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 * mutex_lock(&EXT4_I(inode)->truncate_mutex)
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_grpblk_t colour
;
408 /* Try to find previous block */
409 for (p
= ind
->p
- 1; p
>= start
; p
--) {
411 return le32_to_cpu(*p
);
414 /* No such thing, so let's try location of indirect block */
416 return ind
->bh
->b_blocknr
;
419 * It is going to be referred to from the inode itself? OK, just put it
420 * into the same cylinder group then.
422 bg_start
= ext4_group_first_block_no(inode
->i_sb
, ei
->i_block_group
);
423 colour
= (current
->pid
% 16) *
424 (EXT4_BLOCKS_PER_GROUP(inode
->i_sb
) / 16);
425 return bg_start
+ colour
;
429 * ext4_find_goal - find a prefered place for allocation.
431 * @block: block we want
432 * @chain: chain of indirect blocks
433 * @partial: pointer to the last triple within a chain
434 * @goal: place to store the result.
436 * Normally this function find the prefered place for block allocation,
437 * stores it in *@goal and returns zero.
440 static ext4_fsblk_t
ext4_find_goal(struct inode
*inode
, ext4_lblk_t block
,
441 Indirect chain
[4], Indirect
*partial
)
443 struct ext4_block_alloc_info
*block_i
;
445 block_i
= EXT4_I(inode
)->i_block_alloc_info
;
448 * try the heuristic for sequential allocation,
449 * failing that at least try to get decent locality.
451 if (block_i
&& (block
== block_i
->last_alloc_logical_block
+ 1)
452 && (block_i
->last_alloc_physical_block
!= 0)) {
453 return block_i
->last_alloc_physical_block
+ 1;
456 return ext4_find_near(inode
, partial
);
460 * ext4_blks_to_allocate: Look up the block map and count the number
461 * of direct blocks need to be allocated for the given branch.
463 * @branch: chain of indirect blocks
464 * @k: number of blocks need for indirect blocks
465 * @blks: number of data blocks to be mapped.
466 * @blocks_to_boundary: the offset in the indirect block
468 * return the total number of blocks to be allocate, including the
469 * direct and indirect blocks.
471 static int ext4_blks_to_allocate(Indirect
*branch
, int k
, unsigned long blks
,
472 int blocks_to_boundary
)
474 unsigned long count
= 0;
477 * Simple case, [t,d]Indirect block(s) has not allocated yet
478 * then it's clear blocks on that path have not allocated
481 /* right now we don't handle cross boundary allocation */
482 if (blks
< blocks_to_boundary
+ 1)
485 count
+= blocks_to_boundary
+ 1;
490 while (count
< blks
&& count
<= blocks_to_boundary
&&
491 le32_to_cpu(*(branch
[0].p
+ count
)) == 0) {
498 * ext4_alloc_blocks: multiple allocate blocks needed for a branch
499 * @indirect_blks: the number of blocks need to allocate for indirect
502 * @new_blocks: on return it will store the new block numbers for
503 * the indirect blocks(if needed) and the first direct block,
504 * @blks: on return it will store the total number of allocated
507 static int ext4_alloc_blocks(handle_t
*handle
, struct inode
*inode
,
508 ext4_fsblk_t goal
, int indirect_blks
, int blks
,
509 ext4_fsblk_t new_blocks
[4], int *err
)
512 unsigned long count
= 0;
514 ext4_fsblk_t current_block
= 0;
518 * Here we try to allocate the requested multiple blocks at once,
519 * on a best-effort basis.
520 * To build a branch, we should allocate blocks for
521 * the indirect blocks(if not allocated yet), and at least
522 * the first direct block of this branch. That's the
523 * minimum number of blocks need to allocate(required)
525 target
= blks
+ indirect_blks
;
529 /* allocating blocks for indirect blocks and direct blocks */
530 current_block
= ext4_new_blocks(handle
,inode
,goal
,&count
,err
);
535 /* allocate blocks for indirect blocks */
536 while (index
< indirect_blks
&& count
) {
537 new_blocks
[index
++] = current_block
++;
545 /* save the new block number for the first direct block */
546 new_blocks
[index
] = current_block
;
548 /* total number of blocks allocated for direct blocks */
553 for (i
= 0; i
<index
; i
++)
554 ext4_free_blocks(handle
, inode
, new_blocks
[i
], 1);
559 * ext4_alloc_branch - allocate and set up a chain of blocks.
561 * @indirect_blks: number of allocated indirect blocks
562 * @blks: number of allocated direct blocks
563 * @offsets: offsets (in the blocks) to store the pointers to next.
564 * @branch: place to store the chain in.
566 * This function allocates blocks, zeroes out all but the last one,
567 * links them into chain and (if we are synchronous) writes them to disk.
568 * In other words, it prepares a branch that can be spliced onto the
569 * inode. It stores the information about that chain in the branch[], in
570 * the same format as ext4_get_branch() would do. We are calling it after
571 * we had read the existing part of chain and partial points to the last
572 * triple of that (one with zero ->key). Upon the exit we have the same
573 * picture as after the successful ext4_get_block(), except that in one
574 * place chain is disconnected - *branch->p is still zero (we did not
575 * set the last link), but branch->key contains the number that should
576 * be placed into *branch->p to fill that gap.
578 * If allocation fails we free all blocks we've allocated (and forget
579 * their buffer_heads) and return the error value the from failed
580 * ext4_alloc_block() (normally -ENOSPC). Otherwise we set the chain
581 * as described above and return 0.
583 static int ext4_alloc_branch(handle_t
*handle
, struct inode
*inode
,
584 int indirect_blks
, int *blks
, ext4_fsblk_t goal
,
585 ext4_lblk_t
*offsets
, Indirect
*branch
)
587 int blocksize
= inode
->i_sb
->s_blocksize
;
590 struct buffer_head
*bh
;
592 ext4_fsblk_t new_blocks
[4];
593 ext4_fsblk_t current_block
;
595 num
= ext4_alloc_blocks(handle
, inode
, goal
, indirect_blks
,
596 *blks
, new_blocks
, &err
);
600 branch
[0].key
= cpu_to_le32(new_blocks
[0]);
602 * metadata blocks and data blocks are allocated.
604 for (n
= 1; n
<= indirect_blks
; n
++) {
606 * Get buffer_head for parent block, zero it out
607 * and set the pointer to new one, then send
610 bh
= sb_getblk(inode
->i_sb
, new_blocks
[n
-1]);
613 BUFFER_TRACE(bh
, "call get_create_access");
614 err
= ext4_journal_get_create_access(handle
, bh
);
621 memset(bh
->b_data
, 0, blocksize
);
622 branch
[n
].p
= (__le32
*) bh
->b_data
+ offsets
[n
];
623 branch
[n
].key
= cpu_to_le32(new_blocks
[n
]);
624 *branch
[n
].p
= branch
[n
].key
;
625 if ( n
== indirect_blks
) {
626 current_block
= new_blocks
[n
];
628 * End of chain, update the last new metablock of
629 * the chain to point to the new allocated
630 * data blocks numbers
632 for (i
=1; i
< num
; i
++)
633 *(branch
[n
].p
+ i
) = cpu_to_le32(++current_block
);
635 BUFFER_TRACE(bh
, "marking uptodate");
636 set_buffer_uptodate(bh
);
639 BUFFER_TRACE(bh
, "call ext4_journal_dirty_metadata");
640 err
= ext4_journal_dirty_metadata(handle
, bh
);
647 /* Allocation failed, free what we already allocated */
648 for (i
= 1; i
<= n
; i
++) {
649 BUFFER_TRACE(branch
[i
].bh
, "call jbd2_journal_forget");
650 ext4_journal_forget(handle
, branch
[i
].bh
);
652 for (i
= 0; i
<indirect_blks
; i
++)
653 ext4_free_blocks(handle
, inode
, new_blocks
[i
], 1);
655 ext4_free_blocks(handle
, inode
, new_blocks
[i
], num
);
661 * ext4_splice_branch - splice the allocated branch onto inode.
663 * @block: (logical) number of block we are adding
664 * @chain: chain of indirect blocks (with a missing link - see
666 * @where: location of missing link
667 * @num: number of indirect blocks we are adding
668 * @blks: number of direct blocks we are adding
670 * This function fills the missing link and does all housekeeping needed in
671 * inode (->i_blocks, etc.). In case of success we end up with the full
672 * chain to new block and return 0.
674 static int ext4_splice_branch(handle_t
*handle
, struct inode
*inode
,
675 ext4_lblk_t block
, Indirect
*where
, int num
, int blks
)
679 struct ext4_block_alloc_info
*block_i
;
680 ext4_fsblk_t current_block
;
682 block_i
= EXT4_I(inode
)->i_block_alloc_info
;
684 * If we're splicing into a [td]indirect block (as opposed to the
685 * inode) then we need to get write access to the [td]indirect block
689 BUFFER_TRACE(where
->bh
, "get_write_access");
690 err
= ext4_journal_get_write_access(handle
, where
->bh
);
696 *where
->p
= where
->key
;
699 * Update the host buffer_head or inode to point to more just allocated
700 * direct blocks blocks
702 if (num
== 0 && blks
> 1) {
703 current_block
= le32_to_cpu(where
->key
) + 1;
704 for (i
= 1; i
< blks
; i
++)
705 *(where
->p
+ i
) = cpu_to_le32(current_block
++);
709 * update the most recently allocated logical & physical block
710 * in i_block_alloc_info, to assist find the proper goal block for next
714 block_i
->last_alloc_logical_block
= block
+ blks
- 1;
715 block_i
->last_alloc_physical_block
=
716 le32_to_cpu(where
[num
].key
) + blks
- 1;
719 /* We are done with atomic stuff, now do the rest of housekeeping */
721 inode
->i_ctime
= ext4_current_time(inode
);
722 ext4_mark_inode_dirty(handle
, inode
);
724 /* had we spliced it onto indirect block? */
727 * If we spliced it onto an indirect block, we haven't
728 * altered the inode. Note however that if it is being spliced
729 * onto an indirect block at the very end of the file (the
730 * file is growing) then we *will* alter the inode to reflect
731 * the new i_size. But that is not done here - it is done in
732 * generic_commit_write->__mark_inode_dirty->ext4_dirty_inode.
734 jbd_debug(5, "splicing indirect only\n");
735 BUFFER_TRACE(where
->bh
, "call ext4_journal_dirty_metadata");
736 err
= ext4_journal_dirty_metadata(handle
, where
->bh
);
741 * OK, we spliced it into the inode itself on a direct block.
742 * Inode was dirtied above.
744 jbd_debug(5, "splicing direct\n");
749 for (i
= 1; i
<= num
; i
++) {
750 BUFFER_TRACE(where
[i
].bh
, "call jbd2_journal_forget");
751 ext4_journal_forget(handle
, where
[i
].bh
);
752 ext4_free_blocks(handle
,inode
,le32_to_cpu(where
[i
-1].key
),1);
754 ext4_free_blocks(handle
, inode
, le32_to_cpu(where
[num
].key
), blks
);
760 * Allocation strategy is simple: if we have to allocate something, we will
761 * have to go the whole way to leaf. So let's do it before attaching anything
762 * to tree, set linkage between the newborn blocks, write them if sync is
763 * required, recheck the path, free and repeat if check fails, otherwise
764 * set the last missing link (that will protect us from any truncate-generated
765 * removals - all blocks on the path are immune now) and possibly force the
766 * write on the parent block.
767 * That has a nice additional property: no special recovery from the failed
768 * allocations is needed - we simply release blocks and do not touch anything
769 * reachable from inode.
771 * `handle' can be NULL if create == 0.
773 * The BKL may not be held on entry here. Be sure to take it early.
774 * return > 0, # of blocks mapped or allocated.
775 * return = 0, if plain lookup failed.
776 * return < 0, error case.
