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
;
1640 * mpage_da_submit_io - walks through extent of pages and try to write
1641 * them with writepage() call back
1643 * @mpd->inode: inode
1644 * @mpd->first_page: first page of the extent
1645 * @mpd->next_page: page after the last page of the extent
1646 * @mpd->get_block: the filesystem's block mapper function
1648 * By the time mpage_da_submit_io() is called we expect all blocks
1649 * to be allocated. this may be wrong if allocation failed.
1651 * As pages are already locked by write_cache_pages(), we can't use it
1653 static int mpage_da_submit_io(struct mpage_da_data
*mpd
)
1655 struct address_space
*mapping
= mpd
->inode
->i_mapping
;
1656 int ret
= 0, err
, nr_pages
, i
;
1657 unsigned long index
, end
;
1658 struct pagevec pvec
;
1660 BUG_ON(mpd
->next_page
<= mpd
->first_page
);
1661 pagevec_init(&pvec
, 0);
1662 index
= mpd
->first_page
;
1663 end
= mpd
->next_page
- 1;
1665 while (index
<= end
) {
1666 /* XXX: optimize tail */
1667 nr_pages
= pagevec_lookup(&pvec
, mapping
, index
, PAGEVEC_SIZE
);
1670 for (i
= 0; i
< nr_pages
; i
++) {
1671 struct page
*page
= pvec
.pages
[i
];
1673 index
= page
->index
;
1678 err
= mapping
->a_ops
->writepage(page
, mpd
->wbc
);
1680 mpd
->pages_written
++;
1682 * In error case, we have to continue because
1683 * remaining pages are still locked
1684 * XXX: unlock and re-dirty them?
1689 pagevec_release(&pvec
);
1695 * mpage_put_bnr_to_bhs - walk blocks and assign them actual numbers
1697 * @mpd->inode - inode to walk through
1698 * @exbh->b_blocknr - first block on a disk
1699 * @exbh->b_size - amount of space in bytes
1700 * @logical - first logical block to start assignment with
1702 * the function goes through all passed space and put actual disk
1703 * block numbers into buffer heads, dropping BH_Delay
1705 static void mpage_put_bnr_to_bhs(struct mpage_da_data
*mpd
, sector_t logical
,
1706 struct buffer_head
*exbh
)
1708 struct inode
*inode
= mpd
->inode
;
1709 struct address_space
*mapping
= inode
->i_mapping
;
1710 int blocks
= exbh
->b_size
>> inode
->i_blkbits
;
1711 sector_t pblock
= exbh
->b_blocknr
, cur_logical
;
1712 struct buffer_head
*head
, *bh
;
1714 struct pagevec pvec
;
1717 index
= logical
>> (PAGE_CACHE_SHIFT
- inode
->i_blkbits
);
1718 end
= (logical
+ blocks
- 1) >> (PAGE_CACHE_SHIFT
- inode
->i_blkbits
);
1719 cur_logical
= index
<< (PAGE_CACHE_SHIFT
- inode
->i_blkbits
);
1721 pagevec_init(&pvec
, 0);
1723 while (index
<= end
) {
1724 /* XXX: optimize tail */
1725 nr_pages
= pagevec_lookup(&pvec
, mapping
, index
, PAGEVEC_SIZE
);
1728 for (i
= 0; i
< nr_pages
; i
++) {
1729 struct page
*page
= pvec
.pages
[i
];
1731 index
= page
->index
;
1736 BUG_ON(!PageLocked(page
));
1737 BUG_ON(PageWriteback(page
));
1738 BUG_ON(!page_has_buffers(page
));
1740 bh
= page_buffers(page
);
1743 /* skip blocks out of the range */
1745 if (cur_logical
>= logical
)
1748 } while ((bh
= bh
->b_this_page
) != head
);
1751 if (cur_logical
>= logical
+ blocks
)
1753 if (buffer_delay(bh
)) {
1754 bh
->b_blocknr
= pblock
;
1755 clear_buffer_delay(bh
);
1756 bh
->b_bdev
= inode
->i_sb
->s_bdev
;
1757 } else if (buffer_unwritten(bh
)) {
1758 bh
->b_blocknr
= pblock
;
1759 clear_buffer_unwritten(bh
);
1760 set_buffer_mapped(bh
);
1762 bh
->b_bdev
= inode
->i_sb
->s_bdev
;
1763 } else if (buffer_mapped(bh
))
1764 BUG_ON(bh
->b_blocknr
!= pblock
);
1768 } while ((bh
= bh
->b_this_page
) != head
);
1770 pagevec_release(&pvec
);
1776 * __unmap_underlying_blocks - just a helper function to unmap
1777 * set of blocks described by @bh
1779 static inline void __unmap_underlying_blocks(struct inode
*inode
,
1780 struct buffer_head
*bh
)
1782 struct block_device
*bdev
= inode
->i_sb
->s_bdev
;
1785 blocks
= bh
->b_size
>> inode
->i_blkbits
;
1786 for (i
= 0; i
< blocks
; i
++)
1787 unmap_underlying_metadata(bdev
, bh
->b_blocknr
+ i
);
1790 static void ext4_da_block_invalidatepages(struct mpage_da_data
*mpd
,
1791 sector_t logical
, long blk_cnt
)
1795 struct pagevec pvec
;
1796 struct inode
*inode
= mpd
->inode
;
1797 struct address_space
*mapping
= inode
->i_mapping
;
1799 index
= logical
>> (PAGE_CACHE_SHIFT
- inode
->i_blkbits
);
1800 end
= (logical
+ blk_cnt
- 1) >>
1801 (PAGE_CACHE_SHIFT
- inode
->i_blkbits
);
1802 while (index
<= end
) {
1803 nr_pages
= pagevec_lookup(&pvec
, mapping
, index
, PAGEVEC_SIZE
);
1806 for (i
= 0; i
< nr_pages
; i
++) {
1807 struct page
*page
= pvec
.pages
[i
];
1808 index
= page
->index
;
1813 BUG_ON(!PageLocked(page
));
1814 BUG_ON(PageWriteback(page
));
1815 block_invalidatepage(page
, 0);
1816 ClearPageUptodate(page
);
1824 * mpage_da_map_blocks - go through given space
1826 * @mpd->lbh - bh describing space
1827 * @mpd->get_block - the filesystem's block mapper function
1829 * The function skips space we know is already mapped to disk blocks.
1832 static int mpage_da_map_blocks(struct mpage_da_data
*mpd
)
1835 struct buffer_head
new;
1836 struct buffer_head
*lbh
= &mpd
->lbh
;
1837 sector_t next
= lbh
->b_blocknr
;
1840 * We consider only non-mapped and non-allocated blocks
1842 if (buffer_mapped(lbh
) && !buffer_delay(lbh
))
1844 new.b_state
= lbh
->b_state
;
1846 new.b_size
= lbh
->b_size
;
1848 * If we didn't accumulate anything
1849 * to write simply return
1853 err
= mpd
->get_block(mpd
->inode
, next
, &new, 1);
1856 /* If get block returns with error
1857 * we simply return. Later writepage
1858 * will redirty the page and writepages
1859 * will find the dirty page again
1864 * get block failure will cause us
1865 * to loop in writepages. Because
1866 * a_ops->writepage won't be able to
1867 * make progress. The page will be redirtied
1868 * by writepage and writepages will again
1869 * try to write the same.
1871 printk(KERN_EMERG
"%s block allocation failed for inode %lu "
1872 "at logical offset %llu with max blocks "
1873 "%zd with error %d\n",
1874 __func__
, mpd
->inode
->i_ino
,
1875 (unsigned long long)next
,
1876 lbh
->b_size
>> mpd
->inode
->i_blkbits
, err
);
1877 printk(KERN_EMERG
"This should not happen.!! "
1878 "Data will be lost\n");
1879 if (err
== -ENOSPC
) {
1880 printk(KERN_CRIT
"Total free blocks count %lld\n",
1881 ext4_count_free_blocks(mpd
->inode
->i_sb
));
1883 /* invlaidate all the pages */
1884 ext4_da_block_invalidatepages(mpd
, next
,
1885 lbh
->b_size
>> mpd
->inode
->i_blkbits
);
1888 BUG_ON(new.b_size
== 0);
1890 if (buffer_new(&new))
1891 __unmap_underlying_blocks(mpd
->inode
, &new);
1894 * If blocks are delayed marked, we need to
1895 * put actual blocknr and drop delayed bit
1897 if (buffer_delay(lbh
) || buffer_unwritten(lbh
))
1898 mpage_put_bnr_to_bhs(mpd
, next
, &new);
1903 #define BH_FLAGS ((1 << BH_Uptodate) | (1 << BH_Mapped) | \
1904 (1 << BH_Delay) | (1 << BH_Unwritten))
1907 * mpage_add_bh_to_extent - try to add one more block to extent of blocks
1909 * @mpd->lbh - extent of blocks
1910 * @logical - logical number of the block in the file
1911 * @bh - bh of the block (used to access block's state)
1913 * the function is used to collect contig. blocks in same state
1915 static void mpage_add_bh_to_extent(struct mpage_da_data
*mpd
,
1916 sector_t logical
, struct buffer_head
*bh
)
1919 size_t b_size
= bh
->b_size
;
1920 struct buffer_head
*lbh
= &mpd
->lbh
;
1921 int nrblocks
= lbh
->b_size
>> mpd
->inode
->i_blkbits
;
1923 /* check if thereserved journal credits might overflow */
1924 if (!(EXT4_I(mpd
->inode
)->i_flags
& EXT4_EXTENTS_FL
)) {
1925 if (nrblocks
>= EXT4_MAX_TRANS_DATA
) {
1927 * With non-extent format we are limited by the journal
1928 * credit available. Total credit needed to insert
1929 * nrblocks contiguous blocks is dependent on the
1930 * nrblocks. So limit nrblocks.
1933 } else if ((nrblocks
+ (b_size
>> mpd
->inode
->i_blkbits
)) >
1934 EXT4_MAX_TRANS_DATA
) {
1936 * Adding the new buffer_head would make it cross the
1937 * allowed limit for which we have journal credit
1938 * reserved. So limit the new bh->b_size
1940 b_size
= (EXT4_MAX_TRANS_DATA
- nrblocks
) <<
1941 mpd
->inode
->i_blkbits
;
1942 /* we will do mpage_da_submit_io in the next loop */
1946 * First block in the extent
1948 if (lbh
->b_size
== 0) {
1949 lbh
->b_blocknr
= logical
;
1950 lbh
->b_size
= b_size
;
1951 lbh
->b_state
= bh
->b_state
& BH_FLAGS
;
1955 next
= lbh
->b_blocknr
+ nrblocks
;
1957 * Can we merge the block to our big extent?
1959 if (logical
== next
&& (bh
->b_state
& BH_FLAGS
) == lbh
->b_state
) {
1960 lbh
->b_size
+= b_size
;
1966 * We couldn't merge the block to our extent, so we
1967 * need to flush current extent and start new one
1969 if (mpage_da_map_blocks(mpd
) == 0)
1970 mpage_da_submit_io(mpd
);
1976 * __mpage_da_writepage - finds extent of pages and blocks
1978 * @page: page to consider
1979 * @wbc: not used, we just follow rules
1982 * The function finds extents of pages and scan them for all blocks.
1984 static int __mpage_da_writepage(struct page
*page
,
1985 struct writeback_control
*wbc
, void *data
)
1987 struct mpage_da_data
*mpd
= data
;
1988 struct inode
*inode
= mpd
->inode
;
1989 struct buffer_head
*bh
, *head
, fake
;
1994 * Rest of the page in the page_vec
1995 * redirty then and skip then. We will
1996 * try to to write them again after
1997 * starting a new transaction
1999 redirty_page_for_writepage(wbc
, page
);
2001 return MPAGE_DA_EXTENT_TAIL
;
2004 * Can we merge this page to current extent?
2006 if (mpd
->next_page
!= page
->index
) {
2008 * Nope, we can't. So, we map non-allocated blocks
2009 * and start IO on them using writepage()
2011 if (mpd
->next_page
!= mpd
->first_page
) {
2012 if (mpage_da_map_blocks(mpd
) == 0)
2013 mpage_da_submit_io(mpd
);
2015 * skip rest of the page in the page_vec
2018 redirty_page_for_writepage(wbc
, page
);
2020 return MPAGE_DA_EXTENT_TAIL
;
2024 * Start next extent of pages ...
2026 mpd
->first_page
= page
->index
;
2031 mpd
->lbh
.b_size
= 0;
2032 mpd
->lbh
.b_state
= 0;
2033 mpd
->lbh
.b_blocknr
= 0;
2036 mpd
->next_page
= page
->index
+ 1;
2037 logical
= (sector_t
) page
->index
<<
2038 (PAGE_CACHE_SHIFT
- inode
->i_blkbits
);
2040 if (!page_has_buffers(page
)) {
2042 * There is no attached buffer heads yet (mmap?)
2043 * we treat the page asfull of dirty blocks
2046 bh
->b_size
= PAGE_CACHE_SIZE
;
2048 set_buffer_dirty(bh
);
2049 set_buffer_uptodate(bh
);
2050 mpage_add_bh_to_extent(mpd
, logical
, bh
);
2052 return MPAGE_DA_EXTENT_TAIL
;
2055 * Page with regular buffer heads, just add all dirty ones
2057 head
= page_buffers(page
);
2060 BUG_ON(buffer_locked(bh
));
2061 if (buffer_dirty(bh
) &&
2062 (!buffer_mapped(bh
) || buffer_delay(bh
))) {
2063 mpage_add_bh_to_extent(mpd
, logical
, bh
);
2065 return MPAGE_DA_EXTENT_TAIL
;
2068 } while ((bh
= bh
->b_this_page
) != head
);
2075 * mpage_da_writepages - walk the list of dirty pages of the given
2076 * address space, allocates non-allocated blocks, maps newly-allocated
2077 * blocks to existing bhs and issue IO them
2079 * @mapping: address space structure to write
2080 * @wbc: subtract the number of written pages from *@wbc->nr_to_write
2081 * @get_block: the filesystem's block mapper function.
2083 * This is a library function, which implements the writepages()
2084 * address_space_operation.
