2 * linux/fs/ext3/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 ext3_get_block() by Al Viro, 2000
25 #include <linux/module.h>
27 #include <linux/time.h>
28 #include <linux/ext3_jbd.h>
29 #include <linux/jbd.h>
30 #include <linux/highuid.h>
31 #include <linux/pagemap.h>
32 #include <linux/quotaops.h>
33 #include <linux/string.h>
34 #include <linux/buffer_head.h>
35 #include <linux/writeback.h>
36 #include <linux/mpage.h>
37 #include <linux/uio.h>
38 #include <linux/bio.h>
39 #include <linux/fiemap.h>
40 #include <linux/namei.h>
44 static int ext3_writepage_trans_blocks(struct inode
*inode
);
47 * Test whether an inode is a fast symlink.
49 static int ext3_inode_is_fast_symlink(struct inode
*inode
)
51 int ea_blocks
= EXT3_I(inode
)->i_file_acl
?
52 (inode
->i_sb
->s_blocksize
>> 9) : 0;
54 return (S_ISLNK(inode
->i_mode
) && inode
->i_blocks
- ea_blocks
== 0);
58 * The ext3 forget function must perform a revoke if we are freeing data
59 * which has been journaled. Metadata (eg. indirect blocks) must be
60 * revoked in all cases.
62 * "bh" may be NULL: a metadata block may have been freed from memory
63 * but there may still be a record of it in the journal, and that record
64 * still needs to be revoked.
66 int ext3_forget(handle_t
*handle
, int is_metadata
, struct inode
*inode
,
67 struct buffer_head
*bh
, ext3_fsblk_t blocknr
)
73 BUFFER_TRACE(bh
, "enter");
75 jbd_debug(4, "forgetting bh %p: is_metadata = %d, mode %o, "
77 bh
, is_metadata
, inode
->i_mode
,
78 test_opt(inode
->i_sb
, DATA_FLAGS
));
80 /* Never use the revoke function if we are doing full data
81 * journaling: there is no need to, and a V1 superblock won't
82 * support it. Otherwise, only skip the revoke on un-journaled
85 if (test_opt(inode
->i_sb
, DATA_FLAGS
) == EXT3_MOUNT_JOURNAL_DATA
||
86 (!is_metadata
&& !ext3_should_journal_data(inode
))) {
88 BUFFER_TRACE(bh
, "call journal_forget");
89 return ext3_journal_forget(handle
, bh
);
95 * data!=journal && (is_metadata || should_journal_data(inode))
97 BUFFER_TRACE(bh
, "call ext3_journal_revoke");
98 err
= ext3_journal_revoke(handle
, blocknr
, bh
);
100 ext3_abort(inode
->i_sb
, __func__
,
101 "error %d when attempting revoke", err
);
102 BUFFER_TRACE(bh
, "exit");
107 * Work out how many blocks we need to proceed with the next chunk of a
108 * truncate transaction.
110 static unsigned long blocks_for_truncate(struct inode
*inode
)
112 unsigned long needed
;
114 needed
= inode
->i_blocks
>> (inode
->i_sb
->s_blocksize_bits
- 9);
116 /* Give ourselves just enough room to cope with inodes in which
117 * i_blocks is corrupt: we've seen disk corruptions in the past
118 * which resulted in random data in an inode which looked enough
119 * like a regular file for ext3 to try to delete it. Things
120 * will go a bit crazy if that happens, but at least we should
121 * try not to panic the whole kernel. */
125 /* But we need to bound the transaction so we don't overflow the
127 if (needed
> EXT3_MAX_TRANS_DATA
)
128 needed
= EXT3_MAX_TRANS_DATA
;
130 return EXT3_DATA_TRANS_BLOCKS(inode
->i_sb
) + needed
;
134 * Truncate transactions can be complex and absolutely huge. So we need to
135 * be able to restart the transaction at a conventient checkpoint to make
136 * sure we don't overflow the journal.
138 * start_transaction gets us a new handle for a truncate transaction,
139 * and extend_transaction tries to extend the existing one a bit. If
140 * extend fails, we need to propagate the failure up and restart the
141 * transaction in the top-level truncate loop. --sct
143 static handle_t
*start_transaction(struct inode
*inode
)
147 result
= ext3_journal_start(inode
, blocks_for_truncate(inode
));
151 ext3_std_error(inode
->i_sb
, PTR_ERR(result
));
156 * Try to extend this transaction for the purposes of truncation.
158 * Returns 0 if we managed to create more room. If we can't create more
159 * room, and the transaction must be restarted we return 1.
161 static int try_to_extend_transaction(handle_t
*handle
, struct inode
*inode
)
163 if (handle
->h_buffer_credits
> EXT3_RESERVE_TRANS_BLOCKS
)
165 if (!ext3_journal_extend(handle
, blocks_for_truncate(inode
)))
171 * Restart the transaction associated with *handle. This does a commit,
172 * so before we call here everything must be consistently dirtied against
175 static int truncate_restart_transaction(handle_t
*handle
, struct inode
*inode
)
179 jbd_debug(2, "restarting handle %p\n", handle
);
181 * Drop truncate_mutex to avoid deadlock with ext3_get_blocks_handle
182 * At this moment, get_block can be called only for blocks inside
183 * i_size since page cache has been already dropped and writes are
184 * blocked by i_mutex. So we can safely drop the truncate_mutex.
186 mutex_unlock(&EXT3_I(inode
)->truncate_mutex
);
187 ret
= ext3_journal_restart(handle
, blocks_for_truncate(inode
));
188 mutex_lock(&EXT3_I(inode
)->truncate_mutex
);
193 * Called at the last iput() if i_nlink is zero.
195 void ext3_delete_inode (struct inode
* inode
)
199 if (!is_bad_inode(inode
))
200 dquot_initialize(inode
);
202 truncate_inode_pages(&inode
->i_data
, 0);
204 if (is_bad_inode(inode
))
207 handle
= start_transaction(inode
);
208 if (IS_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 ext3_orphan_del(NULL
, inode
);
222 ext3_truncate(inode
);
224 * Kill off the orphan record which ext3_truncate created.
225 * AKPM: I think this can be inside the above `if'.
226 * Note that ext3_orphan_del() has to be able to cope with the
227 * deletion of a non-existent orphan - this is because we don't
228 * know if ext3_truncate() actually created an orphan record.
229 * (Well, we could do this if we need to, but heck - it works)
231 ext3_orphan_del(handle
, inode
);
232 EXT3_I(inode
)->i_dtime
= get_seconds();
235 * One subtle ordering requirement: if anything has gone wrong
236 * (transaction abort, IO errors, whatever), then we can still
237 * do these next steps (the fs will already have been marked as
238 * having errors), but we can't free the inode if the mark_dirty
241 if (ext3_mark_inode_dirty(handle
, inode
))
242 /* If that failed, just do the required in-core inode clear. */
245 ext3_free_inode(handle
, inode
);
246 ext3_journal_stop(handle
);
249 clear_inode(inode
); /* We must guarantee clearing of inode... */
255 struct buffer_head
*bh
;
258 static inline void add_chain(Indirect
*p
, struct buffer_head
*bh
, __le32
*v
)
260 p
->key
= *(p
->p
= v
);
264 static int verify_chain(Indirect
*from
, Indirect
*to
)
266 while (from
<= to
&& from
->key
== *from
->p
)
272 * ext3_block_to_path - parse the block number into array of offsets
273 * @inode: inode in question (we are only interested in its superblock)
274 * @i_block: block number to be parsed
275 * @offsets: array to store the offsets in
276 * @boundary: set this non-zero if the referred-to block is likely to be
277 * followed (on disk) by an indirect block.
279 * To store the locations of file's data ext3 uses a data structure common
280 * for UNIX filesystems - tree of pointers anchored in the inode, with
281 * data blocks at leaves and indirect blocks in intermediate nodes.
282 * This function translates the block number into path in that tree -
283 * return value is the path length and @offsets[n] is the offset of
284 * pointer to (n+1)th node in the nth one. If @block is out of range
285 * (negative or too large) warning is printed and zero returned.
287 * Note: function doesn't find node addresses, so no IO is needed. All
288 * we need to know is the capacity of indirect blocks (taken from the
293 * Portability note: the last comparison (check that we fit into triple
294 * indirect block) is spelled differently, because otherwise on an
295 * architecture with 32-bit longs and 8Kb pages we might get into trouble
296 * if our filesystem had 8Kb blocks. We might use long long, but that would
297 * kill us on x86. Oh, well, at least the sign propagation does not matter -
298 * i_block would have to be negative in the very beginning, so we would not
302 static int ext3_block_to_path(struct inode
*inode
,
303 long i_block
, int offsets
[4], int *boundary
)
305 int ptrs
= EXT3_ADDR_PER_BLOCK(inode
->i_sb
);
306 int ptrs_bits
= EXT3_ADDR_PER_BLOCK_BITS(inode
->i_sb
);
307 const long direct_blocks
= EXT3_NDIR_BLOCKS
,
308 indirect_blocks
= ptrs
,
309 double_blocks
= (1 << (ptrs_bits
* 2));
314 ext3_warning (inode
->i_sb
, "ext3_block_to_path", "block < 0");
315 } else if (i_block
< direct_blocks
) {
316 offsets
[n
++] = i_block
;
317 final
= direct_blocks
;
318 } else if ( (i_block
-= direct_blocks
) < indirect_blocks
) {
319 offsets
[n
++] = EXT3_IND_BLOCK
;
320 offsets
[n
++] = i_block
;
322 } else if ((i_block
-= indirect_blocks
) < double_blocks
) {
323 offsets
[n
++] = EXT3_DIND_BLOCK
;
324 offsets
[n
++] = i_block
>> ptrs_bits
;
325 offsets
[n
++] = i_block
& (ptrs
- 1);
327 } else if (((i_block
-= double_blocks
) >> (ptrs_bits
* 2)) < ptrs
) {
328 offsets
[n
++] = EXT3_TIND_BLOCK
;
329 offsets
[n
++] = i_block
>> (ptrs_bits
* 2);
330 offsets
[n
++] = (i_block
>> ptrs_bits
) & (ptrs
- 1);
331 offsets
[n
++] = i_block
& (ptrs
- 1);
334 ext3_warning(inode
->i_sb
, "ext3_block_to_path", "block > big");
337 *boundary
= final
- 1 - (i_block
& (ptrs
- 1));
342 * ext3_get_branch - read the chain of indirect blocks leading to data
343 * @inode: inode in question
344 * @depth: depth of the chain (1 - direct pointer, etc.)
345 * @offsets: offsets of pointers in inode/indirect blocks
346 * @chain: place to store the result
347 * @err: here we store the error value
349 * Function fills the array of triples <key, p, bh> and returns %NULL
350 * if everything went OK or the pointer to the last filled triple
351 * (incomplete one) otherwise. Upon the return chain[i].key contains
352 * the number of (i+1)-th block in the chain (as it is stored in memory,
353 * i.e. little-endian 32-bit), chain[i].p contains the address of that
354 * number (it points into struct inode for i==0 and into the bh->b_data
355 * for i>0) and chain[i].bh points to the buffer_head of i-th indirect
356 * block for i>0 and NULL for i==0. In other words, it holds the block
357 * numbers of the chain, addresses they were taken from (and where we can
358 * verify that chain did not change) and buffer_heads hosting these
361 * Function stops when it stumbles upon zero pointer (absent block)
362 * (pointer to last triple returned, *@err == 0)
363 * or when it gets an IO error reading an indirect block
364 * (ditto, *@err == -EIO)
365 * or when it notices that chain had been changed while it was reading
366 * (ditto, *@err == -EAGAIN)
367 * or when it reads all @depth-1 indirect blocks successfully and finds
368 * the whole chain, all way to the data (returns %NULL, *err == 0).
370 static Indirect
*ext3_get_branch(struct inode
*inode
, int depth
, int *offsets
,
371 Indirect chain
[4], int *err
)
373 struct super_block
*sb
= inode
->i_sb
;
375 struct buffer_head
*bh
;
378 /* i_data is not going away, no lock needed */
379 add_chain (chain
, NULL
, EXT3_I(inode
)->i_data
+ *offsets
);
383 bh
= sb_bread(sb
, le32_to_cpu(p
->key
));
386 /* Reader: pointers */
387 if (!verify_chain(chain
, p
))
389 add_chain(++p
, bh
, (__le32
*)bh
->b_data
+ *++offsets
);
407 * ext3_find_near - find a place for allocation with sufficient locality
409 * @ind: descriptor of indirect block.
411 * This function returns the preferred place for block allocation.
412 * It is used when heuristic for sequential allocation fails.
414 * + if there is a block to the left of our position - allocate near it.
415 * + if pointer will live in indirect block - allocate near that block.
416 * + if pointer will live in inode - allocate in the same
419 * In the latter case we colour the starting block by the callers PID to
420 * prevent it from clashing with concurrent allocations for a different inode
421 * in the same block group. The PID is used here so that functionally related
422 * files will be close-by on-disk.
424 * Caller must make sure that @ind is valid and will stay that way.
426 static ext3_fsblk_t
ext3_find_near(struct inode
*inode
, Indirect
*ind
)
428 struct ext3_inode_info
*ei
= EXT3_I(inode
);
429 __le32
*start
= ind
->bh
? (__le32
*) ind
->bh
->b_data
: ei
->i_data
;
431 ext3_fsblk_t bg_start
;
432 ext3_grpblk_t colour
;
434 /* Try to find previous block */
435 for (p
= ind
->p
- 1; p
>= start
; p
--) {
437 return le32_to_cpu(*p
);
440 /* No such thing, so let's try location of indirect block */
442 return ind
->bh
->b_blocknr
;
445 * It is going to be referred to from the inode itself? OK, just put it
446 * into the same cylinder group then.
448 bg_start
= ext3_group_first_block_no(inode
->i_sb
, ei
->i_block_group
);
449 colour
= (current
->pid
% 16) *
450 (EXT3_BLOCKS_PER_GROUP(inode
->i_sb
) / 16);
451 return bg_start
+ colour
;
455 * ext3_find_goal - find a preferred place for allocation.
457 * @block: block we want
458 * @partial: pointer to the last triple within a chain
460 * Normally this function find the preferred place for block allocation,
464 static ext3_fsblk_t
ext3_find_goal(struct inode
*inode
, long block
,
467 struct ext3_block_alloc_info
*block_i
;
469 block_i
= EXT3_I(inode
)->i_block_alloc_info
;
472 * try the heuristic for sequential allocation,
473 * failing that at least try to get decent locality.
