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 inode eviction from icache
195 void ext3_evict_inode (struct inode
*inode
)
197 struct ext3_block_alloc_info
*rsv
;
201 if (!inode
->i_nlink
&& !is_bad_inode(inode
)) {
202 dquot_initialize(inode
);
206 truncate_inode_pages(&inode
->i_data
, 0);
208 ext3_discard_reservation(inode
);
209 rsv
= EXT3_I(inode
)->i_block_alloc_info
;
210 EXT3_I(inode
)->i_block_alloc_info
= NULL
;
217 handle
= start_transaction(inode
);
218 if (IS_ERR(handle
)) {
220 * If we're going to skip the normal cleanup, we still need to
221 * make sure that the in-core orphan linked list is properly
224 ext3_orphan_del(NULL
, inode
);
232 ext3_truncate(inode
);
234 * Kill off the orphan record which ext3_truncate created.
235 * AKPM: I think this can be inside the above `if'.
236 * Note that ext3_orphan_del() has to be able to cope with the
237 * deletion of a non-existent orphan - this is because we don't
238 * know if ext3_truncate() actually created an orphan record.
239 * (Well, we could do this if we need to, but heck - it works)
241 ext3_orphan_del(handle
, inode
);
242 EXT3_I(inode
)->i_dtime
= get_seconds();
245 * One subtle ordering requirement: if anything has gone wrong
246 * (transaction abort, IO errors, whatever), then we can still
247 * do these next steps (the fs will already have been marked as
248 * having errors), but we can't free the inode if the mark_dirty
251 if (ext3_mark_inode_dirty(handle
, inode
)) {
252 /* If that failed, just dquot_drop() and be done with that */
254 end_writeback(inode
);
256 ext3_xattr_delete_inode(handle
, inode
);
257 dquot_free_inode(inode
);
259 end_writeback(inode
);
260 ext3_free_inode(handle
, inode
);
262 ext3_journal_stop(handle
);
265 end_writeback(inode
);
272 struct buffer_head
*bh
;
275 static inline void add_chain(Indirect
*p
, struct buffer_head
*bh
, __le32
*v
)
277 p
->key
= *(p
->p
= v
);
281 static int verify_chain(Indirect
*from
, Indirect
*to
)
283 while (from
<= to
&& from
->key
== *from
->p
)
289 * ext3_block_to_path - parse the block number into array of offsets
290 * @inode: inode in question (we are only interested in its superblock)
291 * @i_block: block number to be parsed
292 * @offsets: array to store the offsets in
293 * @boundary: set this non-zero if the referred-to block is likely to be
294 * followed (on disk) by an indirect block.
296 * To store the locations of file's data ext3 uses a data structure common
297 * for UNIX filesystems - tree of pointers anchored in the inode, with
298 * data blocks at leaves and indirect blocks in intermediate nodes.
299 * This function translates the block number into path in that tree -
300 * return value is the path length and @offsets[n] is the offset of
301 * pointer to (n+1)th node in the nth one. If @block is out of range
302 * (negative or too large) warning is printed and zero returned.
304 * Note: function doesn't find node addresses, so no IO is needed. All
305 * we need to know is the capacity of indirect blocks (taken from the
310 * Portability note: the last comparison (check that we fit into triple
311 * indirect block) is spelled differently, because otherwise on an
312 * architecture with 32-bit longs and 8Kb pages we might get into trouble
313 * if our filesystem had 8Kb blocks. We might use long long, but that would
314 * kill us on x86. Oh, well, at least the sign propagation does not matter -
315 * i_block would have to be negative in the very beginning, so we would not
319 static int ext3_block_to_path(struct inode
*inode
,
320 long i_block
, int offsets
[4], int *boundary
)
322 int ptrs
= EXT3_ADDR_PER_BLOCK(inode
->i_sb
);
323 int ptrs_bits
= EXT3_ADDR_PER_BLOCK_BITS(inode
->i_sb
);
324 const long direct_blocks
= EXT3_NDIR_BLOCKS
,
325 indirect_blocks
= ptrs
,
326 double_blocks
= (1 << (ptrs_bits
* 2));
331 ext3_warning (inode
->i_sb
, "ext3_block_to_path", "block < 0");
332 } else if (i_block
< direct_blocks
) {
333 offsets
[n
++] = i_block
;
334 final
= direct_blocks
;
335 } else if ( (i_block
-= direct_blocks
) < indirect_blocks
) {
336 offsets
[n
++] = EXT3_IND_BLOCK
;
337 offsets
[n
++] = i_block
;
339 } else if ((i_block
-= indirect_blocks
) < double_blocks
) {
340 offsets
[n
++] = EXT3_DIND_BLOCK
;
341 offsets
[n
++] = i_block
>> ptrs_bits
;
342 offsets
[n
++] = i_block
& (ptrs
- 1);
344 } else if (((i_block
-= double_blocks
) >> (ptrs_bits
* 2)) < ptrs
) {
345 offsets
[n
++] = EXT3_TIND_BLOCK
;
346 offsets
[n
++] = i_block
>> (ptrs_bits
* 2);
347 offsets
[n
++] = (i_block
>> ptrs_bits
) & (ptrs
- 1);
348 offsets
[n
++] = i_block
& (ptrs
- 1);
351 ext3_warning(inode
->i_sb
, "ext3_block_to_path", "block > big");
354 *boundary
= final
- 1 - (i_block
& (ptrs
- 1));
359 * ext3_get_branch - read the chain of indirect blocks leading to data
360 * @inode: inode in question
361 * @depth: depth of the chain (1 - direct pointer, etc.)
362 * @offsets: offsets of pointers in inode/indirect blocks
363 * @chain: place to store the result
364 * @err: here we store the error value
366 * Function fills the array of triples <key, p, bh> and returns %NULL
367 * if everything went OK or the pointer to the last filled triple
368 * (incomplete one) otherwise. Upon the return chain[i].key contains
369 * the number of (i+1)-th block in the chain (as it is stored in memory,
370 * i.e. little-endian 32-bit), chain[i].p contains the address of that
371 * number (it points into struct inode for i==0 and into the bh->b_data
372 * for i>0) and chain[i].bh points to the buffer_head of i-th indirect
373 * block for i>0 and NULL for i==0. In other words, it holds the block
374 * numbers of the chain, addresses they were taken from (and where we can
375 * verify that chain did not change) and buffer_heads hosting these
378 * Function stops when it stumbles upon zero pointer (absent block)
379 * (pointer to last triple returned, *@err == 0)
380 * or when it gets an IO error reading an indirect block
381 * (ditto, *@err == -EIO)
382 * or when it notices that chain had been changed while it was reading
383 * (ditto, *@err == -EAGAIN)
384 * or when it reads all @depth-1 indirect blocks successfully and finds
385 * the whole chain, all way to the data (returns %NULL, *err == 0).
387 static Indirect
*ext3_get_branch(struct inode
*inode
, int depth
, int *offsets
,
388 Indirect chain
[4], int *err
)
390 struct super_block
*sb
= inode
->i_sb
;
392 struct buffer_head
*bh
;
395 /* i_data is not going away, no lock needed */
396 add_chain (chain
, NULL
, EXT3_I(inode
)->i_data
+ *offsets
);
400 bh
= sb_bread(sb
, le32_to_cpu(p
->key
));
403 /* Reader: pointers */
404 if (!verify_chain(chain
, p
))
406 add_chain(++p
, bh
, (__le32
*)bh
->b_data
+ *++offsets
);
424 * ext3_find_near - find a place for allocation with sufficient locality
426 * @ind: descriptor of indirect block.
428 * This function returns the preferred place for block allocation.
429 * It is used when heuristic for sequential allocation fails.
431 * + if there is a block to the left of our position - allocate near it.
432 * + if pointer will live in indirect block - allocate near that block.
433 * + if pointer will live in inode - allocate in the same
436 * In the latter case we colour the starting block by the callers PID to
437 * prevent it from clashing with concurrent allocations for a different inode
438 * in the same block group. The PID is used here so that functionally related
439 * files will be close-by on-disk.
441 * Caller must make sure that @ind is valid and will stay that way.
443 static ext3_fsblk_t
ext3_find_near(struct inode
*inode
, Indirect
*ind
)
445 struct ext3_inode_info
*ei
= EXT3_I(inode
);
446 __le32
*start
= ind
->bh
? (__le32
*) ind
->bh
->b_data
: ei
->i_data
;
448 ext3_fsblk_t bg_start
;
449 ext3_grpblk_t colour
;
451 /* Try to find previous block */
452 for (p
= ind
->p
- 1; p
>= start
; p
--) {
454 return le32_to_cpu(*p
);
457 /* No such thing, so let's try location of indirect block */
459 return ind
->bh
->b_blocknr
;
462 * It is going to be referred to from the inode itself? OK, just put it
463 * into the same cylinder group then.
465 bg_start
= ext3_group_first_block_no(inode
->i_sb
, ei
->i_block_group
);
466 colour
= (current
->pid
% 16) *
467 (EXT3_BLOCKS_PER_GROUP(inode
->i_sb
) / 16);
468 return bg_start
+ colour
;
472 * ext3_find_goal - find a preferred place for allocation.
474 * @block: block we want
475 * @partial: pointer to the last triple within a chain
477 * Normally this function find the preferred place for block allocation,
481 static ext3_fsblk_t
ext3_find_goal(struct inode
*inode
, long block
,
484 struct ext3_block_alloc_info
*block_i
;
486 block_i
= EXT3_I(inode
)->i_block_alloc_info
;
489 * try the heuristic for sequential allocation,
490 * failing that at least try to get decent locality.
492 if (block_i
&& (block
== block_i
->last_alloc_logical_block
+ 1)
493 && (block_i
->last_alloc_physical_block
!= 0)) {
494 return block_i
->last_alloc_physical_block
+ 1;
497 return ext3_find_near(inode
, partial
);
501 * ext3_blks_to_allocate: Look up the block map and count the number
502 * of direct blocks need to be allocated for the given branch.
504 * @branch: chain of indirect blocks
505 * @k: number of blocks need for indirect blocks
506 * @blks: number of data blocks to be mapped.
507 * @blocks_to_boundary: the offset in the indirect block
509 * return the total number of blocks to be allocate, including the
510 * direct and indirect blocks.
512 static int ext3_blks_to_allocate(Indirect
*branch
, int k
, unsigned long blks
,
513 int blocks_to_boundary
)
515 unsigned long count
= 0;
518 * Simple case, [t,d]Indirect block(s) has not allocated yet
519 * then it's clear blocks on that path have not allocated
522 /* right now we don't handle cross boundary allocation */
523 if (blks
< blocks_to_boundary
+ 1)
526 count
+= blocks_to_boundary
+ 1;
531 while (count
< blks
&& count
<= blocks_to_boundary
&&
532 le32_to_cpu(*(branch
[0].p
+ count
)) == 0) {
539 * ext3_alloc_blocks: multiple allocate blocks needed for a branch
540 * @indirect_blks: the number of blocks need to allocate for indirect
543 * @new_blocks: on return it will store the new block numbers for
544 * the indirect blocks(if needed) and the first direct block,
545 * @blks: on return it will store the total number of allocated
548 static int ext3_alloc_blocks(handle_t
*handle
, struct inode
*inode
,
549 ext3_fsblk_t goal
, int indirect_blks
, int blks
,
550 ext3_fsblk_t new_blocks
[4], int *err
)
553 unsigned long count
= 0;
555 ext3_fsblk_t current_block
= 0;
559 * Here we try to allocate the requested multiple blocks at once,
560 * on a best-effort basis.
561 * To build a branch, we should allocate blocks for
562 * the indirect blocks(if not allocated yet), and at least
563 * the first direct block of this branch. That's the
564 * minimum number of blocks need to allocate(required)
566 target
= blks
+ indirect_blks
;
570 /* allocating blocks for indirect blocks and direct blocks */
571 current_block
= ext3_new_blocks(handle
,inode
,goal
,&count
,err
);
576 /* allocate blocks for indirect blocks */
577 while (index
< indirect_blks
&& count
) {
578 new_blocks
[index
++] = current_block
++;
586 /* save the new block number for the first direct block */
587 new_blocks
[index
] = current_block
;
589 /* total number of blocks allocated for direct blocks */
594 for (i
= 0; i
<index
; i
++)
595 ext3_free_blocks(handle
, inode
, new_blocks
[i
], 1);
600 * ext3_alloc_branch - allocate and set up a chain of blocks.
602 * @indirect_blks: number of allocated indirect blocks
603 * @blks: number of allocated direct blocks
604 * @offsets: offsets (in the blocks) to store the pointers to next.
605 * @branch: place to store the chain in.
607 * This function allocates blocks, zeroes out all but the last one,
608 * links them into chain and (if we are synchronous) writes them to disk.
609 * In other words, it prepares a branch that can be spliced onto the
610 * inode. It stores the information about that chain in the branch[], in
611 * the same format as ext3_get_branch() would do. We are calling it after
612 * we had read the existing part of chain and partial points to the last
613 * triple of that (one with zero ->key). Upon the exit we have the same
614 * picture as after the successful ext3_get_block(), except that in one
615 * place chain is disconnected - *branch->p is still zero (we did not
616 * set the last link), but branch->key contains the number that should
617 * be placed into *branch->p to fill that gap.
619 * If allocation fails we free all blocks we've allocated (and forget
620 * their buffer_heads) and return the error value the from failed
621 * ext3_alloc_block() (normally -ENOSPC). Otherwise we set the chain
622 * as described above and return 0.
