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;
936 struct ext4_inode_info
*ei
= EXT4_I(inode
);
938 ext4_fsblk_t first_block
= 0;
942 J_ASSERT(!(EXT4_I(inode
)->i_flags
& EXT4_EXTENTS_FL
));
943 J_ASSERT(handle
!= NULL
|| (flags
& EXT4_GET_BLOCKS_CREATE
) == 0);
944 depth
= ext4_block_to_path(inode
, iblock
, offsets
,
945 &blocks_to_boundary
);
950 partial
= ext4_get_branch(inode
, depth
, offsets
, chain
, &err
);
952 /* Simplest case - block found, no allocation needed */
954 first_block
= le32_to_cpu(chain
[depth
- 1].key
);
955 clear_buffer_new(bh_result
);
958 while (count
< maxblocks
&& count
<= blocks_to_boundary
) {
961 blk
= le32_to_cpu(*(chain
[depth
-1].p
+ count
));
963 if (blk
== first_block
+ count
)
971 /* Next simple case - plain lookup or failed read of indirect block */
972 if ((flags
& EXT4_GET_BLOCKS_CREATE
) == 0 || err
== -EIO
)
976 * Okay, we need to do block allocation.
978 goal
= ext4_find_goal(inode
, iblock
, partial
);
980 /* the number of blocks need to allocate for [d,t]indirect blocks */
981 indirect_blks
= (chain
+ depth
) - partial
- 1;
984 * Next look up the indirect map to count the totoal number of
985 * direct blocks to allocate for this branch.
987 count
= ext4_blks_to_allocate(partial
, indirect_blks
,
988 maxblocks
, blocks_to_boundary
);
990 * Block out ext4_truncate while we alter the tree
992 err
= ext4_alloc_branch(handle
, inode
, iblock
, indirect_blks
,
994 offsets
+ (partial
- chain
), partial
);
997 * The ext4_splice_branch call will free and forget any buffers
998 * on the new chain if there is a failure, but that risks using
999 * up transaction credits, especially for bitmaps where the
1000 * credits cannot be returned. Can we handle this somehow? We
1001 * may need to return -EAGAIN upwards in the worst case. --sct
1004 err
= ext4_splice_branch(handle
, inode
, iblock
,
1005 partial
, indirect_blks
, count
);
1007 * i_disksize growing is protected by i_data_sem. Don't forget to
1008 * protect it if you're about to implement concurrent
1009 * ext4_get_block() -bzzz
1011 if (!err
&& (flags
& EXT4_GET_BLOCKS_EXTEND_DISKSIZE
)) {
1012 disksize
= ((loff_t
) iblock
+ count
) << inode
->i_blkbits
;
1013 if (disksize
> i_size_read(inode
))
1014 disksize
= i_size_read(inode
);
1015 if (disksize
> ei
->i_disksize
)
1016 ei
->i_disksize
= disksize
;
1021 set_buffer_new(bh_result
);
1023 map_bh(bh_result
, inode
->i_sb
, le32_to_cpu(chain
[depth
-1].key
));
1024 if (count
> blocks_to_boundary
)
1025 set_buffer_boundary(bh_result
);
1027 /* Clean up and exit */
1028 partial
= chain
+ depth
- 1; /* the whole chain */
1030 while (partial
> chain
) {
1031 BUFFER_TRACE(partial
->bh
, "call brelse");
1032 brelse(partial
->bh
);
1035 BUFFER_TRACE(bh_result
, "returned");
1040 qsize_t
ext4_get_reserved_space(struct inode
*inode
)
1042 unsigned long long total
;
1044 spin_lock(&EXT4_I(inode
)->i_block_reservation_lock
);
1045 total
= EXT4_I(inode
)->i_reserved_data_blocks
+
1046 EXT4_I(inode
)->i_reserved_meta_blocks
;
1047 spin_unlock(&EXT4_I(inode
)->i_block_reservation_lock
);
1052 * Calculate the number of metadata blocks need to reserve
1053 * to allocate @blocks for non extent file based file
1055 static int ext4_indirect_calc_metadata_amount(struct inode
*inode
, int blocks
)
1057 int icap
= EXT4_ADDR_PER_BLOCK(inode
->i_sb
);
1058 int ind_blks
, dind_blks
, tind_blks
;
1060 /* number of new indirect blocks needed */
1061 ind_blks
= (blocks
+ icap
- 1) / icap
;
1063 dind_blks
= (ind_blks
+ icap
- 1) / icap
;
1067 return ind_blks
+ dind_blks
+ tind_blks
;
1071 * Calculate the number of metadata blocks need to reserve
1072 * to allocate given number of blocks
1074 static int ext4_calc_metadata_amount(struct inode
*inode
, int blocks
)
1079 if (EXT4_I(inode
)->i_flags
& EXT4_EXTENTS_FL
)
1080 return ext4_ext_calc_metadata_amount(inode
, blocks
);
1082 return ext4_indirect_calc_metadata_amount(inode
, blocks
);
1085 static void ext4_da_update_reserve_space(struct inode
*inode
, int used
)
1087 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
1088 int total
, mdb
, mdb_free
;
1090 spin_lock(&EXT4_I(inode
)->i_block_reservation_lock
);
1091 /* recalculate the number of metablocks still need to be reserved */
1092 total
= EXT4_I(inode
)->i_reserved_data_blocks
- used
;
1093 mdb
= ext4_calc_metadata_amount(inode
, total
);
1095 /* figure out how many metablocks to release */
1096 BUG_ON(mdb
> EXT4_I(inode
)->i_reserved_meta_blocks
);
1097 mdb_free
= EXT4_I(inode
)->i_reserved_meta_blocks
- mdb
;
1100 /* Account for allocated meta_blocks */
1101 mdb_free
-= EXT4_I(inode
)->i_allocated_meta_blocks
;
1103 /* update fs dirty blocks counter */
1104 percpu_counter_sub(&sbi
->s_dirtyblocks_counter
, mdb_free
);
1105 EXT4_I(inode
)->i_allocated_meta_blocks
= 0;
1106 EXT4_I(inode
)->i_reserved_meta_blocks
= mdb
;
1109 /* update per-inode reservations */
1110 BUG_ON(used
> EXT4_I(inode
)->i_reserved_data_blocks
);
1111 EXT4_I(inode
)->i_reserved_data_blocks
-= used
;
1112 spin_unlock(&EXT4_I(inode
)->i_block_reservation_lock
);
1115 * free those over-booking quota for metadata blocks
1118 vfs_dq_release_reservation_block(inode
, mdb_free
);
1121 * If we have done all the pending block allocations and if
1122 * there aren't any writers on the inode, we can discard the
1123 * inode's preallocations.
