2 * linux/fs/ext4/inode.c
4 * Copyright (C) 1992, 1993, 1994, 1995
5 * Remy Card (card@masi.ibp.fr)
6 * Laboratoire MASI - Institut Blaise Pascal
7 * Universite Pierre et Marie Curie (Paris VI)
11 * linux/fs/minix/inode.c
13 * Copyright (C) 1991, 1992 Linus Torvalds
15 * Goal-directed block allocation by Stephen Tweedie
16 * (sct@redhat.com), 1993, 1998
17 * Big-endian to little-endian byte-swapping/bitmaps by
18 * David S. Miller (davem@caip.rutgers.edu), 1995
19 * 64-bit file support on 64-bit platforms by Jakub Jelinek
20 * (jj@sunsite.ms.mff.cuni.cz)
22 * Assorted race fixes, rewrite of ext4_get_block() by Al Viro, 2000
25 #include <linux/module.h>
27 #include <linux/time.h>
28 #include <linux/jbd2.h>
29 #include <linux/highuid.h>
30 #include <linux/pagemap.h>
31 #include <linux/quotaops.h>
32 #include <linux/string.h>
33 #include <linux/buffer_head.h>
34 #include <linux/writeback.h>
35 #include <linux/pagevec.h>
36 #include <linux/mpage.h>
37 #include <linux/uio.h>
38 #include <linux/bio.h>
39 #include "ext4_jbd2.h"
42 #include "ext4_extents.h"
44 #define MPAGE_DA_EXTENT_TAIL 0x01
46 static inline int ext4_begin_ordered_truncate(struct inode
*inode
,
49 return jbd2_journal_begin_ordered_truncate(&EXT4_I(inode
)->jinode
,
53 static void ext4_invalidatepage(struct page
*page
, unsigned long offset
);
56 * Test whether an inode is a fast symlink.
58 static int ext4_inode_is_fast_symlink(struct inode
*inode
)
60 int ea_blocks
= EXT4_I(inode
)->i_file_acl
?
61 (inode
->i_sb
->s_blocksize
>> 9) : 0;
63 return (S_ISLNK(inode
->i_mode
) && inode
->i_blocks
- ea_blocks
== 0);
67 * The ext4 forget function must perform a revoke if we are freeing data
68 * which has been journaled. Metadata (eg. indirect blocks) must be
69 * revoked in all cases.
71 * "bh" may be NULL: a metadata block may have been freed from memory
72 * but there may still be a record of it in the journal, and that record
73 * still needs to be revoked.
75 int ext4_forget(handle_t
*handle
, int is_metadata
, struct inode
*inode
,
76 struct buffer_head
*bh
, ext4_fsblk_t blocknr
)
82 BUFFER_TRACE(bh
, "enter");
84 jbd_debug(4, "forgetting bh %p: is_metadata = %d, mode %o, "
86 bh
, is_metadata
, inode
->i_mode
,
87 test_opt(inode
->i_sb
, DATA_FLAGS
));
89 /* Never use the revoke function if we are doing full data
90 * journaling: there is no need to, and a V1 superblock won't
91 * support it. Otherwise, only skip the revoke on un-journaled
94 if (test_opt(inode
->i_sb
, DATA_FLAGS
) == EXT4_MOUNT_JOURNAL_DATA
||
95 (!is_metadata
&& !ext4_should_journal_data(inode
))) {
97 BUFFER_TRACE(bh
, "call jbd2_journal_forget");
98 return ext4_journal_forget(handle
, bh
);
104 * data!=journal && (is_metadata || should_journal_data(inode))
106 BUFFER_TRACE(bh
, "call ext4_journal_revoke");
107 err
= ext4_journal_revoke(handle
, blocknr
, bh
);
109 ext4_abort(inode
->i_sb
, __func__
,
110 "error %d when attempting revoke", err
);
111 BUFFER_TRACE(bh
, "exit");
116 * Work out how many blocks we need to proceed with the next chunk of a
117 * truncate transaction.
119 static unsigned long blocks_for_truncate(struct inode
*inode
)
123 needed
= inode
->i_blocks
>> (inode
->i_sb
->s_blocksize_bits
- 9);
125 /* Give ourselves just enough room to cope with inodes in which
126 * i_blocks is corrupt: we've seen disk corruptions in the past
127 * which resulted in random data in an inode which looked enough
128 * like a regular file for ext4 to try to delete it. Things
129 * will go a bit crazy if that happens, but at least we should
130 * try not to panic the whole kernel. */
134 /* But we need to bound the transaction so we don't overflow the
136 if (needed
> EXT4_MAX_TRANS_DATA
)
137 needed
= EXT4_MAX_TRANS_DATA
;
139 return EXT4_DATA_TRANS_BLOCKS(inode
->i_sb
) + needed
;
143 * Truncate transactions can be complex and absolutely huge. So we need to
144 * be able to restart the transaction at a conventient checkpoint to make
145 * sure we don't overflow the journal.
147 * start_transaction gets us a new handle for a truncate transaction,
148 * and extend_transaction tries to extend the existing one a bit. If
149 * extend fails, we need to propagate the failure up and restart the
150 * transaction in the top-level truncate loop. --sct
152 static handle_t
*start_transaction(struct inode
*inode
)
156 result
= ext4_journal_start(inode
, blocks_for_truncate(inode
));
160 ext4_std_error(inode
->i_sb
, PTR_ERR(result
));
165 * Try to extend this transaction for the purposes of truncation.
167 * Returns 0 if we managed to create more room. If we can't create more
168 * room, and the transaction must be restarted we return 1.
170 static int try_to_extend_transaction(handle_t
*handle
, struct inode
*inode
)
172 if (handle
->h_buffer_credits
> EXT4_RESERVE_TRANS_BLOCKS
)
174 if (!ext4_journal_extend(handle
, blocks_for_truncate(inode
)))
180 * Restart the transaction associated with *handle. This does a commit,
181 * so before we call here everything must be consistently dirtied against
184 static int ext4_journal_test_restart(handle_t
*handle
, struct inode
*inode
)
186 jbd_debug(2, "restarting handle %p\n", handle
);
187 return ext4_journal_restart(handle
, blocks_for_truncate(inode
));
191 * Called at the last iput() if i_nlink is zero.
193 void ext4_delete_inode(struct inode
*inode
)
198 if (ext4_should_order_data(inode
))
199 ext4_begin_ordered_truncate(inode
, 0);
200 truncate_inode_pages(&inode
->i_data
, 0);
202 if (is_bad_inode(inode
))
205 handle
= ext4_journal_start(inode
, blocks_for_truncate(inode
)+3);
206 if (IS_ERR(handle
)) {
207 ext4_std_error(inode
->i_sb
, PTR_ERR(handle
));
209 * If we're going to skip the normal cleanup, we still need to
210 * make sure that the in-core orphan linked list is properly
213 ext4_orphan_del(NULL
, inode
);
220 err
= ext4_mark_inode_dirty(handle
, inode
);
222 ext4_warning(inode
->i_sb
, __func__
,
223 "couldn't mark inode dirty (err %d)", err
);
227 ext4_truncate(inode
);
230 * ext4_ext_truncate() doesn't reserve any slop when it
231 * restarts journal transactions; therefore there may not be
232 * enough credits left in the handle to remove the inode from
233 * the orphan list and set the dtime field.
235 if (handle
->h_buffer_credits
< 3) {
236 err
= ext4_journal_extend(handle
, 3);
238 err
= ext4_journal_restart(handle
, 3);
240 ext4_warning(inode
->i_sb
, __func__
,
241 "couldn't extend journal (err %d)", err
);
243 ext4_journal_stop(handle
);
249 * Kill off the orphan record which ext4_truncate created.
250 * AKPM: I think this can be inside the above `if'.
251 * Note that ext4_orphan_del() has to be able to cope with the
252 * deletion of a non-existent orphan - this is because we don't
253 * know if ext4_truncate() actually created an orphan record.
254 * (Well, we could do this if we need to, but heck - it works)
256 ext4_orphan_del(handle
, inode
);
257 EXT4_I(inode
)->i_dtime
= get_seconds();
260 * One subtle ordering requirement: if anything has gone wrong
261 * (transaction abort, IO errors, whatever), then we can still
262 * do these next steps (the fs will already have been marked as
263 * having errors), but we can't free the inode if the mark_dirty
266 if (ext4_mark_inode_dirty(handle
, inode
))
267 /* If that failed, just do the required in-core inode clear. */
270 ext4_free_inode(handle
, inode
);
271 ext4_journal_stop(handle
);
274 clear_inode(inode
); /* We must guarantee clearing of inode... */
280 struct buffer_head
*bh
;
283 static inline void add_chain(Indirect
*p
, struct buffer_head
*bh
, __le32
*v
)
285 p
->key
= *(p
->p
= v
);
290 * ext4_block_to_path - parse the block number into array of offsets
291 * @inode: inode in question (we are only interested in its superblock)
292 * @i_block: block number to be parsed
293 * @offsets: array to store the offsets in
294 * @boundary: set this non-zero if the referred-to block is likely to be
295 * followed (on disk) by an indirect block.
297 * To store the locations of file's data ext4 uses a data structure common
298 * for UNIX filesystems - tree of pointers anchored in the inode, with
299 * data blocks at leaves and indirect blocks in intermediate nodes.
300 * This function translates the block number into path in that tree -
301 * return value is the path length and @offsets[n] is the offset of
302 * pointer to (n+1)th node in the nth one. If @block is out of range
303 * (negative or too large) warning is printed and zero returned.
305 * Note: function doesn't find node addresses, so no IO is needed. All
306 * we need to know is the capacity of indirect blocks (taken from the
311 * Portability note: the last comparison (check that we fit into triple
312 * indirect block) is spelled differently, because otherwise on an
313 * architecture with 32-bit longs and 8Kb pages we might get into trouble
314 * if our filesystem had 8Kb blocks. We might use long long, but that would
315 * kill us on x86. Oh, well, at least the sign propagation does not matter -
316 * i_block would have to be negative in the very beginning, so we would not
320 static int ext4_block_to_path(struct inode
*inode
,
322 ext4_lblk_t offsets
[4], int *boundary
)
324 int ptrs
= EXT4_ADDR_PER_BLOCK(inode
->i_sb
);
325 int ptrs_bits
= EXT4_ADDR_PER_BLOCK_BITS(inode
->i_sb
);
326 const long direct_blocks
= EXT4_NDIR_BLOCKS
,
327 indirect_blocks
= ptrs
,
328 double_blocks
= (1 << (ptrs_bits
* 2));
333 ext4_warning(inode
->i_sb
, "ext4_block_to_path", "block < 0");
334 } else if (i_block
< direct_blocks
) {
335 offsets
[n
++] = i_block
;
336 final
= direct_blocks
;
337 } else if ((i_block
-= direct_blocks
) < indirect_blocks
) {
338 offsets
[n
++] = EXT4_IND_BLOCK
;
339 offsets
[n
++] = i_block
;
341 } else if ((i_block
-= indirect_blocks
) < double_blocks
) {
342 offsets
[n
++] = EXT4_DIND_BLOCK
;
343 offsets
[n
++] = i_block
>> ptrs_bits
;
344 offsets
[n
++] = i_block
& (ptrs
- 1);
346 } else if (((i_block
-= double_blocks
) >> (ptrs_bits
* 2)) < ptrs
) {
347 offsets
[n
++] = EXT4_TIND_BLOCK
;
348 offsets
[n
++] = i_block
>> (ptrs_bits
* 2);
349 offsets
[n
++] = (i_block
>> ptrs_bits
) & (ptrs
- 1);
350 offsets
[n
++] = i_block
& (ptrs
- 1);
353 ext4_warning(inode
->i_sb
, "ext4_block_to_path",
355 i_block
+ direct_blocks
+
356 indirect_blocks
+ double_blocks
);
359 *boundary
= final
- 1 - (i_block
& (ptrs
- 1));
364 * ext4_get_branch - read the chain of indirect blocks leading to data
365 * @inode: inode in question
366 * @depth: depth of the chain (1 - direct pointer, etc.)
367 * @offsets: offsets of pointers in inode/indirect blocks
368 * @chain: place to store the result
369 * @err: here we store the error value
371 * Function fills the array of triples <key, p, bh> and returns %NULL
372 * if everything went OK or the pointer to the last filled triple
373 * (incomplete one) otherwise. Upon the return chain[i].key contains
374 * the number of (i+1)-th block in the chain (as it is stored in memory,
375 * i.e. little-endian 32-bit), chain[i].p contains the address of that
376 * number (it points into struct inode for i==0 and into the bh->b_data
377 * for i>0) and chain[i].bh points to the buffer_head of i-th indirect
378 * block for i>0 and NULL for i==0. In other words, it holds the block
379 * numbers of the chain, addresses they were taken from (and where we can
380 * verify that chain did not change) and buffer_heads hosting these
383 * Function stops when it stumbles upon zero pointer (absent block)
384 * (pointer to last triple returned, *@err == 0)
385 * or when it gets an IO error reading an indirect block
386 * (ditto, *@err == -EIO)
387 * or when it reads all @depth-1 indirect blocks successfully and finds
388 * the whole chain, all way to the data (returns %NULL, *err == 0).
390 * Need to be called with
391 * down_read(&EXT4_I(inode)->i_data_sem)
393 static Indirect
*ext4_get_branch(struct inode
*inode
, int depth
,
394 ext4_lblk_t
*offsets
,
395 Indirect chain
[4], int *err
)
397 struct super_block
*sb
= inode
->i_sb
;
399 struct buffer_head
*bh
;
402 /* i_data is not going away, no lock needed */
403 add_chain(chain
, NULL
, EXT4_I(inode
)->i_data
+ *offsets
);
407 bh
= sb_bread(sb
, le32_to_cpu(p
->key
));
410 add_chain(++p
, bh
, (__le32
*)bh
->b_data
+ *++offsets
);
424 * ext4_find_near - find a place for allocation with sufficient locality
426 * @ind: descriptor of indirect block.
428 * This function returns the preferred place for block allocation.
429 * It is used when heuristic for sequential allocation fails.
431 * + if there is a block to the left of our position - allocate near it.
432 * + if pointer will live in indirect block - allocate near that block.
433 * + if pointer will live in inode - allocate in the same
436 * In the latter case we colour the starting block by the callers PID to
437 * prevent it from clashing with concurrent allocations for a different inode
438 * in the same block group. The PID is used here so that functionally related
439 * files will be close-by on-disk.
441 * Caller must make sure that @ind is valid and will stay that way.
443 static ext4_fsblk_t
ext4_find_near(struct inode
*inode
, Indirect
*ind
)
445 struct ext4_inode_info
*ei
= EXT4_I(inode
);
446 __le32
*start
= ind
->bh
? (__le32
*) ind
->bh
->b_data
: ei
->i_data
;
448 ext4_fsblk_t bg_start
;
449 ext4_fsblk_t last_block
;
450 ext4_grpblk_t colour
;
452 /* Try to find previous block */
453 for (p
= ind
->p
- 1; p
>= start
; p
--) {
455 return le32_to_cpu(*p
);
458 /* No such thing, so let's try location of indirect block */
460 return ind
->bh
->b_blocknr
;
463 * It is going to be referred to from the inode itself? OK, just put it
464 * into the same cylinder group then.
466 bg_start
= ext4_group_first_block_no(inode
->i_sb
, ei
->i_block_group
);
467 last_block
= ext4_blocks_count(EXT4_SB(inode
->i_sb
)->s_es
) - 1;
469 if (bg_start
+ EXT4_BLOCKS_PER_GROUP(inode
->i_sb
) <= last_block
)
470 colour
= (current
->pid
% 16) *
471 (EXT4_BLOCKS_PER_GROUP(inode
->i_sb
) / 16);
473 colour
= (current
->pid
% 16) * ((last_block
- bg_start
) / 16);
474 return bg_start
+ colour
;
478 * ext4_find_goal - find a preferred place for allocation.
480 * @block: block we want
481 * @partial: pointer to the last triple within a chain
483 * Normally this function find the preferred place for block allocation,
486 static ext4_fsblk_t
ext4_find_goal(struct inode
*inode
, ext4_lblk_t block
,
489 struct ext4_block_alloc_info
*block_i
;
491 block_i
= EXT4_I(inode
)->i_block_alloc_info
;
494 * try the heuristic for sequential allocation,
495 * failing that at least try to get decent locality.
497 if (block_i
&& (block
== block_i
->last_alloc_logical_block
+ 1)
498 && (block_i
->last_alloc_physical_block
!= 0)) {
499 return block_i
->last_alloc_physical_block
+ 1;
502 return ext4_find_near(inode
, partial
);
506 * ext4_blks_to_allocate: Look up the block map and count the number
507 * of direct blocks need to be allocated for the given branch.
509 * @branch: chain of indirect blocks
510 * @k: number of blocks need for indirect blocks
511 * @blks: number of data blocks to be mapped.
512 * @blocks_to_boundary: the offset in the indirect block
514 * return the total number of blocks to be allocate, including the
515 * direct and indirect blocks.
517 static int ext4_blks_to_allocate(Indirect
*branch
, int k
, unsigned long blks
,
518 int blocks_to_boundary
)
520 unsigned long count
= 0;
523 * Simple case, [t,d]Indirect block(s) has not allocated yet
524 * then it's clear blocks on that path have not allocated
527 /* right now we don't handle cross boundary allocation */
528 if (blks
< blocks_to_boundary
+ 1)
531 count
+= blocks_to_boundary
+ 1;
536 while (count
< blks
&& count
<= blocks_to_boundary
&&
537 le32_to_cpu(*(branch
[0].p
+ count
)) == 0) {
544 * ext4_alloc_blocks: multiple allocate blocks needed for a branch
545 * @indirect_blks: the number of blocks need to allocate for indirect
548 * @new_blocks: on return it will store the new block numbers for
549 * the indirect blocks(if needed) and the first direct block,
550 * @blks: on return it will store the total number of allocated
553 static int ext4_alloc_blocks(handle_t
*handle
, struct inode
*inode
,
554 ext4_lblk_t iblock
, ext4_fsblk_t goal
,
555 int indirect_blks
, int blks
,
556 ext4_fsblk_t new_blocks
[4], int *err
)
559 unsigned long count
= 0, blk_allocated
= 0;
561 ext4_fsblk_t current_block
= 0;
565 * Here we try to allocate the requested multiple blocks at once,
566 * on a best-effort basis.
567 * To build a branch, we should allocate blocks for
568 * the indirect blocks(if not allocated yet), and at least
569 * the first direct block of this branch. That's the
570 * minimum number of blocks need to allocate(required)
572 /* first we try to allocate the indirect blocks */
573 target
= indirect_blks
;
576 /* allocating blocks for indirect blocks and direct blocks */
577 current_block
= ext4_new_meta_blocks(handle
, inode
,
583 /* allocate blocks for indirect blocks */
584 while (index
< indirect_blks
&& count
) {
585 new_blocks
[index
++] = current_block
++;
590 * save the new block number
591 * for the first direct block
593 new_blocks
[index
] = current_block
;
594 printk(KERN_INFO
"%s returned more blocks than "
595 "requested\n", __func__
);
601 target
= blks
- count
;
602 blk_allocated
= count
;
605 /* Now allocate data blocks */
607 /* allocating blocks for data blocks */
608 current_block
= ext4_new_blocks(handle
, inode
, iblock
,
610 if (*err
&& (target
== blks
)) {
612 * if the allocation failed and we didn't allocate
618 if (target
== blks
) {
620 * save the new block number
621 * for the first direct block
623 new_blocks
[index
] = current_block
;
625 blk_allocated
+= count
;
628 /* total number of blocks allocated for direct blocks */
633 for (i
= 0; i
< index
; i
++)
634 ext4_free_blocks(handle
, inode
, new_blocks
[i
], 1, 0);
639 * ext4_alloc_branch - allocate and set up a chain of blocks.
641 * @indirect_blks: number of allocated indirect blocks
642 * @blks: number of allocated direct blocks
643 * @offsets: offsets (in the blocks) to store the pointers to next.
644 * @branch: place to store the chain in.
646 * This function allocates blocks, zeroes out all but the last one,
647 * links them into chain and (if we are synchronous) writes them to disk.
648 * In other words, it prepares a branch that can be spliced onto the
649 * inode. It stores the information about that chain in the branch[], in
650 * the same format as ext4_get_branch() would do. We are calling it after
651 * we had read the existing part of chain and partial points to the last
652 * triple of that (one with zero ->key). Upon the exit we have the same
653 * picture as after the successful ext4_get_block(), except that in one
654 * place chain is disconnected - *branch->p is still zero (we did not
655 * set the last link), but branch->key contains the number that should
656 * be placed into *branch->p to fill that gap.
658 * If allocation fails we free all blocks we've allocated (and forget
659 * their buffer_heads) and return the error value the from failed
660 * ext4_alloc_block() (normally -ENOSPC). Otherwise we set the chain
661 * as described above and return 0.