624 static int ext3_alloc_branch(handle_t
*handle
, struct inode
*inode
,
625 int indirect_blks
, int *blks
, ext3_fsblk_t goal
,
626 int *offsets
, Indirect
*branch
)
628 int blocksize
= inode
->i_sb
->s_blocksize
;
631 struct buffer_head
*bh
;
633 ext3_fsblk_t new_blocks
[4];
634 ext3_fsblk_t current_block
;
636 num
= ext3_alloc_blocks(handle
, inode
, goal
, indirect_blks
,
637 *blks
, new_blocks
, &err
);
641 branch
[0].key
= cpu_to_le32(new_blocks
[0]);
643 * metadata blocks and data blocks are allocated.
645 for (n
= 1; n
<= indirect_blks
; n
++) {
647 * Get buffer_head for parent block, zero it out
648 * and set the pointer to new one, then send
651 bh
= sb_getblk(inode
->i_sb
, new_blocks
[n
-1]);
654 BUFFER_TRACE(bh
, "call get_create_access");
655 err
= ext3_journal_get_create_access(handle
, bh
);
662 memset(bh
->b_data
, 0, blocksize
);
663 branch
[n
].p
= (__le32
*) bh
->b_data
+ offsets
[n
];
664 branch
[n
].key
= cpu_to_le32(new_blocks
[n
]);
665 *branch
[n
].p
= branch
[n
].key
;
666 if ( n
== indirect_blks
) {
667 current_block
= new_blocks
[n
];
669 * End of chain, update the last new metablock of
670 * the chain to point to the new allocated
671 * data blocks numbers
673 for (i
=1; i
< num
; i
++)
674 *(branch
[n
].p
+ i
) = cpu_to_le32(++current_block
);
676 BUFFER_TRACE(bh
, "marking uptodate");
677 set_buffer_uptodate(bh
);
680 BUFFER_TRACE(bh
, "call ext3_journal_dirty_metadata");
681 err
= ext3_journal_dirty_metadata(handle
, bh
);
688 /* Allocation failed, free what we already allocated */
689 for (i
= 1; i
<= n
; i
++) {
690 BUFFER_TRACE(branch
[i
].bh
, "call journal_forget");
691 ext3_journal_forget(handle
, branch
[i
].bh
);
693 for (i
= 0; i
<indirect_blks
; i
++)
694 ext3_free_blocks(handle
, inode
, new_blocks
[i
], 1);
696 ext3_free_blocks(handle
, inode
, new_blocks
[i
], num
);
702 * ext3_splice_branch - splice the allocated branch onto inode.
704 * @block: (logical) number of block we are adding
705 * @chain: chain of indirect blocks (with a missing link - see
707 * @where: location of missing link
708 * @num: number of indirect blocks we are adding
709 * @blks: number of direct blocks we are adding
711 * This function fills the missing link and does all housekeeping needed in
712 * inode (->i_blocks, etc.). In case of success we end up with the full
713 * chain to new block and return 0.
715 static int ext3_splice_branch(handle_t
*handle
, struct inode
*inode
,
716 long block
, Indirect
*where
, int num
, int blks
)
720 struct ext3_block_alloc_info
*block_i
;
721 ext3_fsblk_t current_block
;
722 struct ext3_inode_info
*ei
= EXT3_I(inode
);
724 block_i
= ei
->i_block_alloc_info
;
726 * If we're splicing into a [td]indirect block (as opposed to the
727 * inode) then we need to get write access to the [td]indirect block
731 BUFFER_TRACE(where
->bh
, "get_write_access");
732 err
= ext3_journal_get_write_access(handle
, where
->bh
);
738 *where
->p
= where
->key
;
741 * Update the host buffer_head or inode to point to more just allocated
742 * direct blocks blocks
744 if (num
== 0 && blks
> 1) {
745 current_block
= le32_to_cpu(where
->key
) + 1;
746 for (i
= 1; i
< blks
; i
++)
747 *(where
->p
+ i
) = cpu_to_le32(current_block
++);
751 * update the most recently allocated logical & physical block
752 * in i_block_alloc_info, to assist find the proper goal block for next
756 block_i
->last_alloc_logical_block
= block
+ blks
- 1;
757 block_i
->last_alloc_physical_block
=
758 le32_to_cpu(where
[num
].key
) + blks
- 1;
761 /* We are done with atomic stuff, now do the rest of housekeeping */
763 inode
->i_ctime
= CURRENT_TIME_SEC
;
764 ext3_mark_inode_dirty(handle
, inode
);
765 /* ext3_mark_inode_dirty already updated i_sync_tid */
766 atomic_set(&ei
->i_datasync_tid
, handle
->h_transaction
->t_tid
);
768 /* had we spliced it onto indirect block? */
771 * If we spliced it onto an indirect block, we haven't
772 * altered the inode. Note however that if it is being spliced
773 * onto an indirect block at the very end of the file (the
774 * file is growing) then we *will* alter the inode to reflect
775 * the new i_size. But that is not done here - it is done in
776 * generic_commit_write->__mark_inode_dirty->ext3_dirty_inode.
778 jbd_debug(5, "splicing indirect only\n");
779 BUFFER_TRACE(where
->bh
, "call ext3_journal_dirty_metadata");
780 err
= ext3_journal_dirty_metadata(handle
, where
->bh
);
785 * OK, we spliced it into the inode itself on a direct block.
786 * Inode was dirtied above.
788 jbd_debug(5, "splicing direct\n");
793 for (i
= 1; i
<= num
; i
++) {
794 BUFFER_TRACE(where
[i
].bh
, "call journal_forget");
795 ext3_journal_forget(handle
, where
[i
].bh
);
796 ext3_free_blocks(handle
,inode
,le32_to_cpu(where
[i
-1].key
),1);
798 ext3_free_blocks(handle
, inode
, le32_to_cpu(where
[num
].key
), blks
);
804 * Allocation strategy is simple: if we have to allocate something, we will
805 * have to go the whole way to leaf. So let's do it before attaching anything
806 * to tree, set linkage between the newborn blocks, write them if sync is
807 * required, recheck the path, free and repeat if check fails, otherwise
808 * set the last missing link (that will protect us from any truncate-generated
809 * removals - all blocks on the path are immune now) and possibly force the
810 * write on the parent block.
811 * That has a nice additional property: no special recovery from the failed
812 * allocations is needed - we simply release blocks and do not touch anything
813 * reachable from inode.
815 * `handle' can be NULL if create == 0.
817 * The BKL may not be held on entry here. Be sure to take it early.
818 * return > 0, # of blocks mapped or allocated.
819 * return = 0, if plain lookup failed.
820 * return < 0, error case.
822 int ext3_get_blocks_handle(handle_t
*handle
, struct inode
*inode
,
823 sector_t iblock
, unsigned long maxblocks
,
824 struct buffer_head
*bh_result
,
833 int blocks_to_boundary
= 0;
835 struct ext3_inode_info
*ei
= EXT3_I(inode
);
837 ext3_fsblk_t first_block
= 0;
840 J_ASSERT(handle
!= NULL
|| create
== 0);
841 depth
= ext3_block_to_path(inode
,iblock
,offsets
,&blocks_to_boundary
);
846 partial
= ext3_get_branch(inode
, depth
, offsets
, chain
, &err
);
848 /* Simplest case - block found, no allocation needed */
850 first_block
= le32_to_cpu(chain
[depth
- 1].key
);
851 clear_buffer_new(bh_result
);
854 while (count
< maxblocks
&& count
<= blocks_to_boundary
) {
857 if (!verify_chain(chain
, chain
+ depth
- 1)) {
859 * Indirect block might be removed by
860 * truncate while we were reading it.
861 * Handling of that case: forget what we've
862 * got now. Flag the err as EAGAIN, so it
869 blk
= le32_to_cpu(*(chain
[depth
-1].p
+ count
));
871 if (blk
== first_block
+ count
)
880 /* Next simple case - plain lookup or failed read of indirect block */
881 if (!create
|| err
== -EIO
)
884 mutex_lock(&ei
->truncate_mutex
);
887 * If the indirect block is missing while we are reading
888 * the chain(ext3_get_branch() returns -EAGAIN err), or
889 * if the chain has been changed after we grab the semaphore,
890 * (either because another process truncated this branch, or
891 * another get_block allocated this branch) re-grab the chain to see if
892 * the request block has been allocated or not.
894 * Since we already block the truncate/other get_block
895 * at this point, we will have the current copy of the chain when we
896 * splice the branch into the tree.
898 if (err
== -EAGAIN
|| !verify_chain(chain
, partial
)) {
899 while (partial
> chain
) {
903 partial
= ext3_get_branch(inode
, depth
, offsets
, chain
, &err
);
906 mutex_unlock(&ei
->truncate_mutex
);
909 clear_buffer_new(bh_result
);
915 * Okay, we need to do block allocation. Lazily initialize the block
916 * allocation info here if necessary
918 if (S_ISREG(inode
->i_mode
) && (!ei
->i_block_alloc_info
))
919 ext3_init_block_alloc_info(inode
);
921 goal
= ext3_find_goal(inode
, iblock
, partial
);
923 /* the number of blocks need to allocate for [d,t]indirect blocks */
924 indirect_blks
= (chain
+ depth
) - partial
- 1;
927 * Next look up the indirect map to count the totoal number of
928 * direct blocks to allocate for this branch.
930 count
= ext3_blks_to_allocate(partial
, indirect_blks
,
931 maxblocks
, blocks_to_boundary
);
933 * Block out ext3_truncate while we alter the tree
935 err
= ext3_alloc_branch(handle
, inode
, indirect_blks
, &count
, goal
,
936 offsets
+ (partial
- chain
), partial
);
939 * The ext3_splice_branch call will free and forget any buffers
940 * on the new chain if there is a failure, but that risks using
941 * up transaction credits, especially for bitmaps where the
942 * credits cannot be returned. Can we handle this somehow? We
943 * may need to return -EAGAIN upwards in the worst case. --sct
946 err
= ext3_splice_branch(handle
, inode
, iblock
,
947 partial
, indirect_blks
, count
);
948 mutex_unlock(&ei
->truncate_mutex
);
952 set_buffer_new(bh_result
);
954 map_bh(bh_result
, inode
->i_sb
, le32_to_cpu(chain
[depth
-1].key
));
955 if (count
> blocks_to_boundary
)
956 set_buffer_boundary(bh_result
);
958 /* Clean up and exit */
959 partial
= chain
+ depth
- 1; /* the whole chain */
961 while (partial
> chain
) {
962 BUFFER_TRACE(partial
->bh
, "call brelse");
966 BUFFER_TRACE(bh_result
, "returned");
971 /* Maximum number of blocks we map for direct IO at once. */
972 #define DIO_MAX_BLOCKS 4096
974 * Number of credits we need for writing DIO_MAX_BLOCKS:
975 * We need sb + group descriptor + bitmap + inode -> 4
976 * For B blocks with A block pointers per block we need:
977 * 1 (triple ind.) + (B/A/A + 2) (doubly ind.) + (B/A + 2) (indirect).
978 * If we plug in 4096 for B and 256 for A (for 1KB block size), we get 25.
980 #define DIO_CREDITS 25
982 static int ext3_get_block(struct inode
*inode
, sector_t iblock
,
983 struct buffer_head
*bh_result
, int create
)
985 handle_t
*handle
= ext3_journal_current_handle();
986 int ret
= 0, started
= 0;
987 unsigned max_blocks
= bh_result
->b_size
>> inode
->i_blkbits
;
989 if (create
&& !handle
) { /* Direct IO write... */
990 if (max_blocks
> DIO_MAX_BLOCKS
)
991 max_blocks
= DIO_MAX_BLOCKS
;
992 handle
= ext3_journal_start(inode
, DIO_CREDITS
+
993 EXT3_MAXQUOTAS_TRANS_BLOCKS(inode
->i_sb
));
994 if (IS_ERR(handle
)) {
995 ret
= PTR_ERR(handle
);
1001 ret
= ext3_get_blocks_handle(handle
, inode
, iblock
,
1002 max_blocks
, bh_result
, create
);
1004 bh_result
->b_size
= (ret
<< inode
->i_blkbits
);
1008 ext3_journal_stop(handle
);
1013 int ext3_fiemap(struct inode
*inode
, struct fiemap_extent_info
*fieinfo
,
1016 return generic_block_fiemap(inode
, fieinfo
, start
, len
,
1021 * `handle' can be NULL if create is zero
1023 struct buffer_head
*ext3_getblk(handle_t
*handle
, struct inode
*inode
,
1024 long block
, int create
, int *errp
)
1026 struct buffer_head dummy
;
1029 J_ASSERT(handle
!= NULL
|| create
== 0);
1032 dummy
.b_blocknr
= -1000;
1033 buffer_trace_init(&dummy
.b_history
);
1034 err
= ext3_get_blocks_handle(handle
, inode
, block
, 1,
1037 * ext3_get_blocks_handle() returns number of blocks
1038 * mapped. 0 in case of a HOLE.
