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(
50 EXT4_SB(inode
->i_sb
)->s_journal
,
51 &EXT4_I(inode
)->jinode
,
55 static void ext4_invalidatepage(struct page
*page
, unsigned long offset
);
58 * Test whether an inode is a fast symlink.
60 static int ext4_inode_is_fast_symlink(struct inode
*inode
)
62 int ea_blocks
= EXT4_I(inode
)->i_file_acl
?
63 (inode
->i_sb
->s_blocksize
>> 9) : 0;
65 return (S_ISLNK(inode
->i_mode
) && inode
->i_blocks
- ea_blocks
== 0);
69 * The ext4 forget function must perform a revoke if we are freeing data
70 * which has been journaled. Metadata (eg. indirect blocks) must be
71 * revoked in all cases.
73 * "bh" may be NULL: a metadata block may have been freed from memory
74 * but there may still be a record of it in the journal, and that record
75 * still needs to be revoked.
77 int ext4_forget(handle_t
*handle
, int is_metadata
, struct inode
*inode
,
78 struct buffer_head
*bh
, ext4_fsblk_t blocknr
)
84 BUFFER_TRACE(bh
, "enter");
86 jbd_debug(4, "forgetting bh %p: is_metadata = %d, mode %o, "
88 bh
, is_metadata
, inode
->i_mode
,
89 test_opt(inode
->i_sb
, DATA_FLAGS
));
91 /* Never use the revoke function if we are doing full data
92 * journaling: there is no need to, and a V1 superblock won't
93 * support it. Otherwise, only skip the revoke on un-journaled
96 if (test_opt(inode
->i_sb
, DATA_FLAGS
) == EXT4_MOUNT_JOURNAL_DATA
||
97 (!is_metadata
&& !ext4_should_journal_data(inode
))) {
99 BUFFER_TRACE(bh
, "call jbd2_journal_forget");
100 return ext4_journal_forget(handle
, bh
);
106 * data!=journal && (is_metadata || should_journal_data(inode))
108 BUFFER_TRACE(bh
, "call ext4_journal_revoke");
109 err
= ext4_journal_revoke(handle
, blocknr
, bh
);
111 ext4_abort(inode
->i_sb
, __func__
,
112 "error %d when attempting revoke", err
);
113 BUFFER_TRACE(bh
, "exit");
118 * Work out how many blocks we need to proceed with the next chunk of a
119 * truncate transaction.
121 static unsigned long blocks_for_truncate(struct inode
*inode
)
125 needed
= inode
->i_blocks
>> (inode
->i_sb
->s_blocksize_bits
- 9);
127 /* Give ourselves just enough room to cope with inodes in which
128 * i_blocks is corrupt: we've seen disk corruptions in the past
129 * which resulted in random data in an inode which looked enough
130 * like a regular file for ext4 to try to delete it. Things
131 * will go a bit crazy if that happens, but at least we should
132 * try not to panic the whole kernel. */
136 /* But we need to bound the transaction so we don't overflow the
138 if (needed
> EXT4_MAX_TRANS_DATA
)
139 needed
= EXT4_MAX_TRANS_DATA
;
141 return EXT4_DATA_TRANS_BLOCKS(inode
->i_sb
) + needed
;
145 * Truncate transactions can be complex and absolutely huge. So we need to
146 * be able to restart the transaction at a conventient checkpoint to make
147 * sure we don't overflow the journal.
149 * start_transaction gets us a new handle for a truncate transaction,
150 * and extend_transaction tries to extend the existing one a bit. If
151 * extend fails, we need to propagate the failure up and restart the
152 * transaction in the top-level truncate loop. --sct
154 static handle_t
*start_transaction(struct inode
*inode
)
158 result
= ext4_journal_start(inode
, blocks_for_truncate(inode
));
162 ext4_std_error(inode
->i_sb
, PTR_ERR(result
));
167 * Try to extend this transaction for the purposes of truncation.
169 * Returns 0 if we managed to create more room. If we can't create more
170 * room, and the transaction must be restarted we return 1.
172 static int try_to_extend_transaction(handle_t
*handle
, struct inode
*inode
)
174 if (handle
->h_buffer_credits
> EXT4_RESERVE_TRANS_BLOCKS
)
176 if (!ext4_journal_extend(handle
, blocks_for_truncate(inode
)))
182 * Restart the transaction associated with *handle. This does a commit,
183 * so before we call here everything must be consistently dirtied against
186 static int ext4_journal_test_restart(handle_t
*handle
, struct inode
*inode
)
188 jbd_debug(2, "restarting handle %p\n", handle
);
189 return ext4_journal_restart(handle
, blocks_for_truncate(inode
));
193 * Called at the last iput() if i_nlink is zero.
195 void ext4_delete_inode (struct inode
* inode
)
200 if (ext4_should_order_data(inode
))
201 ext4_begin_ordered_truncate(inode
, 0);
202 truncate_inode_pages(&inode
->i_data
, 0);
204 if (is_bad_inode(inode
))
207 handle
= ext4_journal_start(inode
, blocks_for_truncate(inode
)+3);
208 if (IS_ERR(handle
)) {
209 ext4_std_error(inode
->i_sb
, PTR_ERR(handle
));
211 * If we're going to skip the normal cleanup, we still need to
212 * make sure that the in-core orphan linked list is properly
215 ext4_orphan_del(NULL
, inode
);
222 err
= ext4_mark_inode_dirty(handle
, inode
);
224 ext4_warning(inode
->i_sb
, __func__
,
225 "couldn't mark inode dirty (err %d)", err
);
229 ext4_truncate(inode
);
232 * ext4_ext_truncate() doesn't reserve any slop when it
233 * restarts journal transactions; therefore there may not be
234 * enough credits left in the handle to remove the inode from
235 * the orphan list and set the dtime field.
237 if (handle
->h_buffer_credits
< 3) {
238 err
= ext4_journal_extend(handle
, 3);
240 err
= ext4_journal_restart(handle
, 3);
242 ext4_warning(inode
->i_sb
, __func__
,
243 "couldn't extend journal (err %d)", err
);
245 ext4_journal_stop(handle
);
251 * Kill off the orphan record which ext4_truncate created.
252 * AKPM: I think this can be inside the above `if'.
253 * Note that ext4_orphan_del() has to be able to cope with the
254 * deletion of a non-existent orphan - this is because we don't
255 * know if ext4_truncate() actually created an orphan record.
256 * (Well, we could do this if we need to, but heck - it works)
258 ext4_orphan_del(handle
, inode
);
259 EXT4_I(inode
)->i_dtime
= get_seconds();
262 * One subtle ordering requirement: if anything has gone wrong
263 * (transaction abort, IO errors, whatever), then we can still
264 * do these next steps (the fs will already have been marked as
265 * having errors), but we can't free the inode if the mark_dirty
268 if (ext4_mark_inode_dirty(handle
, inode
))
269 /* If that failed, just do the required in-core inode clear. */
272 ext4_free_inode(handle
, inode
);
273 ext4_journal_stop(handle
);
276 clear_inode(inode
); /* We must guarantee clearing of inode... */
282 struct buffer_head
*bh
;
285 static inline void add_chain(Indirect
*p
, struct buffer_head
*bh
, __le32
*v
)
287 p
->key
= *(p
->p
= v
);
292 * ext4_block_to_path - parse the block number into array of offsets
293 * @inode: inode in question (we are only interested in its superblock)
294 * @i_block: block number to be parsed
295 * @offsets: array to store the offsets in
296 * @boundary: set this non-zero if the referred-to block is likely to be
297 * followed (on disk) by an indirect block.
299 * To store the locations of file's data ext4 uses a data structure common
300 * for UNIX filesystems - tree of pointers anchored in the inode, with
301 * data blocks at leaves and indirect blocks in intermediate nodes.
302 * This function translates the block number into path in that tree -
303 * return value is the path length and @offsets[n] is the offset of
304 * pointer to (n+1)th node in the nth one. If @block is out of range
305 * (negative or too large) warning is printed and zero returned.
307 * Note: function doesn't find node addresses, so no IO is needed. All
308 * we need to know is the capacity of indirect blocks (taken from the
313 * Portability note: the last comparison (check that we fit into triple
314 * indirect block) is spelled differently, because otherwise on an
315 * architecture with 32-bit longs and 8Kb pages we might get into trouble
316 * if our filesystem had 8Kb blocks. We might use long long, but that would
317 * kill us on x86. Oh, well, at least the sign propagation does not matter -
318 * i_block would have to be negative in the very beginning, so we would not
322 static int ext4_block_to_path(struct inode
*inode
,
324 ext4_lblk_t offsets
[4], int *boundary
)
326 int ptrs
= EXT4_ADDR_PER_BLOCK(inode
->i_sb
);
327 int ptrs_bits
= EXT4_ADDR_PER_BLOCK_BITS(inode
->i_sb
);
328 const long direct_blocks
= EXT4_NDIR_BLOCKS
,
329 indirect_blocks
= ptrs
,
330 double_blocks
= (1 << (ptrs_bits
* 2));
335 ext4_warning (inode
->i_sb
, "ext4_block_to_path", "block < 0");
336 } else if (i_block
< direct_blocks
) {
337 offsets
[n
++] = i_block
;
338 final
= direct_blocks
;
339 } else if ( (i_block
-= direct_blocks
) < indirect_blocks
) {
340 offsets
[n
++] = EXT4_IND_BLOCK
;
341 offsets
[n
++] = i_block
;
343 } else if ((i_block
-= indirect_blocks
) < double_blocks
) {
344 offsets
[n
++] = EXT4_DIND_BLOCK
;
345 offsets
[n
++] = i_block
>> ptrs_bits
;
346 offsets
[n
++] = i_block
& (ptrs
- 1);
348 } else if (((i_block
-= double_blocks
) >> (ptrs_bits
* 2)) < ptrs
) {
349 offsets
[n
++] = EXT4_TIND_BLOCK
;
350 offsets
[n
++] = i_block
>> (ptrs_bits
* 2);
351 offsets
[n
++] = (i_block
>> ptrs_bits
) & (ptrs
- 1);
352 offsets
[n
++] = i_block
& (ptrs
- 1);
355 ext4_warning(inode
->i_sb
, "ext4_block_to_path",
356 "block %lu > max in inode %lu",
357 i_block
+ direct_blocks
+
358 indirect_blocks
+ double_blocks
, inode
->i_ino
);
361 *boundary
= final
- 1 - (i_block
& (ptrs
- 1));
366 * ext4_get_branch - read the chain of indirect blocks leading to data
367 * @inode: inode in question
368 * @depth: depth of the chain (1 - direct pointer, etc.)
369 * @offsets: offsets of pointers in inode/indirect blocks
370 * @chain: place to store the result
371 * @err: here we store the error value
373 * Function fills the array of triples <key, p, bh> and returns %NULL
374 * if everything went OK or the pointer to the last filled triple
375 * (incomplete one) otherwise. Upon the return chain[i].key contains
376 * the number of (i+1)-th block in the chain (as it is stored in memory,
377 * i.e. little-endian 32-bit), chain[i].p contains the address of that
378 * number (it points into struct inode for i==0 and into the bh->b_data
379 * for i>0) and chain[i].bh points to the buffer_head of i-th indirect
380 * block for i>0 and NULL for i==0. In other words, it holds the block
381 * numbers of the chain, addresses they were taken from (and where we can
382 * verify that chain did not change) and buffer_heads hosting these
385 * Function stops when it stumbles upon zero pointer (absent block)
386 * (pointer to last triple returned, *@err == 0)
387 * or when it gets an IO error reading an indirect block
388 * (ditto, *@err == -EIO)
389 * or when it reads all @depth-1 indirect blocks successfully and finds
390 * the whole chain, all way to the data (returns %NULL, *err == 0).
392 * Need to be called with
393 * down_read(&EXT4_I(inode)->i_data_sem)
395 static Indirect
*ext4_get_branch(struct inode
*inode
, int depth
,
396 ext4_lblk_t
*offsets
,
397 Indirect chain
[4], int *err
)
399 struct super_block
*sb
= inode
->i_sb
;
401 struct buffer_head
*bh
;
404 /* i_data is not going away, no lock needed */
405 add_chain (chain
, NULL
, EXT4_I(inode
)->i_data
+ *offsets
);
409 bh
= sb_bread(sb
, le32_to_cpu(p
->key
));
412 add_chain(++p
, bh
, (__le32
*)bh
->b_data
+ *++offsets
);
426 * ext4_find_near - find a place for allocation with sufficient locality
428 * @ind: descriptor of indirect block.
430 * This function returns the preferred place for block allocation.
431 * It is used when heuristic for sequential allocation fails.
433 * + if there is a block to the left of our position - allocate near it.
434 * + if pointer will live in indirect block - allocate near that block.
435 * + if pointer will live in inode - allocate in the same
438 * In the latter case we colour the starting block by the callers PID to
439 * prevent it from clashing with concurrent allocations for a different inode
440 * in the same block group. The PID is used here so that functionally related
441 * files will be close-by on-disk.
443 * Caller must make sure that @ind is valid and will stay that way.
445 static ext4_fsblk_t
ext4_find_near(struct inode
*inode
, Indirect
*ind
)
447 struct ext4_inode_info
*ei
= EXT4_I(inode
);
448 __le32
*start
= ind
->bh
? (__le32
*) ind
->bh
->b_data
: ei
->i_data
;
450 ext4_fsblk_t bg_start
;
451 ext4_fsblk_t last_block
;
452 ext4_grpblk_t colour
;
454 /* Try to find previous block */
455 for (p
= ind
->p
- 1; p
>= start
; p
--) {
457 return le32_to_cpu(*p
);
460 /* No such thing, so let's try location of indirect block */
462 return ind
->bh
->b_blocknr
;
465 * It is going to be referred to from the inode itself? OK, just put it
466 * into the same cylinder group then.
468 bg_start
= ext4_group_first_block_no(inode
->i_sb
, ei
->i_block_group
);
469 last_block
= ext4_blocks_count(EXT4_SB(inode
->i_sb
)->s_es
) - 1;
471 if (bg_start
+ EXT4_BLOCKS_PER_GROUP(inode
->i_sb
) <= last_block
)
472 colour
= (current
->pid
% 16) *
473 (EXT4_BLOCKS_PER_GROUP(inode
->i_sb
) / 16);
475 colour
= (current
->pid
% 16) * ((last_block
- bg_start
) / 16);
476 return bg_start
+ colour
;
480 * ext4_find_goal - find a preferred place for allocation.
482 * @block: block we want
483 * @partial: pointer to the last triple within a chain
485 * Normally this function find the preferred place for block allocation,
488 static ext4_fsblk_t
ext4_find_goal(struct inode
*inode
, ext4_lblk_t block
,
491 struct ext4_block_alloc_info
*block_i
;
493 block_i
= EXT4_I(inode
)->i_block_alloc_info
;
496 * try the heuristic for sequential allocation,
497 * failing that at least try to get decent locality.
499 if (block_i
&& (block
== block_i
->last_alloc_logical_block
+ 1)
500 && (block_i
->last_alloc_physical_block
!= 0)) {
501 return block_i
->last_alloc_physical_block
+ 1;
504 return ext4_find_near(inode
, partial
);
508 * ext4_blks_to_allocate: Look up the block map and count the number
509 * of direct blocks need to be allocated for the given branch.
511 * @branch: chain of indirect blocks
512 * @k: number of blocks need for indirect blocks
513 * @blks: number of data blocks to be mapped.
514 * @blocks_to_boundary: the offset in the indirect block
516 * return the total number of blocks to be allocate, including the
517 * direct and indirect blocks.
519 static int ext4_blks_to_allocate(Indirect
*branch
, int k
, unsigned long blks
,
520 int blocks_to_boundary
)
522 unsigned long count
= 0;
525 * Simple case, [t,d]Indirect block(s) has not allocated yet
526 * then it's clear blocks on that path have not allocated
529 /* right now we don't handle cross boundary allocation */
530 if (blks
< blocks_to_boundary
+ 1)
533 count
+= blocks_to_boundary
+ 1;
538 while (count
< blks
&& count
<= blocks_to_boundary
&&
539 le32_to_cpu(*(branch
[0].p
+ count
)) == 0) {
546 * ext4_alloc_blocks: multiple allocate blocks needed for a branch
547 * @indirect_blks: the number of blocks need to allocate for indirect
550 * @new_blocks: on return it will store the new block numbers for
551 * the indirect blocks(if needed) and the first direct block,
552 * @blks: on return it will store the total number of allocated
555 static int ext4_alloc_blocks(handle_t
*handle
, struct inode
*inode
,
556 ext4_lblk_t iblock
, ext4_fsblk_t goal
,
557 int indirect_blks
, int blks
,
558 ext4_fsblk_t new_blocks
[4], int *err
)
561 unsigned long count
= 0, blk_allocated
= 0;
563 ext4_fsblk_t current_block
= 0;
567 * Here we try to allocate the requested multiple blocks at once,
568 * on a best-effort basis.
569 * To build a branch, we should allocate blocks for
570 * the indirect blocks(if not allocated yet), and at least
571 * the first direct block of this branch. That's the
572 * minimum number of blocks need to allocate(required)
574 /* first we try to allocate the indirect blocks */
575 target
= indirect_blks
;
578 /* allocating blocks for indirect blocks and direct blocks */
579 current_block
= ext4_new_meta_blocks(handle
, inode
,
585 /* allocate blocks for indirect blocks */
586 while (index
< indirect_blks
&& count
) {
587 new_blocks
[index
++] = current_block
++;
592 * save the new block number
593 * for the first direct block
595 new_blocks
[index
] = current_block
;
596 printk(KERN_INFO
"%s returned more blocks than "
597 "requested\n", __func__
);
603 target
= blks
- count
;
604 blk_allocated
= count
;
607 /* Now allocate data blocks */
609 /* allocating blocks for data blocks */
610 current_block
= ext4_new_blocks(handle
, inode
, iblock
,
612 if (*err
&& (target
== blks
)) {
614 * if the allocation failed and we didn't allocate
620 if (target
== blks
) {
622 * save the new block number
623 * for the first direct block
625 new_blocks
[index
] = current_block
;
627 blk_allocated
+= count
;
630 /* total number of blocks allocated for direct blocks */
635 for (i
= 0; i
<index
; i
++)
636 ext4_free_blocks(handle
, inode
, new_blocks
[i
], 1, 0);
641 * ext4_alloc_branch - allocate and set up a chain of blocks.
643 * @indirect_blks: number of allocated indirect blocks
644 * @blks: number of allocated direct blocks
645 * @offsets: offsets (in the blocks) to store the pointers to next.
646 * @branch: place to store the chain in.
648 * This function allocates blocks, zeroes out all but the last one,
649 * links them into chain and (if we are synchronous) writes them to disk.
650 * In other words, it prepares a branch that can be spliced onto the
651 * inode. It stores the information about that chain in the branch[], in
652 * the same format as ext4_get_branch() would do. We are calling it after
653 * we had read the existing part of chain and partial points to the last
654 * triple of that (one with zero ->key). Upon the exit we have the same
655 * picture as after the successful ext4_get_block(), except that in one
656 * place chain is disconnected - *branch->p is still zero (we did not
657 * set the last link), but branch->key contains the number that should
658 * be placed into *branch->p to fill that gap.
660 * If allocation fails we free all blocks we've allocated (and forget
661 * their buffer_heads) and return the error value the from failed
662 * ext4_alloc_block() (normally -ENOSPC). Otherwise we set the chain
663 * as described above and return 0.
665 static int ext4_alloc_branch(handle_t
*handle
, struct inode
*inode
,
666 ext4_lblk_t iblock
, int indirect_blks
,
667 int *blks
, ext4_fsblk_t goal
,
668 ext4_lblk_t
*offsets
, Indirect
*branch
)
670 int blocksize
= inode
->i_sb
->s_blocksize
;
673 struct buffer_head
*bh
;
675 ext4_fsblk_t new_blocks
[4];
676 ext4_fsblk_t current_block
;
678 num
= ext4_alloc_blocks(handle
, inode
, iblock
, goal
, indirect_blks
,
679 *blks
, new_blocks
, &err
);
683 branch
[0].key
= cpu_to_le32(new_blocks
[0]);
685 * metadata blocks and data blocks are allocated.
