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
;
1653 BUG_ON(mpd
->next_page
<= mpd
->first_page
);
1654 pagevec_init(&pvec
, 0);
1655 index
= mpd
->first_page
;
1656 end
= mpd
->next_page
- 1;
1658 while (index
<= end
) {
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 pages_skipped
= mpd
->wbc
->pages_skipped
;
1677 err
= mapping
->a_ops
->writepage(page
, mpd
->wbc
);
1678 if (!err
&& (pages_skipped
== mpd
->wbc
->pages_skipped
))
1680 * have successfully written the page
1681 * without skipping the same
1683 mpd
->pages_written
++;
1685 * In error case, we have to continue because
1686 * remaining pages are still locked
1687 * XXX: unlock and re-dirty them?
1692 pagevec_release(&pvec
);
1698 * mpage_put_bnr_to_bhs - walk blocks and assign them actual numbers
1700 * @mpd->inode - inode to walk through
1701 * @exbh->b_blocknr - first block on a disk
1702 * @exbh->b_size - amount of space in bytes
1703 * @logical - first logical block to start assignment with
1705 * the function goes through all passed space and put actual disk
1706 * block numbers into buffer heads, dropping BH_Delay
1708 static void mpage_put_bnr_to_bhs(struct mpage_da_data
*mpd
, sector_t logical
,
1709 struct buffer_head
*exbh
)
1711 struct inode
*inode
= mpd
->inode
;
1712 struct address_space
*mapping
= inode
->i_mapping
;
1713 int blocks
= exbh
->b_size
>> inode
->i_blkbits
;
1714 sector_t pblock
= exbh
->b_blocknr
, cur_logical
;
1715 struct buffer_head
*head
, *bh
;
1717 struct pagevec pvec
;
1720 index
= logical
>> (PAGE_CACHE_SHIFT
- inode
->i_blkbits
);
1721 end
= (logical
+ blocks
- 1) >> (PAGE_CACHE_SHIFT
- inode
->i_blkbits
);
1722 cur_logical
= index
<< (PAGE_CACHE_SHIFT
- inode
->i_blkbits
);
1724 pagevec_init(&pvec
, 0);
1726 while (index
<= end
) {
1727 /* XXX: optimize tail */
1728 nr_pages
= pagevec_lookup(&pvec
, mapping
, index
, PAGEVEC_SIZE
);
1731 for (i
= 0; i
< nr_pages
; i
++) {
1732 struct page
*page
= pvec
.pages
[i
];
1734 index
= page
->index
;
1739 BUG_ON(!PageLocked(page
));
1740 BUG_ON(PageWriteback(page
));
1741 BUG_ON(!page_has_buffers(page
));
1743 bh
= page_buffers(page
);
1746 /* skip blocks out of the range */
1748 if (cur_logical
>= logical
)
1751 } while ((bh
= bh
->b_this_page
) != head
);
1754 if (cur_logical
>= logical
+ blocks
)
1756 if (buffer_delay(bh
)) {
1757 bh
->b_blocknr
= pblock
;
1758 clear_buffer_delay(bh
);
1759 bh
->b_bdev
= inode
->i_sb
->s_bdev
;
1760 } else if (buffer_unwritten(bh
)) {
1761 bh
->b_blocknr
= pblock
;
1762 clear_buffer_unwritten(bh
);
1763 set_buffer_mapped(bh
);
1765 bh
->b_bdev
= inode
->i_sb
->s_bdev
;
1766 } else if (buffer_mapped(bh
))
1767 BUG_ON(bh
->b_blocknr
!= pblock
);
1771 } while ((bh
= bh
->b_this_page
) != head
);
1773 pagevec_release(&pvec
);
1779 * __unmap_underlying_blocks - just a helper function to unmap
1780 * set of blocks described by @bh
1782 static inline void __unmap_underlying_blocks(struct inode
*inode
,
1783 struct buffer_head
*bh
)
1785 struct block_device
*bdev
= inode
->i_sb
->s_bdev
;
1788 blocks
= bh
->b_size
>> inode
->i_blkbits
;
1789 for (i
= 0; i
< blocks
; i
++)
1790 unmap_underlying_metadata(bdev
, bh
->b_blocknr
+ i
);
1793 static void ext4_da_block_invalidatepages(struct mpage_da_data
*mpd
,
1794 sector_t logical
, long blk_cnt
)
1798 struct pagevec pvec
;
1799 struct inode
*inode
= mpd
->inode
;
1800 struct address_space
*mapping
= inode
->i_mapping
;
1802 index
= logical
>> (PAGE_CACHE_SHIFT
- inode
->i_blkbits
);
1803 end
= (logical
+ blk_cnt
- 1) >>
1804 (PAGE_CACHE_SHIFT
- inode
->i_blkbits
);
1805 while (index
<= end
) {
1806 nr_pages
= pagevec_lookup(&pvec
, mapping
, index
, PAGEVEC_SIZE
);
1809 for (i
= 0; i
< nr_pages
; i
++) {
1810 struct page
*page
= pvec
.pages
[i
];
1811 index
= page
->index
;
1816 BUG_ON(!PageLocked(page
));
1817 BUG_ON(PageWriteback(page
));
1818 block_invalidatepage(page
, 0);
1819 ClearPageUptodate(page
);
1826 static void ext4_print_free_blocks(struct inode
*inode
)
1828 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
1829 printk(KERN_EMERG
"Total free blocks count %lld\n",
1830 ext4_count_free_blocks(inode
->i_sb
));
1831 printk(KERN_EMERG
"Free/Dirty block details\n");
1832 printk(KERN_EMERG
"free_blocks=%lld\n",
1833 percpu_counter_sum(&sbi
->s_freeblocks_counter
));
1834 printk(KERN_EMERG
"dirty_blocks=%lld\n",
1835 percpu_counter_sum(&sbi
->s_dirtyblocks_counter
));
1836 printk(KERN_EMERG
"Block reservation details\n");
1837 printk(KERN_EMERG
"i_reserved_data_blocks=%lu\n",
1838 EXT4_I(inode
)->i_reserved_data_blocks
);
1839 printk(KERN_EMERG
"i_reserved_meta_blocks=%lu\n",
1840 EXT4_I(inode
)->i_reserved_meta_blocks
);
1845 * mpage_da_map_blocks - go through given space
1847 * @mpd->lbh - bh describing space
1848 * @mpd->get_block - the filesystem's block mapper function
1850 * The function skips space we know is already mapped to disk blocks.
1853 static int mpage_da_map_blocks(struct mpage_da_data
*mpd
)
1856 struct buffer_head
new;
1857 struct buffer_head
*lbh
= &mpd
->lbh
;
1861 * We consider only non-mapped and non-allocated blocks
1863 if (buffer_mapped(lbh
) && !buffer_delay(lbh
))
1865 new.b_state
= lbh
->b_state
;
1867 new.b_size
= lbh
->b_size
;
1868 next
= lbh
->b_blocknr
;
1870 * If we didn't accumulate anything
1871 * to write simply return
1875 err
= mpd
->get_block(mpd
->inode
, next
, &new, 1);
1878 /* If get block returns with error
1879 * we simply return. Later writepage
1880 * will redirty the page and writepages
1881 * will find the dirty page again
1886 if (err
== -ENOSPC
&&
1887 ext4_count_free_blocks(mpd
->inode
->i_sb
)) {
1893 * get block failure will cause us
1894 * to loop in writepages. Because
1895 * a_ops->writepage won't be able to
1896 * make progress. The page will be redirtied
1897 * by writepage and writepages will again
1898 * try to write the same.
1900 printk(KERN_EMERG
"%s block allocation failed for inode %lu "
1901 "at logical offset %llu with max blocks "
1902 "%zd with error %d\n",
1903 __func__
, mpd
->inode
->i_ino
,
1904 (unsigned long long)next
,
1905 lbh
->b_size
>> mpd
->inode
->i_blkbits
, err
);
1906 printk(KERN_EMERG
"This should not happen.!! "
1907 "Data will be lost\n");
1908 if (err
== -ENOSPC
) {
1909 ext4_print_free_blocks(mpd
->inode
);
1911 /* invlaidate all the pages */
1912 ext4_da_block_invalidatepages(mpd
, next
,
1913 lbh
->b_size
>> mpd
->inode
->i_blkbits
);
1916 BUG_ON(new.b_size
== 0);
1918 if (buffer_new(&new))
1919 __unmap_underlying_blocks(mpd
->inode
, &new);
1922 * If blocks are delayed marked, we need to
1923 * put actual blocknr and drop delayed bit
1925 if (buffer_delay(lbh
) || buffer_unwritten(lbh
))
1926 mpage_put_bnr_to_bhs(mpd
, next
, &new);
1931 #define BH_FLAGS ((1 << BH_Uptodate) | (1 << BH_Mapped) | \
1932 (1 << BH_Delay) | (1 << BH_Unwritten))
1935 * mpage_add_bh_to_extent - try to add one more block to extent of blocks
1937 * @mpd->lbh - extent of blocks
1938 * @logical - logical number of the block in the file
1939 * @bh - bh of the block (used to access block's state)
1941 * the function is used to collect contig. blocks in same state
1943 static void mpage_add_bh_to_extent(struct mpage_da_data
*mpd
,
1944 sector_t logical
, struct buffer_head
*bh
)
1947 size_t b_size
= bh
->b_size
;
1948 struct buffer_head
*lbh
= &mpd
->lbh
;
1949 int nrblocks
= lbh
->b_size
>> mpd
->inode
->i_blkbits
;
1951 /* check if thereserved journal credits might overflow */
1952 if (!(EXT4_I(mpd
->inode
)->i_flags
& EXT4_EXTENTS_FL
)) {
1953 if (nrblocks
>= EXT4_MAX_TRANS_DATA
) {
1955 * With non-extent format we are limited by the journal
1956 * credit available. Total credit needed to insert
1957 * nrblocks contiguous blocks is dependent on the
1958 * nrblocks. So limit nrblocks.
1961 } else if ((nrblocks
+ (b_size
>> mpd
->inode
->i_blkbits
)) >
1962 EXT4_MAX_TRANS_DATA
) {
1964 * Adding the new buffer_head would make it cross the
1965 * allowed limit for which we have journal credit
1966 * reserved. So limit the new bh->b_size
1968 b_size
= (EXT4_MAX_TRANS_DATA
- nrblocks
) <<
1969 mpd
->inode
->i_blkbits
;
1970 /* we will do mpage_da_submit_io in the next loop */
1974 * First block in the extent
1976 if (lbh
->b_size
== 0) {
1977 lbh
->b_blocknr
= logical
;
1978 lbh
->b_size
= b_size
;
1979 lbh
->b_state
= bh
->b_state
& BH_FLAGS
;
1983 next
= lbh
->b_blocknr
+ nrblocks
;
1985 * Can we merge the block to our big extent?
1987 if (logical
== next
&& (bh
->b_state
& BH_FLAGS
) == lbh
->b_state
) {
1988 lbh
->b_size
+= b_size
;
1994 * We couldn't merge the block to our extent, so we
1995 * need to flush current extent and start new one
1997 if (mpage_da_map_blocks(mpd
) == 0)
1998 mpage_da_submit_io(mpd
);
2004 * __mpage_da_writepage - finds extent of pages and blocks
2006 * @page: page to consider
2007 * @wbc: not used, we just follow rules
2010 * The function finds extents of pages and scan them for all blocks.
2012 static int __mpage_da_writepage(struct page
*page
,
2013 struct writeback_control
*wbc
, void *data
)
2015 struct mpage_da_data
*mpd
= data
;
2016 struct inode
*inode
= mpd
->inode
;
2017 struct buffer_head
*bh
, *head
, fake
;
2022 * Rest of the page in the page_vec
2023 * redirty then and skip then. We will
2024 * try to to write them again after
2025 * starting a new transaction
2027 redirty_page_for_writepage(wbc
, page
);
2029 return MPAGE_DA_EXTENT_TAIL
;
2032 * Can we merge this page to current extent?
2034 if (mpd
->next_page
!= page
->index
) {
2036 * Nope, we can't. So, we map non-allocated blocks
2037 * and start IO on them using writepage()
2039 if (mpd
->next_page
!= mpd
->first_page
) {
2040 if (mpage_da_map_blocks(mpd
) == 0)
2041 mpage_da_submit_io(mpd
);
2043 * skip rest of the page in the page_vec
2046 redirty_page_for_writepage(wbc
, page
);
2048 return MPAGE_DA_EXTENT_TAIL
;
2052 * Start next extent of pages ...
2054 mpd
->first_page
= page
->index
;
2059 mpd
->lbh
.b_size
= 0;
2060 mpd
->lbh
.b_state
= 0;
2061 mpd
->lbh
.b_blocknr
= 0;
2064 mpd
->next_page
= page
->index
+ 1;
2065 logical
= (sector_t
) page
->index
<<
2066 (PAGE_CACHE_SHIFT
- inode
->i_blkbits
);
2068 if (!page_has_buffers(page
)) {
2070 * There is no attached buffer heads yet (mmap?)
2071 * we treat the page asfull of dirty blocks
2074 bh
->b_size
= PAGE_CACHE_SIZE
;
2076 set_buffer_dirty(bh
);
2077 set_buffer_uptodate(bh
);
2078 mpage_add_bh_to_extent(mpd
, logical
, bh
);
2080 return MPAGE_DA_EXTENT_TAIL
;
2083 * Page with regular buffer heads, just add all dirty ones
2085 head
= page_buffers(page
);
2088 BUG_ON(buffer_locked(bh
));
2089 if (buffer_dirty(bh
) &&
2090 (!buffer_mapped(bh
) || buffer_delay(bh
))) {
2091 mpage_add_bh_to_extent(mpd
, logical
, bh
);
2093 return MPAGE_DA_EXTENT_TAIL
;
2096 } while ((bh
= bh
->b_this_page
) != head
);
2103 * mpage_da_writepages - walk the list of dirty pages of the given
2104 * address space, allocates non-allocated blocks, maps newly-allocated
2105 * blocks to existing bhs and issue IO them
2107 * @mapping: address space structure to write
2108 * @wbc: subtract the number of written pages from *@wbc->nr_to_write
2109 * @get_block: the filesystem's block mapper function.
2111 * This is a library function, which implements the writepages()
2112 * address_space_operation.
