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
,
490 * XXX need to get goal block from mballoc's data structures
493 return ext4_find_near(inode
, partial
);
497 * ext4_blks_to_allocate: Look up the block map and count the number
498 * of direct blocks need to be allocated for the given branch.
500 * @branch: chain of indirect blocks
501 * @k: number of blocks need for indirect blocks
502 * @blks: number of data blocks to be mapped.
503 * @blocks_to_boundary: the offset in the indirect block
505 * return the total number of blocks to be allocate, including the
506 * direct and indirect blocks.
508 static int ext4_blks_to_allocate(Indirect
*branch
, int k
, unsigned long blks
,
509 int blocks_to_boundary
)
511 unsigned long count
= 0;
514 * Simple case, [t,d]Indirect block(s) has not allocated yet
515 * then it's clear blocks on that path have not allocated
518 /* right now we don't handle cross boundary allocation */
519 if (blks
< blocks_to_boundary
+ 1)
522 count
+= blocks_to_boundary
+ 1;
527 while (count
< blks
&& count
<= blocks_to_boundary
&&
528 le32_to_cpu(*(branch
[0].p
+ count
)) == 0) {
535 * ext4_alloc_blocks: multiple allocate blocks needed for a branch
536 * @indirect_blks: the number of blocks need to allocate for indirect
539 * @new_blocks: on return it will store the new block numbers for
540 * the indirect blocks(if needed) and the first direct block,
541 * @blks: on return it will store the total number of allocated
544 static int ext4_alloc_blocks(handle_t
*handle
, struct inode
*inode
,
545 ext4_lblk_t iblock
, ext4_fsblk_t goal
,
546 int indirect_blks
, int blks
,
547 ext4_fsblk_t new_blocks
[4], int *err
)
550 unsigned long count
= 0, blk_allocated
= 0;
552 ext4_fsblk_t current_block
= 0;
556 * Here we try to allocate the requested multiple blocks at once,
557 * on a best-effort basis.
558 * To build a branch, we should allocate blocks for
559 * the indirect blocks(if not allocated yet), and at least
560 * the first direct block of this branch. That's the
561 * minimum number of blocks need to allocate(required)
563 /* first we try to allocate the indirect blocks */
564 target
= indirect_blks
;
567 /* allocating blocks for indirect blocks and direct blocks */
568 current_block
= ext4_new_meta_blocks(handle
, inode
,
574 /* allocate blocks for indirect blocks */
575 while (index
< indirect_blks
&& count
) {
576 new_blocks
[index
++] = current_block
++;
581 * save the new block number
582 * for the first direct block
584 new_blocks
[index
] = current_block
;
585 printk(KERN_INFO
"%s returned more blocks than "
586 "requested\n", __func__
);
592 target
= blks
- count
;
593 blk_allocated
= count
;
596 /* Now allocate data blocks */
598 /* allocating blocks for data blocks */
599 current_block
= ext4_new_blocks(handle
, inode
, iblock
,
601 if (*err
&& (target
== blks
)) {
603 * if the allocation failed and we didn't allocate
609 if (target
== blks
) {
611 * save the new block number
612 * for the first direct block
614 new_blocks
[index
] = current_block
;
616 blk_allocated
+= count
;
619 /* total number of blocks allocated for direct blocks */
624 for (i
= 0; i
< index
; i
++)
625 ext4_free_blocks(handle
, inode
, new_blocks
[i
], 1, 0);
630 * ext4_alloc_branch - allocate and set up a chain of blocks.
632 * @indirect_blks: number of allocated indirect blocks
633 * @blks: number of allocated direct blocks
634 * @offsets: offsets (in the blocks) to store the pointers to next.
635 * @branch: place to store the chain in.
637 * This function allocates blocks, zeroes out all but the last one,
638 * links them into chain and (if we are synchronous) writes them to disk.
639 * In other words, it prepares a branch that can be spliced onto the
640 * inode. It stores the information about that chain in the branch[], in
641 * the same format as ext4_get_branch() would do. We are calling it after
642 * we had read the existing part of chain and partial points to the last
643 * triple of that (one with zero ->key). Upon the exit we have the same
644 * picture as after the successful ext4_get_block(), except that in one
645 * place chain is disconnected - *branch->p is still zero (we did not
646 * set the last link), but branch->key contains the number that should
647 * be placed into *branch->p to fill that gap.
649 * If allocation fails we free all blocks we've allocated (and forget
650 * their buffer_heads) and return the error value the from failed
651 * ext4_alloc_block() (normally -ENOSPC). Otherwise we set the chain
652 * as described above and return 0.
654 static int ext4_alloc_branch(handle_t
*handle
, struct inode
*inode
,
655 ext4_lblk_t iblock
, int indirect_blks
,
656 int *blks
, ext4_fsblk_t goal
,
657 ext4_lblk_t
*offsets
, Indirect
*branch
)
659 int blocksize
= inode
->i_sb
->s_blocksize
;
662 struct buffer_head
*bh
;
664 ext4_fsblk_t new_blocks
[4];
665 ext4_fsblk_t current_block
;
667 num
= ext4_alloc_blocks(handle
, inode
, iblock
, goal
, indirect_blks
,
668 *blks
, new_blocks
, &err
);
672 branch
[0].key
= cpu_to_le32(new_blocks
[0]);
674 * metadata blocks and data blocks are allocated.
676 for (n
= 1; n
<= indirect_blks
; n
++) {
678 * Get buffer_head for parent block, zero it out
679 * and set the pointer to new one, then send
682 bh
= sb_getblk(inode
->i_sb
, new_blocks
[n
-1]);
685 BUFFER_TRACE(bh
, "call get_create_access");
686 err
= ext4_journal_get_create_access(handle
, bh
);
693 memset(bh
->b_data
, 0, blocksize
);
694 branch
[n
].p
= (__le32
*) bh
->b_data
+ offsets
[n
];
695 branch
[n
].key
= cpu_to_le32(new_blocks
[n
]);
696 *branch
[n
].p
= branch
[n
].key
;
697 if (n
== indirect_blks
) {
698 current_block
= new_blocks
[n
];
700 * End of chain, update the last new metablock of
701 * the chain to point to the new allocated
702 * data blocks numbers
704 for (i
=1; i
< num
; i
++)
705 *(branch
[n
].p
+ i
) = cpu_to_le32(++current_block
);
707 BUFFER_TRACE(bh
, "marking uptodate");
708 set_buffer_uptodate(bh
);
711 BUFFER_TRACE(bh
, "call ext4_journal_dirty_metadata");
712 err
= ext4_journal_dirty_metadata(handle
, bh
);
719 /* Allocation failed, free what we already allocated */
720 for (i
= 1; i
<= n
; i
++) {
721 BUFFER_TRACE(branch
[i
].bh
, "call jbd2_journal_forget");
722 ext4_journal_forget(handle
, branch
[i
].bh
);
724 for (i
= 0; i
< indirect_blks
; i
++)
725 ext4_free_blocks(handle
, inode
, new_blocks
[i
], 1, 0);
727 ext4_free_blocks(handle
, inode
, new_blocks
[i
], num
, 0);
733 * ext4_splice_branch - splice the allocated branch onto inode.
735 * @block: (logical) number of block we are adding
736 * @chain: chain of indirect blocks (with a missing link - see
738 * @where: location of missing link
739 * @num: number of indirect blocks we are adding
740 * @blks: number of direct blocks we are adding
742 * This function fills the missing link and does all housekeeping needed in
743 * inode (->i_blocks, etc.). In case of success we end up with the full
744 * chain to new block and return 0.
746 static int ext4_splice_branch(handle_t
*handle
, struct inode
*inode
,
747 ext4_lblk_t block
, Indirect
*where
, int num
, int blks
)
751 ext4_fsblk_t current_block
;
754 * If we're splicing into a [td]indirect block (as opposed to the
755 * inode) then we need to get write access to the [td]indirect block
759 BUFFER_TRACE(where
->bh
, "get_write_access");
760 err
= ext4_journal_get_write_access(handle
, where
->bh
);
766 *where
->p
= where
->key
;
769 * Update the host buffer_head or inode to point to more just allocated
770 * direct blocks blocks
772 if (num
== 0 && blks
> 1) {
773 current_block
= le32_to_cpu(where
->key
) + 1;
774 for (i
= 1; i
< blks
; i
++)
775 *(where
->p
+ i
) = cpu_to_le32(current_block
++);
778 /* We are done with atomic stuff, now do the rest of housekeeping */
780 inode
->i_ctime
= ext4_current_time(inode
);
781 ext4_mark_inode_dirty(handle
, inode
);
783 /* had we spliced it onto indirect block? */
786 * If we spliced it onto an indirect block, we haven't
787 * altered the inode. Note however that if it is being spliced
788 * onto an indirect block at the very end of the file (the
789 * file is growing) then we *will* alter the inode to reflect
790 * the new i_size. But that is not done here - it is done in
791 * generic_commit_write->__mark_inode_dirty->ext4_dirty_inode.
793 jbd_debug(5, "splicing indirect only\n");
794 BUFFER_TRACE(where
->bh
, "call ext4_journal_dirty_metadata");
795 err
= ext4_journal_dirty_metadata(handle
, where
->bh
);
800 * OK, we spliced it into the inode itself on a direct block.
801 * Inode was dirtied above.
803 jbd_debug(5, "splicing direct\n");
808 for (i
= 1; i
<= num
; i
++) {
809 BUFFER_TRACE(where
[i
].bh
, "call jbd2_journal_forget");
810 ext4_journal_forget(handle
, where
[i
].bh
);
811 ext4_free_blocks(handle
, inode
,
812 le32_to_cpu(where
[i
-1].key
), 1, 0);
814 ext4_free_blocks(handle
, inode
, le32_to_cpu(where
[num
].key
), blks
, 0);
820 * Allocation strategy is simple: if we have to allocate something, we will
821 * have to go the whole way to leaf. So let's do it before attaching anything
822 * to tree, set linkage between the newborn blocks, write them if sync is
823 * required, recheck the path, free and repeat if check fails, otherwise
824 * set the last missing link (that will protect us from any truncate-generated
825 * removals - all blocks on the path are immune now) and possibly force the
826 * write on the parent block.
827 * That has a nice additional property: no special recovery from the failed
828 * allocations is needed - we simply release blocks and do not touch anything
829 * reachable from inode.
831 * `handle' can be NULL if create == 0.
833 * return > 0, # of blocks mapped or allocated.
834 * return = 0, if plain lookup failed.
835 * return < 0, error case.
838 * Need to be called with
839 * down_read(&EXT4_I(inode)->i_data_sem) if not allocating file system block
840 * (ie, create is zero). Otherwise down_write(&EXT4_I(inode)->i_data_sem)
842 int ext4_get_blocks_handle(handle_t
*handle
, struct inode
*inode
,
843 ext4_lblk_t iblock
, unsigned long maxblocks
,
844 struct buffer_head
*bh_result
,
845 int create
, int extend_disksize
)
848 ext4_lblk_t offsets
[4];
853 int blocks_to_boundary
= 0;
855 struct ext4_inode_info
*ei
= EXT4_I(inode
);
857 ext4_fsblk_t first_block
= 0;
861 J_ASSERT(!(EXT4_I(inode
)->i_flags
& EXT4_EXTENTS_FL
));
862 J_ASSERT(handle
!= NULL
|| create
== 0);
863 depth
= ext4_block_to_path(inode
, iblock
, offsets
,
864 &blocks_to_boundary
);
869 partial
= ext4_get_branch(inode
, depth
, offsets
, chain
, &err
);
871 /* Simplest case - block found, no allocation needed */
873 first_block
= le32_to_cpu(chain
[depth
- 1].key
);
874 clear_buffer_new(bh_result
);
877 while (count
< maxblocks
&& count
<= blocks_to_boundary
) {
880 blk
= le32_to_cpu(*(chain
[depth
-1].p
+ count
));
882 if (blk
== first_block
+ count
)
890 /* Next simple case - plain lookup or failed read of indirect block */
891 if (!create
|| err
== -EIO
)
895 * Okay, we need to do block allocation.
897 goal
= ext4_find_goal(inode
, iblock
, partial
);
899 /* the number of blocks need to allocate for [d,t]indirect blocks */
900 indirect_blks
= (chain
+ depth
) - partial
- 1;
903 * Next look up the indirect map to count the totoal number of
904 * direct blocks to allocate for this branch.
906 count
= ext4_blks_to_allocate(partial
, indirect_blks
,
907 maxblocks
, blocks_to_boundary
);
909 * Block out ext4_truncate while we alter the tree
911 err
= ext4_alloc_branch(handle
, inode
, iblock
, indirect_blks
,
913 offsets
+ (partial
- chain
), partial
);
916 * The ext4_splice_branch call will free and forget any buffers
917 * on the new chain if there is a failure, but that risks using
918 * up transaction credits, especially for bitmaps where the
919 * credits cannot be returned. Can we handle this somehow? We
920 * may need to return -EAGAIN upwards in the worst case. --sct
923 err
= ext4_splice_branch(handle
, inode
, iblock
,
924 partial
, indirect_blks
, count
);
926 * i_disksize growing is protected by i_data_sem. Don't forget to
927 * protect it if you're about to implement concurrent
928 * ext4_get_block() -bzzz
930 if (!err
&& extend_disksize
) {
931 disksize
= ((loff_t
) iblock
+ count
) << inode
->i_blkbits
;
932 if (disksize
> i_size_read(inode
))
933 disksize
= i_size_read(inode
);
934 if (disksize
> ei
->i_disksize
)
935 ei
->i_disksize
= disksize
;
940 set_buffer_new(bh_result
);
942 map_bh(bh_result
, inode
->i_sb
, le32_to_cpu(chain
[depth
-1].key
));
943 if (count
> blocks_to_boundary
)
944 set_buffer_boundary(bh_result
);
946 /* Clean up and exit */
947 partial
= chain
+ depth
- 1; /* the whole chain */
949 while (partial
> chain
) {
950 BUFFER_TRACE(partial
->bh
, "call brelse");
954 BUFFER_TRACE(bh_result
, "returned");
960 * Calculate the number of metadata blocks need to reserve
961 * to allocate @blocks for non extent file based file
963 static int ext4_indirect_calc_metadata_amount(struct inode
*inode
, int blocks
)
965 int icap
= EXT4_ADDR_PER_BLOCK(inode
->i_sb
);
966 int ind_blks
, dind_blks
, tind_blks
;
968 /* number of new indirect blocks needed */
969 ind_blks
= (blocks
+ icap
- 1) / icap
;
971 dind_blks
= (ind_blks
+ icap
- 1) / icap
;
975 return ind_blks
+ dind_blks
+ tind_blks
;
979 * Calculate the number of metadata blocks need to reserve
980 * to allocate given number of blocks
982 static int ext4_calc_metadata_amount(struct inode
*inode
, int blocks
)
987 if (EXT4_I(inode
)->i_flags
& EXT4_EXTENTS_FL
)
988 return ext4_ext_calc_metadata_amount(inode
, blocks
);
990 return ext4_indirect_calc_metadata_amount(inode
, blocks
);
993 static void ext4_da_update_reserve_space(struct inode
*inode
, int used
)
995 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
996 int total
, mdb
, mdb_free
;
998 spin_lock(&EXT4_I(inode
)->i_block_reservation_lock
);
999 /* recalculate the number of metablocks still need to be reserved */
1000 total
= EXT4_I(inode
)->i_reserved_data_blocks
- used
;
1001 mdb
= ext4_calc_metadata_amount(inode
, total
);
1003 /* figure out how many metablocks to release */
1004 BUG_ON(mdb
> EXT4_I(inode
)->i_reserved_meta_blocks
);
1005 mdb_free
= EXT4_I(inode
)->i_reserved_meta_blocks
- mdb
;
1008 /* Account for allocated meta_blocks */
1009 mdb_free
-= EXT4_I(inode
)->i_allocated_meta_blocks
;
1011 /* update fs dirty blocks counter */
1012 percpu_counter_sub(&sbi
->s_dirtyblocks_counter
, mdb_free
);
1013 EXT4_I(inode
)->i_allocated_meta_blocks
= 0;
1014 EXT4_I(inode
)->i_reserved_meta_blocks
= mdb
;
1017 /* update per-inode reservations */
1018 BUG_ON(used
> EXT4_I(inode
)->i_reserved_data_blocks
);
1019 EXT4_I(inode
)->i_reserved_data_blocks
-= used
;
1021 spin_unlock(&EXT4_I(inode
)->i_block_reservation_lock
);
1025 * The ext4_get_blocks_wrap() function try to look up the requested blocks,
1026 * and returns if the blocks are already mapped.
1028 * Otherwise it takes the write lock of the i_data_sem and allocate blocks
1029 * and store the allocated blocks in the result buffer head and mark it
1032 * If file type is extents based, it will call ext4_ext_get_blocks(),
1033 * Otherwise, call with ext4_get_blocks_handle() to handle indirect mapping
1036 * On success, it returns the number of blocks being mapped or allocate.
1037 * if create==0 and the blocks are pre-allocated and uninitialized block,
1038 * the result buffer head is unmapped. If the create ==1, it will make sure
1039 * the buffer head is mapped.
1041 * It returns 0 if plain look up failed (blocks have not been allocated), in
1042 * that casem, buffer head is unmapped
1044 * It returns the error in case of allocation failure.
1046 int ext4_get_blocks_wrap(handle_t
*handle
, struct inode
*inode
, sector_t block
,
1047 unsigned long max_blocks
, struct buffer_head
*bh
,
1048 int create
, int extend_disksize
, int flag
)
1052 clear_buffer_mapped(bh
);
1055 * Try to see if we can get the block without requesting
1056 * for new file system block.
1058 down_read((&EXT4_I(inode
)->i_data_sem
));
1059 if (EXT4_I(inode
)->i_flags
& EXT4_EXTENTS_FL
) {
1060 retval
= ext4_ext_get_blocks(handle
, inode
, block
, max_blocks
,
1063 retval
= ext4_get_blocks_handle(handle
,
1064 inode
, block
, max_blocks
, bh
, 0, 0);
1066 up_read((&EXT4_I(inode
)->i_data_sem
));
1068 /* If it is only a block(s) look up */
1073 * Returns if the blocks have already allocated
1075 * Note that if blocks have been preallocated
1076 * ext4_ext_get_block() returns th create = 0
1077 * with buffer head unmapped.
1079 if (retval
> 0 && buffer_mapped(bh
))
1083 * New blocks allocate and/or writing to uninitialized extent
1084 * will possibly result in updating i_data, so we take
1085 * the write lock of i_data_sem, and call get_blocks()
1086 * with create == 1 flag.