475 if (block_i
&& (block
== block_i
->last_alloc_logical_block
+ 1)
476 && (block_i
->last_alloc_physical_block
!= 0)) {
477 return block_i
->last_alloc_physical_block
+ 1;
480 return ext3_find_near(inode
, partial
);
484 * ext3_blks_to_allocate: Look up the block map and count the number
485 * of direct blocks need to be allocated for the given branch.
487 * @branch: chain of indirect blocks
488 * @k: number of blocks need for indirect blocks
489 * @blks: number of data blocks to be mapped.
490 * @blocks_to_boundary: the offset in the indirect block
492 * return the total number of blocks to be allocate, including the
493 * direct and indirect blocks.
495 static int ext3_blks_to_allocate(Indirect
*branch
, int k
, unsigned long blks
,
496 int blocks_to_boundary
)
498 unsigned long count
= 0;
501 * Simple case, [t,d]Indirect block(s) has not allocated yet
502 * then it's clear blocks on that path have not allocated
505 /* right now we don't handle cross boundary allocation */
506 if (blks
< blocks_to_boundary
+ 1)
509 count
+= blocks_to_boundary
+ 1;
514 while (count
< blks
&& count
<= blocks_to_boundary
&&
515 le32_to_cpu(*(branch
[0].p
+ count
)) == 0) {
522 * ext3_alloc_blocks: multiple allocate blocks needed for a branch
523 * @indirect_blks: the number of blocks need to allocate for indirect
526 * @new_blocks: on return it will store the new block numbers for
527 * the indirect blocks(if needed) and the first direct block,
528 * @blks: on return it will store the total number of allocated
531 static int ext3_alloc_blocks(handle_t
*handle
, struct inode
*inode
,
532 ext3_fsblk_t goal
, int indirect_blks
, int blks
,
533 ext3_fsblk_t new_blocks
[4], int *err
)
536 unsigned long count
= 0;
538 ext3_fsblk_t current_block
= 0;
542 * Here we try to allocate the requested multiple blocks at once,
543 * on a best-effort basis.
544 * To build a branch, we should allocate blocks for
545 * the indirect blocks(if not allocated yet), and at least
546 * the first direct block of this branch. That's the
547 * minimum number of blocks need to allocate(required)
549 target
= blks
+ indirect_blks
;
553 /* allocating blocks for indirect blocks and direct blocks */
554 current_block
= ext3_new_blocks(handle
,inode
,goal
,&count
,err
);
559 /* allocate blocks for indirect blocks */
560 while (index
< indirect_blks
&& count
) {
561 new_blocks
[index
++] = current_block
++;
569 /* save the new block number for the first direct block */
570 new_blocks
[index
] = current_block
;
572 /* total number of blocks allocated for direct blocks */
577 for (i
= 0; i
<index
; i
++)
578 ext3_free_blocks(handle
, inode
, new_blocks
[i
], 1);
583 * ext3_alloc_branch - allocate and set up a chain of blocks.
585 * @indirect_blks: number of allocated indirect blocks
586 * @blks: number of allocated direct blocks
587 * @offsets: offsets (in the blocks) to store the pointers to next.
588 * @branch: place to store the chain in.
590 * This function allocates blocks, zeroes out all but the last one,
591 * links them into chain and (if we are synchronous) writes them to disk.
592 * In other words, it prepares a branch that can be spliced onto the
593 * inode. It stores the information about that chain in the branch[], in
594 * the same format as ext3_get_branch() would do. We are calling it after
595 * we had read the existing part of chain and partial points to the last
596 * triple of that (one with zero ->key). Upon the exit we have the same
597 * picture as after the successful ext3_get_block(), except that in one
598 * place chain is disconnected - *branch->p is still zero (we did not
599 * set the last link), but branch->key contains the number that should
600 * be placed into *branch->p to fill that gap.
602 * If allocation fails we free all blocks we've allocated (and forget
603 * their buffer_heads) and return the error value the from failed
604 * ext3_alloc_block() (normally -ENOSPC). Otherwise we set the chain
605 * as described above and return 0.
607 static int ext3_alloc_branch(handle_t
*handle
, struct inode
*inode
,
608 int indirect_blks
, int *blks
, ext3_fsblk_t goal
,
609 int *offsets
, Indirect
*branch
)
611 int blocksize
= inode
->i_sb
->s_blocksize
;
614 struct buffer_head
*bh
;
616 ext3_fsblk_t new_blocks
[4];
617 ext3_fsblk_t current_block
;
619 num
= ext3_alloc_blocks(handle
, inode
, goal
, indirect_blks
,
620 *blks
, new_blocks
, &err
);
624 branch
[0].key
= cpu_to_le32(new_blocks
[0]);
626 * metadata blocks and data blocks are allocated.
628 for (n
= 1; n
<= indirect_blks
; n
++) {
630 * Get buffer_head for parent block, zero it out
631 * and set the pointer to new one, then send
634 bh
= sb_getblk(inode
->i_sb
, new_blocks
[n
-1]);
637 BUFFER_TRACE(bh
, "call get_create_access");
638 err
= ext3_journal_get_create_access(handle
, bh
);
645 memset(bh
->b_data
, 0, blocksize
);
646 branch
[n
].p
= (__le32
*) bh
->b_data
+ offsets
[n
];
647 branch
[n
].key
= cpu_to_le32(new_blocks
[n
]);
648 *branch
[n
].p
= branch
[n
].key
;
649 if ( n
== indirect_blks
) {
650 current_block
= new_blocks
[n
];
652 * End of chain, update the last new metablock of
653 * the chain to point to the new allocated
654 * data blocks numbers
656 for (i
=1; i
< num
; i
++)
657 *(branch
[n
].p
+ i
) = cpu_to_le32(++current_block
);
659 BUFFER_TRACE(bh
, "marking uptodate");
660 set_buffer_uptodate(bh
);
663 BUFFER_TRACE(bh
, "call ext3_journal_dirty_metadata");
664 err
= ext3_journal_dirty_metadata(handle
, bh
);
671 /* Allocation failed, free what we already allocated */
672 for (i
= 1; i
<= n
; i
++) {
673 BUFFER_TRACE(branch
[i
].bh
, "call journal_forget");
674 ext3_journal_forget(handle
, branch
[i
].bh
);
676 for (i
= 0; i
<indirect_blks
; i
++)
677 ext3_free_blocks(handle
, inode
, new_blocks
[i
], 1);
679 ext3_free_blocks(handle
, inode
, new_blocks
[i
], num
);
685 * ext3_splice_branch - splice the allocated branch onto inode.
687 * @block: (logical) number of block we are adding
688 * @chain: chain of indirect blocks (with a missing link - see
690 * @where: location of missing link
691 * @num: number of indirect blocks we are adding
692 * @blks: number of direct blocks we are adding
694 * This function fills the missing link and does all housekeeping needed in
695 * inode (->i_blocks, etc.). In case of success we end up with the full
696 * chain to new block and return 0.
698 static int ext3_splice_branch(handle_t
*handle
, struct inode
*inode
,
699 long block
, Indirect
*where
, int num
, int blks
)
703 struct ext3_block_alloc_info
*block_i
;
704 ext3_fsblk_t current_block
;
705 struct ext3_inode_info
*ei
= EXT3_I(inode
);
707 block_i
= ei
->i_block_alloc_info
;
709 * If we're splicing into a [td]indirect block (as opposed to the
710 * inode) then we need to get write access to the [td]indirect block
714 BUFFER_TRACE(where
->bh
, "get_write_access");
715 err
= ext3_journal_get_write_access(handle
, where
->bh
);
721 *where
->p
= where
->key
;
724 * Update the host buffer_head or inode to point to more just allocated
725 * direct blocks blocks
727 if (num
== 0 && blks
> 1) {
728 current_block
= le32_to_cpu(where
->key
) + 1;
729 for (i
= 1; i
< blks
; i
++)
730 *(where
->p
+ i
) = cpu_to_le32(current_block
++);
734 * update the most recently allocated logical & physical block
735 * in i_block_alloc_info, to assist find the proper goal block for next
739 block_i
->last_alloc_logical_block
= block
+ blks
- 1;
740 block_i
->last_alloc_physical_block
=
741 le32_to_cpu(where
[num
].key
) + blks
- 1;
744 /* We are done with atomic stuff, now do the rest of housekeeping */
746 inode
->i_ctime
= CURRENT_TIME_SEC
;
747 ext3_mark_inode_dirty(handle
, inode
);
748 /* ext3_mark_inode_dirty already updated i_sync_tid */
749 atomic_set(&ei
->i_datasync_tid
, handle
->h_transaction
->t_tid
);
751 /* had we spliced it onto indirect block? */
754 * If we spliced it onto an indirect block, we haven't
755 * altered the inode. Note however that if it is being spliced
756 * onto an indirect block at the very end of the file (the
757 * file is growing) then we *will* alter the inode to reflect
758 * the new i_size. But that is not done here - it is done in
759 * generic_commit_write->__mark_inode_dirty->ext3_dirty_inode.
761 jbd_debug(5, "splicing indirect only\n");
762 BUFFER_TRACE(where
->bh
, "call ext3_journal_dirty_metadata");
763 err
= ext3_journal_dirty_metadata(handle
, where
->bh
);
768 * OK, we spliced it into the inode itself on a direct block.
769 * Inode was dirtied above.
771 jbd_debug(5, "splicing direct\n");
776 for (i
= 1; i
<= num
; i
++) {
777 BUFFER_TRACE(where
[i
].bh
, "call journal_forget");
778 ext3_journal_forget(handle
, where
[i
].bh
);
779 ext3_free_blocks(handle
,inode
,le32_to_cpu(where
[i
-1].key
),1);
781 ext3_free_blocks(handle
, inode
, le32_to_cpu(where
[num
].key
), blks
);
787 * Allocation strategy is simple: if we have to allocate something, we will
788 * have to go the whole way to leaf. So let's do it before attaching anything
789 * to tree, set linkage between the newborn blocks, write them if sync is
790 * required, recheck the path, free and repeat if check fails, otherwise
791 * set the last missing link (that will protect us from any truncate-generated
792 * removals - all blocks on the path are immune now) and possibly force the
793 * write on the parent block.
794 * That has a nice additional property: no special recovery from the failed
795 * allocations is needed - we simply release blocks and do not touch anything
796 * reachable from inode.
798 * `handle' can be NULL if create == 0.
800 * The BKL may not be held on entry here. Be sure to take it early.
801 * return > 0, # of blocks mapped or allocated.
802 * return = 0, if plain lookup failed.
803 * return < 0, error case.
805 int ext3_get_blocks_handle(handle_t
*handle
, struct inode
*inode
,
806 sector_t iblock
, unsigned long maxblocks
,
807 struct buffer_head
*bh_result
,
816 int blocks_to_boundary
= 0;
818 struct ext3_inode_info
*ei
= EXT3_I(inode
);
820 ext3_fsblk_t first_block
= 0;
823 J_ASSERT(handle
!= NULL
|| create
== 0);
824 depth
= ext3_block_to_path(inode
,iblock
,offsets
,&blocks_to_boundary
);
829 partial
= ext3_get_branch(inode
, depth
, offsets
, chain
, &err
);
831 /* Simplest case - block found, no allocation needed */
833 first_block
= le32_to_cpu(chain
[depth
- 1].key
);
834 clear_buffer_new(bh_result
);
837 while (count
< maxblocks
&& count
<= blocks_to_boundary
) {
840 if (!verify_chain(chain
, chain
+ depth
- 1)) {
842 * Indirect block might be removed by
843 * truncate while we were reading it.
844 * Handling of that case: forget what we've
845 * got now. Flag the err as EAGAIN, so it
852 blk
= le32_to_cpu(*(chain
[depth
-1].p
+ count
));
854 if (blk
== first_block
+ count
)
863 /* Next simple case - plain lookup or failed read of indirect block */
864 if (!create
|| err
== -EIO
)
867 mutex_lock(&ei
->truncate_mutex
);
870 * If the indirect block is missing while we are reading
871 * the chain(ext3_get_branch() returns -EAGAIN err), or
872 * if the chain has been changed after we grab the semaphore,
873 * (either because another process truncated this branch, or
874 * another get_block allocated this branch) re-grab the chain to see if
875 * the request block has been allocated or not.
877 * Since we already block the truncate/other get_block
878 * at this point, we will have the current copy of the chain when we
879 * splice the branch into the tree.
881 if (err
== -EAGAIN
|| !verify_chain(chain
, partial
)) {
882 while (partial
> chain
) {
886 partial
= ext3_get_branch(inode
, depth
, offsets
, chain
, &err
);
889 mutex_unlock(&ei
->truncate_mutex
);
892 clear_buffer_new(bh_result
);
898 * Okay, we need to do block allocation. Lazily initialize the block
899 * allocation info here if necessary
901 if (S_ISREG(inode
->i_mode
) && (!ei
->i_block_alloc_info
))
902 ext3_init_block_alloc_info(inode
);
904 goal
= ext3_find_goal(inode
, iblock
, partial
);
906 /* the number of blocks need to allocate for [d,t]indirect blocks */
907 indirect_blks
= (chain
+ depth
) - partial
- 1;
910 * Next look up the indirect map to count the totoal number of
911 * direct blocks to allocate for this branch.
913 count
= ext3_blks_to_allocate(partial
, indirect_blks
,
914 maxblocks
, blocks_to_boundary
);
916 * Block out ext3_truncate while we alter the tree
918 err
= ext3_alloc_branch(handle
, inode
, indirect_blks
, &count
, goal
,
919 offsets
+ (partial
- chain
), partial
);
922 * The ext3_splice_branch call will free and forget any buffers
923 * on the new chain if there is a failure, but that risks using
924 * up transaction credits, especially for bitmaps where the
925 * credits cannot be returned. Can we handle this somehow? We
926 * may need to return -EAGAIN upwards in the worst case. --sct
929 err
= ext3_splice_branch(handle
, inode
, iblock
,
930 partial
, indirect_blks
, count
);
931 mutex_unlock(&ei
->truncate_mutex
);
935 set_buffer_new(bh_result
);
937 map_bh(bh_result
, inode
->i_sb
, le32_to_cpu(chain
[depth
-1].key
));
938 if (count
> blocks_to_boundary
)
939 set_buffer_boundary(bh_result
);
941 /* Clean up and exit */
942 partial
= chain
+ depth
- 1; /* the whole chain */
944 while (partial
> chain
) {
945 BUFFER_TRACE(partial
->bh
, "call brelse");
949 BUFFER_TRACE(bh_result
, "returned");
954 /* Maximum number of blocks we map for direct IO at once. */
955 #define DIO_MAX_BLOCKS 4096
957 * Number of credits we need for writing DIO_MAX_BLOCKS:
958 * We need sb + group descriptor + bitmap + inode -> 4
959 * For B blocks with A block pointers per block we need:
960 * 1 (triple ind.) + (B/A/A + 2) (doubly ind.) + (B/A + 2) (indirect).
