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/namei.h>
38 #include <linux/uio.h>
39 #include <linux/bio.h>
40 #include "ext4_jbd2.h"
43 #include "ext4_extents.h"
45 #define MPAGE_DA_EXTENT_TAIL 0x01
47 static inline int ext4_begin_ordered_truncate(struct inode
*inode
,
50 return jbd2_journal_begin_ordered_truncate(
51 EXT4_SB(inode
->i_sb
)->s_journal
,
52 &EXT4_I(inode
)->jinode
,
56 static void ext4_invalidatepage(struct page
*page
, unsigned long offset
);
59 * Test whether an inode is a fast symlink.
61 static int ext4_inode_is_fast_symlink(struct inode
*inode
)
63 int ea_blocks
= EXT4_I(inode
)->i_file_acl
?
64 (inode
->i_sb
->s_blocksize
>> 9) : 0;
66 return (S_ISLNK(inode
->i_mode
) && inode
->i_blocks
- ea_blocks
== 0);
70 * The ext4 forget function must perform a revoke if we are freeing data
71 * which has been journaled. Metadata (eg. indirect blocks) must be
72 * revoked in all cases.
74 * "bh" may be NULL: a metadata block may have been freed from memory
75 * but there may still be a record of it in the journal, and that record
76 * still needs to be revoked.
78 * If the handle isn't valid we're not journaling so there's nothing to do.
80 int ext4_forget(handle_t
*handle
, int is_metadata
, struct inode
*inode
,
81 struct buffer_head
*bh
, ext4_fsblk_t blocknr
)
85 if (!ext4_handle_valid(handle
))
90 BUFFER_TRACE(bh
, "enter");
92 jbd_debug(4, "forgetting bh %p: is_metadata = %d, mode %o, "
94 bh
, is_metadata
, inode
->i_mode
,
95 test_opt(inode
->i_sb
, DATA_FLAGS
));
97 /* Never use the revoke function if we are doing full data
98 * journaling: there is no need to, and a V1 superblock won't
99 * support it. Otherwise, only skip the revoke on un-journaled
102 if (test_opt(inode
->i_sb
, DATA_FLAGS
) == EXT4_MOUNT_JOURNAL_DATA
||
103 (!is_metadata
&& !ext4_should_journal_data(inode
))) {
105 BUFFER_TRACE(bh
, "call jbd2_journal_forget");
106 return ext4_journal_forget(handle
, bh
);
112 * data!=journal && (is_metadata || should_journal_data(inode))
114 BUFFER_TRACE(bh
, "call ext4_journal_revoke");
115 err
= ext4_journal_revoke(handle
, blocknr
, bh
);
117 ext4_abort(inode
->i_sb
, __func__
,
118 "error %d when attempting revoke", err
);
119 BUFFER_TRACE(bh
, "exit");
124 * Work out how many blocks we need to proceed with the next chunk of a
125 * truncate transaction.
127 static unsigned long blocks_for_truncate(struct inode
*inode
)
131 needed
= inode
->i_blocks
>> (inode
->i_sb
->s_blocksize_bits
- 9);
133 /* Give ourselves just enough room to cope with inodes in which
134 * i_blocks is corrupt: we've seen disk corruptions in the past
135 * which resulted in random data in an inode which looked enough
136 * like a regular file for ext4 to try to delete it. Things
137 * will go a bit crazy if that happens, but at least we should
138 * try not to panic the whole kernel. */
142 /* But we need to bound the transaction so we don't overflow the
144 if (needed
> EXT4_MAX_TRANS_DATA
)
145 needed
= EXT4_MAX_TRANS_DATA
;
147 return EXT4_DATA_TRANS_BLOCKS(inode
->i_sb
) + needed
;
151 * Truncate transactions can be complex and absolutely huge. So we need to
152 * be able to restart the transaction at a conventient checkpoint to make
153 * sure we don't overflow the journal.
155 * start_transaction gets us a new handle for a truncate transaction,
156 * and extend_transaction tries to extend the existing one a bit. If
157 * extend fails, we need to propagate the failure up and restart the
158 * transaction in the top-level truncate loop. --sct
160 static handle_t
*start_transaction(struct inode
*inode
)
164 result
= ext4_journal_start(inode
, blocks_for_truncate(inode
));
168 ext4_std_error(inode
->i_sb
, PTR_ERR(result
));
173 * Try to extend this transaction for the purposes of truncation.
175 * Returns 0 if we managed to create more room. If we can't create more
176 * room, and the transaction must be restarted we return 1.
178 static int try_to_extend_transaction(handle_t
*handle
, struct inode
*inode
)
180 if (!ext4_handle_valid(handle
))
182 if (ext4_handle_has_enough_credits(handle
, EXT4_RESERVE_TRANS_BLOCKS
+1))
184 if (!ext4_journal_extend(handle
, blocks_for_truncate(inode
)))
190 * Restart the transaction associated with *handle. This does a commit,
191 * so before we call here everything must be consistently dirtied against
194 static int ext4_journal_test_restart(handle_t
*handle
, struct inode
*inode
)
196 BUG_ON(EXT4_JOURNAL(inode
) == NULL
);
197 jbd_debug(2, "restarting handle %p\n", handle
);
198 return ext4_journal_restart(handle
, blocks_for_truncate(inode
));
202 * Called at the last iput() if i_nlink is zero.
204 void ext4_delete_inode(struct inode
*inode
)
209 if (ext4_should_order_data(inode
))
210 ext4_begin_ordered_truncate(inode
, 0);
211 truncate_inode_pages(&inode
->i_data
, 0);
213 if (is_bad_inode(inode
))
216 handle
= ext4_journal_start(inode
, blocks_for_truncate(inode
)+3);
217 if (IS_ERR(handle
)) {
218 ext4_std_error(inode
->i_sb
, PTR_ERR(handle
));
220 * If we're going to skip the normal cleanup, we still need to
221 * make sure that the in-core orphan linked list is properly
224 ext4_orphan_del(NULL
, inode
);
229 ext4_handle_sync(handle
);
231 err
= ext4_mark_inode_dirty(handle
, inode
);
233 ext4_warning(inode
->i_sb
, __func__
,
234 "couldn't mark inode dirty (err %d)", err
);
238 ext4_truncate(inode
);
241 * ext4_ext_truncate() doesn't reserve any slop when it
242 * restarts journal transactions; therefore there may not be
243 * enough credits left in the handle to remove the inode from
244 * the orphan list and set the dtime field.
246 if (!ext4_handle_has_enough_credits(handle
, 3)) {
247 err
= ext4_journal_extend(handle
, 3);
249 err
= ext4_journal_restart(handle
, 3);
251 ext4_warning(inode
->i_sb
, __func__
,
252 "couldn't extend journal (err %d)", err
);
254 ext4_journal_stop(handle
);
260 * Kill off the orphan record which ext4_truncate created.
261 * AKPM: I think this can be inside the above `if'.
262 * Note that ext4_orphan_del() has to be able to cope with the
263 * deletion of a non-existent orphan - this is because we don't
264 * know if ext4_truncate() actually created an orphan record.
265 * (Well, we could do this if we need to, but heck - it works)
267 ext4_orphan_del(handle
, inode
);
268 EXT4_I(inode
)->i_dtime
= get_seconds();
271 * One subtle ordering requirement: if anything has gone wrong
272 * (transaction abort, IO errors, whatever), then we can still
273 * do these next steps (the fs will already have been marked as
274 * having errors), but we can't free the inode if the mark_dirty
277 if (ext4_mark_inode_dirty(handle
, inode
))
278 /* If that failed, just do the required in-core inode clear. */
281 ext4_free_inode(handle
, inode
);
282 ext4_journal_stop(handle
);
285 clear_inode(inode
); /* We must guarantee clearing of inode... */
291 struct buffer_head
*bh
;
294 static inline void add_chain(Indirect
*p
, struct buffer_head
*bh
, __le32
*v
)
296 p
->key
= *(p
->p
= v
);
301 * ext4_block_to_path - parse the block number into array of offsets
302 * @inode: inode in question (we are only interested in its superblock)
303 * @i_block: block number to be parsed
304 * @offsets: array to store the offsets in
305 * @boundary: set this non-zero if the referred-to block is likely to be
306 * followed (on disk) by an indirect block.
308 * To store the locations of file's data ext4 uses a data structure common
309 * for UNIX filesystems - tree of pointers anchored in the inode, with
310 * data blocks at leaves and indirect blocks in intermediate nodes.
311 * This function translates the block number into path in that tree -
312 * return value is the path length and @offsets[n] is the offset of
313 * pointer to (n+1)th node in the nth one. If @block is out of range
314 * (negative or too large) warning is printed and zero returned.
316 * Note: function doesn't find node addresses, so no IO is needed. All
317 * we need to know is the capacity of indirect blocks (taken from the
322 * Portability note: the last comparison (check that we fit into triple
323 * indirect block) is spelled differently, because otherwise on an
324 * architecture with 32-bit longs and 8Kb pages we might get into trouble
325 * if our filesystem had 8Kb blocks. We might use long long, but that would
326 * kill us on x86. Oh, well, at least the sign propagation does not matter -
327 * i_block would have to be negative in the very beginning, so we would not
331 static int ext4_block_to_path(struct inode
*inode
,
333 ext4_lblk_t offsets
[4], int *boundary
)
335 int ptrs
= EXT4_ADDR_PER_BLOCK(inode
->i_sb
);
336 int ptrs_bits
= EXT4_ADDR_PER_BLOCK_BITS(inode
->i_sb
);
337 const long direct_blocks
= EXT4_NDIR_BLOCKS
,
338 indirect_blocks
= ptrs
,
339 double_blocks
= (1 << (ptrs_bits
* 2));
344 ext4_warning(inode
->i_sb
, "ext4_block_to_path", "block < 0");
345 } else if (i_block
< direct_blocks
) {
346 offsets
[n
++] = i_block
;
347 final
= direct_blocks
;
348 } else if ((i_block
-= direct_blocks
) < indirect_blocks
) {
349 offsets
[n
++] = EXT4_IND_BLOCK
;
350 offsets
[n
++] = i_block
;
352 } else if ((i_block
-= indirect_blocks
) < double_blocks
) {
353 offsets
[n
++] = EXT4_DIND_BLOCK
;
354 offsets
[n
++] = i_block
>> ptrs_bits
;
355 offsets
[n
++] = i_block
& (ptrs
- 1);
357 } else if (((i_block
-= double_blocks
) >> (ptrs_bits
* 2)) < ptrs
) {
358 offsets
[n
++] = EXT4_TIND_BLOCK
;
359 offsets
[n
++] = i_block
>> (ptrs_bits
* 2);
360 offsets
[n
++] = (i_block
>> ptrs_bits
) & (ptrs
- 1);
361 offsets
[n
++] = i_block
& (ptrs
- 1);
364 ext4_warning(inode
->i_sb
, "ext4_block_to_path",
365 "block %lu > max in inode %lu",
366 i_block
+ direct_blocks
+
367 indirect_blocks
+ double_blocks
, inode
->i_ino
);
370 *boundary
= final
- 1 - (i_block
& (ptrs
- 1));
374 static int __ext4_check_blockref(const char *function
, struct inode
*inode
,
375 __le32
*p
, unsigned int max
)
380 while (bref
< p
+max
) {
381 blk
= le32_to_cpu(*bref
++);
383 unlikely(!ext4_data_block_valid(EXT4_SB(inode
->i_sb
),
385 ext4_error(inode
->i_sb
, function
,
386 "invalid block reference %u "
387 "in inode #%lu", blk
, inode
->i_ino
);
395 #define ext4_check_indirect_blockref(inode, bh) \
396 __ext4_check_blockref(__func__, inode, (__le32 *)(bh)->b_data, \
397 EXT4_ADDR_PER_BLOCK((inode)->i_sb))
399 #define ext4_check_inode_blockref(inode) \
400 __ext4_check_blockref(__func__, inode, EXT4_I(inode)->i_data, \
404 * ext4_get_branch - read the chain of indirect blocks leading to data
405 * @inode: inode in question
406 * @depth: depth of the chain (1 - direct pointer, etc.)
407 * @offsets: offsets of pointers in inode/indirect blocks
408 * @chain: place to store the result
409 * @err: here we store the error value
411 * Function fills the array of triples <key, p, bh> and returns %NULL
412 * if everything went OK or the pointer to the last filled triple
413 * (incomplete one) otherwise. Upon the return chain[i].key contains
414 * the number of (i+1)-th block in the chain (as it is stored in memory,
415 * i.e. little-endian 32-bit), chain[i].p contains the address of that
416 * number (it points into struct inode for i==0 and into the bh->b_data
417 * for i>0) and chain[i].bh points to the buffer_head of i-th indirect
418 * block for i>0 and NULL for i==0. In other words, it holds the block
419 * numbers of the chain, addresses they were taken from (and where we can
420 * verify that chain did not change) and buffer_heads hosting these
423 * Function stops when it stumbles upon zero pointer (absent block)
424 * (pointer to last triple returned, *@err == 0)
425 * or when it gets an IO error reading an indirect block
426 * (ditto, *@err == -EIO)
427 * or when it reads all @depth-1 indirect blocks successfully and finds
428 * the whole chain, all way to the data (returns %NULL, *err == 0).
430 * Need to be called with
431 * down_read(&EXT4_I(inode)->i_data_sem)
433 static Indirect
*ext4_get_branch(struct inode
*inode
, int depth
,
434 ext4_lblk_t
*offsets
,
435 Indirect chain
[4], int *err
)
437 struct super_block
*sb
= inode
->i_sb
;
439 struct buffer_head
*bh
;
442 /* i_data is not going away, no lock needed */
443 add_chain(chain
, NULL
, EXT4_I(inode
)->i_data
+ *offsets
);
447 bh
= sb_getblk(sb
, le32_to_cpu(p
->key
));
451 if (!bh_uptodate_or_lock(bh
)) {
452 if (bh_submit_read(bh
) < 0) {
456 /* validate block references */
457 if (ext4_check_indirect_blockref(inode
, bh
)) {
463 add_chain(++p
, bh
, (__le32
*)bh
->b_data
+ *++offsets
);
477 * ext4_find_near - find a place for allocation with sufficient locality
479 * @ind: descriptor of indirect block.
481 * This function returns the preferred place for block allocation.
482 * It is used when heuristic for sequential allocation fails.
484 * + if there is a block to the left of our position - allocate near it.
485 * + if pointer will live in indirect block - allocate near that block.
486 * + if pointer will live in inode - allocate in the same
489 * In the latter case we colour the starting block by the callers PID to
490 * prevent it from clashing with concurrent allocations for a different inode
491 * in the same block group. The PID is used here so that functionally related
492 * files will be close-by on-disk.
494 * Caller must make sure that @ind is valid and will stay that way.
496 static ext4_fsblk_t
ext4_find_near(struct inode
*inode
, Indirect
*ind
)
498 struct ext4_inode_info
*ei
= EXT4_I(inode
);
499 __le32
*start
= ind
->bh
? (__le32
*) ind
->bh
->b_data
: ei
->i_data
;
501 ext4_fsblk_t bg_start
;
502 ext4_fsblk_t last_block
;
503 ext4_grpblk_t colour
;
504 ext4_group_t block_group
;
505 int flex_size
= ext4_flex_bg_size(EXT4_SB(inode
->i_sb
));
507 /* Try to find previous block */
508 for (p
= ind
->p
- 1; p
>= start
; p
--) {
510 return le32_to_cpu(*p
);
513 /* No such thing, so let's try location of indirect block */
515 return ind
->bh
->b_blocknr
;
518 * It is going to be referred to from the inode itself? OK, just put it
519 * into the same cylinder group then.
521 block_group
= ei
->i_block_group
;
522 if (flex_size
>= EXT4_FLEX_SIZE_DIR_ALLOC_SCHEME
) {
523 block_group
&= ~(flex_size
-1);
524 if (S_ISREG(inode
->i_mode
))
527 bg_start
= ext4_group_first_block_no(inode
->i_sb
, block_group
);
528 last_block
= ext4_blocks_count(EXT4_SB(inode
->i_sb
)->s_es
) - 1;
531 * If we are doing delayed allocation, we don't need take
532 * colour into account.
534 if (test_opt(inode
->i_sb
, DELALLOC
))
537 if (bg_start
+ EXT4_BLOCKS_PER_GROUP(inode
->i_sb
) <= last_block
)
538 colour
= (current
->pid
% 16) *
539 (EXT4_BLOCKS_PER_GROUP(inode
->i_sb
) / 16);
541 colour
= (current
->pid
% 16) * ((last_block
- bg_start
) / 16);
542 return bg_start
+ colour
;
546 * ext4_find_goal - find a preferred place for allocation.
548 * @block: block we want
549 * @partial: pointer to the last triple within a chain
551 * Normally this function find the preferred place for block allocation,
554 static ext4_fsblk_t
ext4_find_goal(struct inode
*inode
, ext4_lblk_t block
,
558 * XXX need to get goal block from mballoc's data structures
561 return ext4_find_near(inode
, partial
);
565 * ext4_blks_to_allocate: Look up the block map and count the number
566 * of direct blocks need to be allocated for the given branch.
568 * @branch: chain of indirect blocks
569 * @k: number of blocks need for indirect blocks
570 * @blks: number of data blocks to be mapped.
571 * @blocks_to_boundary: the offset in the indirect block
573 * return the total number of blocks to be allocate, including the
574 * direct and indirect blocks.
576 static int ext4_blks_to_allocate(Indirect
*branch
, int k
, unsigned int blks
,
577 int blocks_to_boundary
)
579 unsigned int count
= 0;
582 * Simple case, [t,d]Indirect block(s) has not allocated yet
583 * then it's clear blocks on that path have not allocated
586 /* right now we don't handle cross boundary allocation */
587 if (blks
< blocks_to_boundary
+ 1)
590 count
+= blocks_to_boundary
+ 1;
595 while (count
< blks
&& count
<= blocks_to_boundary
&&
596 le32_to_cpu(*(branch
[0].p
+ count
)) == 0) {
603 * ext4_alloc_blocks: multiple allocate blocks needed for a branch
604 * @indirect_blks: the number of blocks need to allocate for indirect
607 * @new_blocks: on return it will store the new block numbers for
608 * the indirect blocks(if needed) and the first direct block,
609 * @blks: on return it will store the total number of allocated
612 static int ext4_alloc_blocks(handle_t
*handle
, struct inode
*inode
,
613 ext4_lblk_t iblock
, ext4_fsblk_t goal
,
614 int indirect_blks
, int blks
,
615 ext4_fsblk_t new_blocks
[4], int *err
)
617 struct ext4_allocation_request ar
;
619 unsigned long count
= 0, blk_allocated
= 0;
621 ext4_fsblk_t current_block
= 0;
625 * Here we try to allocate the requested multiple blocks at once,
626 * on a best-effort basis.
627 * To build a branch, we should allocate blocks for
628 * the indirect blocks(if not allocated yet), and at least
629 * the first direct block of this branch. That's the
630 * minimum number of blocks need to allocate(required)
632 /* first we try to allocate the indirect blocks */
633 target
= indirect_blks
;
636 /* allocating blocks for indirect blocks and direct blocks */
637 current_block
= ext4_new_meta_blocks(handle
, inode
,
643 /* allocate blocks for indirect blocks */
644 while (index
< indirect_blks
&& count
) {
645 new_blocks
[index
++] = current_block
++;
650 * save the new block number
651 * for the first direct block
653 new_blocks
[index
] = current_block
;
654 printk(KERN_INFO
"%s returned more blocks than "
655 "requested\n", __func__
);
661 target
= blks
- count
;
662 blk_allocated
= count
;
665 /* Now allocate data blocks */
666 memset(&ar
, 0, sizeof(ar
));
671 if (S_ISREG(inode
->i_mode
))
672 /* enable in-core preallocation only for regular files */
673 ar
.flags
= EXT4_MB_HINT_DATA
;
675 current_block
= ext4_mb_new_blocks(handle
, &ar
, err
);
677 if (*err
&& (target
== blks
)) {
679 * if the allocation failed and we didn't allocate
685 if (target
== blks
) {
687 * save the new block number
688 * for the first direct block
690 new_blocks
[index
] = current_block
;
692 blk_allocated
+= ar
.len
;
695 /* total number of blocks allocated for direct blocks */
700 for (i
= 0; i
< index
; i
++)
701 ext4_free_blocks(handle
, inode
, new_blocks
[i
], 1, 0);
706 * ext4_alloc_branch - allocate and set up a chain of blocks.
708 * @indirect_blks: number of allocated indirect blocks
709 * @blks: number of allocated direct blocks
710 * @offsets: offsets (in the blocks) to store the pointers to next.
711 * @branch: place to store the chain in.
713 * This function allocates blocks, zeroes out all but the last one,
714 * links them into chain and (if we are synchronous) writes them to disk.
715 * In other words, it prepares a branch that can be spliced onto the
716 * inode. It stores the information about that chain in the branch[], in
717 * the same format as ext4_get_branch() would do. We are calling it after
718 * we had read the existing part of chain and partial points to the last
719 * triple of that (one with zero ->key). Upon the exit we have the same
720 * picture as after the successful ext4_get_block(), except that in one
721 * place chain is disconnected - *branch->p is still zero (we did not
722 * set the last link), but branch->key contains the number that should
723 * be placed into *branch->p to fill that gap.
725 * If allocation fails we free all blocks we've allocated (and forget
726 * their buffer_heads) and return the error value the from failed
727 * ext4_alloc_block() (normally -ENOSPC). Otherwise we set the chain
728 * as described above and return 0.
730 static int ext4_alloc_branch(handle_t
*handle
, struct inode
*inode
,
731 ext4_lblk_t iblock
, int indirect_blks
,
732 int *blks
, ext4_fsblk_t goal
,
733 ext4_lblk_t
*offsets
, Indirect
*branch
)
735 int blocksize
= inode
->i_sb
->s_blocksize
;
738 struct buffer_head
*bh
;
740 ext4_fsblk_t new_blocks
[4];
741 ext4_fsblk_t current_block
;
743 num
= ext4_alloc_blocks(handle
, inode
, iblock
, goal
, indirect_blks
,
744 *blks
, new_blocks
, &err
);
748 branch
[0].key
= cpu_to_le32(new_blocks
[0]);
750 * metadata blocks and data blocks are allocated.
