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 truncate_inode_pages(&inode
->i_data
, 0);
201 if (is_bad_inode(inode
))
204 handle
= start_transaction(inode
);
205 if (IS_ERR(handle
)) {
207 * If we're going to skip the normal cleanup, we still need to
208 * make sure that the in-core orphan linked list is properly
211 ext3_orphan_del(NULL
, inode
);
219 ext3_truncate(inode
);
221 * Kill off the orphan record which ext3_truncate created.
222 * AKPM: I think this can be inside the above `if'.
223 * Note that ext3_orphan_del() has to be able to cope with the
224 * deletion of a non-existent orphan - this is because we don't
225 * know if ext3_truncate() actually created an orphan record.
226 * (Well, we could do this if we need to, but heck - it works)
228 ext3_orphan_del(handle
, inode
);
229 EXT3_I(inode
)->i_dtime
= get_seconds();
232 * One subtle ordering requirement: if anything has gone wrong
233 * (transaction abort, IO errors, whatever), then we can still
234 * do these next steps (the fs will already have been marked as
235 * having errors), but we can't free the inode if the mark_dirty
238 if (ext3_mark_inode_dirty(handle
, inode
))
239 /* If that failed, just do the required in-core inode clear. */
242 ext3_free_inode(handle
, inode
);
243 ext3_journal_stop(handle
);
246 clear_inode(inode
); /* We must guarantee clearing of inode... */
252 struct buffer_head
*bh
;
255 static inline void add_chain(Indirect
*p
, struct buffer_head
*bh
, __le32
*v
)
257 p
->key
= *(p
->p
= v
);
261 static int verify_chain(Indirect
*from
, Indirect
*to
)
263 while (from
<= to
&& from
->key
== *from
->p
)
269 * ext3_block_to_path - parse the block number into array of offsets
270 * @inode: inode in question (we are only interested in its superblock)
271 * @i_block: block number to be parsed
272 * @offsets: array to store the offsets in
273 * @boundary: set this non-zero if the referred-to block is likely to be
274 * followed (on disk) by an indirect block.
276 * To store the locations of file's data ext3 uses a data structure common
277 * for UNIX filesystems - tree of pointers anchored in the inode, with
278 * data blocks at leaves and indirect blocks in intermediate nodes.
279 * This function translates the block number into path in that tree -
280 * return value is the path length and @offsets[n] is the offset of
281 * pointer to (n+1)th node in the nth one. If @block is out of range
282 * (negative or too large) warning is printed and zero returned.
284 * Note: function doesn't find node addresses, so no IO is needed. All
285 * we need to know is the capacity of indirect blocks (taken from the
290 * Portability note: the last comparison (check that we fit into triple
291 * indirect block) is spelled differently, because otherwise on an
292 * architecture with 32-bit longs and 8Kb pages we might get into trouble
293 * if our filesystem had 8Kb blocks. We might use long long, but that would
294 * kill us on x86. Oh, well, at least the sign propagation does not matter -
295 * i_block would have to be negative in the very beginning, so we would not
299 static int ext3_block_to_path(struct inode
*inode
,
300 long i_block
, int offsets
[4], int *boundary
)
302 int ptrs
= EXT3_ADDR_PER_BLOCK(inode
->i_sb
);
303 int ptrs_bits
= EXT3_ADDR_PER_BLOCK_BITS(inode
->i_sb
);
304 const long direct_blocks
= EXT3_NDIR_BLOCKS
,
305 indirect_blocks
= ptrs
,
306 double_blocks
= (1 << (ptrs_bits
* 2));
311 ext3_warning (inode
->i_sb
, "ext3_block_to_path", "block < 0");
312 } else if (i_block
< direct_blocks
) {
313 offsets
[n
++] = i_block
;
314 final
= direct_blocks
;
315 } else if ( (i_block
-= direct_blocks
) < indirect_blocks
) {
316 offsets
[n
++] = EXT3_IND_BLOCK
;
317 offsets
[n
++] = i_block
;
319 } else if ((i_block
-= indirect_blocks
) < double_blocks
) {
320 offsets
[n
++] = EXT3_DIND_BLOCK
;
321 offsets
[n
++] = i_block
>> ptrs_bits
;
322 offsets
[n
++] = i_block
& (ptrs
- 1);
324 } else if (((i_block
-= double_blocks
) >> (ptrs_bits
* 2)) < ptrs
) {
325 offsets
[n
++] = EXT3_TIND_BLOCK
;
326 offsets
[n
++] = i_block
>> (ptrs_bits
* 2);
327 offsets
[n
++] = (i_block
>> ptrs_bits
) & (ptrs
- 1);
328 offsets
[n
++] = i_block
& (ptrs
- 1);
331 ext3_warning(inode
->i_sb
, "ext3_block_to_path", "block > big");
334 *boundary
= final
- 1 - (i_block
& (ptrs
- 1));
339 * ext3_get_branch - read the chain of indirect blocks leading to data
340 * @inode: inode in question
341 * @depth: depth of the chain (1 - direct pointer, etc.)
342 * @offsets: offsets of pointers in inode/indirect blocks
343 * @chain: place to store the result
344 * @err: here we store the error value
346 * Function fills the array of triples <key, p, bh> and returns %NULL
347 * if everything went OK or the pointer to the last filled triple
348 * (incomplete one) otherwise. Upon the return chain[i].key contains
349 * the number of (i+1)-th block in the chain (as it is stored in memory,
350 * i.e. little-endian 32-bit), chain[i].p contains the address of that
351 * number (it points into struct inode for i==0 and into the bh->b_data
352 * for i>0) and chain[i].bh points to the buffer_head of i-th indirect
353 * block for i>0 and NULL for i==0. In other words, it holds the block
354 * numbers of the chain, addresses they were taken from (and where we can
355 * verify that chain did not change) and buffer_heads hosting these
358 * Function stops when it stumbles upon zero pointer (absent block)
359 * (pointer to last triple returned, *@err == 0)
360 * or when it gets an IO error reading an indirect block
361 * (ditto, *@err == -EIO)
362 * or when it notices that chain had been changed while it was reading
363 * (ditto, *@err == -EAGAIN)
364 * or when it reads all @depth-1 indirect blocks successfully and finds
365 * the whole chain, all way to the data (returns %NULL, *err == 0).
367 static Indirect
*ext3_get_branch(struct inode
*inode
, int depth
, int *offsets
,
368 Indirect chain
[4], int *err
)
370 struct super_block
*sb
= inode
->i_sb
;
372 struct buffer_head
*bh
;
375 /* i_data is not going away, no lock needed */
376 add_chain (chain
, NULL
, EXT3_I(inode
)->i_data
+ *offsets
);
380 bh
= sb_bread(sb
, le32_to_cpu(p
->key
));
383 /* Reader: pointers */
384 if (!verify_chain(chain
, p
))
386 add_chain(++p
, bh
, (__le32
*)bh
->b_data
+ *++offsets
);
404 * ext3_find_near - find a place for allocation with sufficient locality
406 * @ind: descriptor of indirect block.
408 * This function returns the preferred place for block allocation.
409 * It is used when heuristic for sequential allocation fails.
411 * + if there is a block to the left of our position - allocate near it.
412 * + if pointer will live in indirect block - allocate near that block.
413 * + if pointer will live in inode - allocate in the same
416 * In the latter case we colour the starting block by the callers PID to
417 * prevent it from clashing with concurrent allocations for a different inode
418 * in the same block group. The PID is used here so that functionally related
419 * files will be close-by on-disk.
421 * Caller must make sure that @ind is valid and will stay that way.
423 static ext3_fsblk_t
ext3_find_near(struct inode
*inode
, Indirect
*ind
)
425 struct ext3_inode_info
*ei
= EXT3_I(inode
);
426 __le32
*start
= ind
->bh
? (__le32
*) ind
->bh
->b_data
: ei
->i_data
;
428 ext3_fsblk_t bg_start
;
429 ext3_grpblk_t colour
;
431 /* Try to find previous block */
432 for (p
= ind
->p
- 1; p
>= start
; p
--) {
434 return le32_to_cpu(*p
);
437 /* No such thing, so let's try location of indirect block */
439 return ind
->bh
->b_blocknr
;
442 * It is going to be referred to from the inode itself? OK, just put it
443 * into the same cylinder group then.
445 bg_start
= ext3_group_first_block_no(inode
->i_sb
, ei
->i_block_group
);
446 colour
= (current
->pid
% 16) *
447 (EXT3_BLOCKS_PER_GROUP(inode
->i_sb
) / 16);
448 return bg_start
+ colour
;
452 * ext3_find_goal - find a preferred place for allocation.
454 * @block: block we want
455 * @partial: pointer to the last triple within a chain
457 * Normally this function find the preferred place for block allocation,
461 static ext3_fsblk_t
ext3_find_goal(struct inode
*inode
, long block
,
464 struct ext3_block_alloc_info
*block_i
;
466 block_i
= EXT3_I(inode
)->i_block_alloc_info
;
469 * try the heuristic for sequential allocation,
470 * failing that at least try to get decent locality.
472 if (block_i
&& (block
== block_i
->last_alloc_logical_block
+ 1)
473 && (block_i
->last_alloc_physical_block
!= 0)) {
474 return block_i
->last_alloc_physical_block
+ 1;
477 return ext3_find_near(inode
, partial
);
481 * ext3_blks_to_allocate: Look up the block map and count the number
482 * of direct blocks need to be allocated for the given branch.
484 * @branch: chain of indirect blocks
485 * @k: number of blocks need for indirect blocks
486 * @blks: number of data blocks to be mapped.
487 * @blocks_to_boundary: the offset in the indirect block
489 * return the total number of blocks to be allocate, including the
490 * direct and indirect blocks.
492 static int ext3_blks_to_allocate(Indirect
*branch
, int k
, unsigned long blks
,
493 int blocks_to_boundary
)
495 unsigned long count
= 0;
498 * Simple case, [t,d]Indirect block(s) has not allocated yet
499 * then it's clear blocks on that path have not allocated
502 /* right now we don't handle cross boundary allocation */
503 if (blks
< blocks_to_boundary
+ 1)
506 count
+= blocks_to_boundary
+ 1;
511 while (count
< blks
&& count
<= blocks_to_boundary
&&
512 le32_to_cpu(*(branch
[0].p
+ count
)) == 0) {
519 * ext3_alloc_blocks: multiple allocate blocks needed for a branch
520 * @indirect_blks: the number of blocks need to allocate for indirect
523 * @new_blocks: on return it will store the new block numbers for
524 * the indirect blocks(if needed) and the first direct block,
525 * @blks: on return it will store the total number of allocated
528 static int ext3_alloc_blocks(handle_t
*handle
, struct inode
*inode
,
529 ext3_fsblk_t goal
, int indirect_blks
, int blks
,
530 ext3_fsblk_t new_blocks
[4], int *err
)
533 unsigned long count
= 0;
535 ext3_fsblk_t current_block
= 0;
539 * Here we try to allocate the requested multiple blocks at once,
540 * on a best-effort basis.
541 * To build a branch, we should allocate blocks for
542 * the indirect blocks(if not allocated yet), and at least
543 * the first direct block of this branch. That's the
544 * minimum number of blocks need to allocate(required)
546 target
= blks
+ indirect_blks
;
550 /* allocating blocks for indirect blocks and direct blocks */
551 current_block
= ext3_new_blocks(handle
,inode
,goal
,&count
,err
);
556 /* allocate blocks for indirect blocks */
557 while (index
< indirect_blks
&& count
) {
558 new_blocks
[index
++] = current_block
++;
566 /* save the new block number for the first direct block */
567 new_blocks
[index
] = current_block
;
569 /* total number of blocks allocated for direct blocks */
574 for (i
= 0; i
<index
; i
++)
575 ext3_free_blocks(handle
, inode
, new_blocks
[i
], 1);
580 * ext3_alloc_branch - allocate and set up a chain of blocks.
582 * @indirect_blks: number of allocated indirect blocks
583 * @blks: number of allocated direct blocks
584 * @offsets: offsets (in the blocks) to store the pointers to next.
585 * @branch: place to store the chain in.
587 * This function allocates blocks, zeroes out all but the last one,
588 * links them into chain and (if we are synchronous) writes them to disk.
589 * In other words, it prepares a branch that can be spliced onto the
590 * inode. It stores the information about that chain in the branch[], in
591 * the same format as ext3_get_branch() would do. We are calling it after
592 * we had read the existing part of chain and partial points to the last
593 * triple of that (one with zero ->key). Upon the exit we have the same
594 * picture as after the successful ext3_get_block(), except that in one
595 * place chain is disconnected - *branch->p is still zero (we did not
596 * set the last link), but branch->key contains the number that should
597 * be placed into *branch->p to fill that gap.
599 * If allocation fails we free all blocks we've allocated (and forget
600 * their buffer_heads) and return the error value the from failed
601 * ext3_alloc_block() (normally -ENOSPC). Otherwise we set the chain
602 * as described above and return 0.
604 static int ext3_alloc_branch(handle_t
*handle
, struct inode
*inode
,
605 int indirect_blks
, int *blks
, ext3_fsblk_t goal
,
606 int *offsets
, Indirect
*branch
)
608 int blocksize
= inode
->i_sb
->s_blocksize
;
611 struct buffer_head
*bh
;
613 ext3_fsblk_t new_blocks
[4];
614 ext3_fsblk_t current_block
;
616 num
= ext3_alloc_blocks(handle
, inode
, goal
, indirect_blks
,
617 *blks
, new_blocks
, &err
);
621 branch
[0].key
= cpu_to_le32(new_blocks
[0]);
623 * metadata blocks and data blocks are allocated.