1125 if (!total
&& (atomic_read(&inode
->i_writecount
) == 0))
1126 ext4_discard_preallocations(inode
);
1129 static int check_block_validity(struct inode
*inode
, sector_t logical
,
1130 sector_t phys
, int len
)
1132 if (!ext4_data_block_valid(EXT4_SB(inode
->i_sb
), phys
, len
)) {
1133 ext4_error(inode
->i_sb
, "check_block_validity",
1134 "inode #%lu logical block %llu mapped to %llu "
1135 "(size %d)", inode
->i_ino
,
1136 (unsigned long long) logical
,
1137 (unsigned long long) phys
, len
);
1145 * The ext4_get_blocks() function tries to look up the requested blocks,
1146 * and returns if the blocks are already mapped.
1148 * Otherwise it takes the write lock of the i_data_sem and allocate blocks
1149 * and store the allocated blocks in the result buffer head and mark it
1152 * If file type is extents based, it will call ext4_ext_get_blocks(),
1153 * Otherwise, call with ext4_ind_get_blocks() to handle indirect mapping
1156 * On success, it returns the number of blocks being mapped or allocate.
1157 * if create==0 and the blocks are pre-allocated and uninitialized block,
1158 * the result buffer head is unmapped. If the create ==1, it will make sure
1159 * the buffer head is mapped.
1161 * It returns 0 if plain look up failed (blocks have not been allocated), in
1162 * that casem, buffer head is unmapped
1164 * It returns the error in case of allocation failure.
1166 int ext4_get_blocks(handle_t
*handle
, struct inode
*inode
, sector_t block
,
1167 unsigned int max_blocks
, struct buffer_head
*bh
,
1172 clear_buffer_mapped(bh
);
1173 clear_buffer_unwritten(bh
);
1176 * Try to see if we can get the block without requesting a new
1177 * file system block.
1179 down_read((&EXT4_I(inode
)->i_data_sem
));
1180 if (EXT4_I(inode
)->i_flags
& EXT4_EXTENTS_FL
) {
1181 retval
= ext4_ext_get_blocks(handle
, inode
, block
, max_blocks
,
1184 retval
= ext4_ind_get_blocks(handle
, inode
, block
, max_blocks
,
1187 up_read((&EXT4_I(inode
)->i_data_sem
));
1189 if (retval
> 0 && buffer_mapped(bh
)) {
1190 int ret
= check_block_validity(inode
, block
,
1191 bh
->b_blocknr
, retval
);
1196 /* If it is only a block(s) look up */
1197 if ((flags
& EXT4_GET_BLOCKS_CREATE
) == 0)
1201 * Returns if the blocks have already allocated
1203 * Note that if blocks have been preallocated
1204 * ext4_ext_get_block() returns th create = 0
1205 * with buffer head unmapped.
1207 if (retval
> 0 && buffer_mapped(bh
))
1211 * When we call get_blocks without the create flag, the
1212 * BH_Unwritten flag could have gotten set if the blocks
1213 * requested were part of a uninitialized extent. We need to
1214 * clear this flag now that we are committed to convert all or
1215 * part of the uninitialized extent to be an initialized
1216 * extent. This is because we need to avoid the combination
1217 * of BH_Unwritten and BH_Mapped flags being simultaneously
1218 * set on the buffer_head.
1220 clear_buffer_unwritten(bh
);
1223 * New blocks allocate and/or writing to uninitialized extent
1224 * will possibly result in updating i_data, so we take
1225 * the write lock of i_data_sem, and call get_blocks()
1226 * with create == 1 flag.