1088 down_write((&EXT4_I(inode
)->i_data_sem
));
1091 * if the caller is from delayed allocation writeout path
1092 * we have already reserved fs blocks for allocation
1093 * let the underlying get_block() function know to
1094 * avoid double accounting
1097 EXT4_I(inode
)->i_delalloc_reserved_flag
= 1;
1099 * We need to check for EXT4 here because migrate
1100 * could have changed the inode type in between
1102 if (EXT4_I(inode
)->i_flags
& EXT4_EXTENTS_FL
) {
1103 retval
= ext4_ext_get_blocks(handle
, inode
, block
, max_blocks
,
1104 bh
, create
, extend_disksize
);
1106 retval
= ext4_get_blocks_handle(handle
, inode
, block
,
1107 max_blocks
, bh
, create
, extend_disksize
);
1109 if (retval
> 0 && buffer_new(bh
)) {
1111 * We allocated new blocks which will result in
1112 * i_data's format changing. Force the migrate
1113 * to fail by clearing migrate flags
1115 EXT4_I(inode
)->i_flags
= EXT4_I(inode
)->i_flags
&
1121 EXT4_I(inode
)->i_delalloc_reserved_flag
= 0;
1123 * Update reserved blocks/metadata blocks
1124 * after successful block allocation
1125 * which were deferred till now
1127 if ((retval
> 0) && buffer_delay(bh
))
1128 ext4_da_update_reserve_space(inode
, retval
);
1131 up_write((&EXT4_I(inode
)->i_data_sem
));
1135 /* Maximum number of blocks we map for direct IO at once. */
1136 #define DIO_MAX_BLOCKS 4096
1138 int ext4_get_block(struct inode
*inode
, sector_t iblock
,
1139 struct buffer_head
*bh_result
, int create
)
1141 handle_t
*handle
= ext4_journal_current_handle();
1142 int ret
= 0, started
= 0;
1143 unsigned max_blocks
= bh_result
->b_size
>> inode
->i_blkbits
;
1146 if (create
&& !handle
) {
1147 /* Direct IO write... */
1148 if (max_blocks
> DIO_MAX_BLOCKS
)
1149 max_blocks
= DIO_MAX_BLOCKS
;
1150 dio_credits
= ext4_chunk_trans_blocks(inode
, max_blocks
);
1151 handle
= ext4_journal_start(inode
, dio_credits
);
1152 if (IS_ERR(handle
)) {
1153 ret
= PTR_ERR(handle
);
1159 ret
= ext4_get_blocks_wrap(handle
, inode
, iblock
,
1160 max_blocks
, bh_result
, create
, 0, 0);
1162 bh_result
->b_size
= (ret
<< inode
->i_blkbits
);
1166 ext4_journal_stop(handle
);
1172 * `handle' can be NULL if create is zero
1174 struct buffer_head
*ext4_getblk(handle_t
*handle
, struct inode
*inode
,
1175 ext4_lblk_t block
, int create
, int *errp
)
1177 struct buffer_head dummy
;
1180 J_ASSERT(handle
!= NULL
|| create
== 0);
1183 dummy
.b_blocknr
= -1000;
1184 buffer_trace_init(&dummy
.b_history
);
1185 err
= ext4_get_blocks_wrap(handle
, inode
, block
, 1,
1186 &dummy
, create
, 1, 0);
1188 * ext4_get_blocks_handle() returns number of blocks
1189 * mapped. 0 in case of a HOLE.
1197 if (!err
&& buffer_mapped(&dummy
)) {
1198 struct buffer_head
*bh
;
1199 bh
= sb_getblk(inode
->i_sb
, dummy
.b_blocknr
);
1204 if (buffer_new(&dummy
)) {
1205 J_ASSERT(create
!= 0);
1206 J_ASSERT(handle
!= NULL
);
1209 * Now that we do not always journal data, we should
1210 * keep in mind whether this should always journal the
1211 * new buffer as metadata. For now, regular file
1212 * writes use ext4_get_block instead, so it's not a
1216 BUFFER_TRACE(bh
, "call get_create_access");
1217 fatal
= ext4_journal_get_create_access(handle
, bh
);
1218 if (!fatal
&& !buffer_uptodate(bh
)) {
1219 memset(bh
->b_data
, 0, inode
->i_sb
->s_blocksize
);
1220 set_buffer_uptodate(bh
);
1223 BUFFER_TRACE(bh
, "call ext4_journal_dirty_metadata");
1224 err
= ext4_journal_dirty_metadata(handle
, bh
);
1228 BUFFER_TRACE(bh
, "not a new buffer");
1241 struct buffer_head
*ext4_bread(handle_t
*handle
, struct inode
*inode
,
1242 ext4_lblk_t block
, int create
, int *err
)
1244 struct buffer_head
*bh
;
1246 bh
= ext4_getblk(handle
, inode
, block
, create
, err
);
1249 if (buffer_uptodate(bh
))
1251 ll_rw_block(READ_META
, 1, &bh
);
1253 if (buffer_uptodate(bh
))
1260 static int walk_page_buffers(handle_t
*handle
,
1261 struct buffer_head
*head
,
1265 int (*fn
)(handle_t
*handle
,
1266 struct buffer_head
*bh
))
1268 struct buffer_head
*bh
;
1269 unsigned block_start
, block_end
;
1270 unsigned blocksize
= head
->b_size
;
1272 struct buffer_head
*next
;
1274 for (bh
= head
, block_start
= 0;
1275 ret
== 0 && (bh
!= head
|| !block_start
);
1276 block_start
= block_end
, bh
= next
)
1278 next
= bh
->b_this_page
;
1279 block_end
= block_start
+ blocksize
;
1280 if (block_end
<= from
|| block_start
>= to
) {
1281 if (partial
&& !buffer_uptodate(bh
))
1285 err
= (*fn
)(handle
, bh
);
1293 * To preserve ordering, it is essential that the hole instantiation and
1294 * the data write be encapsulated in a single transaction. We cannot
1295 * close off a transaction and start a new one between the ext4_get_block()
1296 * and the commit_write(). So doing the jbd2_journal_start at the start of
1297 * prepare_write() is the right place.
1299 * Also, this function can nest inside ext4_writepage() ->
1300 * block_write_full_page(). In that case, we *know* that ext4_writepage()
1301 * has generated enough buffer credits to do the whole page. So we won't
1302 * block on the journal in that case, which is good, because the caller may
1305 * By accident, ext4 can be reentered when a transaction is open via
1306 * quota file writes. If we were to commit the transaction while thus
1307 * reentered, there can be a deadlock - we would be holding a quota
1308 * lock, and the commit would never complete if another thread had a
1309 * transaction open and was blocking on the quota lock - a ranking
1312 * So what we do is to rely on the fact that jbd2_journal_stop/journal_start
1313 * will _not_ run commit under these circumstances because handle->h_ref
1314 * is elevated. We'll still have enough credits for the tiny quotafile
1317 static int do_journal_get_write_access(handle_t
*handle
,
1318 struct buffer_head
*bh
)
1320 if (!buffer_mapped(bh
) || buffer_freed(bh
))
1322 return ext4_journal_get_write_access(handle
, bh
);
1325 static int ext4_write_begin(struct file
*file
, struct address_space
*mapping
,
1326 loff_t pos
, unsigned len
, unsigned flags
,
1327 struct page
**pagep
, void **fsdata
)
1329 struct inode
*inode
= mapping
->host
;
1330 int ret
, needed_blocks
= ext4_writepage_trans_blocks(inode
);
1337 index
= pos
>> PAGE_CACHE_SHIFT
;
1338 from
= pos
& (PAGE_CACHE_SIZE
- 1);
1342 handle
= ext4_journal_start(inode
, needed_blocks
);
1343 if (IS_ERR(handle
)) {
1344 ret
= PTR_ERR(handle
);
1348 page
= __grab_cache_page(mapping
, index
);
1350 ext4_journal_stop(handle
);
1356 ret
= block_write_begin(file
, mapping
, pos
, len
, flags
, pagep
, fsdata
,
1359 if (!ret
&& ext4_should_journal_data(inode
)) {
1360 ret
= walk_page_buffers(handle
, page_buffers(page
),
1361 from
, to
, NULL
, do_journal_get_write_access
);
1366 ext4_journal_stop(handle
);
1367 page_cache_release(page
);
1369 * block_write_begin may have instantiated a few blocks
1370 * outside i_size. Trim these off again. Don't need
1371 * i_size_read because we hold i_mutex.
1373 if (pos
+ len
> inode
->i_size
)
1374 vmtruncate(inode
, inode
->i_size
);
1377 if (ret
== -ENOSPC
&& ext4_should_retry_alloc(inode
->i_sb
, &retries
))
1383 /* For write_end() in data=journal mode */
1384 static int write_end_fn(handle_t
*handle
, struct buffer_head
*bh
)
1386 if (!buffer_mapped(bh
) || buffer_freed(bh
))
1388 set_buffer_uptodate(bh
);
1389 return ext4_journal_dirty_metadata(handle
, bh
);
1393 * We need to pick up the new inode size which generic_commit_write gave us
1394 * `file' can be NULL - eg, when called from page_symlink().
1396 * ext4 never places buffers on inode->i_mapping->private_list. metadata
1397 * buffers are managed internally.
1399 static int ext4_ordered_write_end(struct file
*file
,
1400 struct address_space
*mapping
,
1401 loff_t pos
, unsigned len
, unsigned copied
,
1402 struct page
*page
, void *fsdata
)
1404 handle_t
*handle
= ext4_journal_current_handle();
1405 struct inode
*inode
= mapping
->host
;
1408 ret
= ext4_jbd2_file_inode(handle
, inode
);
1413 new_i_size
= pos
+ copied
;
1414 if (new_i_size
> EXT4_I(inode
)->i_disksize
) {
1415 ext4_update_i_disksize(inode
, new_i_size
);
1416 /* We need to mark inode dirty even if
1417 * new_i_size is less that inode->i_size
1418 * bu greater than i_disksize.(hint delalloc)
1420 ext4_mark_inode_dirty(handle
, inode
);
1423 ret2
= generic_write_end(file
, mapping
, pos
, len
, copied
,
1429 ret2
= ext4_journal_stop(handle
);
1433 return ret
? ret
: copied
;
1436 static int ext4_writeback_write_end(struct file
*file
,
1437 struct address_space
*mapping
,
1438 loff_t pos
, unsigned len
, unsigned copied
,
1439 struct page
*page
, void *fsdata
)
1441 handle_t
*handle
= ext4_journal_current_handle();
1442 struct inode
*inode
= mapping
->host
;
1446 new_i_size
= pos
+ copied
;
1447 if (new_i_size
> EXT4_I(inode
)->i_disksize
) {
1448 ext4_update_i_disksize(inode
, new_i_size
);
1449 /* We need to mark inode dirty even if
1450 * new_i_size is less that inode->i_size
1451 * bu greater than i_disksize.(hint delalloc)
1453 ext4_mark_inode_dirty(handle
, inode
);
1456 ret2
= generic_write_end(file
, mapping
, pos
, len
, copied
,
1462 ret2
= ext4_journal_stop(handle
);
1466 return ret
? ret
: copied
;
1469 static int ext4_journalled_write_end(struct file
*file
,
1470 struct address_space
*mapping
,
1471 loff_t pos
, unsigned len
, unsigned copied
,
1472 struct page
*page
, void *fsdata
)
1474 handle_t
*handle
= ext4_journal_current_handle();
1475 struct inode
*inode
= mapping
->host
;
1481 from
= pos
& (PAGE_CACHE_SIZE
- 1);
1485 if (!PageUptodate(page
))
1487 page_zero_new_buffers(page
, from
+copied
, to
);
1490 ret
= walk_page_buffers(handle
, page_buffers(page
), from
,
1491 to
, &partial
, write_end_fn
);
1493 SetPageUptodate(page
);
1494 new_i_size
= pos
+ copied
;
1495 if (new_i_size
> inode
->i_size
)
1496 i_size_write(inode
, pos
+copied
);
1497 EXT4_I(inode
)->i_state
|= EXT4_STATE_JDATA
;
1498 if (new_i_size
> EXT4_I(inode
)->i_disksize
) {
1499 ext4_update_i_disksize(inode
, new_i_size
);
1500 ret2
= ext4_mark_inode_dirty(handle
, inode
);
1506 ret2
= ext4_journal_stop(handle
);
1509 page_cache_release(page
);
1511 return ret
? ret
: copied
;
1514 static int ext4_da_reserve_space(struct inode
*inode
, int nrblocks
)
1517 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
1518 unsigned long md_needed
, mdblocks
, total
= 0;
1521 * recalculate the amount of metadata blocks to reserve
1522 * in order to allocate nrblocks
1523 * worse case is one extent per block
1526 spin_lock(&EXT4_I(inode
)->i_block_reservation_lock
);
1527 total
= EXT4_I(inode
)->i_reserved_data_blocks
+ nrblocks
;
1528 mdblocks
= ext4_calc_metadata_amount(inode
, total
);
1529 BUG_ON(mdblocks
< EXT4_I(inode
)->i_reserved_meta_blocks
);
1531 md_needed
= mdblocks
- EXT4_I(inode
)->i_reserved_meta_blocks
;
1532 total
= md_needed
+ nrblocks
;
1534 if (ext4_claim_free_blocks(sbi
, total
)) {
1535 spin_unlock(&EXT4_I(inode
)->i_block_reservation_lock
);
1536 if (ext4_should_retry_alloc(inode
->i_sb
, &retries
)) {
1542 EXT4_I(inode
)->i_reserved_data_blocks
+= nrblocks
;
1543 EXT4_I(inode
)->i_reserved_meta_blocks
= mdblocks
;
1545 spin_unlock(&EXT4_I(inode
)->i_block_reservation_lock
);
1546 return 0; /* success */
1549 static void ext4_da_release_space(struct inode
*inode
, int to_free
)
1551 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
1552 int total
, mdb
, mdb_free
, release
;
1555 return; /* Nothing to release, exit */
1557 spin_lock(&EXT4_I(inode
)->i_block_reservation_lock
);
1559 if (!EXT4_I(inode
)->i_reserved_data_blocks
) {
1561 * if there is no reserved blocks, but we try to free some
1562 * then the counter is messed up somewhere.
1563 * but since this function is called from invalidate
1564 * page, it's harmless to return without any action
1566 printk(KERN_INFO
"ext4 delalloc try to release %d reserved "
1567 "blocks for inode %lu, but there is no reserved "
1568 "data blocks\n", to_free
, inode
->i_ino
);
1569 spin_unlock(&EXT4_I(inode
)->i_block_reservation_lock
);
1573 /* recalculate the number of metablocks still need to be reserved */
1574 total
= EXT4_I(inode
)->i_reserved_data_blocks
- to_free
;
1575 mdb
= ext4_calc_metadata_amount(inode
, total
);
1577 /* figure out how many metablocks to release */
1578 BUG_ON(mdb
> EXT4_I(inode
)->i_reserved_meta_blocks
);
1579 mdb_free
= EXT4_I(inode
)->i_reserved_meta_blocks
- mdb
;
1581 release
= to_free
+ mdb_free
;
1583 /* update fs dirty blocks counter for truncate case */
1584 percpu_counter_sub(&sbi
->s_dirtyblocks_counter
, release
);
1586 /* update per-inode reservations */
1587 BUG_ON(to_free
> EXT4_I(inode
)->i_reserved_data_blocks
);
1588 EXT4_I(inode
)->i_reserved_data_blocks
-= to_free
;
1590 BUG_ON(mdb
> EXT4_I(inode
)->i_reserved_meta_blocks
);
1591 EXT4_I(inode
)->i_reserved_meta_blocks
= mdb
;
1592 spin_unlock(&EXT4_I(inode
)->i_block_reservation_lock
);
1595 static void ext4_da_page_release_reservation(struct page
*page
,
1596 unsigned long offset
)
1599 struct buffer_head
*head
, *bh
;
1600 unsigned int curr_off
= 0;
1602 head
= page_buffers(page
);
1605 unsigned int next_off
= curr_off
+ bh
->b_size
;
1607 if ((offset
<= curr_off
) && (buffer_delay(bh
))) {
1609 clear_buffer_delay(bh
);
1611 curr_off
= next_off
;
1612 } while ((bh
= bh
->b_this_page
) != head
);
1613 ext4_da_release_space(page
->mapping
->host
, to_release
);
1617 * Delayed allocation stuff
1620 struct mpage_da_data
{
1621 struct inode
*inode
;
1622 struct buffer_head lbh
; /* extent of blocks */
1623 unsigned long first_page
, next_page
; /* extent of pages */
1624 get_block_t
*get_block
;
1625 struct writeback_control
*wbc
;
1632 * mpage_da_submit_io - walks through extent of pages and try to write
1633 * them with writepage() call back
1635 * @mpd->inode: inode
1636 * @mpd->first_page: first page of the extent
1637 * @mpd->next_page: page after the last page of the extent
1638 * @mpd->get_block: the filesystem's block mapper function
1640 * By the time mpage_da_submit_io() is called we expect all blocks
1641 * to be allocated. this may be wrong if allocation failed.
1643 * As pages are already locked by write_cache_pages(), we can't use it
1645 static int mpage_da_submit_io(struct mpage_da_data
*mpd
)
1647 struct address_space
*mapping
= mpd
->inode
->i_mapping
;
1648 int ret
= 0, err
, nr_pages
, i
;
1649 unsigned long index
, end
;
1650 struct pagevec pvec
;
1652 BUG_ON(mpd
->next_page
<= mpd
->first_page
);
1653 pagevec_init(&pvec
, 0);
1654 index
= mpd
->first_page
;
1655 end
= mpd
->next_page
- 1;
1657 while (index
<= end
) {
1658 /* XXX: optimize tail */
1660 * We can use PAGECACHE_TAG_DIRTY lookup here because
1661 * even though we have cleared the dirty flag on the page
1662 * We still keep the page in the radix tree with tag
1663 * PAGECACHE_TAG_DIRTY. See clear_page_dirty_for_io.
1664 * The PAGECACHE_TAG_DIRTY is cleared in set_page_writeback
1665 * which is called via the below writepage callback.
1667 nr_pages
= pagevec_lookup_tag(&pvec
, mapping
, &index
,
1668 PAGECACHE_TAG_DIRTY
,
1670 (pgoff_t
)PAGEVEC_SIZE
-1) + 1);
1673 for (i
= 0; i
< nr_pages
; i
++) {
1674 struct page
*page
= pvec
.pages
[i
];
1676 err
= mapping
->a_ops
->writepage(page
, mpd
->wbc
);
1678 mpd
->pages_written
++;
1680 * In error case, we have to continue because
1681 * remaining pages are still locked
1682 * XXX: unlock and re-dirty them?
