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>
41 #include <trace/events/ext3.h>
45 static int ext3_writepage_trans_blocks(struct inode
*inode
);
48 * Test whether an inode is a fast symlink.
50 static int ext3_inode_is_fast_symlink(struct inode
*inode
)
52 int ea_blocks
= EXT3_I(inode
)->i_file_acl
?
53 (inode
->i_sb
->s_blocksize
>> 9) : 0;
55 return (S_ISLNK(inode
->i_mode
) && inode
->i_blocks
- ea_blocks
== 0);
59 * The ext3 forget function must perform a revoke if we are freeing data
60 * which has been journaled. Metadata (eg. indirect blocks) must be
61 * revoked in all cases.
63 * "bh" may be NULL: a metadata block may have been freed from memory
64 * but there may still be a record of it in the journal, and that record
65 * still needs to be revoked.
67 int ext3_forget(handle_t
*handle
, int is_metadata
, struct inode
*inode
,
68 struct buffer_head
*bh
, ext3_fsblk_t blocknr
)
74 trace_ext3_forget(inode
, is_metadata
, blocknr
);
75 BUFFER_TRACE(bh
, "enter");
77 jbd_debug(4, "forgetting bh %p: is_metadata = %d, mode %o, "
79 bh
, is_metadata
, inode
->i_mode
,
80 test_opt(inode
->i_sb
, DATA_FLAGS
));
82 /* Never use the revoke function if we are doing full data
83 * journaling: there is no need to, and a V1 superblock won't
84 * support it. Otherwise, only skip the revoke on un-journaled
87 if (test_opt(inode
->i_sb
, DATA_FLAGS
) == EXT3_MOUNT_JOURNAL_DATA
||
88 (!is_metadata
&& !ext3_should_journal_data(inode
))) {
90 BUFFER_TRACE(bh
, "call journal_forget");
91 return ext3_journal_forget(handle
, bh
);
97 * data!=journal && (is_metadata || should_journal_data(inode))
99 BUFFER_TRACE(bh
, "call ext3_journal_revoke");
100 err
= ext3_journal_revoke(handle
, blocknr
, bh
);
102 ext3_abort(inode
->i_sb
, __func__
,
103 "error %d when attempting revoke", err
);
104 BUFFER_TRACE(bh
, "exit");
109 * Work out how many blocks we need to proceed with the next chunk of a
110 * truncate transaction.
112 static unsigned long blocks_for_truncate(struct inode
*inode
)
114 unsigned long needed
;
116 needed
= inode
->i_blocks
>> (inode
->i_sb
->s_blocksize_bits
- 9);
118 /* Give ourselves just enough room to cope with inodes in which
119 * i_blocks is corrupt: we've seen disk corruptions in the past
120 * which resulted in random data in an inode which looked enough
121 * like a regular file for ext3 to try to delete it. Things
122 * will go a bit crazy if that happens, but at least we should
123 * try not to panic the whole kernel. */
127 /* But we need to bound the transaction so we don't overflow the
129 if (needed
> EXT3_MAX_TRANS_DATA
)
130 needed
= EXT3_MAX_TRANS_DATA
;
132 return EXT3_DATA_TRANS_BLOCKS(inode
->i_sb
) + needed
;
136 * Truncate transactions can be complex and absolutely huge. So we need to
137 * be able to restart the transaction at a conventient checkpoint to make
138 * sure we don't overflow the journal.
140 * start_transaction gets us a new handle for a truncate transaction,
141 * and extend_transaction tries to extend the existing one a bit. If
142 * extend fails, we need to propagate the failure up and restart the
143 * transaction in the top-level truncate loop. --sct
145 static handle_t
*start_transaction(struct inode
*inode
)
149 result
= ext3_journal_start(inode
, blocks_for_truncate(inode
));
153 ext3_std_error(inode
->i_sb
, PTR_ERR(result
));
158 * Try to extend this transaction for the purposes of truncation.
160 * Returns 0 if we managed to create more room. If we can't create more
161 * room, and the transaction must be restarted we return 1.
163 static int try_to_extend_transaction(handle_t
*handle
, struct inode
*inode
)
165 if (handle
->h_buffer_credits
> EXT3_RESERVE_TRANS_BLOCKS
)
167 if (!ext3_journal_extend(handle
, blocks_for_truncate(inode
)))
173 * Restart the transaction associated with *handle. This does a commit,
174 * so before we call here everything must be consistently dirtied against
177 static int truncate_restart_transaction(handle_t
*handle
, struct inode
*inode
)
181 jbd_debug(2, "restarting handle %p\n", handle
);
183 * Drop truncate_mutex to avoid deadlock with ext3_get_blocks_handle
184 * At this moment, get_block can be called only for blocks inside
185 * i_size since page cache has been already dropped and writes are
186 * blocked by i_mutex. So we can safely drop the truncate_mutex.
188 mutex_unlock(&EXT3_I(inode
)->truncate_mutex
);
189 ret
= ext3_journal_restart(handle
, blocks_for_truncate(inode
));
190 mutex_lock(&EXT3_I(inode
)->truncate_mutex
);
195 * Called at inode eviction from icache
197 void ext3_evict_inode (struct inode
*inode
)
199 struct ext3_block_alloc_info
*rsv
;
203 trace_ext3_evict_inode(inode
);
204 if (!inode
->i_nlink
&& !is_bad_inode(inode
)) {
205 dquot_initialize(inode
);
209 truncate_inode_pages(&inode
->i_data
, 0);
211 ext3_discard_reservation(inode
);
212 rsv
= EXT3_I(inode
)->i_block_alloc_info
;
213 EXT3_I(inode
)->i_block_alloc_info
= NULL
;
220 handle
= start_transaction(inode
);
221 if (IS_ERR(handle
)) {
223 * If we're going to skip the normal cleanup, we still need to
224 * make sure that the in-core orphan linked list is properly
227 ext3_orphan_del(NULL
, inode
);
235 ext3_truncate(inode
);
237 * Kill off the orphan record created when the inode lost the last
238 * link. Note that ext3_orphan_del() has to be able to cope with the
239 * deletion of a non-existent orphan - ext3_truncate() could
240 * have removed the record.
242 ext3_orphan_del(handle
, inode
);
243 EXT3_I(inode
)->i_dtime
= get_seconds();
246 * One subtle ordering requirement: if anything has gone wrong
247 * (transaction abort, IO errors, whatever), then we can still
248 * do these next steps (the fs will already have been marked as
249 * having errors), but we can't free the inode if the mark_dirty
252 if (ext3_mark_inode_dirty(handle
, inode
)) {
253 /* If that failed, just dquot_drop() and be done with that */
255 end_writeback(inode
);
257 ext3_xattr_delete_inode(handle
, inode
);
258 dquot_free_inode(inode
);
260 end_writeback(inode
);
261 ext3_free_inode(handle
, inode
);
263 ext3_journal_stop(handle
);
266 end_writeback(inode
);
273 struct buffer_head
*bh
;
276 static inline void add_chain(Indirect
*p
, struct buffer_head
*bh
, __le32
*v
)
278 p
->key
= *(p
->p
= v
);
282 static int verify_chain(Indirect
*from
, Indirect
*to
)
284 while (from
<= to
&& from
->key
== *from
->p
)
290 * ext3_block_to_path - parse the block number into array of offsets
291 * @inode: inode in question (we are only interested in its superblock)
292 * @i_block: block number to be parsed
293 * @offsets: array to store the offsets in
294 * @boundary: set this non-zero if the referred-to block is likely to be
295 * followed (on disk) by an indirect block.
297 * To store the locations of file's data ext3 uses a data structure common
298 * for UNIX filesystems - tree of pointers anchored in the inode, with
299 * data blocks at leaves and indirect blocks in intermediate nodes.
300 * This function translates the block number into path in that tree -
301 * return value is the path length and @offsets[n] is the offset of
302 * pointer to (n+1)th node in the nth one. If @block is out of range
303 * (negative or too large) warning is printed and zero returned.
305 * Note: function doesn't find node addresses, so no IO is needed. All
306 * we need to know is the capacity of indirect blocks (taken from the
311 * Portability note: the last comparison (check that we fit into triple
312 * indirect block) is spelled differently, because otherwise on an
313 * architecture with 32-bit longs and 8Kb pages we might get into trouble
314 * if our filesystem had 8Kb blocks. We might use long long, but that would
315 * kill us on x86. Oh, well, at least the sign propagation does not matter -
316 * i_block would have to be negative in the very beginning, so we would not
320 static int ext3_block_to_path(struct inode
*inode
,
321 long i_block
, int offsets
[4], int *boundary
)
323 int ptrs
= EXT3_ADDR_PER_BLOCK(inode
->i_sb
);
324 int ptrs_bits
= EXT3_ADDR_PER_BLOCK_BITS(inode
->i_sb
);
325 const long direct_blocks
= EXT3_NDIR_BLOCKS
,
326 indirect_blocks
= ptrs
,
327 double_blocks
= (1 << (ptrs_bits
* 2));
332 ext3_warning (inode
->i_sb
, "ext3_block_to_path", "block < 0");
333 } else if (i_block
< direct_blocks
) {
334 offsets
[n
++] = i_block
;
335 final
= direct_blocks
;
336 } else if ( (i_block
-= direct_blocks
) < indirect_blocks
) {
337 offsets
[n
++] = EXT3_IND_BLOCK
;
338 offsets
[n
++] = i_block
;
340 } else if ((i_block
-= indirect_blocks
) < double_blocks
) {
341 offsets
[n
++] = EXT3_DIND_BLOCK
;
342 offsets
[n
++] = i_block
>> ptrs_bits
;
343 offsets
[n
++] = i_block
& (ptrs
- 1);
345 } else if (((i_block
-= double_blocks
) >> (ptrs_bits
* 2)) < ptrs
) {
346 offsets
[n
++] = EXT3_TIND_BLOCK
;
347 offsets
[n
++] = i_block
>> (ptrs_bits
* 2);
348 offsets
[n
++] = (i_block
>> ptrs_bits
) & (ptrs
- 1);
349 offsets
[n
++] = i_block
& (ptrs
- 1);
352 ext3_warning(inode
->i_sb
, "ext3_block_to_path", "block > big");
355 *boundary
= final
- 1 - (i_block
& (ptrs
- 1));
360 * ext3_get_branch - read the chain of indirect blocks leading to data
361 * @inode: inode in question
362 * @depth: depth of the chain (1 - direct pointer, etc.)
363 * @offsets: offsets of pointers in inode/indirect blocks
364 * @chain: place to store the result
365 * @err: here we store the error value
367 * Function fills the array of triples <key, p, bh> and returns %NULL
368 * if everything went OK or the pointer to the last filled triple
369 * (incomplete one) otherwise. Upon the return chain[i].key contains
370 * the number of (i+1)-th block in the chain (as it is stored in memory,
371 * i.e. little-endian 32-bit), chain[i].p contains the address of that
372 * number (it points into struct inode for i==0 and into the bh->b_data
373 * for i>0) and chain[i].bh points to the buffer_head of i-th indirect
374 * block for i>0 and NULL for i==0. In other words, it holds the block
375 * numbers of the chain, addresses they were taken from (and where we can
376 * verify that chain did not change) and buffer_heads hosting these
379 * Function stops when it stumbles upon zero pointer (absent block)
380 * (pointer to last triple returned, *@err == 0)
381 * or when it gets an IO error reading an indirect block
382 * (ditto, *@err == -EIO)
383 * or when it notices that chain had been changed while it was reading
384 * (ditto, *@err == -EAGAIN)
385 * or when it reads all @depth-1 indirect blocks successfully and finds
386 * the whole chain, all way to the data (returns %NULL, *err == 0).
388 static Indirect
*ext3_get_branch(struct inode
*inode
, int depth
, int *offsets
,
389 Indirect chain
[4], int *err
)
391 struct super_block
*sb
= inode
->i_sb
;
393 struct buffer_head
*bh
;
396 /* i_data is not going away, no lock needed */
397 add_chain (chain
, NULL
, EXT3_I(inode
)->i_data
+ *offsets
);
401 bh
= sb_bread(sb
, le32_to_cpu(p
->key
));
404 /* Reader: pointers */
405 if (!verify_chain(chain
, p
))
407 add_chain(++p
, bh
, (__le32
*)bh
->b_data
+ *++offsets
);
425 * ext3_find_near - find a place for allocation with sufficient locality
427 * @ind: descriptor of indirect block.
429 * This function returns the preferred place for block allocation.
430 * It is used when heuristic for sequential allocation fails.
432 * + if there is a block to the left of our position - allocate near it.
433 * + if pointer will live in indirect block - allocate near that block.
434 * + if pointer will live in inode - allocate in the same
437 * In the latter case we colour the starting block by the callers PID to
438 * prevent it from clashing with concurrent allocations for a different inode
439 * in the same block group. The PID is used here so that functionally related
440 * files will be close-by on-disk.
442 * Caller must make sure that @ind is valid and will stay that way.
444 static ext3_fsblk_t
ext3_find_near(struct inode
*inode
, Indirect
*ind
)
446 struct ext3_inode_info
*ei
= EXT3_I(inode
);
447 __le32
*start
= ind
->bh
? (__le32
*) ind
->bh
->b_data
: ei
->i_data
;
449 ext3_fsblk_t bg_start
;
450 ext3_grpblk_t colour
;
452 /* Try to find previous block */
453 for (p
= ind
->p
- 1; p
>= start
; p
--) {
455 return le32_to_cpu(*p
);
458 /* No such thing, so let's try location of indirect block */
460 return ind
->bh
->b_blocknr
;
463 * It is going to be referred to from the inode itself? OK, just put it
464 * into the same cylinder group then.
466 bg_start
= ext3_group_first_block_no(inode
->i_sb
, ei
->i_block_group
);
467 colour
= (current
->pid
% 16) *
468 (EXT3_BLOCKS_PER_GROUP(inode
->i_sb
) / 16);
469 return bg_start
+ colour
;
473 * ext3_find_goal - find a preferred place for allocation.
475 * @block: block we want
476 * @partial: pointer to the last triple within a chain
478 * Normally this function find the preferred place for block allocation,
482 static ext3_fsblk_t
ext3_find_goal(struct inode
*inode
, long block
,
485 struct ext3_block_alloc_info
*block_i
;
487 block_i
= EXT3_I(inode
)->i_block_alloc_info
;
490 * try the heuristic for sequential allocation,
491 * failing that at least try to get decent locality.