779 * Need to be called with
780 * mutex_lock(&EXT4_I(inode)->truncate_mutex)
782 int ext4_get_blocks_handle(handle_t
*handle
, struct inode
*inode
,
783 ext4_lblk_t iblock
, unsigned long maxblocks
,
784 struct buffer_head
*bh_result
,
785 int create
, int extend_disksize
)
788 ext4_lblk_t offsets
[4];
793 int blocks_to_boundary
= 0;
795 struct ext4_inode_info
*ei
= EXT4_I(inode
);
797 ext4_fsblk_t first_block
= 0;
800 J_ASSERT(!(EXT4_I(inode
)->i_flags
& EXT4_EXTENTS_FL
));
801 J_ASSERT(handle
!= NULL
|| create
== 0);
802 depth
= ext4_block_to_path(inode
, iblock
, offsets
,
803 &blocks_to_boundary
);
808 partial
= ext4_get_branch(inode
, depth
, offsets
, chain
, &err
);
810 /* Simplest case - block found, no allocation needed */
812 first_block
= le32_to_cpu(chain
[depth
- 1].key
);
813 clear_buffer_new(bh_result
);
816 while (count
< maxblocks
&& count
<= blocks_to_boundary
) {
819 blk
= le32_to_cpu(*(chain
[depth
-1].p
+ count
));
821 if (blk
== first_block
+ count
)
829 /* Next simple case - plain lookup or failed read of indirect block */
830 if (!create
|| err
== -EIO
)
834 * Okay, we need to do block allocation. Lazily initialize the block
835 * allocation info here if necessary
837 if (S_ISREG(inode
->i_mode
) && (!ei
->i_block_alloc_info
))
838 ext4_init_block_alloc_info(inode
);
840 goal
= ext4_find_goal(inode
, iblock
, chain
, partial
);
842 /* the number of blocks need to allocate for [d,t]indirect blocks */
843 indirect_blks
= (chain
+ depth
) - partial
- 1;
846 * Next look up the indirect map to count the totoal number of
847 * direct blocks to allocate for this branch.
849 count
= ext4_blks_to_allocate(partial
, indirect_blks
,
850 maxblocks
, blocks_to_boundary
);
852 * Block out ext4_truncate while we alter the tree
854 err
= ext4_alloc_branch(handle
, inode
, indirect_blks
, &count
, goal
,
855 offsets
+ (partial
- chain
), partial
);
858 * The ext4_splice_branch call will free and forget any buffers
859 * on the new chain if there is a failure, but that risks using
860 * up transaction credits, especially for bitmaps where the
861 * credits cannot be returned. Can we handle this somehow? We
862 * may need to return -EAGAIN upwards in the worst case. --sct
865 err
= ext4_splice_branch(handle
, inode
, iblock
,
866 partial
, indirect_blks
, count
);
868 * i_disksize growing is protected by truncate_mutex. Don't forget to
869 * protect it if you're about to implement concurrent
870 * ext4_get_block() -bzzz
872 if (!err
&& extend_disksize
&& inode
->i_size
> ei
->i_disksize
)
873 ei
->i_disksize
= inode
->i_size
;
877 set_buffer_new(bh_result
);
879 map_bh(bh_result
, inode
->i_sb
, le32_to_cpu(chain
[depth
-1].key
));
880 if (count
> blocks_to_boundary
)
881 set_buffer_boundary(bh_result
);
883 /* Clean up and exit */
884 partial
= chain
+ depth
- 1; /* the whole chain */
886 while (partial
> chain
) {
887 BUFFER_TRACE(partial
->bh
, "call brelse");
891 BUFFER_TRACE(bh_result
, "returned");
896 #define DIO_CREDITS (EXT4_RESERVE_TRANS_BLOCKS + 32)
898 static int ext4_get_block(struct inode
*inode
, sector_t iblock
,
899 struct buffer_head
*bh_result
, int create
)
901 handle_t
*handle
= ext4_journal_current_handle();
903 unsigned max_blocks
= bh_result
->b_size
>> inode
->i_blkbits
;
906 goto get_block
; /* A read */
909 goto get_block
; /* A single block get */
911 if (handle
->h_transaction
->t_state
== T_LOCKED
) {
913 * Huge direct-io writes can hold off commits for long
914 * periods of time. Let this commit run.
916 ext4_journal_stop(handle
);
917 handle
= ext4_journal_start(inode
, DIO_CREDITS
);
919 ret
= PTR_ERR(handle
);
923 if (handle
->h_buffer_credits
<= EXT4_RESERVE_TRANS_BLOCKS
) {
925 * Getting low on buffer credits...
927 ret
= ext4_journal_extend(handle
, DIO_CREDITS
);
930 * Couldn't extend the transaction. Start a new one.
932 ret
= ext4_journal_restart(handle
, DIO_CREDITS
);
938 ret
= ext4_get_blocks_wrap(handle
, inode
, iblock
,
939 max_blocks
, bh_result
, create
, 0);
941 bh_result
->b_size
= (ret
<< inode
->i_blkbits
);
949 * `handle' can be NULL if create is zero
951 struct buffer_head
*ext4_getblk(handle_t
*handle
, struct inode
*inode
,
952 ext4_lblk_t block
, int create
, int *errp
)
954 struct buffer_head dummy
;
957 J_ASSERT(handle
!= NULL
|| create
== 0);
960 dummy
.b_blocknr
= -1000;
961 buffer_trace_init(&dummy
.b_history
);
962 err
= ext4_get_blocks_wrap(handle
, inode
, block
, 1,
965 * ext4_get_blocks_handle() returns number of blocks
966 * mapped. 0 in case of a HOLE.
974 if (!err
&& buffer_mapped(&dummy
)) {
975 struct buffer_head
*bh
;
976 bh
= sb_getblk(inode
->i_sb
, dummy
.b_blocknr
);
981 if (buffer_new(&dummy
)) {
982 J_ASSERT(create
!= 0);
983 J_ASSERT(handle
!= NULL
);
986 * Now that we do not always journal data, we should
987 * keep in mind whether this should always journal the
988 * new buffer as metadata. For now, regular file
989 * writes use ext4_get_block instead, so it's not a
993 BUFFER_TRACE(bh
, "call get_create_access");
994 fatal
= ext4_journal_get_create_access(handle
, bh
);
995 if (!fatal
&& !buffer_uptodate(bh
)) {
996 memset(bh
->b_data
,0,inode
->i_sb
->s_blocksize
);
997 set_buffer_uptodate(bh
);
1000 BUFFER_TRACE(bh
, "call ext4_journal_dirty_metadata");
1001 err
= ext4_journal_dirty_metadata(handle
, bh
);
1005 BUFFER_TRACE(bh
, "not a new buffer");
1018 struct buffer_head
*ext4_bread(handle_t
*handle
, struct inode
*inode
,
1019 ext4_lblk_t block
, int create
, int *err
)
1021 struct buffer_head
* bh
;
1023 bh
= ext4_getblk(handle
, inode
, block
, create
, err
);
1026 if (buffer_uptodate(bh
))
1028 ll_rw_block(READ_META
, 1, &bh
);
1030 if (buffer_uptodate(bh
))
1037 static int walk_page_buffers( handle_t
*handle
,
1038 struct buffer_head
*head
,
1042 int (*fn
)( handle_t
*handle
,
1043 struct buffer_head
*bh
))
1045 struct buffer_head
*bh
;
1046 unsigned block_start
, block_end
;
1047 unsigned blocksize
= head
->b_size
;
1049 struct buffer_head
*next
;
1051 for ( bh
= head
, block_start
= 0;
1052 ret
== 0 && (bh
!= head
|| !block_start
);
1053 block_start
= block_end
, bh
= next
)
1055 next
= bh
->b_this_page
;
1056 block_end
= block_start
+ blocksize
;
1057 if (block_end
<= from
|| block_start
>= to
) {
1058 if (partial
&& !buffer_uptodate(bh
))
1062 err
= (*fn
)(handle
, bh
);
1070 * To preserve ordering, it is essential that the hole instantiation and
1071 * the data write be encapsulated in a single transaction. We cannot
1072 * close off a transaction and start a new one between the ext4_get_block()
1073 * and the commit_write(). So doing the jbd2_journal_start at the start of
1074 * prepare_write() is the right place.
1076 * Also, this function can nest inside ext4_writepage() ->
1077 * block_write_full_page(). In that case, we *know* that ext4_writepage()
1078 * has generated enough buffer credits to do the whole page. So we won't
1079 * block on the journal in that case, which is good, because the caller may
1082 * By accident, ext4 can be reentered when a transaction is open via
1083 * quota file writes. If we were to commit the transaction while thus
1084 * reentered, there can be a deadlock - we would be holding a quota
1085 * lock, and the commit would never complete if another thread had a
1086 * transaction open and was blocking on the quota lock - a ranking
1089 * So what we do is to rely on the fact that jbd2_journal_stop/journal_start
1090 * will _not_ run commit under these circumstances because handle->h_ref
1091 * is elevated. We'll still have enough credits for the tiny quotafile
1094 static int do_journal_get_write_access(handle_t
*handle
,
1095 struct buffer_head
*bh
)
1097 if (!buffer_mapped(bh
) || buffer_freed(bh
))
1099 return ext4_journal_get_write_access(handle
, bh
);
1102 static int ext4_write_begin(struct file
*file
, struct address_space
*mapping
,
1103 loff_t pos
, unsigned len
, unsigned flags
,
1104 struct page
**pagep
, void **fsdata
)
1106 struct inode
*inode
= mapping
->host
;
1107 int ret
, needed_blocks
= ext4_writepage_trans_blocks(inode
);
1114 index
= pos
>> PAGE_CACHE_SHIFT
;
1115 from
= pos
& (PAGE_CACHE_SIZE
- 1);
1119 page
= __grab_cache_page(mapping
, index
);
1124 handle
= ext4_journal_start(inode
, needed_blocks
);
1125 if (IS_ERR(handle
)) {
1127 page_cache_release(page
);
1128 ret
= PTR_ERR(handle
);
1132 ret
= block_write_begin(file
, mapping
, pos
, len
, flags
, pagep
, fsdata
,
1135 if (!ret
&& ext4_should_journal_data(inode
)) {
1136 ret
= walk_page_buffers(handle
, page_buffers(page
),
1137 from
, to
, NULL
, do_journal_get_write_access
);
1141 ext4_journal_stop(handle
);
1143 page_cache_release(page
);
1146 if (ret
== -ENOSPC
&& ext4_should_retry_alloc(inode
->i_sb
, &retries
))
1152 int ext4_journal_dirty_data(handle_t
*handle
, struct buffer_head
*bh
)
1154 int err
= jbd2_journal_dirty_data(handle
, bh
);
1156 ext4_journal_abort_handle(__FUNCTION__
, __FUNCTION__
,
1161 /* For write_end() in data=journal mode */
1162 static int write_end_fn(handle_t
*handle
, struct buffer_head
*bh
)
1164 if (!buffer_mapped(bh
) || buffer_freed(bh
))
1166 set_buffer_uptodate(bh
);
1167 return ext4_journal_dirty_metadata(handle
, bh
);
1171 * Generic write_end handler for ordered and writeback ext4 journal modes.
1172 * We can't use generic_write_end, because that unlocks the page and we need to
1173 * unlock the page after ext4_journal_stop, but ext4_journal_stop must run
1174 * after block_write_end.
1176 static int ext4_generic_write_end(struct file
*file
,
1177 struct address_space
*mapping
,
1178 loff_t pos
, unsigned len
, unsigned copied
,
1179 struct page
*page
, void *fsdata
)
1181 struct inode
*inode
= file
->f_mapping
->host
;
1183 copied
= block_write_end(file
, mapping
, pos
, len
, copied
, page
, fsdata
);
1185 if (pos
+copied
> inode
->i_size
) {
1186 i_size_write(inode
, pos
+copied
);
1187 mark_inode_dirty(inode
);
1194 * We need to pick up the new inode size which generic_commit_write gave us
1195 * `file' can be NULL - eg, when called from page_symlink().
1197 * ext4 never places buffers on inode->i_mapping->private_list. metadata
1198 * buffers are managed internally.