663 static int ext4_alloc_branch(handle_t
*handle
, struct inode
*inode
,
664 ext4_lblk_t iblock
, int indirect_blks
,
665 int *blks
, ext4_fsblk_t goal
,
666 ext4_lblk_t
*offsets
, Indirect
*branch
)
668 int blocksize
= inode
->i_sb
->s_blocksize
;
671 struct buffer_head
*bh
;
673 ext4_fsblk_t new_blocks
[4];
674 ext4_fsblk_t current_block
;
676 num
= ext4_alloc_blocks(handle
, inode
, iblock
, goal
, indirect_blks
,
677 *blks
, new_blocks
, &err
);
681 branch
[0].key
= cpu_to_le32(new_blocks
[0]);
683 * metadata blocks and data blocks are allocated.
685 for (n
= 1; n
<= indirect_blks
; n
++) {
687 * Get buffer_head for parent block, zero it out
688 * and set the pointer to new one, then send
691 bh
= sb_getblk(inode
->i_sb
, new_blocks
[n
-1]);
694 BUFFER_TRACE(bh
, "call get_create_access");
695 err
= ext4_journal_get_create_access(handle
, bh
);
702 memset(bh
->b_data
, 0, blocksize
);
703 branch
[n
].p
= (__le32
*) bh
->b_data
+ offsets
[n
];
704 branch
[n
].key
= cpu_to_le32(new_blocks
[n
]);
705 *branch
[n
].p
= branch
[n
].key
;
706 if (n
== indirect_blks
) {
707 current_block
= new_blocks
[n
];
709 * End of chain, update the last new metablock of
710 * the chain to point to the new allocated
711 * data blocks numbers
713 for (i
=1; i
< num
; i
++)
714 *(branch
[n
].p
+ i
) = cpu_to_le32(++current_block
);
716 BUFFER_TRACE(bh
, "marking uptodate");
717 set_buffer_uptodate(bh
);
720 BUFFER_TRACE(bh
, "call ext4_journal_dirty_metadata");
721 err
= ext4_journal_dirty_metadata(handle
, bh
);
728 /* Allocation failed, free what we already allocated */
729 for (i
= 1; i
<= n
; i
++) {
730 BUFFER_TRACE(branch
[i
].bh
, "call jbd2_journal_forget");
731 ext4_journal_forget(handle
, branch
[i
].bh
);
733 for (i
= 0; i
< indirect_blks
; i
++)
734 ext4_free_blocks(handle
, inode
, new_blocks
[i
], 1, 0);
736 ext4_free_blocks(handle
, inode
, new_blocks
[i
], num
, 0);
742 * ext4_splice_branch - splice the allocated branch onto inode.
744 * @block: (logical) number of block we are adding
745 * @chain: chain of indirect blocks (with a missing link - see
747 * @where: location of missing link
748 * @num: number of indirect blocks we are adding
749 * @blks: number of direct blocks we are adding
751 * This function fills the missing link and does all housekeeping needed in
752 * inode (->i_blocks, etc.). In case of success we end up with the full
753 * chain to new block and return 0.
755 static int ext4_splice_branch(handle_t
*handle
, struct inode
*inode
,
756 ext4_lblk_t block
, Indirect
*where
, int num
, int blks
)
760 struct ext4_block_alloc_info
*block_i
;
761 ext4_fsblk_t current_block
;
763 block_i
= EXT4_I(inode
)->i_block_alloc_info
;
765 * If we're splicing into a [td]indirect block (as opposed to the
766 * inode) then we need to get write access to the [td]indirect block
770 BUFFER_TRACE(where
->bh
, "get_write_access");
771 err
= ext4_journal_get_write_access(handle
, where
->bh
);
777 *where
->p
= where
->key
;
780 * Update the host buffer_head or inode to point to more just allocated
781 * direct blocks blocks
783 if (num
== 0 && blks
> 1) {
784 current_block
= le32_to_cpu(where
->key
) + 1;
785 for (i
= 1; i
< blks
; i
++)
786 *(where
->p
+ i
) = cpu_to_le32(current_block
++);
790 * update the most recently allocated logical & physical block
791 * in i_block_alloc_info, to assist find the proper goal block for next
795 block_i
->last_alloc_logical_block
= block
+ blks
- 1;
796 block_i
->last_alloc_physical_block
=
797 le32_to_cpu(where
[num
].key
) + blks
- 1;
800 /* We are done with atomic stuff, now do the rest of housekeeping */
802 inode
->i_ctime
= ext4_current_time(inode
);
803 ext4_mark_inode_dirty(handle
, inode
);
805 /* had we spliced it onto indirect block? */
808 * If we spliced it onto an indirect block, we haven't
809 * altered the inode. Note however that if it is being spliced
810 * onto an indirect block at the very end of the file (the
811 * file is growing) then we *will* alter the inode to reflect
812 * the new i_size. But that is not done here - it is done in
813 * generic_commit_write->__mark_inode_dirty->ext4_dirty_inode.
815 jbd_debug(5, "splicing indirect only\n");
816 BUFFER_TRACE(where
->bh
, "call ext4_journal_dirty_metadata");
817 err
= ext4_journal_dirty_metadata(handle
, where
->bh
);
822 * OK, we spliced it into the inode itself on a direct block.
823 * Inode was dirtied above.
825 jbd_debug(5, "splicing direct\n");
830 for (i
= 1; i
<= num
; i
++) {
831 BUFFER_TRACE(where
[i
].bh
, "call jbd2_journal_forget");
832 ext4_journal_forget(handle
, where
[i
].bh
);
833 ext4_free_blocks(handle
, inode
,
834 le32_to_cpu(where
[i
-1].key
), 1, 0);
836 ext4_free_blocks(handle
, inode
, le32_to_cpu(where
[num
].key
), blks
, 0);
842 * Allocation strategy is simple: if we have to allocate something, we will
843 * have to go the whole way to leaf. So let's do it before attaching anything
844 * to tree, set linkage between the newborn blocks, write them if sync is
845 * required, recheck the path, free and repeat if check fails, otherwise
846 * set the last missing link (that will protect us from any truncate-generated
847 * removals - all blocks on the path are immune now) and possibly force the
848 * write on the parent block.
849 * That has a nice additional property: no special recovery from the failed
850 * allocations is needed - we simply release blocks and do not touch anything
851 * reachable from inode.
853 * `handle' can be NULL if create == 0.
855 * return > 0, # of blocks mapped or allocated.
856 * return = 0, if plain lookup failed.
857 * return < 0, error case.
860 * Need to be called with
861 * down_read(&EXT4_I(inode)->i_data_sem) if not allocating file system block
862 * (ie, create is zero). Otherwise down_write(&EXT4_I(inode)->i_data_sem)
864 int ext4_get_blocks_handle(handle_t
*handle
, struct inode
*inode
,
865 ext4_lblk_t iblock
, unsigned long maxblocks
,
866 struct buffer_head
*bh_result
,
867 int create
, int extend_disksize
)
870 ext4_lblk_t offsets
[4];
875 int blocks_to_boundary
= 0;
877 struct ext4_inode_info
*ei
= EXT4_I(inode
);
879 ext4_fsblk_t first_block
= 0;
883 J_ASSERT(!(EXT4_I(inode
)->i_flags
& EXT4_EXTENTS_FL
));
884 J_ASSERT(handle
!= NULL
|| create
== 0);
885 depth
= ext4_block_to_path(inode
, iblock
, offsets
,
886 &blocks_to_boundary
);
891 partial
= ext4_get_branch(inode
, depth
, offsets
, chain
, &err
);
893 /* Simplest case - block found, no allocation needed */
895 first_block
= le32_to_cpu(chain
[depth
- 1].key
);
896 clear_buffer_new(bh_result
);
899 while (count
< maxblocks
&& count
<= blocks_to_boundary
) {
902 blk
= le32_to_cpu(*(chain
[depth
-1].p
+ count
));
904 if (blk
== first_block
+ count
)
912 /* Next simple case - plain lookup or failed read of indirect block */
913 if (!create
|| err
== -EIO
)
917 * Okay, we need to do block allocation. Lazily initialize the block
918 * allocation info here if necessary
920 if (S_ISREG(inode
->i_mode
) && (!ei
->i_block_alloc_info
))
921 ext4_init_block_alloc_info(inode
);
923 goal
= ext4_find_goal(inode
, iblock
, partial
);
925 /* the number of blocks need to allocate for [d,t]indirect blocks */
926 indirect_blks
= (chain
+ depth
) - partial
- 1;
929 * Next look up the indirect map to count the totoal number of
930 * direct blocks to allocate for this branch.
932 count
= ext4_blks_to_allocate(partial
, indirect_blks
,
933 maxblocks
, blocks_to_boundary
);
935 * Block out ext4_truncate while we alter the tree
937 err
= ext4_alloc_branch(handle
, inode
, iblock
, indirect_blks
,
939 offsets
+ (partial
- chain
), partial
);
942 * The ext4_splice_branch call will free and forget any buffers
943 * on the new chain if there is a failure, but that risks using
944 * up transaction credits, especially for bitmaps where the
945 * credits cannot be returned. Can we handle this somehow? We
946 * may need to return -EAGAIN upwards in the worst case. --sct
949 err
= ext4_splice_branch(handle
, inode
, iblock
,
950 partial
, indirect_blks
, count
);
952 * i_disksize growing is protected by i_data_sem. Don't forget to
953 * protect it if you're about to implement concurrent
954 * ext4_get_block() -bzzz
956 if (!err
&& extend_disksize
) {
957 disksize
= ((loff_t
) iblock
+ count
) << inode
->i_blkbits
;
958 if (disksize
> i_size_read(inode
))
959 disksize
= i_size_read(inode
);
960 if (disksize
> ei
->i_disksize
)
961 ei
->i_disksize
= disksize
;
966 set_buffer_new(bh_result
);
968 map_bh(bh_result
, inode
->i_sb
, le32_to_cpu(chain
[depth
-1].key
));
969 if (count
> blocks_to_boundary
)
970 set_buffer_boundary(bh_result
);
972 /* Clean up and exit */
973 partial
= chain
+ depth
- 1; /* the whole chain */
975 while (partial
> chain
) {
976 BUFFER_TRACE(partial
->bh
, "call brelse");
980 BUFFER_TRACE(bh_result
, "returned");
986 * Calculate the number of metadata blocks need to reserve
987 * to allocate @blocks for non extent file based file
989 static int ext4_indirect_calc_metadata_amount(struct inode
*inode
, int blocks
)
991 int icap
= EXT4_ADDR_PER_BLOCK(inode
->i_sb
);
992 int ind_blks
, dind_blks
, tind_blks
;
994 /* number of new indirect blocks needed */
995 ind_blks
= (blocks
+ icap
- 1) / icap
;
997 dind_blks
= (ind_blks
+ icap
- 1) / icap
;
1001 return ind_blks
+ dind_blks
+ tind_blks
;
1005 * Calculate the number of metadata blocks need to reserve
1006 * to allocate given number of blocks
1008 static int ext4_calc_metadata_amount(struct inode
*inode
, int blocks
)
1013 if (EXT4_I(inode
)->i_flags
& EXT4_EXTENTS_FL
)
1014 return ext4_ext_calc_metadata_amount(inode
, blocks
);
1016 return ext4_indirect_calc_metadata_amount(inode
, blocks
);
1019 static void ext4_da_update_reserve_space(struct inode
*inode
, int used
)
1021 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
1022 int total
, mdb
, mdb_free
;
1024 spin_lock(&EXT4_I(inode
)->i_block_reservation_lock
);
1025 /* recalculate the number of metablocks still need to be reserved */
1026 total
= EXT4_I(inode
)->i_reserved_data_blocks
- used
;
1027 mdb
= ext4_calc_metadata_amount(inode
, total
);
1029 /* figure out how many metablocks to release */
1030 BUG_ON(mdb
> EXT4_I(inode
)->i_reserved_meta_blocks
);
1031 mdb_free
= EXT4_I(inode
)->i_reserved_meta_blocks
- mdb
;
1034 /* Account for allocated meta_blocks */
1035 mdb_free
-= EXT4_I(inode
)->i_allocated_meta_blocks
;
1037 /* update fs dirty blocks counter */
1038 percpu_counter_sub(&sbi
->s_dirtyblocks_counter
, mdb_free
);
1039 EXT4_I(inode
)->i_allocated_meta_blocks
= 0;
1040 EXT4_I(inode
)->i_reserved_meta_blocks
= mdb
;
1043 /* update per-inode reservations */
1044 BUG_ON(used
> EXT4_I(inode
)->i_reserved_data_blocks
);
1045 EXT4_I(inode
)->i_reserved_data_blocks
-= used
;
1047 spin_unlock(&EXT4_I(inode
)->i_block_reservation_lock
);
1051 * The ext4_get_blocks_wrap() function try to look up the requested blocks,
1052 * and returns if the blocks are already mapped.
1054 * Otherwise it takes the write lock of the i_data_sem and allocate blocks
1055 * and store the allocated blocks in the result buffer head and mark it
1058 * If file type is extents based, it will call ext4_ext_get_blocks(),
1059 * Otherwise, call with ext4_get_blocks_handle() to handle indirect mapping
1062 * On success, it returns the number of blocks being mapped or allocate.
1063 * if create==0 and the blocks are pre-allocated and uninitialized block,
1064 * the result buffer head is unmapped. If the create ==1, it will make sure
1065 * the buffer head is mapped.
1067 * It returns 0 if plain look up failed (blocks have not been allocated), in
1068 * that casem, buffer head is unmapped
1070 * It returns the error in case of allocation failure.
1072 int ext4_get_blocks_wrap(handle_t
*handle
, struct inode
*inode
, sector_t block
,
1073 unsigned long max_blocks
, struct buffer_head
*bh
,
1074 int create
, int extend_disksize
, int flag
)
1078 clear_buffer_mapped(bh
);
1081 * Try to see if we can get the block without requesting
1082 * for new file system block.
1084 down_read((&EXT4_I(inode
)->i_data_sem
));
1085 if (EXT4_I(inode
)->i_flags
& EXT4_EXTENTS_FL
) {
1086 retval
= ext4_ext_get_blocks(handle
, inode
, block
, max_blocks
,
1089 retval
= ext4_get_blocks_handle(handle
,
1090 inode
, block
, max_blocks
, bh
, 0, 0);
1092 up_read((&EXT4_I(inode
)->i_data_sem
));
1094 /* If it is only a block(s) look up */
1099 * Returns if the blocks have already allocated
1101 * Note that if blocks have been preallocated
1102 * ext4_ext_get_block() returns th create = 0
1103 * with buffer head unmapped.
1105 if (retval
> 0 && buffer_mapped(bh
))
1109 * New blocks allocate and/or writing to uninitialized extent
1110 * will possibly result in updating i_data, so we take
1111 * the write lock of i_data_sem, and call get_blocks()
1112 * with create == 1 flag.
1114 down_write((&EXT4_I(inode
)->i_data_sem
));
1117 * if the caller is from delayed allocation writeout path
1118 * we have already reserved fs blocks for allocation
1119 * let the underlying get_block() function know to
1120 * avoid double accounting
1123 EXT4_I(inode
)->i_delalloc_reserved_flag
= 1;
1125 * We need to check for EXT4 here because migrate
1126 * could have changed the inode type in between
1128 if (EXT4_I(inode
)->i_flags
& EXT4_EXTENTS_FL
) {
1129 retval
= ext4_ext_get_blocks(handle
, inode
, block
, max_blocks
,
1130 bh
, create
, extend_disksize
);
1132 retval
= ext4_get_blocks_handle(handle
, inode
, block
,
1133 max_blocks
, bh
, create
, extend_disksize
);
1135 if (retval
> 0 && buffer_new(bh
)) {
1137 * We allocated new blocks which will result in
1138 * i_data's format changing. Force the migrate
1139 * to fail by clearing migrate flags
1141 EXT4_I(inode
)->i_flags
= EXT4_I(inode
)->i_flags
&
1147 EXT4_I(inode
)->i_delalloc_reserved_flag
= 0;
1149 * Update reserved blocks/metadata blocks
1150 * after successful block allocation
1151 * which were deferred till now
1153 if ((retval
> 0) && buffer_delay(bh
))
1154 ext4_da_update_reserve_space(inode
, retval
);
1157 up_write((&EXT4_I(inode
)->i_data_sem
));
1161 /* Maximum number of blocks we map for direct IO at once. */
1162 #define DIO_MAX_BLOCKS 4096
1164 static int ext4_get_block(struct inode
*inode
, sector_t iblock
,
1165 struct buffer_head
*bh_result
, int create
)
1167 handle_t
*handle
= ext4_journal_current_handle();
1168 int ret
= 0, started
= 0;
1169 unsigned max_blocks
= bh_result
->b_size
>> inode
->i_blkbits
;
1172 if (create
&& !handle
) {
1173 /* Direct IO write... */
1174 if (max_blocks
> DIO_MAX_BLOCKS
)
1175 max_blocks
= DIO_MAX_BLOCKS
;
1176 dio_credits
= ext4_chunk_trans_blocks(inode
, max_blocks
);
1177 handle
= ext4_journal_start(inode
, dio_credits
);
1178 if (IS_ERR(handle
)) {
1179 ret
= PTR_ERR(handle
);
1185 ret
= ext4_get_blocks_wrap(handle
, inode
, iblock
,
1186 max_blocks
, bh_result
, create
, 0, 0);
1188 bh_result
->b_size
= (ret
<< inode
->i_blkbits
);
1192 ext4_journal_stop(handle
);
1198 * `handle' can be NULL if create is zero
1200 struct buffer_head
*ext4_getblk(handle_t
*handle
, struct inode
*inode
,
1201 ext4_lblk_t block
, int create
, int *errp
)
1203 struct buffer_head dummy
;
1206 J_ASSERT(handle
!= NULL
|| create
== 0);
1209 dummy
.b_blocknr
= -1000;
1210 buffer_trace_init(&dummy
.b_history
);
1211 err
= ext4_get_blocks_wrap(handle
, inode
, block
, 1,
1212 &dummy
, create
, 1, 0);
1214 * ext4_get_blocks_handle() returns number of blocks
1215 * mapped. 0 in case of a HOLE.
1223 if (!err
&& buffer_mapped(&dummy
)) {
1224 struct buffer_head
*bh
;
1225 bh
= sb_getblk(inode
->i_sb
, dummy
.b_blocknr
);
1230 if (buffer_new(&dummy
)) {
1231 J_ASSERT(create
!= 0);
1232 J_ASSERT(handle
!= NULL
);
1235 * Now that we do not always journal data, we should
1236 * keep in mind whether this should always journal the
1237 * new buffer as metadata. For now, regular file
1238 * writes use ext4_get_block instead, so it's not a
1242 BUFFER_TRACE(bh
, "call get_create_access");
1243 fatal
= ext4_journal_get_create_access(handle
, bh
);
1244 if (!fatal
&& !buffer_uptodate(bh
)) {
1245 memset(bh
->b_data
, 0, inode
->i_sb
->s_blocksize
);
1246 set_buffer_uptodate(bh
);
1249 BUFFER_TRACE(bh
, "call ext4_journal_dirty_metadata");
1250 err
= ext4_journal_dirty_metadata(handle
, bh
);
1254 BUFFER_TRACE(bh
, "not a new buffer");
1267 struct buffer_head
*ext4_bread(handle_t
*handle
, struct inode
*inode
,
1268 ext4_lblk_t block
, int create
, int *err
)
1270 struct buffer_head
*bh
;
1272 bh
= ext4_getblk(handle
, inode
, block
, create
, err
);
1275 if (buffer_uptodate(bh
))
1277 ll_rw_block(READ_META
, 1, &bh
);
1279 if (buffer_uptodate(bh
))
1286 static int walk_page_buffers(handle_t
*handle
,
1287 struct buffer_head
*head
,
1291 int (*fn
)(handle_t
*handle
,
1292 struct buffer_head
*bh
))
1294 struct buffer_head
*bh
;
1295 unsigned block_start
, block_end
;
1296 unsigned blocksize
= head
->b_size
;
1298 struct buffer_head
*next
;
1300 for (bh
= head
, block_start
= 0;
1301 ret
== 0 && (bh
!= head
|| !block_start
);
1302 block_start
= block_end
, bh
= next
)
1304 next
= bh
->b_this_page
;
1305 block_end
= block_start
+ blocksize
;
1306 if (block_end
<= from
|| block_start
>= to
) {
1307 if (partial
&& !buffer_uptodate(bh
))
1311 err
= (*fn
)(handle
, bh
);
1319 * To preserve ordering, it is essential that the hole instantiation and
1320 * the data write be encapsulated in a single transaction. We cannot
1321 * close off a transaction and start a new one between the ext4_get_block()
1322 * and the commit_write(). So doing the jbd2_journal_start at the start of
1323 * prepare_write() is the right place.
