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
,
492 * XXX need to get goal block from mballoc's data structures
495 return ext4_find_near(inode
, partial
);
499 * ext4_blks_to_allocate: Look up the block map and count the number
500 * of direct blocks need to be allocated for the given branch.
502 * @branch: chain of indirect blocks
503 * @k: number of blocks need for indirect blocks
504 * @blks: number of data blocks to be mapped.
505 * @blocks_to_boundary: the offset in the indirect block
507 * return the total number of blocks to be allocate, including the
508 * direct and indirect blocks.
510 static int ext4_blks_to_allocate(Indirect
*branch
, int k
, unsigned long blks
,
511 int blocks_to_boundary
)
513 unsigned long count
= 0;
516 * Simple case, [t,d]Indirect block(s) has not allocated yet
517 * then it's clear blocks on that path have not allocated
520 /* right now we don't handle cross boundary allocation */
521 if (blks
< blocks_to_boundary
+ 1)
524 count
+= blocks_to_boundary
+ 1;
529 while (count
< blks
&& count
<= blocks_to_boundary
&&
530 le32_to_cpu(*(branch
[0].p
+ count
)) == 0) {
537 * ext4_alloc_blocks: multiple allocate blocks needed for a branch
538 * @indirect_blks: the number of blocks need to allocate for indirect
541 * @new_blocks: on return it will store the new block numbers for
542 * the indirect blocks(if needed) and the first direct block,
543 * @blks: on return it will store the total number of allocated
546 static int ext4_alloc_blocks(handle_t
*handle
, struct inode
*inode
,
547 ext4_lblk_t iblock
, ext4_fsblk_t goal
,
548 int indirect_blks
, int blks
,
549 ext4_fsblk_t new_blocks
[4], int *err
)
552 unsigned long count
= 0, blk_allocated
= 0;
554 ext4_fsblk_t current_block
= 0;
558 * Here we try to allocate the requested multiple blocks at once,
559 * on a best-effort basis.
560 * To build a branch, we should allocate blocks for
561 * the indirect blocks(if not allocated yet), and at least
562 * the first direct block of this branch. That's the
563 * minimum number of blocks need to allocate(required)
565 /* first we try to allocate the indirect blocks */
566 target
= indirect_blks
;
569 /* allocating blocks for indirect blocks and direct blocks */
570 current_block
= ext4_new_meta_blocks(handle
, inode
,
576 /* allocate blocks for indirect blocks */
577 while (index
< indirect_blks
&& count
) {
578 new_blocks
[index
++] = current_block
++;
583 * save the new block number
584 * for the first direct block
586 new_blocks
[index
] = current_block
;
587 printk(KERN_INFO
"%s returned more blocks than "
588 "requested\n", __func__
);
594 target
= blks
- count
;
595 blk_allocated
= count
;
598 /* Now allocate data blocks */
600 /* allocating blocks for data blocks */
601 current_block
= ext4_new_blocks(handle
, inode
, iblock
,
603 if (*err
&& (target
== blks
)) {
605 * if the allocation failed and we didn't allocate
611 if (target
== blks
) {
613 * save the new block number
614 * for the first direct block
616 new_blocks
[index
] = current_block
;
618 blk_allocated
+= count
;
621 /* total number of blocks allocated for direct blocks */
626 for (i
= 0; i
< index
; i
++)
627 ext4_free_blocks(handle
, inode
, new_blocks
[i
], 1, 0);
632 * ext4_alloc_branch - allocate and set up a chain of blocks.
634 * @indirect_blks: number of allocated indirect blocks
635 * @blks: number of allocated direct blocks
636 * @offsets: offsets (in the blocks) to store the pointers to next.
637 * @branch: place to store the chain in.
639 * This function allocates blocks, zeroes out all but the last one,
640 * links them into chain and (if we are synchronous) writes them to disk.
641 * In other words, it prepares a branch that can be spliced onto the
642 * inode. It stores the information about that chain in the branch[], in
643 * the same format as ext4_get_branch() would do. We are calling it after
644 * we had read the existing part of chain and partial points to the last
645 * triple of that (one with zero ->key). Upon the exit we have the same
646 * picture as after the successful ext4_get_block(), except that in one
647 * place chain is disconnected - *branch->p is still zero (we did not
648 * set the last link), but branch->key contains the number that should
649 * be placed into *branch->p to fill that gap.
651 * If allocation fails we free all blocks we've allocated (and forget
652 * their buffer_heads) and return the error value the from failed
653 * ext4_alloc_block() (normally -ENOSPC). Otherwise we set the chain
654 * as described above and return 0.
656 static int ext4_alloc_branch(handle_t
*handle
, struct inode
*inode
,
657 ext4_lblk_t iblock
, int indirect_blks
,
658 int *blks
, ext4_fsblk_t goal
,
659 ext4_lblk_t
*offsets
, Indirect
*branch
)
661 int blocksize
= inode
->i_sb
->s_blocksize
;
664 struct buffer_head
*bh
;
666 ext4_fsblk_t new_blocks
[4];
667 ext4_fsblk_t current_block
;
669 num
= ext4_alloc_blocks(handle
, inode
, iblock
, goal
, indirect_blks
,
670 *blks
, new_blocks
, &err
);
674 branch
[0].key
= cpu_to_le32(new_blocks
[0]);
676 * metadata blocks and data blocks are allocated.
678 for (n
= 1; n
<= indirect_blks
; n
++) {
680 * Get buffer_head for parent block, zero it out
681 * and set the pointer to new one, then send
684 bh
= sb_getblk(inode
->i_sb
, new_blocks
[n
-1]);
687 BUFFER_TRACE(bh
, "call get_create_access");
688 err
= ext4_journal_get_create_access(handle
, bh
);
695 memset(bh
->b_data
, 0, blocksize
);
696 branch
[n
].p
= (__le32
*) bh
->b_data
+ offsets
[n
];
697 branch
[n
].key
= cpu_to_le32(new_blocks
[n
]);
698 *branch
[n
].p
= branch
[n
].key
;
699 if (n
== indirect_blks
) {
700 current_block
= new_blocks
[n
];
702 * End of chain, update the last new metablock of
703 * the chain to point to the new allocated
704 * data blocks numbers
706 for (i
=1; i
< num
; i
++)
707 *(branch
[n
].p
+ i
) = cpu_to_le32(++current_block
);
709 BUFFER_TRACE(bh
, "marking uptodate");
710 set_buffer_uptodate(bh
);
713 BUFFER_TRACE(bh
, "call ext4_journal_dirty_metadata");
714 err
= ext4_journal_dirty_metadata(handle
, bh
);
721 /* Allocation failed, free what we already allocated */
722 for (i
= 1; i
<= n
; i
++) {
723 BUFFER_TRACE(branch
[i
].bh
, "call jbd2_journal_forget");
724 ext4_journal_forget(handle
, branch
[i
].bh
);
726 for (i
= 0; i
< indirect_blks
; i
++)
727 ext4_free_blocks(handle
, inode
, new_blocks
[i
], 1, 0);
729 ext4_free_blocks(handle
, inode
, new_blocks
[i
], num
, 0);
735 * ext4_splice_branch - splice the allocated branch onto inode.
737 * @block: (logical) number of block we are adding
738 * @chain: chain of indirect blocks (with a missing link - see
740 * @where: location of missing link
741 * @num: number of indirect blocks we are adding
742 * @blks: number of direct blocks we are adding
744 * This function fills the missing link and does all housekeeping needed in
745 * inode (->i_blocks, etc.). In case of success we end up with the full
746 * chain to new block and return 0.
748 static int ext4_splice_branch(handle_t
*handle
, struct inode
*inode
,
749 ext4_lblk_t block
, Indirect
*where
, int num
, int blks
)
753 ext4_fsblk_t current_block
;
756 * If we're splicing into a [td]indirect block (as opposed to the
757 * inode) then we need to get write access to the [td]indirect block
761 BUFFER_TRACE(where
->bh
, "get_write_access");
762 err
= ext4_journal_get_write_access(handle
, where
->bh
);
768 *where
->p
= where
->key
;
771 * Update the host buffer_head or inode to point to more just allocated
772 * direct blocks blocks
774 if (num
== 0 && blks
> 1) {
775 current_block
= le32_to_cpu(where
->key
) + 1;
776 for (i
= 1; i
< blks
; i
++)
777 *(where
->p
+ i
) = cpu_to_le32(current_block
++);
780 /* We are done with atomic stuff, now do the rest of housekeeping */
782 inode
->i_ctime
= ext4_current_time(inode
);
783 ext4_mark_inode_dirty(handle
, inode
);
785 /* had we spliced it onto indirect block? */
788 * If we spliced it onto an indirect block, we haven't
789 * altered the inode. Note however that if it is being spliced
790 * onto an indirect block at the very end of the file (the
791 * file is growing) then we *will* alter the inode to reflect
792 * the new i_size. But that is not done here - it is done in
793 * generic_commit_write->__mark_inode_dirty->ext4_dirty_inode.
795 jbd_debug(5, "splicing indirect only\n");
796 BUFFER_TRACE(where
->bh
, "call ext4_journal_dirty_metadata");
797 err
= ext4_journal_dirty_metadata(handle
, where
->bh
);
802 * OK, we spliced it into the inode itself on a direct block.
803 * Inode was dirtied above.
805 jbd_debug(5, "splicing direct\n");
810 for (i
= 1; i
<= num
; i
++) {
811 BUFFER_TRACE(where
[i
].bh
, "call jbd2_journal_forget");
812 ext4_journal_forget(handle
, where
[i
].bh
);
813 ext4_free_blocks(handle
, inode
,
814 le32_to_cpu(where
[i
-1].key
), 1, 0);
816 ext4_free_blocks(handle
, inode
, le32_to_cpu(where
[num
].key
), blks
, 0);
822 * Allocation strategy is simple: if we have to allocate something, we will
823 * have to go the whole way to leaf. So let's do it before attaching anything
824 * to tree, set linkage between the newborn blocks, write them if sync is
825 * required, recheck the path, free and repeat if check fails, otherwise
826 * set the last missing link (that will protect us from any truncate-generated
827 * removals - all blocks on the path are immune now) and possibly force the
828 * write on the parent block.
829 * That has a nice additional property: no special recovery from the failed
830 * allocations is needed - we simply release blocks and do not touch anything
831 * reachable from inode.
833 * `handle' can be NULL if create == 0.
835 * return > 0, # of blocks mapped or allocated.
836 * return = 0, if plain lookup failed.
837 * return < 0, error case.
840 * Need to be called with
841 * down_read(&EXT4_I(inode)->i_data_sem) if not allocating file system block
842 * (ie, create is zero). Otherwise down_write(&EXT4_I(inode)->i_data_sem)
844 int ext4_get_blocks_handle(handle_t
*handle
, struct inode
*inode
,
845 ext4_lblk_t iblock
, unsigned long maxblocks
,
846 struct buffer_head
*bh_result
,
847 int create
, int extend_disksize
)
850 ext4_lblk_t offsets
[4];
855 int blocks_to_boundary
= 0;
857 struct ext4_inode_info
*ei
= EXT4_I(inode
);
859 ext4_fsblk_t first_block
= 0;
863 J_ASSERT(!(EXT4_I(inode
)->i_flags
& EXT4_EXTENTS_FL
));
864 J_ASSERT(handle
!= NULL
|| create
== 0);
865 depth
= ext4_block_to_path(inode
, iblock
, offsets
,
866 &blocks_to_boundary
);
871 partial
= ext4_get_branch(inode
, depth
, offsets
, chain
, &err
);
873 /* Simplest case - block found, no allocation needed */
875 first_block
= le32_to_cpu(chain
[depth
- 1].key
);
876 clear_buffer_new(bh_result
);
879 while (count
< maxblocks
&& count
<= blocks_to_boundary
) {
882 blk
= le32_to_cpu(*(chain
[depth
-1].p
+ count
));
884 if (blk
== first_block
+ count
)
892 /* Next simple case - plain lookup or failed read of indirect block */
893 if (!create
|| err
== -EIO
)
897 * Okay, we need to do block allocation.
899 goal
= ext4_find_goal(inode
, iblock
, partial
);
901 /* the number of blocks need to allocate for [d,t]indirect blocks */
902 indirect_blks
= (chain
+ depth
) - partial
- 1;
905 * Next look up the indirect map to count the totoal number of
906 * direct blocks to allocate for this branch.
908 count
= ext4_blks_to_allocate(partial
, indirect_blks
,
909 maxblocks
, blocks_to_boundary
);
911 * Block out ext4_truncate while we alter the tree
913 err
= ext4_alloc_branch(handle
, inode
, iblock
, indirect_blks
,
915 offsets
+ (partial
- chain
), partial
);
918 * The ext4_splice_branch call will free and forget any buffers
919 * on the new chain if there is a failure, but that risks using
920 * up transaction credits, especially for bitmaps where the
921 * credits cannot be returned. Can we handle this somehow? We
922 * may need to return -EAGAIN upwards in the worst case. --sct
925 err
= ext4_splice_branch(handle
, inode
, iblock
,
926 partial
, indirect_blks
, count
);
928 * i_disksize growing is protected by i_data_sem. Don't forget to
929 * protect it if you're about to implement concurrent
930 * ext4_get_block() -bzzz
932 if (!err
&& extend_disksize
) {
933 disksize
= ((loff_t
) iblock
+ count
) << inode
->i_blkbits
;
934 if (disksize
> i_size_read(inode
))
935 disksize
= i_size_read(inode
);
936 if (disksize
> ei
->i_disksize
)
937 ei
->i_disksize
= disksize
;
942 set_buffer_new(bh_result
);
944 map_bh(bh_result
, inode
->i_sb
, le32_to_cpu(chain
[depth
-1].key
));
945 if (count
> blocks_to_boundary
)
946 set_buffer_boundary(bh_result
);
948 /* Clean up and exit */
949 partial
= chain
+ depth
- 1; /* the whole chain */
951 while (partial
> chain
) {
952 BUFFER_TRACE(partial
->bh
, "call brelse");
956 BUFFER_TRACE(bh_result
, "returned");
962 * Calculate the number of metadata blocks need to reserve
963 * to allocate @blocks for non extent file based file
965 static int ext4_indirect_calc_metadata_amount(struct inode
*inode
, int blocks
)
967 int icap
= EXT4_ADDR_PER_BLOCK(inode
->i_sb
);
968 int ind_blks
, dind_blks
, tind_blks
;
970 /* number of new indirect blocks needed */
971 ind_blks
= (blocks
+ icap
- 1) / icap
;
973 dind_blks
= (ind_blks
+ icap
- 1) / icap
;
977 return ind_blks
+ dind_blks
+ tind_blks
;
981 * Calculate the number of metadata blocks need to reserve
982 * to allocate given number of blocks
984 static int ext4_calc_metadata_amount(struct inode
*inode
, int blocks
)
989 if (EXT4_I(inode
)->i_flags
& EXT4_EXTENTS_FL
)
990 return ext4_ext_calc_metadata_amount(inode
, blocks
);
992 return ext4_indirect_calc_metadata_amount(inode
, blocks
);
995 static void ext4_da_update_reserve_space(struct inode
*inode
, int used
)
997 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
998 int total
, mdb
, mdb_free
;
1000 spin_lock(&EXT4_I(inode
)->i_block_reservation_lock
);
1001 /* recalculate the number of metablocks still need to be reserved */
1002 total
= EXT4_I(inode
)->i_reserved_data_blocks
- used
;
1003 mdb
= ext4_calc_metadata_amount(inode
, total
);
1005 /* figure out how many metablocks to release */
1006 BUG_ON(mdb
> EXT4_I(inode
)->i_reserved_meta_blocks
);
1007 mdb_free
= EXT4_I(inode
)->i_reserved_meta_blocks
- mdb
;
1010 /* Account for allocated meta_blocks */
1011 mdb_free
-= EXT4_I(inode
)->i_allocated_meta_blocks
;
1013 /* update fs dirty blocks counter */
1014 percpu_counter_sub(&sbi
->s_dirtyblocks_counter
, mdb_free
);
1015 EXT4_I(inode
)->i_allocated_meta_blocks
= 0;
1016 EXT4_I(inode
)->i_reserved_meta_blocks
= mdb
;
1019 /* update per-inode reservations */
1020 BUG_ON(used
> EXT4_I(inode
)->i_reserved_data_blocks
);
1021 EXT4_I(inode
)->i_reserved_data_blocks
-= used
;
1023 spin_unlock(&EXT4_I(inode
)->i_block_reservation_lock
);
1027 * The ext4_get_blocks_wrap() function try to look up the requested blocks,
1028 * and returns if the blocks are already mapped.
1030 * Otherwise it takes the write lock of the i_data_sem and allocate blocks
1031 * and store the allocated blocks in the result buffer head and mark it
1034 * If file type is extents based, it will call ext4_ext_get_blocks(),
1035 * Otherwise, call with ext4_get_blocks_handle() to handle indirect mapping
1038 * On success, it returns the number of blocks being mapped or allocate.
1039 * if create==0 and the blocks are pre-allocated and uninitialized block,
1040 * the result buffer head is unmapped. If the create ==1, it will make sure
1041 * the buffer head is mapped.
1043 * It returns 0 if plain look up failed (blocks have not been allocated), in
1044 * that casem, buffer head is unmapped
1046 * It returns the error in case of allocation failure.
1048 int ext4_get_blocks_wrap(handle_t
*handle
, struct inode
*inode
, sector_t block
,
1049 unsigned long max_blocks
, struct buffer_head
*bh
,
1050 int create
, int extend_disksize
, int flag
)
1054 clear_buffer_mapped(bh
);
1057 * Try to see if we can get the block without requesting
1058 * for new file system block.
1060 down_read((&EXT4_I(inode
)->i_data_sem
));
1061 if (EXT4_I(inode
)->i_flags
& EXT4_EXTENTS_FL
) {
1062 retval
= ext4_ext_get_blocks(handle
, inode
, block
, max_blocks
,
1065 retval
= ext4_get_blocks_handle(handle
,
1066 inode
, block
, max_blocks
, bh
, 0, 0);
1068 up_read((&EXT4_I(inode
)->i_data_sem
));
1070 /* If it is only a block(s) look up */
1075 * Returns if the blocks have already allocated
1077 * Note that if blocks have been preallocated
1078 * ext4_ext_get_block() returns th create = 0
1079 * with buffer head unmapped.
1081 if (retval
> 0 && buffer_mapped(bh
))
1085 * New blocks allocate and/or writing to uninitialized extent
1086 * will possibly result in updating i_data, so we take
1087 * the write lock of i_data_sem, and call get_blocks()
1088 * with create == 1 flag.
