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/namei.h>
38 #include <linux/uio.h>
39 #include <linux/bio.h>
40 #include "ext4_jbd2.h"
43 #include "ext4_extents.h"
45 #define MPAGE_DA_EXTENT_TAIL 0x01
47 static inline int ext4_begin_ordered_truncate(struct inode
*inode
,
50 return jbd2_journal_begin_ordered_truncate(&EXT4_I(inode
)->jinode
,
54 static void ext4_invalidatepage(struct page
*page
, unsigned long offset
);
57 * Test whether an inode is a fast symlink.
59 static int ext4_inode_is_fast_symlink(struct inode
*inode
)
61 int ea_blocks
= EXT4_I(inode
)->i_file_acl
?
62 (inode
->i_sb
->s_blocksize
>> 9) : 0;
64 return (S_ISLNK(inode
->i_mode
) && inode
->i_blocks
- ea_blocks
== 0);
68 * The ext4 forget function must perform a revoke if we are freeing data
69 * which has been journaled. Metadata (eg. indirect blocks) must be
70 * revoked in all cases.
72 * "bh" may be NULL: a metadata block may have been freed from memory
73 * but there may still be a record of it in the journal, and that record
74 * still needs to be revoked.
76 int ext4_forget(handle_t
*handle
, int is_metadata
, struct inode
*inode
,
77 struct buffer_head
*bh
, ext4_fsblk_t blocknr
)
83 BUFFER_TRACE(bh
, "enter");
85 jbd_debug(4, "forgetting bh %p: is_metadata = %d, mode %o, "
87 bh
, is_metadata
, inode
->i_mode
,
88 test_opt(inode
->i_sb
, DATA_FLAGS
));
90 /* Never use the revoke function if we are doing full data
91 * journaling: there is no need to, and a V1 superblock won't
92 * support it. Otherwise, only skip the revoke on un-journaled
95 if (test_opt(inode
->i_sb
, DATA_FLAGS
) == EXT4_MOUNT_JOURNAL_DATA
||
96 (!is_metadata
&& !ext4_should_journal_data(inode
))) {
98 BUFFER_TRACE(bh
, "call jbd2_journal_forget");
99 return ext4_journal_forget(handle
, bh
);
105 * data!=journal && (is_metadata || should_journal_data(inode))
107 BUFFER_TRACE(bh
, "call ext4_journal_revoke");
108 err
= ext4_journal_revoke(handle
, blocknr
, bh
);
110 ext4_abort(inode
->i_sb
, __func__
,
111 "error %d when attempting revoke", err
);
112 BUFFER_TRACE(bh
, "exit");
117 * Work out how many blocks we need to proceed with the next chunk of a
118 * truncate transaction.
120 static unsigned long blocks_for_truncate(struct inode
*inode
)
124 needed
= inode
->i_blocks
>> (inode
->i_sb
->s_blocksize_bits
- 9);
126 /* Give ourselves just enough room to cope with inodes in which
127 * i_blocks is corrupt: we've seen disk corruptions in the past
128 * which resulted in random data in an inode which looked enough
129 * like a regular file for ext4 to try to delete it. Things
130 * will go a bit crazy if that happens, but at least we should
131 * try not to panic the whole kernel. */
135 /* But we need to bound the transaction so we don't overflow the
137 if (needed
> EXT4_MAX_TRANS_DATA
)
138 needed
= EXT4_MAX_TRANS_DATA
;
140 return EXT4_DATA_TRANS_BLOCKS(inode
->i_sb
) + needed
;
144 * Truncate transactions can be complex and absolutely huge. So we need to
145 * be able to restart the transaction at a conventient checkpoint to make
146 * sure we don't overflow the journal.
148 * start_transaction gets us a new handle for a truncate transaction,
149 * and extend_transaction tries to extend the existing one a bit. If
150 * extend fails, we need to propagate the failure up and restart the
151 * transaction in the top-level truncate loop. --sct
153 static handle_t
*start_transaction(struct inode
*inode
)
157 result
= ext4_journal_start(inode
, blocks_for_truncate(inode
));
161 ext4_std_error(inode
->i_sb
, PTR_ERR(result
));
166 * Try to extend this transaction for the purposes of truncation.
168 * Returns 0 if we managed to create more room. If we can't create more
169 * room, and the transaction must be restarted we return 1.
171 static int try_to_extend_transaction(handle_t
*handle
, struct inode
*inode
)
173 if (handle
->h_buffer_credits
> EXT4_RESERVE_TRANS_BLOCKS
)
175 if (!ext4_journal_extend(handle
, blocks_for_truncate(inode
)))
181 * Restart the transaction associated with *handle. This does a commit,
182 * so before we call here everything must be consistently dirtied against
185 static int ext4_journal_test_restart(handle_t
*handle
, struct inode
*inode
)
187 jbd_debug(2, "restarting handle %p\n", handle
);
188 return ext4_journal_restart(handle
, blocks_for_truncate(inode
));
192 * Called at the last iput() if i_nlink is zero.
194 void ext4_delete_inode(struct inode
*inode
)
199 if (ext4_should_order_data(inode
))
200 ext4_begin_ordered_truncate(inode
, 0);
201 truncate_inode_pages(&inode
->i_data
, 0);
203 if (is_bad_inode(inode
))
206 handle
= ext4_journal_start(inode
, blocks_for_truncate(inode
)+3);
207 if (IS_ERR(handle
)) {
208 ext4_std_error(inode
->i_sb
, PTR_ERR(handle
));
210 * If we're going to skip the normal cleanup, we still need to
211 * make sure that the in-core orphan linked list is properly
214 ext4_orphan_del(NULL
, inode
);
221 err
= ext4_mark_inode_dirty(handle
, inode
);
223 ext4_warning(inode
->i_sb
, __func__
,
224 "couldn't mark inode dirty (err %d)", err
);
228 ext4_truncate(inode
);
231 * ext4_ext_truncate() doesn't reserve any slop when it
232 * restarts journal transactions; therefore there may not be
233 * enough credits left in the handle to remove the inode from
234 * the orphan list and set the dtime field.
236 if (handle
->h_buffer_credits
< 3) {
237 err
= ext4_journal_extend(handle
, 3);
239 err
= ext4_journal_restart(handle
, 3);
241 ext4_warning(inode
->i_sb
, __func__
,
242 "couldn't extend journal (err %d)", err
);
244 ext4_journal_stop(handle
);
250 * Kill off the orphan record which ext4_truncate created.
251 * AKPM: I think this can be inside the above `if'.
252 * Note that ext4_orphan_del() has to be able to cope with the
253 * deletion of a non-existent orphan - this is because we don't
254 * know if ext4_truncate() actually created an orphan record.
255 * (Well, we could do this if we need to, but heck - it works)
257 ext4_orphan_del(handle
, inode
);
258 EXT4_I(inode
)->i_dtime
= get_seconds();
261 * One subtle ordering requirement: if anything has gone wrong
262 * (transaction abort, IO errors, whatever), then we can still
263 * do these next steps (the fs will already have been marked as
264 * having errors), but we can't free the inode if the mark_dirty
267 if (ext4_mark_inode_dirty(handle
, inode
))
268 /* If that failed, just do the required in-core inode clear. */
271 ext4_free_inode(handle
, inode
);
272 ext4_journal_stop(handle
);
275 clear_inode(inode
); /* We must guarantee clearing of inode... */
281 struct buffer_head
*bh
;
284 static inline void add_chain(Indirect
*p
, struct buffer_head
*bh
, __le32
*v
)
286 p
->key
= *(p
->p
= v
);
291 * ext4_block_to_path - parse the block number into array of offsets
292 * @inode: inode in question (we are only interested in its superblock)
293 * @i_block: block number to be parsed
294 * @offsets: array to store the offsets in
295 * @boundary: set this non-zero if the referred-to block is likely to be
296 * followed (on disk) by an indirect block.
298 * To store the locations of file's data ext4 uses a data structure common
299 * for UNIX filesystems - tree of pointers anchored in the inode, with
300 * data blocks at leaves and indirect blocks in intermediate nodes.
301 * This function translates the block number into path in that tree -
302 * return value is the path length and @offsets[n] is the offset of
303 * pointer to (n+1)th node in the nth one. If @block is out of range
304 * (negative or too large) warning is printed and zero returned.
306 * Note: function doesn't find node addresses, so no IO is needed. All
307 * we need to know is the capacity of indirect blocks (taken from the
312 * Portability note: the last comparison (check that we fit into triple
313 * indirect block) is spelled differently, because otherwise on an
314 * architecture with 32-bit longs and 8Kb pages we might get into trouble
315 * if our filesystem had 8Kb blocks. We might use long long, but that would
316 * kill us on x86. Oh, well, at least the sign propagation does not matter -
317 * i_block would have to be negative in the very beginning, so we would not
321 static int ext4_block_to_path(struct inode
*inode
,
323 ext4_lblk_t offsets
[4], int *boundary
)
325 int ptrs
= EXT4_ADDR_PER_BLOCK(inode
->i_sb
);
326 int ptrs_bits
= EXT4_ADDR_PER_BLOCK_BITS(inode
->i_sb
);
327 const long direct_blocks
= EXT4_NDIR_BLOCKS
,
328 indirect_blocks
= ptrs
,
329 double_blocks
= (1 << (ptrs_bits
* 2));
334 ext4_warning(inode
->i_sb
, "ext4_block_to_path", "block < 0");
335 } else if (i_block
< direct_blocks
) {
336 offsets
[n
++] = i_block
;
337 final
= direct_blocks
;
338 } else if ((i_block
-= direct_blocks
) < indirect_blocks
) {
339 offsets
[n
++] = EXT4_IND_BLOCK
;
340 offsets
[n
++] = i_block
;
342 } else if ((i_block
-= indirect_blocks
) < double_blocks
) {
343 offsets
[n
++] = EXT4_DIND_BLOCK
;
344 offsets
[n
++] = i_block
>> ptrs_bits
;
345 offsets
[n
++] = i_block
& (ptrs
- 1);
347 } else if (((i_block
-= double_blocks
) >> (ptrs_bits
* 2)) < ptrs
) {
348 offsets
[n
++] = EXT4_TIND_BLOCK
;
349 offsets
[n
++] = i_block
>> (ptrs_bits
* 2);
350 offsets
[n
++] = (i_block
>> ptrs_bits
) & (ptrs
- 1);
351 offsets
[n
++] = i_block
& (ptrs
- 1);
354 ext4_warning(inode
->i_sb
, "ext4_block_to_path",
356 i_block
+ direct_blocks
+
357 indirect_blocks
+ double_blocks
);
360 *boundary
= final
- 1 - (i_block
& (ptrs
- 1));
365 * ext4_get_branch - read the chain of indirect blocks leading to data
366 * @inode: inode in question
367 * @depth: depth of the chain (1 - direct pointer, etc.)
368 * @offsets: offsets of pointers in inode/indirect blocks
369 * @chain: place to store the result
370 * @err: here we store the error value
372 * Function fills the array of triples <key, p, bh> and returns %NULL
373 * if everything went OK or the pointer to the last filled triple
374 * (incomplete one) otherwise. Upon the return chain[i].key contains
375 * the number of (i+1)-th block in the chain (as it is stored in memory,
376 * i.e. little-endian 32-bit), chain[i].p contains the address of that
377 * number (it points into struct inode for i==0 and into the bh->b_data
378 * for i>0) and chain[i].bh points to the buffer_head of i-th indirect
379 * block for i>0 and NULL for i==0. In other words, it holds the block
380 * numbers of the chain, addresses they were taken from (and where we can
381 * verify that chain did not change) and buffer_heads hosting these
384 * Function stops when it stumbles upon zero pointer (absent block)
385 * (pointer to last triple returned, *@err == 0)
386 * or when it gets an IO error reading an indirect block
387 * (ditto, *@err == -EIO)
388 * or when it reads all @depth-1 indirect blocks successfully and finds
389 * the whole chain, all way to the data (returns %NULL, *err == 0).
391 * Need to be called with
392 * down_read(&EXT4_I(inode)->i_data_sem)
394 static Indirect
*ext4_get_branch(struct inode
*inode
, int depth
,
395 ext4_lblk_t
*offsets
,
396 Indirect chain
[4], int *err
)
398 struct super_block
*sb
= inode
->i_sb
;
400 struct buffer_head
*bh
;
403 /* i_data is not going away, no lock needed */
404 add_chain(chain
, NULL
, EXT4_I(inode
)->i_data
+ *offsets
);
408 bh
= sb_bread(sb
, le32_to_cpu(p
->key
));
411 add_chain(++p
, bh
, (__le32
*)bh
->b_data
+ *++offsets
);
425 * ext4_find_near - find a place for allocation with sufficient locality
427 * @ind: descriptor of indirect block.
429 * This function returns the preferred place for block allocation.
430 * It is used when heuristic for sequential allocation fails.
432 * + if there is a block to the left of our position - allocate near it.
433 * + if pointer will live in indirect block - allocate near that block.
434 * + if pointer will live in inode - allocate in the same
437 * In the latter case we colour the starting block by the callers PID to
438 * prevent it from clashing with concurrent allocations for a different inode
439 * in the same block group. The PID is used here so that functionally related
440 * files will be close-by on-disk.
442 * Caller must make sure that @ind is valid and will stay that way.
444 static ext4_fsblk_t
ext4_find_near(struct inode
*inode
, Indirect
*ind
)
446 struct ext4_inode_info
*ei
= EXT4_I(inode
);
447 __le32
*start
= ind
->bh
? (__le32
*) ind
->bh
->b_data
: ei
->i_data
;
449 ext4_fsblk_t bg_start
;
450 ext4_fsblk_t last_block
;
451 ext4_grpblk_t colour
;
453 /* Try to find previous block */
454 for (p
= ind
->p
- 1; p
>= start
; p
--) {
456 return le32_to_cpu(*p
);
459 /* No such thing, so let's try location of indirect block */
461 return ind
->bh
->b_blocknr
;
464 * It is going to be referred to from the inode itself? OK, just put it
465 * into the same cylinder group then.
467 bg_start
= ext4_group_first_block_no(inode
->i_sb
, ei
->i_block_group
);
468 last_block
= ext4_blocks_count(EXT4_SB(inode
->i_sb
)->s_es
) - 1;
470 if (bg_start
+ EXT4_BLOCKS_PER_GROUP(inode
->i_sb
) <= last_block
)
471 colour
= (current
->pid
% 16) *
472 (EXT4_BLOCKS_PER_GROUP(inode
->i_sb
) / 16);
474 colour
= (current
->pid
% 16) * ((last_block
- bg_start
) / 16);
475 return bg_start
+ colour
;
479 * ext4_find_goal - find a preferred place for allocation.
481 * @block: block we want
482 * @partial: pointer to the last triple within a chain
484 * Normally this function find the preferred place for block allocation,
487 static ext4_fsblk_t
ext4_find_goal(struct inode
*inode
, ext4_lblk_t block
,
491 * XXX need to get goal block from mballoc's data structures
494 return ext4_find_near(inode
, partial
);
498 * ext4_blks_to_allocate: Look up the block map and count the number
499 * of direct blocks need to be allocated for the given branch.
501 * @branch: chain of indirect blocks
502 * @k: number of blocks need for indirect blocks
503 * @blks: number of data blocks to be mapped.
504 * @blocks_to_boundary: the offset in the indirect block
506 * return the total number of blocks to be allocate, including the
507 * direct and indirect blocks.
509 static int ext4_blks_to_allocate(Indirect
*branch
, int k
, unsigned long blks
,
510 int blocks_to_boundary
)
512 unsigned long count
= 0;
515 * Simple case, [t,d]Indirect block(s) has not allocated yet
516 * then it's clear blocks on that path have not allocated
519 /* right now we don't handle cross boundary allocation */
520 if (blks
< blocks_to_boundary
+ 1)
523 count
+= blocks_to_boundary
+ 1;
528 while (count
< blks
&& count
<= blocks_to_boundary
&&
529 le32_to_cpu(*(branch
[0].p
+ count
)) == 0) {
536 * ext4_alloc_blocks: multiple allocate blocks needed for a branch
537 * @indirect_blks: the number of blocks need to allocate for indirect
540 * @new_blocks: on return it will store the new block numbers for
541 * the indirect blocks(if needed) and the first direct block,
542 * @blks: on return it will store the total number of allocated
545 static int ext4_alloc_blocks(handle_t
*handle
, struct inode
*inode
,
546 ext4_lblk_t iblock
, ext4_fsblk_t goal
,
547 int indirect_blks
, int blks
,
548 ext4_fsblk_t new_blocks
[4], int *err
)
551 unsigned long count
= 0, blk_allocated
= 0;
553 ext4_fsblk_t current_block
= 0;
557 * Here we try to allocate the requested multiple blocks at once,
558 * on a best-effort basis.
559 * To build a branch, we should allocate blocks for
560 * the indirect blocks(if not allocated yet), and at least
561 * the first direct block of this branch. That's the
562 * minimum number of blocks need to allocate(required)
564 /* first we try to allocate the indirect blocks */
565 target
= indirect_blks
;
568 /* allocating blocks for indirect blocks and direct blocks */
569 current_block
= ext4_new_meta_blocks(handle
, inode
,
575 /* allocate blocks for indirect blocks */
576 while (index
< indirect_blks
&& count
) {
577 new_blocks
[index
++] = current_block
++;
582 * save the new block number
583 * for the first direct block
585 new_blocks
[index
] = current_block
;
586 printk(KERN_INFO
"%s returned more blocks than "
587 "requested\n", __func__
);
593 target
= blks
- count
;
594 blk_allocated
= count
;
597 /* Now allocate data blocks */
599 /* allocating blocks for data blocks */
600 current_block
= ext4_new_blocks(handle
, inode
, iblock
,
602 if (*err
&& (target
== blks
)) {
604 * if the allocation failed and we didn't allocate
610 if (target
== blks
) {
612 * save the new block number
613 * for the first direct block
615 new_blocks
[index
] = current_block
;
617 blk_allocated
+= count
;
620 /* total number of blocks allocated for direct blocks */
625 for (i
= 0; i
< index
; i
++)
626 ext4_free_blocks(handle
, inode
, new_blocks
[i
], 1, 0);
631 * ext4_alloc_branch - allocate and set up a chain of blocks.
633 * @indirect_blks: number of allocated indirect blocks
634 * @blks: number of allocated direct blocks
635 * @offsets: offsets (in the blocks) to store the pointers to next.
636 * @branch: place to store the chain in.
638 * This function allocates blocks, zeroes out all but the last one,
639 * links them into chain and (if we are synchronous) writes them to disk.
640 * In other words, it prepares a branch that can be spliced onto the
641 * inode. It stores the information about that chain in the branch[], in
642 * the same format as ext4_get_branch() would do. We are calling it after
643 * we had read the existing part of chain and partial points to the last
644 * triple of that (one with zero ->key). Upon the exit we have the same
645 * picture as after the successful ext4_get_block(), except that in one
646 * place chain is disconnected - *branch->p is still zero (we did not
647 * set the last link), but branch->key contains the number that should
648 * be placed into *branch->p to fill that gap.
650 * If allocation fails we free all blocks we've allocated (and forget
651 * their buffer_heads) and return the error value the from failed
652 * ext4_alloc_block() (normally -ENOSPC). Otherwise we set the chain
653 * as described above and return 0.
655 static int ext4_alloc_branch(handle_t
*handle
, struct inode
*inode
,
656 ext4_lblk_t iblock
, int indirect_blks
,
657 int *blks
, ext4_fsblk_t goal
,
658 ext4_lblk_t
*offsets
, Indirect
*branch
)
660 int blocksize
= inode
->i_sb
->s_blocksize
;
663 struct buffer_head
*bh
;
665 ext4_fsblk_t new_blocks
[4];
666 ext4_fsblk_t current_block
;
668 num
= ext4_alloc_blocks(handle
, inode
, iblock
, goal
, indirect_blks
,
669 *blks
, new_blocks
, &err
);
673 branch
[0].key
= cpu_to_le32(new_blocks
[0]);
675 * metadata blocks and data blocks are allocated.