2086 static int mpage_da_writepages(struct address_space
*mapping
,
2087 struct writeback_control
*wbc
,
2088 get_block_t get_block
)
2090 struct mpage_da_data mpd
;
2095 return generic_writepages(mapping
, wbc
);
2098 mpd
.inode
= mapping
->host
;
2100 mpd
.lbh
.b_state
= 0;
2101 mpd
.lbh
.b_blocknr
= 0;
2104 mpd
.get_block
= get_block
;
2106 mpd
.pages_written
= 0;
2108 to_write
= wbc
->nr_to_write
;
2110 ret
= write_cache_pages(mapping
, wbc
, __mpage_da_writepage
, &mpd
);
2113 * Handle last extent of pages
2115 if (!mpd
.io_done
&& mpd
.next_page
!= mpd
.first_page
) {
2116 if (mpage_da_map_blocks(&mpd
) == 0)
2117 mpage_da_submit_io(&mpd
);
2120 wbc
->nr_to_write
= to_write
- mpd
.pages_written
;
2125 * this is a special callback for ->write_begin() only
2126 * it's intention is to return mapped block or reserve space
2128 static int ext4_da_get_block_prep(struct inode
*inode
, sector_t iblock
,
2129 struct buffer_head
*bh_result
, int create
)
2133 BUG_ON(create
== 0);
2134 BUG_ON(bh_result
->b_size
!= inode
->i_sb
->s_blocksize
);
2137 * first, we need to know whether the block is allocated already
2138 * preallocated blocks are unmapped but should treated
2139 * the same as allocated blocks.
2141 ret
= ext4_get_blocks_wrap(NULL
, inode
, iblock
, 1, bh_result
, 0, 0, 0);
2142 if ((ret
== 0) && !buffer_delay(bh_result
)) {
2143 /* the block isn't (pre)allocated yet, let's reserve space */
2145 * XXX: __block_prepare_write() unmaps passed block,
2148 ret
= ext4_da_reserve_space(inode
, 1);
2150 /* not enough space to reserve */
2153 map_bh(bh_result
, inode
->i_sb
, 0);
2154 set_buffer_new(bh_result
);
2155 set_buffer_delay(bh_result
);
2156 } else if (ret
> 0) {
2157 bh_result
->b_size
= (ret
<< inode
->i_blkbits
);
2163 #define EXT4_DELALLOC_RSVED 1
2164 static int ext4_da_get_block_write(struct inode
*inode
, sector_t iblock
,
2165 struct buffer_head
*bh_result
, int create
)
2168 unsigned max_blocks
= bh_result
->b_size
>> inode
->i_blkbits
;
2169 loff_t disksize
= EXT4_I(inode
)->i_disksize
;
2170 handle_t
*handle
= NULL
;
2172 handle
= ext4_journal_current_handle();
2174 ret
= ext4_get_blocks_wrap(handle
, inode
, iblock
, max_blocks
,
2175 bh_result
, 0, 0, 0);
2178 ret
= ext4_get_blocks_wrap(handle
, inode
, iblock
, max_blocks
,
2179 bh_result
, create
, 0, EXT4_DELALLOC_RSVED
);
2183 bh_result
->b_size
= (ret
<< inode
->i_blkbits
);
2186 * Update on-disk size along with block allocation
2187 * we don't use 'extend_disksize' as size may change
2188 * within already allocated block -bzzz
2190 disksize
= ((loff_t
) iblock
+ ret
) << inode
->i_blkbits
;
2191 if (disksize
> i_size_read(inode
))
2192 disksize
= i_size_read(inode
);
2193 if (disksize
> EXT4_I(inode
)->i_disksize
) {
2195 * XXX: replace with spinlock if seen contended -bzzz
2197 down_write(&EXT4_I(inode
)->i_data_sem
);
2198 if (disksize
> EXT4_I(inode
)->i_disksize
)
2199 EXT4_I(inode
)->i_disksize
= disksize
;
2200 up_write(&EXT4_I(inode
)->i_data_sem
);
2202 if (EXT4_I(inode
)->i_disksize
== disksize
) {
2203 ret
= ext4_mark_inode_dirty(handle
, inode
);
2212 static int ext4_bh_unmapped_or_delay(handle_t
*handle
, struct buffer_head
*bh
)
2215 * unmapped buffer is possible for holes.
2216 * delay buffer is possible with delayed allocation
2218 return ((!buffer_mapped(bh
) || buffer_delay(bh
)) && buffer_dirty(bh
));
2221 static int ext4_normal_get_block_write(struct inode
*inode
, sector_t iblock
,
2222 struct buffer_head
*bh_result
, int create
)
2225 unsigned max_blocks
= bh_result
->b_size
>> inode
->i_blkbits
;
2228 * we don't want to do block allocation in writepage
2229 * so call get_block_wrap with create = 0
2231 ret
= ext4_get_blocks_wrap(NULL
, inode
, iblock
, max_blocks
,
2232 bh_result
, 0, 0, 0);
2234 bh_result
->b_size
= (ret
<< inode
->i_blkbits
);
2241 * get called vi ext4_da_writepages after taking page lock (have journal handle)
2242 * get called via journal_submit_inode_data_buffers (no journal handle)
2243 * get called via shrink_page_list via pdflush (no journal handle)
2244 * or grab_page_cache when doing write_begin (have journal handle)
2246 static int ext4_da_writepage(struct page
*page
,
2247 struct writeback_control
*wbc
)
2252 struct buffer_head
*page_bufs
;
2253 struct inode
*inode
= page
->mapping
->host
;
2255 size
= i_size_read(inode
);
2256 if (page
->index
== size
>> PAGE_CACHE_SHIFT
)
2257 len
= size
& ~PAGE_CACHE_MASK
;
2259 len
= PAGE_CACHE_SIZE
;
2261 if (page_has_buffers(page
)) {
2262 page_bufs
= page_buffers(page
);
2263 if (walk_page_buffers(NULL
, page_bufs
, 0, len
, NULL
,
2264 ext4_bh_unmapped_or_delay
)) {
2266 * We don't want to do block allocation
2267 * So redirty the page and return
2268 * We may reach here when we do a journal commit
2269 * via journal_submit_inode_data_buffers.
2270 * If we don't have mapping block we just ignore
2271 * them. We can also reach here via shrink_page_list
2273 redirty_page_for_writepage(wbc
, page
);
2279 * The test for page_has_buffers() is subtle:
2280 * We know the page is dirty but it lost buffers. That means
2281 * that at some moment in time after write_begin()/write_end()
2282 * has been called all buffers have been clean and thus they
2283 * must have been written at least once. So they are all
2284 * mapped and we can happily proceed with mapping them
2285 * and writing the page.
2287 * Try to initialize the buffer_heads and check whether
2288 * all are mapped and non delay. We don't want to
2289 * do block allocation here.
2291 ret
= block_prepare_write(page
, 0, PAGE_CACHE_SIZE
,
2292 ext4_normal_get_block_write
);
2294 page_bufs
= page_buffers(page
);
2295 /* check whether all are mapped and non delay */
2296 if (walk_page_buffers(NULL
, page_bufs
, 0, len
, NULL
,
2297 ext4_bh_unmapped_or_delay
)) {
2298 redirty_page_for_writepage(wbc
, page
);
2304 * We can't do block allocation here
2305 * so just redity the page and unlock
2308 redirty_page_for_writepage(wbc
, page
);
2314 if (test_opt(inode
->i_sb
, NOBH
) && ext4_should_writeback_data(inode
))
2315 ret
= nobh_writepage(page
, ext4_normal_get_block_write
, wbc
);
2317 ret
= block_write_full_page(page
,
2318 ext4_normal_get_block_write
,
2325 * This is called via ext4_da_writepages() to
2326 * calulate the total number of credits to reserve to fit
2327 * a single extent allocation into a single transaction,
2328 * ext4_da_writpeages() will loop calling this before
2329 * the block allocation.
2332 static int ext4_da_writepages_trans_blocks(struct inode
*inode
)
2334 int max_blocks
= EXT4_I(inode
)->i_reserved_data_blocks
;
2337 * With non-extent format the journal credit needed to
2338 * insert nrblocks contiguous block is dependent on
2339 * number of contiguous block. So we will limit
2340 * number of contiguous block to a sane value
2342 if (!(inode
->i_flags
& EXT4_EXTENTS_FL
) &&
2343 (max_blocks
> EXT4_MAX_TRANS_DATA
))
2344 max_blocks
= EXT4_MAX_TRANS_DATA
;
2346 return ext4_chunk_trans_blocks(inode
, max_blocks
);
2349 static int ext4_da_writepages(struct address_space
*mapping
,
2350 struct writeback_control
*wbc
)
2352 handle_t
*handle
= NULL
;
2353 loff_t range_start
= 0;
2354 struct inode
*inode
= mapping
->host
;
2355 int needed_blocks
, ret
= 0, nr_to_writebump
= 0;
2356 long to_write
, pages_skipped
= 0;
2357 struct ext4_sb_info
*sbi
= EXT4_SB(mapping
->host
->i_sb
);
2360 * No pages to write? This is mainly a kludge to avoid starting
2361 * a transaction for special inodes like journal inode on last iput()
2362 * because that could violate lock ordering on umount
2364 if (!mapping
->nrpages
|| !mapping_tagged(mapping
, PAGECACHE_TAG_DIRTY
))
2367 * Make sure nr_to_write is >= sbi->s_mb_stream_request
2368 * This make sure small files blocks are allocated in
2369 * single attempt. This ensure that small files
2370 * get less fragmented.
2372 if (wbc
->nr_to_write
< sbi
->s_mb_stream_request
) {
2373 nr_to_writebump
= sbi
->s_mb_stream_request
- wbc
->nr_to_write
;
2374 wbc
->nr_to_write
= sbi
->s_mb_stream_request
;
2377 if (!wbc
->range_cyclic
)
2379 * If range_cyclic is not set force range_cont
2380 * and save the old writeback_index
2382 wbc
->range_cont
= 1;
2384 range_start
= wbc
->range_start
;
2385 pages_skipped
= wbc
->pages_skipped
;
2388 to_write
= wbc
->nr_to_write
;
2389 while (!ret
&& to_write
> 0) {
2392 * we insert one extent at a time. So we need
2393 * credit needed for single extent allocation.
2394 * journalled mode is currently not supported
2397 BUG_ON(ext4_should_journal_data(inode
));
2398 needed_blocks
= ext4_da_writepages_trans_blocks(inode
);
2400 /* start a new transaction*/
2401 handle
= ext4_journal_start(inode
, needed_blocks
);
2402 if (IS_ERR(handle
)) {
2403 ret
= PTR_ERR(handle
);
2404 printk(KERN_EMERG
"%s: jbd2_start: "
2405 "%ld pages, ino %lu; err %d\n", __func__
,
2406 wbc
->nr_to_write
, inode
->i_ino
, ret
);
2408 goto out_writepages
;
2410 if (ext4_should_order_data(inode
)) {
2412 * With ordered mode we need to add
2413 * the inode to the journal handl
2414 * when we do block allocation.
2416 ret
= ext4_jbd2_file_inode(handle
, inode
);
2418 ext4_journal_stop(handle
);
2419 goto out_writepages
;
2423 to_write
-= wbc
->nr_to_write
;
2424 ret
= mpage_da_writepages(mapping
, wbc
,
2425 ext4_da_get_block_write
);
2426 ext4_journal_stop(handle
);
2427 if (ret
== MPAGE_DA_EXTENT_TAIL
) {
2429 * got one extent now try with
2432 to_write
+= wbc
->nr_to_write
;
2434 } else if (wbc
->nr_to_write
) {
2436 * There is no more writeout needed
2437 * or we requested for a noblocking writeout
2438 * and we found the device congested
2440 to_write
+= wbc
->nr_to_write
;
2443 wbc
->nr_to_write
= to_write
;
2446 if (wbc
->range_cont
&& (pages_skipped
!= wbc
->pages_skipped
)) {
2447 /* We skipped pages in this loop */
2448 wbc
->range_start
= range_start
;
2449 wbc
->nr_to_write
= to_write
+
2450 wbc
->pages_skipped
- pages_skipped
;
2451 wbc
->pages_skipped
= pages_skipped
;
2456 wbc
->nr_to_write
= to_write
- nr_to_writebump
;
2457 wbc
->range_start
= range_start
;
2461 #define FALL_BACK_TO_NONDELALLOC 1
2462 static int ext4_nonda_switch(struct super_block
*sb
)
2464 s64 free_blocks
, dirty_blocks
;
2465 struct ext4_sb_info
*sbi
= EXT4_SB(sb
);
2468 * switch to non delalloc mode if we are running low
2469 * on free block. The free block accounting via percpu
2470 * counters can get slightly wrong with FBC_BATCH getting
2471 * accumulated on each CPU without updating global counters
2472 * Delalloc need an accurate free block accounting. So switch
2473 * to non delalloc when we are near to error range.
2475 free_blocks
= percpu_counter_read_positive(&sbi
->s_freeblocks_counter
);
2476 dirty_blocks
= percpu_counter_read_positive(&sbi
->s_dirtyblocks_counter
);
2477 if (2 * free_blocks
< 3 * dirty_blocks
||
2478 free_blocks
< (dirty_blocks
+ EXT4_FREEBLOCKS_WATERMARK
)) {
2480 * free block count is less that 150% of dirty blocks
2481 * or free blocks is less that watermark
2488 static int ext4_da_write_begin(struct file
*file
, struct address_space
*mapping
,
2489 loff_t pos
, unsigned len
, unsigned flags
,
2490 struct page
**pagep
, void **fsdata
)
2492 int ret
, retries
= 0;
2496 struct inode
*inode
= mapping
->host
;
2499 index
= pos
>> PAGE_CACHE_SHIFT
;
2500 from
= pos
& (PAGE_CACHE_SIZE
- 1);
2503 if (ext4_nonda_switch(inode
->i_sb
)) {
2504 *fsdata
= (void *)FALL_BACK_TO_NONDELALLOC
;
2505 return ext4_write_begin(file
, mapping
, pos
,
2506 len
, flags
, pagep
, fsdata
);
2508 *fsdata
= (void *)0;
2511 * With delayed allocation, we don't log the i_disksize update
2512 * if there is delayed block allocation. But we still need
2513 * to journalling the i_disksize update if writes to the end
2514 * of file which has an already mapped buffer.