1090 down_write((&EXT4_I(inode
)->i_data_sem
));
1093 * if the caller is from delayed allocation writeout path
1094 * we have already reserved fs blocks for allocation
1095 * let the underlying get_block() function know to
1096 * avoid double accounting
1099 EXT4_I(inode
)->i_delalloc_reserved_flag
= 1;
1101 * We need to check for EXT4 here because migrate
1102 * could have changed the inode type in between
1104 if (EXT4_I(inode
)->i_flags
& EXT4_EXTENTS_FL
) {
1105 retval
= ext4_ext_get_blocks(handle
, inode
, block
, max_blocks
,
1106 bh
, create
, extend_disksize
);
1108 retval
= ext4_get_blocks_handle(handle
, inode
, block
,
1109 max_blocks
, bh
, create
, extend_disksize
);
1111 if (retval
> 0 && buffer_new(bh
)) {
1113 * We allocated new blocks which will result in
1114 * i_data's format changing. Force the migrate
1115 * to fail by clearing migrate flags
1117 EXT4_I(inode
)->i_flags
= EXT4_I(inode
)->i_flags
&
1123 EXT4_I(inode
)->i_delalloc_reserved_flag
= 0;
1125 * Update reserved blocks/metadata blocks
1126 * after successful block allocation
1127 * which were deferred till now
1129 if ((retval
> 0) && buffer_delay(bh
))
1130 ext4_da_update_reserve_space(inode
, retval
);
1133 up_write((&EXT4_I(inode
)->i_data_sem
));
1137 /* Maximum number of blocks we map for direct IO at once. */
1138 #define DIO_MAX_BLOCKS 4096
1140 int ext4_get_block(struct inode
*inode
, sector_t iblock
,
1141 struct buffer_head
*bh_result
, int create
)
1143 handle_t
*handle
= ext4_journal_current_handle();
1144 int ret
= 0, started
= 0;
1145 unsigned max_blocks
= bh_result
->b_size
>> inode
->i_blkbits
;
1148 if (create
&& !handle
) {
1149 /* Direct IO write... */
1150 if (max_blocks
> DIO_MAX_BLOCKS
)
1151 max_blocks
= DIO_MAX_BLOCKS
;
1152 dio_credits
= ext4_chunk_trans_blocks(inode
, max_blocks
);
1153 handle
= ext4_journal_start(inode
, dio_credits
);
1154 if (IS_ERR(handle
)) {
1155 ret
= PTR_ERR(handle
);
1161 ret
= ext4_get_blocks_wrap(handle
, inode
, iblock
,
1162 max_blocks
, bh_result
, create
, 0, 0);
1164 bh_result
->b_size
= (ret
<< inode
->i_blkbits
);
1168 ext4_journal_stop(handle
);
1174 * `handle' can be NULL if create is zero
1176 struct buffer_head
*ext4_getblk(handle_t
*handle
, struct inode
*inode
,
1177 ext4_lblk_t block
, int create
, int *errp
)
1179 struct buffer_head dummy
;
1182 J_ASSERT(handle
!= NULL
|| create
== 0);
1185 dummy
.b_blocknr
= -1000;
1186 buffer_trace_init(&dummy
.b_history
);
1187 err
= ext4_get_blocks_wrap(handle
, inode
, block
, 1,
1188 &dummy
, create
, 1, 0);
1190 * ext4_get_blocks_handle() returns number of blocks
1191 * mapped. 0 in case of a HOLE.
1199 if (!err
&& buffer_mapped(&dummy
)) {
1200 struct buffer_head
*bh
;
1201 bh
= sb_getblk(inode
->i_sb
, dummy
.b_blocknr
);
1206 if (buffer_new(&dummy
)) {
1207 J_ASSERT(create
!= 0);
1208 J_ASSERT(handle
!= NULL
);
1211 * Now that we do not always journal data, we should
1212 * keep in mind whether this should always journal the
1213 * new buffer as metadata. For now, regular file
1214 * writes use ext4_get_block instead, so it's not a
1218 BUFFER_TRACE(bh
, "call get_create_access");
1219 fatal
= ext4_journal_get_create_access(handle
, bh
);
1220 if (!fatal
&& !buffer_uptodate(bh
)) {
1221 memset(bh
->b_data
, 0, inode
->i_sb
->s_blocksize
);
1222 set_buffer_uptodate(bh
);
1225 BUFFER_TRACE(bh
, "call ext4_journal_dirty_metadata");
1226 err
= ext4_journal_dirty_metadata(handle
, bh
);
1230 BUFFER_TRACE(bh
, "not a new buffer");
1243 struct buffer_head
*ext4_bread(handle_t
*handle
, struct inode
*inode
,
1244 ext4_lblk_t block
, int create
, int *err
)
1246 struct buffer_head
*bh
;
1248 bh
= ext4_getblk(handle
, inode
, block
, create
, err
);
1251 if (buffer_uptodate(bh
))
1253 ll_rw_block(READ_META
, 1, &bh
);
1255 if (buffer_uptodate(bh
))
1262 static int walk_page_buffers(handle_t
*handle
,
1263 struct buffer_head
*head
,
1267 int (*fn
)(handle_t
*handle
,
1268 struct buffer_head
*bh
))
1270 struct buffer_head
*bh
;
1271 unsigned block_start
, block_end
;
1272 unsigned blocksize
= head
->b_size
;
1274 struct buffer_head
*next
;
1276 for (bh
= head
, block_start
= 0;
1277 ret
== 0 && (bh
!= head
|| !block_start
);
1278 block_start
= block_end
, bh
= next
)
1280 next
= bh
->b_this_page
;
1281 block_end
= block_start
+ blocksize
;
1282 if (block_end
<= from
|| block_start
>= to
) {
1283 if (partial
&& !buffer_uptodate(bh
))
1287 err
= (*fn
)(handle
, bh
);
1295 * To preserve ordering, it is essential that the hole instantiation and
1296 * the data write be encapsulated in a single transaction. We cannot
1297 * close off a transaction and start a new one between the ext4_get_block()
1298 * and the commit_write(). So doing the jbd2_journal_start at the start of
1299 * prepare_write() is the right place.
1301 * Also, this function can nest inside ext4_writepage() ->
1302 * block_write_full_page(). In that case, we *know* that ext4_writepage()
1303 * has generated enough buffer credits to do the whole page. So we won't
1304 * block on the journal in that case, which is good, because the caller may
1307 * By accident, ext4 can be reentered when a transaction is open via
1308 * quota file writes. If we were to commit the transaction while thus
1309 * reentered, there can be a deadlock - we would be holding a quota
1310 * lock, and the commit would never complete if another thread had a
1311 * transaction open and was blocking on the quota lock - a ranking
1314 * So what we do is to rely on the fact that jbd2_journal_stop/journal_start
1315 * will _not_ run commit under these circumstances because handle->h_ref
1316 * is elevated. We'll still have enough credits for the tiny quotafile
1319 static int do_journal_get_write_access(handle_t
*handle
,
1320 struct buffer_head
*bh
)
1322 if (!buffer_mapped(bh
) || buffer_freed(bh
))
1324 return ext4_journal_get_write_access(handle
, bh
);
1327 static int ext4_write_begin(struct file
*file
, struct address_space
*mapping
,
1328 loff_t pos
, unsigned len
, unsigned flags
,
1329 struct page
**pagep
, void **fsdata
)
1331 struct inode
*inode
= mapping
->host
;
1332 int ret
, needed_blocks
= ext4_writepage_trans_blocks(inode
);
1339 index
= pos
>> PAGE_CACHE_SHIFT
;
1340 from
= pos
& (PAGE_CACHE_SIZE
- 1);
1344 handle
= ext4_journal_start(inode
, needed_blocks
);
1345 if (IS_ERR(handle
)) {
1346 ret
= PTR_ERR(handle
);
1350 /* We cannot recurse into the filesystem as the transaction is already
1352 flags
|= AOP_FLAG_NOFS
;
1354 page
= grab_cache_page_write_begin(mapping
, index
, flags
);
1356 ext4_journal_stop(handle
);
1362 ret
= block_write_begin(file
, mapping
, pos
, len
, flags
, pagep
, fsdata
,
1365 if (!ret
&& ext4_should_journal_data(inode
)) {
1366 ret
= walk_page_buffers(handle
, page_buffers(page
),
1367 from
, to
, NULL
, do_journal_get_write_access
);
1372 ext4_journal_stop(handle
);
1373 page_cache_release(page
);
1375 * block_write_begin may have instantiated a few blocks
1376 * outside i_size. Trim these off again. Don't need
1377 * i_size_read because we hold i_mutex.
1379 if (pos
+ len
> inode
->i_size
)
1380 vmtruncate(inode
, inode
->i_size
);
1383 if (ret
== -ENOSPC
&& ext4_should_retry_alloc(inode
->i_sb
, &retries
))
1389 /* For write_end() in data=journal mode */
1390 static int write_end_fn(handle_t
*handle
, struct buffer_head
*bh
)
1392 if (!buffer_mapped(bh
) || buffer_freed(bh
))
1394 set_buffer_uptodate(bh
);
1395 return ext4_journal_dirty_metadata(handle
, bh
);
1399 * We need to pick up the new inode size which generic_commit_write gave us
1400 * `file' can be NULL - eg, when called from page_symlink().
1402 * ext4 never places buffers on inode->i_mapping->private_list. metadata
1403 * buffers are managed internally.
1405 static int ext4_ordered_write_end(struct file
*file
,
1406 struct address_space
*mapping
,
1407 loff_t pos
, unsigned len
, unsigned copied
,
1408 struct page
*page
, void *fsdata
)
1410 handle_t
*handle
= ext4_journal_current_handle();
1411 struct inode
*inode
= mapping
->host
;
1414 ret
= ext4_jbd2_file_inode(handle
, inode
);
1419 new_i_size
= pos
+ copied
;
1420 if (new_i_size
> EXT4_I(inode
)->i_disksize
) {
1421 ext4_update_i_disksize(inode
, new_i_size
);
1422 /* We need to mark inode dirty even if
1423 * new_i_size is less that inode->i_size
1424 * bu greater than i_disksize.(hint delalloc)
1426 ext4_mark_inode_dirty(handle
, inode
);
1429 ret2
= generic_write_end(file
, mapping
, pos
, len
, copied
,
1435 ret2
= ext4_journal_stop(handle
);
1439 return ret
? ret
: copied
;
1442 static int ext4_writeback_write_end(struct file
*file
,
1443 struct address_space
*mapping
,
1444 loff_t pos
, unsigned len
, unsigned copied
,
1445 struct page
*page
, void *fsdata
)
1447 handle_t
*handle
= ext4_journal_current_handle();
1448 struct inode
*inode
= mapping
->host
;
1452 new_i_size
= pos
+ copied
;
1453 if (new_i_size
> EXT4_I(inode
)->i_disksize
) {
1454 ext4_update_i_disksize(inode
, new_i_size
);
1455 /* We need to mark inode dirty even if
1456 * new_i_size is less that inode->i_size
1457 * bu greater than i_disksize.(hint delalloc)
1459 ext4_mark_inode_dirty(handle
, inode
);
1462 ret2
= generic_write_end(file
, mapping
, pos
, len
, copied
,
1468 ret2
= ext4_journal_stop(handle
);
1472 return ret
? ret
: copied
;
1475 static int ext4_journalled_write_end(struct file
*file
,
1476 struct address_space
*mapping
,
1477 loff_t pos
, unsigned len
, unsigned copied
,
1478 struct page
*page
, void *fsdata
)
1480 handle_t
*handle
= ext4_journal_current_handle();
1481 struct inode
*inode
= mapping
->host
;
1487 from
= pos
& (PAGE_CACHE_SIZE
- 1);
1491 if (!PageUptodate(page
))
1493 page_zero_new_buffers(page
, from
+copied
, to
);
1496 ret
= walk_page_buffers(handle
, page_buffers(page
), from
,
1497 to
, &partial
, write_end_fn
);
1499 SetPageUptodate(page
);
1500 new_i_size
= pos
+ copied
;
1501 if (new_i_size
> inode
->i_size
)
1502 i_size_write(inode
, pos
+copied
);
1503 EXT4_I(inode
)->i_state
|= EXT4_STATE_JDATA
;
1504 if (new_i_size
> EXT4_I(inode
)->i_disksize
) {
1505 ext4_update_i_disksize(inode
, new_i_size
);
1506 ret2
= ext4_mark_inode_dirty(handle
, inode
);
1512 ret2
= ext4_journal_stop(handle
);
1515 page_cache_release(page
);
1517 return ret
? ret
: copied
;
1520 static int ext4_da_reserve_space(struct inode
*inode
, int nrblocks
)
1523 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
1524 unsigned long md_needed
, mdblocks
, total
= 0;
1527 * recalculate the amount of metadata blocks to reserve
1528 * in order to allocate nrblocks
1529 * worse case is one extent per block
1532 spin_lock(&EXT4_I(inode
)->i_block_reservation_lock
);
1533 total
= EXT4_I(inode
)->i_reserved_data_blocks
+ nrblocks
;
1534 mdblocks
= ext4_calc_metadata_amount(inode
, total
);
1535 BUG_ON(mdblocks
< EXT4_I(inode
)->i_reserved_meta_blocks
);
1537 md_needed
= mdblocks
- EXT4_I(inode
)->i_reserved_meta_blocks
;
1538 total
= md_needed
+ nrblocks
;
1540 if (ext4_claim_free_blocks(sbi
, total
)) {
1541 spin_unlock(&EXT4_I(inode
)->i_block_reservation_lock
);
1542 if (ext4_should_retry_alloc(inode
->i_sb
, &retries
)) {
1548 EXT4_I(inode
)->i_reserved_data_blocks
+= nrblocks
;
1549 EXT4_I(inode
)->i_reserved_meta_blocks
= mdblocks
;
1551 spin_unlock(&EXT4_I(inode
)->i_block_reservation_lock
);
1552 return 0; /* success */
1555 static void ext4_da_release_space(struct inode
*inode
, int to_free
)
1557 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
1558 int total
, mdb
, mdb_free
, release
;
1561 return; /* Nothing to release, exit */
1563 spin_lock(&EXT4_I(inode
)->i_block_reservation_lock
);
1565 if (!EXT4_I(inode
)->i_reserved_data_blocks
) {
1567 * if there is no reserved blocks, but we try to free some
1568 * then the counter is messed up somewhere.
1569 * but since this function is called from invalidate
1570 * page, it's harmless to return without any action
1572 printk(KERN_INFO
"ext4 delalloc try to release %d reserved "
1573 "blocks for inode %lu, but there is no reserved "
1574 "data blocks\n", to_free
, inode
->i_ino
);
1575 spin_unlock(&EXT4_I(inode
)->i_block_reservation_lock
);
1579 /* recalculate the number of metablocks still need to be reserved */
1580 total
= EXT4_I(inode
)->i_reserved_data_blocks
- to_free
;
1581 mdb
= ext4_calc_metadata_amount(inode
, total
);
1583 /* figure out how many metablocks to release */
1584 BUG_ON(mdb
> EXT4_I(inode
)->i_reserved_meta_blocks
);
1585 mdb_free
= EXT4_I(inode
)->i_reserved_meta_blocks
- mdb
;
1587 release
= to_free
+ mdb_free
;
1589 /* update fs dirty blocks counter for truncate case */
1590 percpu_counter_sub(&sbi
->s_dirtyblocks_counter
, release
);
1592 /* update per-inode reservations */
1593 BUG_ON(to_free
> EXT4_I(inode
)->i_reserved_data_blocks
);
1594 EXT4_I(inode
)->i_reserved_data_blocks
-= to_free
;
1596 BUG_ON(mdb
> EXT4_I(inode
)->i_reserved_meta_blocks
);
1597 EXT4_I(inode
)->i_reserved_meta_blocks
= mdb
;
1598 spin_unlock(&EXT4_I(inode
)->i_block_reservation_lock
);
1601 static void ext4_da_page_release_reservation(struct page
*page
,
1602 unsigned long offset
)
1605 struct buffer_head
*head
, *bh
;
1606 unsigned int curr_off
= 0;
1608 head
= page_buffers(page
);
1611 unsigned int next_off
= curr_off
+ bh
->b_size
;
1613 if ((offset
<= curr_off
) && (buffer_delay(bh
))) {
1615 clear_buffer_delay(bh
);
1617 curr_off
= next_off
;
1618 } while ((bh
= bh
->b_this_page
) != head
);
1619 ext4_da_release_space(page
->mapping
->host
, to_release
);
1623 * Delayed allocation stuff
1626 struct mpage_da_data
{
1627 struct inode
*inode
;
1628 struct buffer_head lbh
; /* extent of blocks */
1629 unsigned long first_page
, next_page
; /* extent of pages */
1630 get_block_t
*get_block
;
1631 struct writeback_control
*wbc
;
1638 * mpage_da_submit_io - walks through extent of pages and try to write
1639 * them with writepage() call back
1641 * @mpd->inode: inode
1642 * @mpd->first_page: first page of the extent
1643 * @mpd->next_page: page after the last page of the extent
1644 * @mpd->get_block: the filesystem's block mapper function
1646 * By the time mpage_da_submit_io() is called we expect all blocks
1647 * to be allocated. this may be wrong if allocation failed.
1649 * As pages are already locked by write_cache_pages(), we can't use it
1651 static int mpage_da_submit_io(struct mpage_da_data
*mpd
)
1654 struct pagevec pvec
;
1655 unsigned long index
, end
;
1656 int ret
= 0, err
, nr_pages
, i
;
1657 struct inode
*inode
= mpd
->inode
;
1658 struct address_space
*mapping
= inode
->i_mapping
;
1660 BUG_ON(mpd
->next_page
<= mpd
->first_page
);
1662 * We need to start from the first_page to the next_page - 1
1663 * to make sure we also write the mapped dirty buffer_heads.
1664 * If we look at mpd->lbh.b_blocknr we would only be looking
1665 * at the currently mapped buffer_heads.
