EXT4_I(inode
)->i_reserved_meta_blocks
= mdb
;
1043 /* update per-inode reservations */
1044 BUG_ON(used
> EXT4_I(inode
)->i_reserved_data_blocks
);
1045 EXT4_I(inode
)->i_reserved_data_blocks
-= used
;
1047 spin_unlock(&EXT4_I(inode
)->i_block_reservation_lock
);
1051 * The ext4_get_blocks_wrap() function try to look up the requested blocks,
1052 * and returns if the blocks are already mapped.
1054 * Otherwise it takes the write lock of the i_data_sem and allocate blocks
1055 * and store the allocated blocks in the result buffer head and mark it
1058 * If file type is extents based, it will call ext4_ext_get_blocks(),
1059 * Otherwise, call with ext4_get_blocks_handle() to handle indirect mapping
1062 * On success, it returns the number of blocks being mapped or allocate.
1063 * if create==0 and the blocks are pre-allocated and uninitialized block,
1064 * the result buffer head is unmapped. If the create ==1, it will make sure
1065 * the buffer head is mapped.
1067 * It returns 0 if plain look up failed (blocks have not been allocated), in
1068 * that casem, buffer head is unmapped
1070 * It returns the error in case of allocation failure.
1072 int ext4_get_blocks_wrap(handle_t
*handle
, struct inode
*inode
, sector_t block
,
1073 unsigned long max_blocks
, struct buffer_head
*bh
,
1074 int create
, int extend_disksize
, int flag
)
1078 clear_buffer_mapped(bh
);
1081 * Try to see if we can get the block without requesting
1082 * for new file system block.
1084 down_read((&EXT4_I(inode
)->i_data_sem
));
1085 if (EXT4_I(inode
)->i_flags
& EXT4_EXTENTS_FL
) {
1086 retval
= ext4_ext_get_blocks(handle
, inode
, block
, max_blocks
,
1089 retval
= ext4_get_blocks_handle(handle
,
1090 inode
, block
, max_blocks
, bh
, 0, 0);
1092 up_read((&EXT4_I(inode
)->i_data_sem
));
1094 /* If it is only a block(s) look up */
1099 * Returns if the blocks have already allocated
1101 * Note that if blocks have been preallocated
1102 * ext4_ext_get_block() returns th create = 0
1103 * with buffer head unmapped.
1105 if (retval
> 0 && buffer_mapped(bh
))
1109 * New blocks allocate and/or writing to uninitialized extent
1110 * will possibly result in updating i_data, so we take
1111 * the write lock of i_data_sem, and call get_blocks()
1112 * with create == 1 flag.
1114 down_write((&EXT4_I(inode
)->i_data_sem
));
1117 * if the caller is from delayed allocation writeout path
1118 * we have already reserved fs blocks for allocation
1119 * let the underlying get_block() function know to
1120 * avoid double accounting
1123 EXT4_I(inode
)->i_delalloc_reserved_flag
= 1;
1125 * We need to check for EXT4 here because migrate
1126 * could have changed the inode type in between
1128 if (EXT4_I(inode
)->i_flags
& EXT4_EXTENTS_FL
) {
1129 retval
= ext4_ext_get_blocks(handle
, inode
, block
, max_blocks
,
1130 bh
, create
, extend_disksize
);
1132 retval
= ext4_get_blocks_handle(handle
, inode
, block
,
1133 max_blocks
, bh
, create
, extend_disksize
);
1135 if (retval
> 0 && buffer_new(bh
)) {
1137 * We allocated new blocks which will result in
1138 * i_data's format changing. Force the migrate
1139 * to fail by clearing migrate flags
1141 EXT4_I(inode
)->i_flags
= EXT4_I(inode
)->i_flags
&
1147 EXT4_I(inode
)->i_delalloc_reserved_flag
= 0;
1149 * Update reserved blocks/metadata blocks
1150 * after successful block allocation
1151 * which were deferred till now
1153 if ((retval
> 0) && buffer_delay(bh
))
1154 ext4_da_update_reserve_space(inode
, retval
);
1157 up_write((&EXT4_I(inode
)->i_data_sem
));
1161 /* Maximum number of blocks we map for direct IO at once. */
1162 #define DIO_MAX_BLOCKS 4096
1164 static int ext4_get_block(struct inode
*inode
, sector_t iblock
,
1165 struct buffer_head
*bh_result
, int create
)
1167 handle_t
*handle
= ext4_journal_current_handle();
1168 int ret
= 0, started
= 0;
1169 unsigned max_blocks
= bh_result
->b_size
>> inode
->i_blkbits
;
1172 if (create
&& !handle
) {
1173 /* Direct IO write... */
1174 if (max_blocks
> DIO_MAX_BLOCKS
)
1175 max_blocks
= DIO_MAX_BLOCKS
;
1176 dio_credits
= ext4_chunk_trans_blocks(inode
, max_blocks
);
1177 handle
= ext4_journal_start(inode
, dio_credits
);
1178 if (IS_ERR(handle
)) {
1179 ret
= PTR_ERR(handle
);
1185 ret
= ext4_get_blocks_wrap(handle
, inode
, iblock
,
1186 max_blocks
, bh_result
, create
, 0, 0);
1188 bh_result
->b_size
= (ret
<< inode
->i_blkbits
);
1192 ext4_journal_stop(handle
);
1198 * `handle' can be NULL if create is zero
1200 struct buffer_head
*ext4_getblk(handle_t
*handle
, struct inode
*inode
,
1201 ext4_lblk_t block
, int create
, int *errp
)
1203 struct buffer_head dummy
;
1206 J_ASSERT(handle
!= NULL
|| create
== 0);
1209 dummy
.b_blocknr
= -1000;
1210 buffer_trace_init(&dummy
.b_history
);
1211 err
= ext4_get_blocks_wrap(handle
, inode
, block
, 1,
1212 &dummy
, create
, 1, 0);
1214 * ext4_get_blocks_handle() returns number of blocks
1215 * mapped. 0 in case of a HOLE.
1223 if (!err
&& buffer_mapped(&dummy
)) {
1224 struct buffer_head
*bh
;
1225 bh
= sb_getblk(inode
->i_sb
, dummy
.b_blocknr
);
1230 if (buffer_new(&dummy
)) {
1231 J_ASSERT(create
!= 0);
1232 J_ASSERT(handle
!= NULL
);
1235 * Now that we do not always journal data, we should
1236 * keep in mind whether this should always journal the
1237 * new buffer as metadata. For now, regular file
1238 * writes use ext4_get_block instead, so it's not a
1242 BUFFER_TRACE(bh
, "call get_create_access");
1243 fatal
= ext4_journal_get_create_access(handle
, bh
);
1244 if (!fatal
&& !buffer_uptodate(bh
)) {
1245 memset(bh
->b_data
, 0, inode
->i_sb
->s_blocksize
);
1246 set_buffer_uptodate(bh
);
1249 BUFFER_TRACE(bh
, "call ext4_journal_dirty_metadata");
1250 err
= ext4_journal_dirty_metadata(handle
, bh
);
1254 BUFFER_TRACE(bh
, "not a new buffer");
1267 struct buffer_head
*ext4_bread(handle_t
*handle
, struct inode
*inode
,
1268 ext4_lblk_t block
, int create
, int *err
)
1270 struct buffer_head
*bh
;
1272 bh
= ext4_getblk(handle
, inode
, block
, create
, err
);
1275 if (buffer_uptodate(bh
))
1277 ll_rw_block(READ_META
, 1, &bh
);
1279 if (buffer_uptodate(bh
))
1286 static int walk_page_buffers(handle_t
*handle
,
1287 struct buffer_head
*head
,
1291 int (*fn
)(handle_t
*handle
,
1292 struct buffer_head
*bh
))
1294 struct buffer_head
*bh
;
1295 unsigned block_start
, block_end
;
1296 unsigned blocksize
= head
->b_size
;
1298 struct buffer_head
*next
;
1300 for (bh
= head
, block_start
= 0;
1301 ret
== 0 && (bh
!= head
|| !block_start
);
1302 block_start
= block_end
, bh
= next
)
1304 next
= bh
->b_this_page
;
1305 block_end
= block_start
+ blocksize
;
1306 if (block_end
<= from
|| block_start
>= to
) {
1307 if (partial
&& !buffer_uptodate(bh
))
1311 err
= (*fn
)(handle
, bh
);
1319 * To preserve ordering, it is essential that the hole instantiation and
1320 * the data write be encapsulated in a single transaction. We cannot
1321 * close off a transaction and start a new one between the ext4_get_block()
1322 * and the commit_write(). So doing the jbd2_journal_start at the start of
1323 * prepare_write() is the right place.
1325 * Also, this function can nest inside ext4_writepage() ->
1326 * block_write_full_page(). In that case, we *know* that ext4_writepage()
1327 * has generated enough buffer credits to do the whole page. So we won't
1328 * block on the journal in that case, which is good, because the caller may
1331 * By accident, ext4 can be reentered when a transaction is open via
1332 * quota file writes. If we were to commit the transaction while thus
1333 * reentered, there can be a deadlock - we would be holding a quota
1334 * lock, and the commit would never complete if another thread had a
1335 * transaction open and was blocking on the quota lock - a ranking
1338 * So what we do is to rely on the fact that jbd2_journal_stop/journal_start
1339 * will _not_ run commit under these circumstances because handle->h_ref
1340 * is elevated. We'll still have enough credits for the tiny quotafile
1343 static int do_journal_get_write_access(handle_t
*handle
,
1344 struct buffer_head
*bh
)
1346 if (!buffer_mapped(bh
) || buffer_freed(bh
))
1348 return ext4_journal_get_write_access(handle
, bh
);
1351 static int ext4_write_begin(struct file
*file
, struct address_space
*mapping
,
1352 loff_t pos
, unsigned len
, unsigned flags
,
1353 struct page
**pagep
, void **fsdata
)
1355 struct inode
*inode
= mapping
->host
;
1356 int ret
, needed_blocks
= ext4_writepage_trans_blocks(inode
);
1363 index
= pos
>> PAGE_CACHE_SHIFT
;
1364 from
= pos
& (PAGE_CACHE_SIZE
- 1);
1368 handle
= ext4_journal_start(inode
, needed_blocks
);
1369 if (IS_ERR(handle
)) {
1370 ret
= PTR_ERR(handle
);
1374 page
= __grab_cache_page(mapping
, index
);
1376 ext4_journal_stop(handle
);
1382 ret
= block_write_begin(file
, mapping
, pos
, len
, flags
, pagep
, fsdata
,
1385 if (!ret
&& ext4_should_journal_data(inode
)) {
1386 ret
= walk_page_buffers(handle
, page_buffers(page
),
1387 from
, to
, NULL
, do_journal_get_write_access
);
1392 ext4_journal_stop(handle
);
1393 page_cache_release(page
);
1396 if (ret
== -ENOSPC
&& ext4_should_retry_alloc(inode
->i_sb
, &retries
))
1402 /* For write_end() in data=journal mode */
1403 static int write_end_fn(handle_t
*handle
, struct buffer_head
*bh
)
1405 if (!buffer_mapped(bh
) || buffer_freed(bh
))
1407 set_buffer_uptodate(bh
);
1408 return ext4_journal_dirty_metadata(handle
, bh
);
1412 * We need to pick up the new inode size which generic_commit_write gave us
1413 * `file' can be NULL - eg, when called from page_symlink().
1415 * ext4 never places buffers on inode->i_mapping->private_list. metadata
1416 * buffers are managed internally.
1418 static int ext4_ordered_write_end(struct file
*file
,
1419 struct address_space
*mapping
,
1420 loff_t pos
, unsigned len
, unsigned copied
,
1421 struct page
*page
, void *fsdata
)
1423 handle_t
*handle
= ext4_journal_current_handle();
1424 struct inode
*inode
= mapping
->host
;
1427 ret
= ext4_jbd2_file_inode(handle
, inode
);
1431 * generic_write_end() will run mark_inode_dirty() if i_size
1432 * changes. So let's piggyback the i_disksize mark_inode_dirty
1437 new_i_size
= pos
+ copied
;
1438 if (new_i_size
> EXT4_I(inode
)->i_disksize
)
1439 EXT4_I(inode
)->i_disksize
= new_i_size
;
1440 ret2
= generic_write_end(file
, mapping
, pos
, len
, copied
,
1446 ret2
= ext4_journal_stop(handle
);
1450 return ret
? ret
: copied
;
1453 static int ext4_writeback_write_end(struct file
*file
,
1454 struct address_space
*mapping
,
1455 loff_t pos
, unsigned len
, unsigned copied
,
1456 struct page
*page
, void *fsdata
)
1458 handle_t
*handle
= ext4_journal_current_handle();
1459 struct inode
*inode
= mapping
->host
;
1463 new_i_size
= pos
+ copied
;
1464 if (new_i_size
> EXT4_I(inode
)->i_disksize
)
1465 EXT4_I(inode
)->i_disksize
= new_i_size
;
1467 ret2
= generic_write_end(file
, mapping
, pos
, len
, copied
,
1473 ret2
= ext4_journal_stop(handle
);
1477 return ret
? ret
: copied
;
1480 static int ext4_journalled_write_end(struct file
*file
,
1481 struct address_space
*mapping
,
1482 loff_t pos
, unsigned len
, unsigned copied
,
1483 struct page
*page
, void *fsdata
)
1485 handle_t
*handle
= ext4_journal_current_handle();
1486 struct inode
*inode
= mapping
->host
;
1491 from
= pos
& (PAGE_CACHE_SIZE
- 1);
1495 if (!PageUptodate(page
))
1497 page_zero_new_buffers(page
, from
+copied
, to
);
1500 ret
= walk_page_buffers(handle
, page_buffers(page
), from
,
1501 to
, &partial
, write_end_fn
);
1503 SetPageUptodate(page
);
1504 if (pos
+copied
> inode
->i_size
)
1505 i_size_write(inode
, pos
+copied
);
1506 EXT4_I(inode
)->i_state
|= EXT4_STATE_JDATA
;
1507 if (inode
->i_size
> EXT4_I(inode
)->i_disksize
) {
1508 EXT4_I(inode
)->i_disksize
= inode
->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
)
1656 struct address_space
*mapping
= mpd
->inode
->i_mapping
;
1657 int ret
= 0, err
, nr_pages
, i
;
1658 unsigned long index
, end
;
1659 struct pagevec pvec
;
1661 BUG_ON(mpd
->next_page
<= mpd
->first_page
);
1662 pagevec_init(&pvec
, 0);
1663 index
= mpd
->first_page
;
1664 end
= mpd
->next_page
- 1;
1666 while (index
<= end
) {
1667 /* XXX: optimize tail */
1668 nr_pages
= pagevec_lookup(&pvec
, mapping
, index
, PAGEVEC_SIZE
);
1671 for (i
= 0; i
< nr_pages
; i
++) {
1672 struct page
*page
= pvec
.pages
[i
];
1674 index
= page
->index
;
1679 err
= mapping
->a_ops
->writepage(page
, mpd
->wbc
);
1681 mpd
->pages_written
++;
1683 * In error case, we have to continue because
1684 * remaining pages are still locked
1685 * XXX: unlock and re-dirty them?