752 for (n
= 1; n
<= indirect_blks
; n
++) {
754 * Get buffer_head for parent block, zero it out
755 * and set the pointer to new one, then send
758 bh
= sb_getblk(inode
->i_sb
, new_blocks
[n
-1]);
761 BUFFER_TRACE(bh
, "call get_create_access");
762 err
= ext4_journal_get_create_access(handle
, bh
);
769 memset(bh
->b_data
, 0, blocksize
);
770 branch
[n
].p
= (__le32
*) bh
->b_data
+ offsets
[n
];
771 branch
[n
].key
= cpu_to_le32(new_blocks
[n
]);
772 *branch
[n
].p
= branch
[n
].key
;
773 if (n
== indirect_blks
) {
774 current_block
= new_blocks
[n
];
776 * End of chain, update the last new metablock of
777 * the chain to point to the new allocated
778 * data blocks numbers
780 for (i
=1; i
< num
; i
++)
781 *(branch
[n
].p
+ i
) = cpu_to_le32(++current_block
);
783 BUFFER_TRACE(bh
, "marking uptodate");
784 set_buffer_uptodate(bh
);
787 BUFFER_TRACE(bh
, "call ext4_handle_dirty_metadata");
788 err
= ext4_handle_dirty_metadata(handle
, inode
, bh
);
795 /* Allocation failed, free what we already allocated */
796 for (i
= 1; i
<= n
; i
++) {
797 BUFFER_TRACE(branch
[i
].bh
, "call jbd2_journal_forget");
798 ext4_journal_forget(handle
, branch
[i
].bh
);
800 for (i
= 0; i
< indirect_blks
; i
++)
801 ext4_free_blocks(handle
, inode
, new_blocks
[i
], 1, 0);
803 ext4_free_blocks(handle
, inode
, new_blocks
[i
], num
, 0);
809 * ext4_splice_branch - splice the allocated branch onto inode.
811 * @block: (logical) number of block we are adding
812 * @chain: chain of indirect blocks (with a missing link - see
814 * @where: location of missing link
815 * @num: number of indirect blocks we are adding
816 * @blks: number of direct blocks we are adding
818 * This function fills the missing link and does all housekeeping needed in
819 * inode (->i_blocks, etc.). In case of success we end up with the full
820 * chain to new block and return 0.
822 static int ext4_splice_branch(handle_t
*handle
, struct inode
*inode
,
823 ext4_lblk_t block
, Indirect
*where
, int num
, int blks
)
827 ext4_fsblk_t current_block
;
830 * If we're splicing into a [td]indirect block (as opposed to the
831 * inode) then we need to get write access to the [td]indirect block
835 BUFFER_TRACE(where
->bh
, "get_write_access");
836 err
= ext4_journal_get_write_access(handle
, where
->bh
);
842 *where
->p
= where
->key
;
845 * Update the host buffer_head or inode to point to more just allocated
846 * direct blocks blocks
848 if (num
== 0 && blks
> 1) {
849 current_block
= le32_to_cpu(where
->key
) + 1;
850 for (i
= 1; i
< blks
; i
++)
851 *(where
->p
+ i
) = cpu_to_le32(current_block
++);
854 /* We are done with atomic stuff, now do the rest of housekeeping */
856 inode
->i_ctime
= ext4_current_time(inode
);
857 ext4_mark_inode_dirty(handle
, inode
);
859 /* had we spliced it onto indirect block? */
862 * If we spliced it onto an indirect block, we haven't
863 * altered the inode. Note however that if it is being spliced
864 * onto an indirect block at the very end of the file (the
865 * file is growing) then we *will* alter the inode to reflect
866 * the new i_size. But that is not done here - it is done in
867 * generic_commit_write->__mark_inode_dirty->ext4_dirty_inode.
869 jbd_debug(5, "splicing indirect only\n");
870 BUFFER_TRACE(where
->bh
, "call ext4_handle_dirty_metadata");
871 err
= ext4_handle_dirty_metadata(handle
, inode
, where
->bh
);
876 * OK, we spliced it into the inode itself on a direct block.
877 * Inode was dirtied above.
879 jbd_debug(5, "splicing direct\n");
884 for (i
= 1; i
<= num
; i
++) {
885 BUFFER_TRACE(where
[i
].bh
, "call jbd2_journal_forget");
886 ext4_journal_forget(handle
, where
[i
].bh
);
887 ext4_free_blocks(handle
, inode
,
888 le32_to_cpu(where
[i
-1].key
), 1, 0);
890 ext4_free_blocks(handle
, inode
, le32_to_cpu(where
[num
].key
), blks
, 0);
896 * The ext4_ind_get_blocks() function handles non-extents inodes
897 * (i.e., using the traditional indirect/double-indirect i_blocks
898 * scheme) for ext4_get_blocks().
900 * Allocation strategy is simple: if we have to allocate something, we will
901 * have to go the whole way to leaf. So let's do it before attaching anything
902 * to tree, set linkage between the newborn blocks, write them if sync is
903 * required, recheck the path, free and repeat if check fails, otherwise
904 * set the last missing link (that will protect us from any truncate-generated
905 * removals - all blocks on the path are immune now) and possibly force the
906 * write on the parent block.
907 * That has a nice additional property: no special recovery from the failed
908 * allocations is needed - we simply release blocks and do not touch anything
909 * reachable from inode.
911 * `handle' can be NULL if create == 0.
913 * return > 0, # of blocks mapped or allocated.
914 * return = 0, if plain lookup failed.
915 * return < 0, error case.
917 * The ext4_ind_get_blocks() function should be called with
918 * down_write(&EXT4_I(inode)->i_data_sem) if allocating filesystem
919 * blocks (i.e., flags has EXT4_GET_BLOCKS_CREATE set) or
920 * down_read(&EXT4_I(inode)->i_data_sem) if not allocating file system
923 static int ext4_ind_get_blocks(handle_t
*handle
, struct inode
*inode
,
924 ext4_lblk_t iblock
, unsigned int maxblocks
,
925 struct buffer_head
*bh_result
,
929 ext4_lblk_t offsets
[4];
934 int blocks_to_boundary
= 0;
937 ext4_fsblk_t first_block
= 0;
939 J_ASSERT(!(EXT4_I(inode
)->i_flags
& EXT4_EXTENTS_FL
));
940 J_ASSERT(handle
!= NULL
|| (flags
& EXT4_GET_BLOCKS_CREATE
) == 0);
941 depth
= ext4_block_to_path(inode
, iblock
, offsets
,
942 &blocks_to_boundary
);
947 partial
= ext4_get_branch(inode
, depth
, offsets
, chain
, &err
);
949 /* Simplest case - block found, no allocation needed */
951 first_block
= le32_to_cpu(chain
[depth
- 1].key
);
952 clear_buffer_new(bh_result
);
955 while (count
< maxblocks
&& count
<= blocks_to_boundary
) {
958 blk
= le32_to_cpu(*(chain
[depth
-1].p
+ count
));
960 if (blk
== first_block
+ count
)
968 /* Next simple case - plain lookup or failed read of indirect block */
969 if ((flags
& EXT4_GET_BLOCKS_CREATE
) == 0 || err
== -EIO
)
973 * Okay, we need to do block allocation.
975 goal
= ext4_find_goal(inode
, iblock
, partial
);
977 /* the number of blocks need to allocate for [d,t]indirect blocks */
978 indirect_blks
= (chain
+ depth
) - partial
- 1;
981 * Next look up the indirect map to count the totoal number of
982 * direct blocks to allocate for this branch.
984 count
= ext4_blks_to_allocate(partial
, indirect_blks
,
985 maxblocks
, blocks_to_boundary
);
987 * Block out ext4_truncate while we alter the tree
989 err
= ext4_alloc_branch(handle
, inode
, iblock
, indirect_blks
,
991 offsets
+ (partial
- chain
), partial
);
994 * The ext4_splice_branch call will free and forget any buffers
995 * on the new chain if there is a failure, but that risks using
996 * up transaction credits, especially for bitmaps where the
997 * credits cannot be returned. Can we handle this somehow? We
998 * may need to return -EAGAIN upwards in the worst case. --sct
1001 err
= ext4_splice_branch(handle
, inode
, iblock
,
1002 partial
, indirect_blks
, count
);
1006 set_buffer_new(bh_result
);
1008 map_bh(bh_result
, inode
->i_sb
, le32_to_cpu(chain
[depth
-1].key
));
1009 if (count
> blocks_to_boundary
)
1010 set_buffer_boundary(bh_result
);
1012 /* Clean up and exit */
1013 partial
= chain
+ depth
- 1; /* the whole chain */
1015 while (partial
> chain
) {
1016 BUFFER_TRACE(partial
->bh
, "call brelse");
1017 brelse(partial
->bh
);
1020 BUFFER_TRACE(bh_result
, "returned");
1025 qsize_t
ext4_get_reserved_space(struct inode
*inode
)
1027 unsigned long long total
;
1029 spin_lock(&EXT4_I(inode
)->i_block_reservation_lock
);
1030 total
= EXT4_I(inode
)->i_reserved_data_blocks
+
1031 EXT4_I(inode
)->i_reserved_meta_blocks
;
1032 spin_unlock(&EXT4_I(inode
)->i_block_reservation_lock
);
1037 * Calculate the number of metadata blocks need to reserve
1038 * to allocate @blocks for non extent file based file
1040 static int ext4_indirect_calc_metadata_amount(struct inode
*inode
, int blocks
)
1042 int icap
= EXT4_ADDR_PER_BLOCK(inode
->i_sb
);
1043 int ind_blks
, dind_blks
, tind_blks
;
1045 /* number of new indirect blocks needed */
1046 ind_blks
= (blocks
+ icap
- 1) / icap
;
1048 dind_blks
= (ind_blks
+ icap
- 1) / icap
;
1052 return ind_blks
+ dind_blks
+ tind_blks
;
1056 * Calculate the number of metadata blocks need to reserve
1057 * to allocate given number of blocks
1059 static int ext4_calc_metadata_amount(struct inode
*inode
, int blocks
)
1064 if (EXT4_I(inode
)->i_flags
& EXT4_EXTENTS_FL
)
1065 return ext4_ext_calc_metadata_amount(inode
, blocks
);
1067 return ext4_indirect_calc_metadata_amount(inode
, blocks
);
1070 static void ext4_da_update_reserve_space(struct inode
*inode
, int used
)
1072 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
1073 int total
, mdb
, mdb_free
;
1075 spin_lock(&EXT4_I(inode
)->i_block_reservation_lock
);
1076 /* recalculate the number of metablocks still need to be reserved */
1077 total
= EXT4_I(inode
)->i_reserved_data_blocks
- used
;
1078 mdb
= ext4_calc_metadata_amount(inode
, total
);
1080 /* figure out how many metablocks to release */
1081 BUG_ON(mdb
> EXT4_I(inode
)->i_reserved_meta_blocks
);
1082 mdb_free
= EXT4_I(inode
)->i_reserved_meta_blocks
- mdb
;
1085 /* Account for allocated meta_blocks */
1086 mdb_free
-= EXT4_I(inode
)->i_allocated_meta_blocks
;
1088 /* update fs dirty blocks counter */
1089 percpu_counter_sub(&sbi
->s_dirtyblocks_counter
, mdb_free
);
1090 EXT4_I(inode
)->i_allocated_meta_blocks
= 0;
1091 EXT4_I(inode
)->i_reserved_meta_blocks
= mdb
;
1094 /* update per-inode reservations */
1095 BUG_ON(used
> EXT4_I(inode
)->i_reserved_data_blocks
);
1096 EXT4_I(inode
)->i_reserved_data_blocks
-= used
;
1097 spin_unlock(&EXT4_I(inode
)->i_block_reservation_lock
);
1100 * free those over-booking quota for metadata blocks
1103 vfs_dq_release_reservation_block(inode
, mdb_free
);
1106 * If we have done all the pending block allocations and if
1107 * there aren't any writers on the inode, we can discard the
1108 * inode's preallocations.
1110 if (!total
&& (atomic_read(&inode
->i_writecount
) == 0))
1111 ext4_discard_preallocations(inode
);
1114 static int check_block_validity(struct inode
*inode
, sector_t logical
,
1115 sector_t phys
, int len
)
1117 if (!ext4_data_block_valid(EXT4_SB(inode
->i_sb
), phys
, len
)) {
1118 ext4_error(inode
->i_sb
, "check_block_validity",
1119 "inode #%lu logical block %llu mapped to %llu "
1120 "(size %d)", inode
->i_ino
,
1121 (unsigned long long) logical
,
1122 (unsigned long long) phys
, len
);
1130 * The ext4_get_blocks() function tries to look up the requested blocks,
1131 * and returns if the blocks are already mapped.
1133 * Otherwise it takes the write lock of the i_data_sem and allocate blocks
1134 * and store the allocated blocks in the result buffer head and mark it
1137 * If file type is extents based, it will call ext4_ext_get_blocks(),
1138 * Otherwise, call with ext4_ind_get_blocks() to handle indirect mapping
1141 * On success, it returns the number of blocks being mapped or allocate.
1142 * if create==0 and the blocks are pre-allocated and uninitialized block,
1143 * the result buffer head is unmapped. If the create ==1, it will make sure
1144 * the buffer head is mapped.
1146 * It returns 0 if plain look up failed (blocks have not been allocated), in
1147 * that casem, buffer head is unmapped
1149 * It returns the error in case of allocation failure.
1151 int ext4_get_blocks(handle_t
*handle
, struct inode
*inode
, sector_t block
,
1152 unsigned int max_blocks
, struct buffer_head
*bh
,
1157 clear_buffer_mapped(bh
);
1158 clear_buffer_unwritten(bh
);
1161 * Try to see if we can get the block without requesting a new
1162 * file system block.
1164 down_read((&EXT4_I(inode
)->i_data_sem
));
1165 if (EXT4_I(inode
)->i_flags
& EXT4_EXTENTS_FL
) {
1166 retval
= ext4_ext_get_blocks(handle
, inode
, block
, max_blocks
,
1169 retval
= ext4_ind_get_blocks(handle
, inode
, block
, max_blocks
,
1172 up_read((&EXT4_I(inode
)->i_data_sem
));
1174 if (retval
> 0 && buffer_mapped(bh
)) {
1175 int ret
= check_block_validity(inode
, block
,
1176 bh
->b_blocknr
, retval
);
1181 /* If it is only a block(s) look up */
1182 if ((flags
& EXT4_GET_BLOCKS_CREATE
) == 0)
1186 * Returns if the blocks have already allocated
1188 * Note that if blocks have been preallocated
1189 * ext4_ext_get_block() returns th create = 0
1190 * with buffer head unmapped.
1192 if (retval
> 0 && buffer_mapped(bh
))
1196 * When we call get_blocks without the create flag, the
1197 * BH_Unwritten flag could have gotten set if the blocks
1198 * requested were part of a uninitialized extent. We need to
1199 * clear this flag now that we are committed to convert all or
1200 * part of the uninitialized extent to be an initialized
1201 * extent. This is because we need to avoid the combination
1202 * of BH_Unwritten and BH_Mapped flags being simultaneously
1203 * set on the buffer_head.
1205 clear_buffer_unwritten(bh
);
1208 * New blocks allocate and/or writing to uninitialized extent
1209 * will possibly result in updating i_data, so we take
1210 * the write lock of i_data_sem, and call get_blocks()
1211 * with create == 1 flag.
1213 down_write((&EXT4_I(inode
)->i_data_sem
));
1216 * if the caller is from delayed allocation writeout path
1217 * we have already reserved fs blocks for allocation
1218 * let the underlying get_block() function know to
1219 * avoid double accounting
1221 if (flags
& EXT4_GET_BLOCKS_DELALLOC_RESERVE
)
1222 EXT4_I(inode
)->i_delalloc_reserved_flag
= 1;
1224 * We need to check for EXT4 here because migrate
1225 * could have changed the inode type in between
1227 if (EXT4_I(inode
)->i_flags
& EXT4_EXTENTS_FL
) {
1228 retval
= ext4_ext_get_blocks(handle
, inode
, block
, max_blocks
,
1231 retval
= ext4_ind_get_blocks(handle
, inode
, block
,
1232 max_blocks
, bh
, flags
);
1234 if (retval
> 0 && buffer_new(bh
)) {
1236 * We allocated new blocks which will result in
1237 * i_data's format changing. Force the migrate
1238 * to fail by clearing migrate flags
1240 EXT4_I(inode
)->i_flags
= EXT4_I(inode
)->i_flags
&
1245 if (flags
& EXT4_GET_BLOCKS_DELALLOC_RESERVE
)
1246 EXT4_I(inode
)->i_delalloc_reserved_flag
= 0;
1249 * Update reserved blocks/metadata blocks after successful
1250 * block allocation which had been deferred till now.
1252 if ((retval
> 0) && (flags
& EXT4_GET_BLOCKS_UPDATE_RESERVE_SPACE
))
1253 ext4_da_update_reserve_space(inode
, retval
);
1255 up_write((&EXT4_I(inode
)->i_data_sem
));
1256 if (retval
> 0 && buffer_mapped(bh
)) {
1257 int ret
= check_block_validity(inode
, block
,
1258 bh
->b_blocknr
, retval
);
1265 /* Maximum number of blocks we map for direct IO at once. */
1266 #define DIO_MAX_BLOCKS 4096
1268 int ext4_get_block(struct inode
*inode
, sector_t iblock
,
1269 struct buffer_head
*bh_result
, int create
)
1271 handle_t
*handle
= ext4_journal_current_handle();
1272 int ret
= 0, started
= 0;
1273 unsigned max_blocks
= bh_result
->b_size
>> inode
->i_blkbits
;
1276 if (create
&& !handle
) {
1277 /* Direct IO write... */
1278 if (max_blocks
> DIO_MAX_BLOCKS
)
1279 max_blocks
= DIO_MAX_BLOCKS
;
1280 dio_credits
= ext4_chunk_trans_blocks(inode
, max_blocks
);
1281 handle
= ext4_journal_start(inode
, dio_credits
);
1282 if (IS_ERR(handle
)) {
1283 ret
= PTR_ERR(handle
);
1289 ret
= ext4_get_blocks(handle
, inode
, iblock
, max_blocks
, bh_result
,
1290 create
? EXT4_GET_BLOCKS_CREATE
: 0);
1292 bh_result
->b_size
= (ret
<< inode
->i_blkbits
);
1296 ext4_journal_stop(handle
);
1302 * `handle' can be NULL if create is zero
1304 struct buffer_head
*ext4_getblk(handle_t
*handle
, struct inode
*inode
,
1305 ext4_lblk_t block
, int create
, int *errp
)
1307 struct buffer_head dummy
;
1311 J_ASSERT(handle
!= NULL
|| create
== 0);
1314 dummy
.b_blocknr
= -1000;
1315 buffer_trace_init(&dummy
.b_history
);
1317 flags
|= EXT4_GET_BLOCKS_CREATE
;
1318 err
= ext4_get_blocks(handle
, inode
, block
, 1, &dummy
, flags
);
1320 * ext4_get_blocks() returns number of blocks mapped. 0 in
1329 if (!err
&& buffer_mapped(&dummy
)) {
1330 struct buffer_head
*bh
;
1331 bh
= sb_getblk(inode
->i_sb
, dummy
.b_blocknr
);
1336 if (buffer_new(&dummy
)) {
1337 J_ASSERT(create
!= 0);
1338 J_ASSERT(handle
!= NULL
);
1341 * Now that we do not always journal data, we should
1342 * keep in mind whether this should always journal the
1343 * new buffer as metadata. For now, regular file
1344 * writes use ext4_get_block instead, so it's not a
1348 BUFFER_TRACE(bh
, "call get_create_access");
1349 fatal
= ext4_journal_get_create_access(handle
, bh
);
1350 if (!fatal
&& !buffer_uptodate(bh
)) {
1351 memset(bh
->b_data
, 0, inode
->i_sb
->s_blocksize
);
1352 set_buffer_uptodate(bh
);
1355 BUFFER_TRACE(bh
, "call ext4_handle_dirty_metadata");
1356 err
= ext4_handle_dirty_metadata(handle
, inode
, bh
);
1360 BUFFER_TRACE(bh
, "not a new buffer");
1373 struct buffer_head
*ext4_bread(handle_t
*handle
, struct inode
*inode
,
1374 ext4_lblk_t block
, int create
, int *err
)
1376 struct buffer_head
*bh
;
1378 bh
= ext4_getblk(handle
, inode
, block
, create
, err
);
1381 if (buffer_uptodate(bh
))
1383 ll_rw_block(READ_META
, 1, &bh
);
1385 if (buffer_uptodate(bh
))
1392 static int walk_page_buffers(handle_t
*handle
,
1393 struct buffer_head
*head
,
1397 int (*fn
)(handle_t
*handle
,
1398 struct buffer_head
*bh
))
1400 struct buffer_head
*bh
;
1401 unsigned block_start
, block_end
;
1402 unsigned blocksize
= head
->b_size
;
1404 struct buffer_head
*next
;
1406 for (bh
= head
, block_start
= 0;
1407 ret
== 0 && (bh
!= head
|| !block_start
);
1408 block_start
= block_end
, bh
= next
)
1410 next
= bh
->b_this_page
;
1411 block_end
= block_start
+ blocksize
;
1412 if (block_end
<= from
|| block_start
>= to
) {
1413 if (partial
&& !buffer_uptodate(bh
))
1417 err
= (*fn
)(handle
, bh
);
1425 * To preserve ordering, it is essential that the hole instantiation and
1426 * the data write be encapsulated in a single transaction. We cannot
1427 * close off a transaction and start a new one between the ext4_get_block()
1428 * and the commit_write(). So doing the jbd2_journal_start at the start of
1429 * prepare_write() is the right place.
1431 * Also, this function can nest inside ext4_writepage() ->
1432 * block_write_full_page(). In that case, we *know* that ext4_writepage()
1433 * has generated enough buffer credits to do the whole page. So we won't
1434 * block on the journal in that case, which is good, because the caller may
1437 * By accident, ext4 can be reentered when a transaction is open via
1438 * quota file writes. If we were to commit the transaction while thus
1439 * reentered, there can be a deadlock - we would be holding a quota
1440 * lock, and the commit would never complete if another thread had a
1441 * transaction open and was blocking on the quota lock - a ranking
1444 * So what we do is to rely on the fact that jbd2_journal_stop/journal_start
1445 * will _not_ run commit under these circumstances because handle->h_ref
1446 * is elevated. We'll still have enough credits for the tiny quotafile
1449 static int do_journal_get_write_access(handle_t
*handle
,
1450 struct buffer_head
*bh
)
1452 if (!buffer_mapped(bh
) || buffer_freed(bh
))
1454 return ext4_journal_get_write_access(handle
, bh
);
1457 static int ext4_write_begin(struct file
*file
, struct address_space
*mapping
,
1458 loff_t pos
, unsigned len
, unsigned flags
,
1459 struct page
**pagep
, void **fsdata
)
1461 struct inode
*inode
= mapping
->host
;
1462 int ret
, needed_blocks
;
1469 trace_mark(ext4_write_begin
,
1470 "dev %s ino %lu pos %llu len %u flags %u",
1471 inode
->i_sb
->s_id
, inode
->i_ino
,
1472 (unsigned long long) pos
, len
, flags
);
1474 * Reserve one block more for addition to orphan list in case
1475 * we allocate blocks but write fails for some reason
1477 needed_blocks
= ext4_writepage_trans_blocks(inode
) + 1;
1478 index
= pos
>> PAGE_CACHE_SHIFT
;
1479 from
= pos
& (PAGE_CACHE_SIZE
- 1);
1483 handle
= ext4_journal_start(inode
, needed_blocks
);
1484 if (IS_ERR(handle
)) {
1485 ret
= PTR_ERR(handle
);
1489 /* We cannot recurse into the filesystem as the transaction is already
1491 flags
|= AOP_FLAG_NOFS
;
1493 page
= grab_cache_page_write_begin(mapping
, index
, flags
);
1495 ext4_journal_stop(handle
);
1501 ret
= block_write_begin(file
, mapping
, pos
, len
, flags
, pagep
, fsdata
,
1504 if (!ret
&& ext4_should_journal_data(inode
)) {
1505 ret
= walk_page_buffers(handle
, page_buffers(page
),
1506 from
, to
, NULL
, do_journal_get_write_access
);
1511 page_cache_release(page
);
1513 * block_write_begin may have instantiated a few blocks
1514 * outside i_size. Trim these off again. Don't need
1515 * i_size_read because we hold i_mutex.