625 for (n
= 1; n
<= indirect_blks
; n
++) {
627 * Get buffer_head for parent block, zero it out
628 * and set the pointer to new one, then send
631 bh
= sb_getblk(inode
->i_sb
, new_blocks
[n
-1]);
634 BUFFER_TRACE(bh
, "call get_create_access");
635 err
= ext3_journal_get_create_access(handle
, bh
);
642 memset(bh
->b_data
, 0, blocksize
);
643 branch
[n
].p
= (__le32
*) bh
->b_data
+ offsets
[n
];
644 branch
[n
].key
= cpu_to_le32(new_blocks
[n
]);
645 *branch
[n
].p
= branch
[n
].key
;
646 if ( n
== indirect_blks
) {
647 current_block
= new_blocks
[n
];
649 * End of chain, update the last new metablock of
650 * the chain to point to the new allocated
651 * data blocks numbers
653 for (i
=1; i
< num
; i
++)
654 *(branch
[n
].p
+ i
) = cpu_to_le32(++current_block
);
656 BUFFER_TRACE(bh
, "marking uptodate");
657 set_buffer_uptodate(bh
);
660 BUFFER_TRACE(bh
, "call ext3_journal_dirty_metadata");
661 err
= ext3_journal_dirty_metadata(handle
, bh
);
668 /* Allocation failed, free what we already allocated */
669 for (i
= 1; i
<= n
; i
++) {
670 BUFFER_TRACE(branch
[i
].bh
, "call journal_forget");
671 ext3_journal_forget(handle
, branch
[i
].bh
);
673 for (i
= 0; i
<indirect_blks
; i
++)
674 ext3_free_blocks(handle
, inode
, new_blocks
[i
], 1);
676 ext3_free_blocks(handle
, inode
, new_blocks
[i
], num
);
682 * ext3_splice_branch - splice the allocated branch onto inode.
684 * @block: (logical) number of block we are adding
685 * @chain: chain of indirect blocks (with a missing link - see
687 * @where: location of missing link
688 * @num: number of indirect blocks we are adding
689 * @blks: number of direct blocks we are adding
691 * This function fills the missing link and does all housekeeping needed in
692 * inode (->i_blocks, etc.). In case of success we end up with the full
693 * chain to new block and return 0.
695 static int ext3_splice_branch(handle_t
*handle
, struct inode
*inode
,
696 long block
, Indirect
*where
, int num
, int blks
)
700 struct ext3_block_alloc_info
*block_i
;
701 ext3_fsblk_t current_block
;
702 struct ext3_inode_info
*ei
= EXT3_I(inode
);
704 block_i
= ei
->i_block_alloc_info
;
706 * If we're splicing into a [td]indirect block (as opposed to the
707 * inode) then we need to get write access to the [td]indirect block
711 BUFFER_TRACE(where
->bh
, "get_write_access");
712 err
= ext3_journal_get_write_access(handle
, where
->bh
);
718 *where
->p
= where
->key
;
721 * Update the host buffer_head or inode to point to more just allocated
722 * direct blocks blocks
724 if (num
== 0 && blks
> 1) {
725 current_block
= le32_to_cpu(where
->key
) + 1;
726 for (i
= 1; i
< blks
; i
++)
727 *(where
->p
+ i
) = cpu_to_le32(current_block
++);
731 * update the most recently allocated logical & physical block
732 * in i_block_alloc_info, to assist find the proper goal block for next
736 block_i
->last_alloc_logical_block
= block
+ blks
- 1;
737 block_i
->last_alloc_physical_block
=
738 le32_to_cpu(where
[num
].key
) + blks
- 1;
741 /* We are done with atomic stuff, now do the rest of housekeeping */
743 inode
->i_ctime
= CURRENT_TIME_SEC
;
744 ext3_mark_inode_dirty(handle
, inode
);
745 /* ext3_mark_inode_dirty already updated i_sync_tid */
746 atomic_set(&ei
->i_datasync_tid
, handle
->h_transaction
->t_tid
);
748 /* had we spliced it onto indirect block? */
751 * If we spliced it onto an indirect block, we haven't
752 * altered the inode. Note however that if it is being spliced
753 * onto an indirect block at the very end of the file (the
754 * file is growing) then we *will* alter the inode to reflect
755 * the new i_size. But that is not done here - it is done in
756 * generic_commit_write->__mark_inode_dirty->ext3_dirty_inode.
758 jbd_debug(5, "splicing indirect only\n");
759 BUFFER_TRACE(where
->bh
, "call ext3_journal_dirty_metadata");
760 err
= ext3_journal_dirty_metadata(handle
, where
->bh
);
765 * OK, we spliced it into the inode itself on a direct block.
766 * Inode was dirtied above.
768 jbd_debug(5, "splicing direct\n");
773 for (i
= 1; i
<= num
; i
++) {
774 BUFFER_TRACE(where
[i
].bh
, "call journal_forget");
775 ext3_journal_forget(handle
, where
[i
].bh
);
776 ext3_free_blocks(handle
,inode
,le32_to_cpu(where
[i
-1].key
),1);
778 ext3_free_blocks(handle
, inode
, le32_to_cpu(where
[num
].key
), blks
);
784 * Allocation strategy is simple: if we have to allocate something, we will
785 * have to go the whole way to leaf. So let's do it before attaching anything
786 * to tree, set linkage between the newborn blocks, write them if sync is
787 * required, recheck the path, free and repeat if check fails, otherwise
788 * set the last missing link (that will protect us from any truncate-generated
789 * removals - all blocks on the path are immune now) and possibly force the
790 * write on the parent block.
791 * That has a nice additional property: no special recovery from the failed
792 * allocations is needed - we simply release blocks and do not touch anything
793 * reachable from inode.
795 * `handle' can be NULL if create == 0.
797 * The BKL may not be held on entry here. Be sure to take it early.
798 * return > 0, # of blocks mapped or allocated.
799 * return = 0, if plain lookup failed.
800 * return < 0, error case.
802 int ext3_get_blocks_handle(handle_t
*handle
, struct inode
*inode
,
803 sector_t iblock
, unsigned long maxblocks
,
804 struct buffer_head
*bh_result
,
813 int blocks_to_boundary
= 0;
815 struct ext3_inode_info
*ei
= EXT3_I(inode
);
817 ext3_fsblk_t first_block
= 0;
820 J_ASSERT(handle
!= NULL
|| create
== 0);
821 depth
= ext3_block_to_path(inode
,iblock
,offsets
,&blocks_to_boundary
);
826 partial
= ext3_get_branch(inode
, depth
, offsets
, chain
, &err
);
828 /* Simplest case - block found, no allocation needed */
830 first_block
= le32_to_cpu(chain
[depth
- 1].key
);
831 clear_buffer_new(bh_result
);
834 while (count
< maxblocks
&& count
<= blocks_to_boundary
) {
837 if (!verify_chain(chain
, chain
+ depth
- 1)) {
839 * Indirect block might be removed by
840 * truncate while we were reading it.
841 * Handling of that case: forget what we've
842 * got now. Flag the err as EAGAIN, so it
849 blk
= le32_to_cpu(*(chain
[depth
-1].p
+ count
));
851 if (blk
== first_block
+ count
)
860 /* Next simple case - plain lookup or failed read of indirect block */
861 if (!create
|| err
== -EIO
)
864 mutex_lock(&ei
->truncate_mutex
);
867 * If the indirect block is missing while we are reading
868 * the chain(ext3_get_branch() returns -EAGAIN err), or
869 * if the chain has been changed after we grab the semaphore,
870 * (either because another process truncated this branch, or
871 * another get_block allocated this branch) re-grab the chain to see if
872 * the request block has been allocated or not.
874 * Since we already block the truncate/other get_block
875 * at this point, we will have the current copy of the chain when we
876 * splice the branch into the tree.
878 if (err
== -EAGAIN
|| !verify_chain(chain
, partial
)) {
879 while (partial
> chain
) {
883 partial
= ext3_get_branch(inode
, depth
, offsets
, chain
, &err
);
886 mutex_unlock(&ei
->truncate_mutex
);
889 clear_buffer_new(bh_result
);
895 * Okay, we need to do block allocation. Lazily initialize the block
896 * allocation info here if necessary
898 if (S_ISREG(inode
->i_mode
) && (!ei
->i_block_alloc_info
))
899 ext3_init_block_alloc_info(inode
);
901 goal
= ext3_find_goal(inode
, iblock
, partial
);
903 /* the number of blocks need to allocate for [d,t]indirect blocks */
904 indirect_blks
= (chain
+ depth
) - partial
- 1;
907 * Next look up the indirect map to count the totoal number of
908 * direct blocks to allocate for this branch.
910 count
= ext3_blks_to_allocate(partial
, indirect_blks
,
911 maxblocks
, blocks_to_boundary
);
913 * Block out ext3_truncate while we alter the tree
915 err
= ext3_alloc_branch(handle
, inode
, indirect_blks
, &count
, goal
,
916 offsets
+ (partial
- chain
), partial
);
919 * The ext3_splice_branch call will free and forget any buffers
920 * on the new chain if there is a failure, but that risks using
921 * up transaction credits, especially for bitmaps where the
922 * credits cannot be returned. Can we handle this somehow? We
923 * may need to return -EAGAIN upwards in the worst case. --sct
926 err
= ext3_splice_branch(handle
, inode
, iblock
,
927 partial
, indirect_blks
, count
);
928 mutex_unlock(&ei
->truncate_mutex
);
932 set_buffer_new(bh_result
);
934 map_bh(bh_result
, inode
->i_sb
, le32_to_cpu(chain
[depth
-1].key
));
935 if (count
> blocks_to_boundary
)
936 set_buffer_boundary(bh_result
);
938 /* Clean up and exit */
939 partial
= chain
+ depth
- 1; /* the whole chain */
941 while (partial
> chain
) {
942 BUFFER_TRACE(partial
->bh
, "call brelse");
946 BUFFER_TRACE(bh_result
, "returned");
951 /* Maximum number of blocks we map for direct IO at once. */
952 #define DIO_MAX_BLOCKS 4096
954 * Number of credits we need for writing DIO_MAX_BLOCKS:
955 * We need sb + group descriptor + bitmap + inode -> 4
956 * For B blocks with A block pointers per block we need:
957 * 1 (triple ind.) + (B/A/A + 2) (doubly ind.) + (B/A + 2) (indirect).
958 * If we plug in 4096 for B and 256 for A (for 1KB block size), we get 25.
960 #define DIO_CREDITS 25
962 static int ext3_get_block(struct inode
*inode
, sector_t iblock
,
963 struct buffer_head
*bh_result
, int create
)
965 handle_t
*handle
= ext3_journal_current_handle();
966 int ret
= 0, started
= 0;
967 unsigned max_blocks
= bh_result
->b_size
>> inode
->i_blkbits
;
969 if (create
&& !handle
) { /* Direct IO write... */
970 if (max_blocks
> DIO_MAX_BLOCKS
)
971 max_blocks
= DIO_MAX_BLOCKS
;
972 handle
= ext3_journal_start(inode
, DIO_CREDITS
+
973 2 * EXT3_QUOTA_TRANS_BLOCKS(inode
->i_sb
));
974 if (IS_ERR(handle
)) {
975 ret
= PTR_ERR(handle
);
981 ret
= ext3_get_blocks_handle(handle
, inode
, iblock
,
982 max_blocks
, bh_result
, create
);
984 bh_result
->b_size
= (ret
<< inode
->i_blkbits
);
988 ext3_journal_stop(handle
);
993 int ext3_fiemap(struct inode
*inode
, struct fiemap_extent_info
*fieinfo
,
996 return generic_block_fiemap(inode
, fieinfo
, start
, len
,
1001 * `handle' can be NULL if create is zero
1003 struct buffer_head
*ext3_getblk(handle_t
*handle
, struct inode
*inode
,
1004 long block
, int create
, int *errp
)
1006 struct buffer_head dummy
;
1009 J_ASSERT(handle
!= NULL
|| create
== 0);
1012 dummy
.b_blocknr
= -1000;
1013 buffer_trace_init(&dummy
.b_history
);
1014 err
= ext3_get_blocks_handle(handle
, inode
, block
, 1,
1017 * ext3_get_blocks_handle() returns number of blocks
1018 * mapped. 0 in case of a HOLE.