1325 * Also, this function can nest inside ext4_writepage() ->
1326 * block_write_full_page(). In that case, we *know* that ext4_writepage()
1327 * has generated enough buffer credits to do the whole page. So we won't
1328 * block on the journal in that case, which is good, because the caller may
1331 * By accident, ext4 can be reentered when a transaction is open via
1332 * quota file writes. If we were to commit the transaction while thus
1333 * reentered, there can be a deadlock - we would be holding a quota
1334 * lock, and the commit would never complete if another thread had a
1335 * transaction open and was blocking on the quota lock - a ranking
1338 * So what we do is to rely on the fact that jbd2_journal_stop/journal_start
1339 * will _not_ run commit under these circumstances because handle->h_ref
1340 * is elevated. We'll still have enough credits for the tiny quotafile
1343 static int do_journal_get_write_access(handle_t
*handle
,
1344 struct buffer_head
*bh
)
1346 if (!buffer_mapped(bh
) || buffer_freed(bh
))
1348 return ext4_journal_get_write_access(handle
, bh
);
1351 static int ext4_write_begin(struct file
*file
, struct address_space
*mapping
,
1352 loff_t pos
, unsigned len
, unsigned flags
,
1353 struct page
**pagep
, void **fsdata
)
1355 struct inode
*inode
= mapping
->host
;
1356 int ret
, needed_blocks
= ext4_writepage_trans_blocks(inode
);
1363 index
= pos
>> PAGE_CACHE_SHIFT
;
1364 from
= pos
& (PAGE_CACHE_SIZE
- 1);
1368 handle
= ext4_journal_start(inode
, needed_blocks
);
1369 if (IS_ERR(handle
)) {
1370 ret
= PTR_ERR(handle
);
1374 page
= __grab_cache_page(mapping
, index
);
1376 ext4_journal_stop(handle
);
1382 ret
= block_write_begin(file
, mapping
, pos
, len
, flags
, pagep
, fsdata
,
1385 if (!ret
&& ext4_should_journal_data(inode
)) {
1386 ret
= walk_page_buffers(handle
, page_buffers(page
),
1387 from
, to
, NULL
, do_journal_get_write_access
);
1392 ext4_journal_stop(handle
);
1393 page_cache_release(page
);
1395 * block_write_begin may have instantiated a few blocks
1396 * outside i_size. Trim these off again. Don't need
1397 * i_size_read because we hold i_mutex.
1399 if (pos
+ len
> inode
->i_size
)
1400 vmtruncate(inode
, inode
->i_size
);
1403 if (ret
== -ENOSPC
&& ext4_should_retry_alloc(inode
->i_sb
, &retries
))
1409 /* For write_end() in data=journal mode */
1410 static int write_end_fn(handle_t
*handle
, struct buffer_head
*bh
)
1412 if (!buffer_mapped(bh
) || buffer_freed(bh
))
1414 set_buffer_uptodate(bh
);
1415 return ext4_journal_dirty_metadata(handle
, bh
);
1419 * We need to pick up the new inode size which generic_commit_write gave us
1420 * `file' can be NULL - eg, when called from page_symlink().
1422 * ext4 never places buffers on inode->i_mapping->private_list. metadata
1423 * buffers are managed internally.
1425 static int ext4_ordered_write_end(struct file
*file
,
1426 struct address_space
*mapping
,
1427 loff_t pos
, unsigned len
, unsigned copied
,
1428 struct page
*page
, void *fsdata
)
1430 handle_t
*handle
= ext4_journal_current_handle();
1431 struct inode
*inode
= mapping
->host
;
1434 ret
= ext4_jbd2_file_inode(handle
, inode
);
1439 new_i_size
= pos
+ copied
;
1440 if (new_i_size
> EXT4_I(inode
)->i_disksize
) {
1441 ext4_update_i_disksize(inode
, new_i_size
);
1442 /* We need to mark inode dirty even if
1443 * new_i_size is less that inode->i_size
1444 * bu greater than i_disksize.(hint delalloc)
1446 ext4_mark_inode_dirty(handle
, inode
);
1449 ret2
= generic_write_end(file
, mapping
, pos
, len
, copied
,
1455 ret2
= ext4_journal_stop(handle
);
1459 return ret
? ret
: copied
;
1462 static int ext4_writeback_write_end(struct file
*file
,
1463 struct address_space
*mapping
,
1464 loff_t pos
, unsigned len
, unsigned copied
,
1465 struct page
*page
, void *fsdata
)
1467 handle_t
*handle
= ext4_journal_current_handle();
1468 struct inode
*inode
= mapping
->host
;
1472 new_i_size
= pos
+ copied
;
1473 if (new_i_size
> EXT4_I(inode
)->i_disksize
) {
1474 ext4_update_i_disksize(inode
, new_i_size
);
1475 /* We need to mark inode dirty even if
1476 * new_i_size is less that inode->i_size
1477 * bu greater than i_disksize.(hint delalloc)
1479 ext4_mark_inode_dirty(handle
, inode
);
1482 ret2
= generic_write_end(file
, mapping
, pos
, len
, copied
,
1488 ret2
= ext4_journal_stop(handle
);
1492 return ret
? ret
: copied
;
1495 static int ext4_journalled_write_end(struct file
*file
,
1496 struct address_space
*mapping
,
1497 loff_t pos
, unsigned len
, unsigned copied
,
1498 struct page
*page
, void *fsdata
)
1500 handle_t
*handle
= ext4_journal_current_handle();
1501 struct inode
*inode
= mapping
->host
;
1507 from
= pos
& (PAGE_CACHE_SIZE
- 1);
1511 if (!PageUptodate(page
))
1513 page_zero_new_buffers(page
, from
+copied
, to
);
1516 ret
= walk_page_buffers(handle
, page_buffers(page
), from
,
1517 to
, &partial
, write_end_fn
);
1519 SetPageUptodate(page
);
1520 new_i_size
= pos
+ copied
;
1521 if (new_i_size
> inode
->i_size
)
1522 i_size_write(inode
, pos
+copied
);
1523 EXT4_I(inode
)->i_state
|= EXT4_STATE_JDATA
;
1524 if (new_i_size
> EXT4_I(inode
)->i_disksize
) {
1525 ext4_update_i_disksize(inode
, new_i_size
);
1526 ret2
= ext4_mark_inode_dirty(handle
, inode
);
1532 ret2
= ext4_journal_stop(handle
);
1535 page_cache_release(page
);
1537 return ret
? ret
: copied
;
1540 static int ext4_da_reserve_space(struct inode
*inode
, int nrblocks
)
1543 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
1544 unsigned long md_needed
, mdblocks
, total
= 0;
1547 * recalculate the amount of metadata blocks to reserve
1548 * in order to allocate nrblocks
1549 * worse case is one extent per block
1552 spin_lock(&EXT4_I(inode
)->i_block_reservation_lock
);
1553 total
= EXT4_I(inode
)->i_reserved_data_blocks
+ nrblocks
;
1554 mdblocks
= ext4_calc_metadata_amount(inode
, total
);
1555 BUG_ON(mdblocks
< EXT4_I(inode
)->i_reserved_meta_blocks
);
1557 md_needed
= mdblocks
- EXT4_I(inode
)->i_reserved_meta_blocks
;
1558 total
= md_needed
+ nrblocks
;
1560 if (ext4_claim_free_blocks(sbi
, total
)) {
1561 spin_unlock(&EXT4_I(inode
)->i_block_reservation_lock
);
1562 if (ext4_should_retry_alloc(inode
->i_sb
, &retries
)) {
1568 EXT4_I(inode
)->i_reserved_data_blocks
+= nrblocks
;
1569 EXT4_I(inode
)->i_reserved_meta_blocks
= mdblocks
;
1571 spin_unlock(&EXT4_I(inode
)->i_block_reservation_lock
);
1572 return 0; /* success */
1575 static void ext4_da_release_space(struct inode
*inode
, int to_free
)
1577 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
1578 int total
, mdb
, mdb_free
, release
;
1581 return; /* Nothing to release, exit */
1583 spin_lock(&EXT4_I(inode
)->i_block_reservation_lock
);
1585 if (!EXT4_I(inode
)->i_reserved_data_blocks
) {
1587 * if there is no reserved blocks, but we try to free some
1588 * then the counter is messed up somewhere.
1589 * but since this function is called from invalidate
1590 * page, it's harmless to return without any action
1592 printk(KERN_INFO
"ext4 delalloc try to release %d reserved "
1593 "blocks for inode %lu, but there is no reserved "
1594 "data blocks\n", to_free
, inode
->i_ino
);
1595 spin_unlock(&EXT4_I(inode
)->i_block_reservation_lock
);
1599 /* recalculate the number of metablocks still need to be reserved */
1600 total
= EXT4_I(inode
)->i_reserved_data_blocks
- to_free
;
1601 mdb
= ext4_calc_metadata_amount(inode
, total
);
1603 /* figure out how many metablocks to release */
1604 BUG_ON(mdb
> EXT4_I(inode
)->i_reserved_meta_blocks
);
1605 mdb_free
= EXT4_I(inode
)->i_reserved_meta_blocks
- mdb
;
1607 release
= to_free
+ mdb_free
;
1609 /* update fs dirty blocks counter for truncate case */
1610 percpu_counter_sub(&sbi
->s_dirtyblocks_counter
, release
);
1612 /* update per-inode reservations */
1613 BUG_ON(to_free
> EXT4_I(inode
)->i_reserved_data_blocks
);
1614 EXT4_I(inode
)->i_reserved_data_blocks
-= to_free
;
1616 BUG_ON(mdb
> EXT4_I(inode
)->i_reserved_meta_blocks
);
1617 EXT4_I(inode
)->i_reserved_meta_blocks
= mdb
;
1618 spin_unlock(&EXT4_I(inode
)->i_block_reservation_lock
);
1621 static void ext4_da_page_release_reservation(struct page
*page
,
1622 unsigned long offset
)
1625 struct buffer_head
*head
, *bh
;
1626 unsigned int curr_off
= 0;
1628 head
= page_buffers(page
);
1631 unsigned int next_off
= curr_off
+ bh
->b_size
;
1633 if ((offset
<= curr_off
) && (buffer_delay(bh
))) {
1635 clear_buffer_delay(bh
);
1637 curr_off
= next_off
;
1638 } while ((bh
= bh
->b_this_page
) != head
);
1639 ext4_da_release_space(page
->mapping
->host
, to_release
);
1643 * Delayed allocation stuff
1646 struct mpage_da_data
{
1647 struct inode
*inode
;
1648 struct buffer_head lbh
; /* extent of blocks */
1649 unsigned long first_page
, next_page
; /* extent of pages */
1650 get_block_t
*get_block
;
1651 struct writeback_control
*wbc
;
1658 * mpage_da_submit_io - walks through extent of pages and try to write
1659 * them with writepage() call back
1661 * @mpd->inode: inode
1662 * @mpd->first_page: first page of the extent
1663 * @mpd->next_page: page after the last page of the extent
1664 * @mpd->get_block: the filesystem's block mapper function
1666 * By the time mpage_da_submit_io() is called we expect all blocks
1667 * to be allocated. this may be wrong if allocation failed.
1669 * As pages are already locked by write_cache_pages(), we can't use it
1671 static int mpage_da_submit_io(struct mpage_da_data
*mpd
)
1673 struct address_space
*mapping
= mpd
->inode
->i_mapping
;
1674 int ret
= 0, err
, nr_pages
, i
;
1675 unsigned long index
, end
;
1676 struct pagevec pvec
;
1678 BUG_ON(mpd
->next_page
<= mpd
->first_page
);
1679 pagevec_init(&pvec
, 0);
1680 index
= mpd
->first_page
;
1681 end
= mpd
->next_page
- 1;
1683 while (index
<= end
) {
1684 /* XXX: optimize tail */
1685 nr_pages
= pagevec_lookup(&pvec
, mapping
, index
, PAGEVEC_SIZE
);
1688 for (i
= 0; i
< nr_pages
; i
++) {
1689 struct page
*page
= pvec
.pages
[i
];
1691 index
= page
->index
;
1696 err
= mapping
->a_ops
->writepage(page
, mpd
->wbc
);
1698 mpd
->pages_written
++;
1700 * In error case, we have to continue because
1701 * remaining pages are still locked
1702 * XXX: unlock and re-dirty them?
1707 pagevec_release(&pvec
);
1713 * mpage_put_bnr_to_bhs - walk blocks and assign them actual numbers
1715 * @mpd->inode - inode to walk through
1716 * @exbh->b_blocknr - first block on a disk
1717 * @exbh->b_size - amount of space in bytes
1718 * @logical - first logical block to start assignment with
1720 * the function goes through all passed space and put actual disk
1721 * block numbers into buffer heads, dropping BH_Delay
1723 static void mpage_put_bnr_to_bhs(struct mpage_da_data
*mpd
, sector_t logical
,
1724 struct buffer_head
*exbh
)
1726 struct inode
*inode
= mpd
->inode
;
1727 struct address_space
*mapping
= inode
->i_mapping
;
1728 int blocks
= exbh
->b_size
>> inode
->i_blkbits
;
1729 sector_t pblock
= exbh
->b_blocknr
, cur_logical
;
1730 struct buffer_head
*head
, *bh
;
1732 struct pagevec pvec
;
1735 index
= logical
>> (PAGE_CACHE_SHIFT
- inode
->i_blkbits
);
1736 end
= (logical
+ blocks
- 1) >> (PAGE_CACHE_SHIFT
- inode
->i_blkbits
);
1737 cur_logical
= index
<< (PAGE_CACHE_SHIFT
- inode
->i_blkbits
);
1739 pagevec_init(&pvec
, 0);
1741 while (index
<= end
) {
1742 /* XXX: optimize tail */
1743 nr_pages
= pagevec_lookup(&pvec
, mapping
, index
, PAGEVEC_SIZE
);
1746 for (i
= 0; i
< nr_pages
; i
++) {
1747 struct page
*page
= pvec
.pages
[i
];
1749 index
= page
->index
;
1754 BUG_ON(!PageLocked(page
));
1755 BUG_ON(PageWriteback(page
));
1756 BUG_ON(!page_has_buffers(page
));
1758 bh
= page_buffers(page
);
1761 /* skip blocks out of the range */
1763 if (cur_logical
>= logical
)
1766 } while ((bh
= bh
->b_this_page
) != head
);
1769 if (cur_logical
>= logical
+ blocks
)
1771 if (buffer_delay(bh
)) {
1772 bh
->b_blocknr
= pblock
;
1773 clear_buffer_delay(bh
);
1774 bh
->b_bdev
= inode
->i_sb
->s_bdev
;
1775 } else if (buffer_unwritten(bh
)) {
1776 bh
->b_blocknr
= pblock
;
1777 clear_buffer_unwritten(bh
);
1778 set_buffer_mapped(bh
);
1780 bh
->b_bdev
= inode
->i_sb
->s_bdev
;
1781 } else if (buffer_mapped(bh
))
1782 BUG_ON(bh
->b_blocknr
!= pblock
);
1786 } while ((bh
= bh
->b_this_page
) != head
);
1788 pagevec_release(&pvec
);
1794 * __unmap_underlying_blocks - just a helper function to unmap
1795 * set of blocks described by @bh
1797 static inline void __unmap_underlying_blocks(struct inode
*inode
,
1798 struct buffer_head
*bh
)
1800 struct block_device
*bdev
= inode
->i_sb
->s_bdev
;
1803 blocks
= bh
->b_size
>> inode
->i_blkbits
;
1804 for (i
= 0; i
< blocks
; i
++)
1805 unmap_underlying_metadata(bdev
, bh
->b_blocknr
+ i
);
1808 static void ext4_da_block_invalidatepages(struct mpage_da_data
*mpd
,
1809 sector_t logical
, long blk_cnt
)
1813 struct pagevec pvec
;
1814 struct inode
*inode
= mpd
->inode
;
1815 struct address_space
*mapping
= inode
->i_mapping
;
1817 index
= logical
>> (PAGE_CACHE_SHIFT
- inode
->i_blkbits
);
1818 end
= (logical
+ blk_cnt
- 1) >>
1819 (PAGE_CACHE_SHIFT
- inode
->i_blkbits
);
1820 while (index
<= end
) {
1821 nr_pages
= pagevec_lookup(&pvec
, mapping
, index
, PAGEVEC_SIZE
);
1824 for (i
= 0; i
< nr_pages
; i
++) {
1825 struct page
*page
= pvec
.pages
[i
];
1826 index
= page
->index
;
1831 BUG_ON(!PageLocked(page
));
1832 BUG_ON(PageWriteback(page
));
1833 block_invalidatepage(page
, 0);
1834 ClearPageUptodate(page
);
1841 static void ext4_print_free_blocks(struct inode
*inode
)
1843 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
1844 printk(KERN_EMERG
"Total free blocks count %lld\n",
1845 ext4_count_free_blocks(inode
->i_sb
));
1846 printk(KERN_EMERG
"Free/Dirty block details\n");
1847 printk(KERN_EMERG
"free_blocks=%lld\n",
1848 percpu_counter_sum(&sbi
->s_freeblocks_counter
));
1849 printk(KERN_EMERG
"dirty_blocks=%lld\n",
1850 percpu_counter_sum(&sbi
->s_dirtyblocks_counter
));
1851 printk(KERN_EMERG
"Block reservation details\n");
1852 printk(KERN_EMERG
"i_reserved_data_blocks=%lu\n",
1853 EXT4_I(inode
)->i_reserved_data_blocks
);
1854 printk(KERN_EMERG
"i_reserved_meta_blocks=%lu\n",
1855 EXT4_I(inode
)->i_reserved_meta_blocks
);
1860 * mpage_da_map_blocks - go through given space
1862 * @mpd->lbh - bh describing space
1863 * @mpd->get_block - the filesystem's block mapper function
1865 * The function skips space we know is already mapped to disk blocks.
1868 static int mpage_da_map_blocks(struct mpage_da_data
*mpd
)
1871 struct buffer_head
new;
1872 struct buffer_head
*lbh
= &mpd
->lbh
;
1876 * We consider only non-mapped and non-allocated blocks
1878 if (buffer_mapped(lbh
) && !buffer_delay(lbh
))
1880 new.b_state
= lbh
->b_state
;
1882 new.b_size
= lbh
->b_size
;
1883 next
= lbh
->b_blocknr
;
1885 * If we didn't accumulate anything
1886 * to write simply return
1890 err
= mpd
->get_block(mpd
->inode
, next
, &new, 1);
1893 /* If get block returns with error
1894 * we simply return. Later writepage
1895 * will redirty the page and writepages
1896 * will find the dirty page again
1901 if (err
== -ENOSPC
&&
1902 ext4_count_free_blocks(mpd
->inode
->i_sb
)) {
1908 * get block failure will cause us
1909 * to loop in writepages. Because
1910 * a_ops->writepage won't be able to
1911 * make progress. The page will be redirtied
1912 * by writepage and writepages will again
1913 * try to write the same.
1915 printk(KERN_EMERG
"%s block allocation failed for inode %lu "
1916 "at logical offset %llu with max blocks "
1917 "%zd with error %d\n",
1918 __func__
, mpd
->inode
->i_ino
,
1919 (unsigned long long)next
,
1920 lbh
->b_size
>> mpd
->inode
->i_blkbits
, err
);
1921 printk(KERN_EMERG
"This should not happen.!! "
1922 "Data will be lost\n");
1923 if (err
== -ENOSPC
) {
1924 ext4_print_free_blocks(mpd
->inode
);
1926 /* invlaidate all the pages */
1927 ext4_da_block_invalidatepages(mpd
, next
,
1928 lbh
->b_size
>> mpd
->inode
->i_blkbits
);
1931 BUG_ON(new.b_size
== 0);
1933 if (buffer_new(&new))
1934 __unmap_underlying_blocks(mpd
->inode
, &new);
1937 * If blocks are delayed marked, we need to
1938 * put actual blocknr and drop delayed bit
1940 if (buffer_delay(lbh
) || buffer_unwritten(lbh
))
1941 mpage_put_bnr_to_bhs(mpd
, next
, &new);
1946 #define BH_FLAGS ((1 << BH_Uptodate) | (1 << BH_Mapped) | \
1947 (1 << BH_Delay) | (1 << BH_Unwritten))
1950 * mpage_add_bh_to_extent - try to add one more block to extent of blocks
1952 * @mpd->lbh - extent of blocks
1953 * @logical - logical number of the block in the file
1954 * @bh - bh of the block (used to access block's state)
1956 * the function is used to collect contig. blocks in same state
1958 static void mpage_add_bh_to_extent(struct mpage_da_data
*mpd
,
1959 sector_t logical
, struct buffer_head
*bh
)
1962 size_t b_size
= bh
->b_size
;
1963 struct buffer_head
*lbh
= &mpd
->lbh
;
1964 int nrblocks
= lbh
->b_size
>> mpd
->inode
->i_blkbits
;
1966 /* check if thereserved journal credits might overflow */
1967 if (!(EXT4_I(mpd
->inode
)->i_flags
& EXT4_EXTENTS_FL
)) {
1968 if (nrblocks
>= EXT4_MAX_TRANS_DATA
) {
1970 * With non-extent format we are limited by the journal
1971 * credit available. Total credit needed to insert
1972 * nrblocks contiguous blocks is dependent on the
1973 * nrblocks. So limit nrblocks.