1690 pagevec_release(&pvec
);
1696 * mpage_put_bnr_to_bhs - walk blocks and assign them actual numbers
1698 * @mpd->inode - inode to walk through
1699 * @exbh->b_blocknr - first block on a disk
1700 * @exbh->b_size - amount of space in bytes
1701 * @logical - first logical block to start assignment with
1703 * the function goes through all passed space and put actual disk
1704 * block numbers into buffer heads, dropping BH_Delay
1706 static void mpage_put_bnr_to_bhs(struct mpage_da_data
*mpd
, sector_t logical
,
1707 struct buffer_head
*exbh
)
1709 struct inode
*inode
= mpd
->inode
;
1710 struct address_space
*mapping
= inode
->i_mapping
;
1711 int blocks
= exbh
->b_size
>> inode
->i_blkbits
;
1712 sector_t pblock
= exbh
->b_blocknr
, cur_logical
;
1713 struct buffer_head
*head
, *bh
;
1715 struct pagevec pvec
;
1718 index
= logical
>> (PAGE_CACHE_SHIFT
- inode
->i_blkbits
);
1719 end
= (logical
+ blocks
- 1) >> (PAGE_CACHE_SHIFT
- inode
->i_blkbits
);
1720 cur_logical
= index
<< (PAGE_CACHE_SHIFT
- inode
->i_blkbits
);
1722 pagevec_init(&pvec
, 0);
1724 while (index
<= end
) {
1725 /* XXX: optimize tail */
1726 nr_pages
= pagevec_lookup(&pvec
, mapping
, index
, PAGEVEC_SIZE
);
1729 for (i
= 0; i
< nr_pages
; i
++) {
1730 struct page
*page
= pvec
.pages
[i
];
1732 index
= page
->index
;
1737 BUG_ON(!PageLocked(page
));
1738 BUG_ON(PageWriteback(page
));
1739 BUG_ON(!page_has_buffers(page
));
1741 bh
= page_buffers(page
);
1744 /* skip blocks out of the range */
1746 if (cur_logical
>= logical
)
1749 } while ((bh
= bh
->b_this_page
) != head
);
1752 if (cur_logical
>= logical
+ blocks
)
1754 if (buffer_delay(bh
)) {
1755 bh
->b_blocknr
= pblock
;
1756 clear_buffer_delay(bh
);
1757 bh
->b_bdev
= inode
->i_sb
->s_bdev
;
1758 } else if (buffer_unwritten(bh
)) {
1759 bh
->b_blocknr
= pblock
;
1760 clear_buffer_unwritten(bh
);
1761 set_buffer_mapped(bh
);
1763 bh
->b_bdev
= inode
->i_sb
->s_bdev
;
1764 } else if (buffer_mapped(bh
))
1765 BUG_ON(bh
->b_blocknr
!= pblock
);
1769 } while ((bh
= bh
->b_this_page
) != head
);
1771 pagevec_release(&pvec
);
1777 * __unmap_underlying_blocks - just a helper function to unmap
1778 * set of blocks described by @bh
1780 static inline void __unmap_underlying_blocks(struct inode
*inode
,
1781 struct buffer_head
*bh
)
1783 struct block_device
*bdev
= inode
->i_sb
->s_bdev
;
1786 blocks
= bh
->b_size
>> inode
->i_blkbits
;
1787 for (i
= 0; i
< blocks
; i
++)
1788 unmap_underlying_metadata(bdev
, bh
->b_blocknr
+ i
);
1791 static void ext4_da_block_invalidatepages(struct mpage_da_data
*mpd
,
1792 sector_t logical
, long blk_cnt
)
1796 struct pagevec pvec
;
1797 struct inode
*inode
= mpd
->inode
;
1798 struct address_space
*mapping
= inode
->i_mapping
;
1800 index
= logical
>> (PAGE_CACHE_SHIFT
- inode
->i_blkbits
);
1801 end
= (logical
+ blk_cnt
- 1) >>
1802 (PAGE_CACHE_SHIFT
- inode
->i_blkbits
);
1803 while (index
<= end
) {
1804 nr_pages
= pagevec_lookup(&pvec
, mapping
, index
, PAGEVEC_SIZE
);
1807 for (i
= 0; i
< nr_pages
; i
++) {
1808 struct page
*page
= pvec
.pages
[i
];
1809 index
= page
->index
;
1814 BUG_ON(!PageLocked(page
));
1815 BUG_ON(PageWriteback(page
));
1816 block_invalidatepage(page
, 0);
1817 ClearPageUptodate(page
);
1824 static void ext4_print_free_blocks(struct inode
*inode
)
1826 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
1827 printk(KERN_EMERG
"Total free blocks count %lld\n",
1828 ext4_count_free_blocks(inode
->i_sb
));
1829 printk(KERN_EMERG
"Free/Dirty block details\n");
1830 printk(KERN_EMERG
"free_blocks=%lld\n",
1831 percpu_counter_sum(&sbi
->s_freeblocks_counter
));
1832 printk(KERN_EMERG
"dirty_blocks=%lld\n",
1833 percpu_counter_sum(&sbi
->s_dirtyblocks_counter
));
1834 printk(KERN_EMERG
"Block reservation details\n");
1835 printk(KERN_EMERG
"i_reserved_data_blocks=%lu\n",
1836 EXT4_I(inode
)->i_reserved_data_blocks
);
1837 printk(KERN_EMERG
"i_reserved_meta_blocks=%lu\n",
1838 EXT4_I(inode
)->i_reserved_meta_blocks
);
1843 * mpage_da_map_blocks - go through given space
1845 * @mpd->lbh - bh describing space
1846 * @mpd->get_block - the filesystem's block mapper function
1848 * The function skips space we know is already mapped to disk blocks.
1851 static int mpage_da_map_blocks(struct mpage_da_data
*mpd
)
1854 struct buffer_head
new;
1855 struct buffer_head
*lbh
= &mpd
->lbh
;
1859 * We consider only non-mapped and non-allocated blocks
1861 if (buffer_mapped(lbh
) && !buffer_delay(lbh
))
1863 new.b_state
= lbh
->b_state
;
1865 new.b_size
= lbh
->b_size
;
1866 next
= lbh
->b_blocknr
;
1868 * If we didn't accumulate anything
1869 * to write simply return
1873 err
= mpd
->get_block(mpd
->inode
, next
, &new, 1);
1876 /* If get block returns with error
1877 * we simply return. Later writepage
1878 * will redirty the page and writepages
1879 * will find the dirty page again
1884 if (err
== -ENOSPC
&&
1885 ext4_count_free_blocks(mpd
->inode
->i_sb
)) {
1891 * get block failure will cause us
1892 * to loop in writepages. Because
1893 * a_ops->writepage won't be able to
1894 * make progress. The page will be redirtied
1895 * by writepage and writepages will again
1896 * try to write the same.
1898 printk(KERN_EMERG
"%s block allocation failed for inode %lu "
1899 "at logical offset %llu with max blocks "
1900 "%zd with error %d\n",
1901 __func__
, mpd
->inode
->i_ino
,
1902 (unsigned long long)next
,
1903 lbh
->b_size
>> mpd
->inode
->i_blkbits
, err
);
1904 printk(KERN_EMERG
"This should not happen.!! "
1905 "Data will be lost\n");
1906 if (err
== -ENOSPC
) {
1907 ext4_print_free_blocks(mpd
->inode
);
1909 /* invlaidate all the pages */
1910 ext4_da_block_invalidatepages(mpd
, next
,
1911 lbh
->b_size
>> mpd
->inode
->i_blkbits
);
1914 BUG_ON(new.b_size
== 0);
1916 if (buffer_new(&new))
1917 __unmap_underlying_blocks(mpd
->inode
, &new);
1920 * If blocks are delayed marked, we need to
1921 * put actual blocknr and drop delayed bit
1923 if (buffer_delay(lbh
) || buffer_unwritten(lbh
))
1924 mpage_put_bnr_to_bhs(mpd
, next
, &new);
1929 #define BH_FLAGS ((1 << BH_Uptodate) | (1 << BH_Mapped) | \
1930 (1 << BH_Delay) | (1 << BH_Unwritten))
1933 * mpage_add_bh_to_extent - try to add one more block to extent of blocks
1935 * @mpd->lbh - extent of blocks
1936 * @logical - logical number of the block in the file
1937 * @bh - bh of the block (used to access block's state)
1939 * the function is used to collect contig. blocks in same state
1941 static void mpage_add_bh_to_extent(struct mpage_da_data
*mpd
,
1942 sector_t logical
, struct buffer_head
*bh
)
1945 size_t b_size
= bh
->b_size
;
1946 struct buffer_head
*lbh
= &mpd
->lbh
;
1947 int nrblocks
= lbh
->b_size
>> mpd
->inode
->i_blkbits
;
1949 /* check if thereserved journal credits might overflow */
1950 if (!(EXT4_I(mpd
->inode
)->i_flags
& EXT4_EXTENTS_FL
)) {
1951 if (nrblocks
>= EXT4_MAX_TRANS_DATA
) {
1953 * With non-extent format we are limited by the journal
1954 * credit available. Total credit needed to insert
1955 * nrblocks contiguous blocks is dependent on the
1956 * nrblocks. So limit nrblocks.
1959 } else if ((nrblocks
+ (b_size
>> mpd
->inode
->i_blkbits
)) >
1960 EXT4_MAX_TRANS_DATA
) {
1962 * Adding the new buffer_head would make it cross the
1963 * allowed limit for which we have journal credit
1964 * reserved. So limit the new bh->b_size
1966 b_size
= (EXT4_MAX_TRANS_DATA
- nrblocks
) <<
1967 mpd
->inode
->i_blkbits
;
1968 /* we will do mpage_da_submit_io in the next loop */
1972 * First block in the extent
1974 if (lbh
->b_size
== 0) {
1975 lbh
->b_blocknr
= logical
;
1976 lbh
->b_size
= b_size
;
1977 lbh
->b_state
= bh
->b_state
& BH_FLAGS
;
1981 next
= lbh
->b_blocknr
+ nrblocks
;
1983 * Can we merge the block to our big extent?
1985 if (logical
== next
&& (bh
->b_state
& BH_FLAGS
) == lbh
->b_state
) {
1986 lbh
->b_size
+= b_size
;
1992 * We couldn't merge the block to our extent, so we
1993 * need to flush current extent and start new one
1995 if (mpage_da_map_blocks(mpd
) == 0)
1996 mpage_da_submit_io(mpd
);
2002 * __mpage_da_writepage - finds extent of pages and blocks
2004 * @page: page to consider
2005 * @wbc: not used, we just follow rules
2008 * The function finds extents of pages and scan them for all blocks.
2010 static int __mpage_da_writepage(struct page
*page
,
2011 struct writeback_control
*wbc
, void *data
)
2013 struct mpage_da_data
*mpd
= data
;
2014 struct inode
*inode
= mpd
->inode
;
2015 struct buffer_head
*bh
, *head
, fake
;
2020 * Rest of the page in the page_vec
2021 * redirty then and skip then. We will
2022 * try to to write them again after
2023 * starting a new transaction
2025 redirty_page_for_writepage(wbc
, page
);
2027 return MPAGE_DA_EXTENT_TAIL
;
2030 * Can we merge this page to current extent?
2032 if (mpd
->next_page
!= page
->index
) {
2034 * Nope, we can't. So, we map non-allocated blocks
2035 * and start IO on them using writepage()
2037 if (mpd
->next_page
!= mpd
->first_page
) {
2038 if (mpage_da_map_blocks(mpd
) == 0)
2039 mpage_da_submit_io(mpd
);
2041 * skip rest of the page in the page_vec
2044 redirty_page_for_writepage(wbc
, page
);
2046 return MPAGE_DA_EXTENT_TAIL
;
2050 * Start next extent of pages ...
2052 mpd
->first_page
= page
->index
;
2057 mpd
->lbh
.b_size
= 0;
2058 mpd
->lbh
.b_state
= 0;
2059 mpd
->lbh
.b_blocknr
= 0;
2062 mpd
->next_page
= page
->index
+ 1;
2063 logical
= (sector_t
) page
->index
<<
2064 (PAGE_CACHE_SHIFT
- inode
->i_blkbits
);
2066 if (!page_has_buffers(page
)) {
2068 * There is no attached buffer heads yet (mmap?)
2069 * we treat the page asfull of dirty blocks
2072 bh
->b_size
= PAGE_CACHE_SIZE
;
2074 set_buffer_dirty(bh
);
2075 set_buffer_uptodate(bh
);
2076 mpage_add_bh_to_extent(mpd
, logical
, bh
);
2078 return MPAGE_DA_EXTENT_TAIL
;
2081 * Page with regular buffer heads, just add all dirty ones
2083 head
= page_buffers(page
);
2086 BUG_ON(buffer_locked(bh
));
2087 if (buffer_dirty(bh
) &&
2088 (!buffer_mapped(bh
) || buffer_delay(bh
))) {
2089 mpage_add_bh_to_extent(mpd
, logical
, bh
);
2091 return MPAGE_DA_EXTENT_TAIL
;
2094 } while ((bh
= bh
->b_this_page
) != head
);
2101 * mpage_da_writepages - walk the list of dirty pages of the given
2102 * address space, allocates non-allocated blocks, maps newly-allocated
2103 * blocks to existing bhs and issue IO them
2105 * @mapping: address space structure to write
2106 * @wbc: subtract the number of written pages from *@wbc->nr_to_write
2107 * @get_block: the filesystem's block mapper function.
2109 * This is a library function, which implements the writepages()
2110 * address_space_operation.
2112 static int mpage_da_writepages(struct address_space
*mapping
,
2113 struct writeback_control
*wbc
,
2114 struct mpage_da_data
*mpd
)
2119 if (!mpd
->get_block
)
2120 return generic_writepages(mapping
, wbc
);
2122 mpd
->lbh
.b_size
= 0;
2123 mpd
->lbh
.b_state
= 0;
2124 mpd
->lbh
.b_blocknr
= 0;
2125 mpd
->first_page
= 0;
2128 mpd
->pages_written
= 0;
2131 to_write
= wbc
->nr_to_write
;
2133 ret
= write_cache_pages(mapping
, wbc
, __mpage_da_writepage
, mpd
);
2136 * Handle last extent of pages
2138 if (!mpd
->io_done
&& mpd
->next_page
!= mpd
->first_page
) {
2139 if (mpage_da_map_blocks(mpd
) == 0)
2140 mpage_da_submit_io(mpd
);
2143 wbc
->nr_to_write
= to_write
- mpd
->pages_written
;
2148 * this is a special callback for ->write_begin() only
2149 * it's intention is to return mapped block or reserve space
2151 static int ext4_da_get_block_prep(struct inode
*inode
, sector_t iblock
,
2152 struct buffer_head
*bh_result
, int create
)
2156 BUG_ON(create
== 0);
2157 BUG_ON(bh_result
->b_size
!= inode
->i_sb
->s_blocksize
);
2160 * first, we need to know whether the block is allocated already
2161 * preallocated blocks are unmapped but should treated
2162 * the same as allocated blocks.
2164 ret
= ext4_get_blocks_wrap(NULL
, inode
, iblock
, 1, bh_result
, 0, 0, 0);
2165 if ((ret
== 0) && !buffer_delay(bh_result
)) {
2166 /* the block isn't (pre)allocated yet, let's reserve space */
2168 * XXX: __block_prepare_write() unmaps passed block,
2171 ret
= ext4_da_reserve_space(inode
, 1);
2173 /* not enough space to reserve */
2176 map_bh(bh_result
, inode
->i_sb
, 0);
2177 set_buffer_new(bh_result
);
2178 set_buffer_delay(bh_result
);
2179 } else if (ret
> 0) {
2180 bh_result
->b_size
= (ret
<< inode
->i_blkbits
);
2186 #define EXT4_DELALLOC_RSVED 1
2187 static int ext4_da_get_block_write(struct inode
*inode
, sector_t iblock
,
2188 struct buffer_head
*bh_result
, int create
)
2191 unsigned max_blocks
= bh_result
->b_size
>> inode
->i_blkbits
;
2192 loff_t disksize
= EXT4_I(inode
)->i_disksize
;
2193 handle_t
*handle
= NULL
;
2195 handle
= ext4_journal_current_handle();
2197 ret
= ext4_get_blocks_wrap(handle
, inode
, iblock
, max_blocks
,
2198 bh_result
, create
, 0, EXT4_DELALLOC_RSVED
);
2201 bh_result
->b_size
= (ret
<< inode
->i_blkbits
);
2203 if (ext4_should_order_data(inode
)) {
2205 retval
= ext4_jbd2_file_inode(handle
, inode
);
2208 * Failed to add inode for ordered
2209 * mode. Don't update file size
2215 * Update on-disk size along with block allocation
2216 * we don't use 'extend_disksize' as size may change
2217 * within already allocated block -bzzz
2219 disksize
= ((loff_t
) iblock
+ ret
) << inode
->i_blkbits
;
2220 if (disksize
> i_size_read(inode
))
2221 disksize
= i_size_read(inode
);
2222 if (disksize
> EXT4_I(inode
)->i_disksize
) {
2224 * XXX: replace with spinlock if seen contended -bzzz
2226 down_write(&EXT4_I(inode
)->i_data_sem
);
2227 if (disksize
> EXT4_I(inode
)->i_disksize
)
2228 EXT4_I(inode
)->i_disksize
= disksize
;
2229 up_write(&EXT4_I(inode
)->i_data_sem
);
2231 if (EXT4_I(inode
)->i_disksize
== disksize
) {
2232 ret
= ext4_mark_inode_dirty(handle
, inode
);
2241 static int ext4_bh_unmapped_or_delay(handle_t
*handle
, struct buffer_head
*bh
)
2244 * unmapped buffer is possible for holes.