677 for (n
= 1; n
<= indirect_blks
; n
++) {
679 * Get buffer_head for parent block, zero it out
680 * and set the pointer to new one, then send
683 bh
= sb_getblk(inode
->i_sb
, new_blocks
[n
-1]);
686 BUFFER_TRACE(bh
, "call get_create_access");
687 err
= ext4_journal_get_create_access(handle
, bh
);
694 memset(bh
->b_data
, 0, blocksize
);
695 branch
[n
].p
= (__le32
*) bh
->b_data
+ offsets
[n
];
696 branch
[n
].key
= cpu_to_le32(new_blocks
[n
]);
697 *branch
[n
].p
= branch
[n
].key
;
698 if (n
== indirect_blks
) {
699 current_block
= new_blocks
[n
];
701 * End of chain, update the last new metablock of
702 * the chain to point to the new allocated
703 * data blocks numbers
705 for (i
=1; i
< num
; i
++)
706 *(branch
[n
].p
+ i
) = cpu_to_le32(++current_block
);
708 BUFFER_TRACE(bh
, "marking uptodate");
709 set_buffer_uptodate(bh
);
712 BUFFER_TRACE(bh
, "call ext4_journal_dirty_metadata");
713 err
= ext4_journal_dirty_metadata(handle
, bh
);
720 /* Allocation failed, free what we already allocated */
721 for (i
= 1; i
<= n
; i
++) {
722 BUFFER_TRACE(branch
[i
].bh
, "call jbd2_journal_forget");
723 ext4_journal_forget(handle
, branch
[i
].bh
);
725 for (i
= 0; i
< indirect_blks
; i
++)
726 ext4_free_blocks(handle
, inode
, new_blocks
[i
], 1, 0);
728 ext4_free_blocks(handle
, inode
, new_blocks
[i
], num
, 0);
734 * ext4_splice_branch - splice the allocated branch onto inode.
736 * @block: (logical) number of block we are adding
737 * @chain: chain of indirect blocks (with a missing link - see
739 * @where: location of missing link
740 * @num: number of indirect blocks we are adding
741 * @blks: number of direct blocks we are adding
743 * This function fills the missing link and does all housekeeping needed in
744 * inode (->i_blocks, etc.). In case of success we end up with the full
745 * chain to new block and return 0.
747 static int ext4_splice_branch(handle_t
*handle
, struct inode
*inode
,
748 ext4_lblk_t block
, Indirect
*where
, int num
, int blks
)
752 ext4_fsblk_t current_block
;
755 * If we're splicing into a [td]indirect block (as opposed to the
756 * inode) then we need to get write access to the [td]indirect block
760 BUFFER_TRACE(where
->bh
, "get_write_access");
761 err
= ext4_journal_get_write_access(handle
, where
->bh
);
767 *where
->p
= where
->key
;
770 * Update the host buffer_head or inode to point to more just allocated
771 * direct blocks blocks
773 if (num
== 0 && blks
> 1) {
774 current_block
= le32_to_cpu(where
->key
) + 1;
775 for (i
= 1; i
< blks
; i
++)
776 *(where
->p
+ i
) = cpu_to_le32(current_block
++);
779 /* We are done with atomic stuff, now do the rest of housekeeping */
781 inode
->i_ctime
= ext4_current_time(inode
);
782 ext4_mark_inode_dirty(handle
, inode
);
784 /* had we spliced it onto indirect block? */
787 * If we spliced it onto an indirect block, we haven't
788 * altered the inode. Note however that if it is being spliced
789 * onto an indirect block at the very end of the file (the
790 * file is growing) then we *will* alter the inode to reflect
791 * the new i_size. But that is not done here - it is done in
792 * generic_commit_write->__mark_inode_dirty->ext4_dirty_inode.
794 jbd_debug(5, "splicing indirect only\n");
795 BUFFER_TRACE(where
->bh
, "call ext4_journal_dirty_metadata");
796 err
= ext4_journal_dirty_metadata(handle
, where
->bh
);
801 * OK, we spliced it into the inode itself on a direct block.
802 * Inode was dirtied above.
804 jbd_debug(5, "splicing direct\n");
809 for (i
= 1; i
<= num
; i
++) {
810 BUFFER_TRACE(where
[i
].bh
, "call jbd2_journal_forget");
811 ext4_journal_forget(handle
, where
[i
].bh
);
812 ext4_free_blocks(handle
, inode
,
813 le32_to_cpu(where
[i
-1].key
), 1, 0);
815 ext4_free_blocks(handle
, inode
, le32_to_cpu(where
[num
].key
), blks
, 0);
821 * Allocation strategy is simple: if we have to allocate something, we will
822 * have to go the whole way to leaf. So let's do it before attaching anything
823 * to tree, set linkage between the newborn blocks, write them if sync is
824 * required, recheck the path, free and repeat if check fails, otherwise
825 * set the last missing link (that will protect us from any truncate-generated
826 * removals - all blocks on the path are immune now) and possibly force the
827 * write on the parent block.
828 * That has a nice additional property: no special recovery from the failed
829 * allocations is needed - we simply release blocks and do not touch anything
830 * reachable from inode.
832 * `handle' can be NULL if create == 0.
834 * return > 0, # of blocks mapped or allocated.
835 * return = 0, if plain lookup failed.
836 * return < 0, error case.
839 * Need to be called with
840 * down_read(&EXT4_I(inode)->i_data_sem) if not allocating file system block
841 * (ie, create is zero). Otherwise down_write(&EXT4_I(inode)->i_data_sem)
843 int ext4_get_blocks_handle(handle_t
*handle
, struct inode
*inode
,
844 ext4_lblk_t iblock
, unsigned long maxblocks
,
845 struct buffer_head
*bh_result
,
846 int create
, int extend_disksize
)
849 ext4_lblk_t offsets
[4];
854 int blocks_to_boundary
= 0;
856 struct ext4_inode_info
*ei
= EXT4_I(inode
);
858 ext4_fsblk_t first_block
= 0;
862 J_ASSERT(!(EXT4_I(inode
)->i_flags
& EXT4_EXTENTS_FL
));
863 J_ASSERT(handle
!= NULL
|| create
== 0);
864 depth
= ext4_block_to_path(inode
, iblock
, offsets
,
865 &blocks_to_boundary
);
870 partial
= ext4_get_branch(inode
, depth
, offsets
, chain
, &err
);
872 /* Simplest case - block found, no allocation needed */
874 first_block
= le32_to_cpu(chain
[depth
- 1].key
);
875 clear_buffer_new(bh_result
);
878 while (count
< maxblocks
&& count
<= blocks_to_boundary
) {
881 blk
= le32_to_cpu(*(chain
[depth
-1].p
+ count
));
883 if (blk
== first_block
+ count
)
891 /* Next simple case - plain lookup or failed read of indirect block */
892 if (!create
|| err
== -EIO
)
896 * Okay, we need to do block allocation.
898 goal
= ext4_find_goal(inode
, iblock
, partial
);
900 /* the number of blocks need to allocate for [d,t]indirect blocks */
901 indirect_blks
= (chain
+ depth
) - partial
- 1;
904 * Next look up the indirect map to count the totoal number of
905 * direct blocks to allocate for this branch.
907 count
= ext4_blks_to_allocate(partial
, indirect_blks
,
908 maxblocks
, blocks_to_boundary
);
910 * Block out ext4_truncate while we alter the tree
912 err
= ext4_alloc_branch(handle
, inode
, iblock
, indirect_blks
,
914 offsets
+ (partial
- chain
), partial
);
917 * The ext4_splice_branch call will free and forget any buffers
918 * on the new chain if there is a failure, but that risks using
919 * up transaction credits, especially for bitmaps where the
920 * credits cannot be returned. Can we handle this somehow? We
921 * may need to return -EAGAIN upwards in the worst case. --sct
924 err
= ext4_splice_branch(handle
, inode
, iblock
,
925 partial
, indirect_blks
, count
);
927 * i_disksize growing is protected by i_data_sem. Don't forget to
928 * protect it if you're about to implement concurrent
929 * ext4_get_block() -bzzz
931 if (!err
&& extend_disksize
) {
932 disksize
= ((loff_t
) iblock
+ count
) << inode
->i_blkbits
;
933 if (disksize
> i_size_read(inode
))
934 disksize
= i_size_read(inode
);
935 if (disksize
> ei
->i_disksize
)
936 ei
->i_disksize
= disksize
;
941 set_buffer_new(bh_result
);
943 map_bh(bh_result
, inode
->i_sb
, le32_to_cpu(chain
[depth
-1].key
));
944 if (count
> blocks_to_boundary
)
945 set_buffer_boundary(bh_result
);
947 /* Clean up and exit */
948 partial
= chain
+ depth
- 1; /* the whole chain */
950 while (partial
> chain
) {
951 BUFFER_TRACE(partial
->bh
, "call brelse");
955 BUFFER_TRACE(bh_result
, "returned");
961 * Calculate the number of metadata blocks need to reserve
962 * to allocate @blocks for non extent file based file
964 static int ext4_indirect_calc_metadata_amount(struct inode
*inode
, int blocks
)
966 int icap
= EXT4_ADDR_PER_BLOCK(inode
->i_sb
);
967 int ind_blks
, dind_blks
, tind_blks
;
969 /* number of new indirect blocks needed */
970 ind_blks
= (blocks
+ icap
- 1) / icap
;
972 dind_blks
= (ind_blks
+ icap
- 1) / icap
;
976 return ind_blks
+ dind_blks
+ tind_blks
;
980 * Calculate the number of metadata blocks need to reserve
981 * to allocate given number of blocks
983 static int ext4_calc_metadata_amount(struct inode
*inode
, int blocks
)
988 if (EXT4_I(inode
)->i_flags
& EXT4_EXTENTS_FL
)
989 return ext4_ext_calc_metadata_amount(inode
, blocks
);
991 return ext4_indirect_calc_metadata_amount(inode
, blocks
);
994 static void ext4_da_update_reserve_space(struct inode
*inode
, int used
)
996 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
997 int total
, mdb
, mdb_free
;
999 spin_lock(&EXT4_I(inode
)->i_block_reservation_lock
);
1000 /* recalculate the number of metablocks still need to be reserved */
1001 total
= EXT4_I(inode
)->i_reserved_data_blocks
- used
;
1002 mdb
= ext4_calc_metadata_amount(inode
, total
);
1004 /* figure out how many metablocks to release */
1005 BUG_ON(mdb
> EXT4_I(inode
)->i_reserved_meta_blocks
);
1006 mdb_free
= EXT4_I(inode
)->i_reserved_meta_blocks
- mdb
;
1009 /* Account for allocated meta_blocks */
1010 mdb_free
-= EXT4_I(inode
)->i_allocated_meta_blocks
;
1012 /* update fs dirty blocks counter */
1013 percpu_counter_sub(&sbi
->s_dirtyblocks_counter
, mdb_free
);
1014 EXT4_I(inode
)->i_allocated_meta_blocks
= 0;
1015 EXT4_I(inode
)->i_reserved_meta_blocks
= mdb
;
1018 /* update per-inode reservations */
1019 BUG_ON(used
> EXT4_I(inode
)->i_reserved_data_blocks
);
1020 EXT4_I(inode
)->i_reserved_data_blocks
-= used
;
1022 spin_unlock(&EXT4_I(inode
)->i_block_reservation_lock
);
1026 * The ext4_get_blocks_wrap() function try to look up the requested blocks,
1027 * and returns if the blocks are already mapped.
1029 * Otherwise it takes the write lock of the i_data_sem and allocate blocks
1030 * and store the allocated blocks in the result buffer head and mark it
1033 * If file type is extents based, it will call ext4_ext_get_blocks(),
1034 * Otherwise, call with ext4_get_blocks_handle() to handle indirect mapping
1037 * On success, it returns the number of blocks being mapped or allocate.
1038 * if create==0 and the blocks are pre-allocated and uninitialized block,
1039 * the result buffer head is unmapped. If the create ==1, it will make sure
1040 * the buffer head is mapped.
1042 * It returns 0 if plain look up failed (blocks have not been allocated), in
1043 * that casem, buffer head is unmapped
1045 * It returns the error in case of allocation failure.
1047 int ext4_get_blocks_wrap(handle_t
*handle
, struct inode
*inode
, sector_t block
,
1048 unsigned long max_blocks
, struct buffer_head
*bh
,
1049 int create
, int extend_disksize
, int flag
)
1053 clear_buffer_mapped(bh
);
1056 * Try to see if we can get the block without requesting
1057 * for new file system block.
1059 down_read((&EXT4_I(inode
)->i_data_sem
));
1060 if (EXT4_I(inode
)->i_flags
& EXT4_EXTENTS_FL
) {
1061 retval
= ext4_ext_get_blocks(handle
, inode
, block
, max_blocks
,
1064 retval
= ext4_get_blocks_handle(handle
,
1065 inode
, block
, max_blocks
, bh
, 0, 0);
1067 up_read((&EXT4_I(inode
)->i_data_sem
));
1069 /* If it is only a block(s) look up */
1074 * Returns if the blocks have already allocated
1076 * Note that if blocks have been preallocated
1077 * ext4_ext_get_block() returns th create = 0
1078 * with buffer head unmapped.
1080 if (retval
> 0 && buffer_mapped(bh
))
1084 * New blocks allocate and/or writing to uninitialized extent
1085 * will possibly result in updating i_data, so we take
1086 * the write lock of i_data_sem, and call get_blocks()
1087 * with create == 1 flag.
1089 down_write((&EXT4_I(inode
)->i_data_sem
));
1092 * if the caller is from delayed allocation writeout path
1093 * we have already reserved fs blocks for allocation
1094 * let the underlying get_block() function know to
1095 * avoid double accounting
1098 EXT4_I(inode
)->i_delalloc_reserved_flag
= 1;
1100 * We need to check for EXT4 here because migrate
1101 * could have changed the inode type in between
1103 if (EXT4_I(inode
)->i_flags
& EXT4_EXTENTS_FL
) {
1104 retval
= ext4_ext_get_blocks(handle
, inode
, block
, max_blocks
,
1105 bh
, create
, extend_disksize
);
1107 retval
= ext4_get_blocks_handle(handle
, inode
, block
,
1108 max_blocks
, bh
, create
, extend_disksize
);
1110 if (retval
> 0 && buffer_new(bh
)) {
1112 * We allocated new blocks which will result in
1113 * i_data's format changing. Force the migrate
1114 * to fail by clearing migrate flags
1116 EXT4_I(inode
)->i_flags
= EXT4_I(inode
)->i_flags
&
1122 EXT4_I(inode
)->i_delalloc_reserved_flag
= 0;
1124 * Update reserved blocks/metadata blocks
1125 * after successful block allocation
1126 * which were deferred till now
1128 if ((retval
> 0) && buffer_delay(bh
))
1129 ext4_da_update_reserve_space(inode
, retval
);
1132 up_write((&EXT4_I(inode
)->i_data_sem
));
1136 /* Maximum number of blocks we map for direct IO at once. */
1137 #define DIO_MAX_BLOCKS 4096
1139 int ext4_get_block(struct inode
*inode
, sector_t iblock
,
1140 struct buffer_head
*bh_result
, int create
)
1142 handle_t
*handle
= ext4_journal_current_handle();
1143 int ret
= 0, started
= 0;
1144 unsigned max_blocks
= bh_result
->b_size
>> inode
->i_blkbits
;
1147 if (create
&& !handle
) {
1148 /* Direct IO write... */
1149 if (max_blocks
> DIO_MAX_BLOCKS
)
1150 max_blocks
= DIO_MAX_BLOCKS
;
1151 dio_credits
= ext4_chunk_trans_blocks(inode
, max_blocks
);
1152 handle
= ext4_journal_start(inode
, dio_credits
);
1153 if (IS_ERR(handle
)) {
1154 ret
= PTR_ERR(handle
);
1160 ret
= ext4_get_blocks_wrap(handle
, inode
, iblock
,
1161 max_blocks
, bh_result
, create
, 0, 0);
1163 bh_result
->b_size
= (ret
<< inode
->i_blkbits
);
1167 ext4_journal_stop(handle
);
1173 * `handle' can be NULL if create is zero
1175 struct buffer_head
*ext4_getblk(handle_t
*handle
, struct inode
*inode
,
1176 ext4_lblk_t block
, int create
, int *errp
)
1178 struct buffer_head dummy
;
1181 J_ASSERT(handle
!= NULL
|| create
== 0);
1184 dummy
.b_blocknr
= -1000;
1185 buffer_trace_init(&dummy
.b_history
);
1186 err
= ext4_get_blocks_wrap(handle
, inode
, block
, 1,
1187 &dummy
, create
, 1, 0);
1189 * ext4_get_blocks_handle() returns number of blocks
1190 * mapped. 0 in case of a HOLE.
1198 if (!err
&& buffer_mapped(&dummy
)) {
1199 struct buffer_head
*bh
;
1200 bh
= sb_getblk(inode
->i_sb
, dummy
.b_blocknr
);
1205 if (buffer_new(&dummy
)) {
1206 J_ASSERT(create
!= 0);
1207 J_ASSERT(handle
!= NULL
);
1210 * Now that we do not always journal data, we should
1211 * keep in mind whether this should always journal the
1212 * new buffer as metadata. For now, regular file
1213 * writes use ext4_get_block instead, so it's not a
1217 BUFFER_TRACE(bh
, "call get_create_access");
1218 fatal
= ext4_journal_get_create_access(handle
, bh
);
1219 if (!fatal
&& !buffer_uptodate(bh
)) {
1220 memset(bh
->b_data
, 0, inode
->i_sb
->s_blocksize
);
1221 set_buffer_uptodate(bh
);
1224 BUFFER_TRACE(bh
, "call ext4_journal_dirty_metadata");
1225 err
= ext4_journal_dirty_metadata(handle
, bh
);
1229 BUFFER_TRACE(bh
, "not a new buffer");
1242 struct buffer_head
*ext4_bread(handle_t
*handle
, struct inode
*inode
,
1243 ext4_lblk_t block
, int create
, int *err
)
1245 struct buffer_head
*bh
;
1247 bh
= ext4_getblk(handle
, inode
, block
, create
, err
);
1250 if (buffer_uptodate(bh
))
1252 ll_rw_block(READ_META
, 1, &bh
);
1254 if (buffer_uptodate(bh
))
1261 static int walk_page_buffers(handle_t
*handle
,
1262 struct buffer_head
*head
,
1266 int (*fn
)(handle_t
*handle
,
1267 struct buffer_head
*bh
))
1269 struct buffer_head
*bh
;
1270 unsigned block_start
, block_end
;
1271 unsigned blocksize
= head
->b_size
;
1273 struct buffer_head
*next
;
1275 for (bh
= head
, block_start
= 0;
1276 ret
== 0 && (bh
!= head
|| !block_start
);
1277 block_start
= block_end
, bh
= next
)
1279 next
= bh
->b_this_page
;
1280 block_end
= block_start
+ blocksize
;
1281 if (block_end
<= from
|| block_start
>= to
) {
1282 if (partial
&& !buffer_uptodate(bh
))
1286 err
= (*fn
)(handle
, bh
);
1294 * To preserve ordering, it is essential that the hole instantiation and
1295 * the data write be encapsulated in a single transaction. We cannot
1296 * close off a transaction and start a new one between the ext4_get_block()
1297 * and the commit_write(). So doing the jbd2_journal_start at the start of
1298 * prepare_write() is the right place.
1300 * Also, this function can nest inside ext4_writepage() ->
1301 * block_write_full_page(). In that case, we *know* that ext4_writepage()
1302 * has generated enough buffer credits to do the whole page. So we won't
1303 * block on the journal in that case, which is good, because the caller may
1306 * By accident, ext4 can be reentered when a transaction is open via
1307 * quota file writes. If we were to commit the transaction while thus
1308 * reentered, there can be a deadlock - we would be holding a quota
1309 * lock, and the commit would never complete if another thread had a
1310 * transaction open and was blocking on the quota lock - a ranking
1313 * So what we do is to rely on the fact that jbd2_journal_stop/journal_start
1314 * will _not_ run commit under these circumstances because handle->h_ref
1315 * is elevated. We'll still have enough credits for the tiny quotafile
1318 static int do_journal_get_write_access(handle_t
*handle
,
1319 struct buffer_head
*bh
)
1321 if (!buffer_mapped(bh
) || buffer_freed(bh
))
1323 return ext4_journal_get_write_access(handle
, bh
);
1326 static int ext4_write_begin(struct file
*file
, struct address_space
*mapping
,
1327 loff_t pos
, unsigned len
, unsigned flags
,
1328 struct page
**pagep
, void **fsdata
)
1330 struct inode
*inode
= mapping
->host
;
1331 int ret
, needed_blocks
= ext4_writepage_trans_blocks(inode
);
1338 index
= pos
>> PAGE_CACHE_SHIFT
;
1339 from
= pos
& (PAGE_CACHE_SIZE
- 1);
1343 handle
= ext4_journal_start(inode
, needed_blocks
);
1344 if (IS_ERR(handle
)) {
1345 ret
= PTR_ERR(handle
);
1349 page
= __grab_cache_page(mapping
, index
);
1351 ext4_journal_stop(handle
);
1357 ret
= block_write_begin(file
, mapping
, pos
, len
, flags
, pagep
, fsdata
,
1360 if (!ret
&& ext4_should_journal_data(inode
)) {
1361 ret
= walk_page_buffers(handle
, page_buffers(page
),
1362 from
, to
, NULL
, do_journal_get_write_access
);
1367 ext4_journal_stop(handle
);
1368 page_cache_release(page
);
1370 * block_write_begin may have instantiated a few blocks
1371 * outside i_size. Trim these off again. Don't need
1372 * i_size_read because we hold i_mutex.