1026 if (!err
&& buffer_mapped(&dummy
)) {
1027 struct buffer_head
*bh
;
1028 bh
= sb_getblk(inode
->i_sb
, dummy
.b_blocknr
);
1033 if (buffer_new(&dummy
)) {
1034 J_ASSERT(create
!= 0);
1035 J_ASSERT(handle
!= NULL
);
1038 * Now that we do not always journal data, we should
1039 * keep in mind whether this should always journal the
1040 * new buffer as metadata. For now, regular file
1041 * writes use ext3_get_block instead, so it's not a
1045 BUFFER_TRACE(bh
, "call get_create_access");
1046 fatal
= ext3_journal_get_create_access(handle
, bh
);
1047 if (!fatal
&& !buffer_uptodate(bh
)) {
1048 memset(bh
->b_data
,0,inode
->i_sb
->s_blocksize
);
1049 set_buffer_uptodate(bh
);
1052 BUFFER_TRACE(bh
, "call ext3_journal_dirty_metadata");
1053 err
= ext3_journal_dirty_metadata(handle
, bh
);
1057 BUFFER_TRACE(bh
, "not a new buffer");
1070 struct buffer_head
*ext3_bread(handle_t
*handle
, struct inode
*inode
,
1071 int block
, int create
, int *err
)
1073 struct buffer_head
* bh
;
1075 bh
= ext3_getblk(handle
, inode
, block
, create
, err
);
1078 if (buffer_uptodate(bh
))
1080 ll_rw_block(READ_META
, 1, &bh
);
1082 if (buffer_uptodate(bh
))
1089 static int walk_page_buffers( handle_t
*handle
,
1090 struct buffer_head
*head
,
1094 int (*fn
)( handle_t
*handle
,
1095 struct buffer_head
*bh
))
1097 struct buffer_head
*bh
;
1098 unsigned block_start
, block_end
;
1099 unsigned blocksize
= head
->b_size
;
1101 struct buffer_head
*next
;
1103 for ( bh
= head
, block_start
= 0;
1104 ret
== 0 && (bh
!= head
|| !block_start
);
1105 block_start
= block_end
, bh
= next
)
1107 next
= bh
->b_this_page
;
1108 block_end
= block_start
+ blocksize
;
1109 if (block_end
<= from
|| block_start
>= to
) {
1110 if (partial
&& !buffer_uptodate(bh
))
1114 err
= (*fn
)(handle
, bh
);
1122 * To preserve ordering, it is essential that the hole instantiation and
1123 * the data write be encapsulated in a single transaction. We cannot
1124 * close off a transaction and start a new one between the ext3_get_block()
1125 * and the commit_write(). So doing the journal_start at the start of
1126 * prepare_write() is the right place.
1128 * Also, this function can nest inside ext3_writepage() ->
1129 * block_write_full_page(). In that case, we *know* that ext3_writepage()
1130 * has generated enough buffer credits to do the whole page. So we won't
1131 * block on the journal in that case, which is good, because the caller may
1134 * By accident, ext3 can be reentered when a transaction is open via
1135 * quota file writes. If we were to commit the transaction while thus
1136 * reentered, there can be a deadlock - we would be holding a quota
1137 * lock, and the commit would never complete if another thread had a
1138 * transaction open and was blocking on the quota lock - a ranking
1141 * So what we do is to rely on the fact that journal_stop/journal_start
1142 * will _not_ run commit under these circumstances because handle->h_ref
1143 * is elevated. We'll still have enough credits for the tiny quotafile
1146 static int do_journal_get_write_access(handle_t
*handle
,
1147 struct buffer_head
*bh
)
1149 if (!buffer_mapped(bh
) || buffer_freed(bh
))
1151 return ext3_journal_get_write_access(handle
, bh
);
1155 * Truncate blocks that were not used by write. We have to truncate the
1156 * pagecache as well so that corresponding buffers get properly unmapped.
1158 static void ext3_truncate_failed_write(struct inode
*inode
)
1160 truncate_inode_pages(inode
->i_mapping
, inode
->i_size
);
1161 ext3_truncate(inode
);
1164 static int ext3_write_begin(struct file
*file
, struct address_space
*mapping
,
1165 loff_t pos
, unsigned len
, unsigned flags
,
1166 struct page
**pagep
, void **fsdata
)
1168 struct inode
*inode
= mapping
->host
;
1175 /* Reserve one block more for addition to orphan list in case
1176 * we allocate blocks but write fails for some reason */
1177 int needed_blocks
= ext3_writepage_trans_blocks(inode
) + 1;
1179 index
= pos
>> PAGE_CACHE_SHIFT
;
1180 from
= pos
& (PAGE_CACHE_SIZE
- 1);
1184 page
= grab_cache_page_write_begin(mapping
, index
, flags
);
1189 handle
= ext3_journal_start(inode
, needed_blocks
);
1190 if (IS_ERR(handle
)) {
1192 page_cache_release(page
);
1193 ret
= PTR_ERR(handle
);
1196 ret
= block_write_begin(file
, mapping
, pos
, len
, flags
, pagep
, fsdata
,
1199 goto write_begin_failed
;
1201 if (ext3_should_journal_data(inode
)) {
1202 ret
= walk_page_buffers(handle
, page_buffers(page
),
1203 from
, to
, NULL
, do_journal_get_write_access
);
1208 * block_write_begin may have instantiated a few blocks
1209 * outside i_size. Trim these off again. Don't need
1210 * i_size_read because we hold i_mutex.
1212 * Add inode to orphan list in case we crash before truncate
1213 * finishes. Do this only if ext3_can_truncate() agrees so
1214 * that orphan processing code is happy.
1216 if (pos
+ len
> inode
->i_size
&& ext3_can_truncate(inode
))
1217 ext3_orphan_add(handle
, inode
);
1218 ext3_journal_stop(handle
);
1220 page_cache_release(page
);
1221 if (pos
+ len
> inode
->i_size
)
1222 ext3_truncate_failed_write(inode
);
1224 if (ret
== -ENOSPC
&& ext3_should_retry_alloc(inode
->i_sb
, &retries
))
1231 int ext3_journal_dirty_data(handle_t
*handle
, struct buffer_head
*bh
)
1233 int err
= journal_dirty_data(handle
, bh
);
1235 ext3_journal_abort_handle(__func__
, __func__
,
1240 /* For ordered writepage and write_end functions */
1241 static int journal_dirty_data_fn(handle_t
*handle
, struct buffer_head
*bh
)
1244 * Write could have mapped the buffer but it didn't copy the data in
1245 * yet. So avoid filing such buffer into a transaction.
1247 if (buffer_mapped(bh
) && buffer_uptodate(bh
))
1248 return ext3_journal_dirty_data(handle
, bh
);
1252 /* For write_end() in data=journal mode */
1253 static int write_end_fn(handle_t
*handle
, struct buffer_head
*bh
)
1255 if (!buffer_mapped(bh
) || buffer_freed(bh
))
1257 set_buffer_uptodate(bh
);
1258 return ext3_journal_dirty_metadata(handle
, bh
);
1262 * This is nasty and subtle: ext3_write_begin() could have allocated blocks
1263 * for the whole page but later we failed to copy the data in. Update inode
1264 * size according to what we managed to copy. The rest is going to be
1265 * truncated in write_end function.
1267 static void update_file_sizes(struct inode
*inode
, loff_t pos
, unsigned copied
)
1269 /* What matters to us is i_disksize. We don't write i_size anywhere */
1270 if (pos
+ copied
> inode
->i_size
)
1271 i_size_write(inode
, pos
+ copied
);
1272 if (pos
+ copied
> EXT3_I(inode
)->i_disksize
) {
1273 EXT3_I(inode
)->i_disksize
= pos
+ copied
;
1274 mark_inode_dirty(inode
);
1279 * We need to pick up the new inode size which generic_commit_write gave us
1280 * `file' can be NULL - eg, when called from page_symlink().
1282 * ext3 never places buffers on inode->i_mapping->private_list. metadata
1283 * buffers are managed internally.
1285 static int ext3_ordered_write_end(struct file
*file
,
1286 struct address_space
*mapping
,
1287 loff_t pos
, unsigned len
, unsigned copied
,
1288 struct page
*page
, void *fsdata
)
1290 handle_t
*handle
= ext3_journal_current_handle();
1291 struct inode
*inode
= file
->f_mapping
->host
;
1295 copied
= block_write_end(file
, mapping
, pos
, len
, copied
, page
, fsdata
);
1297 from
= pos
& (PAGE_CACHE_SIZE
- 1);
1299 ret
= walk_page_buffers(handle
, page_buffers(page
),
1300 from
, to
, NULL
, journal_dirty_data_fn
);
1303 update_file_sizes(inode
, pos
, copied
);
1305 * There may be allocated blocks outside of i_size because
1306 * we failed to copy some data. Prepare for truncate.
1308 if (pos
+ len
> inode
->i_size
&& ext3_can_truncate(inode
))
1309 ext3_orphan_add(handle
, inode
);
1310 ret2
= ext3_journal_stop(handle
);
1314 page_cache_release(page
);
1316 if (pos
+ len
> inode
->i_size
)
1317 ext3_truncate_failed_write(inode
);
1318 return ret
? ret
: copied
;
1321 static int ext3_writeback_write_end(struct file
*file
,
1322 struct address_space
*mapping
,
1323 loff_t pos
, unsigned len
, unsigned copied
,
1324 struct page
*page
, void *fsdata
)
1326 handle_t
*handle
= ext3_journal_current_handle();
1327 struct inode
*inode
= file
->f_mapping
->host
;
1330 copied
= block_write_end(file
, mapping
, pos
, len
, copied
, page
, fsdata
);
1331 update_file_sizes(inode
, pos
, copied
);
1333 * There may be allocated blocks outside of i_size because
1334 * we failed to copy some data. Prepare for truncate.
1336 if (pos
+ len
> inode
->i_size
&& ext3_can_truncate(inode
))
1337 ext3_orphan_add(handle
, inode
);
1338 ret
= ext3_journal_stop(handle
);
1340 page_cache_release(page
);
1342 if (pos
+ len
> inode
->i_size
)
1343 ext3_truncate_failed_write(inode
);
1344 return ret
? ret
: copied
;
1347 static int ext3_journalled_write_end(struct file
*file
,
1348 struct address_space
*mapping
,
1349 loff_t pos
, unsigned len
, unsigned copied
,
1350 struct page
*page
, void *fsdata
)
1352 handle_t
*handle
= ext3_journal_current_handle();
1353 struct inode
*inode
= mapping
->host
;
1358 from
= pos
& (PAGE_CACHE_SIZE
- 1);
1362 if (!PageUptodate(page
))
1364 page_zero_new_buffers(page
, from
+ copied
, to
);
1368 ret
= walk_page_buffers(handle
, page_buffers(page
), from
,
1369 to
, &partial
, write_end_fn
);
1371 SetPageUptodate(page
);
1373 if (pos
+ copied
> inode
->i_size
)
1374 i_size_write(inode
, pos
+ copied
);
1376 * There may be allocated blocks outside of i_size because
1377 * we failed to copy some data. Prepare for truncate.
1379 if (pos
+ len
> inode
->i_size
&& ext3_can_truncate(inode
))
1380 ext3_orphan_add(handle
, inode
);
1381 EXT3_I(inode
)->i_state
|= EXT3_STATE_JDATA
;
1382 if (inode
->i_size
> EXT3_I(inode
)->i_disksize
) {
1383 EXT3_I(inode
)->i_disksize
= inode
->i_size
;
1384 ret2
= ext3_mark_inode_dirty(handle
, inode
);
1389 ret2
= ext3_journal_stop(handle
);
1393 page_cache_release(page
);
1395 if (pos
+ len
> inode
->i_size
)
1396 ext3_truncate_failed_write(inode
);
1397 return ret
? ret
: copied
;
1401 * bmap() is special. It gets used by applications such as lilo and by
1402 * the swapper to find the on-disk block of a specific piece of data.
1404 * Naturally, this is dangerous if the block concerned is still in the
1405 * journal. If somebody makes a swapfile on an ext3 data-journaling
1406 * filesystem and enables swap, then they may get a nasty shock when the
1407 * data getting swapped to that swapfile suddenly gets overwritten by
1408 * the original zero's written out previously to the journal and
1409 * awaiting writeback in the kernel's buffer cache.
1411 * So, if we see any bmap calls here on a modified, data-journaled file,
1412 * take extra steps to flush any blocks which might be in the cache.
1414 static sector_t
ext3_bmap(struct address_space
*mapping
, sector_t block
)
1416 struct inode
*inode
= mapping
->host
;
1420 if (EXT3_I(inode
)->i_state
& EXT3_STATE_JDATA
) {
1422 * This is a REALLY heavyweight approach, but the use of
1423 * bmap on dirty files is expected to be extremely rare:
1424 * only if we run lilo or swapon on a freshly made file
1425 * do we expect this to happen.
1427 * (bmap requires CAP_SYS_RAWIO so this does not
1428 * represent an unprivileged user DOS attack --- we'd be
1429 * in trouble if mortal users could trigger this path at
1432 * NB. EXT3_STATE_JDATA is not set on files other than
1433 * regular files. If somebody wants to bmap a directory
1434 * or symlink and gets confused because the buffer
1435 * hasn't yet been flushed to disk, they deserve
1436 * everything they get.
1439 EXT3_I(inode
)->i_state
&= ~EXT3_STATE_JDATA
;
1440 journal
= EXT3_JOURNAL(inode
);
1441 journal_lock_updates(journal
);
1442 err
= journal_flush(journal
);
1443 journal_unlock_updates(journal
);
1449 return generic_block_bmap(mapping
,block
,ext3_get_block
);
1452 static int bget_one(handle_t
*handle
, struct buffer_head
*bh
)
1458 static int bput_one(handle_t
*handle
, struct buffer_head
*bh
)
1464 static int buffer_unmapped(handle_t
*handle
, struct buffer_head
*bh
)
1466 return !buffer_mapped(bh
);
1470 * Note that we always start a transaction even if we're not journalling
1471 * data. This is to preserve ordering: any hole instantiation within
1472 * __block_write_full_page -> ext3_get_block() should be journalled
1473 * along with the data so we don't crash and then get metadata which
1474 * refers to old data.
1476 * In all journalling modes block_write_full_page() will start the I/O.
1480 * ext3_writepage() -> kmalloc() -> __alloc_pages() -> page_launder() ->
1485 * ext3_file_write() -> generic_file_write() -> __alloc_pages() -> ...
1487 * Same applies to ext3_get_block(). We will deadlock on various things like
1488 * lock_journal and i_truncate_mutex.