687 for (n
= 1; n
<= indirect_blks
; n
++) {
689 * Get buffer_head for parent block, zero it out
690 * and set the pointer to new one, then send
693 bh
= sb_getblk(inode
->i_sb
, new_blocks
[n
-1]);
696 BUFFER_TRACE(bh
, "call get_create_access");
697 err
= ext4_journal_get_create_access(handle
, bh
);
704 memset(bh
->b_data
, 0, blocksize
);
705 branch
[n
].p
= (__le32
*) bh
->b_data
+ offsets
[n
];
706 branch
[n
].key
= cpu_to_le32(new_blocks
[n
]);
707 *branch
[n
].p
= branch
[n
].key
;
708 if ( n
== indirect_blks
) {
709 current_block
= new_blocks
[n
];
711 * End of chain, update the last new metablock of
712 * the chain to point to the new allocated
713 * data blocks numbers
715 for (i
=1; i
< num
; i
++)
716 *(branch
[n
].p
+ i
) = cpu_to_le32(++current_block
);
718 BUFFER_TRACE(bh
, "marking uptodate");
719 set_buffer_uptodate(bh
);
722 BUFFER_TRACE(bh
, "call ext4_journal_dirty_metadata");
723 err
= ext4_journal_dirty_metadata(handle
, bh
);
730 /* Allocation failed, free what we already allocated */
731 for (i
= 1; i
<= n
; i
++) {
732 BUFFER_TRACE(branch
[i
].bh
, "call jbd2_journal_forget");
733 ext4_journal_forget(handle
, branch
[i
].bh
);
735 for (i
= 0; i
<indirect_blks
; i
++)
736 ext4_free_blocks(handle
, inode
, new_blocks
[i
], 1, 0);
738 ext4_free_blocks(handle
, inode
, new_blocks
[i
], num
, 0);
744 * ext4_splice_branch - splice the allocated branch onto inode.
746 * @block: (logical) number of block we are adding
747 * @chain: chain of indirect blocks (with a missing link - see
749 * @where: location of missing link
750 * @num: number of indirect blocks we are adding
751 * @blks: number of direct blocks we are adding
753 * This function fills the missing link and does all housekeeping needed in
754 * inode (->i_blocks, etc.). In case of success we end up with the full
755 * chain to new block and return 0.
757 static int ext4_splice_branch(handle_t
*handle
, struct inode
*inode
,
758 ext4_lblk_t block
, Indirect
*where
, int num
, int blks
)
762 struct ext4_block_alloc_info
*block_i
;
763 ext4_fsblk_t current_block
;
765 block_i
= EXT4_I(inode
)->i_block_alloc_info
;
767 * If we're splicing into a [td]indirect block (as opposed to the
768 * inode) then we need to get write access to the [td]indirect block
772 BUFFER_TRACE(where
->bh
, "get_write_access");
773 err
= ext4_journal_get_write_access(handle
, where
->bh
);
779 *where
->p
= where
->key
;
782 * Update the host buffer_head or inode to point to more just allocated
783 * direct blocks blocks
785 if (num
== 0 && blks
> 1) {
786 current_block
= le32_to_cpu(where
->key
) + 1;
787 for (i
= 1; i
< blks
; i
++)
788 *(where
->p
+ i
) = cpu_to_le32(current_block
++);
792 * update the most recently allocated logical & physical block
793 * in i_block_alloc_info, to assist find the proper goal block for next
797 block_i
->last_alloc_logical_block
= block
+ blks
- 1;
798 block_i
->last_alloc_physical_block
=
799 le32_to_cpu(where
[num
].key
) + blks
- 1;
802 /* We are done with atomic stuff, now do the rest of housekeeping */
804 inode
->i_ctime
= ext4_current_time(inode
);
805 ext4_mark_inode_dirty(handle
, inode
);
807 /* had we spliced it onto indirect block? */
810 * If we spliced it onto an indirect block, we haven't
811 * altered the inode. Note however that if it is being spliced
812 * onto an indirect block at the very end of the file (the
813 * file is growing) then we *will* alter the inode to reflect
814 * the new i_size. But that is not done here - it is done in
815 * generic_commit_write->__mark_inode_dirty->ext4_dirty_inode.
817 jbd_debug(5, "splicing indirect only\n");
818 BUFFER_TRACE(where
->bh
, "call ext4_journal_dirty_metadata");
819 err
= ext4_journal_dirty_metadata(handle
, where
->bh
);
824 * OK, we spliced it into the inode itself on a direct block.
825 * Inode was dirtied above.
827 jbd_debug(5, "splicing direct\n");
832 for (i
= 1; i
<= num
; i
++) {
833 BUFFER_TRACE(where
[i
].bh
, "call jbd2_journal_forget");
834 ext4_journal_forget(handle
, where
[i
].bh
);
835 ext4_free_blocks(handle
, inode
,
836 le32_to_cpu(where
[i
-1].key
), 1, 0);
838 ext4_free_blocks(handle
, inode
, le32_to_cpu(where
[num
].key
), blks
, 0);
844 * Allocation strategy is simple: if we have to allocate something, we will
845 * have to go the whole way to leaf. So let's do it before attaching anything
846 * to tree, set linkage between the newborn blocks, write them if sync is
847 * required, recheck the path, free and repeat if check fails, otherwise
848 * set the last missing link (that will protect us from any truncate-generated
849 * removals - all blocks on the path are immune now) and possibly force the
850 * write on the parent block.
851 * That has a nice additional property: no special recovery from the failed
852 * allocations is needed - we simply release blocks and do not touch anything
853 * reachable from inode.
855 * `handle' can be NULL if create == 0.
857 * return > 0, # of blocks mapped or allocated.
858 * return = 0, if plain lookup failed.
859 * return < 0, error case.
862 * Need to be called with
863 * down_read(&EXT4_I(inode)->i_data_sem) if not allocating file system block
864 * (ie, create is zero). Otherwise down_write(&EXT4_I(inode)->i_data_sem)
866 int ext4_get_blocks_handle(handle_t
*handle
, struct inode
*inode
,
867 ext4_lblk_t iblock
, unsigned long maxblocks
,
868 struct buffer_head
*bh_result
,
869 int create
, int extend_disksize
)
872 ext4_lblk_t offsets
[4];
877 int blocks_to_boundary
= 0;
879 struct ext4_inode_info
*ei
= EXT4_I(inode
);
881 ext4_fsblk_t first_block
= 0;
885 J_ASSERT(!(EXT4_I(inode
)->i_flags
& EXT4_EXTENTS_FL
));
886 J_ASSERT(handle
!= NULL
|| create
== 0);
887 depth
= ext4_block_to_path(inode
, iblock
, offsets
,
888 &blocks_to_boundary
);
893 partial
= ext4_get_branch(inode
, depth
, offsets
, chain
, &err
);
895 /* Simplest case - block found, no allocation needed */
897 first_block
= le32_to_cpu(chain
[depth
- 1].key
);
898 clear_buffer_new(bh_result
);
901 while (count
< maxblocks
&& count
<= blocks_to_boundary
) {
904 blk
= le32_to_cpu(*(chain
[depth
-1].p
+ count
));
906 if (blk
== first_block
+ count
)
914 /* Next simple case - plain lookup or failed read of indirect block */
915 if (!create
|| err
== -EIO
)
919 * Okay, we need to do block allocation. Lazily initialize the block
920 * allocation info here if necessary
922 if (S_ISREG(inode
->i_mode
) && (!ei
->i_block_alloc_info
))
923 ext4_init_block_alloc_info(inode
);
925 goal
= ext4_find_goal(inode
, iblock
, partial
);
927 /* the number of blocks need to allocate for [d,t]indirect blocks */
928 indirect_blks
= (chain
+ depth
) - partial
- 1;
931 * Next look up the indirect map to count the totoal number of
932 * direct blocks to allocate for this branch.
934 count
= ext4_blks_to_allocate(partial
, indirect_blks
,
935 maxblocks
, blocks_to_boundary
);
937 * Block out ext4_truncate while we alter the tree
939 err
= ext4_alloc_branch(handle
, inode
, iblock
, indirect_blks
,
941 offsets
+ (partial
- chain
), partial
);
944 * The ext4_splice_branch call will free and forget any buffers
945 * on the new chain if there is a failure, but that risks using
946 * up transaction credits, especially for bitmaps where the
947 * credits cannot be returned. Can we handle this somehow? We
948 * may need to return -EAGAIN upwards in the worst case. --sct
951 err
= ext4_splice_branch(handle
, inode
, iblock
,
952 partial
, indirect_blks
, count
);
954 * i_disksize growing is protected by i_data_sem. Don't forget to
955 * protect it if you're about to implement concurrent
956 * ext4_get_block() -bzzz
958 if (!err
&& extend_disksize
) {
959 disksize
= ((loff_t
) iblock
+ count
) << inode
->i_blkbits
;
960 if (disksize
> i_size_read(inode
))
961 disksize
= i_size_read(inode
);
962 if (disksize
> ei
->i_disksize
)
963 ei
->i_disksize
= disksize
;
968 set_buffer_new(bh_result
);
970 map_bh(bh_result
, inode
->i_sb
, le32_to_cpu(chain
[depth
-1].key
));
971 if (count
> blocks_to_boundary
)
972 set_buffer_boundary(bh_result
);
974 /* Clean up and exit */
975 partial
= chain
+ depth
- 1; /* the whole chain */
977 while (partial
> chain
) {
978 BUFFER_TRACE(partial
->bh
, "call brelse");
982 BUFFER_TRACE(bh_result
, "returned");
988 * Calculate the number of metadata blocks need to reserve
989 * to allocate @blocks for non extent file based file
991 static int ext4_indirect_calc_metadata_amount(struct inode
*inode
, int blocks
)
993 int icap
= EXT4_ADDR_PER_BLOCK(inode
->i_sb
);
994 int ind_blks
, dind_blks
, tind_blks
;
996 /* number of new indirect blocks needed */
997 ind_blks
= (blocks
+ icap
- 1) / icap
;
999 dind_blks
= (ind_blks
+ icap
- 1) / icap
;
1003 return ind_blks
+ dind_blks
+ tind_blks
;
1007 * Calculate the number of metadata blocks need to reserve
1008 * to allocate given number of blocks
1010 static int ext4_calc_metadata_amount(struct inode
*inode
, int blocks
)
1015 if (EXT4_I(inode
)->i_flags
& EXT4_EXTENTS_FL
)
1016 return ext4_ext_calc_metadata_amount(inode
, blocks
);
1018 return ext4_indirect_calc_metadata_amount(inode
, blocks
);
1021 static void ext4_da_update_reserve_space(struct inode
*inode
, int used
)
1023 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
1024 int total
, mdb
, mdb_free
;
1026 spin_lock(&EXT4_I(inode
)->i_block_reservation_lock
);
1027 /* recalculate the number of metablocks still need to be reserved */
1028 total
= EXT4_I(inode
)->i_reserved_data_blocks
- used
;
1029 mdb
= ext4_calc_metadata_amount(inode
, total
);
1031 /* figure out how many metablocks to release */
1032 BUG_ON(mdb
> EXT4_I(inode
)->i_reserved_meta_blocks
);
1033 mdb_free
= EXT4_I(inode
)->i_reserved_meta_blocks
- mdb
;
1035 /* Account for allocated meta_blocks */
1036 mdb_free
-= EXT4_I(inode
)->i_allocated_meta_blocks
;
1038 /* update fs free blocks counter for truncate case */
1039 percpu_counter_add(&sbi
->s_freeblocks_counter
, mdb_free
);
1041 /* update per-inode reservations */
1042 BUG_ON(used
> EXT4_I(inode
)->i_reserved_data_blocks
);
1043 EXT4_I(inode
)->i_reserved_data_blocks
-= used
;
1045 BUG_ON(mdb
> EXT4_I(inode
)->i_reserved_meta_blocks
);
1046 EXT4_I(inode
)->i_reserved_meta_blocks
= mdb
;
1047 EXT4_I(inode
)->i_allocated_meta_blocks
= 0;
1048 spin_unlock(&EXT4_I(inode
)->i_block_reservation_lock
);
1052 * The ext4_get_blocks_wrap() function try to look up the requested blocks,
1053 * and returns if the blocks are already mapped.
1055 * Otherwise it takes the write lock of the i_data_sem and allocate blocks
1056 * and store the allocated blocks in the result buffer head and mark it
1059 * If file type is extents based, it will call ext4_ext_get_blocks(),
1060 * Otherwise, call with ext4_get_blocks_handle() to handle indirect mapping
1063 * On success, it returns the number of blocks being mapped or allocate.
1064 * if create==0 and the blocks are pre-allocated and uninitialized block,
1065 * the result buffer head is unmapped. If the create ==1, it will make sure
1066 * the buffer head is mapped.
1068 * It returns 0 if plain look up failed (blocks have not been allocated), in
1069 * that casem, buffer head is unmapped
1071 * It returns the error in case of allocation failure.
1073 int ext4_get_blocks_wrap(handle_t
*handle
, struct inode
*inode
, sector_t block
,
1074 unsigned long max_blocks
, struct buffer_head
*bh
,
1075 int create
, int extend_disksize
, int flag
)
1079 clear_buffer_mapped(bh
);
1082 * Try to see if we can get the block without requesting
1083 * for new file system block.
1085 down_read((&EXT4_I(inode
)->i_data_sem
));
1086 if (EXT4_I(inode
)->i_flags
& EXT4_EXTENTS_FL
) {
1087 retval
= ext4_ext_get_blocks(handle
, inode
, block
, max_blocks
,
1090 retval
= ext4_get_blocks_handle(handle
,
1091 inode
, block
, max_blocks
, bh
, 0, 0);
1093 up_read((&EXT4_I(inode
)->i_data_sem
));
1095 /* If it is only a block(s) look up */
1100 * Returns if the blocks have already allocated
1102 * Note that if blocks have been preallocated
1103 * ext4_ext_get_block() returns th create = 0
1104 * with buffer head unmapped.
1106 if (retval
> 0 && buffer_mapped(bh
))
1110 * New blocks allocate and/or writing to uninitialized extent
1111 * will possibly result in updating i_data, so we take
1112 * the write lock of i_data_sem, and call get_blocks()
1113 * with create == 1 flag.
1115 down_write((&EXT4_I(inode
)->i_data_sem
));
1118 * if the caller is from delayed allocation writeout path
1119 * we have already reserved fs blocks for allocation
1120 * let the underlying get_block() function know to
1121 * avoid double accounting
1124 EXT4_I(inode
)->i_delalloc_reserved_flag
= 1;
1126 * We need to check for EXT4 here because migrate
1127 * could have changed the inode type in between
1129 if (EXT4_I(inode
)->i_flags
& EXT4_EXTENTS_FL
) {
1130 retval
= ext4_ext_get_blocks(handle
, inode
, block
, max_blocks
,
1131 bh
, create
, extend_disksize
);
1133 retval
= ext4_get_blocks_handle(handle
, inode
, block
,
1134 max_blocks
, bh
, create
, extend_disksize
);
1136 if (retval
> 0 && buffer_new(bh
)) {
1138 * We allocated new blocks which will result in
1139 * i_data's format changing. Force the migrate
1140 * to fail by clearing migrate flags
1142 EXT4_I(inode
)->i_flags
= EXT4_I(inode
)->i_flags
&
1148 EXT4_I(inode
)->i_delalloc_reserved_flag
= 0;
1150 * Update reserved blocks/metadata blocks
1151 * after successful block allocation
1152 * which were deferred till now
1154 if ((retval
> 0) && buffer_delay(bh
))
1155 ext4_da_update_reserve_space(inode
, retval
);
1158 up_write((&EXT4_I(inode
)->i_data_sem
));
1162 /* Maximum number of blocks we map for direct IO at once. */
1163 #define DIO_MAX_BLOCKS 4096
1165 static int ext4_get_block(struct inode
*inode
, sector_t iblock
,
1166 struct buffer_head
*bh_result
, int create
)
1168 handle_t
*handle
= ext4_journal_current_handle();
1169 int ret
= 0, started
= 0;
1170 unsigned max_blocks
= bh_result
->b_size
>> inode
->i_blkbits
;
1173 if (create
&& !handle
) {
1174 /* Direct IO write... */
1175 if (max_blocks
> DIO_MAX_BLOCKS
)
1176 max_blocks
= DIO_MAX_BLOCKS
;
1177 dio_credits
= ext4_chunk_trans_blocks(inode
, max_blocks
);
1178 handle
= ext4_journal_start(inode
, dio_credits
);
1179 if (IS_ERR(handle
)) {
1180 ret
= PTR_ERR(handle
);
1186 ret
= ext4_get_blocks_wrap(handle
, inode
, iblock
,
1187 max_blocks
, bh_result
, create
, 0, 0);
1189 bh_result
->b_size
= (ret
<< inode
->i_blkbits
);
1193 ext4_journal_stop(handle
);
1199 * `handle' can be NULL if create is zero
1201 struct buffer_head
*ext4_getblk(handle_t
*handle
, struct inode
*inode
,
1202 ext4_lblk_t block
, int create
, int *errp
)
1204 struct buffer_head dummy
;
1207 J_ASSERT(handle
!= NULL
|| create
== 0);
1210 dummy
.b_blocknr
= -1000;
1211 buffer_trace_init(&dummy
.b_history
);
1212 err
= ext4_get_blocks_wrap(handle
, inode
, block
, 1,
1213 &dummy
, create
, 1, 0);
1215 * ext4_get_blocks_handle() returns number of blocks
1216 * mapped. 0 in case of a HOLE.
1224 if (!err
&& buffer_mapped(&dummy
)) {
1225 struct buffer_head
*bh
;
1226 bh
= sb_getblk(inode
->i_sb
, dummy
.b_blocknr
);
1231 if (buffer_new(&dummy
)) {
1232 J_ASSERT(create
!= 0);
1233 J_ASSERT(handle
!= NULL
);
1236 * Now that we do not always journal data, we should
1237 * keep in mind whether this should always journal the
1238 * new buffer as metadata. For now, regular file
1239 * writes use ext4_get_block instead, so it's not a
1243 BUFFER_TRACE(bh
, "call get_create_access");
1244 fatal
= ext4_journal_get_create_access(handle
, bh
);
1245 if (!fatal
&& !buffer_uptodate(bh
)) {
1246 memset(bh
->b_data
,0,inode
->i_sb
->s_blocksize
);
1247 set_buffer_uptodate(bh
);
1250 BUFFER_TRACE(bh
, "call ext4_journal_dirty_metadata");
1251 err
= ext4_journal_dirty_metadata(handle
, bh
);
1255 BUFFER_TRACE(bh
, "not a new buffer");
1268 struct buffer_head
*ext4_bread(handle_t
*handle
, struct inode
*inode
,
1269 ext4_lblk_t block
, int create
, int *err
)
1271 struct buffer_head
* bh
;
1273 bh
= ext4_getblk(handle
, inode
, block
, create
, err
);
1276 if (buffer_uptodate(bh
))
1278 ll_rw_block(READ_META
, 1, &bh
);
1280 if (buffer_uptodate(bh
))
1287 static int walk_page_buffers( handle_t
*handle
,
1288 struct buffer_head
*head
,
1292 int (*fn
)( handle_t
*handle
,
1293 struct buffer_head
*bh
))
1295 struct buffer_head
*bh
;
1296 unsigned block_start
, block_end
;
1297 unsigned blocksize
= head
->b_size
;
1299 struct buffer_head
*next
;
1301 for ( bh
= head
, block_start
= 0;
1302 ret
== 0 && (bh
!= head
|| !block_start
);
1303 block_start
= block_end
, bh
= next
)
1305 next
= bh
->b_this_page
;
1306 block_end
= block_start
+ blocksize
;
1307 if (block_end
<= from
|| block_start
>= to
) {
1308 if (partial
&& !buffer_uptodate(bh
))
1312 err
= (*fn
)(handle
, bh
);
1320 * To preserve ordering, it is essential that the hole instantiation and
1321 * the data write be encapsulated in a single transaction. We cannot
1322 * close off a transaction and start a new one between the ext4_get_block()
1323 * and the commit_write(). So doing the jbd2_journal_start at the start of
1324 * prepare_write() is the right place.