1046 if (!err
&& buffer_mapped(&dummy
)) {
1047 struct buffer_head
*bh
;
1048 bh
= sb_getblk(inode
->i_sb
, dummy
.b_blocknr
);
1053 if (buffer_new(&dummy
)) {
1054 J_ASSERT(create
!= 0);
1055 J_ASSERT(handle
!= NULL
);
1058 * Now that we do not always journal data, we should
1059 * keep in mind whether this should always journal the
1060 * new buffer as metadata. For now, regular file
1061 * writes use ext3_get_block instead, so it's not a
1065 BUFFER_TRACE(bh
, "call get_create_access");
1066 fatal
= ext3_journal_get_create_access(handle
, bh
);
1067 if (!fatal
&& !buffer_uptodate(bh
)) {
1068 memset(bh
->b_data
,0,inode
->i_sb
->s_blocksize
);
1069 set_buffer_uptodate(bh
);
1072 BUFFER_TRACE(bh
, "call ext3_journal_dirty_metadata");
1073 err
= ext3_journal_dirty_metadata(handle
, bh
);
1077 BUFFER_TRACE(bh
, "not a new buffer");
1090 struct buffer_head
*ext3_bread(handle_t
*handle
, struct inode
*inode
,
1091 int block
, int create
, int *err
)
1093 struct buffer_head
* bh
;
1095 bh
= ext3_getblk(handle
, inode
, block
, create
, err
);
1098 if (buffer_uptodate(bh
))
1100 ll_rw_block(READ_META
, 1, &bh
);
1102 if (buffer_uptodate(bh
))
1109 static int walk_page_buffers( handle_t
*handle
,
1110 struct buffer_head
*head
,
1114 int (*fn
)( handle_t
*handle
,
1115 struct buffer_head
*bh
))
1117 struct buffer_head
*bh
;
1118 unsigned block_start
, block_end
;
1119 unsigned blocksize
= head
->b_size
;
1121 struct buffer_head
*next
;
1123 for ( bh
= head
, block_start
= 0;
1124 ret
== 0 && (bh
!= head
|| !block_start
);
1125 block_start
= block_end
, bh
= next
)
1127 next
= bh
->b_this_page
;
1128 block_end
= block_start
+ blocksize
;
1129 if (block_end
<= from
|| block_start
>= to
) {
1130 if (partial
&& !buffer_uptodate(bh
))
1134 err
= (*fn
)(handle
, bh
);
1142 * To preserve ordering, it is essential that the hole instantiation and
1143 * the data write be encapsulated in a single transaction. We cannot
1144 * close off a transaction and start a new one between the ext3_get_block()
1145 * and the commit_write(). So doing the journal_start at the start of
1146 * prepare_write() is the right place.
1148 * Also, this function can nest inside ext3_writepage() ->
1149 * block_write_full_page(). In that case, we *know* that ext3_writepage()
1150 * has generated enough buffer credits to do the whole page. So we won't
1151 * block on the journal in that case, which is good, because the caller may
1154 * By accident, ext3 can be reentered when a transaction is open via
1155 * quota file writes. If we were to commit the transaction while thus
1156 * reentered, there can be a deadlock - we would be holding a quota
1157 * lock, and the commit would never complete if another thread had a
1158 * transaction open and was blocking on the quota lock - a ranking
1161 * So what we do is to rely on the fact that journal_stop/journal_start
1162 * will _not_ run commit under these circumstances because handle->h_ref
1163 * is elevated. We'll still have enough credits for the tiny quotafile
1166 static int do_journal_get_write_access(handle_t
*handle
,
1167 struct buffer_head
*bh
)
1169 int dirty
= buffer_dirty(bh
);
1172 if (!buffer_mapped(bh
) || buffer_freed(bh
))
1175 * __block_prepare_write() could have dirtied some buffers. Clean
1176 * the dirty bit as jbd2_journal_get_write_access() could complain
1177 * otherwise about fs integrity issues. Setting of the dirty bit
1178 * by __block_prepare_write() isn't a real problem here as we clear
1179 * the bit before releasing a page lock and thus writeback cannot
1180 * ever write the buffer.
1183 clear_buffer_dirty(bh
);
1184 ret
= ext3_journal_get_write_access(handle
, bh
);
1186 ret
= ext3_journal_dirty_metadata(handle
, bh
);
1191 * Truncate blocks that were not used by write. We have to truncate the
1192 * pagecache as well so that corresponding buffers get properly unmapped.
1194 static void ext3_truncate_failed_write(struct inode
*inode
)
1196 truncate_inode_pages(inode
->i_mapping
, inode
->i_size
);
1197 ext3_truncate(inode
);
1200 static int ext3_write_begin(struct file
*file
, struct address_space
*mapping
,
1201 loff_t pos
, unsigned len
, unsigned flags
,
1202 struct page
**pagep
, void **fsdata
)
1204 struct inode
*inode
= mapping
->host
;
1211 /* Reserve one block more for addition to orphan list in case
1212 * we allocate blocks but write fails for some reason */
1213 int needed_blocks
= ext3_writepage_trans_blocks(inode
) + 1;
1215 index
= pos
>> PAGE_CACHE_SHIFT
;
1216 from
= pos
& (PAGE_CACHE_SIZE
- 1);
1220 page
= grab_cache_page_write_begin(mapping
, index
, flags
);
1225 handle
= ext3_journal_start(inode
, needed_blocks
);
1226 if (IS_ERR(handle
)) {
1228 page_cache_release(page
);
1229 ret
= PTR_ERR(handle
);
1232 ret
= __block_write_begin(page
, pos
, len
, ext3_get_block
);
1234 goto write_begin_failed
;
1236 if (ext3_should_journal_data(inode
)) {
1237 ret
= walk_page_buffers(handle
, page_buffers(page
),
1238 from
, to
, NULL
, do_journal_get_write_access
);
1243 * block_write_begin may have instantiated a few blocks
1244 * outside i_size. Trim these off again. Don't need
1245 * i_size_read because we hold i_mutex.
1247 * Add inode to orphan list in case we crash before truncate
1248 * finishes. Do this only if ext3_can_truncate() agrees so
1249 * that orphan processing code is happy.
1251 if (pos
+ len
> inode
->i_size
&& ext3_can_truncate(inode
))
1252 ext3_orphan_add(handle
, inode
);
1253 ext3_journal_stop(handle
);
1255 page_cache_release(page
);
1256 if (pos
+ len
> inode
->i_size
)
1257 ext3_truncate_failed_write(inode
);
1259 if (ret
== -ENOSPC
&& ext3_should_retry_alloc(inode
->i_sb
, &retries
))
1266 int ext3_journal_dirty_data(handle_t
*handle
, struct buffer_head
*bh
)
1268 int err
= journal_dirty_data(handle
, bh
);
1270 ext3_journal_abort_handle(__func__
, __func__
,
1275 /* For ordered writepage and write_end functions */
1276 static int journal_dirty_data_fn(handle_t
*handle
, struct buffer_head
*bh
)
1279 * Write could have mapped the buffer but it didn't copy the data in
1280 * yet. So avoid filing such buffer into a transaction.
1282 if (buffer_mapped(bh
) && buffer_uptodate(bh
))
1283 return ext3_journal_dirty_data(handle
, bh
);
1287 /* For write_end() in data=journal mode */
1288 static int write_end_fn(handle_t
*handle
, struct buffer_head
*bh
)
1290 if (!buffer_mapped(bh
) || buffer_freed(bh
))
1292 set_buffer_uptodate(bh
);
1293 return ext3_journal_dirty_metadata(handle
, bh
);
1297 * This is nasty and subtle: ext3_write_begin() could have allocated blocks
1298 * for the whole page but later we failed to copy the data in. Update inode
1299 * size according to what we managed to copy. The rest is going to be
1300 * truncated in write_end function.
1302 static void update_file_sizes(struct inode
*inode
, loff_t pos
, unsigned copied
)
1304 /* What matters to us is i_disksize. We don't write i_size anywhere */
1305 if (pos
+ copied
> inode
->i_size
)
1306 i_size_write(inode
, pos
+ copied
);
1307 if (pos
+ copied
> EXT3_I(inode
)->i_disksize
) {
1308 EXT3_I(inode
)->i_disksize
= pos
+ copied
;
1309 mark_inode_dirty(inode
);
1314 * We need to pick up the new inode size which generic_commit_write gave us
1315 * `file' can be NULL - eg, when called from page_symlink().
1317 * ext3 never places buffers on inode->i_mapping->private_list. metadata
1318 * buffers are managed internally.
1320 static int ext3_ordered_write_end(struct file
*file
,
1321 struct address_space
*mapping
,
1322 loff_t pos
, unsigned len
, unsigned copied
,
1323 struct page
*page
, void *fsdata
)
1325 handle_t
*handle
= ext3_journal_current_handle();
1326 struct inode
*inode
= file
->f_mapping
->host
;
1330 copied
= block_write_end(file
, mapping
, pos
, len
, copied
, page
, fsdata
);
1332 from
= pos
& (PAGE_CACHE_SIZE
- 1);
1334 ret
= walk_page_buffers(handle
, page_buffers(page
),
1335 from
, to
, NULL
, journal_dirty_data_fn
);
1338 update_file_sizes(inode
, pos
, copied
);
1340 * There may be allocated blocks outside of i_size because
1341 * we failed to copy some data. Prepare for truncate.
1343 if (pos
+ len
> inode
->i_size
&& ext3_can_truncate(inode
))
1344 ext3_orphan_add(handle
, inode
);
1345 ret2
= ext3_journal_stop(handle
);
1349 page_cache_release(page
);
1351 if (pos
+ len
> inode
->i_size
)
1352 ext3_truncate_failed_write(inode
);
1353 return ret
? ret
: copied
;
1356 static int ext3_writeback_write_end(struct file
*file
,
1357 struct address_space
*mapping
,
1358 loff_t pos
, unsigned len
, unsigned copied
,
1359 struct page
*page
, void *fsdata
)
1361 handle_t
*handle
= ext3_journal_current_handle();
1362 struct inode
*inode
= file
->f_mapping
->host
;
1365 copied
= block_write_end(file
, mapping
, pos
, len
, copied
, page
, fsdata
);
1366 update_file_sizes(inode
, pos
, copied
);
1368 * There may be allocated blocks outside of i_size because
1369 * we failed to copy some data. Prepare for truncate.
1371 if (pos
+ len
> inode
->i_size
&& ext3_can_truncate(inode
))
1372 ext3_orphan_add(handle
, inode
);
1373 ret
= ext3_journal_stop(handle
);
1375 page_cache_release(page
);
1377 if (pos
+ len
> inode
->i_size
)
1378 ext3_truncate_failed_write(inode
);
1379 return ret
? ret
: copied
;
1382 static int ext3_journalled_write_end(struct file
*file
,
1383 struct address_space
*mapping
,
1384 loff_t pos
, unsigned len
, unsigned copied
,
1385 struct page
*page
, void *fsdata
)
1387 handle_t
*handle
= ext3_journal_current_handle();
1388 struct inode
*inode
= mapping
->host
;
1393 from
= pos
& (PAGE_CACHE_SIZE
- 1);
1397 if (!PageUptodate(page
))
1399 page_zero_new_buffers(page
, from
+ copied
, to
);
1403 ret
= walk_page_buffers(handle
, page_buffers(page
), from
,
1404 to
, &partial
, write_end_fn
);
1406 SetPageUptodate(page
);
1408 if (pos
+ copied
> inode
->i_size
)
1409 i_size_write(inode
, pos
+ copied
);
1411 * There may be allocated blocks outside of i_size because
1412 * we failed to copy some data. Prepare for truncate.
1414 if (pos
+ len
> inode
->i_size
&& ext3_can_truncate(inode
))
1415 ext3_orphan_add(handle
, inode
);
1416 ext3_set_inode_state(inode
, EXT3_STATE_JDATA
);
1417 if (inode
->i_size
> EXT3_I(inode
)->i_disksize
) {
1418 EXT3_I(inode
)->i_disksize
= inode
->i_size
;
1419 ret2
= ext3_mark_inode_dirty(handle
, inode
);
1424 ret2
= ext3_journal_stop(handle
);
1428 page_cache_release(page
);
1430 if (pos
+ len
> inode
->i_size
)
1431 ext3_truncate_failed_write(inode
);
1432 return ret
? ret
: copied
;
1436 * bmap() is special. It gets used by applications such as lilo and by
1437 * the swapper to find the on-disk block of a specific piece of data.
1439 * Naturally, this is dangerous if the block concerned is still in the
1440 * journal. If somebody makes a swapfile on an ext3 data-journaling
1441 * filesystem and enables swap, then they may get a nasty shock when the
1442 * data getting swapped to that swapfile suddenly gets overwritten by
1443 * the original zero's written out previously to the journal and
1444 * awaiting writeback in the kernel's buffer cache.
1446 * So, if we see any bmap calls here on a modified, data-journaled file,
1447 * take extra steps to flush any blocks which might be in the cache.
1449 static sector_t
ext3_bmap(struct address_space
*mapping
, sector_t block
)
1451 struct inode
*inode
= mapping
->host
;
1455 if (ext3_test_inode_state(inode
, EXT3_STATE_JDATA
)) {
1457 * This is a REALLY heavyweight approach, but the use of
1458 * bmap on dirty files is expected to be extremely rare:
1459 * only if we run lilo or swapon on a freshly made file
1460 * do we expect this to happen.
1462 * (bmap requires CAP_SYS_RAWIO so this does not
1463 * represent an unprivileged user DOS attack --- we'd be
1464 * in trouble if mortal users could trigger this path at
1467 * NB. EXT3_STATE_JDATA is not set on files other than
1468 * regular files. If somebody wants to bmap a directory
1469 * or symlink and gets confused because the buffer
1470 * hasn't yet been flushed to disk, they deserve
1471 * everything they get.