2114 static int mpage_da_writepages(struct address_space
*mapping
,
2115 struct writeback_control
*wbc
,
2116 struct mpage_da_data
*mpd
)
2120 if (!mpd
->get_block
)
2121 return generic_writepages(mapping
, wbc
);
2123 mpd
->lbh
.b_size
= 0;
2124 mpd
->lbh
.b_state
= 0;
2125 mpd
->lbh
.b_blocknr
= 0;
2126 mpd
->first_page
= 0;
2129 mpd
->pages_written
= 0;
2132 ret
= write_cache_pages(mapping
, wbc
, __mpage_da_writepage
, mpd
);
2134 * Handle last extent of pages
2136 if (!mpd
->io_done
&& mpd
->next_page
!= mpd
->first_page
) {
2137 if (mpage_da_map_blocks(mpd
) == 0)
2138 mpage_da_submit_io(mpd
);
2141 ret
= MPAGE_DA_EXTENT_TAIL
;
2143 wbc
->nr_to_write
-= mpd
->pages_written
;
2148 * this is a special callback for ->write_begin() only
2149 * it's intention is to return mapped block or reserve space
2151 static int ext4_da_get_block_prep(struct inode
*inode
, sector_t iblock
,
2152 struct buffer_head
*bh_result
, int create
)
2156 BUG_ON(create
== 0);
2157 BUG_ON(bh_result
->b_size
!= inode
->i_sb
->s_blocksize
);
2160 * first, we need to know whether the block is allocated already
2161 * preallocated blocks are unmapped but should treated
2162 * the same as allocated blocks.
2164 ret
= ext4_get_blocks_wrap(NULL
, inode
, iblock
, 1, bh_result
, 0, 0, 0);
2165 if ((ret
== 0) && !buffer_delay(bh_result
)) {
2166 /* the block isn't (pre)allocated yet, let's reserve space */
2168 * XXX: __block_prepare_write() unmaps passed block,
2171 ret
= ext4_da_reserve_space(inode
, 1);
2173 /* not enough space to reserve */
2176 map_bh(bh_result
, inode
->i_sb
, 0);
2177 set_buffer_new(bh_result
);
2178 set_buffer_delay(bh_result
);
2179 } else if (ret
> 0) {
2180 bh_result
->b_size
= (ret
<< inode
->i_blkbits
);
2186 #define EXT4_DELALLOC_RSVED 1
2187 static int ext4_da_get_block_write(struct inode
*inode
, sector_t iblock
,
2188 struct buffer_head
*bh_result
, int create
)
2191 unsigned max_blocks
= bh_result
->b_size
>> inode
->i_blkbits
;
2192 loff_t disksize
= EXT4_I(inode
)->i_disksize
;
2193 handle_t
*handle
= NULL
;
2195 handle
= ext4_journal_current_handle();
2197 ret
= ext4_get_blocks_wrap(handle
, inode
, iblock
, max_blocks
,
2198 bh_result
, create
, 0, EXT4_DELALLOC_RSVED
);
2201 bh_result
->b_size
= (ret
<< inode
->i_blkbits
);
2203 if (ext4_should_order_data(inode
)) {
2205 retval
= ext4_jbd2_file_inode(handle
, inode
);
2208 * Failed to add inode for ordered
2209 * mode. Don't update file size
2215 * Update on-disk size along with block allocation
2216 * we don't use 'extend_disksize' as size may change
2217 * within already allocated block -bzzz
2219 disksize
= ((loff_t
) iblock
+ ret
) << inode
->i_blkbits
;
2220 if (disksize
> i_size_read(inode
))
2221 disksize
= i_size_read(inode
);
2222 if (disksize
> EXT4_I(inode
)->i_disksize
) {
2223 ext4_update_i_disksize(inode
, disksize
);
2224 ret
= ext4_mark_inode_dirty(handle
, inode
);
2232 static int ext4_bh_unmapped_or_delay(handle_t
*handle
, struct buffer_head
*bh
)
2235 * unmapped buffer is possible for holes.
2236 * delay buffer is possible with delayed allocation
2238 return ((!buffer_mapped(bh
) || buffer_delay(bh
)) && buffer_dirty(bh
));
2241 static int ext4_normal_get_block_write(struct inode
*inode
, sector_t iblock
,
2242 struct buffer_head
*bh_result
, int create
)
2245 unsigned max_blocks
= bh_result
->b_size
>> inode
->i_blkbits
;
2248 * we don't want to do block allocation in writepage
2249 * so call get_block_wrap with create = 0
2251 ret
= ext4_get_blocks_wrap(NULL
, inode
, iblock
, max_blocks
,
2252 bh_result
, 0, 0, 0);
2254 bh_result
->b_size
= (ret
<< inode
->i_blkbits
);
2261 * get called vi ext4_da_writepages after taking page lock (have journal handle)
2262 * get called via journal_submit_inode_data_buffers (no journal handle)
2263 * get called via shrink_page_list via pdflush (no journal handle)
2264 * or grab_page_cache when doing write_begin (have journal handle)
2266 static int ext4_da_writepage(struct page
*page
,
2267 struct writeback_control
*wbc
)
2272 struct buffer_head
*page_bufs
;
2273 struct inode
*inode
= page
->mapping
->host
;
2275 size
= i_size_read(inode
);
2276 if (page
->index
== size
>> PAGE_CACHE_SHIFT
)
2277 len
= size
& ~PAGE_CACHE_MASK
;
2279 len
= PAGE_CACHE_SIZE
;
2281 if (page_has_buffers(page
)) {
2282 page_bufs
= page_buffers(page
);
2283 if (walk_page_buffers(NULL
, page_bufs
, 0, len
, NULL
,
2284 ext4_bh_unmapped_or_delay
)) {
2286 * We don't want to do block allocation
2287 * So redirty the page and return
2288 * We may reach here when we do a journal commit
2289 * via journal_submit_inode_data_buffers.
2290 * If we don't have mapping block we just ignore
2291 * them. We can also reach here via shrink_page_list
2293 redirty_page_for_writepage(wbc
, page
);
2299 * The test for page_has_buffers() is subtle:
2300 * We know the page is dirty but it lost buffers. That means
2301 * that at some moment in time after write_begin()/write_end()
2302 * has been called all buffers have been clean and thus they
2303 * must have been written at least once. So they are all
2304 * mapped and we can happily proceed with mapping them
2305 * and writing the page.
2307 * Try to initialize the buffer_heads and check whether
2308 * all are mapped and non delay. We don't want to
2309 * do block allocation here.
2311 ret
= block_prepare_write(page
, 0, PAGE_CACHE_SIZE
,
2312 ext4_normal_get_block_write
);
2314 page_bufs
= page_buffers(page
);
2315 /* check whether all are mapped and non delay */
2316 if (walk_page_buffers(NULL
, page_bufs
, 0, len
, NULL
,
2317 ext4_bh_unmapped_or_delay
)) {
2318 redirty_page_for_writepage(wbc
, page
);
2324 * We can't do block allocation here
2325 * so just redity the page and unlock
2328 redirty_page_for_writepage(wbc
, page
);
2334 if (test_opt(inode
->i_sb
, NOBH
) && ext4_should_writeback_data(inode
))
2335 ret
= nobh_writepage(page
, ext4_normal_get_block_write
, wbc
);
2337 ret
= block_write_full_page(page
,
2338 ext4_normal_get_block_write
,
2345 * This is called via ext4_da_writepages() to
2346 * calulate the total number of credits to reserve to fit
2347 * a single extent allocation into a single transaction,
2348 * ext4_da_writpeages() will loop calling this before
2349 * the block allocation.
2352 static int ext4_da_writepages_trans_blocks(struct inode
*inode
)
2354 int max_blocks
= EXT4_I(inode
)->i_reserved_data_blocks
;
2357 * With non-extent format the journal credit needed to
2358 * insert nrblocks contiguous block is dependent on
2359 * number of contiguous block. So we will limit
2360 * number of contiguous block to a sane value
2362 if (!(inode
->i_flags
& EXT4_EXTENTS_FL
) &&
2363 (max_blocks
> EXT4_MAX_TRANS_DATA
))
2364 max_blocks
= EXT4_MAX_TRANS_DATA
;
2366 return ext4_chunk_trans_blocks(inode
, max_blocks
);
2369 static int ext4_da_writepages(struct address_space
*mapping
,
2370 struct writeback_control
*wbc
)
2373 int range_whole
= 0;
2374 handle_t
*handle
= NULL
;
2375 struct mpage_da_data mpd
;
2376 struct inode
*inode
= mapping
->host
;
2377 int no_nrwrite_index_update
;
2378 long pages_written
= 0, pages_skipped
;
2379 int needed_blocks
, ret
= 0, nr_to_writebump
= 0;
2380 struct ext4_sb_info
*sbi
= EXT4_SB(mapping
->host
->i_sb
);
2383 * No pages to write? This is mainly a kludge to avoid starting
2384 * a transaction for special inodes like journal inode on last iput()
2385 * because that could violate lock ordering on umount
2387 if (!mapping
->nrpages
|| !mapping_tagged(mapping
, PAGECACHE_TAG_DIRTY
))
2390 * Make sure nr_to_write is >= sbi->s_mb_stream_request
2391 * This make sure small files blocks are allocated in
2392 * single attempt. This ensure that small files
2393 * get less fragmented.
2395 if (wbc
->nr_to_write
< sbi
->s_mb_stream_request
) {
2396 nr_to_writebump
= sbi
->s_mb_stream_request
- wbc
->nr_to_write
;
2397 wbc
->nr_to_write
= sbi
->s_mb_stream_request
;
2399 if (wbc
->range_start
== 0 && wbc
->range_end
== LLONG_MAX
)
2402 if (wbc
->range_cyclic
)
2403 index
= mapping
->writeback_index
;
2405 index
= wbc
->range_start
>> PAGE_CACHE_SHIFT
;
2408 mpd
.inode
= mapping
->host
;
2411 * we don't want write_cache_pages to update
2412 * nr_to_write and writeback_index
2414 no_nrwrite_index_update
= wbc
->no_nrwrite_index_update
;
2415 wbc
->no_nrwrite_index_update
= 1;
2416 pages_skipped
= wbc
->pages_skipped
;
2418 while (!ret
&& wbc
->nr_to_write
> 0) {
2421 * we insert one extent at a time. So we need
2422 * credit needed for single extent allocation.
2423 * journalled mode is currently not supported
2426 BUG_ON(ext4_should_journal_data(inode
));
2427 needed_blocks
= ext4_da_writepages_trans_blocks(inode
);
2429 /* start a new transaction*/
2430 handle
= ext4_journal_start(inode
, needed_blocks
);
2431 if (IS_ERR(handle
)) {
2432 ret
= PTR_ERR(handle
);
2433 printk(KERN_EMERG
"%s: jbd2_start: "
2434 "%ld pages, ino %lu; err %d\n", __func__
,
2435 wbc
->nr_to_write
, inode
->i_ino
, ret
);
2437 goto out_writepages
;
2439 mpd
.get_block
= ext4_da_get_block_write
;
2440 ret
= mpage_da_writepages(mapping
, wbc
, &mpd
);
2442 ext4_journal_stop(handle
);
2444 if (mpd
.retval
== -ENOSPC
) {
2445 /* commit the transaction which would
2446 * free blocks released in the transaction
2449 jbd2_journal_force_commit_nested(sbi
->s_journal
);
2450 wbc
->pages_skipped
= pages_skipped
;
2452 } else if (ret
== MPAGE_DA_EXTENT_TAIL
) {
2454 * got one extent now try with
2457 pages_written
+= mpd
.pages_written
;
2458 wbc
->pages_skipped
= pages_skipped
;
2460 } else if (wbc
->nr_to_write
)
2462 * There is no more writeout needed
2463 * or we requested for a noblocking writeout
2464 * and we found the device congested
2468 if (pages_skipped
!= wbc
->pages_skipped
)
2469 printk(KERN_EMERG
"This should not happen leaving %s "
2470 "with nr_to_write = %ld ret = %d\n",
2471 __func__
, wbc
->nr_to_write
, ret
);
2474 index
+= pages_written
;
2475 if (wbc
->range_cyclic
|| (range_whole
&& wbc
->nr_to_write
> 0))
2477 * set the writeback_index so that range_cyclic
2478 * mode will write it back later
2480 mapping
->writeback_index
= index
;
2483 if (!no_nrwrite_index_update
)
2484 wbc
->no_nrwrite_index_update
= 0;
2485 wbc
->nr_to_write
-= nr_to_writebump
;
2489 #define FALL_BACK_TO_NONDELALLOC 1
2490 static int ext4_nonda_switch(struct super_block
*sb
)
2492 s64 free_blocks
, dirty_blocks
;
2493 struct ext4_sb_info
*sbi
= EXT4_SB(sb
);
2496 * switch to non delalloc mode if we are running low
2497 * on free block. The free block accounting via percpu
2498 * counters can get slightly wrong with FBC_BATCH getting
2499 * accumulated on each CPU without updating global counters
2500 * Delalloc need an accurate free block accounting. So switch
2501 * to non delalloc when we are near to error range.
2503 free_blocks
= percpu_counter_read_positive(&sbi
->s_freeblocks_counter
);
2504 dirty_blocks
= percpu_counter_read_positive(&sbi
->s_dirtyblocks_counter
);
2505 if (2 * free_blocks
< 3 * dirty_blocks
||
2506 free_blocks
< (dirty_blocks
+ EXT4_FREEBLOCKS_WATERMARK
)) {
2508 * free block count is less that 150% of dirty blocks
2509 * or free blocks is less that watermark
2516 static int ext4_da_write_begin(struct file
*file
, struct address_space
*mapping
,
2517 loff_t pos
, unsigned len
, unsigned flags
,
2518 struct page
**pagep
, void **fsdata
)
2520 int ret
, retries
= 0;
2524 struct inode
*inode
= mapping
->host
;
2527 index
= pos
>> PAGE_CACHE_SHIFT
;
2528 from
= pos
& (PAGE_CACHE_SIZE
- 1);
2531 if (ext4_nonda_switch(inode
->i_sb
)) {
2532 *fsdata
= (void *)FALL_BACK_TO_NONDELALLOC
;
2533 return ext4_write_begin(file
, mapping
, pos
,
2534 len
, flags
, pagep
, fsdata
);
2536 *fsdata
= (void *)0;
2539 * With delayed allocation, we don't log the i_disksize update
2540 * if there is delayed block allocation. But we still need
2541 * to journalling the i_disksize update if writes to the end
2542 * of file which has an already mapped buffer.