1200 static int ext4_ordered_write_end(struct file
*file
,
1201 struct address_space
*mapping
,
1202 loff_t pos
, unsigned len
, unsigned copied
,
1203 struct page
*page
, void *fsdata
)
1205 handle_t
*handle
= ext4_journal_current_handle();
1206 struct inode
*inode
= file
->f_mapping
->host
;
1210 from
= pos
& (PAGE_CACHE_SIZE
- 1);
1213 ret
= walk_page_buffers(handle
, page_buffers(page
),
1214 from
, to
, NULL
, ext4_journal_dirty_data
);
1218 * generic_write_end() will run mark_inode_dirty() if i_size
1219 * changes. So let's piggyback the i_disksize mark_inode_dirty
1224 new_i_size
= pos
+ copied
;
1225 if (new_i_size
> EXT4_I(inode
)->i_disksize
)
1226 EXT4_I(inode
)->i_disksize
= new_i_size
;
1227 copied
= ext4_generic_write_end(file
, mapping
, pos
, len
, copied
,
1232 ret2
= ext4_journal_stop(handle
);
1236 page_cache_release(page
);
1238 return ret
? ret
: copied
;
1241 static int ext4_writeback_write_end(struct file
*file
,
1242 struct address_space
*mapping
,
1243 loff_t pos
, unsigned len
, unsigned copied
,
1244 struct page
*page
, void *fsdata
)
1246 handle_t
*handle
= ext4_journal_current_handle();
1247 struct inode
*inode
= file
->f_mapping
->host
;
1251 new_i_size
= pos
+ copied
;
1252 if (new_i_size
> EXT4_I(inode
)->i_disksize
)
1253 EXT4_I(inode
)->i_disksize
= new_i_size
;
1255 copied
= ext4_generic_write_end(file
, mapping
, pos
, len
, copied
,
1260 ret2
= ext4_journal_stop(handle
);
1264 page_cache_release(page
);
1266 return ret
? ret
: copied
;
1269 static int ext4_journalled_write_end(struct file
*file
,
1270 struct address_space
*mapping
,
1271 loff_t pos
, unsigned len
, unsigned copied
,
1272 struct page
*page
, void *fsdata
)
1274 handle_t
*handle
= ext4_journal_current_handle();
1275 struct inode
*inode
= mapping
->host
;
1280 from
= pos
& (PAGE_CACHE_SIZE
- 1);
1284 if (!PageUptodate(page
))
1286 page_zero_new_buffers(page
, from
+copied
, to
);
1289 ret
= walk_page_buffers(handle
, page_buffers(page
), from
,
1290 to
, &partial
, write_end_fn
);
1292 SetPageUptodate(page
);
1293 if (pos
+copied
> inode
->i_size
)
1294 i_size_write(inode
, pos
+copied
);
1295 EXT4_I(inode
)->i_state
|= EXT4_STATE_JDATA
;
1296 if (inode
->i_size
> EXT4_I(inode
)->i_disksize
) {
1297 EXT4_I(inode
)->i_disksize
= inode
->i_size
;
1298 ret2
= ext4_mark_inode_dirty(handle
, inode
);
1303 ret2
= ext4_journal_stop(handle
);
1307 page_cache_release(page
);
1309 return ret
? ret
: copied
;
1313 * bmap() is special. It gets used by applications such as lilo and by
1314 * the swapper to find the on-disk block of a specific piece of data.
1316 * Naturally, this is dangerous if the block concerned is still in the
1317 * journal. If somebody makes a swapfile on an ext4 data-journaling
1318 * filesystem and enables swap, then they may get a nasty shock when the
1319 * data getting swapped to that swapfile suddenly gets overwritten by
1320 * the original zero's written out previously to the journal and
1321 * awaiting writeback in the kernel's buffer cache.
1323 * So, if we see any bmap calls here on a modified, data-journaled file,
1324 * take extra steps to flush any blocks which might be in the cache.
1326 static sector_t
ext4_bmap(struct address_space
*mapping
, sector_t block
)
1328 struct inode
*inode
= mapping
->host
;
1332 if (EXT4_I(inode
)->i_state
& EXT4_STATE_JDATA
) {
1334 * This is a REALLY heavyweight approach, but the use of
1335 * bmap on dirty files is expected to be extremely rare:
1336 * only if we run lilo or swapon on a freshly made file
1337 * do we expect this to happen.
1339 * (bmap requires CAP_SYS_RAWIO so this does not
1340 * represent an unprivileged user DOS attack --- we'd be
1341 * in trouble if mortal users could trigger this path at
1344 * NB. EXT4_STATE_JDATA is not set on files other than
1345 * regular files. If somebody wants to bmap a directory
1346 * or symlink and gets confused because the buffer
1347 * hasn't yet been flushed to disk, they deserve
1348 * everything they get.
1351 EXT4_I(inode
)->i_state
&= ~EXT4_STATE_JDATA
;
1352 journal
= EXT4_JOURNAL(inode
);
1353 jbd2_journal_lock_updates(journal
);
1354 err
= jbd2_journal_flush(journal
);
1355 jbd2_journal_unlock_updates(journal
);
1361 return generic_block_bmap(mapping
,block
,ext4_get_block
);
1364 static int bget_one(handle_t
*handle
, struct buffer_head
*bh
)
1370 static int bput_one(handle_t
*handle
, struct buffer_head
*bh
)
1376 static int jbd2_journal_dirty_data_fn(handle_t
*handle
, struct buffer_head
*bh
)
1378 if (buffer_mapped(bh
))
1379 return ext4_journal_dirty_data(handle
, bh
);
1384 * Note that we always start a transaction even if we're not journalling
1385 * data. This is to preserve ordering: any hole instantiation within
1386 * __block_write_full_page -> ext4_get_block() should be journalled
1387 * along with the data so we don't crash and then get metadata which
1388 * refers to old data.
1390 * In all journalling modes block_write_full_page() will start the I/O.
1394 * ext4_writepage() -> kmalloc() -> __alloc_pages() -> page_launder() ->
1399 * ext4_file_write() -> generic_file_write() -> __alloc_pages() -> ...
1401 * Same applies to ext4_get_block(). We will deadlock on various things like
1402 * lock_journal and i_truncate_mutex.
1404 * Setting PF_MEMALLOC here doesn't work - too many internal memory
1407 * 16May01: If we're reentered then journal_current_handle() will be
1408 * non-zero. We simply *return*.
1410 * 1 July 2001: @@@ FIXME:
1411 * In journalled data mode, a data buffer may be metadata against the
1412 * current transaction. But the same file is part of a shared mapping
1413 * and someone does a writepage() on it.
1415 * We will move the buffer onto the async_data list, but *after* it has
1416 * been dirtied. So there's a small window where we have dirty data on
1419 * Note that this only applies to the last partial page in the file. The
1420 * bit which block_write_full_page() uses prepare/commit for. (That's
1421 * broken code anyway: it's wrong for msync()).
1423 * It's a rare case: affects the final partial page, for journalled data
1424 * where the file is subject to bith write() and writepage() in the same
1425 * transction. To fix it we'll need a custom block_write_full_page().
1426 * We'll probably need that anyway for journalling writepage() output.
1428 * We don't honour synchronous mounts for writepage(). That would be
1429 * disastrous. Any write() or metadata operation will sync the fs for
1432 * AKPM2: if all the page's buffers are mapped to disk and !data=journal,
1433 * we don't need to open a transaction here.
1435 static int ext4_ordered_writepage(struct page
*page
,
1436 struct writeback_control
*wbc
)
1438 struct inode
*inode
= page
->mapping
->host
;
1439 struct buffer_head
*page_bufs
;
1440 handle_t
*handle
= NULL
;
1444 J_ASSERT(PageLocked(page
));
1447 * We give up here if we're reentered, because it might be for a
1448 * different filesystem.
1450 if (ext4_journal_current_handle())
1453 handle
= ext4_journal_start(inode
, ext4_writepage_trans_blocks(inode
));
1455 if (IS_ERR(handle
)) {
1456 ret
= PTR_ERR(handle
);
1460 if (!page_has_buffers(page
)) {
1461 create_empty_buffers(page
, inode
->i_sb
->s_blocksize
,
1462 (1 << BH_Dirty
)|(1 << BH_Uptodate
));
1464 page_bufs
= page_buffers(page
);
1465 walk_page_buffers(handle
, page_bufs
, 0,
1466 PAGE_CACHE_SIZE
, NULL
, bget_one
);
1468 ret
= block_write_full_page(page
, ext4_get_block
, wbc
);
1471 * The page can become unlocked at any point now, and
1472 * truncate can then come in and change things. So we
1473 * can't touch *page from now on. But *page_bufs is
1474 * safe due to elevated refcount.
1478 * And attach them to the current transaction. But only if
1479 * block_write_full_page() succeeded. Otherwise they are unmapped,
1480 * and generally junk.
1483 err
= walk_page_buffers(handle
, page_bufs
, 0, PAGE_CACHE_SIZE
,
1484 NULL
, jbd2_journal_dirty_data_fn
);
1488 walk_page_buffers(handle
, page_bufs
, 0,
1489 PAGE_CACHE_SIZE
, NULL
, bput_one
);
1490 err
= ext4_journal_stop(handle
);
1496 redirty_page_for_writepage(wbc
, page
);
1501 static int ext4_writeback_writepage(struct page
*page
,
1502 struct writeback_control
*wbc
)
1504 struct inode
*inode
= page
->mapping
->host
;
1505 handle_t
*handle
= NULL
;
1509 if (ext4_journal_current_handle())
1512 handle
= ext4_journal_start(inode
, ext4_writepage_trans_blocks(inode
));
1513 if (IS_ERR(handle
)) {
1514 ret
= PTR_ERR(handle
);
1518 if (test_opt(inode
->i_sb
, NOBH
) && ext4_should_writeback_data(inode
))
1519 ret
= nobh_writepage(page
, ext4_get_block
, wbc
);
1521 ret
= block_write_full_page(page
, ext4_get_block
, wbc
);
1523 err
= ext4_journal_stop(handle
);
1529 redirty_page_for_writepage(wbc
, page
);
1534 static int ext4_journalled_writepage(struct page
*page
,
1535 struct writeback_control
*wbc
)
1537 struct inode
*inode
= page
->mapping
->host
;
1538 handle_t
*handle
= NULL
;
1542 if (ext4_journal_current_handle())
1545 handle
= ext4_journal_start(inode
, ext4_writepage_trans_blocks(inode
));
1546 if (IS_ERR(handle
)) {
1547 ret
= PTR_ERR(handle
);
1551 if (!page_has_buffers(page
) || PageChecked(page
)) {
1553 * It's mmapped pagecache. Add buffers and journal it. There
1554 * doesn't seem much point in redirtying the page here.
1556 ClearPageChecked(page
);
1557 ret
= block_prepare_write(page
, 0, PAGE_CACHE_SIZE
,
1560 ext4_journal_stop(handle
);
1563 ret
= walk_page_buffers(handle
, page_buffers(page
), 0,
1564 PAGE_CACHE_SIZE
, NULL
, do_journal_get_write_access
);
1566 err
= walk_page_buffers(handle
, page_buffers(page
), 0,
1567 PAGE_CACHE_SIZE
, NULL
, write_end_fn
);
1570 EXT4_I(inode
)->i_state
|= EXT4_STATE_JDATA
;
1574 * It may be a page full of checkpoint-mode buffers. We don't
1575 * really know unless we go poke around in the buffer_heads.
1576 * But block_write_full_page will do the right thing.
1578 ret
= block_write_full_page(page
, ext4_get_block
, wbc
);
1580 err
= ext4_journal_stop(handle
);
1587 redirty_page_for_writepage(wbc
, page
);
1593 static int ext4_readpage(struct file
*file
, struct page
*page
)
1595 return mpage_readpage(page
, ext4_get_block
);
1599 ext4_readpages(struct file
*file
, struct address_space
*mapping
,
1600 struct list_head
*pages
, unsigned nr_pages
)
1602 return mpage_readpages(mapping
, pages
, nr_pages
, ext4_get_block
);
1605 static void ext4_invalidatepage(struct page
*page
, unsigned long offset
)
1607 journal_t
*journal
= EXT4_JOURNAL(page
->mapping
->host
);
1610 * If it's a full truncate we just forget about the pending dirtying
1613 ClearPageChecked(page
);
1615 jbd2_journal_invalidatepage(journal
, page
, offset
);
1618 static int ext4_releasepage(struct page
*page
, gfp_t wait
)
1620 journal_t
*journal
= EXT4_JOURNAL(page
->mapping
->host
);
1622 WARN_ON(PageChecked(page
));
1623 if (!page_has_buffers(page
))
1625 return jbd2_journal_try_to_free_buffers(journal
, page
, wait
);
1629 * If the O_DIRECT write will extend the file then add this inode to the
1630 * orphan list. So recovery will truncate it back to the original size
1631 * if the machine crashes during the write.
1633 * If the O_DIRECT write is intantiating holes inside i_size and the machine
1634 * crashes then stale disk data _may_ be exposed inside the file.