2516 handle
= ext4_journal_start(inode
, 1);
2517 if (IS_ERR(handle
)) {
2518 ret
= PTR_ERR(handle
);
2522 page
= __grab_cache_page(mapping
, index
);
2524 ext4_journal_stop(handle
);
2530 ret
= block_write_begin(file
, mapping
, pos
, len
, flags
, pagep
, fsdata
,
2531 ext4_da_get_block_prep
);
2534 ext4_journal_stop(handle
);
2535 page_cache_release(page
);
2538 if (ret
== -ENOSPC
&& ext4_should_retry_alloc(inode
->i_sb
, &retries
))
2545 * Check if we should update i_disksize
2546 * when write to the end of file but not require block allocation
2548 static int ext4_da_should_update_i_disksize(struct page
*page
,
2549 unsigned long offset
)
2551 struct buffer_head
*bh
;
2552 struct inode
*inode
= page
->mapping
->host
;
2556 bh
= page_buffers(page
);
2557 idx
= offset
>> inode
->i_blkbits
;
2559 for (i
= 0; i
< idx
; i
++)
2560 bh
= bh
->b_this_page
;
2562 if (!buffer_mapped(bh
) || (buffer_delay(bh
)))
2567 static int ext4_da_write_end(struct file
*file
,
2568 struct address_space
*mapping
,
2569 loff_t pos
, unsigned len
, unsigned copied
,
2570 struct page
*page
, void *fsdata
)
2572 struct inode
*inode
= mapping
->host
;
2574 handle_t
*handle
= ext4_journal_current_handle();
2576 unsigned long start
, end
;
2577 int write_mode
= (int)(unsigned long)fsdata
;
2579 if (write_mode
== FALL_BACK_TO_NONDELALLOC
) {
2580 if (ext4_should_order_data(inode
)) {
2581 return ext4_ordered_write_end(file
, mapping
, pos
,
2582 len
, copied
, page
, fsdata
);
2583 } else if (ext4_should_writeback_data(inode
)) {
2584 return ext4_writeback_write_end(file
, mapping
, pos
,
2585 len
, copied
, page
, fsdata
);
2591 start
= pos
& (PAGE_CACHE_SIZE
- 1);
2592 end
= start
+ copied
- 1;
2595 * generic_write_end() will run mark_inode_dirty() if i_size
2596 * changes. So let's piggyback the i_disksize mark_inode_dirty
2600 new_i_size
= pos
+ copied
;
2601 if (new_i_size
> EXT4_I(inode
)->i_disksize
) {
2602 if (ext4_da_should_update_i_disksize(page
, end
)) {
2603 down_write(&EXT4_I(inode
)->i_data_sem
);
2604 if (new_i_size
> EXT4_I(inode
)->i_disksize
) {
2606 * Updating i_disksize when extending file
2607 * without needing block allocation
2609 if (ext4_should_order_data(inode
))
2610 ret
= ext4_jbd2_file_inode(handle
,
2613 EXT4_I(inode
)->i_disksize
= new_i_size
;
2615 up_write(&EXT4_I(inode
)->i_data_sem
);
2618 ret2
= generic_write_end(file
, mapping
, pos
, len
, copied
,
2623 ret2
= ext4_journal_stop(handle
);
2627 return ret
? ret
: copied
;
2630 static void ext4_da_invalidatepage(struct page
*page
, unsigned long offset
)
2633 * Drop reserved blocks
2635 BUG_ON(!PageLocked(page
));
2636 if (!page_has_buffers(page
))
2639 ext4_da_page_release_reservation(page
, offset
);
2642 ext4_invalidatepage(page
, offset
);
2649 * bmap() is special. It gets used by applications such as lilo and by
2650 * the swapper to find the on-disk block of a specific piece of data.
2652 * Naturally, this is dangerous if the block concerned is still in the
2653 * journal. If somebody makes a swapfile on an ext4 data-journaling
2654 * filesystem and enables swap, then they may get a nasty shock when the
2655 * data getting swapped to that swapfile suddenly gets overwritten by
2656 * the original zero's written out previously to the journal and
2657 * awaiting writeback in the kernel's buffer cache.
2659 * So, if we see any bmap calls here on a modified, data-journaled file,
2660 * take extra steps to flush any blocks which might be in the cache.
2662 static sector_t
ext4_bmap(struct address_space
*mapping
, sector_t block
)
2664 struct inode
*inode
= mapping
->host
;
2668 if (mapping_tagged(mapping
, PAGECACHE_TAG_DIRTY
) &&
2669 test_opt(inode
->i_sb
, DELALLOC
)) {
2671 * With delalloc we want to sync the file
2672 * so that we can make sure we allocate
2675 filemap_write_and_wait(mapping
);
2678 if (EXT4_I(inode
)->i_state
& EXT4_STATE_JDATA
) {
2680 * This is a REALLY heavyweight approach, but the use of
2681 * bmap on dirty files is expected to be extremely rare:
2682 * only if we run lilo or swapon on a freshly made file
2683 * do we expect this to happen.
2685 * (bmap requires CAP_SYS_RAWIO so this does not
2686 * represent an unprivileged user DOS attack --- we'd be
2687 * in trouble if mortal users could trigger this path at
2690 * NB. EXT4_STATE_JDATA is not set on files other than
2691 * regular files. If somebody wants to bmap a directory
2692 * or symlink and gets confused because the buffer
2693 * hasn't yet been flushed to disk, they deserve
2694 * everything they get.
2697 EXT4_I(inode
)->i_state
&= ~EXT4_STATE_JDATA
;
2698 journal
= EXT4_JOURNAL(inode
);
2699 jbd2_journal_lock_updates(journal
);
2700 err
= jbd2_journal_flush(journal
);
2701 jbd2_journal_unlock_updates(journal
);
2707 return generic_block_bmap(mapping
, block
, ext4_get_block
);
2710 static int bget_one(handle_t
*handle
, struct buffer_head
*bh
)
2716 static int bput_one(handle_t
*handle
, struct buffer_head
*bh
)
2723 * Note that we don't need to start a transaction unless we're journaling data
2724 * because we should have holes filled from ext4_page_mkwrite(). We even don't
2725 * need to file the inode to the transaction's list in ordered mode because if
2726 * we are writing back data added by write(), the inode is already there and if
2727 * we are writing back data modified via mmap(), noone guarantees in which
2728 * transaction the data will hit the disk. In case we are journaling data, we
2729 * cannot start transaction directly because transaction start ranks above page
2730 * lock so we have to do some magic.
2732 * In all journaling modes block_write_full_page() will start the I/O.
2736 * ext4_writepage() -> kmalloc() -> __alloc_pages() -> page_launder() ->
2741 * ext4_file_write() -> generic_file_write() -> __alloc_pages() -> ...
2743 * Same applies to ext4_get_block(). We will deadlock on various things like
2744 * lock_journal and i_data_sem
2746 * Setting PF_MEMALLOC here doesn't work - too many internal memory
2749 * 16May01: If we're reentered then journal_current_handle() will be
2750 * non-zero. We simply *return*.
2752 * 1 July 2001: @@@ FIXME:
2753 * In journalled data mode, a data buffer may be metadata against the
2754 * current transaction. But the same file is part of a shared mapping
2755 * and someone does a writepage() on it.
2757 * We will move the buffer onto the async_data list, but *after* it has
2758 * been dirtied. So there's a small window where we have dirty data on
2761 * Note that this only applies to the last partial page in the file. The
2762 * bit which block_write_full_page() uses prepare/commit for. (That's
2763 * broken code anyway: it's wrong for msync()).
2765 * It's a rare case: affects the final partial page, for journalled data
2766 * where the file is subject to bith write() and writepage() in the same
2767 * transction. To fix it we'll need a custom block_write_full_page().
2768 * We'll probably need that anyway for journalling writepage() output.
2770 * We don't honour synchronous mounts for writepage(). That would be
2771 * disastrous. Any write() or metadata operation will sync the fs for
2775 static int __ext4_normal_writepage(struct page
*page
,
2776 struct writeback_control
*wbc
)
2778 struct inode
*inode
= page
->mapping
->host
;
2780 if (test_opt(inode
->i_sb
, NOBH
))
2781 return nobh_writepage(page
,
2782 ext4_normal_get_block_write
, wbc
);
2784 return block_write_full_page(page
,
2785 ext4_normal_get_block_write
,
2789 static int ext4_normal_writepage(struct page
*page
,
2790 struct writeback_control
*wbc
)
2792 struct inode
*inode
= page
->mapping
->host
;
2793 loff_t size
= i_size_read(inode
);
2796 J_ASSERT(PageLocked(page
));
2797 if (page
->index
== size
>> PAGE_CACHE_SHIFT
)
2798 len
= size
& ~PAGE_CACHE_MASK
;
2800 len
= PAGE_CACHE_SIZE
;
2802 if (page_has_buffers(page
)) {
2803 /* if page has buffers it should all be mapped
2804 * and allocated. If there are not buffers attached
2805 * to the page we know the page is dirty but it lost
2806 * buffers. That means that at some moment in time
2807 * after write_begin() / write_end() has been called
2808 * all buffers have been clean and thus they must have been
2809 * written at least once. So they are all mapped and we can
2810 * happily proceed with mapping them and writing the page.
2812 BUG_ON(walk_page_buffers(NULL
, page_buffers(page
), 0, len
, NULL
,
2813 ext4_bh_unmapped_or_delay
));
2816 if (!ext4_journal_current_handle())
2817 return __ext4_normal_writepage(page
, wbc
);
2819 redirty_page_for_writepage(wbc
, page
);
2824 static int __ext4_journalled_writepage(struct page
*page
,
2825 struct writeback_control
*wbc
)
2827 struct address_space
*mapping
= page
->mapping
;
2828 struct inode
*inode
= mapping
->host
;
2829 struct buffer_head
*page_bufs
;
2830 handle_t
*handle
= NULL
;
2834 ret
= block_prepare_write(page
, 0, PAGE_CACHE_SIZE
,
2835 ext4_normal_get_block_write
);
2839 page_bufs
= page_buffers(page
);
2840 walk_page_buffers(handle
, page_bufs
, 0, PAGE_CACHE_SIZE
, NULL
,
2842 /* As soon as we unlock the page, it can go away, but we have
2843 * references to buffers so we are safe */
2846 handle
= ext4_journal_start(inode
, ext4_writepage_trans_blocks(inode
));
2847 if (IS_ERR(handle
)) {
2848 ret
= PTR_ERR(handle
);
2852 ret
= walk_page_buffers(handle
, page_bufs
, 0,
2853 PAGE_CACHE_SIZE
, NULL
, do_journal_get_write_access
);
2855 err
= walk_page_buffers(handle
, page_bufs
, 0,
2856 PAGE_CACHE_SIZE
, NULL
, write_end_fn
);
2859 err
= ext4_journal_stop(handle
);
2863 walk_page_buffers(handle
, page_bufs
, 0,
2864 PAGE_CACHE_SIZE
, NULL
, bput_one
);
2865 EXT4_I(inode
)->i_state
|= EXT4_STATE_JDATA
;
2874 static int ext4_journalled_writepage(struct page
*page
,
2875 struct writeback_control
*wbc
)
2877 struct inode
*inode
= page
->mapping
->host
;
2878 loff_t size
= i_size_read(inode
);
2881 J_ASSERT(PageLocked(page
));
2882 if (page
->index
== size
>> PAGE_CACHE_SHIFT
)
2883 len
= size
& ~PAGE_CACHE_MASK
;
2885 len
= PAGE_CACHE_SIZE
;
2887 if (page_has_buffers(page
)) {
2888 /* if page has buffers it should all be mapped
2889 * and allocated. If there are not buffers attached
2890 * to the page we know the page is dirty but it lost
2891 * buffers. That means that at some moment in time
2892 * after write_begin() / write_end() has been called
2893 * all buffers have been clean and thus they must have been
2894 * written at least once. So they are all mapped and we can
2895 * happily proceed with mapping them and writing the page.
2897 BUG_ON(walk_page_buffers(NULL
, page_buffers(page
), 0, len
, NULL
,
2898 ext4_bh_unmapped_or_delay
));
2901 if (ext4_journal_current_handle())
2904 if (PageChecked(page
)) {
2906 * It's mmapped pagecache. Add buffers and journal it. There
2907 * doesn't seem much point in redirtying the page here.
2909 ClearPageChecked(page
);
2910 return __ext4_journalled_writepage(page
, wbc
);
2913 * It may be a page full of checkpoint-mode buffers. We don't
2914 * really know unless we go poke around in the buffer_heads.
2915 * But block_write_full_page will do the right thing.
2917 return block_write_full_page(page
,
2918 ext4_normal_get_block_write
,
2922 redirty_page_for_writepage(wbc
, page
);
2927 static int ext4_readpage(struct file
*file
, struct page
*page
)
2929 return mpage_readpage(page
, ext4_get_block
);
2933 ext4_readpages(struct file
*file
, struct address_space
*mapping
,
2934 struct list_head
*pages
, unsigned nr_pages
)
2936 return mpage_readpages(mapping
, pages
, nr_pages
, ext4_get_block
);
2939 static void ext4_invalidatepage(struct page
*page
, unsigned long offset
)
2941 journal_t
*journal
= EXT4_JOURNAL(page
->mapping
->host
);
2944 * If it's a full truncate we just forget about the pending dirtying
2947 ClearPageChecked(page
);
2949 jbd2_journal_invalidatepage(journal
, page
, offset
);
2952 static int ext4_releasepage(struct page
*page
, gfp_t wait
)
2954 journal_t
*journal
= EXT4_JOURNAL(page
->mapping
->host
);
2956 WARN_ON(PageChecked(page
));
2957 if (!page_has_buffers(page
))
2959 return jbd2_journal_try_to_free_buffers(journal
, page
, wait
);
2963 * If the O_DIRECT write will extend the file then add this inode to the
2964 * orphan list. So recovery will truncate it back to the original size
2965 * if the machine crashes during the write.
2967 * If the O_DIRECT write is intantiating holes inside i_size and the machine
2968 * crashes then stale disk data _may_ be exposed inside the file. But current
2969 * VFS code falls back into buffered path in that case so we are safe.