1228 down_write((&EXT4_I(inode
)->i_data_sem
));
1231 * if the caller is from delayed allocation writeout path
1232 * we have already reserved fs blocks for allocation
1233 * let the underlying get_block() function know to
1234 * avoid double accounting
1236 if (flags
& EXT4_GET_BLOCKS_DELALLOC_RESERVE
)
1237 EXT4_I(inode
)->i_delalloc_reserved_flag
= 1;
1239 * We need to check for EXT4 here because migrate
1240 * could have changed the inode type in between
1242 if (EXT4_I(inode
)->i_flags
& EXT4_EXTENTS_FL
) {
1243 retval
= ext4_ext_get_blocks(handle
, inode
, block
, max_blocks
,
1246 retval
= ext4_ind_get_blocks(handle
, inode
, block
,
1247 max_blocks
, bh
, flags
);
1249 if (retval
> 0 && buffer_new(bh
)) {
1251 * We allocated new blocks which will result in
1252 * i_data's format changing. Force the migrate
1253 * to fail by clearing migrate flags
1255 EXT4_I(inode
)->i_flags
= EXT4_I(inode
)->i_flags
&
1260 if (flags
& EXT4_GET_BLOCKS_DELALLOC_RESERVE
)
1261 EXT4_I(inode
)->i_delalloc_reserved_flag
= 0;
1264 * Update reserved blocks/metadata blocks after successful
1265 * block allocation which had been deferred till now.
1267 if ((retval
> 0) && (flags
& EXT4_GET_BLOCKS_UPDATE_RESERVE_SPACE
))
1268 ext4_da_update_reserve_space(inode
, retval
);
1270 up_write((&EXT4_I(inode
)->i_data_sem
));
1271 if (retval
> 0 && buffer_mapped(bh
)) {
1272 int ret
= check_block_validity(inode
, block
,
1273 bh
->b_blocknr
, retval
);
1280 /* Maximum number of blocks we map for direct IO at once. */
1281 #define DIO_MAX_BLOCKS 4096
1283 int ext4_get_block(struct inode
*inode
, sector_t iblock
,
1284 struct buffer_head
*bh_result
, int create
)
1286 handle_t
*handle
= ext4_journal_current_handle();
1287 int ret
= 0, started
= 0;
1288 unsigned max_blocks
= bh_result
->b_size
>> inode
->i_blkbits
;
1291 if (create
&& !handle
) {
1292 /* Direct IO write... */
1293 if (max_blocks
> DIO_MAX_BLOCKS
)
1294 max_blocks
= DIO_MAX_BLOCKS
;
1295 dio_credits
= ext4_chunk_trans_blocks(inode
, max_blocks
);
1296 handle
= ext4_journal_start(inode
, dio_credits
);
1297 if (IS_ERR(handle
)) {
1298 ret
= PTR_ERR(handle
);
1304 ret
= ext4_get_blocks(handle
, inode
, iblock
, max_blocks
, bh_result
,
1305 create
? EXT4_GET_BLOCKS_CREATE
: 0);
1307 bh_result
->b_size
= (ret
<< inode
->i_blkbits
);
1311 ext4_journal_stop(handle
);
1317 * `handle' can be NULL if create is zero
1319 struct buffer_head
*ext4_getblk(handle_t
*handle
, struct inode
*inode
,
1320 ext4_lblk_t block
, int create
, int *errp
)
1322 struct buffer_head dummy
;
1324 int flags
= EXT4_GET_BLOCKS_EXTEND_DISKSIZE
;
1326 J_ASSERT(handle
!= NULL
|| create
== 0);
1329 dummy
.b_blocknr
= -1000;
1330 buffer_trace_init(&dummy
.b_history
);
1332 flags
|= EXT4_GET_BLOCKS_CREATE
;
1333 err
= ext4_get_blocks(handle
, inode
, block
, 1, &dummy
, flags
);
1335 * ext4_get_blocks() returns number of blocks mapped. 0 in
1344 if (!err
&& buffer_mapped(&dummy
)) {
1345 struct buffer_head
*bh
;
1346 bh
= sb_getblk(inode
->i_sb
, dummy
.b_blocknr
);
1351 if (buffer_new(&dummy
)) {
1352 J_ASSERT(create
!= 0);
1353 J_ASSERT(handle
!= NULL
);
1356 * Now that we do not always journal data, we should
1357 * keep in mind whether this should always journal the
1358 * new buffer as metadata. For now, regular file
1359 * writes use ext4_get_block instead, so it's not a
1363 BUFFER_TRACE(bh
, "call get_create_access");
1364 fatal
= ext4_journal_get_create_access(handle
, bh
);
1365 if (!fatal
&& !buffer_uptodate(bh
)) {
1366 memset(bh
->b_data
, 0, inode
->i_sb
->s_blocksize
);
1367 set_buffer_uptodate(bh
);
1370 BUFFER_TRACE(bh
, "call ext4_handle_dirty_metadata");
1371 err
= ext4_handle_dirty_metadata(handle
, inode
, bh
);
1375 BUFFER_TRACE(bh
, "not a new buffer");
1388 struct buffer_head
*ext4_bread(handle_t
*handle
, struct inode
*inode
,
1389 ext4_lblk_t block
, int create
, int *err
)
1391 struct buffer_head
*bh
;
1393 bh
= ext4_getblk(handle
, inode
, block
, create
, err
);
1396 if (buffer_uptodate(bh
))
1398 ll_rw_block(READ_META
, 1, &bh
);
1400 if (buffer_uptodate(bh
))
1407 static int walk_page_buffers(handle_t
*handle
,
1408 struct buffer_head
*head
,
1412 int (*fn
)(handle_t
*handle
,
1413 struct buffer_head
*bh
))
1415 struct buffer_head
*bh
;
1416 unsigned block_start
, block_end
;
1417 unsigned blocksize
= head
->b_size
;
1419 struct buffer_head
*next
;
1421 for (bh
= head
, block_start
= 0;
1422 ret
== 0 && (bh
!= head
|| !block_start
);
1423 block_start
= block_end
, bh
= next
)
1425 next
= bh
->b_this_page
;
1426 block_end
= block_start
+ blocksize
;
1427 if (block_end
<= from
|| block_start
>= to
) {
1428 if (partial
&& !buffer_uptodate(bh
))
1432 err
= (*fn
)(handle
, bh
);
1440 * To preserve ordering, it is essential that the hole instantiation and
1441 * the data write be encapsulated in a single transaction. We cannot
1442 * close off a transaction and start a new one between the ext4_get_block()
1443 * and the commit_write(). So doing the jbd2_journal_start at the start of
1444 * prepare_write() is the right place.
