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/smp_lock.h>
31 #include <linux/highuid.h>
32 #include <linux/pagemap.h>
33 #include <linux/quotaops.h>
34 #include <linux/string.h>
35 #include <linux/buffer_head.h>
36 #include <linux/writeback.h>
37 #include <linux/mpage.h>
38 #include <linux/uio.h>
43 * Test whether an inode is a fast symlink.
45 static inline int ext3_inode_is_fast_symlink(struct inode
*inode
)
47 int ea_blocks
= EXT3_I(inode
)->i_file_acl
?
48 (inode
->i_sb
->s_blocksize
>> 9) : 0;
50 return (S_ISLNK(inode
->i_mode
) &&
51 inode
->i_blocks
- ea_blocks
== 0);
54 /* The ext3 forget function must perform a revoke if we are freeing data
55 * which has been journaled. Metadata (eg. indirect blocks) must be
56 * revoked in all cases.
58 * "bh" may be NULL: a metadata block may have been freed from memory
59 * but there may still be a record of it in the journal, and that record
60 * still needs to be revoked.
63 int ext3_forget(handle_t
*handle
, int is_metadata
,
64 struct inode
*inode
, struct buffer_head
*bh
,
69 BUFFER_TRACE(bh
, "enter");
71 jbd_debug(4, "forgetting bh %p: is_metadata = %d, mode %o, "
73 bh
, is_metadata
, inode
->i_mode
,
74 test_opt(inode
->i_sb
, DATA_FLAGS
));
76 /* Never use the revoke function if we are doing full data
77 * journaling: there is no need to, and a V1 superblock won't
78 * support it. Otherwise, only skip the revoke on un-journaled
81 if (test_opt(inode
->i_sb
, DATA_FLAGS
) == EXT3_MOUNT_JOURNAL_DATA
||
82 (!is_metadata
&& !ext3_should_journal_data(inode
))) {
84 BUFFER_TRACE(bh
, "call journal_forget");
85 ext3_journal_forget(handle
, bh
);
91 * data!=journal && (is_metadata || should_journal_data(inode))
93 BUFFER_TRACE(bh
, "call ext3_journal_revoke");
94 err
= ext3_journal_revoke(handle
, blocknr
, bh
);
96 ext3_abort(inode
->i_sb
, __FUNCTION__
,
97 "error %d when attempting revoke", err
);
98 BUFFER_TRACE(bh
, "exit");
103 * Work out how many blocks we need to progress with the next chunk of a
104 * truncate transaction.
107 static unsigned long blocks_for_truncate(struct inode
*inode
)
109 unsigned long needed
;
111 needed
= inode
->i_blocks
>> (inode
->i_sb
->s_blocksize_bits
- 9);
113 /* Give ourselves just enough room to cope with inodes in which
114 * i_blocks is corrupt: we've seen disk corruptions in the past
115 * which resulted in random data in an inode which looked enough
116 * like a regular file for ext3 to try to delete it. Things
117 * will go a bit crazy if that happens, but at least we should
118 * try not to panic the whole kernel. */
122 /* But we need to bound the transaction so we don't overflow the
124 if (needed
> EXT3_MAX_TRANS_DATA
)
125 needed
= EXT3_MAX_TRANS_DATA
;
127 return EXT3_DATA_TRANS_BLOCKS
+ needed
;
131 * Truncate transactions can be complex and absolutely huge. So we need to
132 * be able to restart the transaction at a conventient checkpoint to make
133 * sure we don't overflow the journal.
135 * start_transaction gets us a new handle for a truncate transaction,
136 * and extend_transaction tries to extend the existing one a bit. If
137 * extend fails, we need to propagate the failure up and restart the
138 * transaction in the top-level truncate loop. --sct
141 static handle_t
*start_transaction(struct inode
*inode
)
145 result
= ext3_journal_start(inode
, blocks_for_truncate(inode
));
149 ext3_std_error(inode
->i_sb
, PTR_ERR(result
));
154 * Try to extend this transaction for the purposes of truncation.
156 * Returns 0 if we managed to create more room. If we can't create more
157 * room, and the transaction must be restarted we return 1.
159 static int try_to_extend_transaction(handle_t
*handle
, struct inode
*inode
)
161 if (handle
->h_buffer_credits
> EXT3_RESERVE_TRANS_BLOCKS
)
163 if (!ext3_journal_extend(handle
, blocks_for_truncate(inode
)))
169 * Restart the transaction associated with *handle. This does a commit,
170 * so before we call here everything must be consistently dirtied against
173 static int ext3_journal_test_restart(handle_t
*handle
, struct inode
*inode
)
175 jbd_debug(2, "restarting handle %p\n", handle
);
176 return ext3_journal_restart(handle
, blocks_for_truncate(inode
));
180 * Called at each iput()
182 * The inode may be "bad" if ext3_read_inode() saw an error from
183 * ext3_get_inode(), so we need to check that to avoid freeing random disk
186 void ext3_put_inode(struct inode
*inode
)
188 if (!is_bad_inode(inode
))
189 ext3_discard_prealloc(inode
);
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
))
202 handle
= start_transaction(inode
);
203 if (IS_ERR(handle
)) {
204 /* If we're going to skip the normal cleanup, we still
205 * need to make sure that the in-core orphan linked list
206 * is properly cleaned up. */
207 ext3_orphan_del(NULL
, inode
);
209 ext3_std_error(inode
->i_sb
, PTR_ERR(handle
));
217 ext3_truncate(inode
);
219 * Kill off the orphan record which ext3_truncate created.
220 * AKPM: I think this can be inside the above `if'.
221 * Note that ext3_orphan_del() has to be able to cope with the
222 * deletion of a non-existent orphan - this is because we don't
223 * know if ext3_truncate() actually created an orphan record.
224 * (Well, we could do this if we need to, but heck - it works)
226 ext3_orphan_del(handle
, inode
);
227 EXT3_I(inode
)->i_dtime
= get_seconds();
230 * One subtle ordering requirement: if anything has gone wrong
231 * (transaction abort, IO errors, whatever), then we can still
232 * do these next steps (the fs will already have been marked as
233 * having errors), but we can't free the inode if the mark_dirty
236 if (ext3_mark_inode_dirty(handle
, inode
))
237 /* If that failed, just do the required in-core inode clear. */
240 ext3_free_inode(handle
, inode
);
241 ext3_journal_stop(handle
);
244 clear_inode(inode
); /* We must guarantee clearing of inode... */
247 void ext3_discard_prealloc (struct inode
* inode
)
249 #ifdef EXT3_PREALLOCATE
250 struct ext3_inode_info
*ei
= EXT3_I(inode
);
251 /* Writer: ->i_prealloc* */
252 if (ei
->i_prealloc_count
) {
253 unsigned short total
= ei
->i_prealloc_count
;
254 unsigned long block
= ei
->i_prealloc_block
;
255 ei
->i_prealloc_count
= 0;
256 ei
->i_prealloc_block
= 0;
258 ext3_free_blocks (inode
, block
, total
);
263 static int ext3_alloc_block (handle_t
*handle
,
264 struct inode
* inode
, unsigned long goal
, int *err
)
266 unsigned long result
;
268 #ifdef EXT3_PREALLOCATE
270 static unsigned long alloc_hits
= 0, alloc_attempts
= 0;
272 struct ext3_inode_info
*ei
= EXT3_I(inode
);
273 /* Writer: ->i_prealloc* */
274 if (ei
->i_prealloc_count
&&
275 (goal
== ei
->i_prealloc_block
||
276 goal
+ 1 == ei
->i_prealloc_block
))
278 result
= ei
->i_prealloc_block
++;
279 ei
->i_prealloc_count
--;
281 ext3_debug ("preallocation hit (%lu/%lu).\n",
282 ++alloc_hits
, ++alloc_attempts
);
284 ext3_discard_prealloc (inode
);
285 ext3_debug ("preallocation miss (%lu/%lu).\n",
286 alloc_hits
, ++alloc_attempts
);
287 if (S_ISREG(inode
->i_mode
))
288 result
= ext3_new_block (inode
, goal
,
289 &ei
->i_prealloc_count
,
290 &ei
->i_prealloc_block
, err
);
292 result
= ext3_new_block (inode
, goal
, 0, 0, err
);
294 * AKPM: this is somewhat sticky. I'm not surprised it was
295 * disabled in 2.2's ext3. Need to integrate b_committed_data
296 * guarding with preallocation, if indeed preallocation is
301 result
= ext3_new_block (handle
, inode
, goal
, 0, 0, err
);
310 struct buffer_head
*bh
;
313 static inline void add_chain(Indirect
*p
, struct buffer_head
*bh
, u32
*v
)
315 p
->key
= *(p
->p
= v
);
319 static inline int verify_chain(Indirect
*from
, Indirect
*to
)
321 while (from
<= to
&& from
->key
== *from
->p
)
327 * ext3_block_to_path - parse the block number into array of offsets
328 * @inode: inode in question (we are only interested in its superblock)
329 * @i_block: block number to be parsed
330 * @offsets: array to store the offsets in
331 * @boundary: set this non-zero if the referred-to block is likely to be
332 * followed (on disk) by an indirect block.
334 * To store the locations of file's data ext3 uses a data structure common
335 * for UNIX filesystems - tree of pointers anchored in the inode, with
336 * data blocks at leaves and indirect blocks in intermediate nodes.
337 * This function translates the block number into path in that tree -
338 * return value is the path length and @offsets[n] is the offset of
339 * pointer to (n+1)th node in the nth one. If @block is out of range
340 * (negative or too large) warning is printed and zero returned.
342 * Note: function doesn't find node addresses, so no IO is needed. All
343 * we need to know is the capacity of indirect blocks (taken from the
348 * Portability note: the last comparison (check that we fit into triple
349 * indirect block) is spelled differently, because otherwise on an
350 * architecture with 32-bit longs and 8Kb pages we might get into trouble
351 * if our filesystem had 8Kb blocks. We might use long long, but that would
352 * kill us on x86. Oh, well, at least the sign propagation does not matter -
353 * i_block would have to be negative in the very beginning, so we would not
357 static int ext3_block_to_path(struct inode
*inode
,
358 long i_block
, int offsets
[4], int *boundary
)
360 int ptrs
= EXT3_ADDR_PER_BLOCK(inode
->i_sb
);
361 int ptrs_bits
= EXT3_ADDR_PER_BLOCK_BITS(inode
->i_sb
);
362 const long direct_blocks
= EXT3_NDIR_BLOCKS
,
363 indirect_blocks
= ptrs
,
364 double_blocks
= (1 << (ptrs_bits
* 2));
369 ext3_warning (inode
->i_sb
, "ext3_block_to_path", "block < 0");
370 } else if (i_block
< direct_blocks
) {
371 offsets
[n
++] = i_block
;
372 final
= direct_blocks
;
373 } else if ( (i_block
-= direct_blocks
) < indirect_blocks
) {
374 offsets
[n
++] = EXT3_IND_BLOCK
;
375 offsets
[n
++] = i_block
;
377 } else if ((i_block
-= indirect_blocks
) < double_blocks
) {
378 offsets
[n
++] = EXT3_DIND_BLOCK
;
379 offsets
[n
++] = i_block
>> ptrs_bits
;
380 offsets
[n
++] = i_block
& (ptrs
- 1);
382 } else if (((i_block
-= double_blocks
) >> (ptrs_bits
* 2)) < ptrs
) {
383 offsets
[n
++] = EXT3_TIND_BLOCK
;
384 offsets
[n
++] = i_block
>> (ptrs_bits
* 2);
385 offsets
[n
++] = (i_block
>> ptrs_bits
) & (ptrs
- 1);
386 offsets
[n
++] = i_block
& (ptrs
- 1);
389 ext3_warning (inode
->i_sb
, "ext3_block_to_path", "block > big");
392 *boundary
= (i_block
& (ptrs
- 1)) == (final
- 1);
397 * ext3_get_branch - read the chain of indirect blocks leading to data
398 * @inode: inode in question
399 * @depth: depth of the chain (1 - direct pointer, etc.)