1517 * Add inode to orphan list in case we crash before
1520 if (pos
+ len
> inode
->i_size
)
1521 ext4_orphan_add(handle
, inode
);
1523 ext4_journal_stop(handle
);
1524 if (pos
+ len
> inode
->i_size
) {
1525 vmtruncate(inode
, inode
->i_size
);
1527 * If vmtruncate failed early the inode might
1528 * still be on the orphan list; we need to
1529 * make sure the inode is removed from the
1530 * orphan list in that case.
1533 ext4_orphan_del(NULL
, inode
);
1537 if (ret
== -ENOSPC
&& ext4_should_retry_alloc(inode
->i_sb
, &retries
))
1543 /* For write_end() in data=journal mode */
1544 static int write_end_fn(handle_t
*handle
, struct buffer_head
*bh
)
1546 if (!buffer_mapped(bh
) || buffer_freed(bh
))
1548 set_buffer_uptodate(bh
);
1549 return ext4_handle_dirty_metadata(handle
, NULL
, bh
);
1552 static int ext4_generic_write_end(struct file
*file
,
1553 struct address_space
*mapping
,
1554 loff_t pos
, unsigned len
, unsigned copied
,
1555 struct page
*page
, void *fsdata
)
1557 int i_size_changed
= 0;
1558 struct inode
*inode
= mapping
->host
;
1559 handle_t
*handle
= ext4_journal_current_handle();
1561 copied
= block_write_end(file
, mapping
, pos
, len
, copied
, page
, fsdata
);
1564 * No need to use i_size_read() here, the i_size
1565 * cannot change under us because we hold i_mutex.
1567 * But it's important to update i_size while still holding page lock:
1568 * page writeout could otherwise come in and zero beyond i_size.
1570 if (pos
+ copied
> inode
->i_size
) {
1571 i_size_write(inode
, pos
+ copied
);
1575 if (pos
+ copied
> EXT4_I(inode
)->i_disksize
) {
1576 /* We need to mark inode dirty even if
1577 * new_i_size is less that inode->i_size
1578 * bu greater than i_disksize.(hint delalloc)
1580 ext4_update_i_disksize(inode
, (pos
+ copied
));
1584 page_cache_release(page
);
1587 * Don't mark the inode dirty under page lock. First, it unnecessarily
1588 * makes the holding time of page lock longer. Second, it forces lock
1589 * ordering of page lock and transaction start for journaling
1593 ext4_mark_inode_dirty(handle
, inode
);
1599 * We need to pick up the new inode size which generic_commit_write gave us
1600 * `file' can be NULL - eg, when called from page_symlink().
1602 * ext4 never places buffers on inode->i_mapping->private_list. metadata
1603 * buffers are managed internally.
1605 static int ext4_ordered_write_end(struct file
*file
,
1606 struct address_space
*mapping
,
1607 loff_t pos
, unsigned len
, unsigned copied
,
1608 struct page
*page
, void *fsdata
)
1610 handle_t
*handle
= ext4_journal_current_handle();
1611 struct inode
*inode
= mapping
->host
;
1614 trace_mark(ext4_ordered_write_end
,
1615 "dev %s ino %lu pos %llu len %u copied %u",
1616 inode
->i_sb
->s_id
, inode
->i_ino
,
1617 (unsigned long long) pos
, len
, copied
);
1618 ret
= ext4_jbd2_file_inode(handle
, inode
);
1621 ret2
= ext4_generic_write_end(file
, mapping
, pos
, len
, copied
,
1624 if (pos
+ len
> inode
->i_size
)
1625 /* if we have allocated more blocks and copied
1626 * less. We will have blocks allocated outside
1627 * inode->i_size. So truncate them
1629 ext4_orphan_add(handle
, inode
);
1633 ret2
= ext4_journal_stop(handle
);
1637 if (pos
+ len
> inode
->i_size
) {
1638 vmtruncate(inode
, inode
->i_size
);
1640 * If vmtruncate failed early the inode might still be
1641 * on the orphan list; we need to make sure the inode
1642 * is removed from the orphan list in that case.
1645 ext4_orphan_del(NULL
, inode
);
1649 return ret
? ret
: copied
;
1652 static int ext4_writeback_write_end(struct file
*file
,
1653 struct address_space
*mapping
,
1654 loff_t pos
, unsigned len
, unsigned copied
,
1655 struct page
*page
, void *fsdata
)
1657 handle_t
*handle
= ext4_journal_current_handle();
1658 struct inode
*inode
= mapping
->host
;
1661 trace_mark(ext4_writeback_write_end
,
1662 "dev %s ino %lu pos %llu len %u copied %u",
1663 inode
->i_sb
->s_id
, inode
->i_ino
,
1664 (unsigned long long) pos
, len
, copied
);
1665 ret2
= ext4_generic_write_end(file
, mapping
, pos
, len
, copied
,
1668 if (pos
+ len
> inode
->i_size
)
1669 /* if we have allocated more blocks and copied
1670 * less. We will have blocks allocated outside
1671 * inode->i_size. So truncate them
1673 ext4_orphan_add(handle
, inode
);
1678 ret2
= ext4_journal_stop(handle
);
1682 if (pos
+ len
> inode
->i_size
) {
1683 vmtruncate(inode
, inode
->i_size
);
1685 * If vmtruncate failed early the inode might still be
1686 * on the orphan list; we need to make sure the inode
1687 * is removed from the orphan list in that case.
1690 ext4_orphan_del(NULL
, inode
);
1693 return ret
? ret
: copied
;
1696 static int ext4_journalled_write_end(struct file
*file
,
1697 struct address_space
*mapping
,
1698 loff_t pos
, unsigned len
, unsigned copied
,
1699 struct page
*page
, void *fsdata
)
1701 handle_t
*handle
= ext4_journal_current_handle();
1702 struct inode
*inode
= mapping
->host
;
1708 trace_mark(ext4_journalled_write_end
,
1709 "dev %s ino %lu pos %llu len %u copied %u",
1710 inode
->i_sb
->s_id
, inode
->i_ino
,
1711 (unsigned long long) pos
, len
, copied
);
1712 from
= pos
& (PAGE_CACHE_SIZE
- 1);
1716 if (!PageUptodate(page
))
1718 page_zero_new_buffers(page
, from
+copied
, to
);
1721 ret
= walk_page_buffers(handle
, page_buffers(page
), from
,
1722 to
, &partial
, write_end_fn
);
1724 SetPageUptodate(page
);
1725 new_i_size
= pos
+ copied
;
1726 if (new_i_size
> inode
->i_size
)
1727 i_size_write(inode
, pos
+copied
);
1728 EXT4_I(inode
)->i_state
|= EXT4_STATE_JDATA
;
1729 if (new_i_size
> EXT4_I(inode
)->i_disksize
) {
1730 ext4_update_i_disksize(inode
, new_i_size
);
1731 ret2
= ext4_mark_inode_dirty(handle
, inode
);
1737 page_cache_release(page
);
1738 if (pos
+ len
> inode
->i_size
)
1739 /* if we have allocated more blocks and copied
1740 * less. We will have blocks allocated outside
1741 * inode->i_size. So truncate them
1743 ext4_orphan_add(handle
, inode
);
1745 ret2
= ext4_journal_stop(handle
);
1748 if (pos
+ len
> inode
->i_size
) {
1749 vmtruncate(inode
, inode
->i_size
);
1751 * If vmtruncate failed early the inode might still be
1752 * on the orphan list; we need to make sure the inode
1753 * is removed from the orphan list in that case.
1756 ext4_orphan_del(NULL
, inode
);
1759 return ret
? ret
: copied
;
1762 static int ext4_da_reserve_space(struct inode
*inode
, int nrblocks
)
1765 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
1766 unsigned long md_needed
, mdblocks
, total
= 0;
1769 * recalculate the amount of metadata blocks to reserve
1770 * in order to allocate nrblocks
1771 * worse case is one extent per block
1774 spin_lock(&EXT4_I(inode
)->i_block_reservation_lock
);
1775 total
= EXT4_I(inode
)->i_reserved_data_blocks
+ nrblocks
;
1776 mdblocks
= ext4_calc_metadata_amount(inode
, total
);
1777 BUG_ON(mdblocks
< EXT4_I(inode
)->i_reserved_meta_blocks
);
1779 md_needed
= mdblocks
- EXT4_I(inode
)->i_reserved_meta_blocks
;
1780 total
= md_needed
+ nrblocks
;
1783 * Make quota reservation here to prevent quota overflow
1784 * later. Real quota accounting is done at pages writeout
1787 if (vfs_dq_reserve_block(inode
, total
)) {
1788 spin_unlock(&EXT4_I(inode
)->i_block_reservation_lock
);
1792 if (ext4_claim_free_blocks(sbi
, total
)) {
1793 spin_unlock(&EXT4_I(inode
)->i_block_reservation_lock
);
1794 if (ext4_should_retry_alloc(inode
->i_sb
, &retries
)) {
1798 vfs_dq_release_reservation_block(inode
, total
);
1801 EXT4_I(inode
)->i_reserved_data_blocks
+= nrblocks
;
1802 EXT4_I(inode
)->i_reserved_meta_blocks
= mdblocks
;
1804 spin_unlock(&EXT4_I(inode
)->i_block_reservation_lock
);
1805 return 0; /* success */
1808 static void ext4_da_release_space(struct inode
*inode
, int to_free
)
1810 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
1811 int total
, mdb
, mdb_free
, release
;
1814 return; /* Nothing to release, exit */
1816 spin_lock(&EXT4_I(inode
)->i_block_reservation_lock
);
1818 if (!EXT4_I(inode
)->i_reserved_data_blocks
) {
1820 * if there is no reserved blocks, but we try to free some
1821 * then the counter is messed up somewhere.
1822 * but since this function is called from invalidate
1823 * page, it's harmless to return without any action
1825 printk(KERN_INFO
"ext4 delalloc try to release %d reserved "
1826 "blocks for inode %lu, but there is no reserved "
1827 "data blocks\n", to_free
, inode
->i_ino
);
1828 spin_unlock(&EXT4_I(inode
)->i_block_reservation_lock
);
1832 /* recalculate the number of metablocks still need to be reserved */
1833 total
= EXT4_I(inode
)->i_reserved_data_blocks
- to_free
;
1834 mdb
= ext4_calc_metadata_amount(inode
, total
);
1836 /* figure out how many metablocks to release */
1837 BUG_ON(mdb
> EXT4_I(inode
)->i_reserved_meta_blocks
);
1838 mdb_free
= EXT4_I(inode
)->i_reserved_meta_blocks
- mdb
;
1840 release
= to_free
+ mdb_free
;
1842 /* update fs dirty blocks counter for truncate case */
1843 percpu_counter_sub(&sbi
->s_dirtyblocks_counter
, release
);
1845 /* update per-inode reservations */
1846 BUG_ON(to_free
> EXT4_I(inode
)->i_reserved_data_blocks
);
1847 EXT4_I(inode
)->i_reserved_data_blocks
-= to_free
;
1849 BUG_ON(mdb
> EXT4_I(inode
)->i_reserved_meta_blocks
);
1850 EXT4_I(inode
)->i_reserved_meta_blocks
= mdb
;
1851 spin_unlock(&EXT4_I(inode
)->i_block_reservation_lock
);
1853 vfs_dq_release_reservation_block(inode
, release
);
1856 static void ext4_da_page_release_reservation(struct page
*page
,
1857 unsigned long offset
)
1860 struct buffer_head
*head
, *bh
;
1861 unsigned int curr_off
= 0;
1863 head
= page_buffers(page
);
1866 unsigned int next_off
= curr_off
+ bh
->b_size
;
1868 if ((offset
<= curr_off
) && (buffer_delay(bh
))) {
1870 clear_buffer_delay(bh
);
1872 curr_off
= next_off
;
1873 } while ((bh
= bh
->b_this_page
) != head
);
1874 ext4_da_release_space(page
->mapping
->host
, to_release
);
1878 * Delayed allocation stuff
1881 struct mpage_da_data
{
1882 struct inode
*inode
;
1883 sector_t b_blocknr
; /* start block number of extent */
1884 size_t b_size
; /* size of extent */
1885 unsigned long b_state
; /* state of the extent */
1886 unsigned long first_page
, next_page
; /* extent of pages */
1887 struct writeback_control
*wbc
;
1894 * mpage_da_submit_io - walks through extent of pages and try to write
1895 * them with writepage() call back
1897 * @mpd->inode: inode
1898 * @mpd->first_page: first page of the extent
1899 * @mpd->next_page: page after the last page of the extent
1901 * By the time mpage_da_submit_io() is called we expect all blocks
1902 * to be allocated. this may be wrong if allocation failed.
1904 * As pages are already locked by write_cache_pages(), we can't use it
1906 static int mpage_da_submit_io(struct mpage_da_data
*mpd
)
1909 struct pagevec pvec
;
1910 unsigned long index
, end
;
1911 int ret
= 0, err
, nr_pages
, i
;
1912 struct inode
*inode
= mpd
->inode
;
1913 struct address_space
*mapping
= inode
->i_mapping
;
1915 BUG_ON(mpd
->next_page
<= mpd
->first_page
);
1917 * We need to start from the first_page to the next_page - 1
1918 * to make sure we also write the mapped dirty buffer_heads.
1919 * If we look at mpd->b_blocknr we would only be looking
1920 * at the currently mapped buffer_heads.
1922 index
= mpd
->first_page
;
1923 end
= mpd
->next_page
- 1;
1925 pagevec_init(&pvec
, 0);
1926 while (index
<= end
) {
1927 nr_pages
= pagevec_lookup(&pvec
, mapping
, index
, PAGEVEC_SIZE
);
1930 for (i
= 0; i
< nr_pages
; i
++) {
1931 struct page
*page
= pvec
.pages
[i
];
1933 index
= page
->index
;
1938 BUG_ON(!PageLocked(page
));
1939 BUG_ON(PageWriteback(page
));
1941 pages_skipped
= mpd
->wbc
->pages_skipped
;
1942 err
= mapping
->a_ops
->writepage(page
, mpd
->wbc
);
1943 if (!err
&& (pages_skipped
== mpd
->wbc
->pages_skipped
))
1945 * have successfully written the page
1946 * without skipping the same
1948 mpd
->pages_written
++;
1950 * In error case, we have to continue because
1951 * remaining pages are still locked
1952 * XXX: unlock and re-dirty them?
1957 pagevec_release(&pvec
);
1963 * mpage_put_bnr_to_bhs - walk blocks and assign them actual numbers
1965 * @mpd->inode - inode to walk through
1966 * @exbh->b_blocknr - first block on a disk
1967 * @exbh->b_size - amount of space in bytes
1968 * @logical - first logical block to start assignment with
1970 * the function goes through all passed space and put actual disk
1971 * block numbers into buffer heads, dropping BH_Delay and BH_Unwritten
1973 static void mpage_put_bnr_to_bhs(struct mpage_da_data
*mpd
, sector_t logical
,
1974 struct buffer_head
*exbh
)
1976 struct inode
*inode
= mpd
->inode
;
1977 struct address_space
*mapping
= inode
->i_mapping
;
1978 int blocks
= exbh
->b_size
>> inode
->i_blkbits
;
1979 sector_t pblock
= exbh
->b_blocknr
, cur_logical
;
1980 struct buffer_head
*head
, *bh
;
1982 struct pagevec pvec
;
1985 index
= logical
>> (PAGE_CACHE_SHIFT
- inode
->i_blkbits
);
1986 end
= (logical
+ blocks
- 1) >> (PAGE_CACHE_SHIFT
- inode
->i_blkbits
);
1987 cur_logical
= index
<< (PAGE_CACHE_SHIFT
- inode
->i_blkbits
);
1989 pagevec_init(&pvec
, 0);
1991 while (index
<= end
) {
1992 /* XXX: optimize tail */
1993 nr_pages
= pagevec_lookup(&pvec
, mapping
, index
, PAGEVEC_SIZE
);
1996 for (i
= 0; i
< nr_pages
; i
++) {
1997 struct page
*page
= pvec
.pages
[i
];
1999 index
= page
->index
;
2004 BUG_ON(!PageLocked(page
));
2005 BUG_ON(PageWriteback(page
));
2006 BUG_ON(!page_has_buffers(page
));
2008 bh
= page_buffers(page
);
2011 /* skip blocks out of the range */
2013 if (cur_logical
>= logical
)
2016 } while ((bh
= bh
->b_this_page
) != head
);
2019 if (cur_logical
>= logical
+ blocks
)
2022 if (buffer_delay(bh
) ||
2023 buffer_unwritten(bh
)) {
2025 BUG_ON(bh
->b_bdev
!= inode
->i_sb
->s_bdev
);
2027 if (buffer_delay(bh
)) {
2028 clear_buffer_delay(bh
);
2029 bh
->b_blocknr
= pblock
;
2032 * unwritten already should have
2033 * blocknr assigned. Verify that
2035 clear_buffer_unwritten(bh
);
2036 BUG_ON(bh
->b_blocknr
!= pblock
);
2039 } else if (buffer_mapped(bh
))
2040 BUG_ON(bh
->b_blocknr
!= pblock
);
2044 } while ((bh
= bh
->b_this_page
) != head
);
2046 pagevec_release(&pvec
);
2052 * __unmap_underlying_blocks - just a helper function to unmap
2053 * set of blocks described by @bh
2055 static inline void __unmap_underlying_blocks(struct inode
*inode
,
2056 struct buffer_head
*bh
)
2058 struct block_device
*bdev
= inode
->i_sb
->s_bdev
;
2061 blocks
= bh
->b_size
>> inode
->i_blkbits
;
2062 for (i
= 0; i
< blocks
; i
++)
2063 unmap_underlying_metadata(bdev
, bh
->b_blocknr
+ i
);
2066 static void ext4_da_block_invalidatepages(struct mpage_da_data
*mpd
,
2067 sector_t logical
, long blk_cnt
)
2071 struct pagevec pvec
;
2072 struct inode
*inode
= mpd
->inode
;
2073 struct address_space
*mapping
= inode
->i_mapping
;
2075 index
= logical
>> (PAGE_CACHE_SHIFT
- inode
->i_blkbits
);
2076 end
= (logical
+ blk_cnt
- 1) >>
2077 (PAGE_CACHE_SHIFT
- inode
->i_blkbits
);
2078 while (index
<= end
) {
2079 nr_pages
= pagevec_lookup(&pvec
, mapping
, index
, PAGEVEC_SIZE
);
2082 for (i
= 0; i
< nr_pages
; i
++) {
2083 struct page
*page
= pvec
.pages
[i
];
2084 index
= page
->index
;
2089 BUG_ON(!PageLocked(page
));
2090 BUG_ON(PageWriteback(page
));
2091 block_invalidatepage(page
, 0);
2092 ClearPageUptodate(page
);
2099 static void ext4_print_free_blocks(struct inode
*inode
)
2101 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
2102 printk(KERN_EMERG
"Total free blocks count %lld\n",
2103 ext4_count_free_blocks(inode
->i_sb
));
2104 printk(KERN_EMERG
"Free/Dirty block details\n");
2105 printk(KERN_EMERG
"free_blocks=%lld\n",
2106 (long long)percpu_counter_sum(&sbi
->s_freeblocks_counter
));
2107 printk(KERN_EMERG
"dirty_blocks=%lld\n",
2108 (long long)percpu_counter_sum(&sbi
->s_dirtyblocks_counter
));
2109 printk(KERN_EMERG
"Block reservation details\n");
2110 printk(KERN_EMERG
"i_reserved_data_blocks=%u\n",
2111 EXT4_I(inode
)->i_reserved_data_blocks
);
2112 printk(KERN_EMERG
"i_reserved_meta_blocks=%u\n",
2113 EXT4_I(inode
)->i_reserved_meta_blocks
);
2118 * mpage_da_map_blocks - go through given space
2120 * @mpd - bh describing space
2122 * The function skips space we know is already mapped to disk blocks.
2125 static int mpage_da_map_blocks(struct mpage_da_data
*mpd
)
2127 int err
, blks
, get_blocks_flags
;
2128 struct buffer_head
new;
2129 sector_t next
= mpd
->b_blocknr
;
2130 unsigned max_blocks
= mpd
->b_size
>> mpd
->inode
->i_blkbits
;
2131 loff_t disksize
= EXT4_I(mpd
->inode
)->i_disksize
;
2132 handle_t
*handle
= NULL
;
2135 * We consider only non-mapped and non-allocated blocks
2137 if ((mpd
->b_state
& (1 << BH_Mapped
)) &&
2138 !(mpd
->b_state
& (1 << BH_Delay
)) &&
2139 !(mpd
->b_state
& (1 << BH_Unwritten
)))
2143 * If we didn't accumulate anything to write simply return
2148 handle
= ext4_journal_current_handle();
2152 * Call ext4_get_blocks() to allocate any delayed allocation
2153 * blocks, or to convert an uninitialized extent to be
2154 * initialized (in the case where we have written into
2155 * one or more preallocated blocks).
2157 * We pass in the magic EXT4_GET_BLOCKS_DELALLOC_RESERVE to
2158 * indicate that we are on the delayed allocation path. This
2159 * affects functions in many different parts of the allocation
2160 * call path. This flag exists primarily because we don't
2161 * want to change *many* call functions, so ext4_get_blocks()
2162 * will set the magic i_delalloc_reserved_flag once the
2163 * inode's allocation semaphore is taken.
2165 * If the blocks in questions were delalloc blocks, set
2166 * EXT4_GET_BLOCKS_DELALLOC_RESERVE so the delalloc accounting
2167 * variables are updated after the blocks have been allocated.