1687 pagevec_release(&pvec
);
1693 * mpage_put_bnr_to_bhs - walk blocks and assign them actual numbers
1695 * @mpd->inode - inode to walk through
1696 * @exbh->b_blocknr - first block on a disk
1697 * @exbh->b_size - amount of space in bytes
1698 * @logical - first logical block to start assignment with
1700 * the function goes through all passed space and put actual disk
1701 * block numbers into buffer heads, dropping BH_Delay
1703 static void mpage_put_bnr_to_bhs(struct mpage_da_data
*mpd
, sector_t logical
,
1704 struct buffer_head
*exbh
)
1706 struct inode
*inode
= mpd
->inode
;
1707 struct address_space
*mapping
= inode
->i_mapping
;
1708 int blocks
= exbh
->b_size
>> inode
->i_blkbits
;
1709 sector_t pblock
= exbh
->b_blocknr
, cur_logical
;
1710 struct buffer_head
*head
, *bh
;
1712 struct pagevec pvec
;
1715 index
= logical
>> (PAGE_CACHE_SHIFT
- inode
->i_blkbits
);
1716 end
= (logical
+ blocks
- 1) >> (PAGE_CACHE_SHIFT
- inode
->i_blkbits
);
1717 cur_logical
= index
<< (PAGE_CACHE_SHIFT
- inode
->i_blkbits
);
1719 pagevec_init(&pvec
, 0);
1721 while (index
<= end
) {
1722 /* XXX: optimize tail */
1723 nr_pages
= pagevec_lookup(&pvec
, mapping
, index
, PAGEVEC_SIZE
);
1726 for (i
= 0; i
< nr_pages
; i
++) {
1727 struct page
*page
= pvec
.pages
[i
];
1729 index
= page
->index
;
1734 BUG_ON(!PageLocked(page
));
1735 BUG_ON(PageWriteback(page
));
1736 BUG_ON(!page_has_buffers(page
));
1738 bh
= page_buffers(page
);
1741 /* skip blocks out of the range */
1743 if (cur_logical
>= logical
)
1746 } while ((bh
= bh
->b_this_page
) != head
);
1749 if (cur_logical
>= logical
+ blocks
)
1751 if (buffer_delay(bh
)) {
1752 bh
->b_blocknr
= pblock
;
1753 clear_buffer_delay(bh
);
1754 bh
->b_bdev
= inode
->i_sb
->s_bdev
;
1755 } else if (buffer_unwritten(bh
)) {
1756 bh
->b_blocknr
= pblock
;
1757 clear_buffer_unwritten(bh
);
1758 set_buffer_mapped(bh
);
1760 bh
->b_bdev
= inode
->i_sb
->s_bdev
;
1761 } else if (buffer_mapped(bh
))
1762 BUG_ON(bh
->b_blocknr
!= pblock
);
1766 } while ((bh
= bh
->b_this_page
) != head
);
1768 pagevec_release(&pvec
);
1774 * __unmap_underlying_blocks - just a helper function to unmap
1775 * set of blocks described by @bh
1777 static inline void __unmap_underlying_blocks(struct inode
*inode
,
1778 struct buffer_head
*bh
)
1780 struct block_device
*bdev
= inode
->i_sb
->s_bdev
;
1783 blocks
= bh
->b_size
>> inode
->i_blkbits
;
1784 for (i
= 0; i
< blocks
; i
++)
1785 unmap_underlying_metadata(bdev
, bh
->b_blocknr
+ i
);
1788 static void ext4_da_block_invalidatepages(struct mpage_da_data
*mpd
,
1789 sector_t logical
, long blk_cnt
)
1793 struct pagevec pvec
;
1794 struct inode
*inode
= mpd
->inode
;
1795 struct address_space
*mapping
= inode
->i_mapping
;
1797 index
= logical
>> (PAGE_CACHE_SHIFT
- inode
->i_blkbits
);
1798 end
= (logical
+ blk_cnt
- 1) >>
1799 (PAGE_CACHE_SHIFT
- inode
->i_blkbits
);
1800 while (index
<= end
) {
1801 nr_pages
= pagevec_lookup(&pvec
, mapping
, index
, PAGEVEC_SIZE
);
1804 for (i
= 0; i
< nr_pages
; i
++) {
1805 struct page
*page
= pvec
.pages
[i
];
1806 index
= page
->index
;
1811 BUG_ON(!PageLocked(page
));
1812 BUG_ON(PageWriteback(page
));
1813 block_invalidatepage(page
, 0);
1814 ClearPageUptodate(page
);
1821 static void ext4_print_free_blocks(struct inode
*inode
)
1823 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
1824 printk(KERN_EMERG
"Total free blocks count %lld\n",
1825 ext4_count_free_blocks(inode
->i_sb
));
1826 printk(KERN_EMERG
"Free/Dirty block details\n");
1827 printk(KERN_EMERG
"free_blocks=%lld\n",
1828 percpu_counter_sum(&sbi
->s_freeblocks_counter
));
1829 printk(KERN_EMERG
"dirty_blocks=%lld\n",
1830 percpu_counter_sum(&sbi
->s_dirtyblocks_counter
));
1831 printk(KERN_EMERG
"Block reservation details\n");
1832 printk(KERN_EMERG
"i_reserved_data_blocks=%lu\n",
1833 EXT4_I(inode
)->i_reserved_data_blocks
);
1834 printk(KERN_EMERG
"i_reserved_meta_blocks=%lu\n",
1835 EXT4_I(inode
)->i_reserved_meta_blocks
);
1840 * mpage_da_map_blocks - go through given space
1842 * @mpd->lbh - bh describing space
1843 * @mpd->get_block - the filesystem's block mapper function
1845 * The function skips space we know is already mapped to disk blocks.
1848 static int mpage_da_map_blocks(struct mpage_da_data
*mpd
)
1851 struct buffer_head
new;
1852 struct buffer_head
*lbh
= &mpd
->lbh
;
1856 * We consider only non-mapped and non-allocated blocks
1858 if (buffer_mapped(lbh
) && !buffer_delay(lbh
))
1860 new.b_state
= lbh
->b_state
;
1862 new.b_size
= lbh
->b_size
;
1863 next
= lbh
->b_blocknr
;
1865 * If we didn't accumulate anything
1866 * to write simply return
1870 err
= mpd
->get_block(mpd
->inode
, next
, &new, 1);
1873 /* If get block returns with error
1874 * we simply return. Later writepage
1875 * will redirty the page and writepages
1876 * will find the dirty page again
1881 if (err
== -ENOSPC
&&
1882 ext4_count_free_blocks(mpd
->inode
->i_sb
)) {
1888 * get block failure will cause us
1889 * to loop in writepages. Because
1890 * a_ops->writepage won't be able to
1891 * make progress. The page will be redirtied
1892 * by writepage and writepages will again
1893 * try to write the same.
1895 printk(KERN_EMERG
"%s block allocation failed for inode %lu "
1896 "at logical offset %llu with max blocks "
1897 "%zd with error %d\n",
1898 __func__
, mpd
->inode
->i_ino
,
1899 (unsigned long long)next
,
1900 lbh
->b_size
>> mpd
->inode
->i_blkbits
, err
);
1901 printk(KERN_EMERG
"This should not happen.!! "
1902 "Data will be lost\n");
1903 if (err
== -ENOSPC
) {
1904 ext4_print_free_blocks(mpd
->inode
);
1906 /* invlaidate all the pages */
1907 ext4_da_block_invalidatepages(mpd
, next
,
1908 lbh
->b_size
>> mpd
->inode
->i_blkbits
);
1911 BUG_ON(new.b_size
== 0);
1913 if (buffer_new(&new))
1914 __unmap_underlying_blocks(mpd
->inode
, &new);
1917 * If blocks are delayed marked, we need to
1918 * put actual blocknr and drop delayed bit
1920 if (buffer_delay(lbh
) || buffer_unwritten(lbh
))
1921 mpage_put_bnr_to_bhs(mpd
, next
, &new);
1926 #define BH_FLAGS ((1 << BH_Uptodate) | (1 << BH_Mapped) | \
1927 (1 << BH_Delay) | (1 << BH_Unwritten))
1930 * mpage_add_bh_to_extent - try to add one more block to extent of blocks
1932 * @mpd->lbh - extent of blocks
1933 * @logical - logical number of the block in the file
1934 * @bh - bh of the block (used to access block's state)
1936 * the function is used to collect contig. blocks in same state
1938 static void mpage_add_bh_to_extent(struct mpage_da_data
*mpd
,
1939 sector_t logical
, struct buffer_head
*bh
)
1942 size_t b_size
= bh
->b_size
;
1943 struct buffer_head
*lbh
= &mpd
->lbh
;
1944 int nrblocks
= lbh
->b_size
>> mpd
->inode
->i_blkbits
;
1946 /* check if thereserved journal credits might overflow */
1947 if (!(EXT4_I(mpd
->inode
)->i_flags
& EXT4_EXTENTS_FL
)) {
1948 if (nrblocks
>= EXT4_MAX_TRANS_DATA
) {
1950 * With non-extent format we are limited by the journal
1951 * credit available. Total credit needed to insert
1952 * nrblocks contiguous blocks is dependent on the
1953 * nrblocks. So limit nrblocks.
1956 } else if ((nrblocks
+ (b_size
>> mpd
->inode
->i_blkbits
)) >
1957 EXT4_MAX_TRANS_DATA
) {
1959 * Adding the new buffer_head would make it cross the
1960 * allowed limit for which we have journal credit
1961 * reserved. So limit the new bh->b_size
1963 b_size
= (EXT4_MAX_TRANS_DATA
- nrblocks
) <<
1964 mpd
->inode
->i_blkbits
;
1965 /* we will do mpage_da_submit_io in the next loop */
1969 * First block in the extent
1971 if (lbh
->b_size
== 0) {
1972 lbh
->b_blocknr
= logical
;
1973 lbh
->b_size
= b_size
;
1974 lbh
->b_state
= bh
->b_state
& BH_FLAGS
;
1978 next
= lbh
->b_blocknr
+ nrblocks
;
1980 * Can we merge the block to our big extent?
1982 if (logical
== next
&& (bh
->b_state
& BH_FLAGS
) == lbh
->b_state
) {
1983 lbh
->b_size
+= b_size
;
1989 * We couldn't merge the block to our extent, so we
1990 * need to flush current extent and start new one
1992 if (mpage_da_map_blocks(mpd
) == 0)
1993 mpage_da_submit_io(mpd
);
1999 * __mpage_da_writepage - finds extent of pages and blocks
2001 * @page: page to consider
2002 * @wbc: not used, we just follow rules
2005 * The function finds extents of pages and scan them for all blocks.
2007 static int __mpage_da_writepage(struct page
*page
,
2008 struct writeback_control
*wbc
, void *data
)
2010 struct mpage_da_data
*mpd
= data
;
2011 struct inode
*inode
= mpd
->inode
;
2012 struct buffer_head
*bh
, *head
, fake
;
2017 * Rest of the page in the page_vec
2018 * redirty then and skip then. We will
2019 * try to to write them again after
2020 * starting a new transaction
2022 redirty_page_for_writepage(wbc
, page
);
2024 return MPAGE_DA_EXTENT_TAIL
;
2027 * Can we merge this page to current extent?
2029 if (mpd
->next_page
!= page
->index
) {
2031 * Nope, we can't. So, we map non-allocated blocks
2032 * and start IO on them using writepage()
2034 if (mpd
->next_page
!= mpd
->first_page
) {
2035 if (mpage_da_map_blocks(mpd
) == 0)
2036 mpage_da_submit_io(mpd
);
2038 * skip rest of the page in the page_vec
2041 redirty_page_for_writepage(wbc
, page
);
2043 return MPAGE_DA_EXTENT_TAIL
;
2047 * Start next extent of pages ...
2049 mpd
->first_page
= page
->index
;
2054 mpd
->lbh
.b_size
= 0;
2055 mpd
->lbh
.b_state
= 0;
2056 mpd
->lbh
.b_blocknr
= 0;
2059 mpd
->next_page
= page
->index
+ 1;
2060 logical
= (sector_t
) page
->index
<<
2061 (PAGE_CACHE_SHIFT
- inode
->i_blkbits
);
2063 if (!page_has_buffers(page
)) {
2065 * There is no attached buffer heads yet (mmap?)
2066 * we treat the page asfull of dirty blocks
2069 bh
->b_size
= PAGE_CACHE_SIZE
;
2071 set_buffer_dirty(bh
);
2072 set_buffer_uptodate(bh
);
2073 mpage_add_bh_to_extent(mpd
, logical
, bh
);
2075 return MPAGE_DA_EXTENT_TAIL
;
2078 * Page with regular buffer heads, just add all dirty ones
2080 head
= page_buffers(page
);
2083 BUG_ON(buffer_locked(bh
));
2084 if (buffer_dirty(bh
) &&
2085 (!buffer_mapped(bh
) || buffer_delay(bh
))) {
2086 mpage_add_bh_to_extent(mpd
, logical
, bh
);
2088 return MPAGE_DA_EXTENT_TAIL
;
2091 } while ((bh
= bh
->b_this_page
) != head
);
2098 * mpage_da_writepages - walk the list of dirty pages of the given
2099 * address space, allocates non-allocated blocks, maps newly-allocated
2100 * blocks to existing bhs and issue IO them
2102 * @mapping: address space structure to write
2103 * @wbc: subtract the number of written pages from *@wbc->nr_to_write
2104 * @get_block: the filesystem's block mapper function.
2106 * This is a library function, which implements the writepages()
2107 * address_space_operation.
2109 static int mpage_da_writepages(struct address_space
*mapping
,
2110 struct writeback_control
*wbc
,
2111 struct mpage_da_data
*mpd
)
2116 if (!mpd
->get_block
)
2117 return generic_writepages(mapping
, wbc
);
2119 mpd
->lbh
.b_size
= 0;
2120 mpd
->lbh
.b_state
= 0;
2121 mpd
->lbh
.b_blocknr
= 0;
2122 mpd
->first_page
= 0;
2125 mpd
->pages_written
= 0;
2128 to_write
= wbc
->nr_to_write
;
2130 ret
= write_cache_pages(mapping
, wbc
, __mpage_da_writepage
, mpd
);
2133 * Handle last extent of pages
2135 if (!mpd
->io_done
&& mpd
->next_page
!= mpd
->first_page
) {
2136 if (mpage_da_map_blocks(mpd
) == 0)
2137 mpage_da_submit_io(mpd
);
2140 wbc
->nr_to_write
= to_write
- mpd
->pages_written
;
2145 * this is a special callback for ->write_begin() only
2146 * it's intention is to return mapped block or reserve space
2148 static int ext4_da_get_block_prep(struct inode
*inode
, sector_t iblock
,
2149 struct buffer_head
*bh_result
, int create
)
2153 BUG_ON(create
== 0);
2154 BUG_ON(bh_result
->b_size
!= inode
->i_sb
->s_blocksize
);
2157 * first, we need to know whether the block is allocated already
2158 * preallocated blocks are unmapped but should treated
2159 * the same as allocated blocks.
2161 ret
= ext4_get_blocks_wrap(NULL
, inode
, iblock
, 1, bh_result
, 0, 0, 0);
2162 if ((ret
== 0) && !buffer_delay(bh_result
)) {
2163 /* the block isn't (pre)allocated yet, let's reserve space */
2165 * XXX: __block_prepare_write() unmaps passed block,
2168 ret
= ext4_da_reserve_space(inode
, 1);
2170 /* not enough space to reserve */
2173 map_bh(bh_result
, inode
->i_sb
, 0);
2174 set_buffer_new(bh_result
);
2175 set_buffer_delay(bh_result
);
2176 } else if (ret
> 0) {
2177 bh_result
->b_size
= (ret
<< inode
->i_blkbits
);
2183 #define EXT4_DELALLOC_RSVED 1
2184 static int ext4_da_get_block_write(struct inode
*inode
, sector_t iblock
,
2185 struct buffer_head
*bh_result
, int create
)
2188 unsigned max_blocks
= bh_result
->b_size
>> inode
->i_blkbits
;
2189 loff_t disksize
= EXT4_I(inode
)->i_disksize
;
2190 handle_t
*handle
= NULL
;
2192 handle
= ext4_journal_current_handle();
2194 ret
= ext4_get_blocks_wrap(handle
, inode
, iblock
, max_blocks
,
2195 bh_result
, create
, 0, EXT4_DELALLOC_RSVED
);
2198 bh_result
->b_size
= (ret
<< inode
->i_blkbits
);
2200 if (ext4_should_order_data(inode
)) {
2202 retval
= ext4_jbd2_file_inode(handle
, inode
);
2205 * Failed to add inode for ordered
2206 * mode. Don't update file size
2212 * Update on-disk size along with block allocation
2213 * we don't use 'extend_disksize' as size may change
2214 * within already allocated block -bzzz
2216 disksize
= ((loff_t
) iblock
+ ret
) << inode
->i_blkbits
;
2217 if (disksize
> i_size_read(inode
))
2218 disksize
= i_size_read(inode
);
2219 if (disksize
> EXT4_I(inode
)->i_disksize
) {
2220 ext4_update_i_disksize(inode
, disksize
);
2221 ret
= ext4_mark_inode_dirty(handle
, inode
);
2229 static int ext4_bh_unmapped_or_delay(handle_t
*handle
, struct buffer_head
*bh
)
2232 * unmapped buffer is possible for holes.
2233 * delay buffer is possible with delayed allocation
2235 return ((!buffer_mapped(bh
) || buffer_delay(bh
)) && buffer_dirty(bh
));
2238 static int ext4_normal_get_block_write(struct inode
*inode
, sector_t iblock
,
2239 struct buffer_head
*bh_result
, int create
)
2242 unsigned max_blocks
= bh_result
->b_size
>> inode
->i_blkbits
;
2245 * we don't want to do block allocation in writepage
2246 * so call get_block_wrap with create = 0
2248 ret
= ext4_get_blocks_wrap(NULL
, inode
, iblock
, max_blocks
,
2249 bh_result
, 0, 0, 0);
2251 bh_result
->b_size
= (ret
<< inode
->i_blkbits
);
2258 * get called vi ext4_da_writepages after taking page lock (have journal handle)
2259 * get called via journal_submit_inode_data_buffers (no journal handle)
2260 * get called via shrink_page_list via pdflush (no journal handle)
2261 * or grab_page_cache when doing write_begin (have journal handle)
2263 static int ext4_da_writepage(struct page
*page
,
2264 struct writeback_control
*wbc
)
2269 struct buffer_head
*page_bufs
;
2270 struct inode
*inode
= page
->mapping
->host
;
2272 size
= i_size_read(inode
);
2273 if (page
->index
== size
>> PAGE_CACHE_SHIFT
)
2274 len
= size
& ~PAGE_CACHE_MASK
;
2276 len
= PAGE_CACHE_SIZE
;
2278 if (page_has_buffers(page
)) {
2279 page_bufs
= page_buffers(page
);
2280 if (walk_page_buffers(NULL
, page_bufs
, 0, len
, NULL
,
2281 ext4_bh_unmapped_or_delay
)) {
2283 * We don't want to do block allocation
2284 * So redirty the page and return
2285 * We may reach here when we do a journal commit
2286 * via journal_submit_inode_data_buffers.