400 * @offsets: offsets of pointers in inode/indirect blocks
401 * @chain: place to store the result
402 * @err: here we store the error value
404 * Function fills the array of triples <key, p, bh> and returns %NULL
405 * if everything went OK or the pointer to the last filled triple
406 * (incomplete one) otherwise. Upon the return chain[i].key contains
407 * the number of (i+1)-th block in the chain (as it is stored in memory,
408 * i.e. little-endian 32-bit), chain[i].p contains the address of that
409 * number (it points into struct inode for i==0 and into the bh->b_data
410 * for i>0) and chain[i].bh points to the buffer_head of i-th indirect
411 * block for i>0 and NULL for i==0. In other words, it holds the block
412 * numbers of the chain, addresses they were taken from (and where we can
413 * verify that chain did not change) and buffer_heads hosting these
416 * Function stops when it stumbles upon zero pointer (absent block)
417 * (pointer to last triple returned, *@err == 0)
418 * or when it gets an IO error reading an indirect block
419 * (ditto, *@err == -EIO)
420 * or when it notices that chain had been changed while it was reading
421 * (ditto, *@err == -EAGAIN)
422 * or when it reads all @depth-1 indirect blocks successfully and finds
423 * the whole chain, all way to the data (returns %NULL, *err == 0).
425 static Indirect
*ext3_get_branch(struct inode
*inode
, int depth
, int *offsets
,
426 Indirect chain
[4], int *err
)
428 struct super_block
*sb
= inode
->i_sb
;
430 struct buffer_head
*bh
;
433 /* i_data is not going away, no lock needed */
434 add_chain (chain
, NULL
, EXT3_I(inode
)->i_data
+ *offsets
);
438 bh
= sb_bread(sb
, le32_to_cpu(p
->key
));
441 /* Reader: pointers */
442 if (!verify_chain(chain
, p
))
444 add_chain(++p
, bh
, (u32
*)bh
->b_data
+ *++offsets
);
462 * ext3_find_near - find a place for allocation with sufficient locality
464 * @ind: descriptor of indirect block.
466 * This function returns the prefered place for block allocation.
467 * It is used when heuristic for sequential allocation fails.
469 * + if there is a block to the left of our position - allocate near it.
470 * + if pointer will live in indirect block - allocate near that block.
471 * + if pointer will live in inode - allocate in the same
474 * In the latter case we colour the starting block by the callers PID to
475 * prevent it from clashing with concurrent allocations for a different inode
476 * in the same block group. The PID is used here so that functionally related
477 * files will be close-by on-disk.
479 * Caller must make sure that @ind is valid and will stay that way.
482 static unsigned long ext3_find_near(struct inode
*inode
, Indirect
*ind
)
484 struct ext3_inode_info
*ei
= EXT3_I(inode
);
485 u32
*start
= ind
->bh
? (u32
*) ind
->bh
->b_data
: ei
->i_data
;
487 unsigned long bg_start
;
488 unsigned long colour
;
490 /* Try to find previous block */
491 for (p
= ind
->p
- 1; p
>= start
; p
--)
493 return le32_to_cpu(*p
);
495 /* No such thing, so let's try location of indirect block */
497 return ind
->bh
->b_blocknr
;
500 * It is going to be refered from inode itself? OK, just put it into
501 * the same cylinder group then.
503 bg_start
= (ei
->i_block_group
* EXT3_BLOCKS_PER_GROUP(inode
->i_sb
)) +
504 le32_to_cpu(EXT3_SB(inode
->i_sb
)->s_es
->s_first_data_block
);
505 colour
= (current
->pid
% 16) *
506 (EXT3_BLOCKS_PER_GROUP(inode
->i_sb
) / 16);
507 return bg_start
+ colour
;
511 * ext3_find_goal - find a prefered place for allocation.
513 * @block: block we want
514 * @chain: chain of indirect blocks
515 * @partial: pointer to the last triple within a chain
516 * @goal: place to store the result.
518 * Normally this function find the prefered place for block allocation,
519 * stores it in *@goal and returns zero. If the branch had been changed
520 * under us we return -EAGAIN.
523 static int ext3_find_goal(struct inode
*inode
, long block
, Indirect chain
[4],
524 Indirect
*partial
, unsigned long *goal
)
526 struct ext3_inode_info
*ei
= EXT3_I(inode
);
527 /* Writer: ->i_next_alloc* */
528 if (block
== ei
->i_next_alloc_block
+ 1) {
529 ei
->i_next_alloc_block
++;
530 ei
->i_next_alloc_goal
++;
533 /* Reader: pointers, ->i_next_alloc* */
534 if (verify_chain(chain
, partial
)) {
536 * try the heuristic for sequential allocation,
537 * failing that at least try to get decent locality.
539 if (block
== ei
->i_next_alloc_block
)
540 *goal
= ei
->i_next_alloc_goal
;
542 *goal
= ext3_find_near(inode
, partial
);
550 * ext3_alloc_branch - allocate and set up a chain of blocks.
552 * @num: depth of the chain (number of blocks to allocate)
553 * @offsets: offsets (in the blocks) to store the pointers to next.
554 * @branch: place to store the chain in.
556 * This function allocates @num blocks, zeroes out all but the last one,
557 * links them into chain and (if we are synchronous) writes them to disk.
558 * In other words, it prepares a branch that can be spliced onto the
559 * inode. It stores the information about that chain in the branch[], in
560 * the same format as ext3_get_branch() would do. We are calling it after
561 * we had read the existing part of chain and partial points to the last
562 * triple of that (one with zero ->key). Upon the exit we have the same
563 * picture as after the successful ext3_get_block(), excpet that in one
564 * place chain is disconnected - *branch->p is still zero (we did not
565 * set the last link), but branch->key contains the number that should
566 * be placed into *branch->p to fill that gap.
568 * If allocation fails we free all blocks we've allocated (and forget
569 * their buffer_heads) and return the error value the from failed
570 * ext3_alloc_block() (normally -ENOSPC). Otherwise we set the chain
571 * as described above and return 0.
574 static int ext3_alloc_branch(handle_t
*handle
, struct inode
*inode
,
580 int blocksize
= inode
->i_sb
->s_blocksize
;
584 int parent
= ext3_alloc_block(handle
, inode
, goal
, &err
);
586 branch
[0].key
= cpu_to_le32(parent
);
588 for (n
= 1; n
< num
; n
++) {
589 struct buffer_head
*bh
;
590 /* Allocate the next block */
591 int nr
= ext3_alloc_block(handle
, inode
, parent
, &err
);
594 branch
[n
].key
= cpu_to_le32(nr
);
598 * Get buffer_head for parent block, zero it out
599 * and set the pointer to new one, then send
602 bh
= sb_getblk(inode
->i_sb
, parent
);
605 BUFFER_TRACE(bh
, "call get_create_access");
606 err
= ext3_journal_get_create_access(handle
, bh
);
613 memset(bh
->b_data
, 0, blocksize
);
614 branch
[n
].p
= (u32
*) bh
->b_data
+ offsets
[n
];
615 *branch
[n
].p
= branch
[n
].key
;
616 BUFFER_TRACE(bh
, "marking uptodate");
617 set_buffer_uptodate(bh
);
620 BUFFER_TRACE(bh
, "call ext3_journal_dirty_metadata");
621 err
= ext3_journal_dirty_metadata(handle
, bh
);
631 /* Allocation failed, free what we already allocated */
632 for (i
= 1; i
< keys
; i
++) {
633 BUFFER_TRACE(branch
[i
].bh
, "call journal_forget");
634 ext3_journal_forget(handle
, branch
[i
].bh
);
636 for (i
= 0; i
< keys
; i
++)
637 ext3_free_blocks(handle
, inode
, le32_to_cpu(branch
[i
].key
), 1);
642 * ext3_splice_branch - splice the allocated branch onto inode.
644 * @block: (logical) number of block we are adding
645 * @chain: chain of indirect blocks (with a missing link - see
647 * @where: location of missing link
648 * @num: number of blocks we are adding
650 * This function verifies that chain (up to the missing link) had not
651 * changed, fills the missing link and does all housekeeping needed in
652 * inode (->i_blocks, etc.). In case of success we end up with the full
653 * chain to new block and return 0. Otherwise (== chain had been changed)
654 * we free the new blocks (forgetting their buffer_heads, indeed) and
658 static int ext3_splice_branch(handle_t
*handle
, struct inode
*inode
, long block
,
659 Indirect chain
[4], Indirect
*where
, int num
)
663 struct ext3_inode_info
*ei
= EXT3_I(inode
);
666 * If we're splicing into a [td]indirect block (as opposed to the
667 * inode) then we need to get write access to the [td]indirect block
671 BUFFER_TRACE(where
->bh
, "get_write_access");
672 err
= ext3_journal_get_write_access(handle
, where
->bh
);
676 /* Verify that place we are splicing to is still there and vacant */
678 /* Writer: pointers, ->i_next_alloc* */
679 if (!verify_chain(chain
, where
-1) || *where
->p
)
685 *where
->p
= where
->key
;
686 ei
->i_next_alloc_block
= block
;
687 ei
->i_next_alloc_goal
= le32_to_cpu(where
[num
-1].key
);
690 /* We are done with atomic stuff, now do the rest of housekeeping */
692 inode
->i_ctime
= CURRENT_TIME
;
693 ext3_mark_inode_dirty(handle
, inode
);
695 /* had we spliced it onto indirect block? */
698 * akpm: If we spliced it onto an indirect block, we haven't
699 * altered the inode. Note however that if it is being spliced
700 * onto an indirect block at the very end of the file (the
701 * file is growing) then we *will* alter the inode to reflect
702 * the new i_size. But that is not done here - it is done in
703 * generic_commit_write->__mark_inode_dirty->ext3_dirty_inode.
705 jbd_debug(5, "splicing indirect only\n");
706 BUFFER_TRACE(where
->bh
, "call ext3_journal_dirty_metadata");
707 err
= ext3_journal_dirty_metadata(handle
, where
->bh
);
712 * OK, we spliced it into the inode itself on a direct block.
713 * Inode was dirtied above.
715 jbd_debug(5, "splicing direct\n");
721 * AKPM: if where[i].bh isn't part of the current updating
722 * transaction then we explode nastily. Test this code path.
724 jbd_debug(1, "the chain changed: try again\n");
728 for (i
= 1; i
< num
; i
++) {
729 BUFFER_TRACE(where
[i
].bh
, "call journal_forget");
730 ext3_journal_forget(handle
, where
[i
].bh
);
732 /* For the normal collision cleanup case, we free up the blocks.
733 * On genuine filesystem errors we don't even think about doing
736 for (i
= 0; i
< num
; i
++)
737 ext3_free_blocks(handle
, inode
,
738 le32_to_cpu(where
[i
].key
), 1);
743 * Allocation strategy is simple: if we have to allocate something, we will
744 * have to go the whole way to leaf. So let's do it before attaching anything
745 * to tree, set linkage between the newborn blocks, write them if sync is
746 * required, recheck the path, free and repeat if check fails, otherwise
747 * set the last missing link (that will protect us from any truncate-generated
748 * removals - all blocks on the path are immune now) and possibly force the
749 * write on the parent block.
750 * That has a nice additional property: no special recovery from the failed
751 * allocations is needed - we simply release blocks and do not touch anything
752 * reachable from inode.
754 * akpm: `handle' can be NULL if create == 0.
756 * The BKL may not be held on entry here. Be sure to take it early.
760 ext3_get_block_handle(handle_t
*handle
, struct inode
*inode
, sector_t iblock
,
761 struct buffer_head
*bh_result
, int create
, int extend_disksize
)
770 int depth
= ext3_block_to_path(inode
, iblock
, offsets
, &boundary
);
771 struct ext3_inode_info
*ei
= EXT3_I(inode
);
774 J_ASSERT(handle
!= NULL
|| create
== 0);
780 partial
= ext3_get_branch(inode
, depth
, offsets
, chain
, &err
);
782 /* Simplest case - block found, no allocation needed */
784 clear_buffer_new(bh_result
);
786 map_bh(bh_result
, inode
->i_sb
, le32_to_cpu(chain
[depth
-1].key
));
788 set_buffer_boundary(bh_result
);
789 /* Clean up and exit */
790 partial
= chain
+depth
-1; /* the whole chain */
794 /* Next simple case - plain lookup or failed read of indirect block */
795 if (!create
|| err
== -EIO
) {
797 while (partial
> chain
) {
798 BUFFER_TRACE(partial
->bh
, "call brelse");
802 BUFFER_TRACE(bh_result
, "returned");
808 * Indirect block might be removed by truncate while we were
809 * reading it. Handling of that case (forget what we've got and
810 * reread) is taken out of the main path.