493 if (block_i
&& (block
== block_i
->last_alloc_logical_block
+ 1)
494 && (block_i
->last_alloc_physical_block
!= 0)) {
495 return block_i
->last_alloc_physical_block
+ 1;
498 return ext3_find_near(inode
, partial
);
502 * ext3_blks_to_allocate - Look up the block map and count the number
503 * of direct blocks need to be allocated for the given branch.
505 * @branch: chain of indirect blocks
506 * @k: number of blocks need for indirect blocks
507 * @blks: number of data blocks to be mapped.
508 * @blocks_to_boundary: the offset in the indirect block
510 * return the total number of blocks to be allocate, including the
511 * direct and indirect blocks.
513 static int ext3_blks_to_allocate(Indirect
*branch
, int k
, unsigned long blks
,
514 int blocks_to_boundary
)
516 unsigned long count
= 0;
519 * Simple case, [t,d]Indirect block(s) has not allocated yet
520 * then it's clear blocks on that path have not allocated
523 /* right now we don't handle cross boundary allocation */
524 if (blks
< blocks_to_boundary
+ 1)
527 count
+= blocks_to_boundary
+ 1;
532 while (count
< blks
&& count
<= blocks_to_boundary
&&
533 le32_to_cpu(*(branch
[0].p
+ count
)) == 0) {
540 * ext3_alloc_blocks - multiple allocate blocks needed for a branch
541 * @handle: handle for this transaction
543 * @goal: preferred place for allocation
544 * @indirect_blks: the number of blocks need to allocate for indirect
546 * @blks: number of blocks need to allocated for direct blocks
547 * @new_blocks: on return it will store the new block numbers for
548 * the indirect blocks(if needed) and the first direct block,
549 * @err: here we store the error value
551 * return the number of direct blocks allocated
553 static int ext3_alloc_blocks(handle_t
*handle
, struct inode
*inode
,
554 ext3_fsblk_t goal
, int indirect_blks
, int blks
,
555 ext3_fsblk_t new_blocks
[4], int *err
)
558 unsigned long count
= 0;
560 ext3_fsblk_t current_block
= 0;
564 * Here we try to allocate the requested multiple blocks at once,
565 * on a best-effort basis.
566 * To build a branch, we should allocate blocks for
567 * the indirect blocks(if not allocated yet), and at least
568 * the first direct block of this branch. That's the
569 * minimum number of blocks need to allocate(required)
571 target
= blks
+ indirect_blks
;
575 /* allocating blocks for indirect blocks and direct blocks */
576 current_block
= ext3_new_blocks(handle
,inode
,goal
,&count
,err
);
581 /* allocate blocks for indirect blocks */
582 while (index
< indirect_blks
&& count
) {
583 new_blocks
[index
++] = current_block
++;
591 /* save the new block number for the first direct block */
592 new_blocks
[index
] = current_block
;
594 /* total number of blocks allocated for direct blocks */
599 for (i
= 0; i
<index
; i
++)
600 ext3_free_blocks(handle
, inode
, new_blocks
[i
], 1);
605 * ext3_alloc_branch - allocate and set up a chain of blocks.
606 * @handle: handle for this transaction
608 * @indirect_blks: number of allocated indirect blocks
609 * @blks: number of allocated direct blocks
610 * @goal: preferred place for allocation
611 * @offsets: offsets (in the blocks) to store the pointers to next.
612 * @branch: place to store the chain in.
614 * This function allocates blocks, zeroes out all but the last one,
615 * links them into chain and (if we are synchronous) writes them to disk.
616 * In other words, it prepares a branch that can be spliced onto the
617 * inode. It stores the information about that chain in the branch[], in
618 * the same format as ext3_get_branch() would do. We are calling it after
619 * we had read the existing part of chain and partial points to the last
620 * triple of that (one with zero ->key). Upon the exit we have the same
621 * picture as after the successful ext3_get_block(), except that in one
622 * place chain is disconnected - *branch->p is still zero (we did not
623 * set the last link), but branch->key contains the number that should
624 * be placed into *branch->p to fill that gap.
626 * If allocation fails we free all blocks we've allocated (and forget
627 * their buffer_heads) and return the error value the from failed
628 * ext3_alloc_block() (normally -ENOSPC). Otherwise we set the chain
629 * as described above and return 0.
631 static int ext3_alloc_branch(handle_t
*handle
, struct inode
*inode
,
632 int indirect_blks
, int *blks
, ext3_fsblk_t goal
,
633 int *offsets
, Indirect
*branch
)
635 int blocksize
= inode
->i_sb
->s_blocksize
;
638 struct buffer_head
*bh
;
640 ext3_fsblk_t new_blocks
[4];
641 ext3_fsblk_t current_block
;
643 num
= ext3_alloc_blocks(handle
, inode
, goal
, indirect_blks
,
644 *blks
, new_blocks
, &err
);
648 branch
[0].key
= cpu_to_le32(new_blocks
[0]);
650 * metadata blocks and data blocks are allocated.
652 for (n
= 1; n
<= indirect_blks
; n
++) {
654 * Get buffer_head for parent block, zero it out
655 * and set the pointer to new one, then send
658 bh
= sb_getblk(inode
->i_sb
, new_blocks
[n
-1]);
661 BUFFER_TRACE(bh
, "call get_create_access");
662 err
= ext3_journal_get_create_access(handle
, bh
);
669 memset(bh
->b_data
, 0, blocksize
);
670 branch
[n
].p
= (__le32
*) bh
->b_data
+ offsets
[n
];
671 branch
[n
].key
= cpu_to_le32(new_blocks
[n
]);
672 *branch
[n
].p
= branch
[n
].key
;
673 if ( n
== indirect_blks
) {
674 current_block
= new_blocks
[n
];
676 * End of chain, update the last new metablock of
677 * the chain to point to the new allocated
678 * data blocks numbers
680 for (i
=1; i
< num
; i
++)
681 *(branch
[n
].p
+ i
) = cpu_to_le32(++current_block
);
683 BUFFER_TRACE(bh
, "marking uptodate");
684 set_buffer_uptodate(bh
);
687 BUFFER_TRACE(bh
, "call ext3_journal_dirty_metadata");
688 err
= ext3_journal_dirty_metadata(handle
, bh
);
695 /* Allocation failed, free what we already allocated */
696 for (i
= 1; i
<= n
; i
++) {
697 BUFFER_TRACE(branch
[i
].bh
, "call journal_forget");
698 ext3_journal_forget(handle
, branch
[i
].bh
);
700 for (i
= 0; i
<indirect_blks
; i
++)
701 ext3_free_blocks(handle
, inode
, new_blocks
[i
], 1);
703 ext3_free_blocks(handle
, inode
, new_blocks
[i
], num
);
709 * ext3_splice_branch - splice the allocated branch onto inode.
710 * @handle: handle for this transaction
712 * @block: (logical) number of block we are adding
713 * @where: location of missing link
714 * @num: number of indirect blocks we are adding
715 * @blks: number of direct blocks we are adding
717 * This function fills the missing link and does all housekeeping needed in
718 * inode (->i_blocks, etc.). In case of success we end up with the full
719 * chain to new block and return 0.
721 static int ext3_splice_branch(handle_t
*handle
, struct inode
*inode
,
722 long block
, Indirect
*where
, int num
, int blks
)
726 struct ext3_block_alloc_info
*block_i
;
727 ext3_fsblk_t current_block
;
728 struct ext3_inode_info
*ei
= EXT3_I(inode
);
730 block_i
= ei
->i_block_alloc_info
;
732 * If we're splicing into a [td]indirect block (as opposed to the
733 * inode) then we need to get write access to the [td]indirect block
737 BUFFER_TRACE(where
->bh
, "get_write_access");
738 err
= ext3_journal_get_write_access(handle
, where
->bh
);
744 *where
->p
= where
->key
;
747 * Update the host buffer_head or inode to point to more just allocated
748 * direct blocks blocks
750 if (num
== 0 && blks
> 1) {
751 current_block
= le32_to_cpu(where
->key
) + 1;
752 for (i
= 1; i
< blks
; i
++)
753 *(where
->p
+ i
) = cpu_to_le32(current_block
++);
757 * update the most recently allocated logical & physical block
758 * in i_block_alloc_info, to assist find the proper goal block for next
762 block_i
->last_alloc_logical_block
= block
+ blks
- 1;
763 block_i
->last_alloc_physical_block
=
764 le32_to_cpu(where
[num
].key
) + blks
- 1;
767 /* We are done with atomic stuff, now do the rest of housekeeping */
769 inode
->i_ctime
= CURRENT_TIME_SEC
;
770 ext3_mark_inode_dirty(handle
, inode
);
771 /* ext3_mark_inode_dirty already updated i_sync_tid */
772 atomic_set(&ei
->i_datasync_tid
, handle
->h_transaction
->t_tid
);
774 /* had we spliced it onto indirect block? */
777 * If we spliced it onto an indirect block, we haven't
778 * altered the inode. Note however that if it is being spliced
779 * onto an indirect block at the very end of the file (the
780 * file is growing) then we *will* alter the inode to reflect
781 * the new i_size. But that is not done here - it is done in
782 * generic_commit_write->__mark_inode_dirty->ext3_dirty_inode.
784 jbd_debug(5, "splicing indirect only\n");
785 BUFFER_TRACE(where
->bh
, "call ext3_journal_dirty_metadata");
786 err
= ext3_journal_dirty_metadata(handle
, where
->bh
);
791 * OK, we spliced it into the inode itself on a direct block.
792 * Inode was dirtied above.
794 jbd_debug(5, "splicing direct\n");
799 for (i
= 1; i
<= num
; i
++) {
800 BUFFER_TRACE(where
[i
].bh
, "call journal_forget");
801 ext3_journal_forget(handle
, where
[i
].bh
);
802 ext3_free_blocks(handle
,inode
,le32_to_cpu(where
[i
-1].key
),1);
804 ext3_free_blocks(handle
, inode
, le32_to_cpu(where
[num
].key
), blks
);
810 * Allocation strategy is simple: if we have to allocate something, we will
811 * have to go the whole way to leaf. So let's do it before attaching anything
812 * to tree, set linkage between the newborn blocks, write them if sync is
813 * required, recheck the path, free and repeat if check fails, otherwise
814 * set the last missing link (that will protect us from any truncate-generated
815 * removals - all blocks on the path are immune now) and possibly force the
816 * write on the parent block.
817 * That has a nice additional property: no special recovery from the failed
818 * allocations is needed - we simply release blocks and do not touch anything
819 * reachable from inode.
821 * `handle' can be NULL if create == 0.
823 * The BKL may not be held on entry here. Be sure to take it early.
824 * return > 0, # of blocks mapped or allocated.
825 * return = 0, if plain lookup failed.
826 * return < 0, error case.
828 int ext3_get_blocks_handle(handle_t
*handle
, struct inode
*inode
,
829 sector_t iblock
, unsigned long maxblocks
,
830 struct buffer_head
*bh_result
,
839 int blocks_to_boundary
= 0;
841 struct ext3_inode_info
*ei
= EXT3_I(inode
);
843 ext3_fsblk_t first_block
= 0;
846 trace_ext3_get_blocks_enter(inode
, iblock
, maxblocks
, create
);
847 J_ASSERT(handle
!= NULL
|| create
== 0);
848 depth
= ext3_block_to_path(inode
,iblock
,offsets
,&blocks_to_boundary
);
853 partial
= ext3_get_branch(inode
, depth
, offsets
, chain
, &err
);
855 /* Simplest case - block found, no allocation needed */
857 first_block
= le32_to_cpu(chain
[depth
- 1].key
);
858 clear_buffer_new(bh_result
);
861 while (count
< maxblocks
&& count
<= blocks_to_boundary
) {
864 if (!verify_chain(chain
, chain
+ depth
- 1)) {
866 * Indirect block might be removed by
867 * truncate while we were reading it.
868 * Handling of that case: forget what we've
869 * got now. Flag the err as EAGAIN, so it
876 blk
= le32_to_cpu(*(chain
[depth
-1].p
+ count
));
878 if (blk
== first_block
+ count
)
887 /* Next simple case - plain lookup or failed read of indirect block */
888 if (!create
|| err
== -EIO
)
892 * Block out ext3_truncate while we alter the tree
894 mutex_lock(&ei
->truncate_mutex
);
897 * If the indirect block is missing while we are reading
898 * the chain(ext3_get_branch() returns -EAGAIN err), or
899 * if the chain has been changed after we grab the semaphore,
900 * (either because another process truncated this branch, or
901 * another get_block allocated this branch) re-grab the chain to see if
902 * the request block has been allocated or not.
904 * Since we already block the truncate/other get_block
905 * at this point, we will have the current copy of the chain when we
906 * splice the branch into the tree.
908 if (err
== -EAGAIN
|| !verify_chain(chain
, partial
)) {
909 while (partial
> chain
) {
913 partial
= ext3_get_branch(inode
, depth
, offsets
, chain
, &err
);
916 mutex_unlock(&ei
->truncate_mutex
);
919 clear_buffer_new(bh_result
);
925 * Okay, we need to do block allocation. Lazily initialize the block
926 * allocation info here if necessary
928 if (S_ISREG(inode
->i_mode
) && (!ei
->i_block_alloc_info
))
929 ext3_init_block_alloc_info(inode
);
931 goal
= ext3_find_goal(inode
, iblock
, partial
);
933 /* the number of blocks need to allocate for [d,t]indirect blocks */
934 indirect_blks
= (chain
+ depth
) - partial
- 1;
937 * Next look up the indirect map to count the totoal number of
938 * direct blocks to allocate for this branch.
940 count
= ext3_blks_to_allocate(partial
, indirect_blks
,
941 maxblocks
, blocks_to_boundary
);
942 err
= ext3_alloc_branch(handle
, inode
, indirect_blks
, &count
, goal
,
943 offsets
+ (partial
- chain
), partial
);
946 * The ext3_splice_branch call will free and forget any buffers
947 * on the new chain if there is a failure, but that risks using
948 * up transaction credits, especially for bitmaps where the
949 * credits cannot be returned. Can we handle this somehow? We
950 * may need to return -EAGAIN upwards in the worst case. --sct
953 err
= ext3_splice_branch(handle
, inode
, iblock
,
954 partial
, indirect_blks
, count
);
955 mutex_unlock(&ei
->truncate_mutex
);
959 set_buffer_new(bh_result
);
961 map_bh(bh_result
, inode
->i_sb
, le32_to_cpu(chain
[depth
-1].key
));
962 if (count
> blocks_to_boundary
)
963 set_buffer_boundary(bh_result
);
965 /* Clean up and exit */
966 partial
= chain
+ depth
- 1; /* the whole chain */
968 while (partial
> chain
) {
969 BUFFER_TRACE(partial
->bh
, "call brelse");
973 BUFFER_TRACE(bh_result
, "returned");
975 trace_ext3_get_blocks_exit(inode
, iblock
,
976 depth
? le32_to_cpu(chain
[depth
-1].key
) : 0,
981 /* Maximum number of blocks we map for direct IO at once. */
982 #define DIO_MAX_BLOCKS 4096
984 * Number of credits we need for writing DIO_MAX_BLOCKS:
985 * We need sb + group descriptor + bitmap + inode -> 4
986 * For B blocks with A block pointers per block we need:
987 * 1 (triple ind.) + (B/A/A + 2) (doubly ind.) + (B/A + 2) (indirect).
