2 * linux/fs/ext4/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 ext4_get_block() by Al Viro, 2000
25 #include <linux/module.h>
27 #include <linux/time.h>
28 #include <linux/ext4_jbd2.h>
29 #include <linux/jbd2.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>
43 * Test whether an inode is a fast symlink.
45 static int ext4_inode_is_fast_symlink(struct inode
*inode
)
47 int ea_blocks
= EXT4_I(inode
)->i_file_acl
?
48 (inode
->i_sb
->s_blocksize
>> 9) : 0;
50 return (S_ISLNK(inode
->i_mode
) && inode
->i_blocks
- ea_blocks
== 0);
54 * The ext4 forget function must perform a revoke if we are freeing data
55 * which has been journaled. Metadata (eg. indirect blocks) must be
56 * revoked in all cases.
58 * "bh" may be NULL: a metadata block may have been freed from memory
59 * but there may still be a record of it in the journal, and that record
60 * still needs to be revoked.
62 int ext4_forget(handle_t
*handle
, int is_metadata
, struct inode
*inode
,
63 struct buffer_head
*bh
, ext4_fsblk_t blocknr
)
69 BUFFER_TRACE(bh
, "enter");
71 jbd_debug(4, "forgetting bh %p: is_metadata = %d, mode %o, "
73 bh
, is_metadata
, inode
->i_mode
,
74 test_opt(inode
->i_sb
, DATA_FLAGS
));
76 /* Never use the revoke function if we are doing full data
77 * journaling: there is no need to, and a V1 superblock won't
78 * support it. Otherwise, only skip the revoke on un-journaled
81 if (test_opt(inode
->i_sb
, DATA_FLAGS
) == EXT4_MOUNT_JOURNAL_DATA
||
82 (!is_metadata
&& !ext4_should_journal_data(inode
))) {
84 BUFFER_TRACE(bh
, "call jbd2_journal_forget");
85 return ext4_journal_forget(handle
, bh
);
91 * data!=journal && (is_metadata || should_journal_data(inode))
93 BUFFER_TRACE(bh
, "call ext4_journal_revoke");
94 err
= ext4_journal_revoke(handle
, blocknr
, bh
);
96 ext4_abort(inode
->i_sb
, __FUNCTION__
,
97 "error %d when attempting revoke", err
);
98 BUFFER_TRACE(bh
, "exit");
103 * Work out how many blocks we need to proceed with the next chunk of a
104 * truncate transaction.
106 static unsigned long blocks_for_truncate(struct inode
*inode
)
110 needed
= inode
->i_blocks
>> (inode
->i_sb
->s_blocksize_bits
- 9);
112 /* Give ourselves just enough room to cope with inodes in which
113 * i_blocks is corrupt: we've seen disk corruptions in the past
114 * which resulted in random data in an inode which looked enough
115 * like a regular file for ext4 to try to delete it. Things
116 * will go a bit crazy if that happens, but at least we should
117 * try not to panic the whole kernel. */
121 /* But we need to bound the transaction so we don't overflow the
123 if (needed
> EXT4_MAX_TRANS_DATA
)
124 needed
= EXT4_MAX_TRANS_DATA
;
126 return EXT4_DATA_TRANS_BLOCKS(inode
->i_sb
) + needed
;
130 * Truncate transactions can be complex and absolutely huge. So we need to
131 * be able to restart the transaction at a conventient checkpoint to make
132 * sure we don't overflow the journal.
134 * start_transaction gets us a new handle for a truncate transaction,
135 * and extend_transaction tries to extend the existing one a bit. If
136 * extend fails, we need to propagate the failure up and restart the
137 * transaction in the top-level truncate loop. --sct
139 static handle_t
*start_transaction(struct inode
*inode
)
143 result
= ext4_journal_start(inode
, blocks_for_truncate(inode
));
147 ext4_std_error(inode
->i_sb
, PTR_ERR(result
));
152 * Try to extend this transaction for the purposes of truncation.
154 * Returns 0 if we managed to create more room. If we can't create more
155 * room, and the transaction must be restarted we return 1.
157 static int try_to_extend_transaction(handle_t
*handle
, struct inode
*inode
)
159 if (handle
->h_buffer_credits
> EXT4_RESERVE_TRANS_BLOCKS
)
161 if (!ext4_journal_extend(handle
, blocks_for_truncate(inode
)))
167 * Restart the transaction associated with *handle. This does a commit,
168 * so before we call here everything must be consistently dirtied against
171 static int ext4_journal_test_restart(handle_t
*handle
, struct inode
*inode
)
173 jbd_debug(2, "restarting handle %p\n", handle
);
174 return ext4_journal_restart(handle
, blocks_for_truncate(inode
));
178 * Called at the last iput() if i_nlink is zero.
180 void ext4_delete_inode (struct inode
* inode
)
184 truncate_inode_pages(&inode
->i_data
, 0);
186 if (is_bad_inode(inode
))
189 handle
= start_transaction(inode
);
190 if (IS_ERR(handle
)) {
192 * If we're going to skip the normal cleanup, we still need to
193 * make sure that the in-core orphan linked list is properly
196 ext4_orphan_del(NULL
, inode
);
204 ext4_truncate(inode
);
206 * Kill off the orphan record which ext4_truncate created.
207 * AKPM: I think this can be inside the above `if'.
208 * Note that ext4_orphan_del() has to be able to cope with the
209 * deletion of a non-existent orphan - this is because we don't
210 * know if ext4_truncate() actually created an orphan record.
211 * (Well, we could do this if we need to, but heck - it works)
213 ext4_orphan_del(handle
, inode
);
214 EXT4_I(inode
)->i_dtime
= get_seconds();
217 * One subtle ordering requirement: if anything has gone wrong
218 * (transaction abort, IO errors, whatever), then we can still
219 * do these next steps (the fs will already have been marked as
220 * having errors), but we can't free the inode if the mark_dirty
223 if (ext4_mark_inode_dirty(handle
, inode
))
224 /* If that failed, just do the required in-core inode clear. */
227 ext4_free_inode(handle
, inode
);
228 ext4_journal_stop(handle
);
231 clear_inode(inode
); /* We must guarantee clearing of inode... */
237 struct buffer_head
*bh
;
240 static inline void add_chain(Indirect
*p
, struct buffer_head
*bh
, __le32
*v
)
242 p
->key
= *(p
->p
= v
);
247 * ext4_block_to_path - parse the block number into array of offsets
248 * @inode: inode in question (we are only interested in its superblock)
249 * @i_block: block number to be parsed
250 * @offsets: array to store the offsets in
251 * @boundary: set this non-zero if the referred-to block is likely to be
252 * followed (on disk) by an indirect block.
254 * To store the locations of file's data ext4 uses a data structure common
255 * for UNIX filesystems - tree of pointers anchored in the inode, with
256 * data blocks at leaves and indirect blocks in intermediate nodes.
257 * This function translates the block number into path in that tree -
258 * return value is the path length and @offsets[n] is the offset of
259 * pointer to (n+1)th node in the nth one. If @block is out of range
260 * (negative or too large) warning is printed and zero returned.
262 * Note: function doesn't find node addresses, so no IO is needed. All
263 * we need to know is the capacity of indirect blocks (taken from the
268 * Portability note: the last comparison (check that we fit into triple
269 * indirect block) is spelled differently, because otherwise on an
270 * architecture with 32-bit longs and 8Kb pages we might get into trouble
271 * if our filesystem had 8Kb blocks. We might use long long, but that would
272 * kill us on x86. Oh, well, at least the sign propagation does not matter -
273 * i_block would have to be negative in the very beginning, so we would not
277 static int ext4_block_to_path(struct inode
*inode
,
279 ext4_lblk_t offsets
[4], int *boundary
)
281 int ptrs
= EXT4_ADDR_PER_BLOCK(inode
->i_sb
);
282 int ptrs_bits
= EXT4_ADDR_PER_BLOCK_BITS(inode
->i_sb
);
283 const long direct_blocks
= EXT4_NDIR_BLOCKS
,
284 indirect_blocks
= ptrs
,
285 double_blocks
= (1 << (ptrs_bits
* 2));
290 ext4_warning (inode
->i_sb
, "ext4_block_to_path", "block < 0");
291 } else if (i_block
< direct_blocks
) {
292 offsets
[n
++] = i_block
;
293 final
= direct_blocks
;
294 } else if ( (i_block
-= direct_blocks
) < indirect_blocks
) {
295 offsets
[n
++] = EXT4_IND_BLOCK
;
296 offsets
[n
++] = i_block
;
298 } else if ((i_block
-= indirect_blocks
) < double_blocks
) {
299 offsets
[n
++] = EXT4_DIND_BLOCK
;
300 offsets
[n
++] = i_block
>> ptrs_bits
;
301 offsets
[n
++] = i_block
& (ptrs
- 1);
303 } else if (((i_block
-= double_blocks
) >> (ptrs_bits
* 2)) < ptrs
) {
304 offsets
[n
++] = EXT4_TIND_BLOCK
;
305 offsets
[n
++] = i_block
>> (ptrs_bits
* 2);
306 offsets
[n
++] = (i_block
>> ptrs_bits
) & (ptrs
- 1);
307 offsets
[n
++] = i_block
& (ptrs
- 1);
310 ext4_warning(inode
->i_sb
, "ext4_block_to_path",
312 i_block
+ direct_blocks
+
313 indirect_blocks
+ double_blocks
);
316 *boundary
= final
- 1 - (i_block
& (ptrs
- 1));
321 * ext4_get_branch - read the chain of indirect blocks leading to data
322 * @inode: inode in question
323 * @depth: depth of the chain (1 - direct pointer, etc.)
324 * @offsets: offsets of pointers in inode/indirect blocks
325 * @chain: place to store the result
326 * @err: here we store the error value
328 * Function fills the array of triples <key, p, bh> and returns %NULL
329 * if everything went OK or the pointer to the last filled triple
330 * (incomplete one) otherwise. Upon the return chain[i].key contains
331 * the number of (i+1)-th block in the chain (as it is stored in memory,
332 * i.e. little-endian 32-bit), chain[i].p contains the address of that
333 * number (it points into struct inode for i==0 and into the bh->b_data
334 * for i>0) and chain[i].bh points to the buffer_head of i-th indirect
335 * block for i>0 and NULL for i==0. In other words, it holds the block
336 * numbers of the chain, addresses they were taken from (and where we can
337 * verify that chain did not change) and buffer_heads hosting these
340 * Function stops when it stumbles upon zero pointer (absent block)
341 * (pointer to last triple returned, *@err == 0)
342 * or when it gets an IO error reading an indirect block
343 * (ditto, *@err == -EIO)
344 * or when it reads all @depth-1 indirect blocks successfully and finds
345 * the whole chain, all way to the data (returns %NULL, *err == 0).
347 * Need to be called with
348 * down_read(&EXT4_I(inode)->i_data_sem)
350 static Indirect
*ext4_get_branch(struct inode
*inode
, int depth
,
351 ext4_lblk_t
*offsets
,
352 Indirect chain
[4], int *err
)
354 struct super_block
*sb
= inode
->i_sb
;
356 struct buffer_head
*bh
;
359 /* i_data is not going away, no lock needed */
360 add_chain (chain
, NULL
, EXT4_I(inode
)->i_data
+ *offsets
);
364 bh
= sb_bread(sb
, le32_to_cpu(p
->key
));
367 add_chain(++p
, bh
, (__le32
*)bh
->b_data
+ *++offsets
);
381 * ext4_find_near - find a place for allocation with sufficient locality
383 * @ind: descriptor of indirect block.
385 * This function returns the prefered place for block allocation.
386 * It is used when heuristic for sequential allocation fails.
388 * + if there is a block to the left of our position - allocate near it.
389 * + if pointer will live in indirect block - allocate near that block.
390 * + if pointer will live in inode - allocate in the same
393 * In the latter case we colour the starting block by the callers PID to
394 * prevent it from clashing with concurrent allocations for a different inode
395 * in the same block group. The PID is used here so that functionally related
396 * files will be close-by on-disk.
398 * Caller must make sure that @ind is valid and will stay that way.
400 static ext4_fsblk_t
ext4_find_near(struct inode
*inode
, Indirect
*ind
)
402 struct ext4_inode_info
*ei
= EXT4_I(inode
);
403 __le32
*start
= ind
->bh
? (__le32
*) ind
->bh
->b_data
: ei
->i_data
;
405 ext4_fsblk_t bg_start
;
406 ext4_fsblk_t last_block
;
407 ext4_grpblk_t colour
;
409 /* Try to find previous block */
410 for (p
= ind
->p
- 1; p
>= start
; p
--) {
412 return le32_to_cpu(*p
);
415 /* No such thing, so let's try location of indirect block */
417 return ind
->bh
->b_blocknr
;
420 * It is going to be referred to from the inode itself? OK, just put it
421 * into the same cylinder group then.
423 bg_start
= ext4_group_first_block_no(inode
->i_sb
, ei
->i_block_group
);
424 last_block
= ext4_blocks_count(EXT4_SB(inode
->i_sb
)->s_es
) - 1;
426 if (bg_start
+ EXT4_BLOCKS_PER_GROUP(inode
->i_sb
) <= last_block
)
427 colour
= (current
->pid
% 16) *
428 (EXT4_BLOCKS_PER_GROUP(inode
->i_sb
) / 16);
430 colour
= (current
->pid
% 16) * ((last_block
- bg_start
) / 16);
431 return bg_start
+ colour
;
435 * ext4_find_goal - find a prefered place for allocation.
437 * @block: block we want
438 * @partial: pointer to the last triple within a chain
440 * Normally this function find the prefered place for block allocation,
443 static ext4_fsblk_t
ext4_find_goal(struct inode
*inode
, ext4_lblk_t block
,
446 struct ext4_block_alloc_info
*block_i
;
448 block_i
= EXT4_I(inode
)->i_block_alloc_info
;
451 * try the heuristic for sequential allocation,
452 * failing that at least try to get decent locality.
454 if (block_i
&& (block
== block_i
->last_alloc_logical_block
+ 1)
455 && (block_i
->last_alloc_physical_block
!= 0)) {
456 return block_i
->last_alloc_physical_block
+ 1;
459 return ext4_find_near(inode
, partial
);
463 * ext4_blks_to_allocate: Look up the block map and count the number
464 * of direct blocks need to be allocated for the given branch.
466 * @branch: chain of indirect blocks
467 * @k: number of blocks need for indirect blocks
468 * @blks: number of data blocks to be mapped.