1976 } else if ((nrblocks
+ (b_size
>> mpd
->inode
->i_blkbits
)) >
1977 EXT4_MAX_TRANS_DATA
) {
1979 * Adding the new buffer_head would make it cross the
1980 * allowed limit for which we have journal credit
1981 * reserved. So limit the new bh->b_size
1983 b_size
= (EXT4_MAX_TRANS_DATA
- nrblocks
) <<
1984 mpd
->inode
->i_blkbits
;
1985 /* we will do mpage_da_submit_io in the next loop */
1989 * First block in the extent
1991 if (lbh
->b_size
== 0) {
1992 lbh
->b_blocknr
= logical
;
1993 lbh
->b_size
= b_size
;
1994 lbh
->b_state
= bh
->b_state
& BH_FLAGS
;
1998 next
= lbh
->b_blocknr
+ nrblocks
;
2000 * Can we merge the block to our big extent?
2002 if (logical
== next
&& (bh
->b_state
& BH_FLAGS
) == lbh
->b_state
) {
2003 lbh
->b_size
+= b_size
;
2009 * We couldn't merge the block to our extent, so we
2010 * need to flush current extent and start new one
2012 if (mpage_da_map_blocks(mpd
) == 0)
2013 mpage_da_submit_io(mpd
);
2019 * __mpage_da_writepage - finds extent of pages and blocks
2021 * @page: page to consider
2022 * @wbc: not used, we just follow rules
2025 * The function finds extents of pages and scan them for all blocks.
2027 static int __mpage_da_writepage(struct page
*page
,
2028 struct writeback_control
*wbc
, void *data
)
2030 struct mpage_da_data
*mpd
= data
;
2031 struct inode
*inode
= mpd
->inode
;
2032 struct buffer_head
*bh
, *head
, fake
;
2037 * Rest of the page in the page_vec
2038 * redirty then and skip then. We will
2039 * try to to write them again after
2040 * starting a new transaction
2042 redirty_page_for_writepage(wbc
, page
);
2044 return MPAGE_DA_EXTENT_TAIL
;
2047 * Can we merge this page to current extent?
2049 if (mpd
->next_page
!= page
->index
) {
2051 * Nope, we can't. So, we map non-allocated blocks
2052 * and start IO on them using writepage()
2054 if (mpd
->next_page
!= mpd
->first_page
) {
2055 if (mpage_da_map_blocks(mpd
) == 0)
2056 mpage_da_submit_io(mpd
);
2058 * skip rest of the page in the page_vec
2061 redirty_page_for_writepage(wbc
, page
);
2063 return MPAGE_DA_EXTENT_TAIL
;
2067 * Start next extent of pages ...
2069 mpd
->first_page
= page
->index
;
2074 mpd
->lbh
.b_size
= 0;
2075 mpd
->lbh
.b_state
= 0;
2076 mpd
->lbh
.b_blocknr
= 0;
2079 mpd
->next_page
= page
->index
+ 1;
2080 logical
= (sector_t
) page
->index
<<
2081 (PAGE_CACHE_SHIFT
- inode
->i_blkbits
);
2083 if (!page_has_buffers(page
)) {
2085 * There is no attached buffer heads yet (mmap?)
2086 * we treat the page asfull of dirty blocks
2089 bh
->b_size
= PAGE_CACHE_SIZE
;
2091 set_buffer_dirty(bh
);
2092 set_buffer_uptodate(bh
);
2093 mpage_add_bh_to_extent(mpd
, logical
, bh
);
2095 return MPAGE_DA_EXTENT_TAIL
;
2098 * Page with regular buffer heads, just add all dirty ones
2100 head
= page_buffers(page
);
2103 BUG_ON(buffer_locked(bh
));
2104 if (buffer_dirty(bh
) &&
2105 (!buffer_mapped(bh
) || buffer_delay(bh
))) {
2106 mpage_add_bh_to_extent(mpd
, logical
, bh
);
2108 return MPAGE_DA_EXTENT_TAIL
;
2111 } while ((bh
= bh
->b_this_page
) != head
);
2118 * mpage_da_writepages - walk the list of dirty pages of the given
2119 * address space, allocates non-allocated blocks, maps newly-allocated
2120 * blocks to existing bhs and issue IO them
2122 * @mapping: address space structure to write
2123 * @wbc: subtract the number of written pages from *@wbc->nr_to_write
2124 * @get_block: the filesystem's block mapper function.
2126 * This is a library function, which implements the writepages()
2127 * address_space_operation.
2129 static int mpage_da_writepages(struct address_space
*mapping
,
2130 struct writeback_control
*wbc
,
2131 struct mpage_da_data
*mpd
)
2136 if (!mpd
->get_block
)
2137 return generic_writepages(mapping
, wbc
);
2139 mpd
->lbh
.b_size
= 0;
2140 mpd
->lbh
.b_state
= 0;
2141 mpd
->lbh
.b_blocknr
= 0;
2142 mpd
->first_page
= 0;
2145 mpd
->pages_written
= 0;
2148 to_write
= wbc
->nr_to_write
;
2150 ret
= write_cache_pages(mapping
, wbc
, __mpage_da_writepage
, mpd
);
2153 * Handle last extent of pages
2155 if (!mpd
->io_done
&& mpd
->next_page
!= mpd
->first_page
) {
2156 if (mpage_da_map_blocks(mpd
) == 0)
2157 mpage_da_submit_io(mpd
);
2160 wbc
->nr_to_write
= to_write
- mpd
->pages_written
;
2165 * this is a special callback for ->write_begin() only
2166 * it's intention is to return mapped block or reserve space
2168 static int ext4_da_get_block_prep(struct inode
*inode
, sector_t iblock
,
2169 struct buffer_head
*bh_result
, int create
)
2173 BUG_ON(create
== 0);
2174 BUG_ON(bh_result
->b_size
!= inode
->i_sb
->s_blocksize
);
2177 * first, we need to know whether the block is allocated already
2178 * preallocated blocks are unmapped but should treated
2179 * the same as allocated blocks.
2181 ret
= ext4_get_blocks_wrap(NULL
, inode
, iblock
, 1, bh_result
, 0, 0, 0);
2182 if ((ret
== 0) && !buffer_delay(bh_result
)) {
2183 /* the block isn't (pre)allocated yet, let's reserve space */
2185 * XXX: __block_prepare_write() unmaps passed block,
2188 ret
= ext4_da_reserve_space(inode
, 1);
2190 /* not enough space to reserve */
2193 map_bh(bh_result
, inode
->i_sb
, 0);
2194 set_buffer_new(bh_result
);
2195 set_buffer_delay(bh_result
);
2196 } else if (ret
> 0) {
2197 bh_result
->b_size
= (ret
<< inode
->i_blkbits
);
2203 #define EXT4_DELALLOC_RSVED 1
2204 static int ext4_da_get_block_write(struct inode
*inode
, sector_t iblock
,
2205 struct buffer_head
*bh_result
, int create
)
2208 unsigned max_blocks
= bh_result
->b_size
>> inode
->i_blkbits
;
2209 loff_t disksize
= EXT4_I(inode
)->i_disksize
;
2210 handle_t
*handle
= NULL
;
2212 handle
= ext4_journal_current_handle();
2214 ret
= ext4_get_blocks_wrap(handle
, inode
, iblock
, max_blocks
,
2215 bh_result
, create
, 0, EXT4_DELALLOC_RSVED
);
2218 bh_result
->b_size
= (ret
<< inode
->i_blkbits
);
2220 if (ext4_should_order_data(inode
)) {
2222 retval
= ext4_jbd2_file_inode(handle
, inode
);
2225 * Failed to add inode for ordered
2226 * mode. Don't update file size
2232 * Update on-disk size along with block allocation
2233 * we don't use 'extend_disksize' as size may change
2234 * within already allocated block -bzzz
2236 disksize
= ((loff_t
) iblock
+ ret
) << inode
->i_blkbits
;
2237 if (disksize
> i_size_read(inode
))
2238 disksize
= i_size_read(inode
);
2239 if (disksize
> EXT4_I(inode
)->i_disksize
) {
2240 ext4_update_i_disksize(inode
, disksize
);
2241 ret
= ext4_mark_inode_dirty(handle
, inode
);
2249 static int ext4_bh_unmapped_or_delay(handle_t
*handle
, struct buffer_head
*bh
)
2252 * unmapped buffer is possible for holes.
2253 * delay buffer is possible with delayed allocation
2255 return ((!buffer_mapped(bh
) || buffer_delay(bh
)) && buffer_dirty(bh
));
2258 static int ext4_normal_get_block_write(struct inode
*inode
, sector_t iblock
,
2259 struct buffer_head
*bh_result
, int create
)
2262 unsigned max_blocks
= bh_result
->b_size
>> inode
->i_blkbits
;
2265 * we don't want to do block allocation in writepage
2266 * so call get_block_wrap with create = 0
2268 ret
= ext4_get_blocks_wrap(NULL
, inode
, iblock
, max_blocks
,
2269 bh_result
, 0, 0, 0);
2271 bh_result
->b_size
= (ret
<< inode
->i_blkbits
);
2278 * get called vi ext4_da_writepages after taking page lock (have journal handle)
2279 * get called via journal_submit_inode_data_buffers (no journal handle)
2280 * get called via shrink_page_list via pdflush (no journal handle)
2281 * or grab_page_cache when doing write_begin (have journal handle)
2283 static int ext4_da_writepage(struct page
*page
,
2284 struct writeback_control
*wbc
)
2289 struct buffer_head
*page_bufs
;
2290 struct inode
*inode
= page
->mapping
->host
;
2292 size
= i_size_read(inode
);
2293 if (page
->index
== size
>> PAGE_CACHE_SHIFT
)
2294 len
= size
& ~PAGE_CACHE_MASK
;
2296 len
= PAGE_CACHE_SIZE
;
2298 if (page_has_buffers(page
)) {
2299 page_bufs
= page_buffers(page
);
2300 if (walk_page_buffers(NULL
, page_bufs
, 0, len
, NULL
,
2301 ext4_bh_unmapped_or_delay
)) {
2303 * We don't want to do block allocation
2304 * So redirty the page and return
2305 * We may reach here when we do a journal commit
2306 * via journal_submit_inode_data_buffers.
2307 * If we don't have mapping block we just ignore
2308 * them. We can also reach here via shrink_page_list
2310 redirty_page_for_writepage(wbc
, page
);
2316 * The test for page_has_buffers() is subtle:
2317 * We know the page is dirty but it lost buffers. That means
2318 * that at some moment in time after write_begin()/write_end()
2319 * has been called all buffers have been clean and thus they
2320 * must have been written at least once. So they are all
2321 * mapped and we can happily proceed with mapping them
2322 * and writing the page.
2324 * Try to initialize the buffer_heads and check whether
2325 * all are mapped and non delay. We don't want to
2326 * do block allocation here.
2328 ret
= block_prepare_write(page
, 0, PAGE_CACHE_SIZE
,
2329 ext4_normal_get_block_write
);
2331 page_bufs
= page_buffers(page
);
2332 /* check whether all are mapped and non delay */
2333 if (walk_page_buffers(NULL
, page_bufs
, 0, len
, NULL
,
2334 ext4_bh_unmapped_or_delay
)) {
2335 redirty_page_for_writepage(wbc
, page
);
2341 * We can't do block allocation here
2342 * so just redity the page and unlock
2345 redirty_page_for_writepage(wbc
, page
);
2351 if (test_opt(inode
->i_sb
, NOBH
) && ext4_should_writeback_data(inode
))
2352 ret
= nobh_writepage(page
, ext4_normal_get_block_write
, wbc
);
2354 ret
= block_write_full_page(page
,
2355 ext4_normal_get_block_write
,
2362 * This is called via ext4_da_writepages() to
2363 * calulate the total number of credits to reserve to fit
2364 * a single extent allocation into a single transaction,
2365 * ext4_da_writpeages() will loop calling this before
2366 * the block allocation.
2369 static int ext4_da_writepages_trans_blocks(struct inode
*inode
)
2371 int max_blocks
= EXT4_I(inode
)->i_reserved_data_blocks
;
2374 * With non-extent format the journal credit needed to
2375 * insert nrblocks contiguous block is dependent on
2376 * number of contiguous block. So we will limit
2377 * number of contiguous block to a sane value
2379 if (!(inode
->i_flags
& EXT4_EXTENTS_FL
) &&
2380 (max_blocks
> EXT4_MAX_TRANS_DATA
))
2381 max_blocks
= EXT4_MAX_TRANS_DATA
;
2383 return ext4_chunk_trans_blocks(inode
, max_blocks
);
2386 static int ext4_da_writepages(struct address_space
*mapping
,
2387 struct writeback_control
*wbc
)
2389 handle_t
*handle
= NULL
;
2390 loff_t range_start
= 0;
2391 struct mpage_da_data mpd
;
2392 struct inode
*inode
= mapping
->host
;
2393 int needed_blocks
, ret
= 0, nr_to_writebump
= 0;
2394 long to_write
, pages_skipped
= 0;
2395 struct ext4_sb_info
*sbi
= EXT4_SB(mapping
->host
->i_sb
);
2398 * No pages to write? This is mainly a kludge to avoid starting
2399 * a transaction for special inodes like journal inode on last iput()
2400 * because that could violate lock ordering on umount
2402 if (!mapping
->nrpages
|| !mapping_tagged(mapping
, PAGECACHE_TAG_DIRTY
))
2405 * Make sure nr_to_write is >= sbi->s_mb_stream_request
2406 * This make sure small files blocks are allocated in
2407 * single attempt. This ensure that small files
2408 * get less fragmented.
2410 if (wbc
->nr_to_write
< sbi
->s_mb_stream_request
) {
2411 nr_to_writebump
= sbi
->s_mb_stream_request
- wbc
->nr_to_write
;
2412 wbc
->nr_to_write
= sbi
->s_mb_stream_request
;
2415 if (!wbc
->range_cyclic
)
2417 * If range_cyclic is not set force range_cont
2418 * and save the old writeback_index
2420 wbc
->range_cont
= 1;
2422 range_start
= wbc
->range_start
;
2423 pages_skipped
= wbc
->pages_skipped
;
2426 mpd
.inode
= mapping
->host
;
2429 to_write
= wbc
->nr_to_write
;
2430 while (!ret
&& to_write
> 0) {
2433 * we insert one extent at a time. So we need
2434 * credit needed for single extent allocation.
2435 * journalled mode is currently not supported
2438 BUG_ON(ext4_should_journal_data(inode
));
2439 needed_blocks
= ext4_da_writepages_trans_blocks(inode
);
2441 /* start a new transaction*/
2442 handle
= ext4_journal_start(inode
, needed_blocks
);
2443 if (IS_ERR(handle
)) {
2444 ret
= PTR_ERR(handle
);
2445 printk(KERN_EMERG
"%s: jbd2_start: "
2446 "%ld pages, ino %lu; err %d\n", __func__
,
2447 wbc
->nr_to_write
, inode
->i_ino
, ret
);
2449 goto out_writepages
;
2451 to_write
-= wbc
->nr_to_write
;
2453 mpd
.get_block
= ext4_da_get_block_write
;
2454 ret
= mpage_da_writepages(mapping
, wbc
, &mpd
);
2456 ext4_journal_stop(handle
);
2458 if (mpd
.retval
== -ENOSPC
)
2459 jbd2_journal_force_commit_nested(sbi
->s_journal
);
2461 /* reset the retry count */
2462 if (ret
== MPAGE_DA_EXTENT_TAIL
) {
2464 * got one extent now try with
2467 to_write
+= wbc
->nr_to_write
;
2469 } else if (wbc
->nr_to_write
) {
2471 * There is no more writeout needed
2472 * or we requested for a noblocking writeout
2473 * and we found the device congested
2475 to_write
+= wbc
->nr_to_write
;
2478 wbc
->nr_to_write
= to_write
;
2481 if (wbc
->range_cont
&& (pages_skipped
!= wbc
->pages_skipped
)) {
2482 /* We skipped pages in this loop */
2483 wbc
->range_start
= range_start
;
2484 wbc
->nr_to_write
= to_write
+
2485 wbc
->pages_skipped
- pages_skipped
;
2486 wbc
->pages_skipped
= pages_skipped
;
2491 wbc
->nr_to_write
= to_write
- nr_to_writebump
;
2492 wbc
->range_start
= range_start
;
2496 #define FALL_BACK_TO_NONDELALLOC 1
2497 static int ext4_nonda_switch(struct super_block
*sb
)
2499 s64 free_blocks
, dirty_blocks
;
2500 struct ext4_sb_info
*sbi
= EXT4_SB(sb
);
2503 * switch to non delalloc mode if we are running low
2504 * on free block. The free block accounting via percpu
2505 * counters can get slightly wrong with FBC_BATCH getting
2506 * accumulated on each CPU without updating global counters
2507 * Delalloc need an accurate free block accounting. So switch
2508 * to non delalloc when we are near to error range.
2510 free_blocks
= percpu_counter_read_positive(&sbi
->s_freeblocks_counter
);
2511 dirty_blocks
= percpu_counter_read_positive(&sbi
->s_dirtyblocks_counter
);
2512 if (2 * free_blocks
< 3 * dirty_blocks
||
2513 free_blocks
< (dirty_blocks
+ EXT4_FREEBLOCKS_WATERMARK
)) {
2515 * free block count is less that 150% of dirty blocks
2516 * or free blocks is less that watermark
2523 static int ext4_da_write_begin(struct file
*file
, struct address_space
*mapping
,
2524 loff_t pos
, unsigned len
, unsigned flags
,
2525 struct page
**pagep
, void **fsdata
)
2527 int ret
, retries
= 0;
2531 struct inode
*inode
= mapping
->host
;
2534 index
= pos
>> PAGE_CACHE_SHIFT
;
2535 from
= pos
& (PAGE_CACHE_SIZE
- 1);
2538 if (ext4_nonda_switch(inode
->i_sb
)) {
2539 *fsdata
= (void *)FALL_BACK_TO_NONDELALLOC
;
2540 return ext4_write_begin(file
, mapping
, pos
,
2541 len
, flags
, pagep
, fsdata
);
2543 *fsdata
= (void *)0;
2546 * With delayed allocation, we don't log the i_disksize update
2547 * if there is delayed block allocation. But we still need
2548 * to journalling the i_disksize update if writes to the end
2549 * of file which has an already mapped buffer.
2551 handle
= ext4_journal_start(inode
, 1);
2552 if (IS_ERR(handle
)) {
2553 ret
= PTR_ERR(handle
);
2557 page
= __grab_cache_page(mapping
, index
);
2559 ext4_journal_stop(handle
);
2565 ret
= block_write_begin(file
, mapping
, pos
, len
, flags
, pagep
, fsdata
,
2566 ext4_da_get_block_prep
);
2569 ext4_journal_stop(handle
);
2570 page_cache_release(page
);
2572 * block_write_begin may have instantiated a few blocks
2573 * outside i_size. Trim these off again. Don't need
2574 * i_size_read because we hold i_mutex.