2971 static ssize_t
ext4_direct_IO(int rw
, struct kiocb
*iocb
,
2972 const struct iovec
*iov
, loff_t offset
,
2973 unsigned long nr_segs
)
2975 struct file
*file
= iocb
->ki_filp
;
2976 struct inode
*inode
= file
->f_mapping
->host
;
2977 struct ext4_inode_info
*ei
= EXT4_I(inode
);
2981 size_t count
= iov_length(iov
, nr_segs
);
2984 loff_t final_size
= offset
+ count
;
2986 if (final_size
> inode
->i_size
) {
2987 /* Credits for sb + inode write */
2988 handle
= ext4_journal_start(inode
, 2);
2989 if (IS_ERR(handle
)) {
2990 ret
= PTR_ERR(handle
);
2993 ret
= ext4_orphan_add(handle
, inode
);
2995 ext4_journal_stop(handle
);
2999 ei
->i_disksize
= inode
->i_size
;
3000 ext4_journal_stop(handle
);
3004 ret
= blockdev_direct_IO(rw
, iocb
, inode
, inode
->i_sb
->s_bdev
, iov
,
3006 ext4_get_block
, NULL
);
3011 /* Credits for sb + inode write */
3012 handle
= ext4_journal_start(inode
, 2);
3013 if (IS_ERR(handle
)) {
3014 /* This is really bad luck. We've written the data
3015 * but cannot extend i_size. Bail out and pretend
3016 * the write failed... */
3017 ret
= PTR_ERR(handle
);
3021 ext4_orphan_del(handle
, inode
);
3023 loff_t end
= offset
+ ret
;
3024 if (end
> inode
->i_size
) {
3025 ei
->i_disksize
= end
;
3026 i_size_write(inode
, end
);
3028 * We're going to return a positive `ret'
3029 * here due to non-zero-length I/O, so there's
3030 * no way of reporting error returns from
3031 * ext4_mark_inode_dirty() to userspace. So
3034 ext4_mark_inode_dirty(handle
, inode
);
3037 err
= ext4_journal_stop(handle
);
3046 * Pages can be marked dirty completely asynchronously from ext4's journalling
3047 * activity. By filemap_sync_pte(), try_to_unmap_one(), etc. We cannot do
3048 * much here because ->set_page_dirty is called under VFS locks. The page is
3049 * not necessarily locked.
3051 * We cannot just dirty the page and leave attached buffers clean, because the
3052 * buffers' dirty state is "definitive". We cannot just set the buffers dirty
3053 * or jbddirty because all the journalling code will explode.
3055 * So what we do is to mark the page "pending dirty" and next time writepage
3056 * is called, propagate that into the buffers appropriately.
3058 static int ext4_journalled_set_page_dirty(struct page
*page
)
3060 SetPageChecked(page
);
3061 return __set_page_dirty_nobuffers(page
);
3064 static const struct address_space_operations ext4_ordered_aops
= {
3065 .readpage
= ext4_readpage
,
3066 .readpages
= ext4_readpages
,
3067 .writepage
= ext4_normal_writepage
,
3068 .sync_page
= block_sync_page
,
3069 .write_begin
= ext4_write_begin
,
3070 .write_end
= ext4_ordered_write_end
,
3072 .invalidatepage
= ext4_invalidatepage
,
3073 .releasepage
= ext4_releasepage
,
3074 .direct_IO
= ext4_direct_IO
,
3075 .migratepage
= buffer_migrate_page
,
3076 .is_partially_uptodate
= block_is_partially_uptodate
,
3079 static const struct address_space_operations ext4_writeback_aops
= {
3080 .readpage
= ext4_readpage
,
3081 .readpages
= ext4_readpages
,
3082 .writepage
= ext4_normal_writepage
,
3083 .sync_page
= block_sync_page
,
3084 .write_begin
= ext4_write_begin
,
3085 .write_end
= ext4_writeback_write_end
,
3087 .invalidatepage
= ext4_invalidatepage
,
3088 .releasepage
= ext4_releasepage
,
3089 .direct_IO
= ext4_direct_IO
,
3090 .migratepage
= buffer_migrate_page
,
3091 .is_partially_uptodate
= block_is_partially_uptodate
,
3094 static const struct address_space_operations ext4_journalled_aops
= {
3095 .readpage
= ext4_readpage
,
3096 .readpages
= ext4_readpages
,
3097 .writepage
= ext4_journalled_writepage
,
3098 .sync_page
= block_sync_page
,
3099 .write_begin
= ext4_write_begin
,
3100 .write_end
= ext4_journalled_write_end
,
3101 .set_page_dirty
= ext4_journalled_set_page_dirty
,
3103 .invalidatepage
= ext4_invalidatepage
,
3104 .releasepage
= ext4_releasepage
,
3105 .is_partially_uptodate
= block_is_partially_uptodate
,
3108 static const struct address_space_operations ext4_da_aops
= {
3109 .readpage
= ext4_readpage
,
3110 .readpages
= ext4_readpages
,
3111 .writepage
= ext4_da_writepage
,
3112 .writepages
= ext4_da_writepages
,
3113 .sync_page
= block_sync_page
,
3114 .write_begin
= ext4_da_write_begin
,
3115 .write_end
= ext4_da_write_end
,
3117 .invalidatepage
= ext4_da_invalidatepage
,
3118 .releasepage
= ext4_releasepage
,
3119 .direct_IO
= ext4_direct_IO
,
3120 .migratepage
= buffer_migrate_page
,
3121 .is_partially_uptodate
= block_is_partially_uptodate
,
3124 void ext4_set_aops(struct inode
*inode
)
3126 if (ext4_should_order_data(inode
) &&
3127 test_opt(inode
->i_sb
, DELALLOC
))
3128 inode
->i_mapping
->a_ops
= &ext4_da_aops
;
3129 else if (ext4_should_order_data(inode
))
3130 inode
->i_mapping
->a_ops
= &ext4_ordered_aops
;
3131 else if (ext4_should_writeback_data(inode
) &&
3132 test_opt(inode
->i_sb
, DELALLOC
))
3133 inode
->i_mapping
->a_ops
= &ext4_da_aops
;
3134 else if (ext4_should_writeback_data(inode
))
3135 inode
->i_mapping
->a_ops
= &ext4_writeback_aops
;
3137 inode
->i_mapping
->a_ops
= &ext4_journalled_aops
;
3141 * ext4_block_truncate_page() zeroes out a mapping from file offset `from'
3142 * up to the end of the block which corresponds to `from'.
3143 * This required during truncate. We need to physically zero the tail end
3144 * of that block so it doesn't yield old data if the file is later grown.
3146 int ext4_block_truncate_page(handle_t
*handle
,
3147 struct address_space
*mapping
, loff_t from
)
3149 ext4_fsblk_t index
= from
>> PAGE_CACHE_SHIFT
;
3150 unsigned offset
= from
& (PAGE_CACHE_SIZE
-1);
3151 unsigned blocksize
, length
, pos
;
3153 struct inode
*inode
= mapping
->host
;
3154 struct buffer_head
*bh
;
3158 page
= grab_cache_page(mapping
, from
>> PAGE_CACHE_SHIFT
);
3162 blocksize
= inode
->i_sb
->s_blocksize
;
3163 length
= blocksize
- (offset
& (blocksize
- 1));
3164 iblock
= index
<< (PAGE_CACHE_SHIFT
- inode
->i_sb
->s_blocksize_bits
);
3167 * For "nobh" option, we can only work if we don't need to
3168 * read-in the page - otherwise we create buffers to do the IO.
3170 if (!page_has_buffers(page
) && test_opt(inode
->i_sb
, NOBH
) &&
3171 ext4_should_writeback_data(inode
) && PageUptodate(page
)) {
3172 zero_user(page
, offset
, length
);
3173 set_page_dirty(page
);
3177 if (!page_has_buffers(page
))
3178 create_empty_buffers(page
, blocksize
, 0);
3180 /* Find the buffer that contains "offset" */
3181 bh
= page_buffers(page
);
3183 while (offset
>= pos
) {
3184 bh
= bh
->b_this_page
;
3190 if (buffer_freed(bh
)) {
3191 BUFFER_TRACE(bh
, "freed: skip");
3195 if (!buffer_mapped(bh
)) {
3196 BUFFER_TRACE(bh
, "unmapped");
3197 ext4_get_block(inode
, iblock
, bh
, 0);
3198 /* unmapped? It's a hole - nothing to do */
3199 if (!buffer_mapped(bh
)) {
3200 BUFFER_TRACE(bh
, "still unmapped");
3205 /* Ok, it's mapped. Make sure it's up-to-date */
3206 if (PageUptodate(page
))
3207 set_buffer_uptodate(bh
);
3209 if (!buffer_uptodate(bh
)) {
3211 ll_rw_block(READ
, 1, &bh
);
3213 /* Uhhuh. Read error. Complain and punt. */
3214 if (!buffer_uptodate(bh
))
3218 if (ext4_should_journal_data(inode
)) {
3219 BUFFER_TRACE(bh
, "get write access");
3220 err
= ext4_journal_get_write_access(handle
, bh
);
3225 zero_user(page
, offset
, length
);
3227 BUFFER_TRACE(bh
, "zeroed end of block");
3230 if (ext4_should_journal_data(inode
)) {
3231 err
= ext4_journal_dirty_metadata(handle
, bh
);
3233 if (ext4_should_order_data(inode
))
3234 err
= ext4_jbd2_file_inode(handle
, inode
);
3235 mark_buffer_dirty(bh
);
3240 page_cache_release(page
);
3245 * Probably it should be a library function... search for first non-zero word
3246 * or memcmp with zero_page, whatever is better for particular architecture.
3249 static inline int all_zeroes(__le32
*p
, __le32
*q
)
3258 * ext4_find_shared - find the indirect blocks for partial truncation.
3259 * @inode: inode in question
3260 * @depth: depth of the affected branch
3261 * @offsets: offsets of pointers in that branch (see ext4_block_to_path)
3262 * @chain: place to store the pointers to partial indirect blocks
3263 * @top: place to the (detached) top of branch
3265 * This is a helper function used by ext4_truncate().
3267 * When we do truncate() we may have to clean the ends of several
3268 * indirect blocks but leave the blocks themselves alive. Block is
3269 * partially truncated if some data below the new i_size is refered
3270 * from it (and it is on the path to the first completely truncated
3271 * data block, indeed). We have to free the top of that path along
3272 * with everything to the right of the path. Since no allocation
3273 * past the truncation point is possible until ext4_truncate()
3274 * finishes, we may safely do the latter, but top of branch may
3275 * require special attention - pageout below the truncation point
3276 * might try to populate it.
3278 * We atomically detach the top of branch from the tree, store the
3279 * block number of its root in *@top, pointers to buffer_heads of
3280 * partially truncated blocks - in @chain[].bh and pointers to
3281 * their last elements that should not be removed - in
3282 * @chain[].p. Return value is the pointer to last filled element
3285 * The work left to caller to do the actual freeing of subtrees:
3286 * a) free the subtree starting from *@top
3287 * b) free the subtrees whose roots are stored in
3288 * (@chain[i].p+1 .. end of @chain[i].bh->b_data)
3289 * c) free the subtrees growing from the inode past the @chain[0].
3290 * (no partially truncated stuff there). */
3292 static Indirect
*ext4_find_shared(struct inode
*inode
, int depth
,
3293 ext4_lblk_t offsets
[4], Indirect chain
[4], __le32
*top
)
3295 Indirect
*partial
, *p
;
3299 /* Make k index the deepest non-null offest + 1 */
3300 for (k
= depth
; k
> 1 && !offsets
[k
-1]; k
--)
3302 partial
= ext4_get_branch(inode
, k
, offsets
, chain
, &err
);
3303 /* Writer: pointers */
3305 partial
= chain
+ k
-1;
3307 * If the branch acquired continuation since we've looked at it -
3308 * fine, it should all survive and (new) top doesn't belong to us.
3310 if (!partial
->key
&& *partial
->p
)
3313 for (p
= partial
; (p
> chain
) && all_zeroes((__le32
*) p
->bh
->b_data
, p
->p
); p
--)
3316 * OK, we've found the last block that must survive. The rest of our
3317 * branch should be detached before unlocking. However, if that rest
3318 * of branch is all ours and does not grow immediately from the inode
3319 * it's easier to cheat and just decrement partial->p.
3321 if (p
== chain
+ k
- 1 && p
> chain
) {
3325 /* Nope, don't do this in ext4. Must leave the tree intact */
3332 while (partial
> p
) {
3333 brelse(partial
->bh
);
3341 * Zero a number of block pointers in either an inode or an indirect block.
3342 * If we restart the transaction we must again get write access to the
3343 * indirect block for further modification.
3345 * We release `count' blocks on disk, but (last - first) may be greater
3346 * than `count' because there can be holes in there.
3348 static void ext4_clear_blocks(handle_t
*handle
, struct inode
*inode
,
3349 struct buffer_head
*bh
, ext4_fsblk_t block_to_free
,
3350 unsigned long count
, __le32
*first
, __le32
*last
)
3353 if (try_to_extend_transaction(handle
, inode
)) {
3355 BUFFER_TRACE(bh
, "call ext4_journal_dirty_metadata");
3356 ext4_journal_dirty_metadata(handle
, bh
);
3358 ext4_mark_inode_dirty(handle
, inode
);
3359 ext4_journal_test_restart(handle
, inode
);
3361 BUFFER_TRACE(bh
, "retaking write access");
3362 ext4_journal_get_write_access(handle
, bh
);
3367 * Any buffers which are on the journal will be in memory. We find
3368 * them on the hash table so jbd2_journal_revoke() will run jbd2_journal_forget()
3369 * on them. We've already detached each block from the file, so
3370 * bforget() in jbd2_journal_forget() should be safe.
3372 * AKPM: turn on bforget in jbd2_journal_forget()!!!
3374 for (p
= first
; p
< last
; p
++) {
3375 u32 nr
= le32_to_cpu(*p
);
3377 struct buffer_head
*tbh
;
3380 tbh
= sb_find_get_block(inode
->i_sb
, nr
);
3381 ext4_forget(handle
, 0, inode
, tbh
, nr
);
3385 ext4_free_blocks(handle
, inode
, block_to_free
, count
, 0);
3389 * ext4_free_data - free a list of data blocks
3390 * @handle: handle for this transaction
3391 * @inode: inode we are dealing with
3392 * @this_bh: indirect buffer_head which contains *@first and *@last
3393 * @first: array of block numbers
3394 * @last: points immediately past the end of array
3396 * We are freeing all blocks refered from that array (numbers are stored as
3397 * little-endian 32-bit) and updating @inode->i_blocks appropriately.