1374 if (pos
+ len
> inode
->i_size
)
1375 vmtruncate(inode
, inode
->i_size
);
1378 if (ret
== -ENOSPC
&& ext4_should_retry_alloc(inode
->i_sb
, &retries
))
1384 /* For write_end() in data=journal mode */
1385 static int write_end_fn(handle_t
*handle
, struct buffer_head
*bh
)
1387 if (!buffer_mapped(bh
) || buffer_freed(bh
))
1389 set_buffer_uptodate(bh
);
1390 return ext4_journal_dirty_metadata(handle
, bh
);
1394 * We need to pick up the new inode size which generic_commit_write gave us
1395 * `file' can be NULL - eg, when called from page_symlink().
1397 * ext4 never places buffers on inode->i_mapping->private_list. metadata
1398 * buffers are managed internally.
1400 static int ext4_ordered_write_end(struct file
*file
,
1401 struct address_space
*mapping
,
1402 loff_t pos
, unsigned len
, unsigned copied
,
1403 struct page
*page
, void *fsdata
)
1405 handle_t
*handle
= ext4_journal_current_handle();
1406 struct inode
*inode
= mapping
->host
;
1409 ret
= ext4_jbd2_file_inode(handle
, inode
);
1414 new_i_size
= pos
+ copied
;
1415 if (new_i_size
> EXT4_I(inode
)->i_disksize
) {
1416 ext4_update_i_disksize(inode
, new_i_size
);
1417 /* We need to mark inode dirty even if
1418 * new_i_size is less that inode->i_size
1419 * bu greater than i_disksize.(hint delalloc)
1421 ext4_mark_inode_dirty(handle
, inode
);
1424 ret2
= generic_write_end(file
, mapping
, pos
, len
, copied
,
1430 ret2
= ext4_journal_stop(handle
);
1434 return ret
? ret
: copied
;
1437 static int ext4_writeback_write_end(struct file
*file
,
1438 struct address_space
*mapping
,
1439 loff_t pos
, unsigned len
, unsigned copied
,
1440 struct page
*page
, void *fsdata
)
1442 handle_t
*handle
= ext4_journal_current_handle();
1443 struct inode
*inode
= mapping
->host
;
1447 new_i_size
= pos
+ copied
;
1448 if (new_i_size
> EXT4_I(inode
)->i_disksize
) {
1449 ext4_update_i_disksize(inode
, new_i_size
);
1450 /* We need to mark inode dirty even if
1451 * new_i_size is less that inode->i_size
1452 * bu greater than i_disksize.(hint delalloc)
1454 ext4_mark_inode_dirty(handle
, inode
);
1457 ret2
= generic_write_end(file
, mapping
, pos
, len
, copied
,
1463 ret2
= ext4_journal_stop(handle
);
1467 return ret
? ret
: copied
;
1470 static int ext4_journalled_write_end(struct file
*file
,
1471 struct address_space
*mapping
,
1472 loff_t pos
, unsigned len
, unsigned copied
,
1473 struct page
*page
, void *fsdata
)
1475 handle_t
*handle
= ext4_journal_current_handle();
1476 struct inode
*inode
= mapping
->host
;
1482 from
= pos
& (PAGE_CACHE_SIZE
- 1);
1486 if (!PageUptodate(page
))
1488 page_zero_new_buffers(page
, from
+copied
, to
);
1491 ret
= walk_page_buffers(handle
, page_buffers(page
), from
,
1492 to
, &partial
, write_end_fn
);
1494 SetPageUptodate(page
);
1495 new_i_size
= pos
+ copied
;
1496 if (new_i_size
> inode
->i_size
)
1497 i_size_write(inode
, pos
+copied
);
1498 EXT4_I(inode
)->i_state
|= EXT4_STATE_JDATA
;
1499 if (new_i_size
> EXT4_I(inode
)->i_disksize
) {
1500 ext4_update_i_disksize(inode
, new_i_size
);
1501 ret2
= ext4_mark_inode_dirty(handle
, inode
);
1507 ret2
= ext4_journal_stop(handle
);
1510 page_cache_release(page
);
1512 return ret
? ret
: copied
;
1515 static int ext4_da_reserve_space(struct inode
*inode
, int nrblocks
)
1518 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
1519 unsigned long md_needed
, mdblocks
, total
= 0;
1522 * recalculate the amount of metadata blocks to reserve
1523 * in order to allocate nrblocks
1524 * worse case is one extent per block
1527 spin_lock(&EXT4_I(inode
)->i_block_reservation_lock
);
1528 total
= EXT4_I(inode
)->i_reserved_data_blocks
+ nrblocks
;
1529 mdblocks
= ext4_calc_metadata_amount(inode
, total
);
1530 BUG_ON(mdblocks
< EXT4_I(inode
)->i_reserved_meta_blocks
);
1532 md_needed
= mdblocks
- EXT4_I(inode
)->i_reserved_meta_blocks
;
1533 total
= md_needed
+ nrblocks
;
1535 if (ext4_claim_free_blocks(sbi
, total
)) {
1536 spin_unlock(&EXT4_I(inode
)->i_block_reservation_lock
);
1537 if (ext4_should_retry_alloc(inode
->i_sb
, &retries
)) {
1543 EXT4_I(inode
)->i_reserved_data_blocks
+= nrblocks
;
1544 EXT4_I(inode
)->i_reserved_meta_blocks
= mdblocks
;
1546 spin_unlock(&EXT4_I(inode
)->i_block_reservation_lock
);
1547 return 0; /* success */
1550 static void ext4_da_release_space(struct inode
*inode
, int to_free
)
1552 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
1553 int total
, mdb
, mdb_free
, release
;
1556 return; /* Nothing to release, exit */
1558 spin_lock(&EXT4_I(inode
)->i_block_reservation_lock
);
1560 if (!EXT4_I(inode
)->i_reserved_data_blocks
) {
1562 * if there is no reserved blocks, but we try to free some
1563 * then the counter is messed up somewhere.
1564 * but since this function is called from invalidate
1565 * page, it's harmless to return without any action
1567 printk(KERN_INFO
"ext4 delalloc try to release %d reserved "
1568 "blocks for inode %lu, but there is no reserved "
1569 "data blocks\n", to_free
, inode
->i_ino
);
1570 spin_unlock(&EXT4_I(inode
)->i_block_reservation_lock
);
1574 /* recalculate the number of metablocks still need to be reserved */
1575 total
= EXT4_I(inode
)->i_reserved_data_blocks
- to_free
;
1576 mdb
= ext4_calc_metadata_amount(inode
, total
);
1578 /* figure out how many metablocks to release */
1579 BUG_ON(mdb
> EXT4_I(inode
)->i_reserved_meta_blocks
);
1580 mdb_free
= EXT4_I(inode
)->i_reserved_meta_blocks
- mdb
;
1582 release
= to_free
+ mdb_free
;
1584 /* update fs dirty blocks counter for truncate case */
1585 percpu_counter_sub(&sbi
->s_dirtyblocks_counter
, release
);
1587 /* update per-inode reservations */
1588 BUG_ON(to_free
> EXT4_I(inode
)->i_reserved_data_blocks
);
1589 EXT4_I(inode
)->i_reserved_data_blocks
-= to_free
;
1591 BUG_ON(mdb
> EXT4_I(inode
)->i_reserved_meta_blocks
);
1592 EXT4_I(inode
)->i_reserved_meta_blocks
= mdb
;
1593 spin_unlock(&EXT4_I(inode
)->i_block_reservation_lock
);
1596 static void ext4_da_page_release_reservation(struct page
*page
,
1597 unsigned long offset
)
1600 struct buffer_head
*head
, *bh
;
1601 unsigned int curr_off
= 0;
1603 head
= page_buffers(page
);
1606 unsigned int next_off
= curr_off
+ bh
->b_size
;
1608 if ((offset
<= curr_off
) && (buffer_delay(bh
))) {
1610 clear_buffer_delay(bh
);
1612 curr_off
= next_off
;
1613 } while ((bh
= bh
->b_this_page
) != head
);
1614 ext4_da_release_space(page
->mapping
->host
, to_release
);
1618 * Delayed allocation stuff
1621 struct mpage_da_data
{
1622 struct inode
*inode
;
1623 struct buffer_head lbh
; /* extent of blocks */
1624 unsigned long first_page
, next_page
; /* extent of pages */
1625 get_block_t
*get_block
;
1626 struct writeback_control
*wbc
;
1633 * mpage_da_submit_io - walks through extent of pages and try to write
1634 * them with writepage() call back
1636 * @mpd->inode: inode
1637 * @mpd->first_page: first page of the extent
1638 * @mpd->next_page: page after the last page of the extent
1639 * @mpd->get_block: the filesystem's block mapper function
1641 * By the time mpage_da_submit_io() is called we expect all blocks
1642 * to be allocated. this may be wrong if allocation failed.
1644 * As pages are already locked by write_cache_pages(), we can't use it
1646 static int mpage_da_submit_io(struct mpage_da_data
*mpd
)
1648 struct address_space
*mapping
= mpd
->inode
->i_mapping
;
1649 int ret
= 0, err
, nr_pages
, i
;
1650 unsigned long index
, end
;
1651 struct pagevec pvec
;
1654 BUG_ON(mpd
->next_page
<= mpd
->first_page
);
1655 pagevec_init(&pvec
, 0);
1656 index
= mpd
->first_page
;
1657 end
= mpd
->next_page
- 1;
1659 while (index
<= end
) {
1661 * We can use PAGECACHE_TAG_DIRTY lookup here because
1662 * even though we have cleared the dirty flag on the page
1663 * We still keep the page in the radix tree with tag
1664 * PAGECACHE_TAG_DIRTY. See clear_page_dirty_for_io.
1665 * The PAGECACHE_TAG_DIRTY is cleared in set_page_writeback
1666 * which is called via the below writepage callback.
1668 nr_pages
= pagevec_lookup_tag(&pvec
, mapping
, &index
,
1669 PAGECACHE_TAG_DIRTY
,
1671 (pgoff_t
)PAGEVEC_SIZE
-1) + 1);
1674 for (i
= 0; i
< nr_pages
; i
++) {
1675 struct page
*page
= pvec
.pages
[i
];
1677 pages_skipped
= mpd
->wbc
->pages_skipped
;
1678 err
= mapping
->a_ops
->writepage(page
, mpd
->wbc
);
1679 if (!err
&& (pages_skipped
== mpd
->wbc
->pages_skipped
))
1681 * have successfully written the page
1682 * without skipping the same
1684 mpd
->pages_written
++;
1686 * In error case, we have to continue because
1687 * remaining pages are still locked
1688 * XXX: unlock and re-dirty them?
1693 pagevec_release(&pvec
);
1699 * mpage_put_bnr_to_bhs - walk blocks and assign them actual numbers
1701 * @mpd->inode - inode to walk through
1702 * @exbh->b_blocknr - first block on a disk
1703 * @exbh->b_size - amount of space in bytes
1704 * @logical - first logical block to start assignment with
1706 * the function goes through all passed space and put actual disk
1707 * block numbers into buffer heads, dropping BH_Delay
1709 static void mpage_put_bnr_to_bhs(struct mpage_da_data
*mpd
, sector_t logical
,
1710 struct buffer_head
*exbh
)
1712 struct inode
*inode
= mpd
->inode
;
1713 struct address_space
*mapping
= inode
->i_mapping
;
1714 int blocks
= exbh
->b_size
>> inode
->i_blkbits
;
1715 sector_t pblock
= exbh
->b_blocknr
, cur_logical
;
1716 struct buffer_head
*head
, *bh
;
1718 struct pagevec pvec
;
1721 index
= logical
>> (PAGE_CACHE_SHIFT
- inode
->i_blkbits
);
1722 end
= (logical
+ blocks
- 1) >> (PAGE_CACHE_SHIFT
- inode
->i_blkbits
);
1723 cur_logical
= index
<< (PAGE_CACHE_SHIFT
- inode
->i_blkbits
);
1725 pagevec_init(&pvec
, 0);
1727 while (index
<= end
) {
1728 /* XXX: optimize tail */
1729 nr_pages
= pagevec_lookup(&pvec
, mapping
, index
, PAGEVEC_SIZE
);
1732 for (i
= 0; i
< nr_pages
; i
++) {
1733 struct page
*page
= pvec
.pages
[i
];
1735 index
= page
->index
;
1740 BUG_ON(!PageLocked(page
));
1741 BUG_ON(PageWriteback(page
));
1742 BUG_ON(!page_has_buffers(page
));
1744 bh
= page_buffers(page
);
1747 /* skip blocks out of the range */
1749 if (cur_logical
>= logical
)
1752 } while ((bh
= bh
->b_this_page
) != head
);
1755 if (cur_logical
>= logical
+ blocks
)
1757 if (buffer_delay(bh
)) {
1758 bh
->b_blocknr
= pblock
;
1759 clear_buffer_delay(bh
);
1760 bh
->b_bdev
= inode
->i_sb
->s_bdev
;
1761 } else if (buffer_unwritten(bh
)) {
1762 bh
->b_blocknr
= pblock
;
1763 clear_buffer_unwritten(bh
);
1764 set_buffer_mapped(bh
);
1766 bh
->b_bdev
= inode
->i_sb
->s_bdev
;
1767 } else if (buffer_mapped(bh
))
1768 BUG_ON(bh
->b_blocknr
!= pblock
);
1772 } while ((bh
= bh
->b_this_page
) != head
);
1774 pagevec_release(&pvec
);
1780 * __unmap_underlying_blocks - just a helper function to unmap
1781 * set of blocks described by @bh
1783 static inline void __unmap_underlying_blocks(struct inode
*inode
,
1784 struct buffer_head
*bh
)
1786 struct block_device
*bdev
= inode
->i_sb
->s_bdev
;
1789 blocks
= bh
->b_size
>> inode
->i_blkbits
;
1790 for (i
= 0; i
< blocks
; i
++)
1791 unmap_underlying_metadata(bdev
, bh
->b_blocknr
+ i
);
1794 static void ext4_da_block_invalidatepages(struct mpage_da_data
*mpd
,
1795 sector_t logical
, long blk_cnt
)
1799 struct pagevec pvec
;
1800 struct inode
*inode
= mpd
->inode
;
1801 struct address_space
*mapping
= inode
->i_mapping
;
1803 index
= logical
>> (PAGE_CACHE_SHIFT
- inode
->i_blkbits
);
1804 end
= (logical
+ blk_cnt
- 1) >>
1805 (PAGE_CACHE_SHIFT
- inode
->i_blkbits
);
1806 while (index
<= end
) {
1807 nr_pages
= pagevec_lookup(&pvec
, mapping
, index
, PAGEVEC_SIZE
);
1810 for (i
= 0; i
< nr_pages
; i
++) {
1811 struct page
*page
= pvec
.pages
[i
];
1812 index
= page
->index
;
1817 BUG_ON(!PageLocked(page
));
1818 BUG_ON(PageWriteback(page
));
1819 block_invalidatepage(page
, 0);
1820 ClearPageUptodate(page
);
1827 static void ext4_print_free_blocks(struct inode
*inode
)
1829 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
1830 printk(KERN_EMERG
"Total free blocks count %lld\n",
1831 ext4_count_free_blocks(inode
->i_sb
));
1832 printk(KERN_EMERG
"Free/Dirty block details\n");
1833 printk(KERN_EMERG
"free_blocks=%lld\n",
1834 percpu_counter_sum(&sbi
->s_freeblocks_counter
));
1835 printk(KERN_EMERG
"dirty_blocks=%lld\n",
1836 percpu_counter_sum(&sbi
->s_dirtyblocks_counter
));
1837 printk(KERN_EMERG
"Block reservation details\n");
1838 printk(KERN_EMERG
"i_reserved_data_blocks=%lu\n",
1839 EXT4_I(inode
)->i_reserved_data_blocks
);
1840 printk(KERN_EMERG
"i_reserved_meta_blocks=%lu\n",
1841 EXT4_I(inode
)->i_reserved_meta_blocks
);
1846 * mpage_da_map_blocks - go through given space
1848 * @mpd->lbh - bh describing space
1849 * @mpd->get_block - the filesystem's block mapper function
1851 * The function skips space we know is already mapped to disk blocks.
1854 static int mpage_da_map_blocks(struct mpage_da_data
*mpd
)
1857 struct buffer_head
new;
1858 struct buffer_head
*lbh
= &mpd
->lbh
;
1862 * We consider only non-mapped and non-allocated blocks
1864 if (buffer_mapped(lbh
) && !buffer_delay(lbh
))
1866 new.b_state
= lbh
->b_state
;
1868 new.b_size
= lbh
->b_size
;
1869 next
= lbh
->b_blocknr
;
1871 * If we didn't accumulate anything
1872 * to write simply return
1876 err
= mpd
->get_block(mpd
->inode
, next
, &new, 1);
1879 /* If get block returns with error
1880 * we simply return. Later writepage
1881 * will redirty the page and writepages
1882 * will find the dirty page again
1887 if (err
== -ENOSPC
&&
1888 ext4_count_free_blocks(mpd
->inode
->i_sb
)) {
1894 * get block failure will cause us
1895 * to loop in writepages. Because
1896 * a_ops->writepage won't be able to
1897 * make progress. The page will be redirtied
1898 * by writepage and writepages will again
1899 * try to write the same.
1901 printk(KERN_EMERG
"%s block allocation failed for inode %lu "
1902 "at logical offset %llu with max blocks "
1903 "%zd with error %d\n",
1904 __func__
, mpd
->inode
->i_ino
,
1905 (unsigned long long)next
,
1906 lbh
->b_size
>> mpd
->inode
->i_blkbits
, err
);
1907 printk(KERN_EMERG
"This should not happen.!! "
1908 "Data will be lost\n");
1909 if (err
== -ENOSPC
) {
1910 ext4_print_free_blocks(mpd
->inode
);
1912 /* invlaidate all the pages */
1913 ext4_da_block_invalidatepages(mpd
, next
,
1914 lbh
->b_size
>> mpd
->inode
->i_blkbits
);
1917 BUG_ON(new.b_size
== 0);
1919 if (buffer_new(&new))
1920 __unmap_underlying_blocks(mpd
->inode
, &new);
1923 * If blocks are delayed marked, we need to
1924 * put actual blocknr and drop delayed bit
1926 if (buffer_delay(lbh
) || buffer_unwritten(lbh
))
1927 mpage_put_bnr_to_bhs(mpd
, next
, &new);
1932 #define BH_FLAGS ((1 << BH_Uptodate) | (1 << BH_Mapped) | \
1933 (1 << BH_Delay) | (1 << BH_Unwritten))
1936 * mpage_add_bh_to_extent - try to add one more block to extent of blocks
1938 * @mpd->lbh - extent of blocks
1939 * @logical - logical number of the block in the file
1940 * @bh - bh of the block (used to access block's state)
1942 * the function is used to collect contig. blocks in same state
1944 static void mpage_add_bh_to_extent(struct mpage_da_data
*mpd
,
1945 sector_t logical
, struct buffer_head
*bh
)
1948 size_t b_size
= bh
->b_size
;
1949 struct buffer_head
*lbh
= &mpd
->lbh
;
1950 int nrblocks
= lbh
->b_size
>> mpd
->inode
->i_blkbits
;
1952 /* check if thereserved journal credits might overflow */
1953 if (!(EXT4_I(mpd
->inode
)->i_flags
& EXT4_EXTENTS_FL
)) {
1954 if (nrblocks
>= EXT4_MAX_TRANS_DATA
) {
1956 * With non-extent format we are limited by the journal
1957 * credit available. Total credit needed to insert
1958 * nrblocks contiguous blocks is dependent on the
1959 * nrblocks. So limit nrblocks.
1962 } else if ((nrblocks
+ (b_size
>> mpd
->inode
->i_blkbits
)) >
1963 EXT4_MAX_TRANS_DATA
) {
1965 * Adding the new buffer_head would make it cross the
1966 * allowed limit for which we have journal credit
1967 * reserved. So limit the new bh->b_size
1969 b_size
= (EXT4_MAX_TRANS_DATA
- nrblocks
) <<
1970 mpd
->inode
->i_blkbits
;
1971 /* we will do mpage_da_submit_io in the next loop */
1975 * First block in the extent
1977 if (lbh
->b_size
== 0) {
1978 lbh
->b_blocknr
= logical
;
1979 lbh
->b_size
= b_size
;
1980 lbh
->b_state
= bh
->b_state
& BH_FLAGS
;
1984 next
= lbh
->b_blocknr
+ nrblocks
;
1986 * Can we merge the block to our big extent?