1474 ext3_clear_inode_state(inode
, EXT3_STATE_JDATA
);
1475 journal
= EXT3_JOURNAL(inode
);
1476 journal_lock_updates(journal
);
1477 err
= journal_flush(journal
);
1478 journal_unlock_updates(journal
);
1484 return generic_block_bmap(mapping
,block
,ext3_get_block
);
1487 static int bget_one(handle_t
*handle
, struct buffer_head
*bh
)
1493 static int bput_one(handle_t
*handle
, struct buffer_head
*bh
)
1499 static int buffer_unmapped(handle_t
*handle
, struct buffer_head
*bh
)
1501 return !buffer_mapped(bh
);
1504 static int ext3_ordered_writepage(struct page
*page
,
1505 struct writeback_control
*wbc
)
1507 struct inode
*inode
= page
->mapping
->host
;
1508 struct buffer_head
*page_bufs
;
1509 handle_t
*handle
= NULL
;
1513 J_ASSERT(PageLocked(page
));
1514 WARN_ON_ONCE(IS_RDONLY(inode
));
1517 * We give up here if we're reentered, because it might be for a
1518 * different filesystem.
1520 if (ext3_journal_current_handle())
1523 if (!page_has_buffers(page
)) {
1524 create_empty_buffers(page
, inode
->i_sb
->s_blocksize
,
1525 (1 << BH_Dirty
)|(1 << BH_Uptodate
));
1526 page_bufs
= page_buffers(page
);
1528 page_bufs
= page_buffers(page
);
1529 if (!walk_page_buffers(NULL
, page_bufs
, 0, PAGE_CACHE_SIZE
,
1530 NULL
, buffer_unmapped
)) {
1531 /* Provide NULL get_block() to catch bugs if buffers
1532 * weren't really mapped */
1533 return block_write_full_page(page
, NULL
, wbc
);
1536 handle
= ext3_journal_start(inode
, ext3_writepage_trans_blocks(inode
));
1538 if (IS_ERR(handle
)) {
1539 ret
= PTR_ERR(handle
);
1543 walk_page_buffers(handle
, page_bufs
, 0,
1544 PAGE_CACHE_SIZE
, NULL
, bget_one
);
1546 ret
= block_write_full_page(page
, ext3_get_block
, wbc
);
1549 * The page can become unlocked at any point now, and
1550 * truncate can then come in and change things. So we
1551 * can't touch *page from now on. But *page_bufs is
1552 * safe due to elevated refcount.
1556 * And attach them to the current transaction. But only if
1557 * block_write_full_page() succeeded. Otherwise they are unmapped,
1558 * and generally junk.
1561 err
= walk_page_buffers(handle
, page_bufs
, 0, PAGE_CACHE_SIZE
,
1562 NULL
, journal_dirty_data_fn
);
1566 walk_page_buffers(handle
, page_bufs
, 0,
1567 PAGE_CACHE_SIZE
, NULL
, bput_one
);
1568 err
= ext3_journal_stop(handle
);
1574 redirty_page_for_writepage(wbc
, page
);
1579 static int ext3_writeback_writepage(struct page
*page
,
1580 struct writeback_control
*wbc
)
1582 struct inode
*inode
= page
->mapping
->host
;
1583 handle_t
*handle
= NULL
;
1587 J_ASSERT(PageLocked(page
));
1588 WARN_ON_ONCE(IS_RDONLY(inode
));
1590 if (ext3_journal_current_handle())
1593 if (page_has_buffers(page
)) {
1594 if (!walk_page_buffers(NULL
, page_buffers(page
), 0,
1595 PAGE_CACHE_SIZE
, NULL
, buffer_unmapped
)) {
1596 /* Provide NULL get_block() to catch bugs if buffers
1597 * weren't really mapped */
1598 return block_write_full_page(page
, NULL
, wbc
);
1602 handle
= ext3_journal_start(inode
, ext3_writepage_trans_blocks(inode
));
1603 if (IS_ERR(handle
)) {
1604 ret
= PTR_ERR(handle
);
1608 ret
= block_write_full_page(page
, ext3_get_block
, wbc
);
1610 err
= ext3_journal_stop(handle
);
1616 redirty_page_for_writepage(wbc
, page
);
1621 static int ext3_journalled_writepage(struct page
*page
,
1622 struct writeback_control
*wbc
)
1624 struct inode
*inode
= page
->mapping
->host
;
1625 handle_t
*handle
= NULL
;
1629 J_ASSERT(PageLocked(page
));
1630 WARN_ON_ONCE(IS_RDONLY(inode
));
1632 if (ext3_journal_current_handle())
1635 handle
= ext3_journal_start(inode
, ext3_writepage_trans_blocks(inode
));
1636 if (IS_ERR(handle
)) {
1637 ret
= PTR_ERR(handle
);
1641 if (!page_has_buffers(page
) || PageChecked(page
)) {
1643 * It's mmapped pagecache. Add buffers and journal it. There
1644 * doesn't seem much point in redirtying the page here.
1646 ClearPageChecked(page
);
1647 ret
= block_prepare_write(page
, 0, PAGE_CACHE_SIZE
,
1650 ext3_journal_stop(handle
);
1653 ret
= walk_page_buffers(handle
, page_buffers(page
), 0,
1654 PAGE_CACHE_SIZE
, NULL
, do_journal_get_write_access
);
1656 err
= walk_page_buffers(handle
, page_buffers(page
), 0,
1657 PAGE_CACHE_SIZE
, NULL
, write_end_fn
);
1660 ext3_set_inode_state(inode
, EXT3_STATE_JDATA
);
1664 * It may be a page full of checkpoint-mode buffers. We don't
1665 * really know unless we go poke around in the buffer_heads.
1666 * But block_write_full_page will do the right thing.
1668 ret
= block_write_full_page(page
, ext3_get_block
, wbc
);
1670 err
= ext3_journal_stop(handle
);
1677 redirty_page_for_writepage(wbc
, page
);
1683 static int ext3_readpage(struct file
*file
, struct page
*page
)
1685 return mpage_readpage(page
, ext3_get_block
);
1689 ext3_readpages(struct file
*file
, struct address_space
*mapping
,
1690 struct list_head
*pages
, unsigned nr_pages
)
1692 return mpage_readpages(mapping
, pages
, nr_pages
, ext3_get_block
);
1695 static void ext3_invalidatepage(struct page
*page
, unsigned long offset
)
1697 journal_t
*journal
= EXT3_JOURNAL(page
->mapping
->host
);
1700 * If it's a full truncate we just forget about the pending dirtying
1703 ClearPageChecked(page
);
1705 journal_invalidatepage(journal
, page
, offset
);
1708 static int ext3_releasepage(struct page
*page
, gfp_t wait
)
1710 journal_t
*journal
= EXT3_JOURNAL(page
->mapping
->host
);
1712 WARN_ON(PageChecked(page
));
1713 if (!page_has_buffers(page
))
1715 return journal_try_to_free_buffers(journal
, page
, wait
);
1719 * If the O_DIRECT write will extend the file then add this inode to the
1720 * orphan list. So recovery will truncate it back to the original size
1721 * if the machine crashes during the write.
1723 * If the O_DIRECT write is intantiating holes inside i_size and the machine
1724 * crashes then stale disk data _may_ be exposed inside the file. But current
1725 * VFS code falls back into buffered path in that case so we are safe.
1727 static ssize_t
ext3_direct_IO(int rw
, struct kiocb
*iocb
,
1728 const struct iovec
*iov
, loff_t offset
,
1729 unsigned long nr_segs
)
1731 struct file
*file
= iocb
->ki_filp
;
1732 struct inode
*inode
= file
->f_mapping
->host
;
1733 struct ext3_inode_info
*ei
= EXT3_I(inode
);
1737 size_t count
= iov_length(iov
, nr_segs
);
1741 loff_t final_size
= offset
+ count
;
1743 if (final_size
> inode
->i_size
) {
1744 /* Credits for sb + inode write */
1745 handle
= ext3_journal_start(inode
, 2);
1746 if (IS_ERR(handle
)) {
1747 ret
= PTR_ERR(handle
);
1750 ret
= ext3_orphan_add(handle
, inode
);
1752 ext3_journal_stop(handle
);
1756 ei
->i_disksize
= inode
->i_size
;
1757 ext3_journal_stop(handle
);
1762 ret
= blockdev_direct_IO(rw
, iocb
, inode
, inode
->i_sb
->s_bdev
, iov
,
1764 ext3_get_block
, NULL
);
1766 * In case of error extending write may have instantiated a few
1767 * blocks outside i_size. Trim these off again.
1769 if (unlikely((rw
& WRITE
) && ret
< 0)) {
1770 loff_t isize
= i_size_read(inode
);
1771 loff_t end
= offset
+ iov_length(iov
, nr_segs
);
1774 vmtruncate(inode
, isize
);
1776 if (ret
== -ENOSPC
&& ext3_should_retry_alloc(inode
->i_sb
, &retries
))
1782 /* Credits for sb + inode write */
1783 handle
= ext3_journal_start(inode
, 2);
1784 if (IS_ERR(handle
)) {
1785 /* This is really bad luck. We've written the data
1786 * but cannot extend i_size. Truncate allocated blocks
1787 * and pretend the write failed... */
1788 ext3_truncate(inode
);
1789 ret
= PTR_ERR(handle
);
1793 ext3_orphan_del(handle
, inode
);
1795 loff_t end
= offset
+ ret
;
1796 if (end
> inode
->i_size
) {
1797 ei
->i_disksize
= end
;
1798 i_size_write(inode
, end
);
1800 * We're going to return a positive `ret'
1801 * here due to non-zero-length I/O, so there's
1802 * no way of reporting error returns from
1803 * ext3_mark_inode_dirty() to userspace. So
1806 ext3_mark_inode_dirty(handle
, inode
);
1809 err
= ext3_journal_stop(handle
);
1818 * Pages can be marked dirty completely asynchronously from ext3's journalling
1819 * activity. By filemap_sync_pte(), try_to_unmap_one(), etc. We cannot do
1820 * much here because ->set_page_dirty is called under VFS locks. The page is
1821 * not necessarily locked.
1823 * We cannot just dirty the page and leave attached buffers clean, because the
1824 * buffers' dirty state is "definitive". We cannot just set the buffers dirty
1825 * or jbddirty because all the journalling code will explode.
1827 * So what we do is to mark the page "pending dirty" and next time writepage
1828 * is called, propagate that into the buffers appropriately.
1830 static int ext3_journalled_set_page_dirty(struct page
*page
)
1832 SetPageChecked(page
);
1833 return __set_page_dirty_nobuffers(page
);
1836 static const struct address_space_operations ext3_ordered_aops
= {
1837 .readpage
= ext3_readpage
,
1838 .readpages
= ext3_readpages
,
1839 .writepage
= ext3_ordered_writepage
,
1840 .sync_page
= block_sync_page
,
1841 .write_begin
= ext3_write_begin
,
1842 .write_end
= ext3_ordered_write_end
,
1844 .invalidatepage
= ext3_invalidatepage
,
1845 .releasepage
= ext3_releasepage
,
1846 .direct_IO
= ext3_direct_IO
,
1847 .migratepage
= buffer_migrate_page
,
1848 .is_partially_uptodate
= block_is_partially_uptodate
,
1849 .error_remove_page
= generic_error_remove_page
,
1852 static const struct address_space_operations ext3_writeback_aops
= {
1853 .readpage
= ext3_readpage
,
1854 .readpages
= ext3_readpages
,
1855 .writepage
= ext3_writeback_writepage
,
1856 .sync_page
= block_sync_page
,
1857 .write_begin
= ext3_write_begin
,
1858 .write_end
= ext3_writeback_write_end
,
1860 .invalidatepage
= ext3_invalidatepage
,
1861 .releasepage
= ext3_releasepage
,
1862 .direct_IO
= ext3_direct_IO
,
1863 .migratepage
= buffer_migrate_page
,
1864 .is_partially_uptodate
= block_is_partially_uptodate
,
1865 .error_remove_page
= generic_error_remove_page
,
1868 static const struct address_space_operations ext3_journalled_aops
= {
1869 .readpage
= ext3_readpage
,
1870 .readpages
= ext3_readpages
,
1871 .writepage
= ext3_journalled_writepage
,
1872 .sync_page
= block_sync_page
,
1873 .write_begin
= ext3_write_begin
,
1874 .write_end
= ext3_journalled_write_end
,
1875 .set_page_dirty
= ext3_journalled_set_page_dirty
,
1877 .invalidatepage
= ext3_invalidatepage
,
1878 .releasepage
= ext3_releasepage
,
1879 .is_partially_uptodate
= block_is_partially_uptodate
,
1880 .error_remove_page
= generic_error_remove_page
,
1883 void ext3_set_aops(struct inode
*inode
)
1885 if (ext3_should_order_data(inode
))
1886 inode
->i_mapping
->a_ops
= &ext3_ordered_aops
;
1887 else if (ext3_should_writeback_data(inode
))
1888 inode
->i_mapping
->a_ops
= &ext3_writeback_aops
;
1890 inode
->i_mapping
->a_ops
= &ext3_journalled_aops
;
1894 * ext3_block_truncate_page() zeroes out a mapping from file offset `from'
1895 * up to the end of the block which corresponds to `from'.