3399 * We accumulate contiguous runs of blocks to free. Conveniently, if these
3400 * blocks are contiguous then releasing them at one time will only affect one
3401 * or two bitmap blocks (+ group descriptor(s) and superblock) and we won't
3402 * actually use a lot of journal space.
3404 * @this_bh will be %NULL if @first and @last point into the inode's direct
3407 static void ext4_free_data(handle_t
*handle
, struct inode
*inode
,
3408 struct buffer_head
*this_bh
,
3409 __le32
*first
, __le32
*last
)
3411 ext4_fsblk_t block_to_free
= 0; /* Starting block # of a run */
3412 unsigned long count
= 0; /* Number of blocks in the run */
3413 __le32
*block_to_free_p
= NULL
; /* Pointer into inode/ind
3416 ext4_fsblk_t nr
; /* Current block # */
3417 __le32
*p
; /* Pointer into inode/ind
3418 for current block */
3421 if (this_bh
) { /* For indirect block */
3422 BUFFER_TRACE(this_bh
, "get_write_access");
3423 err
= ext4_journal_get_write_access(handle
, this_bh
);
3424 /* Important: if we can't update the indirect pointers
3425 * to the blocks, we can't free them. */
3430 for (p
= first
; p
< last
; p
++) {
3431 nr
= le32_to_cpu(*p
);
3433 /* accumulate blocks to free if they're contiguous */
3436 block_to_free_p
= p
;
3438 } else if (nr
== block_to_free
+ count
) {
3441 ext4_clear_blocks(handle
, inode
, this_bh
,
3443 count
, block_to_free_p
, p
);
3445 block_to_free_p
= p
;
3452 ext4_clear_blocks(handle
, inode
, this_bh
, block_to_free
,
3453 count
, block_to_free_p
, p
);
3456 BUFFER_TRACE(this_bh
, "call ext4_journal_dirty_metadata");
3459 * The buffer head should have an attached journal head at this
3460 * point. However, if the data is corrupted and an indirect
3461 * block pointed to itself, it would have been detached when
3462 * the block was cleared. Check for this instead of OOPSing.
3465 ext4_journal_dirty_metadata(handle
, this_bh
);
3467 ext4_error(inode
->i_sb
, __func__
,
3468 "circular indirect block detected, "
3469 "inode=%lu, block=%llu",
3471 (unsigned long long) this_bh
->b_blocknr
);
3476 * ext4_free_branches - free an array of branches
3477 * @handle: JBD handle for this transaction
3478 * @inode: inode we are dealing with
3479 * @parent_bh: the buffer_head which contains *@first and *@last
3480 * @first: array of block numbers
3481 * @last: pointer immediately past the end of array
3482 * @depth: depth of the branches to free
3484 * We are freeing all blocks refered from these branches (numbers are
3485 * stored as little-endian 32-bit) and updating @inode->i_blocks
3488 static void ext4_free_branches(handle_t
*handle
, struct inode
*inode
,
3489 struct buffer_head
*parent_bh
,
3490 __le32
*first
, __le32
*last
, int depth
)
3495 if (is_handle_aborted(handle
))
3499 struct buffer_head
*bh
;
3500 int addr_per_block
= EXT4_ADDR_PER_BLOCK(inode
->i_sb
);
3502 while (--p
>= first
) {
3503 nr
= le32_to_cpu(*p
);
3505 continue; /* A hole */
3507 /* Go read the buffer for the next level down */
3508 bh
= sb_bread(inode
->i_sb
, nr
);
3511 * A read failure? Report error and clear slot
3515 ext4_error(inode
->i_sb
, "ext4_free_branches",
3516 "Read failure, inode=%lu, block=%llu",
3521 /* This zaps the entire block. Bottom up. */
3522 BUFFER_TRACE(bh
, "free child branches");
3523 ext4_free_branches(handle
, inode
, bh
,
3524 (__le32
*) bh
->b_data
,
3525 (__le32
*) bh
->b_data
+ addr_per_block
,
3529 * We've probably journalled the indirect block several
3530 * times during the truncate. But it's no longer
3531 * needed and we now drop it from the transaction via
3532 * jbd2_journal_revoke().
3534 * That's easy if it's exclusively part of this
3535 * transaction. But if it's part of the committing
3536 * transaction then jbd2_journal_forget() will simply
3537 * brelse() it. That means that if the underlying
3538 * block is reallocated in ext4_get_block(),
3539 * unmap_underlying_metadata() will find this block
3540 * and will try to get rid of it. damn, damn.
3542 * If this block has already been committed to the
3543 * journal, a revoke record will be written. And
3544 * revoke records must be emitted *before* clearing
3545 * this block's bit in the bitmaps.
3547 ext4_forget(handle
, 1, inode
, bh
, bh
->b_blocknr
);
3550 * Everything below this this pointer has been
3551 * released. Now let this top-of-subtree go.
3553 * We want the freeing of this indirect block to be
3554 * atomic in the journal with the updating of the
3555 * bitmap block which owns it. So make some room in
3558 * We zero the parent pointer *after* freeing its
3559 * pointee in the bitmaps, so if extend_transaction()
3560 * for some reason fails to put the bitmap changes and
3561 * the release into the same transaction, recovery
3562 * will merely complain about releasing a free block,
3563 * rather than leaking blocks.
3565 if (is_handle_aborted(handle
))
3567 if (try_to_extend_transaction(handle
, inode
)) {
3568 ext4_mark_inode_dirty(handle
, inode
);
3569 ext4_journal_test_restart(handle
, inode
);
3572 ext4_free_blocks(handle
, inode
, nr
, 1, 1);
3576 * The block which we have just freed is
3577 * pointed to by an indirect block: journal it
3579 BUFFER_TRACE(parent_bh
, "get_write_access");
3580 if (!ext4_journal_get_write_access(handle
,
3583 BUFFER_TRACE(parent_bh
,
3584 "call ext4_journal_dirty_metadata");
3585 ext4_journal_dirty_metadata(handle
,
3591 /* We have reached the bottom of the tree. */
3592 BUFFER_TRACE(parent_bh
, "free data blocks");
3593 ext4_free_data(handle
, inode
, parent_bh
, first
, last
);
3597 int ext4_can_truncate(struct inode
*inode
)
3599 if (IS_APPEND(inode
) || IS_IMMUTABLE(inode
))
3601 if (S_ISREG(inode
->i_mode
))
3603 if (S_ISDIR(inode
->i_mode
))
3605 if (S_ISLNK(inode
->i_mode
))
3606 return !ext4_inode_is_fast_symlink(inode
);
3613 * We block out ext4_get_block() block instantiations across the entire
3614 * transaction, and VFS/VM ensures that ext4_truncate() cannot run
3615 * simultaneously on behalf of the same inode.
3617 * As we work through the truncate and commmit bits of it to the journal there
3618 * is one core, guiding principle: the file's tree must always be consistent on
3619 * disk. We must be able to restart the truncate after a crash.
3621 * The file's tree may be transiently inconsistent in memory (although it
3622 * probably isn't), but whenever we close off and commit a journal transaction,
3623 * the contents of (the filesystem + the journal) must be consistent and
3624 * restartable. It's pretty simple, really: bottom up, right to left (although
3625 * left-to-right works OK too).
3627 * Note that at recovery time, journal replay occurs *before* the restart of
3628 * truncate against the orphan inode list.
3630 * The committed inode has the new, desired i_size (which is the same as
3631 * i_disksize in this case). After a crash, ext4_orphan_cleanup() will see
3632 * that this inode's truncate did not complete and it will again call
3633 * ext4_truncate() to have another go. So there will be instantiated blocks
3634 * to the right of the truncation point in a crashed ext4 filesystem. But
3635 * that's fine - as long as they are linked from the inode, the post-crash
3636 * ext4_truncate() run will find them and release them.
3638 void ext4_truncate(struct inode
*inode
)
3641 struct ext4_inode_info
*ei
= EXT4_I(inode
);
3642 __le32
*i_data
= ei
->i_data
;
3643 int addr_per_block
= EXT4_ADDR_PER_BLOCK(inode
->i_sb
);
3644 struct address_space
*mapping
= inode
->i_mapping
;
3645 ext4_lblk_t offsets
[4];
3650 ext4_lblk_t last_block
;
3651 unsigned blocksize
= inode
->i_sb
->s_blocksize
;
3653 if (!ext4_can_truncate(inode
))
3656 if (EXT4_I(inode
)->i_flags
& EXT4_EXTENTS_FL
) {
3657 ext4_ext_truncate(inode
);
3661 handle
= start_transaction(inode
);
3663 return; /* AKPM: return what? */
3665 last_block
= (inode
->i_size
+ blocksize
-1)
3666 >> EXT4_BLOCK_SIZE_BITS(inode
->i_sb
);
3668 if (inode
->i_size
& (blocksize
- 1))
3669 if (ext4_block_truncate_page(handle
, mapping
, inode
->i_size
))
3672 n
= ext4_block_to_path(inode
, last_block
, offsets
, NULL
);
3674 goto out_stop
; /* error */
3677 * OK. This truncate is going to happen. We add the inode to the
3678 * orphan list, so that if this truncate spans multiple transactions,
3679 * and we crash, we will resume the truncate when the filesystem
3680 * recovers. It also marks the inode dirty, to catch the new size.
3682 * Implication: the file must always be in a sane, consistent
3683 * truncatable state while each transaction commits.
3685 if (ext4_orphan_add(handle
, inode
))
3689 * From here we block out all ext4_get_block() callers who want to
3690 * modify the block allocation tree.
3692 down_write(&ei
->i_data_sem
);
3694 ext4_discard_reservation(inode
);
3697 * The orphan list entry will now protect us from any crash which
3698 * occurs before the truncate completes, so it is now safe to propagate
3699 * the new, shorter inode size (held for now in i_size) into the
3700 * on-disk inode. We do this via i_disksize, which is the value which
3701 * ext4 *really* writes onto the disk inode.
3703 ei
->i_disksize
= inode
->i_size
;
3705 if (n
== 1) { /* direct blocks */
3706 ext4_free_data(handle
, inode
, NULL
, i_data
+offsets
[0],
3707 i_data
+ EXT4_NDIR_BLOCKS
);
3711 partial
= ext4_find_shared(inode
, n
, offsets
, chain
, &nr
);
3712 /* Kill the top of shared branch (not detached) */
3714 if (partial
== chain
) {
3715 /* Shared branch grows from the inode */
3716 ext4_free_branches(handle
, inode
, NULL
,
3717 &nr
, &nr
+1, (chain
+n
-1) - partial
);
3720 * We mark the inode dirty prior to restart,
3721 * and prior to stop. No need for it here.
3724 /* Shared branch grows from an indirect block */
3725 BUFFER_TRACE(partial
->bh
, "get_write_access");
3726 ext4_free_branches(handle
, inode
, partial
->bh
,
3728 partial
->p
+1, (chain
+n
-1) - partial
);
3731 /* Clear the ends of indirect blocks on the shared branch */
3732 while (partial
> chain
) {
3733 ext4_free_branches(handle
, inode
, partial
->bh
, partial
->p
+ 1,
3734 (__le32
*)partial
->bh
->b_data
+addr_per_block
,
3735 (chain
+n
-1) - partial
);
3736 BUFFER_TRACE(partial
->bh
, "call brelse");
3737 brelse (partial
->bh
);
3741 /* Kill the remaining (whole) subtrees */
3742 switch (offsets
[0]) {
3744 nr
= i_data
[EXT4_IND_BLOCK
];
3746 ext4_free_branches(handle
, inode
, NULL
, &nr
, &nr
+1, 1);
3747 i_data
[EXT4_IND_BLOCK
] = 0;
3749 case EXT4_IND_BLOCK
:
3750 nr
= i_data
[EXT4_DIND_BLOCK
];
3752 ext4_free_branches(handle
, inode
, NULL
, &nr
, &nr
+1, 2);
3753 i_data
[EXT4_DIND_BLOCK
] = 0;
3755 case EXT4_DIND_BLOCK
:
3756 nr
= i_data
[EXT4_TIND_BLOCK
];
3758 ext4_free_branches(handle
, inode
, NULL
, &nr
, &nr
+1, 3);
3759 i_data
[EXT4_TIND_BLOCK
] = 0;
3761 case EXT4_TIND_BLOCK
:
3765 up_write(&ei
->i_data_sem
);
3766 inode
->i_mtime
= inode
->i_ctime
= ext4_current_time(inode
);
3767 ext4_mark_inode_dirty(handle
, inode
);
3770 * In a multi-transaction truncate, we only make the final transaction
3777 * If this was a simple ftruncate(), and the file will remain alive
3778 * then we need to clear up the orphan record which we created above.
3779 * However, if this was a real unlink then we were called by
3780 * ext4_delete_inode(), and we allow that function to clean up the
3781 * orphan info for us.
