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/jbd2.h>
29 #include <linux/highuid.h>
30 #include <linux/pagemap.h>
31 #include <linux/quotaops.h>
32 #include <linux/string.h>
33 #include <linux/buffer_head.h>
34 #include <linux/writeback.h>
35 #include <linux/pagevec.h>
36 #include <linux/mpage.h>
37 #include <linux/uio.h>
38 #include <linux/bio.h>
39 #include "ext4_jbd2.h"
42 #include "ext4_extents.h"
44 static inline int ext4_begin_ordered_truncate(struct inode
*inode
,
47 return jbd2_journal_begin_ordered_truncate(&EXT4_I(inode
)->jinode
,
51 static void ext4_invalidatepage(struct page
*page
, unsigned long offset
);
54 * Test whether an inode is a fast symlink.
56 static int ext4_inode_is_fast_symlink(struct inode
*inode
)
58 int ea_blocks
= EXT4_I(inode
)->i_file_acl
?
59 (inode
->i_sb
->s_blocksize
>> 9) : 0;
61 return (S_ISLNK(inode
->i_mode
) && inode
->i_blocks
- ea_blocks
== 0);
65 * The ext4 forget function must perform a revoke if we are freeing data
66 * which has been journaled. Metadata (eg. indirect blocks) must be
67 * revoked in all cases.
69 * "bh" may be NULL: a metadata block may have been freed from memory
70 * but there may still be a record of it in the journal, and that record
71 * still needs to be revoked.
73 int ext4_forget(handle_t
*handle
, int is_metadata
, struct inode
*inode
,
74 struct buffer_head
*bh
, ext4_fsblk_t blocknr
)
80 BUFFER_TRACE(bh
, "enter");
82 jbd_debug(4, "forgetting bh %p: is_metadata = %d, mode %o, "
84 bh
, is_metadata
, inode
->i_mode
,
85 test_opt(inode
->i_sb
, DATA_FLAGS
));
87 /* Never use the revoke function if we are doing full data
88 * journaling: there is no need to, and a V1 superblock won't
89 * support it. Otherwise, only skip the revoke on un-journaled
92 if (test_opt(inode
->i_sb
, DATA_FLAGS
) == EXT4_MOUNT_JOURNAL_DATA
||
93 (!is_metadata
&& !ext4_should_journal_data(inode
))) {
95 BUFFER_TRACE(bh
, "call jbd2_journal_forget");
96 return ext4_journal_forget(handle
, bh
);
102 * data!=journal && (is_metadata || should_journal_data(inode))
104 BUFFER_TRACE(bh
, "call ext4_journal_revoke");
105 err
= ext4_journal_revoke(handle
, blocknr
, bh
);
107 ext4_abort(inode
->i_sb
, __func__
,
108 "error %d when attempting revoke", err
);
109 BUFFER_TRACE(bh
, "exit");
114 * Work out how many blocks we need to proceed with the next chunk of a
115 * truncate transaction.
117 static unsigned long blocks_for_truncate(struct inode
*inode
)
121 needed
= inode
->i_blocks
>> (inode
->i_sb
->s_blocksize_bits
- 9);
123 /* Give ourselves just enough room to cope with inodes in which
124 * i_blocks is corrupt: we've seen disk corruptions in the past
125 * which resulted in random data in an inode which looked enough
126 * like a regular file for ext4 to try to delete it. Things
127 * will go a bit crazy if that happens, but at least we should
128 * try not to panic the whole kernel. */
132 /* But we need to bound the transaction so we don't overflow the
134 if (needed
> EXT4_MAX_TRANS_DATA
)
135 needed
= EXT4_MAX_TRANS_DATA
;
137 return EXT4_DATA_TRANS_BLOCKS(inode
->i_sb
) + needed
;
141 * Truncate transactions can be complex and absolutely huge. So we need to
142 * be able to restart the transaction at a conventient checkpoint to make
143 * sure we don't overflow the journal.
145 * start_transaction gets us a new handle for a truncate transaction,
146 * and extend_transaction tries to extend the existing one a bit. If
147 * extend fails, we need to propagate the failure up and restart the
148 * transaction in the top-level truncate loop. --sct
150 static handle_t
*start_transaction(struct inode
*inode
)
154 result
= ext4_journal_start(inode
, blocks_for_truncate(inode
));
158 ext4_std_error(inode
->i_sb
, PTR_ERR(result
));
163 * Try to extend this transaction for the purposes of truncation.
165 * Returns 0 if we managed to create more room. If we can't create more
166 * room, and the transaction must be restarted we return 1.
168 static int try_to_extend_transaction(handle_t
*handle
, struct inode
*inode
)
170 if (handle
->h_buffer_credits
> EXT4_RESERVE_TRANS_BLOCKS
)
172 if (!ext4_journal_extend(handle
, blocks_for_truncate(inode
)))
178 * Restart the transaction associated with *handle. This does a commit,
179 * so before we call here everything must be consistently dirtied against
182 static int ext4_journal_test_restart(handle_t
*handle
, struct inode
*inode
)
184 jbd_debug(2, "restarting handle %p\n", handle
);
185 return ext4_journal_restart(handle
, blocks_for_truncate(inode
));
189 * Called at the last iput() if i_nlink is zero.
191 void ext4_delete_inode (struct inode
* inode
)
196 if (ext4_should_order_data(inode
))
197 ext4_begin_ordered_truncate(inode
, 0);
198 truncate_inode_pages(&inode
->i_data
, 0);
200 if (is_bad_inode(inode
))
203 handle
= ext4_journal_start(inode
, blocks_for_truncate(inode
)+3);
204 if (IS_ERR(handle
)) {
205 ext4_std_error(inode
->i_sb
, PTR_ERR(handle
));
207 * If we're going to skip the normal cleanup, we still need to
208 * make sure that the in-core orphan linked list is properly
211 ext4_orphan_del(NULL
, inode
);
218 err
= ext4_mark_inode_dirty(handle
, inode
);
220 ext4_warning(inode
->i_sb
, __func__
,
221 "couldn't mark inode dirty (err %d)", err
);
225 ext4_truncate(inode
);
228 * ext4_ext_truncate() doesn't reserve any slop when it
229 * restarts journal transactions; therefore there may not be
230 * enough credits left in the handle to remove the inode from
231 * the orphan list and set the dtime field.
233 if (handle
->h_buffer_credits
< 3) {
234 err
= ext4_journal_extend(handle
, 3);
236 err
= ext4_journal_restart(handle
, 3);
238 ext4_warning(inode
->i_sb
, __func__
,
239 "couldn't extend journal (err %d)", err
);
241 ext4_journal_stop(handle
);
247 * Kill off the orphan record which ext4_truncate created.
248 * AKPM: I think this can be inside the above `if'.
249 * Note that ext4_orphan_del() has to be able to cope with the
250 * deletion of a non-existent orphan - this is because we don't
251 * know if ext4_truncate() actually created an orphan record.
252 * (Well, we could do this if we need to, but heck - it works)
254 ext4_orphan_del(handle
, inode
);
255 EXT4_I(inode
)->i_dtime
= get_seconds();
258 * One subtle ordering requirement: if anything has gone wrong
259 * (transaction abort, IO errors, whatever), then we can still
260 * do these next steps (the fs will already have been marked as
261 * having errors), but we can't free the inode if the mark_dirty
264 if (ext4_mark_inode_dirty(handle
, inode
))
265 /* If that failed, just do the required in-core inode clear. */
268 ext4_free_inode(handle
, inode
);
269 ext4_journal_stop(handle
);
272 clear_inode(inode
); /* We must guarantee clearing of inode... */
278 struct buffer_head
*bh
;
281 static inline void add_chain(Indirect
*p
, struct buffer_head
*bh
, __le32
*v
)
283 p
->key
= *(p
->p
= v
);
288 * ext4_block_to_path - parse the block number into array of offsets
289 * @inode: inode in question (we are only interested in its superblock)
290 * @i_block: block number to be parsed
291 * @offsets: array to store the offsets in
292 * @boundary: set this non-zero if the referred-to block is likely to be
293 * followed (on disk) by an indirect block.
295 * To store the locations of file's data ext4 uses a data structure common
296 * for UNIX filesystems - tree of pointers anchored in the inode, with
297 * data blocks at leaves and indirect blocks in intermediate nodes.
298 * This function translates the block number into path in that tree -
299 * return value is the path length and @offsets[n] is the offset of
300 * pointer to (n+1)th node in the nth one. If @block is out of range
301 * (negative or too large) warning is printed and zero returned.
303 * Note: function doesn't find node addresses, so no IO is needed. All
304 * we need to know is the capacity of indirect blocks (taken from the
309 * Portability note: the last comparison (check that we fit into triple
310 * indirect block) is spelled differently, because otherwise on an
311 * architecture with 32-bit longs and 8Kb pages we might get into trouble
312 * if our filesystem had 8Kb blocks. We might use long long, but that would
313 * kill us on x86. Oh, well, at least the sign propagation does not matter -
314 * i_block would have to be negative in the very beginning, so we would not
318 static int ext4_block_to_path(struct inode
*inode
,
320 ext4_lblk_t offsets
[4], int *boundary
)
322 int ptrs
= EXT4_ADDR_PER_BLOCK(inode
->i_sb
);
323 int ptrs_bits
= EXT4_ADDR_PER_BLOCK_BITS(inode
->i_sb
);
324 const long direct_blocks
= EXT4_NDIR_BLOCKS
,
325 indirect_blocks
= ptrs
,
326 double_blocks
= (1 << (ptrs_bits
* 2));
331 ext4_warning (inode
->i_sb
, "ext4_block_to_path", "block < 0");
332 } else if (i_block
< direct_blocks
) {
333 offsets
[n
++] = i_block
;
334 final
= direct_blocks
;
335 } else if ( (i_block
-= direct_blocks
) < indirect_blocks
) {
336 offsets
[n
++] = EXT4_IND_BLOCK
;
337 offsets
[n
++] = i_block
;
339 } else if ((i_block
-= indirect_blocks
) < double_blocks
) {
340 offsets
[n
++] = EXT4_DIND_BLOCK
;
341 offsets
[n
++] = i_block
>> ptrs_bits
;
342 offsets
[n
++] = i_block
& (ptrs
- 1);
344 } else if (((i_block
-= double_blocks
) >> (ptrs_bits
* 2)) < ptrs
) {
345 offsets
[n
++] = EXT4_TIND_BLOCK
;
346 offsets
[n
++] = i_block
>> (ptrs_bits
* 2);
347 offsets
[n
++] = (i_block
>> ptrs_bits
) & (ptrs
- 1);
348 offsets
[n
++] = i_block
& (ptrs
- 1);
351 ext4_warning(inode
->i_sb
, "ext4_block_to_path",
353 i_block
+ direct_blocks
+
354 indirect_blocks
+ double_blocks
);
357 *boundary
= final
- 1 - (i_block
& (ptrs
- 1));
362 * ext4_get_branch - read the chain of indirect blocks leading to data
363 * @inode: inode in question
364 * @depth: depth of the chain (1 - direct pointer, etc.)
365 * @offsets: offsets of pointers in inode/indirect blocks
366 * @chain: place to store the result
367 * @err: here we store the error value
369 * Function fills the array of triples <key, p, bh> and returns %NULL
370 * if everything went OK or the pointer to the last filled triple
371 * (incomplete one) otherwise. Upon the return chain[i].key contains
372 * the number of (i+1)-th block in the chain (as it is stored in memory,
373 * i.e. little-endian 32-bit), chain[i].p contains the address of that
374 * number (it points into struct inode for i==0 and into the bh->b_data
375 * for i>0) and chain[i].bh points to the buffer_head of i-th indirect
376 * block for i>0 and NULL for i==0. In other words, it holds the block
377 * numbers of the chain, addresses they were taken from (and where we can
378 * verify that chain did not change) and buffer_heads hosting these
381 * Function stops when it stumbles upon zero pointer (absent block)
382 * (pointer to last triple returned, *@err == 0)
383 * or when it gets an IO error reading an indirect block
384 * (ditto, *@err == -EIO)
385 * or when it reads all @depth-1 indirect blocks successfully and finds
386 * the whole chain, all way to the data (returns %NULL, *err == 0).
388 * Need to be called with
389 * down_read(&EXT4_I(inode)->i_data_sem)
391 static Indirect
*ext4_get_branch(struct inode
*inode
, int depth
,
392 ext4_lblk_t
*offsets
,
393 Indirect chain
[4], int *err
)
395 struct super_block
*sb
= inode
->i_sb
;
397 struct buffer_head
*bh
;
400 /* i_data is not going away, no lock needed */
401 add_chain (chain
, NULL
, EXT4_I(inode
)->i_data
+ *offsets
);
405 bh
= sb_bread(sb
, le32_to_cpu(p
->key
));
408 add_chain(++p
, bh
, (__le32
*)bh
->b_data
+ *++offsets
);
422 * ext4_find_near - find a place for allocation with sufficient locality
424 * @ind: descriptor of indirect block.
426 * This function returns the preferred place for block allocation.
427 * It is used when heuristic for sequential allocation fails.
429 * + if there is a block to the left of our position - allocate near it.
430 * + if pointer will live in indirect block - allocate near that block.
431 * + if pointer will live in inode - allocate in the same
434 * In the latter case we colour the starting block by the callers PID to
435 * prevent it from clashing with concurrent allocations for a different inode
436 * in the same block group. The PID is used here so that functionally related
437 * files will be close-by on-disk.
439 * Caller must make sure that @ind is valid and will stay that way.
441 static ext4_fsblk_t
ext4_find_near(struct inode
*inode
, Indirect
*ind
)
443 struct ext4_inode_info
*ei
= EXT4_I(inode
);
444 __le32
*start
= ind
->bh
? (__le32
*) ind
->bh
->b_data
: ei
->i_data
;
446 ext4_fsblk_t bg_start
;
447 ext4_fsblk_t last_block
;
448 ext4_grpblk_t colour
;
450 /* Try to find previous block */
451 for (p
= ind
->p
- 1; p
>= start
; p
--) {
453 return le32_to_cpu(*p
);
456 /* No such thing, so let's try location of indirect block */
458 return ind
->bh
->b_blocknr
;
461 * It is going to be referred to from the inode itself? OK, just put it
462 * into the same cylinder group then.
464 bg_start
= ext4_group_first_block_no(inode
->i_sb
, ei
->i_block_group
);
465 last_block
= ext4_blocks_count(EXT4_SB(inode
->i_sb
)->s_es
) - 1;
467 if (bg_start
+ EXT4_BLOCKS_PER_GROUP(inode
->i_sb
) <= last_block
)
468 colour
= (current
->pid
% 16) *
469 (EXT4_BLOCKS_PER_GROUP(inode
->i_sb
) / 16);
471 colour
= (current
->pid
% 16) * ((last_block
- bg_start
) / 16);
472 return bg_start
+ colour
;
476 * ext4_find_goal - find a preferred place for allocation.
478 * @block: block we want
479 * @partial: pointer to the last triple within a chain
481 * Normally this function find the preferred place for block allocation,
484 static ext4_fsblk_t
ext4_find_goal(struct inode
*inode
, ext4_lblk_t block
,
487 struct ext4_block_alloc_info
*block_i
;
489 block_i
= EXT4_I(inode
)->i_block_alloc_info
;
492 * try the heuristic for sequential allocation,
493 * failing that at least try to get decent locality.
495 if (block_i
&& (block
== block_i
->last_alloc_logical_block
+ 1)
496 && (block_i
->last_alloc_physical_block
!= 0)) {
497 return block_i
->last_alloc_physical_block
+ 1;
500 return ext4_find_near(inode
, partial
);
504 * ext4_blks_to_allocate: Look up the block map and count the number
505 * of direct blocks need to be allocated for the given branch.
507 * @branch: chain of indirect blocks
508 * @k: number of blocks need for indirect blocks
509 * @blks: number of data blocks to be mapped.
510 * @blocks_to_boundary: the offset in the indirect block
512 * return the total number of blocks to be allocate, including the
513 * direct and indirect blocks.
515 static int ext4_blks_to_allocate(Indirect
*branch
, int k
, unsigned long blks
,
516 int blocks_to_boundary
)
518 unsigned long count
= 0;
521 * Simple case, [t,d]Indirect block(s) has not allocated yet
522 * then it's clear blocks on that path have not allocated
525 /* right now we don't handle cross boundary allocation */
526 if (blks
< blocks_to_boundary
+ 1)
529 count
+= blocks_to_boundary
+ 1;
534 while (count
< blks
&& count
<= blocks_to_boundary
&&
535 le32_to_cpu(*(branch
[0].p
+ count
)) == 0) {
542 * ext4_alloc_blocks: multiple allocate blocks needed for a branch
543 * @indirect_blks: the number of blocks need to allocate for indirect
546 * @new_blocks: on return it will store the new block numbers for
547 * the indirect blocks(if needed) and the first direct block,
548 * @blks: on return it will store the total number of allocated
551 static int ext4_alloc_blocks(handle_t
*handle
, struct inode
*inode
,
552 ext4_lblk_t iblock
, ext4_fsblk_t goal
,
553 int indirect_blks
, int blks
,
554 ext4_fsblk_t new_blocks
[4], int *err
)
557 unsigned long count
= 0, blk_allocated
= 0;
559 ext4_fsblk_t current_block
= 0;
563 * Here we try to allocate the requested multiple blocks at once,
564 * on a best-effort basis.
565 * To build a branch, we should allocate blocks for
566 * the indirect blocks(if not allocated yet), and at least
567 * the first direct block of this branch. That's the
568 * minimum number of blocks need to allocate(required)
570 /* first we try to allocate the indirect blocks */
571 target
= indirect_blks
;
574 /* allocating blocks for indirect blocks and direct blocks */
575 current_block
= ext4_new_meta_blocks(handle
, inode
,
581 /* allocate blocks for indirect blocks */
582 while (index
< indirect_blks
&& count
) {
583 new_blocks
[index
++] = current_block
++;
588 * save the new block number
589 * for the first direct block
591 new_blocks
[index
] = current_block
;
592 printk(KERN_INFO
"%s returned more blocks than "
593 "requested\n", __func__
);
599 target
= blks
- count
;
600 blk_allocated
= count
;
603 /* Now allocate data blocks */
605 /* allocating blocks for data blocks */
606 current_block
= ext4_new_blocks(handle
, inode
, iblock
,
608 if (*err
&& (target
== blks
)) {
610 * if the allocation failed and we didn't allocate
616 if (target
== blks
) {
618 * save the new block number
619 * for the first direct block
621 new_blocks
[index
] = current_block
;
623 blk_allocated
+= count
;
626 /* total number of blocks allocated for direct blocks */
631 for (i
= 0; i
<index
; i
++)
632 ext4_free_blocks(handle
, inode
, new_blocks
[i
], 1, 0);
637 * ext4_alloc_branch - allocate and set up a chain of blocks.
639 * @indirect_blks: number of allocated indirect blocks
640 * @blks: number of allocated direct blocks
641 * @offsets: offsets (in the blocks) to store the pointers to next.
642 * @branch: place to store the chain in.
644 * This function allocates blocks, zeroes out all but the last one,
645 * links them into chain and (if we are synchronous) writes them to disk.
646 * In other words, it prepares a branch that can be spliced onto the
647 * inode. It stores the information about that chain in the branch[], in
648 * the same format as ext4_get_branch() would do. We are calling it after
649 * we had read the existing part of chain and partial points to the last
650 * triple of that (one with zero ->key). Upon the exit we have the same
651 * picture as after the successful ext4_get_block(), except that in one
652 * place chain is disconnected - *branch->p is still zero (we did not
653 * set the last link), but branch->key contains the number that should
654 * be placed into *branch->p to fill that gap.