2576 if (pos
+ len
> inode
->i_size
)
2577 vmtruncate(inode
, inode
->i_size
);
2580 if (ret
== -ENOSPC
&& ext4_should_retry_alloc(inode
->i_sb
, &retries
))
2587 * Check if we should update i_disksize
2588 * when write to the end of file but not require block allocation
2590 static int ext4_da_should_update_i_disksize(struct page
*page
,
2591 unsigned long offset
)
2593 struct buffer_head
*bh
;
2594 struct inode
*inode
= page
->mapping
->host
;
2598 bh
= page_buffers(page
);
2599 idx
= offset
>> inode
->i_blkbits
;
2601 for (i
= 0; i
< idx
; i
++)
2602 bh
= bh
->b_this_page
;
2604 if (!buffer_mapped(bh
) || (buffer_delay(bh
)))
2609 static int ext4_da_write_end(struct file
*file
,
2610 struct address_space
*mapping
,
2611 loff_t pos
, unsigned len
, unsigned copied
,
2612 struct page
*page
, void *fsdata
)
2614 struct inode
*inode
= mapping
->host
;
2616 handle_t
*handle
= ext4_journal_current_handle();
2618 unsigned long start
, end
;
2619 int write_mode
= (int)(unsigned long)fsdata
;
2621 if (write_mode
== FALL_BACK_TO_NONDELALLOC
) {
2622 if (ext4_should_order_data(inode
)) {
2623 return ext4_ordered_write_end(file
, mapping
, pos
,
2624 len
, copied
, page
, fsdata
);
2625 } else if (ext4_should_writeback_data(inode
)) {
2626 return ext4_writeback_write_end(file
, mapping
, pos
,
2627 len
, copied
, page
, fsdata
);
2633 start
= pos
& (PAGE_CACHE_SIZE
- 1);
2634 end
= start
+ copied
- 1;
2637 * generic_write_end() will run mark_inode_dirty() if i_size
2638 * changes. So let's piggyback the i_disksize mark_inode_dirty
2642 new_i_size
= pos
+ copied
;
2643 if (new_i_size
> EXT4_I(inode
)->i_disksize
) {
2644 if (ext4_da_should_update_i_disksize(page
, end
)) {
2645 down_write(&EXT4_I(inode
)->i_data_sem
);
2646 if (new_i_size
> EXT4_I(inode
)->i_disksize
) {
2648 * Updating i_disksize when extending file
2649 * without needing block allocation
2651 if (ext4_should_order_data(inode
))
2652 ret
= ext4_jbd2_file_inode(handle
,
2655 EXT4_I(inode
)->i_disksize
= new_i_size
;
2657 up_write(&EXT4_I(inode
)->i_data_sem
);
2658 /* We need to mark inode dirty even if
2659 * new_i_size is less that inode->i_size
2660 * bu greater than i_disksize.(hint delalloc)
2662 ext4_mark_inode_dirty(handle
, inode
);
2665 ret2
= generic_write_end(file
, mapping
, pos
, len
, copied
,
2670 ret2
= ext4_journal_stop(handle
);
2674 return ret
? ret
: copied
;
2677 static void ext4_da_invalidatepage(struct page
*page
, unsigned long offset
)
2680 * Drop reserved blocks
2682 BUG_ON(!PageLocked(page
));
2683 if (!page_has_buffers(page
))
2686 ext4_da_page_release_reservation(page
, offset
);
2689 ext4_invalidatepage(page
, offset
);
2696 * bmap() is special. It gets used by applications such as lilo and by
2697 * the swapper to find the on-disk block of a specific piece of data.
2699 * Naturally, this is dangerous if the block concerned is still in the
2700 * journal. If somebody makes a swapfile on an ext4 data-journaling
2701 * filesystem and enables swap, then they may get a nasty shock when the
2702 * data getting swapped to that swapfile suddenly gets overwritten by
2703 * the original zero's written out previously to the journal and
2704 * awaiting writeback in the kernel's buffer cache.
2706 * So, if we see any bmap calls here on a modified, data-journaled file,
2707 * take extra steps to flush any blocks which might be in the cache.
2709 static sector_t
ext4_bmap(struct address_space
*mapping
, sector_t block
)
2711 struct inode
*inode
= mapping
->host
;
2715 if (mapping_tagged(mapping
, PAGECACHE_TAG_DIRTY
) &&
2716 test_opt(inode
->i_sb
, DELALLOC
)) {
2718 * With delalloc we want to sync the file
2719 * so that we can make sure we allocate
2722 filemap_write_and_wait(mapping
);
2725 if (EXT4_I(inode
)->i_state
& EXT4_STATE_JDATA
) {
2727 * This is a REALLY heavyweight approach, but the use of
2728 * bmap on dirty files is expected to be extremely rare:
2729 * only if we run lilo or swapon on a freshly made file
2730 * do we expect this to happen.
2732 * (bmap requires CAP_SYS_RAWIO so this does not
2733 * represent an unprivileged user DOS attack --- we'd be
2734 * in trouble if mortal users could trigger this path at
2737 * NB. EXT4_STATE_JDATA is not set on files other than
2738 * regular files. If somebody wants to bmap a directory
2739 * or symlink and gets confused because the buffer
2740 * hasn't yet been flushed to disk, they deserve
2741 * everything they get.
2744 EXT4_I(inode
)->i_state
&= ~EXT4_STATE_JDATA
;
2745 journal
= EXT4_JOURNAL(inode
);
2746 jbd2_journal_lock_updates(journal
);
2747 err
= jbd2_journal_flush(journal
);
2748 jbd2_journal_unlock_updates(journal
);
2754 return generic_block_bmap(mapping
, block
, ext4_get_block
);
2757 static int bget_one(handle_t
*handle
, struct buffer_head
*bh
)
2763 static int bput_one(handle_t
*handle
, struct buffer_head
*bh
)
2770 * Note that we don't need to start a transaction unless we're journaling data
2771 * because we should have holes filled from ext4_page_mkwrite(). We even don't
2772 * need to file the inode to the transaction's list in ordered mode because if
2773 * we are writing back data added by write(), the inode is already there and if
2774 * we are writing back data modified via mmap(), noone guarantees in which
2775 * transaction the data will hit the disk. In case we are journaling data, we
2776 * cannot start transaction directly because transaction start ranks above page
2777 * lock so we have to do some magic.
2779 * In all journaling modes block_write_full_page() will start the I/O.
2783 * ext4_writepage() -> kmalloc() -> __alloc_pages() -> page_launder() ->
2788 * ext4_file_write() -> generic_file_write() -> __alloc_pages() -> ...
2790 * Same applies to ext4_get_block(). We will deadlock on various things like
2791 * lock_journal and i_data_sem
2793 * Setting PF_MEMALLOC here doesn't work - too many internal memory
2796 * 16May01: If we're reentered then journal_current_handle() will be
2797 * non-zero. We simply *return*.
2799 * 1 July 2001: @@@ FIXME:
2800 * In journalled data mode, a data buffer may be metadata against the
2801 * current transaction. But the same file is part of a shared mapping
2802 * and someone does a writepage() on it.
2804 * We will move the buffer onto the async_data list, but *after* it has
2805 * been dirtied. So there's a small window where we have dirty data on
2808 * Note that this only applies to the last partial page in the file. The
2809 * bit which block_write_full_page() uses prepare/commit for. (That's
2810 * broken code anyway: it's wrong for msync()).
2812 * It's a rare case: affects the final partial page, for journalled data
2813 * where the file is subject to bith write() and writepage() in the same
2814 * transction. To fix it we'll need a custom block_write_full_page().
2815 * We'll probably need that anyway for journalling writepage() output.
2817 * We don't honour synchronous mounts for writepage(). That would be
2818 * disastrous. Any write() or metadata operation will sync the fs for
2822 static int __ext4_normal_writepage(struct page
*page
,
2823 struct writeback_control
*wbc
)
2825 struct inode
*inode
= page
->mapping
->host
;
2827 if (test_opt(inode
->i_sb
, NOBH
))
2828 return nobh_writepage(page
,
2829 ext4_normal_get_block_write
, wbc
);
2831 return block_write_full_page(page
,
2832 ext4_normal_get_block_write
,
2836 static int ext4_normal_writepage(struct page
*page
,
2837 struct writeback_control
*wbc
)
2839 struct inode
*inode
= page
->mapping
->host
;
2840 loff_t size
= i_size_read(inode
);
2843 J_ASSERT(PageLocked(page
));
2844 if (page
->index
== size
>> PAGE_CACHE_SHIFT
)
2845 len
= size
& ~PAGE_CACHE_MASK
;
2847 len
= PAGE_CACHE_SIZE
;
2849 if (page_has_buffers(page
)) {
2850 /* if page has buffers it should all be mapped
2851 * and allocated. If there are not buffers attached
2852 * to the page we know the page is dirty but it lost
2853 * buffers. That means that at some moment in time
2854 * after write_begin() / write_end() has been called
2855 * all buffers have been clean and thus they must have been
2856 * written at least once. So they are all mapped and we can
2857 * happily proceed with mapping them and writing the page.
2859 BUG_ON(walk_page_buffers(NULL
, page_buffers(page
), 0, len
, NULL
,
2860 ext4_bh_unmapped_or_delay
));
2863 if (!ext4_journal_current_handle())
2864 return __ext4_normal_writepage(page
, wbc
);
2866 redirty_page_for_writepage(wbc
, page
);
2871 static int __ext4_journalled_writepage(struct page
*page
,
2872 struct writeback_control
*wbc
)
2874 struct address_space
*mapping
= page
->mapping
;
2875 struct inode
*inode
= mapping
->host
;
2876 struct buffer_head
*page_bufs
;
2877 handle_t
*handle
= NULL
;
2881 ret
= block_prepare_write(page
, 0, PAGE_CACHE_SIZE
,
2882 ext4_normal_get_block_write
);
2886 page_bufs
= page_buffers(page
);
2887 walk_page_buffers(handle
, page_bufs
, 0, PAGE_CACHE_SIZE
, NULL
,
2889 /* As soon as we unlock the page, it can go away, but we have
2890 * references to buffers so we are safe */
2893 handle
= ext4_journal_start(inode
, ext4_writepage_trans_blocks(inode
));
2894 if (IS_ERR(handle
)) {
2895 ret
= PTR_ERR(handle
);
2899 ret
= walk_page_buffers(handle
, page_bufs
, 0,
2900 PAGE_CACHE_SIZE
, NULL
, do_journal_get_write_access
);
2902 err
= walk_page_buffers(handle
, page_bufs
, 0,
2903 PAGE_CACHE_SIZE
, NULL
, write_end_fn
);
2906 err
= ext4_journal_stop(handle
);
2910 walk_page_buffers(handle
, page_bufs
, 0,
2911 PAGE_CACHE_SIZE
, NULL
, bput_one
);
2912 EXT4_I(inode
)->i_state
|= EXT4_STATE_JDATA
;
2921 static int ext4_journalled_writepage(struct page
*page
,
2922 struct writeback_control
*wbc
)
2924 struct inode
*inode
= page
->mapping
->host
;
2925 loff_t size
= i_size_read(inode
);
2928 J_ASSERT(PageLocked(page
));
2929 if (page
->index
== size
>> PAGE_CACHE_SHIFT
)
2930 len
= size
& ~PAGE_CACHE_MASK
;
2932 len
= PAGE_CACHE_SIZE
;
2934 if (page_has_buffers(page
)) {
2935 /* if page has buffers it should all be mapped
2936 * and allocated. If there are not buffers attached
2937 * to the page we know the page is dirty but it lost
2938 * buffers. That means that at some moment in time
2939 * after write_begin() / write_end() has been called
2940 * all buffers have been clean and thus they must have been
2941 * written at least once. So they are all mapped and we can
2942 * happily proceed with mapping them and writing the page.
2944 BUG_ON(walk_page_buffers(NULL
, page_buffers(page
), 0, len
, NULL
,
2945 ext4_bh_unmapped_or_delay
));
2948 if (ext4_journal_current_handle())
2951 if (PageChecked(page
)) {
2953 * It's mmapped pagecache. Add buffers and journal it. There
2954 * doesn't seem much point in redirtying the page here.
2956 ClearPageChecked(page
);
2957 return __ext4_journalled_writepage(page
, wbc
);
2960 * It may be a page full of checkpoint-mode buffers. We don't
2961 * really know unless we go poke around in the buffer_heads.
2962 * But block_write_full_page will do the right thing.
2964 return block_write_full_page(page
,
2965 ext4_normal_get_block_write
,
2969 redirty_page_for_writepage(wbc
, page
);
2974 static int ext4_readpage(struct file
*file
, struct page
*page
)
2976 return mpage_readpage(page
, ext4_get_block
);
2980 ext4_readpages(struct file
*file
, struct address_space
*mapping
,
2981 struct list_head
*pages
, unsigned nr_pages
)
2983 return mpage_readpages(mapping
, pages
, nr_pages
, ext4_get_block
);
2986 static void ext4_invalidatepage(struct page
*page
, unsigned long offset
)
2988 journal_t
*journal
= EXT4_JOURNAL(page
->mapping
->host
);
2991 * If it's a full truncate we just forget about the pending dirtying
2994 ClearPageChecked(page
);
2996 jbd2_journal_invalidatepage(journal
, page
, offset
);
2999 static int ext4_releasepage(struct page
*page
, gfp_t wait
)
3001 journal_t
*journal
= EXT4_JOURNAL(page
->mapping
->host
);
3003 WARN_ON(PageChecked(page
));
3004 if (!page_has_buffers(page
))
3006 return jbd2_journal_try_to_free_buffers(journal
, page
, wait
);
3010 * If the O_DIRECT write will extend the file then add this inode to the
3011 * orphan list. So recovery will truncate it back to the original size
3012 * if the machine crashes during the write.
3014 * If the O_DIRECT write is intantiating holes inside i_size and the machine
3015 * crashes then stale disk data _may_ be exposed inside the file. But current
3016 * VFS code falls back into buffered path in that case so we are safe.
3018 static ssize_t
ext4_direct_IO(int rw
, struct kiocb
*iocb
,
3019 const struct iovec
*iov
, loff_t offset
,
3020 unsigned long nr_segs
)
3022 struct file
*file
= iocb
->ki_filp
;
3023 struct inode
*inode
= file
->f_mapping
->host
;
3024 struct ext4_inode_info
*ei
= EXT4_I(inode
);
3028 size_t count
= iov_length(iov
, nr_segs
);
3031 loff_t final_size
= offset
+ count
;
3033 if (final_size
> inode
->i_size
) {
3034 /* Credits for sb + inode write */
3035 handle
= ext4_journal_start(inode
, 2);
3036 if (IS_ERR(handle
)) {
3037 ret
= PTR_ERR(handle
);
3040 ret
= ext4_orphan_add(handle
, inode
);
3042 ext4_journal_stop(handle
);
3046 ei
->i_disksize
= inode
->i_size
;
3047 ext4_journal_stop(handle
);
3051 ret
= blockdev_direct_IO(rw
, iocb
, inode
, inode
->i_sb
->s_bdev
, iov
,
3053 ext4_get_block
, NULL
);
3058 /* Credits for sb + inode write */
3059 handle
= ext4_journal_start(inode
, 2);
3060 if (IS_ERR(handle
)) {
3061 /* This is really bad luck. We've written the data
3062 * but cannot extend i_size. Bail out and pretend
3063 * the write failed... */
3064 ret
= PTR_ERR(handle
);
3068 ext4_orphan_del(handle
, inode
);
3070 loff_t end
= offset
+ ret
;
3071 if (end
> inode
->i_size
) {
3072 ei
->i_disksize
= end
;
3073 i_size_write(inode
, end
);
3075 * We're going to return a positive `ret'
3076 * here due to non-zero-length I/O, so there's
3077 * no way of reporting error returns from
3078 * ext4_mark_inode_dirty() to userspace. So
3081 ext4_mark_inode_dirty(handle
, inode
);
3084 err
= ext4_journal_stop(handle
);
3093 * Pages can be marked dirty completely asynchronously from ext4's journalling
3094 * activity. By filemap_sync_pte(), try_to_unmap_one(), etc. We cannot do
3095 * much here because ->set_page_dirty is called under VFS locks. The page is
3096 * not necessarily locked.
3098 * We cannot just dirty the page and leave attached buffers clean, because the
3099 * buffers' dirty state is "definitive". We cannot just set the buffers dirty
3100 * or jbddirty because all the journalling code will explode.
3102 * So what we do is to mark the page "pending dirty" and next time writepage
3103 * is called, propagate that into the buffers appropriately.
3105 static int ext4_journalled_set_page_dirty(struct page
*page
)
3107 SetPageChecked(page
);
3108 return __set_page_dirty_nobuffers(page
);
3111 static const struct address_space_operations ext4_ordered_aops
= {
3112 .readpage
= ext4_readpage
,
3113 .readpages
= ext4_readpages
,
3114 .writepage
= ext4_normal_writepage
,
3115 .sync_page
= block_sync_page
,
3116 .write_begin
= ext4_write_begin
,
3117 .write_end
= ext4_ordered_write_end
,
3119 .invalidatepage
= ext4_invalidatepage
,
3120 .releasepage
= ext4_releasepage
,
3121 .direct_IO
= ext4_direct_IO
,
3122 .migratepage
= buffer_migrate_page
,
3123 .is_partially_uptodate
= block_is_partially_uptodate
,
3126 static const struct address_space_operations ext4_writeback_aops
= {
3127 .readpage
= ext4_readpage
,
3128 .readpages
= ext4_readpages
,
3129 .writepage
= ext4_normal_writepage
,
3130 .sync_page
= block_sync_page
,
3131 .write_begin
= ext4_write_begin
,
3132 .write_end
= ext4_writeback_write_end
,
3134 .invalidatepage
= ext4_invalidatepage
,
3135 .releasepage
= ext4_releasepage
,
3136 .direct_IO
= ext4_direct_IO
,
3137 .migratepage
= buffer_migrate_page
,
3138 .is_partially_uptodate
= block_is_partially_uptodate
,
3141 static const struct address_space_operations ext4_journalled_aops
= {
3142 .readpage
= ext4_readpage
,
3143 .readpages
= ext4_readpages
,
3144 .writepage
= ext4_journalled_writepage
,
3145 .sync_page
= block_sync_page
,
3146 .write_begin
= ext4_write_begin
,
3147 .write_end
= ext4_journalled_write_end
,
3148 .set_page_dirty
= ext4_journalled_set_page_dirty
,
3150 .invalidatepage
= ext4_invalidatepage
,
3151 .releasepage
= ext4_releasepage
,
3152 .is_partially_uptodate
= block_is_partially_uptodate
,
3155 static const struct address_space_operations ext4_da_aops
= {
3156 .readpage
= ext4_readpage
,
3157 .readpages
= ext4_readpages
,
3158 .writepage
= ext4_da_writepage
,
3159 .writepages
= ext4_da_writepages
,
3160 .sync_page
= block_sync_page
,
3161 .write_begin
= ext4_da_write_begin
,
3162 .write_end
= ext4_da_write_end
,
3164 .invalidatepage
= ext4_da_invalidatepage
,
3165 .releasepage
= ext4_releasepage
,
3166 .direct_IO
= ext4_direct_IO
,
3167 .migratepage
= buffer_migrate_page
,
3168 .is_partially_uptodate
= block_is_partially_uptodate
,
3171 void ext4_set_aops(struct inode
*inode
)
3173 if (ext4_should_order_data(inode
) &&
3174 test_opt(inode
->i_sb
, DELALLOC
))
3175 inode
->i_mapping
->a_ops
= &ext4_da_aops
;
3176 else if (ext4_should_order_data(inode
))
3177 inode
->i_mapping
->a_ops
= &ext4_ordered_aops
;
3178 else if (ext4_should_writeback_data(inode
) &&
3179 test_opt(inode
->i_sb
, DELALLOC
))
3180 inode
->i_mapping
->a_ops
= &ext4_da_aops
;
3181 else if (ext4_should_writeback_data(inode
))
3182 inode
->i_mapping
->a_ops
= &ext4_writeback_aops
;
3184 inode
->i_mapping
->a_ops
= &ext4_journalled_aops
;
3188 * ext4_block_truncate_page() zeroes out a mapping from file offset `from'
3189 * up to the end of the block which corresponds to `from'.
3190 * This required during truncate. We need to physically zero the tail end
3191 * of that block so it doesn't yield old data if the file is later grown.
3193 int ext4_block_truncate_page(handle_t
*handle
,
3194 struct address_space
*mapping
, loff_t from
)
3196 ext4_fsblk_t index
= from
>> PAGE_CACHE_SHIFT
;
3197 unsigned offset
= from
& (PAGE_CACHE_SIZE
-1);
3198 unsigned blocksize
, length
, pos
;
3200 struct inode
*inode
= mapping
->host
;
3201 struct buffer_head
*bh
;
3205 page
= grab_cache_page(mapping
, from
>> PAGE_CACHE_SHIFT
);
3209 blocksize
= inode
->i_sb
->s_blocksize
;
3210 length
= blocksize
- (offset
& (blocksize
- 1));
3211 iblock
= index
<< (PAGE_CACHE_SHIFT
- inode
->i_sb
->s_blocksize_bits
);
3214 * For "nobh" option, we can only work if we don't need to
3215 * read-in the page - otherwise we create buffers to do the IO.