1326 * Also, this function can nest inside ext4_writepage() ->
1327 * block_write_full_page(). In that case, we *know* that ext4_writepage()
1328 * has generated enough buffer credits to do the whole page. So we won't
1329 * block on the journal in that case, which is good, because the caller may
1332 * By accident, ext4 can be reentered when a transaction is open via
1333 * quota file writes. If we were to commit the transaction while thus
1334 * reentered, there can be a deadlock - we would be holding a quota
1335 * lock, and the commit would never complete if another thread had a
1336 * transaction open and was blocking on the quota lock - a ranking
1339 * So what we do is to rely on the fact that jbd2_journal_stop/journal_start
1340 * will _not_ run commit under these circumstances because handle->h_ref
1341 * is elevated. We'll still have enough credits for the tiny quotafile
1344 static int do_journal_get_write_access(handle_t
*handle
,
1345 struct buffer_head
*bh
)
1347 if (!buffer_mapped(bh
) || buffer_freed(bh
))
1349 return ext4_journal_get_write_access(handle
, bh
);
1352 static int ext4_write_begin(struct file
*file
, struct address_space
*mapping
,
1353 loff_t pos
, unsigned len
, unsigned flags
,
1354 struct page
**pagep
, void **fsdata
)
1356 struct inode
*inode
= mapping
->host
;
1357 int ret
, needed_blocks
= ext4_writepage_trans_blocks(inode
);
1364 index
= pos
>> PAGE_CACHE_SHIFT
;
1365 from
= pos
& (PAGE_CACHE_SIZE
- 1);
1369 handle
= ext4_journal_start(inode
, needed_blocks
);
1370 if (IS_ERR(handle
)) {
1371 ret
= PTR_ERR(handle
);
1375 /* We cannot recurse into the filesystem as the transaction is already
1377 flags
|= AOP_FLAG_NOFS
;
1379 page
= grab_cache_page_write_begin(mapping
, index
, flags
);
1381 ext4_journal_stop(handle
);
1387 ret
= block_write_begin(file
, mapping
, pos
, len
, flags
, pagep
, fsdata
,
1390 if (!ret
&& ext4_should_journal_data(inode
)) {
1391 ret
= walk_page_buffers(handle
, page_buffers(page
),
1392 from
, to
, NULL
, do_journal_get_write_access
);
1397 ext4_journal_stop(handle
);
1398 page_cache_release(page
);
1401 if (ret
== -ENOSPC
&& ext4_should_retry_alloc(inode
->i_sb
, &retries
))
1407 /* For write_end() in data=journal mode */
1408 static int write_end_fn(handle_t
*handle
, struct buffer_head
*bh
)
1410 if (!buffer_mapped(bh
) || buffer_freed(bh
))
1412 set_buffer_uptodate(bh
);
1413 return ext4_journal_dirty_metadata(handle
, bh
);
1417 * We need to pick up the new inode size which generic_commit_write gave us
1418 * `file' can be NULL - eg, when called from page_symlink().
1420 * ext4 never places buffers on inode->i_mapping->private_list. metadata
1421 * buffers are managed internally.
1423 static int ext4_ordered_write_end(struct file
*file
,
1424 struct address_space
*mapping
,
1425 loff_t pos
, unsigned len
, unsigned copied
,
1426 struct page
*page
, void *fsdata
)
1428 handle_t
*handle
= ext4_journal_current_handle();
1429 struct inode
*inode
= mapping
->host
;
1432 ret
= ext4_jbd2_file_inode(handle
, inode
);
1436 * generic_write_end() will run mark_inode_dirty() if i_size
1437 * changes. So let's piggyback the i_disksize mark_inode_dirty
1442 new_i_size
= pos
+ copied
;
1443 if (new_i_size
> EXT4_I(inode
)->i_disksize
)
1444 EXT4_I(inode
)->i_disksize
= new_i_size
;
1445 ret2
= generic_write_end(file
, mapping
, pos
, len
, copied
,
1451 ret2
= ext4_journal_stop(handle
);
1455 return ret
? ret
: copied
;
1458 static int ext4_writeback_write_end(struct file
*file
,
1459 struct address_space
*mapping
,
1460 loff_t pos
, unsigned len
, unsigned copied
,
1461 struct page
*page
, void *fsdata
)
1463 handle_t
*handle
= ext4_journal_current_handle();
1464 struct inode
*inode
= mapping
->host
;
1468 new_i_size
= pos
+ copied
;
1469 if (new_i_size
> EXT4_I(inode
)->i_disksize
)
1470 EXT4_I(inode
)->i_disksize
= new_i_size
;
1472 ret2
= generic_write_end(file
, mapping
, pos
, len
, copied
,
1478 ret2
= ext4_journal_stop(handle
);
1482 return ret
? ret
: copied
;
1485 static int ext4_journalled_write_end(struct file
*file
,
1486 struct address_space
*mapping
,
1487 loff_t pos
, unsigned len
, unsigned copied
,
1488 struct page
*page
, void *fsdata
)
1490 handle_t
*handle
= ext4_journal_current_handle();
1491 struct inode
*inode
= mapping
->host
;
1496 from
= pos
& (PAGE_CACHE_SIZE
- 1);
1500 if (!PageUptodate(page
))
1502 page_zero_new_buffers(page
, from
+copied
, to
);
1505 ret
= walk_page_buffers(handle
, page_buffers(page
), from
,
1506 to
, &partial
, write_end_fn
);
1508 SetPageUptodate(page
);
1509 if (pos
+copied
> inode
->i_size
)
1510 i_size_write(inode
, pos
+copied
);
1511 EXT4_I(inode
)->i_state
|= EXT4_STATE_JDATA
;
1512 if (inode
->i_size
> EXT4_I(inode
)->i_disksize
) {
1513 EXT4_I(inode
)->i_disksize
= inode
->i_size
;
1514 ret2
= ext4_mark_inode_dirty(handle
, inode
);
1520 ret2
= ext4_journal_stop(handle
);
1523 page_cache_release(page
);
1525 return ret
? ret
: copied
;
1528 static int ext4_da_reserve_space(struct inode
*inode
, int nrblocks
)
1530 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
1531 unsigned long md_needed
, mdblocks
, total
= 0;
1534 * recalculate the amount of metadata blocks to reserve
1535 * in order to allocate nrblocks
1536 * worse case is one extent per block
1538 spin_lock(&EXT4_I(inode
)->i_block_reservation_lock
);
1539 total
= EXT4_I(inode
)->i_reserved_data_blocks
+ nrblocks
;
1540 mdblocks
= ext4_calc_metadata_amount(inode
, total
);
1541 BUG_ON(mdblocks
< EXT4_I(inode
)->i_reserved_meta_blocks
);
1543 md_needed
= mdblocks
- EXT4_I(inode
)->i_reserved_meta_blocks
;
1544 total
= md_needed
+ nrblocks
;
1546 if (ext4_has_free_blocks(sbi
, total
) < total
) {
1547 spin_unlock(&EXT4_I(inode
)->i_block_reservation_lock
);
1550 /* reduce fs free blocks counter */
1551 percpu_counter_sub(&sbi
->s_freeblocks_counter
, total
);
1553 EXT4_I(inode
)->i_reserved_data_blocks
+= nrblocks
;
1554 EXT4_I(inode
)->i_reserved_meta_blocks
= mdblocks
;
1556 spin_unlock(&EXT4_I(inode
)->i_block_reservation_lock
);
1557 return 0; /* success */
1560 static void ext4_da_release_space(struct inode
*inode
, int to_free
)
1562 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
1563 int total
, mdb
, mdb_free
, release
;
1566 return; /* Nothing to release, exit */
1568 spin_lock(&EXT4_I(inode
)->i_block_reservation_lock
);
1570 if (!EXT4_I(inode
)->i_reserved_data_blocks
) {
1572 * if there is no reserved blocks, but we try to free some
1573 * then the counter is messed up somewhere.
1574 * but since this function is called from invalidate
1575 * page, it's harmless to return without any action
1577 printk(KERN_INFO
"ext4 delalloc try to release %d reserved "
1578 "blocks for inode %lu, but there is no reserved "
1579 "data blocks\n", to_free
, inode
->i_ino
);
1580 spin_unlock(&EXT4_I(inode
)->i_block_reservation_lock
);
1584 /* recalculate the number of metablocks still need to be reserved */
1585 total
= EXT4_I(inode
)->i_reserved_data_blocks
- to_free
;
1586 mdb
= ext4_calc_metadata_amount(inode
, total
);
1588 /* figure out how many metablocks to release */
1589 BUG_ON(mdb
> EXT4_I(inode
)->i_reserved_meta_blocks
);
1590 mdb_free
= EXT4_I(inode
)->i_reserved_meta_blocks
- mdb
;
1592 release
= to_free
+ mdb_free
;
1594 /* update fs free blocks counter for truncate case */
1595 percpu_counter_add(&sbi
->s_freeblocks_counter
, release
);
1597 /* update per-inode reservations */
1598 BUG_ON(to_free
> EXT4_I(inode
)->i_reserved_data_blocks
);
1599 EXT4_I(inode
)->i_reserved_data_blocks
-= to_free
;
1601 BUG_ON(mdb
> EXT4_I(inode
)->i_reserved_meta_blocks
);
1602 EXT4_I(inode
)->i_reserved_meta_blocks
= mdb
;
1603 spin_unlock(&EXT4_I(inode
)->i_block_reservation_lock
);
1606 static void ext4_da_page_release_reservation(struct page
*page
,
1607 unsigned long offset
)
1610 struct buffer_head
*head
, *bh
;
1611 unsigned int curr_off
= 0;
1613 head
= page_buffers(page
);
1616 unsigned int next_off
= curr_off
+ bh
->b_size
;
1618 if ((offset
<= curr_off
) && (buffer_delay(bh
))) {
1620 clear_buffer_delay(bh
);
1622 curr_off
= next_off
;
1623 } while ((bh
= bh
->b_this_page
) != head
);
1624 ext4_da_release_space(page
->mapping
->host
, to_release
);
1628 * Delayed allocation stuff
1631 struct mpage_da_data
{
1632 struct inode
*inode
;
1633 struct buffer_head lbh
; /* extent of blocks */
1634 unsigned long first_page
, next_page
; /* extent of pages */
1635 get_block_t
*get_block
;
1636 struct writeback_control
*wbc
;
1642 * mpage_da_submit_io - walks through extent of pages and try to write
1643 * them with writepage() call back
1645 * @mpd->inode: inode
1646 * @mpd->first_page: first page of the extent
1647 * @mpd->next_page: page after the last page of the extent
1648 * @mpd->get_block: the filesystem's block mapper function
1650 * By the time mpage_da_submit_io() is called we expect all blocks
1651 * to be allocated. this may be wrong if allocation failed.
1653 * As pages are already locked by write_cache_pages(), we can't use it
1655 static int mpage_da_submit_io(struct mpage_da_data
*mpd
)
1658 struct pagevec pvec
;
1659 unsigned long index
, end
;
1660 int ret
= 0, err
, nr_pages
, i
;
1661 struct inode
*inode
= mpd
->inode
;
1662 struct address_space
*mapping
= inode
->i_mapping
;
1664 BUG_ON(mpd
->next_page
<= mpd
->first_page
);
1666 * We need to start from the first_page to the next_page - 1
1667 * to make sure we also write the mapped dirty buffer_heads.
1668 * If we look at mpd->lbh.b_blocknr we would only be looking
1669 * at the currently mapped buffer_heads.
1671 index
= mpd
->first_page
;
1672 end
= mpd
->next_page
- 1;
1674 pagevec_init(&pvec
, 0);
1675 while (index
<= end
) {
1676 nr_pages
= pagevec_lookup(&pvec
, mapping
, index
, PAGEVEC_SIZE
);
1679 for (i
= 0; i
< nr_pages
; i
++) {
1680 struct page
*page
= pvec
.pages
[i
];
1682 index
= page
->index
;
1687 BUG_ON(!PageLocked(page
));
1688 BUG_ON(PageWriteback(page
));
1690 pages_skipped
= mpd
->wbc
->pages_skipped
;
1691 err
= mapping
->a_ops
->writepage(page
, mpd
->wbc
);
1693 mpd
->pages_written
++;
1695 * In error case, we have to continue because
1696 * remaining pages are still locked
1697 * XXX: unlock and re-dirty them?
1702 pagevec_release(&pvec
);
1708 * mpage_put_bnr_to_bhs - walk blocks and assign them actual numbers
1710 * @mpd->inode - inode to walk through
1711 * @exbh->b_blocknr - first block on a disk
1712 * @exbh->b_size - amount of space in bytes
1713 * @logical - first logical block to start assignment with
1715 * the function goes through all passed space and put actual disk
1716 * block numbers into buffer heads, dropping BH_Delay
1718 static void mpage_put_bnr_to_bhs(struct mpage_da_data
*mpd
, sector_t logical
,
1719 struct buffer_head
*exbh
)
1721 struct inode
*inode
= mpd
->inode
;
1722 struct address_space
*mapping
= inode
->i_mapping
;
1723 int blocks
= exbh
->b_size
>> inode
->i_blkbits
;
1724 sector_t pblock
= exbh
->b_blocknr
, cur_logical
;
1725 struct buffer_head
*head
, *bh
;
1727 struct pagevec pvec
;
1730 index
= logical
>> (PAGE_CACHE_SHIFT
- inode
->i_blkbits
);
1731 end
= (logical
+ blocks
- 1) >> (PAGE_CACHE_SHIFT
- inode
->i_blkbits
);
1732 cur_logical
= index
<< (PAGE_CACHE_SHIFT
- inode
->i_blkbits
);
1734 pagevec_init(&pvec
, 0);
1736 while (index
<= end
) {
1737 /* XXX: optimize tail */
1738 nr_pages
= pagevec_lookup(&pvec
, mapping
, index
, PAGEVEC_SIZE
);
1741 for (i
= 0; i
< nr_pages
; i
++) {
1742 struct page
*page
= pvec
.pages
[i
];
1744 index
= page
->index
;
1749 BUG_ON(!PageLocked(page
));
1750 BUG_ON(PageWriteback(page
));
1751 BUG_ON(!page_has_buffers(page
));
1753 bh
= page_buffers(page
);
1756 /* skip blocks out of the range */
1758 if (cur_logical
>= logical
)
1761 } while ((bh
= bh
->b_this_page
) != head
);
1764 if (cur_logical
>= logical
+ blocks
)
1766 if (buffer_delay(bh
)) {
1767 bh
->b_blocknr
= pblock
;
1768 clear_buffer_delay(bh
);
1769 bh
->b_bdev
= inode
->i_sb
->s_bdev
;
1770 } else if (buffer_unwritten(bh
)) {
1771 bh
->b_blocknr
= pblock
;
1772 clear_buffer_unwritten(bh
);
1773 set_buffer_mapped(bh
);
1775 bh
->b_bdev
= inode
->i_sb
->s_bdev
;
1776 } else if (buffer_mapped(bh
))
1777 BUG_ON(bh
->b_blocknr
!= pblock
);
1781 } while ((bh
= bh
->b_this_page
) != head
);
1783 pagevec_release(&pvec
);
1789 * __unmap_underlying_blocks - just a helper function to unmap
1790 * set of blocks described by @bh
1792 static inline void __unmap_underlying_blocks(struct inode
*inode
,
1793 struct buffer_head
*bh
)
1795 struct block_device
*bdev
= inode
->i_sb
->s_bdev
;
1798 blocks
= bh
->b_size
>> inode
->i_blkbits
;
1799 for (i
= 0; i
< blocks
; i
++)
1800 unmap_underlying_metadata(bdev
, bh
->b_blocknr
+ i
);
1804 * mpage_da_map_blocks - go through given space
1806 * @mpd->lbh - bh describing space
1807 * @mpd->get_block - the filesystem's block mapper function
1809 * The function skips space we know is already mapped to disk blocks.
1812 static void mpage_da_map_blocks(struct mpage_da_data
*mpd
)
1815 struct buffer_head
*lbh
= &mpd
->lbh
;
1816 sector_t next
= lbh
->b_blocknr
;
1817 struct buffer_head
new;
1820 * We consider only non-mapped and non-allocated blocks
1822 if (buffer_mapped(lbh
) && !buffer_delay(lbh
))
1825 new.b_state
= lbh
->b_state
;
1827 new.b_size
= lbh
->b_size
;
1830 * If we didn't accumulate anything
1831 * to write simply return
1835 err
= mpd
->get_block(mpd
->inode
, next
, &new, 1);
1838 BUG_ON(new.b_size
== 0);
1840 if (buffer_new(&new))
1841 __unmap_underlying_blocks(mpd
->inode
, &new);
1844 * If blocks are delayed marked, we need to
1845 * put actual blocknr and drop delayed bit
1847 if (buffer_delay(lbh
) || buffer_unwritten(lbh
))
1848 mpage_put_bnr_to_bhs(mpd
, next
, &new);
1853 #define BH_FLAGS ((1 << BH_Uptodate) | (1 << BH_Mapped) | \
1854 (1 << BH_Delay) | (1 << BH_Unwritten))
1857 * mpage_add_bh_to_extent - try to add one more block to extent of blocks
1859 * @mpd->lbh - extent of blocks
1860 * @logical - logical number of the block in the file
1861 * @bh - bh of the block (used to access block's state)
1863 * the function is used to collect contig. blocks in same state
1865 static void mpage_add_bh_to_extent(struct mpage_da_data
*mpd
,
1866 sector_t logical
, struct buffer_head
*bh
)
1869 size_t b_size
= bh
->b_size
;
1870 struct buffer_head
*lbh
= &mpd
->lbh
;
1871 int nrblocks
= lbh
->b_size
>> mpd
->inode
->i_blkbits
;
1873 /* check if thereserved journal credits might overflow */
1874 if (!(EXT4_I(mpd
->inode
)->i_flags
& EXT4_EXTENTS_FL
)) {
1875 if (nrblocks
>= EXT4_MAX_TRANS_DATA
) {
1877 * With non-extent format we are limited by the journal
1878 * credit available. Total credit needed to insert
1879 * nrblocks contiguous blocks is dependent on the
1880 * nrblocks. So limit nrblocks.
1883 } else if ((nrblocks
+ (b_size
>> mpd
->inode
->i_blkbits
)) >
1884 EXT4_MAX_TRANS_DATA
) {
1886 * Adding the new buffer_head would make it cross the
1887 * allowed limit for which we have journal credit
1888 * reserved. So limit the new bh->b_size
1890 b_size
= (EXT4_MAX_TRANS_DATA
- nrblocks
) <<
1891 mpd
->inode
->i_blkbits
;
1892 /* we will do mpage_da_submit_io in the next loop */
1896 * First block in the extent
1898 if (lbh
->b_size
== 0) {
1899 lbh
->b_blocknr
= logical
;
1900 lbh
->b_size
= b_size
;
1901 lbh
->b_state
= bh
->b_state
& BH_FLAGS
;
1905 next
= lbh
->b_blocknr
+ nrblocks
;
1907 * Can we merge the block to our big extent?
1909 if (logical
== next
&& (bh
->b_state
& BH_FLAGS
) == lbh
->b_state
) {
1910 lbh
->b_size
+= b_size
;
1916 * We couldn't merge the block to our extent, so we
1917 * need to flush current extent and start new one
1919 mpage_da_map_blocks(mpd
);
1920 mpage_da_submit_io(mpd
);
1926 * __mpage_da_writepage - finds extent of pages and blocks
1928 * @page: page to consider
1929 * @wbc: not used, we just follow rules
1932 * The function finds extents of pages and scan them for all blocks.
1934 static int __mpage_da_writepage(struct page
*page
,
1935 struct writeback_control
*wbc
, void *data
)
1937 struct mpage_da_data
*mpd
= data
;
1938 struct inode
*inode
= mpd
->inode
;
1939 struct buffer_head
*bh
, *head
, fake
;
1944 * Rest of the page in the page_vec
1945 * redirty then and skip then. We will
1946 * try to to write them again after
1947 * starting a new transaction
1949 redirty_page_for_writepage(wbc
, page
);
1951 return MPAGE_DA_EXTENT_TAIL
;
1954 * Can we merge this page to current extent?
1956 if (mpd
->next_page
!= page
->index
) {
1958 * Nope, we can't. So, we map non-allocated blocks
1959 * and start IO on them using writepage()
1961 if (mpd
->next_page
!= mpd
->first_page
) {
1962 mpage_da_map_blocks(mpd
);
1963 mpage_da_submit_io(mpd
);
1965 * skip rest of the page in the page_vec
1968 redirty_page_for_writepage(wbc
, page
);
1970 return MPAGE_DA_EXTENT_TAIL
;
1974 * Start next extent of pages ...