2287 * If we don't have mapping block we just ignore
2288 * them. We can also reach here via shrink_page_list
2290 redirty_page_for_writepage(wbc
, page
);
2296 * The test for page_has_buffers() is subtle:
2297 * We know the page is dirty but it lost buffers. That means
2298 * that at some moment in time after write_begin()/write_end()
2299 * has been called all buffers have been clean and thus they
2300 * must have been written at least once. So they are all
2301 * mapped and we can happily proceed with mapping them
2302 * and writing the page.
2304 * Try to initialize the buffer_heads and check whether
2305 * all are mapped and non delay. We don't want to
2306 * do block allocation here.
2308 ret
= block_prepare_write(page
, 0, PAGE_CACHE_SIZE
,
2309 ext4_normal_get_block_write
);
2311 page_bufs
= page_buffers(page
);
2312 /* check whether all are mapped and non delay */
2313 if (walk_page_buffers(NULL
, page_bufs
, 0, len
, NULL
,
2314 ext4_bh_unmapped_or_delay
)) {
2315 redirty_page_for_writepage(wbc
, page
);
2321 * We can't do block allocation here
2322 * so just redity the page and unlock
2325 redirty_page_for_writepage(wbc
, page
);
2331 if (test_opt(inode
->i_sb
, NOBH
) && ext4_should_writeback_data(inode
))
2332 ret
= nobh_writepage(page
, ext4_normal_get_block_write
, wbc
);
2334 ret
= block_write_full_page(page
,
2335 ext4_normal_get_block_write
,
2342 * This is called via ext4_da_writepages() to
2343 * calulate the total number of credits to reserve to fit
2344 * a single extent allocation into a single transaction,
2345 * ext4_da_writpeages() will loop calling this before
2346 * the block allocation.
2349 static int ext4_da_writepages_trans_blocks(struct inode
*inode
)
2351 int max_blocks
= EXT4_I(inode
)->i_reserved_data_blocks
;
2354 * With non-extent format the journal credit needed to
2355 * insert nrblocks contiguous block is dependent on
2356 * number of contiguous block. So we will limit
2357 * number of contiguous block to a sane value
2359 if (!(inode
->i_flags
& EXT4_EXTENTS_FL
) &&
2360 (max_blocks
> EXT4_MAX_TRANS_DATA
))
2361 max_blocks
= EXT4_MAX_TRANS_DATA
;
2363 return ext4_chunk_trans_blocks(inode
, max_blocks
);
2366 static int ext4_da_writepages(struct address_space
*mapping
,
2367 struct writeback_control
*wbc
)
2369 handle_t
*handle
= NULL
;
2370 struct mpage_da_data mpd
;
2371 struct inode
*inode
= mapping
->host
;
2372 int needed_blocks
, ret
= 0, nr_to_writebump
= 0;
2373 long to_write
, pages_skipped
= 0;
2374 struct ext4_sb_info
*sbi
= EXT4_SB(mapping
->host
->i_sb
);
2377 * No pages to write? This is mainly a kludge to avoid starting
2378 * a transaction for special inodes like journal inode on last iput()
2379 * because that could violate lock ordering on umount
2381 if (!mapping
->nrpages
|| !mapping_tagged(mapping
, PAGECACHE_TAG_DIRTY
))
2384 * Make sure nr_to_write is >= sbi->s_mb_stream_request
2385 * This make sure small files blocks are allocated in
2386 * single attempt. This ensure that small files
2387 * get less fragmented.
2389 if (wbc
->nr_to_write
< sbi
->s_mb_stream_request
) {
2390 nr_to_writebump
= sbi
->s_mb_stream_request
- wbc
->nr_to_write
;
2391 wbc
->nr_to_write
= sbi
->s_mb_stream_request
;
2395 pages_skipped
= wbc
->pages_skipped
;
2398 mpd
.inode
= mapping
->host
;
2401 to_write
= wbc
->nr_to_write
;
2402 while (!ret
&& to_write
> 0) {
2405 * we insert one extent at a time. So we need
2406 * credit needed for single extent allocation.
2407 * journalled mode is currently not supported
2410 BUG_ON(ext4_should_journal_data(inode
));
2411 needed_blocks
= ext4_da_writepages_trans_blocks(inode
);
2413 /* start a new transaction*/
2414 handle
= ext4_journal_start(inode
, needed_blocks
);
2415 if (IS_ERR(handle
)) {
2416 ret
= PTR_ERR(handle
);
2417 printk(KERN_EMERG
"%s: jbd2_start: "
2418 "%ld pages, ino %lu; err %d\n", __func__
,
2419 wbc
->nr_to_write
, inode
->i_ino
, ret
);
2421 goto out_writepages
;
2423 to_write
-= wbc
->nr_to_write
;
2425 mpd
.get_block
= ext4_da_get_block_write
;
2426 ret
= mpage_da_writepages(mapping
, wbc
, &mpd
);
2428 ext4_journal_stop(handle
);
2430 if (mpd
.retval
== -ENOSPC
)
2431 jbd2_journal_force_commit_nested(sbi
->s_journal
);
2433 /* reset the retry count */
2434 if (ret
== MPAGE_DA_EXTENT_TAIL
) {
2436 * got one extent now try with
2439 to_write
+= wbc
->nr_to_write
;
2441 } else if (wbc
->nr_to_write
) {
2443 * There is no more writeout needed
2444 * or we requested for a noblocking writeout
2445 * and we found the device congested
2447 to_write
+= wbc
->nr_to_write
;
2450 wbc
->nr_to_write
= to_write
;
2453 if (!wbc
->range_cyclic
&& (pages_skipped
!= wbc
->pages_skipped
)) {
2454 /* We skipped pages in this loop */
2455 wbc
->nr_to_write
= to_write
+
2456 wbc
->pages_skipped
- pages_skipped
;
2457 wbc
->pages_skipped
= pages_skipped
;
2462 wbc
->nr_to_write
= to_write
- nr_to_writebump
;
2466 #define FALL_BACK_TO_NONDELALLOC 1
2467 static int ext4_nonda_switch(struct super_block
*sb
)
2469 s64 free_blocks
, dirty_blocks
;
2470 struct ext4_sb_info
*sbi
= EXT4_SB(sb
);
2473 * switch to non delalloc mode if we are running low
2474 * on free block. The free block accounting via percpu
2475 * counters can get slightly wrong with FBC_BATCH getting
2476 * accumulated on each CPU without updating global counters
2477 * Delalloc need an accurate free block accounting. So switch
2478 * to non delalloc when we are near to error range.
2480 free_blocks
= percpu_counter_read_positive(&sbi
->s_freeblocks_counter
);
2481 dirty_blocks
= percpu_counter_read_positive(&sbi
->s_dirtyblocks_counter
);
2482 if (2 * free_blocks
< 3 * dirty_blocks
||
2483 free_blocks
< (dirty_blocks
+ EXT4_FREEBLOCKS_WATERMARK
)) {
2485 * free block count is less that 150% of dirty blocks
2486 * or free blocks is less that watermark
2493 static int ext4_da_write_begin(struct file
*file
, struct address_space
*mapping
,
2494 loff_t pos
, unsigned len
, unsigned flags
,
2495 struct page
**pagep
, void **fsdata
)
2497 int ret
, retries
= 0;
2501 struct inode
*inode
= mapping
->host
;
2504 index
= pos
>> PAGE_CACHE_SHIFT
;
2505 from
= pos
& (PAGE_CACHE_SIZE
- 1);
2508 if (ext4_nonda_switch(inode
->i_sb
)) {
2509 *fsdata
= (void *)FALL_BACK_TO_NONDELALLOC
;
2510 return ext4_write_begin(file
, mapping
, pos
,
2511 len
, flags
, pagep
, fsdata
);
2513 *fsdata
= (void *)0;
2516 * With delayed allocation, we don't log the i_disksize update
2517 * if there is delayed block allocation. But we still need
2518 * to journalling the i_disksize update if writes to the end
2519 * of file which has an already mapped buffer.
2521 handle
= ext4_journal_start(inode
, 1);
2522 if (IS_ERR(handle
)) {
2523 ret
= PTR_ERR(handle
);
2527 page
= __grab_cache_page(mapping
, index
);
2529 ext4_journal_stop(handle
);
2535 ret
= block_write_begin(file
, mapping
, pos
, len
, flags
, pagep
, fsdata
,
2536 ext4_da_get_block_prep
);
2539 ext4_journal_stop(handle
);
2540 page_cache_release(page
);
2542 * block_write_begin may have instantiated a few blocks
2543 * outside i_size. Trim these off again. Don't need
2544 * i_size_read because we hold i_mutex.
2546 if (pos
+ len
> inode
->i_size
)
2547 vmtruncate(inode
, inode
->i_size
);
2550 if (ret
== -ENOSPC
&& ext4_should_retry_alloc(inode
->i_sb
, &retries
))
2557 * Check if we should update i_disksize
2558 * when write to the end of file but not require block allocation
2560 static int ext4_da_should_update_i_disksize(struct page
*page
,
2561 unsigned long offset
)
2563 struct buffer_head
*bh
;
2564 struct inode
*inode
= page
->mapping
->host
;
2568 bh
= page_buffers(page
);
2569 idx
= offset
>> inode
->i_blkbits
;
2571 for (i
= 0; i
< idx
; i
++)
2572 bh
= bh
->b_this_page
;
2574 if (!buffer_mapped(bh
) || (buffer_delay(bh
)))
2579 static int ext4_da_write_end(struct file
*file
,
2580 struct address_space
*mapping
,
2581 loff_t pos
, unsigned len
, unsigned copied
,
2582 struct page
*page
, void *fsdata
)
2584 struct inode
*inode
= mapping
->host
;
2586 handle_t
*handle
= ext4_journal_current_handle();
2588 unsigned long start
, end
;
2589 int write_mode
= (int)(unsigned long)fsdata
;
2591 if (write_mode
== FALL_BACK_TO_NONDELALLOC
) {
2592 if (ext4_should_order_data(inode
)) {
2593 return ext4_ordered_write_end(file
, mapping
, pos
,
2594 len
, copied
, page
, fsdata
);
2595 } else if (ext4_should_writeback_data(inode
)) {
2596 return ext4_writeback_write_end(file
, mapping
, pos
,
2597 len
, copied
, page
, fsdata
);
2603 start
= pos
& (PAGE_CACHE_SIZE
- 1);
2604 end
= start
+ copied
- 1;
2607 * generic_write_end() will run mark_inode_dirty() if i_size
2608 * changes. So let's piggyback the i_disksize mark_inode_dirty
2612 new_i_size
= pos
+ copied
;
2613 if (new_i_size
> EXT4_I(inode
)->i_disksize
) {
2614 if (ext4_da_should_update_i_disksize(page
, end
)) {
2615 down_write(&EXT4_I(inode
)->i_data_sem
);
2616 if (new_i_size
> EXT4_I(inode
)->i_disksize
) {
2618 * Updating i_disksize when extending file
2619 * without needing block allocation
2621 if (ext4_should_order_data(inode
))
2622 ret
= ext4_jbd2_file_inode(handle
,
2625 EXT4_I(inode
)->i_disksize
= new_i_size
;
2627 up_write(&EXT4_I(inode
)->i_data_sem
);
2628 /* We need to mark inode dirty even if
2629 * new_i_size is less that inode->i_size
2630 * bu greater than i_disksize.(hint delalloc)
2632 ext4_mark_inode_dirty(handle
, inode
);
2635 ret2
= generic_write_end(file
, mapping
, pos
, len
, copied
,
2640 ret2
= ext4_journal_stop(handle
);
2644 return ret
? ret
: copied
;
2647 static void ext4_da_invalidatepage(struct page
*page
, unsigned long offset
)
2650 * Drop reserved blocks
2652 BUG_ON(!PageLocked(page
));
2653 if (!page_has_buffers(page
))
2656 ext4_da_page_release_reservation(page
, offset
);
2659 ext4_invalidatepage(page
, offset
);
2666 * bmap() is special. It gets used by applications such as lilo and by
2667 * the swapper to find the on-disk block of a specific piece of data.
2669 * Naturally, this is dangerous if the block concerned is still in the
2670 * journal. If somebody makes a swapfile on an ext4 data-journaling
2671 * filesystem and enables swap, then they may get a nasty shock when the
2672 * data getting swapped to that swapfile suddenly gets overwritten by
2673 * the original zero's written out previously to the journal and
2674 * awaiting writeback in the kernel's buffer cache.
2676 * So, if we see any bmap calls here on a modified, data-journaled file,
2677 * take extra steps to flush any blocks which might be in the cache.
2679 static sector_t
ext4_bmap(struct address_space
*mapping
, sector_t block
)
2681 struct inode
*inode
= mapping
->host
;
2685 if (mapping_tagged(mapping
, PAGECACHE_TAG_DIRTY
) &&
2686 test_opt(inode
->i_sb
, DELALLOC
)) {
2688 * With delalloc we want to sync the file
2689 * so that we can make sure we allocate
2692 filemap_write_and_wait(mapping
);
2695 if (EXT4_I(inode
)->i_state
& EXT4_STATE_JDATA
) {
2697 * This is a REALLY heavyweight approach, but the use of
2698 * bmap on dirty files is expected to be extremely rare:
2699 * only if we run lilo or swapon on a freshly made file
2700 * do we expect this to happen.
2702 * (bmap requires CAP_SYS_RAWIO so this does not
2703 * represent an unprivileged user DOS attack --- we'd be
2704 * in trouble if mortal users could trigger this path at
2707 * NB. EXT4_STATE_JDATA is not set on files other than
2708 * regular files. If somebody wants to bmap a directory
2709 * or symlink and gets confused because the buffer
2710 * hasn't yet been flushed to disk, they deserve
2711 * everything they get.
2714 EXT4_I(inode
)->i_state
&= ~EXT4_STATE_JDATA
;
2715 journal
= EXT4_JOURNAL(inode
);
2716 jbd2_journal_lock_updates(journal
);
2717 err
= jbd2_journal_flush(journal
);
2718 jbd2_journal_unlock_updates(journal
);
2724 return generic_block_bmap(mapping
, block
, ext4_get_block
);
2727 static int bget_one(handle_t
*handle
, struct buffer_head
*bh
)
2733 static int bput_one(handle_t
*handle
, struct buffer_head
*bh
)
2740 * Note that we don't need to start a transaction unless we're journaling data
2741 * because we should have holes filled from ext4_page_mkwrite(). We even don't
2742 * need to file the inode to the transaction's list in ordered mode because if
2743 * we are writing back data added by write(), the inode is already there and if
2744 * we are writing back data modified via mmap(), noone guarantees in which
2745 * transaction the data will hit the disk. In case we are journaling data, we
2746 * cannot start transaction directly because transaction start ranks above page
2747 * lock so we have to do some magic.
2749 * In all journaling modes block_write_full_page() will start the I/O.
2753 * ext4_writepage() -> kmalloc() -> __alloc_pages() -> page_launder() ->
2758 * ext4_file_write() -> generic_file_write() -> __alloc_pages() -> ...
2760 * Same applies to ext4_get_block(). We will deadlock on various things like
2761 * lock_journal and i_data_sem
2763 * Setting PF_MEMALLOC here doesn't work - too many internal memory
2766 * 16May01: If we're reentered then journal_current_handle() will be
2767 * non-zero. We simply *return*.
2769 * 1 July 2001: @@@ FIXME:
2770 * In journalled data mode, a data buffer may be metadata against the
2771 * current transaction. But the same file is part of a shared mapping
2772 * and someone does a writepage() on it.
2774 * We will move the buffer onto the async_data list, but *after* it has
2775 * been dirtied. So there's a small window where we have dirty data on
2778 * Note that this only applies to the last partial page in the file. The
2779 * bit which block_write_full_page() uses prepare/commit for. (That's
2780 * broken code anyway: it's wrong for msync()).
2782 * It's a rare case: affects the final partial page, for journalled data
2783 * where the file is subject to bith write() and writepage() in the same
2784 * transction. To fix it we'll need a custom block_write_full_page().
2785 * We'll probably need that anyway for journalling writepage() output.
2787 * We don't honour synchronous mounts for writepage(). That would be
2788 * disastrous. Any write() or metadata operation will sync the fs for
2792 static int __ext4_normal_writepage(struct page
*page
,
2793 struct writeback_control
*wbc
)
2795 struct inode
*inode
= page
->mapping
->host
;
2797 if (test_opt(inode
->i_sb
, NOBH
))
2798 return nobh_writepage(page
,
2799 ext4_normal_get_block_write
, wbc
);
2801 return block_write_full_page(page
,
2802 ext4_normal_get_block_write
,
2806 static int ext4_normal_writepage(struct page
*page
,
2807 struct writeback_control
*wbc
)
2809 struct inode
*inode
= page
->mapping
->host
;
2810 loff_t size
= i_size_read(inode
);
2813 J_ASSERT(PageLocked(page
));
2814 if (page
->index
== size
>> PAGE_CACHE_SHIFT
)
2815 len
= size
& ~PAGE_CACHE_MASK
;
2817 len
= PAGE_CACHE_SIZE
;
2819 if (page_has_buffers(page
)) {
2820 /* if page has buffers it should all be mapped
2821 * and allocated. If there are not buffers attached
2822 * to the page we know the page is dirty but it lost
2823 * buffers. That means that at some moment in time
2824 * after write_begin() / write_end() has been called
2825 * all buffers have been clean and thus they must have been
2826 * written at least once. So they are all mapped and we can
2827 * happily proceed with mapping them and writing the page.