469 * @blocks_to_boundary: the offset in the indirect block
471 * return the total number of blocks to be allocate, including the
472 * direct and indirect blocks.
474 static int ext4_blks_to_allocate(Indirect
*branch
, int k
, unsigned long blks
,
475 int blocks_to_boundary
)
477 unsigned long count
= 0;
480 * Simple case, [t,d]Indirect block(s) has not allocated yet
481 * then it's clear blocks on that path have not allocated
484 /* right now we don't handle cross boundary allocation */
485 if (blks
< blocks_to_boundary
+ 1)
488 count
+= blocks_to_boundary
+ 1;
493 while (count
< blks
&& count
<= blocks_to_boundary
&&
494 le32_to_cpu(*(branch
[0].p
+ count
)) == 0) {
501 * ext4_alloc_blocks: multiple allocate blocks needed for a branch
502 * @indirect_blks: the number of blocks need to allocate for indirect
505 * @new_blocks: on return it will store the new block numbers for
506 * the indirect blocks(if needed) and the first direct block,
507 * @blks: on return it will store the total number of allocated
510 static int ext4_alloc_blocks(handle_t
*handle
, struct inode
*inode
,
511 ext4_fsblk_t goal
, int indirect_blks
, int blks
,
512 ext4_fsblk_t new_blocks
[4], int *err
)
515 unsigned long count
= 0;
517 ext4_fsblk_t current_block
= 0;
521 * Here we try to allocate the requested multiple blocks at once,
522 * on a best-effort basis.
523 * To build a branch, we should allocate blocks for
524 * the indirect blocks(if not allocated yet), and at least
525 * the first direct block of this branch. That's the
526 * minimum number of blocks need to allocate(required)
528 target
= blks
+ indirect_blks
;
532 /* allocating blocks for indirect blocks and direct blocks */
533 current_block
= ext4_new_blocks(handle
,inode
,goal
,&count
,err
);
538 /* allocate blocks for indirect blocks */
539 while (index
< indirect_blks
&& count
) {
540 new_blocks
[index
++] = current_block
++;
548 /* save the new block number for the first direct block */
549 new_blocks
[index
] = current_block
;
551 /* total number of blocks allocated for direct blocks */
556 for (i
= 0; i
<index
; i
++)
557 ext4_free_blocks(handle
, inode
, new_blocks
[i
], 1, 0);
562 * ext4_alloc_branch - allocate and set up a chain of blocks.
564 * @indirect_blks: number of allocated indirect blocks
565 * @blks: number of allocated direct blocks
566 * @offsets: offsets (in the blocks) to store the pointers to next.
567 * @branch: place to store the chain in.
569 * This function allocates blocks, zeroes out all but the last one,
570 * links them into chain and (if we are synchronous) writes them to disk.
571 * In other words, it prepares a branch that can be spliced onto the
572 * inode. It stores the information about that chain in the branch[], in
573 * the same format as ext4_get_branch() would do. We are calling it after
574 * we had read the existing part of chain and partial points to the last
575 * triple of that (one with zero ->key). Upon the exit we have the same
576 * picture as after the successful ext4_get_block(), except that in one
577 * place chain is disconnected - *branch->p is still zero (we did not
578 * set the last link), but branch->key contains the number that should
579 * be placed into *branch->p to fill that gap.
581 * If allocation fails we free all blocks we've allocated (and forget
582 * their buffer_heads) and return the error value the from failed
583 * ext4_alloc_block() (normally -ENOSPC). Otherwise we set the chain
584 * as described above and return 0.
586 static int ext4_alloc_branch(handle_t
*handle
, struct inode
*inode
,
587 int indirect_blks
, int *blks
, ext4_fsblk_t goal
,
588 ext4_lblk_t
*offsets
, Indirect
*branch
)
590 int blocksize
= inode
->i_sb
->s_blocksize
;
593 struct buffer_head
*bh
;
595 ext4_fsblk_t new_blocks
[4];
596 ext4_fsblk_t current_block
;
598 num
= ext4_alloc_blocks(handle
, inode
, goal
, indirect_blks
,
599 *blks
, new_blocks
, &err
);
603 branch
[0].key
= cpu_to_le32(new_blocks
[0]);
605 * metadata blocks and data blocks are allocated.
607 for (n
= 1; n
<= indirect_blks
; n
++) {
609 * Get buffer_head for parent block, zero it out
610 * and set the pointer to new one, then send
613 bh
= sb_getblk(inode
->i_sb
, new_blocks
[n
-1]);
616 BUFFER_TRACE(bh
, "call get_create_access");
617 err
= ext4_journal_get_create_access(handle
, bh
);
624 memset(bh
->b_data
, 0, blocksize
);
625 branch
[n
].p
= (__le32
*) bh
->b_data
+ offsets
[n
];
626 branch
[n
].key
= cpu_to_le32(new_blocks
[n
]);
627 *branch
[n
].p
= branch
[n
].key
;
628 if ( n
== indirect_blks
) {
629 current_block
= new_blocks
[n
];
631 * End of chain, update the last new metablock of
632 * the chain to point to the new allocated
633 * data blocks numbers
635 for (i
=1; i
< num
; i
++)
636 *(branch
[n
].p
+ i
) = cpu_to_le32(++current_block
);
638 BUFFER_TRACE(bh
, "marking uptodate");
639 set_buffer_uptodate(bh
);
642 BUFFER_TRACE(bh
, "call ext4_journal_dirty_metadata");
643 err
= ext4_journal_dirty_metadata(handle
, bh
);
650 /* Allocation failed, free what we already allocated */
651 for (i
= 1; i
<= n
; i
++) {
652 BUFFER_TRACE(branch
[i
].bh
, "call jbd2_journal_forget");
653 ext4_journal_forget(handle
, branch
[i
].bh
);
655 for (i
= 0; i
<indirect_blks
; i
++)
656 ext4_free_blocks(handle
, inode
, new_blocks
[i
], 1, 0);
658 ext4_free_blocks(handle
, inode
, new_blocks
[i
], num
, 0);
664 * ext4_splice_branch - splice the allocated branch onto inode.
666 * @block: (logical) number of block we are adding
667 * @chain: chain of indirect blocks (with a missing link - see
669 * @where: location of missing link
670 * @num: number of indirect blocks we are adding
671 * @blks: number of direct blocks we are adding
673 * This function fills the missing link and does all housekeeping needed in
674 * inode (->i_blocks, etc.). In case of success we end up with the full
675 * chain to new block and return 0.
677 static int ext4_splice_branch(handle_t
*handle
, struct inode
*inode
,
678 ext4_lblk_t block
, Indirect
*where
, int num
, int blks
)
682 struct ext4_block_alloc_info
*block_i
;
683 ext4_fsblk_t current_block
;
685 block_i
= EXT4_I(inode
)->i_block_alloc_info
;
687 * If we're splicing into a [td]indirect block (as opposed to the
688 * inode) then we need to get write access to the [td]indirect block
692 BUFFER_TRACE(where
->bh
, "get_write_access");
693 err
= ext4_journal_get_write_access(handle
, where
->bh
);
699 *where
->p
= where
->key
;
702 * Update the host buffer_head or inode to point to more just allocated
703 * direct blocks blocks
705 if (num
== 0 && blks
> 1) {
706 current_block
= le32_to_cpu(where
->key
) + 1;
707 for (i
= 1; i
< blks
; i
++)
708 *(where
->p
+ i
) = cpu_to_le32(current_block
++);
712 * update the most recently allocated logical & physical block
713 * in i_block_alloc_info, to assist find the proper goal block for next
717 block_i
->last_alloc_logical_block
= block
+ blks
- 1;
718 block_i
->last_alloc_physical_block
=
719 le32_to_cpu(where
[num
].key
) + blks
- 1;
722 /* We are done with atomic stuff, now do the rest of housekeeping */
724 inode
->i_ctime
= ext4_current_time(inode
);
725 ext4_mark_inode_dirty(handle
, inode
);
727 /* had we spliced it onto indirect block? */
730 * If we spliced it onto an indirect block, we haven't
731 * altered the inode. Note however that if it is being spliced
732 * onto an indirect block at the very end of the file (the
733 * file is growing) then we *will* alter the inode to reflect
734 * the new i_size. But that is not done here - it is done in
735 * generic_commit_write->__mark_inode_dirty->ext4_dirty_inode.
737 jbd_debug(5, "splicing indirect only\n");
738 BUFFER_TRACE(where
->bh
, "call ext4_journal_dirty_metadata");
739 err
= ext4_journal_dirty_metadata(handle
, where
->bh
);
744 * OK, we spliced it into the inode itself on a direct block.
745 * Inode was dirtied above.
747 jbd_debug(5, "splicing direct\n");
752 for (i
= 1; i
<= num
; i
++) {
753 BUFFER_TRACE(where
[i
].bh
, "call jbd2_journal_forget");
754 ext4_journal_forget(handle
, where
[i
].bh
);
755 ext4_free_blocks(handle
, inode
,
756 le32_to_cpu(where
[i
-1].key
), 1, 0);
758 ext4_free_blocks(handle
, inode
, le32_to_cpu(where
[num
].key
), blks
, 0);
764 * Allocation strategy is simple: if we have to allocate something, we will
765 * have to go the whole way to leaf. So let's do it before attaching anything
766 * to tree, set linkage between the newborn blocks, write them if sync is
767 * required, recheck the path, free and repeat if check fails, otherwise
768 * set the last missing link (that will protect us from any truncate-generated
769 * removals - all blocks on the path are immune now) and possibly force the
770 * write on the parent block.
771 * That has a nice additional property: no special recovery from the failed
772 * allocations is needed - we simply release blocks and do not touch anything
773 * reachable from inode.
775 * `handle' can be NULL if create == 0.
777 * return > 0, # of blocks mapped or allocated.
778 * return = 0, if plain lookup failed.
779 * return < 0, error case.
782 * Need to be called with
783 * down_read(&EXT4_I(inode)->i_data_sem) if not allocating file system block
784 * (ie, create is zero). Otherwise down_write(&EXT4_I(inode)->i_data_sem)
786 int ext4_get_blocks_handle(handle_t
*handle
, struct inode
*inode
,
787 ext4_lblk_t iblock
, unsigned long maxblocks
,
788 struct buffer_head
*bh_result
,
789 int create
, int extend_disksize
)
792 ext4_lblk_t offsets
[4];
797 int blocks_to_boundary
= 0;
799 struct ext4_inode_info
*ei
= EXT4_I(inode
);
801 ext4_fsblk_t first_block
= 0;
804 J_ASSERT(!(EXT4_I(inode
)->i_flags
& EXT4_EXTENTS_FL
));
805 J_ASSERT(handle
!= NULL
|| create
== 0);
806 depth
= ext4_block_to_path(inode
, iblock
, offsets
,
807 &blocks_to_boundary
);
812 partial
= ext4_get_branch(inode
, depth
, offsets
, chain
, &err
);
814 /* Simplest case - block found, no allocation needed */
816 first_block
= le32_to_cpu(chain
[depth
- 1].key
);
817 clear_buffer_new(bh_result
);
820 while (count
< maxblocks
&& count
<= blocks_to_boundary
) {
823 blk
= le32_to_cpu(*(chain
[depth
-1].p
+ count
));
825 if (blk
== first_block
+ count
)
833 /* Next simple case - plain lookup or failed read of indirect block */
834 if (!create
|| err
== -EIO
)
838 * Okay, we need to do block allocation. Lazily initialize the block
839 * allocation info here if necessary
841 if (S_ISREG(inode
->i_mode
) && (!ei
->i_block_alloc_info
))
842 ext4_init_block_alloc_info(inode
);
844 goal
= ext4_find_goal(inode
, iblock
, partial
);
846 /* the number of blocks need to allocate for [d,t]indirect blocks */
847 indirect_blks
= (chain
+ depth
) - partial
- 1;
850 * Next look up the indirect map to count the totoal number of
851 * direct blocks to allocate for this branch.
853 count
= ext4_blks_to_allocate(partial
, indirect_blks
,
854 maxblocks
, blocks_to_boundary
);
856 * Block out ext4_truncate while we alter the tree
858 err
= ext4_alloc_branch(handle
, inode
, indirect_blks
, &count
, goal
,
859 offsets
+ (partial
- chain
), partial
);
862 * The ext4_splice_branch call will free and forget any buffers
863 * on the new chain if there is a failure, but that risks using
864 * up transaction credits, especially for bitmaps where the
865 * credits cannot be returned. Can we handle this somehow? We
866 * may need to return -EAGAIN upwards in the worst case. --sct
869 err
= ext4_splice_branch(handle
, inode
, iblock
,
870 partial
, indirect_blks
, count
);
872 * i_disksize growing is protected by i_data_sem. Don't forget to
873 * protect it if you're about to implement concurrent
874 * ext4_get_block() -bzzz
876 if (!err
&& extend_disksize
&& inode
->i_size
> ei
->i_disksize
)
877 ei
->i_disksize
= inode
->i_size
;
881 set_buffer_new(bh_result
);
883 map_bh(bh_result
, inode
->i_sb
, le32_to_cpu(chain
[depth
-1].key
));
884 if (count
> blocks_to_boundary
)
885 set_buffer_boundary(bh_result
);
887 /* Clean up and exit */
888 partial
= chain
+ depth
- 1; /* the whole chain */
890 while (partial
> chain
) {
891 BUFFER_TRACE(partial
->bh
, "call brelse");
895 BUFFER_TRACE(bh_result
, "returned");
900 /* Maximum number of blocks we map for direct IO at once. */
901 #define DIO_MAX_BLOCKS 4096
903 * Number of credits we need for writing DIO_MAX_BLOCKS:
904 * We need sb + group descriptor + bitmap + inode -> 4
905 * For B blocks with A block pointers per block we need:
906 * 1 (triple ind.) + (B/A/A + 2) (doubly ind.) + (B/A + 2) (indirect).
907 * If we plug in 4096 for B and 256 for A (for 1KB block size), we get 25.
909 #define DIO_CREDITS 25
911 int ext4_get_blocks_wrap(handle_t
*handle
, struct inode
*inode
, sector_t block
,
912 unsigned long max_blocks
, struct buffer_head
*bh
,
913 int create
, int extend_disksize
)
917 * Try to see if we can get the block without requesting
918 * for new file system block.