961 * If we plug in 4096 for B and 256 for A (for 1KB block size), we get 25.
963 #define DIO_CREDITS 25
965 static int ext3_get_block(struct inode
*inode
, sector_t iblock
,
966 struct buffer_head
*bh_result
, int create
)
968 handle_t
*handle
= ext3_journal_current_handle();
969 int ret
= 0, started
= 0;
970 unsigned max_blocks
= bh_result
->b_size
>> inode
->i_blkbits
;
972 if (create
&& !handle
) { /* Direct IO write... */
973 if (max_blocks
> DIO_MAX_BLOCKS
)
974 max_blocks
= DIO_MAX_BLOCKS
;
975 handle
= ext3_journal_start(inode
, DIO_CREDITS
+
976 EXT3_MAXQUOTAS_TRANS_BLOCKS(inode
->i_sb
));
977 if (IS_ERR(handle
)) {
978 ret
= PTR_ERR(handle
);
984 ret
= ext3_get_blocks_handle(handle
, inode
, iblock
,
985 max_blocks
, bh_result
, create
);
987 bh_result
->b_size
= (ret
<< inode
->i_blkbits
);
991 ext3_journal_stop(handle
);
996 int ext3_fiemap(struct inode
*inode
, struct fiemap_extent_info
*fieinfo
,
999 return generic_block_fiemap(inode
, fieinfo
, start
, len
,
1004 * `handle' can be NULL if create is zero
1006 struct buffer_head
*ext3_getblk(handle_t
*handle
, struct inode
*inode
,
1007 long block
, int create
, int *errp
)
1009 struct buffer_head dummy
;
1012 J_ASSERT(handle
!= NULL
|| create
== 0);
1015 dummy
.b_blocknr
= -1000;
1016 buffer_trace_init(&dummy
.b_history
);
1017 err
= ext3_get_blocks_handle(handle
, inode
, block
, 1,
1020 * ext3_get_blocks_handle() returns number of blocks
1021 * mapped. 0 in case of a HOLE.
1029 if (!err
&& buffer_mapped(&dummy
)) {
1030 struct buffer_head
*bh
;
1031 bh
= sb_getblk(inode
->i_sb
, dummy
.b_blocknr
);
1036 if (buffer_new(&dummy
)) {
1037 J_ASSERT(create
!= 0);
1038 J_ASSERT(handle
!= NULL
);
1041 * Now that we do not always journal data, we should
1042 * keep in mind whether this should always journal the
1043 * new buffer as metadata. For now, regular file
1044 * writes use ext3_get_block instead, so it's not a
1048 BUFFER_TRACE(bh
, "call get_create_access");
1049 fatal
= ext3_journal_get_create_access(handle
, bh
);
1050 if (!fatal
&& !buffer_uptodate(bh
)) {
1051 memset(bh
->b_data
,0,inode
->i_sb
->s_blocksize
);
1052 set_buffer_uptodate(bh
);
1055 BUFFER_TRACE(bh
, "call ext3_journal_dirty_metadata");
1056 err
= ext3_journal_dirty_metadata(handle
, bh
);
1060 BUFFER_TRACE(bh
, "not a new buffer");
1073 struct buffer_head
*ext3_bread(handle_t
*handle
, struct inode
*inode
,
1074 int block
, int create
, int *err
)
1076 struct buffer_head
* bh
;
1078 bh
= ext3_getblk(handle
, inode
, block
, create
, err
);
1081 if (buffer_uptodate(bh
))
1083 ll_rw_block(READ_META
, 1, &bh
);
1085 if (buffer_uptodate(bh
))
1092 static int walk_page_buffers( handle_t
*handle
,
1093 struct buffer_head
*head
,
1097 int (*fn
)( handle_t
*handle
,
1098 struct buffer_head
*bh
))
1100 struct buffer_head
*bh
;
1101 unsigned block_start
, block_end
;
1102 unsigned blocksize
= head
->b_size
;
1104 struct buffer_head
*next
;
1106 for ( bh
= head
, block_start
= 0;
1107 ret
== 0 && (bh
!= head
|| !block_start
);
1108 block_start
= block_end
, bh
= next
)
1110 next
= bh
->b_this_page
;
1111 block_end
= block_start
+ blocksize
;
1112 if (block_end
<= from
|| block_start
>= to
) {
1113 if (partial
&& !buffer_uptodate(bh
))
1117 err
= (*fn
)(handle
, bh
);
1125 * To preserve ordering, it is essential that the hole instantiation and
1126 * the data write be encapsulated in a single transaction. We cannot
1127 * close off a transaction and start a new one between the ext3_get_block()
1128 * and the commit_write(). So doing the journal_start at the start of
1129 * prepare_write() is the right place.
1131 * Also, this function can nest inside ext3_writepage() ->
1132 * block_write_full_page(). In that case, we *know* that ext3_writepage()
1133 * has generated enough buffer credits to do the whole page. So we won't
1134 * block on the journal in that case, which is good, because the caller may
1137 * By accident, ext3 can be reentered when a transaction is open via
1138 * quota file writes. If we were to commit the transaction while thus
1139 * reentered, there can be a deadlock - we would be holding a quota
1140 * lock, and the commit would never complete if another thread had a
1141 * transaction open and was blocking on the quota lock - a ranking
1144 * So what we do is to rely on the fact that journal_stop/journal_start
1145 * will _not_ run commit under these circumstances because handle->h_ref
1146 * is elevated. We'll still have enough credits for the tiny quotafile
1149 static int do_journal_get_write_access(handle_t
*handle
,
1150 struct buffer_head
*bh
)
1152 if (!buffer_mapped(bh
) || buffer_freed(bh
))
1154 return ext3_journal_get_write_access(handle
, bh
);
1158 * Truncate blocks that were not used by write. We have to truncate the
1159 * pagecache as well so that corresponding buffers get properly unmapped.
1161 static void ext3_truncate_failed_write(struct inode
*inode
)
1163 truncate_inode_pages(inode
->i_mapping
, inode
->i_size
);
1164 ext3_truncate(inode
);
1167 static int ext3_write_begin(struct file
*file
, struct address_space
*mapping
,
1168 loff_t pos
, unsigned len
, unsigned flags
,
1169 struct page
**pagep
, void **fsdata
)
1171 struct inode
*inode
= mapping
->host
;
1178 /* Reserve one block more for addition to orphan list in case
1179 * we allocate blocks but write fails for some reason */
1180 int needed_blocks
= ext3_writepage_trans_blocks(inode
) + 1;
1182 index
= pos
>> PAGE_CACHE_SHIFT
;
1183 from
= pos
& (PAGE_CACHE_SIZE
- 1);
1187 page
= grab_cache_page_write_begin(mapping
, index
, flags
);
1192 handle
= ext3_journal_start(inode
, needed_blocks
);
1193 if (IS_ERR(handle
)) {
1195 page_cache_release(page
);
1196 ret
= PTR_ERR(handle
);
1199 ret
= __block_write_begin(page
, pos
, len
, ext3_get_block
);
1201 goto write_begin_failed
;
1203 if (ext3_should_journal_data(inode
)) {
1204 ret
= walk_page_buffers(handle
, page_buffers(page
),
1205 from
, to
, NULL
, do_journal_get_write_access
);
1210 * block_write_begin may have instantiated a few blocks
1211 * outside i_size. Trim these off again. Don't need
1212 * i_size_read because we hold i_mutex.
1214 * Add inode to orphan list in case we crash before truncate
1215 * finishes. Do this only if ext3_can_truncate() agrees so
1216 * that orphan processing code is happy.
1218 if (pos
+ len
> inode
->i_size
&& ext3_can_truncate(inode
))
1219 ext3_orphan_add(handle
, inode
);
1220 ext3_journal_stop(handle
);
1222 page_cache_release(page
);
1223 if (pos
+ len
> inode
->i_size
)
1224 ext3_truncate_failed_write(inode
);
1226 if (ret
== -ENOSPC
&& ext3_should_retry_alloc(inode
->i_sb
, &retries
))
1233 int ext3_journal_dirty_data(handle_t
*handle
, struct buffer_head
*bh
)
1235 int err
= journal_dirty_data(handle
, bh
);
1237 ext3_journal_abort_handle(__func__
, __func__
,
1242 /* For ordered writepage and write_end functions */
1243 static int journal_dirty_data_fn(handle_t
*handle
, struct buffer_head
*bh
)
1246 * Write could have mapped the buffer but it didn't copy the data in
1247 * yet. So avoid filing such buffer into a transaction.
1249 if (buffer_mapped(bh
) && buffer_uptodate(bh
))
1250 return ext3_journal_dirty_data(handle
, bh
);
1254 /* For write_end() in data=journal mode */
1255 static int write_end_fn(handle_t
*handle
, struct buffer_head
*bh
)
1257 if (!buffer_mapped(bh
) || buffer_freed(bh
))
1259 set_buffer_uptodate(bh
);
1260 return ext3_journal_dirty_metadata(handle
, bh
);
1264 * This is nasty and subtle: ext3_write_begin() could have allocated blocks
1265 * for the whole page but later we failed to copy the data in. Update inode
1266 * size according to what we managed to copy. The rest is going to be
1267 * truncated in write_end function.
1269 static void update_file_sizes(struct inode
*inode
, loff_t pos
, unsigned copied
)
1271 /* What matters to us is i_disksize. We don't write i_size anywhere */
1272 if (pos
+ copied
> inode
->i_size
)
1273 i_size_write(inode
, pos
+ copied
);
1274 if (pos
+ copied
> EXT3_I(inode
)->i_disksize
) {
1275 EXT3_I(inode
)->i_disksize
= pos
+ copied
;
1276 mark_inode_dirty(inode
);
1281 * We need to pick up the new inode size which generic_commit_write gave us
1282 * `file' can be NULL - eg, when called from page_symlink().
1284 * ext3 never places buffers on inode->i_mapping->private_list. metadata
1285 * buffers are managed internally.
1287 static int ext3_ordered_write_end(struct file
*file
,
1288 struct address_space
*mapping
,
1289 loff_t pos
, unsigned len
, unsigned copied
,
1290 struct page
*page
, void *fsdata
)
1292 handle_t
*handle
= ext3_journal_current_handle();
1293 struct inode
*inode
= file
->f_mapping
->host
;
1297 copied
= block_write_end(file
, mapping
, pos
, len
, copied
, page
, fsdata
);
1299 from
= pos
& (PAGE_CACHE_SIZE
- 1);
1301 ret
= walk_page_buffers(handle
, page_buffers(page
),
1302 from
, to
, NULL
, journal_dirty_data_fn
);
1305 update_file_sizes(inode
, pos
, copied
);
1307 * There may be allocated blocks outside of i_size because
1308 * we failed to copy some data. Prepare for truncate.
1310 if (pos
+ len
> inode
->i_size
&& ext3_can_truncate(inode
))
1311 ext3_orphan_add(handle
, inode
);
1312 ret2
= ext3_journal_stop(handle
);
1316 page_cache_release(page
);
1318 if (pos
+ len
> inode
->i_size
)
1319 ext3_truncate_failed_write(inode
);
1320 return ret
? ret
: copied
;
1323 static int ext3_writeback_write_end(struct file
*file
,
1324 struct address_space
*mapping
,
1325 loff_t pos
, unsigned len
, unsigned copied
,
1326 struct page
*page
, void *fsdata
)
1328 handle_t
*handle
= ext3_journal_current_handle();
1329 struct inode
*inode
= file
->f_mapping
->host
;
1332 copied
= block_write_end(file
, mapping
, pos
, len
, copied
, page
, fsdata
);
1333 update_file_sizes(inode
, pos
, copied
);
1335 * There may be allocated blocks outside of i_size because
1336 * we failed to copy some data. Prepare for truncate.
1338 if (pos
+ len
> inode
->i_size
&& ext3_can_truncate(inode
))
1339 ext3_orphan_add(handle
, inode
);
1340 ret
= ext3_journal_stop(handle
);
1342 page_cache_release(page
);
1344 if (pos
+ len
> inode
->i_size
)
1345 ext3_truncate_failed_write(inode
);
1346 return ret
? ret
: copied
;
1349 static int ext3_journalled_write_end(struct file
*file
,
1350 struct address_space
*mapping
,
1351 loff_t pos
, unsigned len
, unsigned copied
,
1352 struct page
*page
, void *fsdata
)
1354 handle_t
*handle
= ext3_journal_current_handle();
1355 struct inode
*inode
= mapping
->host
;
1360 from
= pos
& (PAGE_CACHE_SIZE
- 1);
1364 if (!PageUptodate(page
))
1366 page_zero_new_buffers(page
, from
+ copied
, to
);
1370 ret
= walk_page_buffers(handle
, page_buffers(page
), from
,
1371 to
, &partial
, write_end_fn
);
1373 SetPageUptodate(page
);
1375 if (pos
+ copied
> inode
->i_size
)
1376 i_size_write(inode
, pos
+ copied
);
1378 * There may be allocated blocks outside of i_size because
1379 * we failed to copy some data. Prepare for truncate.
1381 if (pos
+ len
> inode
->i_size
&& ext3_can_truncate(inode
))
1382 ext3_orphan_add(handle
, inode
);
1383 ext3_set_inode_state(inode
, EXT3_STATE_JDATA
);
1384 if (inode
->i_size
> EXT3_I(inode
)->i_disksize
) {
1385 EXT3_I(inode
)->i_disksize
= inode
->i_size
;
1386 ret2
= ext3_mark_inode_dirty(handle
, inode
);
1391 ret2
= ext3_journal_stop(handle
);
1395 page_cache_release(page
);
1397 if (pos
+ len
> inode
->i_size
)
1398 ext3_truncate_failed_write(inode
);
1399 return ret
? ret
: copied
;
1403 * bmap() is special. It gets used by applications such as lilo and by
1404 * the swapper to find the on-disk block of a specific piece of data.