1988 if (logical
== next
&& (bh
->b_state
& BH_FLAGS
) == lbh
->b_state
) {
1989 lbh
->b_size
+= b_size
;
1995 * We couldn't merge the block to our extent, so we
1996 * need to flush current extent and start new one
1998 if (mpage_da_map_blocks(mpd
) == 0)
1999 mpage_da_submit_io(mpd
);
2005 * __mpage_da_writepage - finds extent of pages and blocks
2007 * @page: page to consider
2008 * @wbc: not used, we just follow rules
2011 * The function finds extents of pages and scan them for all blocks.
2013 static int __mpage_da_writepage(struct page
*page
,
2014 struct writeback_control
*wbc
, void *data
)
2016 struct mpage_da_data
*mpd
= data
;
2017 struct inode
*inode
= mpd
->inode
;
2018 struct buffer_head
*bh
, *head
, fake
;
2023 * Rest of the page in the page_vec
2024 * redirty then and skip then. We will
2025 * try to to write them again after
2026 * starting a new transaction
2028 redirty_page_for_writepage(wbc
, page
);
2030 return MPAGE_DA_EXTENT_TAIL
;
2033 * Can we merge this page to current extent?
2035 if (mpd
->next_page
!= page
->index
) {
2037 * Nope, we can't. So, we map non-allocated blocks
2038 * and start IO on them using writepage()
2040 if (mpd
->next_page
!= mpd
->first_page
) {
2041 if (mpage_da_map_blocks(mpd
) == 0)
2042 mpage_da_submit_io(mpd
);
2044 * skip rest of the page in the page_vec
2047 redirty_page_for_writepage(wbc
, page
);
2049 return MPAGE_DA_EXTENT_TAIL
;
2053 * Start next extent of pages ...
2055 mpd
->first_page
= page
->index
;
2060 mpd
->lbh
.b_size
= 0;
2061 mpd
->lbh
.b_state
= 0;
2062 mpd
->lbh
.b_blocknr
= 0;
2065 mpd
->next_page
= page
->index
+ 1;
2066 logical
= (sector_t
) page
->index
<<
2067 (PAGE_CACHE_SHIFT
- inode
->i_blkbits
);
2069 if (!page_has_buffers(page
)) {
2071 * There is no attached buffer heads yet (mmap?)
2072 * we treat the page asfull of dirty blocks
2075 bh
->b_size
= PAGE_CACHE_SIZE
;
2077 set_buffer_dirty(bh
);
2078 set_buffer_uptodate(bh
);
2079 mpage_add_bh_to_extent(mpd
, logical
, bh
);
2081 return MPAGE_DA_EXTENT_TAIL
;
2084 * Page with regular buffer heads, just add all dirty ones
2086 head
= page_buffers(page
);
2089 BUG_ON(buffer_locked(bh
));
2090 if (buffer_dirty(bh
) &&
2091 (!buffer_mapped(bh
) || buffer_delay(bh
))) {
2092 mpage_add_bh_to_extent(mpd
, logical
, bh
);
2094 return MPAGE_DA_EXTENT_TAIL
;
2097 } while ((bh
= bh
->b_this_page
) != head
);
2104 * mpage_da_writepages - walk the list of dirty pages of the given
2105 * address space, allocates non-allocated blocks, maps newly-allocated
2106 * blocks to existing bhs and issue IO them
2108 * @mapping: address space structure to write
2109 * @wbc: subtract the number of written pages from *@wbc->nr_to_write
2110 * @get_block: the filesystem's block mapper function.
2112 * This is a library function, which implements the writepages()
2113 * address_space_operation.
2115 static int mpage_da_writepages(struct address_space
*mapping
,
2116 struct writeback_control
*wbc
,
2117 struct mpage_da_data
*mpd
)
2121 if (!mpd
->get_block
)
2122 return generic_writepages(mapping
, wbc
);
2124 mpd
->lbh
.b_size
= 0;
2125 mpd
->lbh
.b_state
= 0;
2126 mpd
->lbh
.b_blocknr
= 0;
2127 mpd
->first_page
= 0;
2130 mpd
->pages_written
= 0;
2133 ret
= write_cache_pages(mapping
, wbc
, __mpage_da_writepage
, mpd
);
2135 * Handle last extent of pages
2137 if (!mpd
->io_done
&& mpd
->next_page
!= mpd
->first_page
) {
2138 if (mpage_da_map_blocks(mpd
) == 0)
2139 mpage_da_submit_io(mpd
);
2142 ret
= MPAGE_DA_EXTENT_TAIL
;
2144 wbc
->nr_to_write
-= mpd
->pages_written
;
2149 * this is a special callback for ->write_begin() only
2150 * it's intention is to return mapped block or reserve space
2152 static int ext4_da_get_block_prep(struct inode
*inode
, sector_t iblock
,
2153 struct buffer_head
*bh_result
, int create
)
2157 BUG_ON(create
== 0);
2158 BUG_ON(bh_result
->b_size
!= inode
->i_sb
->s_blocksize
);
2161 * first, we need to know whether the block is allocated already
2162 * preallocated blocks are unmapped but should treated
2163 * the same as allocated blocks.
2165 ret
= ext4_get_blocks_wrap(NULL
, inode
, iblock
, 1, bh_result
, 0, 0, 0);
2166 if ((ret
== 0) && !buffer_delay(bh_result
)) {
2167 /* the block isn't (pre)allocated yet, let's reserve space */
2169 * XXX: __block_prepare_write() unmaps passed block,
2172 ret
= ext4_da_reserve_space(inode
, 1);
2174 /* not enough space to reserve */
2177 map_bh(bh_result
, inode
->i_sb
, 0);
2178 set_buffer_new(bh_result
);
2179 set_buffer_delay(bh_result
);
2180 } else if (ret
> 0) {
2181 bh_result
->b_size
= (ret
<< inode
->i_blkbits
);
2187 #define EXT4_DELALLOC_RSVED 1
2188 static int ext4_da_get_block_write(struct inode
*inode
, sector_t iblock
,
2189 struct buffer_head
*bh_result
, int create
)
2192 unsigned max_blocks
= bh_result
->b_size
>> inode
->i_blkbits
;
2193 loff_t disksize
= EXT4_I(inode
)->i_disksize
;
2194 handle_t
*handle
= NULL
;
2196 handle
= ext4_journal_current_handle();
2198 ret
= ext4_get_blocks_wrap(handle
, inode
, iblock
, max_blocks
,
2199 bh_result
, create
, 0, EXT4_DELALLOC_RSVED
);
2202 bh_result
->b_size
= (ret
<< inode
->i_blkbits
);
2204 if (ext4_should_order_data(inode
)) {
2206 retval
= ext4_jbd2_file_inode(handle
, inode
);
2209 * Failed to add inode for ordered
2210 * mode. Don't update file size
2216 * Update on-disk size along with block allocation
2217 * we don't use 'extend_disksize' as size may change
2218 * within already allocated block -bzzz
2220 disksize
= ((loff_t
) iblock
+ ret
) << inode
->i_blkbits
;
2221 if (disksize
> i_size_read(inode
))
2222 disksize
= i_size_read(inode
);
2223 if (disksize
> EXT4_I(inode
)->i_disksize
) {
2224 ext4_update_i_disksize(inode
, disksize
);
2225 ret
= ext4_mark_inode_dirty(handle
, inode
);
2233 static int ext4_bh_unmapped_or_delay(handle_t
*handle
, struct buffer_head
*bh
)
2236 * unmapped buffer is possible for holes.
2237 * delay buffer is possible with delayed allocation
2239 return ((!buffer_mapped(bh
) || buffer_delay(bh
)) && buffer_dirty(bh
));
2242 static int ext4_normal_get_block_write(struct inode
*inode
, sector_t iblock
,
2243 struct buffer_head
*bh_result
, int create
)
2246 unsigned max_blocks
= bh_result
->b_size
>> inode
->i_blkbits
;
2249 * we don't want to do block allocation in writepage
2250 * so call get_block_wrap with create = 0
2252 ret
= ext4_get_blocks_wrap(NULL
, inode
, iblock
, max_blocks
,
2253 bh_result
, 0, 0, 0);
2255 bh_result
->b_size
= (ret
<< inode
->i_blkbits
);
2262 * get called vi ext4_da_writepages after taking page lock (have journal handle)
2263 * get called via journal_submit_inode_data_buffers (no journal handle)
2264 * get called via shrink_page_list via pdflush (no journal handle)
2265 * or grab_page_cache when doing write_begin (have journal handle)
2267 static int ext4_da_writepage(struct page
*page
,
2268 struct writeback_control
*wbc
)
2273 struct buffer_head
*page_bufs
;
2274 struct inode
*inode
= page
->mapping
->host
;
2276 size
= i_size_read(inode
);
2277 if (page
->index
== size
>> PAGE_CACHE_SHIFT
)
2278 len
= size
& ~PAGE_CACHE_MASK
;
2280 len
= PAGE_CACHE_SIZE
;
2282 if (page_has_buffers(page
)) {
2283 page_bufs
= page_buffers(page
);
2284 if (walk_page_buffers(NULL
, page_bufs
, 0, len
, NULL
,
2285 ext4_bh_unmapped_or_delay
)) {
2287 * We don't want to do block allocation
2288 * So redirty the page and return
2289 * We may reach here when we do a journal commit
2290 * via journal_submit_inode_data_buffers.
2291 * If we don't have mapping block we just ignore
2292 * them. We can also reach here via shrink_page_list
2294 redirty_page_for_writepage(wbc
, page
);
2300 * The test for page_has_buffers() is subtle:
2301 * We know the page is dirty but it lost buffers. That means
2302 * that at some moment in time after write_begin()/write_end()
2303 * has been called all buffers have been clean and thus they
2304 * must have been written at least once. So they are all
2305 * mapped and we can happily proceed with mapping them
2306 * and writing the page.
2308 * Try to initialize the buffer_heads and check whether
2309 * all are mapped and non delay. We don't want to
2310 * do block allocation here.
2312 ret
= block_prepare_write(page
, 0, PAGE_CACHE_SIZE
,
2313 ext4_normal_get_block_write
);
2315 page_bufs
= page_buffers(page
);
2316 /* check whether all are mapped and non delay */
2317 if (walk_page_buffers(NULL
, page_bufs
, 0, len
, NULL
,
2318 ext4_bh_unmapped_or_delay
)) {
2319 redirty_page_for_writepage(wbc
, page
);
2325 * We can't do block allocation here
2326 * so just redity the page and unlock
2329 redirty_page_for_writepage(wbc
, page
);
2333 /* now mark the buffer_heads as dirty and uptodate */
2334 block_commit_write(page
, 0, PAGE_CACHE_SIZE
);
2337 if (test_opt(inode
->i_sb
, NOBH
) && ext4_should_writeback_data(inode
))
2338 ret
= nobh_writepage(page
, ext4_normal_get_block_write
, wbc
);
2340 ret
= block_write_full_page(page
,
2341 ext4_normal_get_block_write
,
2348 * This is called via ext4_da_writepages() to
2349 * calulate the total number of credits to reserve to fit
2350 * a single extent allocation into a single transaction,
2351 * ext4_da_writpeages() will loop calling this before
2352 * the block allocation.
2355 static int ext4_da_writepages_trans_blocks(struct inode
*inode
)
2357 int max_blocks
= EXT4_I(inode
)->i_reserved_data_blocks
;
2360 * With non-extent format the journal credit needed to
2361 * insert nrblocks contiguous block is dependent on
2362 * number of contiguous block. So we will limit
2363 * number of contiguous block to a sane value
2365 if (!(inode
->i_flags
& EXT4_EXTENTS_FL
) &&
2366 (max_blocks
> EXT4_MAX_TRANS_DATA
))
2367 max_blocks
= EXT4_MAX_TRANS_DATA
;
2369 return ext4_chunk_trans_blocks(inode
, max_blocks
);
2372 static int ext4_da_writepages(struct address_space
*mapping
,
2373 struct writeback_control
*wbc
)
2376 int range_whole
= 0;
2377 handle_t
*handle
= NULL
;
2378 struct mpage_da_data mpd
;
2379 struct inode
*inode
= mapping
->host
;
2380 int no_nrwrite_index_update
;
2381 long pages_written
= 0, pages_skipped
;
2382 int needed_blocks
, ret
= 0, nr_to_writebump
= 0;
2383 struct ext4_sb_info
*sbi
= EXT4_SB(mapping
->host
->i_sb
);
2386 * No pages to write? This is mainly a kludge to avoid starting
2387 * a transaction for special inodes like journal inode on last iput()
2388 * because that could violate lock ordering on umount
2390 if (!mapping
->nrpages
|| !mapping_tagged(mapping
, PAGECACHE_TAG_DIRTY
))
2393 * Make sure nr_to_write is >= sbi->s_mb_stream_request
2394 * This make sure small files blocks are allocated in
2395 * single attempt. This ensure that small files
2396 * get less fragmented.
2398 if (wbc
->nr_to_write
< sbi
->s_mb_stream_request
) {
2399 nr_to_writebump
= sbi
->s_mb_stream_request
- wbc
->nr_to_write
;
2400 wbc
->nr_to_write
= sbi
->s_mb_stream_request
;
2402 if (wbc
->range_start
== 0 && wbc
->range_end
== LLONG_MAX
)
2405 if (wbc
->range_cyclic
)
2406 index
= mapping
->writeback_index
;
2408 index
= wbc
->range_start
>> PAGE_CACHE_SHIFT
;
2411 mpd
.inode
= mapping
->host
;
2414 * we don't want write_cache_pages to update
2415 * nr_to_write and writeback_index
2417 no_nrwrite_index_update
= wbc
->no_nrwrite_index_update
;
2418 wbc
->no_nrwrite_index_update
= 1;
2419 pages_skipped
= wbc
->pages_skipped
;
2421 while (!ret
&& wbc
->nr_to_write
> 0) {
2424 * we insert one extent at a time. So we need
2425 * credit needed for single extent allocation.
2426 * journalled mode is currently not supported
2429 BUG_ON(ext4_should_journal_data(inode
));
2430 needed_blocks
= ext4_da_writepages_trans_blocks(inode
);
2432 /* start a new transaction*/
2433 handle
= ext4_journal_start(inode
, needed_blocks
);
2434 if (IS_ERR(handle
)) {
2435 ret
= PTR_ERR(handle
);
2436 printk(KERN_EMERG
"%s: jbd2_start: "
2437 "%ld pages, ino %lu; err %d\n", __func__
,
2438 wbc
->nr_to_write
, inode
->i_ino
, ret
);
2440 goto out_writepages
;
2442 mpd
.get_block
= ext4_da_get_block_write
;
2443 ret
= mpage_da_writepages(mapping
, wbc
, &mpd
);
2445 ext4_journal_stop(handle
);
2447 if (mpd
.retval
== -ENOSPC
) {
2448 /* commit the transaction which would
2449 * free blocks released in the transaction
2452 jbd2_journal_force_commit_nested(sbi
->s_journal
);
2453 wbc
->pages_skipped
= pages_skipped
;
2455 } else if (ret
== MPAGE_DA_EXTENT_TAIL
) {
2457 * got one extent now try with
2460 pages_written
+= mpd
.pages_written
;
2461 wbc
->pages_skipped
= pages_skipped
;
2463 } else if (wbc
->nr_to_write
)
2465 * There is no more writeout needed
2466 * or we requested for a noblocking writeout
2467 * and we found the device congested
2471 if (pages_skipped
!= wbc
->pages_skipped
)
2472 printk(KERN_EMERG
"This should not happen leaving %s "
2473 "with nr_to_write = %ld ret = %d\n",
2474 __func__
, wbc
->nr_to_write
, ret
);
2477 index
+= pages_written
;
2478 if (wbc
->range_cyclic
|| (range_whole
&& wbc
->nr_to_write
> 0))
2480 * set the writeback_index so that range_cyclic
2481 * mode will write it back later
2483 mapping
->writeback_index
= index
;
2486 if (!no_nrwrite_index_update
)
2487 wbc
->no_nrwrite_index_update
= 0;
2488 wbc
->nr_to_write
-= nr_to_writebump
;
2492 #define FALL_BACK_TO_NONDELALLOC 1
2493 static int ext4_nonda_switch(struct super_block
*sb
)
2495 s64 free_blocks
, dirty_blocks
;
2496 struct ext4_sb_info
*sbi
= EXT4_SB(sb
);
2499 * switch to non delalloc mode if we are running low
2500 * on free block. The free block accounting via percpu
2501 * counters can get slightly wrong with FBC_BATCH getting
2502 * accumulated on each CPU without updating global counters
2503 * Delalloc need an accurate free block accounting. So switch
2504 * to non delalloc when we are near to error range.
2506 free_blocks
= percpu_counter_read_positive(&sbi
->s_freeblocks_counter
);
2507 dirty_blocks
= percpu_counter_read_positive(&sbi
->s_dirtyblocks_counter
);
2508 if (2 * free_blocks
< 3 * dirty_blocks
||
2509 free_blocks
< (dirty_blocks
+ EXT4_FREEBLOCKS_WATERMARK
)) {
2511 * free block count is less that 150% of dirty blocks
2512 * or free blocks is less that watermark
2519 static int ext4_da_write_begin(struct file
*file
, struct address_space
*mapping
,
2520 loff_t pos
, unsigned len
, unsigned flags
,
2521 struct page
**pagep
, void **fsdata
)
2523 int ret
, retries
= 0;
2527 struct inode
*inode
= mapping
->host
;
2530 index
= pos
>> PAGE_CACHE_SHIFT
;
2531 from
= pos
& (PAGE_CACHE_SIZE
- 1);
2534 if (ext4_nonda_switch(inode
->i_sb
)) {
2535 *fsdata
= (void *)FALL_BACK_TO_NONDELALLOC
;
2536 return ext4_write_begin(file
, mapping
, pos
,
2537 len
, flags
, pagep
, fsdata
);
2539 *fsdata
= (void *)0;
2542 * With delayed allocation, we don't log the i_disksize update
2543 * if there is delayed block allocation. But we still need
2544 * to journalling the i_disksize update if writes to the end
2545 * of file which has an already mapped buffer.
2547 handle
= ext4_journal_start(inode
, 1);
2548 if (IS_ERR(handle
)) {
2549 ret
= PTR_ERR(handle
);
2553 page
= __grab_cache_page(mapping
, index
);
2555 ext4_journal_stop(handle
);
2561 ret
= block_write_begin(file
, mapping
, pos
, len
, flags
, pagep
, fsdata
,
2562 ext4_da_get_block_prep
);
2565 ext4_journal_stop(handle
);
2566 page_cache_release(page
);
2568 * block_write_begin may have instantiated a few blocks
2569 * outside i_size. Trim these off again. Don't need
2570 * i_size_read because we hold i_mutex.
2572 if (pos
+ len
> inode
->i_size
)
2573 vmtruncate(inode
, inode
->i_size
);
2576 if (ret
== -ENOSPC
&& ext4_should_retry_alloc(inode
->i_sb
, &retries
))
2583 * Check if we should update i_disksize
2584 * when write to the end of file but not require block allocation
2586 static int ext4_da_should_update_i_disksize(struct page
*page
,
2587 unsigned long offset
)
2589 struct buffer_head
*bh
;
2590 struct inode
*inode
= page
->mapping
->host
;
2594 bh
= page_buffers(page
);
2595 idx
= offset
>> inode
->i_blkbits
;
2597 for (i
= 0; i
< idx
; i
++)
2598 bh
= bh
->b_this_page
;
2600 if (!buffer_mapped(bh
) || (buffer_delay(bh
)))
2605 static int ext4_da_write_end(struct file
*file
,
2606 struct address_space
*mapping
,
2607 loff_t pos
, unsigned len
, unsigned copied
,
2608 struct page
*page
, void *fsdata
)
2610 struct inode
*inode
= mapping
->host
;
2612 handle_t
*handle
= ext4_journal_current_handle();
2614 unsigned long start
, end
;
2615 int write_mode
= (int)(unsigned long)fsdata
;
2617 if (write_mode
== FALL_BACK_TO_NONDELALLOC
) {
2618 if (ext4_should_order_data(inode
)) {
2619 return ext4_ordered_write_end(file
, mapping
, pos
,
2620 len
, copied
, page
, fsdata
);
2621 } else if (ext4_should_writeback_data(inode
)) {
2622 return ext4_writeback_write_end(file
, mapping
, pos
,
2623 len
, copied
, page
, fsdata
);
2629 start
= pos
& (PAGE_CACHE_SIZE
- 1);
2630 end
= start
+ copied
- 1;
2633 * generic_write_end() will run mark_inode_dirty() if i_size
2634 * changes. So let's piggyback the i_disksize mark_inode_dirty
2638 new_i_size
= pos
+ copied
;
2639 if (new_i_size
> EXT4_I(inode
)->i_disksize
) {
2640 if (ext4_da_should_update_i_disksize(page
, end
)) {
2641 down_write(&EXT4_I(inode
)->i_data_sem
);
2642 if (new_i_size
> EXT4_I(inode
)->i_disksize
) {
2644 * Updating i_disksize when extending file
2645 * without needing block allocation
2647 if (ext4_should_order_data(inode
))
2648 ret
= ext4_jbd2_file_inode(handle
,
2651 EXT4_I(inode
)->i_disksize
= new_i_size
;
2653 up_write(&EXT4_I(inode
)->i_data_sem
);
2654 /* We need to mark inode dirty even if
2655 * new_i_size is less that inode->i_size
2656 * bu greater than i_disksize.(hint delalloc)
2658 ext4_mark_inode_dirty(handle
, inode
);
2661 ret2
= generic_write_end(file
, mapping
, pos
, len
, copied
,
2666 ret2
= ext4_journal_stop(handle
);
2670 return ret
? ret
: copied
;
2673 static void ext4_da_invalidatepage(struct page
*page
, unsigned long offset
)
2676 * Drop reserved blocks
2678 BUG_ON(!PageLocked(page
));
2679 if (!page_has_buffers(page
))
2682 ext4_da_page_release_reservation(page
, offset
);
2685 ext4_invalidatepage(page
, offset
);
2692 * bmap() is special. It gets used by applications such as lilo and by
2693 * the swapper to find the on-disk block of a specific piece of data.