1667 index
= mpd
->first_page
;
1668 end
= mpd
->next_page
- 1;
1670 pagevec_init(&pvec
, 0);
1671 while (index
<= end
) {
1672 nr_pages
= pagevec_lookup(&pvec
, mapping
, index
, PAGEVEC_SIZE
);
1675 for (i
= 0; i
< nr_pages
; i
++) {
1676 struct page
*page
= pvec
.pages
[i
];
1678 index
= page
->index
;
1683 BUG_ON(!PageLocked(page
));
1684 BUG_ON(PageWriteback(page
));
1686 pages_skipped
= mpd
->wbc
->pages_skipped
;
1687 err
= mapping
->a_ops
->writepage(page
, mpd
->wbc
);
1688 if (!err
&& (pages_skipped
== mpd
->wbc
->pages_skipped
))
1690 * have successfully written the page
1691 * without skipping the same
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
);
1803 static void ext4_da_block_invalidatepages(struct mpage_da_data
*mpd
,
1804 sector_t logical
, long blk_cnt
)
1808 struct pagevec pvec
;
1809 struct inode
*inode
= mpd
->inode
;
1810 struct address_space
*mapping
= inode
->i_mapping
;
1812 index
= logical
>> (PAGE_CACHE_SHIFT
- inode
->i_blkbits
);
1813 end
= (logical
+ blk_cnt
- 1) >>
1814 (PAGE_CACHE_SHIFT
- inode
->i_blkbits
);
1815 while (index
<= end
) {
1816 nr_pages
= pagevec_lookup(&pvec
, mapping
, index
, PAGEVEC_SIZE
);
1819 for (i
= 0; i
< nr_pages
; i
++) {
1820 struct page
*page
= pvec
.pages
[i
];
1821 index
= page
->index
;
1826 BUG_ON(!PageLocked(page
));
1827 BUG_ON(PageWriteback(page
));
1828 block_invalidatepage(page
, 0);
1829 ClearPageUptodate(page
);
1836 static void ext4_print_free_blocks(struct inode
*inode
)
1838 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
1839 printk(KERN_EMERG
"Total free blocks count %lld\n",
1840 ext4_count_free_blocks(inode
->i_sb
));
1841 printk(KERN_EMERG
"Free/Dirty block details\n");
1842 printk(KERN_EMERG
"free_blocks=%lld\n",
1843 percpu_counter_sum(&sbi
->s_freeblocks_counter
));
1844 printk(KERN_EMERG
"dirty_blocks=%lld\n",
1845 percpu_counter_sum(&sbi
->s_dirtyblocks_counter
));
1846 printk(KERN_EMERG
"Block reservation details\n");
1847 printk(KERN_EMERG
"i_reserved_data_blocks=%lu\n",
1848 EXT4_I(inode
)->i_reserved_data_blocks
);
1849 printk(KERN_EMERG
"i_reserved_meta_blocks=%lu\n",
1850 EXT4_I(inode
)->i_reserved_meta_blocks
);
1855 * mpage_da_map_blocks - go through given space
1857 * @mpd->lbh - bh describing space
1858 * @mpd->get_block - the filesystem's block mapper function
1860 * The function skips space we know is already mapped to disk blocks.
1863 static int mpage_da_map_blocks(struct mpage_da_data
*mpd
)
1866 struct buffer_head
new;
1867 struct buffer_head
*lbh
= &mpd
->lbh
;
1871 * We consider only non-mapped and non-allocated blocks
1873 if (buffer_mapped(lbh
) && !buffer_delay(lbh
))
1875 new.b_state
= lbh
->b_state
;
1877 new.b_size
= lbh
->b_size
;
1878 next
= lbh
->b_blocknr
;
1880 * If we didn't accumulate anything
1881 * to write simply return
1885 err
= mpd
->get_block(mpd
->inode
, next
, &new, 1);
1888 /* If get block returns with error
1889 * we simply return. Later writepage
1890 * will redirty the page and writepages
1891 * will find the dirty page again
1896 if (err
== -ENOSPC
&&
1897 ext4_count_free_blocks(mpd
->inode
->i_sb
)) {
1903 * get block failure will cause us
1904 * to loop in writepages. Because
1905 * a_ops->writepage won't be able to
1906 * make progress. The page will be redirtied
1907 * by writepage and writepages will again
1908 * try to write the same.
1910 printk(KERN_EMERG
"%s block allocation failed for inode %lu "
1911 "at logical offset %llu with max blocks "
1912 "%zd with error %d\n",
1913 __func__
, mpd
->inode
->i_ino
,
1914 (unsigned long long)next
,
1915 lbh
->b_size
>> mpd
->inode
->i_blkbits
, err
);
1916 printk(KERN_EMERG
"This should not happen.!! "
1917 "Data will be lost\n");
1918 if (err
== -ENOSPC
) {
1919 ext4_print_free_blocks(mpd
->inode
);
1921 /* invlaidate all the pages */
1922 ext4_da_block_invalidatepages(mpd
, next
,
1923 lbh
->b_size
>> mpd
->inode
->i_blkbits
);
1926 BUG_ON(new.b_size
== 0);
1928 if (buffer_new(&new))
1929 __unmap_underlying_blocks(mpd
->inode
, &new);
1932 * If blocks are delayed marked, we need to
1933 * put actual blocknr and drop delayed bit
1935 if (buffer_delay(lbh
) || buffer_unwritten(lbh
))
1936 mpage_put_bnr_to_bhs(mpd
, next
, &new);
1941 #define BH_FLAGS ((1 << BH_Uptodate) | (1 << BH_Mapped) | \
1942 (1 << BH_Delay) | (1 << BH_Unwritten))
1945 * mpage_add_bh_to_extent - try to add one more block to extent of blocks
1947 * @mpd->lbh - extent of blocks
1948 * @logical - logical number of the block in the file
1949 * @bh - bh of the block (used to access block's state)
1951 * the function is used to collect contig. blocks in same state
1953 static void mpage_add_bh_to_extent(struct mpage_da_data
*mpd
,
1954 sector_t logical
, struct buffer_head
*bh
)
1957 size_t b_size
= bh
->b_size
;
1958 struct buffer_head
*lbh
= &mpd
->lbh
;
1959 int nrblocks
= lbh
->b_size
>> mpd
->inode
->i_blkbits
;
1961 /* check if thereserved journal credits might overflow */
1962 if (!(EXT4_I(mpd
->inode
)->i_flags
& EXT4_EXTENTS_FL
)) {
1963 if (nrblocks
>= EXT4_MAX_TRANS_DATA
) {
1965 * With non-extent format we are limited by the journal
1966 * credit available. Total credit needed to insert
1967 * nrblocks contiguous blocks is dependent on the
1968 * nrblocks. So limit nrblocks.
1971 } else if ((nrblocks
+ (b_size
>> mpd
->inode
->i_blkbits
)) >
1972 EXT4_MAX_TRANS_DATA
) {
1974 * Adding the new buffer_head would make it cross the
1975 * allowed limit for which we have journal credit
1976 * reserved. So limit the new bh->b_size
1978 b_size
= (EXT4_MAX_TRANS_DATA
- nrblocks
) <<
1979 mpd
->inode
->i_blkbits
;
1980 /* we will do mpage_da_submit_io in the next loop */
1984 * First block in the extent
1986 if (lbh
->b_size
== 0) {
1987 lbh
->b_blocknr
= logical
;
1988 lbh
->b_size
= b_size
;
1989 lbh
->b_state
= bh
->b_state
& BH_FLAGS
;
1993 next
= lbh
->b_blocknr
+ nrblocks
;
1995 * Can we merge the block to our big extent?
1997 if (logical
== next
&& (bh
->b_state
& BH_FLAGS
) == lbh
->b_state
) {
1998 lbh
->b_size
+= b_size
;
2004 * We couldn't merge the block to our extent, so we
2005 * need to flush current extent and start new one
2007 if (mpage_da_map_blocks(mpd
) == 0)
2008 mpage_da_submit_io(mpd
);
2014 * __mpage_da_writepage - finds extent of pages and blocks
2016 * @page: page to consider
2017 * @wbc: not used, we just follow rules
2020 * The function finds extents of pages and scan them for all blocks.
2022 static int __mpage_da_writepage(struct page
*page
,
2023 struct writeback_control
*wbc
, void *data
)
2025 struct mpage_da_data
*mpd
= data
;
2026 struct inode
*inode
= mpd
->inode
;
2027 struct buffer_head
*bh
, *head
, fake
;
2032 * Rest of the page in the page_vec
2033 * redirty then and skip then. We will
2034 * try to to write them again after
2035 * starting a new transaction
2037 redirty_page_for_writepage(wbc
, page
);
2039 return MPAGE_DA_EXTENT_TAIL
;
2042 * Can we merge this page to current extent?
2044 if (mpd
->next_page
!= page
->index
) {
2046 * Nope, we can't. So, we map non-allocated blocks
2047 * and start IO on them using writepage()
2049 if (mpd
->next_page
!= mpd
->first_page
) {
2050 if (mpage_da_map_blocks(mpd
) == 0)
2051 mpage_da_submit_io(mpd
);
2053 * skip rest of the page in the page_vec
2056 redirty_page_for_writepage(wbc
, page
);
2058 return MPAGE_DA_EXTENT_TAIL
;
2062 * Start next extent of pages ...
2064 mpd
->first_page
= page
->index
;
2069 mpd
->lbh
.b_size
= 0;
2070 mpd
->lbh
.b_state
= 0;
2071 mpd
->lbh
.b_blocknr
= 0;
2074 mpd
->next_page
= page
->index
+ 1;
2075 logical
= (sector_t
) page
->index
<<
2076 (PAGE_CACHE_SHIFT
- inode
->i_blkbits
);
2078 if (!page_has_buffers(page
)) {
2080 * There is no attached buffer heads yet (mmap?)
2081 * we treat the page asfull of dirty blocks
2084 bh
->b_size
= PAGE_CACHE_SIZE
;
2086 set_buffer_dirty(bh
);
2087 set_buffer_uptodate(bh
);
2088 mpage_add_bh_to_extent(mpd
, logical
, bh
);
2090 return MPAGE_DA_EXTENT_TAIL
;
2093 * Page with regular buffer heads, just add all dirty ones
2095 head
= page_buffers(page
);
2098 BUG_ON(buffer_locked(bh
));
2100 * We need to try to allocate
2101 * unmapped blocks in the same page.
2102 * Otherwise we won't make progress
2103 * with the page in ext4_da_writepage
2105 if (buffer_dirty(bh
) &&
2106 (!buffer_mapped(bh
) || buffer_delay(bh
))) {
2107 mpage_add_bh_to_extent(mpd
, logical
, bh
);
2109 return MPAGE_DA_EXTENT_TAIL
;
2110 } else if (buffer_dirty(bh
) && (buffer_mapped(bh
))) {
2112 * mapped dirty buffer. We need to update
2113 * the b_state because we look at
2114 * b_state in mpage_da_map_blocks. We don't
2115 * update b_size because if we find an
2116 * unmapped buffer_head later we need to
2117 * use the b_state flag of that buffer_head.
2119 if (mpd
->lbh
.b_size
== 0)
2121 bh
->b_state
& BH_FLAGS
;
2124 } while ((bh
= bh
->b_this_page
) != head
);
2131 * mpage_da_writepages - walk the list of dirty pages of the given
2132 * address space, allocates non-allocated blocks, maps newly-allocated
2133 * blocks to existing bhs and issue IO them
2135 * @mapping: address space structure to write
2136 * @wbc: subtract the number of written pages from *@wbc->nr_to_write
2137 * @get_block: the filesystem's block mapper function.
2139 * This is a library function, which implements the writepages()
2140 * address_space_operation.
2142 static int mpage_da_writepages(struct address_space
*mapping
,
2143 struct writeback_control
*wbc
,
2144 struct mpage_da_data
*mpd
)
2148 if (!mpd
->get_block
)
2149 return generic_writepages(mapping
, wbc
);
2151 mpd
->lbh
.b_size
= 0;
2152 mpd
->lbh
.b_state
= 0;
2153 mpd
->lbh
.b_blocknr
= 0;
2154 mpd
->first_page
= 0;
2157 mpd
->pages_written
= 0;
2160 ret
= write_cache_pages(mapping
, wbc
, __mpage_da_writepage
, mpd
);
2162 * Handle last extent of pages
2164 if (!mpd
->io_done
&& mpd
->next_page
!= mpd
->first_page
) {
2165 if (mpage_da_map_blocks(mpd
) == 0)
2166 mpage_da_submit_io(mpd
);
2169 ret
= MPAGE_DA_EXTENT_TAIL
;
2171 wbc
->nr_to_write
-= mpd
->pages_written
;
2176 * this is a special callback for ->write_begin() only
2177 * it's intention is to return mapped block or reserve space
2179 static int ext4_da_get_block_prep(struct inode
*inode
, sector_t iblock
,
2180 struct buffer_head
*bh_result
, int create
)
2184 BUG_ON(create
== 0);
2185 BUG_ON(bh_result
->b_size
!= inode
->i_sb
->s_blocksize
);
2188 * first, we need to know whether the block is allocated already
2189 * preallocated blocks are unmapped but should treated
2190 * the same as allocated blocks.
2192 ret
= ext4_get_blocks_wrap(NULL
, inode
, iblock
, 1, bh_result
, 0, 0, 0);
2193 if ((ret
== 0) && !buffer_delay(bh_result
)) {
2194 /* the block isn't (pre)allocated yet, let's reserve space */
2196 * XXX: __block_prepare_write() unmaps passed block,
2199 ret
= ext4_da_reserve_space(inode
, 1);
2201 /* not enough space to reserve */
2204 map_bh(bh_result
, inode
->i_sb
, 0);
2205 set_buffer_new(bh_result
);
2206 set_buffer_delay(bh_result
);
2207 } else if (ret
> 0) {
2208 bh_result
->b_size
= (ret
<< inode
->i_blkbits
);
2214 #define EXT4_DELALLOC_RSVED 1
2215 static int ext4_da_get_block_write(struct inode
*inode
, sector_t iblock
,
2216 struct buffer_head
*bh_result
, int create
)
2219 unsigned max_blocks
= bh_result
->b_size
>> inode
->i_blkbits
;
2220 loff_t disksize
= EXT4_I(inode
)->i_disksize
;
2221 handle_t
*handle
= NULL
;
2223 handle
= ext4_journal_current_handle();
2225 ret
= ext4_get_blocks_wrap(handle
, inode
, iblock
, max_blocks
,
2226 bh_result
, create
, 0, EXT4_DELALLOC_RSVED
);
2229 bh_result
->b_size
= (ret
<< inode
->i_blkbits
);
2231 if (ext4_should_order_data(inode
)) {
2233 retval
= ext4_jbd2_file_inode(handle
, inode
);
2236 * Failed to add inode for ordered
2237 * mode. Don't update file size
2243 * Update on-disk size along with block allocation
2244 * we don't use 'extend_disksize' as size may change
2245 * within already allocated block -bzzz
2247 disksize
= ((loff_t
) iblock
+ ret
) << inode
->i_blkbits
;
2248 if (disksize
> i_size_read(inode
))
2249 disksize
= i_size_read(inode
);
2250 if (disksize
> EXT4_I(inode
)->i_disksize
) {
2251 ext4_update_i_disksize(inode
, disksize
);
2252 ret
= ext4_mark_inode_dirty(handle
, inode
);
2260 static int ext4_bh_unmapped_or_delay(handle_t
*handle
, struct buffer_head
*bh
)
2263 * unmapped buffer is possible for holes.
2264 * delay buffer is possible with delayed allocation
2266 return ((!buffer_mapped(bh
) || buffer_delay(bh
)) && buffer_dirty(bh
));
2269 static int ext4_normal_get_block_write(struct inode
*inode
, sector_t iblock
,
2270 struct buffer_head
*bh_result
, int create
)
2273 unsigned max_blocks
= bh_result
->b_size
>> inode
->i_blkbits
;
2276 * we don't want to do block allocation in writepage
2277 * so call get_block_wrap with create = 0
2279 ret
= ext4_get_blocks_wrap(NULL
, inode
, iblock
, max_blocks
,
2280 bh_result
, 0, 0, 0);
2282 bh_result
->b_size
= (ret
<< inode
->i_blkbits
);
2289 * get called vi ext4_da_writepages after taking page lock (have journal handle)
2290 * get called via journal_submit_inode_data_buffers (no journal handle)
2291 * get called via shrink_page_list via pdflush (no journal handle)
2292 * or grab_page_cache when doing write_begin (have journal handle)
2294 static int ext4_da_writepage(struct page
*page
,
2295 struct writeback_control
*wbc
)
2300 struct buffer_head
*page_bufs
;
2301 struct inode
*inode
= page
->mapping
->host
;
2303 size
= i_size_read(inode
);
2304 if (page
->index
== size
>> PAGE_CACHE_SHIFT
)
2305 len
= size
& ~PAGE_CACHE_MASK
;
2307 len
= PAGE_CACHE_SIZE
;
2309 if (page_has_buffers(page
)) {
2310 page_bufs
= page_buffers(page
);
2311 if (walk_page_buffers(NULL
, page_bufs
, 0, len
, NULL
,
2312 ext4_bh_unmapped_or_delay
)) {
2314 * We don't want to do block allocation
2315 * So redirty the page and return
2316 * We may reach here when we do a journal commit
2317 * via journal_submit_inode_data_buffers.
2318 * If we don't have mapping block we just ignore
2319 * them. We can also reach here via shrink_page_list
2321 redirty_page_for_writepage(wbc
, page
);
2327 * The test for page_has_buffers() is subtle:
2328 * We know the page is dirty but it lost buffers. That means
2329 * that at some moment in time after write_begin()/write_end()
2330 * has been called all buffers have been clean and thus they
2331 * must have been written at least once. So they are all
2332 * mapped and we can happily proceed with mapping them
2333 * and writing the page.
2335 * Try to initialize the buffer_heads and check whether
2336 * all are mapped and non delay. We don't want to
2337 * do block allocation here.
2339 ret
= block_prepare_write(page
, 0, PAGE_CACHE_SIZE
,
2340 ext4_normal_get_block_write
);
2342 page_bufs
= page_buffers(page
);
2343 /* check whether all are mapped and non delay */
2344 if (walk_page_buffers(NULL
, page_bufs
, 0, len
, NULL
,
2345 ext4_bh_unmapped_or_delay
)) {
2346 redirty_page_for_writepage(wbc
, page
);
2352 * We can't do block allocation here
2353 * so just redity the page and unlock
2356 redirty_page_for_writepage(wbc
, page
);
2360 /* now mark the buffer_heads as dirty and uptodate */
2361 block_commit_write(page
, 0, PAGE_CACHE_SIZE
);
2364 if (test_opt(inode
->i_sb
, NOBH
) && ext4_should_writeback_data(inode
))
2365 ret
= nobh_writepage(page
, ext4_normal_get_block_write
, wbc
);
2367 ret
= block_write_full_page(page
,
2368 ext4_normal_get_block_write
,
2375 * This is called via ext4_da_writepages() to
2376 * calulate the total number of credits to reserve to fit
2377 * a single extent allocation into a single transaction,
2378 * ext4_da_writpeages() will loop calling this before
2379 * the block allocation.