1406 * Naturally, this is dangerous if the block concerned is still in the
1407 * journal. If somebody makes a swapfile on an ext3 data-journaling
1408 * filesystem and enables swap, then they may get a nasty shock when the
1409 * data getting swapped to that swapfile suddenly gets overwritten by
1410 * the original zero's written out previously to the journal and
1411 * awaiting writeback in the kernel's buffer cache.
1413 * So, if we see any bmap calls here on a modified, data-journaled file,
1414 * take extra steps to flush any blocks which might be in the cache.
1416 static sector_t
ext3_bmap(struct address_space
*mapping
, sector_t block
)
1418 struct inode
*inode
= mapping
->host
;
1422 if (ext3_test_inode_state(inode
, EXT3_STATE_JDATA
)) {
1424 * This is a REALLY heavyweight approach, but the use of
1425 * bmap on dirty files is expected to be extremely rare:
1426 * only if we run lilo or swapon on a freshly made file
1427 * do we expect this to happen.
1429 * (bmap requires CAP_SYS_RAWIO so this does not
1430 * represent an unprivileged user DOS attack --- we'd be
1431 * in trouble if mortal users could trigger this path at
1434 * NB. EXT3_STATE_JDATA is not set on files other than
1435 * regular files. If somebody wants to bmap a directory
1436 * or symlink and gets confused because the buffer
1437 * hasn't yet been flushed to disk, they deserve
1438 * everything they get.
1441 ext3_clear_inode_state(inode
, EXT3_STATE_JDATA
);
1442 journal
= EXT3_JOURNAL(inode
);
1443 journal_lock_updates(journal
);
1444 err
= journal_flush(journal
);
1445 journal_unlock_updates(journal
);
1451 return generic_block_bmap(mapping
,block
,ext3_get_block
);
1454 static int bget_one(handle_t
*handle
, struct buffer_head
*bh
)
1460 static int bput_one(handle_t
*handle
, struct buffer_head
*bh
)
1466 static int buffer_unmapped(handle_t
*handle
, struct buffer_head
*bh
)
1468 return !buffer_mapped(bh
);
1472 * Note that we always start a transaction even if we're not journalling
1473 * data. This is to preserve ordering: any hole instantiation within
1474 * __block_write_full_page -> ext3_get_block() should be journalled
1475 * along with the data so we don't crash and then get metadata which
1476 * refers to old data.
1478 * In all journalling modes block_write_full_page() will start the I/O.
1482 * ext3_writepage() -> kmalloc() -> __alloc_pages() -> page_launder() ->
1487 * ext3_file_write() -> generic_file_write() -> __alloc_pages() -> ...
1489 * Same applies to ext3_get_block(). We will deadlock on various things like
1490 * lock_journal and i_truncate_mutex.
1492 * Setting PF_MEMALLOC here doesn't work - too many internal memory
1495 * 16May01: If we're reentered then journal_current_handle() will be
1496 * non-zero. We simply *return*.
1498 * 1 July 2001: @@@ FIXME:
1499 * In journalled data mode, a data buffer may be metadata against the
1500 * current transaction. But the same file is part of a shared mapping
1501 * and someone does a writepage() on it.
1503 * We will move the buffer onto the async_data list, but *after* it has
1504 * been dirtied. So there's a small window where we have dirty data on
1507 * Note that this only applies to the last partial page in the file. The
1508 * bit which block_write_full_page() uses prepare/commit for. (That's
1509 * broken code anyway: it's wrong for msync()).
1511 * It's a rare case: affects the final partial page, for journalled data
1512 * where the file is subject to bith write() and writepage() in the same
1513 * transction. To fix it we'll need a custom block_write_full_page().
1514 * We'll probably need that anyway for journalling writepage() output.
1516 * We don't honour synchronous mounts for writepage(). That would be
1517 * disastrous. Any write() or metadata operation will sync the fs for
1520 * AKPM2: if all the page's buffers are mapped to disk and !data=journal,
1521 * we don't need to open a transaction here.
1523 static int ext3_ordered_writepage(struct page
*page
,
1524 struct writeback_control
*wbc
)
1526 struct inode
*inode
= page
->mapping
->host
;
1527 struct buffer_head
*page_bufs
;
1528 handle_t
*handle
= NULL
;
1532 J_ASSERT(PageLocked(page
));
1533 WARN_ON_ONCE(IS_RDONLY(inode
));
1536 * We give up here if we're reentered, because it might be for a
1537 * different filesystem.
1539 if (ext3_journal_current_handle())
1542 if (!page_has_buffers(page
)) {
1543 create_empty_buffers(page
, inode
->i_sb
->s_blocksize
,
1544 (1 << BH_Dirty
)|(1 << BH_Uptodate
));
1545 page_bufs
= page_buffers(page
);
1547 page_bufs
= page_buffers(page
);
1548 if (!walk_page_buffers(NULL
, page_bufs
, 0, PAGE_CACHE_SIZE
,
1549 NULL
, buffer_unmapped
)) {
1550 /* Provide NULL get_block() to catch bugs if buffers
1551 * weren't really mapped */
1552 return block_write_full_page(page
, NULL
, wbc
);
1555 handle
= ext3_journal_start(inode
, ext3_writepage_trans_blocks(inode
));
1557 if (IS_ERR(handle
)) {
1558 ret
= PTR_ERR(handle
);
1562 walk_page_buffers(handle
, page_bufs
, 0,
1563 PAGE_CACHE_SIZE
, NULL
, bget_one
);
1565 ret
= block_write_full_page(page
, ext3_get_block
, wbc
);
1568 * The page can become unlocked at any point now, and
1569 * truncate can then come in and change things. So we
1570 * can't touch *page from now on. But *page_bufs is
1571 * safe due to elevated refcount.
1575 * And attach them to the current transaction. But only if
1576 * block_write_full_page() succeeded. Otherwise they are unmapped,
1577 * and generally junk.
1580 err
= walk_page_buffers(handle
, page_bufs
, 0, PAGE_CACHE_SIZE
,
1581 NULL
, journal_dirty_data_fn
);
1585 walk_page_buffers(handle
, page_bufs
, 0,
1586 PAGE_CACHE_SIZE
, NULL
, bput_one
);
1587 err
= ext3_journal_stop(handle
);
1593 redirty_page_for_writepage(wbc
, page
);
1598 static int ext3_writeback_writepage(struct page
*page
,
1599 struct writeback_control
*wbc
)
1601 struct inode
*inode
= page
->mapping
->host
;
1602 handle_t
*handle
= NULL
;
1606 J_ASSERT(PageLocked(page
));
1607 WARN_ON_ONCE(IS_RDONLY(inode
));
1609 if (ext3_journal_current_handle())
1612 if (page_has_buffers(page
)) {
1613 if (!walk_page_buffers(NULL
, page_buffers(page
), 0,
1614 PAGE_CACHE_SIZE
, NULL
, buffer_unmapped
)) {
1615 /* Provide NULL get_block() to catch bugs if buffers
1616 * weren't really mapped */
1617 return block_write_full_page(page
, NULL
, wbc
);
1621 handle
= ext3_journal_start(inode
, ext3_writepage_trans_blocks(inode
));
1622 if (IS_ERR(handle
)) {
1623 ret
= PTR_ERR(handle
);
1627 if (test_opt(inode
->i_sb
, NOBH
) && ext3_should_writeback_data(inode
))
1628 ret
= nobh_writepage(page
, ext3_get_block
, wbc
);
1630 ret
= block_write_full_page(page
, ext3_get_block
, wbc
);
1632 err
= ext3_journal_stop(handle
);
1638 redirty_page_for_writepage(wbc
, page
);
1643 static int ext3_journalled_writepage(struct page
*page
,
1644 struct writeback_control
*wbc
)
1646 struct inode
*inode
= page
->mapping
->host
;
1647 handle_t
*handle
= NULL
;
1651 J_ASSERT(PageLocked(page
));
1652 WARN_ON_ONCE(IS_RDONLY(inode
));
1654 if (ext3_journal_current_handle())
1657 handle
= ext3_journal_start(inode
, ext3_writepage_trans_blocks(inode
));
1658 if (IS_ERR(handle
)) {
1659 ret
= PTR_ERR(handle
);
1663 if (!page_has_buffers(page
) || PageChecked(page
)) {
1665 * It's mmapped pagecache. Add buffers and journal it. There
1666 * doesn't seem much point in redirtying the page here.
1668 ClearPageChecked(page
);
1669 ret
= block_prepare_write(page
, 0, PAGE_CACHE_SIZE
,
1672 ext3_journal_stop(handle
);
1675 ret
= walk_page_buffers(handle
, page_buffers(page
), 0,
1676 PAGE_CACHE_SIZE
, NULL
, do_journal_get_write_access
);
1678 err
= walk_page_buffers(handle
, page_buffers(page
), 0,
1679 PAGE_CACHE_SIZE
, NULL
, write_end_fn
);
1682 ext3_set_inode_state(inode
, EXT3_STATE_JDATA
);
1686 * It may be a page full of checkpoint-mode buffers. We don't
1687 * really know unless we go poke around in the buffer_heads.
1688 * But block_write_full_page will do the right thing.
1690 ret
= block_write_full_page(page
, ext3_get_block
, wbc
);
1692 err
= ext3_journal_stop(handle
);
1699 redirty_page_for_writepage(wbc
, page
);
1705 static int ext3_readpage(struct file
*file
, struct page
*page
)
1707 return mpage_readpage(page
, ext3_get_block
);
1711 ext3_readpages(struct file
*file
, struct address_space
*mapping
,
1712 struct list_head
*pages
, unsigned nr_pages
)
1714 return mpage_readpages(mapping
, pages
, nr_pages
, ext3_get_block
);
1717 static void ext3_invalidatepage(struct page
*page
, unsigned long offset
)
1719 journal_t
*journal
= EXT3_JOURNAL(page
->mapping
->host
);
1722 * If it's a full truncate we just forget about the pending dirtying
1725 ClearPageChecked(page
);
1727 journal_invalidatepage(journal
, page
, offset
);
1730 static int ext3_releasepage(struct page
*page
, gfp_t wait
)
1732 journal_t
*journal
= EXT3_JOURNAL(page
->mapping
->host
);
1734 WARN_ON(PageChecked(page
));
1735 if (!page_has_buffers(page
))
1737 return journal_try_to_free_buffers(journal
, page
, wait
);
1741 * If the O_DIRECT write will extend the file then add this inode to the
1742 * orphan list. So recovery will truncate it back to the original size
1743 * if the machine crashes during the write.
1745 * If the O_DIRECT write is intantiating holes inside i_size and the machine
1746 * crashes then stale disk data _may_ be exposed inside the file. But current
1747 * VFS code falls back into buffered path in that case so we are safe.
1749 static ssize_t
ext3_direct_IO(int rw
, struct kiocb
*iocb
,
1750 const struct iovec
*iov
, loff_t offset
,
1751 unsigned long nr_segs
)
1753 struct file
*file
= iocb
->ki_filp
;
1754 struct inode
*inode
= file
->f_mapping
->host
;
1755 struct ext3_inode_info
*ei
= EXT3_I(inode
);
1759 size_t count
= iov_length(iov
, nr_segs
);
1763 loff_t final_size
= offset
+ count
;
1765 if (final_size
> inode
->i_size
) {
1766 /* Credits for sb + inode write */
1767 handle
= ext3_journal_start(inode
, 2);
1768 if (IS_ERR(handle
)) {
1769 ret
= PTR_ERR(handle
);
1772 ret
= ext3_orphan_add(handle
, inode
);
1774 ext3_journal_stop(handle
);
1778 ei
->i_disksize
= inode
->i_size
;
1779 ext3_journal_stop(handle
);
1784 ret
= blockdev_direct_IO(rw
, iocb
, inode
, inode
->i_sb
->s_bdev
, iov
,
1786 ext3_get_block
, NULL
);
1788 * In case of error extending write may have instantiated a few
1789 * blocks outside i_size. Trim these off again.
1791 if (unlikely((rw
& WRITE
) && ret
< 0)) {
1792 loff_t isize
= i_size_read(inode
);
1793 loff_t end
= offset
+ iov_length(iov
, nr_segs
);
1796 vmtruncate(inode
, isize
);
1798 if (ret
== -ENOSPC
&& ext3_should_retry_alloc(inode
->i_sb
, &retries
))
1804 /* Credits for sb + inode write */
1805 handle
= ext3_journal_start(inode
, 2);
1806 if (IS_ERR(handle
)) {
1807 /* This is really bad luck. We've written the data
1808 * but cannot extend i_size. Truncate allocated blocks
1809 * and pretend the write failed... */
1810 ext3_truncate(inode
);
1811 ret
= PTR_ERR(handle
);
1815 ext3_orphan_del(handle
, inode
);
1817 loff_t end
= offset
+ ret
;
1818 if (end
> inode
->i_size
) {
1819 ei
->i_disksize
= end
;
1820 i_size_write(inode
, end
);
1822 * We're going to return a positive `ret'
1823 * here due to non-zero-length I/O, so there's
1824 * no way of reporting error returns from
1825 * ext3_mark_inode_dirty() to userspace. So
1828 ext3_mark_inode_dirty(handle
, inode
);
1831 err
= ext3_journal_stop(handle
);
1840 * Pages can be marked dirty completely asynchronously from ext3's journalling
1841 * activity. By filemap_sync_pte(), try_to_unmap_one(), etc. We cannot do
1842 * much here because ->set_page_dirty is called under VFS locks. The page is
1843 * not necessarily locked.
1845 * We cannot just dirty the page and leave attached buffers clean, because the
1846 * buffers' dirty state is "definitive". We cannot just set the buffers dirty
1847 * or jbddirty because all the journalling code will explode.
1849 * So what we do is to mark the page "pending dirty" and next time writepage
1850 * is called, propagate that into the buffers appropriately.