656 * If allocation fails we free all blocks we've allocated (and forget
657 * their buffer_heads) and return the error value the from failed
658 * ext4_alloc_block() (normally -ENOSPC). Otherwise we set the chain
659 * as described above and return 0.
661 static int ext4_alloc_branch(handle_t
*handle
, struct inode
*inode
,
662 ext4_lblk_t iblock
, int indirect_blks
,
663 int *blks
, ext4_fsblk_t goal
,
664 ext4_lblk_t
*offsets
, Indirect
*branch
)
666 int blocksize
= inode
->i_sb
->s_blocksize
;
669 struct buffer_head
*bh
;
671 ext4_fsblk_t new_blocks
[4];
672 ext4_fsblk_t current_block
;
674 num
= ext4_alloc_blocks(handle
, inode
, iblock
, goal
, indirect_blks
,
675 *blks
, new_blocks
, &err
);
679 branch
[0].key
= cpu_to_le32(new_blocks
[0]);
681 * metadata blocks and data blocks are allocated.
683 for (n
= 1; n
<= indirect_blks
; n
++) {
685 * Get buffer_head for parent block, zero it out
686 * and set the pointer to new one, then send
689 bh
= sb_getblk(inode
->i_sb
, new_blocks
[n
-1]);
692 BUFFER_TRACE(bh
, "call get_create_access");
693 err
= ext4_journal_get_create_access(handle
, bh
);
700 memset(bh
->b_data
, 0, blocksize
);
701 branch
[n
].p
= (__le32
*) bh
->b_data
+ offsets
[n
];
702 branch
[n
].key
= cpu_to_le32(new_blocks
[n
]);
703 *branch
[n
].p
= branch
[n
].key
;
704 if ( n
== indirect_blks
) {
705 current_block
= new_blocks
[n
];
707 * End of chain, update the last new metablock of
708 * the chain to point to the new allocated
709 * data blocks numbers
711 for (i
=1; i
< num
; i
++)
712 *(branch
[n
].p
+ i
) = cpu_to_le32(++current_block
);
714 BUFFER_TRACE(bh
, "marking uptodate");
715 set_buffer_uptodate(bh
);
718 BUFFER_TRACE(bh
, "call ext4_journal_dirty_metadata");
719 err
= ext4_journal_dirty_metadata(handle
, bh
);
726 /* Allocation failed, free what we already allocated */
727 for (i
= 1; i
<= n
; i
++) {
728 BUFFER_TRACE(branch
[i
].bh
, "call jbd2_journal_forget");
729 ext4_journal_forget(handle
, branch
[i
].bh
);
731 for (i
= 0; i
<indirect_blks
; i
++)
732 ext4_free_blocks(handle
, inode
, new_blocks
[i
], 1, 0);
734 ext4_free_blocks(handle
, inode
, new_blocks
[i
], num
, 0);
740 * ext4_splice_branch - splice the allocated branch onto inode.
742 * @block: (logical) number of block we are adding
743 * @chain: chain of indirect blocks (with a missing link - see
745 * @where: location of missing link
746 * @num: number of indirect blocks we are adding
747 * @blks: number of direct blocks we are adding
749 * This function fills the missing link and does all housekeeping needed in
750 * inode (->i_blocks, etc.). In case of success we end up with the full
751 * chain to new block and return 0.
753 static int ext4_splice_branch(handle_t
*handle
, struct inode
*inode
,
754 ext4_lblk_t block
, Indirect
*where
, int num
, int blks
)
758 struct ext4_block_alloc_info
*block_i
;
759 ext4_fsblk_t current_block
;
761 block_i
= EXT4_I(inode
)->i_block_alloc_info
;
763 * If we're splicing into a [td]indirect block (as opposed to the
764 * inode) then we need to get write access to the [td]indirect block
768 BUFFER_TRACE(where
->bh
, "get_write_access");
769 err
= ext4_journal_get_write_access(handle
, where
->bh
);
775 *where
->p
= where
->key
;
778 * Update the host buffer_head or inode to point to more just allocated
779 * direct blocks blocks
781 if (num
== 0 && blks
> 1) {
782 current_block
= le32_to_cpu(where
->key
) + 1;
783 for (i
= 1; i
< blks
; i
++)
784 *(where
->p
+ i
) = cpu_to_le32(current_block
++);
788 * update the most recently allocated logical & physical block
789 * in i_block_alloc_info, to assist find the proper goal block for next
793 block_i
->last_alloc_logical_block
= block
+ blks
- 1;
794 block_i
->last_alloc_physical_block
=
795 le32_to_cpu(where
[num
].key
) + blks
- 1;
798 /* We are done with atomic stuff, now do the rest of housekeeping */
800 inode
->i_ctime
= ext4_current_time(inode
);
801 ext4_mark_inode_dirty(handle
, inode
);
803 /* had we spliced it onto indirect block? */
806 * If we spliced it onto an indirect block, we haven't
807 * altered the inode. Note however that if it is being spliced
808 * onto an indirect block at the very end of the file (the
809 * file is growing) then we *will* alter the inode to reflect
810 * the new i_size. But that is not done here - it is done in
811 * generic_commit_write->__mark_inode_dirty->ext4_dirty_inode.
813 jbd_debug(5, "splicing indirect only\n");
814 BUFFER_TRACE(where
->bh
, "call ext4_journal_dirty_metadata");
815 err
= ext4_journal_dirty_metadata(handle
, where
->bh
);
820 * OK, we spliced it into the inode itself on a direct block.
821 * Inode was dirtied above.
823 jbd_debug(5, "splicing direct\n");
828 for (i
= 1; i
<= num
; i
++) {
829 BUFFER_TRACE(where
[i
].bh
, "call jbd2_journal_forget");
830 ext4_journal_forget(handle
, where
[i
].bh
);
831 ext4_free_blocks(handle
, inode
,
832 le32_to_cpu(where
[i
-1].key
), 1, 0);
834 ext4_free_blocks(handle
, inode
, le32_to_cpu(where
[num
].key
), blks
, 0);
840 * Allocation strategy is simple: if we have to allocate something, we will
841 * have to go the whole way to leaf. So let's do it before attaching anything
842 * to tree, set linkage between the newborn blocks, write them if sync is
843 * required, recheck the path, free and repeat if check fails, otherwise
844 * set the last missing link (that will protect us from any truncate-generated
845 * removals - all blocks on the path are immune now) and possibly force the
846 * write on the parent block.
847 * That has a nice additional property: no special recovery from the failed
848 * allocations is needed - we simply release blocks and do not touch anything
849 * reachable from inode.
851 * `handle' can be NULL if create == 0.
853 * return > 0, # of blocks mapped or allocated.
854 * return = 0, if plain lookup failed.
855 * return < 0, error case.
858 * Need to be called with
859 * down_read(&EXT4_I(inode)->i_data_sem) if not allocating file system block
860 * (ie, create is zero). Otherwise down_write(&EXT4_I(inode)->i_data_sem)
862 int ext4_get_blocks_handle(handle_t
*handle
, struct inode
*inode
,
863 ext4_lblk_t iblock
, unsigned long maxblocks
,
864 struct buffer_head
*bh_result
,
865 int create
, int extend_disksize
)
868 ext4_lblk_t offsets
[4];
873 int blocks_to_boundary
= 0;
875 struct ext4_inode_info
*ei
= EXT4_I(inode
);
877 ext4_fsblk_t first_block
= 0;
881 J_ASSERT(!(EXT4_I(inode
)->i_flags
& EXT4_EXTENTS_FL
));
882 J_ASSERT(handle
!= NULL
|| create
== 0);
883 depth
= ext4_block_to_path(inode
, iblock
, offsets
,
884 &blocks_to_boundary
);
889 partial
= ext4_get_branch(inode
, depth
, offsets
, chain
, &err
);
891 /* Simplest case - block found, no allocation needed */
893 first_block
= le32_to_cpu(chain
[depth
- 1].key
);
894 clear_buffer_new(bh_result
);
897 while (count
< maxblocks
&& count
<= blocks_to_boundary
) {
900 blk
= le32_to_cpu(*(chain
[depth
-1].p
+ count
));
902 if (blk
== first_block
+ count
)
910 /* Next simple case - plain lookup or failed read of indirect block */
911 if (!create
|| err
== -EIO
)
915 * Okay, we need to do block allocation. Lazily initialize the block
916 * allocation info here if necessary
918 if (S_ISREG(inode
->i_mode
) && (!ei
->i_block_alloc_info
))
919 ext4_init_block_alloc_info(inode
);
921 goal
= ext4_find_goal(inode
, iblock
, partial
);
923 /* the number of blocks need to allocate for [d,t]indirect blocks */
924 indirect_blks
= (chain
+ depth
) - partial
- 1;
927 * Next look up the indirect map to count the totoal number of
928 * direct blocks to allocate for this branch.
930 count
= ext4_blks_to_allocate(partial
, indirect_blks
,
931 maxblocks
, blocks_to_boundary
);
933 * Block out ext4_truncate while we alter the tree
935 err
= ext4_alloc_branch(handle
, inode
, iblock
, indirect_blks
,
937 offsets
+ (partial
- chain
), partial
);
940 * The ext4_splice_branch call will free and forget any buffers
941 * on the new chain if there is a failure, but that risks using
942 * up transaction credits, especially for bitmaps where the
943 * credits cannot be returned. Can we handle this somehow? We
944 * may need to return -EAGAIN upwards in the worst case. --sct
947 err
= ext4_splice_branch(handle
, inode
, iblock
,
948 partial
, indirect_blks
, count
);
950 * i_disksize growing is protected by i_data_sem. Don't forget to
951 * protect it if you're about to implement concurrent
952 * ext4_get_block() -bzzz
954 if (!err
&& extend_disksize
) {
955 disksize
= ((loff_t
) iblock
+ count
) << inode
->i_blkbits
;
956 if (disksize
> i_size_read(inode
))
957 disksize
= i_size_read(inode
);
958 if (disksize
> ei
->i_disksize
)
959 ei
->i_disksize
= disksize
;
964 set_buffer_new(bh_result
);
966 map_bh(bh_result
, inode
->i_sb
, le32_to_cpu(chain
[depth
-1].key
));
967 if (count
> blocks_to_boundary
)
968 set_buffer_boundary(bh_result
);
970 /* Clean up and exit */
971 partial
= chain
+ depth
- 1; /* the whole chain */
973 while (partial
> chain
) {
974 BUFFER_TRACE(partial
->bh
, "call brelse");
978 BUFFER_TRACE(bh_result
, "returned");
984 * Calculate the number of metadata blocks need to reserve
985 * to allocate @blocks for non extent file based file
987 static int ext4_indirect_calc_metadata_amount(struct inode
*inode
, int blocks
)
989 int icap
= EXT4_ADDR_PER_BLOCK(inode
->i_sb
);
990 int ind_blks
, dind_blks
, tind_blks
;
992 /* number of new indirect blocks needed */
993 ind_blks
= (blocks
+ icap
- 1) / icap
;
995 dind_blks
= (ind_blks
+ icap
- 1) / icap
;
999 return ind_blks
+ dind_blks
+ tind_blks
;
1003 * Calculate the number of metadata blocks need to reserve
1004 * to allocate given number of blocks
1006 static int ext4_calc_metadata_amount(struct inode
*inode
, int blocks
)
1011 if (EXT4_I(inode
)->i_flags
& EXT4_EXTENTS_FL
)
1012 return ext4_ext_calc_metadata_amount(inode
, blocks
);
1014 return ext4_indirect_calc_metadata_amount(inode
, blocks
);
1017 static void ext4_da_update_reserve_space(struct inode
*inode
, int used
)
1019 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
1020 int total
, mdb
, mdb_free
;
1022 spin_lock(&EXT4_I(inode
)->i_block_reservation_lock
);
1023 /* recalculate the number of metablocks still need to be reserved */
1024 total
= EXT4_I(inode
)->i_reserved_data_blocks
- used
;
1025 mdb
= ext4_calc_metadata_amount(inode
, total
);
1027 /* figure out how many metablocks to release */
1028 BUG_ON(mdb
> EXT4_I(inode
)->i_reserved_meta_blocks
);
1029 mdb_free
= EXT4_I(inode
)->i_reserved_meta_blocks
- mdb
;
1031 /* Account for allocated meta_blocks */
1032 mdb_free
-= EXT4_I(inode
)->i_allocated_meta_blocks
;
1034 /* update fs free blocks counter for truncate case */
1035 percpu_counter_add(&sbi
->s_freeblocks_counter
, mdb_free
);
1037 /* update per-inode reservations */
1038 BUG_ON(used
> EXT4_I(inode
)->i_reserved_data_blocks
);
1039 EXT4_I(inode
)->i_reserved_data_blocks
-= used
;
1041 BUG_ON(mdb
> EXT4_I(inode
)->i_reserved_meta_blocks
);
1042 EXT4_I(inode
)->i_reserved_meta_blocks
= mdb
;
1043 EXT4_I(inode
)->i_allocated_meta_blocks
= 0;
1044 spin_unlock(&EXT4_I(inode
)->i_block_reservation_lock
);
1047 /* Maximum number of blocks we map for direct IO at once. */
1048 #define DIO_MAX_BLOCKS 4096
1050 * Number of credits we need for writing DIO_MAX_BLOCKS:
1051 * We need sb + group descriptor + bitmap + inode -> 4
1052 * For B blocks with A block pointers per block we need:
1053 * 1 (triple ind.) + (B/A/A + 2) (doubly ind.) + (B/A + 2) (indirect).
1054 * If we plug in 4096 for B and 256 for A (for 1KB block size), we get 25.
1056 #define DIO_CREDITS 25
1060 * The ext4_get_blocks_wrap() function try to look up the requested blocks,
1061 * and returns if the blocks are already mapped.
1063 * Otherwise it takes the write lock of the i_data_sem and allocate blocks
1064 * and store the allocated blocks in the result buffer head and mark it
1067 * If file type is extents based, it will call ext4_ext_get_blocks(),
1068 * Otherwise, call with ext4_get_blocks_handle() to handle indirect mapping
1071 * On success, it returns the number of blocks being mapped or allocate.
1072 * if create==0 and the blocks are pre-allocated and uninitialized block,
1073 * the result buffer head is unmapped. If the create ==1, it will make sure
1074 * the buffer head is mapped.
1076 * It returns 0 if plain look up failed (blocks have not been allocated), in
1077 * that casem, buffer head is unmapped
1079 * It returns the error in case of allocation failure.
1081 int ext4_get_blocks_wrap(handle_t
*handle
, struct inode
*inode
, sector_t block
,
1082 unsigned long max_blocks
, struct buffer_head
*bh
,
1083 int create
, int extend_disksize
, int flag
)
1087 clear_buffer_mapped(bh
);
1090 * Try to see if we can get the block without requesting
1091 * for new file system block.
1093 down_read((&EXT4_I(inode
)->i_data_sem
));
1094 if (EXT4_I(inode
)->i_flags
& EXT4_EXTENTS_FL
) {
1095 retval
= ext4_ext_get_blocks(handle
, inode
, block
, max_blocks
,
1098 retval
= ext4_get_blocks_handle(handle
,
1099 inode
, block
, max_blocks
, bh
, 0, 0);
1101 up_read((&EXT4_I(inode
)->i_data_sem
));
1103 /* If it is only a block(s) look up */
1108 * Returns if the blocks have already allocated
1110 * Note that if blocks have been preallocated
1111 * ext4_ext_get_block() returns th create = 0
1112 * with buffer head unmapped.
1114 if (retval
> 0 && buffer_mapped(bh
))
1118 * New blocks allocate and/or writing to uninitialized extent
1119 * will possibly result in updating i_data, so we take
1120 * the write lock of i_data_sem, and call get_blocks()
1121 * with create == 1 flag.
1123 down_write((&EXT4_I(inode
)->i_data_sem
));
1126 * if the caller is from delayed allocation writeout path
1127 * we have already reserved fs blocks for allocation
1128 * let the underlying get_block() function know to
1129 * avoid double accounting
1132 EXT4_I(inode
)->i_delalloc_reserved_flag
= 1;
1134 * We need to check for EXT4 here because migrate
1135 * could have changed the inode type in between
1137 if (EXT4_I(inode
)->i_flags
& EXT4_EXTENTS_FL
) {
1138 retval
= ext4_ext_get_blocks(handle
, inode
, block
, max_blocks
,
1139 bh
, create
, extend_disksize
);
1141 retval
= ext4_get_blocks_handle(handle
, inode
, block
,
1142 max_blocks
, bh
, create
, extend_disksize
);
1144 if (retval
> 0 && buffer_new(bh
)) {
1146 * We allocated new blocks which will result in
1147 * i_data's format changing. Force the migrate
1148 * to fail by clearing migrate flags
1150 EXT4_I(inode
)->i_flags
= EXT4_I(inode
)->i_flags
&
1156 EXT4_I(inode
)->i_delalloc_reserved_flag
= 0;
1158 * Update reserved blocks/metadata blocks
1159 * after successful block allocation
1160 * which were deferred till now
1162 if ((retval
> 0) && buffer_delay(bh
))
1163 ext4_da_update_reserve_space(inode
, retval
);
1166 up_write((&EXT4_I(inode
)->i_data_sem
));
1170 static int ext4_get_block(struct inode
*inode
, sector_t iblock
,
1171 struct buffer_head
*bh_result
, int create
)
1173 handle_t
*handle
= ext4_journal_current_handle();
1174 int ret
= 0, started
= 0;
1175 unsigned max_blocks
= bh_result
->b_size
>> inode
->i_blkbits
;
1177 if (create
&& !handle
) {
1178 /* Direct IO write... */
1179 if (max_blocks
> DIO_MAX_BLOCKS
)
1180 max_blocks
= DIO_MAX_BLOCKS
;
1181 handle
= ext4_journal_start(inode
, DIO_CREDITS
+
1182 2 * EXT4_QUOTA_TRANS_BLOCKS(inode
->i_sb
));
1183 if (IS_ERR(handle
)) {
1184 ret
= PTR_ERR(handle
);
1190 ret
= ext4_get_blocks_wrap(handle
, inode
, iblock
,
1191 max_blocks
, bh_result
, create
, 0, 0);
1193 bh_result
->b_size
= (ret
<< inode
->i_blkbits
);
1197 ext4_journal_stop(handle
);
1203 * `handle' can be NULL if create is zero
1205 struct buffer_head
*ext4_getblk(handle_t
*handle
, struct inode
*inode
,
1206 ext4_lblk_t block
, int create
, int *errp
)
1208 struct buffer_head dummy
;
1211 J_ASSERT(handle
!= NULL
|| create
== 0);
1214 dummy
.b_blocknr
= -1000;
1215 buffer_trace_init(&dummy
.b_history
);
1216 err
= ext4_get_blocks_wrap(handle
, inode
, block
, 1,
1217 &dummy
, create
, 1, 0);
1219 * ext4_get_blocks_handle() returns number of blocks
1220 * mapped. 0 in case of a HOLE.