988 * If we plug in 4096 for B and 256 for A (for 1KB block size), we get 25.
990 #define DIO_CREDITS 25
992 static int ext3_get_block(struct inode
*inode
, sector_t iblock
,
993 struct buffer_head
*bh_result
, int create
)
995 handle_t
*handle
= ext3_journal_current_handle();
996 int ret
= 0, started
= 0;
997 unsigned max_blocks
= bh_result
->b_size
>> inode
->i_blkbits
;
999 if (create
&& !handle
) { /* Direct IO write... */
1000 if (max_blocks
> DIO_MAX_BLOCKS
)
1001 max_blocks
= DIO_MAX_BLOCKS
;
1002 handle
= ext3_journal_start(inode
, DIO_CREDITS
+
1003 EXT3_MAXQUOTAS_TRANS_BLOCKS(inode
->i_sb
));
1004 if (IS_ERR(handle
)) {
1005 ret
= PTR_ERR(handle
);
1011 ret
= ext3_get_blocks_handle(handle
, inode
, iblock
,
1012 max_blocks
, bh_result
, create
);
1014 bh_result
->b_size
= (ret
<< inode
->i_blkbits
);
1018 ext3_journal_stop(handle
);
1023 int ext3_fiemap(struct inode
*inode
, struct fiemap_extent_info
*fieinfo
,
1026 return generic_block_fiemap(inode
, fieinfo
, start
, len
,
1031 * `handle' can be NULL if create is zero
1033 struct buffer_head
*ext3_getblk(handle_t
*handle
, struct inode
*inode
,
1034 long block
, int create
, int *errp
)
1036 struct buffer_head dummy
;
1039 J_ASSERT(handle
!= NULL
|| create
== 0);
1042 dummy
.b_blocknr
= -1000;
1043 buffer_trace_init(&dummy
.b_history
);
1044 err
= ext3_get_blocks_handle(handle
, inode
, block
, 1,
1047 * ext3_get_blocks_handle() returns number of blocks
1048 * mapped. 0 in case of a HOLE.
1056 if (!err
&& buffer_mapped(&dummy
)) {
1057 struct buffer_head
*bh
;
1058 bh
= sb_getblk(inode
->i_sb
, dummy
.b_blocknr
);
1063 if (buffer_new(&dummy
)) {
1064 J_ASSERT(create
!= 0);
1065 J_ASSERT(handle
!= NULL
);
1068 * Now that we do not always journal data, we should
1069 * keep in mind whether this should always journal the
1070 * new buffer as metadata. For now, regular file
1071 * writes use ext3_get_block instead, so it's not a
1075 BUFFER_TRACE(bh
, "call get_create_access");
1076 fatal
= ext3_journal_get_create_access(handle
, bh
);
1077 if (!fatal
&& !buffer_uptodate(bh
)) {
1078 memset(bh
->b_data
,0,inode
->i_sb
->s_blocksize
);
1079 set_buffer_uptodate(bh
);
1082 BUFFER_TRACE(bh
, "call ext3_journal_dirty_metadata");
1083 err
= ext3_journal_dirty_metadata(handle
, bh
);
1087 BUFFER_TRACE(bh
, "not a new buffer");
1100 struct buffer_head
*ext3_bread(handle_t
*handle
, struct inode
*inode
,
1101 int block
, int create
, int *err
)
1103 struct buffer_head
* bh
;
1105 bh
= ext3_getblk(handle
, inode
, block
, create
, err
);
1108 if (buffer_uptodate(bh
))
1110 ll_rw_block(READ_META
, 1, &bh
);
1112 if (buffer_uptodate(bh
))
1119 static int walk_page_buffers( handle_t
*handle
,
1120 struct buffer_head
*head
,
1124 int (*fn
)( handle_t
*handle
,
1125 struct buffer_head
*bh
))
1127 struct buffer_head
*bh
;
1128 unsigned block_start
, block_end
;
1129 unsigned blocksize
= head
->b_size
;
1131 struct buffer_head
*next
;
1133 for ( bh
= head
, block_start
= 0;
1134 ret
== 0 && (bh
!= head
|| !block_start
);
1135 block_start
= block_end
, bh
= next
)
1137 next
= bh
->b_this_page
;
1138 block_end
= block_start
+ blocksize
;
1139 if (block_end
<= from
|| block_start
>= to
) {
1140 if (partial
&& !buffer_uptodate(bh
))
1144 err
= (*fn
)(handle
, bh
);
1152 * To preserve ordering, it is essential that the hole instantiation and
1153 * the data write be encapsulated in a single transaction. We cannot
1154 * close off a transaction and start a new one between the ext3_get_block()
1155 * and the commit_write(). So doing the journal_start at the start of
1156 * prepare_write() is the right place.
1158 * Also, this function can nest inside ext3_writepage() ->
1159 * block_write_full_page(). In that case, we *know* that ext3_writepage()
1160 * has generated enough buffer credits to do the whole page. So we won't
1161 * block on the journal in that case, which is good, because the caller may
1164 * By accident, ext3 can be reentered when a transaction is open via
1165 * quota file writes. If we were to commit the transaction while thus
1166 * reentered, there can be a deadlock - we would be holding a quota
1167 * lock, and the commit would never complete if another thread had a
1168 * transaction open and was blocking on the quota lock - a ranking
1171 * So what we do is to rely on the fact that journal_stop/journal_start
1172 * will _not_ run commit under these circumstances because handle->h_ref
1173 * is elevated. We'll still have enough credits for the tiny quotafile
1176 static int do_journal_get_write_access(handle_t
*handle
,
1177 struct buffer_head
*bh
)
1179 int dirty
= buffer_dirty(bh
);
1182 if (!buffer_mapped(bh
) || buffer_freed(bh
))
1185 * __block_prepare_write() could have dirtied some buffers. Clean
1186 * the dirty bit as jbd2_journal_get_write_access() could complain
1187 * otherwise about fs integrity issues. Setting of the dirty bit
1188 * by __block_prepare_write() isn't a real problem here as we clear
1189 * the bit before releasing a page lock and thus writeback cannot
1190 * ever write the buffer.
1193 clear_buffer_dirty(bh
);
1194 ret
= ext3_journal_get_write_access(handle
, bh
);
1196 ret
= ext3_journal_dirty_metadata(handle
, bh
);
1201 * Truncate blocks that were not used by write. We have to truncate the
1202 * pagecache as well so that corresponding buffers get properly unmapped.
1204 static void ext3_truncate_failed_write(struct inode
*inode
)
1206 truncate_inode_pages(inode
->i_mapping
, inode
->i_size
);
1207 ext3_truncate(inode
);
1210 static int ext3_write_begin(struct file
*file
, struct address_space
*mapping
,
1211 loff_t pos
, unsigned len
, unsigned flags
,
1212 struct page
**pagep
, void **fsdata
)
1214 struct inode
*inode
= mapping
->host
;
1221 /* Reserve one block more for addition to orphan list in case
1222 * we allocate blocks but write fails for some reason */
1223 int needed_blocks
= ext3_writepage_trans_blocks(inode
) + 1;
1225 trace_ext3_write_begin(inode
, pos
, len
, flags
);
1227 index
= pos
>> PAGE_CACHE_SHIFT
;
1228 from
= pos
& (PAGE_CACHE_SIZE
- 1);
1232 page
= grab_cache_page_write_begin(mapping
, index
, flags
);
1237 handle
= ext3_journal_start(inode
, needed_blocks
);
1238 if (IS_ERR(handle
)) {
1240 page_cache_release(page
);
1241 ret
= PTR_ERR(handle
);
1244 ret
= __block_write_begin(page
, pos
, len
, ext3_get_block
);
1246 goto write_begin_failed
;
1248 if (ext3_should_journal_data(inode
)) {
1249 ret
= walk_page_buffers(handle
, page_buffers(page
),
1250 from
, to
, NULL
, do_journal_get_write_access
);
1255 * block_write_begin may have instantiated a few blocks
1256 * outside i_size. Trim these off again. Don't need
1257 * i_size_read because we hold i_mutex.
1259 * Add inode to orphan list in case we crash before truncate
1260 * finishes. Do this only if ext3_can_truncate() agrees so
1261 * that orphan processing code is happy.
1263 if (pos
+ len
> inode
->i_size
&& ext3_can_truncate(inode
))
1264 ext3_orphan_add(handle
, inode
);
1265 ext3_journal_stop(handle
);
1267 page_cache_release(page
);
1268 if (pos
+ len
> inode
->i_size
)
1269 ext3_truncate_failed_write(inode
);
1271 if (ret
== -ENOSPC
&& ext3_should_retry_alloc(inode
->i_sb
, &retries
))
1278 int ext3_journal_dirty_data(handle_t
*handle
, struct buffer_head
*bh
)
1280 int err
= journal_dirty_data(handle
, bh
);
1282 ext3_journal_abort_handle(__func__
, __func__
,
1287 /* For ordered writepage and write_end functions */
1288 static int journal_dirty_data_fn(handle_t
*handle
, struct buffer_head
*bh
)
1291 * Write could have mapped the buffer but it didn't copy the data in
1292 * yet. So avoid filing such buffer into a transaction.
1294 if (buffer_mapped(bh
) && buffer_uptodate(bh
))
1295 return ext3_journal_dirty_data(handle
, bh
);
1299 /* For write_end() in data=journal mode */
1300 static int write_end_fn(handle_t
*handle
, struct buffer_head
*bh
)
1302 if (!buffer_mapped(bh
) || buffer_freed(bh
))
1304 set_buffer_uptodate(bh
);
1305 return ext3_journal_dirty_metadata(handle
, bh
);
1309 * This is nasty and subtle: ext3_write_begin() could have allocated blocks
1310 * for the whole page but later we failed to copy the data in. Update inode
1311 * size according to what we managed to copy. The rest is going to be
1312 * truncated in write_end function.
1314 static void update_file_sizes(struct inode
*inode
, loff_t pos
, unsigned copied
)
1316 /* What matters to us is i_disksize. We don't write i_size anywhere */
1317 if (pos
+ copied
> inode
->i_size
)
1318 i_size_write(inode
, pos
+ copied
);
1319 if (pos
+ copied
> EXT3_I(inode
)->i_disksize
) {
1320 EXT3_I(inode
)->i_disksize
= pos
+ copied
;
1321 mark_inode_dirty(inode
);
1326 * We need to pick up the new inode size which generic_commit_write gave us
1327 * `file' can be NULL - eg, when called from page_symlink().
1329 * ext3 never places buffers on inode->i_mapping->private_list. metadata
1330 * buffers are managed internally.
1332 static int ext3_ordered_write_end(struct file
*file
,
1333 struct address_space
*mapping
,
1334 loff_t pos
, unsigned len
, unsigned copied
,
1335 struct page
*page
, void *fsdata
)
1337 handle_t
*handle
= ext3_journal_current_handle();
1338 struct inode
*inode
= file
->f_mapping
->host
;
1342 trace_ext3_ordered_write_end(inode
, pos
, len
, copied
);
1343 copied
= block_write_end(file
, mapping
, pos
, len
, copied
, page
, fsdata
);
1345 from
= pos
& (PAGE_CACHE_SIZE
- 1);
1347 ret
= walk_page_buffers(handle
, page_buffers(page
),
1348 from
, to
, NULL
, journal_dirty_data_fn
);
1351 update_file_sizes(inode
, pos
, copied
);
1353 * There may be allocated blocks outside of i_size because
1354 * we failed to copy some data. Prepare for truncate.
1356 if (pos
+ len
> inode
->i_size
&& ext3_can_truncate(inode
))
1357 ext3_orphan_add(handle
, inode
);
1358 ret2
= ext3_journal_stop(handle
);
1362 page_cache_release(page
);
1364 if (pos
+ len
> inode
->i_size
)
1365 ext3_truncate_failed_write(inode
);
1366 return ret
? ret
: copied
;
1369 static int ext3_writeback_write_end(struct file
*file
,
1370 struct address_space
*mapping
,
1371 loff_t pos
, unsigned len
, unsigned copied
,
1372 struct page
*page
, void *fsdata
)
1374 handle_t
*handle
= ext3_journal_current_handle();
1375 struct inode
*inode
= file
->f_mapping
->host
;
1378 trace_ext3_writeback_write_end(inode
, pos
, len
, copied
);
1379 copied
= block_write_end(file
, mapping
, pos
, len
, copied
, page
, fsdata
);
1380 update_file_sizes(inode
, pos
, copied
);
1382 * There may be allocated blocks outside of i_size because
1383 * we failed to copy some data. Prepare for truncate.
1385 if (pos
+ len
> inode
->i_size
&& ext3_can_truncate(inode
))
1386 ext3_orphan_add(handle
, inode
);
1387 ret
= ext3_journal_stop(handle
);
1389 page_cache_release(page
);
1391 if (pos
+ len
> inode
->i_size
)
1392 ext3_truncate_failed_write(inode
);
1393 return ret
? ret
: copied
;
1396 static int ext3_journalled_write_end(struct file
*file
,
1397 struct address_space
*mapping
,
1398 loff_t pos
, unsigned len
, unsigned copied
,
1399 struct page
*page
, void *fsdata
)
1401 handle_t
*handle
= ext3_journal_current_handle();
1402 struct inode
*inode
= mapping
->host
;
1407 trace_ext3_journalled_write_end(inode
, pos
, len
, copied
);
1408 from
= pos
& (PAGE_CACHE_SIZE
- 1);
1412 if (!PageUptodate(page
))
1414 page_zero_new_buffers(page
, from
+ copied
, to
);
1418 ret
= walk_page_buffers(handle
, page_buffers(page
), from
,
1419 to
, &partial
, write_end_fn
);
1421 SetPageUptodate(page
);
1423 if (pos
+ copied
> inode
->i_size
)
1424 i_size_write(inode
, pos
+ copied
);
1426 * There may be allocated blocks outside of i_size because
1427 * we failed to copy some data. Prepare for truncate.