1896 * This required during truncate. We need to physically zero the tail end
1897 * of that block so it doesn't yield old data if the file is later grown.
1899 static int ext3_block_truncate_page(handle_t
*handle
, struct page
*page
,
1900 struct address_space
*mapping
, loff_t from
)
1902 ext3_fsblk_t index
= from
>> PAGE_CACHE_SHIFT
;
1903 unsigned offset
= from
& (PAGE_CACHE_SIZE
-1);
1904 unsigned blocksize
, iblock
, length
, pos
;
1905 struct inode
*inode
= mapping
->host
;
1906 struct buffer_head
*bh
;
1909 blocksize
= inode
->i_sb
->s_blocksize
;
1910 length
= blocksize
- (offset
& (blocksize
- 1));
1911 iblock
= index
<< (PAGE_CACHE_SHIFT
- inode
->i_sb
->s_blocksize_bits
);
1913 if (!page_has_buffers(page
))
1914 create_empty_buffers(page
, blocksize
, 0);
1916 /* Find the buffer that contains "offset" */
1917 bh
= page_buffers(page
);
1919 while (offset
>= pos
) {
1920 bh
= bh
->b_this_page
;
1926 if (buffer_freed(bh
)) {
1927 BUFFER_TRACE(bh
, "freed: skip");
1931 if (!buffer_mapped(bh
)) {
1932 BUFFER_TRACE(bh
, "unmapped");
1933 ext3_get_block(inode
, iblock
, bh
, 0);
1934 /* unmapped? It's a hole - nothing to do */
1935 if (!buffer_mapped(bh
)) {
1936 BUFFER_TRACE(bh
, "still unmapped");
1941 /* Ok, it's mapped. Make sure it's up-to-date */
1942 if (PageUptodate(page
))
1943 set_buffer_uptodate(bh
);
1945 if (!buffer_uptodate(bh
)) {
1947 ll_rw_block(READ
, 1, &bh
);
1949 /* Uhhuh. Read error. Complain and punt. */
1950 if (!buffer_uptodate(bh
))
1954 if (ext3_should_journal_data(inode
)) {
1955 BUFFER_TRACE(bh
, "get write access");
1956 err
= ext3_journal_get_write_access(handle
, bh
);
1961 zero_user(page
, offset
, length
);
1962 BUFFER_TRACE(bh
, "zeroed end of block");
1965 if (ext3_should_journal_data(inode
)) {
1966 err
= ext3_journal_dirty_metadata(handle
, bh
);
1968 if (ext3_should_order_data(inode
))
1969 err
= ext3_journal_dirty_data(handle
, bh
);
1970 mark_buffer_dirty(bh
);
1975 page_cache_release(page
);
1980 * Probably it should be a library function... search for first non-zero word
1981 * or memcmp with zero_page, whatever is better for particular architecture.
1984 static inline int all_zeroes(__le32
*p
, __le32
*q
)
1993 * ext3_find_shared - find the indirect blocks for partial truncation.
1994 * @inode: inode in question
1995 * @depth: depth of the affected branch
1996 * @offsets: offsets of pointers in that branch (see ext3_block_to_path)
1997 * @chain: place to store the pointers to partial indirect blocks
1998 * @top: place to the (detached) top of branch
2000 * This is a helper function used by ext3_truncate().
2002 * When we do truncate() we may have to clean the ends of several
2003 * indirect blocks but leave the blocks themselves alive. Block is
2004 * partially truncated if some data below the new i_size is refered
2005 * from it (and it is on the path to the first completely truncated
2006 * data block, indeed). We have to free the top of that path along
2007 * with everything to the right of the path. Since no allocation
2008 * past the truncation point is possible until ext3_truncate()
2009 * finishes, we may safely do the latter, but top of branch may
2010 * require special attention - pageout below the truncation point
2011 * might try to populate it.
2013 * We atomically detach the top of branch from the tree, store the
2014 * block number of its root in *@top, pointers to buffer_heads of
2015 * partially truncated blocks - in @chain[].bh and pointers to
2016 * their last elements that should not be removed - in
2017 * @chain[].p. Return value is the pointer to last filled element
2020 * The work left to caller to do the actual freeing of subtrees:
2021 * a) free the subtree starting from *@top
2022 * b) free the subtrees whose roots are stored in
2023 * (@chain[i].p+1 .. end of @chain[i].bh->b_data)
2024 * c) free the subtrees growing from the inode past the @chain[0].
2025 * (no partially truncated stuff there). */
2027 static Indirect
*ext3_find_shared(struct inode
*inode
, int depth
,
2028 int offsets
[4], Indirect chain
[4], __le32
*top
)
2030 Indirect
*partial
, *p
;
2034 /* Make k index the deepest non-null offset + 1 */
2035 for (k
= depth
; k
> 1 && !offsets
[k
-1]; k
--)
2037 partial
= ext3_get_branch(inode
, k
, offsets
, chain
, &err
);
2038 /* Writer: pointers */
2040 partial
= chain
+ k
-1;
2042 * If the branch acquired continuation since we've looked at it -
2043 * fine, it should all survive and (new) top doesn't belong to us.
2045 if (!partial
->key
&& *partial
->p
)
2048 for (p
=partial
; p
>chain
&& all_zeroes((__le32
*)p
->bh
->b_data
,p
->p
); p
--)
2051 * OK, we've found the last block that must survive. The rest of our
2052 * branch should be detached before unlocking. However, if that rest
2053 * of branch is all ours and does not grow immediately from the inode
2054 * it's easier to cheat and just decrement partial->p.
2056 if (p
== chain
+ k
- 1 && p
> chain
) {
2060 /* Nope, don't do this in ext3. Must leave the tree intact */
2064 while(partial
> p
) {
2065 brelse(partial
->bh
);
2073 * Zero a number of block pointers in either an inode or an indirect block.
2074 * If we restart the transaction we must again get write access to the
2075 * indirect block for further modification.
2077 * We release `count' blocks on disk, but (last - first) may be greater
2078 * than `count' because there can be holes in there.
2080 static void ext3_clear_blocks(handle_t
*handle
, struct inode
*inode
,
2081 struct buffer_head
*bh
, ext3_fsblk_t block_to_free
,
2082 unsigned long count
, __le32
*first
, __le32
*last
)
2085 if (try_to_extend_transaction(handle
, inode
)) {
2087 BUFFER_TRACE(bh
, "call ext3_journal_dirty_metadata");
2088 ext3_journal_dirty_metadata(handle
, bh
);
2090 ext3_mark_inode_dirty(handle
, inode
);
2091 truncate_restart_transaction(handle
, inode
);
2093 BUFFER_TRACE(bh
, "retaking write access");
2094 ext3_journal_get_write_access(handle
, bh
);
2099 * Any buffers which are on the journal will be in memory. We find
2100 * them on the hash table so journal_revoke() will run journal_forget()
2101 * on them. We've already detached each block from the file, so
2102 * bforget() in journal_forget() should be safe.
2104 * AKPM: turn on bforget in journal_forget()!!!
2106 for (p
= first
; p
< last
; p
++) {
2107 u32 nr
= le32_to_cpu(*p
);
2109 struct buffer_head
*bh
;
2112 bh
= sb_find_get_block(inode
->i_sb
, nr
);
2113 ext3_forget(handle
, 0, inode
, bh
, nr
);
2117 ext3_free_blocks(handle
, inode
, block_to_free
, count
);
2121 * ext3_free_data - free a list of data blocks
2122 * @handle: handle for this transaction
2123 * @inode: inode we are dealing with
2124 * @this_bh: indirect buffer_head which contains *@first and *@last
2125 * @first: array of block numbers
2126 * @last: points immediately past the end of array
2128 * We are freeing all blocks refered from that array (numbers are stored as
2129 * little-endian 32-bit) and updating @inode->i_blocks appropriately.
2131 * We accumulate contiguous runs of blocks to free. Conveniently, if these
2132 * blocks are contiguous then releasing them at one time will only affect one
2133 * or two bitmap blocks (+ group descriptor(s) and superblock) and we won't
2134 * actually use a lot of journal space.
2136 * @this_bh will be %NULL if @first and @last point into the inode's direct
2139 static void ext3_free_data(handle_t
*handle
, struct inode
*inode
,
2140 struct buffer_head
*this_bh
,
2141 __le32
*first
, __le32
*last
)
2143 ext3_fsblk_t block_to_free
= 0; /* Starting block # of a run */
2144 unsigned long count
= 0; /* Number of blocks in the run */
2145 __le32
*block_to_free_p
= NULL
; /* Pointer into inode/ind
2148 ext3_fsblk_t nr
; /* Current block # */
2149 __le32
*p
; /* Pointer into inode/ind
2150 for current block */
2153 if (this_bh
) { /* For indirect block */
2154 BUFFER_TRACE(this_bh
, "get_write_access");
2155 err
= ext3_journal_get_write_access(handle
, this_bh
);
2156 /* Important: if we can't update the indirect pointers
2157 * to the blocks, we can't free them. */
2162 for (p
= first
; p
< last
; p
++) {
2163 nr
= le32_to_cpu(*p
);
2165 /* accumulate blocks to free if they're contiguous */
2168 block_to_free_p
= p
;
2170 } else if (nr
== block_to_free
+ count
) {
2173 ext3_clear_blocks(handle
, inode
, this_bh
,
2175 count
, block_to_free_p
, p
);
2177 block_to_free_p
= p
;
2184 ext3_clear_blocks(handle
, inode
, this_bh
, block_to_free
,
2185 count
, block_to_free_p
, p
);
2188 BUFFER_TRACE(this_bh
, "call ext3_journal_dirty_metadata");
2191 * The buffer head should have an attached journal head at this
2192 * point. However, if the data is corrupted and an indirect
2193 * block pointed to itself, it would have been detached when
2194 * the block was cleared. Check for this instead of OOPSing.
2197 ext3_journal_dirty_metadata(handle
, this_bh
);
2199 ext3_error(inode
->i_sb
, "ext3_free_data",
2200 "circular indirect block detected, "
2201 "inode=%lu, block=%llu",
2203 (unsigned long long)this_bh
->b_blocknr
);
2208 * ext3_free_branches - free an array of branches
2209 * @handle: JBD handle for this transaction
2210 * @inode: inode we are dealing with
2211 * @parent_bh: the buffer_head which contains *@first and *@last
2212 * @first: array of block numbers
2213 * @last: pointer immediately past the end of array
2214 * @depth: depth of the branches to free
2216 * We are freeing all blocks refered from these branches (numbers are
2217 * stored as little-endian 32-bit) and updating @inode->i_blocks
2220 static void ext3_free_branches(handle_t
*handle
, struct inode
*inode
,
2221 struct buffer_head
*parent_bh
,
2222 __le32
*first
, __le32
*last
, int depth
)
2227 if (is_handle_aborted(handle
))
2231 struct buffer_head
*bh
;
2232 int addr_per_block
= EXT3_ADDR_PER_BLOCK(inode
->i_sb
);
2234 while (--p
>= first
) {
2235 nr
= le32_to_cpu(*p
);
2237 continue; /* A hole */
2239 /* Go read the buffer for the next level down */
2240 bh
= sb_bread(inode
->i_sb
, nr
);
2243 * A read failure? Report error and clear slot
2247 ext3_error(inode
->i_sb
, "ext3_free_branches",
2248 "Read failure, inode=%lu, block="E3FSBLK
,
2253 /* This zaps the entire block. Bottom up. */
2254 BUFFER_TRACE(bh
, "free child branches");
2255 ext3_free_branches(handle
, inode
, bh
,
2256 (__le32
*)bh
->b_data
,
2257 (__le32
*)bh
->b_data
+ addr_per_block
,
2261 * Everything below this this pointer has been
2262 * released. Now let this top-of-subtree go.
2264 * We want the freeing of this indirect block to be
2265 * atomic in the journal with the updating of the
2266 * bitmap block which owns it. So make some room in
2269 * We zero the parent pointer *after* freeing its
2270 * pointee in the bitmaps, so if extend_transaction()
2271 * for some reason fails to put the bitmap changes and
2272 * the release into the same transaction, recovery
2273 * will merely complain about releasing a free block,
2274 * rather than leaking blocks.
2276 if (is_handle_aborted(handle
))
2278 if (try_to_extend_transaction(handle
, inode
)) {
2279 ext3_mark_inode_dirty(handle
, inode
);
2280 truncate_restart_transaction(handle
, inode
);
2284 * We've probably journalled the indirect block several
2285 * times during the truncate. But it's no longer
2286 * needed and we now drop it from the transaction via
2289 * That's easy if it's exclusively part of this
2290 * transaction. But if it's part of the committing
2291 * transaction then journal_forget() will simply
2292 * brelse() it. That means that if the underlying
2293 * block is reallocated in ext3_get_block(),
2294 * unmap_underlying_metadata() will find this block
2295 * and will try to get rid of it. damn, damn. Thus
2296 * we don't allow a block to be reallocated until
2297 * a transaction freeing it has fully committed.