1976 mpd
->first_page
= page
->index
;
1981 mpd
->lbh
.b_size
= 0;
1982 mpd
->lbh
.b_state
= 0;
1983 mpd
->lbh
.b_blocknr
= 0;
1986 mpd
->next_page
= page
->index
+ 1;
1987 logical
= (sector_t
) page
->index
<<
1988 (PAGE_CACHE_SHIFT
- inode
->i_blkbits
);
1990 if (!page_has_buffers(page
)) {
1992 * There is no attached buffer heads yet (mmap?)
1993 * we treat the page asfull of dirty blocks
1996 bh
->b_size
= PAGE_CACHE_SIZE
;
1998 set_buffer_dirty(bh
);
1999 set_buffer_uptodate(bh
);
2000 mpage_add_bh_to_extent(mpd
, logical
, bh
);
2002 return MPAGE_DA_EXTENT_TAIL
;
2005 * Page with regular buffer heads, just add all dirty ones
2007 head
= page_buffers(page
);
2010 BUG_ON(buffer_locked(bh
));
2012 * We need to try to allocate
2013 * unmapped blocks in the same page.
2014 * Otherwise we won't make progress
2015 * with the page in ext4_da_writepage
2017 if (buffer_dirty(bh
) &&
2018 (!buffer_mapped(bh
) || buffer_delay(bh
))) {
2019 mpage_add_bh_to_extent(mpd
, logical
, bh
);
2021 return MPAGE_DA_EXTENT_TAIL
;
2022 } else if (buffer_dirty(bh
) && (buffer_mapped(bh
))) {
2024 * mapped dirty buffer. We need to update
2025 * the b_state because we look at
2026 * b_state in mpage_da_map_blocks. We don't
2027 * update b_size because if we find an
2028 * unmapped buffer_head later we need to
2029 * use the b_state flag of that buffer_head.
2031 if (mpd
->lbh
.b_size
== 0)
2033 bh
->b_state
& BH_FLAGS
;
2036 } while ((bh
= bh
->b_this_page
) != head
);
2043 * mpage_da_writepages - walk the list of dirty pages of the given
2044 * address space, allocates non-allocated blocks, maps newly-allocated
2045 * blocks to existing bhs and issue IO them
2047 * @mapping: address space structure to write
2048 * @wbc: subtract the number of written pages from *@wbc->nr_to_write
2049 * @get_block: the filesystem's block mapper function.
2051 * This is a library function, which implements the writepages()
2052 * address_space_operation.
2054 static int mpage_da_writepages(struct address_space
*mapping
,
2055 struct writeback_control
*wbc
,
2056 get_block_t get_block
)
2058 struct mpage_da_data mpd
;
2063 return generic_writepages(mapping
, wbc
);
2066 mpd
.inode
= mapping
->host
;
2068 mpd
.lbh
.b_state
= 0;
2069 mpd
.lbh
.b_blocknr
= 0;
2072 mpd
.get_block
= get_block
;
2074 mpd
.pages_written
= 0;
2076 to_write
= wbc
->nr_to_write
;
2078 ret
= write_cache_pages(mapping
, wbc
, __mpage_da_writepage
, &mpd
);
2081 * Handle last extent of pages
2083 if (!mpd
.io_done
&& mpd
.next_page
!= mpd
.first_page
) {
2084 mpage_da_map_blocks(&mpd
);
2085 mpage_da_submit_io(&mpd
);
2088 wbc
->nr_to_write
= to_write
- mpd
.pages_written
;
2093 * this is a special callback for ->write_begin() only
2094 * it's intention is to return mapped block or reserve space
2096 static int ext4_da_get_block_prep(struct inode
*inode
, sector_t iblock
,
2097 struct buffer_head
*bh_result
, int create
)
2101 BUG_ON(create
== 0);
2102 BUG_ON(bh_result
->b_size
!= inode
->i_sb
->s_blocksize
);
2105 * first, we need to know whether the block is allocated already
2106 * preallocated blocks are unmapped but should treated
2107 * the same as allocated blocks.
2109 ret
= ext4_get_blocks_wrap(NULL
, inode
, iblock
, 1, bh_result
, 0, 0, 0);
2110 if ((ret
== 0) && !buffer_delay(bh_result
)) {
2111 /* the block isn't (pre)allocated yet, let's reserve space */
2113 * XXX: __block_prepare_write() unmaps passed block,
2116 ret
= ext4_da_reserve_space(inode
, 1);
2118 /* not enough space to reserve */
2121 map_bh(bh_result
, inode
->i_sb
, 0);
2122 set_buffer_new(bh_result
);
2123 set_buffer_delay(bh_result
);
2124 } else if (ret
> 0) {
2125 bh_result
->b_size
= (ret
<< inode
->i_blkbits
);
2131 #define EXT4_DELALLOC_RSVED 1
2132 static int ext4_da_get_block_write(struct inode
*inode
, sector_t iblock
,
2133 struct buffer_head
*bh_result
, int create
)
2136 unsigned max_blocks
= bh_result
->b_size
>> inode
->i_blkbits
;
2137 loff_t disksize
= EXT4_I(inode
)->i_disksize
;
2138 handle_t
*handle
= NULL
;
2140 handle
= ext4_journal_current_handle();
2142 ret
= ext4_get_blocks_wrap(handle
, inode
, iblock
, max_blocks
,
2143 bh_result
, 0, 0, 0);
2146 ret
= ext4_get_blocks_wrap(handle
, inode
, iblock
, max_blocks
,
2147 bh_result
, create
, 0, EXT4_DELALLOC_RSVED
);
2151 bh_result
->b_size
= (ret
<< inode
->i_blkbits
);
2154 * Update on-disk size along with block allocation
2155 * we don't use 'extend_disksize' as size may change
2156 * within already allocated block -bzzz
2158 disksize
= ((loff_t
) iblock
+ ret
) << inode
->i_blkbits
;
2159 if (disksize
> i_size_read(inode
))
2160 disksize
= i_size_read(inode
);
2161 if (disksize
> EXT4_I(inode
)->i_disksize
) {
2163 * XXX: replace with spinlock if seen contended -bzzz
2165 down_write(&EXT4_I(inode
)->i_data_sem
);
2166 if (disksize
> EXT4_I(inode
)->i_disksize
)
2167 EXT4_I(inode
)->i_disksize
= disksize
;
2168 up_write(&EXT4_I(inode
)->i_data_sem
);
2170 if (EXT4_I(inode
)->i_disksize
== disksize
) {
2171 ret
= ext4_mark_inode_dirty(handle
, inode
);
2180 static int ext4_bh_unmapped_or_delay(handle_t
*handle
, struct buffer_head
*bh
)
2183 * unmapped buffer is possible for holes.
2184 * delay buffer is possible with delayed allocation
2186 return ((!buffer_mapped(bh
) || buffer_delay(bh
)) && buffer_dirty(bh
));
2189 static int ext4_normal_get_block_write(struct inode
*inode
, sector_t iblock
,
2190 struct buffer_head
*bh_result
, int create
)
2193 unsigned max_blocks
= bh_result
->b_size
>> inode
->i_blkbits
;
2196 * we don't want to do block allocation in writepage
2197 * so call get_block_wrap with create = 0
2199 ret
= ext4_get_blocks_wrap(NULL
, inode
, iblock
, max_blocks
,
2200 bh_result
, 0, 0, 0);
2202 bh_result
->b_size
= (ret
<< inode
->i_blkbits
);
2209 * get called vi ext4_da_writepages after taking page lock (have journal handle)
2210 * get called via journal_submit_inode_data_buffers (no journal handle)
2211 * get called via shrink_page_list via pdflush (no journal handle)
2212 * or grab_page_cache when doing write_begin (have journal handle)
2214 static int ext4_da_writepage(struct page
*page
,
2215 struct writeback_control
*wbc
)
2220 struct buffer_head
*page_bufs
;
2221 struct inode
*inode
= page
->mapping
->host
;
2223 size
= i_size_read(inode
);
2224 if (page
->index
== size
>> PAGE_CACHE_SHIFT
)
2225 len
= size
& ~PAGE_CACHE_MASK
;
2227 len
= PAGE_CACHE_SIZE
;
2229 if (page_has_buffers(page
)) {
2230 page_bufs
= page_buffers(page
);
2231 if (walk_page_buffers(NULL
, page_bufs
, 0, len
, NULL
,
2232 ext4_bh_unmapped_or_delay
)) {
2234 * We don't want to do block allocation
2235 * So redirty the page and return
2236 * We may reach here when we do a journal commit
2237 * via journal_submit_inode_data_buffers.
2238 * If we don't have mapping block we just ignore
2239 * them. We can also reach here via shrink_page_list
2241 redirty_page_for_writepage(wbc
, page
);
2247 * The test for page_has_buffers() is subtle:
2248 * We know the page is dirty but it lost buffers. That means
2249 * that at some moment in time after write_begin()/write_end()
2250 * has been called all buffers have been clean and thus they
2251 * must have been written at least once. So they are all
2252 * mapped and we can happily proceed with mapping them
2253 * and writing the page.
2255 * Try to initialize the buffer_heads and check whether
2256 * all are mapped and non delay. We don't want to
2257 * do block allocation here.
2259 ret
= block_prepare_write(page
, 0, PAGE_CACHE_SIZE
,
2260 ext4_normal_get_block_write
);
2262 page_bufs
= page_buffers(page
);
2263 /* check whether all are mapped and non delay */
2264 if (walk_page_buffers(NULL
, page_bufs
, 0, len
, NULL
,
2265 ext4_bh_unmapped_or_delay
)) {
2266 redirty_page_for_writepage(wbc
, page
);
2272 * We can't do block allocation here
2273 * so just redity the page and unlock
2276 redirty_page_for_writepage(wbc
, page
);
2280 /* now mark the buffer_heads as dirty and uptodate */
2281 block_commit_write(page
, 0, PAGE_CACHE_SIZE
);
2284 if (test_opt(inode
->i_sb
, NOBH
) && ext4_should_writeback_data(inode
))
2285 ret
= nobh_writepage(page
, ext4_normal_get_block_write
, wbc
);
2287 ret
= block_write_full_page(page
,
2288 ext4_normal_get_block_write
,
2295 * This is called via ext4_da_writepages() to
2296 * calulate the total number of credits to reserve to fit
2297 * a single extent allocation into a single transaction,
2298 * ext4_da_writpeages() will loop calling this before
2299 * the block allocation.
2302 static int ext4_da_writepages_trans_blocks(struct inode
*inode
)
2304 int max_blocks
= EXT4_I(inode
)->i_reserved_data_blocks
;
2307 * With non-extent format the journal credit needed to
2308 * insert nrblocks contiguous block is dependent on
2309 * number of contiguous block. So we will limit
2310 * number of contiguous block to a sane value
2312 if (!(inode
->i_flags
& EXT4_EXTENTS_FL
) &&
2313 (max_blocks
> EXT4_MAX_TRANS_DATA
))
2314 max_blocks
= EXT4_MAX_TRANS_DATA
;
2316 return ext4_chunk_trans_blocks(inode
, max_blocks
);
2319 static int ext4_da_writepages(struct address_space
*mapping
,
2320 struct writeback_control
*wbc
)
2322 handle_t
*handle
= NULL
;
2323 loff_t range_start
= 0;
2324 struct inode
*inode
= mapping
->host
;
2325 int needed_blocks
, ret
= 0, nr_to_writebump
= 0;
2326 long to_write
, pages_skipped
= 0;
2327 struct ext4_sb_info
*sbi
= EXT4_SB(mapping
->host
->i_sb
);
2330 * No pages to write? This is mainly a kludge to avoid starting
2331 * a transaction for special inodes like journal inode on last iput()
2332 * because that could violate lock ordering on umount
2334 if (!mapping
->nrpages
|| !mapping_tagged(mapping
, PAGECACHE_TAG_DIRTY
))
2338 * If the filesystem has aborted, it is read-only, so return
2339 * right away instead of dumping stack traces later on that
2340 * will obscure the real source of the problem. We test
2341 * EXT4_MOUNT_ABORT instead of sb->s_flag's MS_RDONLY because
2342 * the latter could be true if the filesystem is mounted
2343 * read-only, and in that case, ext4_da_writepages should
2344 * *never* be called, so if that ever happens, we would want
2347 if (unlikely(sbi
->s_mount_opt
& EXT4_MOUNT_ABORT
))
2351 * Make sure nr_to_write is >= sbi->s_mb_stream_request
2352 * This make sure small files blocks are allocated in
2353 * single attempt. This ensure that small files
2354 * get less fragmented.
2356 if (wbc
->nr_to_write
< sbi
->s_mb_stream_request
) {
2357 nr_to_writebump
= sbi
->s_mb_stream_request
- wbc
->nr_to_write
;
2358 wbc
->nr_to_write
= sbi
->s_mb_stream_request
;
2361 if (!wbc
->range_cyclic
)
2363 * If range_cyclic is not set force range_cont
2364 * and save the old writeback_index
2366 wbc
->range_cont
= 1;
2368 range_start
= wbc
->range_start
;
2369 pages_skipped
= wbc
->pages_skipped
;
2372 to_write
= wbc
->nr_to_write
;
2373 while (!ret
&& to_write
> 0) {
2376 * we insert one extent at a time. So we need
2377 * credit needed for single extent allocation.
2378 * journalled mode is currently not supported
2381 BUG_ON(ext4_should_journal_data(inode
));
2382 needed_blocks
= ext4_da_writepages_trans_blocks(inode
);
2384 /* start a new transaction*/
2385 handle
= ext4_journal_start(inode
, needed_blocks
);
2386 if (IS_ERR(handle
)) {
2387 ret
= PTR_ERR(handle
);
2388 printk(KERN_CRIT
"%s: jbd2_start: "
2389 "%ld pages, ino %lu; err %d\n", __func__
,
2390 wbc
->nr_to_write
, inode
->i_ino
, ret
);
2392 goto out_writepages
;
2394 if (ext4_should_order_data(inode
)) {
2396 * With ordered mode we need to add
2397 * the inode to the journal handl
2398 * when we do block allocation.
2400 ret
= ext4_jbd2_file_inode(handle
, inode
);
2402 ext4_journal_stop(handle
);
2403 goto out_writepages
;
2407 to_write
-= wbc
->nr_to_write
;
2408 ret
= mpage_da_writepages(mapping
, wbc
,
2409 ext4_da_get_block_write
);
2410 ext4_journal_stop(handle
);
2411 if (ret
== MPAGE_DA_EXTENT_TAIL
) {
2413 * got one extent now try with
2416 to_write
+= wbc
->nr_to_write
;
2418 } else if (wbc
->nr_to_write
) {
2420 * There is no more writeout needed
2421 * or we requested for a noblocking writeout
2422 * and we found the device congested
2424 to_write
+= wbc
->nr_to_write
;
2427 wbc
->nr_to_write
= to_write
;
2430 if (wbc
->range_cont
&& (pages_skipped
!= wbc
->pages_skipped
)) {
2431 /* We skipped pages in this loop */
2432 wbc
->range_start
= range_start
;
2433 wbc
->nr_to_write
= to_write
+
2434 wbc
->pages_skipped
- pages_skipped
;
2435 wbc
->pages_skipped
= pages_skipped
;
2440 wbc
->nr_to_write
= to_write
- nr_to_writebump
;
2441 wbc
->range_start
= range_start
;
2445 static int ext4_da_write_begin(struct file
*file
, struct address_space
*mapping
,
2446 loff_t pos
, unsigned len
, unsigned flags
,
2447 struct page
**pagep
, void **fsdata
)
2449 int ret
, retries
= 0;
2453 struct inode
*inode
= mapping
->host
;
2456 index
= pos
>> PAGE_CACHE_SHIFT
;
2457 from
= pos
& (PAGE_CACHE_SIZE
- 1);
2462 * With delayed allocation, we don't log the i_disksize update
2463 * if there is delayed block allocation. But we still need
2464 * to journalling the i_disksize update if writes to the end
2465 * of file which has an already mapped buffer.
2467 handle
= ext4_journal_start(inode
, 1);
2468 if (IS_ERR(handle
)) {
2469 ret
= PTR_ERR(handle
);
2472 /* We cannot recurse into the filesystem as the transaction is already
2474 flags
|= AOP_FLAG_NOFS
;
2476 page
= grab_cache_page_write_begin(mapping
, index
, flags
);
2478 ext4_journal_stop(handle
);
2484 ret
= block_write_begin(file
, mapping
, pos
, len
, flags
, pagep
, fsdata
,
2485 ext4_da_get_block_prep
);
2488 ext4_journal_stop(handle
);
2489 page_cache_release(page
);
2492 if (ret
== -ENOSPC
&& ext4_should_retry_alloc(inode
->i_sb
, &retries
))
2499 * Check if we should update i_disksize
2500 * when write to the end of file but not require block allocation
2502 static int ext4_da_should_update_i_disksize(struct page
*page
,
2503 unsigned long offset
)
2505 struct buffer_head
*bh
;
2506 struct inode
*inode
= page
->mapping
->host
;
2510 bh
= page_buffers(page
);
2511 idx
= offset
>> inode
->i_blkbits
;
2513 for (i
=0; i
< idx
; i
++)
2514 bh
= bh
->b_this_page
;
2516 if (!buffer_mapped(bh
) || (buffer_delay(bh
)))
2521 static int ext4_da_write_end(struct file
*file
,
2522 struct address_space
*mapping
,
2523 loff_t pos
, unsigned len
, unsigned copied
,
2524 struct page
*page
, void *fsdata
)
2526 struct inode
*inode
= mapping
->host
;
2528 handle_t
*handle
= ext4_journal_current_handle();
2530 unsigned long start
, end
;
2532 start
= pos
& (PAGE_CACHE_SIZE
- 1);
2533 end
= start
+ copied
-1;
2536 * generic_write_end() will run mark_inode_dirty() if i_size
2537 * changes. So let's piggyback the i_disksize mark_inode_dirty
2541 new_i_size
= pos
+ copied
;
2542 if (new_i_size
> EXT4_I(inode
)->i_disksize
) {
2543 if (ext4_da_should_update_i_disksize(page
, end
)) {
2544 down_write(&EXT4_I(inode
)->i_data_sem
);
2545 if (new_i_size
> EXT4_I(inode
)->i_disksize
) {
2547 * Updating i_disksize when extending file
2548 * without needing block allocation
2550 if (ext4_should_order_data(inode
))
2551 ret
= ext4_jbd2_file_inode(handle
,
2554 EXT4_I(inode
)->i_disksize
= new_i_size
;
2556 up_write(&EXT4_I(inode
)->i_data_sem
);
2559 ret2
= generic_write_end(file
, mapping
, pos
, len
, copied
,
2564 ret2
= ext4_journal_stop(handle
);
2568 return ret
? ret
: copied
;
2571 static void ext4_da_invalidatepage(struct page
*page
, unsigned long offset
)
2574 * Drop reserved blocks
2576 BUG_ON(!PageLocked(page
));
2577 if (!page_has_buffers(page
))
2580 ext4_da_page_release_reservation(page
, offset
);
2583 ext4_invalidatepage(page
, offset
);
2589 * Force all delayed allocation blocks to be allocated for a given inode.
2591 int ext4_alloc_da_blocks(struct inode
*inode
)
2593 if (!EXT4_I(inode
)->i_reserved_data_blocks
&&
2594 !EXT4_I(inode
)->i_reserved_meta_blocks
)
2598 * We do something simple for now. The filemap_flush() will
2599 * also start triggering a write of the data blocks, which is
2600 * not strictly speaking necessary (and for users of
2601 * laptop_mode, not even desirable). However, to do otherwise
2602 * would require replicating code paths in:
2604 * ext4_da_writepages() ->
2605 * write_cache_pages() ---> (via passed in callback function)
2606 * __mpage_da_writepage() -->
2607 * mpage_add_bh_to_extent()
2608 * mpage_da_map_blocks()
2610 * The problem is that write_cache_pages(), located in
2611 * mm/page-writeback.c, marks pages clean in preparation for
2612 * doing I/O, which is not desirable if we're not planning on
2615 * We could call write_cache_pages(), and then redirty all of
2616 * the pages by calling redirty_page_for_writeback() but that
2617 * would be ugly in the extreme. So instead we would need to
2618 * replicate parts of the code in the above functions,
2619 * simplifying them becuase we wouldn't actually intend to
2620 * write out the pages, but rather only collect contiguous
2621 * logical block extents, call the multi-block allocator, and
2622 * then update the buffer heads with the block allocations.
2624 * For now, though, we'll cheat by calling filemap_flush(),
2625 * which will map the blocks, and start the I/O, but not
2626 * actually wait for the I/O to complete.