1429 if (pos
+ len
> inode
->i_size
&& ext3_can_truncate(inode
))
1430 ext3_orphan_add(handle
, inode
);
1431 ext3_set_inode_state(inode
, EXT3_STATE_JDATA
);
1432 if (inode
->i_size
> EXT3_I(inode
)->i_disksize
) {
1433 EXT3_I(inode
)->i_disksize
= inode
->i_size
;
1434 ret2
= ext3_mark_inode_dirty(handle
, inode
);
1439 ret2
= ext3_journal_stop(handle
);
1443 page_cache_release(page
);
1445 if (pos
+ len
> inode
->i_size
)
1446 ext3_truncate_failed_write(inode
);
1447 return ret
? ret
: copied
;
1451 * bmap() is special. It gets used by applications such as lilo and by
1452 * the swapper to find the on-disk block of a specific piece of data.
1454 * Naturally, this is dangerous if the block concerned is still in the
1455 * journal. If somebody makes a swapfile on an ext3 data-journaling
1456 * filesystem and enables swap, then they may get a nasty shock when the
1457 * data getting swapped to that swapfile suddenly gets overwritten by
1458 * the original zero's written out previously to the journal and
1459 * awaiting writeback in the kernel's buffer cache.
1461 * So, if we see any bmap calls here on a modified, data-journaled file,
1462 * take extra steps to flush any blocks which might be in the cache.
1464 static sector_t
ext3_bmap(struct address_space
*mapping
, sector_t block
)
1466 struct inode
*inode
= mapping
->host
;
1470 if (ext3_test_inode_state(inode
, EXT3_STATE_JDATA
)) {
1472 * This is a REALLY heavyweight approach, but the use of
1473 * bmap on dirty files is expected to be extremely rare:
1474 * only if we run lilo or swapon on a freshly made file
1475 * do we expect this to happen.
1477 * (bmap requires CAP_SYS_RAWIO so this does not
1478 * represent an unprivileged user DOS attack --- we'd be
1479 * in trouble if mortal users could trigger this path at
1482 * NB. EXT3_STATE_JDATA is not set on files other than
1483 * regular files. If somebody wants to bmap a directory
1484 * or symlink and gets confused because the buffer
1485 * hasn't yet been flushed to disk, they deserve
1486 * everything they get.
1489 ext3_clear_inode_state(inode
, EXT3_STATE_JDATA
);
1490 journal
= EXT3_JOURNAL(inode
);
1491 journal_lock_updates(journal
);
1492 err
= journal_flush(journal
);
1493 journal_unlock_updates(journal
);
1499 return generic_block_bmap(mapping
,block
,ext3_get_block
);
1502 static int bget_one(handle_t
*handle
, struct buffer_head
*bh
)
1508 static int bput_one(handle_t
*handle
, struct buffer_head
*bh
)
1514 static int buffer_unmapped(handle_t
*handle
, struct buffer_head
*bh
)
1516 return !buffer_mapped(bh
);
1520 * Note that we always start a transaction even if we're not journalling
1521 * data. This is to preserve ordering: any hole instantiation within
1522 * __block_write_full_page -> ext3_get_block() should be journalled
1523 * along with the data so we don't crash and then get metadata which
1524 * refers to old data.
1526 * In all journalling modes block_write_full_page() will start the I/O.
1530 * ext3_writepage() -> kmalloc() -> __alloc_pages() -> page_launder() ->
1535 * ext3_file_write() -> generic_file_write() -> __alloc_pages() -> ...
1537 * Same applies to ext3_get_block(). We will deadlock on various things like
1538 * lock_journal and i_truncate_mutex.
1540 * Setting PF_MEMALLOC here doesn't work - too many internal memory
1543 * 16May01: If we're reentered then journal_current_handle() will be
1544 * non-zero. We simply *return*.
1546 * 1 July 2001: @@@ FIXME:
1547 * In journalled data mode, a data buffer may be metadata against the
1548 * current transaction. But the same file is part of a shared mapping
1549 * and someone does a writepage() on it.
1551 * We will move the buffer onto the async_data list, but *after* it has
1552 * been dirtied. So there's a small window where we have dirty data on
1555 * Note that this only applies to the last partial page in the file. The
1556 * bit which block_write_full_page() uses prepare/commit for. (That's
1557 * broken code anyway: it's wrong for msync()).
1559 * It's a rare case: affects the final partial page, for journalled data
1560 * where the file is subject to bith write() and writepage() in the same
1561 * transction. To fix it we'll need a custom block_write_full_page().
1562 * We'll probably need that anyway for journalling writepage() output.
1564 * We don't honour synchronous mounts for writepage(). That would be
1565 * disastrous. Any write() or metadata operation will sync the fs for
1568 * AKPM2: if all the page's buffers are mapped to disk and !data=journal,
1569 * we don't need to open a transaction here.
1571 static int ext3_ordered_writepage(struct page
*page
,
1572 struct writeback_control
*wbc
)
1574 struct inode
*inode
= page
->mapping
->host
;
1575 struct buffer_head
*page_bufs
;
1576 handle_t
*handle
= NULL
;
1580 J_ASSERT(PageLocked(page
));
1581 WARN_ON_ONCE(IS_RDONLY(inode
));
1584 * We give up here if we're reentered, because it might be for a
1585 * different filesystem.
1587 if (ext3_journal_current_handle())
1590 trace_ext3_ordered_writepage(page
);
1591 if (!page_has_buffers(page
)) {
1592 create_empty_buffers(page
, inode
->i_sb
->s_blocksize
,
1593 (1 << BH_Dirty
)|(1 << BH_Uptodate
));
1594 page_bufs
= page_buffers(page
);
1596 page_bufs
= page_buffers(page
);
1597 if (!walk_page_buffers(NULL
, page_bufs
, 0, PAGE_CACHE_SIZE
,
1598 NULL
, buffer_unmapped
)) {
1599 /* Provide NULL get_block() to catch bugs if buffers
1600 * weren't really mapped */
1601 return block_write_full_page(page
, NULL
, wbc
);
1604 handle
= ext3_journal_start(inode
, ext3_writepage_trans_blocks(inode
));
1606 if (IS_ERR(handle
)) {
1607 ret
= PTR_ERR(handle
);
1611 walk_page_buffers(handle
, page_bufs
, 0,
1612 PAGE_CACHE_SIZE
, NULL
, bget_one
);
1614 ret
= block_write_full_page(page
, ext3_get_block
, wbc
);
1617 * The page can become unlocked at any point now, and
1618 * truncate can then come in and change things. So we
1619 * can't touch *page from now on. But *page_bufs is
1620 * safe due to elevated refcount.
1624 * And attach them to the current transaction. But only if
1625 * block_write_full_page() succeeded. Otherwise they are unmapped,
1626 * and generally junk.
1629 err
= walk_page_buffers(handle
, page_bufs
, 0, PAGE_CACHE_SIZE
,
1630 NULL
, journal_dirty_data_fn
);
1634 walk_page_buffers(handle
, page_bufs
, 0,
1635 PAGE_CACHE_SIZE
, NULL
, bput_one
);
1636 err
= ext3_journal_stop(handle
);
1642 redirty_page_for_writepage(wbc
, page
);
1647 static int ext3_writeback_writepage(struct page
*page
,
1648 struct writeback_control
*wbc
)
1650 struct inode
*inode
= page
->mapping
->host
;
1651 handle_t
*handle
= NULL
;
1655 J_ASSERT(PageLocked(page
));
1656 WARN_ON_ONCE(IS_RDONLY(inode
));
1658 if (ext3_journal_current_handle())
1661 trace_ext3_writeback_writepage(page
);
1662 if (page_has_buffers(page
)) {
1663 if (!walk_page_buffers(NULL
, page_buffers(page
), 0,
1664 PAGE_CACHE_SIZE
, NULL
, buffer_unmapped
)) {
1665 /* Provide NULL get_block() to catch bugs if buffers
1666 * weren't really mapped */
1667 return block_write_full_page(page
, NULL
, wbc
);
1671 handle
= ext3_journal_start(inode
, ext3_writepage_trans_blocks(inode
));
1672 if (IS_ERR(handle
)) {
1673 ret
= PTR_ERR(handle
);
1677 ret
= block_write_full_page(page
, ext3_get_block
, wbc
);
1679 err
= ext3_journal_stop(handle
);
1685 redirty_page_for_writepage(wbc
, page
);
1690 static int ext3_journalled_writepage(struct page
*page
,
1691 struct writeback_control
*wbc
)
1693 struct inode
*inode
= page
->mapping
->host
;
1694 handle_t
*handle
= NULL
;
1698 J_ASSERT(PageLocked(page
));
1699 WARN_ON_ONCE(IS_RDONLY(inode
));
1701 if (ext3_journal_current_handle())
1704 trace_ext3_journalled_writepage(page
);
1705 handle
= ext3_journal_start(inode
, ext3_writepage_trans_blocks(inode
));
1706 if (IS_ERR(handle
)) {
1707 ret
= PTR_ERR(handle
);
1711 if (!page_has_buffers(page
) || PageChecked(page
)) {
1713 * It's mmapped pagecache. Add buffers and journal it. There
1714 * doesn't seem much point in redirtying the page here.
1716 ClearPageChecked(page
);
1717 ret
= __block_write_begin(page
, 0, PAGE_CACHE_SIZE
,
1720 ext3_journal_stop(handle
);
1723 ret
= walk_page_buffers(handle
, page_buffers(page
), 0,
1724 PAGE_CACHE_SIZE
, NULL
, do_journal_get_write_access
);
1726 err
= walk_page_buffers(handle
, page_buffers(page
), 0,
1727 PAGE_CACHE_SIZE
, NULL
, write_end_fn
);
1730 ext3_set_inode_state(inode
, EXT3_STATE_JDATA
);
1734 * It may be a page full of checkpoint-mode buffers. We don't
1735 * really know unless we go poke around in the buffer_heads.
1736 * But block_write_full_page will do the right thing.
1738 ret
= block_write_full_page(page
, ext3_get_block
, wbc
);
1740 err
= ext3_journal_stop(handle
);
1747 redirty_page_for_writepage(wbc
, page
);
1753 static int ext3_readpage(struct file
*file
, struct page
*page
)
1755 trace_ext3_readpage(page
);
1756 return mpage_readpage(page
, ext3_get_block
);
1760 ext3_readpages(struct file
*file
, struct address_space
*mapping
,
1761 struct list_head
*pages
, unsigned nr_pages
)
1763 return mpage_readpages(mapping
, pages
, nr_pages
, ext3_get_block
);
1766 static void ext3_invalidatepage(struct page
*page
, unsigned long offset
)
1768 journal_t
*journal
= EXT3_JOURNAL(page
->mapping
->host
);
1770 trace_ext3_invalidatepage(page
, offset
);
1773 * If it's a full truncate we just forget about the pending dirtying
1776 ClearPageChecked(page
);
1778 journal_invalidatepage(journal
, page
, offset
);
1781 static int ext3_releasepage(struct page
*page
, gfp_t wait
)
1783 journal_t
*journal
= EXT3_JOURNAL(page
->mapping
->host
);
1785 trace_ext3_releasepage(page
);
1786 WARN_ON(PageChecked(page
));
1787 if (!page_has_buffers(page
))
1789 return journal_try_to_free_buffers(journal
, page
, wait
);
1793 * If the O_DIRECT write will extend the file then add this inode to the
1794 * orphan list. So recovery will truncate it back to the original size
1795 * if the machine crashes during the write.
1797 * If the O_DIRECT write is intantiating holes inside i_size and the machine
1798 * crashes then stale disk data _may_ be exposed inside the file. But current
1799 * VFS code falls back into buffered path in that case so we are safe.
1801 static ssize_t
ext3_direct_IO(int rw
, struct kiocb
*iocb
,
1802 const struct iovec
*iov
, loff_t offset
,
1803 unsigned long nr_segs
)
1805 struct file
*file
= iocb
->ki_filp
;
1806 struct inode
*inode
= file
->f_mapping
->host
;
1807 struct ext3_inode_info
*ei
= EXT3_I(inode
);
1811 size_t count
= iov_length(iov
, nr_segs
);
1814 trace_ext3_direct_IO_enter(inode
, offset
, iov_length(iov
, nr_segs
), rw
);
1817 loff_t final_size
= offset
+ count
;
1819 if (final_size
> inode
->i_size
) {
1820 /* Credits for sb + inode write */
1821 handle
= ext3_journal_start(inode
, 2);
1822 if (IS_ERR(handle
)) {
1823 ret
= PTR_ERR(handle
);
1826 ret
= ext3_orphan_add(handle
, inode
);
1828 ext3_journal_stop(handle
);
1832 ei
->i_disksize
= inode
->i_size
;
1833 ext3_journal_stop(handle
);
1838 ret
= blockdev_direct_IO(rw
, iocb
, inode
, inode
->i_sb
->s_bdev
, iov
,
1840 ext3_get_block
, NULL
);
1842 * In case of error extending write may have instantiated a few
1843 * blocks outside i_size. Trim these off again.
1845 if (unlikely((rw
& WRITE
) && ret
< 0)) {
1846 loff_t isize
= i_size_read(inode
);
1847 loff_t end
= offset
+ iov_length(iov
, nr_segs
);
1850 ext3_truncate_failed_write(inode
);
1852 if (ret
== -ENOSPC
&& ext3_should_retry_alloc(inode
->i_sb
, &retries
))
1858 /* Credits for sb + inode write */
1859 handle
= ext3_journal_start(inode
, 2);
1860 if (IS_ERR(handle
)) {
1861 /* This is really bad luck. We've written the data
1862 * but cannot extend i_size. Truncate allocated blocks
1863 * and pretend the write failed... */
1864 ext3_truncate_failed_write(inode
);
1865 ret
= PTR_ERR(handle
);
1869 ext3_orphan_del(handle
, inode
);
1871 loff_t end
= offset
+ ret
;
1872 if (end
> inode
->i_size
) {
1873 ei
->i_disksize
= end
;
1874 i_size_write(inode
, end
);
1876 * We're going to return a positive `ret'
1877 * here due to non-zero-length I/O, so there's
1878 * no way of reporting error returns from
1879 * ext3_mark_inode_dirty() to userspace. So
1882 ext3_mark_inode_dirty(handle
, inode
);
1885 err
= ext3_journal_stop(handle
);
1890 trace_ext3_direct_IO_exit(inode
, offset
,
1891 iov_length(iov
, nr_segs
), rw
, ret
);
1896 * Pages can be marked dirty completely asynchronously from ext3's journalling
1897 * activity. By filemap_sync_pte(), try_to_unmap_one(), etc. We cannot do
1898 * much here because ->set_page_dirty is called under VFS locks. The page is
1899 * not necessarily locked.