1446 * Also, this function can nest inside ext4_writepage() ->
1447 * block_write_full_page(). In that case, we *know* that ext4_writepage()
1448 * has generated enough buffer credits to do the whole page. So we won't
1449 * block on the journal in that case, which is good, because the caller may
1452 * By accident, ext4 can be reentered when a transaction is open via
1453 * quota file writes. If we were to commit the transaction while thus
1454 * reentered, there can be a deadlock - we would be holding a quota
1455 * lock, and the commit would never complete if another thread had a
1456 * transaction open and was blocking on the quota lock - a ranking
1459 * So what we do is to rely on the fact that jbd2_journal_stop/journal_start
1460 * will _not_ run commit under these circumstances because handle->h_ref
1461 * is elevated. We'll still have enough credits for the tiny quotafile
1464 static int do_journal_get_write_access(handle_t
*handle
,
1465 struct buffer_head
*bh
)
1467 if (!buffer_mapped(bh
) || buffer_freed(bh
))
1469 return ext4_journal_get_write_access(handle
, bh
);
1472 static int ext4_write_begin(struct file
*file
, struct address_space
*mapping
,
1473 loff_t pos
, unsigned len
, unsigned flags
,
1474 struct page
**pagep
, void **fsdata
)
1476 struct inode
*inode
= mapping
->host
;
1477 int ret
, needed_blocks
= ext4_writepage_trans_blocks(inode
);
1484 trace_mark(ext4_write_begin
,
1485 "dev %s ino %lu pos %llu len %u flags %u",
1486 inode
->i_sb
->s_id
, inode
->i_ino
,
1487 (unsigned long long) pos
, len
, flags
);
1488 index
= pos
>> PAGE_CACHE_SHIFT
;
1489 from
= pos
& (PAGE_CACHE_SIZE
- 1);
1493 handle
= ext4_journal_start(inode
, needed_blocks
);
1494 if (IS_ERR(handle
)) {
1495 ret
= PTR_ERR(handle
);
1499 /* We cannot recurse into the filesystem as the transaction is already
1501 flags
|= AOP_FLAG_NOFS
;
1503 page
= grab_cache_page_write_begin(mapping
, index
, flags
);
1505 ext4_journal_stop(handle
);
1511 ret
= block_write_begin(file
, mapping
, pos
, len
, flags
, pagep
, fsdata
,
1514 if (!ret
&& ext4_should_journal_data(inode
)) {
1515 ret
= walk_page_buffers(handle
, page_buffers(page
),
1516 from
, to
, NULL
, do_journal_get_write_access
);
1521 ext4_journal_stop(handle
);
1522 page_cache_release(page
);
1524 * block_write_begin may have instantiated a few blocks
1525 * outside i_size. Trim these off again. Don't need
1526 * i_size_read because we hold i_mutex.
1528 if (pos
+ len
> inode
->i_size
)
1529 vmtruncate(inode
, inode
->i_size
);
1532 if (ret
== -ENOSPC
&& ext4_should_retry_alloc(inode
->i_sb
, &retries
))
1538 /* For write_end() in data=journal mode */
1539 static int write_end_fn(handle_t
*handle
, struct buffer_head
*bh
)
1541 if (!buffer_mapped(bh
) || buffer_freed(bh
))
1543 set_buffer_uptodate(bh
);
1544 return ext4_handle_dirty_metadata(handle
, NULL
, bh
);
1548 * We need to pick up the new inode size which generic_commit_write gave us
1549 * `file' can be NULL - eg, when called from page_symlink().
1551 * ext4 never places buffers on inode->i_mapping->private_list. metadata
1552 * buffers are managed internally.