3217 if (!page_has_buffers(page
) && test_opt(inode
->i_sb
, NOBH
) &&
3218 ext4_should_writeback_data(inode
) && PageUptodate(page
)) {
3219 zero_user(page
, offset
, length
);
3220 set_page_dirty(page
);
3224 if (!page_has_buffers(page
))
3225 create_empty_buffers(page
, blocksize
, 0);
3227 /* Find the buffer that contains "offset" */
3228 bh
= page_buffers(page
);
3230 while (offset
>= pos
) {
3231 bh
= bh
->b_this_page
;
3237 if (buffer_freed(bh
)) {
3238 BUFFER_TRACE(bh
, "freed: skip");
3242 if (!buffer_mapped(bh
)) {
3243 BUFFER_TRACE(bh
, "unmapped");
3244 ext4_get_block(inode
, iblock
, bh
, 0);
3245 /* unmapped? It's a hole - nothing to do */
3246 if (!buffer_mapped(bh
)) {
3247 BUFFER_TRACE(bh
, "still unmapped");
3252 /* Ok, it's mapped. Make sure it's up-to-date */
3253 if (PageUptodate(page
))
3254 set_buffer_uptodate(bh
);
3256 if (!buffer_uptodate(bh
)) {
3258 ll_rw_block(READ
, 1, &bh
);
3260 /* Uhhuh. Read error. Complain and punt. */
3261 if (!buffer_uptodate(bh
))
3265 if (ext4_should_journal_data(inode
)) {
3266 BUFFER_TRACE(bh
, "get write access");
3267 err
= ext4_journal_get_write_access(handle
, bh
);
3272 zero_user(page
, offset
, length
);
3274 BUFFER_TRACE(bh
, "zeroed end of block");
3277 if (ext4_should_journal_data(inode
)) {
3278 err
= ext4_journal_dirty_metadata(handle
, bh
);
3280 if (ext4_should_order_data(inode
))
3281 err
= ext4_jbd2_file_inode(handle
, inode
);
3282 mark_buffer_dirty(bh
);
3287 page_cache_release(page
);
3292 * Probably it should be a library function... search for first non-zero word
3293 * or memcmp with zero_page, whatever is better for particular architecture.
3296 static inline int all_zeroes(__le32
*p
, __le32
*q
)
3305 * ext4_find_shared - find the indirect blocks for partial truncation.
3306 * @inode: inode in question
3307 * @depth: depth of the affected branch
3308 * @offsets: offsets of pointers in that branch (see ext4_block_to_path)
3309 * @chain: place to store the pointers to partial indirect blocks
3310 * @top: place to the (detached) top of branch
3312 * This is a helper function used by ext4_truncate().
3314 * When we do truncate() we may have to clean the ends of several
3315 * indirect blocks but leave the blocks themselves alive. Block is
3316 * partially truncated if some data below the new i_size is refered
3317 * from it (and it is on the path to the first completely truncated
3318 * data block, indeed). We have to free the top of that path along
3319 * with everything to the right of the path. Since no allocation
3320 * past the truncation point is possible until ext4_truncate()
3321 * finishes, we may safely do the latter, but top of branch may
3322 * require special attention - pageout below the truncation point
3323 * might try to populate it.
3325 * We atomically detach the top of branch from the tree, store the
3326 * block number of its root in *@top, pointers to buffer_heads of
3327 * partially truncated blocks - in @chain[].bh and pointers to
3328 * their last elements that should not be removed - in
3329 * @chain[].p. Return value is the pointer to last filled element
3332 * The work left to caller to do the actual freeing of subtrees:
3333 * a) free the subtree starting from *@top
3334 * b) free the subtrees whose roots are stored in
3335 * (@chain[i].p+1 .. end of @chain[i].bh->b_data)
3336 * c) free the subtrees growing from the inode past the @chain[0].
3337 * (no partially truncated stuff there). */
3339 static Indirect
*ext4_find_shared(struct inode
*inode
, int depth
,
3340 ext4_lblk_t offsets
[4], Indirect chain
[4], __le32
*top
)
3342 Indirect
*partial
, *p
;
3346 /* Make k index the deepest non-null offest + 1 */
3347 for (k
= depth
; k
> 1 && !offsets
[k
-1]; k
--)
3349 partial
= ext4_get_branch(inode
, k
, offsets
, chain
, &err
);
3350 /* Writer: pointers */
3352 partial
= chain
+ k
-1;
3354 * If the branch acquired continuation since we've looked at it -
3355 * fine, it should all survive and (new) top doesn't belong to us.
3357 if (!partial
->key
&& *partial
->p
)
3360 for (p
= partial
; (p
> chain
) && all_zeroes((__le32
*) p
->bh
->b_data
, p
->p
); p
--)
3363 * OK, we've found the last block that must survive. The rest of our
3364 * branch should be detached before unlocking. However, if that rest
3365 * of branch is all ours and does not grow immediately from the inode
3366 * it's easier to cheat and just decrement partial->p.
3368 if (p
== chain
+ k
- 1 && p
> chain
) {
3372 /* Nope, don't do this in ext4. Must leave the tree intact */
3379 while (partial
> p
) {
3380 brelse(partial
->bh
);
3388 * Zero a number of block pointers in either an inode or an indirect block.
3389 * If we restart the transaction we must again get write access to the
3390 * indirect block for further modification.
3392 * We release `count' blocks on disk, but (last - first) may be greater
3393 * than `count' because there can be holes in there.
3395 static void ext4_clear_blocks(handle_t
*handle
, struct inode
*inode
,
3396 struct buffer_head
*bh
, ext4_fsblk_t block_to_free
,
3397 unsigned long count
, __le32
*first
, __le32
*last
)
3400 if (try_to_extend_transaction(handle
, inode
)) {
3402 BUFFER_TRACE(bh
, "call ext4_journal_dirty_metadata");
3403 ext4_journal_dirty_metadata(handle
, bh
);
3405 ext4_mark_inode_dirty(handle
, inode
);
3406 ext4_journal_test_restart(handle
, inode
);
3408 BUFFER_TRACE(bh
, "retaking write access");
3409 ext4_journal_get_write_access(handle
, bh
);
3414 * Any buffers which are on the journal will be in memory. We find
3415 * them on the hash table so jbd2_journal_revoke() will run jbd2_journal_forget()
3416 * on them. We've already detached each block from the file, so
3417 * bforget() in jbd2_journal_forget() should be safe.
3419 * AKPM: turn on bforget in jbd2_journal_forget()!!!
3421 for (p
= first
; p
< last
; p
++) {
3422 u32 nr
= le32_to_cpu(*p
);
3424 struct buffer_head
*tbh
;
3427 tbh
= sb_find_get_block(inode
->i_sb
, nr
);
3428 ext4_forget(handle
, 0, inode
, tbh
, nr
);
3432 ext4_free_blocks(handle
, inode
, block_to_free
, count
, 0);
3436 * ext4_free_data - free a list of data blocks
3437 * @handle: handle for this transaction
3438 * @inode: inode we are dealing with
3439 * @this_bh: indirect buffer_head which contains *@first and *@last
3440 * @first: array of block numbers
3441 * @last: points immediately past the end of array
3443 * We are freeing all blocks refered from that array (numbers are stored as
3444 * little-endian 32-bit) and updating @inode->i_blocks appropriately.
3446 * We accumulate contiguous runs of blocks to free. Conveniently, if these
3447 * blocks are contiguous then releasing them at one time will only affect one
3448 * or two bitmap blocks (+ group descriptor(s) and superblock) and we won't
3449 * actually use a lot of journal space.
3451 * @this_bh will be %NULL if @first and @last point into the inode's direct
3454 static void ext4_free_data(handle_t
*handle
, struct inode
*inode
,
3455 struct buffer_head
*this_bh
,
3456 __le32
*first
, __le32
*last
)
3458 ext4_fsblk_t block_to_free
= 0; /* Starting block # of a run */
3459 unsigned long count
= 0; /* Number of blocks in the run */
3460 __le32
*block_to_free_p
= NULL
; /* Pointer into inode/ind
3463 ext4_fsblk_t nr
; /* Current block # */
3464 __le32
*p
; /* Pointer into inode/ind
3465 for current block */
3468 if (this_bh
) { /* For indirect block */
3469 BUFFER_TRACE(this_bh
, "get_write_access");
3470 err
= ext4_journal_get_write_access(handle
, this_bh
);
3471 /* Important: if we can't update the indirect pointers
3472 * to the blocks, we can't free them. */
3477 for (p
= first
; p
< last
; p
++) {
3478 nr
= le32_to_cpu(*p
);
3480 /* accumulate blocks to free if they're contiguous */
3483 block_to_free_p
= p
;
3485 } else if (nr
== block_to_free
+ count
) {
3488 ext4_clear_blocks(handle
, inode
, this_bh
,
3490 count
, block_to_free_p
, p
);
3492 block_to_free_p
= p
;
3499 ext4_clear_blocks(handle
, inode
, this_bh
, block_to_free
,
3500 count
, block_to_free_p
, p
);
3503 BUFFER_TRACE(this_bh
, "call ext4_journal_dirty_metadata");
3506 * The buffer head should have an attached journal head at this
3507 * point. However, if the data is corrupted and an indirect
3508 * block pointed to itself, it would have been detached when
3509 * the block was cleared. Check for this instead of OOPSing.
3512 ext4_journal_dirty_metadata(handle
, this_bh
);
3514 ext4_error(inode
->i_sb
, __func__
,
3515 "circular indirect block detected, "
3516 "inode=%lu, block=%llu",
3518 (unsigned long long) this_bh
->b_blocknr
);
3523 * ext4_free_branches - free an array of branches
3524 * @handle: JBD handle for this transaction
3525 * @inode: inode we are dealing with
3526 * @parent_bh: the buffer_head which contains *@first and *@last
3527 * @first: array of block numbers
3528 * @last: pointer immediately past the end of array
3529 * @depth: depth of the branches to free
3531 * We are freeing all blocks refered from these branches (numbers are
3532 * stored as little-endian 32-bit) and updating @inode->i_blocks
3535 static void ext4_free_branches(handle_t
*handle
, struct inode
*inode
,
3536 struct buffer_head
*parent_bh
,
3537 __le32
*first
, __le32
*last
, int depth
)
3542 if (is_handle_aborted(handle
))
3546 struct buffer_head
*bh
;
3547 int addr_per_block
= EXT4_ADDR_PER_BLOCK(inode
->i_sb
);
3549 while (--p
>= first
) {
3550 nr
= le32_to_cpu(*p
);
3552 continue; /* A hole */
3554 /* Go read the buffer for the next level down */
3555 bh
= sb_bread(inode
->i_sb
, nr
);
3558 * A read failure? Report error and clear slot
3562 ext4_error(inode
->i_sb
, "ext4_free_branches",
3563 "Read failure, inode=%lu, block=%llu",
3568 /* This zaps the entire block. Bottom up. */
3569 BUFFER_TRACE(bh
, "free child branches");
3570 ext4_free_branches(handle
, inode
, bh
,
3571 (__le32
*) bh
->b_data
,
3572 (__le32
*) bh
->b_data
+ addr_per_block
,
3576 * We've probably journalled the indirect block several
3577 * times during the truncate. But it's no longer
3578 * needed and we now drop it from the transaction via
3579 * jbd2_journal_revoke().
3581 * That's easy if it's exclusively part of this
3582 * transaction. But if it's part of the committing
3583 * transaction then jbd2_journal_forget() will simply
3584 * brelse() it. That means that if the underlying
3585 * block is reallocated in ext4_get_block(),
3586 * unmap_underlying_metadata() will find this block
3587 * and will try to get rid of it. damn, damn.
3589 * If this block has already been committed to the
3590 * journal, a revoke record will be written. And
3591 * revoke records must be emitted *before* clearing
3592 * this block's bit in the bitmaps.
3594 ext4_forget(handle
, 1, inode
, bh
, bh
->b_blocknr
);
3597 * Everything below this this pointer has been
3598 * released. Now let this top-of-subtree go.
3600 * We want the freeing of this indirect block to be
3601 * atomic in the journal with the updating of the
3602 * bitmap block which owns it. So make some room in
3605 * We zero the parent pointer *after* freeing its
3606 * pointee in the bitmaps, so if extend_transaction()
3607 * for some reason fails to put the bitmap changes and
3608 * the release into the same transaction, recovery
3609 * will merely complain about releasing a free block,
3610 * rather than leaking blocks.
3612 if (is_handle_aborted(handle
))
3614 if (try_to_extend_transaction(handle
, inode
)) {
3615 ext4_mark_inode_dirty(handle
, inode
);
3616 ext4_journal_test_restart(handle
, inode
);
3619 ext4_free_blocks(handle
, inode
, nr
, 1, 1);
3623 * The block which we have just freed is
3624 * pointed to by an indirect block: journal it
3626 BUFFER_TRACE(parent_bh
, "get_write_access");
3627 if (!ext4_journal_get_write_access(handle
,
3630 BUFFER_TRACE(parent_bh
,
3631 "call ext4_journal_dirty_metadata");
3632 ext4_journal_dirty_metadata(handle
,
3638 /* We have reached the bottom of the tree. */
3639 BUFFER_TRACE(parent_bh
, "free data blocks");
3640 ext4_free_data(handle
, inode
, parent_bh
, first
, last
);
3644 int ext4_can_truncate(struct inode
*inode
)
3646 if (IS_APPEND(inode
) || IS_IMMUTABLE(inode
))
3648 if (S_ISREG(inode
->i_mode
))
3650 if (S_ISDIR(inode
->i_mode
))
3652 if (S_ISLNK(inode
->i_mode
))
3653 return !ext4_inode_is_fast_symlink(inode
);
3660 * We block out ext4_get_block() block instantiations across the entire
3661 * transaction, and VFS/VM ensures that ext4_truncate() cannot run
3662 * simultaneously on behalf of the same inode.
3664 * As we work through the truncate and commmit bits of it to the journal there
3665 * is one core, guiding principle: the file's tree must always be consistent on
3666 * disk. We must be able to restart the truncate after a crash.
3668 * The file's tree may be transiently inconsistent in memory (although it
3669 * probably isn't), but whenever we close off and commit a journal transaction,
3670 * the contents of (the filesystem + the journal) must be consistent and
3671 * restartable. It's pretty simple, really: bottom up, right to left (although
3672 * left-to-right works OK too).
3674 * Note that at recovery time, journal replay occurs *before* the restart of
3675 * truncate against the orphan inode list.
3677 * The committed inode has the new, desired i_size (which is the same as
3678 * i_disksize in this case). After a crash, ext4_orphan_cleanup() will see
3679 * that this inode's truncate did not complete and it will again call
3680 * ext4_truncate() to have another go. So there will be instantiated blocks
3681 * to the right of the truncation point in a crashed ext4 filesystem. But
3682 * that's fine - as long as they are linked from the inode, the post-crash
3683 * ext4_truncate() run will find them and release them.
3685 void ext4_truncate(struct inode
*inode
)
3688 struct ext4_inode_info
*ei
= EXT4_I(inode
);
3689 __le32
*i_data
= ei
->i_data
;
3690 int addr_per_block
= EXT4_ADDR_PER_BLOCK(inode
->i_sb
);
3691 struct address_space
*mapping
= inode
->i_mapping
;
3692 ext4_lblk_t offsets
[4];
3697 ext4_lblk_t last_block
;
3698 unsigned blocksize
= inode
->i_sb
->s_blocksize
;
3700 if (!ext4_can_truncate(inode
))
3703 if (EXT4_I(inode
)->i_flags
& EXT4_EXTENTS_FL
) {
3704 ext4_ext_truncate(inode
);
3708 handle
= start_transaction(inode
);
3710 return; /* AKPM: return what? */
3712 last_block
= (inode
->i_size
+ blocksize
-1)
3713 >> EXT4_BLOCK_SIZE_BITS(inode
->i_sb
);
3715 if (inode
->i_size
& (blocksize
- 1))
3716 if (ext4_block_truncate_page(handle
, mapping
, inode
->i_size
))
3719 n
= ext4_block_to_path(inode
, last_block
, offsets
, NULL
);
3721 goto out_stop
; /* error */
3724 * OK. This truncate is going to happen. We add the inode to the
3725 * orphan list, so that if this truncate spans multiple transactions,
3726 * and we crash, we will resume the truncate when the filesystem
3727 * recovers. It also marks the inode dirty, to catch the new size.
3729 * Implication: the file must always be in a sane, consistent
3730 * truncatable state while each transaction commits.
3732 if (ext4_orphan_add(handle
, inode
))
3736 * From here we block out all ext4_get_block() callers who want to
3737 * modify the block allocation tree.
3739 down_write(&ei
->i_data_sem
);
3741 ext4_discard_reservation(inode
);
3744 * The orphan list entry will now protect us from any crash which
3745 * occurs before the truncate completes, so it is now safe to propagate
3746 * the new, shorter inode size (held for now in i_size) into the
3747 * on-disk inode. We do this via i_disksize, which is the value which
3748 * ext4 *really* writes onto the disk inode.
3750 ei
->i_disksize
= inode
->i_size
;
3752 if (n
== 1) { /* direct blocks */
3753 ext4_free_data(handle
, inode
, NULL
, i_data
+offsets
[0],
3754 i_data
+ EXT4_NDIR_BLOCKS
);
3758 partial
= ext4_find_shared(inode
, n
, offsets
, chain
, &nr
);
3759 /* Kill the top of shared branch (not detached) */
3761 if (partial
== chain
) {
3762 /* Shared branch grows from the inode */
3763 ext4_free_branches(handle
, inode
, NULL
,
3764 &nr
, &nr
+1, (chain
+n
-1) - partial
);
3767 * We mark the inode dirty prior to restart,
3768 * and prior to stop. No need for it here.
3771 /* Shared branch grows from an indirect block */
3772 BUFFER_TRACE(partial
->bh
, "get_write_access");
3773 ext4_free_branches(handle
, inode
, partial
->bh
,
3775 partial
->p
+1, (chain
+n
-1) - partial
);
3778 /* Clear the ends of indirect blocks on the shared branch */
3779 while (partial
> chain
) {
3780 ext4_free_branches(handle
, inode
, partial
->bh
, partial
->p
+ 1,
3781 (__le32
*)partial
->bh
->b_data
+addr_per_block
,
3782 (chain
+n
-1) - partial
);
3783 BUFFER_TRACE(partial
->bh
, "call brelse");
3784 brelse (partial
->bh
);
3788 /* Kill the remaining (whole) subtrees */
3789 switch (offsets
[0]) {
3791 nr
= i_data
[EXT4_IND_BLOCK
];
3793 ext4_free_branches(handle
, inode
, NULL
, &nr
, &nr
+1, 1);
3794 i_data
[EXT4_IND_BLOCK
] = 0;
3796 case EXT4_IND_BLOCK
:
3797 nr
= i_data
[EXT4_DIND_BLOCK
];
3799 ext4_free_branches(handle
, inode
, NULL
, &nr
, &nr
+1, 2);
3800 i_data
[EXT4_DIND_BLOCK
] = 0;
3802 case EXT4_DIND_BLOCK
:
3803 nr
= i_data
[EXT4_TIND_BLOCK
];
3805 ext4_free_branches(handle
, inode
, NULL
, &nr
, &nr
+1, 3);
3806 i_data
[EXT4_TIND_BLOCK
] = 0;
3808 case EXT4_TIND_BLOCK
:
3812 up_write(&ei
->i_data_sem
);
3813 inode
->i_mtime
= inode
->i_ctime
= ext4_current_time(inode
);
3814 ext4_mark_inode_dirty(handle
, inode
);
3817 * In a multi-transaction truncate, we only make the final transaction
3824 * If this was a simple ftruncate(), and the file will remain alive
3825 * then we need to clear up the orphan record which we created above.
3826 * However, if this was a real unlink then we were called by
3827 * ext4_delete_inode(), and we allow that function to clean up the
3828 * orphan info for us.
3831 ext4_orphan_del(handle
, inode
);
3833 ext4_journal_stop(handle
);
3836 static ext4_fsblk_t
ext4_get_inode_block(struct super_block
*sb
,
3837 unsigned long ino
, struct ext4_iloc
*iloc
)
3839 ext4_group_t block_group
;
3840 unsigned long offset
;
3842 struct ext4_group_desc
*gdp
;
3844 if (!ext4_valid_inum(sb
, ino
)) {
3846 * This error is already checked for in namei.c unless we are
3847 * looking at an NFS filehandle, in which case no error
3853 block_group
= (ino
- 1) / EXT4_INODES_PER_GROUP(sb
);
3854 gdp
= ext4_get_group_desc(sb
, block_group
, NULL
);
3859 * Figure out the offset within the block group inode table
3861 offset
= ((ino
- 1) % EXT4_INODES_PER_GROUP(sb
)) *
3862 EXT4_INODE_SIZE(sb
);
3863 block
= ext4_inode_table(sb
, gdp
) +
3864 (offset
>> EXT4_BLOCK_SIZE_BITS(sb
));
3866 iloc
->block_group
= block_group
;
3867 iloc
->offset
= offset
& (EXT4_BLOCK_SIZE(sb
) - 1);
3872 * ext4_get_inode_loc returns with an extra refcount against the inode's
3873 * underlying buffer_head on success. If 'in_mem' is true, we have all
3874 * data in memory that is needed to recreate the on-disk version of this
3877 static int __ext4_get_inode_loc(struct inode
*inode
,
3878 struct ext4_iloc
*iloc
, int in_mem
)
3881 struct buffer_head
*bh
;
3883 block
= ext4_get_inode_block(inode
->i_sb
, inode
->i_ino
, iloc
);
3887 bh
= sb_getblk(inode
->i_sb
, block
);
3889 ext4_error (inode
->i_sb
, "ext4_get_inode_loc",
3890 "unable to read inode block - "
3891 "inode=%lu, block=%llu",
3892 inode
->i_ino
, block
);
3895 if (!buffer_uptodate(bh
)) {
3899 * If the buffer has the write error flag, we have failed
3900 * to write out another inode in the same block. In this
3901 * case, we don't have to read the block because we may
3902 * read the old inode data successfully.