1901 * We cannot just dirty the page and leave attached buffers clean, because the
1902 * buffers' dirty state is "definitive". We cannot just set the buffers dirty
1903 * or jbddirty because all the journalling code will explode.
1905 * So what we do is to mark the page "pending dirty" and next time writepage
1906 * is called, propagate that into the buffers appropriately.
1908 static int ext3_journalled_set_page_dirty(struct page
*page
)
1910 SetPageChecked(page
);
1911 return __set_page_dirty_nobuffers(page
);
1914 static const struct address_space_operations ext3_ordered_aops
= {
1915 .readpage
= ext3_readpage
,
1916 .readpages
= ext3_readpages
,
1917 .writepage
= ext3_ordered_writepage
,
1918 .write_begin
= ext3_write_begin
,
1919 .write_end
= ext3_ordered_write_end
,
1921 .invalidatepage
= ext3_invalidatepage
,
1922 .releasepage
= ext3_releasepage
,
1923 .direct_IO
= ext3_direct_IO
,
1924 .migratepage
= buffer_migrate_page
,
1925 .is_partially_uptodate
= block_is_partially_uptodate
,
1926 .error_remove_page
= generic_error_remove_page
,
1929 static const struct address_space_operations ext3_writeback_aops
= {
1930 .readpage
= ext3_readpage
,
1931 .readpages
= ext3_readpages
,
1932 .writepage
= ext3_writeback_writepage
,
1933 .write_begin
= ext3_write_begin
,
1934 .write_end
= ext3_writeback_write_end
,
1936 .invalidatepage
= ext3_invalidatepage
,
1937 .releasepage
= ext3_releasepage
,
1938 .direct_IO
= ext3_direct_IO
,
1939 .migratepage
= buffer_migrate_page
,
1940 .is_partially_uptodate
= block_is_partially_uptodate
,
1941 .error_remove_page
= generic_error_remove_page
,
1944 static const struct address_space_operations ext3_journalled_aops
= {
1945 .readpage
= ext3_readpage
,
1946 .readpages
= ext3_readpages
,
1947 .writepage
= ext3_journalled_writepage
,
1948 .write_begin
= ext3_write_begin
,
1949 .write_end
= ext3_journalled_write_end
,
1950 .set_page_dirty
= ext3_journalled_set_page_dirty
,
1952 .invalidatepage
= ext3_invalidatepage
,
1953 .releasepage
= ext3_releasepage
,
1954 .is_partially_uptodate
= block_is_partially_uptodate
,
1955 .error_remove_page
= generic_error_remove_page
,
1958 void ext3_set_aops(struct inode
*inode
)
1960 if (ext3_should_order_data(inode
))
1961 inode
->i_mapping
->a_ops
= &ext3_ordered_aops
;
1962 else if (ext3_should_writeback_data(inode
))
1963 inode
->i_mapping
->a_ops
= &ext3_writeback_aops
;
1965 inode
->i_mapping
->a_ops
= &ext3_journalled_aops
;
1969 * ext3_block_truncate_page() zeroes out a mapping from file offset `from'
1970 * up to the end of the block which corresponds to `from'.
1971 * This required during truncate. We need to physically zero the tail end
1972 * of that block so it doesn't yield old data if the file is later grown.
1974 static int ext3_block_truncate_page(handle_t
*handle
, struct page
*page
,
1975 struct address_space
*mapping
, loff_t from
)
1977 ext3_fsblk_t index
= from
>> PAGE_CACHE_SHIFT
;
1978 unsigned offset
= from
& (PAGE_CACHE_SIZE
-1);
1979 unsigned blocksize
, iblock
, length
, pos
;
1980 struct inode
*inode
= mapping
->host
;
1981 struct buffer_head
*bh
;
1984 blocksize
= inode
->i_sb
->s_blocksize
;
1985 length
= blocksize
- (offset
& (blocksize
- 1));
1986 iblock
= index
<< (PAGE_CACHE_SHIFT
- inode
->i_sb
->s_blocksize_bits
);
1988 if (!page_has_buffers(page
))
1989 create_empty_buffers(page
, blocksize
, 0);
1991 /* Find the buffer that contains "offset" */
1992 bh
= page_buffers(page
);
1994 while (offset
>= pos
) {
1995 bh
= bh
->b_this_page
;
2001 if (buffer_freed(bh
)) {
2002 BUFFER_TRACE(bh
, "freed: skip");
2006 if (!buffer_mapped(bh
)) {
2007 BUFFER_TRACE(bh
, "unmapped");
2008 ext3_get_block(inode
, iblock
, bh
, 0);
2009 /* unmapped? It's a hole - nothing to do */
2010 if (!buffer_mapped(bh
)) {
2011 BUFFER_TRACE(bh
, "still unmapped");
2016 /* Ok, it's mapped. Make sure it's up-to-date */
2017 if (PageUptodate(page
))
2018 set_buffer_uptodate(bh
);
2020 if (!buffer_uptodate(bh
)) {
2022 ll_rw_block(READ
, 1, &bh
);
2024 /* Uhhuh. Read error. Complain and punt. */
2025 if (!buffer_uptodate(bh
))
2029 if (ext3_should_journal_data(inode
)) {
2030 BUFFER_TRACE(bh
, "get write access");
2031 err
= ext3_journal_get_write_access(handle
, bh
);
2036 zero_user(page
, offset
, length
);
2037 BUFFER_TRACE(bh
, "zeroed end of block");
2040 if (ext3_should_journal_data(inode
)) {
2041 err
= ext3_journal_dirty_metadata(handle
, bh
);
2043 if (ext3_should_order_data(inode
))
2044 err
= ext3_journal_dirty_data(handle
, bh
);
2045 mark_buffer_dirty(bh
);
2050 page_cache_release(page
);
2055 * Probably it should be a library function... search for first non-zero word
2056 * or memcmp with zero_page, whatever is better for particular architecture.
2059 static inline int all_zeroes(__le32
*p
, __le32
*q
)
2068 * ext3_find_shared - find the indirect blocks for partial truncation.
2069 * @inode: inode in question
2070 * @depth: depth of the affected branch
2071 * @offsets: offsets of pointers in that branch (see ext3_block_to_path)
2072 * @chain: place to store the pointers to partial indirect blocks
2073 * @top: place to the (detached) top of branch
2075 * This is a helper function used by ext3_truncate().
2077 * When we do truncate() we may have to clean the ends of several
2078 * indirect blocks but leave the blocks themselves alive. Block is
2079 * partially truncated if some data below the new i_size is referred
2080 * from it (and it is on the path to the first completely truncated
2081 * data block, indeed). We have to free the top of that path along
2082 * with everything to the right of the path. Since no allocation
2083 * past the truncation point is possible until ext3_truncate()
2084 * finishes, we may safely do the latter, but top of branch may
2085 * require special attention - pageout below the truncation point
2086 * might try to populate it.
2088 * We atomically detach the top of branch from the tree, store the
2089 * block number of its root in *@top, pointers to buffer_heads of
2090 * partially truncated blocks - in @chain[].bh and pointers to
2091 * their last elements that should not be removed - in
2092 * @chain[].p. Return value is the pointer to last filled element
2095 * The work left to caller to do the actual freeing of subtrees:
2096 * a) free the subtree starting from *@top
2097 * b) free the subtrees whose roots are stored in
2098 * (@chain[i].p+1 .. end of @chain[i].bh->b_data)
2099 * c) free the subtrees growing from the inode past the @chain[0].
2100 * (no partially truncated stuff there). */
2102 static Indirect
*ext3_find_shared(struct inode
*inode
, int depth
,
2103 int offsets
[4], Indirect chain
[4], __le32
*top
)
2105 Indirect
*partial
, *p
;
2109 /* Make k index the deepest non-null offset + 1 */
2110 for (k
= depth
; k
> 1 && !offsets
[k
-1]; k
--)
2112 partial
= ext3_get_branch(inode
, k
, offsets
, chain
, &err
);
2113 /* Writer: pointers */
2115 partial
= chain
+ k
-1;
2117 * If the branch acquired continuation since we've looked at it -
2118 * fine, it should all survive and (new) top doesn't belong to us.
2120 if (!partial
->key
&& *partial
->p
)
2123 for (p
=partial
; p
>chain
&& all_zeroes((__le32
*)p
->bh
->b_data
,p
->p
); p
--)
2126 * OK, we've found the last block that must survive. The rest of our
2127 * branch should be detached before unlocking. However, if that rest
2128 * of branch is all ours and does not grow immediately from the inode
2129 * it's easier to cheat and just decrement partial->p.
2131 if (p
== chain
+ k
- 1 && p
> chain
) {
2135 /* Nope, don't do this in ext3. Must leave the tree intact */
2142 while(partial
> p
) {
2143 brelse(partial
->bh
);
2151 * Zero a number of block pointers in either an inode or an indirect block.
2152 * If we restart the transaction we must again get write access to the
2153 * indirect block for further modification.
2155 * We release `count' blocks on disk, but (last - first) may be greater
2156 * than `count' because there can be holes in there.
2158 static void ext3_clear_blocks(handle_t
*handle
, struct inode
*inode
,
2159 struct buffer_head
*bh
, ext3_fsblk_t block_to_free
,
2160 unsigned long count
, __le32
*first
, __le32
*last
)
2163 if (try_to_extend_transaction(handle
, inode
)) {
2165 BUFFER_TRACE(bh
, "call ext3_journal_dirty_metadata");
2166 if (ext3_journal_dirty_metadata(handle
, bh
))
2169 ext3_mark_inode_dirty(handle
, inode
);
2170 truncate_restart_transaction(handle
, inode
);
2172 BUFFER_TRACE(bh
, "retaking write access");
2173 if (ext3_journal_get_write_access(handle
, bh
))
2179 * Any buffers which are on the journal will be in memory. We find
2180 * them on the hash table so journal_revoke() will run journal_forget()
2181 * on them. We've already detached each block from the file, so
2182 * bforget() in journal_forget() should be safe.
2184 * AKPM: turn on bforget in journal_forget()!!!
2186 for (p
= first
; p
< last
; p
++) {
2187 u32 nr
= le32_to_cpu(*p
);
2189 struct buffer_head
*bh
;
2192 bh
= sb_find_get_block(inode
->i_sb
, nr
);
2193 ext3_forget(handle
, 0, inode
, bh
, nr
);
2197 ext3_free_blocks(handle
, inode
, block_to_free
, count
);
2201 * ext3_free_data - free a list of data blocks
2202 * @handle: handle for this transaction
2203 * @inode: inode we are dealing with
2204 * @this_bh: indirect buffer_head which contains *@first and *@last
2205 * @first: array of block numbers
2206 * @last: points immediately past the end of array
2208 * We are freeing all blocks referred from that array (numbers are stored as
2209 * little-endian 32-bit) and updating @inode->i_blocks appropriately.
2211 * We accumulate contiguous runs of blocks to free. Conveniently, if these
2212 * blocks are contiguous then releasing them at one time will only affect one
2213 * or two bitmap blocks (+ group descriptor(s) and superblock) and we won't
2214 * actually use a lot of journal space.
2216 * @this_bh will be %NULL if @first and @last point into the inode's direct
2219 static void ext3_free_data(handle_t
*handle
, struct inode
*inode
,
2220 struct buffer_head
*this_bh
,
2221 __le32
*first
, __le32
*last
)
2223 ext3_fsblk_t block_to_free
= 0; /* Starting block # of a run */
2224 unsigned long count
= 0; /* Number of blocks in the run */
2225 __le32
*block_to_free_p
= NULL
; /* Pointer into inode/ind
2228 ext3_fsblk_t nr
; /* Current block # */
2229 __le32
*p
; /* Pointer into inode/ind
2230 for current block */
2233 if (this_bh
) { /* For indirect block */
2234 BUFFER_TRACE(this_bh
, "get_write_access");
2235 err
= ext3_journal_get_write_access(handle
, this_bh
);
2236 /* Important: if we can't update the indirect pointers
2237 * to the blocks, we can't free them. */
2242 for (p
= first
; p
< last
; p
++) {
2243 nr
= le32_to_cpu(*p
);
2245 /* accumulate blocks to free if they're contiguous */
2248 block_to_free_p
= p
;
2250 } else if (nr
== block_to_free
+ count
) {
2253 ext3_clear_blocks(handle
, inode
, this_bh
,
2255 count
, block_to_free_p
, p
);
2257 block_to_free_p
= p
;
2264 ext3_clear_blocks(handle
, inode
, this_bh
, block_to_free
,
2265 count
, block_to_free_p
, p
);
2268 BUFFER_TRACE(this_bh
, "call ext3_journal_dirty_metadata");
2271 * The buffer head should have an attached journal head at this
2272 * point. However, if the data is corrupted and an indirect
2273 * block pointed to itself, it would have been detached when
2274 * the block was cleared. Check for this instead of OOPSing.
2277 ext3_journal_dirty_metadata(handle
, this_bh
);
2279 ext3_error(inode
->i_sb
, "ext3_free_data",
2280 "circular indirect block detected, "
2281 "inode=%lu, block=%llu",
2283 (unsigned long long)this_bh
->b_blocknr
);
2288 * ext3_free_branches - free an array of branches
2289 * @handle: JBD handle for this transaction
2290 * @inode: inode we are dealing with
2291 * @parent_bh: the buffer_head which contains *@first and *@last
2292 * @first: array of block numbers
2293 * @last: pointer immediately past the end of array
2294 * @depth: depth of the branches to free
2296 * We are freeing all blocks referred from these branches (numbers are
2297 * stored as little-endian 32-bit) and updating @inode->i_blocks
2300 static void ext3_free_branches(handle_t
*handle
, struct inode
*inode
,
2301 struct buffer_head
*parent_bh
,
2302 __le32
*first
, __le32
*last
, int depth
)
2307 if (is_handle_aborted(handle
))
2311 struct buffer_head
*bh
;
2312 int addr_per_block
= EXT3_ADDR_PER_BLOCK(inode
->i_sb
);
2314 while (--p
>= first
) {
2315 nr
= le32_to_cpu(*p
);
2317 continue; /* A hole */
2319 /* Go read the buffer for the next level down */
2320 bh
= sb_bread(inode
->i_sb
, nr
);
2323 * A read failure? Report error and clear slot
2327 ext3_error(inode
->i_sb
, "ext3_free_branches",
2328 "Read failure, inode=%lu, block="E3FSBLK
,
2333 /* This zaps the entire block. Bottom up. */
2334 BUFFER_TRACE(bh
, "free child branches");
2335 ext3_free_branches(handle
, inode
, bh
,
2336 (__le32
*)bh
->b_data
,
2337 (__le32
*)bh
->b_data
+ addr_per_block
,
2341 * Everything below this this pointer has been
2342 * released. Now let this top-of-subtree go.