2299 * We also have to make sure journal replay after a
2300 * crash does not overwrite non-journaled data blocks
2301 * with old metadata when the block got reallocated for
2302 * data. Thus we have to store a revoke record for a
2303 * block in the same transaction in which we free the
2306 ext3_forget(handle
, 1, inode
, bh
, bh
->b_blocknr
);
2308 ext3_free_blocks(handle
, inode
, nr
, 1);
2312 * The block which we have just freed is
2313 * pointed to by an indirect block: journal it
2315 BUFFER_TRACE(parent_bh
, "get_write_access");
2316 if (!ext3_journal_get_write_access(handle
,
2319 BUFFER_TRACE(parent_bh
,
2320 "call ext3_journal_dirty_metadata");
2321 ext3_journal_dirty_metadata(handle
,
2327 /* We have reached the bottom of the tree. */
2328 BUFFER_TRACE(parent_bh
, "free data blocks");
2329 ext3_free_data(handle
, inode
, parent_bh
, first
, last
);
2333 int ext3_can_truncate(struct inode
*inode
)
2335 if (IS_APPEND(inode
) || IS_IMMUTABLE(inode
))
2337 if (S_ISREG(inode
->i_mode
))
2339 if (S_ISDIR(inode
->i_mode
))
2341 if (S_ISLNK(inode
->i_mode
))
2342 return !ext3_inode_is_fast_symlink(inode
);
2349 * We block out ext3_get_block() block instantiations across the entire
2350 * transaction, and VFS/VM ensures that ext3_truncate() cannot run
2351 * simultaneously on behalf of the same inode.
2353 * As we work through the truncate and commmit bits of it to the journal there
2354 * is one core, guiding principle: the file's tree must always be consistent on
2355 * disk. We must be able to restart the truncate after a crash.
2357 * The file's tree may be transiently inconsistent in memory (although it
2358 * probably isn't), but whenever we close off and commit a journal transaction,
2359 * the contents of (the filesystem + the journal) must be consistent and
2360 * restartable. It's pretty simple, really: bottom up, right to left (although
2361 * left-to-right works OK too).
2363 * Note that at recovery time, journal replay occurs *before* the restart of
2364 * truncate against the orphan inode list.
2366 * The committed inode has the new, desired i_size (which is the same as
2367 * i_disksize in this case). After a crash, ext3_orphan_cleanup() will see
2368 * that this inode's truncate did not complete and it will again call
2369 * ext3_truncate() to have another go. So there will be instantiated blocks
2370 * to the right of the truncation point in a crashed ext3 filesystem. But
2371 * that's fine - as long as they are linked from the inode, the post-crash
2372 * ext3_truncate() run will find them and release them.
2374 void ext3_truncate(struct inode
*inode
)
2377 struct ext3_inode_info
*ei
= EXT3_I(inode
);
2378 __le32
*i_data
= ei
->i_data
;
2379 int addr_per_block
= EXT3_ADDR_PER_BLOCK(inode
->i_sb
);
2380 struct address_space
*mapping
= inode
->i_mapping
;
2387 unsigned blocksize
= inode
->i_sb
->s_blocksize
;
2390 if (!ext3_can_truncate(inode
))
2393 if (inode
->i_size
== 0 && ext3_should_writeback_data(inode
))
2394 ext3_set_inode_state(inode
, EXT3_STATE_FLUSH_ON_CLOSE
);
2397 * We have to lock the EOF page here, because lock_page() nests
2398 * outside journal_start().
2400 if ((inode
->i_size
& (blocksize
- 1)) == 0) {
2401 /* Block boundary? Nothing to do */
2404 page
= grab_cache_page(mapping
,
2405 inode
->i_size
>> PAGE_CACHE_SHIFT
);
2410 handle
= start_transaction(inode
);
2411 if (IS_ERR(handle
)) {
2413 clear_highpage(page
);
2414 flush_dcache_page(page
);
2416 page_cache_release(page
);
2421 last_block
= (inode
->i_size
+ blocksize
-1)
2422 >> EXT3_BLOCK_SIZE_BITS(inode
->i_sb
);
2425 ext3_block_truncate_page(handle
, page
, mapping
, inode
->i_size
);
2427 n
= ext3_block_to_path(inode
, last_block
, offsets
, NULL
);
2429 goto out_stop
; /* error */
2432 * OK. This truncate is going to happen. We add the inode to the
2433 * orphan list, so that if this truncate spans multiple transactions,
2434 * and we crash, we will resume the truncate when the filesystem
2435 * recovers. It also marks the inode dirty, to catch the new size.
2437 * Implication: the file must always be in a sane, consistent
2438 * truncatable state while each transaction commits.
2440 if (ext3_orphan_add(handle
, inode
))
2444 * The orphan list entry will now protect us from any crash which
2445 * occurs before the truncate completes, so it is now safe to propagate
2446 * the new, shorter inode size (held for now in i_size) into the
2447 * on-disk inode. We do this via i_disksize, which is the value which
2448 * ext3 *really* writes onto the disk inode.
2450 ei
->i_disksize
= inode
->i_size
;
2453 * From here we block out all ext3_get_block() callers who want to
2454 * modify the block allocation tree.
2456 mutex_lock(&ei
->truncate_mutex
);
2458 if (n
== 1) { /* direct blocks */
2459 ext3_free_data(handle
, inode
, NULL
, i_data
+offsets
[0],
2460 i_data
+ EXT3_NDIR_BLOCKS
);
2464 partial
= ext3_find_shared(inode
, n
, offsets
, chain
, &nr
);
2465 /* Kill the top of shared branch (not detached) */
2467 if (partial
== chain
) {
2468 /* Shared branch grows from the inode */
2469 ext3_free_branches(handle
, inode
, NULL
,
2470 &nr
, &nr
+1, (chain
+n
-1) - partial
);
2473 * We mark the inode dirty prior to restart,
2474 * and prior to stop. No need for it here.
2477 /* Shared branch grows from an indirect block */
2478 BUFFER_TRACE(partial
->bh
, "get_write_access");
2479 ext3_free_branches(handle
, inode
, partial
->bh
,
2481 partial
->p
+1, (chain
+n
-1) - partial
);
2484 /* Clear the ends of indirect blocks on the shared branch */
2485 while (partial
> chain
) {
2486 ext3_free_branches(handle
, inode
, partial
->bh
, partial
->p
+ 1,
2487 (__le32
*)partial
->bh
->b_data
+addr_per_block
,
2488 (chain
+n
-1) - partial
);
2489 BUFFER_TRACE(partial
->bh
, "call brelse");
2490 brelse (partial
->bh
);
2494 /* Kill the remaining (whole) subtrees */
2495 switch (offsets
[0]) {
2497 nr
= i_data
[EXT3_IND_BLOCK
];
2499 ext3_free_branches(handle
, inode
, NULL
, &nr
, &nr
+1, 1);
2500 i_data
[EXT3_IND_BLOCK
] = 0;
2502 case EXT3_IND_BLOCK
:
2503 nr
= i_data
[EXT3_DIND_BLOCK
];
2505 ext3_free_branches(handle
, inode
, NULL
, &nr
, &nr
+1, 2);
2506 i_data
[EXT3_DIND_BLOCK
] = 0;
2508 case EXT3_DIND_BLOCK
:
2509 nr
= i_data
[EXT3_TIND_BLOCK
];
2511 ext3_free_branches(handle
, inode
, NULL
, &nr
, &nr
+1, 3);
2512 i_data
[EXT3_TIND_BLOCK
] = 0;
2514 case EXT3_TIND_BLOCK
:
2518 ext3_discard_reservation(inode
);
2520 mutex_unlock(&ei
->truncate_mutex
);
2521 inode
->i_mtime
= inode
->i_ctime
= CURRENT_TIME_SEC
;
2522 ext3_mark_inode_dirty(handle
, inode
);
2525 * In a multi-transaction truncate, we only make the final transaction
2532 * If this was a simple ftruncate(), and the file will remain alive
2533 * then we need to clear up the orphan record which we created above.
2534 * However, if this was a real unlink then we were called by
2535 * ext3_evict_inode(), and we allow that function to clean up the
2536 * orphan info for us.
2539 ext3_orphan_del(handle
, inode
);
2541 ext3_journal_stop(handle
);
2545 * Delete the inode from orphan list so that it doesn't stay there
2546 * forever and trigger assertion on umount.
2549 ext3_orphan_del(NULL
, inode
);
2552 static ext3_fsblk_t
ext3_get_inode_block(struct super_block
*sb
,
2553 unsigned long ino
, struct ext3_iloc
*iloc
)
2555 unsigned long block_group
;
2556 unsigned long offset
;
2558 struct ext3_group_desc
*gdp
;
2560 if (!ext3_valid_inum(sb
, ino
)) {
2562 * This error is already checked for in namei.c unless we are
2563 * looking at an NFS filehandle, in which case no error
2569 block_group
= (ino
- 1) / EXT3_INODES_PER_GROUP(sb
);
2570 gdp
= ext3_get_group_desc(sb
, block_group
, NULL
);
2574 * Figure out the offset within the block group inode table
2576 offset
= ((ino
- 1) % EXT3_INODES_PER_GROUP(sb
)) *
2577 EXT3_INODE_SIZE(sb
);
2578 block
= le32_to_cpu(gdp
->bg_inode_table
) +
2579 (offset
>> EXT3_BLOCK_SIZE_BITS(sb
));
2581 iloc
->block_group
= block_group
;
2582 iloc
->offset
= offset
& (EXT3_BLOCK_SIZE(sb
) - 1);
2587 * ext3_get_inode_loc returns with an extra refcount against the inode's
2588 * underlying buffer_head on success. If 'in_mem' is true, we have all
2589 * data in memory that is needed to recreate the on-disk version of this
2592 static int __ext3_get_inode_loc(struct inode
*inode
,
2593 struct ext3_iloc
*iloc
, int in_mem
)
2596 struct buffer_head
*bh
;
2598 block
= ext3_get_inode_block(inode
->i_sb
, inode
->i_ino
, iloc
);
2602 bh
= sb_getblk(inode
->i_sb
, block
);
2604 ext3_error (inode
->i_sb
, "ext3_get_inode_loc",
2605 "unable to read inode block - "
2606 "inode=%lu, block="E3FSBLK
,
2607 inode
->i_ino
, block
);
2610 if (!buffer_uptodate(bh
)) {
2614 * If the buffer has the write error flag, we have failed
2615 * to write out another inode in the same block. In this
2616 * case, we don't have to read the block because we may
2617 * read the old inode data successfully.
2619 if (buffer_write_io_error(bh
) && !buffer_uptodate(bh
))
2620 set_buffer_uptodate(bh
);
2622 if (buffer_uptodate(bh
)) {
2623 /* someone brought it uptodate while we waited */
2629 * If we have all information of the inode in memory and this
2630 * is the only valid inode in the block, we need not read the
2634 struct buffer_head
*bitmap_bh
;
2635 struct ext3_group_desc
*desc
;
2636 int inodes_per_buffer
;
2637 int inode_offset
, i
;
2641 block_group
= (inode
->i_ino
- 1) /
2642 EXT3_INODES_PER_GROUP(inode
->i_sb
);
2643 inodes_per_buffer
= bh
->b_size
/
2644 EXT3_INODE_SIZE(inode
->i_sb
);
2645 inode_offset
= ((inode
->i_ino
- 1) %
2646 EXT3_INODES_PER_GROUP(inode
->i_sb
));
2647 start
= inode_offset
& ~(inodes_per_buffer
- 1);
2649 /* Is the inode bitmap in cache? */
2650 desc
= ext3_get_group_desc(inode
->i_sb
,
2655 bitmap_bh
= sb_getblk(inode
->i_sb
,
2656 le32_to_cpu(desc
->bg_inode_bitmap
));
2661 * If the inode bitmap isn't in cache then the
2662 * optimisation may end up performing two reads instead
2663 * of one, so skip it.
2665 if (!buffer_uptodate(bitmap_bh
)) {
2669 for (i
= start
; i
< start
+ inodes_per_buffer
; i
++) {
2670 if (i
== inode_offset
)
2672 if (ext3_test_bit(i
, bitmap_bh
->b_data
))
2676 if (i
== start
+ inodes_per_buffer
) {
2677 /* all other inodes are free, so skip I/O */
2678 memset(bh
->b_data
, 0, bh
->b_size
);
2679 set_buffer_uptodate(bh
);
2687 * There are other valid inodes in the buffer, this inode
2688 * has in-inode xattrs, or we don't have this inode in memory.
2689 * Read the block from disk.