2544 handle
= ext4_journal_start(inode
, 1);
2545 if (IS_ERR(handle
)) {
2546 ret
= PTR_ERR(handle
);
2550 page
= __grab_cache_page(mapping
, index
);
2552 ext4_journal_stop(handle
);
2558 ret
= block_write_begin(file
, mapping
, pos
, len
, flags
, pagep
, fsdata
,
2559 ext4_da_get_block_prep
);
2562 ext4_journal_stop(handle
);
2563 page_cache_release(page
);
2565 * block_write_begin may have instantiated a few blocks
2566 * outside i_size. Trim these off again. Don't need
2567 * i_size_read because we hold i_mutex.
2569 if (pos
+ len
> inode
->i_size
)
2570 vmtruncate(inode
, inode
->i_size
);
2573 if (ret
== -ENOSPC
&& ext4_should_retry_alloc(inode
->i_sb
, &retries
))
2580 * Check if we should update i_disksize
2581 * when write to the end of file but not require block allocation
2583 static int ext4_da_should_update_i_disksize(struct page
*page
,
2584 unsigned long offset
)
2586 struct buffer_head
*bh
;
2587 struct inode
*inode
= page
->mapping
->host
;
2591 bh
= page_buffers(page
);
2592 idx
= offset
>> inode
->i_blkbits
;
2594 for (i
= 0; i
< idx
; i
++)
2595 bh
= bh
->b_this_page
;
2597 if (!buffer_mapped(bh
) || (buffer_delay(bh
)))
2602 static int ext4_da_write_end(struct file
*file
,
2603 struct address_space
*mapping
,
2604 loff_t pos
, unsigned len
, unsigned copied
,
2605 struct page
*page
, void *fsdata
)
2607 struct inode
*inode
= mapping
->host
;
2609 handle_t
*handle
= ext4_journal_current_handle();
2611 unsigned long start
, end
;
2612 int write_mode
= (int)(unsigned long)fsdata
;
2614 if (write_mode
== FALL_BACK_TO_NONDELALLOC
) {
2615 if (ext4_should_order_data(inode
)) {
2616 return ext4_ordered_write_end(file
, mapping
, pos
,
2617 len
, copied
, page
, fsdata
);
2618 } else if (ext4_should_writeback_data(inode
)) {
2619 return ext4_writeback_write_end(file
, mapping
, pos
,
2620 len
, copied
, page
, fsdata
);
2626 start
= pos
& (PAGE_CACHE_SIZE
- 1);
2627 end
= start
+ copied
- 1;
2630 * generic_write_end() will run mark_inode_dirty() if i_size
2631 * changes. So let's piggyback the i_disksize mark_inode_dirty
2635 new_i_size
= pos
+ copied
;
2636 if (new_i_size
> EXT4_I(inode
)->i_disksize
) {
2637 if (ext4_da_should_update_i_disksize(page
, end
)) {
2638 down_write(&EXT4_I(inode
)->i_data_sem
);
2639 if (new_i_size
> EXT4_I(inode
)->i_disksize
) {
2641 * Updating i_disksize when extending file
2642 * without needing block allocation
2644 if (ext4_should_order_data(inode
))
2645 ret
= ext4_jbd2_file_inode(handle
,
2648 EXT4_I(inode
)->i_disksize
= new_i_size
;
2650 up_write(&EXT4_I(inode
)->i_data_sem
);
2651 /* We need to mark inode dirty even if
2652 * new_i_size is less that inode->i_size
2653 * bu greater than i_disksize.(hint delalloc)
2655 ext4_mark_inode_dirty(handle
, inode
);
2658 ret2
= generic_write_end(file
, mapping
, pos
, len
, copied
,
2663 ret2
= ext4_journal_stop(handle
);
2667 return ret
? ret
: copied
;
2670 static void ext4_da_invalidatepage(struct page
*page
, unsigned long offset
)
2673 * Drop reserved blocks
2675 BUG_ON(!PageLocked(page
));
2676 if (!page_has_buffers(page
))
2679 ext4_da_page_release_reservation(page
, offset
);
2682 ext4_invalidatepage(page
, offset
);
2689 * bmap() is special. It gets used by applications such as lilo and by
2690 * the swapper to find the on-disk block of a specific piece of data.
2692 * Naturally, this is dangerous if the block concerned is still in the
2693 * journal. If somebody makes a swapfile on an ext4 data-journaling
2694 * filesystem and enables swap, then they may get a nasty shock when the
2695 * data getting swapped to that swapfile suddenly gets overwritten by
2696 * the original zero's written out previously to the journal and
2697 * awaiting writeback in the kernel's buffer cache.
2699 * So, if we see any bmap calls here on a modified, data-journaled file,
2700 * take extra steps to flush any blocks which might be in the cache.
2702 static sector_t
ext4_bmap(struct address_space
*mapping
, sector_t block
)
2704 struct inode
*inode
= mapping
->host
;
2708 if (mapping_tagged(mapping
, PAGECACHE_TAG_DIRTY
) &&
2709 test_opt(inode
->i_sb
, DELALLOC
)) {
2711 * With delalloc we want to sync the file
2712 * so that we can make sure we allocate
2715 filemap_write_and_wait(mapping
);
2718 if (EXT4_I(inode
)->i_state
& EXT4_STATE_JDATA
) {
2720 * This is a REALLY heavyweight approach, but the use of
2721 * bmap on dirty files is expected to be extremely rare:
2722 * only if we run lilo or swapon on a freshly made file
2723 * do we expect this to happen.
2725 * (bmap requires CAP_SYS_RAWIO so this does not
2726 * represent an unprivileged user DOS attack --- we'd be
2727 * in trouble if mortal users could trigger this path at
2730 * NB. EXT4_STATE_JDATA is not set on files other than
2731 * regular files. If somebody wants to bmap a directory
2732 * or symlink and gets confused because the buffer
2733 * hasn't yet been flushed to disk, they deserve
2734 * everything they get.
2737 EXT4_I(inode
)->i_state
&= ~EXT4_STATE_JDATA
;
2738 journal
= EXT4_JOURNAL(inode
);
2739 jbd2_journal_lock_updates(journal
);
2740 err
= jbd2_journal_flush(journal
);
2741 jbd2_journal_unlock_updates(journal
);
2747 return generic_block_bmap(mapping
, block
, ext4_get_block
);
2750 static int bget_one(handle_t
*handle
, struct buffer_head
*bh
)
2756 static int bput_one(handle_t
*handle
, struct buffer_head
*bh
)
2763 * Note that we don't need to start a transaction unless we're journaling data
2764 * because we should have holes filled from ext4_page_mkwrite(). We even don't
2765 * need to file the inode to the transaction's list in ordered mode because if
2766 * we are writing back data added by write(), the inode is already there and if
2767 * we are writing back data modified via mmap(), noone guarantees in which
2768 * transaction the data will hit the disk. In case we are journaling data, we
2769 * cannot start transaction directly because transaction start ranks above page
2770 * lock so we have to do some magic.
2772 * In all journaling modes block_write_full_page() will start the I/O.
2776 * ext4_writepage() -> kmalloc() -> __alloc_pages() -> page_launder() ->
2781 * ext4_file_write() -> generic_file_write() -> __alloc_pages() -> ...
2783 * Same applies to ext4_get_block(). We will deadlock on various things like
2784 * lock_journal and i_data_sem
2786 * Setting PF_MEMALLOC here doesn't work - too many internal memory
2789 * 16May01: If we're reentered then journal_current_handle() will be
2790 * non-zero. We simply *return*.
2792 * 1 July 2001: @@@ FIXME:
2793 * In journalled data mode, a data buffer may be metadata against the
2794 * current transaction. But the same file is part of a shared mapping
2795 * and someone does a writepage() on it.
2797 * We will move the buffer onto the async_data list, but *after* it has
2798 * been dirtied. So there's a small window where we have dirty data on
2801 * Note that this only applies to the last partial page in the file. The
2802 * bit which block_write_full_page() uses prepare/commit for. (That's
2803 * broken code anyway: it's wrong for msync()).
2805 * It's a rare case: affects the final partial page, for journalled data
2806 * where the file is subject to bith write() and writepage() in the same
2807 * transction. To fix it we'll need a custom block_write_full_page().
2808 * We'll probably need that anyway for journalling writepage() output.
2810 * We don't honour synchronous mounts for writepage(). That would be
2811 * disastrous. Any write() or metadata operation will sync the fs for
2815 static int __ext4_normal_writepage(struct page
*page
,
2816 struct writeback_control
*wbc
)
2818 struct inode
*inode
= page
->mapping
->host
;
2820 if (test_opt(inode
->i_sb
, NOBH
))
2821 return nobh_writepage(page
,
2822 ext4_normal_get_block_write
, wbc
);
2824 return block_write_full_page(page
,
2825 ext4_normal_get_block_write
,
2829 static int ext4_normal_writepage(struct page
*page
,
2830 struct writeback_control
*wbc
)
2832 struct inode
*inode
= page
->mapping
->host
;
2833 loff_t size
= i_size_read(inode
);
2836 J_ASSERT(PageLocked(page
));
2837 if (page
->index
== size
>> PAGE_CACHE_SHIFT
)
2838 len
= size
& ~PAGE_CACHE_MASK
;
2840 len
= PAGE_CACHE_SIZE
;
2842 if (page_has_buffers(page
)) {
2843 /* if page has buffers it should all be mapped
2844 * and allocated. If there are not buffers attached
2845 * to the page we know the page is dirty but it lost
2846 * buffers. That means that at some moment in time
2847 * after write_begin() / write_end() has been called
2848 * all buffers have been clean and thus they must have been
2849 * written at least once. So they are all mapped and we can
2850 * happily proceed with mapping them and writing the page.
2852 BUG_ON(walk_page_buffers(NULL
, page_buffers(page
), 0, len
, NULL
,
2853 ext4_bh_unmapped_or_delay
));
2856 if (!ext4_journal_current_handle())
2857 return __ext4_normal_writepage(page
, wbc
);
2859 redirty_page_for_writepage(wbc
, page
);
2864 static int __ext4_journalled_writepage(struct page
*page
,
2865 struct writeback_control
*wbc
)
2867 struct address_space
*mapping
= page
->mapping
;
2868 struct inode
*inode
= mapping
->host
;
2869 struct buffer_head
*page_bufs
;
2870 handle_t
*handle
= NULL
;
2874 ret
= block_prepare_write(page
, 0, PAGE_CACHE_SIZE
,
2875 ext4_normal_get_block_write
);
2879 page_bufs
= page_buffers(page
);
2880 walk_page_buffers(handle
, page_bufs
, 0, PAGE_CACHE_SIZE
, NULL
,
2882 /* As soon as we unlock the page, it can go away, but we have
2883 * references to buffers so we are safe */
2886 handle
= ext4_journal_start(inode
, ext4_writepage_trans_blocks(inode
));
2887 if (IS_ERR(handle
)) {
2888 ret
= PTR_ERR(handle
);
2892 ret
= walk_page_buffers(handle
, page_bufs
, 0,
2893 PAGE_CACHE_SIZE
, NULL
, do_journal_get_write_access
);
2895 err
= walk_page_buffers(handle
, page_bufs
, 0,
2896 PAGE_CACHE_SIZE
, NULL
, write_end_fn
);
2899 err
= ext4_journal_stop(handle
);
2903 walk_page_buffers(handle
, page_bufs
, 0,
2904 PAGE_CACHE_SIZE
, NULL
, bput_one
);
2905 EXT4_I(inode
)->i_state
|= EXT4_STATE_JDATA
;
2914 static int ext4_journalled_writepage(struct page
*page
,
2915 struct writeback_control
*wbc
)
2917 struct inode
*inode
= page
->mapping
->host
;
2918 loff_t size
= i_size_read(inode
);
2921 J_ASSERT(PageLocked(page
));
2922 if (page
->index
== size
>> PAGE_CACHE_SHIFT
)
2923 len
= size
& ~PAGE_CACHE_MASK
;
2925 len
= PAGE_CACHE_SIZE
;
2927 if (page_has_buffers(page
)) {
2928 /* if page has buffers it should all be mapped
2929 * and allocated. If there are not buffers attached
2930 * to the page we know the page is dirty but it lost
2931 * buffers. That means that at some moment in time
2932 * after write_begin() / write_end() has been called
2933 * all buffers have been clean and thus they must have been
2934 * written at least once. So they are all mapped and we can
2935 * happily proceed with mapping them and writing the page.
2937 BUG_ON(walk_page_buffers(NULL
, page_buffers(page
), 0, len
, NULL
,
2938 ext4_bh_unmapped_or_delay
));
2941 if (ext4_journal_current_handle())
2944 if (PageChecked(page
)) {
2946 * It's mmapped pagecache. Add buffers and journal it. There
2947 * doesn't seem much point in redirtying the page here.
2949 ClearPageChecked(page
);
2950 return __ext4_journalled_writepage(page
, wbc
);
2953 * It may be a page full of checkpoint-mode buffers. We don't
2954 * really know unless we go poke around in the buffer_heads.
2955 * But block_write_full_page will do the right thing.
2957 return block_write_full_page(page
,
2958 ext4_normal_get_block_write
,
2962 redirty_page_for_writepage(wbc
, page
);
2967 static int ext4_readpage(struct file
*file
, struct page
*page
)
2969 return mpage_readpage(page
, ext4_get_block
);
2973 ext4_readpages(struct file
*file
, struct address_space
*mapping
,
2974 struct list_head
*pages
, unsigned nr_pages
)
2976 return mpage_readpages(mapping
, pages
, nr_pages
, ext4_get_block
);
2979 static void ext4_invalidatepage(struct page
*page
, unsigned long offset
)
2981 journal_t
*journal
= EXT4_JOURNAL(page
->mapping
->host
);
2984 * If it's a full truncate we just forget about the pending dirtying
2987 ClearPageChecked(page
);
2989 jbd2_journal_invalidatepage(journal
, page
, offset
);
2992 static int ext4_releasepage(struct page
*page
, gfp_t wait
)
2994 journal_t
*journal
= EXT4_JOURNAL(page
->mapping
->host
);
2996 WARN_ON(PageChecked(page
));
2997 if (!page_has_buffers(page
))
2999 return jbd2_journal_try_to_free_buffers(journal
, page
, wait
);
3003 * If the O_DIRECT write will extend the file then add this inode to the
3004 * orphan list. So recovery will truncate it back to the original size
3005 * if the machine crashes during the write.