1636 static ssize_t
ext4_direct_IO(int rw
, struct kiocb
*iocb
,
1637 const struct iovec
*iov
, loff_t offset
,
1638 unsigned long nr_segs
)
1640 struct file
*file
= iocb
->ki_filp
;
1641 struct inode
*inode
= file
->f_mapping
->host
;
1642 struct ext4_inode_info
*ei
= EXT4_I(inode
);
1643 handle_t
*handle
= NULL
;
1646 size_t count
= iov_length(iov
, nr_segs
);
1649 loff_t final_size
= offset
+ count
;
1651 handle
= ext4_journal_start(inode
, DIO_CREDITS
);
1652 if (IS_ERR(handle
)) {
1653 ret
= PTR_ERR(handle
);
1656 if (final_size
> inode
->i_size
) {
1657 ret
= ext4_orphan_add(handle
, inode
);
1661 ei
->i_disksize
= inode
->i_size
;
1665 ret
= blockdev_direct_IO(rw
, iocb
, inode
, inode
->i_sb
->s_bdev
, iov
,
1667 ext4_get_block
, NULL
);
1670 * Reacquire the handle: ext4_get_block() can restart the transaction
1672 handle
= ext4_journal_current_handle();
1678 if (orphan
&& inode
->i_nlink
)
1679 ext4_orphan_del(handle
, inode
);
1680 if (orphan
&& ret
> 0) {
1681 loff_t end
= offset
+ ret
;
1682 if (end
> inode
->i_size
) {
1683 ei
->i_disksize
= end
;
1684 i_size_write(inode
, end
);
1686 * We're going to return a positive `ret'
1687 * here due to non-zero-length I/O, so there's
1688 * no way of reporting error returns from
1689 * ext4_mark_inode_dirty() to userspace. So
1692 ext4_mark_inode_dirty(handle
, inode
);
1695 err
= ext4_journal_stop(handle
);
1704 * Pages can be marked dirty completely asynchronously from ext4's journalling
1705 * activity. By filemap_sync_pte(), try_to_unmap_one(), etc. We cannot do
1706 * much here because ->set_page_dirty is called under VFS locks. The page is
1707 * not necessarily locked.
1709 * We cannot just dirty the page and leave attached buffers clean, because the
1710 * buffers' dirty state is "definitive". We cannot just set the buffers dirty
1711 * or jbddirty because all the journalling code will explode.
1713 * So what we do is to mark the page "pending dirty" and next time writepage
1714 * is called, propagate that into the buffers appropriately.
1716 static int ext4_journalled_set_page_dirty(struct page
*page
)
1718 SetPageChecked(page
);
1719 return __set_page_dirty_nobuffers(page
);
1722 static const struct address_space_operations ext4_ordered_aops
= {
1723 .readpage
= ext4_readpage
,
1724 .readpages
= ext4_readpages
,
1725 .writepage
= ext4_ordered_writepage
,
1726 .sync_page
= block_sync_page
,
1727 .write_begin
= ext4_write_begin
,
1728 .write_end
= ext4_ordered_write_end
,
1730 .invalidatepage
= ext4_invalidatepage
,
1731 .releasepage
= ext4_releasepage
,
1732 .direct_IO
= ext4_direct_IO
,
1733 .migratepage
= buffer_migrate_page
,
1736 static const struct address_space_operations ext4_writeback_aops
= {
1737 .readpage
= ext4_readpage
,
1738 .readpages
= ext4_readpages
,
1739 .writepage
= ext4_writeback_writepage
,
1740 .sync_page
= block_sync_page
,
1741 .write_begin
= ext4_write_begin
,
1742 .write_end
= ext4_writeback_write_end
,
1744 .invalidatepage
= ext4_invalidatepage
,
1745 .releasepage
= ext4_releasepage
,
1746 .direct_IO
= ext4_direct_IO
,
1747 .migratepage
= buffer_migrate_page
,
1750 static const struct address_space_operations ext4_journalled_aops
= {
1751 .readpage
= ext4_readpage
,
1752 .readpages
= ext4_readpages
,
1753 .writepage
= ext4_journalled_writepage
,
1754 .sync_page
= block_sync_page
,
1755 .write_begin
= ext4_write_begin
,
1756 .write_end
= ext4_journalled_write_end
,
1757 .set_page_dirty
= ext4_journalled_set_page_dirty
,
1759 .invalidatepage
= ext4_invalidatepage
,
1760 .releasepage
= ext4_releasepage
,
1763 void ext4_set_aops(struct inode
*inode
)
1765 if (ext4_should_order_data(inode
))
1766 inode
->i_mapping
->a_ops
= &ext4_ordered_aops
;
1767 else if (ext4_should_writeback_data(inode
))
1768 inode
->i_mapping
->a_ops
= &ext4_writeback_aops
;
1770 inode
->i_mapping
->a_ops
= &ext4_journalled_aops
;
1774 * ext4_block_truncate_page() zeroes out a mapping from file offset `from'
1775 * up to the end of the block which corresponds to `from'.
1776 * This required during truncate. We need to physically zero the tail end
1777 * of that block so it doesn't yield old data if the file is later grown.
1779 int ext4_block_truncate_page(handle_t
*handle
, struct page
*page
,
1780 struct address_space
*mapping
, loff_t from
)
1782 ext4_fsblk_t index
= from
>> PAGE_CACHE_SHIFT
;
1783 unsigned offset
= from
& (PAGE_CACHE_SIZE
-1);
1784 unsigned blocksize
, length
, pos
;
1786 struct inode
*inode
= mapping
->host
;
1787 struct buffer_head
*bh
;
1790 blocksize
= inode
->i_sb
->s_blocksize
;
1791 length
= blocksize
- (offset
& (blocksize
- 1));
1792 iblock
= index
<< (PAGE_CACHE_SHIFT
- inode
->i_sb
->s_blocksize_bits
);
1795 * For "nobh" option, we can only work if we don't need to
1796 * read-in the page - otherwise we create buffers to do the IO.
1798 if (!page_has_buffers(page
) && test_opt(inode
->i_sb
, NOBH
) &&
1799 ext4_should_writeback_data(inode
) && PageUptodate(page
)) {
1800 zero_user_page(page
, offset
, length
, KM_USER0
);
1801 set_page_dirty(page
);
1805 if (!page_has_buffers(page
))
1806 create_empty_buffers(page
, blocksize
, 0);
1808 /* Find the buffer that contains "offset" */
1809 bh
= page_buffers(page
);
1811 while (offset
>= pos
) {
1812 bh
= bh
->b_this_page
;
1818 if (buffer_freed(bh
)) {
1819 BUFFER_TRACE(bh
, "freed: skip");
1823 if (!buffer_mapped(bh
)) {
1824 BUFFER_TRACE(bh
, "unmapped");
1825 ext4_get_block(inode
, iblock
, bh
, 0);
1826 /* unmapped? It's a hole - nothing to do */
1827 if (!buffer_mapped(bh
)) {
1828 BUFFER_TRACE(bh
, "still unmapped");
1833 /* Ok, it's mapped. Make sure it's up-to-date */
1834 if (PageUptodate(page
))
1835 set_buffer_uptodate(bh
);
1837 if (!buffer_uptodate(bh
)) {
1839 ll_rw_block(READ
, 1, &bh
);
1841 /* Uhhuh. Read error. Complain and punt. */
1842 if (!buffer_uptodate(bh
))
1846 if (ext4_should_journal_data(inode
)) {
1847 BUFFER_TRACE(bh
, "get write access");
1848 err
= ext4_journal_get_write_access(handle
, bh
);
1853 zero_user_page(page
, offset
, length
, KM_USER0
);
1855 BUFFER_TRACE(bh
, "zeroed end of block");
1858 if (ext4_should_journal_data(inode
)) {
1859 err
= ext4_journal_dirty_metadata(handle
, bh
);
1861 if (ext4_should_order_data(inode
))
1862 err
= ext4_journal_dirty_data(handle
, bh
);
1863 mark_buffer_dirty(bh
);
1868 page_cache_release(page
);
1873 * Probably it should be a library function... search for first non-zero word
1874 * or memcmp with zero_page, whatever is better for particular architecture.
1877 static inline int all_zeroes(__le32
*p
, __le32
*q
)
1886 * ext4_find_shared - find the indirect blocks for partial truncation.
1887 * @inode: inode in question
1888 * @depth: depth of the affected branch
1889 * @offsets: offsets of pointers in that branch (see ext4_block_to_path)
1890 * @chain: place to store the pointers to partial indirect blocks
1891 * @top: place to the (detached) top of branch
1893 * This is a helper function used by ext4_truncate().
1895 * When we do truncate() we may have to clean the ends of several
1896 * indirect blocks but leave the blocks themselves alive. Block is
1897 * partially truncated if some data below the new i_size is refered
1898 * from it (and it is on the path to the first completely truncated
1899 * data block, indeed). We have to free the top of that path along
1900 * with everything to the right of the path. Since no allocation
1901 * past the truncation point is possible until ext4_truncate()
1902 * finishes, we may safely do the latter, but top of branch may
1903 * require special attention - pageout below the truncation point
1904 * might try to populate it.
1906 * We atomically detach the top of branch from the tree, store the
1907 * block number of its root in *@top, pointers to buffer_heads of
1908 * partially truncated blocks - in @chain[].bh and pointers to
1909 * their last elements that should not be removed - in
1910 * @chain[].p. Return value is the pointer to last filled element
1913 * The work left to caller to do the actual freeing of subtrees:
1914 * a) free the subtree starting from *@top
1915 * b) free the subtrees whose roots are stored in
1916 * (@chain[i].p+1 .. end of @chain[i].bh->b_data)
1917 * c) free the subtrees growing from the inode past the @chain[0].
1918 * (no partially truncated stuff there). */
1920 static Indirect
*ext4_find_shared(struct inode
*inode
, int depth
,
1921 ext4_lblk_t offsets
[4], Indirect chain
[4], __le32
*top
)
1923 Indirect
*partial
, *p
;
1927 /* Make k index the deepest non-null offest + 1 */
1928 for (k
= depth
; k
> 1 && !offsets
[k
-1]; k
--)
1930 partial
= ext4_get_branch(inode
, k
, offsets
, chain
, &err
);
1931 /* Writer: pointers */
1933 partial
= chain
+ k
-1;
1935 * If the branch acquired continuation since we've looked at it -
1936 * fine, it should all survive and (new) top doesn't belong to us.
1938 if (!partial
->key
&& *partial
->p
)
1941 for (p
=partial
; p
>chain
&& all_zeroes((__le32
*)p
->bh
->b_data
,p
->p
); p
--)
1944 * OK, we've found the last block that must survive. The rest of our
1945 * branch should be detached before unlocking. However, if that rest
1946 * of branch is all ours and does not grow immediately from the inode
1947 * it's easier to cheat and just decrement partial->p.
1949 if (p
== chain
+ k
- 1 && p
> chain
) {
1953 /* Nope, don't do this in ext4. Must leave the tree intact */
1960 while(partial
> p
) {
1961 brelse(partial
->bh
);
1969 * Zero a number of block pointers in either an inode or an indirect block.
1970 * If we restart the transaction we must again get write access to the
1971 * indirect block for further modification.
1973 * We release `count' blocks on disk, but (last - first) may be greater
1974 * than `count' because there can be holes in there.
1976 static void ext4_clear_blocks(handle_t
*handle
, struct inode
*inode
,
1977 struct buffer_head
*bh
, ext4_fsblk_t block_to_free
,
1978 unsigned long count
, __le32
*first
, __le32
*last
)
1981 if (try_to_extend_transaction(handle
, inode
)) {
1983 BUFFER_TRACE(bh
, "call ext4_journal_dirty_metadata");
1984 ext4_journal_dirty_metadata(handle
, bh
);
1986 ext4_mark_inode_dirty(handle
, inode
);
1987 ext4_journal_test_restart(handle
, inode
);
1989 BUFFER_TRACE(bh
, "retaking write access");
1990 ext4_journal_get_write_access(handle
, bh
);
1995 * Any buffers which are on the journal will be in memory. We find
1996 * them on the hash table so jbd2_journal_revoke() will run jbd2_journal_forget()
1997 * on them. We've already detached each block from the file, so
1998 * bforget() in jbd2_journal_forget() should be safe.
2000 * AKPM: turn on bforget in jbd2_journal_forget()!!!
2002 for (p
= first
; p
< last
; p
++) {
2003 u32 nr
= le32_to_cpu(*p
);
2005 struct buffer_head
*tbh
;
2008 tbh
= sb_find_get_block(inode
->i_sb
, nr
);
2009 ext4_forget(handle
, 0, inode
, tbh
, nr
);
2013 ext4_free_blocks(handle
, inode
, block_to_free
, count
);
2017 * ext4_free_data - free a list of data blocks
2018 * @handle: handle for this transaction
2019 * @inode: inode we are dealing with
2020 * @this_bh: indirect buffer_head which contains *@first and *@last
2021 * @first: array of block numbers
2022 * @last: points immediately past the end of array
2024 * We are freeing all blocks refered from that array (numbers are stored as
2025 * little-endian 32-bit) and updating @inode->i_blocks appropriately.