2245 * delay buffer is possible with delayed allocation
2247 return ((!buffer_mapped(bh
) || buffer_delay(bh
)) && buffer_dirty(bh
));
2250 static int ext4_normal_get_block_write(struct inode
*inode
, sector_t iblock
,
2251 struct buffer_head
*bh_result
, int create
)
2254 unsigned max_blocks
= bh_result
->b_size
>> inode
->i_blkbits
;
2257 * we don't want to do block allocation in writepage
2258 * so call get_block_wrap with create = 0
2260 ret
= ext4_get_blocks_wrap(NULL
, inode
, iblock
, max_blocks
,
2261 bh_result
, 0, 0, 0);
2263 bh_result
->b_size
= (ret
<< inode
->i_blkbits
);
2270 * get called vi ext4_da_writepages after taking page lock (have journal handle)
2271 * get called via journal_submit_inode_data_buffers (no journal handle)
2272 * get called via shrink_page_list via pdflush (no journal handle)
2273 * or grab_page_cache when doing write_begin (have journal handle)
2275 static int ext4_da_writepage(struct page
*page
,
2276 struct writeback_control
*wbc
)
2281 struct buffer_head
*page_bufs
;
2282 struct inode
*inode
= page
->mapping
->host
;
2284 size
= i_size_read(inode
);
2285 if (page
->index
== size
>> PAGE_CACHE_SHIFT
)
2286 len
= size
& ~PAGE_CACHE_MASK
;
2288 len
= PAGE_CACHE_SIZE
;
2290 if (page_has_buffers(page
)) {
2291 page_bufs
= page_buffers(page
);
2292 if (walk_page_buffers(NULL
, page_bufs
, 0, len
, NULL
,
2293 ext4_bh_unmapped_or_delay
)) {
2295 * We don't want to do block allocation
2296 * So redirty the page and return
2297 * We may reach here when we do a journal commit
2298 * via journal_submit_inode_data_buffers.
2299 * If we don't have mapping block we just ignore
2300 * them. We can also reach here via shrink_page_list
2302 redirty_page_for_writepage(wbc
, page
);
2308 * The test for page_has_buffers() is subtle:
2309 * We know the page is dirty but it lost buffers. That means
2310 * that at some moment in time after write_begin()/write_end()
2311 * has been called all buffers have been clean and thus they
2312 * must have been written at least once. So they are all
2313 * mapped and we can happily proceed with mapping them
2314 * and writing the page.
2316 * Try to initialize the buffer_heads and check whether
2317 * all are mapped and non delay. We don't want to
2318 * do block allocation here.
2320 ret
= block_prepare_write(page
, 0, PAGE_CACHE_SIZE
,
2321 ext4_normal_get_block_write
);
2323 page_bufs
= page_buffers(page
);
2324 /* check whether all are mapped and non delay */
2325 if (walk_page_buffers(NULL
, page_bufs
, 0, len
, NULL
,
2326 ext4_bh_unmapped_or_delay
)) {
2327 redirty_page_for_writepage(wbc
, page
);
2333 * We can't do block allocation here
2334 * so just redity the page and unlock
2337 redirty_page_for_writepage(wbc
, page
);
2343 if (test_opt(inode
->i_sb
, NOBH
) && ext4_should_writeback_data(inode
))
2344 ret
= nobh_writepage(page
, ext4_normal_get_block_write
, wbc
);
2346 ret
= block_write_full_page(page
,
2347 ext4_normal_get_block_write
,
2354 * This is called via ext4_da_writepages() to
2355 * calulate the total number of credits to reserve to fit
2356 * a single extent allocation into a single transaction,
2357 * ext4_da_writpeages() will loop calling this before
2358 * the block allocation.
2361 static int ext4_da_writepages_trans_blocks(struct inode
*inode
)
2363 int max_blocks
= EXT4_I(inode
)->i_reserved_data_blocks
;
2366 * With non-extent format the journal credit needed to
2367 * insert nrblocks contiguous block is dependent on
2368 * number of contiguous block. So we will limit
2369 * number of contiguous block to a sane value
2371 if (!(inode
->i_flags
& EXT4_EXTENTS_FL
) &&
2372 (max_blocks
> EXT4_MAX_TRANS_DATA
))
2373 max_blocks
= EXT4_MAX_TRANS_DATA
;
2375 return ext4_chunk_trans_blocks(inode
, max_blocks
);
2378 static int ext4_da_writepages(struct address_space
*mapping
,
2379 struct writeback_control
*wbc
)
2381 handle_t
*handle
= NULL
;
2382 loff_t range_start
= 0;
2383 struct mpage_da_data mpd
;
2384 struct inode
*inode
= mapping
->host
;
2385 int needed_blocks
, ret
= 0, nr_to_writebump
= 0;
2386 long to_write
, pages_skipped
= 0;
2387 struct ext4_sb_info
*sbi
= EXT4_SB(mapping
->host
->i_sb
);
2390 * No pages to write? This is mainly a kludge to avoid starting
2391 * a transaction for special inodes like journal inode on last iput()
2392 * because that could violate lock ordering on umount
2394 if (!mapping
->nrpages
|| !mapping_tagged(mapping
, PAGECACHE_TAG_DIRTY
))
2397 * Make sure nr_to_write is >= sbi->s_mb_stream_request
2398 * This make sure small files blocks are allocated in
2399 * single attempt. This ensure that small files
2400 * get less fragmented.
2402 if (wbc
->nr_to_write
< sbi
->s_mb_stream_request
) {
2403 nr_to_writebump
= sbi
->s_mb_stream_request
- wbc
->nr_to_write
;
2404 wbc
->nr_to_write
= sbi
->s_mb_stream_request
;
2407 if (!wbc
->range_cyclic
)
2409 * If range_cyclic is not set force range_cont
2410 * and save the old writeback_index
2412 wbc
->range_cont
= 1;
2414 range_start
= wbc
->range_start
;
2415 pages_skipped
= wbc
->pages_skipped
;
2418 mpd
.inode
= mapping
->host
;
2421 to_write
= wbc
->nr_to_write
;
2422 while (!ret
&& to_write
> 0) {
2425 * we insert one extent at a time. So we need
2426 * credit needed for single extent allocation.
2427 * journalled mode is currently not supported
2430 BUG_ON(ext4_should_journal_data(inode
));
2431 needed_blocks
= ext4_da_writepages_trans_blocks(inode
);
2433 /* start a new transaction*/
2434 handle
= ext4_journal_start(inode
, needed_blocks
);
2435 if (IS_ERR(handle
)) {
2436 ret
= PTR_ERR(handle
);
2437 printk(KERN_EMERG
"%s: jbd2_start: "
2438 "%ld pages, ino %lu; err %d\n", __func__
,
2439 wbc
->nr_to_write
, inode
->i_ino
, ret
);
2441 goto out_writepages
;
2443 to_write
-= wbc
->nr_to_write
;
2445 mpd
.get_block
= ext4_da_get_block_write
;
2446 ret
= mpage_da_writepages(mapping
, wbc
, &mpd
);
2448 ext4_journal_stop(handle
);
2450 if (mpd
.retval
== -ENOSPC
)
2451 jbd2_journal_force_commit_nested(sbi
->s_journal
);
2453 /* reset the retry count */
2454 if (ret
== MPAGE_DA_EXTENT_TAIL
) {
2456 * got one extent now try with
2459 to_write
+= wbc
->nr_to_write
;
2461 } else if (wbc
->nr_to_write
) {
2463 * There is no more writeout needed
2464 * or we requested for a noblocking writeout
2465 * and we found the device congested
2467 to_write
+= wbc
->nr_to_write
;
2470 wbc
->nr_to_write
= to_write
;
2473 if (wbc
->range_cont
&& (pages_skipped
!= wbc
->pages_skipped
)) {
2474 /* We skipped pages in this loop */
2475 wbc
->range_start
= range_start
;
2476 wbc
->nr_to_write
= to_write
+
2477 wbc
->pages_skipped
- pages_skipped
;
2478 wbc
->pages_skipped
= pages_skipped
;
2483 wbc
->nr_to_write
= to_write
- nr_to_writebump
;
2484 wbc
->range_start
= range_start
;
2488 #define FALL_BACK_TO_NONDELALLOC 1
2489 static int ext4_nonda_switch(struct super_block
*sb
)
2491 s64 free_blocks
, dirty_blocks
;
2492 struct ext4_sb_info
*sbi
= EXT4_SB(sb
);
2495 * switch to non delalloc mode if we are running low
2496 * on free block. The free block accounting via percpu
2497 * counters can get slightly wrong with FBC_BATCH getting
2498 * accumulated on each CPU without updating global counters
2499 * Delalloc need an accurate free block accounting. So switch
2500 * to non delalloc when we are near to error range.
2502 free_blocks
= percpu_counter_read_positive(&sbi
->s_freeblocks_counter
);
2503 dirty_blocks
= percpu_counter_read_positive(&sbi
->s_dirtyblocks_counter
);
2504 if (2 * free_blocks
< 3 * dirty_blocks
||
2505 free_blocks
< (dirty_blocks
+ EXT4_FREEBLOCKS_WATERMARK
)) {
2507 * free block count is less that 150% of dirty blocks
2508 * or free blocks is less that watermark
2515 static int ext4_da_write_begin(struct file
*file
, struct address_space
*mapping
,
2516 loff_t pos
, unsigned len
, unsigned flags
,
2517 struct page
**pagep
, void **fsdata
)
2519 int ret
, retries
= 0;
2523 struct inode
*inode
= mapping
->host
;
2526 index
= pos
>> PAGE_CACHE_SHIFT
;
2527 from
= pos
& (PAGE_CACHE_SIZE
- 1);
2530 if (ext4_nonda_switch(inode
->i_sb
)) {
2531 *fsdata
= (void *)FALL_BACK_TO_NONDELALLOC
;
2532 return ext4_write_begin(file
, mapping
, pos
,
2533 len
, flags
, pagep
, fsdata
);
2535 *fsdata
= (void *)0;
2538 * With delayed allocation, we don't log the i_disksize update
2539 * if there is delayed block allocation. But we still need
2540 * to journalling the i_disksize update if writes to the end
2541 * of file which has an already mapped buffer.
2543 handle
= ext4_journal_start(inode
, 1);
2544 if (IS_ERR(handle
)) {
2545 ret
= PTR_ERR(handle
);
2549 page
= __grab_cache_page(mapping
, index
);
2551 ext4_journal_stop(handle
);
2557 ret
= block_write_begin(file
, mapping
, pos
, len
, flags
, pagep
, fsdata
,
2558 ext4_da_get_block_prep
);
2561 ext4_journal_stop(handle
);
2562 page_cache_release(page
);
2565 if (ret
== -ENOSPC
&& ext4_should_retry_alloc(inode
->i_sb
, &retries
))
2572 * Check if we should update i_disksize
2573 * when write to the end of file but not require block allocation
2575 static int ext4_da_should_update_i_disksize(struct page
*page
,
2576 unsigned long offset
)
2578 struct buffer_head
*bh
;
2579 struct inode
*inode
= page
->mapping
->host
;
2583 bh
= page_buffers(page
);
2584 idx
= offset
>> inode
->i_blkbits
;
2586 for (i
= 0; i
< idx
; i
++)
2587 bh
= bh
->b_this_page
;
2589 if (!buffer_mapped(bh
) || (buffer_delay(bh
)))
2594 static int ext4_da_write_end(struct file
*file
,
2595 struct address_space
*mapping
,
2596 loff_t pos
, unsigned len
, unsigned copied
,
2597 struct page
*page
, void *fsdata
)
2599 struct inode
*inode
= mapping
->host
;
2601 handle_t
*handle
= ext4_journal_current_handle();
2603 unsigned long start
, end
;
2604 int write_mode
= (int)(unsigned long)fsdata
;
2606 if (write_mode
== FALL_BACK_TO_NONDELALLOC
) {
2607 if (ext4_should_order_data(inode
)) {
2608 return ext4_ordered_write_end(file
, mapping
, pos
,
2609 len
, copied
, page
, fsdata
);
2610 } else if (ext4_should_writeback_data(inode
)) {
2611 return ext4_writeback_write_end(file
, mapping
, pos
,
2612 len
, copied
, page
, fsdata
);
2618 start
= pos
& (PAGE_CACHE_SIZE
- 1);
2619 end
= start
+ copied
- 1;
2622 * generic_write_end() will run mark_inode_dirty() if i_size
2623 * changes. So let's piggyback the i_disksize mark_inode_dirty
2627 new_i_size
= pos
+ copied
;
2628 if (new_i_size
> EXT4_I(inode
)->i_disksize
) {
2629 if (ext4_da_should_update_i_disksize(page
, end
)) {
2630 down_write(&EXT4_I(inode
)->i_data_sem
);
2631 if (new_i_size
> EXT4_I(inode
)->i_disksize
) {
2633 * Updating i_disksize when extending file
2634 * without needing block allocation
2636 if (ext4_should_order_data(inode
))
2637 ret
= ext4_jbd2_file_inode(handle
,
2640 EXT4_I(inode
)->i_disksize
= new_i_size
;
2642 up_write(&EXT4_I(inode
)->i_data_sem
);
2645 ret2
= generic_write_end(file
, mapping
, pos
, len
, copied
,
2650 ret2
= ext4_journal_stop(handle
);
2654 return ret
? ret
: copied
;
2657 static void ext4_da_invalidatepage(struct page
*page
, unsigned long offset
)
2660 * Drop reserved blocks
2662 BUG_ON(!PageLocked(page
));
2663 if (!page_has_buffers(page
))
2666 ext4_da_page_release_reservation(page
, offset
);
2669 ext4_invalidatepage(page
, offset
);
2676 * bmap() is special. It gets used by applications such as lilo and by
2677 * the swapper to find the on-disk block of a specific piece of data.
2679 * Naturally, this is dangerous if the block concerned is still in the
2680 * journal. If somebody makes a swapfile on an ext4 data-journaling
2681 * filesystem and enables swap, then they may get a nasty shock when the
2682 * data getting swapped to that swapfile suddenly gets overwritten by
2683 * the original zero's written out previously to the journal and
2684 * awaiting writeback in the kernel's buffer cache.
2686 * So, if we see any bmap calls here on a modified, data-journaled file,
2687 * take extra steps to flush any blocks which might be in the cache.
2689 static sector_t
ext4_bmap(struct address_space
*mapping
, sector_t block
)
2691 struct inode
*inode
= mapping
->host
;
2695 if (mapping_tagged(mapping
, PAGECACHE_TAG_DIRTY
) &&
2696 test_opt(inode
->i_sb
, DELALLOC
)) {
2698 * With delalloc we want to sync the file
2699 * so that we can make sure we allocate
2702 filemap_write_and_wait(mapping
);
2705 if (EXT4_I(inode
)->i_state
& EXT4_STATE_JDATA
) {
2707 * This is a REALLY heavyweight approach, but the use of
2708 * bmap on dirty files is expected to be extremely rare:
2709 * only if we run lilo or swapon on a freshly made file
2710 * do we expect this to happen.