1228 if (!err
&& buffer_mapped(&dummy
)) {
1229 struct buffer_head
*bh
;
1230 bh
= sb_getblk(inode
->i_sb
, dummy
.b_blocknr
);
1235 if (buffer_new(&dummy
)) {
1236 J_ASSERT(create
!= 0);
1237 J_ASSERT(handle
!= NULL
);
1240 * Now that we do not always journal data, we should
1241 * keep in mind whether this should always journal the
1242 * new buffer as metadata. For now, regular file
1243 * writes use ext4_get_block instead, so it's not a
1247 BUFFER_TRACE(bh
, "call get_create_access");
1248 fatal
= ext4_journal_get_create_access(handle
, bh
);
1249 if (!fatal
&& !buffer_uptodate(bh
)) {
1250 memset(bh
->b_data
,0,inode
->i_sb
->s_blocksize
);
1251 set_buffer_uptodate(bh
);
1254 BUFFER_TRACE(bh
, "call ext4_journal_dirty_metadata");
1255 err
= ext4_journal_dirty_metadata(handle
, bh
);
1259 BUFFER_TRACE(bh
, "not a new buffer");
1272 struct buffer_head
*ext4_bread(handle_t
*handle
, struct inode
*inode
,
1273 ext4_lblk_t block
, int create
, int *err
)
1275 struct buffer_head
* bh
;
1277 bh
= ext4_getblk(handle
, inode
, block
, create
, err
);
1280 if (buffer_uptodate(bh
))
1282 ll_rw_block(READ_META
, 1, &bh
);
1284 if (buffer_uptodate(bh
))
1291 static int walk_page_buffers( handle_t
*handle
,
1292 struct buffer_head
*head
,
1296 int (*fn
)( handle_t
*handle
,
1297 struct buffer_head
*bh
))
1299 struct buffer_head
*bh
;
1300 unsigned block_start
, block_end
;
1301 unsigned blocksize
= head
->b_size
;
1303 struct buffer_head
*next
;
1305 for ( bh
= head
, block_start
= 0;
1306 ret
== 0 && (bh
!= head
|| !block_start
);
1307 block_start
= block_end
, bh
= next
)
1309 next
= bh
->b_this_page
;
1310 block_end
= block_start
+ blocksize
;
1311 if (block_end
<= from
|| block_start
>= to
) {
1312 if (partial
&& !buffer_uptodate(bh
))
1316 err
= (*fn
)(handle
, bh
);
1324 * To preserve ordering, it is essential that the hole instantiation and
1325 * the data write be encapsulated in a single transaction. We cannot
1326 * close off a transaction and start a new one between the ext4_get_block()
1327 * and the commit_write(). So doing the jbd2_journal_start at the start of
1328 * prepare_write() is the right place.
1330 * Also, this function can nest inside ext4_writepage() ->
1331 * block_write_full_page(). In that case, we *know* that ext4_writepage()
1332 * has generated enough buffer credits to do the whole page. So we won't
1333 * block on the journal in that case, which is good, because the caller may
1336 * By accident, ext4 can be reentered when a transaction is open via
1337 * quota file writes. If we were to commit the transaction while thus
1338 * reentered, there can be a deadlock - we would be holding a quota
1339 * lock, and the commit would never complete if another thread had a
1340 * transaction open and was blocking on the quota lock - a ranking
1343 * So what we do is to rely on the fact that jbd2_journal_stop/journal_start
1344 * will _not_ run commit under these circumstances because handle->h_ref
1345 * is elevated. We'll still have enough credits for the tiny quotafile
1348 static int do_journal_get_write_access(handle_t
*handle
,
1349 struct buffer_head
*bh
)
1351 if (!buffer_mapped(bh
) || buffer_freed(bh
))
1353 return ext4_journal_get_write_access(handle
, bh
);
1356 static int ext4_write_begin(struct file
*file
, struct address_space
*mapping
,
1357 loff_t pos
, unsigned len
, unsigned flags
,
1358 struct page
**pagep
, void **fsdata
)
1360 struct inode
*inode
= mapping
->host
;
1361 int ret
, needed_blocks
= ext4_writepage_trans_blocks(inode
);
1368 index
= pos
>> PAGE_CACHE_SHIFT
;
1369 from
= pos
& (PAGE_CACHE_SIZE
- 1);
1373 handle
= ext4_journal_start(inode
, needed_blocks
);
1374 if (IS_ERR(handle
)) {
1375 ret
= PTR_ERR(handle
);
1379 page
= __grab_cache_page(mapping
, index
);
1381 ext4_journal_stop(handle
);
1387 ret
= block_write_begin(file
, mapping
, pos
, len
, flags
, pagep
, fsdata
,
1390 if (!ret
&& ext4_should_journal_data(inode
)) {
1391 ret
= walk_page_buffers(handle
, page_buffers(page
),
1392 from
, to
, NULL
, do_journal_get_write_access
);
1397 ext4_journal_stop(handle
);
1398 page_cache_release(page
);
1401 if (ret
== -ENOSPC
&& ext4_should_retry_alloc(inode
->i_sb
, &retries
))
1407 /* For write_end() in data=journal mode */
1408 static int write_end_fn(handle_t
*handle
, struct buffer_head
*bh
)
1410 if (!buffer_mapped(bh
) || buffer_freed(bh
))
1412 set_buffer_uptodate(bh
);
1413 return ext4_journal_dirty_metadata(handle
, bh
);
1417 * We need to pick up the new inode size which generic_commit_write gave us
1418 * `file' can be NULL - eg, when called from page_symlink().
1420 * ext4 never places buffers on inode->i_mapping->private_list. metadata
1421 * buffers are managed internally.
1423 static int ext4_ordered_write_end(struct file
*file
,
1424 struct address_space
*mapping
,
1425 loff_t pos
, unsigned len
, unsigned copied
,
1426 struct page
*page
, void *fsdata
)
1428 handle_t
*handle
= ext4_journal_current_handle();
1429 struct inode
*inode
= mapping
->host
;
1432 ret
= ext4_jbd2_file_inode(handle
, inode
);
1436 * generic_write_end() will run mark_inode_dirty() if i_size
1437 * changes. So let's piggyback the i_disksize mark_inode_dirty
1442 new_i_size
= pos
+ copied
;
1443 if (new_i_size
> EXT4_I(inode
)->i_disksize
)
1444 EXT4_I(inode
)->i_disksize
= new_i_size
;
1445 ret2
= generic_write_end(file
, mapping
, pos
, len
, copied
,
1451 ret2
= ext4_journal_stop(handle
);
1455 return ret
? ret
: copied
;
1458 static int ext4_writeback_write_end(struct file
*file
,
1459 struct address_space
*mapping
,
1460 loff_t pos
, unsigned len
, unsigned copied
,
1461 struct page
*page
, void *fsdata
)
1463 handle_t
*handle
= ext4_journal_current_handle();
1464 struct inode
*inode
= mapping
->host
;
1468 new_i_size
= pos
+ copied
;
1469 if (new_i_size
> EXT4_I(inode
)->i_disksize
)
1470 EXT4_I(inode
)->i_disksize
= new_i_size
;
1472 ret2
= generic_write_end(file
, mapping
, pos
, len
, copied
,
1478 ret2
= ext4_journal_stop(handle
);
1482 return ret
? ret
: copied
;
1485 static int ext4_journalled_write_end(struct file
*file
,
1486 struct address_space
*mapping
,
1487 loff_t pos
, unsigned len
, unsigned copied
,
1488 struct page
*page
, void *fsdata
)
1490 handle_t
*handle
= ext4_journal_current_handle();
1491 struct inode
*inode
= mapping
->host
;
1496 from
= pos
& (PAGE_CACHE_SIZE
- 1);
1500 if (!PageUptodate(page
))
1502 page_zero_new_buffers(page
, from
+copied
, to
);
1505 ret
= walk_page_buffers(handle
, page_buffers(page
), from
,
1506 to
, &partial
, write_end_fn
);
1508 SetPageUptodate(page
);
1509 if (pos
+copied
> inode
->i_size
)
1510 i_size_write(inode
, pos
+copied
);
1511 EXT4_I(inode
)->i_state
|= EXT4_STATE_JDATA
;
1512 if (inode
->i_size
> EXT4_I(inode
)->i_disksize
) {
1513 EXT4_I(inode
)->i_disksize
= inode
->i_size
;
1514 ret2
= ext4_mark_inode_dirty(handle
, inode
);
1520 ret2
= ext4_journal_stop(handle
);
1523 page_cache_release(page
);
1525 return ret
? ret
: copied
;
1528 static int ext4_da_reserve_space(struct inode
*inode
, int nrblocks
)
1530 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
1531 unsigned long md_needed
, mdblocks
, total
= 0;
1534 * recalculate the amount of metadata blocks to reserve
1535 * in order to allocate nrblocks
1536 * worse case is one extent per block
1538 spin_lock(&EXT4_I(inode
)->i_block_reservation_lock
);
1539 total
= EXT4_I(inode
)->i_reserved_data_blocks
+ nrblocks
;
1540 mdblocks
= ext4_calc_metadata_amount(inode
, total
);
1541 BUG_ON(mdblocks
< EXT4_I(inode
)->i_reserved_meta_blocks
);
1543 md_needed
= mdblocks
- EXT4_I(inode
)->i_reserved_meta_blocks
;
1544 total
= md_needed
+ nrblocks
;
1546 if (ext4_has_free_blocks(sbi
, total
) < total
) {
1547 spin_unlock(&EXT4_I(inode
)->i_block_reservation_lock
);
1550 /* reduce fs free blocks counter */
1551 percpu_counter_sub(&sbi
->s_freeblocks_counter
, total
);
1553 EXT4_I(inode
)->i_reserved_data_blocks
+= nrblocks
;
1554 EXT4_I(inode
)->i_reserved_meta_blocks
= mdblocks
;
1556 spin_unlock(&EXT4_I(inode
)->i_block_reservation_lock
);
1557 return 0; /* success */
1560 static void ext4_da_release_space(struct inode
*inode
, int to_free
)
1562 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
1563 int total
, mdb
, mdb_free
, release
;
1566 return; /* Nothing to release, exit */
1568 spin_lock(&EXT4_I(inode
)->i_block_reservation_lock
);
1570 if (!EXT4_I(inode
)->i_reserved_data_blocks
) {
1572 * if there is no reserved blocks, but we try to free some
1573 * then the counter is messed up somewhere.
1574 * but since this function is called from invalidate
1575 * page, it's harmless to return without any action
1577 printk(KERN_INFO
"ext4 delalloc try to release %d reserved "
1578 "blocks for inode %lu, but there is no reserved "
1579 "data blocks\n", to_free
, inode
->i_ino
);
1580 spin_unlock(&EXT4_I(inode
)->i_block_reservation_lock
);
1584 /* recalculate the number of metablocks still need to be reserved */
1585 total
= EXT4_I(inode
)->i_reserved_data_blocks
- to_free
;
1586 mdb
= ext4_calc_metadata_amount(inode
, total
);
1588 /* figure out how many metablocks to release */
1589 BUG_ON(mdb
> EXT4_I(inode
)->i_reserved_meta_blocks
);
1590 mdb_free
= EXT4_I(inode
)->i_reserved_meta_blocks
- mdb
;
1592 release
= to_free
+ mdb_free
;
1594 /* update fs free blocks counter for truncate case */
1595 percpu_counter_add(&sbi
->s_freeblocks_counter
, release
);
1597 /* update per-inode reservations */
1598 BUG_ON(to_free
> EXT4_I(inode
)->i_reserved_data_blocks
);
1599 EXT4_I(inode
)->i_reserved_data_blocks
-= to_free
;
1601 BUG_ON(mdb
> EXT4_I(inode
)->i_reserved_meta_blocks
);
1602 EXT4_I(inode
)->i_reserved_meta_blocks
= mdb
;
1603 spin_unlock(&EXT4_I(inode
)->i_block_reservation_lock
);
1606 static void ext4_da_page_release_reservation(struct page
*page
,
1607 unsigned long offset
)
1610 struct buffer_head
*head
, *bh
;
1611 unsigned int curr_off
= 0;
1613 head
= page_buffers(page
);
1616 unsigned int next_off
= curr_off
+ bh
->b_size
;
1618 if ((offset
<= curr_off
) && (buffer_delay(bh
))) {
1620 clear_buffer_delay(bh
);
1622 curr_off
= next_off
;
1623 } while ((bh
= bh
->b_this_page
) != head
);
1624 ext4_da_release_space(page
->mapping
->host
, to_release
);
1628 * Delayed allocation stuff
1631 struct mpage_da_data
{
1632 struct inode
*inode
;
1633 struct buffer_head lbh
; /* extent of blocks */
1634 unsigned long first_page
, next_page
; /* extent of pages */
1635 get_block_t
*get_block
;
1636 struct writeback_control
*wbc
;
1640 * mpage_da_submit_io - walks through extent of pages and try to write
1641 * them with __mpage_writepage()
1643 * @mpd->inode: inode
1644 * @mpd->first_page: first page of the extent
1645 * @mpd->next_page: page after the last page of the extent
1646 * @mpd->get_block: the filesystem's block mapper function
1648 * By the time mpage_da_submit_io() is called we expect all blocks
1649 * to be allocated. this may be wrong if allocation failed.
1651 * As pages are already locked by write_cache_pages(), we can't use it
1653 static int mpage_da_submit_io(struct mpage_da_data
*mpd
)
1655 struct address_space
*mapping
= mpd
->inode
->i_mapping
;
1656 struct mpage_data mpd_pp
= {
1658 .last_block_in_bio
= 0,
1659 .get_block
= mpd
->get_block
,
1662 int ret
= 0, err
, nr_pages
, i
;
1663 unsigned long index
, end
;
1664 struct pagevec pvec
;
1666 BUG_ON(mpd
->next_page
<= mpd
->first_page
);
1668 pagevec_init(&pvec
, 0);
1669 index
= mpd
->first_page
;
1670 end
= mpd
->next_page
- 1;
1672 while (index
<= end
) {
1673 /* XXX: optimize tail */
1674 nr_pages
= pagevec_lookup(&pvec
, mapping
, index
, PAGEVEC_SIZE
);
1677 for (i
= 0; i
< nr_pages
; i
++) {
1678 struct page
*page
= pvec
.pages
[i
];
1680 index
= page
->index
;
1685 err
= __mpage_writepage(page
, mpd
->wbc
, &mpd_pp
);
1688 * In error case, we have to continue because
1689 * remaining pages are still locked
1690 * XXX: unlock and re-dirty them?
1695 pagevec_release(&pvec
);
1698 mpage_bio_submit(WRITE
, mpd_pp
.bio
);
1704 * mpage_put_bnr_to_bhs - walk blocks and assign them actual numbers
1706 * @mpd->inode - inode to walk through
1707 * @exbh->b_blocknr - first block on a disk
1708 * @exbh->b_size - amount of space in bytes
1709 * @logical - first logical block to start assignment with
1711 * the function goes through all passed space and put actual disk
1712 * block numbers into buffer heads, dropping BH_Delay
1714 static void mpage_put_bnr_to_bhs(struct mpage_da_data
*mpd
, sector_t logical
,
1715 struct buffer_head
*exbh
)
1717 struct inode
*inode
= mpd
->inode
;
1718 struct address_space
*mapping
= inode
->i_mapping
;
1719 int blocks
= exbh
->b_size
>> inode
->i_blkbits
;
1720 sector_t pblock
= exbh
->b_blocknr
, cur_logical
;
1721 struct buffer_head
*head
, *bh
;
1722 unsigned long index
, end
;
1723 struct pagevec pvec
;
1726 index
= logical
>> (PAGE_CACHE_SHIFT
- inode
->i_blkbits
);
1727 end
= (logical
+ blocks
- 1) >> (PAGE_CACHE_SHIFT
- inode
->i_blkbits
);
1728 cur_logical
= index
<< (PAGE_CACHE_SHIFT
- inode
->i_blkbits
);
1730 pagevec_init(&pvec
, 0);
1732 while (index
<= end
) {
1733 /* XXX: optimize tail */
1734 nr_pages
= pagevec_lookup(&pvec
, mapping
, index
, PAGEVEC_SIZE
);
1737 for (i
= 0; i
< nr_pages
; i
++) {
1738 struct page
*page
= pvec
.pages
[i
];
1740 index
= page
->index
;
1745 BUG_ON(!PageLocked(page
));
1746 BUG_ON(PageWriteback(page
));
1747 BUG_ON(!page_has_buffers(page
));
1749 bh
= page_buffers(page
);
1752 /* skip blocks out of the range */
1754 if (cur_logical
>= logical
)
1757 } while ((bh
= bh
->b_this_page
) != head
);
1760 if (cur_logical
>= logical
+ blocks
)
1762 if (buffer_delay(bh
)) {
1763 bh
->b_blocknr
= pblock
;
1764 clear_buffer_delay(bh
);
1765 bh
->b_bdev
= inode
->i_sb
->s_bdev
;
1766 } else if (buffer_unwritten(bh
)) {
1767 bh
->b_blocknr
= pblock
;
1768 clear_buffer_unwritten(bh
);
1769 set_buffer_mapped(bh
);
1771 bh
->b_bdev
= inode
->i_sb
->s_bdev
;
1772 } else if (buffer_mapped(bh
))
1773 BUG_ON(bh
->b_blocknr
!= pblock
);
1777 } while ((bh
= bh
->b_this_page
) != head
);
1779 pagevec_release(&pvec
);
1785 * __unmap_underlying_blocks - just a helper function to unmap
1786 * set of blocks described by @bh
1788 static inline void __unmap_underlying_blocks(struct inode
*inode
,
1789 struct buffer_head
*bh
)
1791 struct block_device
*bdev
= inode
->i_sb
->s_bdev
;
1794 blocks
= bh
->b_size
>> inode
->i_blkbits
;
1795 for (i
= 0; i
< blocks
; i
++)
1796 unmap_underlying_metadata(bdev
, bh
->b_blocknr
+ i
);
1800 * mpage_da_map_blocks - go through given space
1802 * @mpd->lbh - bh describing space
1803 * @mpd->get_block - the filesystem's block mapper function
1805 * The function skips space we know is already mapped to disk blocks.
1807 * The function ignores errors ->get_block() returns, thus real
1808 * error handling is postponed to __mpage_writepage()
1810 static void mpage_da_map_blocks(struct mpage_da_data
*mpd
)
1812 struct buffer_head
*lbh
= &mpd
->lbh
;
1813 int err
= 0, remain
= lbh
->b_size
;
1814 sector_t next
= lbh
->b_blocknr
;
1815 struct buffer_head
new;
1818 * We consider only non-mapped and non-allocated blocks
1820 if (buffer_mapped(lbh
) && !buffer_delay(lbh
))
1824 new.b_state
= lbh
->b_state
;
1826 new.b_size
= remain
;
1827 err
= mpd
->get_block(mpd
->inode
, next
, &new, 1);
1830 * Rather than implement own error handling
1831 * here, we just leave remaining blocks
1832 * unallocated and try again with ->writepage()
1836 BUG_ON(new.b_size
== 0);
1838 if (buffer_new(&new))
1839 __unmap_underlying_blocks(mpd
->inode
, &new);
1842 * If blocks are delayed marked, we need to
1843 * put actual blocknr and drop delayed bit
1845 if (buffer_delay(lbh
) || buffer_unwritten(lbh
))
1846 mpage_put_bnr_to_bhs(mpd
, next
, &new);
1848 /* go for the remaining blocks */
1849 next
+= new.b_size
>> mpd
->inode
->i_blkbits
;
1850 remain
-= new.b_size
;
1854 #define BH_FLAGS ((1 << BH_Uptodate) | (1 << BH_Mapped) | \
1855 (1 << BH_Delay) | (1 << BH_Unwritten))
1858 * mpage_add_bh_to_extent - try to add one more block to extent of blocks
1860 * @mpd->lbh - extent of blocks
1861 * @logical - logical number of the block in the file
1862 * @bh - bh of the block (used to access block's state)
1864 * the function is used to collect contig. blocks in same state
1866 static void mpage_add_bh_to_extent(struct mpage_da_data
*mpd
,
1867 sector_t logical
, struct buffer_head
*bh
)
1869 struct buffer_head
*lbh
= &mpd
->lbh
;
1872 next
= lbh
->b_blocknr
+ (lbh
->b_size
>> mpd
->inode
->i_blkbits
);
1875 * First block in the extent
1877 if (lbh
->b_size
== 0) {
1878 lbh
->b_blocknr
= logical
;
1879 lbh
->b_size
= bh
->b_size
;
1880 lbh
->b_state
= bh
->b_state
& BH_FLAGS
;
1885 * Can we merge the block to our big extent?