815 if (ext3_find_goal(inode
, iblock
, chain
, partial
, &goal
) < 0)
818 left
= (chain
+ depth
) - partial
;
821 * Block out ext3_truncate while we alter the tree
823 down_read(&ei
->truncate_sem
);
824 err
= ext3_alloc_branch(handle
, inode
, left
, goal
,
825 offsets
+(partial
-chain
), partial
);
827 /* The ext3_splice_branch call will free and forget any buffers
828 * on the new chain if there is a failure, but that risks using
829 * up transaction credits, especially for bitmaps where the
830 * credits cannot be returned. Can we handle this somehow? We
831 * may need to return -EAGAIN upwards in the worst case. --sct */
833 err
= ext3_splice_branch(handle
, inode
, iblock
, chain
,
835 up_read(&ei
->truncate_sem
);
841 if (extend_disksize
) {
843 * This is not racy against ext3_truncate's modification of
844 * i_disksize because VM/VFS ensures that the file cannot be
845 * extended while truncate is in progress. It is racy between
846 * multiple parallel instances of get_block, but we have BKL.
848 new_size
= inode
->i_size
;
849 if (new_size
> ei
->i_disksize
)
850 ei
->i_disksize
= new_size
;
852 set_buffer_new(bh_result
);
856 while (partial
> chain
) {
857 jbd_debug(1, "buffer chain changed, retrying\n");
858 BUFFER_TRACE(partial
->bh
, "brelsing");
865 static int ext3_get_block(struct inode
*inode
, sector_t iblock
,
866 struct buffer_head
*bh_result
, int create
)
868 handle_t
*handle
= 0;
872 handle
= ext3_journal_current_handle();
873 J_ASSERT(handle
!= 0);
875 ret
= ext3_get_block_handle(handle
, inode
, iblock
,
876 bh_result
, create
, 1);
880 #define DIO_CREDITS (EXT3_RESERVE_TRANS_BLOCKS + 32)
883 ext3_direct_io_get_blocks(struct inode
*inode
, sector_t iblock
,
884 unsigned long max_blocks
, struct buffer_head
*bh_result
,
887 handle_t
*handle
= journal_current_handle();
890 if (handle
&& handle
->h_buffer_credits
<= EXT3_RESERVE_TRANS_BLOCKS
) {
892 * Getting low on buffer credits...
894 if (!ext3_journal_extend(handle
, DIO_CREDITS
)) {
896 * Couldn't extend the transaction. Start a new one
898 ret
= ext3_journal_restart(handle
, DIO_CREDITS
);
902 ret
= ext3_get_block_handle(handle
, inode
, iblock
,
903 bh_result
, create
, 0);
905 bh_result
->b_size
= (1 << inode
->i_blkbits
);
911 * `handle' can be NULL if create is zero
913 struct buffer_head
*ext3_getblk(handle_t
*handle
, struct inode
* inode
,
914 long block
, int create
, int * errp
)
916 struct buffer_head dummy
;
919 J_ASSERT(handle
!= NULL
|| create
== 0);
922 dummy
.b_blocknr
= -1000;
923 buffer_trace_init(&dummy
.b_history
);
924 *errp
= ext3_get_block_handle(handle
, inode
, block
, &dummy
, create
, 1);
925 if (!*errp
&& buffer_mapped(&dummy
)) {
926 struct buffer_head
*bh
;
927 bh
= sb_getblk(inode
->i_sb
, dummy
.b_blocknr
);
928 if (buffer_new(&dummy
)) {
929 J_ASSERT(create
!= 0);
930 J_ASSERT(handle
!= 0);
932 /* Now that we do not always journal data, we
933 should keep in mind whether this should
934 always journal the new buffer as metadata.
935 For now, regular file writes use
936 ext3_get_block instead, so it's not a
939 BUFFER_TRACE(bh
, "call get_create_access");
940 fatal
= ext3_journal_get_create_access(handle
, bh
);
942 memset(bh
->b_data
, 0,
943 inode
->i_sb
->s_blocksize
);
944 set_buffer_uptodate(bh
);
947 BUFFER_TRACE(bh
, "call ext3_journal_dirty_metadata");
948 err
= ext3_journal_dirty_metadata(handle
, bh
);
949 if (!fatal
) fatal
= err
;
951 BUFFER_TRACE(bh
, "not a new buffer");
963 struct buffer_head
*ext3_bread(handle_t
*handle
, struct inode
* inode
,
964 int block
, int create
, int *err
)
966 struct buffer_head
* bh
;
969 prev_blocks
= inode
->i_blocks
;
971 bh
= ext3_getblk (handle
, inode
, block
, create
, err
);
974 #ifdef EXT3_PREALLOCATE
976 * If the inode has grown, and this is a directory, then use a few
977 * more of the preallocated blocks to keep directory fragmentation
978 * down. The preallocated blocks are guaranteed to be contiguous.
981 S_ISDIR(inode
->i_mode
) &&
982 inode
->i_blocks
> prev_blocks
&&
983 EXT3_HAS_COMPAT_FEATURE(inode
->i_sb
,
984 EXT3_FEATURE_COMPAT_DIR_PREALLOC
)) {
986 struct buffer_head
*tmp_bh
;
989 EXT3_I(inode
)->i_prealloc_count
&&
990 i
< EXT3_SB(inode
->i_sb
)->s_es
->s_prealloc_dir_blocks
;
993 * ext3_getblk will zero out the contents of the
996 tmp_bh
= ext3_getblk(handle
, inode
,
997 block
+i
, create
, err
);
1006 if (buffer_uptodate(bh
))
1008 ll_rw_block (READ
, 1, &bh
);
1009 wait_on_buffer (bh
);
1010 if (buffer_uptodate(bh
))
1017 static int walk_page_buffers( handle_t
*handle
,
1018 struct buffer_head
*head
,
1022 int (*fn
)( handle_t
*handle
,
1023 struct buffer_head
*bh
))
1025 struct buffer_head
*bh
;
1026 unsigned block_start
, block_end
;
1027 unsigned blocksize
= head
->b_size
;
1029 struct buffer_head
*next
;
1031 for ( bh
= head
, block_start
= 0;
1032 ret
== 0 && (bh
!= head
|| !block_start
);
1033 block_start
= block_end
, bh
= next
)
1035 next
= bh
->b_this_page
;
1036 block_end
= block_start
+ blocksize
;
1037 if (block_end
<= from
|| block_start
>= to
) {
1038 if (partial
&& !buffer_uptodate(bh
))
1042 err
= (*fn
)(handle
, bh
);
1050 * To preserve ordering, it is essential that the hole instantiation and
1051 * the data write be encapsulated in a single transaction. We cannot
1052 * close off a transaction and start a new one between the ext3_get_block()
1053 * and the commit_write(). So doing the journal_start at the start of
1054 * prepare_write() is the right place.
1056 * Also, this function can nest inside ext3_writepage() ->
1057 * block_write_full_page(). In that case, we *know* that ext3_writepage()
1058 * has generated enough buffer credits to do the whole page. So we won't
1059 * block on the journal in that case, which is good, because the caller may
1062 * By accident, ext3 can be reentered when a transaction is open via
1063 * quota file writes. If we were to commit the transaction while thus
1064 * reentered, there can be a deadlock - we would be holding a quota
1065 * lock, and the commit would never complete if another thread had a
1066 * transaction open and was blocking on the quota lock - a ranking
1069 * So what we do is to rely on the fact that journal_stop/journal_start
1070 * will _not_ run commit under these circumstances because handle->h_ref
1071 * is elevated. We'll still have enough credits for the tiny quotafile
1075 static int do_journal_get_write_access(handle_t
*handle
,
1076 struct buffer_head
*bh
)
1078 if (!buffer_mapped(bh
) || buffer_freed(bh
))
1080 return ext3_journal_get_write_access(handle
, bh
);
1083 static int ext3_prepare_write(struct file
*file
, struct page
*page
,
1084 unsigned from
, unsigned to
)
1086 struct inode
*inode
= page
->mapping
->host
;
1087 int ret
, needed_blocks
= ext3_writepage_trans_blocks(inode
);
1090 handle
= ext3_journal_start(inode
, needed_blocks
);
1091 if (IS_ERR(handle
)) {
1092 ret
= PTR_ERR(handle
);
1095 ret
= block_prepare_write(page
, from
, to
, ext3_get_block
);
1097 goto prepare_write_failed
;
1099 if (ext3_should_journal_data(inode
)) {
1100 ret
= walk_page_buffers(handle
, page_buffers(page
),
1101 from
, to
, NULL
, do_journal_get_write_access
);
1103 prepare_write_failed
:
1105 ext3_journal_stop(handle
);
1111 ext3_journal_dirty_data(handle_t
*handle
, struct buffer_head
*bh
)
1113 int err
= journal_dirty_data(handle
, bh
);
1115 ext3_journal_abort_handle(__FUNCTION__
, __FUNCTION__
,
1120 /* For commit_write() in data=journal mode */
1121 static int commit_write_fn(handle_t
*handle
, struct buffer_head
*bh
)
1123 if (!buffer_mapped(bh
) || buffer_freed(bh
))
1125 set_buffer_uptodate(bh
);
1126 return ext3_journal_dirty_metadata(handle
, bh
);
1130 * We need to pick up the new inode size which generic_commit_write gave us
1131 * `file' can be NULL - eg, when called from page_symlink().
1133 * ext3 never places buffers on inode->i_mapping->private_list. metadata
1134 * buffers are managed internally.
1137 static int ext3_ordered_commit_write(struct file
*file
, struct page
*page
,
1138 unsigned from
, unsigned to
)
1140 handle_t
*handle
= ext3_journal_current_handle();
1141 struct inode
*inode
= page
->mapping
->host
;
1144 ret
= walk_page_buffers(handle
, page_buffers(page
),
1145 from
, to
, NULL
, ext3_journal_dirty_data
);
1149 * generic_commit_write() will run mark_inode_dirty() if i_size
1150 * changes. So let's piggyback the i_disksize mark_inode_dirty
1155 new_i_size
= ((loff_t
)page
->index
<< PAGE_CACHE_SHIFT
) + to
;
1156 if (new_i_size
> EXT3_I(inode
)->i_disksize
)
1157 EXT3_I(inode
)->i_disksize
= new_i_size
;
1158 ret
= generic_commit_write(file
, page
, from
, to
);
1160 ret2
= ext3_journal_stop(handle
);
1166 static int ext3_writeback_commit_write(struct file
*file
, struct page
*page
,
1167 unsigned from
, unsigned to
)
1169 handle_t
*handle
= ext3_journal_current_handle();
1170 struct inode
*inode
= page
->mapping
->host
;
1174 new_i_size
= ((loff_t
)page
->index
<< PAGE_CACHE_SHIFT
) + to
;
1175 if (new_i_size
> EXT3_I(inode
)->i_disksize
)
1176 EXT3_I(inode
)->i_disksize
= new_i_size
;
1177 ret
= generic_commit_write(file
, page
, from
, to
);
1178 ret2
= ext3_journal_stop(handle
);
1184 static int ext3_journalled_commit_write(struct file
*file
,
1185 struct page
*page
, unsigned from
, unsigned to
)
1187 handle_t
*handle
= ext3_journal_current_handle();
1188 struct inode
*inode
= page
->mapping
->host
;
1194 * Here we duplicate the generic_commit_write() functionality
1196 pos
= ((loff_t
)page
->index
<< PAGE_CACHE_SHIFT
) + to
;
1198 ret
= walk_page_buffers(handle
, page_buffers(page
), from
,
1199 to
, &partial
, commit_write_fn
);
1201 SetPageUptodate(page
);
1202 if (pos
> inode
->i_size
)
1203 i_size_write(inode
, pos
);
1204 EXT3_I(inode
)->i_state
|= EXT3_STATE_JDATA
;
1205 if (inode
->i_size
> EXT3_I(inode
)->i_disksize
) {
1206 EXT3_I(inode
)->i_disksize
= inode
->i_size
;
1207 ret2
= ext3_mark_inode_dirty(handle
, inode
);
1211 ret2
= ext3_journal_stop(handle
);
1218 * bmap() is special. It gets used by applications such as lilo and by
1219 * the swapper to find the on-disk block of a specific piece of data.