2692 bh
->b_end_io
= end_buffer_read_sync
;
2693 submit_bh(READ_META
, bh
);
2695 if (!buffer_uptodate(bh
)) {
2696 ext3_error(inode
->i_sb
, "ext3_get_inode_loc",
2697 "unable to read inode block - "
2698 "inode=%lu, block="E3FSBLK
,
2699 inode
->i_ino
, block
);
2709 int ext3_get_inode_loc(struct inode
*inode
, struct ext3_iloc
*iloc
)
2711 /* We have all inode data except xattrs in memory here. */
2712 return __ext3_get_inode_loc(inode
, iloc
,
2713 !ext3_test_inode_state(inode
, EXT3_STATE_XATTR
));
2716 void ext3_set_inode_flags(struct inode
*inode
)
2718 unsigned int flags
= EXT3_I(inode
)->i_flags
;
2720 inode
->i_flags
&= ~(S_SYNC
|S_APPEND
|S_IMMUTABLE
|S_NOATIME
|S_DIRSYNC
);
2721 if (flags
& EXT3_SYNC_FL
)
2722 inode
->i_flags
|= S_SYNC
;
2723 if (flags
& EXT3_APPEND_FL
)
2724 inode
->i_flags
|= S_APPEND
;
2725 if (flags
& EXT3_IMMUTABLE_FL
)
2726 inode
->i_flags
|= S_IMMUTABLE
;
2727 if (flags
& EXT3_NOATIME_FL
)
2728 inode
->i_flags
|= S_NOATIME
;
2729 if (flags
& EXT3_DIRSYNC_FL
)
2730 inode
->i_flags
|= S_DIRSYNC
;
2733 /* Propagate flags from i_flags to EXT3_I(inode)->i_flags */
2734 void ext3_get_inode_flags(struct ext3_inode_info
*ei
)
2736 unsigned int flags
= ei
->vfs_inode
.i_flags
;
2738 ei
->i_flags
&= ~(EXT3_SYNC_FL
|EXT3_APPEND_FL
|
2739 EXT3_IMMUTABLE_FL
|EXT3_NOATIME_FL
|EXT3_DIRSYNC_FL
);
2741 ei
->i_flags
|= EXT3_SYNC_FL
;
2742 if (flags
& S_APPEND
)
2743 ei
->i_flags
|= EXT3_APPEND_FL
;
2744 if (flags
& S_IMMUTABLE
)
2745 ei
->i_flags
|= EXT3_IMMUTABLE_FL
;
2746 if (flags
& S_NOATIME
)
2747 ei
->i_flags
|= EXT3_NOATIME_FL
;
2748 if (flags
& S_DIRSYNC
)
2749 ei
->i_flags
|= EXT3_DIRSYNC_FL
;
2752 struct inode
*ext3_iget(struct super_block
*sb
, unsigned long ino
)
2754 struct ext3_iloc iloc
;
2755 struct ext3_inode
*raw_inode
;
2756 struct ext3_inode_info
*ei
;
2757 struct buffer_head
*bh
;
2758 struct inode
*inode
;
2759 journal_t
*journal
= EXT3_SB(sb
)->s_journal
;
2760 transaction_t
*transaction
;
2764 inode
= iget_locked(sb
, ino
);
2766 return ERR_PTR(-ENOMEM
);
2767 if (!(inode
->i_state
& I_NEW
))
2771 ei
->i_block_alloc_info
= NULL
;
2773 ret
= __ext3_get_inode_loc(inode
, &iloc
, 0);
2777 raw_inode
= ext3_raw_inode(&iloc
);
2778 inode
->i_mode
= le16_to_cpu(raw_inode
->i_mode
);
2779 inode
->i_uid
= (uid_t
)le16_to_cpu(raw_inode
->i_uid_low
);
2780 inode
->i_gid
= (gid_t
)le16_to_cpu(raw_inode
->i_gid_low
);
2781 if(!(test_opt (inode
->i_sb
, NO_UID32
))) {
2782 inode
->i_uid
|= le16_to_cpu(raw_inode
->i_uid_high
) << 16;
2783 inode
->i_gid
|= le16_to_cpu(raw_inode
->i_gid_high
) << 16;
2785 inode
->i_nlink
= le16_to_cpu(raw_inode
->i_links_count
);
2786 inode
->i_size
= le32_to_cpu(raw_inode
->i_size
);
2787 inode
->i_atime
.tv_sec
= (signed)le32_to_cpu(raw_inode
->i_atime
);
2788 inode
->i_ctime
.tv_sec
= (signed)le32_to_cpu(raw_inode
->i_ctime
);
2789 inode
->i_mtime
.tv_sec
= (signed)le32_to_cpu(raw_inode
->i_mtime
);
2790 inode
->i_atime
.tv_nsec
= inode
->i_ctime
.tv_nsec
= inode
->i_mtime
.tv_nsec
= 0;
2792 ei
->i_state_flags
= 0;
2793 ei
->i_dir_start_lookup
= 0;
2794 ei
->i_dtime
= le32_to_cpu(raw_inode
->i_dtime
);
2795 /* We now have enough fields to check if the inode was active or not.
2796 * This is needed because nfsd might try to access dead inodes
2797 * the test is that same one that e2fsck uses
2798 * NeilBrown 1999oct15
2800 if (inode
->i_nlink
== 0) {
2801 if (inode
->i_mode
== 0 ||
2802 !(EXT3_SB(inode
->i_sb
)->s_mount_state
& EXT3_ORPHAN_FS
)) {
2803 /* this inode is deleted */
2808 /* The only unlinked inodes we let through here have
2809 * valid i_mode and are being read by the orphan
2810 * recovery code: that's fine, we're about to complete
2811 * the process of deleting those. */
2813 inode
->i_blocks
= le32_to_cpu(raw_inode
->i_blocks
);
2814 ei
->i_flags
= le32_to_cpu(raw_inode
->i_flags
);
2815 #ifdef EXT3_FRAGMENTS
2816 ei
->i_faddr
= le32_to_cpu(raw_inode
->i_faddr
);
2817 ei
->i_frag_no
= raw_inode
->i_frag
;
2818 ei
->i_frag_size
= raw_inode
->i_fsize
;
2820 ei
->i_file_acl
= le32_to_cpu(raw_inode
->i_file_acl
);
2821 if (!S_ISREG(inode
->i_mode
)) {
2822 ei
->i_dir_acl
= le32_to_cpu(raw_inode
->i_dir_acl
);
2825 ((__u64
)le32_to_cpu(raw_inode
->i_size_high
)) << 32;
2827 ei
->i_disksize
= inode
->i_size
;
2828 inode
->i_generation
= le32_to_cpu(raw_inode
->i_generation
);
2829 ei
->i_block_group
= iloc
.block_group
;
2831 * NOTE! The in-memory inode i_data array is in little-endian order
2832 * even on big-endian machines: we do NOT byteswap the block numbers!
2834 for (block
= 0; block
< EXT3_N_BLOCKS
; block
++)
2835 ei
->i_data
[block
] = raw_inode
->i_block
[block
];
2836 INIT_LIST_HEAD(&ei
->i_orphan
);
2839 * Set transaction id's of transactions that have to be committed
2840 * to finish f[data]sync. We set them to currently running transaction
2841 * as we cannot be sure that the inode or some of its metadata isn't
2842 * part of the transaction - the inode could have been reclaimed and
2843 * now it is reread from disk.
2848 spin_lock(&journal
->j_state_lock
);
2849 if (journal
->j_running_transaction
)
2850 transaction
= journal
->j_running_transaction
;
2852 transaction
= journal
->j_committing_transaction
;
2854 tid
= transaction
->t_tid
;
2856 tid
= journal
->j_commit_sequence
;
2857 spin_unlock(&journal
->j_state_lock
);
2858 atomic_set(&ei
->i_sync_tid
, tid
);
2859 atomic_set(&ei
->i_datasync_tid
, tid
);
2862 if (inode
->i_ino
>= EXT3_FIRST_INO(inode
->i_sb
) + 1 &&
2863 EXT3_INODE_SIZE(inode
->i_sb
) > EXT3_GOOD_OLD_INODE_SIZE
) {
2865 * When mke2fs creates big inodes it does not zero out
2866 * the unused bytes above EXT3_GOOD_OLD_INODE_SIZE,
2867 * so ignore those first few inodes.
2869 ei
->i_extra_isize
= le16_to_cpu(raw_inode
->i_extra_isize
);
2870 if (EXT3_GOOD_OLD_INODE_SIZE
+ ei
->i_extra_isize
>
2871 EXT3_INODE_SIZE(inode
->i_sb
)) {
2876 if (ei
->i_extra_isize
== 0) {
2877 /* The extra space is currently unused. Use it. */
2878 ei
->i_extra_isize
= sizeof(struct ext3_inode
) -
2879 EXT3_GOOD_OLD_INODE_SIZE
;
2881 __le32
*magic
= (void *)raw_inode
+
2882 EXT3_GOOD_OLD_INODE_SIZE
+
2884 if (*magic
== cpu_to_le32(EXT3_XATTR_MAGIC
))
2885 ext3_set_inode_state(inode
, EXT3_STATE_XATTR
);
2888 ei
->i_extra_isize
= 0;
2890 if (S_ISREG(inode
->i_mode
)) {
2891 inode
->i_op
= &ext3_file_inode_operations
;
2892 inode
->i_fop
= &ext3_file_operations
;
2893 ext3_set_aops(inode
);
2894 } else if (S_ISDIR(inode
->i_mode
)) {
2895 inode
->i_op
= &ext3_dir_inode_operations
;
2896 inode
->i_fop
= &ext3_dir_operations
;
2897 } else if (S_ISLNK(inode
->i_mode
)) {
2898 if (ext3_inode_is_fast_symlink(inode
)) {
2899 inode
->i_op
= &ext3_fast_symlink_inode_operations
;
2900 nd_terminate_link(ei
->i_data
, inode
->i_size
,
2901 sizeof(ei
->i_data
) - 1);
2903 inode
->i_op
= &ext3_symlink_inode_operations
;
2904 ext3_set_aops(inode
);
2907 inode
->i_op
= &ext3_special_inode_operations
;
2908 if (raw_inode
->i_block
[0])
2909 init_special_inode(inode
, inode
->i_mode
,
2910 old_decode_dev(le32_to_cpu(raw_inode
->i_block
[0])));
2912 init_special_inode(inode
, inode
->i_mode
,
2913 new_decode_dev(le32_to_cpu(raw_inode
->i_block
[1])));
2916 ext3_set_inode_flags(inode
);
2917 unlock_new_inode(inode
);
2922 return ERR_PTR(ret
);
2926 * Post the struct inode info into an on-disk inode location in the
2927 * buffer-cache. This gobbles the caller's reference to the
2928 * buffer_head in the inode location struct.
2930 * The caller must have write access to iloc->bh.
2932 static int ext3_do_update_inode(handle_t
*handle
,
2933 struct inode
*inode
,
2934 struct ext3_iloc
*iloc
)
2936 struct ext3_inode
*raw_inode
= ext3_raw_inode(iloc
);
2937 struct ext3_inode_info
*ei
= EXT3_I(inode
);
2938 struct buffer_head
*bh
= iloc
->bh
;
2939 int err
= 0, rc
, block
;
2942 /* we can't allow multiple procs in here at once, its a bit racey */
2945 /* For fields not not tracking in the in-memory inode,
2946 * initialise them to zero for new inodes. */
2947 if (ext3_test_inode_state(inode
, EXT3_STATE_NEW
))
2948 memset(raw_inode
, 0, EXT3_SB(inode
->i_sb
)->s_inode_size
);
2950 ext3_get_inode_flags(ei
);
2951 raw_inode
->i_mode
= cpu_to_le16(inode
->i_mode
);
2952 if(!(test_opt(inode
->i_sb
, NO_UID32
))) {
2953 raw_inode
->i_uid_low
= cpu_to_le16(low_16_bits(inode
->i_uid
));
2954 raw_inode
->i_gid_low
= cpu_to_le16(low_16_bits(inode
->i_gid
));
2956 * Fix up interoperability with old kernels. Otherwise, old inodes get
2957 * re-used with the upper 16 bits of the uid/gid intact
2960 raw_inode
->i_uid_high
=
2961 cpu_to_le16(high_16_bits(inode
->i_uid
));
2962 raw_inode
->i_gid_high
=
2963 cpu_to_le16(high_16_bits(inode
->i_gid
));
2965 raw_inode
->i_uid_high
= 0;
2966 raw_inode
->i_gid_high
= 0;
2969 raw_inode
->i_uid_low
=
2970 cpu_to_le16(fs_high2lowuid(inode
->i_uid
));
2971 raw_inode
->i_gid_low
=
2972 cpu_to_le16(fs_high2lowgid(inode
->i_gid
));
2973 raw_inode
->i_uid_high
= 0;
2974 raw_inode
->i_gid_high
= 0;
2976 raw_inode
->i_links_count
= cpu_to_le16(inode
->i_nlink
);
2977 raw_inode
->i_size
= cpu_to_le32(ei
->i_disksize
);
2978 raw_inode
->i_atime
= cpu_to_le32(inode
->i_atime
.tv_sec
);
2979 raw_inode
->i_ctime
= cpu_to_le32(inode
->i_ctime
.tv_sec
);
2980 raw_inode
->i_mtime
= cpu_to_le32(inode
->i_mtime
.tv_sec
);
2981 raw_inode
->i_blocks
= cpu_to_le32(inode
->i_blocks
);
2982 raw_inode
->i_dtime
= cpu_to_le32(ei
->i_dtime
);
2983 raw_inode
->i_flags
= cpu_to_le32(ei
->i_flags
);
2984 #ifdef EXT3_FRAGMENTS
2985 raw_inode
->i_faddr
= cpu_to_le32(ei
->i_faddr
);
2986 raw_inode
->i_frag
= ei
->i_frag_no
;
2987 raw_inode
->i_fsize
= ei
->i_frag_size
;
2989 raw_inode
->i_file_acl
= cpu_to_le32(ei
->i_file_acl
);
2990 if (!S_ISREG(inode
->i_mode
)) {
2991 raw_inode
->i_dir_acl
= cpu_to_le32(ei
->i_dir_acl
);
2993 raw_inode
->i_size_high
=
2994 cpu_to_le32(ei
->i_disksize
>> 32);
2995 if (ei
->i_disksize
> 0x7fffffffULL
) {
2996 struct super_block
*sb
= inode
->i_sb
;
2997 if (!EXT3_HAS_RO_COMPAT_FEATURE(sb
,
2998 EXT3_FEATURE_RO_COMPAT_LARGE_FILE
) ||
2999 EXT3_SB(sb
)->s_es
->s_rev_level
==
3000 cpu_to_le32(EXT3_GOOD_OLD_REV
)) {
3001 /* If this is the first large file
3002 * created, add a flag to the superblock.