2170 get_blocks_flags
= (EXT4_GET_BLOCKS_CREATE
|
2171 EXT4_GET_BLOCKS_DELALLOC_RESERVE
);
2172 if (mpd
->b_state
& (1 << BH_Delay
))
2173 get_blocks_flags
|= EXT4_GET_BLOCKS_UPDATE_RESERVE_SPACE
;
2174 blks
= ext4_get_blocks(handle
, mpd
->inode
, next
, max_blocks
,
2175 &new, get_blocks_flags
);
2179 * If get block returns with error we simply
2180 * return. Later writepage will redirty the page and
2181 * writepages will find the dirty page again
2186 if (err
== -ENOSPC
&&
2187 ext4_count_free_blocks(mpd
->inode
->i_sb
)) {
2193 * get block failure will cause us to loop in
2194 * writepages, because a_ops->writepage won't be able
2195 * to make progress. The page will be redirtied by
2196 * writepage and writepages will again try to write
2199 printk(KERN_EMERG
"%s block allocation failed for inode %lu "
2200 "at logical offset %llu with max blocks "
2201 "%zd with error %d\n",
2202 __func__
, mpd
->inode
->i_ino
,
2203 (unsigned long long)next
,
2204 mpd
->b_size
>> mpd
->inode
->i_blkbits
, err
);
2205 printk(KERN_EMERG
"This should not happen.!! "
2206 "Data will be lost\n");
2207 if (err
== -ENOSPC
) {
2208 ext4_print_free_blocks(mpd
->inode
);
2210 /* invalidate all the pages */
2211 ext4_da_block_invalidatepages(mpd
, next
,
2212 mpd
->b_size
>> mpd
->inode
->i_blkbits
);
2217 new.b_size
= (blks
<< mpd
->inode
->i_blkbits
);
2219 if (buffer_new(&new))
2220 __unmap_underlying_blocks(mpd
->inode
, &new);
2223 * If blocks are delayed marked, we need to
2224 * put actual blocknr and drop delayed bit
2226 if ((mpd
->b_state
& (1 << BH_Delay
)) ||
2227 (mpd
->b_state
& (1 << BH_Unwritten
)))
2228 mpage_put_bnr_to_bhs(mpd
, next
, &new);
2230 if (ext4_should_order_data(mpd
->inode
)) {
2231 err
= ext4_jbd2_file_inode(handle
, mpd
->inode
);
2237 * Update on-disk size along with block allocation.
2239 disksize
= ((loff_t
) next
+ blks
) << mpd
->inode
->i_blkbits
;
2240 if (disksize
> i_size_read(mpd
->inode
))
2241 disksize
= i_size_read(mpd
->inode
);
2242 if (disksize
> EXT4_I(mpd
->inode
)->i_disksize
) {
2243 ext4_update_i_disksize(mpd
->inode
, disksize
);
2244 return ext4_mark_inode_dirty(handle
, mpd
->inode
);
2250 #define BH_FLAGS ((1 << BH_Uptodate) | (1 << BH_Mapped) | \
2251 (1 << BH_Delay) | (1 << BH_Unwritten))
2254 * mpage_add_bh_to_extent - try to add one more block to extent of blocks
2256 * @mpd->lbh - extent of blocks
2257 * @logical - logical number of the block in the file
2258 * @bh - bh of the block (used to access block's state)
2260 * the function is used to collect contig. blocks in same state
2262 static void mpage_add_bh_to_extent(struct mpage_da_data
*mpd
,
2263 sector_t logical
, size_t b_size
,
2264 unsigned long b_state
)
2267 int nrblocks
= mpd
->b_size
>> mpd
->inode
->i_blkbits
;
2269 /* check if thereserved journal credits might overflow */
2270 if (!(EXT4_I(mpd
->inode
)->i_flags
& EXT4_EXTENTS_FL
)) {
2271 if (nrblocks
>= EXT4_MAX_TRANS_DATA
) {
2273 * With non-extent format we are limited by the journal
2274 * credit available. Total credit needed to insert
2275 * nrblocks contiguous blocks is dependent on the
2276 * nrblocks. So limit nrblocks.
2279 } else if ((nrblocks
+ (b_size
>> mpd
->inode
->i_blkbits
)) >
2280 EXT4_MAX_TRANS_DATA
) {
2282 * Adding the new buffer_head would make it cross the
2283 * allowed limit for which we have journal credit
2284 * reserved. So limit the new bh->b_size
2286 b_size
= (EXT4_MAX_TRANS_DATA
- nrblocks
) <<
2287 mpd
->inode
->i_blkbits
;
2288 /* we will do mpage_da_submit_io in the next loop */
2292 * First block in the extent
2294 if (mpd
->b_size
== 0) {
2295 mpd
->b_blocknr
= logical
;
2296 mpd
->b_size
= b_size
;
2297 mpd
->b_state
= b_state
& BH_FLAGS
;
2301 next
= mpd
->b_blocknr
+ nrblocks
;
2303 * Can we merge the block to our big extent?
2305 if (logical
== next
&& (b_state
& BH_FLAGS
) == mpd
->b_state
) {
2306 mpd
->b_size
+= b_size
;
2312 * We couldn't merge the block to our extent, so we
2313 * need to flush current extent and start new one
2315 if (mpage_da_map_blocks(mpd
) == 0)
2316 mpage_da_submit_io(mpd
);
2321 static int ext4_bh_unmapped_or_delay(handle_t
*handle
, struct buffer_head
*bh
)
2324 * unmapped buffer is possible for holes.
2325 * delay buffer is possible with delayed allocation.
2326 * We also need to consider unwritten buffer as unmapped.
2328 return (!buffer_mapped(bh
) || buffer_delay(bh
) ||
2329 buffer_unwritten(bh
)) && buffer_dirty(bh
);
2333 * __mpage_da_writepage - finds extent of pages and blocks
2335 * @page: page to consider
2336 * @wbc: not used, we just follow rules
2339 * The function finds extents of pages and scan them for all blocks.
2341 static int __mpage_da_writepage(struct page
*page
,
2342 struct writeback_control
*wbc
, void *data
)
2344 struct mpage_da_data
*mpd
= data
;
2345 struct inode
*inode
= mpd
->inode
;
2346 struct buffer_head
*bh
, *head
;
2351 * Rest of the page in the page_vec
2352 * redirty then and skip then. We will
2353 * try to to write them again after
2354 * starting a new transaction
2356 redirty_page_for_writepage(wbc
, page
);
2358 return MPAGE_DA_EXTENT_TAIL
;
2361 * Can we merge this page to current extent?
2363 if (mpd
->next_page
!= page
->index
) {
2365 * Nope, we can't. So, we map non-allocated blocks
2366 * and start IO on them using writepage()
2368 if (mpd
->next_page
!= mpd
->first_page
) {
2369 if (mpage_da_map_blocks(mpd
) == 0)
2370 mpage_da_submit_io(mpd
);
2372 * skip rest of the page in the page_vec
2375 redirty_page_for_writepage(wbc
, page
);
2377 return MPAGE_DA_EXTENT_TAIL
;
2381 * Start next extent of pages ...
2383 mpd
->first_page
= page
->index
;
2393 mpd
->next_page
= page
->index
+ 1;
2394 logical
= (sector_t
) page
->index
<<
2395 (PAGE_CACHE_SHIFT
- inode
->i_blkbits
);
2397 if (!page_has_buffers(page
)) {
2398 mpage_add_bh_to_extent(mpd
, logical
, PAGE_CACHE_SIZE
,
2399 (1 << BH_Dirty
) | (1 << BH_Uptodate
));
2401 return MPAGE_DA_EXTENT_TAIL
;
2404 * Page with regular buffer heads, just add all dirty ones
2406 head
= page_buffers(page
);
2409 BUG_ON(buffer_locked(bh
));
2411 * We need to try to allocate
2412 * unmapped blocks in the same page.
2413 * Otherwise we won't make progress
2414 * with the page in ext4_da_writepage
2416 if (ext4_bh_unmapped_or_delay(NULL
, bh
)) {
2417 mpage_add_bh_to_extent(mpd
, logical
,
2421 return MPAGE_DA_EXTENT_TAIL
;
2422 } else if (buffer_dirty(bh
) && (buffer_mapped(bh
))) {
2424 * mapped dirty buffer. We need to update
2425 * the b_state because we look at
2426 * b_state in mpage_da_map_blocks. We don't
2427 * update b_size because if we find an
2428 * unmapped buffer_head later we need to
2429 * use the b_state flag of that buffer_head.
2431 if (mpd
->b_size
== 0)
2432 mpd
->b_state
= bh
->b_state
& BH_FLAGS
;
2435 } while ((bh
= bh
->b_this_page
) != head
);
2442 * This is a special get_blocks_t callback which is used by
2443 * ext4_da_write_begin(). It will either return mapped block or
2444 * reserve space for a single block.
2446 * For delayed buffer_head we have BH_Mapped, BH_New, BH_Delay set.
2447 * We also have b_blocknr = -1 and b_bdev initialized properly
2449 * For unwritten buffer_head we have BH_Mapped, BH_New, BH_Unwritten set.
2450 * We also have b_blocknr = physicalblock mapping unwritten extent and b_bdev
2451 * initialized properly.
2453 static int ext4_da_get_block_prep(struct inode
*inode
, sector_t iblock
,
2454 struct buffer_head
*bh_result
, int create
)
2457 sector_t invalid_block
= ~((sector_t
) 0xffff);
2459 if (invalid_block
< ext4_blocks_count(EXT4_SB(inode
->i_sb
)->s_es
))
2462 BUG_ON(create
== 0);
2463 BUG_ON(bh_result
->b_size
!= inode
->i_sb
->s_blocksize
);
2466 * first, we need to know whether the block is allocated already
2467 * preallocated blocks are unmapped but should treated
2468 * the same as allocated blocks.
2470 ret
= ext4_get_blocks(NULL
, inode
, iblock
, 1, bh_result
, 0);
2471 if ((ret
== 0) && !buffer_delay(bh_result
)) {
2472 /* the block isn't (pre)allocated yet, let's reserve space */
2474 * XXX: __block_prepare_write() unmaps passed block,
2477 ret
= ext4_da_reserve_space(inode
, 1);
2479 /* not enough space to reserve */
2482 map_bh(bh_result
, inode
->i_sb
, invalid_block
);
2483 set_buffer_new(bh_result
);
2484 set_buffer_delay(bh_result
);
2485 } else if (ret
> 0) {
2486 bh_result
->b_size
= (ret
<< inode
->i_blkbits
);
2487 if (buffer_unwritten(bh_result
)) {
2488 /* A delayed write to unwritten bh should
2489 * be marked new and mapped. Mapped ensures
2490 * that we don't do get_block multiple times
2491 * when we write to the same offset and new
2492 * ensures that we do proper zero out for
2495 set_buffer_new(bh_result
);
2496 set_buffer_mapped(bh_result
);
2505 * This function is used as a standard get_block_t calback function
2506 * when there is no desire to allocate any blocks. It is used as a
2507 * callback function for block_prepare_write(), nobh_writepage(), and
2508 * block_write_full_page(). These functions should only try to map a
2509 * single block at a time.
2511 * Since this function doesn't do block allocations even if the caller
2512 * requests it by passing in create=1, it is critically important that
2513 * any caller checks to make sure that any buffer heads are returned
2514 * by this function are either all already mapped or marked for
2515 * delayed allocation before calling nobh_writepage() or
2516 * block_write_full_page(). Otherwise, b_blocknr could be left
2517 * unitialized, and the page write functions will be taken by
2520 static int noalloc_get_block_write(struct inode
*inode
, sector_t iblock
,
2521 struct buffer_head
*bh_result
, int create
)
2524 unsigned max_blocks
= bh_result
->b_size
>> inode
->i_blkbits
;
2526 BUG_ON(bh_result
->b_size
!= inode
->i_sb
->s_blocksize
);
2529 * we don't want to do block allocation in writepage
2530 * so call get_block_wrap with create = 0
2532 ret
= ext4_get_blocks(NULL
, inode
, iblock
, max_blocks
, bh_result
, 0);
2533 BUG_ON(create
&& ret
== 0);
2535 bh_result
->b_size
= (ret
<< inode
->i_blkbits
);
2542 * This function can get called via...
2543 * - ext4_da_writepages after taking page lock (have journal handle)
2544 * - journal_submit_inode_data_buffers (no journal handle)
2545 * - shrink_page_list via pdflush (no journal handle)
2546 * - grab_page_cache when doing write_begin (have journal handle)
2548 static int ext4_da_writepage(struct page
*page
,
2549 struct writeback_control
*wbc
)
2554 struct buffer_head
*page_bufs
;
2555 struct inode
*inode
= page
->mapping
->host
;
2557 trace_mark(ext4_da_writepage
,
2558 "dev %s ino %lu page_index %lu",
2559 inode
->i_sb
->s_id
, inode
->i_ino
, page
->index
);
2560 size
= i_size_read(inode
);
2561 if (page
->index
== size
>> PAGE_CACHE_SHIFT
)
2562 len
= size
& ~PAGE_CACHE_MASK
;
2564 len
= PAGE_CACHE_SIZE
;
2566 if (page_has_buffers(page
)) {
2567 page_bufs
= page_buffers(page
);
2568 if (walk_page_buffers(NULL
, page_bufs
, 0, len
, NULL
,
2569 ext4_bh_unmapped_or_delay
)) {
2571 * We don't want to do block allocation
2572 * So redirty the page and return
2573 * We may reach here when we do a journal commit
2574 * via journal_submit_inode_data_buffers.
2575 * If we don't have mapping block we just ignore
2576 * them. We can also reach here via shrink_page_list
2578 redirty_page_for_writepage(wbc
, page
);
2584 * The test for page_has_buffers() is subtle:
2585 * We know the page is dirty but it lost buffers. That means
2586 * that at some moment in time after write_begin()/write_end()
2587 * has been called all buffers have been clean and thus they
2588 * must have been written at least once. So they are all
2589 * mapped and we can happily proceed with mapping them
2590 * and writing the page.
2592 * Try to initialize the buffer_heads and check whether
2593 * all are mapped and non delay. We don't want to
2594 * do block allocation here.
2596 ret
= block_prepare_write(page
, 0, PAGE_CACHE_SIZE
,
2597 noalloc_get_block_write
);
2599 page_bufs
= page_buffers(page
);
2600 /* check whether all are mapped and non delay */
2601 if (walk_page_buffers(NULL
, page_bufs
, 0, len
, NULL
,
2602 ext4_bh_unmapped_or_delay
)) {
2603 redirty_page_for_writepage(wbc
, page
);
2609 * We can't do block allocation here
2610 * so just redity the page and unlock
2613 redirty_page_for_writepage(wbc
, page
);
2617 /* now mark the buffer_heads as dirty and uptodate */
2618 block_commit_write(page
, 0, PAGE_CACHE_SIZE
);
2621 if (test_opt(inode
->i_sb
, NOBH
) && ext4_should_writeback_data(inode
))
2622 ret
= nobh_writepage(page
, noalloc_get_block_write
, wbc
);
2624 ret
= block_write_full_page(page
, noalloc_get_block_write
,
2631 * This is called via ext4_da_writepages() to
2632 * calulate the total number of credits to reserve to fit
2633 * a single extent allocation into a single transaction,
2634 * ext4_da_writpeages() will loop calling this before
2635 * the block allocation.
2638 static int ext4_da_writepages_trans_blocks(struct inode
*inode
)
2640 int max_blocks
= EXT4_I(inode
)->i_reserved_data_blocks
;
2643 * With non-extent format the journal credit needed to
2644 * insert nrblocks contiguous block is dependent on
2645 * number of contiguous block. So we will limit
2646 * number of contiguous block to a sane value
2648 if (!(inode
->i_flags
& EXT4_EXTENTS_FL
) &&
2649 (max_blocks
> EXT4_MAX_TRANS_DATA
))
2650 max_blocks
= EXT4_MAX_TRANS_DATA
;
2652 return ext4_chunk_trans_blocks(inode
, max_blocks
);
2655 static int ext4_da_writepages(struct address_space
*mapping
,
2656 struct writeback_control
*wbc
)
2659 int range_whole
= 0;
2660 handle_t
*handle
= NULL
;
2661 struct mpage_da_data mpd
;
2662 struct inode
*inode
= mapping
->host
;
2663 int no_nrwrite_index_update
;
2664 int pages_written
= 0;
2666 int range_cyclic
, cycled
= 1, io_done
= 0;
2667 int needed_blocks
, ret
= 0, nr_to_writebump
= 0;
2668 struct ext4_sb_info
*sbi
= EXT4_SB(mapping
->host
->i_sb
);
2670 trace_mark(ext4_da_writepages
,
2671 "dev %s ino %lu nr_t_write %ld "
2672 "pages_skipped %ld range_start %llu "
2673 "range_end %llu nonblocking %d "
2674 "for_kupdate %d for_reclaim %d "
2675 "for_writepages %d range_cyclic %d",
2676 inode
->i_sb
->s_id
, inode
->i_ino
,
2677 wbc
->nr_to_write
, wbc
->pages_skipped
,
2678 (unsigned long long) wbc
->range_start
,
2679 (unsigned long long) wbc
->range_end
,
2680 wbc
->nonblocking
, wbc
->for_kupdate
,
2681 wbc
->for_reclaim
, wbc
->for_writepages
,
2685 * No pages to write? This is mainly a kludge to avoid starting
2686 * a transaction for special inodes like journal inode on last iput()
2687 * because that could violate lock ordering on umount
2689 if (!mapping
->nrpages
|| !mapping_tagged(mapping
, PAGECACHE_TAG_DIRTY
))
2693 * If the filesystem has aborted, it is read-only, so return
2694 * right away instead of dumping stack traces later on that
2695 * will obscure the real source of the problem. We test
2696 * EXT4_MOUNT_ABORT instead of sb->s_flag's MS_RDONLY because
2697 * the latter could be true if the filesystem is mounted
2698 * read-only, and in that case, ext4_da_writepages should
2699 * *never* be called, so if that ever happens, we would want
2702 if (unlikely(sbi
->s_mount_opt
& EXT4_MOUNT_ABORT
))
2706 * Make sure nr_to_write is >= sbi->s_mb_stream_request
2707 * This make sure small files blocks are allocated in
2708 * single attempt. This ensure that small files
2709 * get less fragmented.
2711 if (wbc
->nr_to_write
< sbi
->s_mb_stream_request
) {
2712 nr_to_writebump
= sbi
->s_mb_stream_request
- wbc
->nr_to_write
;
2713 wbc
->nr_to_write
= sbi
->s_mb_stream_request
;
2715 if (wbc
->range_start
== 0 && wbc
->range_end
== LLONG_MAX
)
2718 range_cyclic
= wbc
->range_cyclic
;
2719 if (wbc
->range_cyclic
) {
2720 index
= mapping
->writeback_index
;
2723 wbc
->range_start
= index
<< PAGE_CACHE_SHIFT
;
2724 wbc
->range_end
= LLONG_MAX
;
2725 wbc
->range_cyclic
= 0;
2727 index
= wbc
->range_start
>> PAGE_CACHE_SHIFT
;
2730 mpd
.inode
= mapping
->host
;
2733 * we don't want write_cache_pages to update
2734 * nr_to_write and writeback_index
2736 no_nrwrite_index_update
= wbc
->no_nrwrite_index_update
;
2737 wbc
->no_nrwrite_index_update
= 1;
2738 pages_skipped
= wbc
->pages_skipped
;
2741 while (!ret
&& wbc
->nr_to_write
> 0) {
2744 * we insert one extent at a time. So we need
2745 * credit needed for single extent allocation.
2746 * journalled mode is currently not supported
2749 BUG_ON(ext4_should_journal_data(inode
));
2750 needed_blocks
= ext4_da_writepages_trans_blocks(inode
);
2752 /* start a new transaction*/
2753 handle
= ext4_journal_start(inode
, needed_blocks
);
2754 if (IS_ERR(handle
)) {
2755 ret
= PTR_ERR(handle
);
2756 printk(KERN_CRIT
"%s: jbd2_start: "
2757 "%ld pages, ino %lu; err %d\n", __func__
,
2758 wbc
->nr_to_write
, inode
->i_ino
, ret
);
2760 goto out_writepages
;
2764 * Now call __mpage_da_writepage to find the next
2765 * contiguous region of logical blocks that need
2766 * blocks to be allocated by ext4. We don't actually
2767 * submit the blocks for I/O here, even though
2768 * write_cache_pages thinks it will, and will set the
2769 * pages as clean for write before calling
2770 * __mpage_da_writepage().
2778 mpd
.pages_written
= 0;
2780 ret
= write_cache_pages(mapping
, wbc
, __mpage_da_writepage
,
2783 * If we have a contigous extent of pages and we
2784 * haven't done the I/O yet, map the blocks and submit
2787 if (!mpd
.io_done
&& mpd
.next_page
!= mpd
.first_page
) {
2788 if (mpage_da_map_blocks(&mpd
) == 0)
2789 mpage_da_submit_io(&mpd
);
2791 ret
= MPAGE_DA_EXTENT_TAIL
;
2793 wbc
->nr_to_write
-= mpd
.pages_written
;
2795 ext4_journal_stop(handle
);
2797 if ((mpd
.retval
== -ENOSPC
) && sbi
->s_journal
) {
2798 /* commit the transaction which would
2799 * free blocks released in the transaction
2802 jbd2_journal_force_commit_nested(sbi
->s_journal
);
2803 wbc
->pages_skipped
= pages_skipped
;
2805 } else if (ret
== MPAGE_DA_EXTENT_TAIL
) {
2807 * got one extent now try with
2810 pages_written
+= mpd
.pages_written
;
2811 wbc
->pages_skipped
= pages_skipped
;
2814 } else if (wbc
->nr_to_write
)
2816 * There is no more writeout needed
2817 * or we requested for a noblocking writeout
2818 * and we found the device congested
2822 if (!io_done
&& !cycled
) {
2825 wbc
->range_start
= index
<< PAGE_CACHE_SHIFT
;
2826 wbc
->range_end
= mapping
->writeback_index
- 1;
2829 if (pages_skipped
!= wbc
->pages_skipped
)
2830 printk(KERN_EMERG
"This should not happen leaving %s "
2831 "with nr_to_write = %ld ret = %d\n",
2832 __func__
, wbc
->nr_to_write
, ret
);
2835 index
+= pages_written
;
2836 wbc
->range_cyclic
= range_cyclic
;
2837 if (wbc
->range_cyclic
|| (range_whole
&& wbc
->nr_to_write
> 0))
2839 * set the writeback_index so that range_cyclic
2840 * mode will write it back later
2842 mapping
->writeback_index
= index
;
2845 if (!no_nrwrite_index_update
)
2846 wbc
->no_nrwrite_index_update
= 0;
2847 wbc
->nr_to_write
-= nr_to_writebump
;
2848 trace_mark(ext4_da_writepage_result
,
2849 "dev %s ino %lu ret %d pages_written %d "
2850 "pages_skipped %ld congestion %d "
2851 "more_io %d no_nrwrite_index_update %d",
2852 inode
->i_sb
->s_id
, inode
->i_ino
, ret
,
2853 pages_written
, wbc
->pages_skipped
,
2854 wbc
->encountered_congestion
, wbc
->more_io
,
2855 wbc
->no_nrwrite_index_update
);
2859 #define FALL_BACK_TO_NONDELALLOC 1
2860 static int ext4_nonda_switch(struct super_block
*sb
)
2862 s64 free_blocks
, dirty_blocks
;
2863 struct ext4_sb_info
*sbi
= EXT4_SB(sb
);
2866 * switch to non delalloc mode if we are running low
2867 * on free block. The free block accounting via percpu
2868 * counters can get slightly wrong with percpu_counter_batch getting
2869 * accumulated on each CPU without updating global counters
2870 * Delalloc need an accurate free block accounting. So switch
2871 * to non delalloc when we are near to error range.