1852 static int ext3_journalled_set_page_dirty(struct page
*page
)
1854 SetPageChecked(page
);
1855 return __set_page_dirty_nobuffers(page
);
1858 static const struct address_space_operations ext3_ordered_aops
= {
1859 .readpage
= ext3_readpage
,
1860 .readpages
= ext3_readpages
,
1861 .writepage
= ext3_ordered_writepage
,
1862 .sync_page
= block_sync_page
,
1863 .write_begin
= ext3_write_begin
,
1864 .write_end
= ext3_ordered_write_end
,
1866 .invalidatepage
= ext3_invalidatepage
,
1867 .releasepage
= ext3_releasepage
,
1868 .direct_IO
= ext3_direct_IO
,
1869 .migratepage
= buffer_migrate_page
,
1870 .is_partially_uptodate
= block_is_partially_uptodate
,
1871 .error_remove_page
= generic_error_remove_page
,
1874 static const struct address_space_operations ext3_writeback_aops
= {
1875 .readpage
= ext3_readpage
,
1876 .readpages
= ext3_readpages
,
1877 .writepage
= ext3_writeback_writepage
,
1878 .sync_page
= block_sync_page
,
1879 .write_begin
= ext3_write_begin
,
1880 .write_end
= ext3_writeback_write_end
,
1882 .invalidatepage
= ext3_invalidatepage
,
1883 .releasepage
= ext3_releasepage
,
1884 .direct_IO
= ext3_direct_IO
,
1885 .migratepage
= buffer_migrate_page
,
1886 .is_partially_uptodate
= block_is_partially_uptodate
,
1887 .error_remove_page
= generic_error_remove_page
,
1890 static const struct address_space_operations ext3_journalled_aops
= {
1891 .readpage
= ext3_readpage
,
1892 .readpages
= ext3_readpages
,
1893 .writepage
= ext3_journalled_writepage
,
1894 .sync_page
= block_sync_page
,
1895 .write_begin
= ext3_write_begin
,
1896 .write_end
= ext3_journalled_write_end
,
1897 .set_page_dirty
= ext3_journalled_set_page_dirty
,
1899 .invalidatepage
= ext3_invalidatepage
,
1900 .releasepage
= ext3_releasepage
,
1901 .is_partially_uptodate
= block_is_partially_uptodate
,
1902 .error_remove_page
= generic_error_remove_page
,
1905 void ext3_set_aops(struct inode
*inode
)
1907 if (ext3_should_order_data(inode
))
1908 inode
->i_mapping
->a_ops
= &ext3_ordered_aops
;
1909 else if (ext3_should_writeback_data(inode
))
1910 inode
->i_mapping
->a_ops
= &ext3_writeback_aops
;
1912 inode
->i_mapping
->a_ops
= &ext3_journalled_aops
;
1916 * ext3_block_truncate_page() zeroes out a mapping from file offset `from'
1917 * up to the end of the block which corresponds to `from'.
1918 * This required during truncate. We need to physically zero the tail end
1919 * of that block so it doesn't yield old data if the file is later grown.
1921 static int ext3_block_truncate_page(handle_t
*handle
, struct page
*page
,
1922 struct address_space
*mapping
, loff_t from
)
1924 ext3_fsblk_t index
= from
>> PAGE_CACHE_SHIFT
;
1925 unsigned offset
= from
& (PAGE_CACHE_SIZE
-1);
1926 unsigned blocksize
, iblock
, length
, pos
;
1927 struct inode
*inode
= mapping
->host
;
1928 struct buffer_head
*bh
;
1931 blocksize
= inode
->i_sb
->s_blocksize
;
1932 length
= blocksize
- (offset
& (blocksize
- 1));
1933 iblock
= index
<< (PAGE_CACHE_SHIFT
- inode
->i_sb
->s_blocksize_bits
);
1936 * For "nobh" option, we can only work if we don't need to
1937 * read-in the page - otherwise we create buffers to do the IO.
1939 if (!page_has_buffers(page
) && test_opt(inode
->i_sb
, NOBH
) &&
1940 ext3_should_writeback_data(inode
) && PageUptodate(page
)) {
1941 zero_user(page
, offset
, length
);
1942 set_page_dirty(page
);
1946 if (!page_has_buffers(page
))
1947 create_empty_buffers(page
, blocksize
, 0);
1949 /* Find the buffer that contains "offset" */
1950 bh
= page_buffers(page
);
1952 while (offset
>= pos
) {
1953 bh
= bh
->b_this_page
;
1959 if (buffer_freed(bh
)) {
1960 BUFFER_TRACE(bh
, "freed: skip");
1964 if (!buffer_mapped(bh
)) {
1965 BUFFER_TRACE(bh
, "unmapped");
1966 ext3_get_block(inode
, iblock
, bh
, 0);
1967 /* unmapped? It's a hole - nothing to do */
1968 if (!buffer_mapped(bh
)) {
1969 BUFFER_TRACE(bh
, "still unmapped");
1974 /* Ok, it's mapped. Make sure it's up-to-date */
1975 if (PageUptodate(page
))
1976 set_buffer_uptodate(bh
);
1978 if (!buffer_uptodate(bh
)) {
1980 ll_rw_block(READ
, 1, &bh
);
1982 /* Uhhuh. Read error. Complain and punt. */
1983 if (!buffer_uptodate(bh
))
1987 if (ext3_should_journal_data(inode
)) {
1988 BUFFER_TRACE(bh
, "get write access");
1989 err
= ext3_journal_get_write_access(handle
, bh
);
1994 zero_user(page
, offset
, length
);
1995 BUFFER_TRACE(bh
, "zeroed end of block");
1998 if (ext3_should_journal_data(inode
)) {
1999 err
= ext3_journal_dirty_metadata(handle
, bh
);
2001 if (ext3_should_order_data(inode
))
2002 err
= ext3_journal_dirty_data(handle
, bh
);
2003 mark_buffer_dirty(bh
);
2008 page_cache_release(page
);
2013 * Probably it should be a library function... search for first non-zero word
2014 * or memcmp with zero_page, whatever is better for particular architecture.
2017 static inline int all_zeroes(__le32
*p
, __le32
*q
)
2026 * ext3_find_shared - find the indirect blocks for partial truncation.
2027 * @inode: inode in question
2028 * @depth: depth of the affected branch
2029 * @offsets: offsets of pointers in that branch (see ext3_block_to_path)
2030 * @chain: place to store the pointers to partial indirect blocks
2031 * @top: place to the (detached) top of branch
2033 * This is a helper function used by ext3_truncate().
2035 * When we do truncate() we may have to clean the ends of several
2036 * indirect blocks but leave the blocks themselves alive. Block is
2037 * partially truncated if some data below the new i_size is refered
2038 * from it (and it is on the path to the first completely truncated
2039 * data block, indeed). We have to free the top of that path along
2040 * with everything to the right of the path. Since no allocation
2041 * past the truncation point is possible until ext3_truncate()
2042 * finishes, we may safely do the latter, but top of branch may
2043 * require special attention - pageout below the truncation point
2044 * might try to populate it.
2046 * We atomically detach the top of branch from the tree, store the
2047 * block number of its root in *@top, pointers to buffer_heads of
2048 * partially truncated blocks - in @chain[].bh and pointers to
2049 * their last elements that should not be removed - in
2050 * @chain[].p. Return value is the pointer to last filled element
2053 * The work left to caller to do the actual freeing of subtrees:
2054 * a) free the subtree starting from *@top
2055 * b) free the subtrees whose roots are stored in
2056 * (@chain[i].p+1 .. end of @chain[i].bh->b_data)
2057 * c) free the subtrees growing from the inode past the @chain[0].
2058 * (no partially truncated stuff there). */
2060 static Indirect
*ext3_find_shared(struct inode
*inode
, int depth
,
2061 int offsets
[4], Indirect chain
[4], __le32
*top
)
2063 Indirect
*partial
, *p
;
2067 /* Make k index the deepest non-null offset + 1 */
2068 for (k
= depth
; k
> 1 && !offsets
[k
-1]; k
--)
2070 partial
= ext3_get_branch(inode
, k
, offsets
, chain
, &err
);
2071 /* Writer: pointers */
2073 partial
= chain
+ k
-1;
2075 * If the branch acquired continuation since we've looked at it -
2076 * fine, it should all survive and (new) top doesn't belong to us.
2078 if (!partial
->key
&& *partial
->p
)
2081 for (p
=partial
; p
>chain
&& all_zeroes((__le32
*)p
->bh
->b_data
,p
->p
); p
--)
2084 * OK, we've found the last block that must survive. The rest of our
2085 * branch should be detached before unlocking. However, if that rest
2086 * of branch is all ours and does not grow immediately from the inode
2087 * it's easier to cheat and just decrement partial->p.
2089 if (p
== chain
+ k
- 1 && p
> chain
) {
2093 /* Nope, don't do this in ext3. Must leave the tree intact */
2100 while(partial
> p
) {
2101 brelse(partial
->bh
);
2109 * Zero a number of block pointers in either an inode or an indirect block.
2110 * If we restart the transaction we must again get write access to the
2111 * indirect block for further modification.
2113 * We release `count' blocks on disk, but (last - first) may be greater
2114 * than `count' because there can be holes in there.
2116 static void ext3_clear_blocks(handle_t
*handle
, struct inode
*inode
,
2117 struct buffer_head
*bh
, ext3_fsblk_t block_to_free
,
2118 unsigned long count
, __le32
*first
, __le32
*last
)
2121 if (try_to_extend_transaction(handle
, inode
)) {
2123 BUFFER_TRACE(bh
, "call ext3_journal_dirty_metadata");
2124 ext3_journal_dirty_metadata(handle
, bh
);
2126 ext3_mark_inode_dirty(handle
, inode
);
2127 truncate_restart_transaction(handle
, inode
);
2129 BUFFER_TRACE(bh
, "retaking write access");
2130 ext3_journal_get_write_access(handle
, bh
);
2135 * Any buffers which are on the journal will be in memory. We find
2136 * them on the hash table so journal_revoke() will run journal_forget()
2137 * on them. We've already detached each block from the file, so
2138 * bforget() in journal_forget() should be safe.
2140 * AKPM: turn on bforget in journal_forget()!!!
2142 for (p
= first
; p
< last
; p
++) {
2143 u32 nr
= le32_to_cpu(*p
);
2145 struct buffer_head
*bh
;
2148 bh
= sb_find_get_block(inode
->i_sb
, nr
);
2149 ext3_forget(handle
, 0, inode
, bh
, nr
);
2153 ext3_free_blocks(handle
, inode
, block_to_free
, count
);
2157 * ext3_free_data - free a list of data blocks
2158 * @handle: handle for this transaction
2159 * @inode: inode we are dealing with
2160 * @this_bh: indirect buffer_head which contains *@first and *@last
2161 * @first: array of block numbers
2162 * @last: points immediately past the end of array
2164 * We are freeing all blocks refered from that array (numbers are stored as
2165 * little-endian 32-bit) and updating @inode->i_blocks appropriately.
2167 * We accumulate contiguous runs of blocks to free. Conveniently, if these
2168 * blocks are contiguous then releasing them at one time will only affect one
2169 * or two bitmap blocks (+ group descriptor(s) and superblock) and we won't
2170 * actually use a lot of journal space.
2172 * @this_bh will be %NULL if @first and @last point into the inode's direct
2175 static void ext3_free_data(handle_t
*handle
, struct inode
*inode
,
2176 struct buffer_head
*this_bh
,
2177 __le32
*first
, __le32
*last
)
2179 ext3_fsblk_t block_to_free
= 0; /* Starting block # of a run */
2180 unsigned long count
= 0; /* Number of blocks in the run */
2181 __le32
*block_to_free_p
= NULL
; /* Pointer into inode/ind
2184 ext3_fsblk_t nr
; /* Current block # */
2185 __le32
*p
; /* Pointer into inode/ind
2186 for current block */
2189 if (this_bh
) { /* For indirect block */
2190 BUFFER_TRACE(this_bh
, "get_write_access");
2191 err
= ext3_journal_get_write_access(handle
, this_bh
);
2192 /* Important: if we can't update the indirect pointers
2193 * to the blocks, we can't free them. */
2198 for (p
= first
; p
< last
; p
++) {
2199 nr
= le32_to_cpu(*p
);
2201 /* accumulate blocks to free if they're contiguous */
2204 block_to_free_p
= p
;
2206 } else if (nr
== block_to_free
+ count
) {
2209 ext3_clear_blocks(handle
, inode
, this_bh
,
2211 count
, block_to_free_p
, p
);
2213 block_to_free_p
= p
;
2220 ext3_clear_blocks(handle
, inode
, this_bh
, block_to_free
,
2221 count
, block_to_free_p
, p
);
2224 BUFFER_TRACE(this_bh
, "call ext3_journal_dirty_metadata");
2227 * The buffer head should have an attached journal head at this
2228 * point. However, if the data is corrupted and an indirect
2229 * block pointed to itself, it would have been detached when
2230 * the block was cleared. Check for this instead of OOPSing.
2233 ext3_journal_dirty_metadata(handle
, this_bh
);
2235 ext3_error(inode
->i_sb
, "ext3_free_data",
2236 "circular indirect block detected, "
2237 "inode=%lu, block=%llu",
2239 (unsigned long long)this_bh
->b_blocknr
);
2244 * ext3_free_branches - free an array of branches
2245 * @handle: JBD handle for this transaction
2246 * @inode: inode we are dealing with
2247 * @parent_bh: the buffer_head which contains *@first and *@last
2248 * @first: array of block numbers
2249 * @last: pointer immediately past the end of array
2250 * @depth: depth of the branches to free
2252 * We are freeing all blocks refered from these branches (numbers are
2253 * stored as little-endian 32-bit) and updating @inode->i_blocks
2256 static void ext3_free_branches(handle_t
*handle
, struct inode
*inode
,
2257 struct buffer_head
*parent_bh
,
2258 __le32
*first
, __le32
*last
, int depth
)
2263 if (is_handle_aborted(handle
))
2267 struct buffer_head
*bh
;
2268 int addr_per_block
= EXT3_ADDR_PER_BLOCK(inode
->i_sb
);
2270 while (--p
>= first
) {
2271 nr
= le32_to_cpu(*p
);
2273 continue; /* A hole */
2275 /* Go read the buffer for the next level down */
2276 bh
= sb_bread(inode
->i_sb
, nr
);
2279 * A read failure? Report error and clear slot
2283 ext3_error(inode
->i_sb
, "ext3_free_branches",
2284 "Read failure, inode=%lu, block="E3FSBLK
,
2289 /* This zaps the entire block. Bottom up. */
2290 BUFFER_TRACE(bh
, "free child branches");
2291 ext3_free_branches(handle
, inode
, bh
,
2292 (__le32
*)bh
->b_data
,
2293 (__le32
*)bh
->b_data
+ addr_per_block
,
2297 * We've probably journalled the indirect block several
2298 * times during the truncate. But it's no longer
2299 * needed and we now drop it from the transaction via
2302 * That's easy if it's exclusively part of this
2303 * transaction. But if it's part of the committing
2304 * transaction then journal_forget() will simply
2305 * brelse() it. That means that if the underlying
2306 * block is reallocated in ext3_get_block(),
2307 * unmap_underlying_metadata() will find this block
2308 * and will try to get rid of it. damn, damn.