2829 BUG_ON(walk_page_buffers(NULL
, page_buffers(page
), 0, len
, NULL
,
2830 ext4_bh_unmapped_or_delay
));
2833 if (!ext4_journal_current_handle())
2834 return __ext4_normal_writepage(page
, wbc
);
2836 redirty_page_for_writepage(wbc
, page
);
2841 static int __ext4_journalled_writepage(struct page
*page
,
2842 struct writeback_control
*wbc
)
2844 struct address_space
*mapping
= page
->mapping
;
2845 struct inode
*inode
= mapping
->host
;
2846 struct buffer_head
*page_bufs
;
2847 handle_t
*handle
= NULL
;
2851 ret
= block_prepare_write(page
, 0, PAGE_CACHE_SIZE
,
2852 ext4_normal_get_block_write
);
2856 page_bufs
= page_buffers(page
);
2857 walk_page_buffers(handle
, page_bufs
, 0, PAGE_CACHE_SIZE
, NULL
,
2859 /* As soon as we unlock the page, it can go away, but we have
2860 * references to buffers so we are safe */
2863 handle
= ext4_journal_start(inode
, ext4_writepage_trans_blocks(inode
));
2864 if (IS_ERR(handle
)) {
2865 ret
= PTR_ERR(handle
);
2869 ret
= walk_page_buffers(handle
, page_bufs
, 0,
2870 PAGE_CACHE_SIZE
, NULL
, do_journal_get_write_access
);
2872 err
= walk_page_buffers(handle
, page_bufs
, 0,
2873 PAGE_CACHE_SIZE
, NULL
, write_end_fn
);
2876 err
= ext4_journal_stop(handle
);
2880 walk_page_buffers(handle
, page_bufs
, 0,
2881 PAGE_CACHE_SIZE
, NULL
, bput_one
);
2882 EXT4_I(inode
)->i_state
|= EXT4_STATE_JDATA
;
2891 static int ext4_journalled_writepage(struct page
*page
,
2892 struct writeback_control
*wbc
)
2894 struct inode
*inode
= page
->mapping
->host
;
2895 loff_t size
= i_size_read(inode
);
2898 J_ASSERT(PageLocked(page
));
2899 if (page
->index
== size
>> PAGE_CACHE_SHIFT
)
2900 len
= size
& ~PAGE_CACHE_MASK
;
2902 len
= PAGE_CACHE_SIZE
;
2904 if (page_has_buffers(page
)) {
2905 /* if page has buffers it should all be mapped
2906 * and allocated. If there are not buffers attached
2907 * to the page we know the page is dirty but it lost
2908 * buffers. That means that at some moment in time
2909 * after write_begin() / write_end() has been called
2910 * all buffers have been clean and thus they must have been
2911 * written at least once. So they are all mapped and we can
2912 * happily proceed with mapping them and writing the page.
2914 BUG_ON(walk_page_buffers(NULL
, page_buffers(page
), 0, len
, NULL
,
2915 ext4_bh_unmapped_or_delay
));
2918 if (ext4_journal_current_handle())
2921 if (PageChecked(page
)) {
2923 * It's mmapped pagecache. Add buffers and journal it. There
2924 * doesn't seem much point in redirtying the page here.
2926 ClearPageChecked(page
);
2927 return __ext4_journalled_writepage(page
, wbc
);
2930 * It may be a page full of checkpoint-mode buffers. We don't
2931 * really know unless we go poke around in the buffer_heads.
2932 * But block_write_full_page will do the right thing.
2934 return block_write_full_page(page
,
2935 ext4_normal_get_block_write
,
2939 redirty_page_for_writepage(wbc
, page
);
2944 static int ext4_readpage(struct file
*file
, struct page
*page
)
2946 return mpage_readpage(page
, ext4_get_block
);
2950 ext4_readpages(struct file
*file
, struct address_space
*mapping
,
2951 struct list_head
*pages
, unsigned nr_pages
)
2953 return mpage_readpages(mapping
, pages
, nr_pages
, ext4_get_block
);
2956 static void ext4_invalidatepage(struct page
*page
, unsigned long offset
)
2958 journal_t
*journal
= EXT4_JOURNAL(page
->mapping
->host
);
2961 * If it's a full truncate we just forget about the pending dirtying
2964 ClearPageChecked(page
);
2966 jbd2_journal_invalidatepage(journal
, page
, offset
);
2969 static int ext4_releasepage(struct page
*page
, gfp_t wait
)
2971 journal_t
*journal
= EXT4_JOURNAL(page
->mapping
->host
);
2973 WARN_ON(PageChecked(page
));
2974 if (!page_has_buffers(page
))
2976 return jbd2_journal_try_to_free_buffers(journal
, page
, wait
);
2980 * If the O_DIRECT write will extend the file then add this inode to the
2981 * orphan list. So recovery will truncate it back to the original size
2982 * if the machine crashes during the write.
2984 * If the O_DIRECT write is intantiating holes inside i_size and the machine
2985 * crashes then stale disk data _may_ be exposed inside the file. But current
2986 * VFS code falls back into buffered path in that case so we are safe.
2988 static ssize_t
ext4_direct_IO(int rw
, struct kiocb
*iocb
,
2989 const struct iovec
*iov
, loff_t offset
,
2990 unsigned long nr_segs
)
2992 struct file
*file
= iocb
->ki_filp
;
2993 struct inode
*inode
= file
->f_mapping
->host
;
2994 struct ext4_inode_info
*ei
= EXT4_I(inode
);
2998 size_t count
= iov_length(iov
, nr_segs
);
3001 loff_t final_size
= offset
+ count
;
3003 if (final_size
> inode
->i_size
) {
3004 /* Credits for sb + inode write */
3005 handle
= ext4_journal_start(inode
, 2);
3006 if (IS_ERR(handle
)) {
3007 ret
= PTR_ERR(handle
);
3010 ret
= ext4_orphan_add(handle
, inode
);
3012 ext4_journal_stop(handle
);
3016 ei
->i_disksize
= inode
->i_size
;
3017 ext4_journal_stop(handle
);
3021 ret
= blockdev_direct_IO(rw
, iocb
, inode
, inode
->i_sb
->s_bdev
, iov
,
3023 ext4_get_block
, NULL
);
3028 /* Credits for sb + inode write */
3029 handle
= ext4_journal_start(inode
, 2);
3030 if (IS_ERR(handle
)) {
3031 /* This is really bad luck. We've written the data
3032 * but cannot extend i_size. Bail out and pretend
3033 * the write failed... */
3034 ret
= PTR_ERR(handle
);
3038 ext4_orphan_del(handle
, inode
);
3040 loff_t end
= offset
+ ret
;
3041 if (end
> inode
->i_size
) {
3042 ei
->i_disksize
= end
;
3043 i_size_write(inode
, end
);
3045 * We're going to return a positive `ret'
3046 * here due to non-zero-length I/O, so there's
3047 * no way of reporting error returns from
3048 * ext4_mark_inode_dirty() to userspace. So
3051 ext4_mark_inode_dirty(handle
, inode
);
3054 err
= ext4_journal_stop(handle
);
3063 * Pages can be marked dirty completely asynchronously from ext4's journalling
3064 * activity. By filemap_sync_pte(), try_to_unmap_one(), etc. We cannot do
3065 * much here because ->set_page_dirty is called under VFS locks. The page is
3066 * not necessarily locked.
3068 * We cannot just dirty the page and leave attached buffers clean, because the
3069 * buffers' dirty state is "definitive". We cannot just set the buffers dirty
3070 * or jbddirty because all the journalling code will explode.
3072 * So what we do is to mark the page "pending dirty" and next time writepage
3073 * is called, propagate that into the buffers appropriately.
3075 static int ext4_journalled_set_page_dirty(struct page
*page
)
3077 SetPageChecked(page
);
3078 return __set_page_dirty_nobuffers(page
);
3081 static const struct address_space_operations ext4_ordered_aops
= {
3082 .readpage
= ext4_readpage
,
3083 .readpages
= ext4_readpages
,
3084 .writepage
= ext4_normal_writepage
,
3085 .sync_page
= block_sync_page
,
3086 .write_begin
= ext4_write_begin
,
3087 .write_end
= ext4_ordered_write_end
,
3089 .invalidatepage
= ext4_invalidatepage
,
3090 .releasepage
= ext4_releasepage
,
3091 .direct_IO
= ext4_direct_IO
,
3092 .migratepage
= buffer_migrate_page
,
3093 .is_partially_uptodate
= block_is_partially_uptodate
,
3096 static const struct address_space_operations ext4_writeback_aops
= {
3097 .readpage
= ext4_readpage
,
3098 .readpages
= ext4_readpages
,
3099 .writepage
= ext4_normal_writepage
,
3100 .sync_page
= block_sync_page
,
3101 .write_begin
= ext4_write_begin
,
3102 .write_end
= ext4_writeback_write_end
,
3104 .invalidatepage
= ext4_invalidatepage
,
3105 .releasepage
= ext4_releasepage
,
3106 .direct_IO
= ext4_direct_IO
,
3107 .migratepage
= buffer_migrate_page
,
3108 .is_partially_uptodate
= block_is_partially_uptodate
,
3111 static const struct address_space_operations ext4_journalled_aops
= {
3112 .readpage
= ext4_readpage
,
3113 .readpages
= ext4_readpages
,
3114 .writepage
= ext4_journalled_writepage
,
3115 .sync_page
= block_sync_page
,
3116 .write_begin
= ext4_write_begin
,
3117 .write_end
= ext4_journalled_write_end
,
3118 .set_page_dirty
= ext4_journalled_set_page_dirty
,
3120 .invalidatepage
= ext4_invalidatepage
,
3121 .releasepage
= ext4_releasepage
,
3122 .is_partially_uptodate
= block_is_partially_uptodate
,
3125 static const struct address_space_operations ext4_da_aops
= {
3126 .readpage
= ext4_readpage
,
3127 .readpages
= ext4_readpages
,
3128 .writepage
= ext4_da_writepage
,
3129 .writepages
= ext4_da_writepages
,
3130 .sync_page
= block_sync_page
,
3131 .write_begin
= ext4_da_write_begin
,
3132 .write_end
= ext4_da_write_end
,
3134 .invalidatepage
= ext4_da_invalidatepage
,
3135 .releasepage
= ext4_releasepage
,
3136 .direct_IO
= ext4_direct_IO
,
3137 .migratepage
= buffer_migrate_page
,
3138 .is_partially_uptodate
= block_is_partially_uptodate
,
3141 void ext4_set_aops(struct inode
*inode
)
3143 if (ext4_should_order_data(inode
) &&
3144 test_opt(inode
->i_sb
, DELALLOC
))
3145 inode
->i_mapping
->a_ops
= &ext4_da_aops
;
3146 else if (ext4_should_order_data(inode
))
3147 inode
->i_mapping
->a_ops
= &ext4_ordered_aops
;
3148 else if (ext4_should_writeback_data(inode
) &&
3149 test_opt(inode
->i_sb
, DELALLOC
))
3150 inode
->i_mapping
->a_ops
= &ext4_da_aops
;
3151 else if (ext4_should_writeback_data(inode
))
3152 inode
->i_mapping
->a_ops
= &ext4_writeback_aops
;
3154 inode
->i_mapping
->a_ops
= &ext4_journalled_aops
;
3158 * ext4_block_truncate_page() zeroes out a mapping from file offset `from'
3159 * up to the end of the block which corresponds to `from'.
3160 * This required during truncate. We need to physically zero the tail end
3161 * of that block so it doesn't yield old data if the file is later grown.
3163 int ext4_block_truncate_page(handle_t
*handle
,
3164 struct address_space
*mapping
, loff_t from
)
3166 ext4_fsblk_t index
= from
>> PAGE_CACHE_SHIFT
;
3167 unsigned offset
= from
& (PAGE_CACHE_SIZE
-1);
3168 unsigned blocksize
, length
, pos
;
3170 struct inode
*inode
= mapping
->host
;
3171 struct buffer_head
*bh
;
3175 page
= grab_cache_page(mapping
, from
>> PAGE_CACHE_SHIFT
);
3179 blocksize
= inode
->i_sb
->s_blocksize
;
3180 length
= blocksize
- (offset
& (blocksize
- 1));
3181 iblock
= index
<< (PAGE_CACHE_SHIFT
- inode
->i_sb
->s_blocksize_bits
);
3184 * For "nobh" option, we can only work if we don't need to
3185 * read-in the page - otherwise we create buffers to do the IO.
3187 if (!page_has_buffers(page
) && test_opt(inode
->i_sb
, NOBH
) &&
3188 ext4_should_writeback_data(inode
) && PageUptodate(page
)) {
3189 zero_user(page
, offset
, length
);
3190 set_page_dirty(page
);
3194 if (!page_has_buffers(page
))
3195 create_empty_buffers(page
, blocksize
, 0);
3197 /* Find the buffer that contains "offset" */
3198 bh
= page_buffers(page
);
3200 while (offset
>= pos
) {
3201 bh
= bh
->b_this_page
;
3207 if (buffer_freed(bh
)) {
3208 BUFFER_TRACE(bh
, "freed: skip");
3212 if (!buffer_mapped(bh
)) {
3213 BUFFER_TRACE(bh
, "unmapped");
3214 ext4_get_block(inode
, iblock
, bh
, 0);
3215 /* unmapped? It's a hole - nothing to do */
3216 if (!buffer_mapped(bh
)) {
3217 BUFFER_TRACE(bh
, "still unmapped");
3222 /* Ok, it's mapped. Make sure it's up-to-date */
3223 if (PageUptodate(page
))
3224 set_buffer_uptodate(bh
);
3226 if (!buffer_uptodate(bh
)) {
3228 ll_rw_block(READ
, 1, &bh
);
3230 /* Uhhuh. Read error. Complain and punt. */
3231 if (!buffer_uptodate(bh
))
3235 if (ext4_should_journal_data(inode
)) {
3236 BUFFER_TRACE(bh
, "get write access");
3237 err
= ext4_journal_get_write_access(handle
, bh
);
3242 zero_user(page
, offset
, length
);
3244 BUFFER_TRACE(bh
, "zeroed end of block");
3247 if (ext4_should_journal_data(inode
)) {
3248 err
= ext4_journal_dirty_metadata(handle
, bh
);
3250 if (ext4_should_order_data(inode
))
3251 err
= ext4_jbd2_file_inode(handle
, inode
);
3252 mark_buffer_dirty(bh
);
3257 page_cache_release(page
);
3262 * Probably it should be a library function... search for first non-zero word
3263 * or memcmp with zero_page, whatever is better for particular architecture.
3266 static inline int all_zeroes(__le32
*p
, __le32
*q
)
3275 * ext4_find_shared - find the indirect blocks for partial truncation.
3276 * @inode: inode in question
3277 * @depth: depth of the affected branch
3278 * @offsets: offsets of pointers in that branch (see ext4_block_to_path)
3279 * @chain: place to store the pointers to partial indirect blocks
3280 * @top: place to the (detached) top of branch
3282 * This is a helper function used by ext4_truncate().
3284 * When we do truncate() we may have to clean the ends of several
3285 * indirect blocks but leave the blocks themselves alive. Block is
3286 * partially truncated if some data below the new i_size is refered
3287 * from it (and it is on the path to the first completely truncated
3288 * data block, indeed). We have to free the top of that path along
3289 * with everything to the right of the path. Since no allocation
3290 * past the truncation point is possible until ext4_truncate()
3291 * finishes, we may safely do the latter, but top of branch may
3292 * require special attention - pageout below the truncation point
3293 * might try to populate it.
3295 * We atomically detach the top of branch from the tree, store the
3296 * block number of its root in *@top, pointers to buffer_heads of
3297 * partially truncated blocks - in @chain[].bh and pointers to
3298 * their last elements that should not be removed - in
3299 * @chain[].p. Return value is the pointer to last filled element
3302 * The work left to caller to do the actual freeing of subtrees:
3303 * a) free the subtree starting from *@top
3304 * b) free the subtrees whose roots are stored in
3305 * (@chain[i].p+1 .. end of @chain[i].bh->b_data)
3306 * c) free the subtrees growing from the inode past the @chain[0].
3307 * (no partially truncated stuff there). */
3309 static Indirect
*ext4_find_shared(struct inode
*inode
, int depth
,
3310 ext4_lblk_t offsets
[4], Indirect chain
[4], __le32
*top
)
3312 Indirect
*partial
, *p
;
3316 /* Make k index the deepest non-null offest + 1 */
3317 for (k
= depth
; k
> 1 && !offsets
[k
-1]; k
--)
3319 partial
= ext4_get_branch(inode
, k
, offsets
, chain
, &err
);
3320 /* Writer: pointers */
3322 partial
= chain
+ k
-1;
3324 * If the branch acquired continuation since we've looked at it -
3325 * fine, it should all survive and (new) top doesn't belong to us.
3327 if (!partial
->key
&& *partial
->p
)
3330 for (p
= partial
; (p
> chain
) && all_zeroes((__le32
*) p
->bh
->b_data
, p
->p
); p
--)
3333 * OK, we've found the last block that must survive. The rest of our
3334 * branch should be detached before unlocking. However, if that rest
3335 * of branch is all ours and does not grow immediately from the inode
3336 * it's easier to cheat and just decrement partial->p.
3338 if (p
== chain
+ k
- 1 && p
> chain
) {
3342 /* Nope, don't do this in ext4. Must leave the tree intact */
3349 while (partial
> p
) {
3350 brelse(partial
->bh
);
3358 * Zero a number of block pointers in either an inode or an indirect block.
3359 * If we restart the transaction we must again get write access to the
3360 * indirect block for further modification.
3362 * We release `count' blocks on disk, but (last - first) may be greater
3363 * than `count' because there can be holes in there.
3365 static void ext4_clear_blocks(handle_t
*handle
, struct inode
*inode
,
3366 struct buffer_head
*bh
, ext4_fsblk_t block_to_free
,
3367 unsigned long count
, __le32
*first
, __le32
*last
)
3370 if (try_to_extend_transaction(handle
, inode
)) {
3372 BUFFER_TRACE(bh
, "call ext4_journal_dirty_metadata");
3373 ext4_journal_dirty_metadata(handle
, bh
);
3375 ext4_mark_inode_dirty(handle
, inode
);
3376 ext4_journal_test_restart(handle
, inode
);
3378 BUFFER_TRACE(bh
, "retaking write access");
3379 ext4_journal_get_write_access(handle
, bh
);
3384 * Any buffers which are on the journal will be in memory. We find
3385 * them on the hash table so jbd2_journal_revoke() will run jbd2_journal_forget()
3386 * on them. We've already detached each block from the file, so
3387 * bforget() in jbd2_journal_forget() should be safe.
3389 * AKPM: turn on bforget in jbd2_journal_forget()!!!
3391 for (p
= first
; p
< last
; p
++) {
3392 u32 nr
= le32_to_cpu(*p
);
3394 struct buffer_head
*tbh
;
3397 tbh
= sb_find_get_block(inode
->i_sb
, nr
);
3398 ext4_forget(handle
, 0, inode
, tbh
, nr
);
3402 ext4_free_blocks(handle
, inode
, block_to_free
, count
, 0);
3406 * ext4_free_data - free a list of data blocks
3407 * @handle: handle for this transaction
3408 * @inode: inode we are dealing with
3409 * @this_bh: indirect buffer_head which contains *@first and *@last
3410 * @first: array of block numbers
3411 * @last: points immediately past the end of array
3413 * We are freeing all blocks refered from that array (numbers are stored as
3414 * little-endian 32-bit) and updating @inode->i_blocks appropriately.
3416 * We accumulate contiguous runs of blocks to free. Conveniently, if these
3417 * blocks are contiguous then releasing them at one time will only affect one
3418 * or two bitmap blocks (+ group descriptor(s) and superblock) and we won't
3419 * actually use a lot of journal space.