920 down_read((&EXT4_I(inode
)->i_data_sem
));
921 if (EXT4_I(inode
)->i_flags
& EXT4_EXTENTS_FL
) {
922 retval
= ext4_ext_get_blocks(handle
, inode
, block
, max_blocks
,
925 retval
= ext4_get_blocks_handle(handle
,
926 inode
, block
, max_blocks
, bh
, 0, 0);
928 up_read((&EXT4_I(inode
)->i_data_sem
));
929 if (!create
|| (retval
> 0))
933 * We need to allocate new blocks which will result
936 down_write((&EXT4_I(inode
)->i_data_sem
));
938 * We need to check for EXT4 here because migrate
939 * could have changed the inode type in between
941 if (EXT4_I(inode
)->i_flags
& EXT4_EXTENTS_FL
) {
942 retval
= ext4_ext_get_blocks(handle
, inode
, block
, max_blocks
,
943 bh
, create
, extend_disksize
);
945 retval
= ext4_get_blocks_handle(handle
, inode
, block
,
946 max_blocks
, bh
, create
, extend_disksize
);
948 up_write((&EXT4_I(inode
)->i_data_sem
));
952 static int ext4_get_block(struct inode
*inode
, sector_t iblock
,
953 struct buffer_head
*bh_result
, int create
)
955 handle_t
*handle
= ext4_journal_current_handle();
956 int ret
= 0, started
= 0;
957 unsigned max_blocks
= bh_result
->b_size
>> inode
->i_blkbits
;
959 if (create
&& !handle
) {
960 /* Direct IO write... */
961 if (max_blocks
> DIO_MAX_BLOCKS
)
962 max_blocks
= DIO_MAX_BLOCKS
;
963 handle
= ext4_journal_start(inode
, DIO_CREDITS
+
964 2 * EXT4_QUOTA_TRANS_BLOCKS(inode
->i_sb
));
965 if (IS_ERR(handle
)) {
966 ret
= PTR_ERR(handle
);
972 ret
= ext4_get_blocks_wrap(handle
, inode
, iblock
,
973 max_blocks
, bh_result
, create
, 0);
975 bh_result
->b_size
= (ret
<< inode
->i_blkbits
);
979 ext4_journal_stop(handle
);
985 * `handle' can be NULL if create is zero
987 struct buffer_head
*ext4_getblk(handle_t
*handle
, struct inode
*inode
,
988 ext4_lblk_t block
, int create
, int *errp
)
990 struct buffer_head dummy
;
993 J_ASSERT(handle
!= NULL
|| create
== 0);
996 dummy
.b_blocknr
= -1000;
997 buffer_trace_init(&dummy
.b_history
);
998 err
= ext4_get_blocks_wrap(handle
, inode
, block
, 1,
1001 * ext4_get_blocks_handle() returns number of blocks
1002 * mapped. 0 in case of a HOLE.
1010 if (!err
&& buffer_mapped(&dummy
)) {
1011 struct buffer_head
*bh
;
1012 bh
= sb_getblk(inode
->i_sb
, dummy
.b_blocknr
);
1017 if (buffer_new(&dummy
)) {
1018 J_ASSERT(create
!= 0);
1019 J_ASSERT(handle
!= NULL
);
1022 * Now that we do not always journal data, we should
1023 * keep in mind whether this should always journal the
1024 * new buffer as metadata. For now, regular file
1025 * writes use ext4_get_block instead, so it's not a
1029 BUFFER_TRACE(bh
, "call get_create_access");
1030 fatal
= ext4_journal_get_create_access(handle
, bh
);
1031 if (!fatal
&& !buffer_uptodate(bh
)) {
1032 memset(bh
->b_data
,0,inode
->i_sb
->s_blocksize
);
1033 set_buffer_uptodate(bh
);
1036 BUFFER_TRACE(bh
, "call ext4_journal_dirty_metadata");
1037 err
= ext4_journal_dirty_metadata(handle
, bh
);
1041 BUFFER_TRACE(bh
, "not a new buffer");
1054 struct buffer_head
*ext4_bread(handle_t
*handle
, struct inode
*inode
,
1055 ext4_lblk_t block
, int create
, int *err
)
1057 struct buffer_head
* bh
;
1059 bh
= ext4_getblk(handle
, inode
, block
, create
, err
);
1062 if (buffer_uptodate(bh
))
1064 ll_rw_block(READ_META
, 1, &bh
);
1066 if (buffer_uptodate(bh
))
1073 static int walk_page_buffers( handle_t
*handle
,
1074 struct buffer_head
*head
,
1078 int (*fn
)( handle_t
*handle
,
1079 struct buffer_head
*bh
))
1081 struct buffer_head
*bh
;
1082 unsigned block_start
, block_end
;
1083 unsigned blocksize
= head
->b_size
;
1085 struct buffer_head
*next
;
1087 for ( bh
= head
, block_start
= 0;
1088 ret
== 0 && (bh
!= head
|| !block_start
);
1089 block_start
= block_end
, bh
= next
)
1091 next
= bh
->b_this_page
;
1092 block_end
= block_start
+ blocksize
;
1093 if (block_end
<= from
|| block_start
>= to
) {
1094 if (partial
&& !buffer_uptodate(bh
))
1098 err
= (*fn
)(handle
, bh
);
1106 * To preserve ordering, it is essential that the hole instantiation and
1107 * the data write be encapsulated in a single transaction. We cannot
1108 * close off a transaction and start a new one between the ext4_get_block()
1109 * and the commit_write(). So doing the jbd2_journal_start at the start of
1110 * prepare_write() is the right place.
1112 * Also, this function can nest inside ext4_writepage() ->
1113 * block_write_full_page(). In that case, we *know* that ext4_writepage()
1114 * has generated enough buffer credits to do the whole page. So we won't
1115 * block on the journal in that case, which is good, because the caller may
1118 * By accident, ext4 can be reentered when a transaction is open via
1119 * quota file writes. If we were to commit the transaction while thus
1120 * reentered, there can be a deadlock - we would be holding a quota
1121 * lock, and the commit would never complete if another thread had a
1122 * transaction open and was blocking on the quota lock - a ranking
1125 * So what we do is to rely on the fact that jbd2_journal_stop/journal_start
1126 * will _not_ run commit under these circumstances because handle->h_ref
1127 * is elevated. We'll still have enough credits for the tiny quotafile
1130 static int do_journal_get_write_access(handle_t
*handle
,
1131 struct buffer_head
*bh
)
1133 if (!buffer_mapped(bh
) || buffer_freed(bh
))
1135 return ext4_journal_get_write_access(handle
, bh
);
1138 static int ext4_write_begin(struct file
*file
, struct address_space
*mapping
,
1139 loff_t pos
, unsigned len
, unsigned flags
,
1140 struct page
**pagep
, void **fsdata
)
1142 struct inode
*inode
= mapping
->host
;
1143 int ret
, needed_blocks
= ext4_writepage_trans_blocks(inode
);
1150 index
= pos
>> PAGE_CACHE_SHIFT
;
1151 from
= pos
& (PAGE_CACHE_SIZE
- 1);
1155 page
= __grab_cache_page(mapping
, index
);
1160 handle
= ext4_journal_start(inode
, needed_blocks
);
1161 if (IS_ERR(handle
)) {
1163 page_cache_release(page
);
1164 ret
= PTR_ERR(handle
);
1168 ret
= block_write_begin(file
, mapping
, pos
, len
, flags
, pagep
, fsdata
,
1171 if (!ret
&& ext4_should_journal_data(inode
)) {
1172 ret
= walk_page_buffers(handle
, page_buffers(page
),
1173 from
, to
, NULL
, do_journal_get_write_access
);
1177 ext4_journal_stop(handle
);
1179 page_cache_release(page
);
1182 if (ret
== -ENOSPC
&& ext4_should_retry_alloc(inode
->i_sb
, &retries
))
1188 int ext4_journal_dirty_data(handle_t
*handle
, struct buffer_head
*bh
)
1190 int err
= jbd2_journal_dirty_data(handle
, bh
);
1192 ext4_journal_abort_handle(__FUNCTION__
, __FUNCTION__
,
1197 /* For write_end() in data=journal mode */
1198 static int write_end_fn(handle_t
*handle
, struct buffer_head
*bh
)
1200 if (!buffer_mapped(bh
) || buffer_freed(bh
))
1202 set_buffer_uptodate(bh
);
1203 return ext4_journal_dirty_metadata(handle
, bh
);
1207 * Generic write_end handler for ordered and writeback ext4 journal modes.
1208 * We can't use generic_write_end, because that unlocks the page and we need to
1209 * unlock the page after ext4_journal_stop, but ext4_journal_stop must run
1210 * after block_write_end.
1212 static int ext4_generic_write_end(struct file
*file
,
1213 struct address_space
*mapping
,
1214 loff_t pos
, unsigned len
, unsigned copied
,
1215 struct page
*page
, void *fsdata
)
1217 struct inode
*inode
= file
->f_mapping
->host
;
1219 copied
= block_write_end(file
, mapping
, pos
, len
, copied
, page
, fsdata
);
1221 if (pos
+copied
> inode
->i_size
) {
1222 i_size_write(inode
, pos
+copied
);
1223 mark_inode_dirty(inode
);
1230 * We need to pick up the new inode size which generic_commit_write gave us
1231 * `file' can be NULL - eg, when called from page_symlink().
1233 * ext4 never places buffers on inode->i_mapping->private_list. metadata
1234 * buffers are managed internally.
1236 static int ext4_ordered_write_end(struct file
*file
,
1237 struct address_space
*mapping
,
1238 loff_t pos
, unsigned len
, unsigned copied
,
1239 struct page
*page
, void *fsdata
)
1241 handle_t
*handle
= ext4_journal_current_handle();
1242 struct inode
*inode
= file
->f_mapping
->host
;
1246 from
= pos
& (PAGE_CACHE_SIZE
- 1);
1249 ret
= walk_page_buffers(handle
, page_buffers(page
),
1250 from
, to
, NULL
, ext4_journal_dirty_data
);
1254 * generic_write_end() will run mark_inode_dirty() if i_size
1255 * changes. So let's piggyback the i_disksize mark_inode_dirty
1260 new_i_size
= pos
+ copied
;
1261 if (new_i_size
> EXT4_I(inode
)->i_disksize
)
1262 EXT4_I(inode
)->i_disksize
= new_i_size
;
1263 copied
= ext4_generic_write_end(file
, mapping
, pos
, len
, copied
,
1268 ret2
= ext4_journal_stop(handle
);
1272 page_cache_release(page
);
1274 return ret
? ret
: copied
;
1277 static int ext4_writeback_write_end(struct file
*file
,
1278 struct address_space
*mapping
,
1279 loff_t pos
, unsigned len
, unsigned copied
,
1280 struct page
*page
, void *fsdata
)
1282 handle_t
*handle
= ext4_journal_current_handle();
1283 struct inode
*inode
= file
->f_mapping
->host
;
1287 new_i_size
= pos
+ copied
;
1288 if (new_i_size
> EXT4_I(inode
)->i_disksize
)
1289 EXT4_I(inode
)->i_disksize
= new_i_size
;
1291 copied
= ext4_generic_write_end(file
, mapping
, pos
, len
, copied
,
1296 ret2
= ext4_journal_stop(handle
);
1300 page_cache_release(page
);
1302 return ret
? ret
: copied
;
1305 static int ext4_journalled_write_end(struct file
*file
,
1306 struct address_space
*mapping
,
1307 loff_t pos
, unsigned len
, unsigned copied
,
1308 struct page
*page
, void *fsdata
)
1310 handle_t
*handle
= ext4_journal_current_handle();
1311 struct inode
*inode
= mapping
->host
;
1316 from
= pos
& (PAGE_CACHE_SIZE
- 1);
1320 if (!PageUptodate(page
))
1322 page_zero_new_buffers(page
, from
+copied
, to
);
1325 ret
= walk_page_buffers(handle
, page_buffers(page
), from
,
1326 to
, &partial
, write_end_fn
);
1328 SetPageUptodate(page
);
1329 if (pos
+copied
> inode
->i_size
)
1330 i_size_write(inode
, pos
+copied
);
1331 EXT4_I(inode
)->i_state
|= EXT4_STATE_JDATA
;
1332 if (inode
->i_size
> EXT4_I(inode
)->i_disksize
) {
1333 EXT4_I(inode
)->i_disksize
= inode
->i_size
;
1334 ret2
= ext4_mark_inode_dirty(handle
, inode
);
1339 ret2
= ext4_journal_stop(handle
);
1343 page_cache_release(page
);
1345 return ret
? ret
: copied
;
1349 * bmap() is special. It gets used by applications such as lilo and by
1350 * the swapper to find the on-disk block of a specific piece of data.
1352 * Naturally, this is dangerous if the block concerned is still in the
1353 * journal. If somebody makes a swapfile on an ext4 data-journaling
1354 * filesystem and enables swap, then they may get a nasty shock when the
1355 * data getting swapped to that swapfile suddenly gets overwritten by
1356 * the original zero's written out previously to the journal and
1357 * awaiting writeback in the kernel's buffer cache.
1359 * So, if we see any bmap calls here on a modified, data-journaled file,
1360 * take extra steps to flush any blocks which might be in the cache.
1362 static sector_t
ext4_bmap(struct address_space
*mapping
, sector_t block
)
1364 struct inode
*inode
= mapping
->host
;
1368 if (EXT4_I(inode
)->i_state
& EXT4_STATE_JDATA
) {
1370 * This is a REALLY heavyweight approach, but the use of
1371 * bmap on dirty files is expected to be extremely rare:
1372 * only if we run lilo or swapon on a freshly made file
1373 * do we expect this to happen.
1375 * (bmap requires CAP_SYS_RAWIO so this does not
1376 * represent an unprivileged user DOS attack --- we'd be
1377 * in trouble if mortal users could trigger this path at
1380 * NB. EXT4_STATE_JDATA is not set on files other than
1381 * regular files. If somebody wants to bmap a directory
1382 * or symlink and gets confused because the buffer
1383 * hasn't yet been flushed to disk, they deserve
1384 * everything they get.