2382 static int ext4_da_writepages_trans_blocks(struct inode
*inode
)
2384 int max_blocks
= EXT4_I(inode
)->i_reserved_data_blocks
;
2387 * With non-extent format the journal credit needed to
2388 * insert nrblocks contiguous block is dependent on
2389 * number of contiguous block. So we will limit
2390 * number of contiguous block to a sane value
2392 if (!(inode
->i_flags
& EXT4_EXTENTS_FL
) &&
2393 (max_blocks
> EXT4_MAX_TRANS_DATA
))
2394 max_blocks
= EXT4_MAX_TRANS_DATA
;
2396 return ext4_chunk_trans_blocks(inode
, max_blocks
);
2399 static int ext4_da_writepages(struct address_space
*mapping
,
2400 struct writeback_control
*wbc
)
2403 int range_whole
= 0;
2404 handle_t
*handle
= NULL
;
2405 struct mpage_da_data mpd
;
2406 struct inode
*inode
= mapping
->host
;
2407 int no_nrwrite_index_update
;
2408 long pages_written
= 0, pages_skipped
;
2409 int range_cyclic
, cycled
= 1, io_done
= 0;
2410 int needed_blocks
, ret
= 0, nr_to_writebump
= 0;
2411 struct ext4_sb_info
*sbi
= EXT4_SB(mapping
->host
->i_sb
);
2414 * No pages to write? This is mainly a kludge to avoid starting
2415 * a transaction for special inodes like journal inode on last iput()
2416 * because that could violate lock ordering on umount
2418 if (!mapping
->nrpages
|| !mapping_tagged(mapping
, PAGECACHE_TAG_DIRTY
))
2422 * If the filesystem has aborted, it is read-only, so return
2423 * right away instead of dumping stack traces later on that
2424 * will obscure the real source of the problem. We test
2425 * EXT4_MOUNT_ABORT instead of sb->s_flag's MS_RDONLY because
2426 * the latter could be true if the filesystem is mounted
2427 * read-only, and in that case, ext4_da_writepages should
2428 * *never* be called, so if that ever happens, we would want
2431 if (unlikely(sbi
->s_mount_opt
& EXT4_MOUNT_ABORT
))
2435 * Make sure nr_to_write is >= sbi->s_mb_stream_request
2436 * This make sure small files blocks are allocated in
2437 * single attempt. This ensure that small files
2438 * get less fragmented.
2440 if (wbc
->nr_to_write
< sbi
->s_mb_stream_request
) {
2441 nr_to_writebump
= sbi
->s_mb_stream_request
- wbc
->nr_to_write
;
2442 wbc
->nr_to_write
= sbi
->s_mb_stream_request
;
2444 if (wbc
->range_start
== 0 && wbc
->range_end
== LLONG_MAX
)
2447 range_cyclic
= wbc
->range_cyclic
;
2448 if (wbc
->range_cyclic
) {
2449 index
= mapping
->writeback_index
;
2452 wbc
->range_start
= index
<< PAGE_CACHE_SHIFT
;
2453 wbc
->range_end
= LLONG_MAX
;
2454 wbc
->range_cyclic
= 0;
2456 index
= wbc
->range_start
>> PAGE_CACHE_SHIFT
;
2459 mpd
.inode
= mapping
->host
;
2462 * we don't want write_cache_pages to update
2463 * nr_to_write and writeback_index
2465 no_nrwrite_index_update
= wbc
->no_nrwrite_index_update
;
2466 wbc
->no_nrwrite_index_update
= 1;
2467 pages_skipped
= wbc
->pages_skipped
;
2470 while (!ret
&& wbc
->nr_to_write
> 0) {
2473 * we insert one extent at a time. So we need
2474 * credit needed for single extent allocation.
2475 * journalled mode is currently not supported
2478 BUG_ON(ext4_should_journal_data(inode
));
2479 needed_blocks
= ext4_da_writepages_trans_blocks(inode
);
2481 /* start a new transaction*/
2482 handle
= ext4_journal_start(inode
, needed_blocks
);
2483 if (IS_ERR(handle
)) {
2484 ret
= PTR_ERR(handle
);
2485 printk(KERN_CRIT
"%s: jbd2_start: "
2486 "%ld pages, ino %lu; err %d\n", __func__
,
2487 wbc
->nr_to_write
, inode
->i_ino
, ret
);
2489 goto out_writepages
;
2491 mpd
.get_block
= ext4_da_get_block_write
;
2492 ret
= mpage_da_writepages(mapping
, wbc
, &mpd
);
2494 ext4_journal_stop(handle
);
2496 if (mpd
.retval
== -ENOSPC
) {
2497 /* commit the transaction which would
2498 * free blocks released in the transaction
2501 jbd2_journal_force_commit_nested(sbi
->s_journal
);
2502 wbc
->pages_skipped
= pages_skipped
;
2504 } else if (ret
== MPAGE_DA_EXTENT_TAIL
) {
2506 * got one extent now try with
2509 pages_written
+= mpd
.pages_written
;
2510 wbc
->pages_skipped
= pages_skipped
;
2513 } else if (wbc
->nr_to_write
)
2515 * There is no more writeout needed
2516 * or we requested for a noblocking writeout
2517 * and we found the device congested
2521 if (!io_done
&& !cycled
) {
2524 wbc
->range_start
= index
<< PAGE_CACHE_SHIFT
;
2525 wbc
->range_end
= mapping
->writeback_index
- 1;
2528 if (pages_skipped
!= wbc
->pages_skipped
)
2529 printk(KERN_EMERG
"This should not happen leaving %s "
2530 "with nr_to_write = %ld ret = %d\n",
2531 __func__
, wbc
->nr_to_write
, ret
);
2534 index
+= pages_written
;
2535 wbc
->range_cyclic
= range_cyclic
;
2536 if (wbc
->range_cyclic
|| (range_whole
&& wbc
->nr_to_write
> 0))
2538 * set the writeback_index so that range_cyclic
2539 * mode will write it back later
2541 mapping
->writeback_index
= index
;
2544 if (!no_nrwrite_index_update
)
2545 wbc
->no_nrwrite_index_update
= 0;
2546 wbc
->nr_to_write
-= nr_to_writebump
;
2550 #define FALL_BACK_TO_NONDELALLOC 1
2551 static int ext4_nonda_switch(struct super_block
*sb
)
2553 s64 free_blocks
, dirty_blocks
;
2554 struct ext4_sb_info
*sbi
= EXT4_SB(sb
);
2557 * switch to non delalloc mode if we are running low
2558 * on free block. The free block accounting via percpu
2559 * counters can get slightly wrong with FBC_BATCH getting
2560 * accumulated on each CPU without updating global counters
2561 * Delalloc need an accurate free block accounting. So switch
2562 * to non delalloc when we are near to error range.
2564 free_blocks
= percpu_counter_read_positive(&sbi
->s_freeblocks_counter
);
2565 dirty_blocks
= percpu_counter_read_positive(&sbi
->s_dirtyblocks_counter
);
2566 if (2 * free_blocks
< 3 * dirty_blocks
||
2567 free_blocks
< (dirty_blocks
+ EXT4_FREEBLOCKS_WATERMARK
)) {
2569 * free block count is less that 150% of dirty blocks
2570 * or free blocks is less that watermark
2577 static int ext4_da_write_begin(struct file
*file
, struct address_space
*mapping
,
2578 loff_t pos
, unsigned len
, unsigned flags
,
2579 struct page
**pagep
, void **fsdata
)
2581 int ret
, retries
= 0;
2585 struct inode
*inode
= mapping
->host
;
2588 index
= pos
>> PAGE_CACHE_SHIFT
;
2589 from
= pos
& (PAGE_CACHE_SIZE
- 1);
2592 if (ext4_nonda_switch(inode
->i_sb
)) {
2593 *fsdata
= (void *)FALL_BACK_TO_NONDELALLOC
;
2594 return ext4_write_begin(file
, mapping
, pos
,
2595 len
, flags
, pagep
, fsdata
);
2597 *fsdata
= (void *)0;
2600 * With delayed allocation, we don't log the i_disksize update
2601 * if there is delayed block allocation. But we still need
2602 * to journalling the i_disksize update if writes to the end
2603 * of file which has an already mapped buffer.
2605 handle
= ext4_journal_start(inode
, 1);
2606 if (IS_ERR(handle
)) {
2607 ret
= PTR_ERR(handle
);
2610 /* We cannot recurse into the filesystem as the transaction is already
2612 flags
|= AOP_FLAG_NOFS
;
2614 page
= grab_cache_page_write_begin(mapping
, index
, flags
);
2616 ext4_journal_stop(handle
);
2622 ret
= block_write_begin(file
, mapping
, pos
, len
, flags
, pagep
, fsdata
,
2623 ext4_da_get_block_prep
);
2626 ext4_journal_stop(handle
);
2627 page_cache_release(page
);
2629 * block_write_begin may have instantiated a few blocks
2630 * outside i_size. Trim these off again. Don't need
2631 * i_size_read because we hold i_mutex.
2633 if (pos
+ len
> inode
->i_size
)
2634 vmtruncate(inode
, inode
->i_size
);
2637 if (ret
== -ENOSPC
&& ext4_should_retry_alloc(inode
->i_sb
, &retries
))
2644 * Check if we should update i_disksize
2645 * when write to the end of file but not require block allocation
2647 static int ext4_da_should_update_i_disksize(struct page
*page
,
2648 unsigned long offset
)
2650 struct buffer_head
*bh
;
2651 struct inode
*inode
= page
->mapping
->host
;
2655 bh
= page_buffers(page
);
2656 idx
= offset
>> inode
->i_blkbits
;
2658 for (i
= 0; i
< idx
; i
++)
2659 bh
= bh
->b_this_page
;
2661 if (!buffer_mapped(bh
) || (buffer_delay(bh
)))
2666 static int ext4_da_write_end(struct file
*file
,
2667 struct address_space
*mapping
,
2668 loff_t pos
, unsigned len
, unsigned copied
,
2669 struct page
*page
, void *fsdata
)
2671 struct inode
*inode
= mapping
->host
;
2673 handle_t
*handle
= ext4_journal_current_handle();
2675 unsigned long start
, end
;
2676 int write_mode
= (int)(unsigned long)fsdata
;
2678 if (write_mode
== FALL_BACK_TO_NONDELALLOC
) {
2679 if (ext4_should_order_data(inode
)) {
2680 return ext4_ordered_write_end(file
, mapping
, pos
,
2681 len
, copied
, page
, fsdata
);
2682 } else if (ext4_should_writeback_data(inode
)) {
2683 return ext4_writeback_write_end(file
, mapping
, pos
,
2684 len
, copied
, page
, fsdata
);
2690 start
= pos
& (PAGE_CACHE_SIZE
- 1);
2691 end
= start
+ copied
- 1;
2694 * generic_write_end() will run mark_inode_dirty() if i_size
2695 * changes. So let's piggyback the i_disksize mark_inode_dirty
2699 new_i_size
= pos
+ copied
;
2700 if (new_i_size
> EXT4_I(inode
)->i_disksize
) {
2701 if (ext4_da_should_update_i_disksize(page
, end
)) {
2702 down_write(&EXT4_I(inode
)->i_data_sem
);
2703 if (new_i_size
> EXT4_I(inode
)->i_disksize
) {
2705 * Updating i_disksize when extending file
2706 * without needing block allocation
2708 if (ext4_should_order_data(inode
))
2709 ret
= ext4_jbd2_file_inode(handle
,
2712 EXT4_I(inode
)->i_disksize
= new_i_size
;
2714 up_write(&EXT4_I(inode
)->i_data_sem
);
2715 /* We need to mark inode dirty even if
2716 * new_i_size is less that inode->i_size
2717 * bu greater than i_disksize.(hint delalloc)
2719 ext4_mark_inode_dirty(handle
, inode
);
2722 ret2
= generic_write_end(file
, mapping
, pos
, len
, copied
,
2727 ret2
= ext4_journal_stop(handle
);
2731 return ret
? ret
: copied
;
2734 static void ext4_da_invalidatepage(struct page
*page
, unsigned long offset
)
2737 * Drop reserved blocks
2739 BUG_ON(!PageLocked(page
));
2740 if (!page_has_buffers(page
))
2743 ext4_da_page_release_reservation(page
, offset
);
2746 ext4_invalidatepage(page
, offset
);
2753 * bmap() is special. It gets used by applications such as lilo and by
2754 * the swapper to find the on-disk block of a specific piece of data.
2756 * Naturally, this is dangerous if the block concerned is still in the
2757 * journal. If somebody makes a swapfile on an ext4 data-journaling
2758 * filesystem and enables swap, then they may get a nasty shock when the
2759 * data getting swapped to that swapfile suddenly gets overwritten by
2760 * the original zero's written out previously to the journal and
2761 * awaiting writeback in the kernel's buffer cache.
2763 * So, if we see any bmap calls here on a modified, data-journaled file,
2764 * take extra steps to flush any blocks which might be in the cache.
2766 static sector_t
ext4_bmap(struct address_space
*mapping
, sector_t block
)
2768 struct inode
*inode
= mapping
->host
;
2772 if (mapping_tagged(mapping
, PAGECACHE_TAG_DIRTY
) &&
2773 test_opt(inode
->i_sb
, DELALLOC
)) {
2775 * With delalloc we want to sync the file
2776 * so that we can make sure we allocate
2779 filemap_write_and_wait(mapping
);
2782 if (EXT4_I(inode
)->i_state
& EXT4_STATE_JDATA
) {
2784 * This is a REALLY heavyweight approach, but the use of
2785 * bmap on dirty files is expected to be extremely rare:
2786 * only if we run lilo or swapon on a freshly made file
2787 * do we expect this to happen.
2789 * (bmap requires CAP_SYS_RAWIO so this does not
2790 * represent an unprivileged user DOS attack --- we'd be
2791 * in trouble if mortal users could trigger this path at
2794 * NB. EXT4_STATE_JDATA is not set on files other than
2795 * regular files. If somebody wants to bmap a directory
2796 * or symlink and gets confused because the buffer
2797 * hasn't yet been flushed to disk, they deserve
2798 * everything they get.
2801 EXT4_I(inode
)->i_state
&= ~EXT4_STATE_JDATA
;
2802 journal
= EXT4_JOURNAL(inode
);
2803 jbd2_journal_lock_updates(journal
);
2804 err
= jbd2_journal_flush(journal
);
2805 jbd2_journal_unlock_updates(journal
);
2811 return generic_block_bmap(mapping
, block
, ext4_get_block
);
2814 static int bget_one(handle_t
*handle
, struct buffer_head
*bh
)
2820 static int bput_one(handle_t
*handle
, struct buffer_head
*bh
)
2827 * Note that we don't need to start a transaction unless we're journaling data
2828 * because we should have holes filled from ext4_page_mkwrite(). We even don't
2829 * need to file the inode to the transaction's list in ordered mode because if
2830 * we are writing back data added by write(), the inode is already there and if
2831 * we are writing back data modified via mmap(), noone guarantees in which
2832 * transaction the data will hit the disk. In case we are journaling data, we
2833 * cannot start transaction directly because transaction start ranks above page
2834 * lock so we have to do some magic.
2836 * In all journaling modes block_write_full_page() will start the I/O.
2840 * ext4_writepage() -> kmalloc() -> __alloc_pages() -> page_launder() ->
2845 * ext4_file_write() -> generic_file_write() -> __alloc_pages() -> ...
2847 * Same applies to ext4_get_block(). We will deadlock on various things like
2848 * lock_journal and i_data_sem
2850 * Setting PF_MEMALLOC here doesn't work - too many internal memory
2853 * 16May01: If we're reentered then journal_current_handle() will be
2854 * non-zero. We simply *return*.
2856 * 1 July 2001: @@@ FIXME:
2857 * In journalled data mode, a data buffer may be metadata against the
2858 * current transaction. But the same file is part of a shared mapping
2859 * and someone does a writepage() on it.
2861 * We will move the buffer onto the async_data list, but *after* it has
2862 * been dirtied. So there's a small window where we have dirty data on
2865 * Note that this only applies to the last partial page in the file. The
2866 * bit which block_write_full_page() uses prepare/commit for. (That's
2867 * broken code anyway: it's wrong for msync()).
2869 * It's a rare case: affects the final partial page, for journalled data
2870 * where the file is subject to bith write() and writepage() in the same
2871 * transction. To fix it we'll need a custom block_write_full_page().
2872 * We'll probably need that anyway for journalling writepage() output.
2874 * We don't honour synchronous mounts for writepage(). That would be
2875 * disastrous. Any write() or metadata operation will sync the fs for
2879 static int __ext4_normal_writepage(struct page
*page
,
2880 struct writeback_control
*wbc
)
2882 struct inode
*inode
= page
->mapping
->host
;
2884 if (test_opt(inode
->i_sb
, NOBH
))
2885 return nobh_writepage(page
,
2886 ext4_normal_get_block_write
, wbc
);
2888 return block_write_full_page(page
,
2889 ext4_normal_get_block_write
,
2893 static int ext4_normal_writepage(struct page
*page
,
2894 struct writeback_control
*wbc
)
2896 struct inode
*inode
= page
->mapping
->host
;
2897 loff_t size
= i_size_read(inode
);
2900 J_ASSERT(PageLocked(page
));
2901 if (page
->index
== size
>> PAGE_CACHE_SHIFT
)
2902 len
= size
& ~PAGE_CACHE_MASK
;
2904 len
= PAGE_CACHE_SIZE
;
2906 if (page_has_buffers(page
)) {
2907 /* if page has buffers it should all be mapped
2908 * and allocated. If there are not buffers attached
2909 * to the page we know the page is dirty but it lost
2910 * buffers. That means that at some moment in time
2911 * after write_begin() / write_end() has been called
2912 * all buffers have been clean and thus they must have been
2913 * written at least once. So they are all mapped and we can
2914 * happily proceed with mapping them and writing the page.