3007 * If the O_DIRECT write is intantiating holes inside i_size and the machine
3008 * crashes then stale disk data _may_ be exposed inside the file. But current
3009 * VFS code falls back into buffered path in that case so we are safe.
3011 static ssize_t
ext4_direct_IO(int rw
, struct kiocb
*iocb
,
3012 const struct iovec
*iov
, loff_t offset
,
3013 unsigned long nr_segs
)
3015 struct file
*file
= iocb
->ki_filp
;
3016 struct inode
*inode
= file
->f_mapping
->host
;
3017 struct ext4_inode_info
*ei
= EXT4_I(inode
);
3021 size_t count
= iov_length(iov
, nr_segs
);
3024 loff_t final_size
= offset
+ count
;
3026 if (final_size
> inode
->i_size
) {
3027 /* Credits for sb + inode write */
3028 handle
= ext4_journal_start(inode
, 2);
3029 if (IS_ERR(handle
)) {
3030 ret
= PTR_ERR(handle
);
3033 ret
= ext4_orphan_add(handle
, inode
);
3035 ext4_journal_stop(handle
);
3039 ei
->i_disksize
= inode
->i_size
;
3040 ext4_journal_stop(handle
);
3044 ret
= blockdev_direct_IO(rw
, iocb
, inode
, inode
->i_sb
->s_bdev
, iov
,
3046 ext4_get_block
, NULL
);
3051 /* Credits for sb + inode write */
3052 handle
= ext4_journal_start(inode
, 2);
3053 if (IS_ERR(handle
)) {
3054 /* This is really bad luck. We've written the data
3055 * but cannot extend i_size. Bail out and pretend
3056 * the write failed... */
3057 ret
= PTR_ERR(handle
);
3061 ext4_orphan_del(handle
, inode
);
3063 loff_t end
= offset
+ ret
;
3064 if (end
> inode
->i_size
) {
3065 ei
->i_disksize
= end
;
3066 i_size_write(inode
, end
);
3068 * We're going to return a positive `ret'
3069 * here due to non-zero-length I/O, so there's
3070 * no way of reporting error returns from
3071 * ext4_mark_inode_dirty() to userspace. So
3074 ext4_mark_inode_dirty(handle
, inode
);
3077 err
= ext4_journal_stop(handle
);
3086 * Pages can be marked dirty completely asynchronously from ext4's journalling
3087 * activity. By filemap_sync_pte(), try_to_unmap_one(), etc. We cannot do
3088 * much here because ->set_page_dirty is called under VFS locks. The page is
3089 * not necessarily locked.
3091 * We cannot just dirty the page and leave attached buffers clean, because the
3092 * buffers' dirty state is "definitive". We cannot just set the buffers dirty
3093 * or jbddirty because all the journalling code will explode.
3095 * So what we do is to mark the page "pending dirty" and next time writepage
3096 * is called, propagate that into the buffers appropriately.
3098 static int ext4_journalled_set_page_dirty(struct page
*page
)
3100 SetPageChecked(page
);
3101 return __set_page_dirty_nobuffers(page
);
3104 static const struct address_space_operations ext4_ordered_aops
= {
3105 .readpage
= ext4_readpage
,
3106 .readpages
= ext4_readpages
,
3107 .writepage
= ext4_normal_writepage
,
3108 .sync_page
= block_sync_page
,
3109 .write_begin
= ext4_write_begin
,
3110 .write_end
= ext4_ordered_write_end
,
3112 .invalidatepage
= ext4_invalidatepage
,
3113 .releasepage
= ext4_releasepage
,
3114 .direct_IO
= ext4_direct_IO
,
3115 .migratepage
= buffer_migrate_page
,
3116 .is_partially_uptodate
= block_is_partially_uptodate
,
3119 static const struct address_space_operations ext4_writeback_aops
= {
3120 .readpage
= ext4_readpage
,
3121 .readpages
= ext4_readpages
,
3122 .writepage
= ext4_normal_writepage
,
3123 .sync_page
= block_sync_page
,
3124 .write_begin
= ext4_write_begin
,
3125 .write_end
= ext4_writeback_write_end
,
3127 .invalidatepage
= ext4_invalidatepage
,
3128 .releasepage
= ext4_releasepage
,
3129 .direct_IO
= ext4_direct_IO
,
3130 .migratepage
= buffer_migrate_page
,
3131 .is_partially_uptodate
= block_is_partially_uptodate
,
3134 static const struct address_space_operations ext4_journalled_aops
= {
3135 .readpage
= ext4_readpage
,
3136 .readpages
= ext4_readpages
,
3137 .writepage
= ext4_journalled_writepage
,
3138 .sync_page
= block_sync_page
,
3139 .write_begin
= ext4_write_begin
,
3140 .write_end
= ext4_journalled_write_end
,
3141 .set_page_dirty
= ext4_journalled_set_page_dirty
,
3143 .invalidatepage
= ext4_invalidatepage
,
3144 .releasepage
= ext4_releasepage
,
3145 .is_partially_uptodate
= block_is_partially_uptodate
,
3148 static const struct address_space_operations ext4_da_aops
= {
3149 .readpage
= ext4_readpage
,
3150 .readpages
= ext4_readpages
,
3151 .writepage
= ext4_da_writepage
,
3152 .writepages
= ext4_da_writepages
,
3153 .sync_page
= block_sync_page
,
3154 .write_begin
= ext4_da_write_begin
,
3155 .write_end
= ext4_da_write_end
,
3157 .invalidatepage
= ext4_da_invalidatepage
,
3158 .releasepage
= ext4_releasepage
,
3159 .direct_IO
= ext4_direct_IO
,
3160 .migratepage
= buffer_migrate_page
,
3161 .is_partially_uptodate
= block_is_partially_uptodate
,
3164 void ext4_set_aops(struct inode
*inode
)
3166 if (ext4_should_order_data(inode
) &&
3167 test_opt(inode
->i_sb
, DELALLOC
))
3168 inode
->i_mapping
->a_ops
= &ext4_da_aops
;
3169 else if (ext4_should_order_data(inode
))
3170 inode
->i_mapping
->a_ops
= &ext4_ordered_aops
;
3171 else if (ext4_should_writeback_data(inode
) &&
3172 test_opt(inode
->i_sb
, DELALLOC
))
3173 inode
->i_mapping
->a_ops
= &ext4_da_aops
;
3174 else if (ext4_should_writeback_data(inode
))
3175 inode
->i_mapping
->a_ops
= &ext4_writeback_aops
;
3177 inode
->i_mapping
->a_ops
= &ext4_journalled_aops
;
3181 * ext4_block_truncate_page() zeroes out a mapping from file offset `from'
3182 * up to the end of the block which corresponds to `from'.
3183 * This required during truncate. We need to physically zero the tail end
3184 * of that block so it doesn't yield old data if the file is later grown.
3186 int ext4_block_truncate_page(handle_t
*handle
,
3187 struct address_space
*mapping
, loff_t from
)
3189 ext4_fsblk_t index
= from
>> PAGE_CACHE_SHIFT
;
3190 unsigned offset
= from
& (PAGE_CACHE_SIZE
-1);
3191 unsigned blocksize
, length
, pos
;
3193 struct inode
*inode
= mapping
->host
;
3194 struct buffer_head
*bh
;
3198 page
= grab_cache_page(mapping
, from
>> PAGE_CACHE_SHIFT
);
3202 blocksize
= inode
->i_sb
->s_blocksize
;
3203 length
= blocksize
- (offset
& (blocksize
- 1));
3204 iblock
= index
<< (PAGE_CACHE_SHIFT
- inode
->i_sb
->s_blocksize_bits
);
3207 * For "nobh" option, we can only work if we don't need to
3208 * read-in the page - otherwise we create buffers to do the IO.
3210 if (!page_has_buffers(page
) && test_opt(inode
->i_sb
, NOBH
) &&
3211 ext4_should_writeback_data(inode
) && PageUptodate(page
)) {
3212 zero_user(page
, offset
, length
);
3213 set_page_dirty(page
);
3217 if (!page_has_buffers(page
))
3218 create_empty_buffers(page
, blocksize
, 0);
3220 /* Find the buffer that contains "offset" */
3221 bh
= page_buffers(page
);
3223 while (offset
>= pos
) {
3224 bh
= bh
->b_this_page
;
3230 if (buffer_freed(bh
)) {
3231 BUFFER_TRACE(bh
, "freed: skip");
3235 if (!buffer_mapped(bh
)) {
3236 BUFFER_TRACE(bh
, "unmapped");
3237 ext4_get_block(inode
, iblock
, bh
, 0);
3238 /* unmapped? It's a hole - nothing to do */
3239 if (!buffer_mapped(bh
)) {
3240 BUFFER_TRACE(bh
, "still unmapped");
3245 /* Ok, it's mapped. Make sure it's up-to-date */
3246 if (PageUptodate(page
))
3247 set_buffer_uptodate(bh
);
3249 if (!buffer_uptodate(bh
)) {
3251 ll_rw_block(READ
, 1, &bh
);
3253 /* Uhhuh. Read error. Complain and punt. */
3254 if (!buffer_uptodate(bh
))
3258 if (ext4_should_journal_data(inode
)) {
3259 BUFFER_TRACE(bh
, "get write access");
3260 err
= ext4_journal_get_write_access(handle
, bh
);
3265 zero_user(page
, offset
, length
);
3267 BUFFER_TRACE(bh
, "zeroed end of block");
3270 if (ext4_should_journal_data(inode
)) {
3271 err
= ext4_journal_dirty_metadata(handle
, bh
);
3273 if (ext4_should_order_data(inode
))
3274 err
= ext4_jbd2_file_inode(handle
, inode
);
3275 mark_buffer_dirty(bh
);
3280 page_cache_release(page
);
3285 * Probably it should be a library function... search for first non-zero word
3286 * or memcmp with zero_page, whatever is better for particular architecture.
3289 static inline int all_zeroes(__le32
*p
, __le32
*q
)
3298 * ext4_find_shared - find the indirect blocks for partial truncation.
3299 * @inode: inode in question
3300 * @depth: depth of the affected branch
3301 * @offsets: offsets of pointers in that branch (see ext4_block_to_path)
3302 * @chain: place to store the pointers to partial indirect blocks
3303 * @top: place to the (detached) top of branch
3305 * This is a helper function used by ext4_truncate().
3307 * When we do truncate() we may have to clean the ends of several
3308 * indirect blocks but leave the blocks themselves alive. Block is
3309 * partially truncated if some data below the new i_size is refered
3310 * from it (and it is on the path to the first completely truncated
3311 * data block, indeed). We have to free the top of that path along
3312 * with everything to the right of the path. Since no allocation
3313 * past the truncation point is possible until ext4_truncate()
3314 * finishes, we may safely do the latter, but top of branch may
3315 * require special attention - pageout below the truncation point
3316 * might try to populate it.
3318 * We atomically detach the top of branch from the tree, store the
3319 * block number of its root in *@top, pointers to buffer_heads of
3320 * partially truncated blocks - in @chain[].bh and pointers to
3321 * their last elements that should not be removed - in
3322 * @chain[].p. Return value is the pointer to last filled element
3325 * The work left to caller to do the actual freeing of subtrees:
3326 * a) free the subtree starting from *@top
3327 * b) free the subtrees whose roots are stored in
3328 * (@chain[i].p+1 .. end of @chain[i].bh->b_data)
3329 * c) free the subtrees growing from the inode past the @chain[0].
3330 * (no partially truncated stuff there). */
3332 static Indirect
*ext4_find_shared(struct inode
*inode
, int depth
,
3333 ext4_lblk_t offsets
[4], Indirect chain
[4], __le32
*top
)
3335 Indirect
*partial
, *p
;
3339 /* Make k index the deepest non-null offest + 1 */
3340 for (k
= depth
; k
> 1 && !offsets
[k
-1]; k
--)
3342 partial
= ext4_get_branch(inode
, k
, offsets
, chain
, &err
);
3343 /* Writer: pointers */
3345 partial
= chain
+ k
-1;
3347 * If the branch acquired continuation since we've looked at it -
3348 * fine, it should all survive and (new) top doesn't belong to us.
3350 if (!partial
->key
&& *partial
->p
)
3353 for (p
= partial
; (p
> chain
) && all_zeroes((__le32
*) p
->bh
->b_data
, p
->p
); p
--)
3356 * OK, we've found the last block that must survive. The rest of our
3357 * branch should be detached before unlocking. However, if that rest
3358 * of branch is all ours and does not grow immediately from the inode
3359 * it's easier to cheat and just decrement partial->p.
3361 if (p
== chain
+ k
- 1 && p
> chain
) {
3365 /* Nope, don't do this in ext4. Must leave the tree intact */
3372 while (partial
> p
) {
3373 brelse(partial
->bh
);
3381 * Zero a number of block pointers in either an inode or an indirect block.
3382 * If we restart the transaction we must again get write access to the
3383 * indirect block for further modification.
3385 * We release `count' blocks on disk, but (last - first) may be greater
3386 * than `count' because there can be holes in there.
3388 static void ext4_clear_blocks(handle_t
*handle
, struct inode
*inode
,
3389 struct buffer_head
*bh
, ext4_fsblk_t block_to_free
,
3390 unsigned long count
, __le32
*first
, __le32
*last
)
3393 if (try_to_extend_transaction(handle
, inode
)) {
3395 BUFFER_TRACE(bh
, "call ext4_journal_dirty_metadata");
3396 ext4_journal_dirty_metadata(handle
, bh
);
3398 ext4_mark_inode_dirty(handle
, inode
);
3399 ext4_journal_test_restart(handle
, inode
);
3401 BUFFER_TRACE(bh
, "retaking write access");
3402 ext4_journal_get_write_access(handle
, bh
);
3407 * Any buffers which are on the journal will be in memory. We find
3408 * them on the hash table so jbd2_journal_revoke() will run jbd2_journal_forget()
3409 * on them. We've already detached each block from the file, so
3410 * bforget() in jbd2_journal_forget() should be safe.
3412 * AKPM: turn on bforget in jbd2_journal_forget()!!!