1554 static int ext4_ordered_write_end(struct file
*file
,
1555 struct address_space
*mapping
,
1556 loff_t pos
, unsigned len
, unsigned copied
,
1557 struct page
*page
, void *fsdata
)
1559 handle_t
*handle
= ext4_journal_current_handle();
1560 struct inode
*inode
= mapping
->host
;
1563 trace_mark(ext4_ordered_write_end
,
1564 "dev %s ino %lu pos %llu len %u copied %u",
1565 inode
->i_sb
->s_id
, inode
->i_ino
,
1566 (unsigned long long) pos
, len
, copied
);
1567 ret
= ext4_jbd2_file_inode(handle
, inode
);
1572 new_i_size
= pos
+ copied
;
1573 if (new_i_size
> EXT4_I(inode
)->i_disksize
) {
1574 ext4_update_i_disksize(inode
, new_i_size
);
1575 /* We need to mark inode dirty even if
1576 * new_i_size is less that inode->i_size
1577 * bu greater than i_disksize.(hint delalloc)
1579 ext4_mark_inode_dirty(handle
, inode
);
1582 ret2
= generic_write_end(file
, mapping
, pos
, len
, copied
,
1588 ret2
= ext4_journal_stop(handle
);
1592 return ret
? ret
: copied
;
1595 static int ext4_writeback_write_end(struct file
*file
,
1596 struct address_space
*mapping
,
1597 loff_t pos
, unsigned len
, unsigned copied
,
1598 struct page
*page
, void *fsdata
)
1600 handle_t
*handle
= ext4_journal_current_handle();
1601 struct inode
*inode
= mapping
->host
;
1605 trace_mark(ext4_writeback_write_end
,
1606 "dev %s ino %lu pos %llu len %u copied %u",
1607 inode
->i_sb
->s_id
, inode
->i_ino
,
1608 (unsigned long long) pos
, len
, copied
);
1609 new_i_size
= pos
+ copied
;
1610 if (new_i_size
> EXT4_I(inode
)->i_disksize
) {
1611 ext4_update_i_disksize(inode
, new_i_size
);
1612 /* We need to mark inode dirty even if
1613 * new_i_size is less that inode->i_size
1614 * bu greater than i_disksize.(hint delalloc)
1616 ext4_mark_inode_dirty(handle
, inode
);
1619 ret2
= generic_write_end(file
, mapping
, pos
, len
, copied
,
1625 ret2
= ext4_journal_stop(handle
);
1629 return ret
? ret
: copied
;
1632 static int ext4_journalled_write_end(struct file
*file
,
1633 struct address_space
*mapping
,
1634 loff_t pos
, unsigned len
, unsigned copied
,
1635 struct page
*page
, void *fsdata
)
1637 handle_t
*handle
= ext4_journal_current_handle();
1638 struct inode
*inode
= mapping
->host
;
1644 trace_mark(ext4_journalled_write_end
,
1645 "dev %s ino %lu pos %llu len %u copied %u",
1646 inode
->i_sb
->s_id
, inode
->i_ino
,
1647 (unsigned long long) pos
, len
, copied
);
1648 from
= pos
& (PAGE_CACHE_SIZE
- 1);
1652 if (!PageUptodate(page
))
1654 page_zero_new_buffers(page
, from
+copied
, to
);
1657 ret
= walk_page_buffers(handle
, page_buffers(page
), from
,
1658 to
, &partial
, write_end_fn
);
1660 SetPageUptodate(page
);
1661 new_i_size
= pos
+ copied
;
1662 if (new_i_size
> inode
->i_size
)
1663 i_size_write(inode
, pos
+copied
);
1664 EXT4_I(inode
)->i_state
|= EXT4_STATE_JDATA
;
1665 if (new_i_size
> EXT4_I(inode
)->i_disksize
) {
1666 ext4_update_i_disksize(inode
, new_i_size
);
1667 ret2
= ext4_mark_inode_dirty(handle
, inode
);
1673 ret2
= ext4_journal_stop(handle
);
1676 page_cache_release(page
);
1678 return ret
? ret
: copied
;
1681 static int ext4_da_reserve_space(struct inode
*inode
, int nrblocks
)
1684 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
1685 unsigned long md_needed
, mdblocks
, total
= 0;
1688 * recalculate the amount of metadata blocks to reserve
1689 * in order to allocate nrblocks
1690 * worse case is one extent per block
1693 spin_lock(&EXT4_I(inode
)->i_block_reservation_lock
);
1694 total
= EXT4_I(inode
)->i_reserved_data_blocks
+ nrblocks
;
1695 mdblocks
= ext4_calc_metadata_amount(inode
, total
);
1696 BUG_ON(mdblocks
< EXT4_I(inode
)->i_reserved_meta_blocks
);
1698 md_needed
= mdblocks
- EXT4_I(inode
)->i_reserved_meta_blocks
;
1699 total
= md_needed
+ nrblocks
;
1702 * Make quota reservation here to prevent quota overflow
1703 * later. Real quota accounting is done at pages writeout
1706 if (vfs_dq_reserve_block(inode
, total
)) {
1707 spin_unlock(&EXT4_I(inode
)->i_block_reservation_lock
);
1711 if (ext4_claim_free_blocks(sbi
, total
)) {
1712 spin_unlock(&EXT4_I(inode
)->i_block_reservation_lock
);
1713 if (ext4_should_retry_alloc(inode
->i_sb
, &retries
)) {
1717 vfs_dq_release_reservation_block(inode
, total
);
1720 EXT4_I(inode
)->i_reserved_data_blocks
+= nrblocks
;
1721 EXT4_I(inode
)->i_reserved_meta_blocks
= mdblocks
;
1723 spin_unlock(&EXT4_I(inode
)->i_block_reservation_lock
);
1724 return 0; /* success */
1727 static void ext4_da_release_space(struct inode
*inode
, int to_free
)
1729 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
1730 int total
, mdb
, mdb_free
, release
;
1733 return; /* Nothing to release, exit */
1735 spin_lock(&EXT4_I(inode
)->i_block_reservation_lock
);
1737 if (!EXT4_I(inode
)->i_reserved_data_blocks
) {
1739 * if there is no reserved blocks, but we try to free some
1740 * then the counter is messed up somewhere.