2027 * We accumulate contiguous runs of blocks to free. Conveniently, if these
2028 * blocks are contiguous then releasing them at one time will only affect one
2029 * or two bitmap blocks (+ group descriptor(s) and superblock) and we won't
2030 * actually use a lot of journal space.
2032 * @this_bh will be %NULL if @first and @last point into the inode's direct
2035 static void ext4_free_data(handle_t
*handle
, struct inode
*inode
,
2036 struct buffer_head
*this_bh
,
2037 __le32
*first
, __le32
*last
)
2039 ext4_fsblk_t block_to_free
= 0; /* Starting block # of a run */
2040 unsigned long count
= 0; /* Number of blocks in the run */
2041 __le32
*block_to_free_p
= NULL
; /* Pointer into inode/ind
2044 ext4_fsblk_t nr
; /* Current block # */
2045 __le32
*p
; /* Pointer into inode/ind
2046 for current block */
2049 if (this_bh
) { /* For indirect block */
2050 BUFFER_TRACE(this_bh
, "get_write_access");
2051 err
= ext4_journal_get_write_access(handle
, this_bh
);
2052 /* Important: if we can't update the indirect pointers
2053 * to the blocks, we can't free them. */
2058 for (p
= first
; p
< last
; p
++) {
2059 nr
= le32_to_cpu(*p
);
2061 /* accumulate blocks to free if they're contiguous */
2064 block_to_free_p
= p
;
2066 } else if (nr
== block_to_free
+ count
) {
2069 ext4_clear_blocks(handle
, inode
, this_bh
,
2071 count
, block_to_free_p
, p
);
2073 block_to_free_p
= p
;
2080 ext4_clear_blocks(handle
, inode
, this_bh
, block_to_free
,
2081 count
, block_to_free_p
, p
);
2084 BUFFER_TRACE(this_bh
, "call ext4_journal_dirty_metadata");
2085 ext4_journal_dirty_metadata(handle
, this_bh
);
2090 * ext4_free_branches - free an array of branches
2091 * @handle: JBD handle for this transaction
2092 * @inode: inode we are dealing with
2093 * @parent_bh: the buffer_head which contains *@first and *@last
2094 * @first: array of block numbers
2095 * @last: pointer immediately past the end of array
2096 * @depth: depth of the branches to free
2098 * We are freeing all blocks refered from these branches (numbers are
2099 * stored as little-endian 32-bit) and updating @inode->i_blocks
2102 static void ext4_free_branches(handle_t
*handle
, struct inode
*inode
,
2103 struct buffer_head
*parent_bh
,
2104 __le32
*first
, __le32
*last
, int depth
)
2109 if (is_handle_aborted(handle
))
2113 struct buffer_head
*bh
;
2114 int addr_per_block
= EXT4_ADDR_PER_BLOCK(inode
->i_sb
);
2116 while (--p
>= first
) {
2117 nr
= le32_to_cpu(*p
);
2119 continue; /* A hole */
2121 /* Go read the buffer for the next level down */
2122 bh
= sb_bread(inode
->i_sb
, nr
);
2125 * A read failure? Report error and clear slot
2129 ext4_error(inode
->i_sb
, "ext4_free_branches",
2130 "Read failure, inode=%lu, block=%llu",
2135 /* This zaps the entire block. Bottom up. */
2136 BUFFER_TRACE(bh
, "free child branches");
2137 ext4_free_branches(handle
, inode
, bh
,
2138 (__le32
*)bh
->b_data
,
2139 (__le32
*)bh
->b_data
+ addr_per_block
,
2143 * We've probably journalled the indirect block several
2144 * times during the truncate. But it's no longer
2145 * needed and we now drop it from the transaction via
2146 * jbd2_journal_revoke().
2148 * That's easy if it's exclusively part of this
2149 * transaction. But if it's part of the committing
2150 * transaction then jbd2_journal_forget() will simply
2151 * brelse() it. That means that if the underlying
2152 * block is reallocated in ext4_get_block(),
2153 * unmap_underlying_metadata() will find this block
2154 * and will try to get rid of it. damn, damn.
2156 * If this block has already been committed to the
2157 * journal, a revoke record will be written. And
2158 * revoke records must be emitted *before* clearing
2159 * this block's bit in the bitmaps.
2161 ext4_forget(handle
, 1, inode
, bh
, bh
->b_blocknr
);
2164 * Everything below this this pointer has been
2165 * released. Now let this top-of-subtree go.
2167 * We want the freeing of this indirect block to be
2168 * atomic in the journal with the updating of the
2169 * bitmap block which owns it. So make some room in
2172 * We zero the parent pointer *after* freeing its
2173 * pointee in the bitmaps, so if extend_transaction()
2174 * for some reason fails to put the bitmap changes and
2175 * the release into the same transaction, recovery
2176 * will merely complain about releasing a free block,
2177 * rather than leaking blocks.
2179 if (is_handle_aborted(handle
))
2181 if (try_to_extend_transaction(handle
, inode
)) {
2182 ext4_mark_inode_dirty(handle
, inode
);
2183 ext4_journal_test_restart(handle
, inode
);
2186 ext4_free_blocks(handle
, inode
, nr
, 1);
2190 * The block which we have just freed is
2191 * pointed to by an indirect block: journal it
2193 BUFFER_TRACE(parent_bh
, "get_write_access");
2194 if (!ext4_journal_get_write_access(handle
,
2197 BUFFER_TRACE(parent_bh
,
2198 "call ext4_journal_dirty_metadata");
2199 ext4_journal_dirty_metadata(handle
,
2205 /* We have reached the bottom of the tree. */
2206 BUFFER_TRACE(parent_bh
, "free data blocks");
2207 ext4_free_data(handle
, inode
, parent_bh
, first
, last
);
2214 * We block out ext4_get_block() block instantiations across the entire
2215 * transaction, and VFS/VM ensures that ext4_truncate() cannot run
2216 * simultaneously on behalf of the same inode.
2218 * As we work through the truncate and commmit bits of it to the journal there
2219 * is one core, guiding principle: the file's tree must always be consistent on
2220 * disk. We must be able to restart the truncate after a crash.
2222 * The file's tree may be transiently inconsistent in memory (although it
2223 * probably isn't), but whenever we close off and commit a journal transaction,
2224 * the contents of (the filesystem + the journal) must be consistent and
2225 * restartable. It's pretty simple, really: bottom up, right to left (although
2226 * left-to-right works OK too).
2228 * Note that at recovery time, journal replay occurs *before* the restart of
2229 * truncate against the orphan inode list.
2231 * The committed inode has the new, desired i_size (which is the same as
2232 * i_disksize in this case). After a crash, ext4_orphan_cleanup() will see
2233 * that this inode's truncate did not complete and it will again call
2234 * ext4_truncate() to have another go. So there will be instantiated blocks
2235 * to the right of the truncation point in a crashed ext4 filesystem. But
2236 * that's fine - as long as they are linked from the inode, the post-crash
2237 * ext4_truncate() run will find them and release them.
2239 void ext4_truncate(struct inode
*inode
)
2242 struct ext4_inode_info
*ei
= EXT4_I(inode
);
2243 __le32
*i_data
= ei
->i_data
;
2244 int addr_per_block
= EXT4_ADDR_PER_BLOCK(inode
->i_sb
);
2245 struct address_space
*mapping
= inode
->i_mapping
;
2246 ext4_lblk_t offsets
[4];
2251 ext4_lblk_t last_block
;
2252 unsigned blocksize
= inode
->i_sb
->s_blocksize
;
2255 if (!(S_ISREG(inode
->i_mode
) || S_ISDIR(inode
->i_mode
) ||
2256 S_ISLNK(inode
->i_mode
)))
2258 if (ext4_inode_is_fast_symlink(inode
))
2260 if (IS_APPEND(inode
) || IS_IMMUTABLE(inode
))
2264 * We have to lock the EOF page here, because lock_page() nests
2265 * outside jbd2_journal_start().
2267 if ((inode
->i_size
& (blocksize
- 1)) == 0) {
2268 /* Block boundary? Nothing to do */
2271 page
= grab_cache_page(mapping
,
2272 inode
->i_size
>> PAGE_CACHE_SHIFT
);
2277 if (EXT4_I(inode
)->i_flags
& EXT4_EXTENTS_FL
) {
2278 ext4_ext_truncate(inode
, page
);
2282 handle
= start_transaction(inode
);
2283 if (IS_ERR(handle
)) {
2285 clear_highpage(page
);
2286 flush_dcache_page(page
);
2288 page_cache_release(page
);
2290 return; /* AKPM: return what? */
2293 last_block
= (inode
->i_size
+ blocksize
-1)
2294 >> EXT4_BLOCK_SIZE_BITS(inode
->i_sb
);
2297 ext4_block_truncate_page(handle
, page
, mapping
, inode
->i_size
);
2299 n
= ext4_block_to_path(inode
, last_block
, offsets
, NULL
);
2301 goto out_stop
; /* error */
2304 * OK. This truncate is going to happen. We add the inode to the
2305 * orphan list, so that if this truncate spans multiple transactions,
2306 * and we crash, we will resume the truncate when the filesystem
2307 * recovers. It also marks the inode dirty, to catch the new size.
2309 * Implication: the file must always be in a sane, consistent
2310 * truncatable state while each transaction commits.
2312 if (ext4_orphan_add(handle
, inode
))
2316 * The orphan list entry will now protect us from any crash which
2317 * occurs before the truncate completes, so it is now safe to propagate
2318 * the new, shorter inode size (held for now in i_size) into the
2319 * on-disk inode. We do this via i_disksize, which is the value which
2320 * ext4 *really* writes onto the disk inode.
2322 ei
->i_disksize
= inode
->i_size
;
2325 * From here we block out all ext4_get_block() callers who want to
2326 * modify the block allocation tree.
2328 mutex_lock(&ei
->truncate_mutex
);
2330 if (n
== 1) { /* direct blocks */
2331 ext4_free_data(handle
, inode
, NULL
, i_data
+offsets
[0],
2332 i_data
+ EXT4_NDIR_BLOCKS
);
2336 partial
= ext4_find_shared(inode
, n
, offsets
, chain
, &nr
);
2337 /* Kill the top of shared branch (not detached) */
2339 if (partial
== chain
) {
2340 /* Shared branch grows from the inode */
2341 ext4_free_branches(handle
, inode
, NULL
,
2342 &nr
, &nr
+1, (chain
+n
-1) - partial
);
2345 * We mark the inode dirty prior to restart,
2346 * and prior to stop. No need for it here.
2349 /* Shared branch grows from an indirect block */
2350 BUFFER_TRACE(partial
->bh
, "get_write_access");
2351 ext4_free_branches(handle
, inode
, partial
->bh
,
2353 partial
->p
+1, (chain
+n
-1) - partial
);
2356 /* Clear the ends of indirect blocks on the shared branch */
2357 while (partial
> chain
) {
2358 ext4_free_branches(handle
, inode
, partial
->bh
, partial
->p
+ 1,
2359 (__le32
*)partial
->bh
->b_data
+addr_per_block
,
2360 (chain
+n
-1) - partial
);
2361 BUFFER_TRACE(partial
->bh
, "call brelse");
2362 brelse (partial
->bh
);
2366 /* Kill the remaining (whole) subtrees */
2367 switch (offsets
[0]) {
2369 nr
= i_data
[EXT4_IND_BLOCK
];
2371 ext4_free_branches(handle
, inode
, NULL
, &nr
, &nr
+1, 1);
2372 i_data
[EXT4_IND_BLOCK
] = 0;
2374 case EXT4_IND_BLOCK
:
2375 nr
= i_data
[EXT4_DIND_BLOCK
];
2377 ext4_free_branches(handle
, inode
, NULL
, &nr
, &nr
+1, 2);
2378 i_data
[EXT4_DIND_BLOCK
] = 0;
2380 case EXT4_DIND_BLOCK
:
2381 nr
= i_data
[EXT4_TIND_BLOCK
];
2383 ext4_free_branches(handle
, inode
, NULL
, &nr
, &nr
+1, 3);
2384 i_data
[EXT4_TIND_BLOCK
] = 0;
2386 case EXT4_TIND_BLOCK
:
2390 ext4_discard_reservation(inode
);
2392 mutex_unlock(&ei
->truncate_mutex
);
2393 inode
->i_mtime
= inode
->i_ctime
= ext4_current_time(inode
);
2394 ext4_mark_inode_dirty(handle
, inode
);
2397 * In a multi-transaction truncate, we only make the final transaction
2404 * If this was a simple ftruncate(), and the file will remain alive
2405 * then we need to clear up the orphan record which we created above.