3904 if (buffer_write_io_error(bh
) && !buffer_uptodate(bh
))
3905 set_buffer_uptodate(bh
);
3907 if (buffer_uptodate(bh
)) {
3908 /* someone brought it uptodate while we waited */
3914 * If we have all information of the inode in memory and this
3915 * is the only valid inode in the block, we need not read the
3919 struct buffer_head
*bitmap_bh
;
3920 struct ext4_group_desc
*desc
;
3921 int inodes_per_buffer
;
3922 int inode_offset
, i
;
3923 ext4_group_t block_group
;
3926 block_group
= (inode
->i_ino
- 1) /
3927 EXT4_INODES_PER_GROUP(inode
->i_sb
);
3928 inodes_per_buffer
= bh
->b_size
/
3929 EXT4_INODE_SIZE(inode
->i_sb
);
3930 inode_offset
= ((inode
->i_ino
- 1) %
3931 EXT4_INODES_PER_GROUP(inode
->i_sb
));
3932 start
= inode_offset
& ~(inodes_per_buffer
- 1);
3934 /* Is the inode bitmap in cache? */
3935 desc
= ext4_get_group_desc(inode
->i_sb
,
3940 bitmap_bh
= sb_getblk(inode
->i_sb
,
3941 ext4_inode_bitmap(inode
->i_sb
, desc
));
3946 * If the inode bitmap isn't in cache then the
3947 * optimisation may end up performing two reads instead
3948 * of one, so skip it.
3950 if (!buffer_uptodate(bitmap_bh
)) {
3954 for (i
= start
; i
< start
+ inodes_per_buffer
; i
++) {
3955 if (i
== inode_offset
)
3957 if (ext4_test_bit(i
, bitmap_bh
->b_data
))
3961 if (i
== start
+ inodes_per_buffer
) {
3962 /* all other inodes are free, so skip I/O */
3963 memset(bh
->b_data
, 0, bh
->b_size
);
3964 set_buffer_uptodate(bh
);
3972 * There are other valid inodes in the buffer, this inode
3973 * has in-inode xattrs, or we don't have this inode in memory.
3974 * Read the block from disk.
3977 bh
->b_end_io
= end_buffer_read_sync
;
3978 submit_bh(READ_META
, bh
);
3980 if (!buffer_uptodate(bh
)) {
3981 ext4_error(inode
->i_sb
, "ext4_get_inode_loc",
3982 "unable to read inode block - "
3983 "inode=%lu, block=%llu",
3984 inode
->i_ino
, block
);
3994 int ext4_get_inode_loc(struct inode
*inode
, struct ext4_iloc
*iloc
)
3996 /* We have all inode data except xattrs in memory here. */
3997 return __ext4_get_inode_loc(inode
, iloc
,
3998 !(EXT4_I(inode
)->i_state
& EXT4_STATE_XATTR
));
4001 void ext4_set_inode_flags(struct inode
*inode
)
4003 unsigned int flags
= EXT4_I(inode
)->i_flags
;
4005 inode
->i_flags
&= ~(S_SYNC
|S_APPEND
|S_IMMUTABLE
|S_NOATIME
|S_DIRSYNC
);
4006 if (flags
& EXT4_SYNC_FL
)
4007 inode
->i_flags
|= S_SYNC
;
4008 if (flags
& EXT4_APPEND_FL
)
4009 inode
->i_flags
|= S_APPEND
;
4010 if (flags
& EXT4_IMMUTABLE_FL
)
4011 inode
->i_flags
|= S_IMMUTABLE
;
4012 if (flags
& EXT4_NOATIME_FL
)
4013 inode
->i_flags
|= S_NOATIME
;
4014 if (flags
& EXT4_DIRSYNC_FL
)
4015 inode
->i_flags
|= S_DIRSYNC
;
4018 /* Propagate flags from i_flags to EXT4_I(inode)->i_flags */
4019 void ext4_get_inode_flags(struct ext4_inode_info
*ei
)
4021 unsigned int flags
= ei
->vfs_inode
.i_flags
;
4023 ei
->i_flags
&= ~(EXT4_SYNC_FL
|EXT4_APPEND_FL
|
4024 EXT4_IMMUTABLE_FL
|EXT4_NOATIME_FL
|EXT4_DIRSYNC_FL
);
4026 ei
->i_flags
|= EXT4_SYNC_FL
;
4027 if (flags
& S_APPEND
)
4028 ei
->i_flags
|= EXT4_APPEND_FL
;
4029 if (flags
& S_IMMUTABLE
)
4030 ei
->i_flags
|= EXT4_IMMUTABLE_FL
;
4031 if (flags
& S_NOATIME
)
4032 ei
->i_flags
|= EXT4_NOATIME_FL
;
4033 if (flags
& S_DIRSYNC
)
4034 ei
->i_flags
|= EXT4_DIRSYNC_FL
;
4036 static blkcnt_t
ext4_inode_blocks(struct ext4_inode
*raw_inode
,
4037 struct ext4_inode_info
*ei
)
4040 struct inode
*inode
= &(ei
->vfs_inode
);
4041 struct super_block
*sb
= inode
->i_sb
;
4043 if (EXT4_HAS_RO_COMPAT_FEATURE(sb
,
4044 EXT4_FEATURE_RO_COMPAT_HUGE_FILE
)) {
4045 /* we are using combined 48 bit field */
4046 i_blocks
= ((u64
)le16_to_cpu(raw_inode
->i_blocks_high
)) << 32 |
4047 le32_to_cpu(raw_inode
->i_blocks_lo
);
4048 if (ei
->i_flags
& EXT4_HUGE_FILE_FL
) {
4049 /* i_blocks represent file system block size */
4050 return i_blocks
<< (inode
->i_blkbits
- 9);
4055 return le32_to_cpu(raw_inode
->i_blocks_lo
);
4059 struct inode
*ext4_iget(struct super_block
*sb
, unsigned long ino
)
4061 struct ext4_iloc iloc
;
4062 struct ext4_inode
*raw_inode
;
4063 struct ext4_inode_info
*ei
;
4064 struct buffer_head
*bh
;
4065 struct inode
*inode
;
4069 inode
= iget_locked(sb
, ino
);
4071 return ERR_PTR(-ENOMEM
);
4072 if (!(inode
->i_state
& I_NEW
))
4076 #ifdef CONFIG_EXT4DEV_FS_POSIX_ACL
4077 ei
->i_acl
= EXT4_ACL_NOT_CACHED
;
4078 ei
->i_default_acl
= EXT4_ACL_NOT_CACHED
;
4080 ei
->i_block_alloc_info
= NULL
;
4082 ret
= __ext4_get_inode_loc(inode
, &iloc
, 0);
4086 raw_inode
= ext4_raw_inode(&iloc
);
4087 inode
->i_mode
= le16_to_cpu(raw_inode
->i_mode
);
4088 inode
->i_uid
= (uid_t
)le16_to_cpu(raw_inode
->i_uid_low
);
4089 inode
->i_gid
= (gid_t
)le16_to_cpu(raw_inode
->i_gid_low
);
4090 if (!(test_opt(inode
->i_sb
, NO_UID32
))) {
4091 inode
->i_uid
|= le16_to_cpu(raw_inode
->i_uid_high
) << 16;
4092 inode
->i_gid
|= le16_to_cpu(raw_inode
->i_gid_high
) << 16;
4094 inode
->i_nlink
= le16_to_cpu(raw_inode
->i_links_count
);
4097 ei
->i_dir_start_lookup
= 0;
4098 ei
->i_dtime
= le32_to_cpu(raw_inode
->i_dtime
);
4099 /* We now have enough fields to check if the inode was active or not.
4100 * This is needed because nfsd might try to access dead inodes
4101 * the test is that same one that e2fsck uses
4102 * NeilBrown 1999oct15
4104 if (inode
->i_nlink
== 0) {
4105 if (inode
->i_mode
== 0 ||
4106 !(EXT4_SB(inode
->i_sb
)->s_mount_state
& EXT4_ORPHAN_FS
)) {
4107 /* this inode is deleted */
4112 /* The only unlinked inodes we let through here have
4113 * valid i_mode and are being read by the orphan
4114 * recovery code: that's fine, we're about to complete
4115 * the process of deleting those. */
4117 ei
->i_flags
= le32_to_cpu(raw_inode
->i_flags
);
4118 inode
->i_blocks
= ext4_inode_blocks(raw_inode
, ei
);
4119 ei
->i_file_acl
= le32_to_cpu(raw_inode
->i_file_acl_lo
);
4120 if (EXT4_SB(inode
->i_sb
)->s_es
->s_creator_os
!=
4121 cpu_to_le32(EXT4_OS_HURD
)) {
4123 ((__u64
)le16_to_cpu(raw_inode
->i_file_acl_high
)) << 32;
4125 inode
->i_size
= ext4_isize(raw_inode
);
4126 ei
->i_disksize
= inode
->i_size
;
4127 inode
->i_generation
= le32_to_cpu(raw_inode
->i_generation
);
4128 ei
->i_block_group
= iloc
.block_group
;
4130 * NOTE! The in-memory inode i_data array is in little-endian order
4131 * even on big-endian machines: we do NOT byteswap the block numbers!
4133 for (block
= 0; block
< EXT4_N_BLOCKS
; block
++)
4134 ei
->i_data
[block
] = raw_inode
->i_block
[block
];
4135 INIT_LIST_HEAD(&ei
->i_orphan
);
4137 if (EXT4_INODE_SIZE(inode
->i_sb
) > EXT4_GOOD_OLD_INODE_SIZE
) {
4138 ei
->i_extra_isize
= le16_to_cpu(raw_inode
->i_extra_isize
);
4139 if (EXT4_GOOD_OLD_INODE_SIZE
+ ei
->i_extra_isize
>
4140 EXT4_INODE_SIZE(inode
->i_sb
)) {
4145 if (ei
->i_extra_isize
== 0) {
4146 /* The extra space is currently unused. Use it. */
4147 ei
->i_extra_isize
= sizeof(struct ext4_inode
) -
4148 EXT4_GOOD_OLD_INODE_SIZE
;
4150 __le32
*magic
= (void *)raw_inode
+
4151 EXT4_GOOD_OLD_INODE_SIZE
+
4153 if (*magic
== cpu_to_le32(EXT4_XATTR_MAGIC
))
4154 ei
->i_state
|= EXT4_STATE_XATTR
;
4157 ei
->i_extra_isize
= 0;
4159 EXT4_INODE_GET_XTIME(i_ctime
, inode
, raw_inode
);
4160 EXT4_INODE_GET_XTIME(i_mtime
, inode
, raw_inode
);
4161 EXT4_INODE_GET_XTIME(i_atime
, inode
, raw_inode
);
4162 EXT4_EINODE_GET_XTIME(i_crtime
, ei
, raw_inode
);
4164 inode
->i_version
= le32_to_cpu(raw_inode
->i_disk_version
);
4165 if (EXT4_INODE_SIZE(inode
->i_sb
) > EXT4_GOOD_OLD_INODE_SIZE
) {
4166 if (EXT4_FITS_IN_INODE(raw_inode
, ei
, i_version_hi
))
4168 (__u64
)(le32_to_cpu(raw_inode
->i_version_hi
)) << 32;
4171 if (S_ISREG(inode
->i_mode
)) {
4172 inode
->i_op
= &ext4_file_inode_operations
;
4173 inode
->i_fop
= &ext4_file_operations
;
4174 ext4_set_aops(inode
);
4175 } else if (S_ISDIR(inode
->i_mode
)) {
4176 inode
->i_op
= &ext4_dir_inode_operations
;
4177 inode
->i_fop
= &ext4_dir_operations
;
4178 } else if (S_ISLNK(inode
->i_mode
)) {
4179 if (ext4_inode_is_fast_symlink(inode
))
4180 inode
->i_op
= &ext4_fast_symlink_inode_operations
;
4182 inode
->i_op
= &ext4_symlink_inode_operations
;
4183 ext4_set_aops(inode
);
4186 inode
->i_op
= &ext4_special_inode_operations
;
4187 if (raw_inode
->i_block
[0])
4188 init_special_inode(inode
, inode
->i_mode
,
4189 old_decode_dev(le32_to_cpu(raw_inode
->i_block
[0])));
4191 init_special_inode(inode
, inode
->i_mode
,
4192 new_decode_dev(le32_to_cpu(raw_inode
->i_block
[1])));
4195 ext4_set_inode_flags(inode
);
4196 unlock_new_inode(inode
);
4201 return ERR_PTR(ret
);
4204 static int ext4_inode_blocks_set(handle_t
*handle
,
4205 struct ext4_inode
*raw_inode
,
4206 struct ext4_inode_info
*ei
)
4208 struct inode
*inode
= &(ei
->vfs_inode
);
4209 u64 i_blocks
= inode
->i_blocks
;
4210 struct super_block
*sb
= inode
->i_sb
;
4213 if (i_blocks
<= ~0U) {
4215 * i_blocks can be represnted in a 32 bit variable
4216 * as multiple of 512 bytes
4218 raw_inode
->i_blocks_lo
= cpu_to_le32(i_blocks
);
4219 raw_inode
->i_blocks_high
= 0;
4220 ei
->i_flags
&= ~EXT4_HUGE_FILE_FL
;
4221 } else if (i_blocks
<= 0xffffffffffffULL
) {
4223 * i_blocks can be represented in a 48 bit variable
4224 * as multiple of 512 bytes
4226 err
= ext4_update_rocompat_feature(handle
, sb
,
4227 EXT4_FEATURE_RO_COMPAT_HUGE_FILE
);
4230 /* i_block is stored in the split 48 bit fields */
4231 raw_inode
->i_blocks_lo
= cpu_to_le32(i_blocks
);
4232 raw_inode
->i_blocks_high
= cpu_to_le16(i_blocks
>> 32);
4233 ei
->i_flags
&= ~EXT4_HUGE_FILE_FL
;
4236 * i_blocks should be represented in a 48 bit variable
4237 * as multiple of file system block size
4239 err
= ext4_update_rocompat_feature(handle
, sb
,
4240 EXT4_FEATURE_RO_COMPAT_HUGE_FILE
);
4243 ei
->i_flags
|= EXT4_HUGE_FILE_FL
;
4244 /* i_block is stored in file system block size */
4245 i_blocks
= i_blocks
>> (inode
->i_blkbits
- 9);
4246 raw_inode
->i_blocks_lo
= cpu_to_le32(i_blocks
);
4247 raw_inode
->i_blocks_high
= cpu_to_le16(i_blocks
>> 32);
4254 * Post the struct inode info into an on-disk inode location in the
4255 * buffer-cache. This gobbles the caller's reference to the
4256 * buffer_head in the inode location struct.
4258 * The caller must have write access to iloc->bh.
4260 static int ext4_do_update_inode(handle_t
*handle
,
4261 struct inode
*inode
,
4262 struct ext4_iloc
*iloc
)
4264 struct ext4_inode
*raw_inode
= ext4_raw_inode(iloc
);
4265 struct ext4_inode_info
*ei
= EXT4_I(inode
);
4266 struct buffer_head
*bh
= iloc
->bh
;
4267 int err
= 0, rc
, block
;
4269 /* For fields not not tracking in the in-memory inode,
4270 * initialise them to zero for new inodes. */
4271 if (ei
->i_state
& EXT4_STATE_NEW
)
4272 memset(raw_inode
, 0, EXT4_SB(inode
->i_sb
)->s_inode_size
);
4274 ext4_get_inode_flags(ei
);
4275 raw_inode
->i_mode
= cpu_to_le16(inode
->i_mode
);
4276 if (!(test_opt(inode
->i_sb
, NO_UID32
))) {
4277 raw_inode
->i_uid_low
= cpu_to_le16(low_16_bits(inode
->i_uid
));
4278 raw_inode
->i_gid_low
= cpu_to_le16(low_16_bits(inode
->i_gid
));
4280 * Fix up interoperability with old kernels. Otherwise, old inodes get
4281 * re-used with the upper 16 bits of the uid/gid intact
4284 raw_inode
->i_uid_high
=
4285 cpu_to_le16(high_16_bits(inode
->i_uid
));
4286 raw_inode
->i_gid_high
=
4287 cpu_to_le16(high_16_bits(inode
->i_gid
));
4289 raw_inode
->i_uid_high
= 0;
4290 raw_inode
->i_gid_high
= 0;
4293 raw_inode
->i_uid_low
=
4294 cpu_to_le16(fs_high2lowuid(inode
->i_uid
));
4295 raw_inode
->i_gid_low
=
4296 cpu_to_le16(fs_high2lowgid(inode
->i_gid
));
4297 raw_inode
->i_uid_high
= 0;
4298 raw_inode
->i_gid_high
= 0;
4300 raw_inode
->i_links_count
= cpu_to_le16(inode
->i_nlink
);
4302 EXT4_INODE_SET_XTIME(i_ctime
, inode
, raw_inode
);
4303 EXT4_INODE_SET_XTIME(i_mtime
, inode
, raw_inode
);
4304 EXT4_INODE_SET_XTIME(i_atime
, inode
, raw_inode
);
4305 EXT4_EINODE_SET_XTIME(i_crtime
, ei
, raw_inode
);
4307 if (ext4_inode_blocks_set(handle
, raw_inode
, ei
))
4309 raw_inode
->i_dtime
= cpu_to_le32(ei
->i_dtime
);
4310 /* clear the migrate flag in the raw_inode */
4311 raw_inode
->i_flags
= cpu_to_le32(ei
->i_flags
& ~EXT4_EXT_MIGRATE
);
4312 if (EXT4_SB(inode
->i_sb
)->s_es
->s_creator_os
!=
4313 cpu_to_le32(EXT4_OS_HURD
))
4314 raw_inode
->i_file_acl_high
=
4315 cpu_to_le16(ei
->i_file_acl
>> 32);
4316 raw_inode
->i_file_acl_lo
= cpu_to_le32(ei
->i_file_acl
);
4317 ext4_isize_set(raw_inode
, ei
->i_disksize
);
4318 if (ei
->i_disksize
> 0x7fffffffULL
) {
4319 struct super_block
*sb
= inode
->i_sb
;
4320 if (!EXT4_HAS_RO_COMPAT_FEATURE(sb
,
4321 EXT4_FEATURE_RO_COMPAT_LARGE_FILE
) ||
4322 EXT4_SB(sb
)->s_es
->s_rev_level
==
4323 cpu_to_le32(EXT4_GOOD_OLD_REV
)) {
4324 /* If this is the first large file
4325 * created, add a flag to the superblock.
4327 err
= ext4_journal_get_write_access(handle
,
4328 EXT4_SB(sb
)->s_sbh
);
4331 ext4_update_dynamic_rev(sb
);
4332 EXT4_SET_RO_COMPAT_FEATURE(sb
,
4333 EXT4_FEATURE_RO_COMPAT_LARGE_FILE
);
4336 err
= ext4_journal_dirty_metadata(handle
,
4337 EXT4_SB(sb
)->s_sbh
);
4340 raw_inode
->i_generation
= cpu_to_le32(inode
->i_generation
);
4341 if (S_ISCHR(inode
->i_mode
) || S_ISBLK(inode
->i_mode
)) {
4342 if (old_valid_dev(inode
->i_rdev
)) {
4343 raw_inode
->i_block
[0] =
4344 cpu_to_le32(old_encode_dev(inode
->i_rdev
));
4345 raw_inode
->i_block
[1] = 0;
4347 raw_inode
->i_block
[0] = 0;
4348 raw_inode
->i_block
[1] =
4349 cpu_to_le32(new_encode_dev(inode
->i_rdev
));
4350 raw_inode
->i_block
[2] = 0;
4352 } else for (block
= 0; block
< EXT4_N_BLOCKS
; block
++)
4353 raw_inode
->i_block
[block
] = ei
->i_data
[block
];
4355 raw_inode
->i_disk_version
= cpu_to_le32(inode
->i_version
);
4356 if (ei
->i_extra_isize
) {
4357 if (EXT4_FITS_IN_INODE(raw_inode
, ei
, i_version_hi
))
4358 raw_inode
->i_version_hi
=
4359 cpu_to_le32(inode
->i_version
>> 32);
4360 raw_inode
->i_extra_isize
= cpu_to_le16(ei
->i_extra_isize
);
4364 BUFFER_TRACE(bh
, "call ext4_journal_dirty_metadata");
4365 rc
= ext4_journal_dirty_metadata(handle
, bh
);
4368 ei
->i_state
&= ~EXT4_STATE_NEW
;
4372 ext4_std_error(inode
->i_sb
, err
);
4377 * ext4_write_inode()
4379 * We are called from a few places:
4381 * - Within generic_file_write() for O_SYNC files.