3414 for (p
= first
; p
< last
; p
++) {
3415 u32 nr
= le32_to_cpu(*p
);
3417 struct buffer_head
*tbh
;
3420 tbh
= sb_find_get_block(inode
->i_sb
, nr
);
3421 ext4_forget(handle
, 0, inode
, tbh
, nr
);
3425 ext4_free_blocks(handle
, inode
, block_to_free
, count
, 0);
3429 * ext4_free_data - free a list of data blocks
3430 * @handle: handle for this transaction
3431 * @inode: inode we are dealing with
3432 * @this_bh: indirect buffer_head which contains *@first and *@last
3433 * @first: array of block numbers
3434 * @last: points immediately past the end of array
3436 * We are freeing all blocks refered from that array (numbers are stored as
3437 * little-endian 32-bit) and updating @inode->i_blocks appropriately.
3439 * We accumulate contiguous runs of blocks to free. Conveniently, if these
3440 * blocks are contiguous then releasing them at one time will only affect one
3441 * or two bitmap blocks (+ group descriptor(s) and superblock) and we won't
3442 * actually use a lot of journal space.
3444 * @this_bh will be %NULL if @first and @last point into the inode's direct
3447 static void ext4_free_data(handle_t
*handle
, struct inode
*inode
,
3448 struct buffer_head
*this_bh
,
3449 __le32
*first
, __le32
*last
)
3451 ext4_fsblk_t block_to_free
= 0; /* Starting block # of a run */
3452 unsigned long count
= 0; /* Number of blocks in the run */
3453 __le32
*block_to_free_p
= NULL
; /* Pointer into inode/ind
3456 ext4_fsblk_t nr
; /* Current block # */
3457 __le32
*p
; /* Pointer into inode/ind
3458 for current block */
3461 if (this_bh
) { /* For indirect block */
3462 BUFFER_TRACE(this_bh
, "get_write_access");
3463 err
= ext4_journal_get_write_access(handle
, this_bh
);
3464 /* Important: if we can't update the indirect pointers
3465 * to the blocks, we can't free them. */
3470 for (p
= first
; p
< last
; p
++) {
3471 nr
= le32_to_cpu(*p
);
3473 /* accumulate blocks to free if they're contiguous */
3476 block_to_free_p
= p
;
3478 } else if (nr
== block_to_free
+ count
) {
3481 ext4_clear_blocks(handle
, inode
, this_bh
,
3483 count
, block_to_free_p
, p
);
3485 block_to_free_p
= p
;
3492 ext4_clear_blocks(handle
, inode
, this_bh
, block_to_free
,
3493 count
, block_to_free_p
, p
);
3496 BUFFER_TRACE(this_bh
, "call ext4_journal_dirty_metadata");
3499 * The buffer head should have an attached journal head at this
3500 * point. However, if the data is corrupted and an indirect
3501 * block pointed to itself, it would have been detached when
3502 * the block was cleared. Check for this instead of OOPSing.
3505 ext4_journal_dirty_metadata(handle
, this_bh
);
3507 ext4_error(inode
->i_sb
, __func__
,
3508 "circular indirect block detected, "
3509 "inode=%lu, block=%llu",
3511 (unsigned long long) this_bh
->b_blocknr
);
3516 * ext4_free_branches - free an array of branches
3517 * @handle: JBD handle for this transaction
3518 * @inode: inode we are dealing with
3519 * @parent_bh: the buffer_head which contains *@first and *@last
3520 * @first: array of block numbers
3521 * @last: pointer immediately past the end of array
3522 * @depth: depth of the branches to free
3524 * We are freeing all blocks refered from these branches (numbers are
3525 * stored as little-endian 32-bit) and updating @inode->i_blocks
3528 static void ext4_free_branches(handle_t
*handle
, struct inode
*inode
,
3529 struct buffer_head
*parent_bh
,
3530 __le32
*first
, __le32
*last
, int depth
)
3535 if (is_handle_aborted(handle
))
3539 struct buffer_head
*bh
;
3540 int addr_per_block
= EXT4_ADDR_PER_BLOCK(inode
->i_sb
);
3542 while (--p
>= first
) {
3543 nr
= le32_to_cpu(*p
);
3545 continue; /* A hole */
3547 /* Go read the buffer for the next level down */
3548 bh
= sb_bread(inode
->i_sb
, nr
);
3551 * A read failure? Report error and clear slot
3555 ext4_error(inode
->i_sb
, "ext4_free_branches",
3556 "Read failure, inode=%lu, block=%llu",
3561 /* This zaps the entire block. Bottom up. */
3562 BUFFER_TRACE(bh
, "free child branches");
3563 ext4_free_branches(handle
, inode
, bh
,
3564 (__le32
*) bh
->b_data
,
3565 (__le32
*) bh
->b_data
+ addr_per_block
,
3569 * We've probably journalled the indirect block several
3570 * times during the truncate. But it's no longer
3571 * needed and we now drop it from the transaction via
3572 * jbd2_journal_revoke().
3574 * That's easy if it's exclusively part of this
3575 * transaction. But if it's part of the committing
3576 * transaction then jbd2_journal_forget() will simply
3577 * brelse() it. That means that if the underlying
3578 * block is reallocated in ext4_get_block(),
3579 * unmap_underlying_metadata() will find this block
3580 * and will try to get rid of it. damn, damn.
3582 * If this block has already been committed to the
3583 * journal, a revoke record will be written. And
3584 * revoke records must be emitted *before* clearing
3585 * this block's bit in the bitmaps.
3587 ext4_forget(handle
, 1, inode
, bh
, bh
->b_blocknr
);
3590 * Everything below this this pointer has been
3591 * released. Now let this top-of-subtree go.
3593 * We want the freeing of this indirect block to be
3594 * atomic in the journal with the updating of the
3595 * bitmap block which owns it. So make some room in
3598 * We zero the parent pointer *after* freeing its
3599 * pointee in the bitmaps, so if extend_transaction()
3600 * for some reason fails to put the bitmap changes and
3601 * the release into the same transaction, recovery
3602 * will merely complain about releasing a free block,
3603 * rather than leaking blocks.
3605 if (is_handle_aborted(handle
))
3607 if (try_to_extend_transaction(handle
, inode
)) {
3608 ext4_mark_inode_dirty(handle
, inode
);
3609 ext4_journal_test_restart(handle
, inode
);
3612 ext4_free_blocks(handle
, inode
, nr
, 1, 1);
3616 * The block which we have just freed is
3617 * pointed to by an indirect block: journal it
3619 BUFFER_TRACE(parent_bh
, "get_write_access");
3620 if (!ext4_journal_get_write_access(handle
,
3623 BUFFER_TRACE(parent_bh
,
3624 "call ext4_journal_dirty_metadata");
3625 ext4_journal_dirty_metadata(handle
,
3631 /* We have reached the bottom of the tree. */
3632 BUFFER_TRACE(parent_bh
, "free data blocks");
3633 ext4_free_data(handle
, inode
, parent_bh
, first
, last
);
3637 int ext4_can_truncate(struct inode
*inode
)
3639 if (IS_APPEND(inode
) || IS_IMMUTABLE(inode
))
3641 if (S_ISREG(inode
->i_mode
))
3643 if (S_ISDIR(inode
->i_mode
))
3645 if (S_ISLNK(inode
->i_mode
))
3646 return !ext4_inode_is_fast_symlink(inode
);
3653 * We block out ext4_get_block() block instantiations across the entire
3654 * transaction, and VFS/VM ensures that ext4_truncate() cannot run
3655 * simultaneously on behalf of the same inode.
3657 * As we work through the truncate and commmit bits of it to the journal there
3658 * is one core, guiding principle: the file's tree must always be consistent on
3659 * disk. We must be able to restart the truncate after a crash.
3661 * The file's tree may be transiently inconsistent in memory (although it
3662 * probably isn't), but whenever we close off and commit a journal transaction,
3663 * the contents of (the filesystem + the journal) must be consistent and
3664 * restartable. It's pretty simple, really: bottom up, right to left (although
3665 * left-to-right works OK too).
3667 * Note that at recovery time, journal replay occurs *before* the restart of
3668 * truncate against the orphan inode list.
3670 * The committed inode has the new, desired i_size (which is the same as
3671 * i_disksize in this case). After a crash, ext4_orphan_cleanup() will see
3672 * that this inode's truncate did not complete and it will again call
3673 * ext4_truncate() to have another go. So there will be instantiated blocks
3674 * to the right of the truncation point in a crashed ext4 filesystem. But
3675 * that's fine - as long as they are linked from the inode, the post-crash
3676 * ext4_truncate() run will find them and release them.
3678 void ext4_truncate(struct inode
*inode
)
3681 struct ext4_inode_info
*ei
= EXT4_I(inode
);
3682 __le32
*i_data
= ei
->i_data
;
3683 int addr_per_block
= EXT4_ADDR_PER_BLOCK(inode
->i_sb
);
3684 struct address_space
*mapping
= inode
->i_mapping
;
3685 ext4_lblk_t offsets
[4];
3690 ext4_lblk_t last_block
;
3691 unsigned blocksize
= inode
->i_sb
->s_blocksize
;
3693 if (!ext4_can_truncate(inode
))
3696 if (EXT4_I(inode
)->i_flags
& EXT4_EXTENTS_FL
) {
3697 ext4_ext_truncate(inode
);
3701 handle
= start_transaction(inode
);
3703 return; /* AKPM: return what? */
3705 last_block
= (inode
->i_size
+ blocksize
-1)
3706 >> EXT4_BLOCK_SIZE_BITS(inode
->i_sb
);
3708 if (inode
->i_size
& (blocksize
- 1))
3709 if (ext4_block_truncate_page(handle
, mapping
, inode
->i_size
))
3712 n
= ext4_block_to_path(inode
, last_block
, offsets
, NULL
);
3714 goto out_stop
; /* error */
3717 * OK. This truncate is going to happen. We add the inode to the
3718 * orphan list, so that if this truncate spans multiple transactions,
3719 * and we crash, we will resume the truncate when the filesystem
3720 * recovers. It also marks the inode dirty, to catch the new size.
3722 * Implication: the file must always be in a sane, consistent
3723 * truncatable state while each transaction commits.
3725 if (ext4_orphan_add(handle
, inode
))
3729 * From here we block out all ext4_get_block() callers who want to
3730 * modify the block allocation tree.
3732 down_write(&ei
->i_data_sem
);
3734 ext4_discard_preallocations(inode
);
3737 * The orphan list entry will now protect us from any crash which
3738 * occurs before the truncate completes, so it is now safe to propagate
3739 * the new, shorter inode size (held for now in i_size) into the
3740 * on-disk inode. We do this via i_disksize, which is the value which
3741 * ext4 *really* writes onto the disk inode.
3743 ei
->i_disksize
= inode
->i_size
;
3745 if (n
== 1) { /* direct blocks */
3746 ext4_free_data(handle
, inode
, NULL
, i_data
+offsets
[0],
3747 i_data
+ EXT4_NDIR_BLOCKS
);
3751 partial
= ext4_find_shared(inode
, n
, offsets
, chain
, &nr
);
3752 /* Kill the top of shared branch (not detached) */
3754 if (partial
== chain
) {
3755 /* Shared branch grows from the inode */
3756 ext4_free_branches(handle
, inode
, NULL
,
3757 &nr
, &nr
+1, (chain
+n
-1) - partial
);
3760 * We mark the inode dirty prior to restart,
3761 * and prior to stop. No need for it here.
3764 /* Shared branch grows from an indirect block */
3765 BUFFER_TRACE(partial
->bh
, "get_write_access");
3766 ext4_free_branches(handle
, inode
, partial
->bh
,
3768 partial
->p
+1, (chain
+n
-1) - partial
);
3771 /* Clear the ends of indirect blocks on the shared branch */
3772 while (partial
> chain
) {
3773 ext4_free_branches(handle
, inode
, partial
->bh
, partial
->p
+ 1,
3774 (__le32
*)partial
->bh
->b_data
+addr_per_block
,
3775 (chain
+n
-1) - partial
);
3776 BUFFER_TRACE(partial
->bh
, "call brelse");
3777 brelse (partial
->bh
);
3781 /* Kill the remaining (whole) subtrees */
3782 switch (offsets
[0]) {
3784 nr
= i_data
[EXT4_IND_BLOCK
];
3786 ext4_free_branches(handle
, inode
, NULL
, &nr
, &nr
+1, 1);
3787 i_data
[EXT4_IND_BLOCK
] = 0;
3789 case EXT4_IND_BLOCK
:
3790 nr
= i_data
[EXT4_DIND_BLOCK
];
3792 ext4_free_branches(handle
, inode
, NULL
, &nr
, &nr
+1, 2);
3793 i_data
[EXT4_DIND_BLOCK
] = 0;
3795 case EXT4_DIND_BLOCK
:
3796 nr
= i_data
[EXT4_TIND_BLOCK
];
3798 ext4_free_branches(handle
, inode
, NULL
, &nr
, &nr
+1, 3);
3799 i_data
[EXT4_TIND_BLOCK
] = 0;
3801 case EXT4_TIND_BLOCK
:
3805 up_write(&ei
->i_data_sem
);
3806 inode
->i_mtime
= inode
->i_ctime
= ext4_current_time(inode
);
3807 ext4_mark_inode_dirty(handle
, inode
);
3810 * In a multi-transaction truncate, we only make the final transaction
3817 * If this was a simple ftruncate(), and the file will remain alive
3818 * then we need to clear up the orphan record which we created above.
3819 * However, if this was a real unlink then we were called by
3820 * ext4_delete_inode(), and we allow that function to clean up the
3821 * orphan info for us.