1490 * Setting PF_MEMALLOC here doesn't work - too many internal memory
1493 * 16May01: If we're reentered then journal_current_handle() will be
1494 * non-zero. We simply *return*.
1496 * 1 July 2001: @@@ FIXME:
1497 * In journalled data mode, a data buffer may be metadata against the
1498 * current transaction. But the same file is part of a shared mapping
1499 * and someone does a writepage() on it.
1501 * We will move the buffer onto the async_data list, but *after* it has
1502 * been dirtied. So there's a small window where we have dirty data on
1505 * Note that this only applies to the last partial page in the file. The
1506 * bit which block_write_full_page() uses prepare/commit for. (That's
1507 * broken code anyway: it's wrong for msync()).
1509 * It's a rare case: affects the final partial page, for journalled data
1510 * where the file is subject to bith write() and writepage() in the same
1511 * transction. To fix it we'll need a custom block_write_full_page().
1512 * We'll probably need that anyway for journalling writepage() output.
1514 * We don't honour synchronous mounts for writepage(). That would be
1515 * disastrous. Any write() or metadata operation will sync the fs for
1518 * AKPM2: if all the page's buffers are mapped to disk and !data=journal,
1519 * we don't need to open a transaction here.
1521 static int ext3_ordered_writepage(struct page
*page
,
1522 struct writeback_control
*wbc
)
1524 struct inode
*inode
= page
->mapping
->host
;
1525 struct buffer_head
*page_bufs
;
1526 handle_t
*handle
= NULL
;
1530 J_ASSERT(PageLocked(page
));
1533 * We give up here if we're reentered, because it might be for a
1534 * different filesystem.
1536 if (ext3_journal_current_handle())
1539 if (!page_has_buffers(page
)) {
1540 create_empty_buffers(page
, inode
->i_sb
->s_blocksize
,
1541 (1 << BH_Dirty
)|(1 << BH_Uptodate
));
1542 page_bufs
= page_buffers(page
);
1544 page_bufs
= page_buffers(page
);
1545 if (!walk_page_buffers(NULL
, page_bufs
, 0, PAGE_CACHE_SIZE
,
1546 NULL
, buffer_unmapped
)) {
1547 /* Provide NULL get_block() to catch bugs if buffers
1548 * weren't really mapped */
1549 return block_write_full_page(page
, NULL
, wbc
);
1552 handle
= ext3_journal_start(inode
, ext3_writepage_trans_blocks(inode
));
1554 if (IS_ERR(handle
)) {
1555 ret
= PTR_ERR(handle
);
1559 walk_page_buffers(handle
, page_bufs
, 0,
1560 PAGE_CACHE_SIZE
, NULL
, bget_one
);
1562 ret
= block_write_full_page(page
, ext3_get_block
, wbc
);
1565 * The page can become unlocked at any point now, and
1566 * truncate can then come in and change things. So we
1567 * can't touch *page from now on. But *page_bufs is
1568 * safe due to elevated refcount.
1572 * And attach them to the current transaction. But only if
1573 * block_write_full_page() succeeded. Otherwise they are unmapped,
1574 * and generally junk.
1577 err
= walk_page_buffers(handle
, page_bufs
, 0, PAGE_CACHE_SIZE
,
1578 NULL
, journal_dirty_data_fn
);
1582 walk_page_buffers(handle
, page_bufs
, 0,
1583 PAGE_CACHE_SIZE
, NULL
, bput_one
);
1584 err
= ext3_journal_stop(handle
);
1590 redirty_page_for_writepage(wbc
, page
);
1595 static int ext3_writeback_writepage(struct page
*page
,
1596 struct writeback_control
*wbc
)
1598 struct inode
*inode
= page
->mapping
->host
;
1599 handle_t
*handle
= NULL
;
1603 if (ext3_journal_current_handle())
1606 if (page_has_buffers(page
)) {
1607 if (!walk_page_buffers(NULL
, page_buffers(page
), 0,
1608 PAGE_CACHE_SIZE
, NULL
, buffer_unmapped
)) {
1609 /* Provide NULL get_block() to catch bugs if buffers
1610 * weren't really mapped */
1611 return block_write_full_page(page
, NULL
, wbc
);
1615 handle
= ext3_journal_start(inode
, ext3_writepage_trans_blocks(inode
));
1616 if (IS_ERR(handle
)) {
1617 ret
= PTR_ERR(handle
);
1621 if (test_opt(inode
->i_sb
, NOBH
) && ext3_should_writeback_data(inode
))
1622 ret
= nobh_writepage(page
, ext3_get_block
, wbc
);
1624 ret
= block_write_full_page(page
, ext3_get_block
, wbc
);
1626 err
= ext3_journal_stop(handle
);
1632 redirty_page_for_writepage(wbc
, page
);
1637 static int ext3_journalled_writepage(struct page
*page
,
1638 struct writeback_control
*wbc
)
1640 struct inode
*inode
= page
->mapping
->host
;
1641 handle_t
*handle
= NULL
;
1645 if (ext3_journal_current_handle())
1648 handle
= ext3_journal_start(inode
, ext3_writepage_trans_blocks(inode
));
1649 if (IS_ERR(handle
)) {
1650 ret
= PTR_ERR(handle
);
1654 if (!page_has_buffers(page
) || PageChecked(page
)) {
1656 * It's mmapped pagecache. Add buffers and journal it. There
1657 * doesn't seem much point in redirtying the page here.
1659 ClearPageChecked(page
);
1660 ret
= block_prepare_write(page
, 0, PAGE_CACHE_SIZE
,
1663 ext3_journal_stop(handle
);
1666 ret
= walk_page_buffers(handle
, page_buffers(page
), 0,
1667 PAGE_CACHE_SIZE
, NULL
, do_journal_get_write_access
);
1669 err
= walk_page_buffers(handle
, page_buffers(page
), 0,
1670 PAGE_CACHE_SIZE
, NULL
, write_end_fn
);
1673 EXT3_I(inode
)->i_state
|= EXT3_STATE_JDATA
;
1677 * It may be a page full of checkpoint-mode buffers. We don't
1678 * really know unless we go poke around in the buffer_heads.
1679 * But block_write_full_page will do the right thing.
1681 ret
= block_write_full_page(page
, ext3_get_block
, wbc
);
1683 err
= ext3_journal_stop(handle
);
1690 redirty_page_for_writepage(wbc
, page
);
1696 static int ext3_readpage(struct file
*file
, struct page
*page
)
1698 return mpage_readpage(page
, ext3_get_block
);
1702 ext3_readpages(struct file
*file
, struct address_space
*mapping
,
1703 struct list_head
*pages
, unsigned nr_pages
)
1705 return mpage_readpages(mapping
, pages
, nr_pages
, ext3_get_block
);
1708 static void ext3_invalidatepage(struct page
*page
, unsigned long offset
)
1710 journal_t
*journal
= EXT3_JOURNAL(page
->mapping
->host
);
1713 * If it's a full truncate we just forget about the pending dirtying
1716 ClearPageChecked(page
);
1718 journal_invalidatepage(journal
, page
, offset
);
1721 static int ext3_releasepage(struct page
*page
, gfp_t wait
)
1723 journal_t
*journal
= EXT3_JOURNAL(page
->mapping
->host
);
1725 WARN_ON(PageChecked(page
));
1726 if (!page_has_buffers(page
))
1728 return journal_try_to_free_buffers(journal
, page
, wait
);
1732 * If the O_DIRECT write will extend the file then add this inode to the
1733 * orphan list. So recovery will truncate it back to the original size
1734 * if the machine crashes during the write.
1736 * If the O_DIRECT write is intantiating holes inside i_size and the machine
1737 * crashes then stale disk data _may_ be exposed inside the file. But current
1738 * VFS code falls back into buffered path in that case so we are safe.
1740 static ssize_t
ext3_direct_IO(int rw
, struct kiocb
*iocb
,
1741 const struct iovec
*iov
, loff_t offset
,
1742 unsigned long nr_segs
)
1744 struct file
*file
= iocb
->ki_filp
;
1745 struct inode
*inode
= file
->f_mapping
->host
;
1746 struct ext3_inode_info
*ei
= EXT3_I(inode
);
1750 size_t count
= iov_length(iov
, nr_segs
);
1754 loff_t final_size
= offset
+ count
;
1756 if (final_size
> inode
->i_size
) {
1757 /* Credits for sb + inode write */
1758 handle
= ext3_journal_start(inode
, 2);
1759 if (IS_ERR(handle
)) {
1760 ret
= PTR_ERR(handle
);
1763 ret
= ext3_orphan_add(handle
, inode
);
1765 ext3_journal_stop(handle
);
1769 ei
->i_disksize
= inode
->i_size
;
1770 ext3_journal_stop(handle
);
1775 ret
= blockdev_direct_IO(rw
, iocb
, inode
, inode
->i_sb
->s_bdev
, iov
,
1777 ext3_get_block
, NULL
);
1778 if (ret
== -ENOSPC
&& ext3_should_retry_alloc(inode
->i_sb
, &retries
))
1784 /* Credits for sb + inode write */
1785 handle
= ext3_journal_start(inode
, 2);
1786 if (IS_ERR(handle
)) {
1787 /* This is really bad luck. We've written the data
1788 * but cannot extend i_size. Bail out and pretend
1789 * the write failed... */
1790 ret
= PTR_ERR(handle
);
1794 ext3_orphan_del(handle
, inode
);
1796 loff_t end
= offset
+ ret
;
1797 if (end
> inode
->i_size
) {
1798 ei
->i_disksize
= end
;
1799 i_size_write(inode
, end
);
1801 * We're going to return a positive `ret'
1802 * here due to non-zero-length I/O, so there's
1803 * no way of reporting error returns from
1804 * ext3_mark_inode_dirty() to userspace. So
1807 ext3_mark_inode_dirty(handle
, inode
);
1810 err
= ext3_journal_stop(handle
);
1819 * Pages can be marked dirty completely asynchronously from ext3's journalling
1820 * activity. By filemap_sync_pte(), try_to_unmap_one(), etc. We cannot do
1821 * much here because ->set_page_dirty is called under VFS locks. The page is
1822 * not necessarily locked.
1824 * We cannot just dirty the page and leave attached buffers clean, because the
1825 * buffers' dirty state is "definitive". We cannot just set the buffers dirty
1826 * or jbddirty because all the journalling code will explode.
1828 * So what we do is to mark the page "pending dirty" and next time writepage
1829 * is called, propagate that into the buffers appropriately.
1831 static int ext3_journalled_set_page_dirty(struct page
*page
)
1833 SetPageChecked(page
);
1834 return __set_page_dirty_nobuffers(page
);
1837 static const struct address_space_operations ext3_ordered_aops
= {
1838 .readpage
= ext3_readpage
,
1839 .readpages
= ext3_readpages
,
1840 .writepage
= ext3_ordered_writepage
,
1841 .sync_page
= block_sync_page
,
1842 .write_begin
= ext3_write_begin
,
1843 .write_end
= ext3_ordered_write_end
,
1845 .invalidatepage
= ext3_invalidatepage
,
1846 .releasepage
= ext3_releasepage
,
1847 .direct_IO
= ext3_direct_IO
,
1848 .migratepage
= buffer_migrate_page
,
1849 .is_partially_uptodate
= block_is_partially_uptodate
,
1850 .error_remove_page
= generic_error_remove_page
,
1853 static const struct address_space_operations ext3_writeback_aops
= {
1854 .readpage
= ext3_readpage
,
1855 .readpages
= ext3_readpages
,
1856 .writepage
= ext3_writeback_writepage
,
1857 .sync_page
= block_sync_page
,
1858 .write_begin
= ext3_write_begin
,
1859 .write_end
= ext3_writeback_write_end
,
1861 .invalidatepage
= ext3_invalidatepage
,
1862 .releasepage
= ext3_releasepage
,
1863 .direct_IO
= ext3_direct_IO
,
1864 .migratepage
= buffer_migrate_page
,
1865 .is_partially_uptodate
= block_is_partially_uptodate
,
1866 .error_remove_page
= generic_error_remove_page
,
1869 static const struct address_space_operations ext3_journalled_aops
= {
1870 .readpage
= ext3_readpage
,
1871 .readpages
= ext3_readpages
,
1872 .writepage
= ext3_journalled_writepage
,
1873 .sync_page
= block_sync_page
,
1874 .write_begin
= ext3_write_begin
,
1875 .write_end
= ext3_journalled_write_end
,
1876 .set_page_dirty
= ext3_journalled_set_page_dirty
,
1878 .invalidatepage
= ext3_invalidatepage
,
1879 .releasepage
= ext3_releasepage
,
1880 .is_partially_uptodate
= block_is_partially_uptodate
,
1881 .error_remove_page
= generic_error_remove_page
,
1884 void ext3_set_aops(struct inode
*inode
)
1886 if (ext3_should_order_data(inode
))
1887 inode
->i_mapping
->a_ops
= &ext3_ordered_aops
;
1888 else if (ext3_should_writeback_data(inode
))
1889 inode
->i_mapping
->a_ops
= &ext3_writeback_aops
;
1891 inode
->i_mapping
->a_ops
= &ext3_journalled_aops
;
1895 * ext3_block_truncate_page() zeroes out a mapping from file offset `from'
1896 * up to the end of the block which corresponds to `from'.
1897 * This required during truncate. We need to physically zero the tail end
1898 * of that block so it doesn't yield old data if the file is later grown.