2873 free_blocks
= percpu_counter_read_positive(&sbi
->s_freeblocks_counter
);
2874 dirty_blocks
= percpu_counter_read_positive(&sbi
->s_dirtyblocks_counter
);
2875 if (2 * free_blocks
< 3 * dirty_blocks
||
2876 free_blocks
< (dirty_blocks
+ EXT4_FREEBLOCKS_WATERMARK
)) {
2878 * free block count is less that 150% of dirty blocks
2879 * or free blocks is less that watermark
2886 static int ext4_da_write_begin(struct file
*file
, struct address_space
*mapping
,
2887 loff_t pos
, unsigned len
, unsigned flags
,
2888 struct page
**pagep
, void **fsdata
)
2890 int ret
, retries
= 0;
2894 struct inode
*inode
= mapping
->host
;
2897 index
= pos
>> PAGE_CACHE_SHIFT
;
2898 from
= pos
& (PAGE_CACHE_SIZE
- 1);
2901 if (ext4_nonda_switch(inode
->i_sb
)) {
2902 *fsdata
= (void *)FALL_BACK_TO_NONDELALLOC
;
2903 return ext4_write_begin(file
, mapping
, pos
,
2904 len
, flags
, pagep
, fsdata
);
2906 *fsdata
= (void *)0;
2908 trace_mark(ext4_da_write_begin
,
2909 "dev %s ino %lu pos %llu len %u flags %u",
2910 inode
->i_sb
->s_id
, inode
->i_ino
,
2911 (unsigned long long) pos
, len
, flags
);
2914 * With delayed allocation, we don't log the i_disksize update
2915 * if there is delayed block allocation. But we still need
2916 * to journalling the i_disksize update if writes to the end
2917 * of file which has an already mapped buffer.
2919 handle
= ext4_journal_start(inode
, 1);
2920 if (IS_ERR(handle
)) {
2921 ret
= PTR_ERR(handle
);
2924 /* We cannot recurse into the filesystem as the transaction is already
2926 flags
|= AOP_FLAG_NOFS
;
2928 page
= grab_cache_page_write_begin(mapping
, index
, flags
);
2930 ext4_journal_stop(handle
);
2936 ret
= block_write_begin(file
, mapping
, pos
, len
, flags
, pagep
, fsdata
,
2937 ext4_da_get_block_prep
);
2940 ext4_journal_stop(handle
);
2941 page_cache_release(page
);
2943 * block_write_begin may have instantiated a few blocks
2944 * outside i_size. Trim these off again. Don't need
2945 * i_size_read because we hold i_mutex.
2947 if (pos
+ len
> inode
->i_size
)
2948 vmtruncate(inode
, inode
->i_size
);
2951 if (ret
== -ENOSPC
&& ext4_should_retry_alloc(inode
->i_sb
, &retries
))
2958 * Check if we should update i_disksize
2959 * when write to the end of file but not require block allocation
2961 static int ext4_da_should_update_i_disksize(struct page
*page
,
2962 unsigned long offset
)
2964 struct buffer_head
*bh
;
2965 struct inode
*inode
= page
->mapping
->host
;
2969 bh
= page_buffers(page
);
2970 idx
= offset
>> inode
->i_blkbits
;
2972 for (i
= 0; i
< idx
; i
++)
2973 bh
= bh
->b_this_page
;
2975 if (!buffer_mapped(bh
) || (buffer_delay(bh
)) || buffer_unwritten(bh
))
2980 static int ext4_da_write_end(struct file
*file
,
2981 struct address_space
*mapping
,
2982 loff_t pos
, unsigned len
, unsigned copied
,
2983 struct page
*page
, void *fsdata
)
2985 struct inode
*inode
= mapping
->host
;
2987 handle_t
*handle
= ext4_journal_current_handle();
2989 unsigned long start
, end
;
2990 int write_mode
= (int)(unsigned long)fsdata
;
2992 if (write_mode
== FALL_BACK_TO_NONDELALLOC
) {
2993 if (ext4_should_order_data(inode
)) {
2994 return ext4_ordered_write_end(file
, mapping
, pos
,
2995 len
, copied
, page
, fsdata
);
2996 } else if (ext4_should_writeback_data(inode
)) {
2997 return ext4_writeback_write_end(file
, mapping
, pos
,
2998 len
, copied
, page
, fsdata
);
3004 trace_mark(ext4_da_write_end
,
3005 "dev %s ino %lu pos %llu len %u copied %u",
3006 inode
->i_sb
->s_id
, inode
->i_ino
,
3007 (unsigned long long) pos
, len
, copied
);
3008 start
= pos
& (PAGE_CACHE_SIZE
- 1);
3009 end
= start
+ copied
- 1;
3012 * generic_write_end() will run mark_inode_dirty() if i_size
3013 * changes. So let's piggyback the i_disksize mark_inode_dirty
3017 new_i_size
= pos
+ copied
;
3018 if (new_i_size
> EXT4_I(inode
)->i_disksize
) {
3019 if (ext4_da_should_update_i_disksize(page
, end
)) {
3020 down_write(&EXT4_I(inode
)->i_data_sem
);
3021 if (new_i_size
> EXT4_I(inode
)->i_disksize
) {
3023 * Updating i_disksize when extending file
3024 * without needing block allocation
3026 if (ext4_should_order_data(inode
))
3027 ret
= ext4_jbd2_file_inode(handle
,
3030 EXT4_I(inode
)->i_disksize
= new_i_size
;
3032 up_write(&EXT4_I(inode
)->i_data_sem
);
3033 /* We need to mark inode dirty even if
3034 * new_i_size is less that inode->i_size
3035 * bu greater than i_disksize.(hint delalloc)
3037 ext4_mark_inode_dirty(handle
, inode
);
3040 ret2
= generic_write_end(file
, mapping
, pos
, len
, copied
,
3045 ret2
= ext4_journal_stop(handle
);
3049 return ret
? ret
: copied
;
3052 static void ext4_da_invalidatepage(struct page
*page
, unsigned long offset
)
3055 * Drop reserved blocks
3057 BUG_ON(!PageLocked(page
));
3058 if (!page_has_buffers(page
))
3061 ext4_da_page_release_reservation(page
, offset
);
3064 ext4_invalidatepage(page
, offset
);
3070 * Force all delayed allocation blocks to be allocated for a given inode.
3072 int ext4_alloc_da_blocks(struct inode
*inode
)
3074 if (!EXT4_I(inode
)->i_reserved_data_blocks
&&
3075 !EXT4_I(inode
)->i_reserved_meta_blocks
)
3079 * We do something simple for now. The filemap_flush() will
3080 * also start triggering a write of the data blocks, which is
3081 * not strictly speaking necessary (and for users of
3082 * laptop_mode, not even desirable). However, to do otherwise
3083 * would require replicating code paths in:
3085 * ext4_da_writepages() ->
3086 * write_cache_pages() ---> (via passed in callback function)
3087 * __mpage_da_writepage() -->
3088 * mpage_add_bh_to_extent()
3089 * mpage_da_map_blocks()
3091 * The problem is that write_cache_pages(), located in
3092 * mm/page-writeback.c, marks pages clean in preparation for
3093 * doing I/O, which is not desirable if we're not planning on
3096 * We could call write_cache_pages(), and then redirty all of
3097 * the pages by calling redirty_page_for_writeback() but that
3098 * would be ugly in the extreme. So instead we would need to
3099 * replicate parts of the code in the above functions,
3100 * simplifying them becuase we wouldn't actually intend to
3101 * write out the pages, but rather only collect contiguous
3102 * logical block extents, call the multi-block allocator, and
3103 * then update the buffer heads with the block allocations.
3105 * For now, though, we'll cheat by calling filemap_flush(),
3106 * which will map the blocks, and start the I/O, but not
3107 * actually wait for the I/O to complete.
3109 return filemap_flush(inode
->i_mapping
);
3113 * bmap() is special. It gets used by applications such as lilo and by
3114 * the swapper to find the on-disk block of a specific piece of data.
3116 * Naturally, this is dangerous if the block concerned is still in the
3117 * journal. If somebody makes a swapfile on an ext4 data-journaling
3118 * filesystem and enables swap, then they may get a nasty shock when the
3119 * data getting swapped to that swapfile suddenly gets overwritten by
3120 * the original zero's written out previously to the journal and
3121 * awaiting writeback in the kernel's buffer cache.
3123 * So, if we see any bmap calls here on a modified, data-journaled file,
3124 * take extra steps to flush any blocks which might be in the cache.
3126 static sector_t
ext4_bmap(struct address_space
*mapping
, sector_t block
)
3128 struct inode
*inode
= mapping
->host
;
3132 if (mapping_tagged(mapping
, PAGECACHE_TAG_DIRTY
) &&
3133 test_opt(inode
->i_sb
, DELALLOC
)) {
3135 * With delalloc we want to sync the file
3136 * so that we can make sure we allocate
3139 filemap_write_and_wait(mapping
);
3142 if (EXT4_JOURNAL(inode
) && EXT4_I(inode
)->i_state
& EXT4_STATE_JDATA
) {
3144 * This is a REALLY heavyweight approach, but the use of
3145 * bmap on dirty files is expected to be extremely rare:
3146 * only if we run lilo or swapon on a freshly made file
3147 * do we expect this to happen.
3149 * (bmap requires CAP_SYS_RAWIO so this does not
3150 * represent an unprivileged user DOS attack --- we'd be
3151 * in trouble if mortal users could trigger this path at
3154 * NB. EXT4_STATE_JDATA is not set on files other than
3155 * regular files. If somebody wants to bmap a directory
3156 * or symlink and gets confused because the buffer
3157 * hasn't yet been flushed to disk, they deserve
3158 * everything they get.
3161 EXT4_I(inode
)->i_state
&= ~EXT4_STATE_JDATA
;
3162 journal
= EXT4_JOURNAL(inode
);
3163 jbd2_journal_lock_updates(journal
);
3164 err
= jbd2_journal_flush(journal
);
3165 jbd2_journal_unlock_updates(journal
);
3171 return generic_block_bmap(mapping
, block
, ext4_get_block
);
3174 static int bget_one(handle_t
*handle
, struct buffer_head
*bh
)
3180 static int bput_one(handle_t
*handle
, struct buffer_head
*bh
)
3187 * Note that we don't need to start a transaction unless we're journaling data
3188 * because we should have holes filled from ext4_page_mkwrite(). We even don't
3189 * need to file the inode to the transaction's list in ordered mode because if
3190 * we are writing back data added by write(), the inode is already there and if
3191 * we are writing back data modified via mmap(), noone guarantees in which
3192 * transaction the data will hit the disk. In case we are journaling data, we
3193 * cannot start transaction directly because transaction start ranks above page
3194 * lock so we have to do some magic.
3196 * In all journaling modes block_write_full_page() will start the I/O.
3200 * ext4_writepage() -> kmalloc() -> __alloc_pages() -> page_launder() ->
3205 * ext4_file_write() -> generic_file_write() -> __alloc_pages() -> ...
3207 * Same applies to ext4_get_block(). We will deadlock on various things like
3208 * lock_journal and i_data_sem
3210 * Setting PF_MEMALLOC here doesn't work - too many internal memory
3213 * 16May01: If we're reentered then journal_current_handle() will be
3214 * non-zero. We simply *return*.
3216 * 1 July 2001: @@@ FIXME:
3217 * In journalled data mode, a data buffer may be metadata against the
3218 * current transaction. But the same file is part of a shared mapping
3219 * and someone does a writepage() on it.
3221 * We will move the buffer onto the async_data list, but *after* it has
3222 * been dirtied. So there's a small window where we have dirty data on
3225 * Note that this only applies to the last partial page in the file. The
3226 * bit which block_write_full_page() uses prepare/commit for. (That's
3227 * broken code anyway: it's wrong for msync()).
3229 * It's a rare case: affects the final partial page, for journalled data
3230 * where the file is subject to bith write() and writepage() in the same
3231 * transction. To fix it we'll need a custom block_write_full_page().
3232 * We'll probably need that anyway for journalling writepage() output.
3234 * We don't honour synchronous mounts for writepage(). That would be
3235 * disastrous. Any write() or metadata operation will sync the fs for
3239 static int __ext4_normal_writepage(struct page
*page
,
3240 struct writeback_control
*wbc
)
3242 struct inode
*inode
= page
->mapping
->host
;
3244 if (test_opt(inode
->i_sb
, NOBH
))
3245 return nobh_writepage(page
, noalloc_get_block_write
, wbc
);
3247 return block_write_full_page(page
, noalloc_get_block_write
,
3251 static int ext4_normal_writepage(struct page
*page
,
3252 struct writeback_control
*wbc
)
3254 struct inode
*inode
= page
->mapping
->host
;
3255 loff_t size
= i_size_read(inode
);
3258 trace_mark(ext4_normal_writepage
,
3259 "dev %s ino %lu page_index %lu",
3260 inode
->i_sb
->s_id
, inode
->i_ino
, page
->index
);
3261 J_ASSERT(PageLocked(page
));
3262 if (page
->index
== size
>> PAGE_CACHE_SHIFT
)
3263 len
= size
& ~PAGE_CACHE_MASK
;
3265 len
= PAGE_CACHE_SIZE
;
3267 if (page_has_buffers(page
)) {
3268 /* if page has buffers it should all be mapped
3269 * and allocated. If there are not buffers attached
3270 * to the page we know the page is dirty but it lost
3271 * buffers. That means that at some moment in time
3272 * after write_begin() / write_end() has been called
3273 * all buffers have been clean and thus they must have been
3274 * written at least once. So they are all mapped and we can
3275 * happily proceed with mapping them and writing the page.
3277 BUG_ON(walk_page_buffers(NULL
, page_buffers(page
), 0, len
, NULL
,
3278 ext4_bh_unmapped_or_delay
));
3281 if (!ext4_journal_current_handle())
3282 return __ext4_normal_writepage(page
, wbc
);
3284 redirty_page_for_writepage(wbc
, page
);
3289 static int __ext4_journalled_writepage(struct page
*page
,
3290 struct writeback_control
*wbc
)
3292 struct address_space
*mapping
= page
->mapping
;
3293 struct inode
*inode
= mapping
->host
;
3294 struct buffer_head
*page_bufs
;
3295 handle_t
*handle
= NULL
;
3299 ret
= block_prepare_write(page
, 0, PAGE_CACHE_SIZE
,
3300 noalloc_get_block_write
);
3304 page_bufs
= page_buffers(page
);
3305 walk_page_buffers(handle
, page_bufs
, 0, PAGE_CACHE_SIZE
, NULL
,
3307 /* As soon as we unlock the page, it can go away, but we have
3308 * references to buffers so we are safe */
3311 handle
= ext4_journal_start(inode
, ext4_writepage_trans_blocks(inode
));
3312 if (IS_ERR(handle
)) {
3313 ret
= PTR_ERR(handle
);
3317 ret
= walk_page_buffers(handle
, page_bufs
, 0,
3318 PAGE_CACHE_SIZE
, NULL
, do_journal_get_write_access
);
3320 err
= walk_page_buffers(handle
, page_bufs
, 0,
3321 PAGE_CACHE_SIZE
, NULL
, write_end_fn
);
3324 err
= ext4_journal_stop(handle
);
3328 walk_page_buffers(handle
, page_bufs
, 0,
3329 PAGE_CACHE_SIZE
, NULL
, bput_one
);
3330 EXT4_I(inode
)->i_state
|= EXT4_STATE_JDATA
;
3339 static int ext4_journalled_writepage(struct page
*page
,
3340 struct writeback_control
*wbc
)
3342 struct inode
*inode
= page
->mapping
->host
;
3343 loff_t size
= i_size_read(inode
);
3346 trace_mark(ext4_journalled_writepage
,
3347 "dev %s ino %lu page_index %lu",
3348 inode
->i_sb
->s_id
, inode
->i_ino
, page
->index
);
3349 J_ASSERT(PageLocked(page
));
3350 if (page
->index
== size
>> PAGE_CACHE_SHIFT
)
3351 len
= size
& ~PAGE_CACHE_MASK
;
3353 len
= PAGE_CACHE_SIZE
;
3355 if (page_has_buffers(page
)) {
3356 /* if page has buffers it should all be mapped
3357 * and allocated. If there are not buffers attached
3358 * to the page we know the page is dirty but it lost
3359 * buffers. That means that at some moment in time
3360 * after write_begin() / write_end() has been called
3361 * all buffers have been clean and thus they must have been
3362 * written at least once. So they are all mapped and we can
3363 * happily proceed with mapping them and writing the page.
3365 BUG_ON(walk_page_buffers(NULL
, page_buffers(page
), 0, len
, NULL
,
3366 ext4_bh_unmapped_or_delay
));
3369 if (ext4_journal_current_handle())
3372 if (PageChecked(page
)) {
3374 * It's mmapped pagecache. Add buffers and journal it. There
3375 * doesn't seem much point in redirtying the page here.
3377 ClearPageChecked(page
);
3378 return __ext4_journalled_writepage(page
, wbc
);
3381 * It may be a page full of checkpoint-mode buffers. We don't
3382 * really know unless we go poke around in the buffer_heads.
3383 * But block_write_full_page will do the right thing.
3385 return block_write_full_page(page
, noalloc_get_block_write
,
3389 redirty_page_for_writepage(wbc
, page
);
3394 static int ext4_readpage(struct file
*file
, struct page
*page
)
3396 return mpage_readpage(page
, ext4_get_block
);
3400 ext4_readpages(struct file
*file
, struct address_space
*mapping
,
3401 struct list_head
*pages
, unsigned nr_pages
)
3403 return mpage_readpages(mapping
, pages
, nr_pages
, ext4_get_block
);
3406 static void ext4_invalidatepage(struct page
*page
, unsigned long offset
)
3408 journal_t
*journal
= EXT4_JOURNAL(page
->mapping
->host
);
3411 * If it's a full truncate we just forget about the pending dirtying
3414 ClearPageChecked(page
);
3417 jbd2_journal_invalidatepage(journal
, page
, offset
);
3419 block_invalidatepage(page
, offset
);
3422 static int ext4_releasepage(struct page
*page
, gfp_t wait
)
3424 journal_t
*journal
= EXT4_JOURNAL(page
->mapping
->host
);
3426 WARN_ON(PageChecked(page
));
3427 if (!page_has_buffers(page
))
3430 return jbd2_journal_try_to_free_buffers(journal
, page
, wait
);
3432 return try_to_free_buffers(page
);
3436 * If the O_DIRECT write will extend the file then add this inode to the
3437 * orphan list. So recovery will truncate it back to the original size
3438 * if the machine crashes during the write.
3440 * If the O_DIRECT write is intantiating holes inside i_size and the machine
3441 * crashes then stale disk data _may_ be exposed inside the file. But current
3442 * VFS code falls back into buffered path in that case so we are safe.
3444 static ssize_t
ext4_direct_IO(int rw
, struct kiocb
*iocb
,
3445 const struct iovec
*iov
, loff_t offset
,
3446 unsigned long nr_segs
)
3448 struct file
*file
= iocb
->ki_filp
;
3449 struct inode
*inode
= file
->f_mapping
->host
;
3450 struct ext4_inode_info
*ei
= EXT4_I(inode
);
3454 size_t count
= iov_length(iov
, nr_segs
);
3457 loff_t final_size
= offset
+ count
;
3459 if (final_size
> inode
->i_size
) {
3460 /* Credits for sb + inode write */
3461 handle
= ext4_journal_start(inode
, 2);
3462 if (IS_ERR(handle
)) {
3463 ret
= PTR_ERR(handle
);
3466 ret
= ext4_orphan_add(handle
, inode
);
3468 ext4_journal_stop(handle
);
3472 ei
->i_disksize
= inode
->i_size
;
3473 ext4_journal_stop(handle
);
3477 ret
= blockdev_direct_IO(rw
, iocb
, inode
, inode
->i_sb
->s_bdev
, iov
,
3479 ext4_get_block
, NULL
);
3484 /* Credits for sb + inode write */
3485 handle
= ext4_journal_start(inode
, 2);
3486 if (IS_ERR(handle
)) {
3487 /* This is really bad luck. We've written the data
3488 * but cannot extend i_size. Bail out and pretend
3489 * the write failed... */
3490 ret
= PTR_ERR(handle
);
3494 ext4_orphan_del(handle
, inode
);
3496 loff_t end
= offset
+ ret
;
3497 if (end
> inode
->i_size
) {
3498 ei
->i_disksize
= end
;
3499 i_size_write(inode
, end
);
3501 * We're going to return a positive `ret'
3502 * here due to non-zero-length I/O, so there's
3503 * no way of reporting error returns from
3504 * ext4_mark_inode_dirty() to userspace. So
3507 ext4_mark_inode_dirty(handle
, inode
);
3510 err
= ext4_journal_stop(handle
);
3519 * Pages can be marked dirty completely asynchronously from ext4's journalling
3520 * activity. By filemap_sync_pte(), try_to_unmap_one(), etc. We cannot do
3521 * much here because ->set_page_dirty is called under VFS locks. The page is
3522 * not necessarily locked.
3524 * We cannot just dirty the page and leave attached buffers clean, because the
3525 * buffers' dirty state is "definitive". We cannot just set the buffers dirty
3526 * or jbddirty because all the journalling code will explode.
3528 * So what we do is to mark the page "pending dirty" and next time writepage
3529 * is called, propagate that into the buffers appropriately.