1387 EXT4_I(inode
)->i_state
&= ~EXT4_STATE_JDATA
;
1388 journal
= EXT4_JOURNAL(inode
);
1389 jbd2_journal_lock_updates(journal
);
1390 err
= jbd2_journal_flush(journal
);
1391 jbd2_journal_unlock_updates(journal
);
1397 return generic_block_bmap(mapping
,block
,ext4_get_block
);
1400 static int bget_one(handle_t
*handle
, struct buffer_head
*bh
)
1406 static int bput_one(handle_t
*handle
, struct buffer_head
*bh
)
1412 static int jbd2_journal_dirty_data_fn(handle_t
*handle
, struct buffer_head
*bh
)
1414 if (buffer_mapped(bh
))
1415 return ext4_journal_dirty_data(handle
, bh
);
1420 * Note that we always start a transaction even if we're not journalling
1421 * data. This is to preserve ordering: any hole instantiation within
1422 * __block_write_full_page -> ext4_get_block() should be journalled
1423 * along with the data so we don't crash and then get metadata which
1424 * refers to old data.
1426 * In all journalling modes block_write_full_page() will start the I/O.
1430 * ext4_writepage() -> kmalloc() -> __alloc_pages() -> page_launder() ->
1435 * ext4_file_write() -> generic_file_write() -> __alloc_pages() -> ...
1437 * Same applies to ext4_get_block(). We will deadlock on various things like
1438 * lock_journal and i_data_sem
1440 * Setting PF_MEMALLOC here doesn't work - too many internal memory
1443 * 16May01: If we're reentered then journal_current_handle() will be
1444 * non-zero. We simply *return*.
1446 * 1 July 2001: @@@ FIXME:
1447 * In journalled data mode, a data buffer may be metadata against the
1448 * current transaction. But the same file is part of a shared mapping
1449 * and someone does a writepage() on it.
1451 * We will move the buffer onto the async_data list, but *after* it has
1452 * been dirtied. So there's a small window where we have dirty data on
1455 * Note that this only applies to the last partial page in the file. The
1456 * bit which block_write_full_page() uses prepare/commit for. (That's
1457 * broken code anyway: it's wrong for msync()).
1459 * It's a rare case: affects the final partial page, for journalled data
1460 * where the file is subject to bith write() and writepage() in the same
1461 * transction. To fix it we'll need a custom block_write_full_page().
1462 * We'll probably need that anyway for journalling writepage() output.
1464 * We don't honour synchronous mounts for writepage(). That would be
1465 * disastrous. Any write() or metadata operation will sync the fs for
1468 * AKPM2: if all the page's buffers are mapped to disk and !data=journal,
1469 * we don't need to open a transaction here.
1471 static int ext4_ordered_writepage(struct page
*page
,
1472 struct writeback_control
*wbc
)
1474 struct inode
*inode
= page
->mapping
->host
;
1475 struct buffer_head
*page_bufs
;
1476 handle_t
*handle
= NULL
;
1480 J_ASSERT(PageLocked(page
));
1483 * We give up here if we're reentered, because it might be for a
1484 * different filesystem.
1486 if (ext4_journal_current_handle())
1489 handle
= ext4_journal_start(inode
, ext4_writepage_trans_blocks(inode
));
1491 if (IS_ERR(handle
)) {
1492 ret
= PTR_ERR(handle
);
1496 if (!page_has_buffers(page
)) {
1497 create_empty_buffers(page
, inode
->i_sb
->s_blocksize
,
1498 (1 << BH_Dirty
)|(1 << BH_Uptodate
));
1500 page_bufs
= page_buffers(page
);
1501 walk_page_buffers(handle
, page_bufs
, 0,
1502 PAGE_CACHE_SIZE
, NULL
, bget_one
);
1504 ret
= block_write_full_page(page
, ext4_get_block
, wbc
);
1507 * The page can become unlocked at any point now, and
1508 * truncate can then come in and change things. So we
1509 * can't touch *page from now on. But *page_bufs is
1510 * safe due to elevated refcount.
1514 * And attach them to the current transaction. But only if
1515 * block_write_full_page() succeeded. Otherwise they are unmapped,
1516 * and generally junk.
1519 err
= walk_page_buffers(handle
, page_bufs
, 0, PAGE_CACHE_SIZE
,
1520 NULL
, jbd2_journal_dirty_data_fn
);
1524 walk_page_buffers(handle
, page_bufs
, 0,
1525 PAGE_CACHE_SIZE
, NULL
, bput_one
);
1526 err
= ext4_journal_stop(handle
);
1532 redirty_page_for_writepage(wbc
, page
);
1537 static int ext4_writeback_writepage(struct page
*page
,
1538 struct writeback_control
*wbc
)
1540 struct inode
*inode
= page
->mapping
->host
;
1541 handle_t
*handle
= NULL
;
1545 if (ext4_journal_current_handle())
1548 handle
= ext4_journal_start(inode
, ext4_writepage_trans_blocks(inode
));
1549 if (IS_ERR(handle
)) {
1550 ret
= PTR_ERR(handle
);
1554 if (test_opt(inode
->i_sb
, NOBH
) && ext4_should_writeback_data(inode
))
1555 ret
= nobh_writepage(page
, ext4_get_block
, wbc
);
1557 ret
= block_write_full_page(page
, ext4_get_block
, wbc
);
1559 err
= ext4_journal_stop(handle
);
1565 redirty_page_for_writepage(wbc
, page
);
1570 static int ext4_journalled_writepage(struct page
*page
,
1571 struct writeback_control
*wbc
)
1573 struct inode
*inode
= page
->mapping
->host
;
1574 handle_t
*handle
= NULL
;
1578 if (ext4_journal_current_handle())
1581 handle
= ext4_journal_start(inode
, ext4_writepage_trans_blocks(inode
));
1582 if (IS_ERR(handle
)) {
1583 ret
= PTR_ERR(handle
);
1587 if (!page_has_buffers(page
) || PageChecked(page
)) {
1589 * It's mmapped pagecache. Add buffers and journal it. There
1590 * doesn't seem much point in redirtying the page here.
1592 ClearPageChecked(page
);
1593 ret
= block_prepare_write(page
, 0, PAGE_CACHE_SIZE
,
1596 ext4_journal_stop(handle
);
1599 ret
= walk_page_buffers(handle
, page_buffers(page
), 0,
1600 PAGE_CACHE_SIZE
, NULL
, do_journal_get_write_access
);
1602 err
= walk_page_buffers(handle
, page_buffers(page
), 0,
1603 PAGE_CACHE_SIZE
, NULL
, write_end_fn
);
1606 EXT4_I(inode
)->i_state
|= EXT4_STATE_JDATA
;
1610 * It may be a page full of checkpoint-mode buffers. We don't
1611 * really know unless we go poke around in the buffer_heads.
1612 * But block_write_full_page will do the right thing.
1614 ret
= block_write_full_page(page
, ext4_get_block
, wbc
);
1616 err
= ext4_journal_stop(handle
);
1623 redirty_page_for_writepage(wbc
, page
);
1629 static int ext4_readpage(struct file
*file
, struct page
*page
)
1631 return mpage_readpage(page
, ext4_get_block
);
1635 ext4_readpages(struct file
*file
, struct address_space
*mapping
,
1636 struct list_head
*pages
, unsigned nr_pages
)
1638 return mpage_readpages(mapping
, pages
, nr_pages
, ext4_get_block
);
1641 static void ext4_invalidatepage(struct page
*page
, unsigned long offset
)
1643 journal_t
*journal
= EXT4_JOURNAL(page
->mapping
->host
);
1646 * If it's a full truncate we just forget about the pending dirtying
1649 ClearPageChecked(page
);
1651 jbd2_journal_invalidatepage(journal
, page
, offset
);
1654 static int ext4_releasepage(struct page
*page
, gfp_t wait
)
1656 journal_t
*journal
= EXT4_JOURNAL(page
->mapping
->host
);
1658 WARN_ON(PageChecked(page
));
1659 if (!page_has_buffers(page
))
1661 return jbd2_journal_try_to_free_buffers(journal
, page
, wait
);
1665 * If the O_DIRECT write will extend the file then add this inode to the
1666 * orphan list. So recovery will truncate it back to the original size
1667 * if the machine crashes during the write.
1669 * If the O_DIRECT write is intantiating holes inside i_size and the machine
1670 * crashes then stale disk data _may_ be exposed inside the file. But current
1671 * VFS code falls back into buffered path in that case so we are safe.
1673 static ssize_t
ext4_direct_IO(int rw
, struct kiocb
*iocb
,
1674 const struct iovec
*iov
, loff_t offset
,
1675 unsigned long nr_segs
)
1677 struct file
*file
= iocb
->ki_filp
;
1678 struct inode
*inode
= file
->f_mapping
->host
;
1679 struct ext4_inode_info
*ei
= EXT4_I(inode
);
1683 size_t count
= iov_length(iov
, nr_segs
);
1686 loff_t final_size
= offset
+ count
;
1688 if (final_size
> inode
->i_size
) {
1689 /* Credits for sb + inode write */
1690 handle
= ext4_journal_start(inode
, 2);
1691 if (IS_ERR(handle
)) {
1692 ret
= PTR_ERR(handle
);
1695 ret
= ext4_orphan_add(handle
, inode
);
1697 ext4_journal_stop(handle
);
1701 ei
->i_disksize
= inode
->i_size
;
1702 ext4_journal_stop(handle
);
1706 ret
= blockdev_direct_IO(rw
, iocb
, inode
, inode
->i_sb
->s_bdev
, iov
,
1708 ext4_get_block
, NULL
);
1713 /* Credits for sb + inode write */
1714 handle
= ext4_journal_start(inode
, 2);
1715 if (IS_ERR(handle
)) {
1716 /* This is really bad luck. We've written the data
1717 * but cannot extend i_size. Bail out and pretend
1718 * the write failed... */
1719 ret
= PTR_ERR(handle
);
1723 ext4_orphan_del(handle
, inode
);
1725 loff_t end
= offset
+ ret
;
1726 if (end
> inode
->i_size
) {
1727 ei
->i_disksize
= end
;
1728 i_size_write(inode
, end
);
1730 * We're going to return a positive `ret'
1731 * here due to non-zero-length I/O, so there's
1732 * no way of reporting error returns from
1733 * ext4_mark_inode_dirty() to userspace. So
1736 ext4_mark_inode_dirty(handle
, inode
);
1739 err
= ext4_journal_stop(handle
);
1748 * Pages can be marked dirty completely asynchronously from ext4's journalling
1749 * activity. By filemap_sync_pte(), try_to_unmap_one(), etc. We cannot do
1750 * much here because ->set_page_dirty is called under VFS locks. The page is
1751 * not necessarily locked.
1753 * We cannot just dirty the page and leave attached buffers clean, because the
1754 * buffers' dirty state is "definitive". We cannot just set the buffers dirty
1755 * or jbddirty because all the journalling code will explode.
1757 * So what we do is to mark the page "pending dirty" and next time writepage
1758 * is called, propagate that into the buffers appropriately.
1760 static int ext4_journalled_set_page_dirty(struct page
*page
)
1762 SetPageChecked(page
);
1763 return __set_page_dirty_nobuffers(page
);
1766 static const struct address_space_operations ext4_ordered_aops
= {
1767 .readpage
= ext4_readpage
,
1768 .readpages
= ext4_readpages
,
1769 .writepage
= ext4_ordered_writepage
,
1770 .sync_page
= block_sync_page
,
1771 .write_begin
= ext4_write_begin
,
1772 .write_end
= ext4_ordered_write_end
,
1774 .invalidatepage
= ext4_invalidatepage
,
1775 .releasepage
= ext4_releasepage
,
1776 .direct_IO
= ext4_direct_IO
,
1777 .migratepage
= buffer_migrate_page
,
1780 static const struct address_space_operations ext4_writeback_aops
= {
1781 .readpage
= ext4_readpage
,
1782 .readpages
= ext4_readpages
,
1783 .writepage
= ext4_writeback_writepage
,
1784 .sync_page
= block_sync_page
,
1785 .write_begin
= ext4_write_begin
,
1786 .write_end
= ext4_writeback_write_end
,
1788 .invalidatepage
= ext4_invalidatepage
,
1789 .releasepage
= ext4_releasepage
,
1790 .direct_IO
= ext4_direct_IO
,
1791 .migratepage
= buffer_migrate_page
,
1794 static const struct address_space_operations ext4_journalled_aops
= {
1795 .readpage
= ext4_readpage
,
1796 .readpages
= ext4_readpages
,
1797 .writepage
= ext4_journalled_writepage
,
1798 .sync_page
= block_sync_page
,
1799 .write_begin
= ext4_write_begin
,
1800 .write_end
= ext4_journalled_write_end
,
1801 .set_page_dirty
= ext4_journalled_set_page_dirty
,
1803 .invalidatepage
= ext4_invalidatepage
,
1804 .releasepage
= ext4_releasepage
,
1807 void ext4_set_aops(struct inode
*inode
)
1809 if (ext4_should_order_data(inode
))
1810 inode
->i_mapping
->a_ops
= &ext4_ordered_aops
;
1811 else if (ext4_should_writeback_data(inode
))
1812 inode
->i_mapping
->a_ops
= &ext4_writeback_aops
;
1814 inode
->i_mapping
->a_ops
= &ext4_journalled_aops
;
1818 * ext4_block_truncate_page() zeroes out a mapping from file offset `from'
1819 * up to the end of the block which corresponds to `from'.
1820 * This required during truncate. We need to physically zero the tail end
1821 * of that block so it doesn't yield old data if the file is later grown.
1823 int ext4_block_truncate_page(handle_t
*handle
, struct page
*page
,
1824 struct address_space
*mapping
, loff_t from
)
1826 ext4_fsblk_t index
= from
>> PAGE_CACHE_SHIFT
;
1827 unsigned offset
= from
& (PAGE_CACHE_SIZE
-1);
1828 unsigned blocksize
, length
, pos
;
1830 struct inode
*inode
= mapping
->host
;
1831 struct buffer_head
*bh
;
1834 blocksize
= inode
->i_sb
->s_blocksize
;
1835 length
= blocksize
- (offset
& (blocksize
- 1));
1836 iblock
= index
<< (PAGE_CACHE_SHIFT
- inode
->i_sb
->s_blocksize_bits
);
1839 * For "nobh" option, we can only work if we don't need to
1840 * read-in the page - otherwise we create buffers to do the IO.