2406 * However, if this was a real unlink then we were called by
2407 * ext4_delete_inode(), and we allow that function to clean up the
2408 * orphan info for us.
2411 ext4_orphan_del(handle
, inode
);
2413 ext4_journal_stop(handle
);
2416 static ext4_fsblk_t
ext4_get_inode_block(struct super_block
*sb
,
2417 unsigned long ino
, struct ext4_iloc
*iloc
)
2419 unsigned long desc
, group_desc
;
2420 ext4_group_t block_group
;
2421 unsigned long offset
;
2423 struct buffer_head
*bh
;
2424 struct ext4_group_desc
* gdp
;
2426 if (!ext4_valid_inum(sb
, ino
)) {
2428 * This error is already checked for in namei.c unless we are
2429 * looking at an NFS filehandle, in which case no error
2435 block_group
= (ino
- 1) / EXT4_INODES_PER_GROUP(sb
);
2436 if (block_group
>= EXT4_SB(sb
)->s_groups_count
) {
2437 ext4_error(sb
,"ext4_get_inode_block","group >= groups count");
2441 group_desc
= block_group
>> EXT4_DESC_PER_BLOCK_BITS(sb
);
2442 desc
= block_group
& (EXT4_DESC_PER_BLOCK(sb
) - 1);
2443 bh
= EXT4_SB(sb
)->s_group_desc
[group_desc
];
2445 ext4_error (sb
, "ext4_get_inode_block",
2446 "Descriptor not loaded");
2450 gdp
= (struct ext4_group_desc
*)((__u8
*)bh
->b_data
+
2451 desc
* EXT4_DESC_SIZE(sb
));
2453 * Figure out the offset within the block group inode table
2455 offset
= ((ino
- 1) % EXT4_INODES_PER_GROUP(sb
)) *
2456 EXT4_INODE_SIZE(sb
);
2457 block
= ext4_inode_table(sb
, gdp
) +
2458 (offset
>> EXT4_BLOCK_SIZE_BITS(sb
));
2460 iloc
->block_group
= block_group
;
2461 iloc
->offset
= offset
& (EXT4_BLOCK_SIZE(sb
) - 1);
2466 * ext4_get_inode_loc returns with an extra refcount against the inode's
2467 * underlying buffer_head on success. If 'in_mem' is true, we have all
2468 * data in memory that is needed to recreate the on-disk version of this
2471 static int __ext4_get_inode_loc(struct inode
*inode
,
2472 struct ext4_iloc
*iloc
, int in_mem
)
2475 struct buffer_head
*bh
;
2477 block
= ext4_get_inode_block(inode
->i_sb
, inode
->i_ino
, iloc
);
2481 bh
= sb_getblk(inode
->i_sb
, block
);
2483 ext4_error (inode
->i_sb
, "ext4_get_inode_loc",
2484 "unable to read inode block - "
2485 "inode=%lu, block=%llu",
2486 inode
->i_ino
, block
);
2489 if (!buffer_uptodate(bh
)) {
2491 if (buffer_uptodate(bh
)) {
2492 /* someone brought it uptodate while we waited */
2498 * If we have all information of the inode in memory and this
2499 * is the only valid inode in the block, we need not read the
2503 struct buffer_head
*bitmap_bh
;
2504 struct ext4_group_desc
*desc
;
2505 int inodes_per_buffer
;
2506 int inode_offset
, i
;
2507 ext4_group_t block_group
;
2510 block_group
= (inode
->i_ino
- 1) /
2511 EXT4_INODES_PER_GROUP(inode
->i_sb
);
2512 inodes_per_buffer
= bh
->b_size
/
2513 EXT4_INODE_SIZE(inode
->i_sb
);
2514 inode_offset
= ((inode
->i_ino
- 1) %
2515 EXT4_INODES_PER_GROUP(inode
->i_sb
));
2516 start
= inode_offset
& ~(inodes_per_buffer
- 1);
2518 /* Is the inode bitmap in cache? */
2519 desc
= ext4_get_group_desc(inode
->i_sb
,
2524 bitmap_bh
= sb_getblk(inode
->i_sb
,
2525 ext4_inode_bitmap(inode
->i_sb
, desc
));
2530 * If the inode bitmap isn't in cache then the
2531 * optimisation may end up performing two reads instead
2532 * of one, so skip it.
2534 if (!buffer_uptodate(bitmap_bh
)) {
2538 for (i
= start
; i
< start
+ inodes_per_buffer
; i
++) {
2539 if (i
== inode_offset
)
2541 if (ext4_test_bit(i
, bitmap_bh
->b_data
))
2545 if (i
== start
+ inodes_per_buffer
) {
2546 /* all other inodes are free, so skip I/O */
2547 memset(bh
->b_data
, 0, bh
->b_size
);
2548 set_buffer_uptodate(bh
);
2556 * There are other valid inodes in the buffer, this inode
2557 * has in-inode xattrs, or we don't have this inode in memory.
2558 * Read the block from disk.
2561 bh
->b_end_io
= end_buffer_read_sync
;
2562 submit_bh(READ_META
, bh
);
2564 if (!buffer_uptodate(bh
)) {
2565 ext4_error(inode
->i_sb
, "ext4_get_inode_loc",
2566 "unable to read inode block - "
2567 "inode=%lu, block=%llu",
2568 inode
->i_ino
, block
);
2578 int ext4_get_inode_loc(struct inode
*inode
, struct ext4_iloc
*iloc
)
2580 /* We have all inode data except xattrs in memory here. */
2581 return __ext4_get_inode_loc(inode
, iloc
,
2582 !(EXT4_I(inode
)->i_state
& EXT4_STATE_XATTR
));
2585 void ext4_set_inode_flags(struct inode
*inode
)
2587 unsigned int flags
= EXT4_I(inode
)->i_flags
;
2589 inode
->i_flags
&= ~(S_SYNC
|S_APPEND
|S_IMMUTABLE
|S_NOATIME
|S_DIRSYNC
);
2590 if (flags
& EXT4_SYNC_FL
)
2591 inode
->i_flags
|= S_SYNC
;
2592 if (flags
& EXT4_APPEND_FL
)
2593 inode
->i_flags
|= S_APPEND
;
2594 if (flags
& EXT4_IMMUTABLE_FL
)
2595 inode
->i_flags
|= S_IMMUTABLE
;
2596 if (flags
& EXT4_NOATIME_FL
)
2597 inode
->i_flags
|= S_NOATIME
;
2598 if (flags
& EXT4_DIRSYNC_FL
)
2599 inode
->i_flags
|= S_DIRSYNC
;
2602 /* Propagate flags from i_flags to EXT4_I(inode)->i_flags */
2603 void ext4_get_inode_flags(struct ext4_inode_info
*ei
)
2605 unsigned int flags
= ei
->vfs_inode
.i_flags
;
2607 ei
->i_flags
&= ~(EXT4_SYNC_FL
|EXT4_APPEND_FL
|
2608 EXT4_IMMUTABLE_FL
|EXT4_NOATIME_FL
|EXT4_DIRSYNC_FL
);
2610 ei
->i_flags
|= EXT4_SYNC_FL
;
2611 if (flags
& S_APPEND
)
2612 ei
->i_flags
|= EXT4_APPEND_FL
;
2613 if (flags
& S_IMMUTABLE
)
2614 ei
->i_flags
|= EXT4_IMMUTABLE_FL
;
2615 if (flags
& S_NOATIME
)
2616 ei
->i_flags
|= EXT4_NOATIME_FL
;
2617 if (flags
& S_DIRSYNC
)
2618 ei
->i_flags
|= EXT4_DIRSYNC_FL
;
2620 static blkcnt_t
ext4_inode_blocks(struct ext4_inode
*raw_inode
,
2621 struct ext4_inode_info
*ei
)
2624 struct inode
*inode
= &(ei
->vfs_inode
);
2625 struct super_block
*sb
= inode
->i_sb
;
2627 if (EXT4_HAS_RO_COMPAT_FEATURE(sb
,
2628 EXT4_FEATURE_RO_COMPAT_HUGE_FILE
)) {
2629 /* we are using combined 48 bit field */
2630 i_blocks
= ((u64
)le16_to_cpu(raw_inode
->i_blocks_high
)) << 32 |
2631 le32_to_cpu(raw_inode
->i_blocks_lo
);
2632 if (ei
->i_flags
& EXT4_HUGE_FILE_FL
) {
2633 /* i_blocks represent file system block size */
2634 return i_blocks
<< (inode
->i_blkbits
- 9);
2639 return le32_to_cpu(raw_inode
->i_blocks_lo
);
2643 void ext4_read_inode(struct inode
* inode
)
2645 struct ext4_iloc iloc
;
2646 struct ext4_inode
*raw_inode
;
2647 struct ext4_inode_info
*ei
= EXT4_I(inode
);
2648 struct buffer_head
*bh
;
2651 #ifdef CONFIG_EXT4DEV_FS_POSIX_ACL
2652 ei
->i_acl
= EXT4_ACL_NOT_CACHED
;
2653 ei
->i_default_acl
= EXT4_ACL_NOT_CACHED
;
2655 ei
->i_block_alloc_info
= NULL
;
2657 if (__ext4_get_inode_loc(inode
, &iloc
, 0))
2660 raw_inode
= ext4_raw_inode(&iloc
);
2661 inode
->i_mode
= le16_to_cpu(raw_inode
->i_mode
);
2662 inode
->i_uid
= (uid_t
)le16_to_cpu(raw_inode
->i_uid_low
);
2663 inode
->i_gid
= (gid_t
)le16_to_cpu(raw_inode
->i_gid_low
);
2664 if(!(test_opt (inode
->i_sb
, NO_UID32
))) {
2665 inode
->i_uid
|= le16_to_cpu(raw_inode
->i_uid_high
) << 16;
2666 inode
->i_gid
|= le16_to_cpu(raw_inode
->i_gid_high
) << 16;
2668 inode
->i_nlink
= le16_to_cpu(raw_inode
->i_links_count
);
2671 ei
->i_dir_start_lookup
= 0;
2672 ei
->i_dtime
= le32_to_cpu(raw_inode
->i_dtime
);
2673 /* We now have enough fields to check if the inode was active or not.
2674 * This is needed because nfsd might try to access dead inodes
2675 * the test is that same one that e2fsck uses
2676 * NeilBrown 1999oct15
2678 if (inode
->i_nlink
== 0) {
2679 if (inode
->i_mode
== 0 ||
2680 !(EXT4_SB(inode
->i_sb
)->s_mount_state
& EXT4_ORPHAN_FS
)) {
2681 /* this inode is deleted */
2685 /* The only unlinked inodes we let through here have
2686 * valid i_mode and are being read by the orphan
2687 * recovery code: that's fine, we're about to complete
2688 * the process of deleting those. */
2690 ei
->i_flags
= le32_to_cpu(raw_inode
->i_flags
);
2691 inode
->i_blocks
= ext4_inode_blocks(raw_inode
, ei
);
2692 ei
->i_file_acl
= le32_to_cpu(raw_inode
->i_file_acl_lo
);
2693 if (EXT4_SB(inode
->i_sb
)->s_es
->s_creator_os
!=
2694 cpu_to_le32(EXT4_OS_HURD
)) {
2696 ((__u64
)le16_to_cpu(raw_inode
->i_file_acl_high
)) << 32;
2698 inode
->i_size
= ext4_isize(raw_inode
);
2699 ei
->i_disksize
= inode
->i_size
;
2700 inode
->i_generation
= le32_to_cpu(raw_inode
->i_generation
);
2701 ei
->i_block_group
= iloc
.block_group
;
2703 * NOTE! The in-memory inode i_data array is in little-endian order
2704 * even on big-endian machines: we do NOT byteswap the block numbers!
2706 for (block
= 0; block
< EXT4_N_BLOCKS
; block
++)
2707 ei
->i_data
[block
] = raw_inode
->i_block
[block
];
2708 INIT_LIST_HEAD(&ei
->i_orphan
);
2710 if (inode
->i_ino
>= EXT4_FIRST_INO(inode
->i_sb
) + 1 &&
2711 EXT4_INODE_SIZE(inode
->i_sb
) > EXT4_GOOD_OLD_INODE_SIZE
) {
2713 * When mke2fs creates big inodes it does not zero out
2714 * the unused bytes above EXT4_GOOD_OLD_INODE_SIZE,
2715 * so ignore those first few inodes.