2628 return filemap_flush(inode
->i_mapping
);
2632 * bmap() is special. It gets used by applications such as lilo and by
2633 * the swapper to find the on-disk block of a specific piece of data.
2635 * Naturally, this is dangerous if the block concerned is still in the
2636 * journal. If somebody makes a swapfile on an ext4 data-journaling
2637 * filesystem and enables swap, then they may get a nasty shock when the
2638 * data getting swapped to that swapfile suddenly gets overwritten by
2639 * the original zero's written out previously to the journal and
2640 * awaiting writeback in the kernel's buffer cache.
2642 * So, if we see any bmap calls here on a modified, data-journaled file,
2643 * take extra steps to flush any blocks which might be in the cache.
2645 static sector_t
ext4_bmap(struct address_space
*mapping
, sector_t block
)
2647 struct inode
*inode
= mapping
->host
;
2651 if (mapping_tagged(mapping
, PAGECACHE_TAG_DIRTY
) &&
2652 test_opt(inode
->i_sb
, DELALLOC
)) {
2654 * With delalloc we want to sync the file
2655 * so that we can make sure we allocate
2658 filemap_write_and_wait(mapping
);
2661 if (EXT4_I(inode
)->i_state
& EXT4_STATE_JDATA
) {
2663 * This is a REALLY heavyweight approach, but the use of
2664 * bmap on dirty files is expected to be extremely rare:
2665 * only if we run lilo or swapon on a freshly made file
2666 * do we expect this to happen.
2668 * (bmap requires CAP_SYS_RAWIO so this does not
2669 * represent an unprivileged user DOS attack --- we'd be
2670 * in trouble if mortal users could trigger this path at
2673 * NB. EXT4_STATE_JDATA is not set on files other than
2674 * regular files. If somebody wants to bmap a directory
2675 * or symlink and gets confused because the buffer
2676 * hasn't yet been flushed to disk, they deserve
2677 * everything they get.
2680 EXT4_I(inode
)->i_state
&= ~EXT4_STATE_JDATA
;
2681 journal
= EXT4_JOURNAL(inode
);
2682 jbd2_journal_lock_updates(journal
);
2683 err
= jbd2_journal_flush(journal
);
2684 jbd2_journal_unlock_updates(journal
);
2690 return generic_block_bmap(mapping
,block
,ext4_get_block
);
2693 static int bget_one(handle_t
*handle
, struct buffer_head
*bh
)
2699 static int bput_one(handle_t
*handle
, struct buffer_head
*bh
)
2706 * Note that we don't need to start a transaction unless we're journaling data
2707 * because we should have holes filled from ext4_page_mkwrite(). We even don't
2708 * need to file the inode to the transaction's list in ordered mode because if
2709 * we are writing back data added by write(), the inode is already there and if
2710 * we are writing back data modified via mmap(), noone guarantees in which
2711 * transaction the data will hit the disk. In case we are journaling data, we
2712 * cannot start transaction directly because transaction start ranks above page
2713 * lock so we have to do some magic.
2715 * In all journaling modes block_write_full_page() will start the I/O.
2719 * ext4_writepage() -> kmalloc() -> __alloc_pages() -> page_launder() ->
2724 * ext4_file_write() -> generic_file_write() -> __alloc_pages() -> ...
2726 * Same applies to ext4_get_block(). We will deadlock on various things like
2727 * lock_journal and i_data_sem
2729 * Setting PF_MEMALLOC here doesn't work - too many internal memory
2732 * 16May01: If we're reentered then journal_current_handle() will be
2733 * non-zero. We simply *return*.
2735 * 1 July 2001: @@@ FIXME:
2736 * In journalled data mode, a data buffer may be metadata against the
2737 * current transaction. But the same file is part of a shared mapping
2738 * and someone does a writepage() on it.
2740 * We will move the buffer onto the async_data list, but *after* it has
2741 * been dirtied. So there's a small window where we have dirty data on
2744 * Note that this only applies to the last partial page in the file. The
2745 * bit which block_write_full_page() uses prepare/commit for. (That's
2746 * broken code anyway: it's wrong for msync()).
2748 * It's a rare case: affects the final partial page, for journalled data
2749 * where the file is subject to bith write() and writepage() in the same
2750 * transction. To fix it we'll need a custom block_write_full_page().
2751 * We'll probably need that anyway for journalling writepage() output.
2753 * We don't honour synchronous mounts for writepage(). That would be
2754 * disastrous. Any write() or metadata operation will sync the fs for
2758 static int __ext4_normal_writepage(struct page
*page
,
2759 struct writeback_control
*wbc
)
2761 struct inode
*inode
= page
->mapping
->host
;
2763 if (test_opt(inode
->i_sb
, NOBH
))
2764 return nobh_writepage(page
,
2765 ext4_normal_get_block_write
, wbc
);
2767 return block_write_full_page(page
,
2768 ext4_normal_get_block_write
,
2772 static int ext4_normal_writepage(struct page
*page
,
2773 struct writeback_control
*wbc
)
2775 struct inode
*inode
= page
->mapping
->host
;
2776 loff_t size
= i_size_read(inode
);
2779 J_ASSERT(PageLocked(page
));
2780 if (page
->index
== size
>> PAGE_CACHE_SHIFT
)
2781 len
= size
& ~PAGE_CACHE_MASK
;
2783 len
= PAGE_CACHE_SIZE
;
2785 if (page_has_buffers(page
)) {
2786 /* if page has buffers it should all be mapped
2787 * and allocated. If there are not buffers attached
2788 * to the page we know the page is dirty but it lost
2789 * buffers. That means that at some moment in time
2790 * after write_begin() / write_end() has been called
2791 * all buffers have been clean and thus they must have been
2792 * written at least once. So they are all mapped and we can
2793 * happily proceed with mapping them and writing the page.
2795 BUG_ON(walk_page_buffers(NULL
, page_buffers(page
), 0, len
, NULL
,
2796 ext4_bh_unmapped_or_delay
));
2799 if (!ext4_journal_current_handle())
2800 return __ext4_normal_writepage(page
, wbc
);
2802 redirty_page_for_writepage(wbc
, page
);
2807 static int __ext4_journalled_writepage(struct page
*page
,
2808 struct writeback_control
*wbc
)
2810 struct address_space
*mapping
= page
->mapping
;
2811 struct inode
*inode
= mapping
->host
;
2812 struct buffer_head
*page_bufs
;
2813 handle_t
*handle
= NULL
;
2817 ret
= block_prepare_write(page
, 0, PAGE_CACHE_SIZE
,
2818 ext4_normal_get_block_write
);
2822 page_bufs
= page_buffers(page
);
2823 walk_page_buffers(handle
, page_bufs
, 0, PAGE_CACHE_SIZE
, NULL
,
2825 /* As soon as we unlock the page, it can go away, but we have
2826 * references to buffers so we are safe */
2829 handle
= ext4_journal_start(inode
, ext4_writepage_trans_blocks(inode
));
2830 if (IS_ERR(handle
)) {
2831 ret
= PTR_ERR(handle
);
2835 ret
= walk_page_buffers(handle
, page_bufs
, 0,
2836 PAGE_CACHE_SIZE
, NULL
, do_journal_get_write_access
);
2838 err
= walk_page_buffers(handle
, page_bufs
, 0,
2839 PAGE_CACHE_SIZE
, NULL
, write_end_fn
);
2842 err
= ext4_journal_stop(handle
);
2846 walk_page_buffers(handle
, page_bufs
, 0,
2847 PAGE_CACHE_SIZE
, NULL
, bput_one
);
2848 EXT4_I(inode
)->i_state
|= EXT4_STATE_JDATA
;
2857 static int ext4_journalled_writepage(struct page
*page
,
2858 struct writeback_control
*wbc
)
2860 struct inode
*inode
= page
->mapping
->host
;
2861 loff_t size
= i_size_read(inode
);
2864 J_ASSERT(PageLocked(page
));
2865 if (page
->index
== size
>> PAGE_CACHE_SHIFT
)
2866 len
= size
& ~PAGE_CACHE_MASK
;
2868 len
= PAGE_CACHE_SIZE
;
2870 if (page_has_buffers(page
)) {
2871 /* if page has buffers it should all be mapped
2872 * and allocated. If there are not buffers attached
2873 * to the page we know the page is dirty but it lost
2874 * buffers. That means that at some moment in time
2875 * after write_begin() / write_end() has been called
2876 * all buffers have been clean and thus they must have been
2877 * written at least once. So they are all mapped and we can
2878 * happily proceed with mapping them and writing the page.
2880 BUG_ON(walk_page_buffers(NULL
, page_buffers(page
), 0, len
, NULL
,
2881 ext4_bh_unmapped_or_delay
));
2884 if (ext4_journal_current_handle())
2887 if (PageChecked(page
)) {
2889 * It's mmapped pagecache. Add buffers and journal it. There
2890 * doesn't seem much point in redirtying the page here.
2892 ClearPageChecked(page
);
2893 return __ext4_journalled_writepage(page
, wbc
);
2896 * It may be a page full of checkpoint-mode buffers. We don't
2897 * really know unless we go poke around in the buffer_heads.
2898 * But block_write_full_page will do the right thing.
2900 return block_write_full_page(page
,
2901 ext4_normal_get_block_write
,
2905 redirty_page_for_writepage(wbc
, page
);
2910 static int ext4_readpage(struct file
*file
, struct page
*page
)
2912 return mpage_readpage(page
, ext4_get_block
);
2916 ext4_readpages(struct file
*file
, struct address_space
*mapping
,
2917 struct list_head
*pages
, unsigned nr_pages
)
2919 return mpage_readpages(mapping
, pages
, nr_pages
, ext4_get_block
);
2922 static void ext4_invalidatepage(struct page
*page
, unsigned long offset
)
2924 journal_t
*journal
= EXT4_JOURNAL(page
->mapping
->host
);
2927 * If it's a full truncate we just forget about the pending dirtying
2930 ClearPageChecked(page
);
2932 jbd2_journal_invalidatepage(journal
, page
, offset
);
2935 static int ext4_releasepage(struct page
*page
, gfp_t wait
)
2937 journal_t
*journal
= EXT4_JOURNAL(page
->mapping
->host
);
2939 WARN_ON(PageChecked(page
));
2940 if (!page_has_buffers(page
))
2942 return jbd2_journal_try_to_free_buffers(journal
, page
, wait
);
2946 * If the O_DIRECT write will extend the file then add this inode to the
2947 * orphan list. So recovery will truncate it back to the original size
2948 * if the machine crashes during the write.
2950 * If the O_DIRECT write is intantiating holes inside i_size and the machine
2951 * crashes then stale disk data _may_ be exposed inside the file. But current
2952 * VFS code falls back into buffered path in that case so we are safe.
2954 static ssize_t
ext4_direct_IO(int rw
, struct kiocb
*iocb
,
2955 const struct iovec
*iov
, loff_t offset
,
2956 unsigned long nr_segs
)
2958 struct file
*file
= iocb
->ki_filp
;
2959 struct inode
*inode
= file
->f_mapping
->host
;
2960 struct ext4_inode_info
*ei
= EXT4_I(inode
);
2964 size_t count
= iov_length(iov
, nr_segs
);
2967 loff_t final_size
= offset
+ count
;
2969 if (final_size
> inode
->i_size
) {
2970 /* Credits for sb + inode write */
2971 handle
= ext4_journal_start(inode
, 2);
2972 if (IS_ERR(handle
)) {
2973 ret
= PTR_ERR(handle
);
2976 ret
= ext4_orphan_add(handle
, inode
);
2978 ext4_journal_stop(handle
);
2982 ei
->i_disksize
= inode
->i_size
;
2983 ext4_journal_stop(handle
);
2987 ret
= blockdev_direct_IO(rw
, iocb
, inode
, inode
->i_sb
->s_bdev
, iov
,
2989 ext4_get_block
, NULL
);
2994 /* Credits for sb + inode write */
2995 handle
= ext4_journal_start(inode
, 2);
2996 if (IS_ERR(handle
)) {
2997 /* This is really bad luck. We've written the data
2998 * but cannot extend i_size. Bail out and pretend
2999 * the write failed... */
3000 ret
= PTR_ERR(handle
);
3004 ext4_orphan_del(handle
, inode
);
3006 loff_t end
= offset
+ ret
;
3007 if (end
> inode
->i_size
) {
3008 ei
->i_disksize
= end
;
3009 i_size_write(inode
, end
);
3011 * We're going to return a positive `ret'
3012 * here due to non-zero-length I/O, so there's
3013 * no way of reporting error returns from
3014 * ext4_mark_inode_dirty() to userspace. So
3017 ext4_mark_inode_dirty(handle
, inode
);
3020 err
= ext4_journal_stop(handle
);
3029 * Pages can be marked dirty completely asynchronously from ext4's journalling
3030 * activity. By filemap_sync_pte(), try_to_unmap_one(), etc. We cannot do
3031 * much here because ->set_page_dirty is called under VFS locks. The page is
3032 * not necessarily locked.
3034 * We cannot just dirty the page and leave attached buffers clean, because the
3035 * buffers' dirty state is "definitive". We cannot just set the buffers dirty
3036 * or jbddirty because all the journalling code will explode.
3038 * So what we do is to mark the page "pending dirty" and next time writepage
3039 * is called, propagate that into the buffers appropriately.
3041 static int ext4_journalled_set_page_dirty(struct page
*page
)
3043 SetPageChecked(page
);
3044 return __set_page_dirty_nobuffers(page
);
3047 static const struct address_space_operations ext4_ordered_aops
= {
3048 .readpage
= ext4_readpage
,
3049 .readpages
= ext4_readpages
,
3050 .writepage
= ext4_normal_writepage
,
3051 .sync_page
= block_sync_page
,
3052 .write_begin
= ext4_write_begin
,
3053 .write_end
= ext4_ordered_write_end
,
3055 .invalidatepage
= ext4_invalidatepage
,
3056 .releasepage
= ext4_releasepage
,
3057 .direct_IO
= ext4_direct_IO
,
3058 .migratepage
= buffer_migrate_page
,
3059 .is_partially_uptodate
= block_is_partially_uptodate
,
3062 static const struct address_space_operations ext4_writeback_aops
= {
3063 .readpage
= ext4_readpage
,
3064 .readpages
= ext4_readpages
,
3065 .writepage
= ext4_normal_writepage
,
3066 .sync_page
= block_sync_page
,
3067 .write_begin
= ext4_write_begin
,
3068 .write_end
= ext4_writeback_write_end
,
3070 .invalidatepage
= ext4_invalidatepage
,
3071 .releasepage
= ext4_releasepage
,
3072 .direct_IO
= ext4_direct_IO
,
3073 .migratepage
= buffer_migrate_page
,
3074 .is_partially_uptodate
= block_is_partially_uptodate
,
3077 static const struct address_space_operations ext4_journalled_aops
= {
3078 .readpage
= ext4_readpage
,
3079 .readpages
= ext4_readpages
,
3080 .writepage
= ext4_journalled_writepage
,
3081 .sync_page
= block_sync_page
,
3082 .write_begin
= ext4_write_begin
,
3083 .write_end
= ext4_journalled_write_end
,
3084 .set_page_dirty
= ext4_journalled_set_page_dirty
,
3086 .invalidatepage
= ext4_invalidatepage
,
3087 .releasepage
= ext4_releasepage
,
3088 .is_partially_uptodate
= block_is_partially_uptodate
,
3091 static const struct address_space_operations ext4_da_aops
= {
3092 .readpage
= ext4_readpage
,
3093 .readpages
= ext4_readpages
,
3094 .writepage
= ext4_da_writepage
,
3095 .writepages
= ext4_da_writepages
,
3096 .sync_page
= block_sync_page
,
3097 .write_begin
= ext4_da_write_begin
,
3098 .write_end
= ext4_da_write_end
,
3100 .invalidatepage
= ext4_da_invalidatepage
,
3101 .releasepage
= ext4_releasepage
,
3102 .direct_IO
= ext4_direct_IO
,
3103 .migratepage
= buffer_migrate_page
,
3104 .is_partially_uptodate
= block_is_partially_uptodate
,
3107 void ext4_set_aops(struct inode
*inode
)
3109 if (ext4_should_order_data(inode
) &&
3110 test_opt(inode
->i_sb
, DELALLOC
))
3111 inode
->i_mapping
->a_ops
= &ext4_da_aops
;
3112 else if (ext4_should_order_data(inode
))
3113 inode
->i_mapping
->a_ops
= &ext4_ordered_aops
;
3114 else if (ext4_should_writeback_data(inode
) &&
3115 test_opt(inode
->i_sb
, DELALLOC
))
3116 inode
->i_mapping
->a_ops
= &ext4_da_aops
;
3117 else if (ext4_should_writeback_data(inode
))
3118 inode
->i_mapping
->a_ops
= &ext4_writeback_aops
;
3120 inode
->i_mapping
->a_ops
= &ext4_journalled_aops
;
3124 * ext4_block_truncate_page() zeroes out a mapping from file offset `from'
3125 * up to the end of the block which corresponds to `from'.
3126 * This required during truncate. We need to physically zero the tail end
3127 * of that block so it doesn't yield old data if the file is later grown.
3129 int ext4_block_truncate_page(handle_t
*handle
,
3130 struct address_space
*mapping
, loff_t from
)
3132 ext4_fsblk_t index
= from
>> PAGE_CACHE_SHIFT
;
3133 unsigned offset
= from
& (PAGE_CACHE_SIZE
-1);
3134 unsigned blocksize
, length
, pos
;
3136 struct inode
*inode
= mapping
->host
;
3137 struct buffer_head
*bh
;
3141 page
= grab_cache_page(mapping
, from
>> PAGE_CACHE_SHIFT
);
3145 blocksize
= inode
->i_sb
->s_blocksize
;
3146 length
= blocksize
- (offset
& (blocksize
- 1));
3147 iblock
= index
<< (PAGE_CACHE_SHIFT
- inode
->i_sb
->s_blocksize_bits
);
3150 * For "nobh" option, we can only work if we don't need to
3151 * read-in the page - otherwise we create buffers to do the IO.
3153 if (!page_has_buffers(page
) && test_opt(inode
->i_sb
, NOBH
) &&
3154 ext4_should_writeback_data(inode
) && PageUptodate(page
)) {
3155 zero_user(page
, offset
, length
);
3156 set_page_dirty(page
);
3160 if (!page_has_buffers(page
))
3161 create_empty_buffers(page
, blocksize
, 0);
3163 /* Find the buffer that contains "offset" */
3164 bh
= page_buffers(page
);
3166 while (offset
>= pos
) {
3167 bh
= bh
->b_this_page
;
3173 if (buffer_freed(bh
)) {
3174 BUFFER_TRACE(bh
, "freed: skip");
3178 if (!buffer_mapped(bh
)) {
3179 BUFFER_TRACE(bh
, "unmapped");
3180 ext4_get_block(inode
, iblock
, bh
, 0);
3181 /* unmapped? It's a hole - nothing to do */
3182 if (!buffer_mapped(bh
)) {
3183 BUFFER_TRACE(bh
, "still unmapped");
3188 /* Ok, it's mapped. Make sure it's up-to-date */
3189 if (PageUptodate(page
))
3190 set_buffer_uptodate(bh
);
3192 if (!buffer_uptodate(bh
)) {
3194 ll_rw_block(READ
, 1, &bh
);
3196 /* Uhhuh. Read error. Complain and punt. */
3197 if (!buffer_uptodate(bh
))
3201 if (ext4_should_journal_data(inode
)) {
3202 BUFFER_TRACE(bh
, "get write access");
3203 err
= ext4_journal_get_write_access(handle
, bh
);
3208 zero_user(page
, offset
, length
);
3210 BUFFER_TRACE(bh
, "zeroed end of block");
3213 if (ext4_should_journal_data(inode
)) {
3214 err
= ext4_journal_dirty_metadata(handle
, bh
);
3216 if (ext4_should_order_data(inode
))
3217 err
= ext4_jbd2_file_inode(handle
, inode
);
3218 mark_buffer_dirty(bh
);
3223 page_cache_release(page
);
3228 * Probably it should be a library function... search for first non-zero word
3229 * or memcmp with zero_page, whatever is better for particular architecture.
3232 static inline int all_zeroes(__le32
*p
, __le32
*q
)
3241 * ext4_find_shared - find the indirect blocks for partial truncation.