1842 if (!page_has_buffers(page
) && test_opt(inode
->i_sb
, NOBH
) &&
1843 ext4_should_writeback_data(inode
) && PageUptodate(page
)) {
1844 zero_user(page
, offset
, length
);
1845 set_page_dirty(page
);
1849 if (!page_has_buffers(page
))
1850 create_empty_buffers(page
, blocksize
, 0);
1852 /* Find the buffer that contains "offset" */
1853 bh
= page_buffers(page
);
1855 while (offset
>= pos
) {
1856 bh
= bh
->b_this_page
;
1862 if (buffer_freed(bh
)) {
1863 BUFFER_TRACE(bh
, "freed: skip");
1867 if (!buffer_mapped(bh
)) {
1868 BUFFER_TRACE(bh
, "unmapped");
1869 ext4_get_block(inode
, iblock
, bh
, 0);
1870 /* unmapped? It's a hole - nothing to do */
1871 if (!buffer_mapped(bh
)) {
1872 BUFFER_TRACE(bh
, "still unmapped");
1877 /* Ok, it's mapped. Make sure it's up-to-date */
1878 if (PageUptodate(page
))
1879 set_buffer_uptodate(bh
);
1881 if (!buffer_uptodate(bh
)) {
1883 ll_rw_block(READ
, 1, &bh
);
1885 /* Uhhuh. Read error. Complain and punt. */
1886 if (!buffer_uptodate(bh
))
1890 if (ext4_should_journal_data(inode
)) {
1891 BUFFER_TRACE(bh
, "get write access");
1892 err
= ext4_journal_get_write_access(handle
, bh
);
1897 zero_user(page
, offset
, length
);
1899 BUFFER_TRACE(bh
, "zeroed end of block");
1902 if (ext4_should_journal_data(inode
)) {
1903 err
= ext4_journal_dirty_metadata(handle
, bh
);
1905 if (ext4_should_order_data(inode
))
1906 err
= ext4_journal_dirty_data(handle
, bh
);
1907 mark_buffer_dirty(bh
);
1912 page_cache_release(page
);
1917 * Probably it should be a library function... search for first non-zero word
1918 * or memcmp with zero_page, whatever is better for particular architecture.
1921 static inline int all_zeroes(__le32
*p
, __le32
*q
)
1930 * ext4_find_shared - find the indirect blocks for partial truncation.
1931 * @inode: inode in question
1932 * @depth: depth of the affected branch
1933 * @offsets: offsets of pointers in that branch (see ext4_block_to_path)
1934 * @chain: place to store the pointers to partial indirect blocks
1935 * @top: place to the (detached) top of branch
1937 * This is a helper function used by ext4_truncate().
1939 * When we do truncate() we may have to clean the ends of several
1940 * indirect blocks but leave the blocks themselves alive. Block is
1941 * partially truncated if some data below the new i_size is refered
1942 * from it (and it is on the path to the first completely truncated
1943 * data block, indeed). We have to free the top of that path along
1944 * with everything to the right of the path. Since no allocation
1945 * past the truncation point is possible until ext4_truncate()
1946 * finishes, we may safely do the latter, but top of branch may
1947 * require special attention - pageout below the truncation point
1948 * might try to populate it.
1950 * We atomically detach the top of branch from the tree, store the
1951 * block number of its root in *@top, pointers to buffer_heads of
1952 * partially truncated blocks - in @chain[].bh and pointers to
1953 * their last elements that should not be removed - in
1954 * @chain[].p. Return value is the pointer to last filled element
1957 * The work left to caller to do the actual freeing of subtrees:
1958 * a) free the subtree starting from *@top
1959 * b) free the subtrees whose roots are stored in
1960 * (@chain[i].p+1 .. end of @chain[i].bh->b_data)
1961 * c) free the subtrees growing from the inode past the @chain[0].
1962 * (no partially truncated stuff there). */
1964 static Indirect
*ext4_find_shared(struct inode
*inode
, int depth
,
1965 ext4_lblk_t offsets
[4], Indirect chain
[4], __le32
*top
)
1967 Indirect
*partial
, *p
;
1971 /* Make k index the deepest non-null offest + 1 */
1972 for (k
= depth
; k
> 1 && !offsets
[k
-1]; k
--)
1974 partial
= ext4_get_branch(inode
, k
, offsets
, chain
, &err
);
1975 /* Writer: pointers */
1977 partial
= chain
+ k
-1;
1979 * If the branch acquired continuation since we've looked at it -
1980 * fine, it should all survive and (new) top doesn't belong to us.
1982 if (!partial
->key
&& *partial
->p
)
1985 for (p
=partial
; p
>chain
&& all_zeroes((__le32
*)p
->bh
->b_data
,p
->p
); p
--)
1988 * OK, we've found the last block that must survive. The rest of our
1989 * branch should be detached before unlocking. However, if that rest
1990 * of branch is all ours and does not grow immediately from the inode
1991 * it's easier to cheat and just decrement partial->p.
1993 if (p
== chain
+ k
- 1 && p
> chain
) {
1997 /* Nope, don't do this in ext4. Must leave the tree intact */
2004 while(partial
> p
) {
2005 brelse(partial
->bh
);
2013 * Zero a number of block pointers in either an inode or an indirect block.
2014 * If we restart the transaction we must again get write access to the
2015 * indirect block for further modification.
2017 * We release `count' blocks on disk, but (last - first) may be greater
2018 * than `count' because there can be holes in there.
2020 static void ext4_clear_blocks(handle_t
*handle
, struct inode
*inode
,
2021 struct buffer_head
*bh
, ext4_fsblk_t block_to_free
,
2022 unsigned long count
, __le32
*first
, __le32
*last
)
2025 if (try_to_extend_transaction(handle
, inode
)) {
2027 BUFFER_TRACE(bh
, "call ext4_journal_dirty_metadata");
2028 ext4_journal_dirty_metadata(handle
, bh
);
2030 ext4_mark_inode_dirty(handle
, inode
);
2031 ext4_journal_test_restart(handle
, inode
);
2033 BUFFER_TRACE(bh
, "retaking write access");
2034 ext4_journal_get_write_access(handle
, bh
);
2039 * Any buffers which are on the journal will be in memory. We find
2040 * them on the hash table so jbd2_journal_revoke() will run jbd2_journal_forget()
2041 * on them. We've already detached each block from the file, so
2042 * bforget() in jbd2_journal_forget() should be safe.
2044 * AKPM: turn on bforget in jbd2_journal_forget()!!!
2046 for (p
= first
; p
< last
; p
++) {
2047 u32 nr
= le32_to_cpu(*p
);
2049 struct buffer_head
*tbh
;
2052 tbh
= sb_find_get_block(inode
->i_sb
, nr
);
2053 ext4_forget(handle
, 0, inode
, tbh
, nr
);
2057 ext4_free_blocks(handle
, inode
, block_to_free
, count
, 0);
2061 * ext4_free_data - free a list of data blocks
2062 * @handle: handle for this transaction
2063 * @inode: inode we are dealing with
2064 * @this_bh: indirect buffer_head which contains *@first and *@last
2065 * @first: array of block numbers
2066 * @last: points immediately past the end of array
2068 * We are freeing all blocks refered from that array (numbers are stored as
2069 * little-endian 32-bit) and updating @inode->i_blocks appropriately.
2071 * We accumulate contiguous runs of blocks to free. Conveniently, if these
2072 * blocks are contiguous then releasing them at one time will only affect one
2073 * or two bitmap blocks (+ group descriptor(s) and superblock) and we won't
2074 * actually use a lot of journal space.
2076 * @this_bh will be %NULL if @first and @last point into the inode's direct
2079 static void ext4_free_data(handle_t
*handle
, struct inode
*inode
,
2080 struct buffer_head
*this_bh
,
2081 __le32
*first
, __le32
*last
)
2083 ext4_fsblk_t block_to_free
= 0; /* Starting block # of a run */
2084 unsigned long count
= 0; /* Number of blocks in the run */
2085 __le32
*block_to_free_p
= NULL
; /* Pointer into inode/ind
2088 ext4_fsblk_t nr
; /* Current block # */
2089 __le32
*p
; /* Pointer into inode/ind
2090 for current block */
2093 if (this_bh
) { /* For indirect block */
2094 BUFFER_TRACE(this_bh
, "get_write_access");
2095 err
= ext4_journal_get_write_access(handle
, this_bh
);
2096 /* Important: if we can't update the indirect pointers
2097 * to the blocks, we can't free them. */
2102 for (p
= first
; p
< last
; p
++) {
2103 nr
= le32_to_cpu(*p
);
2105 /* accumulate blocks to free if they're contiguous */
2108 block_to_free_p
= p
;
2110 } else if (nr
== block_to_free
+ count
) {
2113 ext4_clear_blocks(handle
, inode
, this_bh
,
2115 count
, block_to_free_p
, p
);
2117 block_to_free_p
= p
;
2124 ext4_clear_blocks(handle
, inode
, this_bh
, block_to_free
,
2125 count
, block_to_free_p
, p
);
2128 BUFFER_TRACE(this_bh
, "call ext4_journal_dirty_metadata");
2129 ext4_journal_dirty_metadata(handle
, this_bh
);
2134 * ext4_free_branches - free an array of branches
2135 * @handle: JBD handle for this transaction
2136 * @inode: inode we are dealing with
2137 * @parent_bh: the buffer_head which contains *@first and *@last
2138 * @first: array of block numbers
2139 * @last: pointer immediately past the end of array
2140 * @depth: depth of the branches to free
2142 * We are freeing all blocks refered from these branches (numbers are
2143 * stored as little-endian 32-bit) and updating @inode->i_blocks
2146 static void ext4_free_branches(handle_t
*handle
, struct inode
*inode
,
2147 struct buffer_head
*parent_bh
,
2148 __le32
*first
, __le32
*last
, int depth
)
2153 if (is_handle_aborted(handle
))
2157 struct buffer_head
*bh
;
2158 int addr_per_block
= EXT4_ADDR_PER_BLOCK(inode
->i_sb
);
2160 while (--p
>= first
) {
2161 nr
= le32_to_cpu(*p
);
2163 continue; /* A hole */
2165 /* Go read the buffer for the next level down */
2166 bh
= sb_bread(inode
->i_sb
, nr
);
2169 * A read failure? Report error and clear slot
2173 ext4_error(inode
->i_sb
, "ext4_free_branches",
2174 "Read failure, inode=%lu, block=%llu",
2179 /* This zaps the entire block. Bottom up. */
2180 BUFFER_TRACE(bh
, "free child branches");
2181 ext4_free_branches(handle
, inode
, bh
,
2182 (__le32
*)bh
->b_data
,
2183 (__le32
*)bh
->b_data
+ addr_per_block
,
2187 * We've probably journalled the indirect block several
2188 * times during the truncate. But it's no longer
2189 * needed and we now drop it from the transaction via
2190 * jbd2_journal_revoke().
2192 * That's easy if it's exclusively part of this
2193 * transaction. But if it's part of the committing
2194 * transaction then jbd2_journal_forget() will simply
2195 * brelse() it. That means that if the underlying
2196 * block is reallocated in ext4_get_block(),
2197 * unmap_underlying_metadata() will find this block
2198 * and will try to get rid of it. damn, damn.
2200 * If this block has already been committed to the
2201 * journal, a revoke record will be written. And
2202 * revoke records must be emitted *before* clearing
2203 * this block's bit in the bitmaps.
2205 ext4_forget(handle
, 1, inode
, bh
, bh
->b_blocknr
);
2208 * Everything below this this pointer has been
2209 * released. Now let this top-of-subtree go.
2211 * We want the freeing of this indirect block to be
2212 * atomic in the journal with the updating of the
2213 * bitmap block which owns it. So make some room in
2216 * We zero the parent pointer *after* freeing its
2217 * pointee in the bitmaps, so if extend_transaction()
2218 * for some reason fails to put the bitmap changes and
2219 * the release into the same transaction, recovery
2220 * will merely complain about releasing a free block,
2221 * rather than leaking blocks.
2223 if (is_handle_aborted(handle
))
2225 if (try_to_extend_transaction(handle
, inode
)) {
2226 ext4_mark_inode_dirty(handle
, inode
);
2227 ext4_journal_test_restart(handle
, inode
);
2230 ext4_free_blocks(handle
, inode
, nr
, 1, 1);
2234 * The block which we have just freed is
2235 * pointed to by an indirect block: journal it
2237 BUFFER_TRACE(parent_bh
, "get_write_access");
2238 if (!ext4_journal_get_write_access(handle
,
2241 BUFFER_TRACE(parent_bh
,
2242 "call ext4_journal_dirty_metadata");
2243 ext4_journal_dirty_metadata(handle
,
2249 /* We have reached the bottom of the tree. */
2250 BUFFER_TRACE(parent_bh
, "free data blocks");
2251 ext4_free_data(handle
, inode
, parent_bh
, first
, last
);
2258 * We block out ext4_get_block() block instantiations across the entire
2259 * transaction, and VFS/VM ensures that ext4_truncate() cannot run
2260 * simultaneously on behalf of the same inode.
2262 * As we work through the truncate and commmit bits of it to the journal there
2263 * is one core, guiding principle: the file's tree must always be consistent on
2264 * disk. We must be able to restart the truncate after a crash.
2266 * The file's tree may be transiently inconsistent in memory (although it
2267 * probably isn't), but whenever we close off and commit a journal transaction,
2268 * the contents of (the filesystem + the journal) must be consistent and
2269 * restartable. It's pretty simple, really: bottom up, right to left (although
2270 * left-to-right works OK too).
2272 * Note that at recovery time, journal replay occurs *before* the restart of
2273 * truncate against the orphan inode list.
2275 * The committed inode has the new, desired i_size (which is the same as
2276 * i_disksize in this case). After a crash, ext4_orphan_cleanup() will see
2277 * that this inode's truncate did not complete and it will again call
2278 * ext4_truncate() to have another go. So there will be instantiated blocks
2279 * to the right of the truncation point in a crashed ext4 filesystem. But
2280 * that's fine - as long as they are linked from the inode, the post-crash
2281 * ext4_truncate() run will find them and release them.
2283 void ext4_truncate(struct inode
*inode
)
2286 struct ext4_inode_info
*ei
= EXT4_I(inode
);
2287 __le32
*i_data
= ei
->i_data
;
2288 int addr_per_block
= EXT4_ADDR_PER_BLOCK(inode
->i_sb
);
2289 struct address_space
*mapping
= inode
->i_mapping
;
2290 ext4_lblk_t offsets
[4];
2295 ext4_lblk_t last_block
;
2296 unsigned blocksize
= inode
->i_sb
->s_blocksize
;
2299 if (!(S_ISREG(inode
->i_mode
) || S_ISDIR(inode
->i_mode
) ||
2300 S_ISLNK(inode
->i_mode
)))
2302 if (ext4_inode_is_fast_symlink(inode
))
2304 if (IS_APPEND(inode
) || IS_IMMUTABLE(inode
))
2308 * We have to lock the EOF page here, because lock_page() nests
2309 * outside jbd2_journal_start().