3824 ext4_orphan_del(handle
, inode
);
3826 ext4_journal_stop(handle
);
3830 * ext4_get_inode_loc returns with an extra refcount against the inode's
3831 * underlying buffer_head on success. If 'in_mem' is true, we have all
3832 * data in memory that is needed to recreate the on-disk version of this
3835 static int __ext4_get_inode_loc(struct inode
*inode
,
3836 struct ext4_iloc
*iloc
, int in_mem
)
3838 struct ext4_group_desc
*gdp
;
3839 struct buffer_head
*bh
;
3840 struct super_block
*sb
= inode
->i_sb
;
3842 int inodes_per_block
, inode_offset
;
3845 if (!ext4_valid_inum(sb
, inode
->i_ino
))
3848 iloc
->block_group
= (inode
->i_ino
- 1) / EXT4_INODES_PER_GROUP(sb
);
3849 gdp
= ext4_get_group_desc(sb
, iloc
->block_group
, NULL
);
3854 * Figure out the offset within the block group inode table
3856 inodes_per_block
= (EXT4_BLOCK_SIZE(sb
) / EXT4_INODE_SIZE(sb
));
3857 inode_offset
= ((inode
->i_ino
- 1) %
3858 EXT4_INODES_PER_GROUP(sb
));
3859 block
= ext4_inode_table(sb
, gdp
) + (inode_offset
/ inodes_per_block
);
3860 iloc
->offset
= (inode_offset
% inodes_per_block
) * EXT4_INODE_SIZE(sb
);
3862 bh
= sb_getblk(sb
, block
);
3864 ext4_error(sb
, "ext4_get_inode_loc", "unable to read "
3865 "inode block - inode=%lu, block=%llu",
3866 inode
->i_ino
, block
);
3869 if (!buffer_uptodate(bh
)) {
3873 * If the buffer has the write error flag, we have failed
3874 * to write out another inode in the same block. In this
3875 * case, we don't have to read the block because we may
3876 * read the old inode data successfully.
3878 if (buffer_write_io_error(bh
) && !buffer_uptodate(bh
))
3879 set_buffer_uptodate(bh
);
3881 if (buffer_uptodate(bh
)) {
3882 /* someone brought it uptodate while we waited */
3888 * If we have all information of the inode in memory and this
3889 * is the only valid inode in the block, we need not read the
3893 struct buffer_head
*bitmap_bh
;
3896 start
= inode_offset
& ~(inodes_per_block
- 1);
3898 /* Is the inode bitmap in cache? */
3899 bitmap_bh
= sb_getblk(sb
, ext4_inode_bitmap(sb
, gdp
));
3904 * If the inode bitmap isn't in cache then the
3905 * optimisation may end up performing two reads instead
3906 * of one, so skip it.
3908 if (!buffer_uptodate(bitmap_bh
)) {
3912 for (i
= start
; i
< start
+ inodes_per_block
; i
++) {
3913 if (i
== inode_offset
)
3915 if (ext4_test_bit(i
, bitmap_bh
->b_data
))
3919 if (i
== start
+ inodes_per_block
) {
3920 /* all other inodes are free, so skip I/O */
3921 memset(bh
->b_data
, 0, bh
->b_size
);
3922 set_buffer_uptodate(bh
);
3930 * If we need to do any I/O, try to pre-readahead extra
3931 * blocks from the inode table.
3933 if (EXT4_SB(sb
)->s_inode_readahead_blks
) {
3934 ext4_fsblk_t b
, end
, table
;
3937 table
= ext4_inode_table(sb
, gdp
);
3938 /* Make sure s_inode_readahead_blks is a power of 2 */
3939 while (EXT4_SB(sb
)->s_inode_readahead_blks
&
3940 (EXT4_SB(sb
)->s_inode_readahead_blks
-1))
3941 EXT4_SB(sb
)->s_inode_readahead_blks
=
3942 (EXT4_SB(sb
)->s_inode_readahead_blks
&
3943 (EXT4_SB(sb
)->s_inode_readahead_blks
-1));
3944 b
= block
& ~(EXT4_SB(sb
)->s_inode_readahead_blks
-1);
3947 end
= b
+ EXT4_SB(sb
)->s_inode_readahead_blks
;
3948 num
= EXT4_INODES_PER_GROUP(sb
);
3949 if (EXT4_HAS_RO_COMPAT_FEATURE(sb
,
3950 EXT4_FEATURE_RO_COMPAT_GDT_CSUM
))
3951 num
-= le16_to_cpu(gdp
->bg_itable_unused
);
3952 table
+= num
/ inodes_per_block
;
3956 sb_breadahead(sb
, b
++);
3960 * There are other valid inodes in the buffer, this inode
3961 * has in-inode xattrs, or we don't have this inode in memory.
3962 * Read the block from disk.
3965 bh
->b_end_io
= end_buffer_read_sync
;
3966 submit_bh(READ_META
, bh
);
3968 if (!buffer_uptodate(bh
)) {
3969 ext4_error(sb
, __func__
,
3970 "unable to read inode block - inode=%lu, "
3971 "block=%llu", inode
->i_ino
, block
);
3981 int ext4_get_inode_loc(struct inode
*inode
, struct ext4_iloc
*iloc
)
3983 /* We have all inode data except xattrs in memory here. */
3984 return __ext4_get_inode_loc(inode
, iloc
,
3985 !(EXT4_I(inode
)->i_state
& EXT4_STATE_XATTR
));
3988 void ext4_set_inode_flags(struct inode
*inode
)
3990 unsigned int flags
= EXT4_I(inode
)->i_flags
;
3992 inode
->i_flags
&= ~(S_SYNC
|S_APPEND
|S_IMMUTABLE
|S_NOATIME
|S_DIRSYNC
);
3993 if (flags
& EXT4_SYNC_FL
)
3994 inode
->i_flags
|= S_SYNC
;
3995 if (flags
& EXT4_APPEND_FL
)
3996 inode
->i_flags
|= S_APPEND
;
3997 if (flags
& EXT4_IMMUTABLE_FL
)
3998 inode
->i_flags
|= S_IMMUTABLE
;
3999 if (flags
& EXT4_NOATIME_FL
)
4000 inode
->i_flags
|= S_NOATIME
;
4001 if (flags
& EXT4_DIRSYNC_FL
)
4002 inode
->i_flags
|= S_DIRSYNC
;
4005 /* Propagate flags from i_flags to EXT4_I(inode)->i_flags */
4006 void ext4_get_inode_flags(struct ext4_inode_info
*ei
)
4008 unsigned int flags
= ei
->vfs_inode
.i_flags
;
4010 ei
->i_flags
&= ~(EXT4_SYNC_FL
|EXT4_APPEND_FL
|
4011 EXT4_IMMUTABLE_FL
|EXT4_NOATIME_FL
|EXT4_DIRSYNC_FL
);
4013 ei
->i_flags
|= EXT4_SYNC_FL
;
4014 if (flags
& S_APPEND
)
4015 ei
->i_flags
|= EXT4_APPEND_FL
;
4016 if (flags
& S_IMMUTABLE
)
4017 ei
->i_flags
|= EXT4_IMMUTABLE_FL
;
4018 if (flags
& S_NOATIME
)
4019 ei
->i_flags
|= EXT4_NOATIME_FL
;
4020 if (flags
& S_DIRSYNC
)
4021 ei
->i_flags
|= EXT4_DIRSYNC_FL
;
4023 static blkcnt_t
ext4_inode_blocks(struct ext4_inode
*raw_inode
,
4024 struct ext4_inode_info
*ei
)
4027 struct inode
*inode
= &(ei
->vfs_inode
);
4028 struct super_block
*sb
= inode
->i_sb
;
4030 if (EXT4_HAS_RO_COMPAT_FEATURE(sb
,
4031 EXT4_FEATURE_RO_COMPAT_HUGE_FILE
)) {
4032 /* we are using combined 48 bit field */
4033 i_blocks
= ((u64
)le16_to_cpu(raw_inode
->i_blocks_high
)) << 32 |
4034 le32_to_cpu(raw_inode
->i_blocks_lo
);
4035 if (ei
->i_flags
& EXT4_HUGE_FILE_FL
) {
4036 /* i_blocks represent file system block size */
4037 return i_blocks
<< (inode
->i_blkbits
- 9);
4042 return le32_to_cpu(raw_inode
->i_blocks_lo
);
4046 struct inode
*ext4_iget(struct super_block
*sb
, unsigned long ino
)
4048 struct ext4_iloc iloc
;
4049 struct ext4_inode
*raw_inode
;
4050 struct ext4_inode_info
*ei
;
4051 struct buffer_head
*bh
;
4052 struct inode
*inode
;
4056 inode
= iget_locked(sb
, ino
);
4058 return ERR_PTR(-ENOMEM
);
4059 if (!(inode
->i_state
& I_NEW
))
4063 #ifdef CONFIG_EXT4_FS_POSIX_ACL
4064 ei
->i_acl
= EXT4_ACL_NOT_CACHED
;
4065 ei
->i_default_acl
= EXT4_ACL_NOT_CACHED
;
4068 ret
= __ext4_get_inode_loc(inode
, &iloc
, 0);
4072 raw_inode
= ext4_raw_inode(&iloc
);
4073 inode
->i_mode
= le16_to_cpu(raw_inode
->i_mode
);
4074 inode
->i_uid
= (uid_t
)le16_to_cpu(raw_inode
->i_uid_low
);
4075 inode
->i_gid
= (gid_t
)le16_to_cpu(raw_inode
->i_gid_low
);
4076 if (!(test_opt(inode
->i_sb
, NO_UID32
))) {
4077 inode
->i_uid
|= le16_to_cpu(raw_inode
->i_uid_high
) << 16;
4078 inode
->i_gid
|= le16_to_cpu(raw_inode
->i_gid_high
) << 16;
4080 inode
->i_nlink
= le16_to_cpu(raw_inode
->i_links_count
);
4083 ei
->i_dir_start_lookup
= 0;
4084 ei
->i_dtime
= le32_to_cpu(raw_inode
->i_dtime
);
4085 /* We now have enough fields to check if the inode was active or not.
4086 * This is needed because nfsd might try to access dead inodes
4087 * the test is that same one that e2fsck uses
4088 * NeilBrown 1999oct15
4090 if (inode
->i_nlink
== 0) {
4091 if (inode
->i_mode
== 0 ||
4092 !(EXT4_SB(inode
->i_sb
)->s_mount_state
& EXT4_ORPHAN_FS
)) {
4093 /* this inode is deleted */
4098 /* The only unlinked inodes we let through here have
4099 * valid i_mode and are being read by the orphan
4100 * recovery code: that's fine, we're about to complete
4101 * the process of deleting those. */
4103 ei
->i_flags
= le32_to_cpu(raw_inode
->i_flags
);
4104 inode
->i_blocks
= ext4_inode_blocks(raw_inode
, ei
);
4105 ei
->i_file_acl
= le32_to_cpu(raw_inode
->i_file_acl_lo
);
4106 if (EXT4_SB(inode
->i_sb
)->s_es
->s_creator_os
!=
4107 cpu_to_le32(EXT4_OS_HURD
)) {
4109 ((__u64
)le16_to_cpu(raw_inode
->i_file_acl_high
)) << 32;
4111 inode
->i_size
= ext4_isize(raw_inode
);
4112 ei
->i_disksize
= inode
->i_size
;
4113 inode
->i_generation
= le32_to_cpu(raw_inode
->i_generation
);
4114 ei
->i_block_group
= iloc
.block_group
;
4116 * NOTE! The in-memory inode i_data array is in little-endian order
4117 * even on big-endian machines: we do NOT byteswap the block numbers!
4119 for (block
= 0; block
< EXT4_N_BLOCKS
; block
++)
4120 ei
->i_data
[block
] = raw_inode
->i_block
[block
];
4121 INIT_LIST_HEAD(&ei
->i_orphan
);
4123 if (EXT4_INODE_SIZE(inode
->i_sb
) > EXT4_GOOD_OLD_INODE_SIZE
) {
4124 ei
->i_extra_isize
= le16_to_cpu(raw_inode
->i_extra_isize
);
4125 if (EXT4_GOOD_OLD_INODE_SIZE
+ ei
->i_extra_isize
>
4126 EXT4_INODE_SIZE(inode
->i_sb
)) {
4131 if (ei
->i_extra_isize
== 0) {
4132 /* The extra space is currently unused. Use it. */
4133 ei
->i_extra_isize
= sizeof(struct ext4_inode
) -
4134 EXT4_GOOD_OLD_INODE_SIZE
;
4136 __le32
*magic
= (void *)raw_inode
+
4137 EXT4_GOOD_OLD_INODE_SIZE
+
4139 if (*magic
== cpu_to_le32(EXT4_XATTR_MAGIC
))
4140 ei
->i_state
|= EXT4_STATE_XATTR
;
4143 ei
->i_extra_isize
= 0;
4145 EXT4_INODE_GET_XTIME(i_ctime
, inode
, raw_inode
);
4146 EXT4_INODE_GET_XTIME(i_mtime
, inode
, raw_inode
);
4147 EXT4_INODE_GET_XTIME(i_atime
, inode
, raw_inode
);
4148 EXT4_EINODE_GET_XTIME(i_crtime
, ei
, raw_inode
);
4150 inode
->i_version
= le32_to_cpu(raw_inode
->i_disk_version
);
4151 if (EXT4_INODE_SIZE(inode
->i_sb
) > EXT4_GOOD_OLD_INODE_SIZE
) {
4152 if (EXT4_FITS_IN_INODE(raw_inode
, ei
, i_version_hi
))
4154 (__u64
)(le32_to_cpu(raw_inode
->i_version_hi
)) << 32;
4157 if (S_ISREG(inode
->i_mode
)) {
4158 inode
->i_op
= &ext4_file_inode_operations
;
4159 inode
->i_fop
= &ext4_file_operations
;
4160 ext4_set_aops(inode
);
4161 } else if (S_ISDIR(inode
->i_mode
)) {
4162 inode
->i_op
= &ext4_dir_inode_operations
;
4163 inode
->i_fop
= &ext4_dir_operations
;
4164 } else if (S_ISLNK(inode
->i_mode
)) {
4165 if (ext4_inode_is_fast_symlink(inode
))
4166 inode
->i_op
= &ext4_fast_symlink_inode_operations
;
4168 inode
->i_op
= &ext4_symlink_inode_operations
;
4169 ext4_set_aops(inode
);
4172 inode
->i_op
= &ext4_special_inode_operations
;
4173 if (raw_inode
->i_block
[0])
4174 init_special_inode(inode
, inode
->i_mode
,
4175 old_decode_dev(le32_to_cpu(raw_inode
->i_block
[0])));
4177 init_special_inode(inode
, inode
->i_mode
,
4178 new_decode_dev(le32_to_cpu(raw_inode
->i_block
[1])));
4181 ext4_set_inode_flags(inode
);
4182 unlock_new_inode(inode
);
4187 return ERR_PTR(ret
);
4190 static int ext4_inode_blocks_set(handle_t
*handle
,
4191 struct ext4_inode
*raw_inode
,
4192 struct ext4_inode_info
*ei
)
4194 struct inode
*inode
= &(ei
->vfs_inode
);
4195 u64 i_blocks
= inode
->i_blocks
;
4196 struct super_block
*sb
= inode
->i_sb
;
4198 if (i_blocks
<= ~0U) {
4200 * i_blocks can be represnted in a 32 bit variable
4201 * as multiple of 512 bytes
4203 raw_inode
->i_blocks_lo
= cpu_to_le32(i_blocks
);
4204 raw_inode
->i_blocks_high
= 0;
4205 ei
->i_flags
&= ~EXT4_HUGE_FILE_FL
;
4208 if (!EXT4_HAS_RO_COMPAT_FEATURE(sb
, EXT4_FEATURE_RO_COMPAT_HUGE_FILE
))
4211 if (i_blocks
<= 0xffffffffffffULL
) {
4213 * i_blocks can be represented in a 48 bit variable
4214 * as multiple of 512 bytes
4216 raw_inode
->i_blocks_lo
= cpu_to_le32(i_blocks
);
4217 raw_inode
->i_blocks_high
= cpu_to_le16(i_blocks
>> 32);
4218 ei
->i_flags
&= ~EXT4_HUGE_FILE_FL
;
4220 ei
->i_flags
|= EXT4_HUGE_FILE_FL
;
4221 /* i_block is stored in file system block size */
4222 i_blocks
= i_blocks
>> (inode
->i_blkbits
- 9);
4223 raw_inode
->i_blocks_lo
= cpu_to_le32(i_blocks
);
4224 raw_inode
->i_blocks_high
= cpu_to_le16(i_blocks
>> 32);
4230 * Post the struct inode info into an on-disk inode location in the
4231 * buffer-cache. This gobbles the caller's reference to the
4232 * buffer_head in the inode location struct.