3242 * @inode: inode in question
3243 * @depth: depth of the affected branch
3244 * @offsets: offsets of pointers in that branch (see ext4_block_to_path)
3245 * @chain: place to store the pointers to partial indirect blocks
3246 * @top: place to the (detached) top of branch
3248 * This is a helper function used by ext4_truncate().
3250 * When we do truncate() we may have to clean the ends of several
3251 * indirect blocks but leave the blocks themselves alive. Block is
3252 * partially truncated if some data below the new i_size is refered
3253 * from it (and it is on the path to the first completely truncated
3254 * data block, indeed). We have to free the top of that path along
3255 * with everything to the right of the path. Since no allocation
3256 * past the truncation point is possible until ext4_truncate()
3257 * finishes, we may safely do the latter, but top of branch may
3258 * require special attention - pageout below the truncation point
3259 * might try to populate it.
3261 * We atomically detach the top of branch from the tree, store the
3262 * block number of its root in *@top, pointers to buffer_heads of
3263 * partially truncated blocks - in @chain[].bh and pointers to
3264 * their last elements that should not be removed - in
3265 * @chain[].p. Return value is the pointer to last filled element
3268 * The work left to caller to do the actual freeing of subtrees:
3269 * a) free the subtree starting from *@top
3270 * b) free the subtrees whose roots are stored in
3271 * (@chain[i].p+1 .. end of @chain[i].bh->b_data)
3272 * c) free the subtrees growing from the inode past the @chain[0].
3273 * (no partially truncated stuff there). */
3275 static Indirect
*ext4_find_shared(struct inode
*inode
, int depth
,
3276 ext4_lblk_t offsets
[4], Indirect chain
[4], __le32
*top
)
3278 Indirect
*partial
, *p
;
3282 /* Make k index the deepest non-null offest + 1 */
3283 for (k
= depth
; k
> 1 && !offsets
[k
-1]; k
--)
3285 partial
= ext4_get_branch(inode
, k
, offsets
, chain
, &err
);
3286 /* Writer: pointers */
3288 partial
= chain
+ k
-1;
3290 * If the branch acquired continuation since we've looked at it -
3291 * fine, it should all survive and (new) top doesn't belong to us.
3293 if (!partial
->key
&& *partial
->p
)
3296 for (p
=partial
; p
>chain
&& all_zeroes((__le32
*)p
->bh
->b_data
,p
->p
); p
--)
3299 * OK, we've found the last block that must survive. The rest of our
3300 * branch should be detached before unlocking. However, if that rest
3301 * of branch is all ours and does not grow immediately from the inode
3302 * it's easier to cheat and just decrement partial->p.
3304 if (p
== chain
+ k
- 1 && p
> chain
) {
3308 /* Nope, don't do this in ext4. Must leave the tree intact */
3315 while(partial
> p
) {
3316 brelse(partial
->bh
);
3324 * Zero a number of block pointers in either an inode or an indirect block.
3325 * If we restart the transaction we must again get write access to the
3326 * indirect block for further modification.
3328 * We release `count' blocks on disk, but (last - first) may be greater
3329 * than `count' because there can be holes in there.
3331 static void ext4_clear_blocks(handle_t
*handle
, struct inode
*inode
,
3332 struct buffer_head
*bh
, ext4_fsblk_t block_to_free
,
3333 unsigned long count
, __le32
*first
, __le32
*last
)
3336 if (try_to_extend_transaction(handle
, inode
)) {
3338 BUFFER_TRACE(bh
, "call ext4_journal_dirty_metadata");
3339 ext4_journal_dirty_metadata(handle
, bh
);
3341 ext4_mark_inode_dirty(handle
, inode
);
3342 ext4_journal_test_restart(handle
, inode
);
3344 BUFFER_TRACE(bh
, "retaking write access");
3345 ext4_journal_get_write_access(handle
, bh
);
3350 * Any buffers which are on the journal will be in memory. We find
3351 * them on the hash table so jbd2_journal_revoke() will run jbd2_journal_forget()
3352 * on them. We've already detached each block from the file, so
3353 * bforget() in jbd2_journal_forget() should be safe.
3355 * AKPM: turn on bforget in jbd2_journal_forget()!!!
3357 for (p
= first
; p
< last
; p
++) {
3358 u32 nr
= le32_to_cpu(*p
);
3360 struct buffer_head
*tbh
;
3363 tbh
= sb_find_get_block(inode
->i_sb
, nr
);
3364 ext4_forget(handle
, 0, inode
, tbh
, nr
);
3368 ext4_free_blocks(handle
, inode
, block_to_free
, count
, 0);
3372 * ext4_free_data - free a list of data blocks
3373 * @handle: handle for this transaction
3374 * @inode: inode we are dealing with
3375 * @this_bh: indirect buffer_head which contains *@first and *@last
3376 * @first: array of block numbers
3377 * @last: points immediately past the end of array
3379 * We are freeing all blocks refered from that array (numbers are stored as
3380 * little-endian 32-bit) and updating @inode->i_blocks appropriately.
3382 * We accumulate contiguous runs of blocks to free. Conveniently, if these
3383 * blocks are contiguous then releasing them at one time will only affect one
3384 * or two bitmap blocks (+ group descriptor(s) and superblock) and we won't
3385 * actually use a lot of journal space.
3387 * @this_bh will be %NULL if @first and @last point into the inode's direct
3390 static void ext4_free_data(handle_t
*handle
, struct inode
*inode
,
3391 struct buffer_head
*this_bh
,
3392 __le32
*first
, __le32
*last
)
3394 ext4_fsblk_t block_to_free
= 0; /* Starting block # of a run */
3395 unsigned long count
= 0; /* Number of blocks in the run */
3396 __le32
*block_to_free_p
= NULL
; /* Pointer into inode/ind
3399 ext4_fsblk_t nr
; /* Current block # */
3400 __le32
*p
; /* Pointer into inode/ind
3401 for current block */
3404 if (this_bh
) { /* For indirect block */
3405 BUFFER_TRACE(this_bh
, "get_write_access");
3406 err
= ext4_journal_get_write_access(handle
, this_bh
);
3407 /* Important: if we can't update the indirect pointers
3408 * to the blocks, we can't free them. */
3413 for (p
= first
; p
< last
; p
++) {
3414 nr
= le32_to_cpu(*p
);
3416 /* accumulate blocks to free if they're contiguous */
3419 block_to_free_p
= p
;
3421 } else if (nr
== block_to_free
+ count
) {
3424 ext4_clear_blocks(handle
, inode
, this_bh
,
3426 count
, block_to_free_p
, p
);
3428 block_to_free_p
= p
;
3435 ext4_clear_blocks(handle
, inode
, this_bh
, block_to_free
,
3436 count
, block_to_free_p
, p
);
3439 BUFFER_TRACE(this_bh
, "call ext4_journal_dirty_metadata");
3442 * The buffer head should have an attached journal head at this
3443 * point. However, if the data is corrupted and an indirect
3444 * block pointed to itself, it would have been detached when
3445 * the block was cleared. Check for this instead of OOPSing.
3448 ext4_journal_dirty_metadata(handle
, this_bh
);
3450 ext4_error(inode
->i_sb
, __func__
,
3451 "circular indirect block detected, "
3452 "inode=%lu, block=%llu",
3454 (unsigned long long) this_bh
->b_blocknr
);
3459 * ext4_free_branches - free an array of branches
3460 * @handle: JBD handle for this transaction
3461 * @inode: inode we are dealing with
3462 * @parent_bh: the buffer_head which contains *@first and *@last
3463 * @first: array of block numbers
3464 * @last: pointer immediately past the end of array
3465 * @depth: depth of the branches to free
3467 * We are freeing all blocks refered from these branches (numbers are
3468 * stored as little-endian 32-bit) and updating @inode->i_blocks
3471 static void ext4_free_branches(handle_t
*handle
, struct inode
*inode
,
3472 struct buffer_head
*parent_bh
,
3473 __le32
*first
, __le32
*last
, int depth
)
3478 if (is_handle_aborted(handle
))
3482 struct buffer_head
*bh
;
3483 int addr_per_block
= EXT4_ADDR_PER_BLOCK(inode
->i_sb
);
3485 while (--p
>= first
) {
3486 nr
= le32_to_cpu(*p
);
3488 continue; /* A hole */
3490 /* Go read the buffer for the next level down */
3491 bh
= sb_bread(inode
->i_sb
, nr
);
3494 * A read failure? Report error and clear slot
3498 ext4_error(inode
->i_sb
, "ext4_free_branches",
3499 "Read failure, inode=%lu, block=%llu",
3504 /* This zaps the entire block. Bottom up. */
3505 BUFFER_TRACE(bh
, "free child branches");
3506 ext4_free_branches(handle
, inode
, bh
,
3507 (__le32
*)bh
->b_data
,
3508 (__le32
*)bh
->b_data
+ addr_per_block
,
3512 * We've probably journalled the indirect block several
3513 * times during the truncate. But it's no longer
3514 * needed and we now drop it from the transaction via
3515 * jbd2_journal_revoke().
3517 * That's easy if it's exclusively part of this
3518 * transaction. But if it's part of the committing
3519 * transaction then jbd2_journal_forget() will simply
3520 * brelse() it. That means that if the underlying
3521 * block is reallocated in ext4_get_block(),
3522 * unmap_underlying_metadata() will find this block
3523 * and will try to get rid of it. damn, damn.
3525 * If this block has already been committed to the
3526 * journal, a revoke record will be written. And
3527 * revoke records must be emitted *before* clearing
3528 * this block's bit in the bitmaps.
3530 ext4_forget(handle
, 1, inode
, bh
, bh
->b_blocknr
);
3533 * Everything below this this pointer has been
3534 * released. Now let this top-of-subtree go.
3536 * We want the freeing of this indirect block to be
3537 * atomic in the journal with the updating of the
3538 * bitmap block which owns it. So make some room in
3541 * We zero the parent pointer *after* freeing its
3542 * pointee in the bitmaps, so if extend_transaction()
3543 * for some reason fails to put the bitmap changes and
3544 * the release into the same transaction, recovery
3545 * will merely complain about releasing a free block,
3546 * rather than leaking blocks.
3548 if (is_handle_aborted(handle
))
3550 if (try_to_extend_transaction(handle
, inode
)) {
3551 ext4_mark_inode_dirty(handle
, inode
);
3552 ext4_journal_test_restart(handle
, inode
);
3555 ext4_free_blocks(handle
, inode
, nr
, 1, 1);
3559 * The block which we have just freed is
3560 * pointed to by an indirect block: journal it
3562 BUFFER_TRACE(parent_bh
, "get_write_access");
3563 if (!ext4_journal_get_write_access(handle
,
3566 BUFFER_TRACE(parent_bh
,
3567 "call ext4_journal_dirty_metadata");
3568 ext4_journal_dirty_metadata(handle
,
3574 /* We have reached the bottom of the tree. */
3575 BUFFER_TRACE(parent_bh
, "free data blocks");
3576 ext4_free_data(handle
, inode
, parent_bh
, first
, last
);
3580 int ext4_can_truncate(struct inode
*inode
)
3582 if (IS_APPEND(inode
) || IS_IMMUTABLE(inode
))
3584 if (S_ISREG(inode
->i_mode
))
3586 if (S_ISDIR(inode
->i_mode
))
3588 if (S_ISLNK(inode
->i_mode
))
3589 return !ext4_inode_is_fast_symlink(inode
);
3596 * We block out ext4_get_block() block instantiations across the entire
3597 * transaction, and VFS/VM ensures that ext4_truncate() cannot run
3598 * simultaneously on behalf of the same inode.
3600 * As we work through the truncate and commmit bits of it to the journal there
3601 * is one core, guiding principle: the file's tree must always be consistent on
3602 * disk. We must be able to restart the truncate after a crash.
3604 * The file's tree may be transiently inconsistent in memory (although it
3605 * probably isn't), but whenever we close off and commit a journal transaction,
3606 * the contents of (the filesystem + the journal) must be consistent and
3607 * restartable. It's pretty simple, really: bottom up, right to left (although
3608 * left-to-right works OK too).
3610 * Note that at recovery time, journal replay occurs *before* the restart of
3611 * truncate against the orphan inode list.
3613 * The committed inode has the new, desired i_size (which is the same as
3614 * i_disksize in this case). After a crash, ext4_orphan_cleanup() will see
3615 * that this inode's truncate did not complete and it will again call
3616 * ext4_truncate() to have another go. So there will be instantiated blocks
3617 * to the right of the truncation point in a crashed ext4 filesystem. But
3618 * that's fine - as long as they are linked from the inode, the post-crash
3619 * ext4_truncate() run will find them and release them.
3621 void ext4_truncate(struct inode
*inode
)
3624 struct ext4_inode_info
*ei
= EXT4_I(inode
);
3625 __le32
*i_data
= ei
->i_data
;
3626 int addr_per_block
= EXT4_ADDR_PER_BLOCK(inode
->i_sb
);
3627 struct address_space
*mapping
= inode
->i_mapping
;
3628 ext4_lblk_t offsets
[4];
3633 ext4_lblk_t last_block
;
3634 unsigned blocksize
= inode
->i_sb
->s_blocksize
;
3636 if (!ext4_can_truncate(inode
))
3639 if (EXT4_I(inode
)->i_flags
& EXT4_EXTENTS_FL
) {
3640 ext4_ext_truncate(inode
);
3644 handle
= start_transaction(inode
);
3646 return; /* AKPM: return what? */
3648 last_block
= (inode
->i_size
+ blocksize
-1)
3649 >> EXT4_BLOCK_SIZE_BITS(inode
->i_sb
);
3651 if (inode
->i_size
& (blocksize
- 1))
3652 if (ext4_block_truncate_page(handle
, mapping
, inode
->i_size
))
3655 n
= ext4_block_to_path(inode
, last_block
, offsets
, NULL
);
3657 goto out_stop
; /* error */
3660 * OK. This truncate is going to happen. We add the inode to the
3661 * orphan list, so that if this truncate spans multiple transactions,
3662 * and we crash, we will resume the truncate when the filesystem
3663 * recovers. It also marks the inode dirty, to catch the new size.
3665 * Implication: the file must always be in a sane, consistent
3666 * truncatable state while each transaction commits.
3668 if (ext4_orphan_add(handle
, inode
))
3672 * From here we block out all ext4_get_block() callers who want to
3673 * modify the block allocation tree.
3675 down_write(&ei
->i_data_sem
);
3677 ext4_discard_reservation(inode
);
3680 * The orphan list entry will now protect us from any crash which
3681 * occurs before the truncate completes, so it is now safe to propagate
3682 * the new, shorter inode size (held for now in i_size) into the
3683 * on-disk inode. We do this via i_disksize, which is the value which
3684 * ext4 *really* writes onto the disk inode.
3686 ei
->i_disksize
= inode
->i_size
;
3688 if (n
== 1) { /* direct blocks */
3689 ext4_free_data(handle
, inode
, NULL
, i_data
+offsets
[0],
3690 i_data
+ EXT4_NDIR_BLOCKS
);
3694 partial
= ext4_find_shared(inode
, n
, offsets
, chain
, &nr
);
3695 /* Kill the top of shared branch (not detached) */
3697 if (partial
== chain
) {
3698 /* Shared branch grows from the inode */
3699 ext4_free_branches(handle
, inode
, NULL
,
3700 &nr
, &nr
+1, (chain
+n
-1) - partial
);
3703 * We mark the inode dirty prior to restart,
3704 * and prior to stop. No need for it here.
3707 /* Shared branch grows from an indirect block */
3708 BUFFER_TRACE(partial
->bh
, "get_write_access");
3709 ext4_free_branches(handle
, inode
, partial
->bh
,
3711 partial
->p
+1, (chain
+n
-1) - partial
);
3714 /* Clear the ends of indirect blocks on the shared branch */
3715 while (partial
> chain
) {
3716 ext4_free_branches(handle
, inode
, partial
->bh
, partial
->p
+ 1,
3717 (__le32
*)partial
->bh
->b_data
+addr_per_block
,
3718 (chain
+n
-1) - partial
);
3719 BUFFER_TRACE(partial
->bh
, "call brelse");
3720 brelse (partial
->bh
);
3724 /* Kill the remaining (whole) subtrees */
3725 switch (offsets
[0]) {
3727 nr
= i_data
[EXT4_IND_BLOCK
];
3729 ext4_free_branches(handle
, inode
, NULL
, &nr
, &nr
+1, 1);
3730 i_data
[EXT4_IND_BLOCK
] = 0;
3732 case EXT4_IND_BLOCK
:
3733 nr
= i_data
[EXT4_DIND_BLOCK
];
3735 ext4_free_branches(handle
, inode
, NULL
, &nr
, &nr
+1, 2);
3736 i_data
[EXT4_DIND_BLOCK
] = 0;
3738 case EXT4_DIND_BLOCK
:
3739 nr
= i_data
[EXT4_TIND_BLOCK
];
3741 ext4_free_branches(handle
, inode
, NULL
, &nr
, &nr
+1, 3);
3742 i_data
[EXT4_TIND_BLOCK
] = 0;
3744 case EXT4_TIND_BLOCK
:
3748 up_write(&ei
->i_data_sem
);
3749 inode
->i_mtime
= inode
->i_ctime
= ext4_current_time(inode
);
3750 ext4_mark_inode_dirty(handle
, inode
);
3753 * In a multi-transaction truncate, we only make the final transaction
3760 * If this was a simple ftruncate(), and the file will remain alive
3761 * then we need to clear up the orphan record which we created above.
3762 * However, if this was a real unlink then we were called by
3763 * ext4_delete_inode(), and we allow that function to clean up the
3764 * orphan info for us.
3767 ext4_orphan_del(handle
, inode
);
3769 ext4_journal_stop(handle
);
3772 static ext4_fsblk_t
ext4_get_inode_block(struct super_block
*sb
,
3773 unsigned long ino
, struct ext4_iloc
*iloc
)
3775 ext4_group_t block_group
;
3776 unsigned long offset
;
3778 struct ext4_group_desc
*gdp
;
3780 if (!ext4_valid_inum(sb
, ino
)) {
3782 * This error is already checked for in namei.c unless we are
3783 * looking at an NFS filehandle, in which case no error
3789 block_group
= (ino
- 1) / EXT4_INODES_PER_GROUP(sb
);
3790 gdp
= ext4_get_group_desc(sb
, block_group
, NULL
);
3795 * Figure out the offset within the block group inode table
3797 offset
= ((ino
- 1) % EXT4_INODES_PER_GROUP(sb
)) *
3798 EXT4_INODE_SIZE(sb
);
3799 block
= ext4_inode_table(sb
, gdp
) +
3800 (offset
>> EXT4_BLOCK_SIZE_BITS(sb
));
3802 iloc
->block_group
= block_group
;
3803 iloc
->offset
= offset
& (EXT4_BLOCK_SIZE(sb
) - 1);
3808 * ext4_get_inode_loc returns with an extra refcount against the inode's
3809 * underlying buffer_head on success. If 'in_mem' is true, we have all
3810 * data in memory that is needed to recreate the on-disk version of this
3813 static int __ext4_get_inode_loc(struct inode
*inode
,
3814 struct ext4_iloc
*iloc
, int in_mem
)
3817 struct buffer_head
*bh
;
3819 block
= ext4_get_inode_block(inode
->i_sb
, inode
->i_ino
, iloc
);
3823 bh
= sb_getblk(inode
->i_sb
, block
);
3825 ext4_error (inode
->i_sb
, "ext4_get_inode_loc",
3826 "unable to read inode block - "
3827 "inode=%lu, block=%llu",
3828 inode
->i_ino
, block
);
3831 if (!buffer_uptodate(bh
)) {
3835 * If the buffer has the write error flag, we have failed
3836 * to write out another inode in the same block. In this
3837 * case, we don't have to read the block because we may
3838 * read the old inode data successfully.