1900 static int ext3_block_truncate_page(handle_t
*handle
, struct page
*page
,
1901 struct address_space
*mapping
, loff_t from
)
1903 ext3_fsblk_t index
= from
>> PAGE_CACHE_SHIFT
;
1904 unsigned offset
= from
& (PAGE_CACHE_SIZE
-1);
1905 unsigned blocksize
, iblock
, length
, pos
;
1906 struct inode
*inode
= mapping
->host
;
1907 struct buffer_head
*bh
;
1910 blocksize
= inode
->i_sb
->s_blocksize
;
1911 length
= blocksize
- (offset
& (blocksize
- 1));
1912 iblock
= index
<< (PAGE_CACHE_SHIFT
- inode
->i_sb
->s_blocksize_bits
);
1915 * For "nobh" option, we can only work if we don't need to
1916 * read-in the page - otherwise we create buffers to do the IO.
1918 if (!page_has_buffers(page
) && test_opt(inode
->i_sb
, NOBH
) &&
1919 ext3_should_writeback_data(inode
) && PageUptodate(page
)) {
1920 zero_user(page
, offset
, length
);
1921 set_page_dirty(page
);
1925 if (!page_has_buffers(page
))
1926 create_empty_buffers(page
, blocksize
, 0);
1928 /* Find the buffer that contains "offset" */
1929 bh
= page_buffers(page
);
1931 while (offset
>= pos
) {
1932 bh
= bh
->b_this_page
;
1938 if (buffer_freed(bh
)) {
1939 BUFFER_TRACE(bh
, "freed: skip");
1943 if (!buffer_mapped(bh
)) {
1944 BUFFER_TRACE(bh
, "unmapped");
1945 ext3_get_block(inode
, iblock
, bh
, 0);
1946 /* unmapped? It's a hole - nothing to do */
1947 if (!buffer_mapped(bh
)) {
1948 BUFFER_TRACE(bh
, "still unmapped");
1953 /* Ok, it's mapped. Make sure it's up-to-date */
1954 if (PageUptodate(page
))
1955 set_buffer_uptodate(bh
);
1957 if (!buffer_uptodate(bh
)) {
1959 ll_rw_block(READ
, 1, &bh
);
1961 /* Uhhuh. Read error. Complain and punt. */
1962 if (!buffer_uptodate(bh
))
1966 if (ext3_should_journal_data(inode
)) {
1967 BUFFER_TRACE(bh
, "get write access");
1968 err
= ext3_journal_get_write_access(handle
, bh
);
1973 zero_user(page
, offset
, length
);
1974 BUFFER_TRACE(bh
, "zeroed end of block");
1977 if (ext3_should_journal_data(inode
)) {
1978 err
= ext3_journal_dirty_metadata(handle
, bh
);
1980 if (ext3_should_order_data(inode
))
1981 err
= ext3_journal_dirty_data(handle
, bh
);
1982 mark_buffer_dirty(bh
);
1987 page_cache_release(page
);
1992 * Probably it should be a library function... search for first non-zero word
1993 * or memcmp with zero_page, whatever is better for particular architecture.
1996 static inline int all_zeroes(__le32
*p
, __le32
*q
)
2005 * ext3_find_shared - find the indirect blocks for partial truncation.
2006 * @inode: inode in question
2007 * @depth: depth of the affected branch
2008 * @offsets: offsets of pointers in that branch (see ext3_block_to_path)
2009 * @chain: place to store the pointers to partial indirect blocks
2010 * @top: place to the (detached) top of branch
2012 * This is a helper function used by ext3_truncate().
2014 * When we do truncate() we may have to clean the ends of several
2015 * indirect blocks but leave the blocks themselves alive. Block is
2016 * partially truncated if some data below the new i_size is refered
2017 * from it (and it is on the path to the first completely truncated
2018 * data block, indeed). We have to free the top of that path along
2019 * with everything to the right of the path. Since no allocation
2020 * past the truncation point is possible until ext3_truncate()
2021 * finishes, we may safely do the latter, but top of branch may
2022 * require special attention - pageout below the truncation point
2023 * might try to populate it.
2025 * We atomically detach the top of branch from the tree, store the
2026 * block number of its root in *@top, pointers to buffer_heads of
2027 * partially truncated blocks - in @chain[].bh and pointers to
2028 * their last elements that should not be removed - in
2029 * @chain[].p. Return value is the pointer to last filled element
2032 * The work left to caller to do the actual freeing of subtrees:
2033 * a) free the subtree starting from *@top
2034 * b) free the subtrees whose roots are stored in
2035 * (@chain[i].p+1 .. end of @chain[i].bh->b_data)
2036 * c) free the subtrees growing from the inode past the @chain[0].
2037 * (no partially truncated stuff there). */
2039 static Indirect
*ext3_find_shared(struct inode
*inode
, int depth
,
2040 int offsets
[4], Indirect chain
[4], __le32
*top
)
2042 Indirect
*partial
, *p
;
2046 /* Make k index the deepest non-null offset + 1 */
2047 for (k
= depth
; k
> 1 && !offsets
[k
-1]; k
--)
2049 partial
= ext3_get_branch(inode
, k
, offsets
, chain
, &err
);
2050 /* Writer: pointers */
2052 partial
= chain
+ k
-1;
2054 * If the branch acquired continuation since we've looked at it -
2055 * fine, it should all survive and (new) top doesn't belong to us.
2057 if (!partial
->key
&& *partial
->p
)
2060 for (p
=partial
; p
>chain
&& all_zeroes((__le32
*)p
->bh
->b_data
,p
->p
); p
--)
2063 * OK, we've found the last block that must survive. The rest of our
2064 * branch should be detached before unlocking. However, if that rest
2065 * of branch is all ours and does not grow immediately from the inode
2066 * it's easier to cheat and just decrement partial->p.
2068 if (p
== chain
+ k
- 1 && p
> chain
) {
2072 /* Nope, don't do this in ext3. Must leave the tree intact */
2079 while(partial
> p
) {
2080 brelse(partial
->bh
);
2088 * Zero a number of block pointers in either an inode or an indirect block.
2089 * If we restart the transaction we must again get write access to the
2090 * indirect block for further modification.
2092 * We release `count' blocks on disk, but (last - first) may be greater
2093 * than `count' because there can be holes in there.
2095 static void ext3_clear_blocks(handle_t
*handle
, struct inode
*inode
,
2096 struct buffer_head
*bh
, ext3_fsblk_t block_to_free
,
2097 unsigned long count
, __le32
*first
, __le32
*last
)
2100 if (try_to_extend_transaction(handle
, inode
)) {
2102 BUFFER_TRACE(bh
, "call ext3_journal_dirty_metadata");
2103 ext3_journal_dirty_metadata(handle
, bh
);
2105 ext3_mark_inode_dirty(handle
, inode
);
2106 truncate_restart_transaction(handle
, inode
);
2108 BUFFER_TRACE(bh
, "retaking write access");
2109 ext3_journal_get_write_access(handle
, bh
);
2114 * Any buffers which are on the journal will be in memory. We find
2115 * them on the hash table so journal_revoke() will run journal_forget()
2116 * on them. We've already detached each block from the file, so
2117 * bforget() in journal_forget() should be safe.
2119 * AKPM: turn on bforget in journal_forget()!!!
2121 for (p
= first
; p
< last
; p
++) {
2122 u32 nr
= le32_to_cpu(*p
);
2124 struct buffer_head
*bh
;
2127 bh
= sb_find_get_block(inode
->i_sb
, nr
);
2128 ext3_forget(handle
, 0, inode
, bh
, nr
);
2132 ext3_free_blocks(handle
, inode
, block_to_free
, count
);
2136 * ext3_free_data - free a list of data blocks
2137 * @handle: handle for this transaction
2138 * @inode: inode we are dealing with
2139 * @this_bh: indirect buffer_head which contains *@first and *@last
2140 * @first: array of block numbers
2141 * @last: points immediately past the end of array
2143 * We are freeing all blocks refered from that array (numbers are stored as
2144 * little-endian 32-bit) and updating @inode->i_blocks appropriately.
2146 * We accumulate contiguous runs of blocks to free. Conveniently, if these
2147 * blocks are contiguous then releasing them at one time will only affect one
2148 * or two bitmap blocks (+ group descriptor(s) and superblock) and we won't
2149 * actually use a lot of journal space.
2151 * @this_bh will be %NULL if @first and @last point into the inode's direct
2154 static void ext3_free_data(handle_t
*handle
, struct inode
*inode
,
2155 struct buffer_head
*this_bh
,
2156 __le32
*first
, __le32
*last
)
2158 ext3_fsblk_t block_to_free
= 0; /* Starting block # of a run */
2159 unsigned long count
= 0; /* Number of blocks in the run */
2160 __le32
*block_to_free_p
= NULL
; /* Pointer into inode/ind
2163 ext3_fsblk_t nr
; /* Current block # */
2164 __le32
*p
; /* Pointer into inode/ind
2165 for current block */
2168 if (this_bh
) { /* For indirect block */
2169 BUFFER_TRACE(this_bh
, "get_write_access");
2170 err
= ext3_journal_get_write_access(handle
, this_bh
);
2171 /* Important: if we can't update the indirect pointers
2172 * to the blocks, we can't free them. */
2177 for (p
= first
; p
< last
; p
++) {
2178 nr
= le32_to_cpu(*p
);
2180 /* accumulate blocks to free if they're contiguous */
2183 block_to_free_p
= p
;
2185 } else if (nr
== block_to_free
+ count
) {
2188 ext3_clear_blocks(handle
, inode
, this_bh
,
2190 count
, block_to_free_p
, p
);
2192 block_to_free_p
= p
;
2199 ext3_clear_blocks(handle
, inode
, this_bh
, block_to_free
,
2200 count
, block_to_free_p
, p
);
2203 BUFFER_TRACE(this_bh
, "call ext3_journal_dirty_metadata");
2206 * The buffer head should have an attached journal head at this
2207 * point. However, if the data is corrupted and an indirect
2208 * block pointed to itself, it would have been detached when
2209 * the block was cleared. Check for this instead of OOPSing.
2212 ext3_journal_dirty_metadata(handle
, this_bh
);
2214 ext3_error(inode
->i_sb
, "ext3_free_data",
2215 "circular indirect block detected, "
2216 "inode=%lu, block=%llu",
2218 (unsigned long long)this_bh
->b_blocknr
);
2223 * ext3_free_branches - free an array of branches
2224 * @handle: JBD handle for this transaction
2225 * @inode: inode we are dealing with
2226 * @parent_bh: the buffer_head which contains *@first and *@last
2227 * @first: array of block numbers
2228 * @last: pointer immediately past the end of array
2229 * @depth: depth of the branches to free
2231 * We are freeing all blocks refered from these branches (numbers are
2232 * stored as little-endian 32-bit) and updating @inode->i_blocks
2235 static void ext3_free_branches(handle_t
*handle
, struct inode
*inode
,
2236 struct buffer_head
*parent_bh
,
2237 __le32
*first
, __le32
*last
, int depth
)
2242 if (is_handle_aborted(handle
))
2246 struct buffer_head
*bh
;
2247 int addr_per_block
= EXT3_ADDR_PER_BLOCK(inode
->i_sb
);
2249 while (--p
>= first
) {
2250 nr
= le32_to_cpu(*p
);
2252 continue; /* A hole */
2254 /* Go read the buffer for the next level down */
2255 bh
= sb_bread(inode
->i_sb
, nr
);
2258 * A read failure? Report error and clear slot
2262 ext3_error(inode
->i_sb
, "ext3_free_branches",
2263 "Read failure, inode=%lu, block="E3FSBLK
,
2268 /* This zaps the entire block. Bottom up. */
2269 BUFFER_TRACE(bh
, "free child branches");
2270 ext3_free_branches(handle
, inode
, bh
,
2271 (__le32
*)bh
->b_data
,
2272 (__le32
*)bh
->b_data
+ addr_per_block
,
2276 * We've probably journalled the indirect block several
2277 * times during the truncate. But it's no longer
2278 * needed and we now drop it from the transaction via
2281 * That's easy if it's exclusively part of this
2282 * transaction. But if it's part of the committing
2283 * transaction then journal_forget() will simply
2284 * brelse() it. That means that if the underlying
2285 * block is reallocated in ext3_get_block(),
2286 * unmap_underlying_metadata() will find this block
2287 * and will try to get rid of it. damn, damn.
2289 * If this block has already been committed to the
2290 * journal, a revoke record will be written. And
2291 * revoke records must be emitted *before* clearing
2292 * this block's bit in the bitmaps.
2294 ext3_forget(handle
, 1, inode
, bh
, bh
->b_blocknr
);
2297 * Everything below this this pointer has been
2298 * released. Now let this top-of-subtree go.
2300 * We want the freeing of this indirect block to be
2301 * atomic in the journal with the updating of the
2302 * bitmap block which owns it. So make some room in
2305 * We zero the parent pointer *after* freeing its
2306 * pointee in the bitmaps, so if extend_transaction()
2307 * for some reason fails to put the bitmap changes and
2308 * the release into the same transaction, recovery
2309 * will merely complain about releasing a free block,
2310 * rather than leaking blocks.