2717 ei
->i_extra_isize
= le16_to_cpu(raw_inode
->i_extra_isize
);
2718 if (EXT4_GOOD_OLD_INODE_SIZE
+ ei
->i_extra_isize
>
2719 EXT4_INODE_SIZE(inode
->i_sb
)) {
2723 if (ei
->i_extra_isize
== 0) {
2724 /* The extra space is currently unused. Use it. */
2725 ei
->i_extra_isize
= sizeof(struct ext4_inode
) -
2726 EXT4_GOOD_OLD_INODE_SIZE
;
2728 __le32
*magic
= (void *)raw_inode
+
2729 EXT4_GOOD_OLD_INODE_SIZE
+
2731 if (*magic
== cpu_to_le32(EXT4_XATTR_MAGIC
))
2732 ei
->i_state
|= EXT4_STATE_XATTR
;
2735 ei
->i_extra_isize
= 0;
2737 EXT4_INODE_GET_XTIME(i_ctime
, inode
, raw_inode
);
2738 EXT4_INODE_GET_XTIME(i_mtime
, inode
, raw_inode
);
2739 EXT4_INODE_GET_XTIME(i_atime
, inode
, raw_inode
);
2740 EXT4_EINODE_GET_XTIME(i_crtime
, ei
, raw_inode
);
2742 if (S_ISREG(inode
->i_mode
)) {
2743 inode
->i_op
= &ext4_file_inode_operations
;
2744 inode
->i_fop
= &ext4_file_operations
;
2745 ext4_set_aops(inode
);
2746 } else if (S_ISDIR(inode
->i_mode
)) {
2747 inode
->i_op
= &ext4_dir_inode_operations
;
2748 inode
->i_fop
= &ext4_dir_operations
;
2749 } else if (S_ISLNK(inode
->i_mode
)) {
2750 if (ext4_inode_is_fast_symlink(inode
))
2751 inode
->i_op
= &ext4_fast_symlink_inode_operations
;
2753 inode
->i_op
= &ext4_symlink_inode_operations
;
2754 ext4_set_aops(inode
);
2757 inode
->i_op
= &ext4_special_inode_operations
;
2758 if (raw_inode
->i_block
[0])
2759 init_special_inode(inode
, inode
->i_mode
,
2760 old_decode_dev(le32_to_cpu(raw_inode
->i_block
[0])));
2762 init_special_inode(inode
, inode
->i_mode
,
2763 new_decode_dev(le32_to_cpu(raw_inode
->i_block
[1])));
2766 ext4_set_inode_flags(inode
);
2770 make_bad_inode(inode
);
2774 static int ext4_inode_blocks_set(handle_t
*handle
,
2775 struct ext4_inode
*raw_inode
,
2776 struct ext4_inode_info
*ei
)
2778 struct inode
*inode
= &(ei
->vfs_inode
);
2779 u64 i_blocks
= inode
->i_blocks
;
2780 struct super_block
*sb
= inode
->i_sb
;
2783 if (i_blocks
<= ~0U) {
2785 * i_blocks can be represnted in a 32 bit variable
2786 * as multiple of 512 bytes
2788 raw_inode
->i_blocks_lo
= cpu_to_le32(i_blocks
);
2789 raw_inode
->i_blocks_high
= 0;
2790 ei
->i_flags
&= ~EXT4_HUGE_FILE_FL
;
2791 } else if (i_blocks
<= 0xffffffffffffULL
) {
2793 * i_blocks can be represented in a 48 bit variable
2794 * as multiple of 512 bytes
2796 err
= ext4_update_rocompat_feature(handle
, sb
,
2797 EXT4_FEATURE_RO_COMPAT_HUGE_FILE
);
2800 /* i_block is stored in the split 48 bit fields */
2801 raw_inode
->i_blocks_lo
= cpu_to_le32(i_blocks
);
2802 raw_inode
->i_blocks_high
= cpu_to_le16(i_blocks
>> 32);
2803 ei
->i_flags
&= ~EXT4_HUGE_FILE_FL
;
2806 * i_blocks should be represented in a 48 bit variable
2807 * as multiple of file system block size
2809 err
= ext4_update_rocompat_feature(handle
, sb
,
2810 EXT4_FEATURE_RO_COMPAT_HUGE_FILE
);
2813 ei
->i_flags
|= EXT4_HUGE_FILE_FL
;
2814 /* i_block is stored in file system block size */
2815 i_blocks
= i_blocks
>> (inode
->i_blkbits
- 9);
2816 raw_inode
->i_blocks_lo
= cpu_to_le32(i_blocks
);
2817 raw_inode
->i_blocks_high
= cpu_to_le16(i_blocks
>> 32);
2824 * Post the struct inode info into an on-disk inode location in the
2825 * buffer-cache. This gobbles the caller's reference to the
2826 * buffer_head in the inode location struct.
2828 * The caller must have write access to iloc->bh.
2830 static int ext4_do_update_inode(handle_t
*handle
,
2831 struct inode
*inode
,
2832 struct ext4_iloc
*iloc
)
2834 struct ext4_inode
*raw_inode
= ext4_raw_inode(iloc
);
2835 struct ext4_inode_info
*ei
= EXT4_I(inode
);
2836 struct buffer_head
*bh
= iloc
->bh
;
2837 int err
= 0, rc
, block
;
2839 /* For fields not not tracking in the in-memory inode,
2840 * initialise them to zero for new inodes. */
2841 if (ei
->i_state
& EXT4_STATE_NEW
)
2842 memset(raw_inode
, 0, EXT4_SB(inode
->i_sb
)->s_inode_size
);
2844 ext4_get_inode_flags(ei
);
2845 raw_inode
->i_mode
= cpu_to_le16(inode
->i_mode
);
2846 if(!(test_opt(inode
->i_sb
, NO_UID32
))) {
2847 raw_inode
->i_uid_low
= cpu_to_le16(low_16_bits(inode
->i_uid
));
2848 raw_inode
->i_gid_low
= cpu_to_le16(low_16_bits(inode
->i_gid
));
2850 * Fix up interoperability with old kernels. Otherwise, old inodes get
2851 * re-used with the upper 16 bits of the uid/gid intact
2854 raw_inode
->i_uid_high
=
2855 cpu_to_le16(high_16_bits(inode
->i_uid
));
2856 raw_inode
->i_gid_high
=
2857 cpu_to_le16(high_16_bits(inode
->i_gid
));
2859 raw_inode
->i_uid_high
= 0;
2860 raw_inode
->i_gid_high
= 0;
2863 raw_inode
->i_uid_low
=
2864 cpu_to_le16(fs_high2lowuid(inode
->i_uid
));
2865 raw_inode
->i_gid_low
=
2866 cpu_to_le16(fs_high2lowgid(inode
->i_gid
));
2867 raw_inode
->i_uid_high
= 0;
2868 raw_inode
->i_gid_high
= 0;
2870 raw_inode
->i_links_count
= cpu_to_le16(inode
->i_nlink
);
2872 EXT4_INODE_SET_XTIME(i_ctime
, inode
, raw_inode
);
2873 EXT4_INODE_SET_XTIME(i_mtime
, inode
, raw_inode
);
2874 EXT4_INODE_SET_XTIME(i_atime
, inode
, raw_inode
);
2875 EXT4_EINODE_SET_XTIME(i_crtime
, ei
, raw_inode
);
2877 if (ext4_inode_blocks_set(handle
, raw_inode
, ei
))
2879 raw_inode
->i_dtime
= cpu_to_le32(ei
->i_dtime
);
2880 raw_inode
->i_flags
= cpu_to_le32(ei
->i_flags
);
2881 if (EXT4_SB(inode
->i_sb
)->s_es
->s_creator_os
!=
2882 cpu_to_le32(EXT4_OS_HURD
))
2883 raw_inode
->i_file_acl_high
=
2884 cpu_to_le16(ei
->i_file_acl
>> 32);
2885 raw_inode
->i_file_acl_lo
= cpu_to_le32(ei
->i_file_acl
);
2886 ext4_isize_set(raw_inode
, ei
->i_disksize
);
2887 if (ei
->i_disksize
> 0x7fffffffULL
) {
2888 struct super_block
*sb
= inode
->i_sb
;
2889 if (!EXT4_HAS_RO_COMPAT_FEATURE(sb
,
2890 EXT4_FEATURE_RO_COMPAT_LARGE_FILE
) ||
2891 EXT4_SB(sb
)->s_es
->s_rev_level
==
2892 cpu_to_le32(EXT4_GOOD_OLD_REV
)) {
2893 /* If this is the first large file
2894 * created, add a flag to the superblock.
2896 err
= ext4_journal_get_write_access(handle
,
2897 EXT4_SB(sb
)->s_sbh
);
2900 ext4_update_dynamic_rev(sb
);
2901 EXT4_SET_RO_COMPAT_FEATURE(sb
,
2902 EXT4_FEATURE_RO_COMPAT_LARGE_FILE
);
2905 err
= ext4_journal_dirty_metadata(handle
,
2906 EXT4_SB(sb
)->s_sbh
);
2909 raw_inode
->i_generation
= cpu_to_le32(inode
->i_generation
);
2910 if (S_ISCHR(inode
->i_mode
) || S_ISBLK(inode
->i_mode
)) {
2911 if (old_valid_dev(inode
->i_rdev
)) {
2912 raw_inode
->i_block
[0] =
2913 cpu_to_le32(old_encode_dev(inode
->i_rdev
));
2914 raw_inode
->i_block
[1] = 0;
2916 raw_inode
->i_block
[0] = 0;
2917 raw_inode
->i_block
[1] =
2918 cpu_to_le32(new_encode_dev(inode
->i_rdev
));
2919 raw_inode
->i_block
[2] = 0;
2921 } else for (block
= 0; block
< EXT4_N_BLOCKS
; block
++)
2922 raw_inode
->i_block
[block
] = ei
->i_data
[block
];
2924 if (ei
->i_extra_isize
)
2925 raw_inode
->i_extra_isize
= cpu_to_le16(ei
->i_extra_isize
);
2927 BUFFER_TRACE(bh
, "call ext4_journal_dirty_metadata");
2928 rc
= ext4_journal_dirty_metadata(handle
, bh
);
2931 ei
->i_state
&= ~EXT4_STATE_NEW
;
2935 ext4_std_error(inode
->i_sb
, err
);
2940 * ext4_write_inode()
2942 * We are called from a few places:
2944 * - Within generic_file_write() for O_SYNC files.
2945 * Here, there will be no transaction running. We wait for any running
2946 * trasnaction to commit.
2948 * - Within sys_sync(), kupdate and such.
2949 * We wait on commit, if tol to.
2951 * - Within prune_icache() (PF_MEMALLOC == true)
2952 * Here we simply return. We can't afford to block kswapd on the
2955 * In all cases it is actually safe for us to return without doing anything,
2956 * because the inode has been copied into a raw inode buffer in
2957 * ext4_mark_inode_dirty(). This is a correctness thing for O_SYNC and for
2960 * Note that we are absolutely dependent upon all inode dirtiers doing the
2961 * right thing: they *must* call mark_inode_dirty() after dirtying info in
2962 * which we are interested.
2964 * It would be a bug for them to not do this. The code:
2966 * mark_inode_dirty(inode)
2968 * inode->i_size = expr;
2970 * is in error because a kswapd-driven write_inode() could occur while
2971 * `stuff()' is running, and the new i_size will be lost. Plus the inode
2972 * will no longer be on the superblock's dirty inode list.
2974 int ext4_write_inode(struct inode
*inode
, int wait
)
2976 if (current
->flags
& PF_MEMALLOC
)
2979 if (ext4_journal_current_handle()) {
2980 jbd_debug(1, "called recursively, non-PF_MEMALLOC!\n");
2988 return ext4_force_commit(inode
->i_sb
);
2994 * Called from notify_change.
2996 * We want to trap VFS attempts to truncate the file as soon as
2997 * possible. In particular, we want to make sure that when the VFS
2998 * shrinks i_size, we put the inode on the orphan list and modify
2999 * i_disksize immediately, so that during the subsequent flushing of
3000 * dirty pages and freeing of disk blocks, we can guarantee that any
3001 * commit will leave the blocks being flushed in an unused state on
3002 * disk. (On recovery, the inode will get truncated and the blocks will
3003 * be freed, so we have a strong guarantee that no future commit will
3004 * leave these blocks visible to the user.)
3006 * Called with inode->sem down.