1887 if (logical
== next
&& (bh
->b_state
& BH_FLAGS
) == lbh
->b_state
) {
1888 lbh
->b_size
+= bh
->b_size
;
1893 * We couldn't merge the block to our extent, so we
1894 * need to flush current extent and start new one
1896 mpage_da_map_blocks(mpd
);
1899 * Now start a new extent
1901 lbh
->b_size
= bh
->b_size
;
1902 lbh
->b_state
= bh
->b_state
& BH_FLAGS
;
1903 lbh
->b_blocknr
= logical
;
1907 * __mpage_da_writepage - finds extent of pages and blocks
1909 * @page: page to consider
1910 * @wbc: not used, we just follow rules
1913 * The function finds extents of pages and scan them for all blocks.
1915 static int __mpage_da_writepage(struct page
*page
,
1916 struct writeback_control
*wbc
, void *data
)
1918 struct mpage_da_data
*mpd
= data
;
1919 struct inode
*inode
= mpd
->inode
;
1920 struct buffer_head
*bh
, *head
, fake
;
1924 * Can we merge this page to current extent?
1926 if (mpd
->next_page
!= page
->index
) {
1928 * Nope, we can't. So, we map non-allocated blocks
1929 * and start IO on them using __mpage_writepage()
1931 if (mpd
->next_page
!= mpd
->first_page
) {
1932 mpage_da_map_blocks(mpd
);
1933 mpage_da_submit_io(mpd
);
1937 * Start next extent of pages ...
1939 mpd
->first_page
= page
->index
;
1944 mpd
->lbh
.b_size
= 0;
1945 mpd
->lbh
.b_state
= 0;
1946 mpd
->lbh
.b_blocknr
= 0;
1949 mpd
->next_page
= page
->index
+ 1;
1950 logical
= (sector_t
) page
->index
<<
1951 (PAGE_CACHE_SHIFT
- inode
->i_blkbits
);
1953 if (!page_has_buffers(page
)) {
1955 * There is no attached buffer heads yet (mmap?)
1956 * we treat the page asfull of dirty blocks
1959 bh
->b_size
= PAGE_CACHE_SIZE
;
1961 set_buffer_dirty(bh
);
1962 set_buffer_uptodate(bh
);
1963 mpage_add_bh_to_extent(mpd
, logical
, bh
);
1966 * Page with regular buffer heads, just add all dirty ones
1968 head
= page_buffers(page
);
1971 BUG_ON(buffer_locked(bh
));
1972 if (buffer_dirty(bh
))
1973 mpage_add_bh_to_extent(mpd
, logical
, bh
);
1975 } while ((bh
= bh
->b_this_page
) != head
);
1982 * mpage_da_writepages - walk the list of dirty pages of the given
1983 * address space, allocates non-allocated blocks, maps newly-allocated
1984 * blocks to existing bhs and issue IO them
1986 * @mapping: address space structure to write
1987 * @wbc: subtract the number of written pages from *@wbc->nr_to_write
1988 * @get_block: the filesystem's block mapper function.
1990 * This is a library function, which implements the writepages()
1991 * address_space_operation.
1993 * In order to avoid duplication of logic that deals with partial pages,
1994 * multiple bio per page, etc, we find non-allocated blocks, allocate
1995 * them with minimal calls to ->get_block() and re-use __mpage_writepage()
1997 * It's important that we call __mpage_writepage() only once for each
1998 * involved page, otherwise we'd have to implement more complicated logic
1999 * to deal with pages w/o PG_lock or w/ PG_writeback and so on.
2001 * See comments to mpage_writepages()
2003 static int mpage_da_writepages(struct address_space
*mapping
,
2004 struct writeback_control
*wbc
,
2005 get_block_t get_block
)
2007 struct mpage_da_data mpd
;
2011 return generic_writepages(mapping
, wbc
);
2014 mpd
.inode
= mapping
->host
;
2016 mpd
.lbh
.b_state
= 0;
2017 mpd
.lbh
.b_blocknr
= 0;
2020 mpd
.get_block
= get_block
;
2022 ret
= write_cache_pages(mapping
, wbc
, __mpage_da_writepage
, &mpd
);
2025 * Handle last extent of pages
2027 if (mpd
.next_page
!= mpd
.first_page
) {
2028 mpage_da_map_blocks(&mpd
);
2029 mpage_da_submit_io(&mpd
);
2036 * this is a special callback for ->write_begin() only
2037 * it's intention is to return mapped block or reserve space
2039 static int ext4_da_get_block_prep(struct inode
*inode
, sector_t iblock
,
2040 struct buffer_head
*bh_result
, int create
)
2044 BUG_ON(create
== 0);
2045 BUG_ON(bh_result
->b_size
!= inode
->i_sb
->s_blocksize
);
2048 * first, we need to know whether the block is allocated already
2049 * preallocated blocks are unmapped but should treated
2050 * the same as allocated blocks.
2052 ret
= ext4_get_blocks_wrap(NULL
, inode
, iblock
, 1, bh_result
, 0, 0, 0);
2053 if ((ret
== 0) && !buffer_delay(bh_result
)) {
2054 /* the block isn't (pre)allocated yet, let's reserve space */
2056 * XXX: __block_prepare_write() unmaps passed block,
2059 ret
= ext4_da_reserve_space(inode
, 1);
2061 /* not enough space to reserve */
2064 map_bh(bh_result
, inode
->i_sb
, 0);
2065 set_buffer_new(bh_result
);
2066 set_buffer_delay(bh_result
);
2067 } else if (ret
> 0) {
2068 bh_result
->b_size
= (ret
<< inode
->i_blkbits
);
2074 #define EXT4_DELALLOC_RSVED 1
2075 static int ext4_da_get_block_write(struct inode
*inode
, sector_t iblock
,
2076 struct buffer_head
*bh_result
, int create
)
2079 unsigned max_blocks
= bh_result
->b_size
>> inode
->i_blkbits
;
2080 loff_t disksize
= EXT4_I(inode
)->i_disksize
;
2081 handle_t
*handle
= NULL
;
2083 handle
= ext4_journal_current_handle();
2085 ret
= ext4_get_blocks_wrap(handle
, inode
, iblock
, max_blocks
,
2086 bh_result
, 0, 0, 0);
2089 ret
= ext4_get_blocks_wrap(handle
, inode
, iblock
, max_blocks
,
2090 bh_result
, create
, 0, EXT4_DELALLOC_RSVED
);
2094 bh_result
->b_size
= (ret
<< inode
->i_blkbits
);
2097 * Update on-disk size along with block allocation
2098 * we don't use 'extend_disksize' as size may change
2099 * within already allocated block -bzzz
2101 disksize
= ((loff_t
) iblock
+ ret
) << inode
->i_blkbits
;
2102 if (disksize
> i_size_read(inode
))
2103 disksize
= i_size_read(inode
);
2104 if (disksize
> EXT4_I(inode
)->i_disksize
) {
2106 * XXX: replace with spinlock if seen contended -bzzz
2108 down_write(&EXT4_I(inode
)->i_data_sem
);
2109 if (disksize
> EXT4_I(inode
)->i_disksize
)
2110 EXT4_I(inode
)->i_disksize
= disksize
;
2111 up_write(&EXT4_I(inode
)->i_data_sem
);
2113 if (EXT4_I(inode
)->i_disksize
== disksize
) {
2114 ret
= ext4_mark_inode_dirty(handle
, inode
);
2123 static int ext4_bh_unmapped_or_delay(handle_t
*handle
, struct buffer_head
*bh
)
2126 * unmapped buffer is possible for holes.
2127 * delay buffer is possible with delayed allocation
2129 return ((!buffer_mapped(bh
) || buffer_delay(bh
)) && buffer_dirty(bh
));
2132 static int ext4_normal_get_block_write(struct inode
*inode
, sector_t iblock
,
2133 struct buffer_head
*bh_result
, int create
)
2136 unsigned max_blocks
= bh_result
->b_size
>> inode
->i_blkbits
;
2139 * we don't want to do block allocation in writepage
2140 * so call get_block_wrap with create = 0
2142 ret
= ext4_get_blocks_wrap(NULL
, inode
, iblock
, max_blocks
,
2143 bh_result
, 0, 0, 0);
2145 bh_result
->b_size
= (ret
<< inode
->i_blkbits
);
2152 * get called vi ext4_da_writepages after taking page lock (have journal handle)
2153 * get called via journal_submit_inode_data_buffers (no journal handle)
2154 * get called via shrink_page_list via pdflush (no journal handle)
2155 * or grab_page_cache when doing write_begin (have journal handle)
2157 static int ext4_da_writepage(struct page
*page
,
2158 struct writeback_control
*wbc
)
2163 struct buffer_head
*page_bufs
;
2164 struct inode
*inode
= page
->mapping
->host
;
2166 size
= i_size_read(inode
);
2167 if (page
->index
== size
>> PAGE_CACHE_SHIFT
)
2168 len
= size
& ~PAGE_CACHE_MASK
;
2170 len
= PAGE_CACHE_SIZE
;
2172 if (page_has_buffers(page
)) {
2173 page_bufs
= page_buffers(page
);
2174 if (walk_page_buffers(NULL
, page_bufs
, 0, len
, NULL
,
2175 ext4_bh_unmapped_or_delay
)) {
2177 * We don't want to do block allocation
2178 * So redirty the page and return
2179 * We may reach here when we do a journal commit
2180 * via journal_submit_inode_data_buffers.
2181 * If we don't have mapping block we just ignore
2182 * them. We can also reach here via shrink_page_list
2184 redirty_page_for_writepage(wbc
, page
);
2190 * The test for page_has_buffers() is subtle:
2191 * We know the page is dirty but it lost buffers. That means
2192 * that at some moment in time after write_begin()/write_end()
2193 * has been called all buffers have been clean and thus they
2194 * must have been written at least once. So they are all
2195 * mapped and we can happily proceed with mapping them
2196 * and writing the page.
2198 * Try to initialize the buffer_heads and check whether
2199 * all are mapped and non delay. We don't want to
2200 * do block allocation here.
2202 ret
= block_prepare_write(page
, 0, PAGE_CACHE_SIZE
,
2203 ext4_normal_get_block_write
);
2205 page_bufs
= page_buffers(page
);
2206 /* check whether all are mapped and non delay */
2207 if (walk_page_buffers(NULL
, page_bufs
, 0, len
, NULL
,
2208 ext4_bh_unmapped_or_delay
)) {
2209 redirty_page_for_writepage(wbc
, page
);
2215 * We can't do block allocation here
2216 * so just redity the page and unlock
2219 redirty_page_for_writepage(wbc
, page
);
2225 if (test_opt(inode
->i_sb
, NOBH
) && ext4_should_writeback_data(inode
))
2226 ret
= nobh_writepage(page
, ext4_normal_get_block_write
, wbc
);
2228 ret
= block_write_full_page(page
,
2229 ext4_normal_get_block_write
,
2236 * For now just follow the DIO way to estimate the max credits
2237 * needed to write out EXT4_MAX_WRITEBACK_PAGES.
2238 * todo: need to calculate the max credits need for
2239 * extent based files, currently the DIO credits is based on
2240 * indirect-blocks mapping way.
2242 * Probably should have a generic way to calculate credits
2243 * for DIO, writepages, and truncate
2245 #define EXT4_MAX_WRITEBACK_PAGES DIO_MAX_BLOCKS
2246 #define EXT4_MAX_WRITEBACK_CREDITS DIO_CREDITS
2248 static int ext4_da_writepages(struct address_space
*mapping
,
2249 struct writeback_control
*wbc
)
2251 struct inode
*inode
= mapping
->host
;
2252 handle_t
*handle
= NULL
;
2256 loff_t range_start
= 0;
2259 * No pages to write? This is mainly a kludge to avoid starting
2260 * a transaction for special inodes like journal inode on last iput()
2261 * because that could violate lock ordering on umount
2263 if (!mapping
->nrpages
)
2267 * Estimate the worse case needed credits to write out
2268 * EXT4_MAX_BUF_BLOCKS pages
2270 needed_blocks
= EXT4_MAX_WRITEBACK_CREDITS
;
2272 to_write
= wbc
->nr_to_write
;
2273 if (!wbc
->range_cyclic
) {
2275 * If range_cyclic is not set force range_cont
2276 * and save the old writeback_index
2278 wbc
->range_cont
= 1;
2279 range_start
= wbc
->range_start
;
2282 while (!ret
&& to_write
) {
2283 /* start a new transaction*/
2284 handle
= ext4_journal_start(inode
, needed_blocks
);
2285 if (IS_ERR(handle
)) {
2286 ret
= PTR_ERR(handle
);
2287 goto out_writepages
;
2289 if (ext4_should_order_data(inode
)) {
2291 * With ordered mode we need to add
2292 * the inode to the journal handle
2293 * when we do block allocation.
2295 ret
= ext4_jbd2_file_inode(handle
, inode
);
2297 ext4_journal_stop(handle
);
2298 goto out_writepages
;
2303 * set the max dirty pages could be write at a time
2304 * to fit into the reserved transaction credits
2306 if (wbc
->nr_to_write
> EXT4_MAX_WRITEBACK_PAGES
)
2307 wbc
->nr_to_write
= EXT4_MAX_WRITEBACK_PAGES
;
2309 to_write
-= wbc
->nr_to_write
;
2310 ret
= mpage_da_writepages(mapping
, wbc
,
2311 ext4_da_get_block_write
);
2312 ext4_journal_stop(handle
);
2313 if (wbc
->nr_to_write
) {
2315 * There is no more writeout needed
2316 * or we requested for a noblocking writeout
2317 * and we found the device congested
2319 to_write
+= wbc
->nr_to_write
;
2322 wbc
->nr_to_write
= to_write
;
2326 wbc
->nr_to_write
= to_write
;
2328 wbc
->range_start
= range_start
;
2332 static int ext4_da_write_begin(struct file
*file
, struct address_space
*mapping
,
2333 loff_t pos
, unsigned len
, unsigned flags
,
2334 struct page
**pagep
, void **fsdata
)
2336 int ret
, retries
= 0;
2340 struct inode
*inode
= mapping
->host
;
2343 index
= pos
>> PAGE_CACHE_SHIFT
;
2344 from
= pos
& (PAGE_CACHE_SIZE
- 1);
2349 * With delayed allocation, we don't log the i_disksize update
2350 * if there is delayed block allocation. But we still need
2351 * to journalling the i_disksize update if writes to the end
2352 * of file which has an already mapped buffer.
2354 handle
= ext4_journal_start(inode
, 1);
2355 if (IS_ERR(handle
)) {
2356 ret
= PTR_ERR(handle
);
2360 page
= __grab_cache_page(mapping
, index
);
2362 ext4_journal_stop(handle
);
2368 ret
= block_write_begin(file
, mapping
, pos
, len
, flags
, pagep
, fsdata
,
2369 ext4_da_get_block_prep
);
2372 ext4_journal_stop(handle
);
2373 page_cache_release(page
);
2376 if (ret
== -ENOSPC
&& ext4_should_retry_alloc(inode
->i_sb
, &retries
))
2383 * Check if we should update i_disksize
2384 * when write to the end of file but not require block allocation
2386 static int ext4_da_should_update_i_disksize(struct page
*page
,
2387 unsigned long offset
)
2389 struct buffer_head
*bh
;
2390 struct inode
*inode
= page
->mapping
->host
;
2394 bh
= page_buffers(page
);
2395 idx
= offset
>> inode
->i_blkbits
;
2397 for (i
=0; i
< idx
; i
++)
2398 bh
= bh
->b_this_page
;
2400 if (!buffer_mapped(bh
) || (buffer_delay(bh
)))
2405 static int ext4_da_write_end(struct file
*file
,
2406 struct address_space
*mapping
,
2407 loff_t pos
, unsigned len
, unsigned copied
,
2408 struct page
*page
, void *fsdata
)
2410 struct inode
*inode
= mapping
->host
;
2412 handle_t
*handle
= ext4_journal_current_handle();
2414 unsigned long start
, end
;
2416 start
= pos
& (PAGE_CACHE_SIZE
- 1);
2417 end
= start
+ copied
-1;
2420 * generic_write_end() will run mark_inode_dirty() if i_size
2421 * changes. So let's piggyback the i_disksize mark_inode_dirty
2425 new_i_size
= pos
+ copied
;
2426 if (new_i_size
> EXT4_I(inode
)->i_disksize
) {
2427 if (ext4_da_should_update_i_disksize(page
, end
)) {
2428 down_write(&EXT4_I(inode
)->i_data_sem
);
2429 if (new_i_size
> EXT4_I(inode
)->i_disksize
) {
2431 * Updating i_disksize when extending file
2432 * without needing block allocation
2434 if (ext4_should_order_data(inode
))
2435 ret
= ext4_jbd2_file_inode(handle
,
2438 EXT4_I(inode
)->i_disksize
= new_i_size
;
2440 up_write(&EXT4_I(inode
)->i_data_sem
);
2443 ret2
= generic_write_end(file
, mapping
, pos
, len
, copied
,
2448 ret2
= ext4_journal_stop(handle
);
2452 return ret
? ret
: copied
;
2455 static void ext4_da_invalidatepage(struct page
*page
, unsigned long offset
)
2458 * Drop reserved blocks
2460 BUG_ON(!PageLocked(page
));
2461 if (!page_has_buffers(page
))
2464 ext4_da_page_release_reservation(page
, offset
);
2467 ext4_invalidatepage(page
, offset
);
2474 * bmap() is special. It gets used by applications such as lilo and by
2475 * the swapper to find the on-disk block of a specific piece of data.
2477 * Naturally, this is dangerous if the block concerned is still in the
2478 * journal. If somebody makes a swapfile on an ext4 data-journaling
2479 * filesystem and enables swap, then they may get a nasty shock when the
2480 * data getting swapped to that swapfile suddenly gets overwritten by
2481 * the original zero's written out previously to the journal and
2482 * awaiting writeback in the kernel's buffer cache.
2484 * So, if we see any bmap calls here on a modified, data-journaled file,
2485 * take extra steps to flush any blocks which might be in the cache.
2487 static sector_t
ext4_bmap(struct address_space
*mapping
, sector_t block
)
2489 struct inode
*inode
= mapping
->host
;
2493 if (mapping_tagged(mapping
, PAGECACHE_TAG_DIRTY
) &&
2494 test_opt(inode
->i_sb
, DELALLOC
)) {
2496 * With delalloc we want to sync the file
2497 * so that we can make sure we allocate
2500 filemap_write_and_wait(mapping
);
2503 if (EXT4_I(inode
)->i_state
& EXT4_STATE_JDATA
) {
2505 * This is a REALLY heavyweight approach, but the use of
2506 * bmap on dirty files is expected to be extremely rare:
2507 * only if we run lilo or swapon on a freshly made file
2508 * do we expect this to happen.