3784 ext4_orphan_del(handle
, inode
);
3786 ext4_journal_stop(handle
);
3789 static ext4_fsblk_t
ext4_get_inode_block(struct super_block
*sb
,
3790 unsigned long ino
, struct ext4_iloc
*iloc
)
3792 ext4_group_t block_group
;
3793 unsigned long offset
;
3795 struct ext4_group_desc
*gdp
;
3797 if (!ext4_valid_inum(sb
, ino
)) {
3799 * This error is already checked for in namei.c unless we are
3800 * looking at an NFS filehandle, in which case no error
3806 block_group
= (ino
- 1) / EXT4_INODES_PER_GROUP(sb
);
3807 gdp
= ext4_get_group_desc(sb
, block_group
, NULL
);
3812 * Figure out the offset within the block group inode table
3814 offset
= ((ino
- 1) % EXT4_INODES_PER_GROUP(sb
)) *
3815 EXT4_INODE_SIZE(sb
);
3816 block
= ext4_inode_table(sb
, gdp
) +
3817 (offset
>> EXT4_BLOCK_SIZE_BITS(sb
));
3819 iloc
->block_group
= block_group
;
3820 iloc
->offset
= offset
& (EXT4_BLOCK_SIZE(sb
) - 1);
3825 * ext4_get_inode_loc returns with an extra refcount against the inode's
3826 * underlying buffer_head on success. If 'in_mem' is true, we have all
3827 * data in memory that is needed to recreate the on-disk version of this
3830 static int __ext4_get_inode_loc(struct inode
*inode
,
3831 struct ext4_iloc
*iloc
, int in_mem
)
3834 struct buffer_head
*bh
;
3836 block
= ext4_get_inode_block(inode
->i_sb
, inode
->i_ino
, iloc
);
3840 bh
= sb_getblk(inode
->i_sb
, block
);
3842 ext4_error (inode
->i_sb
, "ext4_get_inode_loc",
3843 "unable to read inode block - "
3844 "inode=%lu, block=%llu",
3845 inode
->i_ino
, block
);
3848 if (!buffer_uptodate(bh
)) {
3852 * If the buffer has the write error flag, we have failed
3853 * to write out another inode in the same block. In this
3854 * case, we don't have to read the block because we may
3855 * read the old inode data successfully.
3857 if (buffer_write_io_error(bh
) && !buffer_uptodate(bh
))
3858 set_buffer_uptodate(bh
);
3860 if (buffer_uptodate(bh
)) {
3861 /* someone brought it uptodate while we waited */
3867 * If we have all information of the inode in memory and this
3868 * is the only valid inode in the block, we need not read the
3872 struct buffer_head
*bitmap_bh
;
3873 struct ext4_group_desc
*desc
;
3874 int inodes_per_buffer
;
3875 int inode_offset
, i
;
3876 ext4_group_t block_group
;
3879 block_group
= (inode
->i_ino
- 1) /
3880 EXT4_INODES_PER_GROUP(inode
->i_sb
);
3881 inodes_per_buffer
= bh
->b_size
/
3882 EXT4_INODE_SIZE(inode
->i_sb
);
3883 inode_offset
= ((inode
->i_ino
- 1) %
3884 EXT4_INODES_PER_GROUP(inode
->i_sb
));
3885 start
= inode_offset
& ~(inodes_per_buffer
- 1);
3887 /* Is the inode bitmap in cache? */
3888 desc
= ext4_get_group_desc(inode
->i_sb
,
3893 bitmap_bh
= sb_getblk(inode
->i_sb
,
3894 ext4_inode_bitmap(inode
->i_sb
, desc
));
3899 * If the inode bitmap isn't in cache then the
3900 * optimisation may end up performing two reads instead
3901 * of one, so skip it.
3903 if (!buffer_uptodate(bitmap_bh
)) {
3907 for (i
= start
; i
< start
+ inodes_per_buffer
; i
++) {
3908 if (i
== inode_offset
)
3910 if (ext4_test_bit(i
, bitmap_bh
->b_data
))
3914 if (i
== start
+ inodes_per_buffer
) {
3915 /* all other inodes are free, so skip I/O */
3916 memset(bh
->b_data
, 0, bh
->b_size
);
3917 set_buffer_uptodate(bh
);
3925 * There are other valid inodes in the buffer, this inode
3926 * has in-inode xattrs, or we don't have this inode in memory.
3927 * Read the block from disk.
3930 bh
->b_end_io
= end_buffer_read_sync
;
3931 submit_bh(READ_META
, bh
);
3933 if (!buffer_uptodate(bh
)) {
3934 ext4_error(inode
->i_sb
, "ext4_get_inode_loc",
3935 "unable to read inode block - "
3936 "inode=%lu, block=%llu",
3937 inode
->i_ino
, block
);
3947 int ext4_get_inode_loc(struct inode
*inode
, struct ext4_iloc
*iloc
)
3949 /* We have all inode data except xattrs in memory here. */
3950 return __ext4_get_inode_loc(inode
, iloc
,
3951 !(EXT4_I(inode
)->i_state
& EXT4_STATE_XATTR
));
3954 void ext4_set_inode_flags(struct inode
*inode
)
3956 unsigned int flags
= EXT4_I(inode
)->i_flags
;
3958 inode
->i_flags
&= ~(S_SYNC
|S_APPEND
|S_IMMUTABLE
|S_NOATIME
|S_DIRSYNC
);
3959 if (flags
& EXT4_SYNC_FL
)
3960 inode
->i_flags
|= S_SYNC
;
3961 if (flags
& EXT4_APPEND_FL
)
3962 inode
->i_flags
|= S_APPEND
;
3963 if (flags
& EXT4_IMMUTABLE_FL
)
3964 inode
->i_flags
|= S_IMMUTABLE
;
3965 if (flags
& EXT4_NOATIME_FL
)
3966 inode
->i_flags
|= S_NOATIME
;
3967 if (flags
& EXT4_DIRSYNC_FL
)
3968 inode
->i_flags
|= S_DIRSYNC
;
3971 /* Propagate flags from i_flags to EXT4_I(inode)->i_flags */
3972 void ext4_get_inode_flags(struct ext4_inode_info
*ei
)
3974 unsigned int flags
= ei
->vfs_inode
.i_flags
;
3976 ei
->i_flags
&= ~(EXT4_SYNC_FL
|EXT4_APPEND_FL
|
3977 EXT4_IMMUTABLE_FL
|EXT4_NOATIME_FL
|EXT4_DIRSYNC_FL
);
3979 ei
->i_flags
|= EXT4_SYNC_FL
;
3980 if (flags
& S_APPEND
)
3981 ei
->i_flags
|= EXT4_APPEND_FL
;
3982 if (flags
& S_IMMUTABLE
)
3983 ei
->i_flags
|= EXT4_IMMUTABLE_FL
;
3984 if (flags
& S_NOATIME
)
3985 ei
->i_flags
|= EXT4_NOATIME_FL
;
3986 if (flags
& S_DIRSYNC
)
3987 ei
->i_flags
|= EXT4_DIRSYNC_FL
;
3989 static blkcnt_t
ext4_inode_blocks(struct ext4_inode
*raw_inode
,
3990 struct ext4_inode_info
*ei
)
3993 struct inode
*inode
= &(ei
->vfs_inode
);
3994 struct super_block
*sb
= inode
->i_sb
;
3996 if (EXT4_HAS_RO_COMPAT_FEATURE(sb
,
3997 EXT4_FEATURE_RO_COMPAT_HUGE_FILE
)) {
3998 /* we are using combined 48 bit field */
3999 i_blocks
= ((u64
)le16_to_cpu(raw_inode
->i_blocks_high
)) << 32 |
4000 le32_to_cpu(raw_inode
->i_blocks_lo
);
4001 if (ei
->i_flags
& EXT4_HUGE_FILE_FL
) {
4002 /* i_blocks represent file system block size */
4003 return i_blocks
<< (inode
->i_blkbits
- 9);
4008 return le32_to_cpu(raw_inode
->i_blocks_lo
);
4012 struct inode
*ext4_iget(struct super_block
*sb
, unsigned long ino
)
4014 struct ext4_iloc iloc
;
4015 struct ext4_inode
*raw_inode
;
4016 struct ext4_inode_info
*ei
;
4017 struct buffer_head
*bh
;
4018 struct inode
*inode
;
4022 inode
= iget_locked(sb
, ino
);
4024 return ERR_PTR(-ENOMEM
);
4025 if (!(inode
->i_state
& I_NEW
))
4029 #ifdef CONFIG_EXT4DEV_FS_POSIX_ACL
4030 ei
->i_acl
= EXT4_ACL_NOT_CACHED
;
4031 ei
->i_default_acl
= EXT4_ACL_NOT_CACHED
;
4033 ei
->i_block_alloc_info
= NULL
;
4035 ret
= __ext4_get_inode_loc(inode
, &iloc
, 0);
4039 raw_inode
= ext4_raw_inode(&iloc
);
4040 inode
->i_mode
= le16_to_cpu(raw_inode
->i_mode
);
4041 inode
->i_uid
= (uid_t
)le16_to_cpu(raw_inode
->i_uid_low
);
4042 inode
->i_gid
= (gid_t
)le16_to_cpu(raw_inode
->i_gid_low
);
4043 if (!(test_opt(inode
->i_sb
, NO_UID32
))) {
4044 inode
->i_uid
|= le16_to_cpu(raw_inode
->i_uid_high
) << 16;
4045 inode
->i_gid
|= le16_to_cpu(raw_inode
->i_gid_high
) << 16;
4047 inode
->i_nlink
= le16_to_cpu(raw_inode
->i_links_count
);
4050 ei
->i_dir_start_lookup
= 0;
4051 ei
->i_dtime
= le32_to_cpu(raw_inode
->i_dtime
);
4052 /* We now have enough fields to check if the inode was active or not.
4053 * This is needed because nfsd might try to access dead inodes
4054 * the test is that same one that e2fsck uses
4055 * NeilBrown 1999oct15
4057 if (inode
->i_nlink
== 0) {
4058 if (inode
->i_mode
== 0 ||
4059 !(EXT4_SB(inode
->i_sb
)->s_mount_state
& EXT4_ORPHAN_FS
)) {
4060 /* this inode is deleted */
4065 /* The only unlinked inodes we let through here have
4066 * valid i_mode and are being read by the orphan
4067 * recovery code: that's fine, we're about to complete
4068 * the process of deleting those. */
4070 ei
->i_flags
= le32_to_cpu(raw_inode
->i_flags
);
4071 inode
->i_blocks
= ext4_inode_blocks(raw_inode
, ei
);
4072 ei
->i_file_acl
= le32_to_cpu(raw_inode
->i_file_acl_lo
);
4073 if (EXT4_SB(inode
->i_sb
)->s_es
->s_creator_os
!=
4074 cpu_to_le32(EXT4_OS_HURD
)) {
4076 ((__u64
)le16_to_cpu(raw_inode
->i_file_acl_high
)) << 32;
4078 inode
->i_size
= ext4_isize(raw_inode
);
4079 ei
->i_disksize
= inode
->i_size
;
4080 inode
->i_generation
= le32_to_cpu(raw_inode
->i_generation
);
4081 ei
->i_block_group
= iloc
.block_group
;
4083 * NOTE! The in-memory inode i_data array is in little-endian order
4084 * even on big-endian machines: we do NOT byteswap the block numbers!
4086 for (block
= 0; block
< EXT4_N_BLOCKS
; block
++)
4087 ei
->i_data
[block
] = raw_inode
->i_block
[block
];
4088 INIT_LIST_HEAD(&ei
->i_orphan
);
4090 if (EXT4_INODE_SIZE(inode
->i_sb
) > EXT4_GOOD_OLD_INODE_SIZE
) {
4091 ei
->i_extra_isize
= le16_to_cpu(raw_inode
->i_extra_isize
);
4092 if (EXT4_GOOD_OLD_INODE_SIZE
+ ei
->i_extra_isize
>
4093 EXT4_INODE_SIZE(inode
->i_sb
)) {
4098 if (ei
->i_extra_isize
== 0) {
4099 /* The extra space is currently unused. Use it. */
4100 ei
->i_extra_isize
= sizeof(struct ext4_inode
) -
4101 EXT4_GOOD_OLD_INODE_SIZE
;
4103 __le32
*magic
= (void *)raw_inode
+
4104 EXT4_GOOD_OLD_INODE_SIZE
+
4106 if (*magic
== cpu_to_le32(EXT4_XATTR_MAGIC
))
4107 ei
->i_state
|= EXT4_STATE_XATTR
;
4110 ei
->i_extra_isize
= 0;
4112 EXT4_INODE_GET_XTIME(i_ctime
, inode
, raw_inode
);
4113 EXT4_INODE_GET_XTIME(i_mtime
, inode
, raw_inode
);
4114 EXT4_INODE_GET_XTIME(i_atime
, inode
, raw_inode
);
4115 EXT4_EINODE_GET_XTIME(i_crtime
, ei
, raw_inode
);
4117 inode
->i_version
= le32_to_cpu(raw_inode
->i_disk_version
);
4118 if (EXT4_INODE_SIZE(inode
->i_sb
) > EXT4_GOOD_OLD_INODE_SIZE
) {
4119 if (EXT4_FITS_IN_INODE(raw_inode
, ei
, i_version_hi
))
4121 (__u64
)(le32_to_cpu(raw_inode
->i_version_hi
)) << 32;
4124 if (S_ISREG(inode
->i_mode
)) {
4125 inode
->i_op
= &ext4_file_inode_operations
;
4126 inode
->i_fop
= &ext4_file_operations
;
4127 ext4_set_aops(inode
);
4128 } else if (S_ISDIR(inode
->i_mode
)) {
4129 inode
->i_op
= &ext4_dir_inode_operations
;
4130 inode
->i_fop
= &ext4_dir_operations
;
4131 } else if (S_ISLNK(inode
->i_mode
)) {
4132 if (ext4_inode_is_fast_symlink(inode
))
4133 inode
->i_op
= &ext4_fast_symlink_inode_operations
;
4135 inode
->i_op
= &ext4_symlink_inode_operations
;
4136 ext4_set_aops(inode
);
4139 inode
->i_op
= &ext4_special_inode_operations
;
4140 if (raw_inode
->i_block
[0])
4141 init_special_inode(inode
, inode
->i_mode
,
4142 old_decode_dev(le32_to_cpu(raw_inode
->i_block
[0])));
4144 init_special_inode(inode
, inode
->i_mode
,
4145 new_decode_dev(le32_to_cpu(raw_inode
->i_block
[1])));
4148 ext4_set_inode_flags(inode
);
4149 unlock_new_inode(inode
);
4154 return ERR_PTR(ret
);
4157 static int ext4_inode_blocks_set(handle_t
*handle
,
4158 struct ext4_inode
*raw_inode
,
4159 struct ext4_inode_info
*ei
)
4161 struct inode
*inode
= &(ei
->vfs_inode
);
4162 u64 i_blocks
= inode
->i_blocks
;
4163 struct super_block
*sb
= inode
->i_sb
;
4166 if (i_blocks
<= ~0U) {
4168 * i_blocks can be represnted in a 32 bit variable
4169 * as multiple of 512 bytes
4171 raw_inode
->i_blocks_lo
= cpu_to_le32(i_blocks
);
4172 raw_inode
->i_blocks_high
= 0;
4173 ei
->i_flags
&= ~EXT4_HUGE_FILE_FL
;
4174 } else if (i_blocks
<= 0xffffffffffffULL
) {
4176 * i_blocks can be represented in a 48 bit variable
4177 * as multiple of 512 bytes
4179 err
= ext4_update_rocompat_feature(handle
, sb
,
4180 EXT4_FEATURE_RO_COMPAT_HUGE_FILE
);
4183 /* i_block is stored in the split 48 bit fields */
4184 raw_inode
->i_blocks_lo
= cpu_to_le32(i_blocks
);
4185 raw_inode
->i_blocks_high
= cpu_to_le16(i_blocks
>> 32);
4186 ei
->i_flags
&= ~EXT4_HUGE_FILE_FL
;
4189 * i_blocks should be represented in a 48 bit variable
4190 * as multiple of file system block size
4192 err
= ext4_update_rocompat_feature(handle
, sb
,
4193 EXT4_FEATURE_RO_COMPAT_HUGE_FILE
);
4196 ei
->i_flags
|= EXT4_HUGE_FILE_FL
;
4197 /* i_block is stored in file system block size */
4198 i_blocks
= i_blocks
>> (inode
->i_blkbits
- 9);
4199 raw_inode
->i_blocks_lo
= cpu_to_le32(i_blocks
);
4200 raw_inode
->i_blocks_high
= cpu_to_le16(i_blocks
>> 32);
4207 * Post the struct inode info into an on-disk inode location in the
4208 * buffer-cache. This gobbles the caller's reference to the
4209 * buffer_head in the inode location struct.