1221 * Naturally, this is dangerous if the block concerned is still in the
1222 * journal. If somebody makes a swapfile on an ext3 data-journaling
1223 * filesystem and enables swap, then they may get a nasty shock when the
1224 * data getting swapped to that swapfile suddenly gets overwritten by
1225 * the original zero's written out previously to the journal and
1226 * awaiting writeback in the kernel's buffer cache.
1228 * So, if we see any bmap calls here on a modified, data-journaled file,
1229 * take extra steps to flush any blocks which might be in the cache.
1231 static sector_t
ext3_bmap(struct address_space
*mapping
, sector_t block
)
1233 struct inode
*inode
= mapping
->host
;
1237 if (EXT3_I(inode
)->i_state
& EXT3_STATE_JDATA
) {
1239 * This is a REALLY heavyweight approach, but the use of
1240 * bmap on dirty files is expected to be extremely rare:
1241 * only if we run lilo or swapon on a freshly made file
1242 * do we expect this to happen.
1244 * (bmap requires CAP_SYS_RAWIO so this does not
1245 * represent an unprivileged user DOS attack --- we'd be
1246 * in trouble if mortal users could trigger this path at
1249 * NB. EXT3_STATE_JDATA is not set on files other than
1250 * regular files. If somebody wants to bmap a directory
1251 * or symlink and gets confused because the buffer
1252 * hasn't yet been flushed to disk, they deserve
1253 * everything they get.
1256 EXT3_I(inode
)->i_state
&= ~EXT3_STATE_JDATA
;
1257 journal
= EXT3_JOURNAL(inode
);
1258 journal_lock_updates(journal
);
1259 err
= journal_flush(journal
);
1260 journal_unlock_updates(journal
);
1266 return generic_block_bmap(mapping
,block
,ext3_get_block
);
1269 static int bget_one(handle_t
*handle
, struct buffer_head
*bh
)
1275 static int bput_one(handle_t
*handle
, struct buffer_head
*bh
)
1281 static int journal_dirty_data_fn(handle_t
*handle
, struct buffer_head
*bh
)
1283 if (buffer_mapped(bh
))
1284 return ext3_journal_dirty_data(handle
, bh
);
1289 * Note that we always start a transaction even if we're not journalling
1290 * data. This is to preserve ordering: any hole instantiation within
1291 * __block_write_full_page -> ext3_get_block() should be journalled
1292 * along with the data so we don't crash and then get metadata which
1293 * refers to old data.
1295 * In all journalling modes block_write_full_page() will start the I/O.
1299 * ext3_writepage() -> kmalloc() -> __alloc_pages() -> page_launder() ->
1304 * ext3_file_write() -> generic_file_write() -> __alloc_pages() -> ...
1306 * Same applies to ext3_get_block(). We will deadlock on various things like
1307 * lock_journal and i_truncate_sem.
1309 * Setting PF_MEMALLOC here doesn't work - too many internal memory
1312 * 16May01: If we're reentered then journal_current_handle() will be
1313 * non-zero. We simply *return*.
1315 * 1 July 2001: @@@ FIXME:
1316 * In journalled data mode, a data buffer may be metadata against the
1317 * current transaction. But the same file is part of a shared mapping
1318 * and someone does a writepage() on it.
1320 * We will move the buffer onto the async_data list, but *after* it has
1321 * been dirtied. So there's a small window where we have dirty data on
1324 * Note that this only applies to the last partial page in the file. The
1325 * bit which block_write_full_page() uses prepare/commit for. (That's
1326 * broken code anyway: it's wrong for msync()).
1328 * It's a rare case: affects the final partial page, for journalled data
1329 * where the file is subject to bith write() and writepage() in the same
1330 * transction. To fix it we'll need a custom block_write_full_page().
1331 * We'll probably need that anyway for journalling writepage() output.
1333 * We don't honour synchronous mounts for writepage(). That would be
1334 * disastrous. Any write() or metadata operation will sync the fs for
1337 * AKPM2: if all the page's buffers are mapped to disk and !data=journal,
1338 * we don't need to open a transaction here.
1340 static int ext3_ordered_writepage(struct page
*page
,
1341 struct writeback_control
*wbc
)
1343 struct inode
*inode
= page
->mapping
->host
;
1344 struct buffer_head
*page_bufs
;
1345 handle_t
*handle
= NULL
;
1349 J_ASSERT(PageLocked(page
));
1352 * We give up here if we're reentered, because it might be for a
1353 * different filesystem.
1355 if (ext3_journal_current_handle())
1358 handle
= ext3_journal_start(inode
, ext3_writepage_trans_blocks(inode
));
1360 if (IS_ERR(handle
)) {
1361 ret
= PTR_ERR(handle
);
1365 if (!page_has_buffers(page
)) {
1366 if (!PageUptodate(page
))
1368 create_empty_buffers(page
, inode
->i_sb
->s_blocksize
,
1369 (1 << BH_Dirty
)|(1 << BH_Uptodate
));
1371 page_bufs
= page_buffers(page
);
1372 walk_page_buffers(handle
, page_bufs
, 0,
1373 PAGE_CACHE_SIZE
, NULL
, bget_one
);
1375 ret
= block_write_full_page(page
, ext3_get_block
, wbc
);
1378 * The page can become unlocked at any point now, and
1379 * truncate can then come in and change things. So we
1380 * can't touch *page from now on. But *page_bufs is
1381 * safe due to elevated refcount.
1385 * And attach them to the current transaction. But only if
1386 * block_write_full_page() succeeded. Otherwise they are unmapped,
1387 * and generally junk.
1390 err
= walk_page_buffers(handle
, page_bufs
, 0, PAGE_CACHE_SIZE
,
1391 NULL
, journal_dirty_data_fn
);
1395 walk_page_buffers(handle
, page_bufs
, 0,
1396 PAGE_CACHE_SIZE
, NULL
, bput_one
);
1397 err
= ext3_journal_stop(handle
);
1403 __set_page_dirty_nobuffers(page
);
1408 static int ext3_writeback_writepage(struct page
*page
,
1409 struct writeback_control
*wbc
)
1411 struct inode
*inode
= page
->mapping
->host
;
1412 handle_t
*handle
= NULL
;
1416 if (ext3_journal_current_handle())
1419 handle
= ext3_journal_start(inode
, ext3_writepage_trans_blocks(inode
));
1420 if (IS_ERR(handle
)) {
1421 ret
= PTR_ERR(handle
);
1425 ret
= block_write_full_page(page
, ext3_get_block
, wbc
);
1426 err
= ext3_journal_stop(handle
);
1432 __set_page_dirty_nobuffers(page
);
1437 static int ext3_journalled_writepage(struct page
*page
,
1438 struct writeback_control
*wbc
)
1440 struct inode
*inode
= page
->mapping
->host
;
1441 handle_t
*handle
= NULL
;
1445 if (ext3_journal_current_handle())
1448 handle
= ext3_journal_start(inode
, ext3_writepage_trans_blocks(inode
));
1449 if (IS_ERR(handle
)) {
1450 ret
= PTR_ERR(handle
);
1454 if (!page_has_buffers(page
) || PageChecked(page
)) {
1456 * It's mmapped pagecache. Add buffers and journal it. There
1457 * doesn't seem much point in redirtying the page here.
1459 ClearPageChecked(page
);
1460 ret
= block_prepare_write(page
, 0, PAGE_CACHE_SIZE
,
1464 ret
= walk_page_buffers(handle
, page_buffers(page
), 0,
1465 PAGE_CACHE_SIZE
, NULL
, do_journal_get_write_access
);
1467 err
= walk_page_buffers(handle
, page_buffers(page
), 0,
1468 PAGE_CACHE_SIZE
, NULL
, commit_write_fn
);
1471 EXT3_I(inode
)->i_state
|= EXT3_STATE_JDATA
;
1475 * It may be a page full of checkpoint-mode buffers. We don't
1476 * really know unless we go poke around in the buffer_heads.
1477 * But block_write_full_page will do the right thing.
1479 ret
= block_write_full_page(page
, ext3_get_block
, wbc
);
1481 err
= ext3_journal_stop(handle
);
1488 __set_page_dirty_nobuffers(page
);
1494 static int ext3_readpage(struct file
*file
, struct page
*page
)
1496 return mpage_readpage(page
, ext3_get_block
);
1500 ext3_readpages(struct file
*file
, struct address_space
*mapping
,
1501 struct list_head
*pages
, unsigned nr_pages
)
1503 return mpage_readpages(mapping
, pages
, nr_pages
, ext3_get_block
);
1506 static int ext3_invalidatepage(struct page
*page
, unsigned long offset
)
1508 journal_t
*journal
= EXT3_JOURNAL(page
->mapping
->host
);
1511 * If it's a full truncate we just forget about the pending dirtying
1514 ClearPageChecked(page
);
1516 return journal_invalidatepage(journal
, page
, offset
);
1519 static int ext3_releasepage(struct page
*page
, int wait
)
1521 journal_t
*journal
= EXT3_JOURNAL(page
->mapping
->host
);
1523 WARN_ON(PageChecked(page
));
1524 return journal_try_to_free_buffers(journal
, page
, wait
);
1528 * If the O_DIRECT write will extend the file then add this inode to the
1529 * orphan list. So recovery will truncate it back to the original size
1530 * if the machine crashes during the write.
1532 * If the O_DIRECT write is intantiating holes inside i_size and the machine
1533 * crashes then stale disk data _may_ be exposed inside the file.
1535 static int ext3_direct_IO(int rw
, struct kiocb
*iocb
,
1536 const struct iovec
*iov
, loff_t offset
,
1537 unsigned long nr_segs
)
1539 struct file
*file
= iocb
->ki_filp
;
1540 struct inode
*inode
= file
->f_dentry
->d_inode
->i_mapping
->host
;
1541 struct ext3_inode_info
*ei
= EXT3_I(inode
);
1542 handle_t
*handle
= NULL
;
1545 size_t count
= iov_length(iov
, nr_segs
);
1548 loff_t final_size
= offset
+ count
;
1550 handle
= ext3_journal_start(inode
, DIO_CREDITS
);
1551 if (IS_ERR(handle
)) {
1552 ret
= PTR_ERR(handle
);
1555 if (final_size
> inode
->i_size
) {
1556 ret
= ext3_orphan_add(handle
, inode
);
1560 ei
->i_disksize
= inode
->i_size
;
1564 ret
= blockdev_direct_IO(rw
, iocb
, inode
, inode
->i_sb
->s_bdev
, iov
,
1565 offset
, nr_segs
, ext3_direct_io_get_blocks
);
1572 ext3_orphan_del(handle
, inode
);
1573 if (orphan
&& ret
> 0) {
1574 loff_t end
= offset
+ ret
;
1575 if (end
> inode
->i_size
) {
1576 ei
->i_disksize
= end
;
1577 i_size_write(inode
, end
);
1578 err
= ext3_mark_inode_dirty(handle
, inode
);
1583 err
= ext3_journal_stop(handle
);
1592 * Pages can be marked dirty completely asynchronously from ext3's journalling
1593 * activity. By filemap_sync_pte(), try_to_unmap_one(), etc. We cannot do
1594 * much here because ->set_page_dirty is called under VFS locks. The page is
1595 * not necessarily locked.