3840 if (buffer_write_io_error(bh
) && !buffer_uptodate(bh
))
3841 set_buffer_uptodate(bh
);
3843 if (buffer_uptodate(bh
)) {
3844 /* someone brought it uptodate while we waited */
3850 * If we have all information of the inode in memory and this
3851 * is the only valid inode in the block, we need not read the
3855 struct buffer_head
*bitmap_bh
;
3856 struct ext4_group_desc
*desc
;
3857 int inodes_per_buffer
;
3858 int inode_offset
, i
;
3859 ext4_group_t block_group
;
3862 block_group
= (inode
->i_ino
- 1) /
3863 EXT4_INODES_PER_GROUP(inode
->i_sb
);
3864 inodes_per_buffer
= bh
->b_size
/
3865 EXT4_INODE_SIZE(inode
->i_sb
);
3866 inode_offset
= ((inode
->i_ino
- 1) %
3867 EXT4_INODES_PER_GROUP(inode
->i_sb
));
3868 start
= inode_offset
& ~(inodes_per_buffer
- 1);
3870 /* Is the inode bitmap in cache? */
3871 desc
= ext4_get_group_desc(inode
->i_sb
,
3876 bitmap_bh
= sb_getblk(inode
->i_sb
,
3877 ext4_inode_bitmap(inode
->i_sb
, desc
));
3882 * If the inode bitmap isn't in cache then the
3883 * optimisation may end up performing two reads instead
3884 * of one, so skip it.
3886 if (!buffer_uptodate(bitmap_bh
)) {
3890 for (i
= start
; i
< start
+ inodes_per_buffer
; i
++) {
3891 if (i
== inode_offset
)
3893 if (ext4_test_bit(i
, bitmap_bh
->b_data
))
3897 if (i
== start
+ inodes_per_buffer
) {
3898 /* all other inodes are free, so skip I/O */
3899 memset(bh
->b_data
, 0, bh
->b_size
);
3900 set_buffer_uptodate(bh
);
3908 * There are other valid inodes in the buffer, this inode
3909 * has in-inode xattrs, or we don't have this inode in memory.
3910 * Read the block from disk.
3913 bh
->b_end_io
= end_buffer_read_sync
;
3914 submit_bh(READ_META
, bh
);
3916 if (!buffer_uptodate(bh
)) {
3917 ext4_error(inode
->i_sb
, "ext4_get_inode_loc",
3918 "unable to read inode block - "
3919 "inode=%lu, block=%llu",
3920 inode
->i_ino
, block
);
3930 int ext4_get_inode_loc(struct inode
*inode
, struct ext4_iloc
*iloc
)
3932 /* We have all inode data except xattrs in memory here. */
3933 return __ext4_get_inode_loc(inode
, iloc
,
3934 !(EXT4_I(inode
)->i_state
& EXT4_STATE_XATTR
));
3937 void ext4_set_inode_flags(struct inode
*inode
)
3939 unsigned int flags
= EXT4_I(inode
)->i_flags
;
3941 inode
->i_flags
&= ~(S_SYNC
|S_APPEND
|S_IMMUTABLE
|S_NOATIME
|S_DIRSYNC
);
3942 if (flags
& EXT4_SYNC_FL
)
3943 inode
->i_flags
|= S_SYNC
;
3944 if (flags
& EXT4_APPEND_FL
)
3945 inode
->i_flags
|= S_APPEND
;
3946 if (flags
& EXT4_IMMUTABLE_FL
)
3947 inode
->i_flags
|= S_IMMUTABLE
;
3948 if (flags
& EXT4_NOATIME_FL
)
3949 inode
->i_flags
|= S_NOATIME
;
3950 if (flags
& EXT4_DIRSYNC_FL
)
3951 inode
->i_flags
|= S_DIRSYNC
;
3954 /* Propagate flags from i_flags to EXT4_I(inode)->i_flags */
3955 void ext4_get_inode_flags(struct ext4_inode_info
*ei
)
3957 unsigned int flags
= ei
->vfs_inode
.i_flags
;
3959 ei
->i_flags
&= ~(EXT4_SYNC_FL
|EXT4_APPEND_FL
|
3960 EXT4_IMMUTABLE_FL
|EXT4_NOATIME_FL
|EXT4_DIRSYNC_FL
);
3962 ei
->i_flags
|= EXT4_SYNC_FL
;
3963 if (flags
& S_APPEND
)
3964 ei
->i_flags
|= EXT4_APPEND_FL
;
3965 if (flags
& S_IMMUTABLE
)
3966 ei
->i_flags
|= EXT4_IMMUTABLE_FL
;
3967 if (flags
& S_NOATIME
)
3968 ei
->i_flags
|= EXT4_NOATIME_FL
;
3969 if (flags
& S_DIRSYNC
)
3970 ei
->i_flags
|= EXT4_DIRSYNC_FL
;
3972 static blkcnt_t
ext4_inode_blocks(struct ext4_inode
*raw_inode
,
3973 struct ext4_inode_info
*ei
)
3976 struct inode
*inode
= &(ei
->vfs_inode
);
3977 struct super_block
*sb
= inode
->i_sb
;
3979 if (EXT4_HAS_RO_COMPAT_FEATURE(sb
,
3980 EXT4_FEATURE_RO_COMPAT_HUGE_FILE
)) {
3981 /* we are using combined 48 bit field */
3982 i_blocks
= ((u64
)le16_to_cpu(raw_inode
->i_blocks_high
)) << 32 |
3983 le32_to_cpu(raw_inode
->i_blocks_lo
);
3984 if (ei
->i_flags
& EXT4_HUGE_FILE_FL
) {
3985 /* i_blocks represent file system block size */
3986 return i_blocks
<< (inode
->i_blkbits
- 9);
3991 return le32_to_cpu(raw_inode
->i_blocks_lo
);
3995 struct inode
*ext4_iget(struct super_block
*sb
, unsigned long ino
)
3997 struct ext4_iloc iloc
;
3998 struct ext4_inode
*raw_inode
;
3999 struct ext4_inode_info
*ei
;
4000 struct buffer_head
*bh
;
4001 struct inode
*inode
;
4005 inode
= iget_locked(sb
, ino
);
4007 return ERR_PTR(-ENOMEM
);
4008 if (!(inode
->i_state
& I_NEW
))
4012 #ifdef CONFIG_EXT4DEV_FS_POSIX_ACL
4013 ei
->i_acl
= EXT4_ACL_NOT_CACHED
;
4014 ei
->i_default_acl
= EXT4_ACL_NOT_CACHED
;
4016 ei
->i_block_alloc_info
= NULL
;
4018 ret
= __ext4_get_inode_loc(inode
, &iloc
, 0);
4022 raw_inode
= ext4_raw_inode(&iloc
);
4023 inode
->i_mode
= le16_to_cpu(raw_inode
->i_mode
);
4024 inode
->i_uid
= (uid_t
)le16_to_cpu(raw_inode
->i_uid_low
);
4025 inode
->i_gid
= (gid_t
)le16_to_cpu(raw_inode
->i_gid_low
);
4026 if(!(test_opt (inode
->i_sb
, NO_UID32
))) {
4027 inode
->i_uid
|= le16_to_cpu(raw_inode
->i_uid_high
) << 16;
4028 inode
->i_gid
|= le16_to_cpu(raw_inode
->i_gid_high
) << 16;
4030 inode
->i_nlink
= le16_to_cpu(raw_inode
->i_links_count
);
4033 ei
->i_dir_start_lookup
= 0;
4034 ei
->i_dtime
= le32_to_cpu(raw_inode
->i_dtime
);
4035 /* We now have enough fields to check if the inode was active or not.
4036 * This is needed because nfsd might try to access dead inodes
4037 * the test is that same one that e2fsck uses
4038 * NeilBrown 1999oct15
4040 if (inode
->i_nlink
== 0) {
4041 if (inode
->i_mode
== 0 ||
4042 !(EXT4_SB(inode
->i_sb
)->s_mount_state
& EXT4_ORPHAN_FS
)) {
4043 /* this inode is deleted */
4048 /* The only unlinked inodes we let through here have
4049 * valid i_mode and are being read by the orphan
4050 * recovery code: that's fine, we're about to complete
4051 * the process of deleting those. */
4053 ei
->i_flags
= le32_to_cpu(raw_inode
->i_flags
);
4054 inode
->i_blocks
= ext4_inode_blocks(raw_inode
, ei
);
4055 ei
->i_file_acl
= le32_to_cpu(raw_inode
->i_file_acl_lo
);
4056 if (EXT4_SB(inode
->i_sb
)->s_es
->s_creator_os
!=
4057 cpu_to_le32(EXT4_OS_HURD
)) {
4059 ((__u64
)le16_to_cpu(raw_inode
->i_file_acl_high
)) << 32;
4061 inode
->i_size
= ext4_isize(raw_inode
);
4062 ei
->i_disksize
= inode
->i_size
;
4063 inode
->i_generation
= le32_to_cpu(raw_inode
->i_generation
);
4064 ei
->i_block_group
= iloc
.block_group
;
4066 * NOTE! The in-memory inode i_data array is in little-endian order
4067 * even on big-endian machines: we do NOT byteswap the block numbers!
4069 for (block
= 0; block
< EXT4_N_BLOCKS
; block
++)
4070 ei
->i_data
[block
] = raw_inode
->i_block
[block
];
4071 INIT_LIST_HEAD(&ei
->i_orphan
);
4073 if (EXT4_INODE_SIZE(inode
->i_sb
) > EXT4_GOOD_OLD_INODE_SIZE
) {
4074 ei
->i_extra_isize
= le16_to_cpu(raw_inode
->i_extra_isize
);
4075 if (EXT4_GOOD_OLD_INODE_SIZE
+ ei
->i_extra_isize
>
4076 EXT4_INODE_SIZE(inode
->i_sb
)) {
4081 if (ei
->i_extra_isize
== 0) {
4082 /* The extra space is currently unused. Use it. */
4083 ei
->i_extra_isize
= sizeof(struct ext4_inode
) -
4084 EXT4_GOOD_OLD_INODE_SIZE
;
4086 __le32
*magic
= (void *)raw_inode
+
4087 EXT4_GOOD_OLD_INODE_SIZE
+
4089 if (*magic
== cpu_to_le32(EXT4_XATTR_MAGIC
))
4090 ei
->i_state
|= EXT4_STATE_XATTR
;
4093 ei
->i_extra_isize
= 0;
4095 EXT4_INODE_GET_XTIME(i_ctime
, inode
, raw_inode
);
4096 EXT4_INODE_GET_XTIME(i_mtime
, inode
, raw_inode
);
4097 EXT4_INODE_GET_XTIME(i_atime
, inode
, raw_inode
);
4098 EXT4_EINODE_GET_XTIME(i_crtime
, ei
, raw_inode
);
4100 inode
->i_version
= le32_to_cpu(raw_inode
->i_disk_version
);
4101 if (EXT4_INODE_SIZE(inode
->i_sb
) > EXT4_GOOD_OLD_INODE_SIZE
) {
4102 if (EXT4_FITS_IN_INODE(raw_inode
, ei
, i_version_hi
))
4104 (__u64
)(le32_to_cpu(raw_inode
->i_version_hi
)) << 32;
4107 if (S_ISREG(inode
->i_mode
)) {
4108 inode
->i_op
= &ext4_file_inode_operations
;
4109 inode
->i_fop
= &ext4_file_operations
;
4110 ext4_set_aops(inode
);
4111 } else if (S_ISDIR(inode
->i_mode
)) {
4112 inode
->i_op
= &ext4_dir_inode_operations
;
4113 inode
->i_fop
= &ext4_dir_operations
;
4114 } else if (S_ISLNK(inode
->i_mode
)) {
4115 if (ext4_inode_is_fast_symlink(inode
))
4116 inode
->i_op
= &ext4_fast_symlink_inode_operations
;
4118 inode
->i_op
= &ext4_symlink_inode_operations
;
4119 ext4_set_aops(inode
);
4122 inode
->i_op
= &ext4_special_inode_operations
;
4123 if (raw_inode
->i_block
[0])
4124 init_special_inode(inode
, inode
->i_mode
,
4125 old_decode_dev(le32_to_cpu(raw_inode
->i_block
[0])));
4127 init_special_inode(inode
, inode
->i_mode
,
4128 new_decode_dev(le32_to_cpu(raw_inode
->i_block
[1])));
4131 ext4_set_inode_flags(inode
);
4132 unlock_new_inode(inode
);
4137 return ERR_PTR(ret
);
4140 static int ext4_inode_blocks_set(handle_t
*handle
,
4141 struct ext4_inode
*raw_inode
,
4142 struct ext4_inode_info
*ei
)
4144 struct inode
*inode
= &(ei
->vfs_inode
);
4145 u64 i_blocks
= inode
->i_blocks
;
4146 struct super_block
*sb
= inode
->i_sb
;
4149 if (i_blocks
<= ~0U) {
4151 * i_blocks can be represnted in a 32 bit variable
4152 * as multiple of 512 bytes
4154 raw_inode
->i_blocks_lo
= cpu_to_le32(i_blocks
);
4155 raw_inode
->i_blocks_high
= 0;
4156 ei
->i_flags
&= ~EXT4_HUGE_FILE_FL
;
4157 } else if (i_blocks
<= 0xffffffffffffULL
) {
4159 * i_blocks can be represented in a 48 bit variable
4160 * as multiple of 512 bytes
4162 err
= ext4_update_rocompat_feature(handle
, sb
,
4163 EXT4_FEATURE_RO_COMPAT_HUGE_FILE
);
4166 /* i_block is stored in the split 48 bit fields */
4167 raw_inode
->i_blocks_lo
= cpu_to_le32(i_blocks
);
4168 raw_inode
->i_blocks_high
= cpu_to_le16(i_blocks
>> 32);
4169 ei
->i_flags
&= ~EXT4_HUGE_FILE_FL
;
4172 * i_blocks should be represented in a 48 bit variable
4173 * as multiple of file system block size
4175 err
= ext4_update_rocompat_feature(handle
, sb
,
4176 EXT4_FEATURE_RO_COMPAT_HUGE_FILE
);
4179 ei
->i_flags
|= EXT4_HUGE_FILE_FL
;
4180 /* i_block is stored in file system block size */
4181 i_blocks
= i_blocks
>> (inode
->i_blkbits
- 9);
4182 raw_inode
->i_blocks_lo
= cpu_to_le32(i_blocks
);
4183 raw_inode
->i_blocks_high
= cpu_to_le16(i_blocks
>> 32);
4190 * Post the struct inode info into an on-disk inode location in the
4191 * buffer-cache. This gobbles the caller's reference to the
4192 * buffer_head in the inode location struct.
4194 * The caller must have write access to iloc->bh.
4196 static int ext4_do_update_inode(handle_t
*handle
,
4197 struct inode
*inode
,
4198 struct ext4_iloc
*iloc
)
4200 struct ext4_inode
*raw_inode
= ext4_raw_inode(iloc
);
4201 struct ext4_inode_info
*ei
= EXT4_I(inode
);
4202 struct buffer_head
*bh
= iloc
->bh
;
4203 int err
= 0, rc
, block
;
4205 /* For fields not not tracking in the in-memory inode,
4206 * initialise them to zero for new inodes. */
4207 if (ei
->i_state
& EXT4_STATE_NEW
)
4208 memset(raw_inode
, 0, EXT4_SB(inode
->i_sb
)->s_inode_size
);
4210 ext4_get_inode_flags(ei
);
4211 raw_inode
->i_mode
= cpu_to_le16(inode
->i_mode
);
4212 if(!(test_opt(inode
->i_sb
, NO_UID32
))) {
4213 raw_inode
->i_uid_low
= cpu_to_le16(low_16_bits(inode
->i_uid
));
4214 raw_inode
->i_gid_low
= cpu_to_le16(low_16_bits(inode
->i_gid
));
4216 * Fix up interoperability with old kernels. Otherwise, old inodes get
4217 * re-used with the upper 16 bits of the uid/gid intact
4220 raw_inode
->i_uid_high
=
4221 cpu_to_le16(high_16_bits(inode
->i_uid
));
4222 raw_inode
->i_gid_high
=
4223 cpu_to_le16(high_16_bits(inode
->i_gid
));
4225 raw_inode
->i_uid_high
= 0;
4226 raw_inode
->i_gid_high
= 0;
4229 raw_inode
->i_uid_low
=
4230 cpu_to_le16(fs_high2lowuid(inode
->i_uid
));
4231 raw_inode
->i_gid_low
=
4232 cpu_to_le16(fs_high2lowgid(inode
->i_gid
));
4233 raw_inode
->i_uid_high
= 0;
4234 raw_inode
->i_gid_high
= 0;
4236 raw_inode
->i_links_count
= cpu_to_le16(inode
->i_nlink
);
4238 EXT4_INODE_SET_XTIME(i_ctime
, inode
, raw_inode
);
4239 EXT4_INODE_SET_XTIME(i_mtime
, inode
, raw_inode
);
4240 EXT4_INODE_SET_XTIME(i_atime
, inode
, raw_inode
);
4241 EXT4_EINODE_SET_XTIME(i_crtime
, ei
, raw_inode
);
4243 if (ext4_inode_blocks_set(handle
, raw_inode
, ei
))
4245 raw_inode
->i_dtime
= cpu_to_le32(ei
->i_dtime
);
4246 /* clear the migrate flag in the raw_inode */
4247 raw_inode
->i_flags
= cpu_to_le32(ei
->i_flags
& ~EXT4_EXT_MIGRATE
);
4248 if (EXT4_SB(inode
->i_sb
)->s_es
->s_creator_os
!=
4249 cpu_to_le32(EXT4_OS_HURD
))
4250 raw_inode
->i_file_acl_high
=
4251 cpu_to_le16(ei
->i_file_acl
>> 32);
4252 raw_inode
->i_file_acl_lo
= cpu_to_le32(ei
->i_file_acl
);
4253 ext4_isize_set(raw_inode
, ei
->i_disksize
);
4254 if (ei
->i_disksize
> 0x7fffffffULL
) {
4255 struct super_block
*sb
= inode
->i_sb
;
4256 if (!EXT4_HAS_RO_COMPAT_FEATURE(sb
,
4257 EXT4_FEATURE_RO_COMPAT_LARGE_FILE
) ||
4258 EXT4_SB(sb
)->s_es
->s_rev_level
==
4259 cpu_to_le32(EXT4_GOOD_OLD_REV
)) {
4260 /* If this is the first large file
4261 * created, add a flag to the superblock.
4263 err
= ext4_journal_get_write_access(handle
,
4264 EXT4_SB(sb
)->s_sbh
);
4267 ext4_update_dynamic_rev(sb
);
4268 EXT4_SET_RO_COMPAT_FEATURE(sb
,
4269 EXT4_FEATURE_RO_COMPAT_LARGE_FILE
);
4272 err
= ext4_journal_dirty_metadata(handle
,
4273 EXT4_SB(sb
)->s_sbh
);
4276 raw_inode
->i_generation
= cpu_to_le32(inode
->i_generation
);
4277 if (S_ISCHR(inode
->i_mode
) || S_ISBLK(inode
->i_mode
)) {
4278 if (old_valid_dev(inode
->i_rdev
)) {
4279 raw_inode
->i_block
[0] =
4280 cpu_to_le32(old_encode_dev(inode
->i_rdev
));
4281 raw_inode
->i_block
[1] = 0;
4283 raw_inode
->i_block
[0] = 0;
4284 raw_inode
->i_block
[1] =
4285 cpu_to_le32(new_encode_dev(inode
->i_rdev
));
4286 raw_inode
->i_block
[2] = 0;
4288 } else for (block
= 0; block
< EXT4_N_BLOCKS
; block
++)
4289 raw_inode
->i_block
[block
] = ei
->i_data
[block
];
4291 raw_inode
->i_disk_version
= cpu_to_le32(inode
->i_version
);
4292 if (ei
->i_extra_isize
) {
4293 if (EXT4_FITS_IN_INODE(raw_inode
, ei
, i_version_hi
))
4294 raw_inode
->i_version_hi
=
4295 cpu_to_le32(inode
->i_version
>> 32);
4296 raw_inode
->i_extra_isize
= cpu_to_le16(ei
->i_extra_isize
);
4300 BUFFER_TRACE(bh
, "call ext4_journal_dirty_metadata");
4301 rc
= ext4_journal_dirty_metadata(handle
, bh
);
4304 ei
->i_state
&= ~EXT4_STATE_NEW
;
4308 ext4_std_error(inode
->i_sb
, err
);
4313 * ext4_write_inode()
4315 * We are called from a few places:
4317 * - Within generic_file_write() for O_SYNC files.
4318 * Here, there will be no transaction running. We wait for any running
4319 * trasnaction to commit.
4321 * - Within sys_sync(), kupdate and such.
4322 * We wait on commit, if tol to.