2712 * (bmap requires CAP_SYS_RAWIO so this does not
2713 * represent an unprivileged user DOS attack --- we'd be
2714 * in trouble if mortal users could trigger this path at
2717 * NB. EXT4_STATE_JDATA is not set on files other than
2718 * regular files. If somebody wants to bmap a directory
2719 * or symlink and gets confused because the buffer
2720 * hasn't yet been flushed to disk, they deserve
2721 * everything they get.
2724 EXT4_I(inode
)->i_state
&= ~EXT4_STATE_JDATA
;
2725 journal
= EXT4_JOURNAL(inode
);
2726 jbd2_journal_lock_updates(journal
);
2727 err
= jbd2_journal_flush(journal
);
2728 jbd2_journal_unlock_updates(journal
);
2734 return generic_block_bmap(mapping
, block
, ext4_get_block
);
2737 static int bget_one(handle_t
*handle
, struct buffer_head
*bh
)
2743 static int bput_one(handle_t
*handle
, struct buffer_head
*bh
)
2750 * Note that we don't need to start a transaction unless we're journaling data
2751 * because we should have holes filled from ext4_page_mkwrite(). We even don't
2752 * need to file the inode to the transaction's list in ordered mode because if
2753 * we are writing back data added by write(), the inode is already there and if
2754 * we are writing back data modified via mmap(), noone guarantees in which
2755 * transaction the data will hit the disk. In case we are journaling data, we
2756 * cannot start transaction directly because transaction start ranks above page
2757 * lock so we have to do some magic.
2759 * In all journaling modes block_write_full_page() will start the I/O.
2763 * ext4_writepage() -> kmalloc() -> __alloc_pages() -> page_launder() ->
2768 * ext4_file_write() -> generic_file_write() -> __alloc_pages() -> ...
2770 * Same applies to ext4_get_block(). We will deadlock on various things like
2771 * lock_journal and i_data_sem
2773 * Setting PF_MEMALLOC here doesn't work - too many internal memory
2776 * 16May01: If we're reentered then journal_current_handle() will be
2777 * non-zero. We simply *return*.
2779 * 1 July 2001: @@@ FIXME:
2780 * In journalled data mode, a data buffer may be metadata against the
2781 * current transaction. But the same file is part of a shared mapping
2782 * and someone does a writepage() on it.
2784 * We will move the buffer onto the async_data list, but *after* it has
2785 * been dirtied. So there's a small window where we have dirty data on
2788 * Note that this only applies to the last partial page in the file. The
2789 * bit which block_write_full_page() uses prepare/commit for. (That's
2790 * broken code anyway: it's wrong for msync()).
2792 * It's a rare case: affects the final partial page, for journalled data
2793 * where the file is subject to bith write() and writepage() in the same
2794 * transction. To fix it we'll need a custom block_write_full_page().
2795 * We'll probably need that anyway for journalling writepage() output.
2797 * We don't honour synchronous mounts for writepage(). That would be
2798 * disastrous. Any write() or metadata operation will sync the fs for
2802 static int __ext4_normal_writepage(struct page
*page
,
2803 struct writeback_control
*wbc
)
2805 struct inode
*inode
= page
->mapping
->host
;
2807 if (test_opt(inode
->i_sb
, NOBH
))
2808 return nobh_writepage(page
,
2809 ext4_normal_get_block_write
, wbc
);
2811 return block_write_full_page(page
,
2812 ext4_normal_get_block_write
,
2816 static int ext4_normal_writepage(struct page
*page
,
2817 struct writeback_control
*wbc
)
2819 struct inode
*inode
= page
->mapping
->host
;
2820 loff_t size
= i_size_read(inode
);
2823 J_ASSERT(PageLocked(page
));
2824 if (page
->index
== size
>> PAGE_CACHE_SHIFT
)
2825 len
= size
& ~PAGE_CACHE_MASK
;
2827 len
= PAGE_CACHE_SIZE
;
2829 if (page_has_buffers(page
)) {
2830 /* if page has buffers it should all be mapped
2831 * and allocated. If there are not buffers attached
2832 * to the page we know the page is dirty but it lost
2833 * buffers. That means that at some moment in time
2834 * after write_begin() / write_end() has been called
2835 * all buffers have been clean and thus they must have been
2836 * written at least once. So they are all mapped and we can
2837 * happily proceed with mapping them and writing the page.
2839 BUG_ON(walk_page_buffers(NULL
, page_buffers(page
), 0, len
, NULL
,
2840 ext4_bh_unmapped_or_delay
));
2843 if (!ext4_journal_current_handle())
2844 return __ext4_normal_writepage(page
, wbc
);
2846 redirty_page_for_writepage(wbc
, page
);
2851 static int __ext4_journalled_writepage(struct page
*page
,
2852 struct writeback_control
*wbc
)
2854 struct address_space
*mapping
= page
->mapping
;
2855 struct inode
*inode
= mapping
->host
;
2856 struct buffer_head
*page_bufs
;
2857 handle_t
*handle
= NULL
;
2861 ret
= block_prepare_write(page
, 0, PAGE_CACHE_SIZE
,
2862 ext4_normal_get_block_write
);
2866 page_bufs
= page_buffers(page
);
2867 walk_page_buffers(handle
, page_bufs
, 0, PAGE_CACHE_SIZE
, NULL
,
2869 /* As soon as we unlock the page, it can go away, but we have
2870 * references to buffers so we are safe */
2873 handle
= ext4_journal_start(inode
, ext4_writepage_trans_blocks(inode
));
2874 if (IS_ERR(handle
)) {
2875 ret
= PTR_ERR(handle
);
2879 ret
= walk_page_buffers(handle
, page_bufs
, 0,
2880 PAGE_CACHE_SIZE
, NULL
, do_journal_get_write_access
);
2882 err
= walk_page_buffers(handle
, page_bufs
, 0,
2883 PAGE_CACHE_SIZE
, NULL
, write_end_fn
);
2886 err
= ext4_journal_stop(handle
);
2890 walk_page_buffers(handle
, page_bufs
, 0,
2891 PAGE_CACHE_SIZE
, NULL
, bput_one
);
2892 EXT4_I(inode
)->i_state
|= EXT4_STATE_JDATA
;
2901 static int ext4_journalled_writepage(struct page
*page
,
2902 struct writeback_control
*wbc
)
2904 struct inode
*inode
= page
->mapping
->host
;
2905 loff_t size
= i_size_read(inode
);
2908 J_ASSERT(PageLocked(page
));
2909 if (page
->index
== size
>> PAGE_CACHE_SHIFT
)
2910 len
= size
& ~PAGE_CACHE_MASK
;
2912 len
= PAGE_CACHE_SIZE
;
2914 if (page_has_buffers(page
)) {
2915 /* if page has buffers it should all be mapped
2916 * and allocated. If there are not buffers attached
2917 * to the page we know the page is dirty but it lost
2918 * buffers. That means that at some moment in time
2919 * after write_begin() / write_end() has been called
2920 * all buffers have been clean and thus they must have been
2921 * written at least once. So they are all mapped and we can
2922 * happily proceed with mapping them and writing the page.
2924 BUG_ON(walk_page_buffers(NULL
, page_buffers(page
), 0, len
, NULL
,
2925 ext4_bh_unmapped_or_delay
));
2928 if (ext4_journal_current_handle())
2931 if (PageChecked(page
)) {
2933 * It's mmapped pagecache. Add buffers and journal it. There
2934 * doesn't seem much point in redirtying the page here.
2936 ClearPageChecked(page
);
2937 return __ext4_journalled_writepage(page
, wbc
);
2940 * It may be a page full of checkpoint-mode buffers. We don't
2941 * really know unless we go poke around in the buffer_heads.
2942 * But block_write_full_page will do the right thing.
2944 return block_write_full_page(page
,
2945 ext4_normal_get_block_write
,
2949 redirty_page_for_writepage(wbc
, page
);
2954 static int ext4_readpage(struct file
*file
, struct page
*page
)
2956 return mpage_readpage(page
, ext4_get_block
);
2960 ext4_readpages(struct file
*file
, struct address_space
*mapping
,
2961 struct list_head
*pages
, unsigned nr_pages
)
2963 return mpage_readpages(mapping
, pages
, nr_pages
, ext4_get_block
);
2966 static void ext4_invalidatepage(struct page
*page
, unsigned long offset
)
2968 journal_t
*journal
= EXT4_JOURNAL(page
->mapping
->host
);
2971 * If it's a full truncate we just forget about the pending dirtying
2974 ClearPageChecked(page
);
2976 jbd2_journal_invalidatepage(journal
, page
, offset
);
2979 static int ext4_releasepage(struct page
*page
, gfp_t wait
)
2981 journal_t
*journal
= EXT4_JOURNAL(page
->mapping
->host
);
2983 WARN_ON(PageChecked(page
));
2984 if (!page_has_buffers(page
))
2986 return jbd2_journal_try_to_free_buffers(journal
, page
, wait
);
2990 * If the O_DIRECT write will extend the file then add this inode to the
2991 * orphan list. So recovery will truncate it back to the original size
2992 * if the machine crashes during the write.
2994 * If the O_DIRECT write is intantiating holes inside i_size and the machine
2995 * crashes then stale disk data _may_ be exposed inside the file. But current
2996 * VFS code falls back into buffered path in that case so we are safe.
2998 static ssize_t
ext4_direct_IO(int rw
, struct kiocb
*iocb
,
2999 const struct iovec
*iov
, loff_t offset
,
3000 unsigned long nr_segs
)
3002 struct file
*file
= iocb
->ki_filp
;
3003 struct inode
*inode
= file
->f_mapping
->host
;
3004 struct ext4_inode_info
*ei
= EXT4_I(inode
);
3008 size_t count
= iov_length(iov
, nr_segs
);
3011 loff_t final_size
= offset
+ count
;
3013 if (final_size
> inode
->i_size
) {
3014 /* Credits for sb + inode write */
3015 handle
= ext4_journal_start(inode
, 2);
3016 if (IS_ERR(handle
)) {
3017 ret
= PTR_ERR(handle
);
3020 ret
= ext4_orphan_add(handle
, inode
);
3022 ext4_journal_stop(handle
);
3026 ei
->i_disksize
= inode
->i_size
;
3027 ext4_journal_stop(handle
);
3031 ret
= blockdev_direct_IO(rw
, iocb
, inode
, inode
->i_sb
->s_bdev
, iov
,
3033 ext4_get_block
, NULL
);
3038 /* Credits for sb + inode write */
3039 handle
= ext4_journal_start(inode
, 2);
3040 if (IS_ERR(handle
)) {
3041 /* This is really bad luck. We've written the data
3042 * but cannot extend i_size. Bail out and pretend
3043 * the write failed... */
3044 ret
= PTR_ERR(handle
);
3048 ext4_orphan_del(handle
, inode
);
3050 loff_t end
= offset
+ ret
;
3051 if (end
> inode
->i_size
) {
3052 ei
->i_disksize
= end
;
3053 i_size_write(inode
, end
);
3055 * We're going to return a positive `ret'
3056 * here due to non-zero-length I/O, so there's
3057 * no way of reporting error returns from
3058 * ext4_mark_inode_dirty() to userspace. So
3061 ext4_mark_inode_dirty(handle
, inode
);
3064 err
= ext4_journal_stop(handle
);
3073 * Pages can be marked dirty completely asynchronously from ext4's journalling
3074 * activity. By filemap_sync_pte(), try_to_unmap_one(), etc. We cannot do
3075 * much here because ->set_page_dirty is called under VFS locks. The page is
3076 * not necessarily locked.
3078 * We cannot just dirty the page and leave attached buffers clean, because the
3079 * buffers' dirty state is "definitive". We cannot just set the buffers dirty
3080 * or jbddirty because all the journalling code will explode.
3082 * So what we do is to mark the page "pending dirty" and next time writepage
3083 * is called, propagate that into the buffers appropriately.
3085 static int ext4_journalled_set_page_dirty(struct page
*page
)
3087 SetPageChecked(page
);
3088 return __set_page_dirty_nobuffers(page
);
3091 static const struct address_space_operations ext4_ordered_aops
= {
3092 .readpage
= ext4_readpage
,
3093 .readpages
= ext4_readpages
,
3094 .writepage
= ext4_normal_writepage
,
3095 .sync_page
= block_sync_page
,
3096 .write_begin
= ext4_write_begin
,
3097 .write_end
= ext4_ordered_write_end
,
3099 .invalidatepage
= ext4_invalidatepage
,
3100 .releasepage
= ext4_releasepage
,
3101 .direct_IO
= ext4_direct_IO
,
3102 .migratepage
= buffer_migrate_page
,
3103 .is_partially_uptodate
= block_is_partially_uptodate
,
3106 static const struct address_space_operations ext4_writeback_aops
= {
3107 .readpage
= ext4_readpage
,
3108 .readpages
= ext4_readpages
,
3109 .writepage
= ext4_normal_writepage
,
3110 .sync_page
= block_sync_page
,
3111 .write_begin
= ext4_write_begin
,
3112 .write_end
= ext4_writeback_write_end
,
3114 .invalidatepage
= ext4_invalidatepage
,
3115 .releasepage
= ext4_releasepage
,
3116 .direct_IO
= ext4_direct_IO
,
3117 .migratepage
= buffer_migrate_page
,
3118 .is_partially_uptodate
= block_is_partially_uptodate
,
3121 static const struct address_space_operations ext4_journalled_aops
= {
3122 .readpage
= ext4_readpage
,
3123 .readpages
= ext4_readpages
,
3124 .writepage
= ext4_journalled_writepage
,
3125 .sync_page
= block_sync_page
,
3126 .write_begin
= ext4_write_begin
,
3127 .write_end
= ext4_journalled_write_end
,
3128 .set_page_dirty
= ext4_journalled_set_page_dirty
,
3130 .invalidatepage
= ext4_invalidatepage
,
3131 .releasepage
= ext4_releasepage
,
3132 .is_partially_uptodate
= block_is_partially_uptodate
,
3135 static const struct address_space_operations ext4_da_aops
= {
3136 .readpage
= ext4_readpage
,
3137 .readpages
= ext4_readpages
,
3138 .writepage
= ext4_da_writepage
,
3139 .writepages
= ext4_da_writepages
,
3140 .sync_page
= block_sync_page
,
3141 .write_begin
= ext4_da_write_begin
,
3142 .write_end
= ext4_da_write_end
,
3144 .invalidatepage
= ext4_da_invalidatepage
,
3145 .releasepage
= ext4_releasepage
,
3146 .direct_IO
= ext4_direct_IO
,
3147 .migratepage
= buffer_migrate_page
,
3148 .is_partially_uptodate
= block_is_partially_uptodate
,
3151 void ext4_set_aops(struct inode
*inode
)
3153 if (ext4_should_order_data(inode
) &&
3154 test_opt(inode
->i_sb
, DELALLOC
))
3155 inode
->i_mapping
->a_ops
= &ext4_da_aops
;
3156 else if (ext4_should_order_data(inode
))
3157 inode
->i_mapping
->a_ops
= &ext4_ordered_aops
;
3158 else if (ext4_should_writeback_data(inode
) &&
3159 test_opt(inode
->i_sb
, DELALLOC
))
3160 inode
->i_mapping
->a_ops
= &ext4_da_aops
;
3161 else if (ext4_should_writeback_data(inode
))
3162 inode
->i_mapping
->a_ops
= &ext4_writeback_aops
;
3164 inode
->i_mapping
->a_ops
= &ext4_journalled_aops
;
3168 * ext4_block_truncate_page() zeroes out a mapping from file offset `from'
3169 * up to the end of the block which corresponds to `from'.
3170 * This required during truncate. We need to physically zero the tail end
3171 * of that block so it doesn't yield old data if the file is later grown.