1597 * We cannot just dirty the page and leave attached buffers clean, because the
1598 * buffers' dirty state is "definitive". We cannot just set the buffers dirty
1599 * or jbddirty because all the journalling code will explode.
1601 * So what we do is to mark the page "pending dirty" and next time writepage
1602 * is called, propagate that into the buffers appropriately.
1604 static int ext3_journalled_set_page_dirty(struct page
*page
)
1606 SetPageChecked(page
);
1607 return __set_page_dirty_nobuffers(page
);
1610 static struct address_space_operations ext3_ordered_aops
= {
1611 .readpage
= ext3_readpage
,
1612 .readpages
= ext3_readpages
,
1613 .writepage
= ext3_ordered_writepage
,
1614 .sync_page
= block_sync_page
,
1615 .prepare_write
= ext3_prepare_write
,
1616 .commit_write
= ext3_ordered_commit_write
,
1618 .invalidatepage
= ext3_invalidatepage
,
1619 .releasepage
= ext3_releasepage
,
1620 .direct_IO
= ext3_direct_IO
,
1623 static struct address_space_operations ext3_writeback_aops
= {
1624 .readpage
= ext3_readpage
,
1625 .readpages
= ext3_readpages
,
1626 .writepage
= ext3_writeback_writepage
,
1627 .sync_page
= block_sync_page
,
1628 .prepare_write
= ext3_prepare_write
,
1629 .commit_write
= ext3_writeback_commit_write
,
1631 .invalidatepage
= ext3_invalidatepage
,
1632 .releasepage
= ext3_releasepage
,
1633 .direct_IO
= ext3_direct_IO
,
1636 static struct address_space_operations ext3_journalled_aops
= {
1637 .readpage
= ext3_readpage
,
1638 .readpages
= ext3_readpages
,
1639 .writepage
= ext3_journalled_writepage
,
1640 .sync_page
= block_sync_page
,
1641 .prepare_write
= ext3_prepare_write
,
1642 .commit_write
= ext3_journalled_commit_write
,
1643 .set_page_dirty
= ext3_journalled_set_page_dirty
,
1645 .invalidatepage
= ext3_invalidatepage
,
1646 .releasepage
= ext3_releasepage
,
1649 void ext3_set_aops(struct inode
*inode
)
1651 if (ext3_should_order_data(inode
))
1652 inode
->i_mapping
->a_ops
= &ext3_ordered_aops
;
1653 else if (ext3_should_writeback_data(inode
))
1654 inode
->i_mapping
->a_ops
= &ext3_writeback_aops
;
1656 inode
->i_mapping
->a_ops
= &ext3_journalled_aops
;
1660 * ext3_block_truncate_page() zeroes out a mapping from file offset `from'
1661 * up to the end of the block which corresponds to `from'.
1662 * This required during truncate. We need to physically zero the tail end
1663 * of that block so it doesn't yield old data if the file is later grown.
1665 static int ext3_block_truncate_page(handle_t
*handle
, struct page
*page
,
1666 struct address_space
*mapping
, loff_t from
)
1668 unsigned long index
= from
>> PAGE_CACHE_SHIFT
;
1669 unsigned offset
= from
& (PAGE_CACHE_SIZE
-1);
1670 unsigned blocksize
, iblock
, length
, pos
;
1671 struct inode
*inode
= mapping
->host
;
1672 struct buffer_head
*bh
;
1676 blocksize
= inode
->i_sb
->s_blocksize
;
1677 length
= blocksize
- (offset
& (blocksize
- 1));
1678 iblock
= index
<< (PAGE_CACHE_SHIFT
- inode
->i_sb
->s_blocksize_bits
);
1680 if (!page_has_buffers(page
))
1681 create_empty_buffers(page
, blocksize
, 0);
1683 /* Find the buffer that contains "offset" */
1684 bh
= page_buffers(page
);
1686 while (offset
>= pos
) {
1687 bh
= bh
->b_this_page
;
1693 if (buffer_freed(bh
)) {
1694 BUFFER_TRACE(bh
, "freed: skip");
1698 if (!buffer_mapped(bh
)) {
1699 BUFFER_TRACE(bh
, "unmapped");
1700 ext3_get_block(inode
, iblock
, bh
, 0);
1701 /* unmapped? It's a hole - nothing to do */
1702 if (!buffer_mapped(bh
)) {
1703 BUFFER_TRACE(bh
, "still unmapped");
1708 /* Ok, it's mapped. Make sure it's up-to-date */
1709 if (PageUptodate(page
))
1710 set_buffer_uptodate(bh
);
1712 if (!buffer_uptodate(bh
)) {
1714 ll_rw_block(READ
, 1, &bh
);
1716 /* Uhhuh. Read error. Complain and punt. */
1717 if (!buffer_uptodate(bh
))
1721 if (ext3_should_journal_data(inode
)) {
1722 BUFFER_TRACE(bh
, "get write access");
1723 err
= ext3_journal_get_write_access(handle
, bh
);
1728 kaddr
= kmap_atomic(page
, KM_USER0
);
1729 memset(kaddr
+ offset
, 0, length
);
1730 flush_dcache_page(page
);
1731 kunmap_atomic(kaddr
, KM_USER0
);
1733 BUFFER_TRACE(bh
, "zeroed end of block");
1736 if (ext3_should_journal_data(inode
)) {
1737 err
= ext3_journal_dirty_metadata(handle
, bh
);
1739 if (ext3_should_order_data(inode
))
1740 err
= ext3_journal_dirty_data(handle
, bh
);
1741 mark_buffer_dirty(bh
);
1746 page_cache_release(page
);
1751 * Probably it should be a library function... search for first non-zero word
1752 * or memcmp with zero_page, whatever is better for particular architecture.
1755 static inline int all_zeroes(u32
*p
, u32
*q
)
1764 * ext3_find_shared - find the indirect blocks for partial truncation.
1765 * @inode: inode in question
1766 * @depth: depth of the affected branch
1767 * @offsets: offsets of pointers in that branch (see ext3_block_to_path)
1768 * @chain: place to store the pointers to partial indirect blocks
1769 * @top: place to the (detached) top of branch
1771 * This is a helper function used by ext3_truncate().
1773 * When we do truncate() we may have to clean the ends of several
1774 * indirect blocks but leave the blocks themselves alive. Block is
1775 * partially truncated if some data below the new i_size is refered
1776 * from it (and it is on the path to the first completely truncated
1777 * data block, indeed). We have to free the top of that path along
1778 * with everything to the right of the path. Since no allocation
1779 * past the truncation point is possible until ext3_truncate()
1780 * finishes, we may safely do the latter, but top of branch may
1781 * require special attention - pageout below the truncation point
1782 * might try to populate it.
1784 * We atomically detach the top of branch from the tree, store the
1785 * block number of its root in *@top, pointers to buffer_heads of
1786 * partially truncated blocks - in @chain[].bh and pointers to
1787 * their last elements that should not be removed - in
1788 * @chain[].p. Return value is the pointer to last filled element
1791 * The work left to caller to do the actual freeing of subtrees:
1792 * a) free the subtree starting from *@top
1793 * b) free the subtrees whose roots are stored in
1794 * (@chain[i].p+1 .. end of @chain[i].bh->b_data)
1795 * c) free the subtrees growing from the inode past the @chain[0].
1796 * (no partially truncated stuff there). */
1798 static Indirect
*ext3_find_shared(struct inode
*inode
,
1804 Indirect
*partial
, *p
;
1808 /* Make k index the deepest non-null offest + 1 */
1809 for (k
= depth
; k
> 1 && !offsets
[k
-1]; k
--)
1811 partial
= ext3_get_branch(inode
, k
, offsets
, chain
, &err
);
1812 /* Writer: pointers */
1814 partial
= chain
+ k
-1;
1816 * If the branch acquired continuation since we've looked at it -
1817 * fine, it should all survive and (new) top doesn't belong to us.
1819 if (!partial
->key
&& *partial
->p
)
1822 for (p
=partial
; p
>chain
&& all_zeroes((u32
*)p
->bh
->b_data
,p
->p
); p
--)
1825 * OK, we've found the last block that must survive. The rest of our
1826 * branch should be detached before unlocking. However, if that rest
1827 * of branch is all ours and does not grow immediately from the inode
1828 * it's easier to cheat and just decrement partial->p.
1830 if (p
== chain
+ k
- 1 && p
> chain
) {
1834 /* Nope, don't do this in ext3. Must leave the tree intact */
1843 brelse(partial
->bh
);
1851 * Zero a number of block pointers in either an inode or an indirect block.
1852 * If we restart the transaction we must again get write access to the
1853 * indirect block for further modification.
1855 * We release `count' blocks on disk, but (last - first) may be greater
1856 * than `count' because there can be holes in there.
1859 ext3_clear_blocks(handle_t
*handle
, struct inode
*inode
, struct buffer_head
*bh
,
1860 unsigned long block_to_free
, unsigned long count
,
1861 u32
*first
, u32
*last
)
1864 if (try_to_extend_transaction(handle
, inode
)) {
1866 BUFFER_TRACE(bh
, "call ext3_journal_dirty_metadata");
1867 ext3_journal_dirty_metadata(handle
, bh
);
1869 ext3_mark_inode_dirty(handle
, inode
);
1870 ext3_journal_test_restart(handle
, inode
);
1872 BUFFER_TRACE(bh
, "retaking write access");
1873 ext3_journal_get_write_access(handle
, bh
);
1878 * Any buffers which are on the journal will be in memory. We find
1879 * them on the hash table so journal_revoke() will run journal_forget()
1880 * on them. We've already detached each block from the file, so
1881 * bforget() in journal_forget() should be safe.
1883 * AKPM: turn on bforget in journal_forget()!!!
1885 for (p
= first
; p
< last
; p
++) {
1886 u32 nr
= le32_to_cpu(*p
);
1888 struct buffer_head
*bh
;
1891 bh
= sb_find_get_block(inode
->i_sb
, nr
);
1892 ext3_forget(handle
, 0, inode
, bh
, nr
);
1896 ext3_free_blocks(handle
, inode
, block_to_free
, count
);
1900 * ext3_free_data - free a list of data blocks
1901 * @handle: handle for this transaction
1902 * @inode: inode we are dealing with
1903 * @this_bh: indirect buffer_head which contains *@first and *@last
1904 * @first: array of block numbers
1905 * @last: points immediately past the end of array
1907 * We are freeing all blocks refered from that array (numbers are stored as
1908 * little-endian 32-bit) and updating @inode->i_blocks appropriately.
1910 * We accumulate contiguous runs of blocks to free. Conveniently, if these
1911 * blocks are contiguous then releasing them at one time will only affect one
1912 * or two bitmap blocks (+ group descriptor(s) and superblock) and we won't
1913 * actually use a lot of journal space.