2916 BUG_ON(walk_page_buffers(NULL
, page_buffers(page
), 0, len
, NULL
,
2917 ext4_bh_unmapped_or_delay
));
2920 if (!ext4_journal_current_handle())
2921 return __ext4_normal_writepage(page
, wbc
);
2923 redirty_page_for_writepage(wbc
, page
);
2928 static int __ext4_journalled_writepage(struct page
*page
,
2929 struct writeback_control
*wbc
)
2931 struct address_space
*mapping
= page
->mapping
;
2932 struct inode
*inode
= mapping
->host
;
2933 struct buffer_head
*page_bufs
;
2934 handle_t
*handle
= NULL
;
2938 ret
= block_prepare_write(page
, 0, PAGE_CACHE_SIZE
,
2939 ext4_normal_get_block_write
);
2943 page_bufs
= page_buffers(page
);
2944 walk_page_buffers(handle
, page_bufs
, 0, PAGE_CACHE_SIZE
, NULL
,
2946 /* As soon as we unlock the page, it can go away, but we have
2947 * references to buffers so we are safe */
2950 handle
= ext4_journal_start(inode
, ext4_writepage_trans_blocks(inode
));
2951 if (IS_ERR(handle
)) {
2952 ret
= PTR_ERR(handle
);
2956 ret
= walk_page_buffers(handle
, page_bufs
, 0,
2957 PAGE_CACHE_SIZE
, NULL
, do_journal_get_write_access
);
2959 err
= walk_page_buffers(handle
, page_bufs
, 0,
2960 PAGE_CACHE_SIZE
, NULL
, write_end_fn
);
2963 err
= ext4_journal_stop(handle
);
2967 walk_page_buffers(handle
, page_bufs
, 0,
2968 PAGE_CACHE_SIZE
, NULL
, bput_one
);
2969 EXT4_I(inode
)->i_state
|= EXT4_STATE_JDATA
;
2978 static int ext4_journalled_writepage(struct page
*page
,
2979 struct writeback_control
*wbc
)
2981 struct inode
*inode
= page
->mapping
->host
;
2982 loff_t size
= i_size_read(inode
);
2985 J_ASSERT(PageLocked(page
));
2986 if (page
->index
== size
>> PAGE_CACHE_SHIFT
)
2987 len
= size
& ~PAGE_CACHE_MASK
;
2989 len
= PAGE_CACHE_SIZE
;
2991 if (page_has_buffers(page
)) {
2992 /* if page has buffers it should all be mapped
2993 * and allocated. If there are not buffers attached
2994 * to the page we know the page is dirty but it lost
2995 * buffers. That means that at some moment in time
2996 * after write_begin() / write_end() has been called
2997 * all buffers have been clean and thus they must have been
2998 * written at least once. So they are all mapped and we can
2999 * happily proceed with mapping them and writing the page.
3001 BUG_ON(walk_page_buffers(NULL
, page_buffers(page
), 0, len
, NULL
,
3002 ext4_bh_unmapped_or_delay
));
3005 if (ext4_journal_current_handle())
3008 if (PageChecked(page
)) {
3010 * It's mmapped pagecache. Add buffers and journal it. There
3011 * doesn't seem much point in redirtying the page here.
3013 ClearPageChecked(page
);
3014 return __ext4_journalled_writepage(page
, wbc
);
3017 * It may be a page full of checkpoint-mode buffers. We don't
3018 * really know unless we go poke around in the buffer_heads.
3019 * But block_write_full_page will do the right thing.
3021 return block_write_full_page(page
,
3022 ext4_normal_get_block_write
,
3026 redirty_page_for_writepage(wbc
, page
);
3031 static int ext4_readpage(struct file
*file
, struct page
*page
)
3033 return mpage_readpage(page
, ext4_get_block
);
3037 ext4_readpages(struct file
*file
, struct address_space
*mapping
,
3038 struct list_head
*pages
, unsigned nr_pages
)
3040 return mpage_readpages(mapping
, pages
, nr_pages
, ext4_get_block
);
3043 static void ext4_invalidatepage(struct page
*page
, unsigned long offset
)
3045 journal_t
*journal
= EXT4_JOURNAL(page
->mapping
->host
);
3048 * If it's a full truncate we just forget about the pending dirtying
3051 ClearPageChecked(page
);
3053 jbd2_journal_invalidatepage(journal
, page
, offset
);
3056 static int ext4_releasepage(struct page
*page
, gfp_t wait
)
3058 journal_t
*journal
= EXT4_JOURNAL(page
->mapping
->host
);
3060 WARN_ON(PageChecked(page
));
3061 if (!page_has_buffers(page
))
3063 return jbd2_journal_try_to_free_buffers(journal
, page
, wait
);
3067 * If the O_DIRECT write will extend the file then add this inode to the
3068 * orphan list. So recovery will truncate it back to the original size
3069 * if the machine crashes during the write.
3071 * If the O_DIRECT write is intantiating holes inside i_size and the machine
3072 * crashes then stale disk data _may_ be exposed inside the file. But current
3073 * VFS code falls back into buffered path in that case so we are safe.
3075 static ssize_t
ext4_direct_IO(int rw
, struct kiocb
*iocb
,
3076 const struct iovec
*iov
, loff_t offset
,
3077 unsigned long nr_segs
)
3079 struct file
*file
= iocb
->ki_filp
;
3080 struct inode
*inode
= file
->f_mapping
->host
;
3081 struct ext4_inode_info
*ei
= EXT4_I(inode
);
3085 size_t count
= iov_length(iov
, nr_segs
);
3088 loff_t final_size
= offset
+ count
;
3090 if (final_size
> inode
->i_size
) {
3091 /* Credits for sb + inode write */
3092 handle
= ext4_journal_start(inode
, 2);
3093 if (IS_ERR(handle
)) {
3094 ret
= PTR_ERR(handle
);
3097 ret
= ext4_orphan_add(handle
, inode
);
3099 ext4_journal_stop(handle
);
3103 ei
->i_disksize
= inode
->i_size
;
3104 ext4_journal_stop(handle
);
3108 ret
= blockdev_direct_IO(rw
, iocb
, inode
, inode
->i_sb
->s_bdev
, iov
,
3110 ext4_get_block
, NULL
);
3115 /* Credits for sb + inode write */
3116 handle
= ext4_journal_start(inode
, 2);
3117 if (IS_ERR(handle
)) {
3118 /* This is really bad luck. We've written the data
3119 * but cannot extend i_size. Bail out and pretend
3120 * the write failed... */
3121 ret
= PTR_ERR(handle
);
3125 ext4_orphan_del(handle
, inode
);
3127 loff_t end
= offset
+ ret
;
3128 if (end
> inode
->i_size
) {
3129 ei
->i_disksize
= end
;
3130 i_size_write(inode
, end
);
3132 * We're going to return a positive `ret'
3133 * here due to non-zero-length I/O, so there's
3134 * no way of reporting error returns from
3135 * ext4_mark_inode_dirty() to userspace. So
3138 ext4_mark_inode_dirty(handle
, inode
);
3141 err
= ext4_journal_stop(handle
);
3150 * Pages can be marked dirty completely asynchronously from ext4's journalling
3151 * activity. By filemap_sync_pte(), try_to_unmap_one(), etc. We cannot do
3152 * much here because ->set_page_dirty is called under VFS locks. The page is
3153 * not necessarily locked.
3155 * We cannot just dirty the page and leave attached buffers clean, because the
3156 * buffers' dirty state is "definitive". We cannot just set the buffers dirty
3157 * or jbddirty because all the journalling code will explode.
3159 * So what we do is to mark the page "pending dirty" and next time writepage
3160 * is called, propagate that into the buffers appropriately.
3162 static int ext4_journalled_set_page_dirty(struct page
*page
)
3164 SetPageChecked(page
);
3165 return __set_page_dirty_nobuffers(page
);
3168 static const struct address_space_operations ext4_ordered_aops
= {
3169 .readpage
= ext4_readpage
,
3170 .readpages
= ext4_readpages
,
3171 .writepage
= ext4_normal_writepage
,
3172 .sync_page
= block_sync_page
,
3173 .write_begin
= ext4_write_begin
,
3174 .write_end
= ext4_ordered_write_end
,
3176 .invalidatepage
= ext4_invalidatepage
,
3177 .releasepage
= ext4_releasepage
,
3178 .direct_IO
= ext4_direct_IO
,
3179 .migratepage
= buffer_migrate_page
,
3180 .is_partially_uptodate
= block_is_partially_uptodate
,
3183 static const struct address_space_operations ext4_writeback_aops
= {
3184 .readpage
= ext4_readpage
,
3185 .readpages
= ext4_readpages
,
3186 .writepage
= ext4_normal_writepage
,
3187 .sync_page
= block_sync_page
,
3188 .write_begin
= ext4_write_begin
,
3189 .write_end
= ext4_writeback_write_end
,
3191 .invalidatepage
= ext4_invalidatepage
,
3192 .releasepage
= ext4_releasepage
,
3193 .direct_IO
= ext4_direct_IO
,
3194 .migratepage
= buffer_migrate_page
,
3195 .is_partially_uptodate
= block_is_partially_uptodate
,
3198 static const struct address_space_operations ext4_journalled_aops
= {
3199 .readpage
= ext4_readpage
,
3200 .readpages
= ext4_readpages
,
3201 .writepage
= ext4_journalled_writepage
,
3202 .sync_page
= block_sync_page
,
3203 .write_begin
= ext4_write_begin
,
3204 .write_end
= ext4_journalled_write_end
,
3205 .set_page_dirty
= ext4_journalled_set_page_dirty
,
3207 .invalidatepage
= ext4_invalidatepage
,
3208 .releasepage
= ext4_releasepage
,
3209 .is_partially_uptodate
= block_is_partially_uptodate
,
3212 static const struct address_space_operations ext4_da_aops
= {
3213 .readpage
= ext4_readpage
,
3214 .readpages
= ext4_readpages
,
3215 .writepage
= ext4_da_writepage
,
3216 .writepages
= ext4_da_writepages
,
3217 .sync_page
= block_sync_page
,
3218 .write_begin
= ext4_da_write_begin
,
3219 .write_end
= ext4_da_write_end
,
3221 .invalidatepage
= ext4_da_invalidatepage
,
3222 .releasepage
= ext4_releasepage
,
3223 .direct_IO
= ext4_direct_IO
,
3224 .migratepage
= buffer_migrate_page
,
3225 .is_partially_uptodate
= block_is_partially_uptodate
,
3228 void ext4_set_aops(struct inode
*inode
)
3230 if (ext4_should_order_data(inode
) &&
3231 test_opt(inode
->i_sb
, DELALLOC
))
3232 inode
->i_mapping
->a_ops
= &ext4_da_aops
;
3233 else if (ext4_should_order_data(inode
))
3234 inode
->i_mapping
->a_ops
= &ext4_ordered_aops
;
3235 else if (ext4_should_writeback_data(inode
) &&
3236 test_opt(inode
->i_sb
, DELALLOC
))
3237 inode
->i_mapping
->a_ops
= &ext4_da_aops
;
3238 else if (ext4_should_writeback_data(inode
))
3239 inode
->i_mapping
->a_ops
= &ext4_writeback_aops
;
3241 inode
->i_mapping
->a_ops
= &ext4_journalled_aops
;
3245 * ext4_block_truncate_page() zeroes out a mapping from file offset `from'
3246 * up to the end of the block which corresponds to `from'.
3247 * This required during truncate. We need to physically zero the tail end
3248 * of that block so it doesn't yield old data if the file is later grown.
3250 int ext4_block_truncate_page(handle_t
*handle
,
3251 struct address_space
*mapping
, loff_t from
)
3253 ext4_fsblk_t index
= from
>> PAGE_CACHE_SHIFT
;
3254 unsigned offset
= from
& (PAGE_CACHE_SIZE
-1);
3255 unsigned blocksize
, length
, pos
;
3257 struct inode
*inode
= mapping
->host
;
3258 struct buffer_head
*bh
;
3262 page
= grab_cache_page(mapping
, from
>> PAGE_CACHE_SHIFT
);
3266 blocksize
= inode
->i_sb
->s_blocksize
;
3267 length
= blocksize
- (offset
& (blocksize
- 1));
3268 iblock
= index
<< (PAGE_CACHE_SHIFT
- inode
->i_sb
->s_blocksize_bits
);
3271 * For "nobh" option, we can only work if we don't need to
3272 * read-in the page - otherwise we create buffers to do the IO.
3274 if (!page_has_buffers(page
) && test_opt(inode
->i_sb
, NOBH
) &&
3275 ext4_should_writeback_data(inode
) && PageUptodate(page
)) {
3276 zero_user(page
, offset
, length
);
3277 set_page_dirty(page
);
3281 if (!page_has_buffers(page
))
3282 create_empty_buffers(page
, blocksize
, 0);
3284 /* Find the buffer that contains "offset" */
3285 bh
= page_buffers(page
);
3287 while (offset
>= pos
) {
3288 bh
= bh
->b_this_page
;
3294 if (buffer_freed(bh
)) {
3295 BUFFER_TRACE(bh
, "freed: skip");
3299 if (!buffer_mapped(bh
)) {
3300 BUFFER_TRACE(bh
, "unmapped");
3301 ext4_get_block(inode
, iblock
, bh
, 0);
3302 /* unmapped? It's a hole - nothing to do */
3303 if (!buffer_mapped(bh
)) {
3304 BUFFER_TRACE(bh
, "still unmapped");
3309 /* Ok, it's mapped. Make sure it's up-to-date */
3310 if (PageUptodate(page
))
3311 set_buffer_uptodate(bh
);
3313 if (!buffer_uptodate(bh
)) {
3315 ll_rw_block(READ
, 1, &bh
);
3317 /* Uhhuh. Read error. Complain and punt. */
3318 if (!buffer_uptodate(bh
))
3322 if (ext4_should_journal_data(inode
)) {
3323 BUFFER_TRACE(bh
, "get write access");
3324 err
= ext4_journal_get_write_access(handle
, bh
);
3329 zero_user(page
, offset
, length
);
3331 BUFFER_TRACE(bh
, "zeroed end of block");
3334 if (ext4_should_journal_data(inode
)) {
3335 err
= ext4_journal_dirty_metadata(handle
, bh
);
3337 if (ext4_should_order_data(inode
))
3338 err
= ext4_jbd2_file_inode(handle
, inode
);
3339 mark_buffer_dirty(bh
);
3344 page_cache_release(page
);
3349 * Probably it should be a library function... search for first non-zero word
3350 * or memcmp with zero_page, whatever is better for particular architecture.
3353 static inline int all_zeroes(__le32
*p
, __le32
*q
)
3362 * ext4_find_shared - find the indirect blocks for partial truncation.
3363 * @inode: inode in question
3364 * @depth: depth of the affected branch
3365 * @offsets: offsets of pointers in that branch (see ext4_block_to_path)
3366 * @chain: place to store the pointers to partial indirect blocks
3367 * @top: place to the (detached) top of branch
3369 * This is a helper function used by ext4_truncate().
3371 * When we do truncate() we may have to clean the ends of several
3372 * indirect blocks but leave the blocks themselves alive. Block is
3373 * partially truncated if some data below the new i_size is refered
3374 * from it (and it is on the path to the first completely truncated
3375 * data block, indeed). We have to free the top of that path along
3376 * with everything to the right of the path. Since no allocation
3377 * past the truncation point is possible until ext4_truncate()
3378 * finishes, we may safely do the latter, but top of branch may
3379 * require special attention - pageout below the truncation point
3380 * might try to populate it.
3382 * We atomically detach the top of branch from the tree, store the
3383 * block number of its root in *@top, pointers to buffer_heads of
3384 * partially truncated blocks - in @chain[].bh and pointers to
3385 * their last elements that should not be removed - in
3386 * @chain[].p. Return value is the pointer to last filled element
3389 * The work left to caller to do the actual freeing of subtrees:
3390 * a) free the subtree starting from *@top
3391 * b) free the subtrees whose roots are stored in
3392 * (@chain[i].p+1 .. end of @chain[i].bh->b_data)
3393 * c) free the subtrees growing from the inode past the @chain[0].
3394 * (no partially truncated stuff there). */
3396 static Indirect
*ext4_find_shared(struct inode
*inode
, int depth
,
3397 ext4_lblk_t offsets
[4], Indirect chain
[4], __le32
*top
)
3399 Indirect
*partial
, *p
;
3403 /* Make k index the deepest non-null offest + 1 */
3404 for (k
= depth
; k
> 1 && !offsets
[k
-1]; k
--)
3406 partial
= ext4_get_branch(inode
, k
, offsets
, chain
, &err
);
3407 /* Writer: pointers */
3409 partial
= chain
+ k
-1;
3411 * If the branch acquired continuation since we've looked at it -
3412 * fine, it should all survive and (new) top doesn't belong to us.
3414 if (!partial
->key
&& *partial
->p
)
3417 for (p
= partial
; (p
> chain
) && all_zeroes((__le32
*) p
->bh
->b_data
, p
->p
); p
--)
3420 * OK, we've found the last block that must survive. The rest of our
3421 * branch should be detached before unlocking. However, if that rest
3422 * of branch is all ours and does not grow immediately from the inode
3423 * it's easier to cheat and just decrement partial->p.
3425 if (p
== chain
+ k
- 1 && p
> chain
) {
3429 /* Nope, don't do this in ext4. Must leave the tree intact */
3436 while (partial
> p
) {
3437 brelse(partial
->bh
);
3445 * Zero a number of block pointers in either an inode or an indirect block.
3446 * If we restart the transaction we must again get write access to the
3447 * indirect block for further modification.
3449 * We release `count' blocks on disk, but (last - first) may be greater
3450 * than `count' because there can be holes in there.
3452 static void ext4_clear_blocks(handle_t
*handle
, struct inode
*inode
,
3453 struct buffer_head
*bh
, ext4_fsblk_t block_to_free
,
3454 unsigned long count
, __le32
*first
, __le32
*last
)
3457 if (try_to_extend_transaction(handle
, inode
)) {
3459 BUFFER_TRACE(bh
, "call ext4_journal_dirty_metadata");
3460 ext4_journal_dirty_metadata(handle
, bh
);
3462 ext4_mark_inode_dirty(handle
, inode
);
3463 ext4_journal_test_restart(handle
, inode
);
3465 BUFFER_TRACE(bh
, "retaking write access");
3466 ext4_journal_get_write_access(handle
, bh
);
3471 * Any buffers which are on the journal will be in memory. We find
3472 * them on the hash table so jbd2_journal_revoke() will run jbd2_journal_forget()
3473 * on them. We've already detached each block from the file, so
3474 * bforget() in jbd2_journal_forget() should be safe.
3476 * AKPM: turn on bforget in jbd2_journal_forget()!!!
3478 for (p
= first
; p
< last
; p
++) {
3479 u32 nr
= le32_to_cpu(*p
);
3481 struct buffer_head
*tbh
;
3484 tbh
= sb_find_get_block(inode
->i_sb
, nr
);
3485 ext4_forget(handle
, 0, inode
, tbh
, nr
);
3489 ext4_free_blocks(handle
, inode
, block_to_free
, count
, 0);
3493 * ext4_free_data - free a list of data blocks
3494 * @handle: handle for this transaction
3495 * @inode: inode we are dealing with
3496 * @this_bh: indirect buffer_head which contains *@first and *@last
3497 * @first: array of block numbers
3498 * @last: points immediately past the end of array
3500 * We are freeing all blocks refered from that array (numbers are stored as
3501 * little-endian 32-bit) and updating @inode->i_blocks appropriately.
3503 * We accumulate contiguous runs of blocks to free. Conveniently, if these
3504 * blocks are contiguous then releasing them at one time will only affect one
3505 * or two bitmap blocks (+ group descriptor(s) and superblock) and we won't
3506 * actually use a lot of journal space.