3173 int ext4_block_truncate_page(handle_t
*handle
,
3174 struct address_space
*mapping
, loff_t from
)
3176 ext4_fsblk_t index
= from
>> PAGE_CACHE_SHIFT
;
3177 unsigned offset
= from
& (PAGE_CACHE_SIZE
-1);
3178 unsigned blocksize
, length
, pos
;
3180 struct inode
*inode
= mapping
->host
;
3181 struct buffer_head
*bh
;
3185 page
= grab_cache_page(mapping
, from
>> PAGE_CACHE_SHIFT
);
3189 blocksize
= inode
->i_sb
->s_blocksize
;
3190 length
= blocksize
- (offset
& (blocksize
- 1));
3191 iblock
= index
<< (PAGE_CACHE_SHIFT
- inode
->i_sb
->s_blocksize_bits
);
3194 * For "nobh" option, we can only work if we don't need to
3195 * read-in the page - otherwise we create buffers to do the IO.
3197 if (!page_has_buffers(page
) && test_opt(inode
->i_sb
, NOBH
) &&
3198 ext4_should_writeback_data(inode
) && PageUptodate(page
)) {
3199 zero_user(page
, offset
, length
);
3200 set_page_dirty(page
);
3204 if (!page_has_buffers(page
))
3205 create_empty_buffers(page
, blocksize
, 0);
3207 /* Find the buffer that contains "offset" */
3208 bh
= page_buffers(page
);
3210 while (offset
>= pos
) {
3211 bh
= bh
->b_this_page
;
3217 if (buffer_freed(bh
)) {
3218 BUFFER_TRACE(bh
, "freed: skip");
3222 if (!buffer_mapped(bh
)) {
3223 BUFFER_TRACE(bh
, "unmapped");
3224 ext4_get_block(inode
, iblock
, bh
, 0);
3225 /* unmapped? It's a hole - nothing to do */
3226 if (!buffer_mapped(bh
)) {
3227 BUFFER_TRACE(bh
, "still unmapped");
3232 /* Ok, it's mapped. Make sure it's up-to-date */
3233 if (PageUptodate(page
))
3234 set_buffer_uptodate(bh
);
3236 if (!buffer_uptodate(bh
)) {
3238 ll_rw_block(READ
, 1, &bh
);
3240 /* Uhhuh. Read error. Complain and punt. */
3241 if (!buffer_uptodate(bh
))
3245 if (ext4_should_journal_data(inode
)) {
3246 BUFFER_TRACE(bh
, "get write access");
3247 err
= ext4_journal_get_write_access(handle
, bh
);
3252 zero_user(page
, offset
, length
);
3254 BUFFER_TRACE(bh
, "zeroed end of block");
3257 if (ext4_should_journal_data(inode
)) {
3258 err
= ext4_journal_dirty_metadata(handle
, bh
);
3260 if (ext4_should_order_data(inode
))
3261 err
= ext4_jbd2_file_inode(handle
, inode
);
3262 mark_buffer_dirty(bh
);
3267 page_cache_release(page
);
3272 * Probably it should be a library function... search for first non-zero word
3273 * or memcmp with zero_page, whatever is better for particular architecture.
3276 static inline int all_zeroes(__le32
*p
, __le32
*q
)
3285 * ext4_find_shared - find the indirect blocks for partial truncation.
3286 * @inode: inode in question
3287 * @depth: depth of the affected branch
3288 * @offsets: offsets of pointers in that branch (see ext4_block_to_path)
3289 * @chain: place to store the pointers to partial indirect blocks
3290 * @top: place to the (detached) top of branch
3292 * This is a helper function used by ext4_truncate().
3294 * When we do truncate() we may have to clean the ends of several
3295 * indirect blocks but leave the blocks themselves alive. Block is
3296 * partially truncated if some data below the new i_size is refered
3297 * from it (and it is on the path to the first completely truncated
3298 * data block, indeed). We have to free the top of that path along
3299 * with everything to the right of the path. Since no allocation
3300 * past the truncation point is possible until ext4_truncate()
3301 * finishes, we may safely do the latter, but top of branch may
3302 * require special attention - pageout below the truncation point
3303 * might try to populate it.
3305 * We atomically detach the top of branch from the tree, store the
3306 * block number of its root in *@top, pointers to buffer_heads of
3307 * partially truncated blocks - in @chain[].bh and pointers to
3308 * their last elements that should not be removed - in
3309 * @chain[].p. Return value is the pointer to last filled element
3312 * The work left to caller to do the actual freeing of subtrees:
3313 * a) free the subtree starting from *@top
3314 * b) free the subtrees whose roots are stored in
3315 * (@chain[i].p+1 .. end of @chain[i].bh->b_data)
3316 * c) free the subtrees growing from the inode past the @chain[0].
3317 * (no partially truncated stuff there). */
3319 static Indirect
*ext4_find_shared(struct inode
*inode
, int depth
,
3320 ext4_lblk_t offsets
[4], Indirect chain
[4], __le32
*top
)
3322 Indirect
*partial
, *p
;
3326 /* Make k index the deepest non-null offest + 1 */
3327 for (k
= depth
; k
> 1 && !offsets
[k
-1]; k
--)
3329 partial
= ext4_get_branch(inode
, k
, offsets
, chain
, &err
);
3330 /* Writer: pointers */
3332 partial
= chain
+ k
-1;
3334 * If the branch acquired continuation since we've looked at it -
3335 * fine, it should all survive and (new) top doesn't belong to us.
3337 if (!partial
->key
&& *partial
->p
)
3340 for (p
= partial
; (p
> chain
) && all_zeroes((__le32
*) p
->bh
->b_data
, p
->p
); p
--)
3343 * OK, we've found the last block that must survive. The rest of our
3344 * branch should be detached before unlocking. However, if that rest
3345 * of branch is all ours and does not grow immediately from the inode
3346 * it's easier to cheat and just decrement partial->p.
3348 if (p
== chain
+ k
- 1 && p
> chain
) {
3352 /* Nope, don't do this in ext4. Must leave the tree intact */
3359 while (partial
> p
) {
3360 brelse(partial
->bh
);
3368 * Zero a number of block pointers in either an inode or an indirect block.
3369 * If we restart the transaction we must again get write access to the
3370 * indirect block for further modification.
3372 * We release `count' blocks on disk, but (last - first) may be greater
3373 * than `count' because there can be holes in there.
3375 static void ext4_clear_blocks(handle_t
*handle
, struct inode
*inode
,
3376 struct buffer_head
*bh
, ext4_fsblk_t block_to_free
,
3377 unsigned long count
, __le32
*first
, __le32
*last
)
3380 if (try_to_extend_transaction(handle
, inode
)) {
3382 BUFFER_TRACE(bh
, "call ext4_journal_dirty_metadata");
3383 ext4_journal_dirty_metadata(handle
, bh
);
3385 ext4_mark_inode_dirty(handle
, inode
);
3386 ext4_journal_test_restart(handle
, inode
);
3388 BUFFER_TRACE(bh
, "retaking write access");
3389 ext4_journal_get_write_access(handle
, bh
);
3394 * Any buffers which are on the journal will be in memory. We find
3395 * them on the hash table so jbd2_journal_revoke() will run jbd2_journal_forget()
3396 * on them. We've already detached each block from the file, so
3397 * bforget() in jbd2_journal_forget() should be safe.
3399 * AKPM: turn on bforget in jbd2_journal_forget()!!!
3401 for (p
= first
; p
< last
; p
++) {
3402 u32 nr
= le32_to_cpu(*p
);
3404 struct buffer_head
*tbh
;
3407 tbh
= sb_find_get_block(inode
->i_sb
, nr
);
3408 ext4_forget(handle
, 0, inode
, tbh
, nr
);
3412 ext4_free_blocks(handle
, inode
, block_to_free
, count
, 0);
3416 * ext4_free_data - free a list of data blocks
3417 * @handle: handle for this transaction
3418 * @inode: inode we are dealing with
3419 * @this_bh: indirect buffer_head which contains *@first and *@last
3420 * @first: array of block numbers
3421 * @last: points immediately past the end of array
3423 * We are freeing all blocks refered from that array (numbers are stored as
3424 * little-endian 32-bit) and updating @inode->i_blocks appropriately.
3426 * We accumulate contiguous runs of blocks to free. Conveniently, if these
3427 * blocks are contiguous then releasing them at one time will only affect one
3428 * or two bitmap blocks (+ group descriptor(s) and superblock) and we won't
3429 * actually use a lot of journal space.
3431 * @this_bh will be %NULL if @first and @last point into the inode's direct
3434 static void ext4_free_data(handle_t
*handle
, struct inode
*inode
,
3435 struct buffer_head
*this_bh
,
3436 __le32
*first
, __le32
*last
)
3438 ext4_fsblk_t block_to_free
= 0; /* Starting block # of a run */
3439 unsigned long count
= 0; /* Number of blocks in the run */
3440 __le32
*block_to_free_p
= NULL
; /* Pointer into inode/ind
3443 ext4_fsblk_t nr
; /* Current block # */
3444 __le32
*p
; /* Pointer into inode/ind
3445 for current block */
3448 if (this_bh
) { /* For indirect block */
3449 BUFFER_TRACE(this_bh
, "get_write_access");
3450 err
= ext4_journal_get_write_access(handle
, this_bh
);
3451 /* Important: if we can't update the indirect pointers
3452 * to the blocks, we can't free them. */
3457 for (p
= first
; p
< last
; p
++) {
3458 nr
= le32_to_cpu(*p
);
3460 /* accumulate blocks to free if they're contiguous */
3463 block_to_free_p
= p
;
3465 } else if (nr
== block_to_free
+ count
) {
3468 ext4_clear_blocks(handle
, inode
, this_bh
,
3470 count
, block_to_free_p
, p
);
3472 block_to_free_p
= p
;
3479 ext4_clear_blocks(handle
, inode
, this_bh
, block_to_free
,
3480 count
, block_to_free_p
, p
);
3483 BUFFER_TRACE(this_bh
, "call ext4_journal_dirty_metadata");
3486 * The buffer head should have an attached journal head at this
3487 * point. However, if the data is corrupted and an indirect
3488 * block pointed to itself, it would have been detached when
3489 * the block was cleared. Check for this instead of OOPSing.
3492 ext4_journal_dirty_metadata(handle
, this_bh
);
3494 ext4_error(inode
->i_sb
, __func__
,
3495 "circular indirect block detected, "
3496 "inode=%lu, block=%llu",
3498 (unsigned long long) this_bh
->b_blocknr
);
3503 * ext4_free_branches - free an array of branches
3504 * @handle: JBD handle for this transaction
3505 * @inode: inode we are dealing with
3506 * @parent_bh: the buffer_head which contains *@first and *@last
3507 * @first: array of block numbers
3508 * @last: pointer immediately past the end of array
3509 * @depth: depth of the branches to free
3511 * We are freeing all blocks refered from these branches (numbers are
3512 * stored as little-endian 32-bit) and updating @inode->i_blocks
3515 static void ext4_free_branches(handle_t
*handle
, struct inode
*inode
,
3516 struct buffer_head
*parent_bh
,
3517 __le32
*first
, __le32
*last
, int depth
)
3522 if (is_handle_aborted(handle
))
3526 struct buffer_head
*bh
;
3527 int addr_per_block
= EXT4_ADDR_PER_BLOCK(inode
->i_sb
);
3529 while (--p
>= first
) {
3530 nr
= le32_to_cpu(*p
);
3532 continue; /* A hole */
3534 /* Go read the buffer for the next level down */
3535 bh
= sb_bread(inode
->i_sb
, nr
);
3538 * A read failure? Report error and clear slot
3542 ext4_error(inode
->i_sb
, "ext4_free_branches",
3543 "Read failure, inode=%lu, block=%llu",
3548 /* This zaps the entire block. Bottom up. */
3549 BUFFER_TRACE(bh
, "free child branches");
3550 ext4_free_branches(handle
, inode
, bh
,
3551 (__le32
*) bh
->b_data
,
3552 (__le32
*) bh
->b_data
+ addr_per_block
,
3556 * We've probably journalled the indirect block several
3557 * times during the truncate. But it's no longer
3558 * needed and we now drop it from the transaction via
3559 * jbd2_journal_revoke().
3561 * That's easy if it's exclusively part of this
3562 * transaction. But if it's part of the committing
3563 * transaction then jbd2_journal_forget() will simply
3564 * brelse() it. That means that if the underlying
3565 * block is reallocated in ext4_get_block(),
3566 * unmap_underlying_metadata() will find this block
3567 * and will try to get rid of it. damn, damn.
3569 * If this block has already been committed to the
3570 * journal, a revoke record will be written. And
3571 * revoke records must be emitted *before* clearing
3572 * this block's bit in the bitmaps.
3574 ext4_forget(handle
, 1, inode
, bh
, bh
->b_blocknr
);
3577 * Everything below this this pointer has been
3578 * released. Now let this top-of-subtree go.
3580 * We want the freeing of this indirect block to be
3581 * atomic in the journal with the updating of the
3582 * bitmap block which owns it. So make some room in
3585 * We zero the parent pointer *after* freeing its
3586 * pointee in the bitmaps, so if extend_transaction()
3587 * for some reason fails to put the bitmap changes and
3588 * the release into the same transaction, recovery
3589 * will merely complain about releasing a free block,
3590 * rather than leaking blocks.
3592 if (is_handle_aborted(handle
))
3594 if (try_to_extend_transaction(handle
, inode
)) {
3595 ext4_mark_inode_dirty(handle
, inode
);
3596 ext4_journal_test_restart(handle
, inode
);
3599 ext4_free_blocks(handle
, inode
, nr
, 1, 1);
3603 * The block which we have just freed is
3604 * pointed to by an indirect block: journal it
3606 BUFFER_TRACE(parent_bh
, "get_write_access");
3607 if (!ext4_journal_get_write_access(handle
,
3610 BUFFER_TRACE(parent_bh
,
3611 "call ext4_journal_dirty_metadata");
3612 ext4_journal_dirty_metadata(handle
,
3618 /* We have reached the bottom of the tree. */
3619 BUFFER_TRACE(parent_bh
, "free data blocks");
3620 ext4_free_data(handle
, inode
, parent_bh
, first
, last
);
3624 int ext4_can_truncate(struct inode
*inode
)
3626 if (IS_APPEND(inode
) || IS_IMMUTABLE(inode
))
3628 if (S_ISREG(inode
->i_mode
))
3630 if (S_ISDIR(inode
->i_mode
))
3632 if (S_ISLNK(inode
->i_mode
))
3633 return !ext4_inode_is_fast_symlink(inode
);
3640 * We block out ext4_get_block() block instantiations across the entire
3641 * transaction, and VFS/VM ensures that ext4_truncate() cannot run
3642 * simultaneously on behalf of the same inode.
3644 * As we work through the truncate and commmit bits of it to the journal there
3645 * is one core, guiding principle: the file's tree must always be consistent on
3646 * disk. We must be able to restart the truncate after a crash.
3648 * The file's tree may be transiently inconsistent in memory (although it
3649 * probably isn't), but whenever we close off and commit a journal transaction,
3650 * the contents of (the filesystem + the journal) must be consistent and
3651 * restartable. It's pretty simple, really: bottom up, right to left (although
3652 * left-to-right works OK too).
3654 * Note that at recovery time, journal replay occurs *before* the restart of
3655 * truncate against the orphan inode list.
3657 * The committed inode has the new, desired i_size (which is the same as
3658 * i_disksize in this case). After a crash, ext4_orphan_cleanup() will see
3659 * that this inode's truncate did not complete and it will again call
3660 * ext4_truncate() to have another go. So there will be instantiated blocks
3661 * to the right of the truncation point in a crashed ext4 filesystem. But
3662 * that's fine - as long as they are linked from the inode, the post-crash
3663 * ext4_truncate() run will find them and release them.