3531 static int ext4_journalled_set_page_dirty(struct page
*page
)
3533 SetPageChecked(page
);
3534 return __set_page_dirty_nobuffers(page
);
3537 static const struct address_space_operations ext4_ordered_aops
= {
3538 .readpage
= ext4_readpage
,
3539 .readpages
= ext4_readpages
,
3540 .writepage
= ext4_normal_writepage
,
3541 .sync_page
= block_sync_page
,
3542 .write_begin
= ext4_write_begin
,
3543 .write_end
= ext4_ordered_write_end
,
3545 .invalidatepage
= ext4_invalidatepage
,
3546 .releasepage
= ext4_releasepage
,
3547 .direct_IO
= ext4_direct_IO
,
3548 .migratepage
= buffer_migrate_page
,
3549 .is_partially_uptodate
= block_is_partially_uptodate
,
3552 static const struct address_space_operations ext4_writeback_aops
= {
3553 .readpage
= ext4_readpage
,
3554 .readpages
= ext4_readpages
,
3555 .writepage
= ext4_normal_writepage
,
3556 .sync_page
= block_sync_page
,
3557 .write_begin
= ext4_write_begin
,
3558 .write_end
= ext4_writeback_write_end
,
3560 .invalidatepage
= ext4_invalidatepage
,
3561 .releasepage
= ext4_releasepage
,
3562 .direct_IO
= ext4_direct_IO
,
3563 .migratepage
= buffer_migrate_page
,
3564 .is_partially_uptodate
= block_is_partially_uptodate
,
3567 static const struct address_space_operations ext4_journalled_aops
= {
3568 .readpage
= ext4_readpage
,
3569 .readpages
= ext4_readpages
,
3570 .writepage
= ext4_journalled_writepage
,
3571 .sync_page
= block_sync_page
,
3572 .write_begin
= ext4_write_begin
,
3573 .write_end
= ext4_journalled_write_end
,
3574 .set_page_dirty
= ext4_journalled_set_page_dirty
,
3576 .invalidatepage
= ext4_invalidatepage
,
3577 .releasepage
= ext4_releasepage
,
3578 .is_partially_uptodate
= block_is_partially_uptodate
,
3581 static const struct address_space_operations ext4_da_aops
= {
3582 .readpage
= ext4_readpage
,
3583 .readpages
= ext4_readpages
,
3584 .writepage
= ext4_da_writepage
,
3585 .writepages
= ext4_da_writepages
,
3586 .sync_page
= block_sync_page
,
3587 .write_begin
= ext4_da_write_begin
,
3588 .write_end
= ext4_da_write_end
,
3590 .invalidatepage
= ext4_da_invalidatepage
,
3591 .releasepage
= ext4_releasepage
,
3592 .direct_IO
= ext4_direct_IO
,
3593 .migratepage
= buffer_migrate_page
,
3594 .is_partially_uptodate
= block_is_partially_uptodate
,
3597 void ext4_set_aops(struct inode
*inode
)
3599 if (ext4_should_order_data(inode
) &&
3600 test_opt(inode
->i_sb
, DELALLOC
))
3601 inode
->i_mapping
->a_ops
= &ext4_da_aops
;
3602 else if (ext4_should_order_data(inode
))
3603 inode
->i_mapping
->a_ops
= &ext4_ordered_aops
;
3604 else if (ext4_should_writeback_data(inode
) &&
3605 test_opt(inode
->i_sb
, DELALLOC
))
3606 inode
->i_mapping
->a_ops
= &ext4_da_aops
;
3607 else if (ext4_should_writeback_data(inode
))
3608 inode
->i_mapping
->a_ops
= &ext4_writeback_aops
;
3610 inode
->i_mapping
->a_ops
= &ext4_journalled_aops
;
3614 * ext4_block_truncate_page() zeroes out a mapping from file offset `from'
3615 * up to the end of the block which corresponds to `from'.
3616 * This required during truncate. We need to physically zero the tail end
3617 * of that block so it doesn't yield old data if the file is later grown.
3619 int ext4_block_truncate_page(handle_t
*handle
,
3620 struct address_space
*mapping
, loff_t from
)
3622 ext4_fsblk_t index
= from
>> PAGE_CACHE_SHIFT
;
3623 unsigned offset
= from
& (PAGE_CACHE_SIZE
-1);
3624 unsigned blocksize
, length
, pos
;
3626 struct inode
*inode
= mapping
->host
;
3627 struct buffer_head
*bh
;
3631 page
= grab_cache_page(mapping
, from
>> PAGE_CACHE_SHIFT
);
3635 blocksize
= inode
->i_sb
->s_blocksize
;
3636 length
= blocksize
- (offset
& (blocksize
- 1));
3637 iblock
= index
<< (PAGE_CACHE_SHIFT
- inode
->i_sb
->s_blocksize_bits
);
3640 * For "nobh" option, we can only work if we don't need to
3641 * read-in the page - otherwise we create buffers to do the IO.
3643 if (!page_has_buffers(page
) && test_opt(inode
->i_sb
, NOBH
) &&
3644 ext4_should_writeback_data(inode
) && PageUptodate(page
)) {
3645 zero_user(page
, offset
, length
);
3646 set_page_dirty(page
);
3650 if (!page_has_buffers(page
))
3651 create_empty_buffers(page
, blocksize
, 0);
3653 /* Find the buffer that contains "offset" */
3654 bh
= page_buffers(page
);
3656 while (offset
>= pos
) {
3657 bh
= bh
->b_this_page
;
3663 if (buffer_freed(bh
)) {
3664 BUFFER_TRACE(bh
, "freed: skip");
3668 if (!buffer_mapped(bh
)) {
3669 BUFFER_TRACE(bh
, "unmapped");
3670 ext4_get_block(inode
, iblock
, bh
, 0);
3671 /* unmapped? It's a hole - nothing to do */
3672 if (!buffer_mapped(bh
)) {
3673 BUFFER_TRACE(bh
, "still unmapped");
3678 /* Ok, it's mapped. Make sure it's up-to-date */
3679 if (PageUptodate(page
))
3680 set_buffer_uptodate(bh
);
3682 if (!buffer_uptodate(bh
)) {
3684 ll_rw_block(READ
, 1, &bh
);
3686 /* Uhhuh. Read error. Complain and punt. */
3687 if (!buffer_uptodate(bh
))
3691 if (ext4_should_journal_data(inode
)) {
3692 BUFFER_TRACE(bh
, "get write access");
3693 err
= ext4_journal_get_write_access(handle
, bh
);
3698 zero_user(page
, offset
, length
);
3700 BUFFER_TRACE(bh
, "zeroed end of block");
3703 if (ext4_should_journal_data(inode
)) {
3704 err
= ext4_handle_dirty_metadata(handle
, inode
, bh
);
3706 if (ext4_should_order_data(inode
))
3707 err
= ext4_jbd2_file_inode(handle
, inode
);
3708 mark_buffer_dirty(bh
);
3713 page_cache_release(page
);
3718 * Probably it should be a library function... search for first non-zero word
3719 * or memcmp with zero_page, whatever is better for particular architecture.
3722 static inline int all_zeroes(__le32
*p
, __le32
*q
)
3731 * ext4_find_shared - find the indirect blocks for partial truncation.
3732 * @inode: inode in question
3733 * @depth: depth of the affected branch
3734 * @offsets: offsets of pointers in that branch (see ext4_block_to_path)
3735 * @chain: place to store the pointers to partial indirect blocks
3736 * @top: place to the (detached) top of branch
3738 * This is a helper function used by ext4_truncate().
3740 * When we do truncate() we may have to clean the ends of several
3741 * indirect blocks but leave the blocks themselves alive. Block is
3742 * partially truncated if some data below the new i_size is refered
3743 * from it (and it is on the path to the first completely truncated
3744 * data block, indeed). We have to free the top of that path along
3745 * with everything to the right of the path. Since no allocation
3746 * past the truncation point is possible until ext4_truncate()
3747 * finishes, we may safely do the latter, but top of branch may
3748 * require special attention - pageout below the truncation point
3749 * might try to populate it.
3751 * We atomically detach the top of branch from the tree, store the
3752 * block number of its root in *@top, pointers to buffer_heads of
3753 * partially truncated blocks - in @chain[].bh and pointers to
3754 * their last elements that should not be removed - in
3755 * @chain[].p. Return value is the pointer to last filled element
3758 * The work left to caller to do the actual freeing of subtrees:
3759 * a) free the subtree starting from *@top
3760 * b) free the subtrees whose roots are stored in
3761 * (@chain[i].p+1 .. end of @chain[i].bh->b_data)
3762 * c) free the subtrees growing from the inode past the @chain[0].
3763 * (no partially truncated stuff there). */
3765 static Indirect
*ext4_find_shared(struct inode
*inode
, int depth
,
3766 ext4_lblk_t offsets
[4], Indirect chain
[4], __le32
*top
)
3768 Indirect
*partial
, *p
;
3772 /* Make k index the deepest non-null offest + 1 */
3773 for (k
= depth
; k
> 1 && !offsets
[k
-1]; k
--)
3775 partial
= ext4_get_branch(inode
, k
, offsets
, chain
, &err
);
3776 /* Writer: pointers */
3778 partial
= chain
+ k
-1;
3780 * If the branch acquired continuation since we've looked at it -
3781 * fine, it should all survive and (new) top doesn't belong to us.
3783 if (!partial
->key
&& *partial
->p
)
3786 for (p
= partial
; (p
> chain
) && all_zeroes((__le32
*) p
->bh
->b_data
, p
->p
); p
--)
3789 * OK, we've found the last block that must survive. The rest of our
3790 * branch should be detached before unlocking. However, if that rest
3791 * of branch is all ours and does not grow immediately from the inode
3792 * it's easier to cheat and just decrement partial->p.
3794 if (p
== chain
+ k
- 1 && p
> chain
) {
3798 /* Nope, don't do this in ext4. Must leave the tree intact */
3805 while (partial
> p
) {
3806 brelse(partial
->bh
);
3814 * Zero a number of block pointers in either an inode or an indirect block.
3815 * If we restart the transaction we must again get write access to the
3816 * indirect block for further modification.
3818 * We release `count' blocks on disk, but (last - first) may be greater
3819 * than `count' because there can be holes in there.
3821 static void ext4_clear_blocks(handle_t
*handle
, struct inode
*inode
,
3822 struct buffer_head
*bh
, ext4_fsblk_t block_to_free
,
3823 unsigned long count
, __le32
*first
, __le32
*last
)
3826 if (try_to_extend_transaction(handle
, inode
)) {
3828 BUFFER_TRACE(bh
, "call ext4_handle_dirty_metadata");
3829 ext4_handle_dirty_metadata(handle
, inode
, bh
);
3831 ext4_mark_inode_dirty(handle
, inode
);
3832 ext4_journal_test_restart(handle
, inode
);
3834 BUFFER_TRACE(bh
, "retaking write access");
3835 ext4_journal_get_write_access(handle
, bh
);
3840 * Any buffers which are on the journal will be in memory. We find
3841 * them on the hash table so jbd2_journal_revoke() will run jbd2_journal_forget()
3842 * on them. We've already detached each block from the file, so
3843 * bforget() in jbd2_journal_forget() should be safe.
3845 * AKPM: turn on bforget in jbd2_journal_forget()!!!
3847 for (p
= first
; p
< last
; p
++) {
3848 u32 nr
= le32_to_cpu(*p
);
3850 struct buffer_head
*tbh
;
3853 tbh
= sb_find_get_block(inode
->i_sb
, nr
);
3854 ext4_forget(handle
, 0, inode
, tbh
, nr
);
3858 ext4_free_blocks(handle
, inode
, block_to_free
, count
, 0);
3862 * ext4_free_data - free a list of data blocks
3863 * @handle: handle for this transaction
3864 * @inode: inode we are dealing with
3865 * @this_bh: indirect buffer_head which contains *@first and *@last
3866 * @first: array of block numbers
3867 * @last: points immediately past the end of array
3869 * We are freeing all blocks refered from that array (numbers are stored as
3870 * little-endian 32-bit) and updating @inode->i_blocks appropriately.
3872 * We accumulate contiguous runs of blocks to free. Conveniently, if these
3873 * blocks are contiguous then releasing them at one time will only affect one
3874 * or two bitmap blocks (+ group descriptor(s) and superblock) and we won't
3875 * actually use a lot of journal space.
3877 * @this_bh will be %NULL if @first and @last point into the inode's direct
3880 static void ext4_free_data(handle_t
*handle
, struct inode
*inode
,
3881 struct buffer_head
*this_bh
,
3882 __le32
*first
, __le32
*last
)
3884 ext4_fsblk_t block_to_free
= 0; /* Starting block # of a run */
3885 unsigned long count
= 0; /* Number of blocks in the run */
3886 __le32
*block_to_free_p
= NULL
; /* Pointer into inode/ind
3889 ext4_fsblk_t nr
; /* Current block # */
3890 __le32
*p
; /* Pointer into inode/ind
3891 for current block */
3894 if (this_bh
) { /* For indirect block */
3895 BUFFER_TRACE(this_bh
, "get_write_access");
3896 err
= ext4_journal_get_write_access(handle
, this_bh
);
3897 /* Important: if we can't update the indirect pointers
3898 * to the blocks, we can't free them. */
3903 for (p
= first
; p
< last
; p
++) {
3904 nr
= le32_to_cpu(*p
);
3906 /* accumulate blocks to free if they're contiguous */
3909 block_to_free_p
= p
;
3911 } else if (nr
== block_to_free
+ count
) {
3914 ext4_clear_blocks(handle
, inode
, this_bh
,
3916 count
, block_to_free_p
, p
);
3918 block_to_free_p
= p
;
3925 ext4_clear_blocks(handle
, inode
, this_bh
, block_to_free
,
3926 count
, block_to_free_p
, p
);
3929 BUFFER_TRACE(this_bh
, "call ext4_handle_dirty_metadata");
3932 * The buffer head should have an attached journal head at this
3933 * point. However, if the data is corrupted and an indirect
3934 * block pointed to itself, it would have been detached when
3935 * the block was cleared. Check for this instead of OOPSing.
3937 if ((EXT4_JOURNAL(inode
) == NULL
) || bh2jh(this_bh
))
3938 ext4_handle_dirty_metadata(handle
, inode
, this_bh
);
3940 ext4_error(inode
->i_sb
, __func__
,
3941 "circular indirect block detected, "
3942 "inode=%lu, block=%llu",
3944 (unsigned long long) this_bh
->b_blocknr
);
3949 * ext4_free_branches - free an array of branches
3950 * @handle: JBD handle for this transaction
3951 * @inode: inode we are dealing with
3952 * @parent_bh: the buffer_head which contains *@first and *@last
3953 * @first: array of block numbers
3954 * @last: pointer immediately past the end of array
3955 * @depth: depth of the branches to free
3957 * We are freeing all blocks refered from these branches (numbers are
3958 * stored as little-endian 32-bit) and updating @inode->i_blocks
3961 static void ext4_free_branches(handle_t
*handle
, struct inode
*inode
,
3962 struct buffer_head
*parent_bh
,
3963 __le32
*first
, __le32
*last
, int depth
)
3968 if (ext4_handle_is_aborted(handle
))
3972 struct buffer_head
*bh
;
3973 int addr_per_block
= EXT4_ADDR_PER_BLOCK(inode
->i_sb
);
3975 while (--p
>= first
) {
3976 nr
= le32_to_cpu(*p
);
3978 continue; /* A hole */
3980 /* Go read the buffer for the next level down */
3981 bh
= sb_bread(inode
->i_sb
, nr
);
3984 * A read failure? Report error and clear slot
3988 ext4_error(inode
->i_sb
, "ext4_free_branches",
3989 "Read failure, inode=%lu, block=%llu",
3994 /* This zaps the entire block. Bottom up. */
3995 BUFFER_TRACE(bh
, "free child branches");
3996 ext4_free_branches(handle
, inode
, bh
,
3997 (__le32
*) bh
->b_data
,
3998 (__le32
*) bh
->b_data
+ addr_per_block
,
4002 * We've probably journalled the indirect block several
4003 * times during the truncate. But it's no longer
4004 * needed and we now drop it from the transaction via
4005 * jbd2_journal_revoke().
4007 * That's easy if it's exclusively part of this
4008 * transaction. But if it's part of the committing
4009 * transaction then jbd2_journal_forget() will simply
4010 * brelse() it. That means that if the underlying
4011 * block is reallocated in ext4_get_block(),
4012 * unmap_underlying_metadata() will find this block
4013 * and will try to get rid of it. damn, damn.
4015 * If this block has already been committed to the
4016 * journal, a revoke record will be written. And
4017 * revoke records must be emitted *before* clearing
4018 * this block's bit in the bitmaps.
4020 ext4_forget(handle
, 1, inode
, bh
, bh
->b_blocknr
);
4023 * Everything below this this pointer has been
4024 * released. Now let this top-of-subtree go.
4026 * We want the freeing of this indirect block to be
4027 * atomic in the journal with the updating of the
4028 * bitmap block which owns it. So make some room in
4031 * We zero the parent pointer *after* freeing its
4032 * pointee in the bitmaps, so if extend_transaction()
4033 * for some reason fails to put the bitmap changes and
4034 * the release into the same transaction, recovery
4035 * will merely complain about releasing a free block,
4036 * rather than leaking blocks.
4038 if (ext4_handle_is_aborted(handle
))
4040 if (try_to_extend_transaction(handle
, inode
)) {
4041 ext4_mark_inode_dirty(handle
, inode
);
4042 ext4_journal_test_restart(handle
, inode
);
4045 ext4_free_blocks(handle
, inode
, nr
, 1, 1);
4049 * The block which we have just freed is
4050 * pointed to by an indirect block: journal it
4052 BUFFER_TRACE(parent_bh
, "get_write_access");
4053 if (!ext4_journal_get_write_access(handle
,
4056 BUFFER_TRACE(parent_bh
,
4057 "call ext4_handle_dirty_metadata");
4058 ext4_handle_dirty_metadata(handle
,
4065 /* We have reached the bottom of the tree. */
4066 BUFFER_TRACE(parent_bh
, "free data blocks");
4067 ext4_free_data(handle
, inode
, parent_bh
, first
, last
);
4071 int ext4_can_truncate(struct inode
*inode
)
4073 if (IS_APPEND(inode
) || IS_IMMUTABLE(inode
))
4075 if (S_ISREG(inode
->i_mode
))
4077 if (S_ISDIR(inode
->i_mode
))
4079 if (S_ISLNK(inode
->i_mode
))
4080 return !ext4_inode_is_fast_symlink(inode
);
4087 * We block out ext4_get_block() block instantiations across the entire
4088 * transaction, and VFS/VM ensures that ext4_truncate() cannot run
4089 * simultaneously on behalf of the same inode.
4091 * As we work through the truncate and commmit bits of it to the journal there
4092 * is one core, guiding principle: the file's tree must always be consistent on
4093 * disk. We must be able to restart the truncate after a crash.
4095 * The file's tree may be transiently inconsistent in memory (although it
4096 * probably isn't), but whenever we close off and commit a journal transaction,
4097 * the contents of (the filesystem + the journal) must be consistent and
4098 * restartable. It's pretty simple, really: bottom up, right to left (although
4099 * left-to-right works OK too).
4101 * Note that at recovery time, journal replay occurs *before* the restart of
4102 * truncate against the orphan inode list.
4104 * The committed inode has the new, desired i_size (which is the same as
4105 * i_disksize in this case). After a crash, ext4_orphan_cleanup() will see
4106 * that this inode's truncate did not complete and it will again call
4107 * ext4_truncate() to have another go. So there will be instantiated blocks
4108 * to the right of the truncation point in a crashed ext4 filesystem. But
4109 * that's fine - as long as they are linked from the inode, the post-crash
4110 * ext4_truncate() run will find them and release them.
4112 void ext4_truncate(struct inode
*inode
)
4115 struct ext4_inode_info
*ei
= EXT4_I(inode
);
4116 __le32
*i_data
= ei
->i_data
;
4117 int addr_per_block
= EXT4_ADDR_PER_BLOCK(inode
->i_sb
);
4118 struct address_space
*mapping
= inode
->i_mapping
;
4119 ext4_lblk_t offsets
[4];
4124 ext4_lblk_t last_block
;
4125 unsigned blocksize
= inode
->i_sb
->s_blocksize
;
4127 if (!ext4_can_truncate(inode
))
4130 if (ei
->i_disksize
&& inode
->i_size
== 0 &&
4131 !test_opt(inode
->i_sb
, NO_AUTO_DA_ALLOC
))
4132 ei
->i_state
|= EXT4_STATE_DA_ALLOC_CLOSE
;
4134 if (EXT4_I(inode
)->i_flags
& EXT4_EXTENTS_FL
) {
4135 ext4_ext_truncate(inode
);
4139 handle
= start_transaction(inode
);
4141 return; /* AKPM: return what? */
4143 last_block
= (inode
->i_size
+ blocksize
-1)
4144 >> EXT4_BLOCK_SIZE_BITS(inode
->i_sb
);
4146 if (inode
->i_size
& (blocksize
- 1))
4147 if (ext4_block_truncate_page(handle
, mapping
, inode
->i_size
))
4150 n
= ext4_block_to_path(inode
, last_block
, offsets
, NULL
);
4152 goto out_stop
; /* error */
4155 * OK. This truncate is going to happen. We add the inode to the
4156 * orphan list, so that if this truncate spans multiple transactions,
4157 * and we crash, we will resume the truncate when the filesystem
4158 * recovers. It also marks the inode dirty, to catch the new size.
4160 * Implication: the file must always be in a sane, consistent
4161 * truncatable state while each transaction commits.
4163 if (ext4_orphan_add(handle
, inode
))
4167 * From here we block out all ext4_get_block() callers who want to
4168 * modify the block allocation tree.
4170 down_write(&ei
->i_data_sem
);
4172 ext4_discard_preallocations(inode
);
4175 * The orphan list entry will now protect us from any crash which
4176 * occurs before the truncate completes, so it is now safe to propagate
4177 * the new, shorter inode size (held for now in i_size) into the
4178 * on-disk inode. We do this via i_disksize, which is the value which
4179 * ext4 *really* writes onto the disk inode.
4181 ei
->i_disksize
= inode
->i_size
;
4183 if (n
== 1) { /* direct blocks */
4184 ext4_free_data(handle
, inode
, NULL
, i_data
+offsets
[0],
4185 i_data
+ EXT4_NDIR_BLOCKS
);
4189 partial
= ext4_find_shared(inode
, n
, offsets
, chain
, &nr
);
4190 /* Kill the top of shared branch (not detached) */
4192 if (partial
== chain
) {
4193 /* Shared branch grows from the inode */
4194 ext4_free_branches(handle
, inode
, NULL
,
4195 &nr
, &nr
+1, (chain
+n
-1) - partial
);
4198 * We mark the inode dirty prior to restart,
4199 * and prior to stop. No need for it here.
4202 /* Shared branch grows from an indirect block */
4203 BUFFER_TRACE(partial
->bh
, "get_write_access");
4204 ext4_free_branches(handle
, inode
, partial
->bh
,
4206 partial
->p
+1, (chain
+n
-1) - partial
);
4209 /* Clear the ends of indirect blocks on the shared branch */
4210 while (partial
> chain
) {
4211 ext4_free_branches(handle
, inode
, partial
->bh
, partial
->p
+ 1,
4212 (__le32
*)partial
->bh
->b_data
+addr_per_block
,
4213 (chain
+n
-1) - partial
);
4214 BUFFER_TRACE(partial
->bh
, "call brelse");
4215 brelse (partial
->bh
);
4219 /* Kill the remaining (whole) subtrees */
4220 switch (offsets
[0]) {
4222 nr
= i_data
[EXT4_IND_BLOCK
];
4224 ext4_free_branches(handle
, inode
, NULL
, &nr
, &nr
+1, 1);
4225 i_data
[EXT4_IND_BLOCK
] = 0;
4227 case EXT4_IND_BLOCK
:
4228 nr
= i_data
[EXT4_DIND_BLOCK
];
4230 ext4_free_branches(handle
, inode
, NULL
, &nr
, &nr
+1, 2);
4231 i_data
[EXT4_DIND_BLOCK
] = 0;
4233 case EXT4_DIND_BLOCK
:
4234 nr
= i_data
[EXT4_TIND_BLOCK
];
4236 ext4_free_branches(handle
, inode
, NULL
, &nr
, &nr
+1, 3);
4237 i_data
[EXT4_TIND_BLOCK
] = 0;
4239 case EXT4_TIND_BLOCK
:
4243 up_write(&ei
->i_data_sem
);
4244 inode
->i_mtime
= inode
->i_ctime
= ext4_current_time(inode
);
4245 ext4_mark_inode_dirty(handle
, inode
);
4248 * In a multi-transaction truncate, we only make the final transaction
4252 ext4_handle_sync(handle
);
4255 * If this was a simple ftruncate(), and the file will remain alive
4256 * then we need to clear up the orphan record which we created above.