4211 * The caller must have write access to iloc->bh.
4213 static int ext4_do_update_inode(handle_t
*handle
,
4214 struct inode
*inode
,
4215 struct ext4_iloc
*iloc
)
4217 struct ext4_inode
*raw_inode
= ext4_raw_inode(iloc
);
4218 struct ext4_inode_info
*ei
= EXT4_I(inode
);
4219 struct buffer_head
*bh
= iloc
->bh
;
4220 int err
= 0, rc
, block
;
4222 /* For fields not not tracking in the in-memory inode,
4223 * initialise them to zero for new inodes. */
4224 if (ei
->i_state
& EXT4_STATE_NEW
)
4225 memset(raw_inode
, 0, EXT4_SB(inode
->i_sb
)->s_inode_size
);
4227 ext4_get_inode_flags(ei
);
4228 raw_inode
->i_mode
= cpu_to_le16(inode
->i_mode
);
4229 if (!(test_opt(inode
->i_sb
, NO_UID32
))) {
4230 raw_inode
->i_uid_low
= cpu_to_le16(low_16_bits(inode
->i_uid
));
4231 raw_inode
->i_gid_low
= cpu_to_le16(low_16_bits(inode
->i_gid
));
4233 * Fix up interoperability with old kernels. Otherwise, old inodes get
4234 * re-used with the upper 16 bits of the uid/gid intact
4237 raw_inode
->i_uid_high
=
4238 cpu_to_le16(high_16_bits(inode
->i_uid
));
4239 raw_inode
->i_gid_high
=
4240 cpu_to_le16(high_16_bits(inode
->i_gid
));
4242 raw_inode
->i_uid_high
= 0;
4243 raw_inode
->i_gid_high
= 0;
4246 raw_inode
->i_uid_low
=
4247 cpu_to_le16(fs_high2lowuid(inode
->i_uid
));
4248 raw_inode
->i_gid_low
=
4249 cpu_to_le16(fs_high2lowgid(inode
->i_gid
));
4250 raw_inode
->i_uid_high
= 0;
4251 raw_inode
->i_gid_high
= 0;
4253 raw_inode
->i_links_count
= cpu_to_le16(inode
->i_nlink
);
4255 EXT4_INODE_SET_XTIME(i_ctime
, inode
, raw_inode
);
4256 EXT4_INODE_SET_XTIME(i_mtime
, inode
, raw_inode
);
4257 EXT4_INODE_SET_XTIME(i_atime
, inode
, raw_inode
);
4258 EXT4_EINODE_SET_XTIME(i_crtime
, ei
, raw_inode
);
4260 if (ext4_inode_blocks_set(handle
, raw_inode
, ei
))
4262 raw_inode
->i_dtime
= cpu_to_le32(ei
->i_dtime
);
4263 /* clear the migrate flag in the raw_inode */
4264 raw_inode
->i_flags
= cpu_to_le32(ei
->i_flags
& ~EXT4_EXT_MIGRATE
);
4265 if (EXT4_SB(inode
->i_sb
)->s_es
->s_creator_os
!=
4266 cpu_to_le32(EXT4_OS_HURD
))
4267 raw_inode
->i_file_acl_high
=
4268 cpu_to_le16(ei
->i_file_acl
>> 32);
4269 raw_inode
->i_file_acl_lo
= cpu_to_le32(ei
->i_file_acl
);
4270 ext4_isize_set(raw_inode
, ei
->i_disksize
);
4271 if (ei
->i_disksize
> 0x7fffffffULL
) {
4272 struct super_block
*sb
= inode
->i_sb
;
4273 if (!EXT4_HAS_RO_COMPAT_FEATURE(sb
,
4274 EXT4_FEATURE_RO_COMPAT_LARGE_FILE
) ||
4275 EXT4_SB(sb
)->s_es
->s_rev_level
==
4276 cpu_to_le32(EXT4_GOOD_OLD_REV
)) {
4277 /* If this is the first large file
4278 * created, add a flag to the superblock.
4280 err
= ext4_journal_get_write_access(handle
,
4281 EXT4_SB(sb
)->s_sbh
);
4284 ext4_update_dynamic_rev(sb
);
4285 EXT4_SET_RO_COMPAT_FEATURE(sb
,
4286 EXT4_FEATURE_RO_COMPAT_LARGE_FILE
);
4289 err
= ext4_journal_dirty_metadata(handle
,
4290 EXT4_SB(sb
)->s_sbh
);
4293 raw_inode
->i_generation
= cpu_to_le32(inode
->i_generation
);
4294 if (S_ISCHR(inode
->i_mode
) || S_ISBLK(inode
->i_mode
)) {
4295 if (old_valid_dev(inode
->i_rdev
)) {
4296 raw_inode
->i_block
[0] =
4297 cpu_to_le32(old_encode_dev(inode
->i_rdev
));
4298 raw_inode
->i_block
[1] = 0;
4300 raw_inode
->i_block
[0] = 0;
4301 raw_inode
->i_block
[1] =
4302 cpu_to_le32(new_encode_dev(inode
->i_rdev
));
4303 raw_inode
->i_block
[2] = 0;
4305 } else for (block
= 0; block
< EXT4_N_BLOCKS
; block
++)
4306 raw_inode
->i_block
[block
] = ei
->i_data
[block
];
4308 raw_inode
->i_disk_version
= cpu_to_le32(inode
->i_version
);
4309 if (ei
->i_extra_isize
) {
4310 if (EXT4_FITS_IN_INODE(raw_inode
, ei
, i_version_hi
))
4311 raw_inode
->i_version_hi
=
4312 cpu_to_le32(inode
->i_version
>> 32);
4313 raw_inode
->i_extra_isize
= cpu_to_le16(ei
->i_extra_isize
);
4317 BUFFER_TRACE(bh
, "call ext4_journal_dirty_metadata");
4318 rc
= ext4_journal_dirty_metadata(handle
, bh
);
4321 ei
->i_state
&= ~EXT4_STATE_NEW
;
4325 ext4_std_error(inode
->i_sb
, err
);
4330 * ext4_write_inode()
4332 * We are called from a few places:
4334 * - Within generic_file_write() for O_SYNC files.
4335 * Here, there will be no transaction running. We wait for any running
4336 * trasnaction to commit.
4338 * - Within sys_sync(), kupdate and such.
4339 * We wait on commit, if tol to.
4341 * - Within prune_icache() (PF_MEMALLOC == true)
4342 * Here we simply return. We can't afford to block kswapd on the
4345 * In all cases it is actually safe for us to return without doing anything,
4346 * because the inode has been copied into a raw inode buffer in
4347 * ext4_mark_inode_dirty(). This is a correctness thing for O_SYNC and for
4350 * Note that we are absolutely dependent upon all inode dirtiers doing the
4351 * right thing: they *must* call mark_inode_dirty() after dirtying info in
4352 * which we are interested.
4354 * It would be a bug for them to not do this. The code:
4356 * mark_inode_dirty(inode)
4358 * inode->i_size = expr;
4360 * is in error because a kswapd-driven write_inode() could occur while
4361 * `stuff()' is running, and the new i_size will be lost. Plus the inode
4362 * will no longer be on the superblock's dirty inode list.
4364 int ext4_write_inode(struct inode
*inode
, int wait
)
4366 if (current
->flags
& PF_MEMALLOC
)
4369 if (ext4_journal_current_handle()) {
4370 jbd_debug(1, "called recursively, non-PF_MEMALLOC!\n");
4378 return ext4_force_commit(inode
->i_sb
);
4384 * Called from notify_change.
4386 * We want to trap VFS attempts to truncate the file as soon as
4387 * possible. In particular, we want to make sure that when the VFS
4388 * shrinks i_size, we put the inode on the orphan list and modify
4389 * i_disksize immediately, so that during the subsequent flushing of
4390 * dirty pages and freeing of disk blocks, we can guarantee that any
4391 * commit will leave the blocks being flushed in an unused state on
4392 * disk. (On recovery, the inode will get truncated and the blocks will
4393 * be freed, so we have a strong guarantee that no future commit will
4394 * leave these blocks visible to the user.)
4396 * Another thing we have to assure is that if we are in ordered mode
4397 * and inode is still attached to the committing transaction, we must
4398 * we start writeout of all the dirty pages which are being truncated.
4399 * This way we are sure that all the data written in the previous
4400 * transaction are already on disk (truncate waits for pages under
4403 * Called with inode->i_mutex down.
4405 int ext4_setattr(struct dentry
*dentry
, struct iattr
*attr
)
4407 struct inode
*inode
= dentry
->d_inode
;
4409 const unsigned int ia_valid
= attr
->ia_valid
;
4411 error
= inode_change_ok(inode
, attr
);
4415 if ((ia_valid
& ATTR_UID
&& attr
->ia_uid
!= inode
->i_uid
) ||
4416 (ia_valid
& ATTR_GID
&& attr
->ia_gid
!= inode
->i_gid
)) {
4419 /* (user+group)*(old+new) structure, inode write (sb,
4420 * inode block, ? - but truncate inode update has it) */
4421 handle
= ext4_journal_start(inode
, 2*(EXT4_QUOTA_INIT_BLOCKS(inode
->i_sb
)+
4422 EXT4_QUOTA_DEL_BLOCKS(inode
->i_sb
))+3);
4423 if (IS_ERR(handle
)) {
4424 error
= PTR_ERR(handle
);
4427 error
= DQUOT_TRANSFER(inode
, attr
) ? -EDQUOT
: 0;
4429 ext4_journal_stop(handle
);
4432 /* Update corresponding info in inode so that everything is in
4433 * one transaction */
4434 if (attr
->ia_valid
& ATTR_UID
)
4435 inode
->i_uid
= attr
->ia_uid
;
4436 if (attr
->ia_valid
& ATTR_GID
)
4437 inode
->i_gid
= attr
->ia_gid
;
4438 error
= ext4_mark_inode_dirty(handle
, inode
);
4439 ext4_journal_stop(handle
);
4442 if (attr
->ia_valid
& ATTR_SIZE
) {
4443 if (!(EXT4_I(inode
)->i_flags
& EXT4_EXTENTS_FL
)) {
4444 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
4446 if (attr
->ia_size
> sbi
->s_bitmap_maxbytes
) {
4453 if (S_ISREG(inode
->i_mode
) &&
4454 attr
->ia_valid
& ATTR_SIZE
&& attr
->ia_size
< inode
->i_size
) {
4457 handle
= ext4_journal_start(inode
, 3);
4458 if (IS_ERR(handle
)) {
4459 error
= PTR_ERR(handle
);
4463 error
= ext4_orphan_add(handle
, inode
);
4464 EXT4_I(inode
)->i_disksize
= attr
->ia_size
;
4465 rc
= ext4_mark_inode_dirty(handle
, inode
);
4468 ext4_journal_stop(handle
);
4470 if (ext4_should_order_data(inode
)) {
4471 error
= ext4_begin_ordered_truncate(inode
,
4474 /* Do as much error cleanup as possible */
4475 handle
= ext4_journal_start(inode
, 3);
4476 if (IS_ERR(handle
)) {
4477 ext4_orphan_del(NULL
, inode
);
4480 ext4_orphan_del(handle
, inode
);
4481 ext4_journal_stop(handle
);
4487 rc
= inode_setattr(inode
, attr
);
4489 /* If inode_setattr's call to ext4_truncate failed to get a
4490 * transaction handle at all, we need to clean up the in-core
4491 * orphan list manually. */
4493 ext4_orphan_del(NULL
, inode
);
4495 if (!rc
&& (ia_valid
& ATTR_MODE
))
4496 rc
= ext4_acl_chmod(inode
);
4499 ext4_std_error(inode
->i_sb
, error
);
4505 int ext4_getattr(struct vfsmount
*mnt
, struct dentry
*dentry
,
4508 struct inode
*inode
;
4509 unsigned long delalloc_blocks
;
4511 inode
= dentry
->d_inode
;
4512 generic_fillattr(inode
, stat
);
4515 * We can't update i_blocks if the block allocation is delayed
4516 * otherwise in the case of system crash before the real block
4517 * allocation is done, we will have i_blocks inconsistent with
4518 * on-disk file blocks.
4519 * We always keep i_blocks updated together with real
4520 * allocation. But to not confuse with user, stat
4521 * will return the blocks that include the delayed allocation
4522 * blocks for this file.