2510 * (bmap requires CAP_SYS_RAWIO so this does not
2511 * represent an unprivileged user DOS attack --- we'd be
2512 * in trouble if mortal users could trigger this path at
2515 * NB. EXT4_STATE_JDATA is not set on files other than
2516 * regular files. If somebody wants to bmap a directory
2517 * or symlink and gets confused because the buffer
2518 * hasn't yet been flushed to disk, they deserve
2519 * everything they get.
2522 EXT4_I(inode
)->i_state
&= ~EXT4_STATE_JDATA
;
2523 journal
= EXT4_JOURNAL(inode
);
2524 jbd2_journal_lock_updates(journal
);
2525 err
= jbd2_journal_flush(journal
);
2526 jbd2_journal_unlock_updates(journal
);
2532 return generic_block_bmap(mapping
,block
,ext4_get_block
);
2535 static int bget_one(handle_t
*handle
, struct buffer_head
*bh
)
2541 static int bput_one(handle_t
*handle
, struct buffer_head
*bh
)
2548 * Note that we don't need to start a transaction unless we're journaling data
2549 * because we should have holes filled from ext4_page_mkwrite(). We even don't
2550 * need to file the inode to the transaction's list in ordered mode because if
2551 * we are writing back data added by write(), the inode is already there and if
2552 * we are writing back data modified via mmap(), noone guarantees in which
2553 * transaction the data will hit the disk. In case we are journaling data, we
2554 * cannot start transaction directly because transaction start ranks above page
2555 * lock so we have to do some magic.
2557 * In all journaling modes block_write_full_page() will start the I/O.
2561 * ext4_writepage() -> kmalloc() -> __alloc_pages() -> page_launder() ->
2566 * ext4_file_write() -> generic_file_write() -> __alloc_pages() -> ...
2568 * Same applies to ext4_get_block(). We will deadlock on various things like
2569 * lock_journal and i_data_sem
2571 * Setting PF_MEMALLOC here doesn't work - too many internal memory
2574 * 16May01: If we're reentered then journal_current_handle() will be
2575 * non-zero. We simply *return*.
2577 * 1 July 2001: @@@ FIXME:
2578 * In journalled data mode, a data buffer may be metadata against the
2579 * current transaction. But the same file is part of a shared mapping
2580 * and someone does a writepage() on it.
2582 * We will move the buffer onto the async_data list, but *after* it has
2583 * been dirtied. So there's a small window where we have dirty data on
2586 * Note that this only applies to the last partial page in the file. The
2587 * bit which block_write_full_page() uses prepare/commit for. (That's
2588 * broken code anyway: it's wrong for msync()).
2590 * It's a rare case: affects the final partial page, for journalled data
2591 * where the file is subject to bith write() and writepage() in the same
2592 * transction. To fix it we'll need a custom block_write_full_page().
2593 * We'll probably need that anyway for journalling writepage() output.
2595 * We don't honour synchronous mounts for writepage(). That would be
2596 * disastrous. Any write() or metadata operation will sync the fs for
2600 static int __ext4_normal_writepage(struct page
*page
,
2601 struct writeback_control
*wbc
)
2603 struct inode
*inode
= page
->mapping
->host
;
2605 if (test_opt(inode
->i_sb
, NOBH
))
2606 return nobh_writepage(page
,
2607 ext4_normal_get_block_write
, wbc
);
2609 return block_write_full_page(page
,
2610 ext4_normal_get_block_write
,
2614 static int ext4_normal_writepage(struct page
*page
,
2615 struct writeback_control
*wbc
)
2617 struct inode
*inode
= page
->mapping
->host
;
2618 loff_t size
= i_size_read(inode
);
2621 J_ASSERT(PageLocked(page
));
2622 if (page
->index
== size
>> PAGE_CACHE_SHIFT
)
2623 len
= size
& ~PAGE_CACHE_MASK
;
2625 len
= PAGE_CACHE_SIZE
;
2627 if (page_has_buffers(page
)) {
2628 /* if page has buffers it should all be mapped
2629 * and allocated. If there are not buffers attached
2630 * to the page we know the page is dirty but it lost
2631 * buffers. That means that at some moment in time
2632 * after write_begin() / write_end() has been called
2633 * all buffers have been clean and thus they must have been
2634 * written at least once. So they are all mapped and we can
2635 * happily proceed with mapping them and writing the page.
2637 BUG_ON(walk_page_buffers(NULL
, page_buffers(page
), 0, len
, NULL
,
2638 ext4_bh_unmapped_or_delay
));
2641 if (!ext4_journal_current_handle())
2642 return __ext4_normal_writepage(page
, wbc
);
2644 redirty_page_for_writepage(wbc
, page
);
2649 static int __ext4_journalled_writepage(struct page
*page
,
2650 struct writeback_control
*wbc
)
2652 struct address_space
*mapping
= page
->mapping
;
2653 struct inode
*inode
= mapping
->host
;
2654 struct buffer_head
*page_bufs
;
2655 handle_t
*handle
= NULL
;
2659 ret
= block_prepare_write(page
, 0, PAGE_CACHE_SIZE
,
2660 ext4_normal_get_block_write
);
2664 page_bufs
= page_buffers(page
);
2665 walk_page_buffers(handle
, page_bufs
, 0, PAGE_CACHE_SIZE
, NULL
,
2667 /* As soon as we unlock the page, it can go away, but we have
2668 * references to buffers so we are safe */
2671 handle
= ext4_journal_start(inode
, ext4_writepage_trans_blocks(inode
));
2672 if (IS_ERR(handle
)) {
2673 ret
= PTR_ERR(handle
);
2677 ret
= walk_page_buffers(handle
, page_bufs
, 0,
2678 PAGE_CACHE_SIZE
, NULL
, do_journal_get_write_access
);
2680 err
= walk_page_buffers(handle
, page_bufs
, 0,
2681 PAGE_CACHE_SIZE
, NULL
, write_end_fn
);
2684 err
= ext4_journal_stop(handle
);
2688 walk_page_buffers(handle
, page_bufs
, 0,
2689 PAGE_CACHE_SIZE
, NULL
, bput_one
);
2690 EXT4_I(inode
)->i_state
|= EXT4_STATE_JDATA
;
2699 static int ext4_journalled_writepage(struct page
*page
,
2700 struct writeback_control
*wbc
)
2702 struct inode
*inode
= page
->mapping
->host
;
2703 loff_t size
= i_size_read(inode
);
2706 J_ASSERT(PageLocked(page
));
2707 if (page
->index
== size
>> PAGE_CACHE_SHIFT
)
2708 len
= size
& ~PAGE_CACHE_MASK
;
2710 len
= PAGE_CACHE_SIZE
;
2712 if (page_has_buffers(page
)) {
2713 /* if page has buffers it should all be mapped
2714 * and allocated. If there are not buffers attached
2715 * to the page we know the page is dirty but it lost
2716 * buffers. That means that at some moment in time
2717 * after write_begin() / write_end() has been called
2718 * all buffers have been clean and thus they must have been
2719 * written at least once. So they are all mapped and we can
2720 * happily proceed with mapping them and writing the page.
2722 BUG_ON(walk_page_buffers(NULL
, page_buffers(page
), 0, len
, NULL
,
2723 ext4_bh_unmapped_or_delay
));
2726 if (ext4_journal_current_handle())
2729 if (PageChecked(page
)) {
2731 * It's mmapped pagecache. Add buffers and journal it. There
2732 * doesn't seem much point in redirtying the page here.
2734 ClearPageChecked(page
);
2735 return __ext4_journalled_writepage(page
, wbc
);
2738 * It may be a page full of checkpoint-mode buffers. We don't
2739 * really know unless we go poke around in the buffer_heads.
2740 * But block_write_full_page will do the right thing.
2742 return block_write_full_page(page
,
2743 ext4_normal_get_block_write
,
2747 redirty_page_for_writepage(wbc
, page
);
2752 static int ext4_readpage(struct file
*file
, struct page
*page
)
2754 return mpage_readpage(page
, ext4_get_block
);
2758 ext4_readpages(struct file
*file
, struct address_space
*mapping
,
2759 struct list_head
*pages
, unsigned nr_pages
)
2761 return mpage_readpages(mapping
, pages
, nr_pages
, ext4_get_block
);
2764 static void ext4_invalidatepage(struct page
*page
, unsigned long offset
)
2766 journal_t
*journal
= EXT4_JOURNAL(page
->mapping
->host
);
2769 * If it's a full truncate we just forget about the pending dirtying
2772 ClearPageChecked(page
);
2774 jbd2_journal_invalidatepage(journal
, page
, offset
);
2777 static int ext4_releasepage(struct page
*page
, gfp_t wait
)
2779 journal_t
*journal
= EXT4_JOURNAL(page
->mapping
->host
);
2781 WARN_ON(PageChecked(page
));
2782 if (!page_has_buffers(page
))
2784 return jbd2_journal_try_to_free_buffers(journal
, page
, wait
);
2788 * If the O_DIRECT write will extend the file then add this inode to the
2789 * orphan list. So recovery will truncate it back to the original size
2790 * if the machine crashes during the write.
2792 * If the O_DIRECT write is intantiating holes inside i_size and the machine
2793 * crashes then stale disk data _may_ be exposed inside the file. But current
2794 * VFS code falls back into buffered path in that case so we are safe.
2796 static ssize_t
ext4_direct_IO(int rw
, struct kiocb
*iocb
,
2797 const struct iovec
*iov
, loff_t offset
,
2798 unsigned long nr_segs
)
2800 struct file
*file
= iocb
->ki_filp
;
2801 struct inode
*inode
= file
->f_mapping
->host
;
2802 struct ext4_inode_info
*ei
= EXT4_I(inode
);
2806 size_t count
= iov_length(iov
, nr_segs
);
2809 loff_t final_size
= offset
+ count
;
2811 if (final_size
> inode
->i_size
) {
2812 /* Credits for sb + inode write */
2813 handle
= ext4_journal_start(inode
, 2);
2814 if (IS_ERR(handle
)) {
2815 ret
= PTR_ERR(handle
);
2818 ret
= ext4_orphan_add(handle
, inode
);
2820 ext4_journal_stop(handle
);
2824 ei
->i_disksize
= inode
->i_size
;
2825 ext4_journal_stop(handle
);
2829 ret
= blockdev_direct_IO(rw
, iocb
, inode
, inode
->i_sb
->s_bdev
, iov
,
2831 ext4_get_block
, NULL
);
2836 /* Credits for sb + inode write */
2837 handle
= ext4_journal_start(inode
, 2);
2838 if (IS_ERR(handle
)) {
2839 /* This is really bad luck. We've written the data
2840 * but cannot extend i_size. Bail out and pretend
2841 * the write failed... */
2842 ret
= PTR_ERR(handle
);
2846 ext4_orphan_del(handle
, inode
);
2848 loff_t end
= offset
+ ret
;
2849 if (end
> inode
->i_size
) {
2850 ei
->i_disksize
= end
;
2851 i_size_write(inode
, end
);
2853 * We're going to return a positive `ret'
2854 * here due to non-zero-length I/O, so there's
2855 * no way of reporting error returns from
2856 * ext4_mark_inode_dirty() to userspace. So
2859 ext4_mark_inode_dirty(handle
, inode
);
2862 err
= ext4_journal_stop(handle
);
2871 * Pages can be marked dirty completely asynchronously from ext4's journalling
2872 * activity. By filemap_sync_pte(), try_to_unmap_one(), etc. We cannot do
2873 * much here because ->set_page_dirty is called under VFS locks. The page is
2874 * not necessarily locked.
2876 * We cannot just dirty the page and leave attached buffers clean, because the
2877 * buffers' dirty state is "definitive". We cannot just set the buffers dirty
2878 * or jbddirty because all the journalling code will explode.
2880 * So what we do is to mark the page "pending dirty" and next time writepage
2881 * is called, propagate that into the buffers appropriately.
2883 static int ext4_journalled_set_page_dirty(struct page
*page
)
2885 SetPageChecked(page
);
2886 return __set_page_dirty_nobuffers(page
);
2889 static const struct address_space_operations ext4_ordered_aops
= {
2890 .readpage
= ext4_readpage
,
2891 .readpages
= ext4_readpages
,
2892 .writepage
= ext4_normal_writepage
,
2893 .sync_page
= block_sync_page
,
2894 .write_begin
= ext4_write_begin
,
2895 .write_end
= ext4_ordered_write_end
,
2897 .invalidatepage
= ext4_invalidatepage
,
2898 .releasepage
= ext4_releasepage
,
2899 .direct_IO
= ext4_direct_IO
,
2900 .migratepage
= buffer_migrate_page
,
2901 .is_partially_uptodate
= block_is_partially_uptodate
,
2904 static const struct address_space_operations ext4_writeback_aops
= {
2905 .readpage
= ext4_readpage
,
2906 .readpages
= ext4_readpages
,
2907 .writepage
= ext4_normal_writepage
,
2908 .sync_page
= block_sync_page
,
2909 .write_begin
= ext4_write_begin
,
2910 .write_end
= ext4_writeback_write_end
,
2912 .invalidatepage
= ext4_invalidatepage
,
2913 .releasepage
= ext4_releasepage
,
2914 .direct_IO
= ext4_direct_IO
,
2915 .migratepage
= buffer_migrate_page
,
2916 .is_partially_uptodate
= block_is_partially_uptodate
,
2919 static const struct address_space_operations ext4_journalled_aops
= {
2920 .readpage
= ext4_readpage
,
2921 .readpages
= ext4_readpages
,
2922 .writepage
= ext4_journalled_writepage
,
2923 .sync_page
= block_sync_page
,
2924 .write_begin
= ext4_write_begin
,
2925 .write_end
= ext4_journalled_write_end
,
2926 .set_page_dirty
= ext4_journalled_set_page_dirty
,
2928 .invalidatepage
= ext4_invalidatepage
,
2929 .releasepage
= ext4_releasepage
,
2930 .is_partially_uptodate
= block_is_partially_uptodate
,
2933 static const struct address_space_operations ext4_da_aops
= {
2934 .readpage
= ext4_readpage
,
2935 .readpages
= ext4_readpages
,
2936 .writepage
= ext4_da_writepage
,
2937 .writepages
= ext4_da_writepages
,
2938 .sync_page
= block_sync_page
,
2939 .write_begin
= ext4_da_write_begin
,
2940 .write_end
= ext4_da_write_end
,
2942 .invalidatepage
= ext4_da_invalidatepage
,
2943 .releasepage
= ext4_releasepage
,
2944 .direct_IO
= ext4_direct_IO
,
2945 .migratepage
= buffer_migrate_page
,
2946 .is_partially_uptodate
= block_is_partially_uptodate
,
2949 void ext4_set_aops(struct inode
*inode
)
2951 if (ext4_should_order_data(inode
) &&
2952 test_opt(inode
->i_sb
, DELALLOC
))
2953 inode
->i_mapping
->a_ops
= &ext4_da_aops
;
2954 else if (ext4_should_order_data(inode
))
2955 inode
->i_mapping
->a_ops
= &ext4_ordered_aops
;
2956 else if (ext4_should_writeback_data(inode
) &&
2957 test_opt(inode
->i_sb
, DELALLOC
))
2958 inode
->i_mapping
->a_ops
= &ext4_da_aops
;
2959 else if (ext4_should_writeback_data(inode
))
2960 inode
->i_mapping
->a_ops
= &ext4_writeback_aops
;
2962 inode
->i_mapping
->a_ops
= &ext4_journalled_aops
;
2966 * ext4_block_truncate_page() zeroes out a mapping from file offset `from'
2967 * up to the end of the block which corresponds to `from'.
2968 * This required during truncate. We need to physically zero the tail end
2969 * of that block so it doesn't yield old data if the file is later grown.
2971 int ext4_block_truncate_page(handle_t
*handle
,
2972 struct address_space
*mapping
, loff_t from
)
2974 ext4_fsblk_t index
= from
>> PAGE_CACHE_SHIFT
;
2975 unsigned offset
= from
& (PAGE_CACHE_SIZE
-1);
2976 unsigned blocksize
, length
, pos
;
2978 struct inode
*inode
= mapping
->host
;
2979 struct buffer_head
*bh
;
2983 page
= grab_cache_page(mapping
, from
>> PAGE_CACHE_SHIFT
);
2987 blocksize
= inode
->i_sb
->s_blocksize
;
2988 length
= blocksize
- (offset
& (blocksize
- 1));
2989 iblock
= index
<< (PAGE_CACHE_SHIFT
- inode
->i_sb
->s_blocksize_bits
);
2992 * For "nobh" option, we can only work if we don't need to
2993 * read-in the page - otherwise we create buffers to do the IO.
2995 if (!page_has_buffers(page
) && test_opt(inode
->i_sb
, NOBH
) &&
2996 ext4_should_writeback_data(inode
) && PageUptodate(page
)) {
2997 zero_user(page
, offset
, length
);
2998 set_page_dirty(page
);
3002 if (!page_has_buffers(page
))
3003 create_empty_buffers(page
, blocksize
, 0);
3005 /* Find the buffer that contains "offset" */
3006 bh
= page_buffers(page
);
3008 while (offset
>= pos
) {
3009 bh
= bh
->b_this_page
;
3015 if (buffer_freed(bh
)) {
3016 BUFFER_TRACE(bh
, "freed: skip");
3020 if (!buffer_mapped(bh
)) {
3021 BUFFER_TRACE(bh
, "unmapped");
3022 ext4_get_block(inode
, iblock
, bh
, 0);
3023 /* unmapped? It's a hole - nothing to do */
3024 if (!buffer_mapped(bh
)) {
3025 BUFFER_TRACE(bh
, "still unmapped");
3030 /* Ok, it's mapped. Make sure it's up-to-date */
3031 if (PageUptodate(page
))
3032 set_buffer_uptodate(bh
);
3034 if (!buffer_uptodate(bh
)) {
3036 ll_rw_block(READ
, 1, &bh
);
3038 /* Uhhuh. Read error. Complain and punt. */
3039 if (!buffer_uptodate(bh
))
3043 if (ext4_should_journal_data(inode
)) {
3044 BUFFER_TRACE(bh
, "get write access");
3045 err
= ext4_journal_get_write_access(handle
, bh
);
3050 zero_user(page
, offset
, length
);
3052 BUFFER_TRACE(bh
, "zeroed end of block");
3055 if (ext4_should_journal_data(inode
)) {
3056 err
= ext4_journal_dirty_metadata(handle
, bh
);
3058 if (ext4_should_order_data(inode
))
3059 err
= ext4_jbd2_file_inode(handle
, inode
);
3060 mark_buffer_dirty(bh
);
3065 page_cache_release(page
);
3070 * Probably it should be a library function... search for first non-zero word
3071 * or memcmp with zero_page, whatever is better for particular architecture.
3074 static inline int all_zeroes(__le32
*p
, __le32
*q
)
3083 * ext4_find_shared - find the indirect blocks for partial truncation.
3084 * @inode: inode in question
3085 * @depth: depth of the affected branch
3086 * @offsets: offsets of pointers in that branch (see ext4_block_to_path)
3087 * @chain: place to store the pointers to partial indirect blocks
3088 * @top: place to the (detached) top of branch
3090 * This is a helper function used by ext4_truncate().