4257 * However, if this was a real unlink then we were called by
4258 * ext4_delete_inode(), and we allow that function to clean up the
4259 * orphan info for us.
4262 ext4_orphan_del(handle
, inode
);
4264 ext4_journal_stop(handle
);
4268 * ext4_get_inode_loc returns with an extra refcount against the inode's
4269 * underlying buffer_head on success. If 'in_mem' is true, we have all
4270 * data in memory that is needed to recreate the on-disk version of this
4273 static int __ext4_get_inode_loc(struct inode
*inode
,
4274 struct ext4_iloc
*iloc
, int in_mem
)
4276 struct ext4_group_desc
*gdp
;
4277 struct buffer_head
*bh
;
4278 struct super_block
*sb
= inode
->i_sb
;
4280 int inodes_per_block
, inode_offset
;
4283 if (!ext4_valid_inum(sb
, inode
->i_ino
))
4286 iloc
->block_group
= (inode
->i_ino
- 1) / EXT4_INODES_PER_GROUP(sb
);
4287 gdp
= ext4_get_group_desc(sb
, iloc
->block_group
, NULL
);
4292 * Figure out the offset within the block group inode table
4294 inodes_per_block
= (EXT4_BLOCK_SIZE(sb
) / EXT4_INODE_SIZE(sb
));
4295 inode_offset
= ((inode
->i_ino
- 1) %
4296 EXT4_INODES_PER_GROUP(sb
));
4297 block
= ext4_inode_table(sb
, gdp
) + (inode_offset
/ inodes_per_block
);
4298 iloc
->offset
= (inode_offset
% inodes_per_block
) * EXT4_INODE_SIZE(sb
);
4300 bh
= sb_getblk(sb
, block
);
4302 ext4_error(sb
, "ext4_get_inode_loc", "unable to read "
4303 "inode block - inode=%lu, block=%llu",
4304 inode
->i_ino
, block
);
4307 if (!buffer_uptodate(bh
)) {
4311 * If the buffer has the write error flag, we have failed
4312 * to write out another inode in the same block. In this
4313 * case, we don't have to read the block because we may
4314 * read the old inode data successfully.
4316 if (buffer_write_io_error(bh
) && !buffer_uptodate(bh
))
4317 set_buffer_uptodate(bh
);
4319 if (buffer_uptodate(bh
)) {
4320 /* someone brought it uptodate while we waited */
4326 * If we have all information of the inode in memory and this
4327 * is the only valid inode in the block, we need not read the
4331 struct buffer_head
*bitmap_bh
;
4334 start
= inode_offset
& ~(inodes_per_block
- 1);
4336 /* Is the inode bitmap in cache? */
4337 bitmap_bh
= sb_getblk(sb
, ext4_inode_bitmap(sb
, gdp
));
4342 * If the inode bitmap isn't in cache then the
4343 * optimisation may end up performing two reads instead
4344 * of one, so skip it.
4346 if (!buffer_uptodate(bitmap_bh
)) {
4350 for (i
= start
; i
< start
+ inodes_per_block
; i
++) {
4351 if (i
== inode_offset
)
4353 if (ext4_test_bit(i
, bitmap_bh
->b_data
))
4357 if (i
== start
+ inodes_per_block
) {
4358 /* all other inodes are free, so skip I/O */
4359 memset(bh
->b_data
, 0, bh
->b_size
);
4360 set_buffer_uptodate(bh
);
4368 * If we need to do any I/O, try to pre-readahead extra
4369 * blocks from the inode table.
4371 if (EXT4_SB(sb
)->s_inode_readahead_blks
) {
4372 ext4_fsblk_t b
, end
, table
;
4375 table
= ext4_inode_table(sb
, gdp
);
4376 /* s_inode_readahead_blks is always a power of 2 */
4377 b
= block
& ~(EXT4_SB(sb
)->s_inode_readahead_blks
-1);
4380 end
= b
+ EXT4_SB(sb
)->s_inode_readahead_blks
;
4381 num
= EXT4_INODES_PER_GROUP(sb
);
4382 if (EXT4_HAS_RO_COMPAT_FEATURE(sb
,
4383 EXT4_FEATURE_RO_COMPAT_GDT_CSUM
))
4384 num
-= ext4_itable_unused_count(sb
, gdp
);
4385 table
+= num
/ inodes_per_block
;
4389 sb_breadahead(sb
, b
++);
4393 * There are other valid inodes in the buffer, this inode
4394 * has in-inode xattrs, or we don't have this inode in memory.
4395 * Read the block from disk.
4398 bh
->b_end_io
= end_buffer_read_sync
;
4399 submit_bh(READ_META
, bh
);
4401 if (!buffer_uptodate(bh
)) {
4402 ext4_error(sb
, __func__
,
4403 "unable to read inode block - inode=%lu, "
4404 "block=%llu", inode
->i_ino
, block
);
4414 int ext4_get_inode_loc(struct inode
*inode
, struct ext4_iloc
*iloc
)
4416 /* We have all inode data except xattrs in memory here. */
4417 return __ext4_get_inode_loc(inode
, iloc
,
4418 !(EXT4_I(inode
)->i_state
& EXT4_STATE_XATTR
));
4421 void ext4_set_inode_flags(struct inode
*inode
)
4423 unsigned int flags
= EXT4_I(inode
)->i_flags
;
4425 inode
->i_flags
&= ~(S_SYNC
|S_APPEND
|S_IMMUTABLE
|S_NOATIME
|S_DIRSYNC
);
4426 if (flags
& EXT4_SYNC_FL
)
4427 inode
->i_flags
|= S_SYNC
;
4428 if (flags
& EXT4_APPEND_FL
)
4429 inode
->i_flags
|= S_APPEND
;
4430 if (flags
& EXT4_IMMUTABLE_FL
)
4431 inode
->i_flags
|= S_IMMUTABLE
;
4432 if (flags
& EXT4_NOATIME_FL
)
4433 inode
->i_flags
|= S_NOATIME
;
4434 if (flags
& EXT4_DIRSYNC_FL
)
4435 inode
->i_flags
|= S_DIRSYNC
;
4438 /* Propagate flags from i_flags to EXT4_I(inode)->i_flags */
4439 void ext4_get_inode_flags(struct ext4_inode_info
*ei
)
4441 unsigned int flags
= ei
->vfs_inode
.i_flags
;
4443 ei
->i_flags
&= ~(EXT4_SYNC_FL
|EXT4_APPEND_FL
|
4444 EXT4_IMMUTABLE_FL
|EXT4_NOATIME_FL
|EXT4_DIRSYNC_FL
);
4446 ei
->i_flags
|= EXT4_SYNC_FL
;
4447 if (flags
& S_APPEND
)
4448 ei
->i_flags
|= EXT4_APPEND_FL
;
4449 if (flags
& S_IMMUTABLE
)
4450 ei
->i_flags
|= EXT4_IMMUTABLE_FL
;
4451 if (flags
& S_NOATIME
)
4452 ei
->i_flags
|= EXT4_NOATIME_FL
;
4453 if (flags
& S_DIRSYNC
)
4454 ei
->i_flags
|= EXT4_DIRSYNC_FL
;
4456 static blkcnt_t
ext4_inode_blocks(struct ext4_inode
*raw_inode
,
4457 struct ext4_inode_info
*ei
)
4460 struct inode
*inode
= &(ei
->vfs_inode
);
4461 struct super_block
*sb
= inode
->i_sb
;
4463 if (EXT4_HAS_RO_COMPAT_FEATURE(sb
,
4464 EXT4_FEATURE_RO_COMPAT_HUGE_FILE
)) {
4465 /* we are using combined 48 bit field */
4466 i_blocks
= ((u64
)le16_to_cpu(raw_inode
->i_blocks_high
)) << 32 |
4467 le32_to_cpu(raw_inode
->i_blocks_lo
);
4468 if (ei
->i_flags
& EXT4_HUGE_FILE_FL
) {
4469 /* i_blocks represent file system block size */
4470 return i_blocks
<< (inode
->i_blkbits
- 9);
4475 return le32_to_cpu(raw_inode
->i_blocks_lo
);
4479 struct inode
*ext4_iget(struct super_block
*sb
, unsigned long ino
)
4481 struct ext4_iloc iloc
;
4482 struct ext4_inode
*raw_inode
;
4483 struct ext4_inode_info
*ei
;
4484 struct buffer_head
*bh
;
4485 struct inode
*inode
;
4489 inode
= iget_locked(sb
, ino
);
4491 return ERR_PTR(-ENOMEM
);
4492 if (!(inode
->i_state
& I_NEW
))
4496 #ifdef CONFIG_EXT4_FS_POSIX_ACL
4497 ei
->i_acl
= EXT4_ACL_NOT_CACHED
;
4498 ei
->i_default_acl
= EXT4_ACL_NOT_CACHED
;
4501 ret
= __ext4_get_inode_loc(inode
, &iloc
, 0);
4505 raw_inode
= ext4_raw_inode(&iloc
);
4506 inode
->i_mode
= le16_to_cpu(raw_inode
->i_mode
);
4507 inode
->i_uid
= (uid_t
)le16_to_cpu(raw_inode
->i_uid_low
);
4508 inode
->i_gid
= (gid_t
)le16_to_cpu(raw_inode
->i_gid_low
);
4509 if (!(test_opt(inode
->i_sb
, NO_UID32
))) {
4510 inode
->i_uid
|= le16_to_cpu(raw_inode
->i_uid_high
) << 16;
4511 inode
->i_gid
|= le16_to_cpu(raw_inode
->i_gid_high
) << 16;
4513 inode
->i_nlink
= le16_to_cpu(raw_inode
->i_links_count
);
4516 ei
->i_dir_start_lookup
= 0;
4517 ei
->i_dtime
= le32_to_cpu(raw_inode
->i_dtime
);
4518 /* We now have enough fields to check if the inode was active or not.
4519 * This is needed because nfsd might try to access dead inodes
4520 * the test is that same one that e2fsck uses
4521 * NeilBrown 1999oct15
4523 if (inode
->i_nlink
== 0) {
4524 if (inode
->i_mode
== 0 ||
4525 !(EXT4_SB(inode
->i_sb
)->s_mount_state
& EXT4_ORPHAN_FS
)) {
4526 /* this inode is deleted */
4531 /* The only unlinked inodes we let through here have
4532 * valid i_mode and are being read by the orphan
4533 * recovery code: that's fine, we're about to complete
4534 * the process of deleting those. */
4536 ei
->i_flags
= le32_to_cpu(raw_inode
->i_flags
);
4537 inode
->i_blocks
= ext4_inode_blocks(raw_inode
, ei
);
4538 ei
->i_file_acl
= le32_to_cpu(raw_inode
->i_file_acl_lo
);
4539 if (EXT4_HAS_INCOMPAT_FEATURE(sb
, EXT4_FEATURE_INCOMPAT_64BIT
))
4541 ((__u64
)le16_to_cpu(raw_inode
->i_file_acl_high
)) << 32;
4542 inode
->i_size
= ext4_isize(raw_inode
);
4543 ei
->i_disksize
= inode
->i_size
;
4544 inode
->i_generation
= le32_to_cpu(raw_inode
->i_generation
);
4545 ei
->i_block_group
= iloc
.block_group
;
4546 ei
->i_last_alloc_group
= ~0;
4548 * NOTE! The in-memory inode i_data array is in little-endian order
4549 * even on big-endian machines: we do NOT byteswap the block numbers!
4551 for (block
= 0; block
< EXT4_N_BLOCKS
; block
++)
4552 ei
->i_data
[block
] = raw_inode
->i_block
[block
];
4553 INIT_LIST_HEAD(&ei
->i_orphan
);
4555 if (EXT4_INODE_SIZE(inode
->i_sb
) > EXT4_GOOD_OLD_INODE_SIZE
) {
4556 ei
->i_extra_isize
= le16_to_cpu(raw_inode
->i_extra_isize
);
4557 if (EXT4_GOOD_OLD_INODE_SIZE
+ ei
->i_extra_isize
>
4558 EXT4_INODE_SIZE(inode
->i_sb
)) {
4563 if (ei
->i_extra_isize
== 0) {
4564 /* The extra space is currently unused. Use it. */
4565 ei
->i_extra_isize
= sizeof(struct ext4_inode
) -
4566 EXT4_GOOD_OLD_INODE_SIZE
;
4568 __le32
*magic
= (void *)raw_inode
+
4569 EXT4_GOOD_OLD_INODE_SIZE
+
4571 if (*magic
== cpu_to_le32(EXT4_XATTR_MAGIC
))
4572 ei
->i_state
|= EXT4_STATE_XATTR
;
4575 ei
->i_extra_isize
= 0;
4577 EXT4_INODE_GET_XTIME(i_ctime
, inode
, raw_inode
);
4578 EXT4_INODE_GET_XTIME(i_mtime
, inode
, raw_inode
);
4579 EXT4_INODE_GET_XTIME(i_atime
, inode
, raw_inode
);
4580 EXT4_EINODE_GET_XTIME(i_crtime
, ei
, raw_inode
);
4582 inode
->i_version
= le32_to_cpu(raw_inode
->i_disk_version
);
4583 if (EXT4_INODE_SIZE(inode
->i_sb
) > EXT4_GOOD_OLD_INODE_SIZE
) {
4584 if (EXT4_FITS_IN_INODE(raw_inode
, ei
, i_version_hi
))
4586 (__u64
)(le32_to_cpu(raw_inode
->i_version_hi
)) << 32;
4590 if (ei
->i_file_acl
&&
4592 (le32_to_cpu(EXT4_SB(sb
)->s_es
->s_first_data_block
) +
4593 EXT4_SB(sb
)->s_gdb_count
)) ||
4594 (ei
->i_file_acl
>= ext4_blocks_count(EXT4_SB(sb
)->s_es
)))) {
4595 ext4_error(sb
, __func__
,
4596 "bad extended attribute block %llu in inode #%lu",
4597 ei
->i_file_acl
, inode
->i_ino
);
4600 } else if (ei
->i_flags
& EXT4_EXTENTS_FL
) {
4601 if (S_ISREG(inode
->i_mode
) || S_ISDIR(inode
->i_mode
) ||
4602 (S_ISLNK(inode
->i_mode
) &&
4603 !ext4_inode_is_fast_symlink(inode
)))
4604 /* Validate extent which is part of inode */
4605 ret
= ext4_ext_check_inode(inode
);
4606 } else if (S_ISREG(inode
->i_mode
) || S_ISDIR(inode
->i_mode
) ||
4607 (S_ISLNK(inode
->i_mode
) &&
4608 !ext4_inode_is_fast_symlink(inode
))) {
4609 /* Validate block references which are part of inode */
4610 ret
= ext4_check_inode_blockref(inode
);
4617 if (S_ISREG(inode
->i_mode
)) {
4618 inode
->i_op
= &ext4_file_inode_operations
;
4619 inode
->i_fop
= &ext4_file_operations
;
4620 ext4_set_aops(inode
);
4621 } else if (S_ISDIR(inode
->i_mode
)) {
4622 inode
->i_op
= &ext4_dir_inode_operations
;
4623 inode
->i_fop
= &ext4_dir_operations
;
4624 } else if (S_ISLNK(inode
->i_mode
)) {
4625 if (ext4_inode_is_fast_symlink(inode
)) {
4626 inode
->i_op
= &ext4_fast_symlink_inode_operations
;
4627 nd_terminate_link(ei
->i_data
, inode
->i_size
,
4628 sizeof(ei
->i_data
) - 1);
4630 inode
->i_op
= &ext4_symlink_inode_operations
;
4631 ext4_set_aops(inode
);
4633 } else if (S_ISCHR(inode
->i_mode
) || S_ISBLK(inode
->i_mode
) ||
4634 S_ISFIFO(inode
->i_mode
) || S_ISSOCK(inode
->i_mode
)) {
4635 inode
->i_op
= &ext4_special_inode_operations
;
4636 if (raw_inode
->i_block
[0])
4637 init_special_inode(inode
, inode
->i_mode
,
4638 old_decode_dev(le32_to_cpu(raw_inode
->i_block
[0])));
4640 init_special_inode(inode
, inode
->i_mode
,
4641 new_decode_dev(le32_to_cpu(raw_inode
->i_block
[1])));
4645 ext4_error(inode
->i_sb
, __func__
,
4646 "bogus i_mode (%o) for inode=%lu",
4647 inode
->i_mode
, inode
->i_ino
);
4651 ext4_set_inode_flags(inode
);
4652 unlock_new_inode(inode
);
4657 return ERR_PTR(ret
);
4660 static int ext4_inode_blocks_set(handle_t
*handle
,
4661 struct ext4_inode
*raw_inode
,
4662 struct ext4_inode_info
*ei
)
4664 struct inode
*inode
= &(ei
->vfs_inode
);
4665 u64 i_blocks
= inode
->i_blocks
;
4666 struct super_block
*sb
= inode
->i_sb
;
4668 if (i_blocks
<= ~0U) {
4670 * i_blocks can be represnted in a 32 bit variable
4671 * as multiple of 512 bytes
4673 raw_inode
->i_blocks_lo
= cpu_to_le32(i_blocks
);
4674 raw_inode
->i_blocks_high
= 0;
4675 ei
->i_flags
&= ~EXT4_HUGE_FILE_FL
;
4678 if (!EXT4_HAS_RO_COMPAT_FEATURE(sb
, EXT4_FEATURE_RO_COMPAT_HUGE_FILE
))
4681 if (i_blocks
<= 0xffffffffffffULL
) {
4683 * i_blocks can be represented in a 48 bit variable
4684 * as multiple of 512 bytes
4686 raw_inode
->i_blocks_lo
= cpu_to_le32(i_blocks
);
4687 raw_inode
->i_blocks_high
= cpu_to_le16(i_blocks
>> 32);
4688 ei
->i_flags
&= ~EXT4_HUGE_FILE_FL
;
4690 ei
->i_flags
|= EXT4_HUGE_FILE_FL
;
4691 /* i_block is stored in file system block size */
4692 i_blocks
= i_blocks
>> (inode
->i_blkbits
- 9);
4693 raw_inode
->i_blocks_lo
= cpu_to_le32(i_blocks
);
4694 raw_inode
->i_blocks_high
= cpu_to_le16(i_blocks
>> 32);
4700 * Post the struct inode info into an on-disk inode location in the
4701 * buffer-cache. This gobbles the caller's reference to the
4702 * buffer_head in the inode location struct.
4704 * The caller must have write access to iloc->bh.
4706 static int ext4_do_update_inode(handle_t
*handle
,
4707 struct inode
*inode
,
4708 struct ext4_iloc
*iloc
)
4710 struct ext4_inode
*raw_inode
= ext4_raw_inode(iloc
);
4711 struct ext4_inode_info
*ei
= EXT4_I(inode
);
4712 struct buffer_head
*bh
= iloc
->bh
;
4713 int err
= 0, rc
, block
;
4715 /* For fields not not tracking in the in-memory inode,
4716 * initialise them to zero for new inodes. */
4717 if (ei
->i_state
& EXT4_STATE_NEW
)
4718 memset(raw_inode
, 0, EXT4_SB(inode
->i_sb
)->s_inode_size
);
4720 ext4_get_inode_flags(ei
);
4721 raw_inode
->i_mode
= cpu_to_le16(inode
->i_mode
);
4722 if (!(test_opt(inode
->i_sb
, NO_UID32
))) {
4723 raw_inode
->i_uid_low
= cpu_to_le16(low_16_bits(inode
->i_uid
));
4724 raw_inode
->i_gid_low
= cpu_to_le16(low_16_bits(inode
->i_gid
));
4726 * Fix up interoperability with old kernels. Otherwise, old inodes get
4727 * re-used with the upper 16 bits of the uid/gid intact
4730 raw_inode
->i_uid_high
=
4731 cpu_to_le16(high_16_bits(inode
->i_uid
));
4732 raw_inode
->i_gid_high
=
4733 cpu_to_le16(high_16_bits(inode
->i_gid
));
4735 raw_inode
->i_uid_high
= 0;
4736 raw_inode
->i_gid_high
= 0;
4739 raw_inode
->i_uid_low
=
4740 cpu_to_le16(fs_high2lowuid(inode
->i_uid
));
4741 raw_inode
->i_gid_low
=
4742 cpu_to_le16(fs_high2lowgid(inode
->i_gid
));
4743 raw_inode
->i_uid_high
= 0;
4744 raw_inode
->i_gid_high
= 0;
4746 raw_inode
->i_links_count
= cpu_to_le16(inode
->i_nlink
);
4748 EXT4_INODE_SET_XTIME(i_ctime
, inode
, raw_inode
);
4749 EXT4_INODE_SET_XTIME(i_mtime
, inode
, raw_inode
);
4750 EXT4_INODE_SET_XTIME(i_atime
, inode
, raw_inode
);
4751 EXT4_EINODE_SET_XTIME(i_crtime
, ei
, raw_inode
);
4753 if (ext4_inode_blocks_set(handle
, raw_inode
, ei
))
4755 raw_inode
->i_dtime
= cpu_to_le32(ei
->i_dtime
);
4756 /* clear the migrate flag in the raw_inode */
4757 raw_inode
->i_flags
= cpu_to_le32(ei
->i_flags
& ~EXT4_EXT_MIGRATE
);
4758 if (EXT4_SB(inode
->i_sb
)->s_es
->s_creator_os
!=
4759 cpu_to_le32(EXT4_OS_HURD
))
4760 raw_inode
->i_file_acl_high
=
4761 cpu_to_le16(ei
->i_file_acl
>> 32);
4762 raw_inode
->i_file_acl_lo
= cpu_to_le32(ei
->i_file_acl
);
4763 ext4_isize_set(raw_inode
, ei
->i_disksize
);
4764 if (ei
->i_disksize
> 0x7fffffffULL
) {
4765 struct super_block
*sb
= inode
->i_sb
;
4766 if (!EXT4_HAS_RO_COMPAT_FEATURE(sb
,
4767 EXT4_FEATURE_RO_COMPAT_LARGE_FILE
) ||
4768 EXT4_SB(sb
)->s_es
->s_rev_level
==
4769 cpu_to_le32(EXT4_GOOD_OLD_REV
)) {
4770 /* If this is the first large file
4771 * created, add a flag to the superblock.