1915 * @this_bh will be %NULL if @first and @last point into the inode's direct
1918 static void ext3_free_data(handle_t
*handle
, struct inode
*inode
,
1919 struct buffer_head
*this_bh
, u32
*first
, u32
*last
)
1921 unsigned long block_to_free
= 0; /* Starting block # of a run */
1922 unsigned long count
= 0; /* Number of blocks in the run */
1923 u32
*block_to_free_p
= NULL
; /* Pointer into inode/ind
1926 unsigned long nr
; /* Current block # */
1927 u32
*p
; /* Pointer into inode/ind
1928 for current block */
1931 if (this_bh
) { /* For indirect block */
1932 BUFFER_TRACE(this_bh
, "get_write_access");
1933 err
= ext3_journal_get_write_access(handle
, this_bh
);
1934 /* Important: if we can't update the indirect pointers
1935 * to the blocks, we can't free them. */
1940 for (p
= first
; p
< last
; p
++) {
1941 nr
= le32_to_cpu(*p
);
1943 /* accumulate blocks to free if they're contiguous */
1946 block_to_free_p
= p
;
1948 } else if (nr
== block_to_free
+ count
) {
1951 ext3_clear_blocks(handle
, inode
, this_bh
,
1953 count
, block_to_free_p
, p
);
1955 block_to_free_p
= p
;
1962 ext3_clear_blocks(handle
, inode
, this_bh
, block_to_free
,
1963 count
, block_to_free_p
, p
);
1966 BUFFER_TRACE(this_bh
, "call ext3_journal_dirty_metadata");
1967 ext3_journal_dirty_metadata(handle
, this_bh
);
1972 * ext3_free_branches - free an array of branches
1973 * @handle: JBD handle for this transaction
1974 * @inode: inode we are dealing with
1975 * @parent_bh: the buffer_head which contains *@first and *@last
1976 * @first: array of block numbers
1977 * @last: pointer immediately past the end of array
1978 * @depth: depth of the branches to free
1980 * We are freeing all blocks refered from these branches (numbers are
1981 * stored as little-endian 32-bit) and updating @inode->i_blocks
1984 static void ext3_free_branches(handle_t
*handle
, struct inode
*inode
,
1985 struct buffer_head
*parent_bh
,
1986 u32
*first
, u32
*last
, int depth
)
1991 if (is_handle_aborted(handle
))
1995 struct buffer_head
*bh
;
1996 int addr_per_block
= EXT3_ADDR_PER_BLOCK(inode
->i_sb
);
1998 while (--p
>= first
) {
1999 nr
= le32_to_cpu(*p
);
2001 continue; /* A hole */
2003 /* Go read the buffer for the next level down */
2004 bh
= sb_bread(inode
->i_sb
, nr
);
2007 * A read failure? Report error and clear slot
2011 ext3_error(inode
->i_sb
, "ext3_free_branches",
2012 "Read failure, inode=%ld, block=%ld",
2017 /* This zaps the entire block. Bottom up. */
2018 BUFFER_TRACE(bh
, "free child branches");
2019 ext3_free_branches(handle
, inode
, bh
, (u32
*)bh
->b_data
,
2020 (u32
*)bh
->b_data
+ addr_per_block
,
2024 * We've probably journalled the indirect block several
2025 * times during the truncate. But it's no longer
2026 * needed and we now drop it from the transaction via
2029 * That's easy if it's exclusively part of this
2030 * transaction. But if it's part of the committing
2031 * transaction then journal_forget() will simply
2032 * brelse() it. That means that if the underlying
2033 * block is reallocated in ext3_get_block(),
2034 * unmap_underlying_metadata() will find this block
2035 * and will try to get rid of it. damn, damn.
2037 * If this block has already been committed to the
2038 * journal, a revoke record will be written. And
2039 * revoke records must be emitted *before* clearing
2040 * this block's bit in the bitmaps.
2042 ext3_forget(handle
, 1, inode
, bh
, bh
->b_blocknr
);
2045 * Everything below this this pointer has been
2046 * released. Now let this top-of-subtree go.
2048 * We want the freeing of this indirect block to be
2049 * atomic in the journal with the updating of the
2050 * bitmap block which owns it. So make some room in
2053 * We zero the parent pointer *after* freeing its
2054 * pointee in the bitmaps, so if extend_transaction()
2055 * for some reason fails to put the bitmap changes and
2056 * the release into the same transaction, recovery
2057 * will merely complain about releasing a free block,
2058 * rather than leaking blocks.
2060 if (is_handle_aborted(handle
))
2062 if (try_to_extend_transaction(handle
, inode
)) {
2063 ext3_mark_inode_dirty(handle
, inode
);
2064 ext3_journal_test_restart(handle
, inode
);
2067 ext3_free_blocks(handle
, inode
, nr
, 1);
2071 * The block which we have just freed is
2072 * pointed to by an indirect block: journal it
2074 BUFFER_TRACE(parent_bh
, "get_write_access");
2075 if (!ext3_journal_get_write_access(handle
,
2078 BUFFER_TRACE(parent_bh
,
2079 "call ext3_journal_dirty_metadata");
2080 ext3_journal_dirty_metadata(handle
,
2086 /* We have reached the bottom of the tree. */
2087 BUFFER_TRACE(parent_bh
, "free data blocks");
2088 ext3_free_data(handle
, inode
, parent_bh
, first
, last
);
2095 * We block out ext3_get_block() block instantiations across the entire
2096 * transaction, and VFS/VM ensures that ext3_truncate() cannot run
2097 * simultaneously on behalf of the same inode.
2099 * As we work through the truncate and commmit bits of it to the journal there
2100 * is one core, guiding principle: the file's tree must always be consistent on
2101 * disk. We must be able to restart the truncate after a crash.
2103 * The file's tree may be transiently inconsistent in memory (although it
2104 * probably isn't), but whenever we close off and commit a journal transaction,
2105 * the contents of (the filesystem + the journal) must be consistent and
2106 * restartable. It's pretty simple, really: bottom up, right to left (although
2107 * left-to-right works OK too).
2109 * Note that at recovery time, journal replay occurs *before* the restart of
2110 * truncate against the orphan inode list.
2112 * The committed inode has the new, desired i_size (which is the same as
2113 * i_disksize in this case). After a crash, ext3_orphan_cleanup() will see
2114 * that this inode's truncate did not complete and it will again call
2115 * ext3_truncate() to have another go. So there will be instantiated blocks
2116 * to the right of the truncation point in a crashed ext3 filesystem. But
2117 * that's fine - as long as they are linked from the inode, the post-crash
2118 * ext3_truncate() run will find them and release them.
2121 void ext3_truncate(struct inode
* inode
)
2124 struct ext3_inode_info
*ei
= EXT3_I(inode
);
2125 u32
*i_data
= ei
->i_data
;
2126 int addr_per_block
= EXT3_ADDR_PER_BLOCK(inode
->i_sb
);
2127 struct address_space
*mapping
= inode
->i_mapping
;
2134 unsigned blocksize
= inode
->i_sb
->s_blocksize
;
2137 if (!(S_ISREG(inode
->i_mode
) || S_ISDIR(inode
->i_mode
) ||
2138 S_ISLNK(inode
->i_mode
)))
2140 if (ext3_inode_is_fast_symlink(inode
))
2142 if (IS_APPEND(inode
) || IS_IMMUTABLE(inode
))
2145 ext3_discard_prealloc(inode
);
2148 * We have to lock the EOF page here, because lock_page() nests
2149 * outside journal_start().
2151 if ((inode
->i_size
& (blocksize
- 1)) == 0) {
2152 /* Block boundary? Nothing to do */
2155 page
= grab_cache_page(mapping
,
2156 inode
->i_size
>> PAGE_CACHE_SHIFT
);
2161 handle
= start_transaction(inode
);
2162 if (IS_ERR(handle
)) {
2164 clear_highpage(page
);
2165 flush_dcache_page(page
);
2167 page_cache_release(page
);
2169 return; /* AKPM: return what? */
2172 last_block
= (inode
->i_size
+ blocksize
-1)
2173 >> EXT3_BLOCK_SIZE_BITS(inode
->i_sb
);
2176 ext3_block_truncate_page(handle
, page
, mapping
, inode
->i_size
);
2178 n
= ext3_block_to_path(inode
, last_block
, offsets
, NULL
);
2180 goto out_stop
; /* error */
2183 * OK. This truncate is going to happen. We add the inode to the
2184 * orphan list, so that if this truncate spans multiple transactions,
2185 * and we crash, we will resume the truncate when the filesystem
2186 * recovers. It also marks the inode dirty, to catch the new size.
2188 * Implication: the file must always be in a sane, consistent
2189 * truncatable state while each transaction commits.
2191 if (ext3_orphan_add(handle
, inode
))
2195 * The orphan list entry will now protect us from any crash which
2196 * occurs before the truncate completes, so it is now safe to propagate
2197 * the new, shorter inode size (held for now in i_size) into the
2198 * on-disk inode. We do this via i_disksize, which is the value which
2199 * ext3 *really* writes onto the disk inode.
2201 ei
->i_disksize
= inode
->i_size
;
2204 * From here we block out all ext3_get_block() callers who want to
2205 * modify the block allocation tree.
2207 down_write(&ei
->truncate_sem
);
2209 if (n
== 1) { /* direct blocks */
2210 ext3_free_data(handle
, inode
, NULL
, i_data
+offsets
[0],
2211 i_data
+ EXT3_NDIR_BLOCKS
);
2215 partial
= ext3_find_shared(inode
, n
, offsets
, chain
, &nr
);
2216 /* Kill the top of shared branch (not detached) */
2218 if (partial
== chain
) {
2219 /* Shared branch grows from the inode */
2220 ext3_free_branches(handle
, inode
, NULL
,
2221 &nr
, &nr
+1, (chain
+n
-1) - partial
);
2224 * We mark the inode dirty prior to restart,
2225 * and prior to stop. No need for it here.
2228 /* Shared branch grows from an indirect block */
2229 BUFFER_TRACE(partial
->bh
, "get_write_access");
2230 ext3_free_branches(handle
, inode
, partial
->bh
,
2232 partial
->p
+1, (chain
+n
-1) - partial
);
2235 /* Clear the ends of indirect blocks on the shared branch */
2236 while (partial
> chain
) {
2237 ext3_free_branches(handle
, inode
, partial
->bh
, partial
->p
+ 1,
2238 (u32
*)partial
->bh
->b_data
+ addr_per_block
,
2239 (chain
+n
-1) - partial
);
2240 BUFFER_TRACE(partial
->bh
, "call brelse");
2241 brelse (partial
->bh
);
2245 /* Kill the remaining (whole) subtrees */
2246 switch (offsets
[0]) {
2248 nr
= i_data
[EXT3_IND_BLOCK
];
2250 ext3_free_branches(handle
, inode
, NULL
,
2252 i_data
[EXT3_IND_BLOCK
] = 0;
2254 case EXT3_IND_BLOCK
:
2255 nr
= i_data
[EXT3_DIND_BLOCK
];
2257 ext3_free_branches(handle
, inode
, NULL
,
2259 i_data
[EXT3_DIND_BLOCK
] = 0;
2261 case EXT3_DIND_BLOCK
:
2262 nr
= i_data
[EXT3_TIND_BLOCK
];
2264 ext3_free_branches(handle
, inode
, NULL
,
2266 i_data
[EXT3_TIND_BLOCK
] = 0;
2268 case EXT3_TIND_BLOCK
:
2271 up_write(&ei
->truncate_sem
);
2272 inode
->i_mtime
= inode
->i_ctime
= CURRENT_TIME
;
2273 ext3_mark_inode_dirty(handle
, inode
);
2275 /* In a multi-transaction truncate, we only make the final
2276 * transaction synchronous */
2281 * If this was a simple ftruncate(), and the file will remain alive
2282 * then we need to clear up the orphan record which we created above.
2283 * However, if this was a real unlink then we were called by
2284 * ext3_delete_inode(), and we allow that function to clean up the
2285 * orphan info for us.
2288 ext3_orphan_del(handle
, inode
);
2290 ext3_journal_stop(handle
);
2293 static unsigned long ext3_get_inode_block(struct super_block
*sb
,
2294 unsigned long ino
, struct ext3_iloc
*iloc
)
2296 unsigned long desc
, group_desc
, block_group
;
2297 unsigned long offset
, block
;
2298 struct buffer_head
*bh
;
2299 struct ext3_group_desc
* gdp
;
2301 if ((ino
!= EXT3_ROOT_INO
&&
2302 ino
!= EXT3_JOURNAL_INO
&&
2303 ino
< EXT3_FIRST_INO(sb
)) ||
2305 EXT3_SB(sb
)->s_es
->s_inodes_count
)) {
2306 ext3_error (sb
, "ext3_get_inode_block",
2307 "bad inode number: %lu", ino
);
2310 block_group
= (ino
- 1) / EXT3_INODES_PER_GROUP(sb
);
2311 if (block_group
>= EXT3_SB(sb
)->s_groups_count
) {
2312 ext3_error (sb
, "ext3_get_inode_block",
2313 "group >= groups count");
2316 group_desc
= block_group
>> EXT3_DESC_PER_BLOCK_BITS(sb
);
2317 desc
= block_group
& (EXT3_DESC_PER_BLOCK(sb
) - 1);
2318 bh
= EXT3_SB(sb
)->s_group_desc
[group_desc
];
2320 ext3_error (sb
, "ext3_get_inode_block",
2321 "Descriptor not loaded");
2325 gdp
= (struct ext3_group_desc
*) bh
->b_data
;
2327 * Figure out the offset within the block group inode table
2329 offset
= ((ino
- 1) % EXT3_INODES_PER_GROUP(sb
)) *
2330 EXT3_INODE_SIZE(sb
);
2331 block
= le32_to_cpu(gdp
[desc
].bg_inode_table
) +
2332 (offset
>> EXT3_BLOCK_SIZE_BITS(sb
));
2334 iloc
->block_group
= block_group
;
2335 iloc
->offset
= offset
& (EXT3_BLOCK_SIZE(sb
) - 1);
2340 * ext3_get_inode_loc returns with an extra refcount against the
2341 * inode's underlying buffer_head on success.