2311 if ((inode
->i_size
& (blocksize
- 1)) == 0) {
2312 /* Block boundary? Nothing to do */
2315 page
= grab_cache_page(mapping
,
2316 inode
->i_size
>> PAGE_CACHE_SHIFT
);
2321 if (EXT4_I(inode
)->i_flags
& EXT4_EXTENTS_FL
) {
2322 ext4_ext_truncate(inode
, page
);
2326 handle
= start_transaction(inode
);
2327 if (IS_ERR(handle
)) {
2329 clear_highpage(page
);
2330 flush_dcache_page(page
);
2332 page_cache_release(page
);
2334 return; /* AKPM: return what? */
2337 last_block
= (inode
->i_size
+ blocksize
-1)
2338 >> EXT4_BLOCK_SIZE_BITS(inode
->i_sb
);
2341 ext4_block_truncate_page(handle
, page
, mapping
, inode
->i_size
);
2343 n
= ext4_block_to_path(inode
, last_block
, offsets
, NULL
);
2345 goto out_stop
; /* error */
2348 * OK. This truncate is going to happen. We add the inode to the
2349 * orphan list, so that if this truncate spans multiple transactions,
2350 * and we crash, we will resume the truncate when the filesystem
2351 * recovers. It also marks the inode dirty, to catch the new size.
2353 * Implication: the file must always be in a sane, consistent
2354 * truncatable state while each transaction commits.
2356 if (ext4_orphan_add(handle
, inode
))
2360 * The orphan list entry will now protect us from any crash which
2361 * occurs before the truncate completes, so it is now safe to propagate
2362 * the new, shorter inode size (held for now in i_size) into the
2363 * on-disk inode. We do this via i_disksize, which is the value which
2364 * ext4 *really* writes onto the disk inode.
2366 ei
->i_disksize
= inode
->i_size
;
2369 * From here we block out all ext4_get_block() callers who want to
2370 * modify the block allocation tree.
2372 down_write(&ei
->i_data_sem
);
2374 if (n
== 1) { /* direct blocks */
2375 ext4_free_data(handle
, inode
, NULL
, i_data
+offsets
[0],
2376 i_data
+ EXT4_NDIR_BLOCKS
);
2380 partial
= ext4_find_shared(inode
, n
, offsets
, chain
, &nr
);
2381 /* Kill the top of shared branch (not detached) */
2383 if (partial
== chain
) {
2384 /* Shared branch grows from the inode */
2385 ext4_free_branches(handle
, inode
, NULL
,
2386 &nr
, &nr
+1, (chain
+n
-1) - partial
);
2389 * We mark the inode dirty prior to restart,
2390 * and prior to stop. No need for it here.
2393 /* Shared branch grows from an indirect block */
2394 BUFFER_TRACE(partial
->bh
, "get_write_access");
2395 ext4_free_branches(handle
, inode
, partial
->bh
,
2397 partial
->p
+1, (chain
+n
-1) - partial
);
2400 /* Clear the ends of indirect blocks on the shared branch */
2401 while (partial
> chain
) {
2402 ext4_free_branches(handle
, inode
, partial
->bh
, partial
->p
+ 1,
2403 (__le32
*)partial
->bh
->b_data
+addr_per_block
,
2404 (chain
+n
-1) - partial
);
2405 BUFFER_TRACE(partial
->bh
, "call brelse");
2406 brelse (partial
->bh
);
2410 /* Kill the remaining (whole) subtrees */
2411 switch (offsets
[0]) {
2413 nr
= i_data
[EXT4_IND_BLOCK
];
2415 ext4_free_branches(handle
, inode
, NULL
, &nr
, &nr
+1, 1);
2416 i_data
[EXT4_IND_BLOCK
] = 0;
2418 case EXT4_IND_BLOCK
:
2419 nr
= i_data
[EXT4_DIND_BLOCK
];
2421 ext4_free_branches(handle
, inode
, NULL
, &nr
, &nr
+1, 2);
2422 i_data
[EXT4_DIND_BLOCK
] = 0;
2424 case EXT4_DIND_BLOCK
:
2425 nr
= i_data
[EXT4_TIND_BLOCK
];
2427 ext4_free_branches(handle
, inode
, NULL
, &nr
, &nr
+1, 3);
2428 i_data
[EXT4_TIND_BLOCK
] = 0;
2430 case EXT4_TIND_BLOCK
:
2434 ext4_discard_reservation(inode
);
2436 up_write(&ei
->i_data_sem
);
2437 inode
->i_mtime
= inode
->i_ctime
= ext4_current_time(inode
);
2438 ext4_mark_inode_dirty(handle
, inode
);
2441 * In a multi-transaction truncate, we only make the final transaction
2448 * If this was a simple ftruncate(), and the file will remain alive
2449 * then we need to clear up the orphan record which we created above.
2450 * However, if this was a real unlink then we were called by
2451 * ext4_delete_inode(), and we allow that function to clean up the
2452 * orphan info for us.
2455 ext4_orphan_del(handle
, inode
);
2457 ext4_journal_stop(handle
);
2460 static ext4_fsblk_t
ext4_get_inode_block(struct super_block
*sb
,
2461 unsigned long ino
, struct ext4_iloc
*iloc
)
2463 unsigned long desc
, group_desc
;
2464 ext4_group_t block_group
;
2465 unsigned long offset
;
2467 struct buffer_head
*bh
;
2468 struct ext4_group_desc
* gdp
;
2470 if (!ext4_valid_inum(sb
, ino
)) {
2472 * This error is already checked for in namei.c unless we are
2473 * looking at an NFS filehandle, in which case no error
2479 block_group
= (ino
- 1) / EXT4_INODES_PER_GROUP(sb
);
2480 if (block_group
>= EXT4_SB(sb
)->s_groups_count
) {
2481 ext4_error(sb
,"ext4_get_inode_block","group >= groups count");
2485 group_desc
= block_group
>> EXT4_DESC_PER_BLOCK_BITS(sb
);
2486 desc
= block_group
& (EXT4_DESC_PER_BLOCK(sb
) - 1);
2487 bh
= EXT4_SB(sb
)->s_group_desc
[group_desc
];
2489 ext4_error (sb
, "ext4_get_inode_block",
2490 "Descriptor not loaded");
2494 gdp
= (struct ext4_group_desc
*)((__u8
*)bh
->b_data
+
2495 desc
* EXT4_DESC_SIZE(sb
));
2497 * Figure out the offset within the block group inode table
2499 offset
= ((ino
- 1) % EXT4_INODES_PER_GROUP(sb
)) *
2500 EXT4_INODE_SIZE(sb
);
2501 block
= ext4_inode_table(sb
, gdp
) +
2502 (offset
>> EXT4_BLOCK_SIZE_BITS(sb
));
2504 iloc
->block_group
= block_group
;
2505 iloc
->offset
= offset
& (EXT4_BLOCK_SIZE(sb
) - 1);
2510 * ext4_get_inode_loc returns with an extra refcount against the inode's
2511 * underlying buffer_head on success. If 'in_mem' is true, we have all
2512 * data in memory that is needed to recreate the on-disk version of this
2515 static int __ext4_get_inode_loc(struct inode
*inode
,
2516 struct ext4_iloc
*iloc
, int in_mem
)
2519 struct buffer_head
*bh
;
2521 block
= ext4_get_inode_block(inode
->i_sb
, inode
->i_ino
, iloc
);
2525 bh
= sb_getblk(inode
->i_sb
, block
);
2527 ext4_error (inode
->i_sb
, "ext4_get_inode_loc",
2528 "unable to read inode block - "
2529 "inode=%lu, block=%llu",
2530 inode
->i_ino
, block
);
2533 if (!buffer_uptodate(bh
)) {
2535 if (buffer_uptodate(bh
)) {
2536 /* someone brought it uptodate while we waited */
2542 * If we have all information of the inode in memory and this
2543 * is the only valid inode in the block, we need not read the
2547 struct buffer_head
*bitmap_bh
;
2548 struct ext4_group_desc
*desc
;
2549 int inodes_per_buffer
;
2550 int inode_offset
, i
;
2551 ext4_group_t block_group
;
2554 block_group
= (inode
->i_ino
- 1) /
2555 EXT4_INODES_PER_GROUP(inode
->i_sb
);
2556 inodes_per_buffer
= bh
->b_size
/
2557 EXT4_INODE_SIZE(inode
->i_sb
);
2558 inode_offset
= ((inode
->i_ino
- 1) %
2559 EXT4_INODES_PER_GROUP(inode
->i_sb
));
2560 start
= inode_offset
& ~(inodes_per_buffer
- 1);
2562 /* Is the inode bitmap in cache? */
2563 desc
= ext4_get_group_desc(inode
->i_sb
,
2568 bitmap_bh
= sb_getblk(inode
->i_sb
,
2569 ext4_inode_bitmap(inode
->i_sb
, desc
));
2574 * If the inode bitmap isn't in cache then the
2575 * optimisation may end up performing two reads instead
2576 * of one, so skip it.
2578 if (!buffer_uptodate(bitmap_bh
)) {
2582 for (i
= start
; i
< start
+ inodes_per_buffer
; i
++) {
2583 if (i
== inode_offset
)
2585 if (ext4_test_bit(i
, bitmap_bh
->b_data
))
2589 if (i
== start
+ inodes_per_buffer
) {
2590 /* all other inodes are free, so skip I/O */
2591 memset(bh
->b_data
, 0, bh
->b_size
);
2592 set_buffer_uptodate(bh
);
2600 * There are other valid inodes in the buffer, this inode
2601 * has in-inode xattrs, or we don't have this inode in memory.
2602 * Read the block from disk.
2605 bh
->b_end_io
= end_buffer_read_sync
;
2606 submit_bh(READ_META
, bh
);
2608 if (!buffer_uptodate(bh
)) {
2609 ext4_error(inode
->i_sb
, "ext4_get_inode_loc",
2610 "unable to read inode block - "
2611 "inode=%lu, block=%llu",
2612 inode
->i_ino
, block
);
2622 int ext4_get_inode_loc(struct inode
*inode
, struct ext4_iloc
*iloc
)
2624 /* We have all inode data except xattrs in memory here. */
2625 return __ext4_get_inode_loc(inode
, iloc
,
2626 !(EXT4_I(inode
)->i_state
& EXT4_STATE_XATTR
));
2629 void ext4_set_inode_flags(struct inode
*inode
)
2631 unsigned int flags
= EXT4_I(inode
)->i_flags
;
2633 inode
->i_flags
&= ~(S_SYNC
|S_APPEND
|S_IMMUTABLE
|S_NOATIME
|S_DIRSYNC
);
2634 if (flags
& EXT4_SYNC_FL
)
2635 inode
->i_flags
|= S_SYNC
;
2636 if (flags
& EXT4_APPEND_FL
)
2637 inode
->i_flags
|= S_APPEND
;
2638 if (flags
& EXT4_IMMUTABLE_FL
)
2639 inode
->i_flags
|= S_IMMUTABLE
;
2640 if (flags
& EXT4_NOATIME_FL
)
2641 inode
->i_flags
|= S_NOATIME
;
2642 if (flags
& EXT4_DIRSYNC_FL
)
2643 inode
->i_flags
|= S_DIRSYNC
;
2646 /* Propagate flags from i_flags to EXT4_I(inode)->i_flags */
2647 void ext4_get_inode_flags(struct ext4_inode_info
*ei
)
2649 unsigned int flags
= ei
->vfs_inode
.i_flags
;
2651 ei
->i_flags
&= ~(EXT4_SYNC_FL
|EXT4_APPEND_FL
|
2652 EXT4_IMMUTABLE_FL
|EXT4_NOATIME_FL
|EXT4_DIRSYNC_FL
);
2654 ei
->i_flags
|= EXT4_SYNC_FL
;
2655 if (flags
& S_APPEND
)
2656 ei
->i_flags
|= EXT4_APPEND_FL
;
2657 if (flags
& S_IMMUTABLE
)
2658 ei
->i_flags
|= EXT4_IMMUTABLE_FL
;
2659 if (flags
& S_NOATIME
)
2660 ei
->i_flags
|= EXT4_NOATIME_FL
;
2661 if (flags
& S_DIRSYNC
)
2662 ei
->i_flags
|= EXT4_DIRSYNC_FL
;
2664 static blkcnt_t
ext4_inode_blocks(struct ext4_inode
*raw_inode
,
2665 struct ext4_inode_info
*ei
)
2668 struct inode
*inode
= &(ei
->vfs_inode
);
2669 struct super_block
*sb
= inode
->i_sb
;
2671 if (EXT4_HAS_RO_COMPAT_FEATURE(sb
,
2672 EXT4_FEATURE_RO_COMPAT_HUGE_FILE
)) {
2673 /* we are using combined 48 bit field */
2674 i_blocks
= ((u64
)le16_to_cpu(raw_inode
->i_blocks_high
)) << 32 |
2675 le32_to_cpu(raw_inode
->i_blocks_lo
);
2676 if (ei
->i_flags
& EXT4_HUGE_FILE_FL
) {
2677 /* i_blocks represent file system block size */
2678 return i_blocks
<< (inode
->i_blkbits
- 9);
2683 return le32_to_cpu(raw_inode
->i_blocks_lo
);
2687 struct inode
*ext4_iget(struct super_block
*sb
, unsigned long ino
)
2689 struct ext4_iloc iloc
;
2690 struct ext4_inode
*raw_inode
;
2691 struct ext4_inode_info
*ei
;
2692 struct buffer_head
*bh
;
2693 struct inode
*inode
;
2697 inode
= iget_locked(sb
, ino
);
2699 return ERR_PTR(-ENOMEM
);
2700 if (!(inode
->i_state
& I_NEW
))
2704 #ifdef CONFIG_EXT4DEV_FS_POSIX_ACL
2705 ei
->i_acl
= EXT4_ACL_NOT_CACHED
;
2706 ei
->i_default_acl
= EXT4_ACL_NOT_CACHED
;
2708 ei
->i_block_alloc_info
= NULL
;
2710 ret
= __ext4_get_inode_loc(inode
, &iloc
, 0);
2714 raw_inode
= ext4_raw_inode(&iloc
);
2715 inode
->i_mode
= le16_to_cpu(raw_inode
->i_mode
);
2716 inode
->i_uid
= (uid_t
)le16_to_cpu(raw_inode
->i_uid_low
);
2717 inode
->i_gid
= (gid_t
)le16_to_cpu(raw_inode
->i_gid_low
);
2718 if(!(test_opt (inode
->i_sb
, NO_UID32
))) {
2719 inode
->i_uid
|= le16_to_cpu(raw_inode
->i_uid_high
) << 16;
2720 inode
->i_gid
|= le16_to_cpu(raw_inode
->i_gid_high
) << 16;
2722 inode
->i_nlink
= le16_to_cpu(raw_inode
->i_links_count
);
2725 ei
->i_dir_start_lookup
= 0;
2726 ei
->i_dtime
= le32_to_cpu(raw_inode
->i_dtime
);
2727 /* We now have enough fields to check if the inode was active or not.