3005 err
= ext3_journal_get_write_access(handle
,
3006 EXT3_SB(sb
)->s_sbh
);
3010 ext3_update_dynamic_rev(sb
);
3011 EXT3_SET_RO_COMPAT_FEATURE(sb
,
3012 EXT3_FEATURE_RO_COMPAT_LARGE_FILE
);
3014 err
= ext3_journal_dirty_metadata(handle
,
3015 EXT3_SB(sb
)->s_sbh
);
3016 /* get our lock and start over */
3021 raw_inode
->i_generation
= cpu_to_le32(inode
->i_generation
);
3022 if (S_ISCHR(inode
->i_mode
) || S_ISBLK(inode
->i_mode
)) {
3023 if (old_valid_dev(inode
->i_rdev
)) {
3024 raw_inode
->i_block
[0] =
3025 cpu_to_le32(old_encode_dev(inode
->i_rdev
));
3026 raw_inode
->i_block
[1] = 0;
3028 raw_inode
->i_block
[0] = 0;
3029 raw_inode
->i_block
[1] =
3030 cpu_to_le32(new_encode_dev(inode
->i_rdev
));
3031 raw_inode
->i_block
[2] = 0;
3033 } else for (block
= 0; block
< EXT3_N_BLOCKS
; block
++)
3034 raw_inode
->i_block
[block
] = ei
->i_data
[block
];
3036 if (ei
->i_extra_isize
)
3037 raw_inode
->i_extra_isize
= cpu_to_le16(ei
->i_extra_isize
);
3039 BUFFER_TRACE(bh
, "call ext3_journal_dirty_metadata");
3041 rc
= ext3_journal_dirty_metadata(handle
, bh
);
3044 ext3_clear_inode_state(inode
, EXT3_STATE_NEW
);
3046 atomic_set(&ei
->i_sync_tid
, handle
->h_transaction
->t_tid
);
3049 ext3_std_error(inode
->i_sb
, err
);
3054 * ext3_write_inode()
3056 * We are called from a few places:
3058 * - Within generic_file_write() for O_SYNC files.
3059 * Here, there will be no transaction running. We wait for any running
3060 * trasnaction to commit.
3062 * - Within sys_sync(), kupdate and such.
3063 * We wait on commit, if tol to.
3065 * - Within prune_icache() (PF_MEMALLOC == true)
3066 * Here we simply return. We can't afford to block kswapd on the
3069 * In all cases it is actually safe for us to return without doing anything,
3070 * because the inode has been copied into a raw inode buffer in
3071 * ext3_mark_inode_dirty(). This is a correctness thing for O_SYNC and for
3074 * Note that we are absolutely dependent upon all inode dirtiers doing the
3075 * right thing: they *must* call mark_inode_dirty() after dirtying info in
3076 * which we are interested.
3078 * It would be a bug for them to not do this. The code:
3080 * mark_inode_dirty(inode)
3082 * inode->i_size = expr;
3084 * is in error because a kswapd-driven write_inode() could occur while
3085 * `stuff()' is running, and the new i_size will be lost. Plus the inode
3086 * will no longer be on the superblock's dirty inode list.
3088 int ext3_write_inode(struct inode
*inode
, struct writeback_control
*wbc
)
3090 if (current
->flags
& PF_MEMALLOC
)
3093 if (ext3_journal_current_handle()) {
3094 jbd_debug(1, "called recursively, non-PF_MEMALLOC!\n");
3099 if (wbc
->sync_mode
!= WB_SYNC_ALL
)
3102 return ext3_force_commit(inode
->i_sb
);
3108 * Called from notify_change.
3110 * We want to trap VFS attempts to truncate the file as soon as
3111 * possible. In particular, we want to make sure that when the VFS
3112 * shrinks i_size, we put the inode on the orphan list and modify
3113 * i_disksize immediately, so that during the subsequent flushing of
3114 * dirty pages and freeing of disk blocks, we can guarantee that any
3115 * commit will leave the blocks being flushed in an unused state on
3116 * disk. (On recovery, the inode will get truncated and the blocks will
3117 * be freed, so we have a strong guarantee that no future commit will
3118 * leave these blocks visible to the user.)
3120 * Called with inode->sem down.
3122 int ext3_setattr(struct dentry
*dentry
, struct iattr
*attr
)
3124 struct inode
*inode
= dentry
->d_inode
;
3126 const unsigned int ia_valid
= attr
->ia_valid
;
3128 error
= inode_change_ok(inode
, attr
);
3132 if (is_quota_modification(inode
, attr
))
3133 dquot_initialize(inode
);
3134 if ((ia_valid
& ATTR_UID
&& attr
->ia_uid
!= inode
->i_uid
) ||
3135 (ia_valid
& ATTR_GID
&& attr
->ia_gid
!= inode
->i_gid
)) {
3138 /* (user+group)*(old+new) structure, inode write (sb,
3139 * inode block, ? - but truncate inode update has it) */
3140 handle
= ext3_journal_start(inode
, EXT3_MAXQUOTAS_INIT_BLOCKS(inode
->i_sb
)+
3141 EXT3_MAXQUOTAS_DEL_BLOCKS(inode
->i_sb
)+3);
3142 if (IS_ERR(handle
)) {
3143 error
= PTR_ERR(handle
);
3146 error
= dquot_transfer(inode
, attr
);
3148 ext3_journal_stop(handle
);
3151 /* Update corresponding info in inode so that everything is in
3152 * one transaction */
3153 if (attr
->ia_valid
& ATTR_UID
)
3154 inode
->i_uid
= attr
->ia_uid
;
3155 if (attr
->ia_valid
& ATTR_GID
)
3156 inode
->i_gid
= attr
->ia_gid
;
3157 error
= ext3_mark_inode_dirty(handle
, inode
);
3158 ext3_journal_stop(handle
);
3161 if (S_ISREG(inode
->i_mode
) &&
3162 attr
->ia_valid
& ATTR_SIZE
&& attr
->ia_size
< inode
->i_size
) {
3165 handle
= ext3_journal_start(inode
, 3);
3166 if (IS_ERR(handle
)) {
3167 error
= PTR_ERR(handle
);
3171 error
= ext3_orphan_add(handle
, inode
);
3172 EXT3_I(inode
)->i_disksize
= attr
->ia_size
;
3173 rc
= ext3_mark_inode_dirty(handle
, inode
);
3176 ext3_journal_stop(handle
);
3179 if ((attr
->ia_valid
& ATTR_SIZE
) &&
3180 attr
->ia_size
!= i_size_read(inode
)) {
3181 rc
= vmtruncate(inode
, attr
->ia_size
);
3186 setattr_copy(inode
, attr
);
3187 mark_inode_dirty(inode
);
3189 if (ia_valid
& ATTR_MODE
)
3190 rc
= ext3_acl_chmod(inode
);
3193 ext3_std_error(inode
->i_sb
, error
);
3201 * How many blocks doth make a writepage()?
3203 * With N blocks per page, it may be:
3208 * N+5 bitmap blocks (from the above)
3209 * N+5 group descriptor summary blocks
3212 * 2 * EXT3_SINGLEDATA_TRANS_BLOCKS for the quote files
3214 * 3 * (N + 5) + 2 + 2 * EXT3_SINGLEDATA_TRANS_BLOCKS
3216 * With ordered or writeback data it's the same, less the N data blocks.
3218 * If the inode's direct blocks can hold an integral number of pages then a
3219 * page cannot straddle two indirect blocks, and we can only touch one indirect
3220 * and dindirect block, and the "5" above becomes "3".
3222 * This still overestimates under most circumstances. If we were to pass the
3223 * start and end offsets in here as well we could do block_to_path() on each
3224 * block and work out the exact number of indirects which are touched. Pah.
3227 static int ext3_writepage_trans_blocks(struct inode
*inode
)
3229 int bpp
= ext3_journal_blocks_per_page(inode
);
3230 int indirects
= (EXT3_NDIR_BLOCKS
% bpp
) ? 5 : 3;
3233 if (ext3_should_journal_data(inode
))
3234 ret
= 3 * (bpp
+ indirects
) + 2;
3236 ret
= 2 * (bpp
+ indirects
) + 2;
3239 /* We know that structure was already allocated during dquot_initialize so
3240 * we will be updating only the data blocks + inodes */
3241 ret
+= EXT3_MAXQUOTAS_TRANS_BLOCKS(inode
->i_sb
);
3248 * The caller must have previously called ext3_reserve_inode_write().
3249 * Give this, we know that the caller already has write access to iloc->bh.
3251 int ext3_mark_iloc_dirty(handle_t
*handle
,
3252 struct inode
*inode
, struct ext3_iloc
*iloc
)
3256 /* the do_update_inode consumes one bh->b_count */
3259 /* ext3_do_update_inode() does journal_dirty_metadata */
3260 err
= ext3_do_update_inode(handle
, inode
, iloc
);
3266 * On success, We end up with an outstanding reference count against
3267 * iloc->bh. This _must_ be cleaned up later.
3271 ext3_reserve_inode_write(handle_t
*handle
, struct inode
*inode
,
3272 struct ext3_iloc
*iloc
)
3276 err
= ext3_get_inode_loc(inode
, iloc
);
3278 BUFFER_TRACE(iloc
->bh
, "get_write_access");
3279 err
= ext3_journal_get_write_access(handle
, iloc
->bh
);
3286 ext3_std_error(inode
->i_sb
, err
);
3291 * What we do here is to mark the in-core inode as clean with respect to inode
3292 * dirtiness (it may still be data-dirty).
3293 * This means that the in-core inode may be reaped by prune_icache
3294 * without having to perform any I/O. This is a very good thing,
3295 * because *any* task may call prune_icache - even ones which
3296 * have a transaction open against a different journal.
3298 * Is this cheating? Not really. Sure, we haven't written the
3299 * inode out, but prune_icache isn't a user-visible syncing function.
3300 * Whenever the user wants stuff synced (sys_sync, sys_msync, sys_fsync)
3301 * we start and wait on commits.
3303 * Is this efficient/effective? Well, we're being nice to the system
3304 * by cleaning up our inodes proactively so they can be reaped
3305 * without I/O. But we are potentially leaving up to five seconds'
3306 * worth of inodes floating about which prune_icache wants us to
3307 * write out. One way to fix that would be to get prune_icache()
3308 * to do a write_super() to free up some memory. It has the desired
3311 int ext3_mark_inode_dirty(handle_t
*handle
, struct inode
*inode
)
3313 struct ext3_iloc iloc
;
3317 err
= ext3_reserve_inode_write(handle
, inode
, &iloc
);
3319 err
= ext3_mark_iloc_dirty(handle
, inode
, &iloc
);
3324 * ext3_dirty_inode() is called from __mark_inode_dirty()
3326 * We're really interested in the case where a file is being extended.
3327 * i_size has been changed by generic_commit_write() and we thus need
3328 * to include the updated inode in the current transaction.
3330 * Also, dquot_alloc_space() will always dirty the inode when blocks
3331 * are allocated to the file.
3333 * If the inode is marked synchronous, we don't honour that here - doing
3334 * so would cause a commit on atime updates, which we don't bother doing.
3335 * We handle synchronous inodes at the highest possible level.
3337 void ext3_dirty_inode(struct inode
*inode
)
3339 handle_t
*current_handle
= ext3_journal_current_handle();
3342 handle
= ext3_journal_start(inode
, 2);
3345 if (current_handle
&&
3346 current_handle
->h_transaction
!= handle
->h_transaction
) {
3347 /* This task has a transaction open against a different fs */
3348 printk(KERN_EMERG
"%s: transactions do not match!\n",
3351 jbd_debug(5, "marking dirty. outer handle=%p\n",
3353 ext3_mark_inode_dirty(handle
, inode
);
3355 ext3_journal_stop(handle
);
3361 int ext3_change_inode_journal_flag(struct inode
*inode
, int val
)
3368 * We have to be very careful here: changing a data block's
3369 * journaling status dynamically is dangerous. If we write a
3370 * data block to the journal, change the status and then delete
3371 * that block, we risk forgetting to revoke the old log record
3372 * from the journal and so a subsequent replay can corrupt data.
3373 * So, first we make sure that the journal is empty and that
3374 * nobody is changing anything.
3377 journal
= EXT3_JOURNAL(inode
);
3378 if (is_journal_aborted(journal
))
3381 journal_lock_updates(journal
);
3382 journal_flush(journal
);
3385 * OK, there are no updates running now, and all cached data is
3386 * synced to disk. We are now in a completely consistent state
3387 * which doesn't have anything in the journal, and we know that
3388 * no filesystem updates are running, so it is safe to modify
3389 * the inode's in-core data-journaling state flag now.
3393 EXT3_I(inode
)->i_flags
|= EXT3_JOURNAL_DATA_FL
;
3395 EXT3_I(inode
)->i_flags
&= ~EXT3_JOURNAL_DATA_FL
;
3396 ext3_set_aops(inode
);
3398 journal_unlock_updates(journal
);
3400 /* Finally we can mark the inode as dirty. */
3402 handle
= ext3_journal_start(inode
, 1);
3404 return PTR_ERR(handle
);
3406 err
= ext3_mark_inode_dirty(handle
, inode
);
3408 ext3_journal_stop(handle
);
3409 ext3_std_error(inode
->i_sb
, err
);