2310 * If this block has already been committed to the
2311 * journal, a revoke record will be written. And
2312 * revoke records must be emitted *before* clearing
2313 * this block's bit in the bitmaps.
2315 ext3_forget(handle
, 1, inode
, bh
, bh
->b_blocknr
);
2318 * Everything below this this pointer has been
2319 * released. Now let this top-of-subtree go.
2321 * We want the freeing of this indirect block to be
2322 * atomic in the journal with the updating of the
2323 * bitmap block which owns it. So make some room in
2326 * We zero the parent pointer *after* freeing its
2327 * pointee in the bitmaps, so if extend_transaction()
2328 * for some reason fails to put the bitmap changes and
2329 * the release into the same transaction, recovery
2330 * will merely complain about releasing a free block,
2331 * rather than leaking blocks.
2333 if (is_handle_aborted(handle
))
2335 if (try_to_extend_transaction(handle
, inode
)) {
2336 ext3_mark_inode_dirty(handle
, inode
);
2337 truncate_restart_transaction(handle
, inode
);
2340 ext3_free_blocks(handle
, inode
, nr
, 1);
2344 * The block which we have just freed is
2345 * pointed to by an indirect block: journal it
2347 BUFFER_TRACE(parent_bh
, "get_write_access");
2348 if (!ext3_journal_get_write_access(handle
,
2351 BUFFER_TRACE(parent_bh
,
2352 "call ext3_journal_dirty_metadata");
2353 ext3_journal_dirty_metadata(handle
,
2359 /* We have reached the bottom of the tree. */
2360 BUFFER_TRACE(parent_bh
, "free data blocks");
2361 ext3_free_data(handle
, inode
, parent_bh
, first
, last
);
2365 int ext3_can_truncate(struct inode
*inode
)
2367 if (IS_APPEND(inode
) || IS_IMMUTABLE(inode
))
2369 if (S_ISREG(inode
->i_mode
))
2371 if (S_ISDIR(inode
->i_mode
))
2373 if (S_ISLNK(inode
->i_mode
))
2374 return !ext3_inode_is_fast_symlink(inode
);
2381 * We block out ext3_get_block() block instantiations across the entire
2382 * transaction, and VFS/VM ensures that ext3_truncate() cannot run
2383 * simultaneously on behalf of the same inode.
2385 * As we work through the truncate and commmit bits of it to the journal there
2386 * is one core, guiding principle: the file's tree must always be consistent on
2387 * disk. We must be able to restart the truncate after a crash.
2389 * The file's tree may be transiently inconsistent in memory (although it
2390 * probably isn't), but whenever we close off and commit a journal transaction,
2391 * the contents of (the filesystem + the journal) must be consistent and
2392 * restartable. It's pretty simple, really: bottom up, right to left (although
2393 * left-to-right works OK too).
2395 * Note that at recovery time, journal replay occurs *before* the restart of
2396 * truncate against the orphan inode list.
2398 * The committed inode has the new, desired i_size (which is the same as
2399 * i_disksize in this case). After a crash, ext3_orphan_cleanup() will see
2400 * that this inode's truncate did not complete and it will again call
2401 * ext3_truncate() to have another go. So there will be instantiated blocks
2402 * to the right of the truncation point in a crashed ext3 filesystem. But
2403 * that's fine - as long as they are linked from the inode, the post-crash
2404 * ext3_truncate() run will find them and release them.
2406 void ext3_truncate(struct inode
*inode
)
2409 struct ext3_inode_info
*ei
= EXT3_I(inode
);
2410 __le32
*i_data
= ei
->i_data
;
2411 int addr_per_block
= EXT3_ADDR_PER_BLOCK(inode
->i_sb
);
2412 struct address_space
*mapping
= inode
->i_mapping
;
2419 unsigned blocksize
= inode
->i_sb
->s_blocksize
;
2422 if (!ext3_can_truncate(inode
))
2425 if (inode
->i_size
== 0 && ext3_should_writeback_data(inode
))
2426 ext3_set_inode_state(inode
, EXT3_STATE_FLUSH_ON_CLOSE
);
2429 * We have to lock the EOF page here, because lock_page() nests
2430 * outside journal_start().
2432 if ((inode
->i_size
& (blocksize
- 1)) == 0) {
2433 /* Block boundary? Nothing to do */
2436 page
= grab_cache_page(mapping
,
2437 inode
->i_size
>> PAGE_CACHE_SHIFT
);
2442 handle
= start_transaction(inode
);
2443 if (IS_ERR(handle
)) {
2445 clear_highpage(page
);
2446 flush_dcache_page(page
);
2448 page_cache_release(page
);
2453 last_block
= (inode
->i_size
+ blocksize
-1)
2454 >> EXT3_BLOCK_SIZE_BITS(inode
->i_sb
);
2457 ext3_block_truncate_page(handle
, page
, mapping
, inode
->i_size
);
2459 n
= ext3_block_to_path(inode
, last_block
, offsets
, NULL
);
2461 goto out_stop
; /* error */
2464 * OK. This truncate is going to happen. We add the inode to the
2465 * orphan list, so that if this truncate spans multiple transactions,
2466 * and we crash, we will resume the truncate when the filesystem
2467 * recovers. It also marks the inode dirty, to catch the new size.
2469 * Implication: the file must always be in a sane, consistent
2470 * truncatable state while each transaction commits.
2472 if (ext3_orphan_add(handle
, inode
))
2476 * The orphan list entry will now protect us from any crash which
2477 * occurs before the truncate completes, so it is now safe to propagate
2478 * the new, shorter inode size (held for now in i_size) into the
2479 * on-disk inode. We do this via i_disksize, which is the value which
2480 * ext3 *really* writes onto the disk inode.
2482 ei
->i_disksize
= inode
->i_size
;
2485 * From here we block out all ext3_get_block() callers who want to
2486 * modify the block allocation tree.
2488 mutex_lock(&ei
->truncate_mutex
);
2490 if (n
== 1) { /* direct blocks */
2491 ext3_free_data(handle
, inode
, NULL
, i_data
+offsets
[0],
2492 i_data
+ EXT3_NDIR_BLOCKS
);
2496 partial
= ext3_find_shared(inode
, n
, offsets
, chain
, &nr
);
2497 /* Kill the top of shared branch (not detached) */
2499 if (partial
== chain
) {
2500 /* Shared branch grows from the inode */
2501 ext3_free_branches(handle
, inode
, NULL
,
2502 &nr
, &nr
+1, (chain
+n
-1) - partial
);
2505 * We mark the inode dirty prior to restart,
2506 * and prior to stop. No need for it here.
2509 /* Shared branch grows from an indirect block */
2510 BUFFER_TRACE(partial
->bh
, "get_write_access");
2511 ext3_free_branches(handle
, inode
, partial
->bh
,
2513 partial
->p
+1, (chain
+n
-1) - partial
);
2516 /* Clear the ends of indirect blocks on the shared branch */
2517 while (partial
> chain
) {
2518 ext3_free_branches(handle
, inode
, partial
->bh
, partial
->p
+ 1,
2519 (__le32
*)partial
->bh
->b_data
+addr_per_block
,
2520 (chain
+n
-1) - partial
);
2521 BUFFER_TRACE(partial
->bh
, "call brelse");
2522 brelse (partial
->bh
);
2526 /* Kill the remaining (whole) subtrees */
2527 switch (offsets
[0]) {
2529 nr
= i_data
[EXT3_IND_BLOCK
];
2531 ext3_free_branches(handle
, inode
, NULL
, &nr
, &nr
+1, 1);
2532 i_data
[EXT3_IND_BLOCK
] = 0;
2534 case EXT3_IND_BLOCK
:
2535 nr
= i_data
[EXT3_DIND_BLOCK
];
2537 ext3_free_branches(handle
, inode
, NULL
, &nr
, &nr
+1, 2);
2538 i_data
[EXT3_DIND_BLOCK
] = 0;
2540 case EXT3_DIND_BLOCK
:
2541 nr
= i_data
[EXT3_TIND_BLOCK
];
2543 ext3_free_branches(handle
, inode
, NULL
, &nr
, &nr
+1, 3);
2544 i_data
[EXT3_TIND_BLOCK
] = 0;
2546 case EXT3_TIND_BLOCK
:
2550 ext3_discard_reservation(inode
);
2552 mutex_unlock(&ei
->truncate_mutex
);
2553 inode
->i_mtime
= inode
->i_ctime
= CURRENT_TIME_SEC
;
2554 ext3_mark_inode_dirty(handle
, inode
);
2557 * In a multi-transaction truncate, we only make the final transaction
2564 * If this was a simple ftruncate(), and the file will remain alive
2565 * then we need to clear up the orphan record which we created above.
2566 * However, if this was a real unlink then we were called by
2567 * ext3_delete_inode(), and we allow that function to clean up the
2568 * orphan info for us.
2571 ext3_orphan_del(handle
, inode
);
2573 ext3_journal_stop(handle
);
2577 * Delete the inode from orphan list so that it doesn't stay there
2578 * forever and trigger assertion on umount.
2581 ext3_orphan_del(NULL
, inode
);
2584 static ext3_fsblk_t
ext3_get_inode_block(struct super_block
*sb
,
2585 unsigned long ino
, struct ext3_iloc
*iloc
)
2587 unsigned long block_group
;
2588 unsigned long offset
;
2590 struct ext3_group_desc
*gdp
;
2592 if (!ext3_valid_inum(sb
, ino
)) {
2594 * This error is already checked for in namei.c unless we are
2595 * looking at an NFS filehandle, in which case no error
2601 block_group
= (ino
- 1) / EXT3_INODES_PER_GROUP(sb
);
2602 gdp
= ext3_get_group_desc(sb
, block_group
, NULL
);
2606 * Figure out the offset within the block group inode table
2608 offset
= ((ino
- 1) % EXT3_INODES_PER_GROUP(sb
)) *
2609 EXT3_INODE_SIZE(sb
);
2610 block
= le32_to_cpu(gdp
->bg_inode_table
) +
2611 (offset
>> EXT3_BLOCK_SIZE_BITS(sb
));
2613 iloc
->block_group
= block_group
;
2614 iloc
->offset
= offset
& (EXT3_BLOCK_SIZE(sb
) - 1);
2619 * ext3_get_inode_loc returns with an extra refcount against the inode's
2620 * underlying buffer_head on success. If 'in_mem' is true, we have all
2621 * data in memory that is needed to recreate the on-disk version of this
2624 static int __ext3_get_inode_loc(struct inode
*inode
,
2625 struct ext3_iloc
*iloc
, int in_mem
)
2628 struct buffer_head
*bh
;
2630 block
= ext3_get_inode_block(inode
->i_sb
, inode
->i_ino
, iloc
);
2634 bh
= sb_getblk(inode
->i_sb
, block
);
2636 ext3_error (inode
->i_sb
, "ext3_get_inode_loc",
2637 "unable to read inode block - "
2638 "inode=%lu, block="E3FSBLK
,
2639 inode
->i_ino
, block
);
2642 if (!buffer_uptodate(bh
)) {
2646 * If the buffer has the write error flag, we have failed
2647 * to write out another inode in the same block. In this
2648 * case, we don't have to read the block because we may
2649 * read the old inode data successfully.
2651 if (buffer_write_io_error(bh
) && !buffer_uptodate(bh
))
2652 set_buffer_uptodate(bh
);
2654 if (buffer_uptodate(bh
)) {
2655 /* someone brought it uptodate while we waited */
2661 * If we have all information of the inode in memory and this
2662 * is the only valid inode in the block, we need not read the
2666 struct buffer_head
*bitmap_bh
;
2667 struct ext3_group_desc
*desc
;
2668 int inodes_per_buffer
;
2669 int inode_offset
, i
;
2673 block_group
= (inode
->i_ino
- 1) /
2674 EXT3_INODES_PER_GROUP(inode
->i_sb
);
2675 inodes_per_buffer
= bh
->b_size
/
2676 EXT3_INODE_SIZE(inode
->i_sb
);
2677 inode_offset
= ((inode
->i_ino
- 1) %
2678 EXT3_INODES_PER_GROUP(inode
->i_sb
));
2679 start
= inode_offset
& ~(inodes_per_buffer
- 1);
2681 /* Is the inode bitmap in cache? */
2682 desc
= ext3_get_group_desc(inode
->i_sb
,
2687 bitmap_bh
= sb_getblk(inode
->i_sb
,
2688 le32_to_cpu(desc
->bg_inode_bitmap
));
2693 * If the inode bitmap isn't in cache then the
2694 * optimisation may end up performing two reads instead
2695 * of one, so skip it.
2697 if (!buffer_uptodate(bitmap_bh
)) {
2701 for (i
= start
; i
< start
+ inodes_per_buffer
; i
++) {
2702 if (i
== inode_offset
)
2704 if (ext3_test_bit(i
, bitmap_bh
->b_data
))
2708 if (i
== start
+ inodes_per_buffer
) {
2709 /* all other inodes are free, so skip I/O */
2710 memset(bh
->b_data
, 0, bh
->b_size
);
2711 set_buffer_uptodate(bh
);
2719 * There are other valid inodes in the buffer, this inode
2720 * has in-inode xattrs, or we don't have this inode in memory.
2721 * Read the block from disk.