3008 int ext4_setattr(struct dentry
*dentry
, struct iattr
*attr
)
3010 struct inode
*inode
= dentry
->d_inode
;
3012 const unsigned int ia_valid
= attr
->ia_valid
;
3014 error
= inode_change_ok(inode
, attr
);
3018 if ((ia_valid
& ATTR_UID
&& attr
->ia_uid
!= inode
->i_uid
) ||
3019 (ia_valid
& ATTR_GID
&& attr
->ia_gid
!= inode
->i_gid
)) {
3022 /* (user+group)*(old+new) structure, inode write (sb,
3023 * inode block, ? - but truncate inode update has it) */
3024 handle
= ext4_journal_start(inode
, 2*(EXT4_QUOTA_INIT_BLOCKS(inode
->i_sb
)+
3025 EXT4_QUOTA_DEL_BLOCKS(inode
->i_sb
))+3);
3026 if (IS_ERR(handle
)) {
3027 error
= PTR_ERR(handle
);
3030 error
= DQUOT_TRANSFER(inode
, attr
) ? -EDQUOT
: 0;
3032 ext4_journal_stop(handle
);
3035 /* Update corresponding info in inode so that everything is in
3036 * one transaction */
3037 if (attr
->ia_valid
& ATTR_UID
)
3038 inode
->i_uid
= attr
->ia_uid
;
3039 if (attr
->ia_valid
& ATTR_GID
)
3040 inode
->i_gid
= attr
->ia_gid
;
3041 error
= ext4_mark_inode_dirty(handle
, inode
);
3042 ext4_journal_stop(handle
);
3045 if (attr
->ia_valid
& ATTR_SIZE
) {
3046 if (!(EXT4_I(inode
)->i_flags
& EXT4_EXTENTS_FL
)) {
3047 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
3049 if (attr
->ia_size
> sbi
->s_bitmap_maxbytes
) {
3056 if (S_ISREG(inode
->i_mode
) &&
3057 attr
->ia_valid
& ATTR_SIZE
&& attr
->ia_size
< inode
->i_size
) {
3060 handle
= ext4_journal_start(inode
, 3);
3061 if (IS_ERR(handle
)) {
3062 error
= PTR_ERR(handle
);
3066 error
= ext4_orphan_add(handle
, inode
);
3067 EXT4_I(inode
)->i_disksize
= attr
->ia_size
;
3068 rc
= ext4_mark_inode_dirty(handle
, inode
);
3071 ext4_journal_stop(handle
);
3074 rc
= inode_setattr(inode
, attr
);
3076 /* If inode_setattr's call to ext4_truncate failed to get a
3077 * transaction handle at all, we need to clean up the in-core
3078 * orphan list manually. */
3080 ext4_orphan_del(NULL
, inode
);
3082 if (!rc
&& (ia_valid
& ATTR_MODE
))
3083 rc
= ext4_acl_chmod(inode
);
3086 ext4_std_error(inode
->i_sb
, error
);
3094 * How many blocks doth make a writepage()?
3096 * With N blocks per page, it may be:
3101 * N+5 bitmap blocks (from the above)
3102 * N+5 group descriptor summary blocks
3105 * 2 * EXT4_SINGLEDATA_TRANS_BLOCKS for the quote files
3107 * 3 * (N + 5) + 2 + 2 * EXT4_SINGLEDATA_TRANS_BLOCKS
3109 * With ordered or writeback data it's the same, less the N data blocks.
3111 * If the inode's direct blocks can hold an integral number of pages then a
3112 * page cannot straddle two indirect blocks, and we can only touch one indirect
3113 * and dindirect block, and the "5" above becomes "3".
3115 * This still overestimates under most circumstances. If we were to pass the
3116 * start and end offsets in here as well we could do block_to_path() on each
3117 * block and work out the exact number of indirects which are touched. Pah.
3120 int ext4_writepage_trans_blocks(struct inode
*inode
)
3122 int bpp
= ext4_journal_blocks_per_page(inode
);
3123 int indirects
= (EXT4_NDIR_BLOCKS
% bpp
) ? 5 : 3;
3126 if (EXT4_I(inode
)->i_flags
& EXT4_EXTENTS_FL
)
3127 return ext4_ext_writepage_trans_blocks(inode
, bpp
);
3129 if (ext4_should_journal_data(inode
))
3130 ret
= 3 * (bpp
+ indirects
) + 2;
3132 ret
= 2 * (bpp
+ indirects
) + 2;
3135 /* We know that structure was already allocated during DQUOT_INIT so
3136 * we will be updating only the data blocks + inodes */
3137 ret
+= 2*EXT4_QUOTA_TRANS_BLOCKS(inode
->i_sb
);
3144 * The caller must have previously called ext4_reserve_inode_write().
3145 * Give this, we know that the caller already has write access to iloc->bh.
3147 int ext4_mark_iloc_dirty(handle_t
*handle
,
3148 struct inode
*inode
, struct ext4_iloc
*iloc
)
3152 /* the do_update_inode consumes one bh->b_count */
3155 /* ext4_do_update_inode() does jbd2_journal_dirty_metadata */
3156 err
= ext4_do_update_inode(handle
, inode
, iloc
);
3162 * On success, We end up with an outstanding reference count against
3163 * iloc->bh. This _must_ be cleaned up later.
3167 ext4_reserve_inode_write(handle_t
*handle
, struct inode
*inode
,
3168 struct ext4_iloc
*iloc
)
3172 err
= ext4_get_inode_loc(inode
, iloc
);
3174 BUFFER_TRACE(iloc
->bh
, "get_write_access");
3175 err
= ext4_journal_get_write_access(handle
, iloc
->bh
);
3182 ext4_std_error(inode
->i_sb
, err
);
3187 * Expand an inode by new_extra_isize bytes.
3188 * Returns 0 on success or negative error number on failure.
3190 static int ext4_expand_extra_isize(struct inode
*inode
,
3191 unsigned int new_extra_isize
,
3192 struct ext4_iloc iloc
,
3195 struct ext4_inode
*raw_inode
;
3196 struct ext4_xattr_ibody_header
*header
;
3197 struct ext4_xattr_entry
*entry
;
3199 if (EXT4_I(inode
)->i_extra_isize
>= new_extra_isize
)
3202 raw_inode
= ext4_raw_inode(&iloc
);
3204 header
= IHDR(inode
, raw_inode
);
3205 entry
= IFIRST(header
);
3207 /* No extended attributes present */
3208 if (!(EXT4_I(inode
)->i_state
& EXT4_STATE_XATTR
) ||
3209 header
->h_magic
!= cpu_to_le32(EXT4_XATTR_MAGIC
)) {
3210 memset((void *)raw_inode
+ EXT4_GOOD_OLD_INODE_SIZE
, 0,
3212 EXT4_I(inode
)->i_extra_isize
= new_extra_isize
;
3216 /* try to expand with EAs present */
3217 return ext4_expand_extra_isize_ea(inode
, new_extra_isize
,
3222 * What we do here is to mark the in-core inode as clean with respect to inode
3223 * dirtiness (it may still be data-dirty).
3224 * This means that the in-core inode may be reaped by prune_icache
3225 * without having to perform any I/O. This is a very good thing,
3226 * because *any* task may call prune_icache - even ones which
3227 * have a transaction open against a different journal.
3229 * Is this cheating? Not really. Sure, we haven't written the
3230 * inode out, but prune_icache isn't a user-visible syncing function.
3231 * Whenever the user wants stuff synced (sys_sync, sys_msync, sys_fsync)
3232 * we start and wait on commits.
3234 * Is this efficient/effective? Well, we're being nice to the system
3235 * by cleaning up our inodes proactively so they can be reaped
3236 * without I/O. But we are potentially leaving up to five seconds'
3237 * worth of inodes floating about which prune_icache wants us to
3238 * write out. One way to fix that would be to get prune_icache()
3239 * to do a write_super() to free up some memory. It has the desired
3242 int ext4_mark_inode_dirty(handle_t
*handle
, struct inode
*inode
)
3244 struct ext4_iloc iloc
;
3245 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
3246 static unsigned int mnt_count
;
3250 err
= ext4_reserve_inode_write(handle
, inode
, &iloc
);
3251 if (EXT4_I(inode
)->i_extra_isize
< sbi
->s_want_extra_isize
&&
3252 !(EXT4_I(inode
)->i_state
& EXT4_STATE_NO_EXPAND
)) {
3254 * We need extra buffer credits since we may write into EA block
3255 * with this same handle. If journal_extend fails, then it will
3256 * only result in a minor loss of functionality for that inode.
3257 * If this is felt to be critical, then e2fsck should be run to
3258 * force a large enough s_min_extra_isize.
3260 if ((jbd2_journal_extend(handle
,
3261 EXT4_DATA_TRANS_BLOCKS(inode
->i_sb
))) == 0) {
3262 ret
= ext4_expand_extra_isize(inode
,
3263 sbi
->s_want_extra_isize
,
3266 EXT4_I(inode
)->i_state
|= EXT4_STATE_NO_EXPAND
;
3268 le16_to_cpu(sbi
->s_es
->s_mnt_count
)) {
3269 ext4_warning(inode
->i_sb
, __FUNCTION__
,
3270 "Unable to expand inode %lu. Delete"
3271 " some EAs or run e2fsck.",
3274 le16_to_cpu(sbi
->s_es
->s_mnt_count
);
3280 err
= ext4_mark_iloc_dirty(handle
, inode
, &iloc
);
3285 * ext4_dirty_inode() is called from __mark_inode_dirty()
3287 * We're really interested in the case where a file is being extended.
3288 * i_size has been changed by generic_commit_write() and we thus need
3289 * to include the updated inode in the current transaction.
3291 * Also, DQUOT_ALLOC_SPACE() will always dirty the inode when blocks
3292 * are allocated to the file.
3294 * If the inode is marked synchronous, we don't honour that here - doing
3295 * so would cause a commit on atime updates, which we don't bother doing.
3296 * We handle synchronous inodes at the highest possible level.
3298 void ext4_dirty_inode(struct inode
*inode
)
3300 handle_t
*current_handle
= ext4_journal_current_handle();
3303 handle
= ext4_journal_start(inode
, 2);
3306 if (current_handle
&&
3307 current_handle
->h_transaction
!= handle
->h_transaction
) {
3308 /* This task has a transaction open against a different fs */
3309 printk(KERN_EMERG
"%s: transactions do not match!\n",
3312 jbd_debug(5, "marking dirty. outer handle=%p\n",
3314 ext4_mark_inode_dirty(handle
, inode
);
3316 ext4_journal_stop(handle
);
3323 * Bind an inode's backing buffer_head into this transaction, to prevent
3324 * it from being flushed to disk early. Unlike
3325 * ext4_reserve_inode_write, this leaves behind no bh reference and
3326 * returns no iloc structure, so the caller needs to repeat the iloc
3327 * lookup to mark the inode dirty later.
3329 static int ext4_pin_inode(handle_t
*handle
, struct inode
*inode
)
3331 struct ext4_iloc iloc
;
3335 err
= ext4_get_inode_loc(inode
, &iloc
);
3337 BUFFER_TRACE(iloc
.bh
, "get_write_access");
3338 err
= jbd2_journal_get_write_access(handle
, iloc
.bh
);
3340 err
= ext4_journal_dirty_metadata(handle
,
3345 ext4_std_error(inode
->i_sb
, err
);
3350 int ext4_change_inode_journal_flag(struct inode
*inode
, int val
)
3357 * We have to be very careful here: changing a data block's
3358 * journaling status dynamically is dangerous. If we write a
3359 * data block to the journal, change the status and then delete
3360 * that block, we risk forgetting to revoke the old log record
3361 * from the journal and so a subsequent replay can corrupt data.
3362 * So, first we make sure that the journal is empty and that
3363 * nobody is changing anything.
3366 journal
= EXT4_JOURNAL(inode
);
3367 if (is_journal_aborted(journal
))
3370 jbd2_journal_lock_updates(journal
);
3371 jbd2_journal_flush(journal
);
3374 * OK, there are no updates running now, and all cached data is
3375 * synced to disk. We are now in a completely consistent state
3376 * which doesn't have anything in the journal, and we know that
3377 * no filesystem updates are running, so it is safe to modify
3378 * the inode's in-core data-journaling state flag now.
3382 EXT4_I(inode
)->i_flags
|= EXT4_JOURNAL_DATA_FL
;
3384 EXT4_I(inode
)->i_flags
&= ~EXT4_JOURNAL_DATA_FL
;
3385 ext4_set_aops(inode
);
3387 jbd2_journal_unlock_updates(journal
);
3389 /* Finally we can mark the inode as dirty. */
3391 handle
= ext4_journal_start(inode
, 1);
3393 return PTR_ERR(handle
);
3395 err
= ext4_mark_inode_dirty(handle
, inode
);
3397 ext4_journal_stop(handle
);
3398 ext4_std_error(inode
->i_sb
, err
);