1741 * but since this function is called from invalidate
1742 * page, it's harmless to return without any action
1744 printk(KERN_INFO
"ext4 delalloc try to release %d reserved "
1745 "blocks for inode %lu, but there is no reserved "
1746 "data blocks\n", to_free
, inode
->i_ino
);
1747 spin_unlock(&EXT4_I(inode
)->i_block_reservation_lock
);
1751 /* recalculate the number of metablocks still need to be reserved */
1752 total
= EXT4_I(inode
)->i_reserved_data_blocks
- to_free
;
1753 mdb
= ext4_calc_metadata_amount(inode
, total
);
1755 /* figure out how many metablocks to release */
1756 BUG_ON(mdb
> EXT4_I(inode
)->i_reserved_meta_blocks
);
1757 mdb_free
= EXT4_I(inode
)->i_reserved_meta_blocks
- mdb
;
1759 release
= to_free
+ mdb_free
;
1761 /* update fs dirty blocks counter for truncate case */
1762 percpu_counter_sub(&sbi
->s_dirtyblocks_counter
, release
);
1764 /* update per-inode reservations */
1765 BUG_ON(to_free
> EXT4_I(inode
)->i_reserved_data_blocks
);
1766 EXT4_I(inode
)->i_reserved_data_blocks
-= to_free
;
1768 BUG_ON(mdb
> EXT4_I(inode
)->i_reserved_meta_blocks
);
1769 EXT4_I(inode
)->i_reserved_meta_blocks
= mdb
;
1770 spin_unlock(&EXT4_I(inode
)->i_block_reservation_lock
);
1772 vfs_dq_release_reservation_block(inode
, release
);
1775 static void ext4_da_page_release_reservation(struct page
*page
,
1776 unsigned long offset
)
1779 struct buffer_head
*head
, *bh
;
1780 unsigned int curr_off
= 0;
1782 head
= page_buffers(page
);
1785 unsigned int next_off
= curr_off
+ bh
->b_size
;
1787 if ((offset
<= curr_off
) && (buffer_delay(bh
))) {
1789 clear_buffer_delay(bh
);
1791 curr_off
= next_off
;
1792 } while ((bh
= bh
->b_this_page
) != head
);
1793 ext4_da_release_space(page
->mapping
->host
, to_release
);
1797 * Delayed allocation stuff
1800 struct mpage_da_data
{
1801 struct inode
*inode
;
1802 sector_t b_blocknr
; /* start block number of extent */
1803 size_t b_size
; /* size of extent */
1804 unsigned long b_state
; /* state of the extent */
1805 unsigned long first_page
, next_page
; /* extent of pages */
1806 struct writeback_control
*wbc
;
1813 * mpage_da_submit_io - walks through extent of pages and try to write
1814 * them with writepage() call back
1816 * @mpd->inode: inode
1817 * @mpd->first_page: first page of the extent
1818 * @mpd->next_page: page after the last page of the extent
1820 * By the time mpage_da_submit_io() is called we expect all blocks
1821 * to be allocated. this may be wrong if allocation failed.
1823 * As pages are already locked by write_cache_pages(), we can't use it
1825 static int mpage_da_submit_io(struct mpage_da_data
*mpd
)
1828 struct pagevec pvec
;
1829 unsigned long index
, end
;
1830 int ret
= 0, err
, nr_pages
, i
;
1831 struct inode
*inode
= mpd
->inode
;
1832 struct address_space
*mapping
= inode
->i_mapping
;
1834 BUG_ON(mpd
->next_page
<= mpd
->first_page
);
1836 * We need to start from the first_page to the next_page - 1
1837 * to make sure we also write the mapped dirty buffer_heads.
1838 * If we look at mpd->b_blocknr we would only be looking
1839 * at the currently mapped buffer_heads.
1841 index
= mpd
->first_page
;
1842 end
= mpd
->next_page
- 1;
1844 pagevec_init(&pvec
, 0);
1845 while (index
<= end
) {
1846 nr_pages
= pagevec_lookup(&pvec
, mapping
, index
, PAGEVEC_SIZE
);
1849 for (i
= 0; i
< nr_pages
; i
++) {
1850 struct page
*page
= pvec
.pages
[i
];
1852 index
= page
->index
;
1857 BUG_ON(!PageLocked(page
));
1858 BUG_ON(PageWriteback(page
));
1860 pages_skipped
= mpd
->wbc
->pages_skipped
;
1861 err
= mapping
->a_ops
->writepage(page
, mpd
->wbc
);
1862 if (!err
&& (pages_skipped
== mpd
->wbc
->pages_skipped
))
1864 * have successfully written the page
1865 * without skipping the same
1867 mpd
->pages_written
++;
1869 * In error case, we have to continue because
1870 * remaining pages are still locked
1871 * XXX: unlock and re-dirty them?