4234 * The caller must have write access to iloc->bh.
4236 static int ext4_do_update_inode(handle_t
*handle
,
4237 struct inode
*inode
,
4238 struct ext4_iloc
*iloc
)
4240 struct ext4_inode
*raw_inode
= ext4_raw_inode(iloc
);
4241 struct ext4_inode_info
*ei
= EXT4_I(inode
);
4242 struct buffer_head
*bh
= iloc
->bh
;
4243 int err
= 0, rc
, block
;
4245 /* For fields not not tracking in the in-memory inode,
4246 * initialise them to zero for new inodes. */
4247 if (ei
->i_state
& EXT4_STATE_NEW
)
4248 memset(raw_inode
, 0, EXT4_SB(inode
->i_sb
)->s_inode_size
);
4250 ext4_get_inode_flags(ei
);
4251 raw_inode
->i_mode
= cpu_to_le16(inode
->i_mode
);
4252 if (!(test_opt(inode
->i_sb
, NO_UID32
))) {
4253 raw_inode
->i_uid_low
= cpu_to_le16(low_16_bits(inode
->i_uid
));
4254 raw_inode
->i_gid_low
= cpu_to_le16(low_16_bits(inode
->i_gid
));
4256 * Fix up interoperability with old kernels. Otherwise, old inodes get
4257 * re-used with the upper 16 bits of the uid/gid intact
4260 raw_inode
->i_uid_high
=
4261 cpu_to_le16(high_16_bits(inode
->i_uid
));
4262 raw_inode
->i_gid_high
=
4263 cpu_to_le16(high_16_bits(inode
->i_gid
));
4265 raw_inode
->i_uid_high
= 0;
4266 raw_inode
->i_gid_high
= 0;
4269 raw_inode
->i_uid_low
=
4270 cpu_to_le16(fs_high2lowuid(inode
->i_uid
));
4271 raw_inode
->i_gid_low
=
4272 cpu_to_le16(fs_high2lowgid(inode
->i_gid
));
4273 raw_inode
->i_uid_high
= 0;
4274 raw_inode
->i_gid_high
= 0;
4276 raw_inode
->i_links_count
= cpu_to_le16(inode
->i_nlink
);
4278 EXT4_INODE_SET_XTIME(i_ctime
, inode
, raw_inode
);
4279 EXT4_INODE_SET_XTIME(i_mtime
, inode
, raw_inode
);
4280 EXT4_INODE_SET_XTIME(i_atime
, inode
, raw_inode
);
4281 EXT4_EINODE_SET_XTIME(i_crtime
, ei
, raw_inode
);
4283 if (ext4_inode_blocks_set(handle
, raw_inode
, ei
))
4285 raw_inode
->i_dtime
= cpu_to_le32(ei
->i_dtime
);
4286 /* clear the migrate flag in the raw_inode */
4287 raw_inode
->i_flags
= cpu_to_le32(ei
->i_flags
& ~EXT4_EXT_MIGRATE
);
4288 if (EXT4_SB(inode
->i_sb
)->s_es
->s_creator_os
!=
4289 cpu_to_le32(EXT4_OS_HURD
))
4290 raw_inode
->i_file_acl_high
=
4291 cpu_to_le16(ei
->i_file_acl
>> 32);
4292 raw_inode
->i_file_acl_lo
= cpu_to_le32(ei
->i_file_acl
);
4293 ext4_isize_set(raw_inode
, ei
->i_disksize
);
4294 if (ei
->i_disksize
> 0x7fffffffULL
) {
4295 struct super_block
*sb
= inode
->i_sb
;
4296 if (!EXT4_HAS_RO_COMPAT_FEATURE(sb
,
4297 EXT4_FEATURE_RO_COMPAT_LARGE_FILE
) ||
4298 EXT4_SB(sb
)->s_es
->s_rev_level
==
4299 cpu_to_le32(EXT4_GOOD_OLD_REV
)) {
4300 /* If this is the first large file
4301 * created, add a flag to the superblock.
4303 err
= ext4_journal_get_write_access(handle
,
4304 EXT4_SB(sb
)->s_sbh
);
4307 ext4_update_dynamic_rev(sb
);
4308 EXT4_SET_RO_COMPAT_FEATURE(sb
,
4309 EXT4_FEATURE_RO_COMPAT_LARGE_FILE
);
4312 err
= ext4_journal_dirty_metadata(handle
,
4313 EXT4_SB(sb
)->s_sbh
);
4316 raw_inode
->i_generation
= cpu_to_le32(inode
->i_generation
);
4317 if (S_ISCHR(inode
->i_mode
) || S_ISBLK(inode
->i_mode
)) {
4318 if (old_valid_dev(inode
->i_rdev
)) {
4319 raw_inode
->i_block
[0] =
4320 cpu_to_le32(old_encode_dev(inode
->i_rdev
));
4321 raw_inode
->i_block
[1] = 0;
4323 raw_inode
->i_block
[0] = 0;
4324 raw_inode
->i_block
[1] =
4325 cpu_to_le32(new_encode_dev(inode
->i_rdev
));
4326 raw_inode
->i_block
[2] = 0;
4328 } else for (block
= 0; block
< EXT4_N_BLOCKS
; block
++)
4329 raw_inode
->i_block
[block
] = ei
->i_data
[block
];
4331 raw_inode
->i_disk_version
= cpu_to_le32(inode
->i_version
);
4332 if (ei
->i_extra_isize
) {
4333 if (EXT4_FITS_IN_INODE(raw_inode
, ei
, i_version_hi
))
4334 raw_inode
->i_version_hi
=
4335 cpu_to_le32(inode
->i_version
>> 32);
4336 raw_inode
->i_extra_isize
= cpu_to_le16(ei
->i_extra_isize
);
4340 BUFFER_TRACE(bh
, "call ext4_journal_dirty_metadata");
4341 rc
= ext4_journal_dirty_metadata(handle
, bh
);
4344 ei
->i_state
&= ~EXT4_STATE_NEW
;
4348 ext4_std_error(inode
->i_sb
, err
);
4353 * ext4_write_inode()
4355 * We are called from a few places:
4357 * - Within generic_file_write() for O_SYNC files.
4358 * Here, there will be no transaction running. We wait for any running
4359 * trasnaction to commit.
4361 * - Within sys_sync(), kupdate and such.
4362 * We wait on commit, if tol to.
4364 * - Within prune_icache() (PF_MEMALLOC == true)
4365 * Here we simply return. We can't afford to block kswapd on the
4368 * In all cases it is actually safe for us to return without doing anything,
4369 * because the inode has been copied into a raw inode buffer in
4370 * ext4_mark_inode_dirty(). This is a correctness thing for O_SYNC and for
4373 * Note that we are absolutely dependent upon all inode dirtiers doing the
4374 * right thing: they *must* call mark_inode_dirty() after dirtying info in
4375 * which we are interested.
4377 * It would be a bug for them to not do this. The code:
4379 * mark_inode_dirty(inode)
4381 * inode->i_size = expr;
4383 * is in error because a kswapd-driven write_inode() could occur while
4384 * `stuff()' is running, and the new i_size will be lost. Plus the inode
4385 * will no longer be on the superblock's dirty inode list.
4387 int ext4_write_inode(struct inode
*inode
, int wait
)
4389 if (current
->flags
& PF_MEMALLOC
)
4392 if (ext4_journal_current_handle()) {
4393 jbd_debug(1, "called recursively, non-PF_MEMALLOC!\n");
4401 return ext4_force_commit(inode
->i_sb
);
4407 * Called from notify_change.
4409 * We want to trap VFS attempts to truncate the file as soon as
4410 * possible. In particular, we want to make sure that when the VFS
4411 * shrinks i_size, we put the inode on the orphan list and modify
4412 * i_disksize immediately, so that during the subsequent flushing of
4413 * dirty pages and freeing of disk blocks, we can guarantee that any
4414 * commit will leave the blocks being flushed in an unused state on
4415 * disk. (On recovery, the inode will get truncated and the blocks will
4416 * be freed, so we have a strong guarantee that no future commit will
4417 * leave these blocks visible to the user.)
4419 * Another thing we have to assure is that if we are in ordered mode
4420 * and inode is still attached to the committing transaction, we must
4421 * we start writeout of all the dirty pages which are being truncated.
4422 * This way we are sure that all the data written in the previous
4423 * transaction are already on disk (truncate waits for pages under
4426 * Called with inode->i_mutex down.
4428 int ext4_setattr(struct dentry
*dentry
, struct iattr
*attr
)
4430 struct inode
*inode
= dentry
->d_inode
;
4432 const unsigned int ia_valid
= attr
->ia_valid
;
4434 error
= inode_change_ok(inode
, attr
);
4438 if ((ia_valid
& ATTR_UID
&& attr
->ia_uid
!= inode
->i_uid
) ||
4439 (ia_valid
& ATTR_GID
&& attr
->ia_gid
!= inode
->i_gid
)) {
4442 /* (user+group)*(old+new) structure, inode write (sb,
4443 * inode block, ? - but truncate inode update has it) */
4444 handle
= ext4_journal_start(inode
, 2*(EXT4_QUOTA_INIT_BLOCKS(inode
->i_sb
)+
4445 EXT4_QUOTA_DEL_BLOCKS(inode
->i_sb
))+3);
4446 if (IS_ERR(handle
)) {
4447 error
= PTR_ERR(handle
);
4450 error
= DQUOT_TRANSFER(inode
, attr
) ? -EDQUOT
: 0;
4452 ext4_journal_stop(handle
);
4455 /* Update corresponding info in inode so that everything is in
4456 * one transaction */
4457 if (attr
->ia_valid
& ATTR_UID
)
4458 inode
->i_uid
= attr
->ia_uid
;
4459 if (attr
->ia_valid
& ATTR_GID
)
4460 inode
->i_gid
= attr
->ia_gid
;
4461 error
= ext4_mark_inode_dirty(handle
, inode
);
4462 ext4_journal_stop(handle
);
4465 if (attr
->ia_valid
& ATTR_SIZE
) {
4466 if (!(EXT4_I(inode
)->i_flags
& EXT4_EXTENTS_FL
)) {
4467 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
4469 if (attr
->ia_size
> sbi
->s_bitmap_maxbytes
) {
4476 if (S_ISREG(inode
->i_mode
) &&
4477 attr
->ia_valid
& ATTR_SIZE
&& attr
->ia_size
< inode
->i_size
) {
4480 handle
= ext4_journal_start(inode
, 3);
4481 if (IS_ERR(handle
)) {
4482 error
= PTR_ERR(handle
);
4486 error
= ext4_orphan_add(handle
, inode
);
4487 EXT4_I(inode
)->i_disksize
= attr
->ia_size
;
4488 rc
= ext4_mark_inode_dirty(handle
, inode
);
4491 ext4_journal_stop(handle
);
4493 if (ext4_should_order_data(inode
)) {
4494 error
= ext4_begin_ordered_truncate(inode
,
4497 /* Do as much error cleanup as possible */
4498 handle
= ext4_journal_start(inode
, 3);
4499 if (IS_ERR(handle
)) {
4500 ext4_orphan_del(NULL
, inode
);
4503 ext4_orphan_del(handle
, inode
);
4504 ext4_journal_stop(handle
);
4510 rc
= inode_setattr(inode
, attr
);
4512 /* If inode_setattr's call to ext4_truncate failed to get a
4513 * transaction handle at all, we need to clean up the in-core
4514 * orphan list manually. */
4516 ext4_orphan_del(NULL
, inode
);
4518 if (!rc
&& (ia_valid
& ATTR_MODE
))
4519 rc
= ext4_acl_chmod(inode
);
4522 ext4_std_error(inode
->i_sb
, error
);
4528 int ext4_getattr(struct vfsmount
*mnt
, struct dentry
*dentry
,
4531 struct inode
*inode
;
4532 unsigned long delalloc_blocks
;
4534 inode
= dentry
->d_inode
;
4535 generic_fillattr(inode
, stat
);
4538 * We can't update i_blocks if the block allocation is delayed
4539 * otherwise in the case of system crash before the real block
4540 * allocation is done, we will have i_blocks inconsistent with
4541 * on-disk file blocks.
4542 * We always keep i_blocks updated together with real
4543 * allocation. But to not confuse with user, stat
4544 * will return the blocks that include the delayed allocation
4545 * blocks for this file.