3421 * @this_bh will be %NULL if @first and @last point into the inode's direct
3424 static void ext4_free_data(handle_t
*handle
, struct inode
*inode
,
3425 struct buffer_head
*this_bh
,
3426 __le32
*first
, __le32
*last
)
3428 ext4_fsblk_t block_to_free
= 0; /* Starting block # of a run */
3429 unsigned long count
= 0; /* Number of blocks in the run */
3430 __le32
*block_to_free_p
= NULL
; /* Pointer into inode/ind
3433 ext4_fsblk_t nr
; /* Current block # */
3434 __le32
*p
; /* Pointer into inode/ind
3435 for current block */
3438 if (this_bh
) { /* For indirect block */
3439 BUFFER_TRACE(this_bh
, "get_write_access");
3440 err
= ext4_journal_get_write_access(handle
, this_bh
);
3441 /* Important: if we can't update the indirect pointers
3442 * to the blocks, we can't free them. */
3447 for (p
= first
; p
< last
; p
++) {
3448 nr
= le32_to_cpu(*p
);
3450 /* accumulate blocks to free if they're contiguous */
3453 block_to_free_p
= p
;
3455 } else if (nr
== block_to_free
+ count
) {
3458 ext4_clear_blocks(handle
, inode
, this_bh
,
3460 count
, block_to_free_p
, p
);
3462 block_to_free_p
= p
;
3469 ext4_clear_blocks(handle
, inode
, this_bh
, block_to_free
,
3470 count
, block_to_free_p
, p
);
3473 BUFFER_TRACE(this_bh
, "call ext4_journal_dirty_metadata");
3476 * The buffer head should have an attached journal head at this
3477 * point. However, if the data is corrupted and an indirect
3478 * block pointed to itself, it would have been detached when
3479 * the block was cleared. Check for this instead of OOPSing.
3482 ext4_journal_dirty_metadata(handle
, this_bh
);
3484 ext4_error(inode
->i_sb
, __func__
,
3485 "circular indirect block detected, "
3486 "inode=%lu, block=%llu",
3488 (unsigned long long) this_bh
->b_blocknr
);
3493 * ext4_free_branches - free an array of branches
3494 * @handle: JBD handle for this transaction
3495 * @inode: inode we are dealing with
3496 * @parent_bh: the buffer_head which contains *@first and *@last
3497 * @first: array of block numbers
3498 * @last: pointer immediately past the end of array
3499 * @depth: depth of the branches to free
3501 * We are freeing all blocks refered from these branches (numbers are
3502 * stored as little-endian 32-bit) and updating @inode->i_blocks
3505 static void ext4_free_branches(handle_t
*handle
, struct inode
*inode
,
3506 struct buffer_head
*parent_bh
,
3507 __le32
*first
, __le32
*last
, int depth
)
3512 if (is_handle_aborted(handle
))
3516 struct buffer_head
*bh
;
3517 int addr_per_block
= EXT4_ADDR_PER_BLOCK(inode
->i_sb
);
3519 while (--p
>= first
) {
3520 nr
= le32_to_cpu(*p
);
3522 continue; /* A hole */
3524 /* Go read the buffer for the next level down */
3525 bh
= sb_bread(inode
->i_sb
, nr
);
3528 * A read failure? Report error and clear slot
3532 ext4_error(inode
->i_sb
, "ext4_free_branches",
3533 "Read failure, inode=%lu, block=%llu",
3538 /* This zaps the entire block. Bottom up. */
3539 BUFFER_TRACE(bh
, "free child branches");
3540 ext4_free_branches(handle
, inode
, bh
,
3541 (__le32
*) bh
->b_data
,
3542 (__le32
*) bh
->b_data
+ addr_per_block
,
3546 * We've probably journalled the indirect block several
3547 * times during the truncate. But it's no longer
3548 * needed and we now drop it from the transaction via
3549 * jbd2_journal_revoke().
3551 * That's easy if it's exclusively part of this
3552 * transaction. But if it's part of the committing
3553 * transaction then jbd2_journal_forget() will simply
3554 * brelse() it. That means that if the underlying
3555 * block is reallocated in ext4_get_block(),
3556 * unmap_underlying_metadata() will find this block
3557 * and will try to get rid of it. damn, damn.
3559 * If this block has already been committed to the
3560 * journal, a revoke record will be written. And
3561 * revoke records must be emitted *before* clearing
3562 * this block's bit in the bitmaps.
3564 ext4_forget(handle
, 1, inode
, bh
, bh
->b_blocknr
);
3567 * Everything below this this pointer has been
3568 * released. Now let this top-of-subtree go.
3570 * We want the freeing of this indirect block to be
3571 * atomic in the journal with the updating of the
3572 * bitmap block which owns it. So make some room in
3575 * We zero the parent pointer *after* freeing its
3576 * pointee in the bitmaps, so if extend_transaction()
3577 * for some reason fails to put the bitmap changes and
3578 * the release into the same transaction, recovery
3579 * will merely complain about releasing a free block,
3580 * rather than leaking blocks.
3582 if (is_handle_aborted(handle
))
3584 if (try_to_extend_transaction(handle
, inode
)) {
3585 ext4_mark_inode_dirty(handle
, inode
);
3586 ext4_journal_test_restart(handle
, inode
);
3589 ext4_free_blocks(handle
, inode
, nr
, 1, 1);
3593 * The block which we have just freed is
3594 * pointed to by an indirect block: journal it
3596 BUFFER_TRACE(parent_bh
, "get_write_access");
3597 if (!ext4_journal_get_write_access(handle
,
3600 BUFFER_TRACE(parent_bh
,
3601 "call ext4_journal_dirty_metadata");
3602 ext4_journal_dirty_metadata(handle
,
3608 /* We have reached the bottom of the tree. */
3609 BUFFER_TRACE(parent_bh
, "free data blocks");
3610 ext4_free_data(handle
, inode
, parent_bh
, first
, last
);
3614 int ext4_can_truncate(struct inode
*inode
)
3616 if (IS_APPEND(inode
) || IS_IMMUTABLE(inode
))
3618 if (S_ISREG(inode
->i_mode
))
3620 if (S_ISDIR(inode
->i_mode
))
3622 if (S_ISLNK(inode
->i_mode
))
3623 return !ext4_inode_is_fast_symlink(inode
);
3630 * We block out ext4_get_block() block instantiations across the entire
3631 * transaction, and VFS/VM ensures that ext4_truncate() cannot run
3632 * simultaneously on behalf of the same inode.
3634 * As we work through the truncate and commmit bits of it to the journal there
3635 * is one core, guiding principle: the file's tree must always be consistent on
3636 * disk. We must be able to restart the truncate after a crash.
3638 * The file's tree may be transiently inconsistent in memory (although it
3639 * probably isn't), but whenever we close off and commit a journal transaction,
3640 * the contents of (the filesystem + the journal) must be consistent and
3641 * restartable. It's pretty simple, really: bottom up, right to left (although
3642 * left-to-right works OK too).
3644 * Note that at recovery time, journal replay occurs *before* the restart of
3645 * truncate against the orphan inode list.
3647 * The committed inode has the new, desired i_size (which is the same as
3648 * i_disksize in this case). After a crash, ext4_orphan_cleanup() will see
3649 * that this inode's truncate did not complete and it will again call
3650 * ext4_truncate() to have another go. So there will be instantiated blocks
3651 * to the right of the truncation point in a crashed ext4 filesystem. But
3652 * that's fine - as long as they are linked from the inode, the post-crash
3653 * ext4_truncate() run will find them and release them.
3655 void ext4_truncate(struct inode
*inode
)
3658 struct ext4_inode_info
*ei
= EXT4_I(inode
);
3659 __le32
*i_data
= ei
->i_data
;
3660 int addr_per_block
= EXT4_ADDR_PER_BLOCK(inode
->i_sb
);
3661 struct address_space
*mapping
= inode
->i_mapping
;
3662 ext4_lblk_t offsets
[4];
3667 ext4_lblk_t last_block
;
3668 unsigned blocksize
= inode
->i_sb
->s_blocksize
;
3670 if (!ext4_can_truncate(inode
))
3673 if (EXT4_I(inode
)->i_flags
& EXT4_EXTENTS_FL
) {
3674 ext4_ext_truncate(inode
);
3678 handle
= start_transaction(inode
);
3680 return; /* AKPM: return what? */
3682 last_block
= (inode
->i_size
+ blocksize
-1)
3683 >> EXT4_BLOCK_SIZE_BITS(inode
->i_sb
);
3685 if (inode
->i_size
& (blocksize
- 1))
3686 if (ext4_block_truncate_page(handle
, mapping
, inode
->i_size
))
3689 n
= ext4_block_to_path(inode
, last_block
, offsets
, NULL
);
3691 goto out_stop
; /* error */
3694 * OK. This truncate is going to happen. We add the inode to the
3695 * orphan list, so that if this truncate spans multiple transactions,
3696 * and we crash, we will resume the truncate when the filesystem
3697 * recovers. It also marks the inode dirty, to catch the new size.
3699 * Implication: the file must always be in a sane, consistent
3700 * truncatable state while each transaction commits.
3702 if (ext4_orphan_add(handle
, inode
))
3706 * From here we block out all ext4_get_block() callers who want to
3707 * modify the block allocation tree.
3709 down_write(&ei
->i_data_sem
);
3711 ext4_discard_preallocations(inode
);
3714 * The orphan list entry will now protect us from any crash which
3715 * occurs before the truncate completes, so it is now safe to propagate
3716 * the new, shorter inode size (held for now in i_size) into the
3717 * on-disk inode. We do this via i_disksize, which is the value which
3718 * ext4 *really* writes onto the disk inode.
3720 ei
->i_disksize
= inode
->i_size
;
3722 if (n
== 1) { /* direct blocks */
3723 ext4_free_data(handle
, inode
, NULL
, i_data
+offsets
[0],
3724 i_data
+ EXT4_NDIR_BLOCKS
);
3728 partial
= ext4_find_shared(inode
, n
, offsets
, chain
, &nr
);
3729 /* Kill the top of shared branch (not detached) */
3731 if (partial
== chain
) {
3732 /* Shared branch grows from the inode */
3733 ext4_free_branches(handle
, inode
, NULL
,
3734 &nr
, &nr
+1, (chain
+n
-1) - partial
);
3737 * We mark the inode dirty prior to restart,
3738 * and prior to stop. No need for it here.
3741 /* Shared branch grows from an indirect block */
3742 BUFFER_TRACE(partial
->bh
, "get_write_access");
3743 ext4_free_branches(handle
, inode
, partial
->bh
,
3745 partial
->p
+1, (chain
+n
-1) - partial
);
3748 /* Clear the ends of indirect blocks on the shared branch */
3749 while (partial
> chain
) {
3750 ext4_free_branches(handle
, inode
, partial
->bh
, partial
->p
+ 1,
3751 (__le32
*)partial
->bh
->b_data
+addr_per_block
,
3752 (chain
+n
-1) - partial
);
3753 BUFFER_TRACE(partial
->bh
, "call brelse");
3754 brelse (partial
->bh
);
3758 /* Kill the remaining (whole) subtrees */
3759 switch (offsets
[0]) {
3761 nr
= i_data
[EXT4_IND_BLOCK
];
3763 ext4_free_branches(handle
, inode
, NULL
, &nr
, &nr
+1, 1);
3764 i_data
[EXT4_IND_BLOCK
] = 0;
3766 case EXT4_IND_BLOCK
:
3767 nr
= i_data
[EXT4_DIND_BLOCK
];
3769 ext4_free_branches(handle
, inode
, NULL
, &nr
, &nr
+1, 2);
3770 i_data
[EXT4_DIND_BLOCK
] = 0;
3772 case EXT4_DIND_BLOCK
:
3773 nr
= i_data
[EXT4_TIND_BLOCK
];
3775 ext4_free_branches(handle
, inode
, NULL
, &nr
, &nr
+1, 3);
3776 i_data
[EXT4_TIND_BLOCK
] = 0;
3778 case EXT4_TIND_BLOCK
:
3782 up_write(&ei
->i_data_sem
);
3783 inode
->i_mtime
= inode
->i_ctime
= ext4_current_time(inode
);
3784 ext4_mark_inode_dirty(handle
, inode
);
3787 * In a multi-transaction truncate, we only make the final transaction
3794 * If this was a simple ftruncate(), and the file will remain alive
3795 * then we need to clear up the orphan record which we created above.
3796 * However, if this was a real unlink then we were called by
3797 * ext4_delete_inode(), and we allow that function to clean up the
3798 * orphan info for us.
3801 ext4_orphan_del(handle
, inode
);
3803 ext4_journal_stop(handle
);
3807 * ext4_get_inode_loc returns with an extra refcount against the inode's
3808 * underlying buffer_head on success. If 'in_mem' is true, we have all
3809 * data in memory that is needed to recreate the on-disk version of this
3812 static int __ext4_get_inode_loc(struct inode
*inode
,
3813 struct ext4_iloc
*iloc
, int in_mem
)
3815 struct ext4_group_desc
*gdp
;
3816 struct buffer_head
*bh
;
3817 struct super_block
*sb
= inode
->i_sb
;
3819 int inodes_per_block
, inode_offset
;
3822 if (!ext4_valid_inum(sb
, inode
->i_ino
))
3825 iloc
->block_group
= (inode
->i_ino
- 1) / EXT4_INODES_PER_GROUP(sb
);
3826 gdp
= ext4_get_group_desc(sb
, iloc
->block_group
, NULL
);
3831 * Figure out the offset within the block group inode table
3833 inodes_per_block
= (EXT4_BLOCK_SIZE(sb
) / EXT4_INODE_SIZE(sb
));
3834 inode_offset
= ((inode
->i_ino
- 1) %
3835 EXT4_INODES_PER_GROUP(sb
));
3836 block
= ext4_inode_table(sb
, gdp
) + (inode_offset
/ inodes_per_block
);
3837 iloc
->offset
= (inode_offset
% inodes_per_block
) * EXT4_INODE_SIZE(sb
);
3839 bh
= sb_getblk(sb
, block
);
3841 ext4_error(sb
, "ext4_get_inode_loc", "unable to read "
3842 "inode block - inode=%lu, block=%llu",
3843 inode
->i_ino
, block
);
3846 if (!buffer_uptodate(bh
)) {
3850 * If the buffer has the write error flag, we have failed
3851 * to write out another inode in the same block. In this
3852 * case, we don't have to read the block because we may
3853 * read the old inode data successfully.
3855 if (buffer_write_io_error(bh
) && !buffer_uptodate(bh
))
3856 set_buffer_uptodate(bh
);
3858 if (buffer_uptodate(bh
)) {
3859 /* someone brought it uptodate while we waited */
3865 * If we have all information of the inode in memory and this
3866 * is the only valid inode in the block, we need not read the
3870 struct buffer_head
*bitmap_bh
;
3873 start
= inode_offset
& ~(inodes_per_block
- 1);
3875 /* Is the inode bitmap in cache? */
3876 bitmap_bh
= sb_getblk(sb
, ext4_inode_bitmap(sb
, gdp
));
3881 * If the inode bitmap isn't in cache then the
3882 * optimisation may end up performing two reads instead
3883 * of one, so skip it.
3885 if (!buffer_uptodate(bitmap_bh
)) {
3889 for (i
= start
; i
< start
+ inodes_per_block
; i
++) {
3890 if (i
== inode_offset
)
3892 if (ext4_test_bit(i
, bitmap_bh
->b_data
))
3896 if (i
== start
+ inodes_per_block
) {
3897 /* all other inodes are free, so skip I/O */
3898 memset(bh
->b_data
, 0, bh
->b_size
);
3899 set_buffer_uptodate(bh
);
3907 * If we need to do any I/O, try to pre-readahead extra
3908 * blocks from the inode table.
3910 if (EXT4_SB(sb
)->s_inode_readahead_blks
) {
3911 ext4_fsblk_t b
, end
, table
;
3914 table
= ext4_inode_table(sb
, gdp
);
3915 /* Make sure s_inode_readahead_blks is a power of 2 */
3916 while (EXT4_SB(sb
)->s_inode_readahead_blks
&
3917 (EXT4_SB(sb
)->s_inode_readahead_blks
-1))
3918 EXT4_SB(sb
)->s_inode_readahead_blks
=
3919 (EXT4_SB(sb
)->s_inode_readahead_blks
&
3920 (EXT4_SB(sb
)->s_inode_readahead_blks
-1));
3921 b
= block
& ~(EXT4_SB(sb
)->s_inode_readahead_blks
-1);
3924 end
= b
+ EXT4_SB(sb
)->s_inode_readahead_blks
;
3925 num
= EXT4_INODES_PER_GROUP(sb
);
3926 if (EXT4_HAS_RO_COMPAT_FEATURE(sb
,
3927 EXT4_FEATURE_RO_COMPAT_GDT_CSUM
))
3928 num
-= le16_to_cpu(gdp
->bg_itable_unused
);
3929 table
+= num
/ inodes_per_block
;
3933 sb_breadahead(sb
, b
++);
3937 * There are other valid inodes in the buffer, this inode
3938 * has in-inode xattrs, or we don't have this inode in memory.
3939 * Read the block from disk.