1876 pagevec_release(&pvec
);
1882 * mpage_put_bnr_to_bhs - walk blocks and assign them actual numbers
1884 * @mpd->inode - inode to walk through
1885 * @exbh->b_blocknr - first block on a disk
1886 * @exbh->b_size - amount of space in bytes
1887 * @logical - first logical block to start assignment with
1889 * the function goes through all passed space and put actual disk
1890 * block numbers into buffer heads, dropping BH_Delay and BH_Unwritten
1892 static void mpage_put_bnr_to_bhs(struct mpage_da_data
*mpd
, sector_t logical
,
1893 struct buffer_head
*exbh
)
1895 struct inode
*inode
= mpd
->inode
;
1896 struct address_space
*mapping
= inode
->i_mapping
;
1897 int blocks
= exbh
->b_size
>> inode
->i_blkbits
;
1898 sector_t pblock
= exbh
->b_blocknr
, cur_logical
;
1899 struct buffer_head
*head
, *bh
;
1901 struct pagevec pvec
;
1904 index
= logical
>> (PAGE_CACHE_SHIFT
- inode
->i_blkbits
);
1905 end
= (logical
+ blocks
- 1) >> (PAGE_CACHE_SHIFT
- inode
->i_blkbits
);
1906 cur_logical
= index
<< (PAGE_CACHE_SHIFT
- inode
->i_blkbits
);
1908 pagevec_init(&pvec
, 0);
1910 while (index
<= end
) {
1911 /* XXX: optimize tail */
1912 nr_pages
= pagevec_lookup(&pvec
, mapping
, index
, PAGEVEC_SIZE
);
1915 for (i
= 0; i
< nr_pages
; i
++) {
1916 struct page
*page
= pvec
.pages
[i
];
1918 index
= page
->index
;
1923 BUG_ON(!PageLocked(page
));
1924 BUG_ON(PageWriteback(page
));
1925 BUG_ON(!page_has_buffers(page
));
1927 bh
= page_buffers(page
);
1930 /* skip blocks out of the range */
1932 if (cur_logical
>= logical
)
1935 } while ((bh
= bh
->b_this_page
) != head
);
1938 if (cur_logical
>= logical
+ blocks
)
1941 if (buffer_delay(bh
) ||
1942 buffer_unwritten(bh
)) {
1944 BUG_ON(bh
->b_bdev
!= inode
->i_sb
->s_bdev
);
1946 if (buffer_delay(bh
)) {
1947 clear_buffer_delay(bh
);
1948 bh
->b_blocknr
= pblock
;
1951 * unwritten already should have
1952 * blocknr assigned. Verify that
1954 clear_buffer_unwritten(bh
);
1955 BUG_ON(bh
->b_blocknr
!= pblock
);
1958 } else if (buffer_mapped(bh
))
1959 BUG_ON(bh
->b_blocknr
!= pblock
);
1963 } while ((bh
= bh
->b_this_page
) != head
);
1965 pagevec_release(&pvec
);
1971 * __unmap_underlying_blocks - just a helper function to unmap
1972 * set of blocks described by @bh
1974 static inline void __unmap_underlying_blocks(struct inode
*inode
,
1975 struct buffer_head
*bh
)
1977 struct block_device
*bdev
= inode
->i_sb
->s_bdev
;
1980 blocks
= bh
->b_size
>> inode
->i_blkbits
;
1981 for (i
= 0; i
< blocks
; i
++)
1982 unmap_underlying_metadata(bdev
, bh
->b_blocknr
+ i
);
1985 static void ext4_da_block_invalidatepages(struct mpage_da_data
*mpd
,
1986 sector_t logical
, long blk_cnt
)
1990 struct pagevec pvec
;
1991 struct inode
*inode
= mpd
->inode
;
1992 struct address_space
*mapping
= inode
->i_mapping
;
1994 index
= logical
>> (PAGE_CACHE_SHIFT
- inode
->i_blkbits
);
1995 end
= (logical
+ blk_cnt
- 1) >>
1996 (PAGE_CACHE_SHIFT
- inode
->i_blkbits
);
1997 while (index
<= end
) {
1998 nr_pages
= pagevec_lookup(&pvec
, mapping
, index
, PAGEVEC_SIZE
);
2001 for (i
= 0; i
< nr_pages
; i
++) {
2002 struct page
*page
= pvec
.pages
[i
];
2003 index
= page
->index
;
2008 BUG_ON(!PageLocked(page
));
2009 BUG_ON(PageWriteback(page
));
2010 block_invalidatepage(page
, 0);
2011 ClearPageUptodate(page
);
2018 static void ext4_print_free_blocks(struct inode
*inode
)
2020 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
2021 printk(KERN_EMERG
"Total free blocks count %lld\n",
2022 ext4_count_free_blocks(inode
->i_sb
));
2023 printk(KERN_EMERG
"Free/Dirty block details\n");
2024 printk(KERN_EMERG
"free_blocks=%lld\n",
2025 (long long)percpu_counter_sum(&sbi
->s_freeblocks_counter
));
2026 printk(KERN_EMERG
"dirty_blocks=%lld\n",
2027 (long long)percpu_counter_sum(&sbi
->s_dirtyblocks_counter
));
2028 printk(KERN_EMERG
"Block reservation details\n");
2029 printk(KERN_EMERG
"i_reserved_data_blocks=%u\n",
2030 EXT4_I(inode
)->i_reserved_data_blocks
);
2031 printk(KERN_EMERG
"i_reserved_meta_blocks=%u\n",
2032 EXT4_I(inode
)->i_reserved_meta_blocks
);
2037 * mpage_da_map_blocks - go through given space
2039 * @mpd - bh describing space
2041 * The function skips space we know is already mapped to disk blocks.