3665 void ext4_truncate(struct inode
*inode
)
3668 struct ext4_inode_info
*ei
= EXT4_I(inode
);
3669 __le32
*i_data
= ei
->i_data
;
3670 int addr_per_block
= EXT4_ADDR_PER_BLOCK(inode
->i_sb
);
3671 struct address_space
*mapping
= inode
->i_mapping
;
3672 ext4_lblk_t offsets
[4];
3677 ext4_lblk_t last_block
;
3678 unsigned blocksize
= inode
->i_sb
->s_blocksize
;
3680 if (!ext4_can_truncate(inode
))
3683 if (EXT4_I(inode
)->i_flags
& EXT4_EXTENTS_FL
) {
3684 ext4_ext_truncate(inode
);
3688 handle
= start_transaction(inode
);
3690 return; /* AKPM: return what? */
3692 last_block
= (inode
->i_size
+ blocksize
-1)
3693 >> EXT4_BLOCK_SIZE_BITS(inode
->i_sb
);
3695 if (inode
->i_size
& (blocksize
- 1))
3696 if (ext4_block_truncate_page(handle
, mapping
, inode
->i_size
))
3699 n
= ext4_block_to_path(inode
, last_block
, offsets
, NULL
);
3701 goto out_stop
; /* error */
3704 * OK. This truncate is going to happen. We add the inode to the
3705 * orphan list, so that if this truncate spans multiple transactions,
3706 * and we crash, we will resume the truncate when the filesystem
3707 * recovers. It also marks the inode dirty, to catch the new size.
3709 * Implication: the file must always be in a sane, consistent
3710 * truncatable state while each transaction commits.
3712 if (ext4_orphan_add(handle
, inode
))
3716 * From here we block out all ext4_get_block() callers who want to
3717 * modify the block allocation tree.
3719 down_write(&ei
->i_data_sem
);
3721 ext4_discard_reservation(inode
);
3724 * The orphan list entry will now protect us from any crash which
3725 * occurs before the truncate completes, so it is now safe to propagate
3726 * the new, shorter inode size (held for now in i_size) into the
3727 * on-disk inode. We do this via i_disksize, which is the value which
3728 * ext4 *really* writes onto the disk inode.
3730 ei
->i_disksize
= inode
->i_size
;
3732 if (n
== 1) { /* direct blocks */
3733 ext4_free_data(handle
, inode
, NULL
, i_data
+offsets
[0],
3734 i_data
+ EXT4_NDIR_BLOCKS
);
3738 partial
= ext4_find_shared(inode
, n
, offsets
, chain
, &nr
);
3739 /* Kill the top of shared branch (not detached) */
3741 if (partial
== chain
) {
3742 /* Shared branch grows from the inode */
3743 ext4_free_branches(handle
, inode
, NULL
,
3744 &nr
, &nr
+1, (chain
+n
-1) - partial
);
3747 * We mark the inode dirty prior to restart,
3748 * and prior to stop. No need for it here.
3751 /* Shared branch grows from an indirect block */
3752 BUFFER_TRACE(partial
->bh
, "get_write_access");
3753 ext4_free_branches(handle
, inode
, partial
->bh
,
3755 partial
->p
+1, (chain
+n
-1) - partial
);
3758 /* Clear the ends of indirect blocks on the shared branch */
3759 while (partial
> chain
) {
3760 ext4_free_branches(handle
, inode
, partial
->bh
, partial
->p
+ 1,
3761 (__le32
*)partial
->bh
->b_data
+addr_per_block
,
3762 (chain
+n
-1) - partial
);
3763 BUFFER_TRACE(partial
->bh
, "call brelse");
3764 brelse (partial
->bh
);
3768 /* Kill the remaining (whole) subtrees */
3769 switch (offsets
[0]) {
3771 nr
= i_data
[EXT4_IND_BLOCK
];
3773 ext4_free_branches(handle
, inode
, NULL
, &nr
, &nr
+1, 1);
3774 i_data
[EXT4_IND_BLOCK
] = 0;
3776 case EXT4_IND_BLOCK
:
3777 nr
= i_data
[EXT4_DIND_BLOCK
];
3779 ext4_free_branches(handle
, inode
, NULL
, &nr
, &nr
+1, 2);
3780 i_data
[EXT4_DIND_BLOCK
] = 0;
3782 case EXT4_DIND_BLOCK
:
3783 nr
= i_data
[EXT4_TIND_BLOCK
];
3785 ext4_free_branches(handle
, inode
, NULL
, &nr
, &nr
+1, 3);
3786 i_data
[EXT4_TIND_BLOCK
] = 0;
3788 case EXT4_TIND_BLOCK
:
3792 up_write(&ei
->i_data_sem
);
3793 inode
->i_mtime
= inode
->i_ctime
= ext4_current_time(inode
);
3794 ext4_mark_inode_dirty(handle
, inode
);
3797 * In a multi-transaction truncate, we only make the final transaction
3804 * If this was a simple ftruncate(), and the file will remain alive
3805 * then we need to clear up the orphan record which we created above.
3806 * However, if this was a real unlink then we were called by
3807 * ext4_delete_inode(), and we allow that function to clean up the
3808 * orphan info for us.
3811 ext4_orphan_del(handle
, inode
);
3813 ext4_journal_stop(handle
);
3816 static ext4_fsblk_t
ext4_get_inode_block(struct super_block
*sb
,
3817 unsigned long ino
, struct ext4_iloc
*iloc
)
3819 ext4_group_t block_group
;
3820 unsigned long offset
;
3822 struct ext4_group_desc
*gdp
;
3824 if (!ext4_valid_inum(sb
, ino
)) {
3826 * This error is already checked for in namei.c unless we are
3827 * looking at an NFS filehandle, in which case no error
3833 block_group
= (ino
- 1) / EXT4_INODES_PER_GROUP(sb
);
3834 gdp
= ext4_get_group_desc(sb
, block_group
, NULL
);
3839 * Figure out the offset within the block group inode table
3841 offset
= ((ino
- 1) % EXT4_INODES_PER_GROUP(sb
)) *
3842 EXT4_INODE_SIZE(sb
);
3843 block
= ext4_inode_table(sb
, gdp
) +
3844 (offset
>> EXT4_BLOCK_SIZE_BITS(sb
));
3846 iloc
->block_group
= block_group
;
3847 iloc
->offset
= offset
& (EXT4_BLOCK_SIZE(sb
) - 1);
3852 * ext4_get_inode_loc returns with an extra refcount against the inode's
3853 * underlying buffer_head on success. If 'in_mem' is true, we have all
3854 * data in memory that is needed to recreate the on-disk version of this
3857 static int __ext4_get_inode_loc(struct inode
*inode
,
3858 struct ext4_iloc
*iloc
, int in_mem
)
3861 struct buffer_head
*bh
;
3863 block
= ext4_get_inode_block(inode
->i_sb
, inode
->i_ino
, iloc
);
3867 bh
= sb_getblk(inode
->i_sb
, block
);
3869 ext4_error (inode
->i_sb
, "ext4_get_inode_loc",
3870 "unable to read inode block - "
3871 "inode=%lu, block=%llu",
3872 inode
->i_ino
, block
);
3875 if (!buffer_uptodate(bh
)) {
3879 * If the buffer has the write error flag, we have failed
3880 * to write out another inode in the same block. In this
3881 * case, we don't have to read the block because we may
3882 * read the old inode data successfully.
3884 if (buffer_write_io_error(bh
) && !buffer_uptodate(bh
))
3885 set_buffer_uptodate(bh
);
3887 if (buffer_uptodate(bh
)) {
3888 /* someone brought it uptodate while we waited */
3894 * If we have all information of the inode in memory and this
3895 * is the only valid inode in the block, we need not read the
3899 struct buffer_head
*bitmap_bh
;
3900 struct ext4_group_desc
*desc
;
3901 int inodes_per_buffer
;
3902 int inode_offset
, i
;
3903 ext4_group_t block_group
;
3906 block_group
= (inode
->i_ino
- 1) /
3907 EXT4_INODES_PER_GROUP(inode
->i_sb
);
3908 inodes_per_buffer
= bh
->b_size
/
3909 EXT4_INODE_SIZE(inode
->i_sb
);
3910 inode_offset
= ((inode
->i_ino
- 1) %
3911 EXT4_INODES_PER_GROUP(inode
->i_sb
));
3912 start
= inode_offset
& ~(inodes_per_buffer
- 1);
3914 /* Is the inode bitmap in cache? */
3915 desc
= ext4_get_group_desc(inode
->i_sb
,
3920 bitmap_bh
= sb_getblk(inode
->i_sb
,
3921 ext4_inode_bitmap(inode
->i_sb
, desc
));
3926 * If the inode bitmap isn't in cache then the
3927 * optimisation may end up performing two reads instead
3928 * of one, so skip it.
3930 if (!buffer_uptodate(bitmap_bh
)) {
3934 for (i
= start
; i
< start
+ inodes_per_buffer
; i
++) {
3935 if (i
== inode_offset
)
3937 if (ext4_test_bit(i
, bitmap_bh
->b_data
))
3941 if (i
== start
+ inodes_per_buffer
) {
3942 /* all other inodes are free, so skip I/O */
3943 memset(bh
->b_data
, 0, bh
->b_size
);
3944 set_buffer_uptodate(bh
);
3952 * There are other valid inodes in the buffer, this inode
3953 * has in-inode xattrs, or we don't have this inode in memory.
3954 * Read the block from disk.
3957 bh
->b_end_io
= end_buffer_read_sync
;
3958 submit_bh(READ_META
, bh
);
3960 if (!buffer_uptodate(bh
)) {
3961 ext4_error(inode
->i_sb
, "ext4_get_inode_loc",
3962 "unable to read inode block - "
3963 "inode=%lu, block=%llu",
3964 inode
->i_ino
, block
);
3974 int ext4_get_inode_loc(struct inode
*inode
, struct ext4_iloc
*iloc
)
3976 /* We have all inode data except xattrs in memory here. */
3977 return __ext4_get_inode_loc(inode
, iloc
,
3978 !(EXT4_I(inode
)->i_state
& EXT4_STATE_XATTR
));
3981 void ext4_set_inode_flags(struct inode
*inode
)
3983 unsigned int flags
= EXT4_I(inode
)->i_flags
;
3985 inode
->i_flags
&= ~(S_SYNC
|S_APPEND
|S_IMMUTABLE
|S_NOATIME
|S_DIRSYNC
);
3986 if (flags
& EXT4_SYNC_FL
)
3987 inode
->i_flags
|= S_SYNC
;
3988 if (flags
& EXT4_APPEND_FL
)
3989 inode
->i_flags
|= S_APPEND
;
3990 if (flags
& EXT4_IMMUTABLE_FL
)
3991 inode
->i_flags
|= S_IMMUTABLE
;
3992 if (flags
& EXT4_NOATIME_FL
)
3993 inode
->i_flags
|= S_NOATIME
;
3994 if (flags
& EXT4_DIRSYNC_FL
)
3995 inode
->i_flags
|= S_DIRSYNC
;
3998 /* Propagate flags from i_flags to EXT4_I(inode)->i_flags */
3999 void ext4_get_inode_flags(struct ext4_inode_info
*ei
)
4001 unsigned int flags
= ei
->vfs_inode
.i_flags
;
4003 ei
->i_flags
&= ~(EXT4_SYNC_FL
|EXT4_APPEND_FL
|
4004 EXT4_IMMUTABLE_FL
|EXT4_NOATIME_FL
|EXT4_DIRSYNC_FL
);
4006 ei
->i_flags
|= EXT4_SYNC_FL
;
4007 if (flags
& S_APPEND
)
4008 ei
->i_flags
|= EXT4_APPEND_FL
;
4009 if (flags
& S_IMMUTABLE
)
4010 ei
->i_flags
|= EXT4_IMMUTABLE_FL
;
4011 if (flags
& S_NOATIME
)
4012 ei
->i_flags
|= EXT4_NOATIME_FL
;
4013 if (flags
& S_DIRSYNC
)
4014 ei
->i_flags
|= EXT4_DIRSYNC_FL
;
4016 static blkcnt_t
ext4_inode_blocks(struct ext4_inode
*raw_inode
,
4017 struct ext4_inode_info
*ei
)
4020 struct inode
*inode
= &(ei
->vfs_inode
);
4021 struct super_block
*sb
= inode
->i_sb
;
4023 if (EXT4_HAS_RO_COMPAT_FEATURE(sb
,
4024 EXT4_FEATURE_RO_COMPAT_HUGE_FILE
)) {
4025 /* we are using combined 48 bit field */
4026 i_blocks
= ((u64
)le16_to_cpu(raw_inode
->i_blocks_high
)) << 32 |
4027 le32_to_cpu(raw_inode
->i_blocks_lo
);
4028 if (ei
->i_flags
& EXT4_HUGE_FILE_FL
) {
4029 /* i_blocks represent file system block size */
4030 return i_blocks
<< (inode
->i_blkbits
- 9);
4035 return le32_to_cpu(raw_inode
->i_blocks_lo
);
4039 struct inode
*ext4_iget(struct super_block
*sb
, unsigned long ino
)
4041 struct ext4_iloc iloc
;
4042 struct ext4_inode
*raw_inode
;
4043 struct ext4_inode_info
*ei
;
4044 struct buffer_head
*bh
;
4045 struct inode
*inode
;
4049 inode
= iget_locked(sb
, ino
);
4051 return ERR_PTR(-ENOMEM
);
4052 if (!(inode
->i_state
& I_NEW
))
4056 #ifdef CONFIG_EXT4DEV_FS_POSIX_ACL
4057 ei
->i_acl
= EXT4_ACL_NOT_CACHED
;
4058 ei
->i_default_acl
= EXT4_ACL_NOT_CACHED
;
4060 ei
->i_block_alloc_info
= NULL
;
4062 ret
= __ext4_get_inode_loc(inode
, &iloc
, 0);
4066 raw_inode
= ext4_raw_inode(&iloc
);
4067 inode
->i_mode
= le16_to_cpu(raw_inode
->i_mode
);
4068 inode
->i_uid
= (uid_t
)le16_to_cpu(raw_inode
->i_uid_low
);
4069 inode
->i_gid
= (gid_t
)le16_to_cpu(raw_inode
->i_gid_low
);
4070 if (!(test_opt(inode
->i_sb
, NO_UID32
))) {
4071 inode
->i_uid
|= le16_to_cpu(raw_inode
->i_uid_high
) << 16;
4072 inode
->i_gid
|= le16_to_cpu(raw_inode
->i_gid_high
) << 16;
4074 inode
->i_nlink
= le16_to_cpu(raw_inode
->i_links_count
);
4077 ei
->i_dir_start_lookup
= 0;
4078 ei
->i_dtime
= le32_to_cpu(raw_inode
->i_dtime
);
4079 /* We now have enough fields to check if the inode was active or not.
4080 * This is needed because nfsd might try to access dead inodes
4081 * the test is that same one that e2fsck uses
4082 * NeilBrown 1999oct15
4084 if (inode
->i_nlink
== 0) {
4085 if (inode
->i_mode
== 0 ||
4086 !(EXT4_SB(inode
->i_sb
)->s_mount_state
& EXT4_ORPHAN_FS
)) {
4087 /* this inode is deleted */
4092 /* The only unlinked inodes we let through here have
4093 * valid i_mode and are being read by the orphan
4094 * recovery code: that's fine, we're about to complete
4095 * the process of deleting those. */
4097 ei
->i_flags
= le32_to_cpu(raw_inode
->i_flags
);
4098 inode
->i_blocks
= ext4_inode_blocks(raw_inode
, ei
);
4099 ei
->i_file_acl
= le32_to_cpu(raw_inode
->i_file_acl_lo
);
4100 if (EXT4_SB(inode
->i_sb
)->s_es
->s_creator_os
!=
4101 cpu_to_le32(EXT4_OS_HURD
)) {
4103 ((__u64
)le16_to_cpu(raw_inode
->i_file_acl_high
)) << 32;
4105 inode
->i_size
= ext4_isize(raw_inode
);
4106 ei
->i_disksize
= inode
->i_size
;
4107 inode
->i_generation
= le32_to_cpu(raw_inode
->i_generation
);
4108 ei
->i_block_group
= iloc
.block_group
;
4110 * NOTE! The in-memory inode i_data array is in little-endian order
4111 * even on big-endian machines: we do NOT byteswap the block numbers!