4547 spin_lock(&EXT4_I(inode
)->i_block_reservation_lock
);
4548 delalloc_blocks
= EXT4_I(inode
)->i_reserved_data_blocks
;
4549 spin_unlock(&EXT4_I(inode
)->i_block_reservation_lock
);
4551 stat
->blocks
+= (delalloc_blocks
<< inode
->i_sb
->s_blocksize_bits
)>>9;
4555 static int ext4_indirect_trans_blocks(struct inode
*inode
, int nrblocks
,
4560 /* if nrblocks are contiguous */
4563 * With N contiguous data blocks, it need at most
4564 * N/EXT4_ADDR_PER_BLOCK(inode->i_sb) indirect blocks
4565 * 2 dindirect blocks
4568 indirects
= nrblocks
/ EXT4_ADDR_PER_BLOCK(inode
->i_sb
);
4569 return indirects
+ 3;
4572 * if nrblocks are not contiguous, worse case, each block touch
4573 * a indirect block, and each indirect block touch a double indirect
4574 * block, plus a triple indirect block
4576 indirects
= nrblocks
* 2 + 1;
4580 static int ext4_index_trans_blocks(struct inode
*inode
, int nrblocks
, int chunk
)
4582 if (!(EXT4_I(inode
)->i_flags
& EXT4_EXTENTS_FL
))
4583 return ext4_indirect_trans_blocks(inode
, nrblocks
, 0);
4584 return ext4_ext_index_trans_blocks(inode
, nrblocks
, 0);
4587 * Account for index blocks, block groups bitmaps and block group
4588 * descriptor blocks if modify datablocks and index blocks
4589 * worse case, the indexs blocks spread over different block groups
4591 * If datablocks are discontiguous, they are possible to spread over
4592 * different block groups too. If they are contiugous, with flexbg,
4593 * they could still across block group boundary.
4595 * Also account for superblock, inode, quota and xattr blocks
4597 int ext4_meta_trans_blocks(struct inode
*inode
, int nrblocks
, int chunk
)
4599 int groups
, gdpblocks
;
4604 * How many index blocks need to touch to modify nrblocks?
4605 * The "Chunk" flag indicating whether the nrblocks is
4606 * physically contiguous on disk
4608 * For Direct IO and fallocate, they calls get_block to allocate
4609 * one single extent at a time, so they could set the "Chunk" flag
4611 idxblocks
= ext4_index_trans_blocks(inode
, nrblocks
, chunk
);
4616 * Now let's see how many group bitmaps and group descriptors need
4626 if (groups
> EXT4_SB(inode
->i_sb
)->s_groups_count
)
4627 groups
= EXT4_SB(inode
->i_sb
)->s_groups_count
;
4628 if (groups
> EXT4_SB(inode
->i_sb
)->s_gdb_count
)
4629 gdpblocks
= EXT4_SB(inode
->i_sb
)->s_gdb_count
;
4631 /* bitmaps and block group descriptor blocks */
4632 ret
+= groups
+ gdpblocks
;
4634 /* Blocks for super block, inode, quota and xattr blocks */
4635 ret
+= EXT4_META_TRANS_BLOCKS(inode
->i_sb
);
4641 * Calulate the total number of credits to reserve to fit
4642 * the modification of a single pages into a single transaction,
4643 * which may include multiple chunks of block allocations.
4645 * This could be called via ext4_write_begin()
4647 * We need to consider the worse case, when
4648 * one new block per extent.
4650 int ext4_writepage_trans_blocks(struct inode
*inode
)
4652 int bpp
= ext4_journal_blocks_per_page(inode
);
4655 ret
= ext4_meta_trans_blocks(inode
, bpp
, 0);
4657 /* Account for data blocks for journalled mode */
4658 if (ext4_should_journal_data(inode
))
4664 * Calculate the journal credits for a chunk of data modification.
4666 * This is called from DIO, fallocate or whoever calling
4667 * ext4_get_blocks_wrap() to map/allocate a chunk of contigous disk blocks.
4669 * journal buffers for data blocks are not included here, as DIO
4670 * and fallocate do no need to journal data buffers.
4672 int ext4_chunk_trans_blocks(struct inode
*inode
, int nrblocks
)
4674 return ext4_meta_trans_blocks(inode
, nrblocks
, 1);
4678 * The caller must have previously called ext4_reserve_inode_write().
4679 * Give this, we know that the caller already has write access to iloc->bh.
4681 int ext4_mark_iloc_dirty(handle_t
*handle
,
4682 struct inode
*inode
, struct ext4_iloc
*iloc
)
4686 if (test_opt(inode
->i_sb
, I_VERSION
))
4687 inode_inc_iversion(inode
);
4689 /* the do_update_inode consumes one bh->b_count */
4692 /* ext4_do_update_inode() does jbd2_journal_dirty_metadata */
4693 err
= ext4_do_update_inode(handle
, inode
, iloc
);
4699 * On success, We end up with an outstanding reference count against
4700 * iloc->bh. This _must_ be cleaned up later.
4704 ext4_reserve_inode_write(handle_t
*handle
, struct inode
*inode
,
4705 struct ext4_iloc
*iloc
)
4709 err
= ext4_get_inode_loc(inode
, iloc
);
4711 BUFFER_TRACE(iloc
->bh
, "get_write_access");
4712 err
= ext4_journal_get_write_access(handle
, iloc
->bh
);
4719 ext4_std_error(inode
->i_sb
, err
);
4724 * Expand an inode by new_extra_isize bytes.
4725 * Returns 0 on success or negative error number on failure.
4727 static int ext4_expand_extra_isize(struct inode
*inode
,
4728 unsigned int new_extra_isize
,
4729 struct ext4_iloc iloc
,
4732 struct ext4_inode
*raw_inode
;
4733 struct ext4_xattr_ibody_header
*header
;
4734 struct ext4_xattr_entry
*entry
;
4736 if (EXT4_I(inode
)->i_extra_isize
>= new_extra_isize
)
4739 raw_inode
= ext4_raw_inode(&iloc
);
4741 header
= IHDR(inode
, raw_inode
);
4742 entry
= IFIRST(header
);
4744 /* No extended attributes present */
4745 if (!(EXT4_I(inode
)->i_state
& EXT4_STATE_XATTR
) ||
4746 header
->h_magic
!= cpu_to_le32(EXT4_XATTR_MAGIC
)) {
4747 memset((void *)raw_inode
+ EXT4_GOOD_OLD_INODE_SIZE
, 0,
4749 EXT4_I(inode
)->i_extra_isize
= new_extra_isize
;
4753 /* try to expand with EAs present */
4754 return ext4_expand_extra_isize_ea(inode
, new_extra_isize
,
4759 * What we do here is to mark the in-core inode as clean with respect to inode
4760 * dirtiness (it may still be data-dirty).
4761 * This means that the in-core inode may be reaped by prune_icache
4762 * without having to perform any I/O. This is a very good thing,
4763 * because *any* task may call prune_icache - even ones which
4764 * have a transaction open against a different journal.
4766 * Is this cheating? Not really. Sure, we haven't written the
4767 * inode out, but prune_icache isn't a user-visible syncing function.
4768 * Whenever the user wants stuff synced (sys_sync, sys_msync, sys_fsync)
4769 * we start and wait on commits.
4771 * Is this efficient/effective? Well, we're being nice to the system
4772 * by cleaning up our inodes proactively so they can be reaped
4773 * without I/O. But we are potentially leaving up to five seconds'
4774 * worth of inodes floating about which prune_icache wants us to
4775 * write out. One way to fix that would be to get prune_icache()
4776 * to do a write_super() to free up some memory. It has the desired
4779 int ext4_mark_inode_dirty(handle_t
*handle
, struct inode
*inode
)
4781 struct ext4_iloc iloc
;
4782 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
4783 static unsigned int mnt_count
;
4787 err
= ext4_reserve_inode_write(handle
, inode
, &iloc
);
4788 if (EXT4_I(inode
)->i_extra_isize
< sbi
->s_want_extra_isize
&&
4789 !(EXT4_I(inode
)->i_state
& EXT4_STATE_NO_EXPAND
)) {
4791 * We need extra buffer credits since we may write into EA block
4792 * with this same handle. If journal_extend fails, then it will
4793 * only result in a minor loss of functionality for that inode.
4794 * If this is felt to be critical, then e2fsck should be run to
4795 * force a large enough s_min_extra_isize.
4797 if ((jbd2_journal_extend(handle
,
4798 EXT4_DATA_TRANS_BLOCKS(inode
->i_sb
))) == 0) {
4799 ret
= ext4_expand_extra_isize(inode
,
4800 sbi
->s_want_extra_isize
,
4803 EXT4_I(inode
)->i_state
|= EXT4_STATE_NO_EXPAND
;
4805 le16_to_cpu(sbi
->s_es
->s_mnt_count
)) {
4806 ext4_warning(inode
->i_sb
, __func__
,
4807 "Unable to expand inode %lu. Delete"
4808 " some EAs or run e2fsck.",
4811 le16_to_cpu(sbi
->s_es
->s_mnt_count
);
4817 err
= ext4_mark_iloc_dirty(handle
, inode
, &iloc
);
4822 * ext4_dirty_inode() is called from __mark_inode_dirty()
4824 * We're really interested in the case where a file is being extended.
4825 * i_size has been changed by generic_commit_write() and we thus need
4826 * to include the updated inode in the current transaction.
4828 * Also, DQUOT_ALLOC_SPACE() will always dirty the inode when blocks
4829 * are allocated to the file.
4831 * If the inode is marked synchronous, we don't honour that here - doing
4832 * so would cause a commit on atime updates, which we don't bother doing.
4833 * We handle synchronous inodes at the highest possible level.
4835 void ext4_dirty_inode(struct inode
*inode
)
4837 handle_t
*current_handle
= ext4_journal_current_handle();
4840 handle
= ext4_journal_start(inode
, 2);
4843 if (current_handle
&&
4844 current_handle
->h_transaction
!= handle
->h_transaction
) {
4845 /* This task has a transaction open against a different fs */
4846 printk(KERN_EMERG
"%s: transactions do not match!\n",
4849 jbd_debug(5, "marking dirty. outer handle=%p\n",
4851 ext4_mark_inode_dirty(handle
, inode
);
4853 ext4_journal_stop(handle
);
4860 * Bind an inode's backing buffer_head into this transaction, to prevent
4861 * it from being flushed to disk early. Unlike
4862 * ext4_reserve_inode_write, this leaves behind no bh reference and
4863 * returns no iloc structure, so the caller needs to repeat the iloc
4864 * lookup to mark the inode dirty later.
4866 static int ext4_pin_inode(handle_t
*handle
, struct inode
*inode
)
4868 struct ext4_iloc iloc
;
4872 err
= ext4_get_inode_loc(inode
, &iloc
);
4874 BUFFER_TRACE(iloc
.bh
, "get_write_access");
4875 err
= jbd2_journal_get_write_access(handle
, iloc
.bh
);
4877 err
= ext4_journal_dirty_metadata(handle
,
4882 ext4_std_error(inode
->i_sb
, err
);
4887 int ext4_change_inode_journal_flag(struct inode
*inode
, int val
)
4894 * We have to be very careful here: changing a data block's
4895 * journaling status dynamically is dangerous. If we write a
4896 * data block to the journal, change the status and then delete
4897 * that block, we risk forgetting to revoke the old log record
4898 * from the journal and so a subsequent replay can corrupt data.
4899 * So, first we make sure that the journal is empty and that
4900 * nobody is changing anything.
4903 journal
= EXT4_JOURNAL(inode
);
4904 if (is_journal_aborted(journal
))
4907 jbd2_journal_lock_updates(journal
);
4908 jbd2_journal_flush(journal
);
4911 * OK, there are no updates running now, and all cached data is
4912 * synced to disk. We are now in a completely consistent state
4913 * which doesn't have anything in the journal, and we know that
4914 * no filesystem updates are running, so it is safe to modify
4915 * the inode's in-core data-journaling state flag now.
4919 EXT4_I(inode
)->i_flags
|= EXT4_JOURNAL_DATA_FL
;
4921 EXT4_I(inode
)->i_flags
&= ~EXT4_JOURNAL_DATA_FL
;
4922 ext4_set_aops(inode
);
4924 jbd2_journal_unlock_updates(journal
);
4926 /* Finally we can mark the inode as dirty. */
4928 handle
= ext4_journal_start(inode
, 1);
4930 return PTR_ERR(handle
);
4932 err
= ext4_mark_inode_dirty(handle
, inode
);
4934 ext4_journal_stop(handle
);
4935 ext4_std_error(inode
->i_sb
, err
);
4940 static int ext4_bh_unmapped(handle_t
*handle
, struct buffer_head
*bh
)
4942 return !buffer_mapped(bh
);
4945 int ext4_page_mkwrite(struct vm_area_struct
*vma
, struct page
*page
)
4951 struct file
*file
= vma
->vm_file
;
4952 struct inode
*inode
= file
->f_path
.dentry
->d_inode
;
4953 struct address_space
*mapping
= inode
->i_mapping
;
4956 * Get i_alloc_sem to stop truncates messing with the inode. We cannot
4957 * get i_mutex because we are already holding mmap_sem.
4959 down_read(&inode
->i_alloc_sem
);
4960 size
= i_size_read(inode
);
4961 if (page
->mapping
!= mapping
|| size
<= page_offset(page
)
4962 || !PageUptodate(page
)) {
4963 /* page got truncated from under us? */
4967 if (PageMappedToDisk(page
))
4970 if (page
->index
== size
>> PAGE_CACHE_SHIFT
)
4971 len
= size
& ~PAGE_CACHE_MASK
;
4973 len
= PAGE_CACHE_SIZE
;
4975 if (page_has_buffers(page
)) {
4976 /* return if we have all the buffers mapped */
4977 if (!walk_page_buffers(NULL
, page_buffers(page
), 0, len
, NULL
,
4982 * OK, we need to fill the hole... Do write_begin write_end
4983 * to do block allocation/reservation.We are not holding
4984 * inode.i__mutex here. That allow * parallel write_begin,
4985 * write_end call. lock_page prevent this from happening
4986 * on the same page though
4988 ret
= mapping
->a_ops
->write_begin(file
, mapping
, page_offset(page
),
4989 len
, AOP_FLAG_UNINTERRUPTIBLE
, &page
, &fsdata
);
4992 ret
= mapping
->a_ops
->write_end(file
, mapping
, page_offset(page
),
4993 len
, len
, page
, fsdata
);
4998 up_read(&inode
->i_alloc_sem
);