2724 bh
->b_end_io
= end_buffer_read_sync
;
2725 submit_bh(READ_META
, bh
);
2727 if (!buffer_uptodate(bh
)) {
2728 ext3_error(inode
->i_sb
, "ext3_get_inode_loc",
2729 "unable to read inode block - "
2730 "inode=%lu, block="E3FSBLK
,
2731 inode
->i_ino
, block
);
2741 int ext3_get_inode_loc(struct inode
*inode
, struct ext3_iloc
*iloc
)
2743 /* We have all inode data except xattrs in memory here. */
2744 return __ext3_get_inode_loc(inode
, iloc
,
2745 !ext3_test_inode_state(inode
, EXT3_STATE_XATTR
));
2748 void ext3_set_inode_flags(struct inode
*inode
)
2750 unsigned int flags
= EXT3_I(inode
)->i_flags
;
2752 inode
->i_flags
&= ~(S_SYNC
|S_APPEND
|S_IMMUTABLE
|S_NOATIME
|S_DIRSYNC
);
2753 if (flags
& EXT3_SYNC_FL
)
2754 inode
->i_flags
|= S_SYNC
;
2755 if (flags
& EXT3_APPEND_FL
)
2756 inode
->i_flags
|= S_APPEND
;
2757 if (flags
& EXT3_IMMUTABLE_FL
)
2758 inode
->i_flags
|= S_IMMUTABLE
;
2759 if (flags
& EXT3_NOATIME_FL
)
2760 inode
->i_flags
|= S_NOATIME
;
2761 if (flags
& EXT3_DIRSYNC_FL
)
2762 inode
->i_flags
|= S_DIRSYNC
;
2765 /* Propagate flags from i_flags to EXT3_I(inode)->i_flags */
2766 void ext3_get_inode_flags(struct ext3_inode_info
*ei
)
2768 unsigned int flags
= ei
->vfs_inode
.i_flags
;
2770 ei
->i_flags
&= ~(EXT3_SYNC_FL
|EXT3_APPEND_FL
|
2771 EXT3_IMMUTABLE_FL
|EXT3_NOATIME_FL
|EXT3_DIRSYNC_FL
);
2773 ei
->i_flags
|= EXT3_SYNC_FL
;
2774 if (flags
& S_APPEND
)
2775 ei
->i_flags
|= EXT3_APPEND_FL
;
2776 if (flags
& S_IMMUTABLE
)
2777 ei
->i_flags
|= EXT3_IMMUTABLE_FL
;
2778 if (flags
& S_NOATIME
)
2779 ei
->i_flags
|= EXT3_NOATIME_FL
;
2780 if (flags
& S_DIRSYNC
)
2781 ei
->i_flags
|= EXT3_DIRSYNC_FL
;
2784 struct inode
*ext3_iget(struct super_block
*sb
, unsigned long ino
)
2786 struct ext3_iloc iloc
;
2787 struct ext3_inode
*raw_inode
;
2788 struct ext3_inode_info
*ei
;
2789 struct buffer_head
*bh
;
2790 struct inode
*inode
;
2791 journal_t
*journal
= EXT3_SB(sb
)->s_journal
;
2792 transaction_t
*transaction
;
2796 inode
= iget_locked(sb
, ino
);
2798 return ERR_PTR(-ENOMEM
);
2799 if (!(inode
->i_state
& I_NEW
))
2803 ei
->i_block_alloc_info
= NULL
;
2805 ret
= __ext3_get_inode_loc(inode
, &iloc
, 0);
2809 raw_inode
= ext3_raw_inode(&iloc
);
2810 inode
->i_mode
= le16_to_cpu(raw_inode
->i_mode
);
2811 inode
->i_uid
= (uid_t
)le16_to_cpu(raw_inode
->i_uid_low
);
2812 inode
->i_gid
= (gid_t
)le16_to_cpu(raw_inode
->i_gid_low
);
2813 if(!(test_opt (inode
->i_sb
, NO_UID32
))) {
2814 inode
->i_uid
|= le16_to_cpu(raw_inode
->i_uid_high
) << 16;
2815 inode
->i_gid
|= le16_to_cpu(raw_inode
->i_gid_high
) << 16;
2817 inode
->i_nlink
= le16_to_cpu(raw_inode
->i_links_count
);
2818 inode
->i_size
= le32_to_cpu(raw_inode
->i_size
);
2819 inode
->i_atime
.tv_sec
= (signed)le32_to_cpu(raw_inode
->i_atime
);
2820 inode
->i_ctime
.tv_sec
= (signed)le32_to_cpu(raw_inode
->i_ctime
);
2821 inode
->i_mtime
.tv_sec
= (signed)le32_to_cpu(raw_inode
->i_mtime
);
2822 inode
->i_atime
.tv_nsec
= inode
->i_ctime
.tv_nsec
= inode
->i_mtime
.tv_nsec
= 0;
2824 ei
->i_state_flags
= 0;
2825 ei
->i_dir_start_lookup
= 0;
2826 ei
->i_dtime
= le32_to_cpu(raw_inode
->i_dtime
);
2827 /* We now have enough fields to check if the inode was active or not.
2828 * This is needed because nfsd might try to access dead inodes
2829 * the test is that same one that e2fsck uses
2830 * NeilBrown 1999oct15
2832 if (inode
->i_nlink
== 0) {
2833 if (inode
->i_mode
== 0 ||
2834 !(EXT3_SB(inode
->i_sb
)->s_mount_state
& EXT3_ORPHAN_FS
)) {
2835 /* this inode is deleted */
2840 /* The only unlinked inodes we let through here have
2841 * valid i_mode and are being read by the orphan
2842 * recovery code: that's fine, we're about to complete
2843 * the process of deleting those. */
2845 inode
->i_blocks
= le32_to_cpu(raw_inode
->i_blocks
);
2846 ei
->i_flags
= le32_to_cpu(raw_inode
->i_flags
);
2847 #ifdef EXT3_FRAGMENTS
2848 ei
->i_faddr
= le32_to_cpu(raw_inode
->i_faddr
);
2849 ei
->i_frag_no
= raw_inode
->i_frag
;
2850 ei
->i_frag_size
= raw_inode
->i_fsize
;
2852 ei
->i_file_acl
= le32_to_cpu(raw_inode
->i_file_acl
);
2853 if (!S_ISREG(inode
->i_mode
)) {
2854 ei
->i_dir_acl
= le32_to_cpu(raw_inode
->i_dir_acl
);
2857 ((__u64
)le32_to_cpu(raw_inode
->i_size_high
)) << 32;
2859 ei
->i_disksize
= inode
->i_size
;
2860 inode
->i_generation
= le32_to_cpu(raw_inode
->i_generation
);
2861 ei
->i_block_group
= iloc
.block_group
;
2863 * NOTE! The in-memory inode i_data array is in little-endian order
2864 * even on big-endian machines: we do NOT byteswap the block numbers!
2866 for (block
= 0; block
< EXT3_N_BLOCKS
; block
++)
2867 ei
->i_data
[block
] = raw_inode
->i_block
[block
];
2868 INIT_LIST_HEAD(&ei
->i_orphan
);
2871 * Set transaction id's of transactions that have to be committed
2872 * to finish f[data]sync. We set them to currently running transaction
2873 * as we cannot be sure that the inode or some of its metadata isn't
2874 * part of the transaction - the inode could have been reclaimed and
2875 * now it is reread from disk.
2880 spin_lock(&journal
->j_state_lock
);
2881 if (journal
->j_running_transaction
)
2882 transaction
= journal
->j_running_transaction
;
2884 transaction
= journal
->j_committing_transaction
;
2886 tid
= transaction
->t_tid
;
2888 tid
= journal
->j_commit_sequence
;
2889 spin_unlock(&journal
->j_state_lock
);
2890 atomic_set(&ei
->i_sync_tid
, tid
);
2891 atomic_set(&ei
->i_datasync_tid
, tid
);
2894 if (inode
->i_ino
>= EXT3_FIRST_INO(inode
->i_sb
) + 1 &&
2895 EXT3_INODE_SIZE(inode
->i_sb
) > EXT3_GOOD_OLD_INODE_SIZE
) {
2897 * When mke2fs creates big inodes it does not zero out
2898 * the unused bytes above EXT3_GOOD_OLD_INODE_SIZE,
2899 * so ignore those first few inodes.
2901 ei
->i_extra_isize
= le16_to_cpu(raw_inode
->i_extra_isize
);
2902 if (EXT3_GOOD_OLD_INODE_SIZE
+ ei
->i_extra_isize
>
2903 EXT3_INODE_SIZE(inode
->i_sb
)) {
2908 if (ei
->i_extra_isize
== 0) {
2909 /* The extra space is currently unused. Use it. */
2910 ei
->i_extra_isize
= sizeof(struct ext3_inode
) -
2911 EXT3_GOOD_OLD_INODE_SIZE
;
2913 __le32
*magic
= (void *)raw_inode
+
2914 EXT3_GOOD_OLD_INODE_SIZE
+
2916 if (*magic
== cpu_to_le32(EXT3_XATTR_MAGIC
))
2917 ext3_set_inode_state(inode
, EXT3_STATE_XATTR
);
2920 ei
->i_extra_isize
= 0;
2922 if (S_ISREG(inode
->i_mode
)) {
2923 inode
->i_op
= &ext3_file_inode_operations
;
2924 inode
->i_fop
= &ext3_file_operations
;
2925 ext3_set_aops(inode
);
2926 } else if (S_ISDIR(inode
->i_mode
)) {
2927 inode
->i_op
= &ext3_dir_inode_operations
;
2928 inode
->i_fop
= &ext3_dir_operations
;
2929 } else if (S_ISLNK(inode
->i_mode
)) {
2930 if (ext3_inode_is_fast_symlink(inode
)) {
2931 inode
->i_op
= &ext3_fast_symlink_inode_operations
;
2932 nd_terminate_link(ei
->i_data
, inode
->i_size
,
2933 sizeof(ei
->i_data
) - 1);
2935 inode
->i_op
= &ext3_symlink_inode_operations
;
2936 ext3_set_aops(inode
);
2939 inode
->i_op
= &ext3_special_inode_operations
;
2940 if (raw_inode
->i_block
[0])
2941 init_special_inode(inode
, inode
->i_mode
,
2942 old_decode_dev(le32_to_cpu(raw_inode
->i_block
[0])));
2944 init_special_inode(inode
, inode
->i_mode
,
2945 new_decode_dev(le32_to_cpu(raw_inode
->i_block
[1])));
2948 ext3_set_inode_flags(inode
);
2949 unlock_new_inode(inode
);
2954 return ERR_PTR(ret
);
2958 * Post the struct inode info into an on-disk inode location in the
2959 * buffer-cache. This gobbles the caller's reference to the
2960 * buffer_head in the inode location struct.
2962 * The caller must have write access to iloc->bh.
2964 static int ext3_do_update_inode(handle_t
*handle
,
2965 struct inode
*inode
,
2966 struct ext3_iloc
*iloc
)
2968 struct ext3_inode
*raw_inode
= ext3_raw_inode(iloc
);
2969 struct ext3_inode_info
*ei
= EXT3_I(inode
);
2970 struct buffer_head
*bh
= iloc
->bh
;
2971 int err
= 0, rc
, block
;
2974 /* we can't allow multiple procs in here at once, its a bit racey */
2977 /* For fields not not tracking in the in-memory inode,
2978 * initialise them to zero for new inodes. */
2979 if (ext3_test_inode_state(inode
, EXT3_STATE_NEW
))
2980 memset(raw_inode
, 0, EXT3_SB(inode
->i_sb
)->s_inode_size
);
2982 ext3_get_inode_flags(ei
);
2983 raw_inode
->i_mode
= cpu_to_le16(inode
->i_mode
);
2984 if(!(test_opt(inode
->i_sb
, NO_UID32
))) {
2985 raw_inode
->i_uid_low
= cpu_to_le16(low_16_bits(inode
->i_uid
));
2986 raw_inode
->i_gid_low
= cpu_to_le16(low_16_bits(inode
->i_gid
));
2988 * Fix up interoperability with old kernels. Otherwise, old inodes get
2989 * re-used with the upper 16 bits of the uid/gid intact
2992 raw_inode
->i_uid_high
=
2993 cpu_to_le16(high_16_bits(inode
->i_uid
));
2994 raw_inode
->i_gid_high
=
2995 cpu_to_le16(high_16_bits(inode
->i_gid
));
2997 raw_inode
->i_uid_high
= 0;
2998 raw_inode
->i_gid_high
= 0;
3001 raw_inode
->i_uid_low
=
3002 cpu_to_le16(fs_high2lowuid(inode
->i_uid
));
3003 raw_inode
->i_gid_low
=
3004 cpu_to_le16(fs_high2lowgid(inode
->i_gid
));
3005 raw_inode
->i_uid_high
= 0;
3006 raw_inode
->i_gid_high
= 0;
3008 raw_inode
->i_links_count
= cpu_to_le16(inode
->i_nlink
);
3009 raw_inode
->i_size
= cpu_to_le32(ei
->i_disksize
);
3010 raw_inode
->i_atime
= cpu_to_le32(inode
->i_atime
.tv_sec
);
3011 raw_inode
->i_ctime
= cpu_to_le32(inode
->i_ctime
.tv_sec
);
3012 raw_inode
->i_mtime
= cpu_to_le32(inode
->i_mtime
.tv_sec
);
3013 raw_inode
->i_blocks
= cpu_to_le32(inode
->i_blocks
);
3014 raw_inode
->i_dtime
= cpu_to_le32(ei
->i_dtime
);
3015 raw_inode
->i_flags
= cpu_to_le32(ei
->i_flags
);
3016 #ifdef EXT3_FRAGMENTS
3017 raw_inode
->i_faddr
= cpu_to_le32(ei
->i_faddr
);
3018 raw_inode
->i_frag
= ei
->i_frag_no
;
3019 raw_inode
->i_fsize
= ei
->i_frag_size
;
3021 raw_inode
->i_file_acl
= cpu_to_le32(ei
->i_file_acl
);
3022 if (!S_ISREG(inode
->i_mode
)) {
3023 raw_inode
->i_dir_acl
= cpu_to_le32(ei
->i_dir_acl
);
3025 raw_inode
->i_size_high
=
3026 cpu_to_le32(ei
->i_disksize
>> 32);
3027 if (ei
->i_disksize
> 0x7fffffffULL
) {
3028 struct super_block
*sb
= inode
->i_sb
;
3029 if (!EXT3_HAS_RO_COMPAT_FEATURE(sb
,
3030 EXT3_FEATURE_RO_COMPAT_LARGE_FILE
) ||
3031 EXT3_SB(sb
)->s_es
->s_rev_level
==
3032 cpu_to_le32(EXT3_GOOD_OLD_REV
)) {
3033 /* If this is the first large file
3034 * created, add a flag to the superblock.