4324 * - Within prune_icache() (PF_MEMALLOC == true)
4325 * Here we simply return. We can't afford to block kswapd on the
4328 * In all cases it is actually safe for us to return without doing anything,
4329 * because the inode has been copied into a raw inode buffer in
4330 * ext4_mark_inode_dirty(). This is a correctness thing for O_SYNC and for
4333 * Note that we are absolutely dependent upon all inode dirtiers doing the
4334 * right thing: they *must* call mark_inode_dirty() after dirtying info in
4335 * which we are interested.
4337 * It would be a bug for them to not do this. The code:
4339 * mark_inode_dirty(inode)
4341 * inode->i_size = expr;
4343 * is in error because a kswapd-driven write_inode() could occur while
4344 * `stuff()' is running, and the new i_size will be lost. Plus the inode
4345 * will no longer be on the superblock's dirty inode list.
4347 int ext4_write_inode(struct inode
*inode
, int wait
)
4349 if (current
->flags
& PF_MEMALLOC
)
4352 if (ext4_journal_current_handle()) {
4353 jbd_debug(1, "called recursively, non-PF_MEMALLOC!\n");
4361 return ext4_force_commit(inode
->i_sb
);
4367 * Called from notify_change.
4369 * We want to trap VFS attempts to truncate the file as soon as
4370 * possible. In particular, we want to make sure that when the VFS
4371 * shrinks i_size, we put the inode on the orphan list and modify
4372 * i_disksize immediately, so that during the subsequent flushing of
4373 * dirty pages and freeing of disk blocks, we can guarantee that any
4374 * commit will leave the blocks being flushed in an unused state on
4375 * disk. (On recovery, the inode will get truncated and the blocks will
4376 * be freed, so we have a strong guarantee that no future commit will
4377 * leave these blocks visible to the user.)
4379 * Another thing we have to assure is that if we are in ordered mode
4380 * and inode is still attached to the committing transaction, we must
4381 * we start writeout of all the dirty pages which are being truncated.
4382 * This way we are sure that all the data written in the previous
4383 * transaction are already on disk (truncate waits for pages under
4386 * Called with inode->i_mutex down.
4388 int ext4_setattr(struct dentry
*dentry
, struct iattr
*attr
)
4390 struct inode
*inode
= dentry
->d_inode
;
4392 const unsigned int ia_valid
= attr
->ia_valid
;
4394 error
= inode_change_ok(inode
, attr
);
4398 if ((ia_valid
& ATTR_UID
&& attr
->ia_uid
!= inode
->i_uid
) ||
4399 (ia_valid
& ATTR_GID
&& attr
->ia_gid
!= inode
->i_gid
)) {
4402 /* (user+group)*(old+new) structure, inode write (sb,
4403 * inode block, ? - but truncate inode update has it) */
4404 handle
= ext4_journal_start(inode
, 2*(EXT4_QUOTA_INIT_BLOCKS(inode
->i_sb
)+
4405 EXT4_QUOTA_DEL_BLOCKS(inode
->i_sb
))+3);
4406 if (IS_ERR(handle
)) {
4407 error
= PTR_ERR(handle
);
4410 error
= DQUOT_TRANSFER(inode
, attr
) ? -EDQUOT
: 0;
4412 ext4_journal_stop(handle
);
4415 /* Update corresponding info in inode so that everything is in
4416 * one transaction */
4417 if (attr
->ia_valid
& ATTR_UID
)
4418 inode
->i_uid
= attr
->ia_uid
;
4419 if (attr
->ia_valid
& ATTR_GID
)
4420 inode
->i_gid
= attr
->ia_gid
;
4421 error
= ext4_mark_inode_dirty(handle
, inode
);
4422 ext4_journal_stop(handle
);
4425 if (attr
->ia_valid
& ATTR_SIZE
) {
4426 if (!(EXT4_I(inode
)->i_flags
& EXT4_EXTENTS_FL
)) {
4427 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
4429 if (attr
->ia_size
> sbi
->s_bitmap_maxbytes
) {
4436 if (S_ISREG(inode
->i_mode
) &&
4437 attr
->ia_valid
& ATTR_SIZE
&& attr
->ia_size
< inode
->i_size
) {
4440 handle
= ext4_journal_start(inode
, 3);
4441 if (IS_ERR(handle
)) {
4442 error
= PTR_ERR(handle
);
4446 error
= ext4_orphan_add(handle
, inode
);
4447 EXT4_I(inode
)->i_disksize
= attr
->ia_size
;
4448 rc
= ext4_mark_inode_dirty(handle
, inode
);
4451 ext4_journal_stop(handle
);
4453 if (ext4_should_order_data(inode
)) {
4454 error
= ext4_begin_ordered_truncate(inode
,
4457 /* Do as much error cleanup as possible */
4458 handle
= ext4_journal_start(inode
, 3);
4459 if (IS_ERR(handle
)) {
4460 ext4_orphan_del(NULL
, inode
);
4463 ext4_orphan_del(handle
, inode
);
4464 ext4_journal_stop(handle
);
4470 rc
= inode_setattr(inode
, attr
);
4472 /* If inode_setattr's call to ext4_truncate failed to get a
4473 * transaction handle at all, we need to clean up the in-core
4474 * orphan list manually. */
4476 ext4_orphan_del(NULL
, inode
);
4478 if (!rc
&& (ia_valid
& ATTR_MODE
))
4479 rc
= ext4_acl_chmod(inode
);
4482 ext4_std_error(inode
->i_sb
, error
);
4488 int ext4_getattr(struct vfsmount
*mnt
, struct dentry
*dentry
,
4491 struct inode
*inode
;
4492 unsigned long delalloc_blocks
;
4494 inode
= dentry
->d_inode
;
4495 generic_fillattr(inode
, stat
);
4498 * We can't update i_blocks if the block allocation is delayed
4499 * otherwise in the case of system crash before the real block
4500 * allocation is done, we will have i_blocks inconsistent with
4501 * on-disk file blocks.
4502 * We always keep i_blocks updated together with real
4503 * allocation. But to not confuse with user, stat
4504 * will return the blocks that include the delayed allocation
4505 * blocks for this file.
4507 spin_lock(&EXT4_I(inode
)->i_block_reservation_lock
);
4508 delalloc_blocks
= EXT4_I(inode
)->i_reserved_data_blocks
;
4509 spin_unlock(&EXT4_I(inode
)->i_block_reservation_lock
);
4511 stat
->blocks
+= (delalloc_blocks
<< inode
->i_sb
->s_blocksize_bits
)>>9;
4515 static int ext4_indirect_trans_blocks(struct inode
*inode
, int nrblocks
,
4520 /* if nrblocks are contiguous */
4523 * With N contiguous data blocks, it need at most
4524 * N/EXT4_ADDR_PER_BLOCK(inode->i_sb) indirect blocks
4525 * 2 dindirect blocks
4528 indirects
= nrblocks
/ EXT4_ADDR_PER_BLOCK(inode
->i_sb
);
4529 return indirects
+ 3;
4532 * if nrblocks are not contiguous, worse case, each block touch
4533 * a indirect block, and each indirect block touch a double indirect
4534 * block, plus a triple indirect block
4536 indirects
= nrblocks
* 2 + 1;
4540 static int ext4_index_trans_blocks(struct inode
*inode
, int nrblocks
, int chunk
)
4542 if (!(EXT4_I(inode
)->i_flags
& EXT4_EXTENTS_FL
))
4543 return ext4_indirect_trans_blocks(inode
, nrblocks
, chunk
);
4544 return ext4_ext_index_trans_blocks(inode
, nrblocks
, chunk
);
4548 * Account for index blocks, block groups bitmaps and block group
4549 * descriptor blocks if modify datablocks and index blocks
4550 * worse case, the indexs blocks spread over different block groups
4552 * If datablocks are discontiguous, they are possible to spread over
4553 * different block groups too. If they are contiugous, with flexbg,
4554 * they could still across block group boundary.
4556 * Also account for superblock, inode, quota and xattr blocks
4558 int ext4_meta_trans_blocks(struct inode
*inode
, int nrblocks
, int chunk
)
4560 int groups
, gdpblocks
;
4565 * How many index blocks need to touch to modify nrblocks?
4566 * The "Chunk" flag indicating whether the nrblocks is
4567 * physically contiguous on disk
4569 * For Direct IO and fallocate, they calls get_block to allocate
4570 * one single extent at a time, so they could set the "Chunk" flag
4572 idxblocks
= ext4_index_trans_blocks(inode
, nrblocks
, chunk
);
4577 * Now let's see how many group bitmaps and group descriptors need
4587 if (groups
> EXT4_SB(inode
->i_sb
)->s_groups_count
)
4588 groups
= EXT4_SB(inode
->i_sb
)->s_groups_count
;
4589 if (groups
> EXT4_SB(inode
->i_sb
)->s_gdb_count
)
4590 gdpblocks
= EXT4_SB(inode
->i_sb
)->s_gdb_count
;
4592 /* bitmaps and block group descriptor blocks */
4593 ret
+= groups
+ gdpblocks
;
4595 /* Blocks for super block, inode, quota and xattr blocks */
4596 ret
+= EXT4_META_TRANS_BLOCKS(inode
->i_sb
);
4602 * Calulate the total number of credits to reserve to fit
4603 * the modification of a single pages into a single transaction,
4604 * which may include multiple chunks of block allocations.
4606 * This could be called via ext4_write_begin()
4608 * We need to consider the worse case, when
4609 * one new block per extent.
4611 int ext4_writepage_trans_blocks(struct inode
*inode
)
4613 int bpp
= ext4_journal_blocks_per_page(inode
);
4616 ret
= ext4_meta_trans_blocks(inode
, bpp
, 0);
4618 /* Account for data blocks for journalled mode */
4619 if (ext4_should_journal_data(inode
))
4625 * Calculate the journal credits for a chunk of data modification.
4627 * This is called from DIO, fallocate or whoever calling
4628 * ext4_get_blocks_wrap() to map/allocate a chunk of contigous disk blocks.
4630 * journal buffers for data blocks are not included here, as DIO
4631 * and fallocate do no need to journal data buffers.
4633 int ext4_chunk_trans_blocks(struct inode
*inode
, int nrblocks
)
4635 return ext4_meta_trans_blocks(inode
, nrblocks
, 1);
4639 * The caller must have previously called ext4_reserve_inode_write().
4640 * Give this, we know that the caller already has write access to iloc->bh.
4642 int ext4_mark_iloc_dirty(handle_t
*handle
,
4643 struct inode
*inode
, struct ext4_iloc
*iloc
)
4647 if (test_opt(inode
->i_sb
, I_VERSION
))
4648 inode_inc_iversion(inode
);
4650 /* the do_update_inode consumes one bh->b_count */
4653 /* ext4_do_update_inode() does jbd2_journal_dirty_metadata */
4654 err
= ext4_do_update_inode(handle
, inode
, iloc
);
4660 * On success, We end up with an outstanding reference count against
4661 * iloc->bh. This _must_ be cleaned up later.
4665 ext4_reserve_inode_write(handle_t
*handle
, struct inode
*inode
,
4666 struct ext4_iloc
*iloc
)
4670 err
= ext4_get_inode_loc(inode
, iloc
);
4672 BUFFER_TRACE(iloc
->bh
, "get_write_access");
4673 err
= ext4_journal_get_write_access(handle
, iloc
->bh
);
4680 ext4_std_error(inode
->i_sb
, err
);
4685 * Expand an inode by new_extra_isize bytes.
4686 * Returns 0 on success or negative error number on failure.
4688 static int ext4_expand_extra_isize(struct inode
*inode
,
4689 unsigned int new_extra_isize
,
4690 struct ext4_iloc iloc
,
4693 struct ext4_inode
*raw_inode
;
4694 struct ext4_xattr_ibody_header
*header
;
4695 struct ext4_xattr_entry
*entry
;
4697 if (EXT4_I(inode
)->i_extra_isize
>= new_extra_isize
)
4700 raw_inode
= ext4_raw_inode(&iloc
);
4702 header
= IHDR(inode
, raw_inode
);
4703 entry
= IFIRST(header
);
4705 /* No extended attributes present */
4706 if (!(EXT4_I(inode
)->i_state
& EXT4_STATE_XATTR
) ||
4707 header
->h_magic
!= cpu_to_le32(EXT4_XATTR_MAGIC
)) {
4708 memset((void *)raw_inode
+ EXT4_GOOD_OLD_INODE_SIZE
, 0,
4710 EXT4_I(inode
)->i_extra_isize
= new_extra_isize
;
4714 /* try to expand with EAs present */
4715 return ext4_expand_extra_isize_ea(inode
, new_extra_isize
,
4720 * What we do here is to mark the in-core inode as clean with respect to inode
4721 * dirtiness (it may still be data-dirty).
4722 * This means that the in-core inode may be reaped by prune_icache
4723 * without having to perform any I/O. This is a very good thing,
4724 * because *any* task may call prune_icache - even ones which
4725 * have a transaction open against a different journal.
4727 * Is this cheating? Not really. Sure, we haven't written the
4728 * inode out, but prune_icache isn't a user-visible syncing function.
4729 * Whenever the user wants stuff synced (sys_sync, sys_msync, sys_fsync)
4730 * we start and wait on commits.
4732 * Is this efficient/effective? Well, we're being nice to the system
4733 * by cleaning up our inodes proactively so they can be reaped
4734 * without I/O. But we are potentially leaving up to five seconds'
4735 * worth of inodes floating about which prune_icache wants us to
4736 * write out. One way to fix that would be to get prune_icache()
4737 * to do a write_super() to free up some memory. It has the desired
4740 int ext4_mark_inode_dirty(handle_t
*handle
, struct inode
*inode
)
4742 struct ext4_iloc iloc
;
4743 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
4744 static unsigned int mnt_count
;
4748 err
= ext4_reserve_inode_write(handle
, inode
, &iloc
);
4749 if (EXT4_I(inode
)->i_extra_isize
< sbi
->s_want_extra_isize
&&
4750 !(EXT4_I(inode
)->i_state
& EXT4_STATE_NO_EXPAND
)) {
4752 * We need extra buffer credits since we may write into EA block
4753 * with this same handle. If journal_extend fails, then it will
4754 * only result in a minor loss of functionality for that inode.
4755 * If this is felt to be critical, then e2fsck should be run to
4756 * force a large enough s_min_extra_isize.
4758 if ((jbd2_journal_extend(handle
,
4759 EXT4_DATA_TRANS_BLOCKS(inode
->i_sb
))) == 0) {
4760 ret
= ext4_expand_extra_isize(inode
,
4761 sbi
->s_want_extra_isize
,
4764 EXT4_I(inode
)->i_state
|= EXT4_STATE_NO_EXPAND
;
4766 le16_to_cpu(sbi
->s_es
->s_mnt_count
)) {
4767 ext4_warning(inode
->i_sb
, __func__
,
4768 "Unable to expand inode %lu. Delete"
4769 " some EAs or run e2fsck.",
4772 le16_to_cpu(sbi
->s_es
->s_mnt_count
);
4778 err
= ext4_mark_iloc_dirty(handle
, inode
, &iloc
);
4783 * ext4_dirty_inode() is called from __mark_inode_dirty()
4785 * We're really interested in the case where a file is being extended.
4786 * i_size has been changed by generic_commit_write() and we thus need
4787 * to include the updated inode in the current transaction.
4789 * Also, DQUOT_ALLOC_SPACE() will always dirty the inode when blocks
4790 * are allocated to the file.
4792 * If the inode is marked synchronous, we don't honour that here - doing
4793 * so would cause a commit on atime updates, which we don't bother doing.
4794 * We handle synchronous inodes at the highest possible level.
4796 void ext4_dirty_inode(struct inode
*inode
)
4798 handle_t
*current_handle
= ext4_journal_current_handle();
4801 handle
= ext4_journal_start(inode
, 2);
4804 if (current_handle
&&
4805 current_handle
->h_transaction
!= handle
->h_transaction
) {
4806 /* This task has a transaction open against a different fs */
4807 printk(KERN_EMERG
"%s: transactions do not match!\n",
4810 jbd_debug(5, "marking dirty. outer handle=%p\n",
4812 ext4_mark_inode_dirty(handle
, inode
);
4814 ext4_journal_stop(handle
);
4821 * Bind an inode's backing buffer_head into this transaction, to prevent
4822 * it from being flushed to disk early. Unlike
4823 * ext4_reserve_inode_write, this leaves behind no bh reference and
4824 * returns no iloc structure, so the caller needs to repeat the iloc
4825 * lookup to mark the inode dirty later.
4827 static int ext4_pin_inode(handle_t
*handle
, struct inode
*inode
)
4829 struct ext4_iloc iloc
;
4833 err
= ext4_get_inode_loc(inode
, &iloc
);
4835 BUFFER_TRACE(iloc
.bh
, "get_write_access");
4836 err
= jbd2_journal_get_write_access(handle
, iloc
.bh
);
4838 err
= ext4_journal_dirty_metadata(handle
,
4843 ext4_std_error(inode
->i_sb
, err
);
4848 int ext4_change_inode_journal_flag(struct inode
*inode
, int val
)
4855 * We have to be very careful here: changing a data block's
4856 * journaling status dynamically is dangerous. If we write a
4857 * data block to the journal, change the status and then delete
4858 * that block, we risk forgetting to revoke the old log record
4859 * from the journal and so a subsequent replay can corrupt data.
4860 * So, first we make sure that the journal is empty and that
4861 * nobody is changing anything.
4864 journal
= EXT4_JOURNAL(inode
);
4865 if (is_journal_aborted(journal
))
4868 jbd2_journal_lock_updates(journal
);
4869 jbd2_journal_flush(journal
);
4872 * OK, there are no updates running now, and all cached data is
4873 * synced to disk. We are now in a completely consistent state
4874 * which doesn't have anything in the journal, and we know that
4875 * no filesystem updates are running, so it is safe to modify
4876 * the inode's in-core data-journaling state flag now.
4880 EXT4_I(inode
)->i_flags
|= EXT4_JOURNAL_DATA_FL
;
4882 EXT4_I(inode
)->i_flags
&= ~EXT4_JOURNAL_DATA_FL
;
4883 ext4_set_aops(inode
);
4885 jbd2_journal_unlock_updates(journal
);
4887 /* Finally we can mark the inode as dirty. */
4889 handle
= ext4_journal_start(inode
, 1);
4891 return PTR_ERR(handle
);
4893 err
= ext4_mark_inode_dirty(handle
, inode
);
4895 ext4_journal_stop(handle
);
4896 ext4_std_error(inode
->i_sb
, err
);
4901 static int ext4_bh_unmapped(handle_t
*handle
, struct buffer_head
*bh
)
4903 return !buffer_mapped(bh
);
4906 int ext4_page_mkwrite(struct vm_area_struct
*vma
, struct vm_fault
*vmf
)
4908 struct page
*page
= vmf
->page
;
4912 struct file
*file
= vma
->vm_file
;
4913 struct inode
*inode
= file
->f_path
.dentry
->d_inode
;
4914 struct address_space
*mapping
= inode
->i_mapping
;
4917 * Get i_alloc_sem to stop truncates messing with the inode. We cannot
4918 * get i_mutex because we are already holding mmap_sem.
4920 down_read(&inode
->i_alloc_sem
);
4921 size
= i_size_read(inode
);
4922 if (page
->mapping
!= mapping
|| size
<= page_offset(page
)
4923 || !PageUptodate(page
)) {
4924 /* page got truncated from under us? */
4928 if (PageMappedToDisk(page
))
4931 if (page
->index
== size
>> PAGE_CACHE_SHIFT
)
4932 len
= size
& ~PAGE_CACHE_MASK
;
4934 len
= PAGE_CACHE_SIZE
;
4936 if (page_has_buffers(page
)) {
4937 /* return if we have all the buffers mapped */
4938 if (!walk_page_buffers(NULL
, page_buffers(page
), 0, len
, NULL
,
4943 * OK, we need to fill the hole... Do write_begin write_end
4944 * to do block allocation/reservation.We are not holding
4945 * inode.i__mutex here. That allow * parallel write_begin,
4946 * write_end call. lock_page prevent this from happening
4947 * on the same page though
4949 ret
= mapping
->a_ops
->write_begin(file
, mapping
, page_offset(page
),
4950 len
, AOP_FLAG_UNINTERRUPTIBLE
, &page
, NULL
);
4953 ret
= mapping
->a_ops
->write_end(file
, mapping
, page_offset(page
),
4954 len
, len
, page
, NULL
);
4960 ret
= VM_FAULT_SIGBUS
;
4961 up_read(&inode
->i_alloc_sem
);