2695 * Naturally, this is dangerous if the block concerned is still in the
2696 * journal. If somebody makes a swapfile on an ext4 data-journaling
2697 * filesystem and enables swap, then they may get a nasty shock when the
2698 * data getting swapped to that swapfile suddenly gets overwritten by
2699 * the original zero's written out previously to the journal and
2700 * awaiting writeback in the kernel's buffer cache.
2702 * So, if we see any bmap calls here on a modified, data-journaled file,
2703 * take extra steps to flush any blocks which might be in the cache.
2705 static sector_t
ext4_bmap(struct address_space
*mapping
, sector_t block
)
2707 struct inode
*inode
= mapping
->host
;
2711 if (mapping_tagged(mapping
, PAGECACHE_TAG_DIRTY
) &&
2712 test_opt(inode
->i_sb
, DELALLOC
)) {
2714 * With delalloc we want to sync the file
2715 * so that we can make sure we allocate
2718 filemap_write_and_wait(mapping
);
2721 if (EXT4_I(inode
)->i_state
& EXT4_STATE_JDATA
) {
2723 * This is a REALLY heavyweight approach, but the use of
2724 * bmap on dirty files is expected to be extremely rare:
2725 * only if we run lilo or swapon on a freshly made file
2726 * do we expect this to happen.
2728 * (bmap requires CAP_SYS_RAWIO so this does not
2729 * represent an unprivileged user DOS attack --- we'd be
2730 * in trouble if mortal users could trigger this path at
2733 * NB. EXT4_STATE_JDATA is not set on files other than
2734 * regular files. If somebody wants to bmap a directory
2735 * or symlink and gets confused because the buffer
2736 * hasn't yet been flushed to disk, they deserve
2737 * everything they get.
2740 EXT4_I(inode
)->i_state
&= ~EXT4_STATE_JDATA
;
2741 journal
= EXT4_JOURNAL(inode
);
2742 jbd2_journal_lock_updates(journal
);
2743 err
= jbd2_journal_flush(journal
);
2744 jbd2_journal_unlock_updates(journal
);
2750 return generic_block_bmap(mapping
, block
, ext4_get_block
);
2753 static int bget_one(handle_t
*handle
, struct buffer_head
*bh
)
2759 static int bput_one(handle_t
*handle
, struct buffer_head
*bh
)
2766 * Note that we don't need to start a transaction unless we're journaling data
2767 * because we should have holes filled from ext4_page_mkwrite(). We even don't
2768 * need to file the inode to the transaction's list in ordered mode because if
2769 * we are writing back data added by write(), the inode is already there and if
2770 * we are writing back data modified via mmap(), noone guarantees in which
2771 * transaction the data will hit the disk. In case we are journaling data, we
2772 * cannot start transaction directly because transaction start ranks above page
2773 * lock so we have to do some magic.
2775 * In all journaling modes block_write_full_page() will start the I/O.
2779 * ext4_writepage() -> kmalloc() -> __alloc_pages() -> page_launder() ->
2784 * ext4_file_write() -> generic_file_write() -> __alloc_pages() -> ...
2786 * Same applies to ext4_get_block(). We will deadlock on various things like
2787 * lock_journal and i_data_sem
2789 * Setting PF_MEMALLOC here doesn't work - too many internal memory
2792 * 16May01: If we're reentered then journal_current_handle() will be
2793 * non-zero. We simply *return*.
2795 * 1 July 2001: @@@ FIXME:
2796 * In journalled data mode, a data buffer may be metadata against the
2797 * current transaction. But the same file is part of a shared mapping
2798 * and someone does a writepage() on it.
2800 * We will move the buffer onto the async_data list, but *after* it has
2801 * been dirtied. So there's a small window where we have dirty data on
2804 * Note that this only applies to the last partial page in the file. The
2805 * bit which block_write_full_page() uses prepare/commit for. (That's
2806 * broken code anyway: it's wrong for msync()).
2808 * It's a rare case: affects the final partial page, for journalled data
2809 * where the file is subject to bith write() and writepage() in the same
2810 * transction. To fix it we'll need a custom block_write_full_page().
2811 * We'll probably need that anyway for journalling writepage() output.
2813 * We don't honour synchronous mounts for writepage(). That would be
2814 * disastrous. Any write() or metadata operation will sync the fs for
2818 static int __ext4_normal_writepage(struct page
*page
,
2819 struct writeback_control
*wbc
)
2821 struct inode
*inode
= page
->mapping
->host
;
2823 if (test_opt(inode
->i_sb
, NOBH
))
2824 return nobh_writepage(page
,
2825 ext4_normal_get_block_write
, wbc
);
2827 return block_write_full_page(page
,
2828 ext4_normal_get_block_write
,
2832 static int ext4_normal_writepage(struct page
*page
,
2833 struct writeback_control
*wbc
)
2835 struct inode
*inode
= page
->mapping
->host
;
2836 loff_t size
= i_size_read(inode
);
2839 J_ASSERT(PageLocked(page
));
2840 if (page
->index
== size
>> PAGE_CACHE_SHIFT
)
2841 len
= size
& ~PAGE_CACHE_MASK
;
2843 len
= PAGE_CACHE_SIZE
;
2845 if (page_has_buffers(page
)) {
2846 /* if page has buffers it should all be mapped
2847 * and allocated. If there are not buffers attached
2848 * to the page we know the page is dirty but it lost
2849 * buffers. That means that at some moment in time
2850 * after write_begin() / write_end() has been called
2851 * all buffers have been clean and thus they must have been
2852 * written at least once. So they are all mapped and we can
2853 * happily proceed with mapping them and writing the page.
2855 BUG_ON(walk_page_buffers(NULL
, page_buffers(page
), 0, len
, NULL
,
2856 ext4_bh_unmapped_or_delay
));
2859 if (!ext4_journal_current_handle())
2860 return __ext4_normal_writepage(page
, wbc
);
2862 redirty_page_for_writepage(wbc
, page
);
2867 static int __ext4_journalled_writepage(struct page
*page
,
2868 struct writeback_control
*wbc
)
2870 struct address_space
*mapping
= page
->mapping
;
2871 struct inode
*inode
= mapping
->host
;
2872 struct buffer_head
*page_bufs
;
2873 handle_t
*handle
= NULL
;
2877 ret
= block_prepare_write(page
, 0, PAGE_CACHE_SIZE
,
2878 ext4_normal_get_block_write
);
2882 page_bufs
= page_buffers(page
);
2883 walk_page_buffers(handle
, page_bufs
, 0, PAGE_CACHE_SIZE
, NULL
,
2885 /* As soon as we unlock the page, it can go away, but we have
2886 * references to buffers so we are safe */
2889 handle
= ext4_journal_start(inode
, ext4_writepage_trans_blocks(inode
));
2890 if (IS_ERR(handle
)) {
2891 ret
= PTR_ERR(handle
);
2895 ret
= walk_page_buffers(handle
, page_bufs
, 0,
2896 PAGE_CACHE_SIZE
, NULL
, do_journal_get_write_access
);
2898 err
= walk_page_buffers(handle
, page_bufs
, 0,
2899 PAGE_CACHE_SIZE
, NULL
, write_end_fn
);
2902 err
= ext4_journal_stop(handle
);
2906 walk_page_buffers(handle
, page_bufs
, 0,
2907 PAGE_CACHE_SIZE
, NULL
, bput_one
);
2908 EXT4_I(inode
)->i_state
|= EXT4_STATE_JDATA
;
2917 static int ext4_journalled_writepage(struct page
*page
,
2918 struct writeback_control
*wbc
)
2920 struct inode
*inode
= page
->mapping
->host
;
2921 loff_t size
= i_size_read(inode
);
2924 J_ASSERT(PageLocked(page
));
2925 if (page
->index
== size
>> PAGE_CACHE_SHIFT
)
2926 len
= size
& ~PAGE_CACHE_MASK
;
2928 len
= PAGE_CACHE_SIZE
;
2930 if (page_has_buffers(page
)) {
2931 /* if page has buffers it should all be mapped
2932 * and allocated. If there are not buffers attached
2933 * to the page we know the page is dirty but it lost
2934 * buffers. That means that at some moment in time
2935 * after write_begin() / write_end() has been called
2936 * all buffers have been clean and thus they must have been
2937 * written at least once. So they are all mapped and we can
2938 * happily proceed with mapping them and writing the page.
2940 BUG_ON(walk_page_buffers(NULL
, page_buffers(page
), 0, len
, NULL
,
2941 ext4_bh_unmapped_or_delay
));
2944 if (ext4_journal_current_handle())
2947 if (PageChecked(page
)) {
2949 * It's mmapped pagecache. Add buffers and journal it. There
2950 * doesn't seem much point in redirtying the page here.
2952 ClearPageChecked(page
);
2953 return __ext4_journalled_writepage(page
, wbc
);
2956 * It may be a page full of checkpoint-mode buffers. We don't
2957 * really know unless we go poke around in the buffer_heads.
2958 * But block_write_full_page will do the right thing.
2960 return block_write_full_page(page
,
2961 ext4_normal_get_block_write
,
2965 redirty_page_for_writepage(wbc
, page
);
2970 static int ext4_readpage(struct file
*file
, struct page
*page
)
2972 return mpage_readpage(page
, ext4_get_block
);
2976 ext4_readpages(struct file
*file
, struct address_space
*mapping
,
2977 struct list_head
*pages
, unsigned nr_pages
)
2979 return mpage_readpages(mapping
, pages
, nr_pages
, ext4_get_block
);
2982 static void ext4_invalidatepage(struct page
*page
, unsigned long offset
)
2984 journal_t
*journal
= EXT4_JOURNAL(page
->mapping
->host
);
2987 * If it's a full truncate we just forget about the pending dirtying
2990 ClearPageChecked(page
);
2992 jbd2_journal_invalidatepage(journal
, page
, offset
);
2995 static int ext4_releasepage(struct page
*page
, gfp_t wait
)
2997 journal_t
*journal
= EXT4_JOURNAL(page
->mapping
->host
);
2999 WARN_ON(PageChecked(page
));
3000 if (!page_has_buffers(page
))
3002 return jbd2_journal_try_to_free_buffers(journal
, page
, wait
);
3006 * If the O_DIRECT write will extend the file then add this inode to the
3007 * orphan list. So recovery will truncate it back to the original size
3008 * if the machine crashes during the write.
3010 * If the O_DIRECT write is intantiating holes inside i_size and the machine
3011 * crashes then stale disk data _may_ be exposed inside the file. But current
3012 * VFS code falls back into buffered path in that case so we are safe.
3014 static ssize_t
ext4_direct_IO(int rw
, struct kiocb
*iocb
,
3015 const struct iovec
*iov
, loff_t offset
,
3016 unsigned long nr_segs
)
3018 struct file
*file
= iocb
->ki_filp
;
3019 struct inode
*inode
= file
->f_mapping
->host
;
3020 struct ext4_inode_info
*ei
= EXT4_I(inode
);
3024 size_t count
= iov_length(iov
, nr_segs
);
3027 loff_t final_size
= offset
+ count
;
3029 if (final_size
> inode
->i_size
) {
3030 /* Credits for sb + inode write */
3031 handle
= ext4_journal_start(inode
, 2);
3032 if (IS_ERR(handle
)) {
3033 ret
= PTR_ERR(handle
);
3036 ret
= ext4_orphan_add(handle
, inode
);
3038 ext4_journal_stop(handle
);
3042 ei
->i_disksize
= inode
->i_size
;
3043 ext4_journal_stop(handle
);
3047 ret
= blockdev_direct_IO(rw
, iocb
, inode
, inode
->i_sb
->s_bdev
, iov
,
3049 ext4_get_block
, NULL
);
3054 /* Credits for sb + inode write */
3055 handle
= ext4_journal_start(inode
, 2);
3056 if (IS_ERR(handle
)) {
3057 /* This is really bad luck. We've written the data
3058 * but cannot extend i_size. Bail out and pretend
3059 * the write failed... */
3060 ret
= PTR_ERR(handle
);
3064 ext4_orphan_del(handle
, inode
);
3066 loff_t end
= offset
+ ret
;
3067 if (end
> inode
->i_size
) {
3068 ei
->i_disksize
= end
;
3069 i_size_write(inode
, end
);
3071 * We're going to return a positive `ret'
3072 * here due to non-zero-length I/O, so there's
3073 * no way of reporting error returns from
3074 * ext4_mark_inode_dirty() to userspace. So
3077 ext4_mark_inode_dirty(handle
, inode
);
3080 err
= ext4_journal_stop(handle
);
3089 * Pages can be marked dirty completely asynchronously from ext4's journalling
3090 * activity. By filemap_sync_pte(), try_to_unmap_one(), etc. We cannot do
3091 * much here because ->set_page_dirty is called under VFS locks. The page is
3092 * not necessarily locked.
3094 * We cannot just dirty the page and leave attached buffers clean, because the
3095 * buffers' dirty state is "definitive". We cannot just set the buffers dirty
3096 * or jbddirty because all the journalling code will explode.
3098 * So what we do is to mark the page "pending dirty" and next time writepage
3099 * is called, propagate that into the buffers appropriately.
3101 static int ext4_journalled_set_page_dirty(struct page
*page
)
3103 SetPageChecked(page
);
3104 return __set_page_dirty_nobuffers(page
);
3107 static const struct address_space_operations ext4_ordered_aops
= {
3108 .readpage
= ext4_readpage
,
3109 .readpages
= ext4_readpages
,
3110 .writepage
= ext4_normal_writepage
,
3111 .sync_page
= block_sync_page
,
3112 .write_begin
= ext4_write_begin
,
3113 .write_end
= ext4_ordered_write_end
,
3115 .invalidatepage
= ext4_invalidatepage
,
3116 .releasepage
= ext4_releasepage
,
3117 .direct_IO
= ext4_direct_IO
,
3118 .migratepage
= buffer_migrate_page
,
3119 .is_partially_uptodate
= block_is_partially_uptodate
,
3122 static const struct address_space_operations ext4_writeback_aops
= {
3123 .readpage
= ext4_readpage
,
3124 .readpages
= ext4_readpages
,
3125 .writepage
= ext4_normal_writepage
,
3126 .sync_page
= block_sync_page
,
3127 .write_begin
= ext4_write_begin
,
3128 .write_end
= ext4_writeback_write_end
,
3130 .invalidatepage
= ext4_invalidatepage
,
3131 .releasepage
= ext4_releasepage
,
3132 .direct_IO
= ext4_direct_IO
,
3133 .migratepage
= buffer_migrate_page
,
3134 .is_partially_uptodate
= block_is_partially_uptodate
,
3137 static const struct address_space_operations ext4_journalled_aops
= {
3138 .readpage
= ext4_readpage
,
3139 .readpages
= ext4_readpages
,
3140 .writepage
= ext4_journalled_writepage
,
3141 .sync_page
= block_sync_page
,
3142 .write_begin
= ext4_write_begin
,
3143 .write_end
= ext4_journalled_write_end
,
3144 .set_page_dirty
= ext4_journalled_set_page_dirty
,
3146 .invalidatepage
= ext4_invalidatepage
,
3147 .releasepage
= ext4_releasepage
,
3148 .is_partially_uptodate
= block_is_partially_uptodate
,
3151 static const struct address_space_operations ext4_da_aops
= {
3152 .readpage
= ext4_readpage
,
3153 .readpages
= ext4_readpages
,
3154 .writepage
= ext4_da_writepage
,
3155 .writepages
= ext4_da_writepages
,
3156 .sync_page
= block_sync_page
,
3157 .write_begin
= ext4_da_write_begin
,
3158 .write_end
= ext4_da_write_end
,
3160 .invalidatepage
= ext4_da_invalidatepage
,
3161 .releasepage
= ext4_releasepage
,
3162 .direct_IO
= ext4_direct_IO
,
3163 .migratepage
= buffer_migrate_page
,
3164 .is_partially_uptodate
= block_is_partially_uptodate
,
3167 void ext4_set_aops(struct inode
*inode
)
3169 if (ext4_should_order_data(inode
) &&
3170 test_opt(inode
->i_sb
, DELALLOC
))
3171 inode
->i_mapping
->a_ops
= &ext4_da_aops
;
3172 else if (ext4_should_order_data(inode
))
3173 inode
->i_mapping
->a_ops
= &ext4_ordered_aops
;
3174 else if (ext4_should_writeback_data(inode
) &&
3175 test_opt(inode
->i_sb
, DELALLOC
))
3176 inode
->i_mapping
->a_ops
= &ext4_da_aops
;
3177 else if (ext4_should_writeback_data(inode
))
3178 inode
->i_mapping
->a_ops
= &ext4_writeback_aops
;
3180 inode
->i_mapping
->a_ops
= &ext4_journalled_aops
;
3184 * ext4_block_truncate_page() zeroes out a mapping from file offset `from'
3185 * up to the end of the block which corresponds to `from'.
3186 * This required during truncate. We need to physically zero the tail end
3187 * of that block so it doesn't yield old data if the file is later grown.
3189 int ext4_block_truncate_page(handle_t
*handle
,
3190 struct address_space
*mapping
, loff_t from
)
3192 ext4_fsblk_t index
= from
>> PAGE_CACHE_SHIFT
;
3193 unsigned offset
= from
& (PAGE_CACHE_SIZE
-1);
3194 unsigned blocksize
, length
, pos
;
3196 struct inode
*inode
= mapping
->host
;
3197 struct buffer_head
*bh
;
3201 page
= grab_cache_page(mapping
, from
>> PAGE_CACHE_SHIFT
);
3205 blocksize
= inode
->i_sb
->s_blocksize
;
3206 length
= blocksize
- (offset
& (blocksize
- 1));
3207 iblock
= index
<< (PAGE_CACHE_SHIFT
- inode
->i_sb
->s_blocksize_bits
);
3210 * For "nobh" option, we can only work if we don't need to
3211 * read-in the page - otherwise we create buffers to do the IO.
3213 if (!page_has_buffers(page
) && test_opt(inode
->i_sb
, NOBH
) &&
3214 ext4_should_writeback_data(inode
) && PageUptodate(page
)) {
3215 zero_user(page
, offset
, length
);
3216 set_page_dirty(page
);
3220 if (!page_has_buffers(page
))
3221 create_empty_buffers(page
, blocksize
, 0);
3223 /* Find the buffer that contains "offset" */
3224 bh
= page_buffers(page
);
3226 while (offset
>= pos
) {
3227 bh
= bh
->b_this_page
;
3233 if (buffer_freed(bh
)) {
3234 BUFFER_TRACE(bh
, "freed: skip");
3238 if (!buffer_mapped(bh
)) {
3239 BUFFER_TRACE(bh
, "unmapped");
3240 ext4_get_block(inode
, iblock
, bh
, 0);
3241 /* unmapped? It's a hole - nothing to do */
3242 if (!buffer_mapped(bh
)) {
3243 BUFFER_TRACE(bh
, "still unmapped");
3248 /* Ok, it's mapped. Make sure it's up-to-date */
3249 if (PageUptodate(page
))
3250 set_buffer_uptodate(bh
);
3252 if (!buffer_uptodate(bh
)) {
3254 ll_rw_block(READ
, 1, &bh
);
3256 /* Uhhuh. Read error. Complain and punt. */
3257 if (!buffer_uptodate(bh
))
3261 if (ext4_should_journal_data(inode
)) {
3262 BUFFER_TRACE(bh
, "get write access");
3263 err
= ext4_journal_get_write_access(handle
, bh
);
3268 zero_user(page
, offset
, length
);
3270 BUFFER_TRACE(bh
, "zeroed end of block");
3273 if (ext4_should_journal_data(inode
)) {
3274 err
= ext4_journal_dirty_metadata(handle
, bh
);
3276 if (ext4_should_order_data(inode
))
3277 err
= ext4_jbd2_file_inode(handle
, inode
);
3278 mark_buffer_dirty(bh
);
3283 page_cache_release(page
);
3288 * Probably it should be a library function... search for first non-zero word
3289 * or memcmp with zero_page, whatever is better for particular architecture.