3508 * @this_bh will be %NULL if @first and @last point into the inode's direct
3511 static void ext4_free_data(handle_t
*handle
, struct inode
*inode
,
3512 struct buffer_head
*this_bh
,
3513 __le32
*first
, __le32
*last
)
3515 ext4_fsblk_t block_to_free
= 0; /* Starting block # of a run */
3516 unsigned long count
= 0; /* Number of blocks in the run */
3517 __le32
*block_to_free_p
= NULL
; /* Pointer into inode/ind
3520 ext4_fsblk_t nr
; /* Current block # */
3521 __le32
*p
; /* Pointer into inode/ind
3522 for current block */
3525 if (this_bh
) { /* For indirect block */
3526 BUFFER_TRACE(this_bh
, "get_write_access");
3527 err
= ext4_journal_get_write_access(handle
, this_bh
);
3528 /* Important: if we can't update the indirect pointers
3529 * to the blocks, we can't free them. */
3534 for (p
= first
; p
< last
; p
++) {
3535 nr
= le32_to_cpu(*p
);
3537 /* accumulate blocks to free if they're contiguous */
3540 block_to_free_p
= p
;
3542 } else if (nr
== block_to_free
+ count
) {
3545 ext4_clear_blocks(handle
, inode
, this_bh
,
3547 count
, block_to_free_p
, p
);
3549 block_to_free_p
= p
;
3556 ext4_clear_blocks(handle
, inode
, this_bh
, block_to_free
,
3557 count
, block_to_free_p
, p
);
3560 BUFFER_TRACE(this_bh
, "call ext4_journal_dirty_metadata");
3563 * The buffer head should have an attached journal head at this
3564 * point. However, if the data is corrupted and an indirect
3565 * block pointed to itself, it would have been detached when
3566 * the block was cleared. Check for this instead of OOPSing.
3569 ext4_journal_dirty_metadata(handle
, this_bh
);
3571 ext4_error(inode
->i_sb
, __func__
,
3572 "circular indirect block detected, "
3573 "inode=%lu, block=%llu",
3575 (unsigned long long) this_bh
->b_blocknr
);
3580 * ext4_free_branches - free an array of branches
3581 * @handle: JBD handle for this transaction
3582 * @inode: inode we are dealing with
3583 * @parent_bh: the buffer_head which contains *@first and *@last
3584 * @first: array of block numbers
3585 * @last: pointer immediately past the end of array
3586 * @depth: depth of the branches to free
3588 * We are freeing all blocks refered from these branches (numbers are
3589 * stored as little-endian 32-bit) and updating @inode->i_blocks
3592 static void ext4_free_branches(handle_t
*handle
, struct inode
*inode
,
3593 struct buffer_head
*parent_bh
,
3594 __le32
*first
, __le32
*last
, int depth
)
3599 if (is_handle_aborted(handle
))
3603 struct buffer_head
*bh
;
3604 int addr_per_block
= EXT4_ADDR_PER_BLOCK(inode
->i_sb
);
3606 while (--p
>= first
) {
3607 nr
= le32_to_cpu(*p
);
3609 continue; /* A hole */
3611 /* Go read the buffer for the next level down */
3612 bh
= sb_bread(inode
->i_sb
, nr
);
3615 * A read failure? Report error and clear slot
3619 ext4_error(inode
->i_sb
, "ext4_free_branches",
3620 "Read failure, inode=%lu, block=%llu",
3625 /* This zaps the entire block. Bottom up. */
3626 BUFFER_TRACE(bh
, "free child branches");
3627 ext4_free_branches(handle
, inode
, bh
,
3628 (__le32
*) bh
->b_data
,
3629 (__le32
*) bh
->b_data
+ addr_per_block
,
3633 * We've probably journalled the indirect block several
3634 * times during the truncate. But it's no longer
3635 * needed and we now drop it from the transaction via
3636 * jbd2_journal_revoke().
3638 * That's easy if it's exclusively part of this
3639 * transaction. But if it's part of the committing
3640 * transaction then jbd2_journal_forget() will simply
3641 * brelse() it. That means that if the underlying
3642 * block is reallocated in ext4_get_block(),
3643 * unmap_underlying_metadata() will find this block
3644 * and will try to get rid of it. damn, damn.
3646 * If this block has already been committed to the
3647 * journal, a revoke record will be written. And
3648 * revoke records must be emitted *before* clearing
3649 * this block's bit in the bitmaps.
3651 ext4_forget(handle
, 1, inode
, bh
, bh
->b_blocknr
);
3654 * Everything below this this pointer has been
3655 * released. Now let this top-of-subtree go.
3657 * We want the freeing of this indirect block to be
3658 * atomic in the journal with the updating of the
3659 * bitmap block which owns it. So make some room in
3662 * We zero the parent pointer *after* freeing its
3663 * pointee in the bitmaps, so if extend_transaction()
3664 * for some reason fails to put the bitmap changes and
3665 * the release into the same transaction, recovery
3666 * will merely complain about releasing a free block,
3667 * rather than leaking blocks.
3669 if (is_handle_aborted(handle
))
3671 if (try_to_extend_transaction(handle
, inode
)) {
3672 ext4_mark_inode_dirty(handle
, inode
);
3673 ext4_journal_test_restart(handle
, inode
);
3676 ext4_free_blocks(handle
, inode
, nr
, 1, 1);
3680 * The block which we have just freed is
3681 * pointed to by an indirect block: journal it
3683 BUFFER_TRACE(parent_bh
, "get_write_access");
3684 if (!ext4_journal_get_write_access(handle
,
3687 BUFFER_TRACE(parent_bh
,
3688 "call ext4_journal_dirty_metadata");
3689 ext4_journal_dirty_metadata(handle
,
3695 /* We have reached the bottom of the tree. */
3696 BUFFER_TRACE(parent_bh
, "free data blocks");
3697 ext4_free_data(handle
, inode
, parent_bh
, first
, last
);
3701 int ext4_can_truncate(struct inode
*inode
)
3703 if (IS_APPEND(inode
) || IS_IMMUTABLE(inode
))
3705 if (S_ISREG(inode
->i_mode
))
3707 if (S_ISDIR(inode
->i_mode
))
3709 if (S_ISLNK(inode
->i_mode
))
3710 return !ext4_inode_is_fast_symlink(inode
);
3717 * We block out ext4_get_block() block instantiations across the entire
3718 * transaction, and VFS/VM ensures that ext4_truncate() cannot run
3719 * simultaneously on behalf of the same inode.
3721 * As we work through the truncate and commmit bits of it to the journal there
3722 * is one core, guiding principle: the file's tree must always be consistent on
3723 * disk. We must be able to restart the truncate after a crash.
3725 * The file's tree may be transiently inconsistent in memory (although it
3726 * probably isn't), but whenever we close off and commit a journal transaction,
3727 * the contents of (the filesystem + the journal) must be consistent and
3728 * restartable. It's pretty simple, really: bottom up, right to left (although
3729 * left-to-right works OK too).
3731 * Note that at recovery time, journal replay occurs *before* the restart of
3732 * truncate against the orphan inode list.
3734 * The committed inode has the new, desired i_size (which is the same as
3735 * i_disksize in this case). After a crash, ext4_orphan_cleanup() will see
3736 * that this inode's truncate did not complete and it will again call
3737 * ext4_truncate() to have another go. So there will be instantiated blocks
3738 * to the right of the truncation point in a crashed ext4 filesystem. But
3739 * that's fine - as long as they are linked from the inode, the post-crash
3740 * ext4_truncate() run will find them and release them.
3742 void ext4_truncate(struct inode
*inode
)
3745 struct ext4_inode_info
*ei
= EXT4_I(inode
);
3746 __le32
*i_data
= ei
->i_data
;
3747 int addr_per_block
= EXT4_ADDR_PER_BLOCK(inode
->i_sb
);
3748 struct address_space
*mapping
= inode
->i_mapping
;
3749 ext4_lblk_t offsets
[4];
3754 ext4_lblk_t last_block
;
3755 unsigned blocksize
= inode
->i_sb
->s_blocksize
;
3757 if (!ext4_can_truncate(inode
))
3760 if (EXT4_I(inode
)->i_flags
& EXT4_EXTENTS_FL
) {
3761 ext4_ext_truncate(inode
);
3765 handle
= start_transaction(inode
);
3767 return; /* AKPM: return what? */
3769 last_block
= (inode
->i_size
+ blocksize
-1)
3770 >> EXT4_BLOCK_SIZE_BITS(inode
->i_sb
);
3772 if (inode
->i_size
& (blocksize
- 1))
3773 if (ext4_block_truncate_page(handle
, mapping
, inode
->i_size
))
3776 n
= ext4_block_to_path(inode
, last_block
, offsets
, NULL
);
3778 goto out_stop
; /* error */
3781 * OK. This truncate is going to happen. We add the inode to the
3782 * orphan list, so that if this truncate spans multiple transactions,
3783 * and we crash, we will resume the truncate when the filesystem
3784 * recovers. It also marks the inode dirty, to catch the new size.
3786 * Implication: the file must always be in a sane, consistent
3787 * truncatable state while each transaction commits.
3789 if (ext4_orphan_add(handle
, inode
))
3793 * From here we block out all ext4_get_block() callers who want to
3794 * modify the block allocation tree.
3796 down_write(&ei
->i_data_sem
);
3798 ext4_discard_preallocations(inode
);
3801 * The orphan list entry will now protect us from any crash which
3802 * occurs before the truncate completes, so it is now safe to propagate
3803 * the new, shorter inode size (held for now in i_size) into the
3804 * on-disk inode. We do this via i_disksize, which is the value which
3805 * ext4 *really* writes onto the disk inode.
3807 ei
->i_disksize
= inode
->i_size
;
3809 if (n
== 1) { /* direct blocks */
3810 ext4_free_data(handle
, inode
, NULL
, i_data
+offsets
[0],
3811 i_data
+ EXT4_NDIR_BLOCKS
);
3815 partial
= ext4_find_shared(inode
, n
, offsets
, chain
, &nr
);
3816 /* Kill the top of shared branch (not detached) */
3818 if (partial
== chain
) {
3819 /* Shared branch grows from the inode */
3820 ext4_free_branches(handle
, inode
, NULL
,
3821 &nr
, &nr
+1, (chain
+n
-1) - partial
);
3824 * We mark the inode dirty prior to restart,
3825 * and prior to stop. No need for it here.
3828 /* Shared branch grows from an indirect block */
3829 BUFFER_TRACE(partial
->bh
, "get_write_access");
3830 ext4_free_branches(handle
, inode
, partial
->bh
,
3832 partial
->p
+1, (chain
+n
-1) - partial
);
3835 /* Clear the ends of indirect blocks on the shared branch */
3836 while (partial
> chain
) {
3837 ext4_free_branches(handle
, inode
, partial
->bh
, partial
->p
+ 1,
3838 (__le32
*)partial
->bh
->b_data
+addr_per_block
,
3839 (chain
+n
-1) - partial
);
3840 BUFFER_TRACE(partial
->bh
, "call brelse");
3841 brelse (partial
->bh
);
3845 /* Kill the remaining (whole) subtrees */
3846 switch (offsets
[0]) {
3848 nr
= i_data
[EXT4_IND_BLOCK
];
3850 ext4_free_branches(handle
, inode
, NULL
, &nr
, &nr
+1, 1);
3851 i_data
[EXT4_IND_BLOCK
] = 0;
3853 case EXT4_IND_BLOCK
:
3854 nr
= i_data
[EXT4_DIND_BLOCK
];
3856 ext4_free_branches(handle
, inode
, NULL
, &nr
, &nr
+1, 2);
3857 i_data
[EXT4_DIND_BLOCK
] = 0;
3859 case EXT4_DIND_BLOCK
:
3860 nr
= i_data
[EXT4_TIND_BLOCK
];
3862 ext4_free_branches(handle
, inode
, NULL
, &nr
, &nr
+1, 3);
3863 i_data
[EXT4_TIND_BLOCK
] = 0;
3865 case EXT4_TIND_BLOCK
:
3869 up_write(&ei
->i_data_sem
);
3870 inode
->i_mtime
= inode
->i_ctime
= ext4_current_time(inode
);
3871 ext4_mark_inode_dirty(handle
, inode
);
3874 * In a multi-transaction truncate, we only make the final transaction
3881 * If this was a simple ftruncate(), and the file will remain alive
3882 * then we need to clear up the orphan record which we created above.
3883 * However, if this was a real unlink then we were called by
3884 * ext4_delete_inode(), and we allow that function to clean up the
3885 * orphan info for us.
3888 ext4_orphan_del(handle
, inode
);
3890 ext4_journal_stop(handle
);
3894 * ext4_get_inode_loc returns with an extra refcount against the inode's
3895 * underlying buffer_head on success. If 'in_mem' is true, we have all
3896 * data in memory that is needed to recreate the on-disk version of this
3899 static int __ext4_get_inode_loc(struct inode
*inode
,
3900 struct ext4_iloc
*iloc
, int in_mem
)
3902 struct ext4_group_desc
*gdp
;
3903 struct buffer_head
*bh
;
3904 struct super_block
*sb
= inode
->i_sb
;
3906 int inodes_per_block
, inode_offset
;
3909 if (!ext4_valid_inum(sb
, inode
->i_ino
))
3912 iloc
->block_group
= (inode
->i_ino
- 1) / EXT4_INODES_PER_GROUP(sb
);
3913 gdp
= ext4_get_group_desc(sb
, iloc
->block_group
, NULL
);
3918 * Figure out the offset within the block group inode table
3920 inodes_per_block
= (EXT4_BLOCK_SIZE(sb
) / EXT4_INODE_SIZE(sb
));
3921 inode_offset
= ((inode
->i_ino
- 1) %
3922 EXT4_INODES_PER_GROUP(sb
));
3923 block
= ext4_inode_table(sb
, gdp
) + (inode_offset
/ inodes_per_block
);
3924 iloc
->offset
= (inode_offset
% inodes_per_block
) * EXT4_INODE_SIZE(sb
);
3926 bh
= sb_getblk(sb
, block
);
3928 ext4_error(sb
, "ext4_get_inode_loc", "unable to read "
3929 "inode block - inode=%lu, block=%llu",
3930 inode
->i_ino
, block
);
3933 if (!buffer_uptodate(bh
)) {
3937 * If the buffer has the write error flag, we have failed
3938 * to write out another inode in the same block. In this
3939 * case, we don't have to read the block because we may
3940 * read the old inode data successfully.
3942 if (buffer_write_io_error(bh
) && !buffer_uptodate(bh
))
3943 set_buffer_uptodate(bh
);
3945 if (buffer_uptodate(bh
)) {
3946 /* someone brought it uptodate while we waited */
3952 * If we have all information of the inode in memory and this
3953 * is the only valid inode in the block, we need not read the
3957 struct buffer_head
*bitmap_bh
;
3960 start
= inode_offset
& ~(inodes_per_block
- 1);
3962 /* Is the inode bitmap in cache? */
3963 bitmap_bh
= sb_getblk(sb
, ext4_inode_bitmap(sb
, gdp
));
3968 * If the inode bitmap isn't in cache then the
3969 * optimisation may end up performing two reads instead
3970 * of one, so skip it.
3972 if (!buffer_uptodate(bitmap_bh
)) {
3976 for (i
= start
; i
< start
+ inodes_per_block
; i
++) {
3977 if (i
== inode_offset
)
3979 if (ext4_test_bit(i
, bitmap_bh
->b_data
))
3983 if (i
== start
+ inodes_per_block
) {
3984 /* all other inodes are free, so skip I/O */
3985 memset(bh
->b_data
, 0, bh
->b_size
);
3986 set_buffer_uptodate(bh
);
3994 * If we need to do any I/O, try to pre-readahead extra
3995 * blocks from the inode table.
3997 if (EXT4_SB(sb
)->s_inode_readahead_blks
) {
3998 ext4_fsblk_t b
, end
, table
;
4001 table
= ext4_inode_table(sb
, gdp
);
4002 /* Make sure s_inode_readahead_blks is a power of 2 */
4003 while (EXT4_SB(sb
)->s_inode_readahead_blks
&
4004 (EXT4_SB(sb
)->s_inode_readahead_blks
-1))
4005 EXT4_SB(sb
)->s_inode_readahead_blks
=
4006 (EXT4_SB(sb
)->s_inode_readahead_blks
&
4007 (EXT4_SB(sb
)->s_inode_readahead_blks
-1));
4008 b
= block
& ~(EXT4_SB(sb
)->s_inode_readahead_blks
-1);
4011 end
= b
+ EXT4_SB(sb
)->s_inode_readahead_blks
;
4012 num
= EXT4_INODES_PER_GROUP(sb
);
4013 if (EXT4_HAS_RO_COMPAT_FEATURE(sb
,
4014 EXT4_FEATURE_RO_COMPAT_GDT_CSUM
))
4015 num
-= le16_to_cpu(gdp
->bg_itable_unused
);
4016 table
+= num
/ inodes_per_block
;
4020 sb_breadahead(sb
, b
++);
4024 * There are other valid inodes in the buffer, this inode
4025 * has in-inode xattrs, or we don't have this inode in memory.
4026 * Read the block from disk.