2728 * This is needed because nfsd might try to access dead inodes
2729 * the test is that same one that e2fsck uses
2730 * NeilBrown 1999oct15
2732 if (inode
->i_nlink
== 0) {
2733 if (inode
->i_mode
== 0 ||
2734 !(EXT4_SB(inode
->i_sb
)->s_mount_state
& EXT4_ORPHAN_FS
)) {
2735 /* this inode is deleted */
2740 /* The only unlinked inodes we let through here have
2741 * valid i_mode and are being read by the orphan
2742 * recovery code: that's fine, we're about to complete
2743 * the process of deleting those. */
2745 ei
->i_flags
= le32_to_cpu(raw_inode
->i_flags
);
2746 inode
->i_blocks
= ext4_inode_blocks(raw_inode
, ei
);
2747 ei
->i_file_acl
= le32_to_cpu(raw_inode
->i_file_acl_lo
);
2748 if (EXT4_SB(inode
->i_sb
)->s_es
->s_creator_os
!=
2749 cpu_to_le32(EXT4_OS_HURD
)) {
2751 ((__u64
)le16_to_cpu(raw_inode
->i_file_acl_high
)) << 32;
2753 inode
->i_size
= ext4_isize(raw_inode
);
2754 ei
->i_disksize
= inode
->i_size
;
2755 inode
->i_generation
= le32_to_cpu(raw_inode
->i_generation
);
2756 ei
->i_block_group
= iloc
.block_group
;
2758 * NOTE! The in-memory inode i_data array is in little-endian order
2759 * even on big-endian machines: we do NOT byteswap the block numbers!
2761 for (block
= 0; block
< EXT4_N_BLOCKS
; block
++)
2762 ei
->i_data
[block
] = raw_inode
->i_block
[block
];
2763 INIT_LIST_HEAD(&ei
->i_orphan
);
2765 if (EXT4_INODE_SIZE(inode
->i_sb
) > EXT4_GOOD_OLD_INODE_SIZE
) {
2766 ei
->i_extra_isize
= le16_to_cpu(raw_inode
->i_extra_isize
);
2767 if (EXT4_GOOD_OLD_INODE_SIZE
+ ei
->i_extra_isize
>
2768 EXT4_INODE_SIZE(inode
->i_sb
)) {
2773 if (ei
->i_extra_isize
== 0) {
2774 /* The extra space is currently unused. Use it. */
2775 ei
->i_extra_isize
= sizeof(struct ext4_inode
) -
2776 EXT4_GOOD_OLD_INODE_SIZE
;
2778 __le32
*magic
= (void *)raw_inode
+
2779 EXT4_GOOD_OLD_INODE_SIZE
+
2781 if (*magic
== cpu_to_le32(EXT4_XATTR_MAGIC
))
2782 ei
->i_state
|= EXT4_STATE_XATTR
;
2785 ei
->i_extra_isize
= 0;
2787 EXT4_INODE_GET_XTIME(i_ctime
, inode
, raw_inode
);
2788 EXT4_INODE_GET_XTIME(i_mtime
, inode
, raw_inode
);
2789 EXT4_INODE_GET_XTIME(i_atime
, inode
, raw_inode
);
2790 EXT4_EINODE_GET_XTIME(i_crtime
, ei
, raw_inode
);
2792 inode
->i_version
= le32_to_cpu(raw_inode
->i_disk_version
);
2793 if (EXT4_INODE_SIZE(inode
->i_sb
) > EXT4_GOOD_OLD_INODE_SIZE
) {
2794 if (EXT4_FITS_IN_INODE(raw_inode
, ei
, i_version_hi
))
2796 (__u64
)(le32_to_cpu(raw_inode
->i_version_hi
)) << 32;
2799 if (S_ISREG(inode
->i_mode
)) {
2800 inode
->i_op
= &ext4_file_inode_operations
;
2801 inode
->i_fop
= &ext4_file_operations
;
2802 ext4_set_aops(inode
);
2803 } else if (S_ISDIR(inode
->i_mode
)) {
2804 inode
->i_op
= &ext4_dir_inode_operations
;
2805 inode
->i_fop
= &ext4_dir_operations
;
2806 } else if (S_ISLNK(inode
->i_mode
)) {
2807 if (ext4_inode_is_fast_symlink(inode
))
2808 inode
->i_op
= &ext4_fast_symlink_inode_operations
;
2810 inode
->i_op
= &ext4_symlink_inode_operations
;
2811 ext4_set_aops(inode
);
2814 inode
->i_op
= &ext4_special_inode_operations
;
2815 if (raw_inode
->i_block
[0])
2816 init_special_inode(inode
, inode
->i_mode
,
2817 old_decode_dev(le32_to_cpu(raw_inode
->i_block
[0])));
2819 init_special_inode(inode
, inode
->i_mode
,
2820 new_decode_dev(le32_to_cpu(raw_inode
->i_block
[1])));
2823 ext4_set_inode_flags(inode
);
2824 unlock_new_inode(inode
);
2829 return ERR_PTR(ret
);
2832 static int ext4_inode_blocks_set(handle_t
*handle
,
2833 struct ext4_inode
*raw_inode
,
2834 struct ext4_inode_info
*ei
)
2836 struct inode
*inode
= &(ei
->vfs_inode
);
2837 u64 i_blocks
= inode
->i_blocks
;
2838 struct super_block
*sb
= inode
->i_sb
;
2841 if (i_blocks
<= ~0U) {
2843 * i_blocks can be represnted in a 32 bit variable
2844 * as multiple of 512 bytes
2846 raw_inode
->i_blocks_lo
= cpu_to_le32(i_blocks
);
2847 raw_inode
->i_blocks_high
= 0;
2848 ei
->i_flags
&= ~EXT4_HUGE_FILE_FL
;
2849 } else if (i_blocks
<= 0xffffffffffffULL
) {
2851 * i_blocks can be represented in a 48 bit variable
2852 * as multiple of 512 bytes
2854 err
= ext4_update_rocompat_feature(handle
, sb
,
2855 EXT4_FEATURE_RO_COMPAT_HUGE_FILE
);
2858 /* i_block is stored in the split 48 bit fields */
2859 raw_inode
->i_blocks_lo
= cpu_to_le32(i_blocks
);
2860 raw_inode
->i_blocks_high
= cpu_to_le16(i_blocks
>> 32);
2861 ei
->i_flags
&= ~EXT4_HUGE_FILE_FL
;
2864 * i_blocks should be represented in a 48 bit variable
2865 * as multiple of file system block size
2867 err
= ext4_update_rocompat_feature(handle
, sb
,
2868 EXT4_FEATURE_RO_COMPAT_HUGE_FILE
);
2871 ei
->i_flags
|= EXT4_HUGE_FILE_FL
;
2872 /* i_block is stored in file system block size */
2873 i_blocks
= i_blocks
>> (inode
->i_blkbits
- 9);
2874 raw_inode
->i_blocks_lo
= cpu_to_le32(i_blocks
);
2875 raw_inode
->i_blocks_high
= cpu_to_le16(i_blocks
>> 32);
2882 * Post the struct inode info into an on-disk inode location in the
2883 * buffer-cache. This gobbles the caller's reference to the
2884 * buffer_head in the inode location struct.
2886 * The caller must have write access to iloc->bh.
2888 static int ext4_do_update_inode(handle_t
*handle
,
2889 struct inode
*inode
,
2890 struct ext4_iloc
*iloc
)
2892 struct ext4_inode
*raw_inode
= ext4_raw_inode(iloc
);
2893 struct ext4_inode_info
*ei
= EXT4_I(inode
);
2894 struct buffer_head
*bh
= iloc
->bh
;
2895 int err
= 0, rc
, block
;
2897 /* For fields not not tracking in the in-memory inode,
2898 * initialise them to zero for new inodes. */
2899 if (ei
->i_state
& EXT4_STATE_NEW
)
2900 memset(raw_inode
, 0, EXT4_SB(inode
->i_sb
)->s_inode_size
);
2902 ext4_get_inode_flags(ei
);
2903 raw_inode
->i_mode
= cpu_to_le16(inode
->i_mode
);
2904 if(!(test_opt(inode
->i_sb
, NO_UID32
))) {
2905 raw_inode
->i_uid_low
= cpu_to_le16(low_16_bits(inode
->i_uid
));
2906 raw_inode
->i_gid_low
= cpu_to_le16(low_16_bits(inode
->i_gid
));
2908 * Fix up interoperability with old kernels. Otherwise, old inodes get
2909 * re-used with the upper 16 bits of the uid/gid intact
2912 raw_inode
->i_uid_high
=
2913 cpu_to_le16(high_16_bits(inode
->i_uid
));
2914 raw_inode
->i_gid_high
=
2915 cpu_to_le16(high_16_bits(inode
->i_gid
));
2917 raw_inode
->i_uid_high
= 0;
2918 raw_inode
->i_gid_high
= 0;
2921 raw_inode
->i_uid_low
=
2922 cpu_to_le16(fs_high2lowuid(inode
->i_uid
));
2923 raw_inode
->i_gid_low
=
2924 cpu_to_le16(fs_high2lowgid(inode
->i_gid
));
2925 raw_inode
->i_uid_high
= 0;
2926 raw_inode
->i_gid_high
= 0;
2928 raw_inode
->i_links_count
= cpu_to_le16(inode
->i_nlink
);
2930 EXT4_INODE_SET_XTIME(i_ctime
, inode
, raw_inode
);
2931 EXT4_INODE_SET_XTIME(i_mtime
, inode
, raw_inode
);
2932 EXT4_INODE_SET_XTIME(i_atime
, inode
, raw_inode
);
2933 EXT4_EINODE_SET_XTIME(i_crtime
, ei
, raw_inode
);
2935 if (ext4_inode_blocks_set(handle
, raw_inode
, ei
))
2937 raw_inode
->i_dtime
= cpu_to_le32(ei
->i_dtime
);
2938 raw_inode
->i_flags
= cpu_to_le32(ei
->i_flags
);
2939 if (EXT4_SB(inode
->i_sb
)->s_es
->s_creator_os
!=
2940 cpu_to_le32(EXT4_OS_HURD
))
2941 raw_inode
->i_file_acl_high
=
2942 cpu_to_le16(ei
->i_file_acl
>> 32);
2943 raw_inode
->i_file_acl_lo
= cpu_to_le32(ei
->i_file_acl
);
2944 ext4_isize_set(raw_inode
, ei
->i_disksize
);
2945 if (ei
->i_disksize
> 0x7fffffffULL
) {
2946 struct super_block
*sb
= inode
->i_sb
;
2947 if (!EXT4_HAS_RO_COMPAT_FEATURE(sb
,
2948 EXT4_FEATURE_RO_COMPAT_LARGE_FILE
) ||
2949 EXT4_SB(sb
)->s_es
->s_rev_level
==
2950 cpu_to_le32(EXT4_GOOD_OLD_REV
)) {
2951 /* If this is the first large file
2952 * created, add a flag to the superblock.
2954 err
= ext4_journal_get_write_access(handle
,
2955 EXT4_SB(sb
)->s_sbh
);
2958 ext4_update_dynamic_rev(sb
);
2959 EXT4_SET_RO_COMPAT_FEATURE(sb
,
2960 EXT4_FEATURE_RO_COMPAT_LARGE_FILE
);
2963 err
= ext4_journal_dirty_metadata(handle
,
2964 EXT4_SB(sb
)->s_sbh
);
2967 raw_inode
->i_generation
= cpu_to_le32(inode
->i_generation
);
2968 if (S_ISCHR(inode
->i_mode
) || S_ISBLK(inode
->i_mode
)) {
2969 if (old_valid_dev(inode
->i_rdev
)) {
2970 raw_inode
->i_block
[0] =
2971 cpu_to_le32(old_encode_dev(inode
->i_rdev
));
2972 raw_inode
->i_block
[1] = 0;
2974 raw_inode
->i_block
[0] = 0;
2975 raw_inode
->i_block
[1] =
2976 cpu_to_le32(new_encode_dev(inode
->i_rdev
));
2977 raw_inode
->i_block
[2] = 0;
2979 } else for (block
= 0; block
< EXT4_N_BLOCKS
; block
++)
2980 raw_inode
->i_block
[block
] = ei
->i_data
[block
];
2982 raw_inode
->i_disk_version
= cpu_to_le32(inode
->i_version
);
2983 if (ei
->i_extra_isize
) {
2984 if (EXT4_FITS_IN_INODE(raw_inode
, ei
, i_version_hi
))
2985 raw_inode
->i_version_hi
=
2986 cpu_to_le32(inode
->i_version
>> 32);
2987 raw_inode
->i_extra_isize
= cpu_to_le16(ei
->i_extra_isize
);
2991 BUFFER_TRACE(bh
, "call ext4_journal_dirty_metadata");
2992 rc
= ext4_journal_dirty_metadata(handle
, bh
);
2995 ei
->i_state
&= ~EXT4_STATE_NEW
;
2999 ext4_std_error(inode
->i_sb
, err
);
3004 * ext4_write_inode()
3006 * We are called from a few places:
3008 * - Within generic_file_write() for O_SYNC files.
3009 * Here, there will be no transaction running. We wait for any running
3010 * trasnaction to commit.
3012 * - Within sys_sync(), kupdate and such.
3013 * We wait on commit, if tol to.
3015 * - Within prune_icache() (PF_MEMALLOC == true)
3016 * Here we simply return. We can't afford to block kswapd on the
3019 * In all cases it is actually safe for us to return without doing anything,
3020 * because the inode has been copied into a raw inode buffer in
3021 * ext4_mark_inode_dirty(). This is a correctness thing for O_SYNC and for
3024 * Note that we are absolutely dependent upon all inode dirtiers doing the
3025 * right thing: they *must* call mark_inode_dirty() after dirtying info in
3026 * which we are interested.