2344 int ext3_get_inode_loc (struct inode
*inode
, struct ext3_iloc
*iloc
)
2346 unsigned long block
;
2348 block
= ext3_get_inode_block(inode
->i_sb
, inode
->i_ino
, iloc
);
2350 struct buffer_head
*bh
= sb_bread(inode
->i_sb
, block
);
2355 ext3_error (inode
->i_sb
, "ext3_get_inode_loc",
2356 "unable to read inode block - "
2357 "inode=%lu, block=%lu", inode
->i_ino
, block
);
2362 void ext3_set_inode_flags(struct inode
*inode
)
2364 unsigned int flags
= EXT3_I(inode
)->i_flags
;
2366 inode
->i_flags
&= ~(S_SYNC
|S_APPEND
|S_IMMUTABLE
|S_NOATIME
|S_DIRSYNC
);
2367 if (flags
& EXT3_SYNC_FL
)
2368 inode
->i_flags
|= S_SYNC
;
2369 if (flags
& EXT3_APPEND_FL
)
2370 inode
->i_flags
|= S_APPEND
;
2371 if (flags
& EXT3_IMMUTABLE_FL
)
2372 inode
->i_flags
|= S_IMMUTABLE
;
2373 if (flags
& EXT3_NOATIME_FL
)
2374 inode
->i_flags
|= S_NOATIME
;
2375 if (flags
& EXT3_DIRSYNC_FL
)
2376 inode
->i_flags
|= S_DIRSYNC
;
2380 void ext3_read_inode(struct inode
* inode
)
2382 struct ext3_iloc iloc
;
2383 struct ext3_inode
*raw_inode
;
2384 struct ext3_inode_info
*ei
= EXT3_I(inode
);
2385 struct buffer_head
*bh
;
2388 #ifdef CONFIG_EXT3_FS_POSIX_ACL
2389 ei
->i_acl
= EXT3_ACL_NOT_CACHED
;
2390 ei
->i_default_acl
= EXT3_ACL_NOT_CACHED
;
2392 if (ext3_get_inode_loc(inode
, &iloc
))
2395 raw_inode
= ext3_raw_inode(&iloc
);
2396 inode
->i_mode
= le16_to_cpu(raw_inode
->i_mode
);
2397 inode
->i_uid
= (uid_t
)le16_to_cpu(raw_inode
->i_uid_low
);
2398 inode
->i_gid
= (gid_t
)le16_to_cpu(raw_inode
->i_gid_low
);
2399 if(!(test_opt (inode
->i_sb
, NO_UID32
))) {
2400 inode
->i_uid
|= le16_to_cpu(raw_inode
->i_uid_high
) << 16;
2401 inode
->i_gid
|= le16_to_cpu(raw_inode
->i_gid_high
) << 16;
2403 inode
->i_nlink
= le16_to_cpu(raw_inode
->i_links_count
);
2404 inode
->i_size
= le32_to_cpu(raw_inode
->i_size
);
2405 inode
->i_atime
.tv_sec
= le32_to_cpu(raw_inode
->i_atime
);
2406 inode
->i_ctime
.tv_sec
= le32_to_cpu(raw_inode
->i_ctime
);
2407 inode
->i_mtime
.tv_sec
= le32_to_cpu(raw_inode
->i_mtime
);
2408 inode
->i_atime
.tv_nsec
= inode
->i_ctime
.tv_nsec
= inode
->i_mtime
.tv_nsec
= 0;
2411 ei
->i_next_alloc_block
= 0;
2412 ei
->i_next_alloc_goal
= 0;
2413 ei
->i_dir_start_lookup
= 0;
2414 ei
->i_dtime
= le32_to_cpu(raw_inode
->i_dtime
);
2415 /* We now have enough fields to check if the inode was active or not.
2416 * This is needed because nfsd might try to access dead inodes
2417 * the test is that same one that e2fsck uses
2418 * NeilBrown 1999oct15
2420 if (inode
->i_nlink
== 0) {
2421 if (inode
->i_mode
== 0 ||
2422 !(EXT3_SB(inode
->i_sb
)->s_mount_state
& EXT3_ORPHAN_FS
)) {
2423 /* this inode is deleted */
2427 /* The only unlinked inodes we let through here have
2428 * valid i_mode and are being read by the orphan
2429 * recovery code: that's fine, we're about to complete
2430 * the process of deleting those. */
2432 inode
->i_blksize
= PAGE_SIZE
; /* This is the optimal IO size
2433 * (for stat), not the fs block
2435 inode
->i_blocks
= le32_to_cpu(raw_inode
->i_blocks
);
2436 ei
->i_flags
= le32_to_cpu(raw_inode
->i_flags
);
2437 #ifdef EXT3_FRAGMENTS
2438 ei
->i_faddr
= le32_to_cpu(raw_inode
->i_faddr
);
2439 ei
->i_frag_no
= raw_inode
->i_frag
;
2440 ei
->i_frag_size
= raw_inode
->i_fsize
;
2442 ei
->i_file_acl
= le32_to_cpu(raw_inode
->i_file_acl
);
2443 if (!S_ISREG(inode
->i_mode
)) {
2444 ei
->i_dir_acl
= le32_to_cpu(raw_inode
->i_dir_acl
);
2447 ((__u64
)le32_to_cpu(raw_inode
->i_size_high
)) << 32;
2449 ei
->i_disksize
= inode
->i_size
;
2450 inode
->i_generation
= le32_to_cpu(raw_inode
->i_generation
);
2451 #ifdef EXT3_PREALLOCATE
2452 ei
->i_prealloc_count
= 0;
2454 ei
->i_block_group
= iloc
.block_group
;
2457 * NOTE! The in-memory inode i_data array is in little-endian order
2458 * even on big-endian machines: we do NOT byteswap the block numbers!
2460 for (block
= 0; block
< EXT3_N_BLOCKS
; block
++)
2461 ei
->i_data
[block
] = raw_inode
->i_block
[block
];
2462 INIT_LIST_HEAD(&ei
->i_orphan
);
2464 if (S_ISREG(inode
->i_mode
)) {
2465 inode
->i_op
= &ext3_file_inode_operations
;
2466 inode
->i_fop
= &ext3_file_operations
;
2467 ext3_set_aops(inode
);
2468 } else if (S_ISDIR(inode
->i_mode
)) {
2469 inode
->i_op
= &ext3_dir_inode_operations
;
2470 inode
->i_fop
= &ext3_dir_operations
;
2471 } else if (S_ISLNK(inode
->i_mode
)) {
2472 if (ext3_inode_is_fast_symlink(inode
))
2473 inode
->i_op
= &ext3_fast_symlink_inode_operations
;
2475 inode
->i_op
= &ext3_symlink_inode_operations
;
2476 ext3_set_aops(inode
);
2479 inode
->i_op
= &ext3_special_inode_operations
;
2480 init_special_inode(inode
, inode
->i_mode
,
2481 le32_to_cpu(raw_inode
->i_block
[0]));
2484 ext3_set_inode_flags(inode
);
2488 make_bad_inode(inode
);
2493 * Post the struct inode info into an on-disk inode location in the
2494 * buffer-cache. This gobbles the caller's reference to the
2495 * buffer_head in the inode location struct.
2497 * The caller must have write access to iloc->bh.
2499 static int ext3_do_update_inode(handle_t
*handle
,
2500 struct inode
*inode
,
2501 struct ext3_iloc
*iloc
)
2503 struct ext3_inode
*raw_inode
= ext3_raw_inode(iloc
);
2504 struct ext3_inode_info
*ei
= EXT3_I(inode
);
2505 struct buffer_head
*bh
= iloc
->bh
;
2506 int err
= 0, rc
, block
;
2508 /* For fields not not tracking in the in-memory inode,
2509 * initialise them to zero for new inodes. */
2510 if (ei
->i_state
& EXT3_STATE_NEW
)
2511 memset(raw_inode
, 0, EXT3_SB(inode
->i_sb
)->s_inode_size
);
2513 raw_inode
->i_mode
= cpu_to_le16(inode
->i_mode
);
2514 if(!(test_opt(inode
->i_sb
, NO_UID32
))) {
2515 raw_inode
->i_uid_low
= cpu_to_le16(low_16_bits(inode
->i_uid
));
2516 raw_inode
->i_gid_low
= cpu_to_le16(low_16_bits(inode
->i_gid
));
2518 * Fix up interoperability with old kernels. Otherwise, old inodes get
2519 * re-used with the upper 16 bits of the uid/gid intact
2522 raw_inode
->i_uid_high
=
2523 cpu_to_le16(high_16_bits(inode
->i_uid
));
2524 raw_inode
->i_gid_high
=
2525 cpu_to_le16(high_16_bits(inode
->i_gid
));
2527 raw_inode
->i_uid_high
= 0;
2528 raw_inode
->i_gid_high
= 0;
2531 raw_inode
->i_uid_low
=
2532 cpu_to_le16(fs_high2lowuid(inode
->i_uid
));
2533 raw_inode
->i_gid_low
=
2534 cpu_to_le16(fs_high2lowgid(inode
->i_gid
));
2535 raw_inode
->i_uid_high
= 0;
2536 raw_inode
->i_gid_high
= 0;
2538 raw_inode
->i_links_count
= cpu_to_le16(inode
->i_nlink
);
2539 raw_inode
->i_size
= cpu_to_le32(ei
->i_disksize
);
2540 raw_inode
->i_atime
= cpu_to_le32(inode
->i_atime
.tv_sec
);
2541 raw_inode
->i_ctime
= cpu_to_le32(inode
->i_ctime
.tv_sec
);
2542 raw_inode
->i_mtime
= cpu_to_le32(inode
->i_mtime
.tv_sec
);
2543 raw_inode
->i_blocks
= cpu_to_le32(inode
->i_blocks
);
2544 raw_inode
->i_dtime
= cpu_to_le32(ei
->i_dtime
);
2545 raw_inode
->i_flags
= cpu_to_le32(ei
->i_flags
);
2546 #ifdef EXT3_FRAGMENTS
2547 raw_inode
->i_faddr
= cpu_to_le32(ei
->i_faddr
);
2548 raw_inode
->i_frag
= ei
->i_frag_no
;
2549 raw_inode
->i_fsize
= ei
->i_frag_size
;
2551 raw_inode
->i_file_acl
= cpu_to_le32(ei
->i_file_acl
);
2552 if (!S_ISREG(inode
->i_mode
)) {
2553 raw_inode
->i_dir_acl
= cpu_to_le32(ei
->i_dir_acl
);
2555 raw_inode
->i_size_high
=
2556 cpu_to_le32(ei
->i_disksize
>> 32);
2557 if (ei
->i_disksize
> 0x7fffffffULL
) {
2558 struct super_block
*sb
= inode
->i_sb
;
2559 if (!EXT3_HAS_RO_COMPAT_FEATURE(sb
,
2560 EXT3_FEATURE_RO_COMPAT_LARGE_FILE
) ||
2561 EXT3_SB(sb
)->s_es
->s_rev_level
==
2562 cpu_to_le32(EXT3_GOOD_OLD_REV
)) {
2563 /* If this is the first large file
2564 * created, add a flag to the superblock.