4029 bh
->b_end_io
= end_buffer_read_sync
;
4030 submit_bh(READ_META
, bh
);
4032 if (!buffer_uptodate(bh
)) {
4033 ext4_error(sb
, __func__
,
4034 "unable to read inode block - inode=%lu, "
4035 "block=%llu", inode
->i_ino
, block
);
4045 int ext4_get_inode_loc(struct inode
*inode
, struct ext4_iloc
*iloc
)
4047 /* We have all inode data except xattrs in memory here. */
4048 return __ext4_get_inode_loc(inode
, iloc
,
4049 !(EXT4_I(inode
)->i_state
& EXT4_STATE_XATTR
));
4052 void ext4_set_inode_flags(struct inode
*inode
)
4054 unsigned int flags
= EXT4_I(inode
)->i_flags
;
4056 inode
->i_flags
&= ~(S_SYNC
|S_APPEND
|S_IMMUTABLE
|S_NOATIME
|S_DIRSYNC
);
4057 if (flags
& EXT4_SYNC_FL
)
4058 inode
->i_flags
|= S_SYNC
;
4059 if (flags
& EXT4_APPEND_FL
)
4060 inode
->i_flags
|= S_APPEND
;
4061 if (flags
& EXT4_IMMUTABLE_FL
)
4062 inode
->i_flags
|= S_IMMUTABLE
;
4063 if (flags
& EXT4_NOATIME_FL
)
4064 inode
->i_flags
|= S_NOATIME
;
4065 if (flags
& EXT4_DIRSYNC_FL
)
4066 inode
->i_flags
|= S_DIRSYNC
;
4069 /* Propagate flags from i_flags to EXT4_I(inode)->i_flags */
4070 void ext4_get_inode_flags(struct ext4_inode_info
*ei
)
4072 unsigned int flags
= ei
->vfs_inode
.i_flags
;
4074 ei
->i_flags
&= ~(EXT4_SYNC_FL
|EXT4_APPEND_FL
|
4075 EXT4_IMMUTABLE_FL
|EXT4_NOATIME_FL
|EXT4_DIRSYNC_FL
);
4077 ei
->i_flags
|= EXT4_SYNC_FL
;
4078 if (flags
& S_APPEND
)
4079 ei
->i_flags
|= EXT4_APPEND_FL
;
4080 if (flags
& S_IMMUTABLE
)
4081 ei
->i_flags
|= EXT4_IMMUTABLE_FL
;
4082 if (flags
& S_NOATIME
)
4083 ei
->i_flags
|= EXT4_NOATIME_FL
;
4084 if (flags
& S_DIRSYNC
)
4085 ei
->i_flags
|= EXT4_DIRSYNC_FL
;
4087 static blkcnt_t
ext4_inode_blocks(struct ext4_inode
*raw_inode
,
4088 struct ext4_inode_info
*ei
)
4091 struct inode
*inode
= &(ei
->vfs_inode
);
4092 struct super_block
*sb
= inode
->i_sb
;
4094 if (EXT4_HAS_RO_COMPAT_FEATURE(sb
,
4095 EXT4_FEATURE_RO_COMPAT_HUGE_FILE
)) {
4096 /* we are using combined 48 bit field */
4097 i_blocks
= ((u64
)le16_to_cpu(raw_inode
->i_blocks_high
)) << 32 |
4098 le32_to_cpu(raw_inode
->i_blocks_lo
);
4099 if (ei
->i_flags
& EXT4_HUGE_FILE_FL
) {
4100 /* i_blocks represent file system block size */
4101 return i_blocks
<< (inode
->i_blkbits
- 9);
4106 return le32_to_cpu(raw_inode
->i_blocks_lo
);
4110 struct inode
*ext4_iget(struct super_block
*sb
, unsigned long ino
)
4112 struct ext4_iloc iloc
;
4113 struct ext4_inode
*raw_inode
;
4114 struct ext4_inode_info
*ei
;
4115 struct buffer_head
*bh
;
4116 struct inode
*inode
;
4120 inode
= iget_locked(sb
, ino
);
4122 return ERR_PTR(-ENOMEM
);
4123 if (!(inode
->i_state
& I_NEW
))
4127 #ifdef CONFIG_EXT4_FS_POSIX_ACL
4128 ei
->i_acl
= EXT4_ACL_NOT_CACHED
;
4129 ei
->i_default_acl
= EXT4_ACL_NOT_CACHED
;
4132 ret
= __ext4_get_inode_loc(inode
, &iloc
, 0);
4136 raw_inode
= ext4_raw_inode(&iloc
);
4137 inode
->i_mode
= le16_to_cpu(raw_inode
->i_mode
);
4138 inode
->i_uid
= (uid_t
)le16_to_cpu(raw_inode
->i_uid_low
);
4139 inode
->i_gid
= (gid_t
)le16_to_cpu(raw_inode
->i_gid_low
);
4140 if (!(test_opt(inode
->i_sb
, NO_UID32
))) {
4141 inode
->i_uid
|= le16_to_cpu(raw_inode
->i_uid_high
) << 16;
4142 inode
->i_gid
|= le16_to_cpu(raw_inode
->i_gid_high
) << 16;
4144 inode
->i_nlink
= le16_to_cpu(raw_inode
->i_links_count
);
4147 ei
->i_dir_start_lookup
= 0;
4148 ei
->i_dtime
= le32_to_cpu(raw_inode
->i_dtime
);
4149 /* We now have enough fields to check if the inode was active or not.
4150 * This is needed because nfsd might try to access dead inodes
4151 * the test is that same one that e2fsck uses
4152 * NeilBrown 1999oct15
4154 if (inode
->i_nlink
== 0) {
4155 if (inode
->i_mode
== 0 ||
4156 !(EXT4_SB(inode
->i_sb
)->s_mount_state
& EXT4_ORPHAN_FS
)) {
4157 /* this inode is deleted */
4162 /* The only unlinked inodes we let through here have
4163 * valid i_mode and are being read by the orphan
4164 * recovery code: that's fine, we're about to complete
4165 * the process of deleting those. */
4167 ei
->i_flags
= le32_to_cpu(raw_inode
->i_flags
);
4168 inode
->i_blocks
= ext4_inode_blocks(raw_inode
, ei
);
4169 ei
->i_file_acl
= le32_to_cpu(raw_inode
->i_file_acl_lo
);
4170 if (EXT4_SB(inode
->i_sb
)->s_es
->s_creator_os
!=
4171 cpu_to_le32(EXT4_OS_HURD
)) {
4173 ((__u64
)le16_to_cpu(raw_inode
->i_file_acl_high
)) << 32;
4175 inode
->i_size
= ext4_isize(raw_inode
);
4176 ei
->i_disksize
= inode
->i_size
;
4177 inode
->i_generation
= le32_to_cpu(raw_inode
->i_generation
);
4178 ei
->i_block_group
= iloc
.block_group
;
4180 * NOTE! The in-memory inode i_data array is in little-endian order
4181 * even on big-endian machines: we do NOT byteswap the block numbers!
4183 for (block
= 0; block
< EXT4_N_BLOCKS
; block
++)
4184 ei
->i_data
[block
] = raw_inode
->i_block
[block
];
4185 INIT_LIST_HEAD(&ei
->i_orphan
);
4187 if (EXT4_INODE_SIZE(inode
->i_sb
) > EXT4_GOOD_OLD_INODE_SIZE
) {
4188 ei
->i_extra_isize
= le16_to_cpu(raw_inode
->i_extra_isize
);
4189 if (EXT4_GOOD_OLD_INODE_SIZE
+ ei
->i_extra_isize
>
4190 EXT4_INODE_SIZE(inode
->i_sb
)) {
4195 if (ei
->i_extra_isize
== 0) {
4196 /* The extra space is currently unused. Use it. */
4197 ei
->i_extra_isize
= sizeof(struct ext4_inode
) -
4198 EXT4_GOOD_OLD_INODE_SIZE
;
4200 __le32
*magic
= (void *)raw_inode
+
4201 EXT4_GOOD_OLD_INODE_SIZE
+
4203 if (*magic
== cpu_to_le32(EXT4_XATTR_MAGIC
))
4204 ei
->i_state
|= EXT4_STATE_XATTR
;
4207 ei
->i_extra_isize
= 0;
4209 EXT4_INODE_GET_XTIME(i_ctime
, inode
, raw_inode
);
4210 EXT4_INODE_GET_XTIME(i_mtime
, inode
, raw_inode
);
4211 EXT4_INODE_GET_XTIME(i_atime
, inode
, raw_inode
);
4212 EXT4_EINODE_GET_XTIME(i_crtime
, ei
, raw_inode
);
4214 inode
->i_version
= le32_to_cpu(raw_inode
->i_disk_version
);
4215 if (EXT4_INODE_SIZE(inode
->i_sb
) > EXT4_GOOD_OLD_INODE_SIZE
) {
4216 if (EXT4_FITS_IN_INODE(raw_inode
, ei
, i_version_hi
))
4218 (__u64
)(le32_to_cpu(raw_inode
->i_version_hi
)) << 32;
4221 if (S_ISREG(inode
->i_mode
)) {
4222 inode
->i_op
= &ext4_file_inode_operations
;
4223 inode
->i_fop
= &ext4_file_operations
;
4224 ext4_set_aops(inode
);
4225 } else if (S_ISDIR(inode
->i_mode
)) {
4226 inode
->i_op
= &ext4_dir_inode_operations
;
4227 inode
->i_fop
= &ext4_dir_operations
;
4228 } else if (S_ISLNK(inode
->i_mode
)) {
4229 if (ext4_inode_is_fast_symlink(inode
))
4230 inode
->i_op
= &ext4_fast_symlink_inode_operations
;
4232 inode
->i_op
= &ext4_symlink_inode_operations
;
4233 ext4_set_aops(inode
);
4236 inode
->i_op
= &ext4_special_inode_operations
;
4237 if (raw_inode
->i_block
[0])
4238 init_special_inode(inode
, inode
->i_mode
,
4239 old_decode_dev(le32_to_cpu(raw_inode
->i_block
[0])));
4241 init_special_inode(inode
, inode
->i_mode
,
4242 new_decode_dev(le32_to_cpu(raw_inode
->i_block
[1])));
4245 ext4_set_inode_flags(inode
);
4246 unlock_new_inode(inode
);
4251 return ERR_PTR(ret
);
4254 static int ext4_inode_blocks_set(handle_t
*handle
,
4255 struct ext4_inode
*raw_inode
,
4256 struct ext4_inode_info
*ei
)
4258 struct inode
*inode
= &(ei
->vfs_inode
);
4259 u64 i_blocks
= inode
->i_blocks
;
4260 struct super_block
*sb
= inode
->i_sb
;
4262 if (i_blocks
<= ~0U) {
4264 * i_blocks can be represnted in a 32 bit variable
4265 * as multiple of 512 bytes
4267 raw_inode
->i_blocks_lo
= cpu_to_le32(i_blocks
);
4268 raw_inode
->i_blocks_high
= 0;
4269 ei
->i_flags
&= ~EXT4_HUGE_FILE_FL
;
4272 if (!EXT4_HAS_RO_COMPAT_FEATURE(sb
, EXT4_FEATURE_RO_COMPAT_HUGE_FILE
))
4275 if (i_blocks
<= 0xffffffffffffULL
) {
4277 * i_blocks can be represented in a 48 bit variable
4278 * as multiple of 512 bytes
4280 raw_inode
->i_blocks_lo
= cpu_to_le32(i_blocks
);
4281 raw_inode
->i_blocks_high
= cpu_to_le16(i_blocks
>> 32);
4282 ei
->i_flags
&= ~EXT4_HUGE_FILE_FL
;
4284 ei
->i_flags
|= EXT4_HUGE_FILE_FL
;
4285 /* i_block is stored in file system block size */
4286 i_blocks
= i_blocks
>> (inode
->i_blkbits
- 9);
4287 raw_inode
->i_blocks_lo
= cpu_to_le32(i_blocks
);
4288 raw_inode
->i_blocks_high
= cpu_to_le16(i_blocks
>> 32);
4294 * Post the struct inode info into an on-disk inode location in the
4295 * buffer-cache. This gobbles the caller's reference to the
4296 * buffer_head in the inode location struct.
4298 * The caller must have write access to iloc->bh.
4300 static int ext4_do_update_inode(handle_t
*handle
,
4301 struct inode
*inode
,
4302 struct ext4_iloc
*iloc
)
4304 struct ext4_inode
*raw_inode
= ext4_raw_inode(iloc
);
4305 struct ext4_inode_info
*ei
= EXT4_I(inode
);
4306 struct buffer_head
*bh
= iloc
->bh
;
4307 int err
= 0, rc
, block
;
4309 /* For fields not not tracking in the in-memory inode,
4310 * initialise them to zero for new inodes. */
4311 if (ei
->i_state
& EXT4_STATE_NEW
)
4312 memset(raw_inode
, 0, EXT4_SB(inode
->i_sb
)->s_inode_size
);
4314 ext4_get_inode_flags(ei
);
4315 raw_inode
->i_mode
= cpu_to_le16(inode
->i_mode
);
4316 if (!(test_opt(inode
->i_sb
, NO_UID32
))) {
4317 raw_inode
->i_uid_low
= cpu_to_le16(low_16_bits(inode
->i_uid
));
4318 raw_inode
->i_gid_low
= cpu_to_le16(low_16_bits(inode
->i_gid
));
4320 * Fix up interoperability with old kernels. Otherwise, old inodes get
4321 * re-used with the upper 16 bits of the uid/gid intact
4324 raw_inode
->i_uid_high
=
4325 cpu_to_le16(high_16_bits(inode
->i_uid
));
4326 raw_inode
->i_gid_high
=
4327 cpu_to_le16(high_16_bits(inode
->i_gid
));
4329 raw_inode
->i_uid_high
= 0;
4330 raw_inode
->i_gid_high
= 0;
4333 raw_inode
->i_uid_low
=
4334 cpu_to_le16(fs_high2lowuid(inode
->i_uid
));
4335 raw_inode
->i_gid_low
=
4336 cpu_to_le16(fs_high2lowgid(inode
->i_gid
));
4337 raw_inode
->i_uid_high
= 0;
4338 raw_inode
->i_gid_high
= 0;
4340 raw_inode
->i_links_count
= cpu_to_le16(inode
->i_nlink
);
4342 EXT4_INODE_SET_XTIME(i_ctime
, inode
, raw_inode
);
4343 EXT4_INODE_SET_XTIME(i_mtime
, inode
, raw_inode
);
4344 EXT4_INODE_SET_XTIME(i_atime
, inode
, raw_inode
);
4345 EXT4_EINODE_SET_XTIME(i_crtime
, ei
, raw_inode
);
4347 if (ext4_inode_blocks_set(handle
, raw_inode
, ei
))
4349 raw_inode
->i_dtime
= cpu_to_le32(ei
->i_dtime
);
4350 /* clear the migrate flag in the raw_inode */
4351 raw_inode
->i_flags
= cpu_to_le32(ei
->i_flags
& ~EXT4_EXT_MIGRATE
);
4352 if (EXT4_SB(inode
->i_sb
)->s_es
->s_creator_os
!=
4353 cpu_to_le32(EXT4_OS_HURD
))
4354 raw_inode
->i_file_acl_high
=
4355 cpu_to_le16(ei
->i_file_acl
>> 32);
4356 raw_inode
->i_file_acl_lo
= cpu_to_le32(ei
->i_file_acl
);
4357 ext4_isize_set(raw_inode
, ei
->i_disksize
);
4358 if (ei
->i_disksize
> 0x7fffffffULL
) {
4359 struct super_block
*sb
= inode
->i_sb
;
4360 if (!EXT4_HAS_RO_COMPAT_FEATURE(sb
,
4361 EXT4_FEATURE_RO_COMPAT_LARGE_FILE
) ||
4362 EXT4_SB(sb
)->s_es
->s_rev_level
==
4363 cpu_to_le32(EXT4_GOOD_OLD_REV
)) {
4364 /* If this is the first large file
4365 * created, add a flag to the superblock.
4367 err
= ext4_journal_get_write_access(handle
,
4368 EXT4_SB(sb
)->s_sbh
);
4371 ext4_update_dynamic_rev(sb
);
4372 EXT4_SET_RO_COMPAT_FEATURE(sb
,
4373 EXT4_FEATURE_RO_COMPAT_LARGE_FILE
);
4376 err
= ext4_journal_dirty_metadata(handle
,
4377 EXT4_SB(sb
)->s_sbh
);
4380 raw_inode
->i_generation
= cpu_to_le32(inode
->i_generation
);
4381 if (S_ISCHR(inode
->i_mode
) || S_ISBLK(inode
->i_mode
)) {
4382 if (old_valid_dev(inode
->i_rdev
)) {
4383 raw_inode
->i_block
[0] =
4384 cpu_to_le32(old_encode_dev(inode
->i_rdev
));
4385 raw_inode
->i_block
[1] = 0;
4387 raw_inode
->i_block
[0] = 0;
4388 raw_inode
->i_block
[1] =
4389 cpu_to_le32(new_encode_dev(inode
->i_rdev
));
4390 raw_inode
->i_block
[2] = 0;
4392 } else for (block
= 0; block
< EXT4_N_BLOCKS
; block
++)
4393 raw_inode
->i_block
[block
] = ei
->i_data
[block
];
4395 raw_inode
->i_disk_version
= cpu_to_le32(inode
->i_version
);
4396 if (ei
->i_extra_isize
) {
4397 if (EXT4_FITS_IN_INODE(raw_inode
, ei
, i_version_hi
))
4398 raw_inode
->i_version_hi
=
4399 cpu_to_le32(inode
->i_version
>> 32);
4400 raw_inode
->i_extra_isize
= cpu_to_le16(ei
->i_extra_isize
);
4404 BUFFER_TRACE(bh
, "call ext4_journal_dirty_metadata");
4405 rc
= ext4_journal_dirty_metadata(handle
, bh
);
4408 ei
->i_state
&= ~EXT4_STATE_NEW
;
4412 ext4_std_error(inode
->i_sb
, err
);
4417 * ext4_write_inode()
4419 * We are called from a few places:
4421 * - Within generic_file_write() for O_SYNC files.
4422 * Here, there will be no transaction running. We wait for any running
4423 * trasnaction to commit.
4425 * - Within sys_sync(), kupdate and such.
4426 * We wait on commit, if tol to.
4428 * - Within prune_icache() (PF_MEMALLOC == true)
4429 * Here we simply return. We can't afford to block kswapd on the
4432 * In all cases it is actually safe for us to return without doing anything,
4433 * because the inode has been copied into a raw inode buffer in
4434 * ext4_mark_inode_dirty(). This is a correctness thing for O_SYNC and for
4437 * Note that we are absolutely dependent upon all inode dirtiers doing the
4438 * right thing: they *must* call mark_inode_dirty() after dirtying info in
4439 * which we are interested.
4441 * It would be a bug for them to not do this. The code:
4443 * mark_inode_dirty(inode)
4445 * inode->i_size = expr;
4447 * is in error because a kswapd-driven write_inode() could occur while
4448 * `stuff()' is running, and the new i_size will be lost. Plus the inode
4449 * will no longer be on the superblock's dirty inode list.
4451 int ext4_write_inode(struct inode
*inode
, int wait
)
4453 if (current
->flags
& PF_MEMALLOC
)
4456 if (ext4_journal_current_handle()) {
4457 jbd_debug(1, "called recursively, non-PF_MEMALLOC!\n");
4465 return ext4_force_commit(inode
->i_sb
);
4471 * Called from notify_change.
4473 * We want to trap VFS attempts to truncate the file as soon as
4474 * possible. In particular, we want to make sure that when the VFS
4475 * shrinks i_size, we put the inode on the orphan list and modify
4476 * i_disksize immediately, so that during the subsequent flushing of
4477 * dirty pages and freeing of disk blocks, we can guarantee that any
4478 * commit will leave the blocks being flushed in an unused state on
4479 * disk. (On recovery, the inode will get truncated and the blocks will
4480 * be freed, so we have a strong guarantee that no future commit will
4481 * leave these blocks visible to the user.)