3942 bh
->b_end_io
= end_buffer_read_sync
;
3943 submit_bh(READ_META
, bh
);
3945 if (!buffer_uptodate(bh
)) {
3946 ext4_error(sb
, __func__
,
3947 "unable to read inode block - inode=%lu, "
3948 "block=%llu", inode
->i_ino
, block
);
3958 int ext4_get_inode_loc(struct inode
*inode
, struct ext4_iloc
*iloc
)
3960 /* We have all inode data except xattrs in memory here. */
3961 return __ext4_get_inode_loc(inode
, iloc
,
3962 !(EXT4_I(inode
)->i_state
& EXT4_STATE_XATTR
));
3965 void ext4_set_inode_flags(struct inode
*inode
)
3967 unsigned int flags
= EXT4_I(inode
)->i_flags
;
3969 inode
->i_flags
&= ~(S_SYNC
|S_APPEND
|S_IMMUTABLE
|S_NOATIME
|S_DIRSYNC
);
3970 if (flags
& EXT4_SYNC_FL
)
3971 inode
->i_flags
|= S_SYNC
;
3972 if (flags
& EXT4_APPEND_FL
)
3973 inode
->i_flags
|= S_APPEND
;
3974 if (flags
& EXT4_IMMUTABLE_FL
)
3975 inode
->i_flags
|= S_IMMUTABLE
;
3976 if (flags
& EXT4_NOATIME_FL
)
3977 inode
->i_flags
|= S_NOATIME
;
3978 if (flags
& EXT4_DIRSYNC_FL
)
3979 inode
->i_flags
|= S_DIRSYNC
;
3982 /* Propagate flags from i_flags to EXT4_I(inode)->i_flags */
3983 void ext4_get_inode_flags(struct ext4_inode_info
*ei
)
3985 unsigned int flags
= ei
->vfs_inode
.i_flags
;
3987 ei
->i_flags
&= ~(EXT4_SYNC_FL
|EXT4_APPEND_FL
|
3988 EXT4_IMMUTABLE_FL
|EXT4_NOATIME_FL
|EXT4_DIRSYNC_FL
);
3990 ei
->i_flags
|= EXT4_SYNC_FL
;
3991 if (flags
& S_APPEND
)
3992 ei
->i_flags
|= EXT4_APPEND_FL
;
3993 if (flags
& S_IMMUTABLE
)
3994 ei
->i_flags
|= EXT4_IMMUTABLE_FL
;
3995 if (flags
& S_NOATIME
)
3996 ei
->i_flags
|= EXT4_NOATIME_FL
;
3997 if (flags
& S_DIRSYNC
)
3998 ei
->i_flags
|= EXT4_DIRSYNC_FL
;
4000 static blkcnt_t
ext4_inode_blocks(struct ext4_inode
*raw_inode
,
4001 struct ext4_inode_info
*ei
)
4004 struct inode
*inode
= &(ei
->vfs_inode
);
4005 struct super_block
*sb
= inode
->i_sb
;
4007 if (EXT4_HAS_RO_COMPAT_FEATURE(sb
,
4008 EXT4_FEATURE_RO_COMPAT_HUGE_FILE
)) {
4009 /* we are using combined 48 bit field */
4010 i_blocks
= ((u64
)le16_to_cpu(raw_inode
->i_blocks_high
)) << 32 |
4011 le32_to_cpu(raw_inode
->i_blocks_lo
);
4012 if (ei
->i_flags
& EXT4_HUGE_FILE_FL
) {
4013 /* i_blocks represent file system block size */
4014 return i_blocks
<< (inode
->i_blkbits
- 9);
4019 return le32_to_cpu(raw_inode
->i_blocks_lo
);
4023 struct inode
*ext4_iget(struct super_block
*sb
, unsigned long ino
)
4025 struct ext4_iloc iloc
;
4026 struct ext4_inode
*raw_inode
;
4027 struct ext4_inode_info
*ei
;
4028 struct buffer_head
*bh
;
4029 struct inode
*inode
;
4033 inode
= iget_locked(sb
, ino
);
4035 return ERR_PTR(-ENOMEM
);
4036 if (!(inode
->i_state
& I_NEW
))
4040 #ifdef CONFIG_EXT4_FS_POSIX_ACL
4041 ei
->i_acl
= EXT4_ACL_NOT_CACHED
;
4042 ei
->i_default_acl
= EXT4_ACL_NOT_CACHED
;
4045 ret
= __ext4_get_inode_loc(inode
, &iloc
, 0);
4049 raw_inode
= ext4_raw_inode(&iloc
);
4050 inode
->i_mode
= le16_to_cpu(raw_inode
->i_mode
);
4051 inode
->i_uid
= (uid_t
)le16_to_cpu(raw_inode
->i_uid_low
);
4052 inode
->i_gid
= (gid_t
)le16_to_cpu(raw_inode
->i_gid_low
);
4053 if (!(test_opt(inode
->i_sb
, NO_UID32
))) {
4054 inode
->i_uid
|= le16_to_cpu(raw_inode
->i_uid_high
) << 16;
4055 inode
->i_gid
|= le16_to_cpu(raw_inode
->i_gid_high
) << 16;
4057 inode
->i_nlink
= le16_to_cpu(raw_inode
->i_links_count
);
4060 ei
->i_dir_start_lookup
= 0;
4061 ei
->i_dtime
= le32_to_cpu(raw_inode
->i_dtime
);
4062 /* We now have enough fields to check if the inode was active or not.
4063 * This is needed because nfsd might try to access dead inodes
4064 * the test is that same one that e2fsck uses
4065 * NeilBrown 1999oct15
4067 if (inode
->i_nlink
== 0) {
4068 if (inode
->i_mode
== 0 ||
4069 !(EXT4_SB(inode
->i_sb
)->s_mount_state
& EXT4_ORPHAN_FS
)) {
4070 /* this inode is deleted */
4075 /* The only unlinked inodes we let through here have
4076 * valid i_mode and are being read by the orphan
4077 * recovery code: that's fine, we're about to complete
4078 * the process of deleting those. */
4080 ei
->i_flags
= le32_to_cpu(raw_inode
->i_flags
);
4081 inode
->i_blocks
= ext4_inode_blocks(raw_inode
, ei
);
4082 ei
->i_file_acl
= le32_to_cpu(raw_inode
->i_file_acl_lo
);
4083 if (EXT4_SB(inode
->i_sb
)->s_es
->s_creator_os
!=
4084 cpu_to_le32(EXT4_OS_HURD
)) {
4086 ((__u64
)le16_to_cpu(raw_inode
->i_file_acl_high
)) << 32;
4088 inode
->i_size
= ext4_isize(raw_inode
);
4089 ei
->i_disksize
= inode
->i_size
;
4090 inode
->i_generation
= le32_to_cpu(raw_inode
->i_generation
);
4091 ei
->i_block_group
= iloc
.block_group
;
4093 * NOTE! The in-memory inode i_data array is in little-endian order
4094 * even on big-endian machines: we do NOT byteswap the block numbers!
4096 for (block
= 0; block
< EXT4_N_BLOCKS
; block
++)
4097 ei
->i_data
[block
] = raw_inode
->i_block
[block
];
4098 INIT_LIST_HEAD(&ei
->i_orphan
);
4100 if (EXT4_INODE_SIZE(inode
->i_sb
) > EXT4_GOOD_OLD_INODE_SIZE
) {
4101 ei
->i_extra_isize
= le16_to_cpu(raw_inode
->i_extra_isize
);
4102 if (EXT4_GOOD_OLD_INODE_SIZE
+ ei
->i_extra_isize
>
4103 EXT4_INODE_SIZE(inode
->i_sb
)) {
4108 if (ei
->i_extra_isize
== 0) {
4109 /* The extra space is currently unused. Use it. */
4110 ei
->i_extra_isize
= sizeof(struct ext4_inode
) -
4111 EXT4_GOOD_OLD_INODE_SIZE
;
4113 __le32
*magic
= (void *)raw_inode
+
4114 EXT4_GOOD_OLD_INODE_SIZE
+
4116 if (*magic
== cpu_to_le32(EXT4_XATTR_MAGIC
))
4117 ei
->i_state
|= EXT4_STATE_XATTR
;
4120 ei
->i_extra_isize
= 0;
4122 EXT4_INODE_GET_XTIME(i_ctime
, inode
, raw_inode
);
4123 EXT4_INODE_GET_XTIME(i_mtime
, inode
, raw_inode
);
4124 EXT4_INODE_GET_XTIME(i_atime
, inode
, raw_inode
);
4125 EXT4_EINODE_GET_XTIME(i_crtime
, ei
, raw_inode
);
4127 inode
->i_version
= le32_to_cpu(raw_inode
->i_disk_version
);
4128 if (EXT4_INODE_SIZE(inode
->i_sb
) > EXT4_GOOD_OLD_INODE_SIZE
) {
4129 if (EXT4_FITS_IN_INODE(raw_inode
, ei
, i_version_hi
))
4131 (__u64
)(le32_to_cpu(raw_inode
->i_version_hi
)) << 32;
4134 if (S_ISREG(inode
->i_mode
)) {
4135 inode
->i_op
= &ext4_file_inode_operations
;
4136 inode
->i_fop
= &ext4_file_operations
;
4137 ext4_set_aops(inode
);
4138 } else if (S_ISDIR(inode
->i_mode
)) {
4139 inode
->i_op
= &ext4_dir_inode_operations
;
4140 inode
->i_fop
= &ext4_dir_operations
;
4141 } else if (S_ISLNK(inode
->i_mode
)) {
4142 if (ext4_inode_is_fast_symlink(inode
))
4143 inode
->i_op
= &ext4_fast_symlink_inode_operations
;
4145 inode
->i_op
= &ext4_symlink_inode_operations
;
4146 ext4_set_aops(inode
);
4149 inode
->i_op
= &ext4_special_inode_operations
;
4150 if (raw_inode
->i_block
[0])
4151 init_special_inode(inode
, inode
->i_mode
,
4152 old_decode_dev(le32_to_cpu(raw_inode
->i_block
[0])));
4154 init_special_inode(inode
, inode
->i_mode
,
4155 new_decode_dev(le32_to_cpu(raw_inode
->i_block
[1])));
4158 ext4_set_inode_flags(inode
);
4159 unlock_new_inode(inode
);
4164 return ERR_PTR(ret
);
4167 static int ext4_inode_blocks_set(handle_t
*handle
,
4168 struct ext4_inode
*raw_inode
,
4169 struct ext4_inode_info
*ei
)
4171 struct inode
*inode
= &(ei
->vfs_inode
);
4172 u64 i_blocks
= inode
->i_blocks
;
4173 struct super_block
*sb
= inode
->i_sb
;
4176 if (i_blocks
<= ~0U) {
4178 * i_blocks can be represnted in a 32 bit variable
4179 * as multiple of 512 bytes
4181 raw_inode
->i_blocks_lo
= cpu_to_le32(i_blocks
);
4182 raw_inode
->i_blocks_high
= 0;
4183 ei
->i_flags
&= ~EXT4_HUGE_FILE_FL
;
4184 } else if (i_blocks
<= 0xffffffffffffULL
) {
4186 * i_blocks can be represented in a 48 bit variable
4187 * as multiple of 512 bytes
4189 err
= ext4_update_rocompat_feature(handle
, sb
,
4190 EXT4_FEATURE_RO_COMPAT_HUGE_FILE
);
4193 /* i_block is stored in the split 48 bit fields */
4194 raw_inode
->i_blocks_lo
= cpu_to_le32(i_blocks
);
4195 raw_inode
->i_blocks_high
= cpu_to_le16(i_blocks
>> 32);
4196 ei
->i_flags
&= ~EXT4_HUGE_FILE_FL
;
4199 * i_blocks should be represented in a 48 bit variable
4200 * as multiple of file system block size
4202 err
= ext4_update_rocompat_feature(handle
, sb
,
4203 EXT4_FEATURE_RO_COMPAT_HUGE_FILE
);
4206 ei
->i_flags
|= EXT4_HUGE_FILE_FL
;
4207 /* i_block is stored in file system block size */
4208 i_blocks
= i_blocks
>> (inode
->i_blkbits
- 9);
4209 raw_inode
->i_blocks_lo
= cpu_to_le32(i_blocks
);
4210 raw_inode
->i_blocks_high
= cpu_to_le16(i_blocks
>> 32);
4217 * Post the struct inode info into an on-disk inode location in the
4218 * buffer-cache. This gobbles the caller's reference to the
4219 * buffer_head in the inode location struct.
4221 * The caller must have write access to iloc->bh.
4223 static int ext4_do_update_inode(handle_t
*handle
,
4224 struct inode
*inode
,
4225 struct ext4_iloc
*iloc
)
4227 struct ext4_inode
*raw_inode
= ext4_raw_inode(iloc
);
4228 struct ext4_inode_info
*ei
= EXT4_I(inode
);
4229 struct buffer_head
*bh
= iloc
->bh
;
4230 int err
= 0, rc
, block
;
4232 /* For fields not not tracking in the in-memory inode,
4233 * initialise them to zero for new inodes. */
4234 if (ei
->i_state
& EXT4_STATE_NEW
)
4235 memset(raw_inode
, 0, EXT4_SB(inode
->i_sb
)->s_inode_size
);
4237 ext4_get_inode_flags(ei
);
4238 raw_inode
->i_mode
= cpu_to_le16(inode
->i_mode
);
4239 if (!(test_opt(inode
->i_sb
, NO_UID32
))) {
4240 raw_inode
->i_uid_low
= cpu_to_le16(low_16_bits(inode
->i_uid
));
4241 raw_inode
->i_gid_low
= cpu_to_le16(low_16_bits(inode
->i_gid
));
4243 * Fix up interoperability with old kernels. Otherwise, old inodes get
4244 * re-used with the upper 16 bits of the uid/gid intact
4247 raw_inode
->i_uid_high
=
4248 cpu_to_le16(high_16_bits(inode
->i_uid
));
4249 raw_inode
->i_gid_high
=
4250 cpu_to_le16(high_16_bits(inode
->i_gid
));
4252 raw_inode
->i_uid_high
= 0;
4253 raw_inode
->i_gid_high
= 0;
4256 raw_inode
->i_uid_low
=
4257 cpu_to_le16(fs_high2lowuid(inode
->i_uid
));
4258 raw_inode
->i_gid_low
=
4259 cpu_to_le16(fs_high2lowgid(inode
->i_gid
));
4260 raw_inode
->i_uid_high
= 0;
4261 raw_inode
->i_gid_high
= 0;
4263 raw_inode
->i_links_count
= cpu_to_le16(inode
->i_nlink
);
4265 EXT4_INODE_SET_XTIME(i_ctime
, inode
, raw_inode
);
4266 EXT4_INODE_SET_XTIME(i_mtime
, inode
, raw_inode
);
4267 EXT4_INODE_SET_XTIME(i_atime
, inode
, raw_inode
);
4268 EXT4_EINODE_SET_XTIME(i_crtime
, ei
, raw_inode
);
4270 if (ext4_inode_blocks_set(handle
, raw_inode
, ei
))
4272 raw_inode
->i_dtime
= cpu_to_le32(ei
->i_dtime
);
4273 /* clear the migrate flag in the raw_inode */
4274 raw_inode
->i_flags
= cpu_to_le32(ei
->i_flags
& ~EXT4_EXT_MIGRATE
);
4275 if (EXT4_SB(inode
->i_sb
)->s_es
->s_creator_os
!=
4276 cpu_to_le32(EXT4_OS_HURD
))
4277 raw_inode
->i_file_acl_high
=
4278 cpu_to_le16(ei
->i_file_acl
>> 32);
4279 raw_inode
->i_file_acl_lo
= cpu_to_le32(ei
->i_file_acl
);
4280 ext4_isize_set(raw_inode
, ei
->i_disksize
);
4281 if (ei
->i_disksize
> 0x7fffffffULL
) {
4282 struct super_block
*sb
= inode
->i_sb
;
4283 if (!EXT4_HAS_RO_COMPAT_FEATURE(sb
,
4284 EXT4_FEATURE_RO_COMPAT_LARGE_FILE
) ||
4285 EXT4_SB(sb
)->s_es
->s_rev_level
==
4286 cpu_to_le32(EXT4_GOOD_OLD_REV
)) {
4287 /* If this is the first large file
4288 * created, add a flag to the superblock.
4290 err
= ext4_journal_get_write_access(handle
,
4291 EXT4_SB(sb
)->s_sbh
);
4294 ext4_update_dynamic_rev(sb
);
4295 EXT4_SET_RO_COMPAT_FEATURE(sb
,
4296 EXT4_FEATURE_RO_COMPAT_LARGE_FILE
);
4299 err
= ext4_journal_dirty_metadata(handle
,
4300 EXT4_SB(sb
)->s_sbh
);
4303 raw_inode
->i_generation
= cpu_to_le32(inode
->i_generation
);
4304 if (S_ISCHR(inode
->i_mode
) || S_ISBLK(inode
->i_mode
)) {
4305 if (old_valid_dev(inode
->i_rdev
)) {
4306 raw_inode
->i_block
[0] =
4307 cpu_to_le32(old_encode_dev(inode
->i_rdev
));
4308 raw_inode
->i_block
[1] = 0;
4310 raw_inode
->i_block
[0] = 0;
4311 raw_inode
->i_block
[1] =
4312 cpu_to_le32(new_encode_dev(inode
->i_rdev
));
4313 raw_inode
->i_block
[2] = 0;
4315 } else for (block
= 0; block
< EXT4_N_BLOCKS
; block
++)
4316 raw_inode
->i_block
[block
] = ei
->i_data
[block
];
4318 raw_inode
->i_disk_version
= cpu_to_le32(inode
->i_version
);
4319 if (ei
->i_extra_isize
) {
4320 if (EXT4_FITS_IN_INODE(raw_inode
, ei
, i_version_hi
))
4321 raw_inode
->i_version_hi
=
4322 cpu_to_le32(inode
->i_version
>> 32);
4323 raw_inode
->i_extra_isize
= cpu_to_le16(ei
->i_extra_isize
);
4327 BUFFER_TRACE(bh
, "call ext4_journal_dirty_metadata");
4328 rc
= ext4_journal_dirty_metadata(handle
, bh
);
4331 ei
->i_state
&= ~EXT4_STATE_NEW
;
4335 ext4_std_error(inode
->i_sb
, err
);
4340 * ext4_write_inode()
4342 * We are called from a few places:
4344 * - Within generic_file_write() for O_SYNC files.
4345 * Here, there will be no transaction running. We wait for any running
4346 * trasnaction to commit.
4348 * - Within sys_sync(), kupdate and such.
4349 * We wait on commit, if tol to.
4351 * - Within prune_icache() (PF_MEMALLOC == true)
4352 * Here we simply return. We can't afford to block kswapd on the
4355 * In all cases it is actually safe for us to return without doing anything,
4356 * because the inode has been copied into a raw inode buffer in
4357 * ext4_mark_inode_dirty(). This is a correctness thing for O_SYNC and for
4360 * Note that we are absolutely dependent upon all inode dirtiers doing the
4361 * right thing: they *must* call mark_inode_dirty() after dirtying info in
4362 * which we are interested.
4364 * It would be a bug for them to not do this. The code:
4366 * mark_inode_dirty(inode)
4368 * inode->i_size = expr;
4370 * is in error because a kswapd-driven write_inode() could occur while
4371 * `stuff()' is running, and the new i_size will be lost. Plus the inode
4372 * will no longer be on the superblock's dirty inode list.
4374 int ext4_write_inode(struct inode
*inode
, int wait
)
4376 if (current
->flags
& PF_MEMALLOC
)
4379 if (ext4_journal_current_handle()) {
4380 jbd_debug(1, "called recursively, non-PF_MEMALLOC!\n");
4388 return ext4_force_commit(inode
->i_sb
);
4394 * Called from notify_change.
4396 * We want to trap VFS attempts to truncate the file as soon as
4397 * possible. In particular, we want to make sure that when the VFS
4398 * shrinks i_size, we put the inode on the orphan list and modify
4399 * i_disksize immediately, so that during the subsequent flushing of
4400 * dirty pages and freeing of disk blocks, we can guarantee that any
4401 * commit will leave the blocks being flushed in an unused state on
4402 * disk. (On recovery, the inode will get truncated and the blocks will
4403 * be freed, so we have a strong guarantee that no future commit will
4404 * leave these blocks visible to the user.)