4382 * Here, there will be no transaction running. We wait for any running
4383 * trasnaction to commit.
4385 * - Within sys_sync(), kupdate and such.
4386 * We wait on commit, if tol to.
4388 * - Within prune_icache() (PF_MEMALLOC == true)
4389 * Here we simply return. We can't afford to block kswapd on the
4392 * In all cases it is actually safe for us to return without doing anything,
4393 * because the inode has been copied into a raw inode buffer in
4394 * ext4_mark_inode_dirty(). This is a correctness thing for O_SYNC and for
4397 * Note that we are absolutely dependent upon all inode dirtiers doing the
4398 * right thing: they *must* call mark_inode_dirty() after dirtying info in
4399 * which we are interested.
4401 * It would be a bug for them to not do this. The code:
4403 * mark_inode_dirty(inode)
4405 * inode->i_size = expr;
4407 * is in error because a kswapd-driven write_inode() could occur while
4408 * `stuff()' is running, and the new i_size will be lost. Plus the inode
4409 * will no longer be on the superblock's dirty inode list.
4411 int ext4_write_inode(struct inode
*inode
, int wait
)
4413 if (current
->flags
& PF_MEMALLOC
)
4416 if (ext4_journal_current_handle()) {
4417 jbd_debug(1, "called recursively, non-PF_MEMALLOC!\n");
4425 return ext4_force_commit(inode
->i_sb
);
4431 * Called from notify_change.
4433 * We want to trap VFS attempts to truncate the file as soon as
4434 * possible. In particular, we want to make sure that when the VFS
4435 * shrinks i_size, we put the inode on the orphan list and modify
4436 * i_disksize immediately, so that during the subsequent flushing of
4437 * dirty pages and freeing of disk blocks, we can guarantee that any
4438 * commit will leave the blocks being flushed in an unused state on
4439 * disk. (On recovery, the inode will get truncated and the blocks will
4440 * be freed, so we have a strong guarantee that no future commit will
4441 * leave these blocks visible to the user.)
4443 * Another thing we have to assure is that if we are in ordered mode
4444 * and inode is still attached to the committing transaction, we must
4445 * we start writeout of all the dirty pages which are being truncated.
4446 * This way we are sure that all the data written in the previous
4447 * transaction are already on disk (truncate waits for pages under
4450 * Called with inode->i_mutex down.
4452 int ext4_setattr(struct dentry
*dentry
, struct iattr
*attr
)
4454 struct inode
*inode
= dentry
->d_inode
;
4456 const unsigned int ia_valid
= attr
->ia_valid
;
4458 error
= inode_change_ok(inode
, attr
);
4462 if ((ia_valid
& ATTR_UID
&& attr
->ia_uid
!= inode
->i_uid
) ||
4463 (ia_valid
& ATTR_GID
&& attr
->ia_gid
!= inode
->i_gid
)) {
4466 /* (user+group)*(old+new) structure, inode write (sb,
4467 * inode block, ? - but truncate inode update has it) */
4468 handle
= ext4_journal_start(inode
, 2*(EXT4_QUOTA_INIT_BLOCKS(inode
->i_sb
)+
4469 EXT4_QUOTA_DEL_BLOCKS(inode
->i_sb
))+3);
4470 if (IS_ERR(handle
)) {
4471 error
= PTR_ERR(handle
);
4474 error
= DQUOT_TRANSFER(inode
, attr
) ? -EDQUOT
: 0;
4476 ext4_journal_stop(handle
);
4479 /* Update corresponding info in inode so that everything is in
4480 * one transaction */
4481 if (attr
->ia_valid
& ATTR_UID
)
4482 inode
->i_uid
= attr
->ia_uid
;
4483 if (attr
->ia_valid
& ATTR_GID
)
4484 inode
->i_gid
= attr
->ia_gid
;
4485 error
= ext4_mark_inode_dirty(handle
, inode
);
4486 ext4_journal_stop(handle
);
4489 if (attr
->ia_valid
& ATTR_SIZE
) {
4490 if (!(EXT4_I(inode
)->i_flags
& EXT4_EXTENTS_FL
)) {
4491 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
4493 if (attr
->ia_size
> sbi
->s_bitmap_maxbytes
) {
4500 if (S_ISREG(inode
->i_mode
) &&
4501 attr
->ia_valid
& ATTR_SIZE
&& attr
->ia_size
< inode
->i_size
) {
4504 handle
= ext4_journal_start(inode
, 3);
4505 if (IS_ERR(handle
)) {
4506 error
= PTR_ERR(handle
);
4510 error
= ext4_orphan_add(handle
, inode
);
4511 EXT4_I(inode
)->i_disksize
= attr
->ia_size
;
4512 rc
= ext4_mark_inode_dirty(handle
, inode
);
4515 ext4_journal_stop(handle
);
4517 if (ext4_should_order_data(inode
)) {
4518 error
= ext4_begin_ordered_truncate(inode
,
4521 /* Do as much error cleanup as possible */
4522 handle
= ext4_journal_start(inode
, 3);
4523 if (IS_ERR(handle
)) {
4524 ext4_orphan_del(NULL
, inode
);
4527 ext4_orphan_del(handle
, inode
);
4528 ext4_journal_stop(handle
);
4534 rc
= inode_setattr(inode
, attr
);
4536 /* If inode_setattr's call to ext4_truncate failed to get a
4537 * transaction handle at all, we need to clean up the in-core
4538 * orphan list manually. */
4540 ext4_orphan_del(NULL
, inode
);
4542 if (!rc
&& (ia_valid
& ATTR_MODE
))
4543 rc
= ext4_acl_chmod(inode
);
4546 ext4_std_error(inode
->i_sb
, error
);
4552 int ext4_getattr(struct vfsmount
*mnt
, struct dentry
*dentry
,
4555 struct inode
*inode
;
4556 unsigned long delalloc_blocks
;
4558 inode
= dentry
->d_inode
;
4559 generic_fillattr(inode
, stat
);
4562 * We can't update i_blocks if the block allocation is delayed
4563 * otherwise in the case of system crash before the real block
4564 * allocation is done, we will have i_blocks inconsistent with
4565 * on-disk file blocks.
4566 * We always keep i_blocks updated together with real
4567 * allocation. But to not confuse with user, stat
4568 * will return the blocks that include the delayed allocation
4569 * blocks for this file.
4571 spin_lock(&EXT4_I(inode
)->i_block_reservation_lock
);
4572 delalloc_blocks
= EXT4_I(inode
)->i_reserved_data_blocks
;
4573 spin_unlock(&EXT4_I(inode
)->i_block_reservation_lock
);
4575 stat
->blocks
+= (delalloc_blocks
<< inode
->i_sb
->s_blocksize_bits
)>>9;
4579 static int ext4_indirect_trans_blocks(struct inode
*inode
, int nrblocks
,
4584 /* if nrblocks are contiguous */
4587 * With N contiguous data blocks, it need at most
4588 * N/EXT4_ADDR_PER_BLOCK(inode->i_sb) indirect blocks
4589 * 2 dindirect blocks
4592 indirects
= nrblocks
/ EXT4_ADDR_PER_BLOCK(inode
->i_sb
);
4593 return indirects
+ 3;
4596 * if nrblocks are not contiguous, worse case, each block touch
4597 * a indirect block, and each indirect block touch a double indirect
4598 * block, plus a triple indirect block
4600 indirects
= nrblocks
* 2 + 1;
4604 static int ext4_index_trans_blocks(struct inode
*inode
, int nrblocks
, int chunk
)
4606 if (!(EXT4_I(inode
)->i_flags
& EXT4_EXTENTS_FL
))
4607 return ext4_indirect_trans_blocks(inode
, nrblocks
, 0);
4608 return ext4_ext_index_trans_blocks(inode
, nrblocks
, 0);
4611 * Account for index blocks, block groups bitmaps and block group
4612 * descriptor blocks if modify datablocks and index blocks
4613 * worse case, the indexs blocks spread over different block groups
4615 * If datablocks are discontiguous, they are possible to spread over
4616 * different block groups too. If they are contiugous, with flexbg,
4617 * they could still across block group boundary.
4619 * Also account for superblock, inode, quota and xattr blocks
4621 int ext4_meta_trans_blocks(struct inode
*inode
, int nrblocks
, int chunk
)
4623 int groups
, gdpblocks
;
4628 * How many index blocks need to touch to modify nrblocks?
4629 * The "Chunk" flag indicating whether the nrblocks is
4630 * physically contiguous on disk
4632 * For Direct IO and fallocate, they calls get_block to allocate
4633 * one single extent at a time, so they could set the "Chunk" flag
4635 idxblocks
= ext4_index_trans_blocks(inode
, nrblocks
, chunk
);
4640 * Now let's see how many group bitmaps and group descriptors need
4650 if (groups
> EXT4_SB(inode
->i_sb
)->s_groups_count
)
4651 groups
= EXT4_SB(inode
->i_sb
)->s_groups_count
;
4652 if (groups
> EXT4_SB(inode
->i_sb
)->s_gdb_count
)
4653 gdpblocks
= EXT4_SB(inode
->i_sb
)->s_gdb_count
;
4655 /* bitmaps and block group descriptor blocks */
4656 ret
+= groups
+ gdpblocks
;
4658 /* Blocks for super block, inode, quota and xattr blocks */
4659 ret
+= EXT4_META_TRANS_BLOCKS(inode
->i_sb
);
4665 * Calulate the total number of credits to reserve to fit
4666 * the modification of a single pages into a single transaction,
4667 * which may include multiple chunks of block allocations.
4669 * This could be called via ext4_write_begin()
4671 * We need to consider the worse case, when
4672 * one new block per extent.
4674 int ext4_writepage_trans_blocks(struct inode
*inode
)
4676 int bpp
= ext4_journal_blocks_per_page(inode
);
4679 ret
= ext4_meta_trans_blocks(inode
, bpp
, 0);
4681 /* Account for data blocks for journalled mode */
4682 if (ext4_should_journal_data(inode
))
4688 * Calculate the journal credits for a chunk of data modification.
4690 * This is called from DIO, fallocate or whoever calling
4691 * ext4_get_blocks_wrap() to map/allocate a chunk of contigous disk blocks.
4693 * journal buffers for data blocks are not included here, as DIO
4694 * and fallocate do no need to journal data buffers.
4696 int ext4_chunk_trans_blocks(struct inode
*inode
, int nrblocks
)
4698 return ext4_meta_trans_blocks(inode
, nrblocks
, 1);
4702 * The caller must have previously called ext4_reserve_inode_write().
4703 * Give this, we know that the caller already has write access to iloc->bh.
4705 int ext4_mark_iloc_dirty(handle_t
*handle
,
4706 struct inode
*inode
, struct ext4_iloc
*iloc
)
4710 if (test_opt(inode
->i_sb
, I_VERSION
))
4711 inode_inc_iversion(inode
);
4713 /* the do_update_inode consumes one bh->b_count */
4716 /* ext4_do_update_inode() does jbd2_journal_dirty_metadata */
4717 err
= ext4_do_update_inode(handle
, inode
, iloc
);
4723 * On success, We end up with an outstanding reference count against
4724 * iloc->bh. This _must_ be cleaned up later.
4728 ext4_reserve_inode_write(handle_t
*handle
, struct inode
*inode
,
4729 struct ext4_iloc
*iloc
)
4733 err
= ext4_get_inode_loc(inode
, iloc
);
4735 BUFFER_TRACE(iloc
->bh
, "get_write_access");
4736 err
= ext4_journal_get_write_access(handle
, iloc
->bh
);
4743 ext4_std_error(inode
->i_sb
, err
);
4748 * Expand an inode by new_extra_isize bytes.
4749 * Returns 0 on success or negative error number on failure.
4751 static int ext4_expand_extra_isize(struct inode
*inode
,
4752 unsigned int new_extra_isize
,
4753 struct ext4_iloc iloc
,
4756 struct ext4_inode
*raw_inode
;
4757 struct ext4_xattr_ibody_header
*header
;
4758 struct ext4_xattr_entry
*entry
;
4760 if (EXT4_I(inode
)->i_extra_isize
>= new_extra_isize
)
4763 raw_inode
= ext4_raw_inode(&iloc
);
4765 header
= IHDR(inode
, raw_inode
);
4766 entry
= IFIRST(header
);
4768 /* No extended attributes present */
4769 if (!(EXT4_I(inode
)->i_state
& EXT4_STATE_XATTR
) ||
4770 header
->h_magic
!= cpu_to_le32(EXT4_XATTR_MAGIC
)) {
4771 memset((void *)raw_inode
+ EXT4_GOOD_OLD_INODE_SIZE
, 0,
4773 EXT4_I(inode
)->i_extra_isize
= new_extra_isize
;
4777 /* try to expand with EAs present */
4778 return ext4_expand_extra_isize_ea(inode
, new_extra_isize
,
4783 * What we do here is to mark the in-core inode as clean with respect to inode
4784 * dirtiness (it may still be data-dirty).
4785 * This means that the in-core inode may be reaped by prune_icache
4786 * without having to perform any I/O. This is a very good thing,
4787 * because *any* task may call prune_icache - even ones which
4788 * have a transaction open against a different journal.
4790 * Is this cheating? Not really. Sure, we haven't written the
4791 * inode out, but prune_icache isn't a user-visible syncing function.
4792 * Whenever the user wants stuff synced (sys_sync, sys_msync, sys_fsync)
4793 * we start and wait on commits.
4795 * Is this efficient/effective? Well, we're being nice to the system
4796 * by cleaning up our inodes proactively so they can be reaped
4797 * without I/O. But we are potentially leaving up to five seconds'
4798 * worth of inodes floating about which prune_icache wants us to
4799 * write out. One way to fix that would be to get prune_icache()
4800 * to do a write_super() to free up some memory. It has the desired
4803 int ext4_mark_inode_dirty(handle_t
*handle
, struct inode
*inode
)
4805 struct ext4_iloc iloc
;
4806 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
4807 static unsigned int mnt_count
;
4811 err
= ext4_reserve_inode_write(handle
, inode
, &iloc
);
4812 if (EXT4_I(inode
)->i_extra_isize
< sbi
->s_want_extra_isize
&&
4813 !(EXT4_I(inode
)->i_state
& EXT4_STATE_NO_EXPAND
)) {
4815 * We need extra buffer credits since we may write into EA block
4816 * with this same handle. If journal_extend fails, then it will
4817 * only result in a minor loss of functionality for that inode.
4818 * If this is felt to be critical, then e2fsck should be run to
4819 * force a large enough s_min_extra_isize.
4821 if ((jbd2_journal_extend(handle
,
4822 EXT4_DATA_TRANS_BLOCKS(inode
->i_sb
))) == 0) {
4823 ret
= ext4_expand_extra_isize(inode
,
4824 sbi
->s_want_extra_isize
,
4827 EXT4_I(inode
)->i_state
|= EXT4_STATE_NO_EXPAND
;
4829 le16_to_cpu(sbi
->s_es
->s_mnt_count
)) {
4830 ext4_warning(inode
->i_sb
, __func__
,
4831 "Unable to expand inode %lu. Delete"
4832 " some EAs or run e2fsck.",
4835 le16_to_cpu(sbi
->s_es
->s_mnt_count
);
4841 err
= ext4_mark_iloc_dirty(handle
, inode
, &iloc
);
4846 * ext4_dirty_inode() is called from __mark_inode_dirty()
4848 * We're really interested in the case where a file is being extended.
4849 * i_size has been changed by generic_commit_write() and we thus need
4850 * to include the updated inode in the current transaction.
4852 * Also, DQUOT_ALLOC_SPACE() will always dirty the inode when blocks
4853 * are allocated to the file.
4855 * If the inode is marked synchronous, we don't honour that here - doing
4856 * so would cause a commit on atime updates, which we don't bother doing.
4857 * We handle synchronous inodes at the highest possible level.
4859 void ext4_dirty_inode(struct inode
*inode
)
4861 handle_t
*current_handle
= ext4_journal_current_handle();
4864 handle
= ext4_journal_start(inode
, 2);
4867 if (current_handle
&&
4868 current_handle
->h_transaction
!= handle
->h_transaction
) {
4869 /* This task has a transaction open against a different fs */
4870 printk(KERN_EMERG
"%s: transactions do not match!\n",
4873 jbd_debug(5, "marking dirty. outer handle=%p\n",
4875 ext4_mark_inode_dirty(handle
, inode
);
4877 ext4_journal_stop(handle
);
4884 * Bind an inode's backing buffer_head into this transaction, to prevent
4885 * it from being flushed to disk early. Unlike
4886 * ext4_reserve_inode_write, this leaves behind no bh reference and
4887 * returns no iloc structure, so the caller needs to repeat the iloc
4888 * lookup to mark the inode dirty later.
4890 static int ext4_pin_inode(handle_t
*handle
, struct inode
*inode
)
4892 struct ext4_iloc iloc
;
4896 err
= ext4_get_inode_loc(inode
, &iloc
);
4898 BUFFER_TRACE(iloc
.bh
, "get_write_access");
4899 err
= jbd2_journal_get_write_access(handle
, iloc
.bh
);
4901 err
= ext4_journal_dirty_metadata(handle
,
4906 ext4_std_error(inode
->i_sb
, err
);
4911 int ext4_change_inode_journal_flag(struct inode
*inode
, int val
)
4918 * We have to be very careful here: changing a data block's
4919 * journaling status dynamically is dangerous. If we write a
4920 * data block to the journal, change the status and then delete
4921 * that block, we risk forgetting to revoke the old log record
4922 * from the journal and so a subsequent replay can corrupt data.
4923 * So, first we make sure that the journal is empty and that
4924 * nobody is changing anything.
4927 journal
= EXT4_JOURNAL(inode
);
4928 if (is_journal_aborted(journal
))
4931 jbd2_journal_lock_updates(journal
);
4932 jbd2_journal_flush(journal
);
4935 * OK, there are no updates running now, and all cached data is
4936 * synced to disk. We are now in a completely consistent state
4937 * which doesn't have anything in the journal, and we know that
4938 * no filesystem updates are running, so it is safe to modify
4939 * the inode's in-core data-journaling state flag now.
4943 EXT4_I(inode
)->i_flags
|= EXT4_JOURNAL_DATA_FL
;
4945 EXT4_I(inode
)->i_flags
&= ~EXT4_JOURNAL_DATA_FL
;
4946 ext4_set_aops(inode
);
4948 jbd2_journal_unlock_updates(journal
);
4950 /* Finally we can mark the inode as dirty. */
4952 handle
= ext4_journal_start(inode
, 1);
4954 return PTR_ERR(handle
);
4956 err
= ext4_mark_inode_dirty(handle
, inode
);
4958 ext4_journal_stop(handle
);
4959 ext4_std_error(inode
->i_sb
, err
);
4964 static int ext4_bh_unmapped(handle_t
*handle
, struct buffer_head
*bh
)
4966 return !buffer_mapped(bh
);
4969 int ext4_page_mkwrite(struct vm_area_struct
*vma
, struct page
*page
)
4975 struct file
*file
= vma
->vm_file
;
4976 struct inode
*inode
= file
->f_path
.dentry
->d_inode
;
4977 struct address_space
*mapping
= inode
->i_mapping
;
4980 * Get i_alloc_sem to stop truncates messing with the inode. We cannot
4981 * get i_mutex because we are already holding mmap_sem.
4983 down_read(&inode
->i_alloc_sem
);
4984 size
= i_size_read(inode
);
4985 if (page
->mapping
!= mapping
|| size
<= page_offset(page
)
4986 || !PageUptodate(page
)) {
4987 /* page got truncated from under us? */
4991 if (PageMappedToDisk(page
))
4994 if (page
->index
== size
>> PAGE_CACHE_SHIFT
)
4995 len
= size
& ~PAGE_CACHE_MASK
;
4997 len
= PAGE_CACHE_SIZE
;
4999 if (page_has_buffers(page
)) {
5000 /* return if we have all the buffers mapped */
5001 if (!walk_page_buffers(NULL
, page_buffers(page
), 0, len
, NULL
,
5006 * OK, we need to fill the hole... Do write_begin write_end
5007 * to do block allocation/reservation.We are not holding
5008 * inode.i__mutex here. That allow * parallel write_begin,
5009 * write_end call. lock_page prevent this from happening
5010 * on the same page though
5012 ret
= mapping
->a_ops
->write_begin(file
, mapping
, page_offset(page
),
5013 len
, AOP_FLAG_UNINTERRUPTIBLE
, &page
, &fsdata
);
5016 ret
= mapping
->a_ops
->write_end(file
, mapping
, page_offset(page
),
5017 len
, len
, page
, fsdata
);
5022 up_read(&inode
->i_alloc_sem
);