4483 * Another thing we have to assure is that if we are in ordered mode
4484 * and inode is still attached to the committing transaction, we must
4485 * we start writeout of all the dirty pages which are being truncated.
4486 * This way we are sure that all the data written in the previous
4487 * transaction are already on disk (truncate waits for pages under
4490 * Called with inode->i_mutex down.
4492 int ext4_setattr(struct dentry
*dentry
, struct iattr
*attr
)
4494 struct inode
*inode
= dentry
->d_inode
;
4496 const unsigned int ia_valid
= attr
->ia_valid
;
4498 error
= inode_change_ok(inode
, attr
);
4502 if ((ia_valid
& ATTR_UID
&& attr
->ia_uid
!= inode
->i_uid
) ||
4503 (ia_valid
& ATTR_GID
&& attr
->ia_gid
!= inode
->i_gid
)) {
4506 /* (user+group)*(old+new) structure, inode write (sb,
4507 * inode block, ? - but truncate inode update has it) */
4508 handle
= ext4_journal_start(inode
, 2*(EXT4_QUOTA_INIT_BLOCKS(inode
->i_sb
)+
4509 EXT4_QUOTA_DEL_BLOCKS(inode
->i_sb
))+3);
4510 if (IS_ERR(handle
)) {
4511 error
= PTR_ERR(handle
);
4514 error
= DQUOT_TRANSFER(inode
, attr
) ? -EDQUOT
: 0;
4516 ext4_journal_stop(handle
);
4519 /* Update corresponding info in inode so that everything is in
4520 * one transaction */
4521 if (attr
->ia_valid
& ATTR_UID
)
4522 inode
->i_uid
= attr
->ia_uid
;
4523 if (attr
->ia_valid
& ATTR_GID
)
4524 inode
->i_gid
= attr
->ia_gid
;
4525 error
= ext4_mark_inode_dirty(handle
, inode
);
4526 ext4_journal_stop(handle
);
4529 if (attr
->ia_valid
& ATTR_SIZE
) {
4530 if (!(EXT4_I(inode
)->i_flags
& EXT4_EXTENTS_FL
)) {
4531 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
4533 if (attr
->ia_size
> sbi
->s_bitmap_maxbytes
) {
4540 if (S_ISREG(inode
->i_mode
) &&
4541 attr
->ia_valid
& ATTR_SIZE
&& attr
->ia_size
< inode
->i_size
) {
4544 handle
= ext4_journal_start(inode
, 3);
4545 if (IS_ERR(handle
)) {
4546 error
= PTR_ERR(handle
);
4550 error
= ext4_orphan_add(handle
, inode
);
4551 EXT4_I(inode
)->i_disksize
= attr
->ia_size
;
4552 rc
= ext4_mark_inode_dirty(handle
, inode
);
4555 ext4_journal_stop(handle
);
4557 if (ext4_should_order_data(inode
)) {
4558 error
= ext4_begin_ordered_truncate(inode
,
4561 /* Do as much error cleanup as possible */
4562 handle
= ext4_journal_start(inode
, 3);
4563 if (IS_ERR(handle
)) {
4564 ext4_orphan_del(NULL
, inode
);
4567 ext4_orphan_del(handle
, inode
);
4568 ext4_journal_stop(handle
);
4574 rc
= inode_setattr(inode
, attr
);
4576 /* If inode_setattr's call to ext4_truncate failed to get a
4577 * transaction handle at all, we need to clean up the in-core
4578 * orphan list manually. */
4580 ext4_orphan_del(NULL
, inode
);
4582 if (!rc
&& (ia_valid
& ATTR_MODE
))
4583 rc
= ext4_acl_chmod(inode
);
4586 ext4_std_error(inode
->i_sb
, error
);
4592 int ext4_getattr(struct vfsmount
*mnt
, struct dentry
*dentry
,
4595 struct inode
*inode
;
4596 unsigned long delalloc_blocks
;
4598 inode
= dentry
->d_inode
;
4599 generic_fillattr(inode
, stat
);
4602 * We can't update i_blocks if the block allocation is delayed
4603 * otherwise in the case of system crash before the real block
4604 * allocation is done, we will have i_blocks inconsistent with
4605 * on-disk file blocks.
4606 * We always keep i_blocks updated together with real
4607 * allocation. But to not confuse with user, stat
4608 * will return the blocks that include the delayed allocation
4609 * blocks for this file.
4611 spin_lock(&EXT4_I(inode
)->i_block_reservation_lock
);
4612 delalloc_blocks
= EXT4_I(inode
)->i_reserved_data_blocks
;
4613 spin_unlock(&EXT4_I(inode
)->i_block_reservation_lock
);
4615 stat
->blocks
+= (delalloc_blocks
<< inode
->i_sb
->s_blocksize_bits
)>>9;
4619 static int ext4_indirect_trans_blocks(struct inode
*inode
, int nrblocks
,
4624 /* if nrblocks are contiguous */
4627 * With N contiguous data blocks, it need at most
4628 * N/EXT4_ADDR_PER_BLOCK(inode->i_sb) indirect blocks
4629 * 2 dindirect blocks
4632 indirects
= nrblocks
/ EXT4_ADDR_PER_BLOCK(inode
->i_sb
);
4633 return indirects
+ 3;
4636 * if nrblocks are not contiguous, worse case, each block touch
4637 * a indirect block, and each indirect block touch a double indirect
4638 * block, plus a triple indirect block
4640 indirects
= nrblocks
* 2 + 1;
4644 static int ext4_index_trans_blocks(struct inode
*inode
, int nrblocks
, int chunk
)
4646 if (!(EXT4_I(inode
)->i_flags
& EXT4_EXTENTS_FL
))
4647 return ext4_indirect_trans_blocks(inode
, nrblocks
, chunk
);
4648 return ext4_ext_index_trans_blocks(inode
, nrblocks
, chunk
);
4652 * Account for index blocks, block groups bitmaps and block group
4653 * descriptor blocks if modify datablocks and index blocks
4654 * worse case, the indexs blocks spread over different block groups
4656 * If datablocks are discontiguous, they are possible to spread over
4657 * different block groups too. If they are contiugous, with flexbg,
4658 * they could still across block group boundary.
4660 * Also account for superblock, inode, quota and xattr blocks
4662 int ext4_meta_trans_blocks(struct inode
*inode
, int nrblocks
, int chunk
)
4664 int groups
, gdpblocks
;
4669 * How many index blocks need to touch to modify nrblocks?
4670 * The "Chunk" flag indicating whether the nrblocks is
4671 * physically contiguous on disk
4673 * For Direct IO and fallocate, they calls get_block to allocate
4674 * one single extent at a time, so they could set the "Chunk" flag
4676 idxblocks
= ext4_index_trans_blocks(inode
, nrblocks
, chunk
);
4681 * Now let's see how many group bitmaps and group descriptors need
4691 if (groups
> EXT4_SB(inode
->i_sb
)->s_groups_count
)
4692 groups
= EXT4_SB(inode
->i_sb
)->s_groups_count
;
4693 if (groups
> EXT4_SB(inode
->i_sb
)->s_gdb_count
)
4694 gdpblocks
= EXT4_SB(inode
->i_sb
)->s_gdb_count
;
4696 /* bitmaps and block group descriptor blocks */
4697 ret
+= groups
+ gdpblocks
;
4699 /* Blocks for super block, inode, quota and xattr blocks */
4700 ret
+= EXT4_META_TRANS_BLOCKS(inode
->i_sb
);
4706 * Calulate the total number of credits to reserve to fit
4707 * the modification of a single pages into a single transaction,
4708 * which may include multiple chunks of block allocations.
4710 * This could be called via ext4_write_begin()
4712 * We need to consider the worse case, when
4713 * one new block per extent.
4715 int ext4_writepage_trans_blocks(struct inode
*inode
)
4717 int bpp
= ext4_journal_blocks_per_page(inode
);
4720 ret
= ext4_meta_trans_blocks(inode
, bpp
, 0);
4722 /* Account for data blocks for journalled mode */
4723 if (ext4_should_journal_data(inode
))
4729 * Calculate the journal credits for a chunk of data modification.
4731 * This is called from DIO, fallocate or whoever calling
4732 * ext4_get_blocks_wrap() to map/allocate a chunk of contigous disk blocks.
4734 * journal buffers for data blocks are not included here, as DIO
4735 * and fallocate do no need to journal data buffers.
4737 int ext4_chunk_trans_blocks(struct inode
*inode
, int nrblocks
)
4739 return ext4_meta_trans_blocks(inode
, nrblocks
, 1);
4743 * The caller must have previously called ext4_reserve_inode_write().
4744 * Give this, we know that the caller already has write access to iloc->bh.
4746 int ext4_mark_iloc_dirty(handle_t
*handle
,
4747 struct inode
*inode
, struct ext4_iloc
*iloc
)
4751 if (test_opt(inode
->i_sb
, I_VERSION
))
4752 inode_inc_iversion(inode
);
4754 /* the do_update_inode consumes one bh->b_count */
4757 /* ext4_do_update_inode() does jbd2_journal_dirty_metadata */
4758 err
= ext4_do_update_inode(handle
, inode
, iloc
);
4764 * On success, We end up with an outstanding reference count against
4765 * iloc->bh. This _must_ be cleaned up later.
4769 ext4_reserve_inode_write(handle_t
*handle
, struct inode
*inode
,
4770 struct ext4_iloc
*iloc
)
4774 err
= ext4_get_inode_loc(inode
, iloc
);
4776 BUFFER_TRACE(iloc
->bh
, "get_write_access");
4777 err
= ext4_journal_get_write_access(handle
, iloc
->bh
);
4784 ext4_std_error(inode
->i_sb
, err
);
4789 * Expand an inode by new_extra_isize bytes.
4790 * Returns 0 on success or negative error number on failure.
4792 static int ext4_expand_extra_isize(struct inode
*inode
,
4793 unsigned int new_extra_isize
,
4794 struct ext4_iloc iloc
,
4797 struct ext4_inode
*raw_inode
;
4798 struct ext4_xattr_ibody_header
*header
;
4799 struct ext4_xattr_entry
*entry
;
4801 if (EXT4_I(inode
)->i_extra_isize
>= new_extra_isize
)
4804 raw_inode
= ext4_raw_inode(&iloc
);
4806 header
= IHDR(inode
, raw_inode
);
4807 entry
= IFIRST(header
);
4809 /* No extended attributes present */
4810 if (!(EXT4_I(inode
)->i_state
& EXT4_STATE_XATTR
) ||
4811 header
->h_magic
!= cpu_to_le32(EXT4_XATTR_MAGIC
)) {
4812 memset((void *)raw_inode
+ EXT4_GOOD_OLD_INODE_SIZE
, 0,
4814 EXT4_I(inode
)->i_extra_isize
= new_extra_isize
;
4818 /* try to expand with EAs present */
4819 return ext4_expand_extra_isize_ea(inode
, new_extra_isize
,
4824 * What we do here is to mark the in-core inode as clean with respect to inode
4825 * dirtiness (it may still be data-dirty).
4826 * This means that the in-core inode may be reaped by prune_icache
4827 * without having to perform any I/O. This is a very good thing,
4828 * because *any* task may call prune_icache - even ones which
4829 * have a transaction open against a different journal.
4831 * Is this cheating? Not really. Sure, we haven't written the
4832 * inode out, but prune_icache isn't a user-visible syncing function.
4833 * Whenever the user wants stuff synced (sys_sync, sys_msync, sys_fsync)
4834 * we start and wait on commits.
4836 * Is this efficient/effective? Well, we're being nice to the system
4837 * by cleaning up our inodes proactively so they can be reaped
4838 * without I/O. But we are potentially leaving up to five seconds'
4839 * worth of inodes floating about which prune_icache wants us to
4840 * write out. One way to fix that would be to get prune_icache()
4841 * to do a write_super() to free up some memory. It has the desired
4844 int ext4_mark_inode_dirty(handle_t
*handle
, struct inode
*inode
)
4846 struct ext4_iloc iloc
;
4847 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
4848 static unsigned int mnt_count
;
4852 err
= ext4_reserve_inode_write(handle
, inode
, &iloc
);
4853 if (EXT4_I(inode
)->i_extra_isize
< sbi
->s_want_extra_isize
&&
4854 !(EXT4_I(inode
)->i_state
& EXT4_STATE_NO_EXPAND
)) {
4856 * We need extra buffer credits since we may write into EA block
4857 * with this same handle. If journal_extend fails, then it will
4858 * only result in a minor loss of functionality for that inode.
4859 * If this is felt to be critical, then e2fsck should be run to
4860 * force a large enough s_min_extra_isize.
4862 if ((jbd2_journal_extend(handle
,
4863 EXT4_DATA_TRANS_BLOCKS(inode
->i_sb
))) == 0) {
4864 ret
= ext4_expand_extra_isize(inode
,
4865 sbi
->s_want_extra_isize
,
4868 EXT4_I(inode
)->i_state
|= EXT4_STATE_NO_EXPAND
;
4870 le16_to_cpu(sbi
->s_es
->s_mnt_count
)) {
4871 ext4_warning(inode
->i_sb
, __func__
,
4872 "Unable to expand inode %lu. Delete"
4873 " some EAs or run e2fsck.",
4876 le16_to_cpu(sbi
->s_es
->s_mnt_count
);
4882 err
= ext4_mark_iloc_dirty(handle
, inode
, &iloc
);
4887 * ext4_dirty_inode() is called from __mark_inode_dirty()
4889 * We're really interested in the case where a file is being extended.
4890 * i_size has been changed by generic_commit_write() and we thus need
4891 * to include the updated inode in the current transaction.
4893 * Also, DQUOT_ALLOC_SPACE() will always dirty the inode when blocks
4894 * are allocated to the file.
4896 * If the inode is marked synchronous, we don't honour that here - doing
4897 * so would cause a commit on atime updates, which we don't bother doing.
4898 * We handle synchronous inodes at the highest possible level.
4900 void ext4_dirty_inode(struct inode
*inode
)
4902 handle_t
*current_handle
= ext4_journal_current_handle();
4905 handle
= ext4_journal_start(inode
, 2);
4908 if (current_handle
&&
4909 current_handle
->h_transaction
!= handle
->h_transaction
) {
4910 /* This task has a transaction open against a different fs */
4911 printk(KERN_EMERG
"%s: transactions do not match!\n",
4914 jbd_debug(5, "marking dirty. outer handle=%p\n",
4916 ext4_mark_inode_dirty(handle
, inode
);
4918 ext4_journal_stop(handle
);
4925 * Bind an inode's backing buffer_head into this transaction, to prevent
4926 * it from being flushed to disk early. Unlike
4927 * ext4_reserve_inode_write, this leaves behind no bh reference and
4928 * returns no iloc structure, so the caller needs to repeat the iloc
4929 * lookup to mark the inode dirty later.
4931 static int ext4_pin_inode(handle_t
*handle
, struct inode
*inode
)
4933 struct ext4_iloc iloc
;
4937 err
= ext4_get_inode_loc(inode
, &iloc
);
4939 BUFFER_TRACE(iloc
.bh
, "get_write_access");
4940 err
= jbd2_journal_get_write_access(handle
, iloc
.bh
);
4942 err
= ext4_journal_dirty_metadata(handle
,
4947 ext4_std_error(inode
->i_sb
, err
);
4952 int ext4_change_inode_journal_flag(struct inode
*inode
, int val
)
4959 * We have to be very careful here: changing a data block's
4960 * journaling status dynamically is dangerous. If we write a
4961 * data block to the journal, change the status and then delete
4962 * that block, we risk forgetting to revoke the old log record
4963 * from the journal and so a subsequent replay can corrupt data.
4964 * So, first we make sure that the journal is empty and that
4965 * nobody is changing anything.
4968 journal
= EXT4_JOURNAL(inode
);
4969 if (is_journal_aborted(journal
))
4972 jbd2_journal_lock_updates(journal
);
4973 jbd2_journal_flush(journal
);
4976 * OK, there are no updates running now, and all cached data is
4977 * synced to disk. We are now in a completely consistent state
4978 * which doesn't have anything in the journal, and we know that
4979 * no filesystem updates are running, so it is safe to modify
4980 * the inode's in-core data-journaling state flag now.
4984 EXT4_I(inode
)->i_flags
|= EXT4_JOURNAL_DATA_FL
;
4986 EXT4_I(inode
)->i_flags
&= ~EXT4_JOURNAL_DATA_FL
;
4987 ext4_set_aops(inode
);
4989 jbd2_journal_unlock_updates(journal
);
4991 /* Finally we can mark the inode as dirty. */
4993 handle
= ext4_journal_start(inode
, 1);
4995 return PTR_ERR(handle
);
4997 err
= ext4_mark_inode_dirty(handle
, inode
);
4999 ext4_journal_stop(handle
);
5000 ext4_std_error(inode
->i_sb
, err
);
5005 static int ext4_bh_unmapped(handle_t
*handle
, struct buffer_head
*bh
)
5007 return !buffer_mapped(bh
);
5010 int ext4_page_mkwrite(struct vm_area_struct
*vma
, struct page
*page
)
5016 struct file
*file
= vma
->vm_file
;
5017 struct inode
*inode
= file
->f_path
.dentry
->d_inode
;
5018 struct address_space
*mapping
= inode
->i_mapping
;
5021 * Get i_alloc_sem to stop truncates messing with the inode. We cannot
5022 * get i_mutex because we are already holding mmap_sem.
5024 down_read(&inode
->i_alloc_sem
);
5025 size
= i_size_read(inode
);
5026 if (page
->mapping
!= mapping
|| size
<= page_offset(page
)
5027 || !PageUptodate(page
)) {
5028 /* page got truncated from under us? */
5032 if (PageMappedToDisk(page
))
5035 if (page
->index
== size
>> PAGE_CACHE_SHIFT
)
5036 len
= size
& ~PAGE_CACHE_MASK
;
5038 len
= PAGE_CACHE_SIZE
;
5040 if (page_has_buffers(page
)) {
5041 /* return if we have all the buffers mapped */
5042 if (!walk_page_buffers(NULL
, page_buffers(page
), 0, len
, NULL
,
5047 * OK, we need to fill the hole... Do write_begin write_end
5048 * to do block allocation/reservation.We are not holding
5049 * inode.i__mutex here. That allow * parallel write_begin,
5050 * write_end call. lock_page prevent this from happening
5051 * on the same page though
5053 ret
= mapping
->a_ops
->write_begin(file
, mapping
, page_offset(page
),
5054 len
, AOP_FLAG_UNINTERRUPTIBLE
, &page
, &fsdata
);
5057 ret
= mapping
->a_ops
->write_end(file
, mapping
, page_offset(page
),
5058 len
, len
, page
, fsdata
);
5063 up_read(&inode
->i_alloc_sem
);