3092 * When we do truncate() we may have to clean the ends of several
3093 * indirect blocks but leave the blocks themselves alive. Block is
3094 * partially truncated if some data below the new i_size is refered
3095 * from it (and it is on the path to the first completely truncated
3096 * data block, indeed). We have to free the top of that path along
3097 * with everything to the right of the path. Since no allocation
3098 * past the truncation point is possible until ext4_truncate()
3099 * finishes, we may safely do the latter, but top of branch may
3100 * require special attention - pageout below the truncation point
3101 * might try to populate it.
3103 * We atomically detach the top of branch from the tree, store the
3104 * block number of its root in *@top, pointers to buffer_heads of
3105 * partially truncated blocks - in @chain[].bh and pointers to
3106 * their last elements that should not be removed - in
3107 * @chain[].p. Return value is the pointer to last filled element
3110 * The work left to caller to do the actual freeing of subtrees:
3111 * a) free the subtree starting from *@top
3112 * b) free the subtrees whose roots are stored in
3113 * (@chain[i].p+1 .. end of @chain[i].bh->b_data)
3114 * c) free the subtrees growing from the inode past the @chain[0].
3115 * (no partially truncated stuff there). */
3117 static Indirect
*ext4_find_shared(struct inode
*inode
, int depth
,
3118 ext4_lblk_t offsets
[4], Indirect chain
[4], __le32
*top
)
3120 Indirect
*partial
, *p
;
3124 /* Make k index the deepest non-null offest + 1 */
3125 for (k
= depth
; k
> 1 && !offsets
[k
-1]; k
--)
3127 partial
= ext4_get_branch(inode
, k
, offsets
, chain
, &err
);
3128 /* Writer: pointers */
3130 partial
= chain
+ k
-1;
3132 * If the branch acquired continuation since we've looked at it -
3133 * fine, it should all survive and (new) top doesn't belong to us.
3135 if (!partial
->key
&& *partial
->p
)
3138 for (p
=partial
; p
>chain
&& all_zeroes((__le32
*)p
->bh
->b_data
,p
->p
); p
--)
3141 * OK, we've found the last block that must survive. The rest of our
3142 * branch should be detached before unlocking. However, if that rest
3143 * of branch is all ours and does not grow immediately from the inode
3144 * it's easier to cheat and just decrement partial->p.
3146 if (p
== chain
+ k
- 1 && p
> chain
) {
3150 /* Nope, don't do this in ext4. Must leave the tree intact */
3157 while(partial
> p
) {
3158 brelse(partial
->bh
);
3166 * Zero a number of block pointers in either an inode or an indirect block.
3167 * If we restart the transaction we must again get write access to the
3168 * indirect block for further modification.
3170 * We release `count' blocks on disk, but (last - first) may be greater
3171 * than `count' because there can be holes in there.
3173 static void ext4_clear_blocks(handle_t
*handle
, struct inode
*inode
,
3174 struct buffer_head
*bh
, ext4_fsblk_t block_to_free
,
3175 unsigned long count
, __le32
*first
, __le32
*last
)
3178 if (try_to_extend_transaction(handle
, inode
)) {
3180 BUFFER_TRACE(bh
, "call ext4_journal_dirty_metadata");
3181 ext4_journal_dirty_metadata(handle
, bh
);
3183 ext4_mark_inode_dirty(handle
, inode
);
3184 ext4_journal_test_restart(handle
, inode
);
3186 BUFFER_TRACE(bh
, "retaking write access");
3187 ext4_journal_get_write_access(handle
, bh
);
3192 * Any buffers which are on the journal will be in memory. We find
3193 * them on the hash table so jbd2_journal_revoke() will run jbd2_journal_forget()
3194 * on them. We've already detached each block from the file, so
3195 * bforget() in jbd2_journal_forget() should be safe.
3197 * AKPM: turn on bforget in jbd2_journal_forget()!!!
3199 for (p
= first
; p
< last
; p
++) {
3200 u32 nr
= le32_to_cpu(*p
);
3202 struct buffer_head
*tbh
;
3205 tbh
= sb_find_get_block(inode
->i_sb
, nr
);
3206 ext4_forget(handle
, 0, inode
, tbh
, nr
);
3210 ext4_free_blocks(handle
, inode
, block_to_free
, count
, 0);
3214 * ext4_free_data - free a list of data blocks
3215 * @handle: handle for this transaction
3216 * @inode: inode we are dealing with
3217 * @this_bh: indirect buffer_head which contains *@first and *@last
3218 * @first: array of block numbers
3219 * @last: points immediately past the end of array
3221 * We are freeing all blocks refered from that array (numbers are stored as
3222 * little-endian 32-bit) and updating @inode->i_blocks appropriately.
3224 * We accumulate contiguous runs of blocks to free. Conveniently, if these
3225 * blocks are contiguous then releasing them at one time will only affect one
3226 * or two bitmap blocks (+ group descriptor(s) and superblock) and we won't
3227 * actually use a lot of journal space.
3229 * @this_bh will be %NULL if @first and @last point into the inode's direct
3232 static void ext4_free_data(handle_t
*handle
, struct inode
*inode
,
3233 struct buffer_head
*this_bh
,
3234 __le32
*first
, __le32
*last
)
3236 ext4_fsblk_t block_to_free
= 0; /* Starting block # of a run */
3237 unsigned long count
= 0; /* Number of blocks in the run */
3238 __le32
*block_to_free_p
= NULL
; /* Pointer into inode/ind
3241 ext4_fsblk_t nr
; /* Current block # */
3242 __le32
*p
; /* Pointer into inode/ind
3243 for current block */
3246 if (this_bh
) { /* For indirect block */
3247 BUFFER_TRACE(this_bh
, "get_write_access");
3248 err
= ext4_journal_get_write_access(handle
, this_bh
);
3249 /* Important: if we can't update the indirect pointers
3250 * to the blocks, we can't free them. */
3255 for (p
= first
; p
< last
; p
++) {
3256 nr
= le32_to_cpu(*p
);
3258 /* accumulate blocks to free if they're contiguous */
3261 block_to_free_p
= p
;
3263 } else if (nr
== block_to_free
+ count
) {
3266 ext4_clear_blocks(handle
, inode
, this_bh
,
3268 count
, block_to_free_p
, p
);
3270 block_to_free_p
= p
;
3277 ext4_clear_blocks(handle
, inode
, this_bh
, block_to_free
,
3278 count
, block_to_free_p
, p
);
3281 BUFFER_TRACE(this_bh
, "call ext4_journal_dirty_metadata");
3284 * The buffer head should have an attached journal head at this
3285 * point. However, if the data is corrupted and an indirect
3286 * block pointed to itself, it would have been detached when
3287 * the block was cleared. Check for this instead of OOPSing.
3290 ext4_journal_dirty_metadata(handle
, this_bh
);
3292 ext4_error(inode
->i_sb
, __func__
,
3293 "circular indirect block detected, "
3294 "inode=%lu, block=%llu",
3296 (unsigned long long) this_bh
->b_blocknr
);
3301 * ext4_free_branches - free an array of branches
3302 * @handle: JBD handle for this transaction
3303 * @inode: inode we are dealing with
3304 * @parent_bh: the buffer_head which contains *@first and *@last
3305 * @first: array of block numbers
3306 * @last: pointer immediately past the end of array
3307 * @depth: depth of the branches to free
3309 * We are freeing all blocks refered from these branches (numbers are
3310 * stored as little-endian 32-bit) and updating @inode->i_blocks
3313 static void ext4_free_branches(handle_t
*handle
, struct inode
*inode
,
3314 struct buffer_head
*parent_bh
,
3315 __le32
*first
, __le32
*last
, int depth
)
3320 if (is_handle_aborted(handle
))
3324 struct buffer_head
*bh
;
3325 int addr_per_block
= EXT4_ADDR_PER_BLOCK(inode
->i_sb
);
3327 while (--p
>= first
) {
3328 nr
= le32_to_cpu(*p
);
3330 continue; /* A hole */
3332 /* Go read the buffer for the next level down */
3333 bh
= sb_bread(inode
->i_sb
, nr
);
3336 * A read failure? Report error and clear slot
3340 ext4_error(inode
->i_sb
, "ext4_free_branches",
3341 "Read failure, inode=%lu, block=%llu",
3346 /* This zaps the entire block. Bottom up. */
3347 BUFFER_TRACE(bh
, "free child branches");
3348 ext4_free_branches(handle
, inode
, bh
,
3349 (__le32
*)bh
->b_data
,
3350 (__le32
*)bh
->b_data
+ addr_per_block
,
3354 * We've probably journalled the indirect block several
3355 * times during the truncate. But it's no longer
3356 * needed and we now drop it from the transaction via
3357 * jbd2_journal_revoke().
3359 * That's easy if it's exclusively part of this
3360 * transaction. But if it's part of the committing
3361 * transaction then jbd2_journal_forget() will simply
3362 * brelse() it. That means that if the underlying
3363 * block is reallocated in ext4_get_block(),
3364 * unmap_underlying_metadata() will find this block
3365 * and will try to get rid of it. damn, damn.
3367 * If this block has already been committed to the
3368 * journal, a revoke record will be written. And
3369 * revoke records must be emitted *before* clearing
3370 * this block's bit in the bitmaps.
3372 ext4_forget(handle
, 1, inode
, bh
, bh
->b_blocknr
);
3375 * Everything below this this pointer has been
3376 * released. Now let this top-of-subtree go.
3378 * We want the freeing of this indirect block to be
3379 * atomic in the journal with the updating of the
3380 * bitmap block which owns it. So make some room in
3383 * We zero the parent pointer *after* freeing its
3384 * pointee in the bitmaps, so if extend_transaction()
3385 * for some reason fails to put the bitmap changes and
3386 * the release into the same transaction, recovery
3387 * will merely complain about releasing a free block,
3388 * rather than leaking blocks.
3390 if (is_handle_aborted(handle
))
3392 if (try_to_extend_transaction(handle
, inode
)) {
3393 ext4_mark_inode_dirty(handle
, inode
);
3394 ext4_journal_test_restart(handle
, inode
);
3397 ext4_free_blocks(handle
, inode
, nr
, 1, 1);
3401 * The block which we have just freed is
3402 * pointed to by an indirect block: journal it
3404 BUFFER_TRACE(parent_bh
, "get_write_access");
3405 if (!ext4_journal_get_write_access(handle
,
3408 BUFFER_TRACE(parent_bh
,
3409 "call ext4_journal_dirty_metadata");
3410 ext4_journal_dirty_metadata(handle
,
3416 /* We have reached the bottom of the tree. */
3417 BUFFER_TRACE(parent_bh
, "free data blocks");
3418 ext4_free_data(handle
, inode
, parent_bh
, first
, last
);
3422 int ext4_can_truncate(struct inode
*inode
)
3424 if (IS_APPEND(inode
) || IS_IMMUTABLE(inode
))
3426 if (S_ISREG(inode
->i_mode
))
3428 if (S_ISDIR(inode
->i_mode
))
3430 if (S_ISLNK(inode
->i_mode
))
3431 return !ext4_inode_is_fast_symlink(inode
);
3438 * We block out ext4_get_block() block instantiations across the entire
3439 * transaction, and VFS/VM ensures that ext4_truncate() cannot run
3440 * simultaneously on behalf of the same inode.
3442 * As we work through the truncate and commmit bits of it to the journal there
3443 * is one core, guiding principle: the file's tree must always be consistent on
3444 * disk. We must be able to restart the truncate after a crash.
3446 * The file's tree may be transiently inconsistent in memory (although it
3447 * probably isn't), but whenever we close off and commit a journal transaction,
3448 * the contents of (the filesystem + the journal) must be consistent and
3449 * restartable. It's pretty simple, really: bottom up, right to left (although
3450 * left-to-right works OK too).
3452 * Note that at recovery time, journal replay occurs *before* the restart of
3453 * truncate against the orphan inode list.
3455 * The committed inode has the new, desired i_size (which is the same as
3456 * i_disksize in this case). After a crash, ext4_orphan_cleanup() will see
3457 * that this inode's truncate did not complete and it will again call
3458 * ext4_truncate() to have another go. So there will be instantiated blocks
3459 * to the right of the truncation point in a crashed ext4 filesystem. But
3460 * that's fine - as long as they are linked from the inode, the post-crash
3461 * ext4_truncate() run will find them and release them.
3463 void ext4_truncate(struct inode
*inode
)
3466 struct ext4_inode_info
*ei
= EXT4_I(inode
);
3467 __le32
*i_data
= ei
->i_data
;
3468 int addr_per_block
= EXT4_ADDR_PER_BLOCK(inode
->i_sb
);
3469 struct address_space
*mapping
= inode
->i_mapping
;
3470 ext4_lblk_t offsets
[4];
3475 ext4_lblk_t last_block
;
3476 unsigned blocksize
= inode
->i_sb
->s_blocksize
;
3478 if (!ext4_can_truncate(inode
))
3481 if (EXT4_I(inode
)->i_flags
& EXT4_EXTENTS_FL
) {
3482 ext4_ext_truncate(inode
);
3486 handle
= start_transaction(inode
);
3488 return; /* AKPM: return what? */
3490 last_block
= (inode
->i_size
+ blocksize
-1)
3491 >> EXT4_BLOCK_SIZE_BITS(inode
->i_sb
);
3493 if (inode
->i_size
& (blocksize
- 1))
3494 if (ext4_block_truncate_page(handle
, mapping
, inode
->i_size
))
3497 n
= ext4_block_to_path(inode
, last_block
, offsets
, NULL
);
3499 goto out_stop
; /* error */
3502 * OK. This truncate is going to happen. We add the inode to the
3503 * orphan list, so that if this truncate spans multiple transactions,
3504 * and we crash, we will resume the truncate when the filesystem
3505 * recovers. It also marks the inode dirty, to catch the new size.
3507 * Implication: the file must always be in a sane, consistent
3508 * truncatable state while each transaction commits.
3510 if (ext4_orphan_add(handle
, inode
))
3514 * From here we block out all ext4_get_block() callers who want to
3515 * modify the block allocation tree.
3517 down_write(&ei
->i_data_sem
);
3519 ext4_discard_reservation(inode
);
3522 * The orphan list entry will now protect us from any crash which
3523 * occurs before the truncate completes, so it is now safe to propagate
3524 * the new, shorter inode size (held for now in i_size) into the
3525 * on-disk inode. We do this via i_disksize, which is the value which
3526 * ext4 *really* writes onto the disk inode.
3528 ei
->i_disksize
= inode
->i_size
;
3530 if (n
== 1) { /* direct blocks */
3531 ext4_free_data(handle
, inode
, NULL
, i_data
+offsets
[0],
3532 i_data
+ EXT4_NDIR_BLOCKS
);
3536 partial
= ext4_find_shared(inode
, n
, offsets
, chain
, &nr
);
3537 /* Kill the top of shared branch (not detached) */
3539 if (partial
== chain
) {
3540 /* Shared branch grows from the inode */
3541 ext4_free_branches(handle
, inode
, NULL
,
3542 &nr
, &nr
+1, (chain
+n
-1) - partial
);
3545 * We mark the inode dirty prior to restart,
3546 * and prior to stop. No need for it here.
3549 /* Shared branch grows from an indirect block */
3550 BUFFER_TRACE(partial
->bh
, "get_write_access");
3551 ext4_free_branches(handle
, inode
, partial
->bh
,
3553 partial
->p
+1, (chain
+n
-1) - partial
);
3556 /* Clear the ends of indirect blocks on the shared branch */
3557 while (partial
> chain
) {
3558 ext4_free_branches(handle
, inode
, partial
->bh
, partial
->p
+ 1,
3559 (__le32
*)partial
->bh
->b_data
+addr_per_block
,
3560 (chain
+n
-1) - partial
);
3561 BUFFER_TRACE(partial
->bh
, "call brelse");
3562 brelse (partial
->bh
);
3566 /* Kill the remaining (whole) subtrees */
3567 switch (offsets
[0]) {
3569 nr
= i_data
[EXT4_IND_BLOCK
];
3571 ext4_free_branches(handle
, inode
, NULL
, &nr
, &nr
+1, 1);
3572 i_data
[EXT4_IND_BLOCK
] = 0;
3574 case EXT4_IND_BLOCK
:
3575 nr
= i_data
[EXT4_DIND_BLOCK
];
3577 ext4_free_branches(handle
, inode
, NULL
, &nr
, &nr
+1, 2);
3578 i_data
[EXT4_DIND_BLOCK
] = 0;
3580 case EXT4_DIND_BLOCK
:
3581 nr
= i_data
[EXT4_TIND_BLOCK
];
3583 ext4_free_branches(handle
, inode
, NULL
, &nr
, &nr
+1, 3);
3584 i_data
[EXT4_TIND_BLOCK
] = 0;
3586 case EXT4_TIND_BLOCK
:
3590 up_write(&ei
->i_data_sem
);
3591 inode
->i_mtime
= inode
->i_ctime
= ext4_current_time(inode
);
3592 ext4_mark_inode_dirty(handle
, inode
);
3595 * In a multi-transaction truncate, we only make the final transaction
3602 * If this was a simple ftruncate(), and the file will remain alive
3603 * then we need to clear up the orphan record which we created above.
3604 * However, if this was a real unlink then we were called by
3605 * ext4_delete_inode(), and we allow that function to clean up the
3606 * orphan info for us.
3609 ext4_orphan_del(handle
, inode
);
3611 ext4_journal_stop(handle
);
3614 static ext4_fsblk_t
ext4_get_inode_block(struct super_block
*sb
,
3615 unsigned long ino
, struct ext4_iloc
*iloc
)
3617 ext4_group_t block_group
;
3618 unsigned long offset
;
3620 struct ext4_group_desc
*gdp
;
3622 if (!ext4_valid_inum(sb
, ino
)) {
3624 * This error is already checked for in namei.c unless we are
3625 * looking at an NFS filehandle, in which case no error
3631 block_group
= (ino
- 1) / EXT4_INODES_PER_GROUP(sb
);
3632 gdp
= ext4_get_group_desc(sb
, block_group
, NULL
);
3637 * Figure out the offset within the block group inode table
3639 offset
= ((ino
- 1) % EXT4_INODES_PER_GROUP(sb
)) *
3640 EXT4_INODE_SIZE(sb
);
3641 block
= ext4_inode_table(sb
, gdp
) +
3642 (offset
>> EXT4_BLOCK_SIZE_BITS(sb
));
3644 iloc
->block_group
= block_group
;
3645 iloc
->offset
= offset
& (EXT4_BLOCK_SIZE(sb
) - 1);
3650 * ext4_get_inode_loc returns with an extra refcount against the inode's
3651 * underlying buffer_head on success. If 'in_mem' is true, we have all
3652 * data in memory that is needed to recreate the on-disk version of this
3655 static int __ext4_get_inode_loc(struct inode
*inode
,
3656 struct ext4_iloc
*iloc
, int in_mem
)
3659 struct buffer_head
*bh
;
3661 block
= ext4_get_inode_block(inode
->i_sb
, inode
->i_ino
, iloc
);
3665 bh
= sb_getblk(inode
->i_sb
, block
);
3667 ext4_error (inode
->i_sb
, "ext4_get_inode_loc",
3668 "unable to read inode block - "
3669 "inode=%lu, block=%llu",
3670 inode
->i_ino
, block
);
3673 if (!buffer_uptodate(bh
)) {
3677 * If the buffer has the write error flag, we have failed
3678 * to write out another inode in the same block. In this
3679 * case, we don't have to read the block because we may
3680 * read the old inode data successfully.