4773 err
= ext4_journal_get_write_access(handle
,
4774 EXT4_SB(sb
)->s_sbh
);
4777 ext4_update_dynamic_rev(sb
);
4778 EXT4_SET_RO_COMPAT_FEATURE(sb
,
4779 EXT4_FEATURE_RO_COMPAT_LARGE_FILE
);
4781 ext4_handle_sync(handle
);
4782 err
= ext4_handle_dirty_metadata(handle
, inode
,
4783 EXT4_SB(sb
)->s_sbh
);
4786 raw_inode
->i_generation
= cpu_to_le32(inode
->i_generation
);
4787 if (S_ISCHR(inode
->i_mode
) || S_ISBLK(inode
->i_mode
)) {
4788 if (old_valid_dev(inode
->i_rdev
)) {
4789 raw_inode
->i_block
[0] =
4790 cpu_to_le32(old_encode_dev(inode
->i_rdev
));
4791 raw_inode
->i_block
[1] = 0;
4793 raw_inode
->i_block
[0] = 0;
4794 raw_inode
->i_block
[1] =
4795 cpu_to_le32(new_encode_dev(inode
->i_rdev
));
4796 raw_inode
->i_block
[2] = 0;
4798 } else for (block
= 0; block
< EXT4_N_BLOCKS
; block
++)
4799 raw_inode
->i_block
[block
] = ei
->i_data
[block
];
4801 raw_inode
->i_disk_version
= cpu_to_le32(inode
->i_version
);
4802 if (ei
->i_extra_isize
) {
4803 if (EXT4_FITS_IN_INODE(raw_inode
, ei
, i_version_hi
))
4804 raw_inode
->i_version_hi
=
4805 cpu_to_le32(inode
->i_version
>> 32);
4806 raw_inode
->i_extra_isize
= cpu_to_le16(ei
->i_extra_isize
);
4809 BUFFER_TRACE(bh
, "call ext4_handle_dirty_metadata");
4810 rc
= ext4_handle_dirty_metadata(handle
, inode
, bh
);
4813 ei
->i_state
&= ~EXT4_STATE_NEW
;
4817 ext4_std_error(inode
->i_sb
, err
);
4822 * ext4_write_inode()
4824 * We are called from a few places:
4826 * - Within generic_file_write() for O_SYNC files.
4827 * Here, there will be no transaction running. We wait for any running
4828 * trasnaction to commit.
4830 * - Within sys_sync(), kupdate and such.
4831 * We wait on commit, if tol to.
4833 * - Within prune_icache() (PF_MEMALLOC == true)
4834 * Here we simply return. We can't afford to block kswapd on the
4837 * In all cases it is actually safe for us to return without doing anything,
4838 * because the inode has been copied into a raw inode buffer in
4839 * ext4_mark_inode_dirty(). This is a correctness thing for O_SYNC and for
4842 * Note that we are absolutely dependent upon all inode dirtiers doing the
4843 * right thing: they *must* call mark_inode_dirty() after dirtying info in
4844 * which we are interested.
4846 * It would be a bug for them to not do this. The code:
4848 * mark_inode_dirty(inode)
4850 * inode->i_size = expr;
4852 * is in error because a kswapd-driven write_inode() could occur while
4853 * `stuff()' is running, and the new i_size will be lost. Plus the inode
4854 * will no longer be on the superblock's dirty inode list.
4856 int ext4_write_inode(struct inode
*inode
, int wait
)
4858 if (current
->flags
& PF_MEMALLOC
)
4861 if (ext4_journal_current_handle()) {
4862 jbd_debug(1, "called recursively, non-PF_MEMALLOC!\n");
4870 return ext4_force_commit(inode
->i_sb
);
4876 * Called from notify_change.
4878 * We want to trap VFS attempts to truncate the file as soon as
4879 * possible. In particular, we want to make sure that when the VFS
4880 * shrinks i_size, we put the inode on the orphan list and modify
4881 * i_disksize immediately, so that during the subsequent flushing of
4882 * dirty pages and freeing of disk blocks, we can guarantee that any
4883 * commit will leave the blocks being flushed in an unused state on
4884 * disk. (On recovery, the inode will get truncated and the blocks will
4885 * be freed, so we have a strong guarantee that no future commit will
4886 * leave these blocks visible to the user.)
4888 * Another thing we have to assure is that if we are in ordered mode
4889 * and inode is still attached to the committing transaction, we must
4890 * we start writeout of all the dirty pages which are being truncated.
4891 * This way we are sure that all the data written in the previous
4892 * transaction are already on disk (truncate waits for pages under
4895 * Called with inode->i_mutex down.
4897 int ext4_setattr(struct dentry
*dentry
, struct iattr
*attr
)
4899 struct inode
*inode
= dentry
->d_inode
;
4901 const unsigned int ia_valid
= attr
->ia_valid
;
4903 error
= inode_change_ok(inode
, attr
);
4907 if ((ia_valid
& ATTR_UID
&& attr
->ia_uid
!= inode
->i_uid
) ||
4908 (ia_valid
& ATTR_GID
&& attr
->ia_gid
!= inode
->i_gid
)) {
4911 /* (user+group)*(old+new) structure, inode write (sb,
4912 * inode block, ? - but truncate inode update has it) */
4913 handle
= ext4_journal_start(inode
, 2*(EXT4_QUOTA_INIT_BLOCKS(inode
->i_sb
)+
4914 EXT4_QUOTA_DEL_BLOCKS(inode
->i_sb
))+3);
4915 if (IS_ERR(handle
)) {
4916 error
= PTR_ERR(handle
);
4919 error
= vfs_dq_transfer(inode
, attr
) ? -EDQUOT
: 0;
4921 ext4_journal_stop(handle
);
4924 /* Update corresponding info in inode so that everything is in
4925 * one transaction */
4926 if (attr
->ia_valid
& ATTR_UID
)
4927 inode
->i_uid
= attr
->ia_uid
;
4928 if (attr
->ia_valid
& ATTR_GID
)
4929 inode
->i_gid
= attr
->ia_gid
;
4930 error
= ext4_mark_inode_dirty(handle
, inode
);
4931 ext4_journal_stop(handle
);
4934 if (attr
->ia_valid
& ATTR_SIZE
) {
4935 if (!(EXT4_I(inode
)->i_flags
& EXT4_EXTENTS_FL
)) {
4936 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
4938 if (attr
->ia_size
> sbi
->s_bitmap_maxbytes
) {
4945 if (S_ISREG(inode
->i_mode
) &&
4946 attr
->ia_valid
& ATTR_SIZE
&& attr
->ia_size
< inode
->i_size
) {
4949 handle
= ext4_journal_start(inode
, 3);
4950 if (IS_ERR(handle
)) {
4951 error
= PTR_ERR(handle
);
4955 error
= ext4_orphan_add(handle
, inode
);
4956 EXT4_I(inode
)->i_disksize
= attr
->ia_size
;
4957 rc
= ext4_mark_inode_dirty(handle
, inode
);
4960 ext4_journal_stop(handle
);
4962 if (ext4_should_order_data(inode
)) {
4963 error
= ext4_begin_ordered_truncate(inode
,
4966 /* Do as much error cleanup as possible */
4967 handle
= ext4_journal_start(inode
, 3);
4968 if (IS_ERR(handle
)) {
4969 ext4_orphan_del(NULL
, inode
);
4972 ext4_orphan_del(handle
, inode
);
4973 ext4_journal_stop(handle
);
4979 rc
= inode_setattr(inode
, attr
);
4981 /* If inode_setattr's call to ext4_truncate failed to get a
4982 * transaction handle at all, we need to clean up the in-core
4983 * orphan list manually. */
4985 ext4_orphan_del(NULL
, inode
);
4987 if (!rc
&& (ia_valid
& ATTR_MODE
))
4988 rc
= ext4_acl_chmod(inode
);
4991 ext4_std_error(inode
->i_sb
, error
);
4997 int ext4_getattr(struct vfsmount
*mnt
, struct dentry
*dentry
,
5000 struct inode
*inode
;
5001 unsigned long delalloc_blocks
;
5003 inode
= dentry
->d_inode
;
5004 generic_fillattr(inode
, stat
);
5007 * We can't update i_blocks if the block allocation is delayed
5008 * otherwise in the case of system crash before the real block
5009 * allocation is done, we will have i_blocks inconsistent with
5010 * on-disk file blocks.
5011 * We always keep i_blocks updated together with real
5012 * allocation. But to not confuse with user, stat
5013 * will return the blocks that include the delayed allocation
5014 * blocks for this file.
5016 spin_lock(&EXT4_I(inode
)->i_block_reservation_lock
);
5017 delalloc_blocks
= EXT4_I(inode
)->i_reserved_data_blocks
;
5018 spin_unlock(&EXT4_I(inode
)->i_block_reservation_lock
);
5020 stat
->blocks
+= (delalloc_blocks
<< inode
->i_sb
->s_blocksize_bits
)>>9;
5024 static int ext4_indirect_trans_blocks(struct inode
*inode
, int nrblocks
,
5029 /* if nrblocks are contiguous */
5032 * With N contiguous data blocks, it need at most
5033 * N/EXT4_ADDR_PER_BLOCK(inode->i_sb) indirect blocks
5034 * 2 dindirect blocks
5037 indirects
= nrblocks
/ EXT4_ADDR_PER_BLOCK(inode
->i_sb
);
5038 return indirects
+ 3;
5041 * if nrblocks are not contiguous, worse case, each block touch
5042 * a indirect block, and each indirect block touch a double indirect
5043 * block, plus a triple indirect block
5045 indirects
= nrblocks
* 2 + 1;
5049 static int ext4_index_trans_blocks(struct inode
*inode
, int nrblocks
, int chunk
)
5051 if (!(EXT4_I(inode
)->i_flags
& EXT4_EXTENTS_FL
))
5052 return ext4_indirect_trans_blocks(inode
, nrblocks
, chunk
);
5053 return ext4_ext_index_trans_blocks(inode
, nrblocks
, chunk
);
5057 * Account for index blocks, block groups bitmaps and block group
5058 * descriptor blocks if modify datablocks and index blocks
5059 * worse case, the indexs blocks spread over different block groups
5061 * If datablocks are discontiguous, they are possible to spread over
5062 * different block groups too. If they are contiugous, with flexbg,
5063 * they could still across block group boundary.
5065 * Also account for superblock, inode, quota and xattr blocks
5067 int ext4_meta_trans_blocks(struct inode
*inode
, int nrblocks
, int chunk
)
5069 ext4_group_t groups
, ngroups
= ext4_get_groups_count(inode
->i_sb
);
5075 * How many index blocks need to touch to modify nrblocks?
5076 * The "Chunk" flag indicating whether the nrblocks is
5077 * physically contiguous on disk
5079 * For Direct IO and fallocate, they calls get_block to allocate
5080 * one single extent at a time, so they could set the "Chunk" flag
5082 idxblocks
= ext4_index_trans_blocks(inode
, nrblocks
, chunk
);
5087 * Now let's see how many group bitmaps and group descriptors need
5097 if (groups
> ngroups
)
5099 if (groups
> EXT4_SB(inode
->i_sb
)->s_gdb_count
)
5100 gdpblocks
= EXT4_SB(inode
->i_sb
)->s_gdb_count
;
5102 /* bitmaps and block group descriptor blocks */
5103 ret
+= groups
+ gdpblocks
;
5105 /* Blocks for super block, inode, quota and xattr blocks */
5106 ret
+= EXT4_META_TRANS_BLOCKS(inode
->i_sb
);
5112 * Calulate the total number of credits to reserve to fit
5113 * the modification of a single pages into a single transaction,
5114 * which may include multiple chunks of block allocations.
5116 * This could be called via ext4_write_begin()
5118 * We need to consider the worse case, when
5119 * one new block per extent.
5121 int ext4_writepage_trans_blocks(struct inode
*inode
)
5123 int bpp
= ext4_journal_blocks_per_page(inode
);
5126 ret
= ext4_meta_trans_blocks(inode
, bpp
, 0);
5128 /* Account for data blocks for journalled mode */
5129 if (ext4_should_journal_data(inode
))
5135 * Calculate the journal credits for a chunk of data modification.
5137 * This is called from DIO, fallocate or whoever calling
5138 * ext4_get_blocks() to map/allocate a chunk of contigous disk blocks.
5140 * journal buffers for data blocks are not included here, as DIO
5141 * and fallocate do no need to journal data buffers.
5143 int ext4_chunk_trans_blocks(struct inode
*inode
, int nrblocks
)
5145 return ext4_meta_trans_blocks(inode
, nrblocks
, 1);
5149 * The caller must have previously called ext4_reserve_inode_write().
5150 * Give this, we know that the caller already has write access to iloc->bh.
5152 int ext4_mark_iloc_dirty(handle_t
*handle
,
5153 struct inode
*inode
, struct ext4_iloc
*iloc
)
5157 if (test_opt(inode
->i_sb
, I_VERSION
))
5158 inode_inc_iversion(inode
);
5160 /* the do_update_inode consumes one bh->b_count */
5163 /* ext4_do_update_inode() does jbd2_journal_dirty_metadata */
5164 err
= ext4_do_update_inode(handle
, inode
, iloc
);
5170 * On success, We end up with an outstanding reference count against
5171 * iloc->bh. This _must_ be cleaned up later.
5175 ext4_reserve_inode_write(handle_t
*handle
, struct inode
*inode
,
5176 struct ext4_iloc
*iloc
)
5180 err
= ext4_get_inode_loc(inode
, iloc
);
5182 BUFFER_TRACE(iloc
->bh
, "get_write_access");
5183 err
= ext4_journal_get_write_access(handle
, iloc
->bh
);
5189 ext4_std_error(inode
->i_sb
, err
);
5194 * Expand an inode by new_extra_isize bytes.
5195 * Returns 0 on success or negative error number on failure.
5197 static int ext4_expand_extra_isize(struct inode
*inode
,
5198 unsigned int new_extra_isize
,
5199 struct ext4_iloc iloc
,
5202 struct ext4_inode
*raw_inode
;
5203 struct ext4_xattr_ibody_header
*header
;
5204 struct ext4_xattr_entry
*entry
;
5206 if (EXT4_I(inode
)->i_extra_isize
>= new_extra_isize
)
5209 raw_inode
= ext4_raw_inode(&iloc
);
5211 header
= IHDR(inode
, raw_inode
);
5212 entry
= IFIRST(header
);
5214 /* No extended attributes present */
5215 if (!(EXT4_I(inode
)->i_state
& EXT4_STATE_XATTR
) ||
5216 header
->h_magic
!= cpu_to_le32(EXT4_XATTR_MAGIC
)) {
5217 memset((void *)raw_inode
+ EXT4_GOOD_OLD_INODE_SIZE
, 0,
5219 EXT4_I(inode
)->i_extra_isize
= new_extra_isize
;
5223 /* try to expand with EAs present */
5224 return ext4_expand_extra_isize_ea(inode
, new_extra_isize
,
5229 * What we do here is to mark the in-core inode as clean with respect to inode
5230 * dirtiness (it may still be data-dirty).
5231 * This means that the in-core inode may be reaped by prune_icache
5232 * without having to perform any I/O. This is a very good thing,
5233 * because *any* task may call prune_icache - even ones which
5234 * have a transaction open against a different journal.
5236 * Is this cheating? Not really. Sure, we haven't written the
5237 * inode out, but prune_icache isn't a user-visible syncing function.
5238 * Whenever the user wants stuff synced (sys_sync, sys_msync, sys_fsync)
5239 * we start and wait on commits.
5241 * Is this efficient/effective? Well, we're being nice to the system
5242 * by cleaning up our inodes proactively so they can be reaped
5243 * without I/O. But we are potentially leaving up to five seconds'
5244 * worth of inodes floating about which prune_icache wants us to
5245 * write out. One way to fix that would be to get prune_icache()
5246 * to do a write_super() to free up some memory. It has the desired
5249 int ext4_mark_inode_dirty(handle_t
*handle
, struct inode
*inode
)
5251 struct ext4_iloc iloc
;
5252 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
5253 static unsigned int mnt_count
;
5257 err
= ext4_reserve_inode_write(handle
, inode
, &iloc
);
5258 if (ext4_handle_valid(handle
) &&
5259 EXT4_I(inode
)->i_extra_isize
< sbi
->s_want_extra_isize
&&
5260 !(EXT4_I(inode
)->i_state
& EXT4_STATE_NO_EXPAND
)) {
5262 * We need extra buffer credits since we may write into EA block
5263 * with this same handle. If journal_extend fails, then it will
5264 * only result in a minor loss of functionality for that inode.
5265 * If this is felt to be critical, then e2fsck should be run to
5266 * force a large enough s_min_extra_isize.
5268 if ((jbd2_journal_extend(handle
,
5269 EXT4_DATA_TRANS_BLOCKS(inode
->i_sb
))) == 0) {
5270 ret
= ext4_expand_extra_isize(inode
,
5271 sbi
->s_want_extra_isize
,
5274 EXT4_I(inode
)->i_state
|= EXT4_STATE_NO_EXPAND
;
5276 le16_to_cpu(sbi
->s_es
->s_mnt_count
)) {
5277 ext4_warning(inode
->i_sb
, __func__
,
5278 "Unable to expand inode %lu. Delete"
5279 " some EAs or run e2fsck.",
5282 le16_to_cpu(sbi
->s_es
->s_mnt_count
);
5288 err
= ext4_mark_iloc_dirty(handle
, inode
, &iloc
);
5293 * ext4_dirty_inode() is called from __mark_inode_dirty()
5295 * We're really interested in the case where a file is being extended.
5296 * i_size has been changed by generic_commit_write() and we thus need
5297 * to include the updated inode in the current transaction.
5299 * Also, vfs_dq_alloc_block() will always dirty the inode when blocks
5300 * are allocated to the file.
5302 * If the inode is marked synchronous, we don't honour that here - doing
5303 * so would cause a commit on atime updates, which we don't bother doing.
5304 * We handle synchronous inodes at the highest possible level.
5306 void ext4_dirty_inode(struct inode
*inode
)
5308 handle_t
*current_handle
= ext4_journal_current_handle();
5311 if (!ext4_handle_valid(current_handle
)) {
5312 ext4_mark_inode_dirty(current_handle
, inode
);
5316 handle
= ext4_journal_start(inode
, 2);
5319 if (current_handle
&&
5320 current_handle
->h_transaction
!= handle
->h_transaction
) {
5321 /* This task has a transaction open against a different fs */
5322 printk(KERN_EMERG
"%s: transactions do not match!\n",
5325 jbd_debug(5, "marking dirty. outer handle=%p\n",
5327 ext4_mark_inode_dirty(handle
, inode
);
5329 ext4_journal_stop(handle
);
5336 * Bind an inode's backing buffer_head into this transaction, to prevent
5337 * it from being flushed to disk early. Unlike
5338 * ext4_reserve_inode_write, this leaves behind no bh reference and
5339 * returns no iloc structure, so the caller needs to repeat the iloc
5340 * lookup to mark the inode dirty later.
5342 static int ext4_pin_inode(handle_t
*handle
, struct inode
*inode
)
5344 struct ext4_iloc iloc
;
5348 err
= ext4_get_inode_loc(inode
, &iloc
);
5350 BUFFER_TRACE(iloc
.bh
, "get_write_access");
5351 err
= jbd2_journal_get_write_access(handle
, iloc
.bh
);
5353 err
= ext4_handle_dirty_metadata(handle
,
5359 ext4_std_error(inode
->i_sb
, err
);
5364 int ext4_change_inode_journal_flag(struct inode
*inode
, int val
)
5371 * We have to be very careful here: changing a data block's
5372 * journaling status dynamically is dangerous. If we write a
5373 * data block to the journal, change the status and then delete
5374 * that block, we risk forgetting to revoke the old log record
5375 * from the journal and so a subsequent replay can corrupt data.
5376 * So, first we make sure that the journal is empty and that
5377 * nobody is changing anything.
5380 journal
= EXT4_JOURNAL(inode
);
5383 if (is_journal_aborted(journal
))
5386 jbd2_journal_lock_updates(journal
);
5387 jbd2_journal_flush(journal
);
5390 * OK, there are no updates running now, and all cached data is
5391 * synced to disk. We are now in a completely consistent state
5392 * which doesn't have anything in the journal, and we know that
5393 * no filesystem updates are running, so it is safe to modify
5394 * the inode's in-core data-journaling state flag now.
5398 EXT4_I(inode
)->i_flags
|= EXT4_JOURNAL_DATA_FL
;
5400 EXT4_I(inode
)->i_flags
&= ~EXT4_JOURNAL_DATA_FL
;
5401 ext4_set_aops(inode
);
5403 jbd2_journal_unlock_updates(journal
);
5405 /* Finally we can mark the inode as dirty. */
5407 handle
= ext4_journal_start(inode
, 1);
5409 return PTR_ERR(handle
);
5411 err
= ext4_mark_inode_dirty(handle
, inode
);
5412 ext4_handle_sync(handle
);
5413 ext4_journal_stop(handle
);
5414 ext4_std_error(inode
->i_sb
, err
);
5419 static int ext4_bh_unmapped(handle_t
*handle
, struct buffer_head
*bh
)
5421 return !buffer_mapped(bh
);
5424 int ext4_page_mkwrite(struct vm_area_struct
*vma
, struct vm_fault
*vmf
)
5426 struct page
*page
= vmf
->page
;
5431 struct file
*file
= vma
->vm_file
;
5432 struct inode
*inode
= file
->f_path
.dentry
->d_inode
;
5433 struct address_space
*mapping
= inode
->i_mapping
;
5436 * Get i_alloc_sem to stop truncates messing with the inode. We cannot
5437 * get i_mutex because we are already holding mmap_sem.
5439 down_read(&inode
->i_alloc_sem
);
5440 size
= i_size_read(inode
);
5441 if (page
->mapping
!= mapping
|| size
<= page_offset(page
)
5442 || !PageUptodate(page
)) {
5443 /* page got truncated from under us? */
5447 if (PageMappedToDisk(page
))
5450 if (page
->index
== size
>> PAGE_CACHE_SHIFT
)
5451 len
= size
& ~PAGE_CACHE_MASK
;
5453 len
= PAGE_CACHE_SIZE
;
5455 if (page_has_buffers(page
)) {
5456 /* return if we have all the buffers mapped */
5457 if (!walk_page_buffers(NULL
, page_buffers(page
), 0, len
, NULL
,
5462 * OK, we need to fill the hole... Do write_begin write_end
5463 * to do block allocation/reservation.We are not holding
5464 * inode.i__mutex here. That allow * parallel write_begin,
5465 * write_end call. lock_page prevent this from happening
5466 * on the same page though
5468 ret
= mapping
->a_ops
->write_begin(file
, mapping
, page_offset(page
),
5469 len
, AOP_FLAG_UNINTERRUPTIBLE
, &page
, &fsdata
);
5472 ret
= mapping
->a_ops
->write_end(file
, mapping
, page_offset(page
),
5473 len
, len
, page
, fsdata
);
5479 ret
= VM_FAULT_SIGBUS
;
5480 up_read(&inode
->i_alloc_sem
);