2312 if (is_handle_aborted(handle
))
2314 if (try_to_extend_transaction(handle
, inode
)) {
2315 ext3_mark_inode_dirty(handle
, inode
);
2316 truncate_restart_transaction(handle
, inode
);
2319 ext3_free_blocks(handle
, inode
, nr
, 1);
2323 * The block which we have just freed is
2324 * pointed to by an indirect block: journal it
2326 BUFFER_TRACE(parent_bh
, "get_write_access");
2327 if (!ext3_journal_get_write_access(handle
,
2330 BUFFER_TRACE(parent_bh
,
2331 "call ext3_journal_dirty_metadata");
2332 ext3_journal_dirty_metadata(handle
,
2338 /* We have reached the bottom of the tree. */
2339 BUFFER_TRACE(parent_bh
, "free data blocks");
2340 ext3_free_data(handle
, inode
, parent_bh
, first
, last
);
2344 int ext3_can_truncate(struct inode
*inode
)
2346 if (IS_APPEND(inode
) || IS_IMMUTABLE(inode
))
2348 if (S_ISREG(inode
->i_mode
))
2350 if (S_ISDIR(inode
->i_mode
))
2352 if (S_ISLNK(inode
->i_mode
))
2353 return !ext3_inode_is_fast_symlink(inode
);
2360 * We block out ext3_get_block() block instantiations across the entire
2361 * transaction, and VFS/VM ensures that ext3_truncate() cannot run
2362 * simultaneously on behalf of the same inode.
2364 * As we work through the truncate and commmit bits of it to the journal there
2365 * is one core, guiding principle: the file's tree must always be consistent on
2366 * disk. We must be able to restart the truncate after a crash.
2368 * The file's tree may be transiently inconsistent in memory (although it
2369 * probably isn't), but whenever we close off and commit a journal transaction,
2370 * the contents of (the filesystem + the journal) must be consistent and
2371 * restartable. It's pretty simple, really: bottom up, right to left (although
2372 * left-to-right works OK too).
2374 * Note that at recovery time, journal replay occurs *before* the restart of
2375 * truncate against the orphan inode list.
2377 * The committed inode has the new, desired i_size (which is the same as
2378 * i_disksize in this case). After a crash, ext3_orphan_cleanup() will see
2379 * that this inode's truncate did not complete and it will again call
2380 * ext3_truncate() to have another go. So there will be instantiated blocks
2381 * to the right of the truncation point in a crashed ext3 filesystem. But
2382 * that's fine - as long as they are linked from the inode, the post-crash
2383 * ext3_truncate() run will find them and release them.
2385 void ext3_truncate(struct inode
*inode
)
2388 struct ext3_inode_info
*ei
= EXT3_I(inode
);
2389 __le32
*i_data
= ei
->i_data
;
2390 int addr_per_block
= EXT3_ADDR_PER_BLOCK(inode
->i_sb
);
2391 struct address_space
*mapping
= inode
->i_mapping
;
2398 unsigned blocksize
= inode
->i_sb
->s_blocksize
;
2401 if (!ext3_can_truncate(inode
))
2404 if (inode
->i_size
== 0 && ext3_should_writeback_data(inode
))
2405 ei
->i_state
|= EXT3_STATE_FLUSH_ON_CLOSE
;
2408 * We have to lock the EOF page here, because lock_page() nests
2409 * outside journal_start().
2411 if ((inode
->i_size
& (blocksize
- 1)) == 0) {
2412 /* Block boundary? Nothing to do */
2415 page
= grab_cache_page(mapping
,
2416 inode
->i_size
>> PAGE_CACHE_SHIFT
);
2421 handle
= start_transaction(inode
);
2422 if (IS_ERR(handle
)) {
2424 clear_highpage(page
);
2425 flush_dcache_page(page
);
2427 page_cache_release(page
);
2432 last_block
= (inode
->i_size
+ blocksize
-1)
2433 >> EXT3_BLOCK_SIZE_BITS(inode
->i_sb
);
2436 ext3_block_truncate_page(handle
, page
, mapping
, inode
->i_size
);
2438 n
= ext3_block_to_path(inode
, last_block
, offsets
, NULL
);
2440 goto out_stop
; /* error */
2443 * OK. This truncate is going to happen. We add the inode to the
2444 * orphan list, so that if this truncate spans multiple transactions,
2445 * and we crash, we will resume the truncate when the filesystem
2446 * recovers. It also marks the inode dirty, to catch the new size.
2448 * Implication: the file must always be in a sane, consistent
2449 * truncatable state while each transaction commits.
2451 if (ext3_orphan_add(handle
, inode
))
2455 * The orphan list entry will now protect us from any crash which
2456 * occurs before the truncate completes, so it is now safe to propagate
2457 * the new, shorter inode size (held for now in i_size) into the
2458 * on-disk inode. We do this via i_disksize, which is the value which
2459 * ext3 *really* writes onto the disk inode.
2461 ei
->i_disksize
= inode
->i_size
;
2464 * From here we block out all ext3_get_block() callers who want to
2465 * modify the block allocation tree.
2467 mutex_lock(&ei
->truncate_mutex
);
2469 if (n
== 1) { /* direct blocks */
2470 ext3_free_data(handle
, inode
, NULL
, i_data
+offsets
[0],
2471 i_data
+ EXT3_NDIR_BLOCKS
);
2475 partial
= ext3_find_shared(inode
, n
, offsets
, chain
, &nr
);
2476 /* Kill the top of shared branch (not detached) */
2478 if (partial
== chain
) {
2479 /* Shared branch grows from the inode */
2480 ext3_free_branches(handle
, inode
, NULL
,
2481 &nr
, &nr
+1, (chain
+n
-1) - partial
);
2484 * We mark the inode dirty prior to restart,
2485 * and prior to stop. No need for it here.
2488 /* Shared branch grows from an indirect block */
2489 BUFFER_TRACE(partial
->bh
, "get_write_access");
2490 ext3_free_branches(handle
, inode
, partial
->bh
,
2492 partial
->p
+1, (chain
+n
-1) - partial
);
2495 /* Clear the ends of indirect blocks on the shared branch */
2496 while (partial
> chain
) {
2497 ext3_free_branches(handle
, inode
, partial
->bh
, partial
->p
+ 1,
2498 (__le32
*)partial
->bh
->b_data
+addr_per_block
,
2499 (chain
+n
-1) - partial
);
2500 BUFFER_TRACE(partial
->bh
, "call brelse");
2501 brelse (partial
->bh
);
2505 /* Kill the remaining (whole) subtrees */
2506 switch (offsets
[0]) {
2508 nr
= i_data
[EXT3_IND_BLOCK
];
2510 ext3_free_branches(handle
, inode
, NULL
, &nr
, &nr
+1, 1);
2511 i_data
[EXT3_IND_BLOCK
] = 0;
2513 case EXT3_IND_BLOCK
:
2514 nr
= i_data
[EXT3_DIND_BLOCK
];
2516 ext3_free_branches(handle
, inode
, NULL
, &nr
, &nr
+1, 2);
2517 i_data
[EXT3_DIND_BLOCK
] = 0;
2519 case EXT3_DIND_BLOCK
:
2520 nr
= i_data
[EXT3_TIND_BLOCK
];
2522 ext3_free_branches(handle
, inode
, NULL
, &nr
, &nr
+1, 3);
2523 i_data
[EXT3_TIND_BLOCK
] = 0;
2525 case EXT3_TIND_BLOCK
:
2529 ext3_discard_reservation(inode
);
2531 mutex_unlock(&ei
->truncate_mutex
);
2532 inode
->i_mtime
= inode
->i_ctime
= CURRENT_TIME_SEC
;
2533 ext3_mark_inode_dirty(handle
, inode
);
2536 * In a multi-transaction truncate, we only make the final transaction
2543 * If this was a simple ftruncate(), and the file will remain alive
2544 * then we need to clear up the orphan record which we created above.
2545 * However, if this was a real unlink then we were called by
2546 * ext3_delete_inode(), and we allow that function to clean up the
2547 * orphan info for us.
2550 ext3_orphan_del(handle
, inode
);
2552 ext3_journal_stop(handle
);
2556 * Delete the inode from orphan list so that it doesn't stay there
2557 * forever and trigger assertion on umount.
2560 ext3_orphan_del(NULL
, inode
);
2563 static ext3_fsblk_t
ext3_get_inode_block(struct super_block
*sb
,
2564 unsigned long ino
, struct ext3_iloc
*iloc
)
2566 unsigned long block_group
;
2567 unsigned long offset
;
2569 struct ext3_group_desc
*gdp
;
2571 if (!ext3_valid_inum(sb
, ino
)) {
2573 * This error is already checked for in namei.c unless we are
2574 * looking at an NFS filehandle, in which case no error
2580 block_group
= (ino
- 1) / EXT3_INODES_PER_GROUP(sb
);
2581 gdp
= ext3_get_group_desc(sb
, block_group
, NULL
);
2585 * Figure out the offset within the block group inode table
2587 offset
= ((ino
- 1) % EXT3_INODES_PER_GROUP(sb
)) *
2588 EXT3_INODE_SIZE(sb
);
2589 block
= le32_to_cpu(gdp
->bg_inode_table
) +
2590 (offset
>> EXT3_BLOCK_SIZE_BITS(sb
));
2592 iloc
->block_group
= block_group
;
2593 iloc
->offset
= offset
& (EXT3_BLOCK_SIZE(sb
) - 1);
2598 * ext3_get_inode_loc returns with an extra refcount against the inode's
2599 * underlying buffer_head on success. If 'in_mem' is true, we have all
2600 * data in memory that is needed to recreate the on-disk version of this
2603 static int __ext3_get_inode_loc(struct inode
*inode
,
2604 struct ext3_iloc
*iloc
, int in_mem
)
2607 struct buffer_head
*bh
;
2609 block
= ext3_get_inode_block(inode
->i_sb
, inode
->i_ino
, iloc
);
2613 bh
= sb_getblk(inode
->i_sb
, block
);
2615 ext3_error (inode
->i_sb
, "ext3_get_inode_loc",
2616 "unable to read inode block - "
2617 "inode=%lu, block="E3FSBLK
,
2618 inode
->i_ino
, block
);
2621 if (!buffer_uptodate(bh
)) {
2625 * If the buffer has the write error flag, we have failed
2626 * to write out another inode in the same block. In this
2627 * case, we don't have to read the block because we may
2628 * read the old inode data successfully.
2630 if (buffer_write_io_error(bh
) && !buffer_uptodate(bh
))
2631 set_buffer_uptodate(bh
);
2633 if (buffer_uptodate(bh
)) {
2634 /* someone brought it uptodate while we waited */
2640 * If we have all information of the inode in memory and this
2641 * is the only valid inode in the block, we need not read the
2645 struct buffer_head
*bitmap_bh
;
2646 struct ext3_group_desc
*desc
;
2647 int inodes_per_buffer
;
2648 int inode_offset
, i
;
2652 block_group
= (inode
->i_ino
- 1) /
2653 EXT3_INODES_PER_GROUP(inode
->i_sb
);
2654 inodes_per_buffer
= bh
->b_size
/
2655 EXT3_INODE_SIZE(inode
->i_sb
);
2656 inode_offset
= ((inode
->i_ino
- 1) %
2657 EXT3_INODES_PER_GROUP(inode
->i_sb
));
2658 start
= inode_offset
& ~(inodes_per_buffer
- 1);
2660 /* Is the inode bitmap in cache? */
2661 desc
= ext3_get_group_desc(inode
->i_sb
,
2666 bitmap_bh
= sb_getblk(inode
->i_sb
,
2667 le32_to_cpu(desc
->bg_inode_bitmap
));
2672 * If the inode bitmap isn't in cache then the
2673 * optimisation may end up performing two reads instead
2674 * of one, so skip it.
2676 if (!buffer_uptodate(bitmap_bh
)) {
2680 for (i
= start
; i
< start
+ inodes_per_buffer
; i
++) {
2681 if (i
== inode_offset
)
2683 if (ext3_test_bit(i
, bitmap_bh
->b_data
))
2687 if (i
== start
+ inodes_per_buffer
) {
2688 /* all other inodes are free, so skip I/O */
2689 memset(bh
->b_data
, 0, bh
->b_size
);
2690 set_buffer_uptodate(bh
);
2698 * There are other valid inodes in the buffer, this inode
2699 * has in-inode xattrs, or we don't have this inode in memory.
2700 * Read the block from disk.