4113 for (block
= 0; block
< EXT4_N_BLOCKS
; block
++)
4114 ei
->i_data
[block
] = raw_inode
->i_block
[block
];
4115 INIT_LIST_HEAD(&ei
->i_orphan
);
4117 if (EXT4_INODE_SIZE(inode
->i_sb
) > EXT4_GOOD_OLD_INODE_SIZE
) {
4118 ei
->i_extra_isize
= le16_to_cpu(raw_inode
->i_extra_isize
);
4119 if (EXT4_GOOD_OLD_INODE_SIZE
+ ei
->i_extra_isize
>
4120 EXT4_INODE_SIZE(inode
->i_sb
)) {
4125 if (ei
->i_extra_isize
== 0) {
4126 /* The extra space is currently unused. Use it. */
4127 ei
->i_extra_isize
= sizeof(struct ext4_inode
) -
4128 EXT4_GOOD_OLD_INODE_SIZE
;
4130 __le32
*magic
= (void *)raw_inode
+
4131 EXT4_GOOD_OLD_INODE_SIZE
+
4133 if (*magic
== cpu_to_le32(EXT4_XATTR_MAGIC
))
4134 ei
->i_state
|= EXT4_STATE_XATTR
;
4137 ei
->i_extra_isize
= 0;
4139 EXT4_INODE_GET_XTIME(i_ctime
, inode
, raw_inode
);
4140 EXT4_INODE_GET_XTIME(i_mtime
, inode
, raw_inode
);
4141 EXT4_INODE_GET_XTIME(i_atime
, inode
, raw_inode
);
4142 EXT4_EINODE_GET_XTIME(i_crtime
, ei
, raw_inode
);
4144 inode
->i_version
= le32_to_cpu(raw_inode
->i_disk_version
);
4145 if (EXT4_INODE_SIZE(inode
->i_sb
) > EXT4_GOOD_OLD_INODE_SIZE
) {
4146 if (EXT4_FITS_IN_INODE(raw_inode
, ei
, i_version_hi
))
4148 (__u64
)(le32_to_cpu(raw_inode
->i_version_hi
)) << 32;
4151 if (S_ISREG(inode
->i_mode
)) {
4152 inode
->i_op
= &ext4_file_inode_operations
;
4153 inode
->i_fop
= &ext4_file_operations
;
4154 ext4_set_aops(inode
);
4155 } else if (S_ISDIR(inode
->i_mode
)) {
4156 inode
->i_op
= &ext4_dir_inode_operations
;
4157 inode
->i_fop
= &ext4_dir_operations
;
4158 } else if (S_ISLNK(inode
->i_mode
)) {
4159 if (ext4_inode_is_fast_symlink(inode
))
4160 inode
->i_op
= &ext4_fast_symlink_inode_operations
;
4162 inode
->i_op
= &ext4_symlink_inode_operations
;
4163 ext4_set_aops(inode
);
4166 inode
->i_op
= &ext4_special_inode_operations
;
4167 if (raw_inode
->i_block
[0])
4168 init_special_inode(inode
, inode
->i_mode
,
4169 old_decode_dev(le32_to_cpu(raw_inode
->i_block
[0])));
4171 init_special_inode(inode
, inode
->i_mode
,
4172 new_decode_dev(le32_to_cpu(raw_inode
->i_block
[1])));
4175 ext4_set_inode_flags(inode
);
4176 unlock_new_inode(inode
);
4181 return ERR_PTR(ret
);
4184 static int ext4_inode_blocks_set(handle_t
*handle
,
4185 struct ext4_inode
*raw_inode
,
4186 struct ext4_inode_info
*ei
)
4188 struct inode
*inode
= &(ei
->vfs_inode
);
4189 u64 i_blocks
= inode
->i_blocks
;
4190 struct super_block
*sb
= inode
->i_sb
;
4193 if (i_blocks
<= ~0U) {
4195 * i_blocks can be represnted in a 32 bit variable
4196 * as multiple of 512 bytes
4198 raw_inode
->i_blocks_lo
= cpu_to_le32(i_blocks
);
4199 raw_inode
->i_blocks_high
= 0;
4200 ei
->i_flags
&= ~EXT4_HUGE_FILE_FL
;
4201 } else if (i_blocks
<= 0xffffffffffffULL
) {
4203 * i_blocks can be represented in a 48 bit variable
4204 * as multiple of 512 bytes
4206 err
= ext4_update_rocompat_feature(handle
, sb
,
4207 EXT4_FEATURE_RO_COMPAT_HUGE_FILE
);
4210 /* i_block is stored in the split 48 bit fields */
4211 raw_inode
->i_blocks_lo
= cpu_to_le32(i_blocks
);
4212 raw_inode
->i_blocks_high
= cpu_to_le16(i_blocks
>> 32);
4213 ei
->i_flags
&= ~EXT4_HUGE_FILE_FL
;
4216 * i_blocks should be represented in a 48 bit variable
4217 * as multiple of file system block size
4219 err
= ext4_update_rocompat_feature(handle
, sb
,
4220 EXT4_FEATURE_RO_COMPAT_HUGE_FILE
);
4223 ei
->i_flags
|= EXT4_HUGE_FILE_FL
;
4224 /* i_block is stored in file system block size */
4225 i_blocks
= i_blocks
>> (inode
->i_blkbits
- 9);
4226 raw_inode
->i_blocks_lo
= cpu_to_le32(i_blocks
);
4227 raw_inode
->i_blocks_high
= cpu_to_le16(i_blocks
>> 32);
4234 * Post the struct inode info into an on-disk inode location in the
4235 * buffer-cache. This gobbles the caller's reference to the
4236 * buffer_head in the inode location struct.
4238 * The caller must have write access to iloc->bh.
4240 static int ext4_do_update_inode(handle_t
*handle
,
4241 struct inode
*inode
,
4242 struct ext4_iloc
*iloc
)
4244 struct ext4_inode
*raw_inode
= ext4_raw_inode(iloc
);
4245 struct ext4_inode_info
*ei
= EXT4_I(inode
);
4246 struct buffer_head
*bh
= iloc
->bh
;
4247 int err
= 0, rc
, block
;
4249 /* For fields not not tracking in the in-memory inode,
4250 * initialise them to zero for new inodes. */
4251 if (ei
->i_state
& EXT4_STATE_NEW
)
4252 memset(raw_inode
, 0, EXT4_SB(inode
->i_sb
)->s_inode_size
);
4254 ext4_get_inode_flags(ei
);
4255 raw_inode
->i_mode
= cpu_to_le16(inode
->i_mode
);
4256 if (!(test_opt(inode
->i_sb
, NO_UID32
))) {
4257 raw_inode
->i_uid_low
= cpu_to_le16(low_16_bits(inode
->i_uid
));
4258 raw_inode
->i_gid_low
= cpu_to_le16(low_16_bits(inode
->i_gid
));
4260 * Fix up interoperability with old kernels. Otherwise, old inodes get
4261 * re-used with the upper 16 bits of the uid/gid intact
4264 raw_inode
->i_uid_high
=
4265 cpu_to_le16(high_16_bits(inode
->i_uid
));
4266 raw_inode
->i_gid_high
=
4267 cpu_to_le16(high_16_bits(inode
->i_gid
));
4269 raw_inode
->i_uid_high
= 0;
4270 raw_inode
->i_gid_high
= 0;
4273 raw_inode
->i_uid_low
=
4274 cpu_to_le16(fs_high2lowuid(inode
->i_uid
));
4275 raw_inode
->i_gid_low
=
4276 cpu_to_le16(fs_high2lowgid(inode
->i_gid
));
4277 raw_inode
->i_uid_high
= 0;
4278 raw_inode
->i_gid_high
= 0;
4280 raw_inode
->i_links_count
= cpu_to_le16(inode
->i_nlink
);
4282 EXT4_INODE_SET_XTIME(i_ctime
, inode
, raw_inode
);
4283 EXT4_INODE_SET_XTIME(i_mtime
, inode
, raw_inode
);
4284 EXT4_INODE_SET_XTIME(i_atime
, inode
, raw_inode
);
4285 EXT4_EINODE_SET_XTIME(i_crtime
, ei
, raw_inode
);
4287 if (ext4_inode_blocks_set(handle
, raw_inode
, ei
))
4289 raw_inode
->i_dtime
= cpu_to_le32(ei
->i_dtime
);
4290 /* clear the migrate flag in the raw_inode */
4291 raw_inode
->i_flags
= cpu_to_le32(ei
->i_flags
& ~EXT4_EXT_MIGRATE
);
4292 if (EXT4_SB(inode
->i_sb
)->s_es
->s_creator_os
!=
4293 cpu_to_le32(EXT4_OS_HURD
))
4294 raw_inode
->i_file_acl_high
=
4295 cpu_to_le16(ei
->i_file_acl
>> 32);
4296 raw_inode
->i_file_acl_lo
= cpu_to_le32(ei
->i_file_acl
);
4297 ext4_isize_set(raw_inode
, ei
->i_disksize
);
4298 if (ei
->i_disksize
> 0x7fffffffULL
) {
4299 struct super_block
*sb
= inode
->i_sb
;
4300 if (!EXT4_HAS_RO_COMPAT_FEATURE(sb
,
4301 EXT4_FEATURE_RO_COMPAT_LARGE_FILE
) ||
4302 EXT4_SB(sb
)->s_es
->s_rev_level
==
4303 cpu_to_le32(EXT4_GOOD_OLD_REV
)) {
4304 /* If this is the first large file
4305 * created, add a flag to the superblock.
4307 err
= ext4_journal_get_write_access(handle
,
4308 EXT4_SB(sb
)->s_sbh
);
4311 ext4_update_dynamic_rev(sb
);
4312 EXT4_SET_RO_COMPAT_FEATURE(sb
,
4313 EXT4_FEATURE_RO_COMPAT_LARGE_FILE
);
4316 err
= ext4_journal_dirty_metadata(handle
,
4317 EXT4_SB(sb
)->s_sbh
);
4320 raw_inode
->i_generation
= cpu_to_le32(inode
->i_generation
);
4321 if (S_ISCHR(inode
->i_mode
) || S_ISBLK(inode
->i_mode
)) {
4322 if (old_valid_dev(inode
->i_rdev
)) {
4323 raw_inode
->i_block
[0] =
4324 cpu_to_le32(old_encode_dev(inode
->i_rdev
));
4325 raw_inode
->i_block
[1] = 0;
4327 raw_inode
->i_block
[0] = 0;
4328 raw_inode
->i_block
[1] =
4329 cpu_to_le32(new_encode_dev(inode
->i_rdev
));
4330 raw_inode
->i_block
[2] = 0;
4332 } else for (block
= 0; block
< EXT4_N_BLOCKS
; block
++)
4333 raw_inode
->i_block
[block
] = ei
->i_data
[block
];
4335 raw_inode
->i_disk_version
= cpu_to_le32(inode
->i_version
);
4336 if (ei
->i_extra_isize
) {
4337 if (EXT4_FITS_IN_INODE(raw_inode
, ei
, i_version_hi
))
4338 raw_inode
->i_version_hi
=
4339 cpu_to_le32(inode
->i_version
>> 32);
4340 raw_inode
->i_extra_isize
= cpu_to_le16(ei
->i_extra_isize
);
4344 BUFFER_TRACE(bh
, "call ext4_journal_dirty_metadata");
4345 rc
= ext4_journal_dirty_metadata(handle
, bh
);
4348 ei
->i_state
&= ~EXT4_STATE_NEW
;
4352 ext4_std_error(inode
->i_sb
, err
);
4357 * ext4_write_inode()
4359 * We are called from a few places:
4361 * - Within generic_file_write() for O_SYNC files.
4362 * Here, there will be no transaction running. We wait for any running
4363 * trasnaction to commit.
4365 * - Within sys_sync(), kupdate and such.
4366 * We wait on commit, if tol to.
4368 * - Within prune_icache() (PF_MEMALLOC == true)
4369 * Here we simply return. We can't afford to block kswapd on the
4372 * In all cases it is actually safe for us to return without doing anything,
4373 * because the inode has been copied into a raw inode buffer in
4374 * ext4_mark_inode_dirty(). This is a correctness thing for O_SYNC and for
4377 * Note that we are absolutely dependent upon all inode dirtiers doing the
4378 * right thing: they *must* call mark_inode_dirty() after dirtying info in
4379 * which we are interested.
4381 * It would be a bug for them to not do this. The code:
4383 * mark_inode_dirty(inode)
4385 * inode->i_size = expr;
4387 * is in error because a kswapd-driven write_inode() could occur while
4388 * `stuff()' is running, and the new i_size will be lost. Plus the inode
4389 * will no longer be on the superblock's dirty inode list.
4391 int ext4_write_inode(struct inode
*inode
, int wait
)
4393 if (current
->flags
& PF_MEMALLOC
)
4396 if (ext4_journal_current_handle()) {
4397 jbd_debug(1, "called recursively, non-PF_MEMALLOC!\n");
4405 return ext4_force_commit(inode
->i_sb
);
4411 * Called from notify_change.
4413 * We want to trap VFS attempts to truncate the file as soon as
4414 * possible. In particular, we want to make sure that when the VFS
4415 * shrinks i_size, we put the inode on the orphan list and modify
4416 * i_disksize immediately, so that during the subsequent flushing of
4417 * dirty pages and freeing of disk blocks, we can guarantee that any
4418 * commit will leave the blocks being flushed in an unused state on
4419 * disk. (On recovery, the inode will get truncated and the blocks will
4420 * be freed, so we have a strong guarantee that no future commit will
4421 * leave these blocks visible to the user.)
4423 * Another thing we have to assure is that if we are in ordered mode
4424 * and inode is still attached to the committing transaction, we must
4425 * we start writeout of all the dirty pages which are being truncated.
4426 * This way we are sure that all the data written in the previous
4427 * transaction are already on disk (truncate waits for pages under
4430 * Called with inode->i_mutex down.
4432 int ext4_setattr(struct dentry
*dentry
, struct iattr
*attr
)
4434 struct inode
*inode
= dentry
->d_inode
;
4436 const unsigned int ia_valid
= attr
->ia_valid
;
4438 error
= inode_change_ok(inode
, attr
);
4442 if ((ia_valid
& ATTR_UID
&& attr
->ia_uid
!= inode
->i_uid
) ||
4443 (ia_valid
& ATTR_GID
&& attr
->ia_gid
!= inode
->i_gid
)) {
4446 /* (user+group)*(old+new) structure, inode write (sb,
4447 * inode block, ? - but truncate inode update has it) */
4448 handle
= ext4_journal_start(inode
, 2*(EXT4_QUOTA_INIT_BLOCKS(inode
->i_sb
)+
4449 EXT4_QUOTA_DEL_BLOCKS(inode
->i_sb
))+3);
4450 if (IS_ERR(handle
)) {
4451 error
= PTR_ERR(handle
);
4454 error
= DQUOT_TRANSFER(inode
, attr
) ? -EDQUOT
: 0;
4456 ext4_journal_stop(handle
);
4459 /* Update corresponding info in inode so that everything is in
4460 * one transaction */
4461 if (attr
->ia_valid
& ATTR_UID
)
4462 inode
->i_uid
= attr
->ia_uid
;
4463 if (attr
->ia_valid
& ATTR_GID
)
4464 inode
->i_gid
= attr
->ia_gid
;
4465 error
= ext4_mark_inode_dirty(handle
, inode
);
4466 ext4_journal_stop(handle
);
4469 if (attr
->ia_valid
& ATTR_SIZE
) {
4470 if (!(EXT4_I(inode
)->i_flags
& EXT4_EXTENTS_FL
)) {
4471 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
4473 if (attr
->ia_size
> sbi
->s_bitmap_maxbytes
) {
4480 if (S_ISREG(inode
->i_mode
) &&
4481 attr
->ia_valid
& ATTR_SIZE
&& attr
->ia_size
< inode
->i_size
) {
4484 handle
= ext4_journal_start(inode
, 3);
4485 if (IS_ERR(handle
)) {
4486 error
= PTR_ERR(handle
);
4490 error
= ext4_orphan_add(handle
, inode
);
4491 EXT4_I(inode
)->i_disksize
= attr
->ia_size
;
4492 rc
= ext4_mark_inode_dirty(handle
, inode
);
4495 ext4_journal_stop(handle
);
4497 if (ext4_should_order_data(inode
)) {
4498 error
= ext4_begin_ordered_truncate(inode
,
4501 /* Do as much error cleanup as possible */
4502 handle
= ext4_journal_start(inode
, 3);
4503 if (IS_ERR(handle
)) {
4504 ext4_orphan_del(NULL
, inode
);
4507 ext4_orphan_del(handle
, inode
);
4508 ext4_journal_stop(handle
);
4514 rc
= inode_setattr(inode
, attr
);
4516 /* If inode_setattr's call to ext4_truncate failed to get a
4517 * transaction handle at all, we need to clean up the in-core
4518 * orphan list manually. */
4520 ext4_orphan_del(NULL
, inode
);
4522 if (!rc
&& (ia_valid
& ATTR_MODE
))
4523 rc
= ext4_acl_chmod(inode
);
4526 ext4_std_error(inode
->i_sb
, error
);
4532 int ext4_getattr(struct vfsmount
*mnt
, struct dentry
*dentry
,
4535 struct inode
*inode
;
4536 unsigned long delalloc_blocks
;
4538 inode
= dentry
->d_inode
;
4539 generic_fillattr(inode
, stat
);
4542 * We can't update i_blocks if the block allocation is delayed
4543 * otherwise in the case of system crash before the real block
4544 * allocation is done, we will have i_blocks inconsistent with
4545 * on-disk file blocks.