3037 err
= ext3_journal_get_write_access(handle
,
3038 EXT3_SB(sb
)->s_sbh
);
3042 ext3_update_dynamic_rev(sb
);
3043 EXT3_SET_RO_COMPAT_FEATURE(sb
,
3044 EXT3_FEATURE_RO_COMPAT_LARGE_FILE
);
3046 err
= ext3_journal_dirty_metadata(handle
,
3047 EXT3_SB(sb
)->s_sbh
);
3048 /* get our lock and start over */
3053 raw_inode
->i_generation
= cpu_to_le32(inode
->i_generation
);
3054 if (S_ISCHR(inode
->i_mode
) || S_ISBLK(inode
->i_mode
)) {
3055 if (old_valid_dev(inode
->i_rdev
)) {
3056 raw_inode
->i_block
[0] =
3057 cpu_to_le32(old_encode_dev(inode
->i_rdev
));
3058 raw_inode
->i_block
[1] = 0;
3060 raw_inode
->i_block
[0] = 0;
3061 raw_inode
->i_block
[1] =
3062 cpu_to_le32(new_encode_dev(inode
->i_rdev
));
3063 raw_inode
->i_block
[2] = 0;
3065 } else for (block
= 0; block
< EXT3_N_BLOCKS
; block
++)
3066 raw_inode
->i_block
[block
] = ei
->i_data
[block
];
3068 if (ei
->i_extra_isize
)
3069 raw_inode
->i_extra_isize
= cpu_to_le16(ei
->i_extra_isize
);
3071 BUFFER_TRACE(bh
, "call ext3_journal_dirty_metadata");
3073 rc
= ext3_journal_dirty_metadata(handle
, bh
);
3076 ext3_clear_inode_state(inode
, EXT3_STATE_NEW
);
3078 atomic_set(&ei
->i_sync_tid
, handle
->h_transaction
->t_tid
);
3081 ext3_std_error(inode
->i_sb
, err
);
3086 * ext3_write_inode()
3088 * We are called from a few places:
3090 * - Within generic_file_write() for O_SYNC files.
3091 * Here, there will be no transaction running. We wait for any running
3092 * trasnaction to commit.
3094 * - Within sys_sync(), kupdate and such.
3095 * We wait on commit, if tol to.
3097 * - Within prune_icache() (PF_MEMALLOC == true)
3098 * Here we simply return. We can't afford to block kswapd on the
3101 * In all cases it is actually safe for us to return without doing anything,
3102 * because the inode has been copied into a raw inode buffer in
3103 * ext3_mark_inode_dirty(). This is a correctness thing for O_SYNC and for
3106 * Note that we are absolutely dependent upon all inode dirtiers doing the
3107 * right thing: they *must* call mark_inode_dirty() after dirtying info in
3108 * which we are interested.
3110 * It would be a bug for them to not do this. The code:
3112 * mark_inode_dirty(inode)
3114 * inode->i_size = expr;
3116 * is in error because a kswapd-driven write_inode() could occur while
3117 * `stuff()' is running, and the new i_size will be lost. Plus the inode
3118 * will no longer be on the superblock's dirty inode list.
3120 int ext3_write_inode(struct inode
*inode
, struct writeback_control
*wbc
)
3122 if (current
->flags
& PF_MEMALLOC
)
3125 if (ext3_journal_current_handle()) {
3126 jbd_debug(1, "called recursively, non-PF_MEMALLOC!\n");
3131 if (wbc
->sync_mode
!= WB_SYNC_ALL
)
3134 return ext3_force_commit(inode
->i_sb
);
3140 * Called from notify_change.
3142 * We want to trap VFS attempts to truncate the file as soon as
3143 * possible. In particular, we want to make sure that when the VFS
3144 * shrinks i_size, we put the inode on the orphan list and modify
3145 * i_disksize immediately, so that during the subsequent flushing of
3146 * dirty pages and freeing of disk blocks, we can guarantee that any
3147 * commit will leave the blocks being flushed in an unused state on
3148 * disk. (On recovery, the inode will get truncated and the blocks will
3149 * be freed, so we have a strong guarantee that no future commit will
3150 * leave these blocks visible to the user.)
3152 * Called with inode->sem down.
3154 int ext3_setattr(struct dentry
*dentry
, struct iattr
*attr
)
3156 struct inode
*inode
= dentry
->d_inode
;
3158 const unsigned int ia_valid
= attr
->ia_valid
;
3160 error
= inode_change_ok(inode
, attr
);
3164 if (is_quota_modification(inode
, attr
))
3165 dquot_initialize(inode
);
3166 if ((ia_valid
& ATTR_UID
&& attr
->ia_uid
!= inode
->i_uid
) ||
3167 (ia_valid
& ATTR_GID
&& attr
->ia_gid
!= inode
->i_gid
)) {
3170 /* (user+group)*(old+new) structure, inode write (sb,
3171 * inode block, ? - but truncate inode update has it) */
3172 handle
= ext3_journal_start(inode
, EXT3_MAXQUOTAS_INIT_BLOCKS(inode
->i_sb
)+
3173 EXT3_MAXQUOTAS_DEL_BLOCKS(inode
->i_sb
)+3);
3174 if (IS_ERR(handle
)) {
3175 error
= PTR_ERR(handle
);
3178 error
= dquot_transfer(inode
, attr
);
3180 ext3_journal_stop(handle
);
3183 /* Update corresponding info in inode so that everything is in
3184 * one transaction */
3185 if (attr
->ia_valid
& ATTR_UID
)
3186 inode
->i_uid
= attr
->ia_uid
;
3187 if (attr
->ia_valid
& ATTR_GID
)
3188 inode
->i_gid
= attr
->ia_gid
;
3189 error
= ext3_mark_inode_dirty(handle
, inode
);
3190 ext3_journal_stop(handle
);
3193 if (S_ISREG(inode
->i_mode
) &&
3194 attr
->ia_valid
& ATTR_SIZE
&& attr
->ia_size
< inode
->i_size
) {
3197 handle
= ext3_journal_start(inode
, 3);
3198 if (IS_ERR(handle
)) {
3199 error
= PTR_ERR(handle
);
3203 error
= ext3_orphan_add(handle
, inode
);
3204 EXT3_I(inode
)->i_disksize
= attr
->ia_size
;
3205 rc
= ext3_mark_inode_dirty(handle
, inode
);
3208 ext3_journal_stop(handle
);
3211 if ((attr
->ia_valid
& ATTR_SIZE
) &&
3212 attr
->ia_size
!= i_size_read(inode
)) {
3213 rc
= vmtruncate(inode
, attr
->ia_size
);
3218 setattr_copy(inode
, attr
);
3219 mark_inode_dirty(inode
);
3221 if (ia_valid
& ATTR_MODE
)
3222 rc
= ext3_acl_chmod(inode
);
3225 ext3_std_error(inode
->i_sb
, error
);
3233 * How many blocks doth make a writepage()?
3235 * With N blocks per page, it may be:
3240 * N+5 bitmap blocks (from the above)
3241 * N+5 group descriptor summary blocks
3244 * 2 * EXT3_SINGLEDATA_TRANS_BLOCKS for the quote files
3246 * 3 * (N + 5) + 2 + 2 * EXT3_SINGLEDATA_TRANS_BLOCKS
3248 * With ordered or writeback data it's the same, less the N data blocks.
3250 * If the inode's direct blocks can hold an integral number of pages then a
3251 * page cannot straddle two indirect blocks, and we can only touch one indirect
3252 * and dindirect block, and the "5" above becomes "3".
3254 * This still overestimates under most circumstances. If we were to pass the
3255 * start and end offsets in here as well we could do block_to_path() on each
3256 * block and work out the exact number of indirects which are touched. Pah.
3259 static int ext3_writepage_trans_blocks(struct inode
*inode
)
3261 int bpp
= ext3_journal_blocks_per_page(inode
);
3262 int indirects
= (EXT3_NDIR_BLOCKS
% bpp
) ? 5 : 3;
3265 if (ext3_should_journal_data(inode
))
3266 ret
= 3 * (bpp
+ indirects
) + 2;
3268 ret
= 2 * (bpp
+ indirects
) + 2;
3271 /* We know that structure was already allocated during dquot_initialize so
3272 * we will be updating only the data blocks + inodes */
3273 ret
+= EXT3_MAXQUOTAS_TRANS_BLOCKS(inode
->i_sb
);
3280 * The caller must have previously called ext3_reserve_inode_write().
3281 * Give this, we know that the caller already has write access to iloc->bh.
3283 int ext3_mark_iloc_dirty(handle_t
*handle
,
3284 struct inode
*inode
, struct ext3_iloc
*iloc
)
3288 /* the do_update_inode consumes one bh->b_count */
3291 /* ext3_do_update_inode() does journal_dirty_metadata */
3292 err
= ext3_do_update_inode(handle
, inode
, iloc
);
3298 * On success, We end up with an outstanding reference count against
3299 * iloc->bh. This _must_ be cleaned up later.
3303 ext3_reserve_inode_write(handle_t
*handle
, struct inode
*inode
,
3304 struct ext3_iloc
*iloc
)
3308 err
= ext3_get_inode_loc(inode
, iloc
);
3310 BUFFER_TRACE(iloc
->bh
, "get_write_access");
3311 err
= ext3_journal_get_write_access(handle
, iloc
->bh
);
3318 ext3_std_error(inode
->i_sb
, err
);
3323 * What we do here is to mark the in-core inode as clean with respect to inode
3324 * dirtiness (it may still be data-dirty).
3325 * This means that the in-core inode may be reaped by prune_icache
3326 * without having to perform any I/O. This is a very good thing,
3327 * because *any* task may call prune_icache - even ones which
3328 * have a transaction open against a different journal.
3330 * Is this cheating? Not really. Sure, we haven't written the
3331 * inode out, but prune_icache isn't a user-visible syncing function.
3332 * Whenever the user wants stuff synced (sys_sync, sys_msync, sys_fsync)
3333 * we start and wait on commits.
3335 * Is this efficient/effective? Well, we're being nice to the system
3336 * by cleaning up our inodes proactively so they can be reaped
3337 * without I/O. But we are potentially leaving up to five seconds'
3338 * worth of inodes floating about which prune_icache wants us to
3339 * write out. One way to fix that would be to get prune_icache()
3340 * to do a write_super() to free up some memory. It has the desired
3343 int ext3_mark_inode_dirty(handle_t
*handle
, struct inode
*inode
)
3345 struct ext3_iloc iloc
;
3349 err
= ext3_reserve_inode_write(handle
, inode
, &iloc
);
3351 err
= ext3_mark_iloc_dirty(handle
, inode
, &iloc
);
3356 * ext3_dirty_inode() is called from __mark_inode_dirty()
3358 * We're really interested in the case where a file is being extended.
3359 * i_size has been changed by generic_commit_write() and we thus need
3360 * to include the updated inode in the current transaction.
3362 * Also, dquot_alloc_space() will always dirty the inode when blocks
3363 * are allocated to the file.
3365 * If the inode is marked synchronous, we don't honour that here - doing
3366 * so would cause a commit on atime updates, which we don't bother doing.
3367 * We handle synchronous inodes at the highest possible level.
3369 void ext3_dirty_inode(struct inode
*inode
)
3371 handle_t
*current_handle
= ext3_journal_current_handle();
3374 handle
= ext3_journal_start(inode
, 2);
3377 if (current_handle
&&
3378 current_handle
->h_transaction
!= handle
->h_transaction
) {
3379 /* This task has a transaction open against a different fs */
3380 printk(KERN_EMERG
"%s: transactions do not match!\n",
3383 jbd_debug(5, "marking dirty. outer handle=%p\n",
3385 ext3_mark_inode_dirty(handle
, inode
);
3387 ext3_journal_stop(handle
);
3394 * Bind an inode's backing buffer_head into this transaction, to prevent
3395 * it from being flushed to disk early. Unlike
3396 * ext3_reserve_inode_write, this leaves behind no bh reference and
3397 * returns no iloc structure, so the caller needs to repeat the iloc
3398 * lookup to mark the inode dirty later.
3400 static int ext3_pin_inode(handle_t
*handle
, struct inode
*inode
)
3402 struct ext3_iloc iloc
;
3406 err
= ext3_get_inode_loc(inode
, &iloc
);
3408 BUFFER_TRACE(iloc
.bh
, "get_write_access");
3409 err
= journal_get_write_access(handle
, iloc
.bh
);
3411 err
= ext3_journal_dirty_metadata(handle
,
3416 ext3_std_error(inode
->i_sb
, err
);
3421 int ext3_change_inode_journal_flag(struct inode
*inode
, int val
)
3428 * We have to be very careful here: changing a data block's
3429 * journaling status dynamically is dangerous. If we write a
3430 * data block to the journal, change the status and then delete
3431 * that block, we risk forgetting to revoke the old log record
3432 * from the journal and so a subsequent replay can corrupt data.
3433 * So, first we make sure that the journal is empty and that
3434 * nobody is changing anything.
3437 journal
= EXT3_JOURNAL(inode
);
3438 if (is_journal_aborted(journal
))
3441 journal_lock_updates(journal
);
3442 journal_flush(journal
);
3445 * OK, there are no updates running now, and all cached data is
3446 * synced to disk. We are now in a completely consistent state
3447 * which doesn't have anything in the journal, and we know that
3448 * no filesystem updates are running, so it is safe to modify
3449 * the inode's in-core data-journaling state flag now.
3453 EXT3_I(inode
)->i_flags
|= EXT3_JOURNAL_DATA_FL
;
3455 EXT3_I(inode
)->i_flags
&= ~EXT3_JOURNAL_DATA_FL
;
3456 ext3_set_aops(inode
);
3458 journal_unlock_updates(journal
);
3460 /* Finally we can mark the inode as dirty. */
3462 handle
= ext3_journal_start(inode
, 1);
3464 return PTR_ERR(handle
);
3466 err
= ext3_mark_inode_dirty(handle
, inode
);
3468 ext3_journal_stop(handle
);
3469 ext3_std_error(inode
->i_sb
, err
);