3292 static inline int all_zeroes(__le32
*p
, __le32
*q
)
3301 * ext4_find_shared - find the indirect blocks for partial truncation.
3302 * @inode: inode in question
3303 * @depth: depth of the affected branch
3304 * @offsets: offsets of pointers in that branch (see ext4_block_to_path)
3305 * @chain: place to store the pointers to partial indirect blocks
3306 * @top: place to the (detached) top of branch
3308 * This is a helper function used by ext4_truncate().
3310 * When we do truncate() we may have to clean the ends of several
3311 * indirect blocks but leave the blocks themselves alive. Block is
3312 * partially truncated if some data below the new i_size is refered
3313 * from it (and it is on the path to the first completely truncated
3314 * data block, indeed). We have to free the top of that path along
3315 * with everything to the right of the path. Since no allocation
3316 * past the truncation point is possible until ext4_truncate()
3317 * finishes, we may safely do the latter, but top of branch may
3318 * require special attention - pageout below the truncation point
3319 * might try to populate it.
3321 * We atomically detach the top of branch from the tree, store the
3322 * block number of its root in *@top, pointers to buffer_heads of
3323 * partially truncated blocks - in @chain[].bh and pointers to
3324 * their last elements that should not be removed - in
3325 * @chain[].p. Return value is the pointer to last filled element
3328 * The work left to caller to do the actual freeing of subtrees:
3329 * a) free the subtree starting from *@top
3330 * b) free the subtrees whose roots are stored in
3331 * (@chain[i].p+1 .. end of @chain[i].bh->b_data)
3332 * c) free the subtrees growing from the inode past the @chain[0].
3333 * (no partially truncated stuff there). */
3335 static Indirect
*ext4_find_shared(struct inode
*inode
, int depth
,
3336 ext4_lblk_t offsets
[4], Indirect chain
[4], __le32
*top
)
3338 Indirect
*partial
, *p
;
3342 /* Make k index the deepest non-null offest + 1 */
3343 for (k
= depth
; k
> 1 && !offsets
[k
-1]; k
--)
3345 partial
= ext4_get_branch(inode
, k
, offsets
, chain
, &err
);
3346 /* Writer: pointers */
3348 partial
= chain
+ k
-1;
3350 * If the branch acquired continuation since we've looked at it -
3351 * fine, it should all survive and (new) top doesn't belong to us.
3353 if (!partial
->key
&& *partial
->p
)
3356 for (p
= partial
; (p
> chain
) && all_zeroes((__le32
*) p
->bh
->b_data
, p
->p
); p
--)
3359 * OK, we've found the last block that must survive. The rest of our
3360 * branch should be detached before unlocking. However, if that rest
3361 * of branch is all ours and does not grow immediately from the inode
3362 * it's easier to cheat and just decrement partial->p.
3364 if (p
== chain
+ k
- 1 && p
> chain
) {
3368 /* Nope, don't do this in ext4. Must leave the tree intact */
3375 while (partial
> p
) {
3376 brelse(partial
->bh
);
3384 * Zero a number of block pointers in either an inode or an indirect block.
3385 * If we restart the transaction we must again get write access to the
3386 * indirect block for further modification.
3388 * We release `count' blocks on disk, but (last - first) may be greater
3389 * than `count' because there can be holes in there.
3391 static void ext4_clear_blocks(handle_t
*handle
, struct inode
*inode
,
3392 struct buffer_head
*bh
, ext4_fsblk_t block_to_free
,
3393 unsigned long count
, __le32
*first
, __le32
*last
)
3396 if (try_to_extend_transaction(handle
, inode
)) {
3398 BUFFER_TRACE(bh
, "call ext4_journal_dirty_metadata");
3399 ext4_journal_dirty_metadata(handle
, bh
);
3401 ext4_mark_inode_dirty(handle
, inode
);
3402 ext4_journal_test_restart(handle
, inode
);
3404 BUFFER_TRACE(bh
, "retaking write access");
3405 ext4_journal_get_write_access(handle
, bh
);
3410 * Any buffers which are on the journal will be in memory. We find
3411 * them on the hash table so jbd2_journal_revoke() will run jbd2_journal_forget()
3412 * on them. We've already detached each block from the file, so
3413 * bforget() in jbd2_journal_forget() should be safe.
3415 * AKPM: turn on bforget in jbd2_journal_forget()!!!
3417 for (p
= first
; p
< last
; p
++) {
3418 u32 nr
= le32_to_cpu(*p
);
3420 struct buffer_head
*tbh
;
3423 tbh
= sb_find_get_block(inode
->i_sb
, nr
);
3424 ext4_forget(handle
, 0, inode
, tbh
, nr
);
3428 ext4_free_blocks(handle
, inode
, block_to_free
, count
, 0);
3432 * ext4_free_data - free a list of data blocks
3433 * @handle: handle for this transaction
3434 * @inode: inode we are dealing with
3435 * @this_bh: indirect buffer_head which contains *@first and *@last
3436 * @first: array of block numbers
3437 * @last: points immediately past the end of array
3439 * We are freeing all blocks refered from that array (numbers are stored as
3440 * little-endian 32-bit) and updating @inode->i_blocks appropriately.
3442 * We accumulate contiguous runs of blocks to free. Conveniently, if these
3443 * blocks are contiguous then releasing them at one time will only affect one
3444 * or two bitmap blocks (+ group descriptor(s) and superblock) and we won't
3445 * actually use a lot of journal space.
3447 * @this_bh will be %NULL if @first and @last point into the inode's direct
3450 static void ext4_free_data(handle_t
*handle
, struct inode
*inode
,
3451 struct buffer_head
*this_bh
,
3452 __le32
*first
, __le32
*last
)
3454 ext4_fsblk_t block_to_free
= 0; /* Starting block # of a run */
3455 unsigned long count
= 0; /* Number of blocks in the run */
3456 __le32
*block_to_free_p
= NULL
; /* Pointer into inode/ind
3459 ext4_fsblk_t nr
; /* Current block # */
3460 __le32
*p
; /* Pointer into inode/ind
3461 for current block */
3464 if (this_bh
) { /* For indirect block */
3465 BUFFER_TRACE(this_bh
, "get_write_access");
3466 err
= ext4_journal_get_write_access(handle
, this_bh
);
3467 /* Important: if we can't update the indirect pointers
3468 * to the blocks, we can't free them. */
3473 for (p
= first
; p
< last
; p
++) {
3474 nr
= le32_to_cpu(*p
);
3476 /* accumulate blocks to free if they're contiguous */
3479 block_to_free_p
= p
;
3481 } else if (nr
== block_to_free
+ count
) {
3484 ext4_clear_blocks(handle
, inode
, this_bh
,
3486 count
, block_to_free_p
, p
);
3488 block_to_free_p
= p
;
3495 ext4_clear_blocks(handle
, inode
, this_bh
, block_to_free
,
3496 count
, block_to_free_p
, p
);
3499 BUFFER_TRACE(this_bh
, "call ext4_journal_dirty_metadata");
3502 * The buffer head should have an attached journal head at this
3503 * point. However, if the data is corrupted and an indirect
3504 * block pointed to itself, it would have been detached when
3505 * the block was cleared. Check for this instead of OOPSing.
3508 ext4_journal_dirty_metadata(handle
, this_bh
);
3510 ext4_error(inode
->i_sb
, __func__
,
3511 "circular indirect block detected, "
3512 "inode=%lu, block=%llu",
3514 (unsigned long long) this_bh
->b_blocknr
);
3519 * ext4_free_branches - free an array of branches
3520 * @handle: JBD handle for this transaction
3521 * @inode: inode we are dealing with
3522 * @parent_bh: the buffer_head which contains *@first and *@last
3523 * @first: array of block numbers
3524 * @last: pointer immediately past the end of array
3525 * @depth: depth of the branches to free
3527 * We are freeing all blocks refered from these branches (numbers are
3528 * stored as little-endian 32-bit) and updating @inode->i_blocks
3531 static void ext4_free_branches(handle_t
*handle
, struct inode
*inode
,
3532 struct buffer_head
*parent_bh
,
3533 __le32
*first
, __le32
*last
, int depth
)
3538 if (is_handle_aborted(handle
))
3542 struct buffer_head
*bh
;
3543 int addr_per_block
= EXT4_ADDR_PER_BLOCK(inode
->i_sb
);
3545 while (--p
>= first
) {
3546 nr
= le32_to_cpu(*p
);
3548 continue; /* A hole */
3550 /* Go read the buffer for the next level down */
3551 bh
= sb_bread(inode
->i_sb
, nr
);
3554 * A read failure? Report error and clear slot
3558 ext4_error(inode
->i_sb
, "ext4_free_branches",
3559 "Read failure, inode=%lu, block=%llu",
3564 /* This zaps the entire block. Bottom up. */
3565 BUFFER_TRACE(bh
, "free child branches");
3566 ext4_free_branches(handle
, inode
, bh
,
3567 (__le32
*) bh
->b_data
,
3568 (__le32
*) bh
->b_data
+ addr_per_block
,
3572 * We've probably journalled the indirect block several
3573 * times during the truncate. But it's no longer
3574 * needed and we now drop it from the transaction via
3575 * jbd2_journal_revoke().
3577 * That's easy if it's exclusively part of this
3578 * transaction. But if it's part of the committing
3579 * transaction then jbd2_journal_forget() will simply
3580 * brelse() it. That means that if the underlying
3581 * block is reallocated in ext4_get_block(),
3582 * unmap_underlying_metadata() will find this block
3583 * and will try to get rid of it. damn, damn.
3585 * If this block has already been committed to the
3586 * journal, a revoke record will be written. And
3587 * revoke records must be emitted *before* clearing
3588 * this block's bit in the bitmaps.
3590 ext4_forget(handle
, 1, inode
, bh
, bh
->b_blocknr
);
3593 * Everything below this this pointer has been
3594 * released. Now let this top-of-subtree go.
3596 * We want the freeing of this indirect block to be
3597 * atomic in the journal with the updating of the
3598 * bitmap block which owns it. So make some room in
3601 * We zero the parent pointer *after* freeing its
3602 * pointee in the bitmaps, so if extend_transaction()
3603 * for some reason fails to put the bitmap changes and
3604 * the release into the same transaction, recovery
3605 * will merely complain about releasing a free block,
3606 * rather than leaking blocks.
3608 if (is_handle_aborted(handle
))
3610 if (try_to_extend_transaction(handle
, inode
)) {
3611 ext4_mark_inode_dirty(handle
, inode
);
3612 ext4_journal_test_restart(handle
, inode
);
3615 ext4_free_blocks(handle
, inode
, nr
, 1, 1);
3619 * The block which we have just freed is
3620 * pointed to by an indirect block: journal it
3622 BUFFER_TRACE(parent_bh
, "get_write_access");
3623 if (!ext4_journal_get_write_access(handle
,
3626 BUFFER_TRACE(parent_bh
,
3627 "call ext4_journal_dirty_metadata");
3628 ext4_journal_dirty_metadata(handle
,
3634 /* We have reached the bottom of the tree. */
3635 BUFFER_TRACE(parent_bh
, "free data blocks");
3636 ext4_free_data(handle
, inode
, parent_bh
, first
, last
);
3640 int ext4_can_truncate(struct inode
*inode
)
3642 if (IS_APPEND(inode
) || IS_IMMUTABLE(inode
))
3644 if (S_ISREG(inode
->i_mode
))
3646 if (S_ISDIR(inode
->i_mode
))
3648 if (S_ISLNK(inode
->i_mode
))
3649 return !ext4_inode_is_fast_symlink(inode
);
3656 * We block out ext4_get_block() block instantiations across the entire
3657 * transaction, and VFS/VM ensures that ext4_truncate() cannot run
3658 * simultaneously on behalf of the same inode.
3660 * As we work through the truncate and commmit bits of it to the journal there
3661 * is one core, guiding principle: the file's tree must always be consistent on
3662 * disk. We must be able to restart the truncate after a crash.
3664 * The file's tree may be transiently inconsistent in memory (although it
3665 * probably isn't), but whenever we close off and commit a journal transaction,
3666 * the contents of (the filesystem + the journal) must be consistent and
3667 * restartable. It's pretty simple, really: bottom up, right to left (although
3668 * left-to-right works OK too).
3670 * Note that at recovery time, journal replay occurs *before* the restart of
3671 * truncate against the orphan inode list.
3673 * The committed inode has the new, desired i_size (which is the same as
3674 * i_disksize in this case). After a crash, ext4_orphan_cleanup() will see
3675 * that this inode's truncate did not complete and it will again call
3676 * ext4_truncate() to have another go. So there will be instantiated blocks
3677 * to the right of the truncation point in a crashed ext4 filesystem. But
3678 * that's fine - as long as they are linked from the inode, the post-crash
3679 * ext4_truncate() run will find them and release them.
3681 void ext4_truncate(struct inode
*inode
)
3684 struct ext4_inode_info
*ei
= EXT4_I(inode
);
3685 __le32
*i_data
= ei
->i_data
;
3686 int addr_per_block
= EXT4_ADDR_PER_BLOCK(inode
->i_sb
);
3687 struct address_space
*mapping
= inode
->i_mapping
;
3688 ext4_lblk_t offsets
[4];
3693 ext4_lblk_t last_block
;
3694 unsigned blocksize
= inode
->i_sb
->s_blocksize
;
3696 if (!ext4_can_truncate(inode
))
3699 if (EXT4_I(inode
)->i_flags
& EXT4_EXTENTS_FL
) {
3700 ext4_ext_truncate(inode
);
3704 handle
= start_transaction(inode
);
3706 return; /* AKPM: return what? */
3708 last_block
= (inode
->i_size
+ blocksize
-1)
3709 >> EXT4_BLOCK_SIZE_BITS(inode
->i_sb
);
3711 if (inode
->i_size
& (blocksize
- 1))
3712 if (ext4_block_truncate_page(handle
, mapping
, inode
->i_size
))
3715 n
= ext4_block_to_path(inode
, last_block
, offsets
, NULL
);
3717 goto out_stop
; /* error */
3720 * OK. This truncate is going to happen. We add the inode to the
3721 * orphan list, so that if this truncate spans multiple transactions,
3722 * and we crash, we will resume the truncate when the filesystem
3723 * recovers. It also marks the inode dirty, to catch the new size.
3725 * Implication: the file must always be in a sane, consistent
3726 * truncatable state while each transaction commits.
3728 if (ext4_orphan_add(handle
, inode
))
3732 * From here we block out all ext4_get_block() callers who want to
3733 * modify the block allocation tree.
3735 down_write(&ei
->i_data_sem
);
3737 ext4_discard_preallocations(inode
);
3740 * The orphan list entry will now protect us from any crash which
3741 * occurs before the truncate completes, so it is now safe to propagate
3742 * the new, shorter inode size (held for now in i_size) into the
3743 * on-disk inode. We do this via i_disksize, which is the value which
3744 * ext4 *really* writes onto the disk inode.
3746 ei
->i_disksize
= inode
->i_size
;
3748 if (n
== 1) { /* direct blocks */
3749 ext4_free_data(handle
, inode
, NULL
, i_data
+offsets
[0],
3750 i_data
+ EXT4_NDIR_BLOCKS
);
3754 partial
= ext4_find_shared(inode
, n
, offsets
, chain
, &nr
);
3755 /* Kill the top of shared branch (not detached) */
3757 if (partial
== chain
) {
3758 /* Shared branch grows from the inode */
3759 ext4_free_branches(handle
, inode
, NULL
,
3760 &nr
, &nr
+1, (chain
+n
-1) - partial
);
3763 * We mark the inode dirty prior to restart,
3764 * and prior to stop. No need for it here.
3767 /* Shared branch grows from an indirect block */
3768 BUFFER_TRACE(partial
->bh
, "get_write_access");
3769 ext4_free_branches(handle
, inode
, partial
->bh
,
3771 partial
->p
+1, (chain
+n
-1) - partial
);
3774 /* Clear the ends of indirect blocks on the shared branch */
3775 while (partial
> chain
) {
3776 ext4_free_branches(handle
, inode
, partial
->bh
, partial
->p
+ 1,
3777 (__le32
*)partial
->bh
->b_data
+addr_per_block
,
3778 (chain
+n
-1) - partial
);
3779 BUFFER_TRACE(partial
->bh
, "call brelse");
3780 brelse (partial
->bh
);
3784 /* Kill the remaining (whole) subtrees */
3785 switch (offsets
[0]) {
3787 nr
= i_data
[EXT4_IND_BLOCK
];
3789 ext4_free_branches(handle
, inode
, NULL
, &nr
, &nr
+1, 1);
3790 i_data
[EXT4_IND_BLOCK
] = 0;
3792 case EXT4_IND_BLOCK
:
3793 nr
= i_data
[EXT4_DIND_BLOCK
];
3795 ext4_free_branches(handle
, inode
, NULL
, &nr
, &nr
+1, 2);
3796 i_data
[EXT4_DIND_BLOCK
] = 0;
3798 case EXT4_DIND_BLOCK
:
3799 nr
= i_data
[EXT4_TIND_BLOCK
];
3801 ext4_free_branches(handle
, inode
, NULL
, &nr
, &nr
+1, 3);
3802 i_data
[EXT4_TIND_BLOCK
] = 0;
3804 case EXT4_TIND_BLOCK
:
3808 up_write(&ei
->i_data_sem
);
3809 inode
->i_mtime
= inode
->i_ctime
= ext4_current_time(inode
);
3810 ext4_mark_inode_dirty(handle
, inode
);
3813 * In a multi-transaction truncate, we only make the final transaction
3820 * If this was a simple ftruncate(), and the file will remain alive
3821 * then we need to clear up the orphan record which we created above.
3822 * However, if this was a real unlink then we were called by
3823 * ext4_delete_inode(), and we allow that function to clean up the
3824 * orphan info for us.
3827 ext4_orphan_del(handle
, inode
);
3829 ext4_journal_stop(handle
);
3833 * ext4_get_inode_loc returns with an extra refcount against the inode's
3834 * underlying buffer_head on success. If 'in_mem' is true, we have all
3835 * data in memory that is needed to recreate the on-disk version of this
3838 static int __ext4_get_inode_loc(struct inode
*inode
,
3839 struct ext4_iloc
*iloc
, int in_mem
)
3841 struct ext4_group_desc
*gdp
;
3842 struct buffer_head
*bh
;
3843 struct super_block
*sb
= inode
->i_sb
;
3845 int inodes_per_block
, inode_offset
;
3848 if (!ext4_valid_inum(sb
, inode
->i_ino
))
3851 iloc
->block_group
= (inode
->i_ino
- 1) / EXT4_INODES_PER_GROUP(sb
);
3852 gdp
= ext4_get_group_desc(sb
, iloc
->block_group
, NULL
);
3857 * Figure out the offset within the block group inode table
3859 inodes_per_block
= (EXT4_BLOCK_SIZE(sb
) / EXT4_INODE_SIZE(sb
));
3860 inode_offset
= ((inode
->i_ino
- 1) %
3861 EXT4_INODES_PER_GROUP(sb
));
3862 block
= ext4_inode_table(sb
, gdp
) + (inode_offset
/ inodes_per_block
);
3863 iloc
->offset
= (inode_offset
% inodes_per_block
) * EXT4_INODE_SIZE(sb
);
3865 bh
= sb_getblk(sb
, block
);
3867 ext4_error(sb
, "ext4_get_inode_loc", "unable to read "
3868 "inode block - inode=%lu, block=%llu",
3869 inode
->i_ino
, block
);
3872 if (!buffer_uptodate(bh
)) {
3876 * If the buffer has the write error flag, we have failed
3877 * to write out another inode in the same block. In this
3878 * case, we don't have to read the block because we may
3879 * read the old inode data successfully.