3028 * It would be a bug for them to not do this. The code:
3030 * mark_inode_dirty(inode)
3032 * inode->i_size = expr;
3034 * is in error because a kswapd-driven write_inode() could occur while
3035 * `stuff()' is running, and the new i_size will be lost. Plus the inode
3036 * will no longer be on the superblock's dirty inode list.
3038 int ext4_write_inode(struct inode
*inode
, int wait
)
3040 if (current
->flags
& PF_MEMALLOC
)
3043 if (ext4_journal_current_handle()) {
3044 jbd_debug(1, "called recursively, non-PF_MEMALLOC!\n");
3052 return ext4_force_commit(inode
->i_sb
);
3058 * Called from notify_change.
3060 * We want to trap VFS attempts to truncate the file as soon as
3061 * possible. In particular, we want to make sure that when the VFS
3062 * shrinks i_size, we put the inode on the orphan list and modify
3063 * i_disksize immediately, so that during the subsequent flushing of
3064 * dirty pages and freeing of disk blocks, we can guarantee that any
3065 * commit will leave the blocks being flushed in an unused state on
3066 * disk. (On recovery, the inode will get truncated and the blocks will
3067 * be freed, so we have a strong guarantee that no future commit will
3068 * leave these blocks visible to the user.)
3070 * Called with inode->sem down.
3072 int ext4_setattr(struct dentry
*dentry
, struct iattr
*attr
)
3074 struct inode
*inode
= dentry
->d_inode
;
3076 const unsigned int ia_valid
= attr
->ia_valid
;
3078 error
= inode_change_ok(inode
, attr
);
3082 if ((ia_valid
& ATTR_UID
&& attr
->ia_uid
!= inode
->i_uid
) ||
3083 (ia_valid
& ATTR_GID
&& attr
->ia_gid
!= inode
->i_gid
)) {
3086 /* (user+group)*(old+new) structure, inode write (sb,
3087 * inode block, ? - but truncate inode update has it) */
3088 handle
= ext4_journal_start(inode
, 2*(EXT4_QUOTA_INIT_BLOCKS(inode
->i_sb
)+
3089 EXT4_QUOTA_DEL_BLOCKS(inode
->i_sb
))+3);
3090 if (IS_ERR(handle
)) {
3091 error
= PTR_ERR(handle
);
3094 error
= DQUOT_TRANSFER(inode
, attr
) ? -EDQUOT
: 0;
3096 ext4_journal_stop(handle
);
3099 /* Update corresponding info in inode so that everything is in
3100 * one transaction */
3101 if (attr
->ia_valid
& ATTR_UID
)
3102 inode
->i_uid
= attr
->ia_uid
;
3103 if (attr
->ia_valid
& ATTR_GID
)
3104 inode
->i_gid
= attr
->ia_gid
;
3105 error
= ext4_mark_inode_dirty(handle
, inode
);
3106 ext4_journal_stop(handle
);
3109 if (attr
->ia_valid
& ATTR_SIZE
) {
3110 if (!(EXT4_I(inode
)->i_flags
& EXT4_EXTENTS_FL
)) {
3111 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
3113 if (attr
->ia_size
> sbi
->s_bitmap_maxbytes
) {
3120 if (S_ISREG(inode
->i_mode
) &&
3121 attr
->ia_valid
& ATTR_SIZE
&& attr
->ia_size
< inode
->i_size
) {
3124 handle
= ext4_journal_start(inode
, 3);
3125 if (IS_ERR(handle
)) {
3126 error
= PTR_ERR(handle
);
3130 error
= ext4_orphan_add(handle
, inode
);
3131 EXT4_I(inode
)->i_disksize
= attr
->ia_size
;
3132 rc
= ext4_mark_inode_dirty(handle
, inode
);
3135 ext4_journal_stop(handle
);
3138 rc
= inode_setattr(inode
, attr
);
3140 /* If inode_setattr's call to ext4_truncate failed to get a
3141 * transaction handle at all, we need to clean up the in-core
3142 * orphan list manually. */
3144 ext4_orphan_del(NULL
, inode
);
3146 if (!rc
&& (ia_valid
& ATTR_MODE
))
3147 rc
= ext4_acl_chmod(inode
);
3150 ext4_std_error(inode
->i_sb
, error
);
3158 * How many blocks doth make a writepage()?
3160 * With N blocks per page, it may be:
3165 * N+5 bitmap blocks (from the above)
3166 * N+5 group descriptor summary blocks
3169 * 2 * EXT4_SINGLEDATA_TRANS_BLOCKS for the quote files
3171 * 3 * (N + 5) + 2 + 2 * EXT4_SINGLEDATA_TRANS_BLOCKS
3173 * With ordered or writeback data it's the same, less the N data blocks.
3175 * If the inode's direct blocks can hold an integral number of pages then a
3176 * page cannot straddle two indirect blocks, and we can only touch one indirect
3177 * and dindirect block, and the "5" above becomes "3".
3179 * This still overestimates under most circumstances. If we were to pass the
3180 * start and end offsets in here as well we could do block_to_path() on each
3181 * block and work out the exact number of indirects which are touched. Pah.
3184 int ext4_writepage_trans_blocks(struct inode
*inode
)
3186 int bpp
= ext4_journal_blocks_per_page(inode
);
3187 int indirects
= (EXT4_NDIR_BLOCKS
% bpp
) ? 5 : 3;
3190 if (EXT4_I(inode
)->i_flags
& EXT4_EXTENTS_FL
)
3191 return ext4_ext_writepage_trans_blocks(inode
, bpp
);
3193 if (ext4_should_journal_data(inode
))
3194 ret
= 3 * (bpp
+ indirects
) + 2;
3196 ret
= 2 * (bpp
+ indirects
) + 2;
3199 /* We know that structure was already allocated during DQUOT_INIT so
3200 * we will be updating only the data blocks + inodes */
3201 ret
+= 2*EXT4_QUOTA_TRANS_BLOCKS(inode
->i_sb
);
3208 * The caller must have previously called ext4_reserve_inode_write().
3209 * Give this, we know that the caller already has write access to iloc->bh.
3211 int ext4_mark_iloc_dirty(handle_t
*handle
,
3212 struct inode
*inode
, struct ext4_iloc
*iloc
)
3216 if (test_opt(inode
->i_sb
, I_VERSION
))
3217 inode_inc_iversion(inode
);
3219 /* the do_update_inode consumes one bh->b_count */
3222 /* ext4_do_update_inode() does jbd2_journal_dirty_metadata */
3223 err
= ext4_do_update_inode(handle
, inode
, iloc
);
3229 * On success, We end up with an outstanding reference count against
3230 * iloc->bh. This _must_ be cleaned up later.
3234 ext4_reserve_inode_write(handle_t
*handle
, struct inode
*inode
,
3235 struct ext4_iloc
*iloc
)
3239 err
= ext4_get_inode_loc(inode
, iloc
);
3241 BUFFER_TRACE(iloc
->bh
, "get_write_access");
3242 err
= ext4_journal_get_write_access(handle
, iloc
->bh
);
3249 ext4_std_error(inode
->i_sb
, err
);
3254 * Expand an inode by new_extra_isize bytes.
3255 * Returns 0 on success or negative error number on failure.
3257 static int ext4_expand_extra_isize(struct inode
*inode
,
3258 unsigned int new_extra_isize
,
3259 struct ext4_iloc iloc
,
3262 struct ext4_inode
*raw_inode
;
3263 struct ext4_xattr_ibody_header
*header
;
3264 struct ext4_xattr_entry
*entry
;
3266 if (EXT4_I(inode
)->i_extra_isize
>= new_extra_isize
)
3269 raw_inode
= ext4_raw_inode(&iloc
);
3271 header
= IHDR(inode
, raw_inode
);
3272 entry
= IFIRST(header
);
3274 /* No extended attributes present */
3275 if (!(EXT4_I(inode
)->i_state
& EXT4_STATE_XATTR
) ||
3276 header
->h_magic
!= cpu_to_le32(EXT4_XATTR_MAGIC
)) {
3277 memset((void *)raw_inode
+ EXT4_GOOD_OLD_INODE_SIZE
, 0,
3279 EXT4_I(inode
)->i_extra_isize
= new_extra_isize
;
3283 /* try to expand with EAs present */
3284 return ext4_expand_extra_isize_ea(inode
, new_extra_isize
,
3289 * What we do here is to mark the in-core inode as clean with respect to inode
3290 * dirtiness (it may still be data-dirty).
3291 * This means that the in-core inode may be reaped by prune_icache
3292 * without having to perform any I/O. This is a very good thing,
3293 * because *any* task may call prune_icache - even ones which
3294 * have a transaction open against a different journal.
3296 * Is this cheating? Not really. Sure, we haven't written the
3297 * inode out, but prune_icache isn't a user-visible syncing function.
3298 * Whenever the user wants stuff synced (sys_sync, sys_msync, sys_fsync)
3299 * we start and wait on commits.
3301 * Is this efficient/effective? Well, we're being nice to the system
3302 * by cleaning up our inodes proactively so they can be reaped
3303 * without I/O. But we are potentially leaving up to five seconds'
3304 * worth of inodes floating about which prune_icache wants us to
3305 * write out. One way to fix that would be to get prune_icache()
3306 * to do a write_super() to free up some memory. It has the desired
3309 int ext4_mark_inode_dirty(handle_t
*handle
, struct inode
*inode
)
3311 struct ext4_iloc iloc
;
3312 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
3313 static unsigned int mnt_count
;
3317 err
= ext4_reserve_inode_write(handle
, inode
, &iloc
);
3318 if (EXT4_I(inode
)->i_extra_isize
< sbi
->s_want_extra_isize
&&
3319 !(EXT4_I(inode
)->i_state
& EXT4_STATE_NO_EXPAND
)) {
3321 * We need extra buffer credits since we may write into EA block
3322 * with this same handle. If journal_extend fails, then it will
3323 * only result in a minor loss of functionality for that inode.
3324 * If this is felt to be critical, then e2fsck should be run to
3325 * force a large enough s_min_extra_isize.
3327 if ((jbd2_journal_extend(handle
,
3328 EXT4_DATA_TRANS_BLOCKS(inode
->i_sb
))) == 0) {
3329 ret
= ext4_expand_extra_isize(inode
,
3330 sbi
->s_want_extra_isize
,
3333 EXT4_I(inode
)->i_state
|= EXT4_STATE_NO_EXPAND
;
3335 le16_to_cpu(sbi
->s_es
->s_mnt_count
)) {
3336 ext4_warning(inode
->i_sb
, __FUNCTION__
,
3337 "Unable to expand inode %lu. Delete"
3338 " some EAs or run e2fsck.",
3341 le16_to_cpu(sbi
->s_es
->s_mnt_count
);
3347 err
= ext4_mark_iloc_dirty(handle
, inode
, &iloc
);
3352 * ext4_dirty_inode() is called from __mark_inode_dirty()
3354 * We're really interested in the case where a file is being extended.
3355 * i_size has been changed by generic_commit_write() and we thus need
3356 * to include the updated inode in the current transaction.
3358 * Also, DQUOT_ALLOC_SPACE() will always dirty the inode when blocks
3359 * are allocated to the file.
3361 * If the inode is marked synchronous, we don't honour that here - doing
3362 * so would cause a commit on atime updates, which we don't bother doing.
3363 * We handle synchronous inodes at the highest possible level.
3365 void ext4_dirty_inode(struct inode
*inode
)
3367 handle_t
*current_handle
= ext4_journal_current_handle();
3370 handle
= ext4_journal_start(inode
, 2);
3373 if (current_handle
&&
3374 current_handle
->h_transaction
!= handle
->h_transaction
) {
3375 /* This task has a transaction open against a different fs */
3376 printk(KERN_EMERG
"%s: transactions do not match!\n",
3379 jbd_debug(5, "marking dirty. outer handle=%p\n",
3381 ext4_mark_inode_dirty(handle
, inode
);
3383 ext4_journal_stop(handle
);
3390 * Bind an inode's backing buffer_head into this transaction, to prevent
3391 * it from being flushed to disk early. Unlike
3392 * ext4_reserve_inode_write, this leaves behind no bh reference and
3393 * returns no iloc structure, so the caller needs to repeat the iloc
3394 * lookup to mark the inode dirty later.
3396 static int ext4_pin_inode(handle_t
*handle
, struct inode
*inode
)
3398 struct ext4_iloc iloc
;
3402 err
= ext4_get_inode_loc(inode
, &iloc
);
3404 BUFFER_TRACE(iloc
.bh
, "get_write_access");
3405 err
= jbd2_journal_get_write_access(handle
, iloc
.bh
);
3407 err
= ext4_journal_dirty_metadata(handle
,
3412 ext4_std_error(inode
->i_sb
, err
);
3417 int ext4_change_inode_journal_flag(struct inode
*inode
, int val
)
3424 * We have to be very careful here: changing a data block's
3425 * journaling status dynamically is dangerous. If we write a
3426 * data block to the journal, change the status and then delete
3427 * that block, we risk forgetting to revoke the old log record
3428 * from the journal and so a subsequent replay can corrupt data.
3429 * So, first we make sure that the journal is empty and that
3430 * nobody is changing anything.
3433 journal
= EXT4_JOURNAL(inode
);
3434 if (is_journal_aborted(journal
))
3437 jbd2_journal_lock_updates(journal
);
3438 jbd2_journal_flush(journal
);
3441 * OK, there are no updates running now, and all cached data is
3442 * synced to disk. We are now in a completely consistent state
3443 * which doesn't have anything in the journal, and we know that
3444 * no filesystem updates are running, so it is safe to modify
3445 * the inode's in-core data-journaling state flag now.
3449 EXT4_I(inode
)->i_flags
|= EXT4_JOURNAL_DATA_FL
;
3451 EXT4_I(inode
)->i_flags
&= ~EXT4_JOURNAL_DATA_FL
;
3452 ext4_set_aops(inode
);
3454 jbd2_journal_unlock_updates(journal
);
3456 /* Finally we can mark the inode as dirty. */
3458 handle
= ext4_journal_start(inode
, 1);
3460 return PTR_ERR(handle
);
3462 err
= ext4_mark_inode_dirty(handle
, inode
);
3464 ext4_journal_stop(handle
);
3465 ext4_std_error(inode
->i_sb
, err
);