2703 bh
->b_end_io
= end_buffer_read_sync
;
2704 submit_bh(READ_META
, bh
);
2706 if (!buffer_uptodate(bh
)) {
2707 ext3_error(inode
->i_sb
, "ext3_get_inode_loc",
2708 "unable to read inode block - "
2709 "inode=%lu, block="E3FSBLK
,
2710 inode
->i_ino
, block
);
2720 int ext3_get_inode_loc(struct inode
*inode
, struct ext3_iloc
*iloc
)
2722 /* We have all inode data except xattrs in memory here. */
2723 return __ext3_get_inode_loc(inode
, iloc
,
2724 !(EXT3_I(inode
)->i_state
& EXT3_STATE_XATTR
));
2727 void ext3_set_inode_flags(struct inode
*inode
)
2729 unsigned int flags
= EXT3_I(inode
)->i_flags
;
2731 inode
->i_flags
&= ~(S_SYNC
|S_APPEND
|S_IMMUTABLE
|S_NOATIME
|S_DIRSYNC
);
2732 if (flags
& EXT3_SYNC_FL
)
2733 inode
->i_flags
|= S_SYNC
;
2734 if (flags
& EXT3_APPEND_FL
)
2735 inode
->i_flags
|= S_APPEND
;
2736 if (flags
& EXT3_IMMUTABLE_FL
)
2737 inode
->i_flags
|= S_IMMUTABLE
;
2738 if (flags
& EXT3_NOATIME_FL
)
2739 inode
->i_flags
|= S_NOATIME
;
2740 if (flags
& EXT3_DIRSYNC_FL
)
2741 inode
->i_flags
|= S_DIRSYNC
;
2744 /* Propagate flags from i_flags to EXT3_I(inode)->i_flags */
2745 void ext3_get_inode_flags(struct ext3_inode_info
*ei
)
2747 unsigned int flags
= ei
->vfs_inode
.i_flags
;
2749 ei
->i_flags
&= ~(EXT3_SYNC_FL
|EXT3_APPEND_FL
|
2750 EXT3_IMMUTABLE_FL
|EXT3_NOATIME_FL
|EXT3_DIRSYNC_FL
);
2752 ei
->i_flags
|= EXT3_SYNC_FL
;
2753 if (flags
& S_APPEND
)
2754 ei
->i_flags
|= EXT3_APPEND_FL
;
2755 if (flags
& S_IMMUTABLE
)
2756 ei
->i_flags
|= EXT3_IMMUTABLE_FL
;
2757 if (flags
& S_NOATIME
)
2758 ei
->i_flags
|= EXT3_NOATIME_FL
;
2759 if (flags
& S_DIRSYNC
)
2760 ei
->i_flags
|= EXT3_DIRSYNC_FL
;
2763 struct inode
*ext3_iget(struct super_block
*sb
, unsigned long ino
)
2765 struct ext3_iloc iloc
;
2766 struct ext3_inode
*raw_inode
;
2767 struct ext3_inode_info
*ei
;
2768 struct buffer_head
*bh
;
2769 struct inode
*inode
;
2770 journal_t
*journal
= EXT3_SB(sb
)->s_journal
;
2771 transaction_t
*transaction
;
2775 inode
= iget_locked(sb
, ino
);
2777 return ERR_PTR(-ENOMEM
);
2778 if (!(inode
->i_state
& I_NEW
))
2782 ei
->i_block_alloc_info
= NULL
;
2784 ret
= __ext3_get_inode_loc(inode
, &iloc
, 0);
2788 raw_inode
= ext3_raw_inode(&iloc
);
2789 inode
->i_mode
= le16_to_cpu(raw_inode
->i_mode
);
2790 inode
->i_uid
= (uid_t
)le16_to_cpu(raw_inode
->i_uid_low
);
2791 inode
->i_gid
= (gid_t
)le16_to_cpu(raw_inode
->i_gid_low
);
2792 if(!(test_opt (inode
->i_sb
, NO_UID32
))) {
2793 inode
->i_uid
|= le16_to_cpu(raw_inode
->i_uid_high
) << 16;
2794 inode
->i_gid
|= le16_to_cpu(raw_inode
->i_gid_high
) << 16;
2796 inode
->i_nlink
= le16_to_cpu(raw_inode
->i_links_count
);
2797 inode
->i_size
= le32_to_cpu(raw_inode
->i_size
);
2798 inode
->i_atime
.tv_sec
= (signed)le32_to_cpu(raw_inode
->i_atime
);
2799 inode
->i_ctime
.tv_sec
= (signed)le32_to_cpu(raw_inode
->i_ctime
);
2800 inode
->i_mtime
.tv_sec
= (signed)le32_to_cpu(raw_inode
->i_mtime
);
2801 inode
->i_atime
.tv_nsec
= inode
->i_ctime
.tv_nsec
= inode
->i_mtime
.tv_nsec
= 0;
2804 ei
->i_dir_start_lookup
= 0;
2805 ei
->i_dtime
= le32_to_cpu(raw_inode
->i_dtime
);
2806 /* We now have enough fields to check if the inode was active or not.
2807 * This is needed because nfsd might try to access dead inodes
2808 * the test is that same one that e2fsck uses
2809 * NeilBrown 1999oct15
2811 if (inode
->i_nlink
== 0) {
2812 if (inode
->i_mode
== 0 ||
2813 !(EXT3_SB(inode
->i_sb
)->s_mount_state
& EXT3_ORPHAN_FS
)) {
2814 /* this inode is deleted */
2819 /* The only unlinked inodes we let through here have
2820 * valid i_mode and are being read by the orphan
2821 * recovery code: that's fine, we're about to complete
2822 * the process of deleting those. */
2824 inode
->i_blocks
= le32_to_cpu(raw_inode
->i_blocks
);
2825 ei
->i_flags
= le32_to_cpu(raw_inode
->i_flags
);
2826 #ifdef EXT3_FRAGMENTS
2827 ei
->i_faddr
= le32_to_cpu(raw_inode
->i_faddr
);
2828 ei
->i_frag_no
= raw_inode
->i_frag
;
2829 ei
->i_frag_size
= raw_inode
->i_fsize
;
2831 ei
->i_file_acl
= le32_to_cpu(raw_inode
->i_file_acl
);
2832 if (!S_ISREG(inode
->i_mode
)) {
2833 ei
->i_dir_acl
= le32_to_cpu(raw_inode
->i_dir_acl
);
2836 ((__u64
)le32_to_cpu(raw_inode
->i_size_high
)) << 32;
2838 ei
->i_disksize
= inode
->i_size
;
2839 inode
->i_generation
= le32_to_cpu(raw_inode
->i_generation
);
2840 ei
->i_block_group
= iloc
.block_group
;
2842 * NOTE! The in-memory inode i_data array is in little-endian order
2843 * even on big-endian machines: we do NOT byteswap the block numbers!
2845 for (block
= 0; block
< EXT3_N_BLOCKS
; block
++)
2846 ei
->i_data
[block
] = raw_inode
->i_block
[block
];
2847 INIT_LIST_HEAD(&ei
->i_orphan
);
2850 * Set transaction id's of transactions that have to be committed
2851 * to finish f[data]sync. We set them to currently running transaction
2852 * as we cannot be sure that the inode or some of its metadata isn't
2853 * part of the transaction - the inode could have been reclaimed and
2854 * now it is reread from disk.
2859 spin_lock(&journal
->j_state_lock
);
2860 if (journal
->j_running_transaction
)
2861 transaction
= journal
->j_running_transaction
;
2863 transaction
= journal
->j_committing_transaction
;
2865 tid
= transaction
->t_tid
;
2867 tid
= journal
->j_commit_sequence
;
2868 spin_unlock(&journal
->j_state_lock
);
2869 atomic_set(&ei
->i_sync_tid
, tid
);
2870 atomic_set(&ei
->i_datasync_tid
, tid
);
2873 if (inode
->i_ino
>= EXT3_FIRST_INO(inode
->i_sb
) + 1 &&
2874 EXT3_INODE_SIZE(inode
->i_sb
) > EXT3_GOOD_OLD_INODE_SIZE
) {
2876 * When mke2fs creates big inodes it does not zero out
2877 * the unused bytes above EXT3_GOOD_OLD_INODE_SIZE,
2878 * so ignore those first few inodes.
2880 ei
->i_extra_isize
= le16_to_cpu(raw_inode
->i_extra_isize
);
2881 if (EXT3_GOOD_OLD_INODE_SIZE
+ ei
->i_extra_isize
>
2882 EXT3_INODE_SIZE(inode
->i_sb
)) {
2887 if (ei
->i_extra_isize
== 0) {
2888 /* The extra space is currently unused. Use it. */
2889 ei
->i_extra_isize
= sizeof(struct ext3_inode
) -
2890 EXT3_GOOD_OLD_INODE_SIZE
;
2892 __le32
*magic
= (void *)raw_inode
+
2893 EXT3_GOOD_OLD_INODE_SIZE
+
2895 if (*magic
== cpu_to_le32(EXT3_XATTR_MAGIC
))
2896 ei
->i_state
|= EXT3_STATE_XATTR
;
2899 ei
->i_extra_isize
= 0;
2901 if (S_ISREG(inode
->i_mode
)) {
2902 inode
->i_op
= &ext3_file_inode_operations
;
2903 inode
->i_fop
= &ext3_file_operations
;
2904 ext3_set_aops(inode
);
2905 } else if (S_ISDIR(inode
->i_mode
)) {
2906 inode
->i_op
= &ext3_dir_inode_operations
;
2907 inode
->i_fop
= &ext3_dir_operations
;
2908 } else if (S_ISLNK(inode
->i_mode
)) {
2909 if (ext3_inode_is_fast_symlink(inode
)) {
2910 inode
->i_op
= &ext3_fast_symlink_inode_operations
;
2911 nd_terminate_link(ei
->i_data
, inode
->i_size
,
2912 sizeof(ei
->i_data
) - 1);
2914 inode
->i_op
= &ext3_symlink_inode_operations
;
2915 ext3_set_aops(inode
);
2918 inode
->i_op
= &ext3_special_inode_operations
;
2919 if (raw_inode
->i_block
[0])
2920 init_special_inode(inode
, inode
->i_mode
,
2921 old_decode_dev(le32_to_cpu(raw_inode
->i_block
[0])));
2923 init_special_inode(inode
, inode
->i_mode
,
2924 new_decode_dev(le32_to_cpu(raw_inode
->i_block
[1])));
2927 ext3_set_inode_flags(inode
);
2928 unlock_new_inode(inode
);
2933 return ERR_PTR(ret
);
2937 * Post the struct inode info into an on-disk inode location in the
2938 * buffer-cache. This gobbles the caller's reference to the
2939 * buffer_head in the inode location struct.
2941 * The caller must have write access to iloc->bh.
2943 static int ext3_do_update_inode(handle_t
*handle
,
2944 struct inode
*inode
,
2945 struct ext3_iloc
*iloc
)
2947 struct ext3_inode
*raw_inode
= ext3_raw_inode(iloc
);
2948 struct ext3_inode_info
*ei
= EXT3_I(inode
);
2949 struct buffer_head
*bh
= iloc
->bh
;
2950 int err
= 0, rc
, block
;
2953 /* we can't allow multiple procs in here at once, its a bit racey */
2956 /* For fields not not tracking in the in-memory inode,
2957 * initialise them to zero for new inodes. */
2958 if (ei
->i_state
& EXT3_STATE_NEW
)
2959 memset(raw_inode
, 0, EXT3_SB(inode
->i_sb
)->s_inode_size
);
2961 ext3_get_inode_flags(ei
);
2962 raw_inode
->i_mode
= cpu_to_le16(inode
->i_mode
);
2963 if(!(test_opt(inode
->i_sb
, NO_UID32
))) {
2964 raw_inode
->i_uid_low
= cpu_to_le16(low_16_bits(inode
->i_uid
));
2965 raw_inode
->i_gid_low
= cpu_to_le16(low_16_bits(inode
->i_gid
));
2967 * Fix up interoperability with old kernels. Otherwise, old inodes get
2968 * re-used with the upper 16 bits of the uid/gid intact
2971 raw_inode
->i_uid_high
=
2972 cpu_to_le16(high_16_bits(inode
->i_uid
));
2973 raw_inode
->i_gid_high
=
2974 cpu_to_le16(high_16_bits(inode
->i_gid
));
2976 raw_inode
->i_uid_high
= 0;
2977 raw_inode
->i_gid_high
= 0;
2980 raw_inode
->i_uid_low
=
2981 cpu_to_le16(fs_high2lowuid(inode
->i_uid
));
2982 raw_inode
->i_gid_low
=
2983 cpu_to_le16(fs_high2lowgid(inode
->i_gid
));
2984 raw_inode
->i_uid_high
= 0;
2985 raw_inode
->i_gid_high
= 0;
2987 raw_inode
->i_links_count
= cpu_to_le16(inode
->i_nlink
);
2988 raw_inode
->i_size
= cpu_to_le32(ei
->i_disksize
);
2989 raw_inode
->i_atime
= cpu_to_le32(inode
->i_atime
.tv_sec
);
2990 raw_inode
->i_ctime
= cpu_to_le32(inode
->i_ctime
.tv_sec
);
2991 raw_inode
->i_mtime
= cpu_to_le32(inode
->i_mtime
.tv_sec
);
2992 raw_inode
->i_blocks
= cpu_to_le32(inode
->i_blocks
);
2993 raw_inode
->i_dtime
= cpu_to_le32(ei
->i_dtime
);
2994 raw_inode
->i_flags
= cpu_to_le32(ei
->i_flags
);
2995 #ifdef EXT3_FRAGMENTS
2996 raw_inode
->i_faddr
= cpu_to_le32(ei
->i_faddr
);
2997 raw_inode
->i_frag
= ei
->i_frag_no
;
2998 raw_inode
->i_fsize
= ei
->i_frag_size
;
3000 raw_inode
->i_file_acl
= cpu_to_le32(ei
->i_file_acl
);
3001 if (!S_ISREG(inode
->i_mode
)) {
3002 raw_inode
->i_dir_acl
= cpu_to_le32(ei
->i_dir_acl
);
3004 raw_inode
->i_size_high
=
3005 cpu_to_le32(ei
->i_disksize
>> 32);
3006 if (ei
->i_disksize
> 0x7fffffffULL
) {
3007 struct super_block
*sb
= inode
->i_sb
;
3008 if (!EXT3_HAS_RO_COMPAT_FEATURE(sb
,
3009 EXT3_FEATURE_RO_COMPAT_LARGE_FILE
) ||
3010 EXT3_SB(sb
)->s_es
->s_rev_level
==
3011 cpu_to_le32(EXT3_GOOD_OLD_REV
)) {
3012 /* If this is the first large file
3013 * created, add a flag to the superblock.