2344 * We want the freeing of this indirect block to be
2345 * atomic in the journal with the updating of the
2346 * bitmap block which owns it. So make some room in
2349 * We zero the parent pointer *after* freeing its
2350 * pointee in the bitmaps, so if extend_transaction()
2351 * for some reason fails to put the bitmap changes and
2352 * the release into the same transaction, recovery
2353 * will merely complain about releasing a free block,
2354 * rather than leaking blocks.
2356 if (is_handle_aborted(handle
))
2358 if (try_to_extend_transaction(handle
, inode
)) {
2359 ext3_mark_inode_dirty(handle
, inode
);
2360 truncate_restart_transaction(handle
, inode
);
2364 * We've probably journalled the indirect block several
2365 * times during the truncate. But it's no longer
2366 * needed and we now drop it from the transaction via
2369 * That's easy if it's exclusively part of this
2370 * transaction. But if it's part of the committing
2371 * transaction then journal_forget() will simply
2372 * brelse() it. That means that if the underlying
2373 * block is reallocated in ext3_get_block(),
2374 * unmap_underlying_metadata() will find this block
2375 * and will try to get rid of it. damn, damn. Thus
2376 * we don't allow a block to be reallocated until
2377 * a transaction freeing it has fully committed.
2379 * We also have to make sure journal replay after a
2380 * crash does not overwrite non-journaled data blocks
2381 * with old metadata when the block got reallocated for
2382 * data. Thus we have to store a revoke record for a
2383 * block in the same transaction in which we free the
2386 ext3_forget(handle
, 1, inode
, bh
, bh
->b_blocknr
);
2388 ext3_free_blocks(handle
, inode
, nr
, 1);
2392 * The block which we have just freed is
2393 * pointed to by an indirect block: journal it
2395 BUFFER_TRACE(parent_bh
, "get_write_access");
2396 if (!ext3_journal_get_write_access(handle
,
2399 BUFFER_TRACE(parent_bh
,
2400 "call ext3_journal_dirty_metadata");
2401 ext3_journal_dirty_metadata(handle
,
2407 /* We have reached the bottom of the tree. */
2408 BUFFER_TRACE(parent_bh
, "free data blocks");
2409 ext3_free_data(handle
, inode
, parent_bh
, first
, last
);
2413 int ext3_can_truncate(struct inode
*inode
)
2415 if (S_ISREG(inode
->i_mode
))
2417 if (S_ISDIR(inode
->i_mode
))
2419 if (S_ISLNK(inode
->i_mode
))
2420 return !ext3_inode_is_fast_symlink(inode
);
2427 * We block out ext3_get_block() block instantiations across the entire
2428 * transaction, and VFS/VM ensures that ext3_truncate() cannot run
2429 * simultaneously on behalf of the same inode.
2431 * As we work through the truncate and commmit bits of it to the journal there
2432 * is one core, guiding principle: the file's tree must always be consistent on
2433 * disk. We must be able to restart the truncate after a crash.
2435 * The file's tree may be transiently inconsistent in memory (although it
2436 * probably isn't), but whenever we close off and commit a journal transaction,
2437 * the contents of (the filesystem + the journal) must be consistent and
2438 * restartable. It's pretty simple, really: bottom up, right to left (although
2439 * left-to-right works OK too).
2441 * Note that at recovery time, journal replay occurs *before* the restart of
2442 * truncate against the orphan inode list.
2444 * The committed inode has the new, desired i_size (which is the same as
2445 * i_disksize in this case). After a crash, ext3_orphan_cleanup() will see
2446 * that this inode's truncate did not complete and it will again call
2447 * ext3_truncate() to have another go. So there will be instantiated blocks
2448 * to the right of the truncation point in a crashed ext3 filesystem. But
2449 * that's fine - as long as they are linked from the inode, the post-crash
2450 * ext3_truncate() run will find them and release them.
2452 void ext3_truncate(struct inode
*inode
)
2455 struct ext3_inode_info
*ei
= EXT3_I(inode
);
2456 __le32
*i_data
= ei
->i_data
;
2457 int addr_per_block
= EXT3_ADDR_PER_BLOCK(inode
->i_sb
);
2458 struct address_space
*mapping
= inode
->i_mapping
;
2465 unsigned blocksize
= inode
->i_sb
->s_blocksize
;
2468 trace_ext3_truncate_enter(inode
);
2470 if (!ext3_can_truncate(inode
))
2473 if (inode
->i_size
== 0 && ext3_should_writeback_data(inode
))
2474 ext3_set_inode_state(inode
, EXT3_STATE_FLUSH_ON_CLOSE
);
2477 * We have to lock the EOF page here, because lock_page() nests
2478 * outside journal_start().
2480 if ((inode
->i_size
& (blocksize
- 1)) == 0) {
2481 /* Block boundary? Nothing to do */
2484 page
= grab_cache_page(mapping
,
2485 inode
->i_size
>> PAGE_CACHE_SHIFT
);
2490 handle
= start_transaction(inode
);
2491 if (IS_ERR(handle
)) {
2493 clear_highpage(page
);
2494 flush_dcache_page(page
);
2496 page_cache_release(page
);
2501 last_block
= (inode
->i_size
+ blocksize
-1)
2502 >> EXT3_BLOCK_SIZE_BITS(inode
->i_sb
);
2505 ext3_block_truncate_page(handle
, page
, mapping
, inode
->i_size
);
2507 n
= ext3_block_to_path(inode
, last_block
, offsets
, NULL
);
2509 goto out_stop
; /* error */
2512 * OK. This truncate is going to happen. We add the inode to the
2513 * orphan list, so that if this truncate spans multiple transactions,
2514 * and we crash, we will resume the truncate when the filesystem
2515 * recovers. It also marks the inode dirty, to catch the new size.
2517 * Implication: the file must always be in a sane, consistent
2518 * truncatable state while each transaction commits.
2520 if (ext3_orphan_add(handle
, inode
))
2524 * The orphan list entry will now protect us from any crash which
2525 * occurs before the truncate completes, so it is now safe to propagate
2526 * the new, shorter inode size (held for now in i_size) into the
2527 * on-disk inode. We do this via i_disksize, which is the value which
2528 * ext3 *really* writes onto the disk inode.
2530 ei
->i_disksize
= inode
->i_size
;
2533 * From here we block out all ext3_get_block() callers who want to
2534 * modify the block allocation tree.
2536 mutex_lock(&ei
->truncate_mutex
);
2538 if (n
== 1) { /* direct blocks */
2539 ext3_free_data(handle
, inode
, NULL
, i_data
+offsets
[0],
2540 i_data
+ EXT3_NDIR_BLOCKS
);
2544 partial
= ext3_find_shared(inode
, n
, offsets
, chain
, &nr
);
2545 /* Kill the top of shared branch (not detached) */
2547 if (partial
== chain
) {
2548 /* Shared branch grows from the inode */
2549 ext3_free_branches(handle
, inode
, NULL
,
2550 &nr
, &nr
+1, (chain
+n
-1) - partial
);
2553 * We mark the inode dirty prior to restart,
2554 * and prior to stop. No need for it here.
2557 /* Shared branch grows from an indirect block */
2558 ext3_free_branches(handle
, inode
, partial
->bh
,
2560 partial
->p
+1, (chain
+n
-1) - partial
);
2563 /* Clear the ends of indirect blocks on the shared branch */
2564 while (partial
> chain
) {
2565 ext3_free_branches(handle
, inode
, partial
->bh
, partial
->p
+ 1,
2566 (__le32
*)partial
->bh
->b_data
+addr_per_block
,
2567 (chain
+n
-1) - partial
);
2568 BUFFER_TRACE(partial
->bh
, "call brelse");
2569 brelse (partial
->bh
);
2573 /* Kill the remaining (whole) subtrees */
2574 switch (offsets
[0]) {
2576 nr
= i_data
[EXT3_IND_BLOCK
];
2578 ext3_free_branches(handle
, inode
, NULL
, &nr
, &nr
+1, 1);
2579 i_data
[EXT3_IND_BLOCK
] = 0;
2581 case EXT3_IND_BLOCK
:
2582 nr
= i_data
[EXT3_DIND_BLOCK
];
2584 ext3_free_branches(handle
, inode
, NULL
, &nr
, &nr
+1, 2);
2585 i_data
[EXT3_DIND_BLOCK
] = 0;
2587 case EXT3_DIND_BLOCK
:
2588 nr
= i_data
[EXT3_TIND_BLOCK
];
2590 ext3_free_branches(handle
, inode
, NULL
, &nr
, &nr
+1, 3);
2591 i_data
[EXT3_TIND_BLOCK
] = 0;
2593 case EXT3_TIND_BLOCK
:
2597 ext3_discard_reservation(inode
);
2599 mutex_unlock(&ei
->truncate_mutex
);
2600 inode
->i_mtime
= inode
->i_ctime
= CURRENT_TIME_SEC
;
2601 ext3_mark_inode_dirty(handle
, inode
);
2604 * In a multi-transaction truncate, we only make the final transaction
2611 * If this was a simple ftruncate(), and the file will remain alive
2612 * then we need to clear up the orphan record which we created above.
2613 * However, if this was a real unlink then we were called by
2614 * ext3_evict_inode(), and we allow that function to clean up the
2615 * orphan info for us.
2618 ext3_orphan_del(handle
, inode
);
2620 ext3_journal_stop(handle
);
2621 trace_ext3_truncate_exit(inode
);
2625 * Delete the inode from orphan list so that it doesn't stay there
2626 * forever and trigger assertion on umount.
2629 ext3_orphan_del(NULL
, inode
);
2630 trace_ext3_truncate_exit(inode
);
2633 static ext3_fsblk_t
ext3_get_inode_block(struct super_block
*sb
,
2634 unsigned long ino
, struct ext3_iloc
*iloc
)
2636 unsigned long block_group
;
2637 unsigned long offset
;
2639 struct ext3_group_desc
*gdp
;
2641 if (!ext3_valid_inum(sb
, ino
)) {
2643 * This error is already checked for in namei.c unless we are
2644 * looking at an NFS filehandle, in which case no error
2650 block_group
= (ino
- 1) / EXT3_INODES_PER_GROUP(sb
);
2651 gdp
= ext3_get_group_desc(sb
, block_group
, NULL
);
2655 * Figure out the offset within the block group inode table
2657 offset
= ((ino
- 1) % EXT3_INODES_PER_GROUP(sb
)) *
2658 EXT3_INODE_SIZE(sb
);
2659 block
= le32_to_cpu(gdp
->bg_inode_table
) +
2660 (offset
>> EXT3_BLOCK_SIZE_BITS(sb
));
2662 iloc
->block_group
= block_group
;
2663 iloc
->offset
= offset
& (EXT3_BLOCK_SIZE(sb
) - 1);
2668 * ext3_get_inode_loc returns with an extra refcount against the inode's
2669 * underlying buffer_head on success. If 'in_mem' is true, we have all
2670 * data in memory that is needed to recreate the on-disk version of this
2673 static int __ext3_get_inode_loc(struct inode
*inode
,
2674 struct ext3_iloc
*iloc
, int in_mem
)
2677 struct buffer_head
*bh
;
2679 block
= ext3_get_inode_block(inode
->i_sb
, inode
->i_ino
, iloc
);
2683 bh
= sb_getblk(inode
->i_sb
, block
);
2685 ext3_error (inode
->i_sb
, "ext3_get_inode_loc",
2686 "unable to read inode block - "
2687 "inode=%lu, block="E3FSBLK
,
2688 inode
->i_ino
, block
);
2691 if (!buffer_uptodate(bh
)) {
2695 * If the buffer has the write error flag, we have failed
2696 * to write out another inode in the same block. In this
2697 * case, we don't have to read the block because we may
2698 * read the old inode data successfully.
2700 if (buffer_write_io_error(bh
) && !buffer_uptodate(bh
))
2701 set_buffer_uptodate(bh
);
2703 if (buffer_uptodate(bh
)) {
2704 /* someone brought it uptodate while we waited */
2710 * If we have all information of the inode in memory and this
2711 * is the only valid inode in the block, we need not read the
2715 struct buffer_head
*bitmap_bh
;
2716 struct ext3_group_desc
*desc
;
2717 int inodes_per_buffer
;
2718 int inode_offset
, i
;
2722 block_group
= (inode
->i_ino
- 1) /
2723 EXT3_INODES_PER_GROUP(inode
->i_sb
);
2724 inodes_per_buffer
= bh
->b_size
/
2725 EXT3_INODE_SIZE(inode
->i_sb
);
2726 inode_offset
= ((inode
->i_ino
- 1) %
2727 EXT3_INODES_PER_GROUP(inode
->i_sb
));
2728 start
= inode_offset
& ~(inodes_per_buffer
- 1);
2730 /* Is the inode bitmap in cache? */
2731 desc
= ext3_get_group_desc(inode
->i_sb
,
2736 bitmap_bh
= sb_getblk(inode
->i_sb
,
2737 le32_to_cpu(desc
->bg_inode_bitmap
));
2742 * If the inode bitmap isn't in cache then the
2743 * optimisation may end up performing two reads instead
2744 * of one, so skip it.
2746 if (!buffer_uptodate(bitmap_bh
)) {
2750 for (i
= start
; i
< start
+ inodes_per_buffer
; i
++) {
2751 if (i
== inode_offset
)
2753 if (ext3_test_bit(i
, bitmap_bh
->b_data
))
2757 if (i
== start
+ inodes_per_buffer
) {
2758 /* all other inodes are free, so skip I/O */
2759 memset(bh
->b_data
, 0, bh
->b_size
);
2760 set_buffer_uptodate(bh
);
2768 * There are other valid inodes in the buffer, this inode
2769 * has in-inode xattrs, or we don't have this inode in memory.
2770 * Read the block from disk.