4406 * Another thing we have to assure is that if we are in ordered mode
4407 * and inode is still attached to the committing transaction, we must
4408 * we start writeout of all the dirty pages which are being truncated.
4409 * This way we are sure that all the data written in the previous
4410 * transaction are already on disk (truncate waits for pages under
4413 * Called with inode->i_mutex down.
4415 int ext4_setattr(struct dentry
*dentry
, struct iattr
*attr
)
4417 struct inode
*inode
= dentry
->d_inode
;
4419 const unsigned int ia_valid
= attr
->ia_valid
;
4421 error
= inode_change_ok(inode
, attr
);
4425 if ((ia_valid
& ATTR_UID
&& attr
->ia_uid
!= inode
->i_uid
) ||
4426 (ia_valid
& ATTR_GID
&& attr
->ia_gid
!= inode
->i_gid
)) {
4429 /* (user+group)*(old+new) structure, inode write (sb,
4430 * inode block, ? - but truncate inode update has it) */
4431 handle
= ext4_journal_start(inode
, 2*(EXT4_QUOTA_INIT_BLOCKS(inode
->i_sb
)+
4432 EXT4_QUOTA_DEL_BLOCKS(inode
->i_sb
))+3);
4433 if (IS_ERR(handle
)) {
4434 error
= PTR_ERR(handle
);
4437 error
= DQUOT_TRANSFER(inode
, attr
) ? -EDQUOT
: 0;
4439 ext4_journal_stop(handle
);
4442 /* Update corresponding info in inode so that everything is in
4443 * one transaction */
4444 if (attr
->ia_valid
& ATTR_UID
)
4445 inode
->i_uid
= attr
->ia_uid
;
4446 if (attr
->ia_valid
& ATTR_GID
)
4447 inode
->i_gid
= attr
->ia_gid
;
4448 error
= ext4_mark_inode_dirty(handle
, inode
);
4449 ext4_journal_stop(handle
);
4452 if (attr
->ia_valid
& ATTR_SIZE
) {
4453 if (!(EXT4_I(inode
)->i_flags
& EXT4_EXTENTS_FL
)) {
4454 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
4456 if (attr
->ia_size
> sbi
->s_bitmap_maxbytes
) {
4463 if (S_ISREG(inode
->i_mode
) &&
4464 attr
->ia_valid
& ATTR_SIZE
&& attr
->ia_size
< inode
->i_size
) {
4467 handle
= ext4_journal_start(inode
, 3);
4468 if (IS_ERR(handle
)) {
4469 error
= PTR_ERR(handle
);
4473 error
= ext4_orphan_add(handle
, inode
);
4474 EXT4_I(inode
)->i_disksize
= attr
->ia_size
;
4475 rc
= ext4_mark_inode_dirty(handle
, inode
);
4478 ext4_journal_stop(handle
);
4480 if (ext4_should_order_data(inode
)) {
4481 error
= ext4_begin_ordered_truncate(inode
,
4484 /* Do as much error cleanup as possible */
4485 handle
= ext4_journal_start(inode
, 3);
4486 if (IS_ERR(handle
)) {
4487 ext4_orphan_del(NULL
, inode
);
4490 ext4_orphan_del(handle
, inode
);
4491 ext4_journal_stop(handle
);
4497 rc
= inode_setattr(inode
, attr
);
4499 /* If inode_setattr's call to ext4_truncate failed to get a
4500 * transaction handle at all, we need to clean up the in-core
4501 * orphan list manually. */
4503 ext4_orphan_del(NULL
, inode
);
4505 if (!rc
&& (ia_valid
& ATTR_MODE
))
4506 rc
= ext4_acl_chmod(inode
);
4509 ext4_std_error(inode
->i_sb
, error
);
4515 int ext4_getattr(struct vfsmount
*mnt
, struct dentry
*dentry
,
4518 struct inode
*inode
;
4519 unsigned long delalloc_blocks
;
4521 inode
= dentry
->d_inode
;
4522 generic_fillattr(inode
, stat
);
4525 * We can't update i_blocks if the block allocation is delayed
4526 * otherwise in the case of system crash before the real block
4527 * allocation is done, we will have i_blocks inconsistent with
4528 * on-disk file blocks.
4529 * We always keep i_blocks updated together with real
4530 * allocation. But to not confuse with user, stat
4531 * will return the blocks that include the delayed allocation
4532 * blocks for this file.
4534 spin_lock(&EXT4_I(inode
)->i_block_reservation_lock
);
4535 delalloc_blocks
= EXT4_I(inode
)->i_reserved_data_blocks
;
4536 spin_unlock(&EXT4_I(inode
)->i_block_reservation_lock
);
4538 stat
->blocks
+= (delalloc_blocks
<< inode
->i_sb
->s_blocksize_bits
)>>9;
4542 static int ext4_indirect_trans_blocks(struct inode
*inode
, int nrblocks
,
4547 /* if nrblocks are contiguous */
4550 * With N contiguous data blocks, it need at most
4551 * N/EXT4_ADDR_PER_BLOCK(inode->i_sb) indirect blocks
4552 * 2 dindirect blocks
4555 indirects
= nrblocks
/ EXT4_ADDR_PER_BLOCK(inode
->i_sb
);
4556 return indirects
+ 3;
4559 * if nrblocks are not contiguous, worse case, each block touch
4560 * a indirect block, and each indirect block touch a double indirect
4561 * block, plus a triple indirect block
4563 indirects
= nrblocks
* 2 + 1;
4567 static int ext4_index_trans_blocks(struct inode
*inode
, int nrblocks
, int chunk
)
4569 if (!(EXT4_I(inode
)->i_flags
& EXT4_EXTENTS_FL
))
4570 return ext4_indirect_trans_blocks(inode
, nrblocks
, 0);
4571 return ext4_ext_index_trans_blocks(inode
, nrblocks
, 0);
4574 * Account for index blocks, block groups bitmaps and block group
4575 * descriptor blocks if modify datablocks and index blocks
4576 * worse case, the indexs blocks spread over different block groups
4578 * If datablocks are discontiguous, they are possible to spread over
4579 * different block groups too. If they are contiugous, with flexbg,
4580 * they could still across block group boundary.
4582 * Also account for superblock, inode, quota and xattr blocks
4584 int ext4_meta_trans_blocks(struct inode
*inode
, int nrblocks
, int chunk
)
4586 int groups
, gdpblocks
;
4591 * How many index blocks need to touch to modify nrblocks?
4592 * The "Chunk" flag indicating whether the nrblocks is
4593 * physically contiguous on disk
4595 * For Direct IO and fallocate, they calls get_block to allocate
4596 * one single extent at a time, so they could set the "Chunk" flag
4598 idxblocks
= ext4_index_trans_blocks(inode
, nrblocks
, chunk
);
4603 * Now let's see how many group bitmaps and group descriptors need
4613 if (groups
> EXT4_SB(inode
->i_sb
)->s_groups_count
)
4614 groups
= EXT4_SB(inode
->i_sb
)->s_groups_count
;
4615 if (groups
> EXT4_SB(inode
->i_sb
)->s_gdb_count
)
4616 gdpblocks
= EXT4_SB(inode
->i_sb
)->s_gdb_count
;
4618 /* bitmaps and block group descriptor blocks */
4619 ret
+= groups
+ gdpblocks
;
4621 /* Blocks for super block, inode, quota and xattr blocks */
4622 ret
+= EXT4_META_TRANS_BLOCKS(inode
->i_sb
);
4628 * Calulate the total number of credits to reserve to fit
4629 * the modification of a single pages into a single transaction,
4630 * which may include multiple chunks of block allocations.
4632 * This could be called via ext4_write_begin()
4634 * We need to consider the worse case, when
4635 * one new block per extent.
4637 int ext4_writepage_trans_blocks(struct inode
*inode
)
4639 int bpp
= ext4_journal_blocks_per_page(inode
);
4642 ret
= ext4_meta_trans_blocks(inode
, bpp
, 0);
4644 /* Account for data blocks for journalled mode */
4645 if (ext4_should_journal_data(inode
))
4651 * Calculate the journal credits for a chunk of data modification.
4653 * This is called from DIO, fallocate or whoever calling
4654 * ext4_get_blocks_wrap() to map/allocate a chunk of contigous disk blocks.
4656 * journal buffers for data blocks are not included here, as DIO
4657 * and fallocate do no need to journal data buffers.
4659 int ext4_chunk_trans_blocks(struct inode
*inode
, int nrblocks
)
4661 return ext4_meta_trans_blocks(inode
, nrblocks
, 1);
4665 * The caller must have previously called ext4_reserve_inode_write().
4666 * Give this, we know that the caller already has write access to iloc->bh.
4668 int ext4_mark_iloc_dirty(handle_t
*handle
,
4669 struct inode
*inode
, struct ext4_iloc
*iloc
)
4673 if (test_opt(inode
->i_sb
, I_VERSION
))
4674 inode_inc_iversion(inode
);
4676 /* the do_update_inode consumes one bh->b_count */
4679 /* ext4_do_update_inode() does jbd2_journal_dirty_metadata */
4680 err
= ext4_do_update_inode(handle
, inode
, iloc
);
4686 * On success, We end up with an outstanding reference count against
4687 * iloc->bh. This _must_ be cleaned up later.
4691 ext4_reserve_inode_write(handle_t
*handle
, struct inode
*inode
,
4692 struct ext4_iloc
*iloc
)
4696 err
= ext4_get_inode_loc(inode
, iloc
);
4698 BUFFER_TRACE(iloc
->bh
, "get_write_access");
4699 err
= ext4_journal_get_write_access(handle
, iloc
->bh
);
4706 ext4_std_error(inode
->i_sb
, err
);
4711 * Expand an inode by new_extra_isize bytes.
4712 * Returns 0 on success or negative error number on failure.
4714 static int ext4_expand_extra_isize(struct inode
*inode
,
4715 unsigned int new_extra_isize
,
4716 struct ext4_iloc iloc
,
4719 struct ext4_inode
*raw_inode
;
4720 struct ext4_xattr_ibody_header
*header
;
4721 struct ext4_xattr_entry
*entry
;
4723 if (EXT4_I(inode
)->i_extra_isize
>= new_extra_isize
)
4726 raw_inode
= ext4_raw_inode(&iloc
);
4728 header
= IHDR(inode
, raw_inode
);
4729 entry
= IFIRST(header
);
4731 /* No extended attributes present */
4732 if (!(EXT4_I(inode
)->i_state
& EXT4_STATE_XATTR
) ||
4733 header
->h_magic
!= cpu_to_le32(EXT4_XATTR_MAGIC
)) {
4734 memset((void *)raw_inode
+ EXT4_GOOD_OLD_INODE_SIZE
, 0,
4736 EXT4_I(inode
)->i_extra_isize
= new_extra_isize
;
4740 /* try to expand with EAs present */
4741 return ext4_expand_extra_isize_ea(inode
, new_extra_isize
,
4746 * What we do here is to mark the in-core inode as clean with respect to inode
4747 * dirtiness (it may still be data-dirty).
4748 * This means that the in-core inode may be reaped by prune_icache
4749 * without having to perform any I/O. This is a very good thing,
4750 * because *any* task may call prune_icache - even ones which
4751 * have a transaction open against a different journal.
4753 * Is this cheating? Not really. Sure, we haven't written the
4754 * inode out, but prune_icache isn't a user-visible syncing function.
4755 * Whenever the user wants stuff synced (sys_sync, sys_msync, sys_fsync)
4756 * we start and wait on commits.
4758 * Is this efficient/effective? Well, we're being nice to the system
4759 * by cleaning up our inodes proactively so they can be reaped
4760 * without I/O. But we are potentially leaving up to five seconds'
4761 * worth of inodes floating about which prune_icache wants us to
4762 * write out. One way to fix that would be to get prune_icache()
4763 * to do a write_super() to free up some memory. It has the desired
4766 int ext4_mark_inode_dirty(handle_t
*handle
, struct inode
*inode
)
4768 struct ext4_iloc iloc
;
4769 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
4770 static unsigned int mnt_count
;
4774 err
= ext4_reserve_inode_write(handle
, inode
, &iloc
);
4775 if (EXT4_I(inode
)->i_extra_isize
< sbi
->s_want_extra_isize
&&
4776 !(EXT4_I(inode
)->i_state
& EXT4_STATE_NO_EXPAND
)) {
4778 * We need extra buffer credits since we may write into EA block
4779 * with this same handle. If journal_extend fails, then it will
4780 * only result in a minor loss of functionality for that inode.
4781 * If this is felt to be critical, then e2fsck should be run to
4782 * force a large enough s_min_extra_isize.
4784 if ((jbd2_journal_extend(handle
,
4785 EXT4_DATA_TRANS_BLOCKS(inode
->i_sb
))) == 0) {
4786 ret
= ext4_expand_extra_isize(inode
,
4787 sbi
->s_want_extra_isize
,
4790 EXT4_I(inode
)->i_state
|= EXT4_STATE_NO_EXPAND
;
4792 le16_to_cpu(sbi
->s_es
->s_mnt_count
)) {
4793 ext4_warning(inode
->i_sb
, __func__
,
4794 "Unable to expand inode %lu. Delete"
4795 " some EAs or run e2fsck.",
4798 le16_to_cpu(sbi
->s_es
->s_mnt_count
);
4804 err
= ext4_mark_iloc_dirty(handle
, inode
, &iloc
);
4809 * ext4_dirty_inode() is called from __mark_inode_dirty()
4811 * We're really interested in the case where a file is being extended.
4812 * i_size has been changed by generic_commit_write() and we thus need
4813 * to include the updated inode in the current transaction.
4815 * Also, DQUOT_ALLOC_SPACE() will always dirty the inode when blocks
4816 * are allocated to the file.
4818 * If the inode is marked synchronous, we don't honour that here - doing
4819 * so would cause a commit on atime updates, which we don't bother doing.
4820 * We handle synchronous inodes at the highest possible level.
4822 void ext4_dirty_inode(struct inode
*inode
)
4824 handle_t
*current_handle
= ext4_journal_current_handle();
4827 handle
= ext4_journal_start(inode
, 2);
4830 if (current_handle
&&
4831 current_handle
->h_transaction
!= handle
->h_transaction
) {
4832 /* This task has a transaction open against a different fs */
4833 printk(KERN_EMERG
"%s: transactions do not match!\n",
4836 jbd_debug(5, "marking dirty. outer handle=%p\n",
4838 ext4_mark_inode_dirty(handle
, inode
);
4840 ext4_journal_stop(handle
);
4847 * Bind an inode's backing buffer_head into this transaction, to prevent
4848 * it from being flushed to disk early. Unlike
4849 * ext4_reserve_inode_write, this leaves behind no bh reference and
4850 * returns no iloc structure, so the caller needs to repeat the iloc
4851 * lookup to mark the inode dirty later.
4853 static int ext4_pin_inode(handle_t
*handle
, struct inode
*inode
)
4855 struct ext4_iloc iloc
;
4859 err
= ext4_get_inode_loc(inode
, &iloc
);
4861 BUFFER_TRACE(iloc
.bh
, "get_write_access");
4862 err
= jbd2_journal_get_write_access(handle
, iloc
.bh
);
4864 err
= ext4_journal_dirty_metadata(handle
,
4869 ext4_std_error(inode
->i_sb
, err
);
4874 int ext4_change_inode_journal_flag(struct inode
*inode
, int val
)
4881 * We have to be very careful here: changing a data block's
4882 * journaling status dynamically is dangerous. If we write a
4883 * data block to the journal, change the status and then delete
4884 * that block, we risk forgetting to revoke the old log record
4885 * from the journal and so a subsequent replay can corrupt data.
4886 * So, first we make sure that the journal is empty and that
4887 * nobody is changing anything.
4890 journal
= EXT4_JOURNAL(inode
);
4891 if (is_journal_aborted(journal
))
4894 jbd2_journal_lock_updates(journal
);
4895 jbd2_journal_flush(journal
);
4898 * OK, there are no updates running now, and all cached data is
4899 * synced to disk. We are now in a completely consistent state
4900 * which doesn't have anything in the journal, and we know that
4901 * no filesystem updates are running, so it is safe to modify
4902 * the inode's in-core data-journaling state flag now.
4906 EXT4_I(inode
)->i_flags
|= EXT4_JOURNAL_DATA_FL
;
4908 EXT4_I(inode
)->i_flags
&= ~EXT4_JOURNAL_DATA_FL
;
4909 ext4_set_aops(inode
);
4911 jbd2_journal_unlock_updates(journal
);
4913 /* Finally we can mark the inode as dirty. */
4915 handle
= ext4_journal_start(inode
, 1);
4917 return PTR_ERR(handle
);
4919 err
= ext4_mark_inode_dirty(handle
, inode
);
4921 ext4_journal_stop(handle
);
4922 ext4_std_error(inode
->i_sb
, err
);
4927 static int ext4_bh_unmapped(handle_t
*handle
, struct buffer_head
*bh
)
4929 return !buffer_mapped(bh
);
4932 int ext4_page_mkwrite(struct vm_area_struct
*vma
, struct page
*page
)
4938 struct file
*file
= vma
->vm_file
;
4939 struct inode
*inode
= file
->f_path
.dentry
->d_inode
;
4940 struct address_space
*mapping
= inode
->i_mapping
;
4943 * Get i_alloc_sem to stop truncates messing with the inode. We cannot
4944 * get i_mutex because we are already holding mmap_sem.
4946 down_read(&inode
->i_alloc_sem
);
4947 size
= i_size_read(inode
);
4948 if (page
->mapping
!= mapping
|| size
<= page_offset(page
)
4949 || !PageUptodate(page
)) {
4950 /* page got truncated from under us? */
4954 if (PageMappedToDisk(page
))
4957 if (page
->index
== size
>> PAGE_CACHE_SHIFT
)
4958 len
= size
& ~PAGE_CACHE_MASK
;
4960 len
= PAGE_CACHE_SIZE
;
4962 if (page_has_buffers(page
)) {
4963 /* return if we have all the buffers mapped */
4964 if (!walk_page_buffers(NULL
, page_buffers(page
), 0, len
, NULL
,
4969 * OK, we need to fill the hole... Do write_begin write_end
4970 * to do block allocation/reservation.We are not holding
4971 * inode.i__mutex here. That allow * parallel write_begin,
4972 * write_end call. lock_page prevent this from happening
4973 * on the same page though
4975 ret
= mapping
->a_ops
->write_begin(file
, mapping
, page_offset(page
),
4976 len
, AOP_FLAG_UNINTERRUPTIBLE
, &page
, &fsdata
);
4979 ret
= mapping
->a_ops
->write_end(file
, mapping
, page_offset(page
),
4980 len
, len
, page
, fsdata
);
4985 up_read(&inode
->i_alloc_sem
);