4524 spin_lock(&EXT4_I(inode
)->i_block_reservation_lock
);
4525 delalloc_blocks
= EXT4_I(inode
)->i_reserved_data_blocks
;
4526 spin_unlock(&EXT4_I(inode
)->i_block_reservation_lock
);
4528 stat
->blocks
+= (delalloc_blocks
<< inode
->i_sb
->s_blocksize_bits
)>>9;
4532 static int ext4_indirect_trans_blocks(struct inode
*inode
, int nrblocks
,
4537 /* if nrblocks are contiguous */
4540 * With N contiguous data blocks, it need at most
4541 * N/EXT4_ADDR_PER_BLOCK(inode->i_sb) indirect blocks
4542 * 2 dindirect blocks
4545 indirects
= nrblocks
/ EXT4_ADDR_PER_BLOCK(inode
->i_sb
);
4546 return indirects
+ 3;
4549 * if nrblocks are not contiguous, worse case, each block touch
4550 * a indirect block, and each indirect block touch a double indirect
4551 * block, plus a triple indirect block
4553 indirects
= nrblocks
* 2 + 1;
4557 static int ext4_index_trans_blocks(struct inode
*inode
, int nrblocks
, int chunk
)
4559 if (!(EXT4_I(inode
)->i_flags
& EXT4_EXTENTS_FL
))
4560 return ext4_indirect_trans_blocks(inode
, nrblocks
, 0);
4561 return ext4_ext_index_trans_blocks(inode
, nrblocks
, 0);
4564 * Account for index blocks, block groups bitmaps and block group
4565 * descriptor blocks if modify datablocks and index blocks
4566 * worse case, the indexs blocks spread over different block groups
4568 * If datablocks are discontiguous, they are possible to spread over
4569 * different block groups too. If they are contiugous, with flexbg,
4570 * they could still across block group boundary.
4572 * Also account for superblock, inode, quota and xattr blocks
4574 int ext4_meta_trans_blocks(struct inode
*inode
, int nrblocks
, int chunk
)
4576 int groups
, gdpblocks
;
4581 * How many index blocks need to touch to modify nrblocks?
4582 * The "Chunk" flag indicating whether the nrblocks is
4583 * physically contiguous on disk
4585 * For Direct IO and fallocate, they calls get_block to allocate
4586 * one single extent at a time, so they could set the "Chunk" flag
4588 idxblocks
= ext4_index_trans_blocks(inode
, nrblocks
, chunk
);
4593 * Now let's see how many group bitmaps and group descriptors need
4603 if (groups
> EXT4_SB(inode
->i_sb
)->s_groups_count
)
4604 groups
= EXT4_SB(inode
->i_sb
)->s_groups_count
;
4605 if (groups
> EXT4_SB(inode
->i_sb
)->s_gdb_count
)
4606 gdpblocks
= EXT4_SB(inode
->i_sb
)->s_gdb_count
;
4608 /* bitmaps and block group descriptor blocks */
4609 ret
+= groups
+ gdpblocks
;
4611 /* Blocks for super block, inode, quota and xattr blocks */
4612 ret
+= EXT4_META_TRANS_BLOCKS(inode
->i_sb
);
4618 * Calulate the total number of credits to reserve to fit
4619 * the modification of a single pages into a single transaction,
4620 * which may include multiple chunks of block allocations.
4622 * This could be called via ext4_write_begin()
4624 * We need to consider the worse case, when
4625 * one new block per extent.
4627 int ext4_writepage_trans_blocks(struct inode
*inode
)
4629 int bpp
= ext4_journal_blocks_per_page(inode
);
4632 ret
= ext4_meta_trans_blocks(inode
, bpp
, 0);
4634 /* Account for data blocks for journalled mode */
4635 if (ext4_should_journal_data(inode
))
4641 * Calculate the journal credits for a chunk of data modification.
4643 * This is called from DIO, fallocate or whoever calling
4644 * ext4_get_blocks_wrap() to map/allocate a chunk of contigous disk blocks.
4646 * journal buffers for data blocks are not included here, as DIO
4647 * and fallocate do no need to journal data buffers.
4649 int ext4_chunk_trans_blocks(struct inode
*inode
, int nrblocks
)
4651 return ext4_meta_trans_blocks(inode
, nrblocks
, 1);
4655 * The caller must have previously called ext4_reserve_inode_write().
4656 * Give this, we know that the caller already has write access to iloc->bh.
4658 int ext4_mark_iloc_dirty(handle_t
*handle
,
4659 struct inode
*inode
, struct ext4_iloc
*iloc
)
4663 if (test_opt(inode
->i_sb
, I_VERSION
))
4664 inode_inc_iversion(inode
);
4666 /* the do_update_inode consumes one bh->b_count */
4669 /* ext4_do_update_inode() does jbd2_journal_dirty_metadata */
4670 err
= ext4_do_update_inode(handle
, inode
, iloc
);
4676 * On success, We end up with an outstanding reference count against
4677 * iloc->bh. This _must_ be cleaned up later.
4681 ext4_reserve_inode_write(handle_t
*handle
, struct inode
*inode
,
4682 struct ext4_iloc
*iloc
)
4686 err
= ext4_get_inode_loc(inode
, iloc
);
4688 BUFFER_TRACE(iloc
->bh
, "get_write_access");
4689 err
= ext4_journal_get_write_access(handle
, iloc
->bh
);
4696 ext4_std_error(inode
->i_sb
, err
);
4701 * Expand an inode by new_extra_isize bytes.
4702 * Returns 0 on success or negative error number on failure.
4704 static int ext4_expand_extra_isize(struct inode
*inode
,
4705 unsigned int new_extra_isize
,
4706 struct ext4_iloc iloc
,
4709 struct ext4_inode
*raw_inode
;
4710 struct ext4_xattr_ibody_header
*header
;
4711 struct ext4_xattr_entry
*entry
;
4713 if (EXT4_I(inode
)->i_extra_isize
>= new_extra_isize
)
4716 raw_inode
= ext4_raw_inode(&iloc
);
4718 header
= IHDR(inode
, raw_inode
);
4719 entry
= IFIRST(header
);
4721 /* No extended attributes present */
4722 if (!(EXT4_I(inode
)->i_state
& EXT4_STATE_XATTR
) ||
4723 header
->h_magic
!= cpu_to_le32(EXT4_XATTR_MAGIC
)) {
4724 memset((void *)raw_inode
+ EXT4_GOOD_OLD_INODE_SIZE
, 0,
4726 EXT4_I(inode
)->i_extra_isize
= new_extra_isize
;
4730 /* try to expand with EAs present */
4731 return ext4_expand_extra_isize_ea(inode
, new_extra_isize
,
4736 * What we do here is to mark the in-core inode as clean with respect to inode
4737 * dirtiness (it may still be data-dirty).
4738 * This means that the in-core inode may be reaped by prune_icache
4739 * without having to perform any I/O. This is a very good thing,
4740 * because *any* task may call prune_icache - even ones which
4741 * have a transaction open against a different journal.
4743 * Is this cheating? Not really. Sure, we haven't written the
4744 * inode out, but prune_icache isn't a user-visible syncing function.
4745 * Whenever the user wants stuff synced (sys_sync, sys_msync, sys_fsync)
4746 * we start and wait on commits.
4748 * Is this efficient/effective? Well, we're being nice to the system
4749 * by cleaning up our inodes proactively so they can be reaped
4750 * without I/O. But we are potentially leaving up to five seconds'
4751 * worth of inodes floating about which prune_icache wants us to
4752 * write out. One way to fix that would be to get prune_icache()
4753 * to do a write_super() to free up some memory. It has the desired
4756 int ext4_mark_inode_dirty(handle_t
*handle
, struct inode
*inode
)
4758 struct ext4_iloc iloc
;
4759 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
4760 static unsigned int mnt_count
;
4764 err
= ext4_reserve_inode_write(handle
, inode
, &iloc
);
4765 if (EXT4_I(inode
)->i_extra_isize
< sbi
->s_want_extra_isize
&&
4766 !(EXT4_I(inode
)->i_state
& EXT4_STATE_NO_EXPAND
)) {
4768 * We need extra buffer credits since we may write into EA block
4769 * with this same handle. If journal_extend fails, then it will
4770 * only result in a minor loss of functionality for that inode.
4771 * If this is felt to be critical, then e2fsck should be run to
4772 * force a large enough s_min_extra_isize.
4774 if ((jbd2_journal_extend(handle
,
4775 EXT4_DATA_TRANS_BLOCKS(inode
->i_sb
))) == 0) {
4776 ret
= ext4_expand_extra_isize(inode
,
4777 sbi
->s_want_extra_isize
,
4780 EXT4_I(inode
)->i_state
|= EXT4_STATE_NO_EXPAND
;
4782 le16_to_cpu(sbi
->s_es
->s_mnt_count
)) {
4783 ext4_warning(inode
->i_sb
, __func__
,
4784 "Unable to expand inode %lu. Delete"
4785 " some EAs or run e2fsck.",
4788 le16_to_cpu(sbi
->s_es
->s_mnt_count
);
4794 err
= ext4_mark_iloc_dirty(handle
, inode
, &iloc
);
4799 * ext4_dirty_inode() is called from __mark_inode_dirty()
4801 * We're really interested in the case where a file is being extended.
4802 * i_size has been changed by generic_commit_write() and we thus need
4803 * to include the updated inode in the current transaction.
4805 * Also, DQUOT_ALLOC_SPACE() will always dirty the inode when blocks
4806 * are allocated to the file.
4808 * If the inode is marked synchronous, we don't honour that here - doing
4809 * so would cause a commit on atime updates, which we don't bother doing.
4810 * We handle synchronous inodes at the highest possible level.
4812 void ext4_dirty_inode(struct inode
*inode
)
4814 handle_t
*current_handle
= ext4_journal_current_handle();
4817 handle
= ext4_journal_start(inode
, 2);
4820 if (current_handle
&&
4821 current_handle
->h_transaction
!= handle
->h_transaction
) {
4822 /* This task has a transaction open against a different fs */
4823 printk(KERN_EMERG
"%s: transactions do not match!\n",
4826 jbd_debug(5, "marking dirty. outer handle=%p\n",
4828 ext4_mark_inode_dirty(handle
, inode
);
4830 ext4_journal_stop(handle
);
4837 * Bind an inode's backing buffer_head into this transaction, to prevent
4838 * it from being flushed to disk early. Unlike
4839 * ext4_reserve_inode_write, this leaves behind no bh reference and
4840 * returns no iloc structure, so the caller needs to repeat the iloc
4841 * lookup to mark the inode dirty later.
4843 static int ext4_pin_inode(handle_t
*handle
, struct inode
*inode
)
4845 struct ext4_iloc iloc
;
4849 err
= ext4_get_inode_loc(inode
, &iloc
);
4851 BUFFER_TRACE(iloc
.bh
, "get_write_access");
4852 err
= jbd2_journal_get_write_access(handle
, iloc
.bh
);
4854 err
= ext4_journal_dirty_metadata(handle
,
4859 ext4_std_error(inode
->i_sb
, err
);
4864 int ext4_change_inode_journal_flag(struct inode
*inode
, int val
)
4871 * We have to be very careful here: changing a data block's
4872 * journaling status dynamically is dangerous. If we write a
4873 * data block to the journal, change the status and then delete
4874 * that block, we risk forgetting to revoke the old log record
4875 * from the journal and so a subsequent replay can corrupt data.
4876 * So, first we make sure that the journal is empty and that
4877 * nobody is changing anything.
4880 journal
= EXT4_JOURNAL(inode
);
4881 if (is_journal_aborted(journal
))
4884 jbd2_journal_lock_updates(journal
);
4885 jbd2_journal_flush(journal
);
4888 * OK, there are no updates running now, and all cached data is
4889 * synced to disk. We are now in a completely consistent state
4890 * which doesn't have anything in the journal, and we know that
4891 * no filesystem updates are running, so it is safe to modify
4892 * the inode's in-core data-journaling state flag now.
4896 EXT4_I(inode
)->i_flags
|= EXT4_JOURNAL_DATA_FL
;
4898 EXT4_I(inode
)->i_flags
&= ~EXT4_JOURNAL_DATA_FL
;
4899 ext4_set_aops(inode
);
4901 jbd2_journal_unlock_updates(journal
);
4903 /* Finally we can mark the inode as dirty. */
4905 handle
= ext4_journal_start(inode
, 1);
4907 return PTR_ERR(handle
);
4909 err
= ext4_mark_inode_dirty(handle
, inode
);
4911 ext4_journal_stop(handle
);
4912 ext4_std_error(inode
->i_sb
, err
);
4917 static int ext4_bh_unmapped(handle_t
*handle
, struct buffer_head
*bh
)
4919 return !buffer_mapped(bh
);
4922 int ext4_page_mkwrite(struct vm_area_struct
*vma
, struct page
*page
)
4928 struct file
*file
= vma
->vm_file
;
4929 struct inode
*inode
= file
->f_path
.dentry
->d_inode
;
4930 struct address_space
*mapping
= inode
->i_mapping
;
4933 * Get i_alloc_sem to stop truncates messing with the inode. We cannot
4934 * get i_mutex because we are already holding mmap_sem.
4936 down_read(&inode
->i_alloc_sem
);
4937 size
= i_size_read(inode
);
4938 if (page
->mapping
!= mapping
|| size
<= page_offset(page
)
4939 || !PageUptodate(page
)) {
4940 /* page got truncated from under us? */
4944 if (PageMappedToDisk(page
))
4947 if (page
->index
== size
>> PAGE_CACHE_SHIFT
)
4948 len
= size
& ~PAGE_CACHE_MASK
;
4950 len
= PAGE_CACHE_SIZE
;
4952 if (page_has_buffers(page
)) {
4953 /* return if we have all the buffers mapped */
4954 if (!walk_page_buffers(NULL
, page_buffers(page
), 0, len
, NULL
,
4959 * OK, we need to fill the hole... Do write_begin write_end
4960 * to do block allocation/reservation.We are not holding
4961 * inode.i__mutex here. That allow * parallel write_begin,
4962 * write_end call. lock_page prevent this from happening
4963 * on the same page though
4965 ret
= mapping
->a_ops
->write_begin(file
, mapping
, page_offset(page
),
4966 len
, AOP_FLAG_UNINTERRUPTIBLE
, &page
, &fsdata
);
4969 ret
= mapping
->a_ops
->write_end(file
, mapping
, page_offset(page
),
4970 len
, len
, page
, fsdata
);
4975 up_read(&inode
->i_alloc_sem
);