2772 trace_ext3_load_inode(inode
);
2774 bh
->b_end_io
= end_buffer_read_sync
;
2775 submit_bh(READ_META
, bh
);
2777 if (!buffer_uptodate(bh
)) {
2778 ext3_error(inode
->i_sb
, "ext3_get_inode_loc",
2779 "unable to read inode block - "
2780 "inode=%lu, block="E3FSBLK
,
2781 inode
->i_ino
, block
);
2791 int ext3_get_inode_loc(struct inode
*inode
, struct ext3_iloc
*iloc
)
2793 /* We have all inode data except xattrs in memory here. */
2794 return __ext3_get_inode_loc(inode
, iloc
,
2795 !ext3_test_inode_state(inode
, EXT3_STATE_XATTR
));
2798 void ext3_set_inode_flags(struct inode
*inode
)
2800 unsigned int flags
= EXT3_I(inode
)->i_flags
;
2802 inode
->i_flags
&= ~(S_SYNC
|S_APPEND
|S_IMMUTABLE
|S_NOATIME
|S_DIRSYNC
);
2803 if (flags
& EXT3_SYNC_FL
)
2804 inode
->i_flags
|= S_SYNC
;
2805 if (flags
& EXT3_APPEND_FL
)
2806 inode
->i_flags
|= S_APPEND
;
2807 if (flags
& EXT3_IMMUTABLE_FL
)
2808 inode
->i_flags
|= S_IMMUTABLE
;
2809 if (flags
& EXT3_NOATIME_FL
)
2810 inode
->i_flags
|= S_NOATIME
;
2811 if (flags
& EXT3_DIRSYNC_FL
)
2812 inode
->i_flags
|= S_DIRSYNC
;
2815 /* Propagate flags from i_flags to EXT3_I(inode)->i_flags */
2816 void ext3_get_inode_flags(struct ext3_inode_info
*ei
)
2818 unsigned int flags
= ei
->vfs_inode
.i_flags
;
2820 ei
->i_flags
&= ~(EXT3_SYNC_FL
|EXT3_APPEND_FL
|
2821 EXT3_IMMUTABLE_FL
|EXT3_NOATIME_FL
|EXT3_DIRSYNC_FL
);
2823 ei
->i_flags
|= EXT3_SYNC_FL
;
2824 if (flags
& S_APPEND
)
2825 ei
->i_flags
|= EXT3_APPEND_FL
;
2826 if (flags
& S_IMMUTABLE
)
2827 ei
->i_flags
|= EXT3_IMMUTABLE_FL
;
2828 if (flags
& S_NOATIME
)
2829 ei
->i_flags
|= EXT3_NOATIME_FL
;
2830 if (flags
& S_DIRSYNC
)
2831 ei
->i_flags
|= EXT3_DIRSYNC_FL
;
2834 struct inode
*ext3_iget(struct super_block
*sb
, unsigned long ino
)
2836 struct ext3_iloc iloc
;
2837 struct ext3_inode
*raw_inode
;
2838 struct ext3_inode_info
*ei
;
2839 struct buffer_head
*bh
;
2840 struct inode
*inode
;
2841 journal_t
*journal
= EXT3_SB(sb
)->s_journal
;
2842 transaction_t
*transaction
;
2846 inode
= iget_locked(sb
, ino
);
2848 return ERR_PTR(-ENOMEM
);
2849 if (!(inode
->i_state
& I_NEW
))
2853 ei
->i_block_alloc_info
= NULL
;
2855 ret
= __ext3_get_inode_loc(inode
, &iloc
, 0);
2859 raw_inode
= ext3_raw_inode(&iloc
);
2860 inode
->i_mode
= le16_to_cpu(raw_inode
->i_mode
);
2861 inode
->i_uid
= (uid_t
)le16_to_cpu(raw_inode
->i_uid_low
);
2862 inode
->i_gid
= (gid_t
)le16_to_cpu(raw_inode
->i_gid_low
);
2863 if(!(test_opt (inode
->i_sb
, NO_UID32
))) {
2864 inode
->i_uid
|= le16_to_cpu(raw_inode
->i_uid_high
) << 16;
2865 inode
->i_gid
|= le16_to_cpu(raw_inode
->i_gid_high
) << 16;
2867 inode
->i_nlink
= le16_to_cpu(raw_inode
->i_links_count
);
2868 inode
->i_size
= le32_to_cpu(raw_inode
->i_size
);
2869 inode
->i_atime
.tv_sec
= (signed)le32_to_cpu(raw_inode
->i_atime
);
2870 inode
->i_ctime
.tv_sec
= (signed)le32_to_cpu(raw_inode
->i_ctime
);
2871 inode
->i_mtime
.tv_sec
= (signed)le32_to_cpu(raw_inode
->i_mtime
);
2872 inode
->i_atime
.tv_nsec
= inode
->i_ctime
.tv_nsec
= inode
->i_mtime
.tv_nsec
= 0;
2874 ei
->i_state_flags
= 0;
2875 ei
->i_dir_start_lookup
= 0;
2876 ei
->i_dtime
= le32_to_cpu(raw_inode
->i_dtime
);
2877 /* We now have enough fields to check if the inode was active or not.
2878 * This is needed because nfsd might try to access dead inodes
2879 * the test is that same one that e2fsck uses
2880 * NeilBrown 1999oct15
2882 if (inode
->i_nlink
== 0) {
2883 if (inode
->i_mode
== 0 ||
2884 !(EXT3_SB(inode
->i_sb
)->s_mount_state
& EXT3_ORPHAN_FS
)) {
2885 /* this inode is deleted */
2890 /* The only unlinked inodes we let through here have
2891 * valid i_mode and are being read by the orphan
2892 * recovery code: that's fine, we're about to complete
2893 * the process of deleting those. */
2895 inode
->i_blocks
= le32_to_cpu(raw_inode
->i_blocks
);
2896 ei
->i_flags
= le32_to_cpu(raw_inode
->i_flags
);
2897 #ifdef EXT3_FRAGMENTS
2898 ei
->i_faddr
= le32_to_cpu(raw_inode
->i_faddr
);
2899 ei
->i_frag_no
= raw_inode
->i_frag
;
2900 ei
->i_frag_size
= raw_inode
->i_fsize
;
2902 ei
->i_file_acl
= le32_to_cpu(raw_inode
->i_file_acl
);
2903 if (!S_ISREG(inode
->i_mode
)) {
2904 ei
->i_dir_acl
= le32_to_cpu(raw_inode
->i_dir_acl
);
2907 ((__u64
)le32_to_cpu(raw_inode
->i_size_high
)) << 32;
2909 ei
->i_disksize
= inode
->i_size
;
2910 inode
->i_generation
= le32_to_cpu(raw_inode
->i_generation
);
2911 ei
->i_block_group
= iloc
.block_group
;
2913 * NOTE! The in-memory inode i_data array is in little-endian order
2914 * even on big-endian machines: we do NOT byteswap the block numbers!
2916 for (block
= 0; block
< EXT3_N_BLOCKS
; block
++)
2917 ei
->i_data
[block
] = raw_inode
->i_block
[block
];
2918 INIT_LIST_HEAD(&ei
->i_orphan
);
2921 * Set transaction id's of transactions that have to be committed
2922 * to finish f[data]sync. We set them to currently running transaction
2923 * as we cannot be sure that the inode or some of its metadata isn't
2924 * part of the transaction - the inode could have been reclaimed and
2925 * now it is reread from disk.
2930 spin_lock(&journal
->j_state_lock
);
2931 if (journal
->j_running_transaction
)
2932 transaction
= journal
->j_running_transaction
;
2934 transaction
= journal
->j_committing_transaction
;
2936 tid
= transaction
->t_tid
;
2938 tid
= journal
->j_commit_sequence
;
2939 spin_unlock(&journal
->j_state_lock
);
2940 atomic_set(&ei
->i_sync_tid
, tid
);
2941 atomic_set(&ei
->i_datasync_tid
, tid
);
2944 if (inode
->i_ino
>= EXT3_FIRST_INO(inode
->i_sb
) + 1 &&
2945 EXT3_INODE_SIZE(inode
->i_sb
) > EXT3_GOOD_OLD_INODE_SIZE
) {
2947 * When mke2fs creates big inodes it does not zero out
2948 * the unused bytes above EXT3_GOOD_OLD_INODE_SIZE,
2949 * so ignore those first few inodes.
2951 ei
->i_extra_isize
= le16_to_cpu(raw_inode
->i_extra_isize
);
2952 if (EXT3_GOOD_OLD_INODE_SIZE
+ ei
->i_extra_isize
>
2953 EXT3_INODE_SIZE(inode
->i_sb
)) {
2958 if (ei
->i_extra_isize
== 0) {
2959 /* The extra space is currently unused. Use it. */
2960 ei
->i_extra_isize
= sizeof(struct ext3_inode
) -
2961 EXT3_GOOD_OLD_INODE_SIZE
;
2963 __le32
*magic
= (void *)raw_inode
+
2964 EXT3_GOOD_OLD_INODE_SIZE
+
2966 if (*magic
== cpu_to_le32(EXT3_XATTR_MAGIC
))
2967 ext3_set_inode_state(inode
, EXT3_STATE_XATTR
);
2970 ei
->i_extra_isize
= 0;
2972 if (S_ISREG(inode
->i_mode
)) {
2973 inode
->i_op
= &ext3_file_inode_operations
;
2974 inode
->i_fop
= &ext3_file_operations
;
2975 ext3_set_aops(inode
);
2976 } else if (S_ISDIR(inode
->i_mode
)) {
2977 inode
->i_op
= &ext3_dir_inode_operations
;
2978 inode
->i_fop
= &ext3_dir_operations
;
2979 } else if (S_ISLNK(inode
->i_mode
)) {
2980 if (ext3_inode_is_fast_symlink(inode
)) {
2981 inode
->i_op
= &ext3_fast_symlink_inode_operations
;
2982 nd_terminate_link(ei
->i_data
, inode
->i_size
,
2983 sizeof(ei
->i_data
) - 1);
2985 inode
->i_op
= &ext3_symlink_inode_operations
;
2986 ext3_set_aops(inode
);
2989 inode
->i_op
= &ext3_special_inode_operations
;
2990 if (raw_inode
->i_block
[0])
2991 init_special_inode(inode
, inode
->i_mode
,
2992 old_decode_dev(le32_to_cpu(raw_inode
->i_block
[0])));
2994 init_special_inode(inode
, inode
->i_mode
,
2995 new_decode_dev(le32_to_cpu(raw_inode
->i_block
[1])));
2998 ext3_set_inode_flags(inode
);
2999 unlock_new_inode(inode
);
3004 return ERR_PTR(ret
);
3008 * Post the struct inode info into an on-disk inode location in the
3009 * buffer-cache. This gobbles the caller's reference to the
3010 * buffer_head in the inode location struct.
3012 * The caller must have write access to iloc->bh.
3014 static int ext3_do_update_inode(handle_t
*handle
,
3015 struct inode
*inode
,
3016 struct ext3_iloc
*iloc
)
3018 struct ext3_inode
*raw_inode
= ext3_raw_inode(iloc
);
3019 struct ext3_inode_info
*ei
= EXT3_I(inode
);
3020 struct buffer_head
*bh
= iloc
->bh
;
3021 int err
= 0, rc
, block
;
3024 /* we can't allow multiple procs in here at once, its a bit racey */
3027 /* For fields not not tracking in the in-memory inode,
3028 * initialise them to zero for new inodes. */
3029 if (ext3_test_inode_state(inode
, EXT3_STATE_NEW
))
3030 memset(raw_inode
, 0, EXT3_SB(inode
->i_sb
)->s_inode_size
);
3032 ext3_get_inode_flags(ei
);
3033 raw_inode
->i_mode
= cpu_to_le16(inode
->i_mode
);
3034 if(!(test_opt(inode
->i_sb
, NO_UID32
))) {
3035 raw_inode
->i_uid_low
= cpu_to_le16(low_16_bits(inode
->i_uid
));
3036 raw_inode
->i_gid_low
= cpu_to_le16(low_16_bits(inode
->i_gid
));
3038 * Fix up interoperability with old kernels. Otherwise, old inodes get
3039 * re-used with the upper 16 bits of the uid/gid intact
3042 raw_inode
->i_uid_high
=
3043 cpu_to_le16(high_16_bits(inode
->i_uid
));
3044 raw_inode
->i_gid_high
=
3045 cpu_to_le16(high_16_bits(inode
->i_gid
));
3047 raw_inode
->i_uid_high
= 0;
3048 raw_inode
->i_gid_high
= 0;
3051 raw_inode
->i_uid_low
=
3052 cpu_to_le16(fs_high2lowuid(inode
->i_uid
));
3053 raw_inode
->i_gid_low
=
3054 cpu_to_le16(fs_high2lowgid(inode
->i_gid
));
3055 raw_inode
->i_uid_high
= 0;
3056 raw_inode
->i_gid_high
= 0;
3058 raw_inode
->i_links_count
= cpu_to_le16(inode
->i_nlink
);
3059 raw_inode
->i_size
= cpu_to_le32(ei
->i_disksize
);
3060 raw_inode
->i_atime
= cpu_to_le32(inode
->i_atime
.tv_sec
);
3061 raw_inode
->i_ctime
= cpu_to_le32(inode
->i_ctime
.tv_sec
);
3062 raw_inode
->i_mtime
= cpu_to_le32(inode
->i_mtime
.tv_sec
);
3063 raw_inode
->i_blocks
= cpu_to_le32(inode
->i_blocks
);
3064 raw_inode
->i_dtime
= cpu_to_le32(ei
->i_dtime
);
3065 raw_inode
->i_flags
= cpu_to_le32(ei
->i_flags
);
3066 #ifdef EXT3_FRAGMENTS
3067 raw_inode
->i_faddr
= cpu_to_le32(ei
->i_faddr
);
3068 raw_inode
->i_frag
= ei
->i_frag_no
;
3069 raw_inode
->i_fsize
= ei
->i_frag_size
;
3071 raw_inode
->i_file_acl
= cpu_to_le32(ei
->i_file_acl
);
3072 if (!S_ISREG(inode
->i_mode
)) {
3073 raw_inode
->i_dir_acl
= cpu_to_le32(ei
->i_dir_acl
);
3075 raw_inode
->i_size_high
=
3076 cpu_to_le32(ei
->i_disksize
>> 32);
3077 if (ei
->i_disksize
> 0x7fffffffULL
) {
3078 struct super_block
*sb
= inode
->i_sb
;
3079 if (!EXT3_HAS_RO_COMPAT_FEATURE(sb
,
3080 EXT3_FEATURE_RO_COMPAT_LARGE_FILE
) ||
3081 EXT3_SB(sb
)->s_es
->s_rev_level
==
3082 cpu_to_le32(EXT3_GOOD_OLD_REV
)) {
3083 /* If this is the first large file
3084 * created, add a flag to the superblock.