3881 if (buffer_write_io_error(bh
) && !buffer_uptodate(bh
))
3882 set_buffer_uptodate(bh
);
3884 if (buffer_uptodate(bh
)) {
3885 /* someone brought it uptodate while we waited */
3891 * If we have all information of the inode in memory and this
3892 * is the only valid inode in the block, we need not read the
3896 struct buffer_head
*bitmap_bh
;
3899 start
= inode_offset
& ~(inodes_per_block
- 1);
3901 /* Is the inode bitmap in cache? */
3902 bitmap_bh
= sb_getblk(sb
, ext4_inode_bitmap(sb
, gdp
));
3907 * If the inode bitmap isn't in cache then the
3908 * optimisation may end up performing two reads instead
3909 * of one, so skip it.
3911 if (!buffer_uptodate(bitmap_bh
)) {
3915 for (i
= start
; i
< start
+ inodes_per_block
; i
++) {
3916 if (i
== inode_offset
)
3918 if (ext4_test_bit(i
, bitmap_bh
->b_data
))
3922 if (i
== start
+ inodes_per_block
) {
3923 /* all other inodes are free, so skip I/O */
3924 memset(bh
->b_data
, 0, bh
->b_size
);
3925 set_buffer_uptodate(bh
);
3933 * If we need to do any I/O, try to pre-readahead extra
3934 * blocks from the inode table.
3936 if (EXT4_SB(sb
)->s_inode_readahead_blks
) {
3937 ext4_fsblk_t b
, end
, table
;
3940 table
= ext4_inode_table(sb
, gdp
);
3941 /* Make sure s_inode_readahead_blks is a power of 2 */
3942 while (EXT4_SB(sb
)->s_inode_readahead_blks
&
3943 (EXT4_SB(sb
)->s_inode_readahead_blks
-1))
3944 EXT4_SB(sb
)->s_inode_readahead_blks
=
3945 (EXT4_SB(sb
)->s_inode_readahead_blks
&
3946 (EXT4_SB(sb
)->s_inode_readahead_blks
-1));
3947 b
= block
& ~(EXT4_SB(sb
)->s_inode_readahead_blks
-1);
3950 end
= b
+ EXT4_SB(sb
)->s_inode_readahead_blks
;
3951 num
= EXT4_INODES_PER_GROUP(sb
);
3952 if (EXT4_HAS_RO_COMPAT_FEATURE(sb
,
3953 EXT4_FEATURE_RO_COMPAT_GDT_CSUM
))
3954 num
-= le16_to_cpu(gdp
->bg_itable_unused
);
3955 table
+= num
/ inodes_per_block
;
3959 sb_breadahead(sb
, b
++);
3963 * There are other valid inodes in the buffer, this inode
3964 * has in-inode xattrs, or we don't have this inode in memory.
3965 * Read the block from disk.
3968 bh
->b_end_io
= end_buffer_read_sync
;
3969 submit_bh(READ_META
, bh
);
3971 if (!buffer_uptodate(bh
)) {
3972 ext4_error(sb
, __func__
,
3973 "unable to read inode block - inode=%lu, "
3974 "block=%llu", inode
->i_ino
, block
);
3984 int ext4_get_inode_loc(struct inode
*inode
, struct ext4_iloc
*iloc
)
3986 /* We have all inode data except xattrs in memory here. */
3987 return __ext4_get_inode_loc(inode
, iloc
,
3988 !(EXT4_I(inode
)->i_state
& EXT4_STATE_XATTR
));
3991 void ext4_set_inode_flags(struct inode
*inode
)
3993 unsigned int flags
= EXT4_I(inode
)->i_flags
;
3995 inode
->i_flags
&= ~(S_SYNC
|S_APPEND
|S_IMMUTABLE
|S_NOATIME
|S_DIRSYNC
);
3996 if (flags
& EXT4_SYNC_FL
)
3997 inode
->i_flags
|= S_SYNC
;
3998 if (flags
& EXT4_APPEND_FL
)
3999 inode
->i_flags
|= S_APPEND
;
4000 if (flags
& EXT4_IMMUTABLE_FL
)
4001 inode
->i_flags
|= S_IMMUTABLE
;
4002 if (flags
& EXT4_NOATIME_FL
)
4003 inode
->i_flags
|= S_NOATIME
;
4004 if (flags
& EXT4_DIRSYNC_FL
)
4005 inode
->i_flags
|= S_DIRSYNC
;
4008 /* Propagate flags from i_flags to EXT4_I(inode)->i_flags */
4009 void ext4_get_inode_flags(struct ext4_inode_info
*ei
)
4011 unsigned int flags
= ei
->vfs_inode
.i_flags
;
4013 ei
->i_flags
&= ~(EXT4_SYNC_FL
|EXT4_APPEND_FL
|
4014 EXT4_IMMUTABLE_FL
|EXT4_NOATIME_FL
|EXT4_DIRSYNC_FL
);
4016 ei
->i_flags
|= EXT4_SYNC_FL
;
4017 if (flags
& S_APPEND
)
4018 ei
->i_flags
|= EXT4_APPEND_FL
;
4019 if (flags
& S_IMMUTABLE
)
4020 ei
->i_flags
|= EXT4_IMMUTABLE_FL
;
4021 if (flags
& S_NOATIME
)
4022 ei
->i_flags
|= EXT4_NOATIME_FL
;
4023 if (flags
& S_DIRSYNC
)
4024 ei
->i_flags
|= EXT4_DIRSYNC_FL
;
4026 static blkcnt_t
ext4_inode_blocks(struct ext4_inode
*raw_inode
,
4027 struct ext4_inode_info
*ei
)
4030 struct inode
*inode
= &(ei
->vfs_inode
);
4031 struct super_block
*sb
= inode
->i_sb
;
4033 if (EXT4_HAS_RO_COMPAT_FEATURE(sb
,
4034 EXT4_FEATURE_RO_COMPAT_HUGE_FILE
)) {
4035 /* we are using combined 48 bit field */
4036 i_blocks
= ((u64
)le16_to_cpu(raw_inode
->i_blocks_high
)) << 32 |
4037 le32_to_cpu(raw_inode
->i_blocks_lo
);
4038 if (ei
->i_flags
& EXT4_HUGE_FILE_FL
) {
4039 /* i_blocks represent file system block size */
4040 return i_blocks
<< (inode
->i_blkbits
- 9);
4045 return le32_to_cpu(raw_inode
->i_blocks_lo
);
4049 struct inode
*ext4_iget(struct super_block
*sb
, unsigned long ino
)
4051 struct ext4_iloc iloc
;
4052 struct ext4_inode
*raw_inode
;
4053 struct ext4_inode_info
*ei
;
4054 struct buffer_head
*bh
;
4055 struct inode
*inode
;
4059 inode
= iget_locked(sb
, ino
);
4061 return ERR_PTR(-ENOMEM
);
4062 if (!(inode
->i_state
& I_NEW
))
4066 #ifdef CONFIG_EXT4_FS_POSIX_ACL
4067 ei
->i_acl
= EXT4_ACL_NOT_CACHED
;
4068 ei
->i_default_acl
= EXT4_ACL_NOT_CACHED
;
4071 ret
= __ext4_get_inode_loc(inode
, &iloc
, 0);
4075 raw_inode
= ext4_raw_inode(&iloc
);
4076 inode
->i_mode
= le16_to_cpu(raw_inode
->i_mode
);
4077 inode
->i_uid
= (uid_t
)le16_to_cpu(raw_inode
->i_uid_low
);
4078 inode
->i_gid
= (gid_t
)le16_to_cpu(raw_inode
->i_gid_low
);
4079 if (!(test_opt(inode
->i_sb
, NO_UID32
))) {
4080 inode
->i_uid
|= le16_to_cpu(raw_inode
->i_uid_high
) << 16;
4081 inode
->i_gid
|= le16_to_cpu(raw_inode
->i_gid_high
) << 16;
4083 inode
->i_nlink
= le16_to_cpu(raw_inode
->i_links_count
);
4086 ei
->i_dir_start_lookup
= 0;
4087 ei
->i_dtime
= le32_to_cpu(raw_inode
->i_dtime
);
4088 /* We now have enough fields to check if the inode was active or not.
4089 * This is needed because nfsd might try to access dead inodes
4090 * the test is that same one that e2fsck uses
4091 * NeilBrown 1999oct15
4093 if (inode
->i_nlink
== 0) {
4094 if (inode
->i_mode
== 0 ||
4095 !(EXT4_SB(inode
->i_sb
)->s_mount_state
& EXT4_ORPHAN_FS
)) {
4096 /* this inode is deleted */
4101 /* The only unlinked inodes we let through here have
4102 * valid i_mode and are being read by the orphan
4103 * recovery code: that's fine, we're about to complete
4104 * the process of deleting those. */
4106 ei
->i_flags
= le32_to_cpu(raw_inode
->i_flags
);
4107 inode
->i_blocks
= ext4_inode_blocks(raw_inode
, ei
);
4108 ei
->i_file_acl
= le32_to_cpu(raw_inode
->i_file_acl_lo
);
4109 if (EXT4_SB(inode
->i_sb
)->s_es
->s_creator_os
!=
4110 cpu_to_le32(EXT4_OS_HURD
)) {
4112 ((__u64
)le16_to_cpu(raw_inode
->i_file_acl_high
)) << 32;
4114 inode
->i_size
= ext4_isize(raw_inode
);
4115 ei
->i_disksize
= inode
->i_size
;
4116 inode
->i_generation
= le32_to_cpu(raw_inode
->i_generation
);
4117 ei
->i_block_group
= iloc
.block_group
;
4119 * NOTE! The in-memory inode i_data array is in little-endian order
4120 * even on big-endian machines: we do NOT byteswap the block numbers!
4122 for (block
= 0; block
< EXT4_N_BLOCKS
; block
++)
4123 ei
->i_data
[block
] = raw_inode
->i_block
[block
];
4124 INIT_LIST_HEAD(&ei
->i_orphan
);
4126 if (EXT4_INODE_SIZE(inode
->i_sb
) > EXT4_GOOD_OLD_INODE_SIZE
) {
4127 ei
->i_extra_isize
= le16_to_cpu(raw_inode
->i_extra_isize
);
4128 if (EXT4_GOOD_OLD_INODE_SIZE
+ ei
->i_extra_isize
>
4129 EXT4_INODE_SIZE(inode
->i_sb
)) {
4134 if (ei
->i_extra_isize
== 0) {
4135 /* The extra space is currently unused. Use it. */
4136 ei
->i_extra_isize
= sizeof(struct ext4_inode
) -
4137 EXT4_GOOD_OLD_INODE_SIZE
;
4139 __le32
*magic
= (void *)raw_inode
+
4140 EXT4_GOOD_OLD_INODE_SIZE
+
4142 if (*magic
== cpu_to_le32(EXT4_XATTR_MAGIC
))
4143 ei
->i_state
|= EXT4_STATE_XATTR
;
4146 ei
->i_extra_isize
= 0;
4148 EXT4_INODE_GET_XTIME(i_ctime
, inode
, raw_inode
);
4149 EXT4_INODE_GET_XTIME(i_mtime
, inode
, raw_inode
);
4150 EXT4_INODE_GET_XTIME(i_atime
, inode
, raw_inode
);
4151 EXT4_EINODE_GET_XTIME(i_crtime
, ei
, raw_inode
);
4153 inode
->i_version
= le32_to_cpu(raw_inode
->i_disk_version
);
4154 if (EXT4_INODE_SIZE(inode
->i_sb
) > EXT4_GOOD_OLD_INODE_SIZE
) {
4155 if (EXT4_FITS_IN_INODE(raw_inode
, ei
, i_version_hi
))
4157 (__u64
)(le32_to_cpu(raw_inode
->i_version_hi
)) << 32;
4160 if (S_ISREG(inode
->i_mode
)) {
4161 inode
->i_op
= &ext4_file_inode_operations
;
4162 inode
->i_fop
= &ext4_file_operations
;
4163 ext4_set_aops(inode
);
4164 } else if (S_ISDIR(inode
->i_mode
)) {
4165 inode
->i_op
= &ext4_dir_inode_operations
;
4166 inode
->i_fop
= &ext4_dir_operations
;
4167 } else if (S_ISLNK(inode
->i_mode
)) {
4168 if (ext4_inode_is_fast_symlink(inode
)) {
4169 inode
->i_op
= &ext4_fast_symlink_inode_operations
;
4170 nd_terminate_link(ei
->i_data
, inode
->i_size
,
4171 sizeof(ei
->i_data
) - 1);
4173 inode
->i_op
= &ext4_symlink_inode_operations
;
4174 ext4_set_aops(inode
);
4177 inode
->i_op
= &ext4_special_inode_operations
;
4178 if (raw_inode
->i_block
[0])
4179 init_special_inode(inode
, inode
->i_mode
,
4180 old_decode_dev(le32_to_cpu(raw_inode
->i_block
[0])));
4182 init_special_inode(inode
, inode
->i_mode
,
4183 new_decode_dev(le32_to_cpu(raw_inode
->i_block
[1])));
4186 ext4_set_inode_flags(inode
);
4187 unlock_new_inode(inode
);
4192 return ERR_PTR(ret
);
4195 static int ext4_inode_blocks_set(handle_t
*handle
,
4196 struct ext4_inode
*raw_inode
,
4197 struct ext4_inode_info
*ei
)
4199 struct inode
*inode
= &(ei
->vfs_inode
);
4200 u64 i_blocks
= inode
->i_blocks
;
4201 struct super_block
*sb
= inode
->i_sb
;
4203 if (i_blocks
<= ~0U) {
4205 * i_blocks can be represnted in a 32 bit variable
4206 * as multiple of 512 bytes
4208 raw_inode
->i_blocks_lo
= cpu_to_le32(i_blocks
);
4209 raw_inode
->i_blocks_high
= 0;
4210 ei
->i_flags
&= ~EXT4_HUGE_FILE_FL
;
4213 if (!EXT4_HAS_RO_COMPAT_FEATURE(sb
, EXT4_FEATURE_RO_COMPAT_HUGE_FILE
))
4216 if (i_blocks
<= 0xffffffffffffULL
) {
4218 * i_blocks can be represented in a 48 bit variable
4219 * as multiple of 512 bytes
4221 raw_inode
->i_blocks_lo
= cpu_to_le32(i_blocks
);
4222 raw_inode
->i_blocks_high
= cpu_to_le16(i_blocks
>> 32);
4223 ei
->i_flags
&= ~EXT4_HUGE_FILE_FL
;
4225 ei
->i_flags
|= EXT4_HUGE_FILE_FL
;
4226 /* i_block is stored in file system block size */
4227 i_blocks
= i_blocks
>> (inode
->i_blkbits
- 9);
4228 raw_inode
->i_blocks_lo
= cpu_to_le32(i_blocks
);
4229 raw_inode
->i_blocks_high
= cpu_to_le16(i_blocks
>> 32);
4235 * Post the struct inode info into an on-disk inode location in the
4236 * buffer-cache. This gobbles the caller's reference to the
4237 * buffer_head in the inode location struct.
4239 * The caller must have write access to iloc->bh.
4241 static int ext4_do_update_inode(handle_t
*handle
,
4242 struct inode
*inode
,
4243 struct ext4_iloc
*iloc
)
4245 struct ext4_inode
*raw_inode
= ext4_raw_inode(iloc
);
4246 struct ext4_inode_info
*ei
= EXT4_I(inode
);
4247 struct buffer_head
*bh
= iloc
->bh
;
4248 int err
= 0, rc
, block
;
4250 /* For fields not not tracking in the in-memory inode,
4251 * initialise them to zero for new inodes. */
4252 if (ei
->i_state
& EXT4_STATE_NEW
)
4253 memset(raw_inode
, 0, EXT4_SB(inode
->i_sb
)->s_inode_size
);
4255 ext4_get_inode_flags(ei
);
4256 raw_inode
->i_mode
= cpu_to_le16(inode
->i_mode
);
4257 if (!(test_opt(inode
->i_sb
, NO_UID32
))) {
4258 raw_inode
->i_uid_low
= cpu_to_le16(low_16_bits(inode
->i_uid
));
4259 raw_inode
->i_gid_low
= cpu_to_le16(low_16_bits(inode
->i_gid
));
4261 * Fix up interoperability with old kernels. Otherwise, old inodes get
4262 * re-used with the upper 16 bits of the uid/gid intact
4265 raw_inode
->i_uid_high
=
4266 cpu_to_le16(high_16_bits(inode
->i_uid
));
4267 raw_inode
->i_gid_high
=
4268 cpu_to_le16(high_16_bits(inode
->i_gid
));
4270 raw_inode
->i_uid_high
= 0;
4271 raw_inode
->i_gid_high
= 0;
4274 raw_inode
->i_uid_low
=
4275 cpu_to_le16(fs_high2lowuid(inode
->i_uid
));
4276 raw_inode
->i_gid_low
=
4277 cpu_to_le16(fs_high2lowgid(inode
->i_gid
));
4278 raw_inode
->i_uid_high
= 0;
4279 raw_inode
->i_gid_high
= 0;
4281 raw_inode
->i_links_count
= cpu_to_le16(inode
->i_nlink
);
4283 EXT4_INODE_SET_XTIME(i_ctime
, inode
, raw_inode
);
4284 EXT4_INODE_SET_XTIME(i_mtime
, inode
, raw_inode
);
4285 EXT4_INODE_SET_XTIME(i_atime
, inode
, raw_inode
);
4286 EXT4_EINODE_SET_XTIME(i_crtime
, ei
, raw_inode
);
4288 if (ext4_inode_blocks_set(handle
, raw_inode
, ei
))
4290 raw_inode
->i_dtime
= cpu_to_le32(ei
->i_dtime
);
4291 /* clear the migrate flag in the raw_inode */
4292 raw_inode
->i_flags
= cpu_to_le32(ei
->i_flags
& ~EXT4_EXT_MIGRATE
);
4293 if (EXT4_SB(inode
->i_sb
)->s_es
->s_creator_os
!=
4294 cpu_to_le32(EXT4_OS_HURD
))
4295 raw_inode
->i_file_acl_high
=
4296 cpu_to_le16(ei
->i_file_acl
>> 32);
4297 raw_inode
->i_file_acl_lo
= cpu_to_le32(ei
->i_file_acl
);
4298 ext4_isize_set(raw_inode
, ei
->i_disksize
);
4299 if (ei
->i_disksize
> 0x7fffffffULL
) {
4300 struct super_block
*sb
= inode
->i_sb
;
4301 if (!EXT4_HAS_RO_COMPAT_FEATURE(sb
,
4302 EXT4_FEATURE_RO_COMPAT_LARGE_FILE
) ||
4303 EXT4_SB(sb
)->s_es
->s_rev_level
==
4304 cpu_to_le32(EXT4_GOOD_OLD_REV
)) {
4305 /* If this is the first large file
4306 * created, add a flag to the superblock.
4308 err
= ext4_journal_get_write_access(handle
,
4309 EXT4_SB(sb
)->s_sbh
);
4312 ext4_update_dynamic_rev(sb
);
4313 EXT4_SET_RO_COMPAT_FEATURE(sb
,
4314 EXT4_FEATURE_RO_COMPAT_LARGE_FILE
);
4317 err
= ext4_journal_dirty_metadata(handle
,
4318 EXT4_SB(sb
)->s_sbh
);
4321 raw_inode
->i_generation
= cpu_to_le32(inode
->i_generation
);
4322 if (S_ISCHR(inode
->i_mode
) || S_ISBLK(inode
->i_mode
)) {
4323 if (old_valid_dev(inode
->i_rdev
)) {
4324 raw_inode
->i_block
[0] =
4325 cpu_to_le32(old_encode_dev(inode
->i_rdev
));
4326 raw_inode
->i_block
[1] = 0;
4328 raw_inode
->i_block
[0] = 0;
4329 raw_inode
->i_block
[1] =
4330 cpu_to_le32(new_encode_dev(inode
->i_rdev
));
4331 raw_inode
->i_block
[2] = 0;
4333 } else for (block
= 0; block
< EXT4_N_BLOCKS
; block
++)
4334 raw_inode
->i_block
[block
] = ei
->i_data
[block
];
4336 raw_inode
->i_disk_version
= cpu_to_le32(inode
->i_version
);
4337 if (ei
->i_extra_isize
) {
4338 if (EXT4_FITS_IN_INODE(raw_inode
, ei
, i_version_hi
))
4339 raw_inode
->i_version_hi
=
4340 cpu_to_le32(inode
->i_version
>> 32);
4341 raw_inode
->i_extra_isize
= cpu_to_le16(ei
->i_extra_isize
);
4345 BUFFER_TRACE(bh
, "call ext4_journal_dirty_metadata");
4346 rc
= ext4_journal_dirty_metadata(handle
, bh
);
4349 ei
->i_state
&= ~EXT4_STATE_NEW
;
4353 ext4_std_error(inode
->i_sb
, err
);
4358 * ext4_write_inode()
4360 * We are called from a few places:
4362 * - Within generic_file_write() for O_SYNC files.