4546 * We always keep i_blocks updated together with real
4547 * allocation. But to not confuse with user, stat
4548 * will return the blocks that include the delayed allocation
4549 * blocks for this file.
4551 spin_lock(&EXT4_I(inode
)->i_block_reservation_lock
);
4552 delalloc_blocks
= EXT4_I(inode
)->i_reserved_data_blocks
;
4553 spin_unlock(&EXT4_I(inode
)->i_block_reservation_lock
);
4555 stat
->blocks
+= (delalloc_blocks
<< inode
->i_sb
->s_blocksize_bits
)>>9;
4559 static int ext4_indirect_trans_blocks(struct inode
*inode
, int nrblocks
,
4564 /* if nrblocks are contiguous */
4567 * With N contiguous data blocks, it need at most
4568 * N/EXT4_ADDR_PER_BLOCK(inode->i_sb) indirect blocks
4569 * 2 dindirect blocks
4572 indirects
= nrblocks
/ EXT4_ADDR_PER_BLOCK(inode
->i_sb
);
4573 return indirects
+ 3;
4576 * if nrblocks are not contiguous, worse case, each block touch
4577 * a indirect block, and each indirect block touch a double indirect
4578 * block, plus a triple indirect block
4580 indirects
= nrblocks
* 2 + 1;
4584 static int ext4_index_trans_blocks(struct inode
*inode
, int nrblocks
, int chunk
)
4586 if (!(EXT4_I(inode
)->i_flags
& EXT4_EXTENTS_FL
))
4587 return ext4_indirect_trans_blocks(inode
, nrblocks
, 0);
4588 return ext4_ext_index_trans_blocks(inode
, nrblocks
, 0);
4591 * Account for index blocks, block groups bitmaps and block group
4592 * descriptor blocks if modify datablocks and index blocks
4593 * worse case, the indexs blocks spread over different block groups
4595 * If datablocks are discontiguous, they are possible to spread over
4596 * different block groups too. If they are contiugous, with flexbg,
4597 * they could still across block group boundary.
4599 * Also account for superblock, inode, quota and xattr blocks
4601 int ext4_meta_trans_blocks(struct inode
*inode
, int nrblocks
, int chunk
)
4603 int groups
, gdpblocks
;
4608 * How many index blocks need to touch to modify nrblocks?
4609 * The "Chunk" flag indicating whether the nrblocks is
4610 * physically contiguous on disk
4612 * For Direct IO and fallocate, they calls get_block to allocate
4613 * one single extent at a time, so they could set the "Chunk" flag
4615 idxblocks
= ext4_index_trans_blocks(inode
, nrblocks
, chunk
);
4620 * Now let's see how many group bitmaps and group descriptors need
4630 if (groups
> EXT4_SB(inode
->i_sb
)->s_groups_count
)
4631 groups
= EXT4_SB(inode
->i_sb
)->s_groups_count
;
4632 if (groups
> EXT4_SB(inode
->i_sb
)->s_gdb_count
)
4633 gdpblocks
= EXT4_SB(inode
->i_sb
)->s_gdb_count
;
4635 /* bitmaps and block group descriptor blocks */
4636 ret
+= groups
+ gdpblocks
;
4638 /* Blocks for super block, inode, quota and xattr blocks */
4639 ret
+= EXT4_META_TRANS_BLOCKS(inode
->i_sb
);
4645 * Calulate the total number of credits to reserve to fit
4646 * the modification of a single pages into a single transaction,
4647 * which may include multiple chunks of block allocations.
4649 * This could be called via ext4_write_begin()
4651 * We need to consider the worse case, when
4652 * one new block per extent.
4654 int ext4_writepage_trans_blocks(struct inode
*inode
)
4656 int bpp
= ext4_journal_blocks_per_page(inode
);
4659 ret
= ext4_meta_trans_blocks(inode
, bpp
, 0);
4661 /* Account for data blocks for journalled mode */
4662 if (ext4_should_journal_data(inode
))
4668 * Calculate the journal credits for a chunk of data modification.
4670 * This is called from DIO, fallocate or whoever calling
4671 * ext4_get_blocks_wrap() to map/allocate a chunk of contigous disk blocks.
4673 * journal buffers for data blocks are not included here, as DIO
4674 * and fallocate do no need to journal data buffers.
4676 int ext4_chunk_trans_blocks(struct inode
*inode
, int nrblocks
)
4678 return ext4_meta_trans_blocks(inode
, nrblocks
, 1);
4682 * The caller must have previously called ext4_reserve_inode_write().
4683 * Give this, we know that the caller already has write access to iloc->bh.
4685 int ext4_mark_iloc_dirty(handle_t
*handle
,
4686 struct inode
*inode
, struct ext4_iloc
*iloc
)
4690 if (test_opt(inode
->i_sb
, I_VERSION
))
4691 inode_inc_iversion(inode
);
4693 /* the do_update_inode consumes one bh->b_count */
4696 /* ext4_do_update_inode() does jbd2_journal_dirty_metadata */
4697 err
= ext4_do_update_inode(handle
, inode
, iloc
);
4703 * On success, We end up with an outstanding reference count against
4704 * iloc->bh. This _must_ be cleaned up later.
4708 ext4_reserve_inode_write(handle_t
*handle
, struct inode
*inode
,
4709 struct ext4_iloc
*iloc
)
4713 err
= ext4_get_inode_loc(inode
, iloc
);
4715 BUFFER_TRACE(iloc
->bh
, "get_write_access");
4716 err
= ext4_journal_get_write_access(handle
, iloc
->bh
);
4723 ext4_std_error(inode
->i_sb
, err
);
4728 * Expand an inode by new_extra_isize bytes.
4729 * Returns 0 on success or negative error number on failure.
4731 static int ext4_expand_extra_isize(struct inode
*inode
,
4732 unsigned int new_extra_isize
,
4733 struct ext4_iloc iloc
,
4736 struct ext4_inode
*raw_inode
;
4737 struct ext4_xattr_ibody_header
*header
;
4738 struct ext4_xattr_entry
*entry
;
4740 if (EXT4_I(inode
)->i_extra_isize
>= new_extra_isize
)
4743 raw_inode
= ext4_raw_inode(&iloc
);
4745 header
= IHDR(inode
, raw_inode
);
4746 entry
= IFIRST(header
);
4748 /* No extended attributes present */
4749 if (!(EXT4_I(inode
)->i_state
& EXT4_STATE_XATTR
) ||
4750 header
->h_magic
!= cpu_to_le32(EXT4_XATTR_MAGIC
)) {
4751 memset((void *)raw_inode
+ EXT4_GOOD_OLD_INODE_SIZE
, 0,
4753 EXT4_I(inode
)->i_extra_isize
= new_extra_isize
;
4757 /* try to expand with EAs present */
4758 return ext4_expand_extra_isize_ea(inode
, new_extra_isize
,
4763 * What we do here is to mark the in-core inode as clean with respect to inode
4764 * dirtiness (it may still be data-dirty).
4765 * This means that the in-core inode may be reaped by prune_icache
4766 * without having to perform any I/O. This is a very good thing,
4767 * because *any* task may call prune_icache - even ones which
4768 * have a transaction open against a different journal.
4770 * Is this cheating? Not really. Sure, we haven't written the
4771 * inode out, but prune_icache isn't a user-visible syncing function.
4772 * Whenever the user wants stuff synced (sys_sync, sys_msync, sys_fsync)
4773 * we start and wait on commits.
4775 * Is this efficient/effective? Well, we're being nice to the system
4776 * by cleaning up our inodes proactively so they can be reaped
4777 * without I/O. But we are potentially leaving up to five seconds'
4778 * worth of inodes floating about which prune_icache wants us to
4779 * write out. One way to fix that would be to get prune_icache()
4780 * to do a write_super() to free up some memory. It has the desired
4783 int ext4_mark_inode_dirty(handle_t
*handle
, struct inode
*inode
)
4785 struct ext4_iloc iloc
;
4786 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
4787 static unsigned int mnt_count
;
4791 err
= ext4_reserve_inode_write(handle
, inode
, &iloc
);
4792 if (EXT4_I(inode
)->i_extra_isize
< sbi
->s_want_extra_isize
&&
4793 !(EXT4_I(inode
)->i_state
& EXT4_STATE_NO_EXPAND
)) {
4795 * We need extra buffer credits since we may write into EA block
4796 * with this same handle. If journal_extend fails, then it will
4797 * only result in a minor loss of functionality for that inode.
4798 * If this is felt to be critical, then e2fsck should be run to
4799 * force a large enough s_min_extra_isize.
4801 if ((jbd2_journal_extend(handle
,
4802 EXT4_DATA_TRANS_BLOCKS(inode
->i_sb
))) == 0) {
4803 ret
= ext4_expand_extra_isize(inode
,
4804 sbi
->s_want_extra_isize
,
4807 EXT4_I(inode
)->i_state
|= EXT4_STATE_NO_EXPAND
;
4809 le16_to_cpu(sbi
->s_es
->s_mnt_count
)) {
4810 ext4_warning(inode
->i_sb
, __func__
,
4811 "Unable to expand inode %lu. Delete"
4812 " some EAs or run e2fsck.",
4815 le16_to_cpu(sbi
->s_es
->s_mnt_count
);
4821 err
= ext4_mark_iloc_dirty(handle
, inode
, &iloc
);
4826 * ext4_dirty_inode() is called from __mark_inode_dirty()
4828 * We're really interested in the case where a file is being extended.
4829 * i_size has been changed by generic_commit_write() and we thus need
4830 * to include the updated inode in the current transaction.
4832 * Also, DQUOT_ALLOC_SPACE() will always dirty the inode when blocks
4833 * are allocated to the file.
4835 * If the inode is marked synchronous, we don't honour that here - doing
4836 * so would cause a commit on atime updates, which we don't bother doing.
4837 * We handle synchronous inodes at the highest possible level.
4839 void ext4_dirty_inode(struct inode
*inode
)
4841 handle_t
*current_handle
= ext4_journal_current_handle();
4844 handle
= ext4_journal_start(inode
, 2);
4847 if (current_handle
&&
4848 current_handle
->h_transaction
!= handle
->h_transaction
) {
4849 /* This task has a transaction open against a different fs */
4850 printk(KERN_EMERG
"%s: transactions do not match!\n",
4853 jbd_debug(5, "marking dirty. outer handle=%p\n",
4855 ext4_mark_inode_dirty(handle
, inode
);
4857 ext4_journal_stop(handle
);
4864 * Bind an inode's backing buffer_head into this transaction, to prevent
4865 * it from being flushed to disk early. Unlike
4866 * ext4_reserve_inode_write, this leaves behind no bh reference and
4867 * returns no iloc structure, so the caller needs to repeat the iloc
4868 * lookup to mark the inode dirty later.
4870 static int ext4_pin_inode(handle_t
*handle
, struct inode
*inode
)
4872 struct ext4_iloc iloc
;
4876 err
= ext4_get_inode_loc(inode
, &iloc
);
4878 BUFFER_TRACE(iloc
.bh
, "get_write_access");
4879 err
= jbd2_journal_get_write_access(handle
, iloc
.bh
);
4881 err
= ext4_journal_dirty_metadata(handle
,
4886 ext4_std_error(inode
->i_sb
, err
);
4891 int ext4_change_inode_journal_flag(struct inode
*inode
, int val
)
4898 * We have to be very careful here: changing a data block's
4899 * journaling status dynamically is dangerous. If we write a
4900 * data block to the journal, change the status and then delete
4901 * that block, we risk forgetting to revoke the old log record
4902 * from the journal and so a subsequent replay can corrupt data.
4903 * So, first we make sure that the journal is empty and that
4904 * nobody is changing anything.
4907 journal
= EXT4_JOURNAL(inode
);
4908 if (is_journal_aborted(journal
))
4911 jbd2_journal_lock_updates(journal
);
4912 jbd2_journal_flush(journal
);
4915 * OK, there are no updates running now, and all cached data is
4916 * synced to disk. We are now in a completely consistent state
4917 * which doesn't have anything in the journal, and we know that
4918 * no filesystem updates are running, so it is safe to modify
4919 * the inode's in-core data-journaling state flag now.
4923 EXT4_I(inode
)->i_flags
|= EXT4_JOURNAL_DATA_FL
;
4925 EXT4_I(inode
)->i_flags
&= ~EXT4_JOURNAL_DATA_FL
;
4926 ext4_set_aops(inode
);
4928 jbd2_journal_unlock_updates(journal
);
4930 /* Finally we can mark the inode as dirty. */
4932 handle
= ext4_journal_start(inode
, 1);
4934 return PTR_ERR(handle
);
4936 err
= ext4_mark_inode_dirty(handle
, inode
);
4938 ext4_journal_stop(handle
);
4939 ext4_std_error(inode
->i_sb
, err
);
4944 static int ext4_bh_unmapped(handle_t
*handle
, struct buffer_head
*bh
)
4946 return !buffer_mapped(bh
);
4949 int ext4_page_mkwrite(struct vm_area_struct
*vma
, struct page
*page
)
4955 struct file
*file
= vma
->vm_file
;
4956 struct inode
*inode
= file
->f_path
.dentry
->d_inode
;
4957 struct address_space
*mapping
= inode
->i_mapping
;
4960 * Get i_alloc_sem to stop truncates messing with the inode. We cannot
4961 * get i_mutex because we are already holding mmap_sem.
4963 down_read(&inode
->i_alloc_sem
);
4964 size
= i_size_read(inode
);
4965 if (page
->mapping
!= mapping
|| size
<= page_offset(page
)
4966 || !PageUptodate(page
)) {
4967 /* page got truncated from under us? */
4971 if (PageMappedToDisk(page
))
4974 if (page
->index
== size
>> PAGE_CACHE_SHIFT
)
4975 len
= size
& ~PAGE_CACHE_MASK
;
4977 len
= PAGE_CACHE_SIZE
;
4979 if (page_has_buffers(page
)) {
4980 /* return if we have all the buffers mapped */
4981 if (!walk_page_buffers(NULL
, page_buffers(page
), 0, len
, NULL
,
4986 * OK, we need to fill the hole... Do write_begin write_end
4987 * to do block allocation/reservation.We are not holding
4988 * inode.i__mutex here. That allow * parallel write_begin,
4989 * write_end call. lock_page prevent this from happening
4990 * on the same page though
4992 ret
= mapping
->a_ops
->write_begin(file
, mapping
, page_offset(page
),
4993 len
, AOP_FLAG_UNINTERRUPTIBLE
, &page
, &fsdata
);
4996 ret
= mapping
->a_ops
->write_end(file
, mapping
, page_offset(page
),
4997 len
, len
, page
, fsdata
);
5002 up_read(&inode
->i_alloc_sem
);