2566 err
= ext3_journal_get_write_access(handle
,
2567 EXT3_SB(sb
)->s_sbh
);
2570 ext3_update_dynamic_rev(sb
);
2571 EXT3_SET_RO_COMPAT_FEATURE(sb
,
2572 EXT3_FEATURE_RO_COMPAT_LARGE_FILE
);
2575 err
= ext3_journal_dirty_metadata(handle
,
2576 EXT3_SB(sb
)->s_sbh
);
2580 raw_inode
->i_generation
= cpu_to_le32(inode
->i_generation
);
2581 if (S_ISCHR(inode
->i_mode
) || S_ISBLK(inode
->i_mode
))
2582 raw_inode
->i_block
[0] =
2583 cpu_to_le32(kdev_t_to_nr(inode
->i_rdev
));
2584 else for (block
= 0; block
< EXT3_N_BLOCKS
; block
++)
2585 raw_inode
->i_block
[block
] = ei
->i_data
[block
];
2587 BUFFER_TRACE(bh
, "call ext3_journal_dirty_metadata");
2588 rc
= ext3_journal_dirty_metadata(handle
, bh
);
2591 ei
->i_state
&= ~EXT3_STATE_NEW
;
2595 ext3_std_error(inode
->i_sb
, err
);
2600 * ext3_write_inode()
2602 * We are called from a few places:
2604 * - Within generic_file_write() for O_SYNC files.
2605 * Here, there will be no transaction running. We wait for any running
2606 * trasnaction to commit.
2608 * - Within sys_sync(), kupdate and such.
2609 * We wait on commit, if tol to.
2611 * - Within prune_icache() (PF_MEMALLOC == true)
2612 * Here we simply return. We can't afford to block kswapd on the
2615 * In all cases it is actually safe for us to return without doing anything,
2616 * because the inode has been copied into a raw inode buffer in
2617 * ext3_mark_inode_dirty(). This is a correctness thing for O_SYNC and for
2620 * Note that we are absolutely dependent upon all inode dirtiers doing the
2621 * right thing: they *must* call mark_inode_dirty() after dirtying info in
2622 * which we are interested.
2624 * It would be a bug for them to not do this. The code:
2626 * mark_inode_dirty(inode)
2628 * inode->i_size = expr;
2630 * is in error because a kswapd-driven write_inode() could occur while
2631 * `stuff()' is running, and the new i_size will be lost. Plus the inode
2632 * will no longer be on the superblock's dirty inode list.
2634 void ext3_write_inode(struct inode
*inode
, int wait
)
2636 if (current
->flags
& PF_MEMALLOC
)
2639 if (ext3_journal_current_handle()) {
2640 jbd_debug(0, "called recursively, non-PF_MEMALLOC!\n");
2648 ext3_force_commit(inode
->i_sb
);
2654 * Called from notify_change.
2656 * We want to trap VFS attempts to truncate the file as soon as
2657 * possible. In particular, we want to make sure that when the VFS
2658 * shrinks i_size, we put the inode on the orphan list and modify
2659 * i_disksize immediately, so that during the subsequent flushing of
2660 * dirty pages and freeing of disk blocks, we can guarantee that any
2661 * commit will leave the blocks being flushed in an unused state on
2662 * disk. (On recovery, the inode will get truncated and the blocks will
2663 * be freed, so we have a strong guarantee that no future commit will
2664 * leave these blocks visible to the user.)
2666 * Called with inode->sem down.
2668 int ext3_setattr(struct dentry
*dentry
, struct iattr
*attr
)
2670 struct inode
*inode
= dentry
->d_inode
;
2672 const unsigned int ia_valid
= attr
->ia_valid
;
2674 error
= inode_change_ok(inode
, attr
);
2678 if ((ia_valid
& ATTR_UID
&& attr
->ia_uid
!= inode
->i_uid
) ||
2679 (ia_valid
& ATTR_GID
&& attr
->ia_gid
!= inode
->i_gid
)) {
2680 error
= DQUOT_TRANSFER(inode
, attr
) ? -EDQUOT
: 0;
2685 if (S_ISREG(inode
->i_mode
) &&
2686 attr
->ia_valid
& ATTR_SIZE
&& attr
->ia_size
< inode
->i_size
) {
2689 handle
= ext3_journal_start(inode
, 3);
2690 if (IS_ERR(handle
)) {
2691 error
= PTR_ERR(handle
);
2695 error
= ext3_orphan_add(handle
, inode
);
2696 EXT3_I(inode
)->i_disksize
= attr
->ia_size
;
2697 rc
= ext3_mark_inode_dirty(handle
, inode
);
2700 ext3_journal_stop(handle
);
2703 rc
= inode_setattr(inode
, attr
);
2705 /* If inode_setattr's call to ext3_truncate failed to get a
2706 * transaction handle at all, we need to clean up the in-core
2707 * orphan list manually. */
2709 ext3_orphan_del(NULL
, inode
);
2711 if (!rc
&& (ia_valid
& ATTR_MODE
))
2712 rc
= ext3_acl_chmod(inode
);
2715 ext3_std_error(inode
->i_sb
, error
);
2723 * akpm: how many blocks doth make a writepage()?
2725 * With N blocks per page, it may be:
2730 * N+5 bitmap blocks (from the above)
2731 * N+5 group descriptor summary blocks
2734 * 2 * EXT3_SINGLEDATA_TRANS_BLOCKS for the quote files
2736 * 3 * (N + 5) + 2 + 2 * EXT3_SINGLEDATA_TRANS_BLOCKS
2738 * With ordered or writeback data it's the same, less the N data blocks.
2740 * If the inode's direct blocks can hold an integral number of pages then a
2741 * page cannot straddle two indirect blocks, and we can only touch one indirect
2742 * and dindirect block, and the "5" above becomes "3".
2744 * This still overestimates under most circumstances. If we were to pass the
2745 * start and end offsets in here as well we could do block_to_path() on each
2746 * block and work out the exact number of indirects which are touched. Pah.
2749 int ext3_writepage_trans_blocks(struct inode
*inode
)
2751 int bpp
= ext3_journal_blocks_per_page(inode
);
2752 int indirects
= (EXT3_NDIR_BLOCKS
% bpp
) ? 5 : 3;
2755 if (ext3_should_journal_data(inode
))
2756 ret
= 3 * (bpp
+ indirects
) + 2;
2758 ret
= 2 * (bpp
+ indirects
) + 2;
2761 ret
+= 2 * EXT3_SINGLEDATA_TRANS_BLOCKS
;
2768 * The caller must have previously called ext3_reserve_inode_write().
2769 * Give this, we know that the caller already has write access to iloc->bh.
2771 int ext3_mark_iloc_dirty(handle_t
*handle
,
2772 struct inode
*inode
, struct ext3_iloc
*iloc
)
2776 /* the do_update_inode consumes one bh->b_count */
2779 /* ext3_do_update_inode() does journal_dirty_metadata */
2780 err
= ext3_do_update_inode(handle
, inode
, iloc
);
2786 * On success, We end up with an outstanding reference count against
2787 * iloc->bh. This _must_ be cleaned up later.
2791 ext3_reserve_inode_write(handle_t
*handle
, struct inode
*inode
,
2792 struct ext3_iloc
*iloc
)
2796 err
= ext3_get_inode_loc(inode
, iloc
);
2798 BUFFER_TRACE(iloc
->bh
, "get_write_access");
2799 err
= ext3_journal_get_write_access(handle
, iloc
->bh
);
2806 ext3_std_error(inode
->i_sb
, err
);
2811 * akpm: What we do here is to mark the in-core inode as clean
2812 * with respect to inode dirtiness (it may still be data-dirty).
2813 * This means that the in-core inode may be reaped by prune_icache
2814 * without having to perform any I/O. This is a very good thing,
2815 * because *any* task may call prune_icache - even ones which
2816 * have a transaction open against a different journal.
2818 * Is this cheating? Not really. Sure, we haven't written the
2819 * inode out, but prune_icache isn't a user-visible syncing function.
2820 * Whenever the user wants stuff synced (sys_sync, sys_msync, sys_fsync)
2821 * we start and wait on commits.
2823 * Is this efficient/effective? Well, we're being nice to the system
2824 * by cleaning up our inodes proactively so they can be reaped
2825 * without I/O. But we are potentially leaving up to five seconds'
2826 * worth of inodes floating about which prune_icache wants us to
2827 * write out. One way to fix that would be to get prune_icache()
2828 * to do a write_super() to free up some memory. It has the desired
2831 int ext3_mark_inode_dirty(handle_t
*handle
, struct inode
*inode
)
2833 struct ext3_iloc iloc
;
2836 err
= ext3_reserve_inode_write(handle
, inode
, &iloc
);
2838 err
= ext3_mark_iloc_dirty(handle
, inode
, &iloc
);
2843 * akpm: ext3_dirty_inode() is called from __mark_inode_dirty()
2845 * We're really interested in the case where a file is being extended.
2846 * i_size has been changed by generic_commit_write() and we thus need
2847 * to include the updated inode in the current transaction.
2849 * Also, DQUOT_ALLOC_SPACE() will always dirty the inode when blocks
2850 * are allocated to the file.
2852 * If the inode is marked synchronous, we don't honour that here - doing
2853 * so would cause a commit on atime updates, which we don't bother doing.
2854 * We handle synchronous inodes at the highest possible level.
2856 void ext3_dirty_inode(struct inode
*inode
)
2858 handle_t
*current_handle
= ext3_journal_current_handle();
2861 handle
= ext3_journal_start(inode
, 2);
2864 if (current_handle
&&
2865 current_handle
->h_transaction
!= handle
->h_transaction
) {
2866 /* This task has a transaction open against a different fs */
2867 printk(KERN_EMERG
"%s: transactions do not match!\n",
2870 jbd_debug(5, "marking dirty. outer handle=%p\n",
2872 ext3_mark_inode_dirty(handle
, inode
);
2874 ext3_journal_stop(handle
);
2881 * Bind an inode's backing buffer_head into this transaction, to prevent
2882 * it from being flushed to disk early. Unlike
2883 * ext3_reserve_inode_write, this leaves behind no bh reference and
2884 * returns no iloc structure, so the caller needs to repeat the iloc
2885 * lookup to mark the inode dirty later.
2888 ext3_pin_inode(handle_t
*handle
, struct inode
*inode
)
2890 struct ext3_iloc iloc
;
2894 err
= ext3_get_inode_loc(inode
, &iloc
);
2896 BUFFER_TRACE(iloc
.bh
, "get_write_access");
2897 err
= journal_get_write_access(handle
, iloc
.bh
);
2899 err
= ext3_journal_dirty_metadata(handle
,
2904 ext3_std_error(inode
->i_sb
, err
);
2909 int ext3_change_inode_journal_flag(struct inode
*inode
, int val
)
2916 * We have to be very careful here: changing a data block's
2917 * journaling status dynamically is dangerous. If we write a
2918 * data block to the journal, change the status and then delete
2919 * that block, we risk forgetting to revoke the old log record
2920 * from the journal and so a subsequent replay can corrupt data.
2921 * So, first we make sure that the journal is empty and that
2922 * nobody is changing anything.
2925 journal
= EXT3_JOURNAL(inode
);
2926 if (is_journal_aborted(journal
) || IS_RDONLY(inode
))
2929 journal_lock_updates(journal
);
2930 journal_flush(journal
);
2933 * OK, there are no updates running now, and all cached data is
2934 * synced to disk. We are now in a completely consistent state
2935 * which doesn't have anything in the journal, and we know that
2936 * no filesystem updates are running, so it is safe to modify
2937 * the inode's in-core data-journaling state flag now.
2941 EXT3_I(inode
)->i_flags
|= EXT3_JOURNAL_DATA_FL
;
2943 EXT3_I(inode
)->i_flags
&= ~EXT3_JOURNAL_DATA_FL
;
2945 journal_unlock_updates(journal
);
2947 /* Finally we can mark the inode as dirty. */
2949 handle
= ext3_journal_start(inode
, 1);
2951 return PTR_ERR(handle
);
2953 err
= ext3_mark_inode_dirty(handle
, inode
);
2955 ext3_journal_stop(handle
);
2956 ext3_std_error(inode
->i_sb
, err
);