4363 * Here, there will be no transaction running. We wait for any running
4364 * trasnaction to commit.
4366 * - Within sys_sync(), kupdate and such.
4367 * We wait on commit, if tol to.
4369 * - Within prune_icache() (PF_MEMALLOC == true)
4370 * Here we simply return. We can't afford to block kswapd on the
4373 * In all cases it is actually safe for us to return without doing anything,
4374 * because the inode has been copied into a raw inode buffer in
4375 * ext4_mark_inode_dirty(). This is a correctness thing for O_SYNC and for
4378 * Note that we are absolutely dependent upon all inode dirtiers doing the
4379 * right thing: they *must* call mark_inode_dirty() after dirtying info in
4380 * which we are interested.
4382 * It would be a bug for them to not do this. The code:
4384 * mark_inode_dirty(inode)
4386 * inode->i_size = expr;
4388 * is in error because a kswapd-driven write_inode() could occur while
4389 * `stuff()' is running, and the new i_size will be lost. Plus the inode
4390 * will no longer be on the superblock's dirty inode list.
4392 int ext4_write_inode(struct inode
*inode
, int wait
)
4394 if (current
->flags
& PF_MEMALLOC
)
4397 if (ext4_journal_current_handle()) {
4398 jbd_debug(1, "called recursively, non-PF_MEMALLOC!\n");
4406 return ext4_force_commit(inode
->i_sb
);
4412 * Called from notify_change.
4414 * We want to trap VFS attempts to truncate the file as soon as
4415 * possible. In particular, we want to make sure that when the VFS
4416 * shrinks i_size, we put the inode on the orphan list and modify
4417 * i_disksize immediately, so that during the subsequent flushing of
4418 * dirty pages and freeing of disk blocks, we can guarantee that any
4419 * commit will leave the blocks being flushed in an unused state on
4420 * disk. (On recovery, the inode will get truncated and the blocks will
4421 * be freed, so we have a strong guarantee that no future commit will
4422 * leave these blocks visible to the user.)
4424 * Another thing we have to assure is that if we are in ordered mode
4425 * and inode is still attached to the committing transaction, we must
4426 * we start writeout of all the dirty pages which are being truncated.
4427 * This way we are sure that all the data written in the previous
4428 * transaction are already on disk (truncate waits for pages under
4431 * Called with inode->i_mutex down.
4433 int ext4_setattr(struct dentry
*dentry
, struct iattr
*attr
)
4435 struct inode
*inode
= dentry
->d_inode
;
4437 const unsigned int ia_valid
= attr
->ia_valid
;
4439 error
= inode_change_ok(inode
, attr
);
4443 if ((ia_valid
& ATTR_UID
&& attr
->ia_uid
!= inode
->i_uid
) ||
4444 (ia_valid
& ATTR_GID
&& attr
->ia_gid
!= inode
->i_gid
)) {
4447 /* (user+group)*(old+new) structure, inode write (sb,
4448 * inode block, ? - but truncate inode update has it) */
4449 handle
= ext4_journal_start(inode
, 2*(EXT4_QUOTA_INIT_BLOCKS(inode
->i_sb
)+
4450 EXT4_QUOTA_DEL_BLOCKS(inode
->i_sb
))+3);
4451 if (IS_ERR(handle
)) {
4452 error
= PTR_ERR(handle
);
4455 error
= DQUOT_TRANSFER(inode
, attr
) ? -EDQUOT
: 0;
4457 ext4_journal_stop(handle
);
4460 /* Update corresponding info in inode so that everything is in
4461 * one transaction */
4462 if (attr
->ia_valid
& ATTR_UID
)
4463 inode
->i_uid
= attr
->ia_uid
;
4464 if (attr
->ia_valid
& ATTR_GID
)
4465 inode
->i_gid
= attr
->ia_gid
;
4466 error
= ext4_mark_inode_dirty(handle
, inode
);
4467 ext4_journal_stop(handle
);
4470 if (attr
->ia_valid
& ATTR_SIZE
) {
4471 if (!(EXT4_I(inode
)->i_flags
& EXT4_EXTENTS_FL
)) {
4472 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
4474 if (attr
->ia_size
> sbi
->s_bitmap_maxbytes
) {
4481 if (S_ISREG(inode
->i_mode
) &&
4482 attr
->ia_valid
& ATTR_SIZE
&& attr
->ia_size
< inode
->i_size
) {
4485 handle
= ext4_journal_start(inode
, 3);
4486 if (IS_ERR(handle
)) {
4487 error
= PTR_ERR(handle
);
4491 error
= ext4_orphan_add(handle
, inode
);
4492 EXT4_I(inode
)->i_disksize
= attr
->ia_size
;
4493 rc
= ext4_mark_inode_dirty(handle
, inode
);
4496 ext4_journal_stop(handle
);
4498 if (ext4_should_order_data(inode
)) {
4499 error
= ext4_begin_ordered_truncate(inode
,
4502 /* Do as much error cleanup as possible */
4503 handle
= ext4_journal_start(inode
, 3);
4504 if (IS_ERR(handle
)) {
4505 ext4_orphan_del(NULL
, inode
);
4508 ext4_orphan_del(handle
, inode
);
4509 ext4_journal_stop(handle
);
4515 rc
= inode_setattr(inode
, attr
);
4517 /* If inode_setattr's call to ext4_truncate failed to get a
4518 * transaction handle at all, we need to clean up the in-core
4519 * orphan list manually. */
4521 ext4_orphan_del(NULL
, inode
);
4523 if (!rc
&& (ia_valid
& ATTR_MODE
))
4524 rc
= ext4_acl_chmod(inode
);
4527 ext4_std_error(inode
->i_sb
, error
);
4533 int ext4_getattr(struct vfsmount
*mnt
, struct dentry
*dentry
,
4536 struct inode
*inode
;
4537 unsigned long delalloc_blocks
;
4539 inode
= dentry
->d_inode
;
4540 generic_fillattr(inode
, stat
);
4543 * We can't update i_blocks if the block allocation is delayed
4544 * otherwise in the case of system crash before the real block
4545 * allocation is done, we will have i_blocks inconsistent with
4546 * on-disk file blocks.
4547 * We always keep i_blocks updated together with real
4548 * allocation. But to not confuse with user, stat
4549 * will return the blocks that include the delayed allocation
4550 * blocks for this file.
4552 spin_lock(&EXT4_I(inode
)->i_block_reservation_lock
);
4553 delalloc_blocks
= EXT4_I(inode
)->i_reserved_data_blocks
;
4554 spin_unlock(&EXT4_I(inode
)->i_block_reservation_lock
);
4556 stat
->blocks
+= (delalloc_blocks
<< inode
->i_sb
->s_blocksize_bits
)>>9;
4560 static int ext4_indirect_trans_blocks(struct inode
*inode
, int nrblocks
,
4565 /* if nrblocks are contiguous */
4568 * With N contiguous data blocks, it need at most
4569 * N/EXT4_ADDR_PER_BLOCK(inode->i_sb) indirect blocks
4570 * 2 dindirect blocks
4573 indirects
= nrblocks
/ EXT4_ADDR_PER_BLOCK(inode
->i_sb
);
4574 return indirects
+ 3;
4577 * if nrblocks are not contiguous, worse case, each block touch
4578 * a indirect block, and each indirect block touch a double indirect
4579 * block, plus a triple indirect block
4581 indirects
= nrblocks
* 2 + 1;
4585 static int ext4_index_trans_blocks(struct inode
*inode
, int nrblocks
, int chunk
)
4587 if (!(EXT4_I(inode
)->i_flags
& EXT4_EXTENTS_FL
))
4588 return ext4_indirect_trans_blocks(inode
, nrblocks
, chunk
);
4589 return ext4_ext_index_trans_blocks(inode
, nrblocks
, chunk
);
4593 * Account for index blocks, block groups bitmaps and block group
4594 * descriptor blocks if modify datablocks and index blocks
4595 * worse case, the indexs blocks spread over different block groups
4597 * If datablocks are discontiguous, they are possible to spread over
4598 * different block groups too. If they are contiugous, with flexbg,
4599 * they could still across block group boundary.
4601 * Also account for superblock, inode, quota and xattr blocks
4603 int ext4_meta_trans_blocks(struct inode
*inode
, int nrblocks
, int chunk
)
4605 int groups
, gdpblocks
;
4610 * How many index blocks need to touch to modify nrblocks?
4611 * The "Chunk" flag indicating whether the nrblocks is
4612 * physically contiguous on disk
4614 * For Direct IO and fallocate, they calls get_block to allocate
4615 * one single extent at a time, so they could set the "Chunk" flag
4617 idxblocks
= ext4_index_trans_blocks(inode
, nrblocks
, chunk
);
4622 * Now let's see how many group bitmaps and group descriptors need
4632 if (groups
> EXT4_SB(inode
->i_sb
)->s_groups_count
)
4633 groups
= EXT4_SB(inode
->i_sb
)->s_groups_count
;
4634 if (groups
> EXT4_SB(inode
->i_sb
)->s_gdb_count
)
4635 gdpblocks
= EXT4_SB(inode
->i_sb
)->s_gdb_count
;
4637 /* bitmaps and block group descriptor blocks */
4638 ret
+= groups
+ gdpblocks
;
4640 /* Blocks for super block, inode, quota and xattr blocks */
4641 ret
+= EXT4_META_TRANS_BLOCKS(inode
->i_sb
);
4647 * Calulate the total number of credits to reserve to fit
4648 * the modification of a single pages into a single transaction,
4649 * which may include multiple chunks of block allocations.
4651 * This could be called via ext4_write_begin()
4653 * We need to consider the worse case, when
4654 * one new block per extent.
4656 int ext4_writepage_trans_blocks(struct inode
*inode
)
4658 int bpp
= ext4_journal_blocks_per_page(inode
);
4661 ret
= ext4_meta_trans_blocks(inode
, bpp
, 0);
4663 /* Account for data blocks for journalled mode */
4664 if (ext4_should_journal_data(inode
))
4670 * Calculate the journal credits for a chunk of data modification.
4672 * This is called from DIO, fallocate or whoever calling
4673 * ext4_get_blocks_wrap() to map/allocate a chunk of contigous disk blocks.
4675 * journal buffers for data blocks are not included here, as DIO
4676 * and fallocate do no need to journal data buffers.
4678 int ext4_chunk_trans_blocks(struct inode
*inode
, int nrblocks
)
4680 return ext4_meta_trans_blocks(inode
, nrblocks
, 1);
4684 * The caller must have previously called ext4_reserve_inode_write().
4685 * Give this, we know that the caller already has write access to iloc->bh.
4687 int ext4_mark_iloc_dirty(handle_t
*handle
,
4688 struct inode
*inode
, struct ext4_iloc
*iloc
)
4692 if (test_opt(inode
->i_sb
, I_VERSION
))
4693 inode_inc_iversion(inode
);
4695 /* the do_update_inode consumes one bh->b_count */
4698 /* ext4_do_update_inode() does jbd2_journal_dirty_metadata */
4699 err
= ext4_do_update_inode(handle
, inode
, iloc
);
4705 * On success, We end up with an outstanding reference count against
4706 * iloc->bh. This _must_ be cleaned up later.
4710 ext4_reserve_inode_write(handle_t
*handle
, struct inode
*inode
,
4711 struct ext4_iloc
*iloc
)
4715 err
= ext4_get_inode_loc(inode
, iloc
);
4717 BUFFER_TRACE(iloc
->bh
, "get_write_access");
4718 err
= ext4_journal_get_write_access(handle
, iloc
->bh
);
4725 ext4_std_error(inode
->i_sb
, err
);
4730 * Expand an inode by new_extra_isize bytes.
4731 * Returns 0 on success or negative error number on failure.
4733 static int ext4_expand_extra_isize(struct inode
*inode
,
4734 unsigned int new_extra_isize
,
4735 struct ext4_iloc iloc
,
4738 struct ext4_inode
*raw_inode
;
4739 struct ext4_xattr_ibody_header
*header
;
4740 struct ext4_xattr_entry
*entry
;
4742 if (EXT4_I(inode
)->i_extra_isize
>= new_extra_isize
)
4745 raw_inode
= ext4_raw_inode(&iloc
);
4747 header
= IHDR(inode
, raw_inode
);
4748 entry
= IFIRST(header
);
4750 /* No extended attributes present */
4751 if (!(EXT4_I(inode
)->i_state
& EXT4_STATE_XATTR
) ||
4752 header
->h_magic
!= cpu_to_le32(EXT4_XATTR_MAGIC
)) {
4753 memset((void *)raw_inode
+ EXT4_GOOD_OLD_INODE_SIZE
, 0,
4755 EXT4_I(inode
)->i_extra_isize
= new_extra_isize
;
4759 /* try to expand with EAs present */
4760 return ext4_expand_extra_isize_ea(inode
, new_extra_isize
,
4765 * What we do here is to mark the in-core inode as clean with respect to inode
4766 * dirtiness (it may still be data-dirty).
4767 * This means that the in-core inode may be reaped by prune_icache
4768 * without having to perform any I/O. This is a very good thing,
4769 * because *any* task may call prune_icache - even ones which
4770 * have a transaction open against a different journal.
4772 * Is this cheating? Not really. Sure, we haven't written the
4773 * inode out, but prune_icache isn't a user-visible syncing function.
4774 * Whenever the user wants stuff synced (sys_sync, sys_msync, sys_fsync)
4775 * we start and wait on commits.
4777 * Is this efficient/effective? Well, we're being nice to the system
4778 * by cleaning up our inodes proactively so they can be reaped
4779 * without I/O. But we are potentially leaving up to five seconds'
4780 * worth of inodes floating about which prune_icache wants us to
4781 * write out. One way to fix that would be to get prune_icache()
4782 * to do a write_super() to free up some memory. It has the desired
4785 int ext4_mark_inode_dirty(handle_t
*handle
, struct inode
*inode
)
4787 struct ext4_iloc iloc
;
4788 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
4789 static unsigned int mnt_count
;
4793 err
= ext4_reserve_inode_write(handle
, inode
, &iloc
);
4794 if (EXT4_I(inode
)->i_extra_isize
< sbi
->s_want_extra_isize
&&
4795 !(EXT4_I(inode
)->i_state
& EXT4_STATE_NO_EXPAND
)) {
4797 * We need extra buffer credits since we may write into EA block
4798 * with this same handle. If journal_extend fails, then it will
4799 * only result in a minor loss of functionality for that inode.
4800 * If this is felt to be critical, then e2fsck should be run to
4801 * force a large enough s_min_extra_isize.
4803 if ((jbd2_journal_extend(handle
,
4804 EXT4_DATA_TRANS_BLOCKS(inode
->i_sb
))) == 0) {
4805 ret
= ext4_expand_extra_isize(inode
,
4806 sbi
->s_want_extra_isize
,
4809 EXT4_I(inode
)->i_state
|= EXT4_STATE_NO_EXPAND
;
4811 le16_to_cpu(sbi
->s_es
->s_mnt_count
)) {
4812 ext4_warning(inode
->i_sb
, __func__
,
4813 "Unable to expand inode %lu. Delete"
4814 " some EAs or run e2fsck.",
4817 le16_to_cpu(sbi
->s_es
->s_mnt_count
);
4823 err
= ext4_mark_iloc_dirty(handle
, inode
, &iloc
);
4828 * ext4_dirty_inode() is called from __mark_inode_dirty()
4830 * We're really interested in the case where a file is being extended.
4831 * i_size has been changed by generic_commit_write() and we thus need
4832 * to include the updated inode in the current transaction.
4834 * Also, DQUOT_ALLOC_SPACE() will always dirty the inode when blocks
4835 * are allocated to the file.
4837 * If the inode is marked synchronous, we don't honour that here - doing
4838 * so would cause a commit on atime updates, which we don't bother doing.
4839 * We handle synchronous inodes at the highest possible level.
4841 void ext4_dirty_inode(struct inode
*inode
)
4843 handle_t
*current_handle
= ext4_journal_current_handle();
4846 handle
= ext4_journal_start(inode
, 2);
4849 if (current_handle
&&
4850 current_handle
->h_transaction
!= handle
->h_transaction
) {
4851 /* This task has a transaction open against a different fs */
4852 printk(KERN_EMERG
"%s: transactions do not match!\n",
4855 jbd_debug(5, "marking dirty. outer handle=%p\n",
4857 ext4_mark_inode_dirty(handle
, inode
);
4859 ext4_journal_stop(handle
);
4866 * Bind an inode's backing buffer_head into this transaction, to prevent
4867 * it from being flushed to disk early. Unlike
4868 * ext4_reserve_inode_write, this leaves behind no bh reference and
4869 * returns no iloc structure, so the caller needs to repeat the iloc
4870 * lookup to mark the inode dirty later.
4872 static int ext4_pin_inode(handle_t
*handle
, struct inode
*inode
)
4874 struct ext4_iloc iloc
;
4878 err
= ext4_get_inode_loc(inode
, &iloc
);
4880 BUFFER_TRACE(iloc
.bh
, "get_write_access");
4881 err
= jbd2_journal_get_write_access(handle
, iloc
.bh
);
4883 err
= ext4_journal_dirty_metadata(handle
,
4888 ext4_std_error(inode
->i_sb
, err
);
4893 int ext4_change_inode_journal_flag(struct inode
*inode
, int val
)
4900 * We have to be very careful here: changing a data block's
4901 * journaling status dynamically is dangerous. If we write a
4902 * data block to the journal, change the status and then delete
4903 * that block, we risk forgetting to revoke the old log record
4904 * from the journal and so a subsequent replay can corrupt data.
4905 * So, first we make sure that the journal is empty and that
4906 * nobody is changing anything.
4909 journal
= EXT4_JOURNAL(inode
);
4910 if (is_journal_aborted(journal
))
4913 jbd2_journal_lock_updates(journal
);
4914 jbd2_journal_flush(journal
);
4917 * OK, there are no updates running now, and all cached data is
4918 * synced to disk. We are now in a completely consistent state
4919 * which doesn't have anything in the journal, and we know that
4920 * no filesystem updates are running, so it is safe to modify
4921 * the inode's in-core data-journaling state flag now.
4925 EXT4_I(inode
)->i_flags
|= EXT4_JOURNAL_DATA_FL
;
4927 EXT4_I(inode
)->i_flags
&= ~EXT4_JOURNAL_DATA_FL
;
4928 ext4_set_aops(inode
);
4930 jbd2_journal_unlock_updates(journal
);
4932 /* Finally we can mark the inode as dirty. */
4934 handle
= ext4_journal_start(inode
, 1);
4936 return PTR_ERR(handle
);
4938 err
= ext4_mark_inode_dirty(handle
, inode
);
4940 ext4_journal_stop(handle
);
4941 ext4_std_error(inode
->i_sb
, err
);
4946 static int ext4_bh_unmapped(handle_t
*handle
, struct buffer_head
*bh
)
4948 return !buffer_mapped(bh
);
4951 int ext4_page_mkwrite(struct vm_area_struct
*vma
, struct page
*page
)
4957 struct file
*file
= vma
->vm_file
;
4958 struct inode
*inode
= file
->f_path
.dentry
->d_inode
;
4959 struct address_space
*mapping
= inode
->i_mapping
;
4962 * Get i_alloc_sem to stop truncates messing with the inode. We cannot
4963 * get i_mutex because we are already holding mmap_sem.
4965 down_read(&inode
->i_alloc_sem
);
4966 size
= i_size_read(inode
);
4967 if (page
->mapping
!= mapping
|| size
<= page_offset(page
)
4968 || !PageUptodate(page
)) {
4969 /* page got truncated from under us? */
4973 if (PageMappedToDisk(page
))
4976 if (page
->index
== size
>> PAGE_CACHE_SHIFT
)
4977 len
= size
& ~PAGE_CACHE_MASK
;
4979 len
= PAGE_CACHE_SIZE
;
4981 if (page_has_buffers(page
)) {
4982 /* return if we have all the buffers mapped */
4983 if (!walk_page_buffers(NULL
, page_buffers(page
), 0, len
, NULL
,
4988 * OK, we need to fill the hole... Do write_begin write_end
4989 * to do block allocation/reservation.We are not holding
4990 * inode.i__mutex here. That allow * parallel write_begin,
4991 * write_end call. lock_page prevent this from happening
4992 * on the same page though
4994 ret
= mapping
->a_ops
->write_begin(file
, mapping
, page_offset(page
),
4995 len
, AOP_FLAG_UNINTERRUPTIBLE
, &page
, &fsdata
);
4998 ret
= mapping
->a_ops
->write_end(file
, mapping
, page_offset(page
),
4999 len
, len
, page
, fsdata
);
5004 up_read(&inode
->i_alloc_sem
);