3682 if (buffer_write_io_error(bh
) && !buffer_uptodate(bh
))
3683 set_buffer_uptodate(bh
);
3685 if (buffer_uptodate(bh
)) {
3686 /* someone brought it uptodate while we waited */
3692 * If we have all information of the inode in memory and this
3693 * is the only valid inode in the block, we need not read the
3697 struct buffer_head
*bitmap_bh
;
3698 struct ext4_group_desc
*desc
;
3699 int inodes_per_buffer
;
3700 int inode_offset
, i
;
3701 ext4_group_t block_group
;
3704 block_group
= (inode
->i_ino
- 1) /
3705 EXT4_INODES_PER_GROUP(inode
->i_sb
);
3706 inodes_per_buffer
= bh
->b_size
/
3707 EXT4_INODE_SIZE(inode
->i_sb
);
3708 inode_offset
= ((inode
->i_ino
- 1) %
3709 EXT4_INODES_PER_GROUP(inode
->i_sb
));
3710 start
= inode_offset
& ~(inodes_per_buffer
- 1);
3712 /* Is the inode bitmap in cache? */
3713 desc
= ext4_get_group_desc(inode
->i_sb
,
3718 bitmap_bh
= sb_getblk(inode
->i_sb
,
3719 ext4_inode_bitmap(inode
->i_sb
, desc
));
3724 * If the inode bitmap isn't in cache then the
3725 * optimisation may end up performing two reads instead
3726 * of one, so skip it.
3728 if (!buffer_uptodate(bitmap_bh
)) {
3732 for (i
= start
; i
< start
+ inodes_per_buffer
; i
++) {
3733 if (i
== inode_offset
)
3735 if (ext4_test_bit(i
, bitmap_bh
->b_data
))
3739 if (i
== start
+ inodes_per_buffer
) {
3740 /* all other inodes are free, so skip I/O */
3741 memset(bh
->b_data
, 0, bh
->b_size
);
3742 set_buffer_uptodate(bh
);
3750 * There are other valid inodes in the buffer, this inode
3751 * has in-inode xattrs, or we don't have this inode in memory.
3752 * Read the block from disk.
3755 bh
->b_end_io
= end_buffer_read_sync
;
3756 submit_bh(READ_META
, bh
);
3758 if (!buffer_uptodate(bh
)) {
3759 ext4_error(inode
->i_sb
, "ext4_get_inode_loc",
3760 "unable to read inode block - "
3761 "inode=%lu, block=%llu",
3762 inode
->i_ino
, block
);
3772 int ext4_get_inode_loc(struct inode
*inode
, struct ext4_iloc
*iloc
)
3774 /* We have all inode data except xattrs in memory here. */
3775 return __ext4_get_inode_loc(inode
, iloc
,
3776 !(EXT4_I(inode
)->i_state
& EXT4_STATE_XATTR
));
3779 void ext4_set_inode_flags(struct inode
*inode
)
3781 unsigned int flags
= EXT4_I(inode
)->i_flags
;
3783 inode
->i_flags
&= ~(S_SYNC
|S_APPEND
|S_IMMUTABLE
|S_NOATIME
|S_DIRSYNC
);
3784 if (flags
& EXT4_SYNC_FL
)
3785 inode
->i_flags
|= S_SYNC
;
3786 if (flags
& EXT4_APPEND_FL
)
3787 inode
->i_flags
|= S_APPEND
;
3788 if (flags
& EXT4_IMMUTABLE_FL
)
3789 inode
->i_flags
|= S_IMMUTABLE
;
3790 if (flags
& EXT4_NOATIME_FL
)
3791 inode
->i_flags
|= S_NOATIME
;
3792 if (flags
& EXT4_DIRSYNC_FL
)
3793 inode
->i_flags
|= S_DIRSYNC
;
3796 /* Propagate flags from i_flags to EXT4_I(inode)->i_flags */
3797 void ext4_get_inode_flags(struct ext4_inode_info
*ei
)
3799 unsigned int flags
= ei
->vfs_inode
.i_flags
;
3801 ei
->i_flags
&= ~(EXT4_SYNC_FL
|EXT4_APPEND_FL
|
3802 EXT4_IMMUTABLE_FL
|EXT4_NOATIME_FL
|EXT4_DIRSYNC_FL
);
3804 ei
->i_flags
|= EXT4_SYNC_FL
;
3805 if (flags
& S_APPEND
)
3806 ei
->i_flags
|= EXT4_APPEND_FL
;
3807 if (flags
& S_IMMUTABLE
)
3808 ei
->i_flags
|= EXT4_IMMUTABLE_FL
;
3809 if (flags
& S_NOATIME
)
3810 ei
->i_flags
|= EXT4_NOATIME_FL
;
3811 if (flags
& S_DIRSYNC
)
3812 ei
->i_flags
|= EXT4_DIRSYNC_FL
;
3814 static blkcnt_t
ext4_inode_blocks(struct ext4_inode
*raw_inode
,
3815 struct ext4_inode_info
*ei
)
3818 struct inode
*inode
= &(ei
->vfs_inode
);
3819 struct super_block
*sb
= inode
->i_sb
;
3821 if (EXT4_HAS_RO_COMPAT_FEATURE(sb
,
3822 EXT4_FEATURE_RO_COMPAT_HUGE_FILE
)) {
3823 /* we are using combined 48 bit field */
3824 i_blocks
= ((u64
)le16_to_cpu(raw_inode
->i_blocks_high
)) << 32 |
3825 le32_to_cpu(raw_inode
->i_blocks_lo
);
3826 if (ei
->i_flags
& EXT4_HUGE_FILE_FL
) {
3827 /* i_blocks represent file system block size */
3828 return i_blocks
<< (inode
->i_blkbits
- 9);
3833 return le32_to_cpu(raw_inode
->i_blocks_lo
);
3837 struct inode
*ext4_iget(struct super_block
*sb
, unsigned long ino
)
3839 struct ext4_iloc iloc
;
3840 struct ext4_inode
*raw_inode
;
3841 struct ext4_inode_info
*ei
;
3842 struct buffer_head
*bh
;
3843 struct inode
*inode
;
3847 inode
= iget_locked(sb
, ino
);
3849 return ERR_PTR(-ENOMEM
);
3850 if (!(inode
->i_state
& I_NEW
))
3854 #ifdef CONFIG_EXT4DEV_FS_POSIX_ACL
3855 ei
->i_acl
= EXT4_ACL_NOT_CACHED
;
3856 ei
->i_default_acl
= EXT4_ACL_NOT_CACHED
;
3858 ei
->i_block_alloc_info
= NULL
;
3860 ret
= __ext4_get_inode_loc(inode
, &iloc
, 0);
3864 raw_inode
= ext4_raw_inode(&iloc
);
3865 inode
->i_mode
= le16_to_cpu(raw_inode
->i_mode
);
3866 inode
->i_uid
= (uid_t
)le16_to_cpu(raw_inode
->i_uid_low
);
3867 inode
->i_gid
= (gid_t
)le16_to_cpu(raw_inode
->i_gid_low
);
3868 if(!(test_opt (inode
->i_sb
, NO_UID32
))) {
3869 inode
->i_uid
|= le16_to_cpu(raw_inode
->i_uid_high
) << 16;
3870 inode
->i_gid
|= le16_to_cpu(raw_inode
->i_gid_high
) << 16;
3872 inode
->i_nlink
= le16_to_cpu(raw_inode
->i_links_count
);
3875 ei
->i_dir_start_lookup
= 0;
3876 ei
->i_dtime
= le32_to_cpu(raw_inode
->i_dtime
);
3877 /* We now have enough fields to check if the inode was active or not.
3878 * This is needed because nfsd might try to access dead inodes
3879 * the test is that same one that e2fsck uses
3880 * NeilBrown 1999oct15
3882 if (inode
->i_nlink
== 0) {
3883 if (inode
->i_mode
== 0 ||
3884 !(EXT4_SB(inode
->i_sb
)->s_mount_state
& EXT4_ORPHAN_FS
)) {
3885 /* this inode is deleted */
3890 /* The only unlinked inodes we let through here have
3891 * valid i_mode and are being read by the orphan
3892 * recovery code: that's fine, we're about to complete
3893 * the process of deleting those. */
3895 ei
->i_flags
= le32_to_cpu(raw_inode
->i_flags
);
3896 inode
->i_blocks
= ext4_inode_blocks(raw_inode
, ei
);
3897 ei
->i_file_acl
= le32_to_cpu(raw_inode
->i_file_acl_lo
);
3898 if (EXT4_SB(inode
->i_sb
)->s_es
->s_creator_os
!=
3899 cpu_to_le32(EXT4_OS_HURD
)) {
3901 ((__u64
)le16_to_cpu(raw_inode
->i_file_acl_high
)) << 32;
3903 inode
->i_size
= ext4_isize(raw_inode
);
3904 ei
->i_disksize
= inode
->i_size
;
3905 inode
->i_generation
= le32_to_cpu(raw_inode
->i_generation
);
3906 ei
->i_block_group
= iloc
.block_group
;
3908 * NOTE! The in-memory inode i_data array is in little-endian order
3909 * even on big-endian machines: we do NOT byteswap the block numbers!
3911 for (block
= 0; block
< EXT4_N_BLOCKS
; block
++)
3912 ei
->i_data
[block
] = raw_inode
->i_block
[block
];
3913 INIT_LIST_HEAD(&ei
->i_orphan
);
3915 if (EXT4_INODE_SIZE(inode
->i_sb
) > EXT4_GOOD_OLD_INODE_SIZE
) {
3916 ei
->i_extra_isize
= le16_to_cpu(raw_inode
->i_extra_isize
);
3917 if (EXT4_GOOD_OLD_INODE_SIZE
+ ei
->i_extra_isize
>
3918 EXT4_INODE_SIZE(inode
->i_sb
)) {
3923 if (ei
->i_extra_isize
== 0) {
3924 /* The extra space is currently unused. Use it. */
3925 ei
->i_extra_isize
= sizeof(struct ext4_inode
) -
3926 EXT4_GOOD_OLD_INODE_SIZE
;
3928 __le32
*magic
= (void *)raw_inode
+
3929 EXT4_GOOD_OLD_INODE_SIZE
+
3931 if (*magic
== cpu_to_le32(EXT4_XATTR_MAGIC
))
3932 ei
->i_state
|= EXT4_STATE_XATTR
;
3935 ei
->i_extra_isize
= 0;
3937 EXT4_INODE_GET_XTIME(i_ctime
, inode
, raw_inode
);
3938 EXT4_INODE_GET_XTIME(i_mtime
, inode
, raw_inode
);
3939 EXT4_INODE_GET_XTIME(i_atime
, inode
, raw_inode
);
3940 EXT4_EINODE_GET_XTIME(i_crtime
, ei
, raw_inode
);
3942 inode
->i_version
= le32_to_cpu(raw_inode
->i_disk_version
);
3943 if (EXT4_INODE_SIZE(inode
->i_sb
) > EXT4_GOOD_OLD_INODE_SIZE
) {
3944 if (EXT4_FITS_IN_INODE(raw_inode
, ei
, i_version_hi
))
3946 (__u64
)(le32_to_cpu(raw_inode
->i_version_hi
)) << 32;
3949 if (S_ISREG(inode
->i_mode
)) {
3950 inode
->i_op
= &ext4_file_inode_operations
;
3951 inode
->i_fop
= &ext4_file_operations
;
3952 ext4_set_aops(inode
);
3953 } else if (S_ISDIR(inode
->i_mode
)) {
3954 inode
->i_op
= &ext4_dir_inode_operations
;
3955 inode
->i_fop
= &ext4_dir_operations
;
3956 } else if (S_ISLNK(inode
->i_mode
)) {
3957 if (ext4_inode_is_fast_symlink(inode
))
3958 inode
->i_op
= &ext4_fast_symlink_inode_operations
;
3960 inode
->i_op
= &ext4_symlink_inode_operations
;
3961 ext4_set_aops(inode
);
3964 inode
->i_op
= &ext4_special_inode_operations
;
3965 if (raw_inode
->i_block
[0])
3966 init_special_inode(inode
, inode
->i_mode
,
3967 old_decode_dev(le32_to_cpu(raw_inode
->i_block
[0])));
3969 init_special_inode(inode
, inode
->i_mode
,
3970 new_decode_dev(le32_to_cpu(raw_inode
->i_block
[1])));
3973 ext4_set_inode_flags(inode
);
3974 unlock_new_inode(inode
);
3979 return ERR_PTR(ret
);
3982 static int ext4_inode_blocks_set(handle_t
*handle
,
3983 struct ext4_inode
*raw_inode
,
3984 struct ext4_inode_info
*ei
)
3986 struct inode
*inode
= &(ei
->vfs_inode
);
3987 u64 i_blocks
= inode
->i_blocks
;
3988 struct super_block
*sb
= inode
->i_sb
;
3991 if (i_blocks
<= ~0U) {
3993 * i_blocks can be represnted in a 32 bit variable
3994 * as multiple of 512 bytes
3996 raw_inode
->i_blocks_lo
= cpu_to_le32(i_blocks
);
3997 raw_inode
->i_blocks_high
= 0;
3998 ei
->i_flags
&= ~EXT4_HUGE_FILE_FL
;
3999 } else if (i_blocks
<= 0xffffffffffffULL
) {
4001 * i_blocks can be represented in a 48 bit variable
4002 * as multiple of 512 bytes
4004 err
= ext4_update_rocompat_feature(handle
, sb
,
4005 EXT4_FEATURE_RO_COMPAT_HUGE_FILE
);
4008 /* i_block is stored in the split 48 bit fields */
4009 raw_inode
->i_blocks_lo
= cpu_to_le32(i_blocks
);
4010 raw_inode
->i_blocks_high
= cpu_to_le16(i_blocks
>> 32);
4011 ei
->i_flags
&= ~EXT4_HUGE_FILE_FL
;
4014 * i_blocks should be represented in a 48 bit variable
4015 * as multiple of file system block size
4017 err
= ext4_update_rocompat_feature(handle
, sb
,
4018 EXT4_FEATURE_RO_COMPAT_HUGE_FILE
);
4021 ei
->i_flags
|= EXT4_HUGE_FILE_FL
;
4022 /* i_block is stored in file system block size */
4023 i_blocks
= i_blocks
>> (inode
->i_blkbits
- 9);
4024 raw_inode
->i_blocks_lo
= cpu_to_le32(i_blocks
);
4025 raw_inode
->i_blocks_high
= cpu_to_le16(i_blocks
>> 32);
4032 * Post the struct inode info into an on-disk inode location in the
4033 * buffer-cache. This gobbles the caller's reference to the
4034 * buffer_head in the inode location struct.
4036 * The caller must have write access to iloc->bh.
4038 static int ext4_do_update_inode(handle_t
*handle
,
4039 struct inode
*inode
,
4040 struct ext4_iloc
*iloc
)
4042 struct ext4_inode
*raw_inode
= ext4_raw_inode(iloc
);
4043 struct ext4_inode_info
*ei
= EXT4_I(inode
);
4044 struct buffer_head
*bh
= iloc
->bh
;
4045 int err
= 0, rc
, block
;
4047 /* For fields not not tracking in the in-memory inode,
4048 * initialise them to zero for new inodes. */
4049 if (ei
->i_state
& EXT4_STATE_NEW
)
4050 memset(raw_inode
, 0, EXT4_SB(inode
->i_sb
)->s_inode_size
);
4052 ext4_get_inode_flags(ei
);
4053 raw_inode
->i_mode
= cpu_to_le16(inode
->i_mode
);
4054 if(!(test_opt(inode
->i_sb
, NO_UID32
))) {
4055 raw_inode
->i_uid_low
= cpu_to_le16(low_16_bits(inode
->i_uid
));
4056 raw_inode
->i_gid_low
= cpu_to_le16(low_16_bits(inode
->i_gid
));
4058 * Fix up interoperability with old kernels. Otherwise, old inodes get
4059 * re-used with the upper 16 bits of the uid/gid intact
4062 raw_inode
->i_uid_high
=
4063 cpu_to_le16(high_16_bits(inode
->i_uid
));
4064 raw_inode
->i_gid_high
=
4065 cpu_to_le16(high_16_bits(inode
->i_gid
));
4067 raw_inode
->i_uid_high
= 0;
4068 raw_inode
->i_gid_high
= 0;
4071 raw_inode
->i_uid_low
=
4072 cpu_to_le16(fs_high2lowuid(inode
->i_uid
));
4073 raw_inode
->i_gid_low
=
4074 cpu_to_le16(fs_high2lowgid(inode
->i_gid
));
4075 raw_inode
->i_uid_high
= 0;
4076 raw_inode
->i_gid_high
= 0;
4078 raw_inode
->i_links_count
= cpu_to_le16(inode
->i_nlink
);
4080 EXT4_INODE_SET_XTIME(i_ctime
, inode
, raw_inode
);
4081 EXT4_INODE_SET_XTIME(i_mtime
, inode
, raw_inode
);
4082 EXT4_INODE_SET_XTIME(i_atime
, inode
, raw_inode
);
4083 EXT4_EINODE_SET_XTIME(i_crtime
, ei
, raw_inode
);
4085 if (ext4_inode_blocks_set(handle
, raw_inode
, ei
))
4087 raw_inode
->i_dtime
= cpu_to_le32(ei
->i_dtime
);
4088 /* clear the migrate flag in the raw_inode */
4089 raw_inode
->i_flags
= cpu_to_le32(ei
->i_flags
& ~EXT4_EXT_MIGRATE
);
4090 if (EXT4_SB(inode
->i_sb
)->s_es
->s_creator_os
!=
4091 cpu_to_le32(EXT4_OS_HURD
))
4092 raw_inode
->i_file_acl_high
=
4093 cpu_to_le16(ei
->i_file_acl
>> 32);
4094 raw_inode
->i_file_acl_lo
= cpu_to_le32(ei
->i_file_acl
);
4095 ext4_isize_set(raw_inode
, ei
->i_disksize
);
4096 if (ei
->i_disksize
> 0x7fffffffULL
) {
4097 struct super_block
*sb
= inode
->i_sb
;
4098 if (!EXT4_HAS_RO_COMPAT_FEATURE(sb
,
4099 EXT4_FEATURE_RO_COMPAT_LARGE_FILE
) ||
4100 EXT4_SB(sb
)->s_es
->s_rev_level
==
4101 cpu_to_le32(EXT4_GOOD_OLD_REV
)) {
4102 /* If this is the first large file
4103 * created, add a flag to the superblock.
4105 err
= ext4_journal_get_write_access(handle
,
4106 EXT4_SB(sb
)->s_sbh
);
4109 ext4_update_dynamic_rev(sb
);
4110 EXT4_SET_RO_COMPAT_FEATURE(sb
,
4111 EXT4_FEATURE_RO_COMPAT_LARGE_FILE
);
4114 err
= ext4_journal_dirty_metadata(handle
,
4115 EXT4_SB(sb
)->s_sbh
);
4118 raw_inode
->i_generation
= cpu_to_le32(inode
->i_generation
);
4119 if (S_ISCHR(inode
->i_mode
) || S_ISBLK(inode
->i_mode
)) {
4120 if (old_valid_dev(inode
->i_rdev
)) {
4121 raw_inode
->i_block
[0] =
4122 cpu_to_le32(old_encode_dev(inode
->i_rdev
));
4123 raw_inode
->i_block
[1] = 0;
4125 raw_inode
->i_block
[0] = 0;
4126 raw_inode
->i_block
[1] =
4127 cpu_to_le32(new_encode_dev(inode
->i_rdev
));
4128 raw_inode
->i_block
[2] = 0;
4130 } else for (block
= 0; block
< EXT4_N_BLOCKS
; block
++)
4131 raw_inode
->i_block
[block
] = ei
->i_data
[block
];
4133 raw_inode
->i_disk_version
= cpu_to_le32(inode
->i_version
);
4134 if (ei
->i_extra_isize
) {
4135 if (EXT4_FITS_IN_INODE(raw_inode
, ei
, i_version_hi
))
4136 raw_inode
->i_version_hi
=
4137 cpu_to_le32(inode
->i_version
>> 32);
4138 raw_inode
->i_extra_isize
= cpu_to_le16(ei
->i_extra_isize
);
4142 BUFFER_TRACE(bh
, "call ext4_journal_dirty_metadata");
4143 rc
= ext4_journal_dirty_metadata(handle
, bh
);
4146 ei
->i_state
&= ~EXT4_STATE_NEW
;
4150 ext4_std_error(inode
->i_sb
, err
);
4155 * ext4_write_inode()
4157 * We are called from a few places:
4159 * - Within generic_file_write() for O_SYNC files.