3087 err
= ext3_journal_get_write_access(handle
,
3088 EXT3_SB(sb
)->s_sbh
);
3092 ext3_update_dynamic_rev(sb
);
3093 EXT3_SET_RO_COMPAT_FEATURE(sb
,
3094 EXT3_FEATURE_RO_COMPAT_LARGE_FILE
);
3096 err
= ext3_journal_dirty_metadata(handle
,
3097 EXT3_SB(sb
)->s_sbh
);
3098 /* get our lock and start over */
3103 raw_inode
->i_generation
= cpu_to_le32(inode
->i_generation
);
3104 if (S_ISCHR(inode
->i_mode
) || S_ISBLK(inode
->i_mode
)) {
3105 if (old_valid_dev(inode
->i_rdev
)) {
3106 raw_inode
->i_block
[0] =
3107 cpu_to_le32(old_encode_dev(inode
->i_rdev
));
3108 raw_inode
->i_block
[1] = 0;
3110 raw_inode
->i_block
[0] = 0;
3111 raw_inode
->i_block
[1] =
3112 cpu_to_le32(new_encode_dev(inode
->i_rdev
));
3113 raw_inode
->i_block
[2] = 0;
3115 } else for (block
= 0; block
< EXT3_N_BLOCKS
; block
++)
3116 raw_inode
->i_block
[block
] = ei
->i_data
[block
];
3118 if (ei
->i_extra_isize
)
3119 raw_inode
->i_extra_isize
= cpu_to_le16(ei
->i_extra_isize
);
3121 BUFFER_TRACE(bh
, "call ext3_journal_dirty_metadata");
3123 rc
= ext3_journal_dirty_metadata(handle
, bh
);
3126 ext3_clear_inode_state(inode
, EXT3_STATE_NEW
);
3128 atomic_set(&ei
->i_sync_tid
, handle
->h_transaction
->t_tid
);
3131 ext3_std_error(inode
->i_sb
, err
);
3136 * ext3_write_inode()
3138 * We are called from a few places:
3140 * - Within generic_file_write() for O_SYNC files.
3141 * Here, there will be no transaction running. We wait for any running
3142 * trasnaction to commit.
3144 * - Within sys_sync(), kupdate and such.
3145 * We wait on commit, if tol to.
3147 * - Within prune_icache() (PF_MEMALLOC == true)
3148 * Here we simply return. We can't afford to block kswapd on the
3151 * In all cases it is actually safe for us to return without doing anything,
3152 * because the inode has been copied into a raw inode buffer in
3153 * ext3_mark_inode_dirty(). This is a correctness thing for O_SYNC and for
3156 * Note that we are absolutely dependent upon all inode dirtiers doing the
3157 * right thing: they *must* call mark_inode_dirty() after dirtying info in
3158 * which we are interested.
3160 * It would be a bug for them to not do this. The code:
3162 * mark_inode_dirty(inode)
3164 * inode->i_size = expr;
3166 * is in error because a kswapd-driven write_inode() could occur while
3167 * `stuff()' is running, and the new i_size will be lost. Plus the inode
3168 * will no longer be on the superblock's dirty inode list.
3170 int ext3_write_inode(struct inode
*inode
, struct writeback_control
*wbc
)
3172 if (current
->flags
& PF_MEMALLOC
)
3175 if (ext3_journal_current_handle()) {
3176 jbd_debug(1, "called recursively, non-PF_MEMALLOC!\n");
3181 if (wbc
->sync_mode
!= WB_SYNC_ALL
)
3184 return ext3_force_commit(inode
->i_sb
);
3190 * Called from notify_change.
3192 * We want to trap VFS attempts to truncate the file as soon as
3193 * possible. In particular, we want to make sure that when the VFS
3194 * shrinks i_size, we put the inode on the orphan list and modify
3195 * i_disksize immediately, so that during the subsequent flushing of
3196 * dirty pages and freeing of disk blocks, we can guarantee that any
3197 * commit will leave the blocks being flushed in an unused state on
3198 * disk. (On recovery, the inode will get truncated and the blocks will
3199 * be freed, so we have a strong guarantee that no future commit will
3200 * leave these blocks visible to the user.)
3202 * Called with inode->sem down.
3204 int ext3_setattr(struct dentry
*dentry
, struct iattr
*attr
)
3206 struct inode
*inode
= dentry
->d_inode
;
3208 const unsigned int ia_valid
= attr
->ia_valid
;
3210 error
= inode_change_ok(inode
, attr
);
3214 if (is_quota_modification(inode
, attr
))
3215 dquot_initialize(inode
);
3216 if ((ia_valid
& ATTR_UID
&& attr
->ia_uid
!= inode
->i_uid
) ||
3217 (ia_valid
& ATTR_GID
&& attr
->ia_gid
!= inode
->i_gid
)) {
3220 /* (user+group)*(old+new) structure, inode write (sb,
3221 * inode block, ? - but truncate inode update has it) */
3222 handle
= ext3_journal_start(inode
, EXT3_MAXQUOTAS_INIT_BLOCKS(inode
->i_sb
)+
3223 EXT3_MAXQUOTAS_DEL_BLOCKS(inode
->i_sb
)+3);
3224 if (IS_ERR(handle
)) {
3225 error
= PTR_ERR(handle
);
3228 error
= dquot_transfer(inode
, attr
);
3230 ext3_journal_stop(handle
);
3233 /* Update corresponding info in inode so that everything is in
3234 * one transaction */
3235 if (attr
->ia_valid
& ATTR_UID
)
3236 inode
->i_uid
= attr
->ia_uid
;
3237 if (attr
->ia_valid
& ATTR_GID
)
3238 inode
->i_gid
= attr
->ia_gid
;
3239 error
= ext3_mark_inode_dirty(handle
, inode
);
3240 ext3_journal_stop(handle
);
3243 if (S_ISREG(inode
->i_mode
) &&
3244 attr
->ia_valid
& ATTR_SIZE
&& attr
->ia_size
< inode
->i_size
) {
3247 handle
= ext3_journal_start(inode
, 3);
3248 if (IS_ERR(handle
)) {
3249 error
= PTR_ERR(handle
);
3253 error
= ext3_orphan_add(handle
, inode
);
3254 EXT3_I(inode
)->i_disksize
= attr
->ia_size
;
3255 rc
= ext3_mark_inode_dirty(handle
, inode
);
3258 ext3_journal_stop(handle
);
3261 if ((attr
->ia_valid
& ATTR_SIZE
) &&
3262 attr
->ia_size
!= i_size_read(inode
)) {
3263 truncate_setsize(inode
, attr
->ia_size
);
3264 ext3_truncate(inode
);
3267 setattr_copy(inode
, attr
);
3268 mark_inode_dirty(inode
);
3270 if (ia_valid
& ATTR_MODE
)
3271 rc
= ext3_acl_chmod(inode
);
3274 ext3_std_error(inode
->i_sb
, error
);
3282 * How many blocks doth make a writepage()?
3284 * With N blocks per page, it may be:
3289 * N+5 bitmap blocks (from the above)
3290 * N+5 group descriptor summary blocks
3293 * 2 * EXT3_SINGLEDATA_TRANS_BLOCKS for the quote files
3295 * 3 * (N + 5) + 2 + 2 * EXT3_SINGLEDATA_TRANS_BLOCKS
3297 * With ordered or writeback data it's the same, less the N data blocks.
3299 * If the inode's direct blocks can hold an integral number of pages then a
3300 * page cannot straddle two indirect blocks, and we can only touch one indirect
3301 * and dindirect block, and the "5" above becomes "3".
3303 * This still overestimates under most circumstances. If we were to pass the
3304 * start and end offsets in here as well we could do block_to_path() on each
3305 * block and work out the exact number of indirects which are touched. Pah.
3308 static int ext3_writepage_trans_blocks(struct inode
*inode
)
3310 int bpp
= ext3_journal_blocks_per_page(inode
);
3311 int indirects
= (EXT3_NDIR_BLOCKS
% bpp
) ? 5 : 3;
3314 if (ext3_should_journal_data(inode
))
3315 ret
= 3 * (bpp
+ indirects
) + 2;
3317 ret
= 2 * (bpp
+ indirects
) + indirects
+ 2;
3320 /* We know that structure was already allocated during dquot_initialize so
3321 * we will be updating only the data blocks + inodes */
3322 ret
+= EXT3_MAXQUOTAS_TRANS_BLOCKS(inode
->i_sb
);
3329 * The caller must have previously called ext3_reserve_inode_write().
3330 * Give this, we know that the caller already has write access to iloc->bh.
3332 int ext3_mark_iloc_dirty(handle_t
*handle
,
3333 struct inode
*inode
, struct ext3_iloc
*iloc
)
3337 /* the do_update_inode consumes one bh->b_count */
3340 /* ext3_do_update_inode() does journal_dirty_metadata */
3341 err
= ext3_do_update_inode(handle
, inode
, iloc
);
3347 * On success, We end up with an outstanding reference count against
3348 * iloc->bh. This _must_ be cleaned up later.
3352 ext3_reserve_inode_write(handle_t
*handle
, struct inode
*inode
,
3353 struct ext3_iloc
*iloc
)
3357 err
= ext3_get_inode_loc(inode
, iloc
);
3359 BUFFER_TRACE(iloc
->bh
, "get_write_access");
3360 err
= ext3_journal_get_write_access(handle
, iloc
->bh
);
3367 ext3_std_error(inode
->i_sb
, err
);
3372 * What we do here is to mark the in-core inode as clean with respect to inode
3373 * dirtiness (it may still be data-dirty).
3374 * This means that the in-core inode may be reaped by prune_icache
3375 * without having to perform any I/O. This is a very good thing,
3376 * because *any* task may call prune_icache - even ones which
3377 * have a transaction open against a different journal.
3379 * Is this cheating? Not really. Sure, we haven't written the
3380 * inode out, but prune_icache isn't a user-visible syncing function.
3381 * Whenever the user wants stuff synced (sys_sync, sys_msync, sys_fsync)
3382 * we start and wait on commits.
3384 * Is this efficient/effective? Well, we're being nice to the system
3385 * by cleaning up our inodes proactively so they can be reaped
3386 * without I/O. But we are potentially leaving up to five seconds'
3387 * worth of inodes floating about which prune_icache wants us to
3388 * write out. One way to fix that would be to get prune_icache()
3389 * to do a write_super() to free up some memory. It has the desired
3392 int ext3_mark_inode_dirty(handle_t
*handle
, struct inode
*inode
)
3394 struct ext3_iloc iloc
;
3398 trace_ext3_mark_inode_dirty(inode
, _RET_IP_
);
3399 err
= ext3_reserve_inode_write(handle
, inode
, &iloc
);
3401 err
= ext3_mark_iloc_dirty(handle
, inode
, &iloc
);
3406 * ext3_dirty_inode() is called from __mark_inode_dirty()
3408 * We're really interested in the case where a file is being extended.
3409 * i_size has been changed by generic_commit_write() and we thus need
3410 * to include the updated inode in the current transaction.
3412 * Also, dquot_alloc_space() will always dirty the inode when blocks
3413 * are allocated to the file.
3415 * If the inode is marked synchronous, we don't honour that here - doing
3416 * so would cause a commit on atime updates, which we don't bother doing.
3417 * We handle synchronous inodes at the highest possible level.
3419 void ext3_dirty_inode(struct inode
*inode
, int flags
)
3421 handle_t
*current_handle
= ext3_journal_current_handle();
3424 handle
= ext3_journal_start(inode
, 2);
3427 if (current_handle
&&
3428 current_handle
->h_transaction
!= handle
->h_transaction
) {
3429 /* This task has a transaction open against a different fs */
3430 printk(KERN_EMERG
"%s: transactions do not match!\n",
3433 jbd_debug(5, "marking dirty. outer handle=%p\n",
3435 ext3_mark_inode_dirty(handle
, inode
);
3437 ext3_journal_stop(handle
);
3444 * Bind an inode's backing buffer_head into this transaction, to prevent
3445 * it from being flushed to disk early. Unlike
3446 * ext3_reserve_inode_write, this leaves behind no bh reference and
3447 * returns no iloc structure, so the caller needs to repeat the iloc
3448 * lookup to mark the inode dirty later.
3450 static int ext3_pin_inode(handle_t
*handle
, struct inode
*inode
)
3452 struct ext3_iloc iloc
;
3456 err
= ext3_get_inode_loc(inode
, &iloc
);
3458 BUFFER_TRACE(iloc
.bh
, "get_write_access");
3459 err
= journal_get_write_access(handle
, iloc
.bh
);
3461 err
= ext3_journal_dirty_metadata(handle
,
3466 ext3_std_error(inode
->i_sb
, err
);
3471 int ext3_change_inode_journal_flag(struct inode
*inode
, int val
)
3478 * We have to be very careful here: changing a data block's
3479 * journaling status dynamically is dangerous. If we write a
3480 * data block to the journal, change the status and then delete
3481 * that block, we risk forgetting to revoke the old log record
3482 * from the journal and so a subsequent replay can corrupt data.
3483 * So, first we make sure that the journal is empty and that
3484 * nobody is changing anything.
3487 journal
= EXT3_JOURNAL(inode
);
3488 if (is_journal_aborted(journal
))
3491 journal_lock_updates(journal
);
3492 journal_flush(journal
);
3495 * OK, there are no updates running now, and all cached data is
3496 * synced to disk. We are now in a completely consistent state
3497 * which doesn't have anything in the journal, and we know that
3498 * no filesystem updates are running, so it is safe to modify
3499 * the inode's in-core data-journaling state flag now.
3503 EXT3_I(inode
)->i_flags
|= EXT3_JOURNAL_DATA_FL
;
3505 EXT3_I(inode
)->i_flags
&= ~EXT3_JOURNAL_DATA_FL
;
3506 ext3_set_aops(inode
);
3508 journal_unlock_updates(journal
);
3510 /* Finally we can mark the inode as dirty. */
3512 handle
= ext3_journal_start(inode
, 1);
3514 return PTR_ERR(handle
);
3516 err
= ext3_mark_inode_dirty(handle
, inode
);
3518 ext3_journal_stop(handle
);
3519 ext3_std_error(inode
->i_sb
, err
);