3016 err
= ext3_journal_get_write_access(handle
,
3017 EXT3_SB(sb
)->s_sbh
);
3021 ext3_update_dynamic_rev(sb
);
3022 EXT3_SET_RO_COMPAT_FEATURE(sb
,
3023 EXT3_FEATURE_RO_COMPAT_LARGE_FILE
);
3025 err
= ext3_journal_dirty_metadata(handle
,
3026 EXT3_SB(sb
)->s_sbh
);
3027 /* get our lock and start over */
3032 raw_inode
->i_generation
= cpu_to_le32(inode
->i_generation
);
3033 if (S_ISCHR(inode
->i_mode
) || S_ISBLK(inode
->i_mode
)) {
3034 if (old_valid_dev(inode
->i_rdev
)) {
3035 raw_inode
->i_block
[0] =
3036 cpu_to_le32(old_encode_dev(inode
->i_rdev
));
3037 raw_inode
->i_block
[1] = 0;
3039 raw_inode
->i_block
[0] = 0;
3040 raw_inode
->i_block
[1] =
3041 cpu_to_le32(new_encode_dev(inode
->i_rdev
));
3042 raw_inode
->i_block
[2] = 0;
3044 } else for (block
= 0; block
< EXT3_N_BLOCKS
; block
++)
3045 raw_inode
->i_block
[block
] = ei
->i_data
[block
];
3047 if (ei
->i_extra_isize
)
3048 raw_inode
->i_extra_isize
= cpu_to_le16(ei
->i_extra_isize
);
3050 BUFFER_TRACE(bh
, "call ext3_journal_dirty_metadata");
3052 rc
= ext3_journal_dirty_metadata(handle
, bh
);
3055 ei
->i_state
&= ~EXT3_STATE_NEW
;
3057 atomic_set(&ei
->i_sync_tid
, handle
->h_transaction
->t_tid
);
3060 ext3_std_error(inode
->i_sb
, err
);
3065 * ext3_write_inode()
3067 * We are called from a few places:
3069 * - Within generic_file_write() for O_SYNC files.
3070 * Here, there will be no transaction running. We wait for any running
3071 * trasnaction to commit.
3073 * - Within sys_sync(), kupdate and such.
3074 * We wait on commit, if tol to.
3076 * - Within prune_icache() (PF_MEMALLOC == true)
3077 * Here we simply return. We can't afford to block kswapd on the
3080 * In all cases it is actually safe for us to return without doing anything,
3081 * because the inode has been copied into a raw inode buffer in
3082 * ext3_mark_inode_dirty(). This is a correctness thing for O_SYNC and for
3085 * Note that we are absolutely dependent upon all inode dirtiers doing the
3086 * right thing: they *must* call mark_inode_dirty() after dirtying info in
3087 * which we are interested.
3089 * It would be a bug for them to not do this. The code:
3091 * mark_inode_dirty(inode)
3093 * inode->i_size = expr;
3095 * is in error because a kswapd-driven write_inode() could occur while
3096 * `stuff()' is running, and the new i_size will be lost. Plus the inode
3097 * will no longer be on the superblock's dirty inode list.
3099 int ext3_write_inode(struct inode
*inode
, int wait
)
3101 if (current
->flags
& PF_MEMALLOC
)
3104 if (ext3_journal_current_handle()) {
3105 jbd_debug(1, "called recursively, non-PF_MEMALLOC!\n");
3113 return ext3_force_commit(inode
->i_sb
);
3119 * Called from notify_change.
3121 * We want to trap VFS attempts to truncate the file as soon as
3122 * possible. In particular, we want to make sure that when the VFS
3123 * shrinks i_size, we put the inode on the orphan list and modify
3124 * i_disksize immediately, so that during the subsequent flushing of
3125 * dirty pages and freeing of disk blocks, we can guarantee that any
3126 * commit will leave the blocks being flushed in an unused state on
3127 * disk. (On recovery, the inode will get truncated and the blocks will
3128 * be freed, so we have a strong guarantee that no future commit will
3129 * leave these blocks visible to the user.)
3131 * Called with inode->sem down.
3133 int ext3_setattr(struct dentry
*dentry
, struct iattr
*attr
)
3135 struct inode
*inode
= dentry
->d_inode
;
3137 const unsigned int ia_valid
= attr
->ia_valid
;
3139 error
= inode_change_ok(inode
, attr
);
3143 if ((ia_valid
& ATTR_UID
&& attr
->ia_uid
!= inode
->i_uid
) ||
3144 (ia_valid
& ATTR_GID
&& attr
->ia_gid
!= inode
->i_gid
)) {
3147 /* (user+group)*(old+new) structure, inode write (sb,
3148 * inode block, ? - but truncate inode update has it) */
3149 handle
= ext3_journal_start(inode
, 2*(EXT3_QUOTA_INIT_BLOCKS(inode
->i_sb
)+
3150 EXT3_QUOTA_DEL_BLOCKS(inode
->i_sb
))+3);
3151 if (IS_ERR(handle
)) {
3152 error
= PTR_ERR(handle
);
3155 error
= vfs_dq_transfer(inode
, attr
) ? -EDQUOT
: 0;
3157 ext3_journal_stop(handle
);
3160 /* Update corresponding info in inode so that everything is in
3161 * one transaction */
3162 if (attr
->ia_valid
& ATTR_UID
)
3163 inode
->i_uid
= attr
->ia_uid
;
3164 if (attr
->ia_valid
& ATTR_GID
)
3165 inode
->i_gid
= attr
->ia_gid
;
3166 error
= ext3_mark_inode_dirty(handle
, inode
);
3167 ext3_journal_stop(handle
);
3170 if (S_ISREG(inode
->i_mode
) &&
3171 attr
->ia_valid
& ATTR_SIZE
&& attr
->ia_size
< inode
->i_size
) {
3174 handle
= ext3_journal_start(inode
, 3);
3175 if (IS_ERR(handle
)) {
3176 error
= PTR_ERR(handle
);
3180 error
= ext3_orphan_add(handle
, inode
);
3181 EXT3_I(inode
)->i_disksize
= attr
->ia_size
;
3182 rc
= ext3_mark_inode_dirty(handle
, inode
);
3185 ext3_journal_stop(handle
);
3188 rc
= inode_setattr(inode
, attr
);
3190 if (!rc
&& (ia_valid
& ATTR_MODE
))
3191 rc
= ext3_acl_chmod(inode
);
3194 ext3_std_error(inode
->i_sb
, error
);
3202 * How many blocks doth make a writepage()?
3204 * With N blocks per page, it may be:
3209 * N+5 bitmap blocks (from the above)
3210 * N+5 group descriptor summary blocks
3213 * 2 * EXT3_SINGLEDATA_TRANS_BLOCKS for the quote files
3215 * 3 * (N + 5) + 2 + 2 * EXT3_SINGLEDATA_TRANS_BLOCKS
3217 * With ordered or writeback data it's the same, less the N data blocks.
3219 * If the inode's direct blocks can hold an integral number of pages then a
3220 * page cannot straddle two indirect blocks, and we can only touch one indirect
3221 * and dindirect block, and the "5" above becomes "3".
3223 * This still overestimates under most circumstances. If we were to pass the
3224 * start and end offsets in here as well we could do block_to_path() on each
3225 * block and work out the exact number of indirects which are touched. Pah.
3228 static int ext3_writepage_trans_blocks(struct inode
*inode
)
3230 int bpp
= ext3_journal_blocks_per_page(inode
);
3231 int indirects
= (EXT3_NDIR_BLOCKS
% bpp
) ? 5 : 3;
3234 if (ext3_should_journal_data(inode
))
3235 ret
= 3 * (bpp
+ indirects
) + 2;
3237 ret
= 2 * (bpp
+ indirects
) + 2;
3240 /* We know that structure was already allocated during vfs_dq_init so
3241 * we will be updating only the data blocks + inodes */
3242 ret
+= 2*EXT3_QUOTA_TRANS_BLOCKS(inode
->i_sb
);
3249 * The caller must have previously called ext3_reserve_inode_write().
3250 * Give this, we know that the caller already has write access to iloc->bh.
3252 int ext3_mark_iloc_dirty(handle_t
*handle
,
3253 struct inode
*inode
, struct ext3_iloc
*iloc
)
3257 /* the do_update_inode consumes one bh->b_count */
3260 /* ext3_do_update_inode() does journal_dirty_metadata */
3261 err
= ext3_do_update_inode(handle
, inode
, iloc
);
3267 * On success, We end up with an outstanding reference count against
3268 * iloc->bh. This _must_ be cleaned up later.
3272 ext3_reserve_inode_write(handle_t
*handle
, struct inode
*inode
,
3273 struct ext3_iloc
*iloc
)
3277 err
= ext3_get_inode_loc(inode
, iloc
);
3279 BUFFER_TRACE(iloc
->bh
, "get_write_access");
3280 err
= ext3_journal_get_write_access(handle
, iloc
->bh
);
3287 ext3_std_error(inode
->i_sb
, err
);
3292 * What we do here is to mark the in-core inode as clean with respect to inode
3293 * dirtiness (it may still be data-dirty).
3294 * This means that the in-core inode may be reaped by prune_icache
3295 * without having to perform any I/O. This is a very good thing,
3296 * because *any* task may call prune_icache - even ones which
3297 * have a transaction open against a different journal.
3299 * Is this cheating? Not really. Sure, we haven't written the
3300 * inode out, but prune_icache isn't a user-visible syncing function.
3301 * Whenever the user wants stuff synced (sys_sync, sys_msync, sys_fsync)
3302 * we start and wait on commits.
3304 * Is this efficient/effective? Well, we're being nice to the system
3305 * by cleaning up our inodes proactively so they can be reaped
3306 * without I/O. But we are potentially leaving up to five seconds'
3307 * worth of inodes floating about which prune_icache wants us to
3308 * write out. One way to fix that would be to get prune_icache()
3309 * to do a write_super() to free up some memory. It has the desired
3312 int ext3_mark_inode_dirty(handle_t
*handle
, struct inode
*inode
)
3314 struct ext3_iloc iloc
;
3318 err
= ext3_reserve_inode_write(handle
, inode
, &iloc
);
3320 err
= ext3_mark_iloc_dirty(handle
, inode
, &iloc
);
3325 * ext3_dirty_inode() is called from __mark_inode_dirty()
3327 * We're really interested in the case where a file is being extended.
3328 * i_size has been changed by generic_commit_write() and we thus need
3329 * to include the updated inode in the current transaction.
3331 * Also, vfs_dq_alloc_space() will always dirty the inode when blocks
3332 * are allocated to the file.
3334 * If the inode is marked synchronous, we don't honour that here - doing
3335 * so would cause a commit on atime updates, which we don't bother doing.
3336 * We handle synchronous inodes at the highest possible level.
3338 void ext3_dirty_inode(struct inode
*inode
)
3340 handle_t
*current_handle
= ext3_journal_current_handle();
3343 handle
= ext3_journal_start(inode
, 2);
3346 if (current_handle
&&
3347 current_handle
->h_transaction
!= handle
->h_transaction
) {
3348 /* This task has a transaction open against a different fs */
3349 printk(KERN_EMERG
"%s: transactions do not match!\n",
3352 jbd_debug(5, "marking dirty. outer handle=%p\n",
3354 ext3_mark_inode_dirty(handle
, inode
);
3356 ext3_journal_stop(handle
);
3363 * Bind an inode's backing buffer_head into this transaction, to prevent
3364 * it from being flushed to disk early. Unlike
3365 * ext3_reserve_inode_write, this leaves behind no bh reference and
3366 * returns no iloc structure, so the caller needs to repeat the iloc
3367 * lookup to mark the inode dirty later.
3369 static int ext3_pin_inode(handle_t
*handle
, struct inode
*inode
)
3371 struct ext3_iloc iloc
;
3375 err
= ext3_get_inode_loc(inode
, &iloc
);
3377 BUFFER_TRACE(iloc
.bh
, "get_write_access");
3378 err
= journal_get_write_access(handle
, iloc
.bh
);
3380 err
= ext3_journal_dirty_metadata(handle
,
3385 ext3_std_error(inode
->i_sb
, err
);
3390 int ext3_change_inode_journal_flag(struct inode
*inode
, int val
)
3397 * We have to be very careful here: changing a data block's
3398 * journaling status dynamically is dangerous. If we write a
3399 * data block to the journal, change the status and then delete
3400 * that block, we risk forgetting to revoke the old log record
3401 * from the journal and so a subsequent replay can corrupt data.
3402 * So, first we make sure that the journal is empty and that
3403 * nobody is changing anything.
3406 journal
= EXT3_JOURNAL(inode
);
3407 if (is_journal_aborted(journal
))
3410 journal_lock_updates(journal
);
3411 journal_flush(journal
);
3414 * OK, there are no updates running now, and all cached data is
3415 * synced to disk. We are now in a completely consistent state
3416 * which doesn't have anything in the journal, and we know that
3417 * no filesystem updates are running, so it is safe to modify
3418 * the inode's in-core data-journaling state flag now.
3422 EXT3_I(inode
)->i_flags
|= EXT3_JOURNAL_DATA_FL
;
3424 EXT3_I(inode
)->i_flags
&= ~EXT3_JOURNAL_DATA_FL
;
3425 ext3_set_aops(inode
);
3427 journal_unlock_updates(journal
);
3429 /* Finally we can mark the inode as dirty. */
3431 handle
= ext3_journal_start(inode
, 1);
3433 return PTR_ERR(handle
);
3435 err
= ext3_mark_inode_dirty(handle
, inode
);
3437 ext3_journal_stop(handle
);
3438 ext3_std_error(inode
->i_sb
, err
);