4160 * Here, there will be no transaction running. We wait for any running
4161 * trasnaction to commit.
4163 * - Within sys_sync(), kupdate and such.
4164 * We wait on commit, if tol to.
4166 * - Within prune_icache() (PF_MEMALLOC == true)
4167 * Here we simply return. We can't afford to block kswapd on the
4170 * In all cases it is actually safe for us to return without doing anything,
4171 * because the inode has been copied into a raw inode buffer in
4172 * ext4_mark_inode_dirty(). This is a correctness thing for O_SYNC and for
4175 * Note that we are absolutely dependent upon all inode dirtiers doing the
4176 * right thing: they *must* call mark_inode_dirty() after dirtying info in
4177 * which we are interested.
4179 * It would be a bug for them to not do this. The code:
4181 * mark_inode_dirty(inode)
4183 * inode->i_size = expr;
4185 * is in error because a kswapd-driven write_inode() could occur while
4186 * `stuff()' is running, and the new i_size will be lost. Plus the inode
4187 * will no longer be on the superblock's dirty inode list.
4189 int ext4_write_inode(struct inode
*inode
, int wait
)
4191 if (current
->flags
& PF_MEMALLOC
)
4194 if (ext4_journal_current_handle()) {
4195 jbd_debug(1, "called recursively, non-PF_MEMALLOC!\n");
4203 return ext4_force_commit(inode
->i_sb
);
4209 * Called from notify_change.
4211 * We want to trap VFS attempts to truncate the file as soon as
4212 * possible. In particular, we want to make sure that when the VFS
4213 * shrinks i_size, we put the inode on the orphan list and modify
4214 * i_disksize immediately, so that during the subsequent flushing of
4215 * dirty pages and freeing of disk blocks, we can guarantee that any
4216 * commit will leave the blocks being flushed in an unused state on
4217 * disk. (On recovery, the inode will get truncated and the blocks will
4218 * be freed, so we have a strong guarantee that no future commit will
4219 * leave these blocks visible to the user.)
4221 * Another thing we have to assure is that if we are in ordered mode
4222 * and inode is still attached to the committing transaction, we must
4223 * we start writeout of all the dirty pages which are being truncated.
4224 * This way we are sure that all the data written in the previous
4225 * transaction are already on disk (truncate waits for pages under
4228 * Called with inode->i_mutex down.
4230 int ext4_setattr(struct dentry
*dentry
, struct iattr
*attr
)
4232 struct inode
*inode
= dentry
->d_inode
;
4234 const unsigned int ia_valid
= attr
->ia_valid
;
4236 error
= inode_change_ok(inode
, attr
);
4240 if ((ia_valid
& ATTR_UID
&& attr
->ia_uid
!= inode
->i_uid
) ||
4241 (ia_valid
& ATTR_GID
&& attr
->ia_gid
!= inode
->i_gid
)) {
4244 /* (user+group)*(old+new) structure, inode write (sb,
4245 * inode block, ? - but truncate inode update has it) */
4246 handle
= ext4_journal_start(inode
, 2*(EXT4_QUOTA_INIT_BLOCKS(inode
->i_sb
)+
4247 EXT4_QUOTA_DEL_BLOCKS(inode
->i_sb
))+3);
4248 if (IS_ERR(handle
)) {
4249 error
= PTR_ERR(handle
);
4252 error
= DQUOT_TRANSFER(inode
, attr
) ? -EDQUOT
: 0;
4254 ext4_journal_stop(handle
);
4257 /* Update corresponding info in inode so that everything is in
4258 * one transaction */
4259 if (attr
->ia_valid
& ATTR_UID
)
4260 inode
->i_uid
= attr
->ia_uid
;
4261 if (attr
->ia_valid
& ATTR_GID
)
4262 inode
->i_gid
= attr
->ia_gid
;
4263 error
= ext4_mark_inode_dirty(handle
, inode
);
4264 ext4_journal_stop(handle
);
4267 if (attr
->ia_valid
& ATTR_SIZE
) {
4268 if (!(EXT4_I(inode
)->i_flags
& EXT4_EXTENTS_FL
)) {
4269 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
4271 if (attr
->ia_size
> sbi
->s_bitmap_maxbytes
) {
4278 if (S_ISREG(inode
->i_mode
) &&
4279 attr
->ia_valid
& ATTR_SIZE
&& attr
->ia_size
< inode
->i_size
) {
4282 handle
= ext4_journal_start(inode
, 3);
4283 if (IS_ERR(handle
)) {
4284 error
= PTR_ERR(handle
);
4288 error
= ext4_orphan_add(handle
, inode
);
4289 EXT4_I(inode
)->i_disksize
= attr
->ia_size
;
4290 rc
= ext4_mark_inode_dirty(handle
, inode
);
4293 ext4_journal_stop(handle
);
4295 if (ext4_should_order_data(inode
)) {
4296 error
= ext4_begin_ordered_truncate(inode
,
4299 /* Do as much error cleanup as possible */
4300 handle
= ext4_journal_start(inode
, 3);
4301 if (IS_ERR(handle
)) {
4302 ext4_orphan_del(NULL
, inode
);
4305 ext4_orphan_del(handle
, inode
);
4306 ext4_journal_stop(handle
);
4312 rc
= inode_setattr(inode
, attr
);
4314 /* If inode_setattr's call to ext4_truncate failed to get a
4315 * transaction handle at all, we need to clean up the in-core
4316 * orphan list manually. */
4318 ext4_orphan_del(NULL
, inode
);
4320 if (!rc
&& (ia_valid
& ATTR_MODE
))
4321 rc
= ext4_acl_chmod(inode
);
4324 ext4_std_error(inode
->i_sb
, error
);
4330 int ext4_getattr(struct vfsmount
*mnt
, struct dentry
*dentry
,
4333 struct inode
*inode
;
4334 unsigned long delalloc_blocks
;
4336 inode
= dentry
->d_inode
;
4337 generic_fillattr(inode
, stat
);
4340 * We can't update i_blocks if the block allocation is delayed
4341 * otherwise in the case of system crash before the real block
4342 * allocation is done, we will have i_blocks inconsistent with
4343 * on-disk file blocks.
4344 * We always keep i_blocks updated together with real
4345 * allocation. But to not confuse with user, stat
4346 * will return the blocks that include the delayed allocation
4347 * blocks for this file.
4349 spin_lock(&EXT4_I(inode
)->i_block_reservation_lock
);
4350 delalloc_blocks
= EXT4_I(inode
)->i_reserved_data_blocks
;
4351 spin_unlock(&EXT4_I(inode
)->i_block_reservation_lock
);
4353 stat
->blocks
+= (delalloc_blocks
<< inode
->i_sb
->s_blocksize_bits
)>>9;
4358 * How many blocks doth make a writepage()?
4360 * With N blocks per page, it may be:
4365 * N+5 bitmap blocks (from the above)
4366 * N+5 group descriptor summary blocks
4369 * 2 * EXT4_SINGLEDATA_TRANS_BLOCKS for the quote files
4371 * 3 * (N + 5) + 2 + 2 * EXT4_SINGLEDATA_TRANS_BLOCKS
4373 * With ordered or writeback data it's the same, less the N data blocks.
4375 * If the inode's direct blocks can hold an integral number of pages then a
4376 * page cannot straddle two indirect blocks, and we can only touch one indirect
4377 * and dindirect block, and the "5" above becomes "3".
4379 * This still overestimates under most circumstances. If we were to pass the
4380 * start and end offsets in here as well we could do block_to_path() on each
4381 * block and work out the exact number of indirects which are touched. Pah.
4384 int ext4_writepage_trans_blocks(struct inode
*inode
)
4386 int bpp
= ext4_journal_blocks_per_page(inode
);
4387 int indirects
= (EXT4_NDIR_BLOCKS
% bpp
) ? 5 : 3;
4390 if (EXT4_I(inode
)->i_flags
& EXT4_EXTENTS_FL
)
4391 return ext4_ext_writepage_trans_blocks(inode
, bpp
);
4393 if (ext4_should_journal_data(inode
))
4394 ret
= 3 * (bpp
+ indirects
) + 2;
4396 ret
= 2 * (bpp
+ indirects
) + 2;
4399 /* We know that structure was already allocated during DQUOT_INIT so
4400 * we will be updating only the data blocks + inodes */
4401 ret
+= 2*EXT4_QUOTA_TRANS_BLOCKS(inode
->i_sb
);
4408 * The caller must have previously called ext4_reserve_inode_write().
4409 * Give this, we know that the caller already has write access to iloc->bh.
4411 int ext4_mark_iloc_dirty(handle_t
*handle
,
4412 struct inode
*inode
, struct ext4_iloc
*iloc
)
4416 if (test_opt(inode
->i_sb
, I_VERSION
))
4417 inode_inc_iversion(inode
);
4419 /* the do_update_inode consumes one bh->b_count */
4422 /* ext4_do_update_inode() does jbd2_journal_dirty_metadata */
4423 err
= ext4_do_update_inode(handle
, inode
, iloc
);
4429 * On success, We end up with an outstanding reference count against
4430 * iloc->bh. This _must_ be cleaned up later.
4434 ext4_reserve_inode_write(handle_t
*handle
, struct inode
*inode
,
4435 struct ext4_iloc
*iloc
)
4439 err
= ext4_get_inode_loc(inode
, iloc
);
4441 BUFFER_TRACE(iloc
->bh
, "get_write_access");
4442 err
= ext4_journal_get_write_access(handle
, iloc
->bh
);
4449 ext4_std_error(inode
->i_sb
, err
);
4454 * Expand an inode by new_extra_isize bytes.
4455 * Returns 0 on success or negative error number on failure.
4457 static int ext4_expand_extra_isize(struct inode
*inode
,
4458 unsigned int new_extra_isize
,
4459 struct ext4_iloc iloc
,
4462 struct ext4_inode
*raw_inode
;
4463 struct ext4_xattr_ibody_header
*header
;
4464 struct ext4_xattr_entry
*entry
;
4466 if (EXT4_I(inode
)->i_extra_isize
>= new_extra_isize
)
4469 raw_inode
= ext4_raw_inode(&iloc
);
4471 header
= IHDR(inode
, raw_inode
);
4472 entry
= IFIRST(header
);
4474 /* No extended attributes present */
4475 if (!(EXT4_I(inode
)->i_state
& EXT4_STATE_XATTR
) ||
4476 header
->h_magic
!= cpu_to_le32(EXT4_XATTR_MAGIC
)) {
4477 memset((void *)raw_inode
+ EXT4_GOOD_OLD_INODE_SIZE
, 0,
4479 EXT4_I(inode
)->i_extra_isize
= new_extra_isize
;
4483 /* try to expand with EAs present */
4484 return ext4_expand_extra_isize_ea(inode
, new_extra_isize
,
4489 * What we do here is to mark the in-core inode as clean with respect to inode
4490 * dirtiness (it may still be data-dirty).
4491 * This means that the in-core inode may be reaped by prune_icache
4492 * without having to perform any I/O. This is a very good thing,
4493 * because *any* task may call prune_icache - even ones which
4494 * have a transaction open against a different journal.
4496 * Is this cheating? Not really. Sure, we haven't written the
4497 * inode out, but prune_icache isn't a user-visible syncing function.
4498 * Whenever the user wants stuff synced (sys_sync, sys_msync, sys_fsync)
4499 * we start and wait on commits.
4501 * Is this efficient/effective? Well, we're being nice to the system
4502 * by cleaning up our inodes proactively so they can be reaped
4503 * without I/O. But we are potentially leaving up to five seconds'
4504 * worth of inodes floating about which prune_icache wants us to
4505 * write out. One way to fix that would be to get prune_icache()
4506 * to do a write_super() to free up some memory. It has the desired
4509 int ext4_mark_inode_dirty(handle_t
*handle
, struct inode
*inode
)
4511 struct ext4_iloc iloc
;
4512 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
4513 static unsigned int mnt_count
;
4517 err
= ext4_reserve_inode_write(handle
, inode
, &iloc
);
4518 if (EXT4_I(inode
)->i_extra_isize
< sbi
->s_want_extra_isize
&&
4519 !(EXT4_I(inode
)->i_state
& EXT4_STATE_NO_EXPAND
)) {
4521 * We need extra buffer credits since we may write into EA block
4522 * with this same handle. If journal_extend fails, then it will
4523 * only result in a minor loss of functionality for that inode.
4524 * If this is felt to be critical, then e2fsck should be run to
4525 * force a large enough s_min_extra_isize.
4527 if ((jbd2_journal_extend(handle
,
4528 EXT4_DATA_TRANS_BLOCKS(inode
->i_sb
))) == 0) {
4529 ret
= ext4_expand_extra_isize(inode
,
4530 sbi
->s_want_extra_isize
,
4533 EXT4_I(inode
)->i_state
|= EXT4_STATE_NO_EXPAND
;
4535 le16_to_cpu(sbi
->s_es
->s_mnt_count
)) {
4536 ext4_warning(inode
->i_sb
, __func__
,
4537 "Unable to expand inode %lu. Delete"
4538 " some EAs or run e2fsck.",
4541 le16_to_cpu(sbi
->s_es
->s_mnt_count
);
4547 err
= ext4_mark_iloc_dirty(handle
, inode
, &iloc
);
4552 * ext4_dirty_inode() is called from __mark_inode_dirty()
4554 * We're really interested in the case where a file is being extended.
4555 * i_size has been changed by generic_commit_write() and we thus need
4556 * to include the updated inode in the current transaction.
4558 * Also, DQUOT_ALLOC_SPACE() will always dirty the inode when blocks
4559 * are allocated to the file.
4561 * If the inode is marked synchronous, we don't honour that here - doing
4562 * so would cause a commit on atime updates, which we don't bother doing.
4563 * We handle synchronous inodes at the highest possible level.
4565 void ext4_dirty_inode(struct inode
*inode
)
4567 handle_t
*current_handle
= ext4_journal_current_handle();
4570 handle
= ext4_journal_start(inode
, 2);
4573 if (current_handle
&&
4574 current_handle
->h_transaction
!= handle
->h_transaction
) {
4575 /* This task has a transaction open against a different fs */
4576 printk(KERN_EMERG
"%s: transactions do not match!\n",
4579 jbd_debug(5, "marking dirty. outer handle=%p\n",
4581 ext4_mark_inode_dirty(handle
, inode
);
4583 ext4_journal_stop(handle
);
4590 * Bind an inode's backing buffer_head into this transaction, to prevent
4591 * it from being flushed to disk early. Unlike
4592 * ext4_reserve_inode_write, this leaves behind no bh reference and
4593 * returns no iloc structure, so the caller needs to repeat the iloc
4594 * lookup to mark the inode dirty later.
4596 static int ext4_pin_inode(handle_t
*handle
, struct inode
*inode
)
4598 struct ext4_iloc iloc
;
4602 err
= ext4_get_inode_loc(inode
, &iloc
);
4604 BUFFER_TRACE(iloc
.bh
, "get_write_access");
4605 err
= jbd2_journal_get_write_access(handle
, iloc
.bh
);
4607 err
= ext4_journal_dirty_metadata(handle
,
4612 ext4_std_error(inode
->i_sb
, err
);
4617 int ext4_change_inode_journal_flag(struct inode
*inode
, int val
)
4624 * We have to be very careful here: changing a data block's
4625 * journaling status dynamically is dangerous. If we write a
4626 * data block to the journal, change the status and then delete
4627 * that block, we risk forgetting to revoke the old log record
4628 * from the journal and so a subsequent replay can corrupt data.
4629 * So, first we make sure that the journal is empty and that
4630 * nobody is changing anything.
4633 journal
= EXT4_JOURNAL(inode
);
4634 if (is_journal_aborted(journal
))
4637 jbd2_journal_lock_updates(journal
);
4638 jbd2_journal_flush(journal
);
4641 * OK, there are no updates running now, and all cached data is
4642 * synced to disk. We are now in a completely consistent state
4643 * which doesn't have anything in the journal, and we know that
4644 * no filesystem updates are running, so it is safe to modify
4645 * the inode's in-core data-journaling state flag now.
4649 EXT4_I(inode
)->i_flags
|= EXT4_JOURNAL_DATA_FL
;
4651 EXT4_I(inode
)->i_flags
&= ~EXT4_JOURNAL_DATA_FL
;
4652 ext4_set_aops(inode
);
4654 jbd2_journal_unlock_updates(journal
);
4656 /* Finally we can mark the inode as dirty. */
4658 handle
= ext4_journal_start(inode
, 1);
4660 return PTR_ERR(handle
);
4662 err
= ext4_mark_inode_dirty(handle
, inode
);
4664 ext4_journal_stop(handle
);
4665 ext4_std_error(inode
->i_sb
, err
);
4670 static int ext4_bh_unmapped(handle_t
*handle
, struct buffer_head
*bh
)
4672 return !buffer_mapped(bh
);
4675 int ext4_page_mkwrite(struct vm_area_struct
*vma
, struct page
*page
)
4680 struct file
*file
= vma
->vm_file
;
4681 struct inode
*inode
= file
->f_path
.dentry
->d_inode
;
4682 struct address_space
*mapping
= inode
->i_mapping
;
4685 * Get i_alloc_sem to stop truncates messing with the inode. We cannot
4686 * get i_mutex because we are already holding mmap_sem.
4688 down_read(&inode
->i_alloc_sem
);
4689 size
= i_size_read(inode
);
4690 if (page
->mapping
!= mapping
|| size
<= page_offset(page
)
4691 || !PageUptodate(page
)) {
4692 /* page got truncated from under us? */
4696 if (PageMappedToDisk(page
))
4699 if (page
->index
== size
>> PAGE_CACHE_SHIFT
)
4700 len
= size
& ~PAGE_CACHE_MASK
;
4702 len
= PAGE_CACHE_SIZE
;
4704 if (page_has_buffers(page
)) {
4705 /* return if we have all the buffers mapped */
4706 if (!walk_page_buffers(NULL
, page_buffers(page
), 0, len
, NULL
,
4711 * OK, we need to fill the hole... Do write_begin write_end
4712 * to do block allocation/reservation.We are not holding
4713 * inode.i__mutex here. That allow * parallel write_begin,
4714 * write_end call. lock_page prevent this from happening
4715 * on the same page though
4717 ret
= mapping
->a_ops
->write_begin(file
, mapping
, page_offset(page
),
4718 len
, AOP_FLAG_UNINTERRUPTIBLE
, &page
, NULL
);
4721 ret
= mapping
->a_ops
->write_end(file
, mapping
, page_offset(page
),
4722 len
, len
, page
, NULL
);
4727 up_read(&inode
->i_alloc_sem
);