2044 static int mpage_da_map_blocks(struct mpage_da_data
*mpd
)
2046 int err
, blks
, get_blocks_flags
;
2047 struct buffer_head
new;
2048 sector_t next
= mpd
->b_blocknr
;
2049 unsigned max_blocks
= mpd
->b_size
>> mpd
->inode
->i_blkbits
;
2050 loff_t disksize
= EXT4_I(mpd
->inode
)->i_disksize
;
2051 handle_t
*handle
= NULL
;
2054 * We consider only non-mapped and non-allocated blocks
2056 if ((mpd
->b_state
& (1 << BH_Mapped
)) &&
2057 !(mpd
->b_state
& (1 << BH_Delay
)) &&
2058 !(mpd
->b_state
& (1 << BH_Unwritten
)))
2062 * If we didn't accumulate anything to write simply return
2067 handle
= ext4_journal_current_handle();
2071 * Call ext4_get_blocks() to allocate any delayed allocation
2072 * blocks, or to convert an uninitialized extent to be
2073 * initialized (in the case where we have written into
2074 * one or more preallocated blocks).
2076 * We pass in the magic EXT4_GET_BLOCKS_DELALLOC_RESERVE to
2077 * indicate that we are on the delayed allocation path. This
2078 * affects functions in many different parts of the allocation
2079 * call path. This flag exists primarily because we don't
2080 * want to change *many* call functions, so ext4_get_blocks()
2081 * will set the magic i_delalloc_reserved_flag once the
2082 * inode's allocation semaphore is taken.
2084 * If the blocks in questions were delalloc blocks, set
2085 * EXT4_GET_BLOCKS_DELALLOC_RESERVE so the delalloc accounting
2086 * variables are updated after the blocks have been allocated.
2089 get_blocks_flags
= (EXT4_GET_BLOCKS_CREATE
|
2090 EXT4_GET_BLOCKS_DELALLOC_RESERVE
);
2091 if (mpd
->b_state
& (1 << BH_Delay
))
2092 get_blocks_flags
|= EXT4_GET_BLOCKS_UPDATE_RESERVE_SPACE
;
2093 blks
= ext4_get_blocks(handle
, mpd
->inode
, next
, max_blocks
,
2094 &new, get_blocks_flags
);
2098 * If get block returns with error we simply
2099 * return. Later writepage will redirty the page and
2100 * writepages will find the dirty page again
2105 if (err
== -ENOSPC
&&
2106 ext4_count_free_blocks(mpd
->inode
->i_sb
)) {
2112 * get block failure will cause us to loop in
2113 * writepages, because a_ops->writepage won't be able
2114 * to make progress. The page will be redirtied by
2115 * writepage and writepages will again try to write
2118 printk(KERN_EMERG
"%s block allocation failed for inode %lu "
2119 "at logical offset %llu with max blocks "
2120 "%zd with error %d\n",
2121 __func__
, mpd
->inode
->i_ino
,
2122 (unsigned long long)next
,
2123 mpd
->b_size
>> mpd
->inode
->i_blkbits
, err
);
2124 printk(KERN_EMERG
"This should not happen.!! "
2125 "Data will be lost\n");
2126 if (err
== -ENOSPC
) {
2127 ext4_print_free_blocks(mpd
->inode
);
2129 /* invalidate all the pages */
2130 ext4_da_block_invalidatepages(mpd
, next
,
2131 mpd
->b_size
>> mpd
->inode
->i_blkbits
);
2136 new.b_size
= (blks
<< mpd
->inode
->i_blkbits
);
2138 if (buffer_new(&new))
2139 __unmap_underlying_blocks(mpd
->inode
, &new);
2142 * If blocks are delayed marked, we need to
2143 * put actual blocknr and drop delayed bit
2145 if ((mpd
->b_state
& (1 << BH_Delay
)) ||
2146 (mpd
->b_state
& (1 << BH_Unwritten
)))
2147 mpage_put_bnr_to_bhs(mpd
, next
, &new);
2149 if (ext4_should_order_data(mpd
->inode
)) {
2150 err
= ext4_jbd2_file_inode(handle
, mpd
->inode
);
2156 * Update on-disk size along with block allocation we don't
2157 * use EXT4_GET_BLOCKS_EXTEND_DISKSIZE as size may change
2158 * within already allocated block -bzzz
2160 disksize
= ((loff_t
) next
+ blks
) << mpd
->inode
->i_blkbits
;
2161 if (disksize
> i_size_read(mpd
->inode
))
2162 disksize
= i_size_read(mpd
->inode
);
2163 if (disksize
> EXT4_I(mpd
->inode
)->i_disksize
) {
2164 ext4_update_i_disksize(mpd
->inode
, disksize
);
2165 return ext4_mark_inode_dirty(handle
, mpd
->inode
);
2171 #define BH_FLAGS ((1 << BH_Uptodate) | (1 << BH_Mapped) | \
2172 (1 << BH_Delay) | (1 << BH_Unwritten))
2175 * mpage_add_bh_to_extent - try to add one more block to extent of blocks
2177 * @mpd->lbh - extent of blocks
2178 * @logical - logical number of the block in the file
2179 * @bh - bh of the block (used to access block's state)
2181 * the function is used to collect contig. blocks in same state
2183 static void mpage_add_bh_to_extent(struct mpage_da_data
*mpd
,
2184 sector_t logical
, size_t b_size
,
2185 unsigned long b_state
)
2188 int nrblocks
= mpd
->b_size
>> mpd
->inode
->i_blkbits
;
2190 /* check if thereserved journal credits might overflow */
2191 if (!(EXT4_I(mpd
->inode
)->i_flags
& EXT4_EXTENTS_FL
)) {
2192 if (nrblocks
>= EXT4_MAX_TRANS_DATA
) {
2194 * With non-extent format we are limited by the journal
2195 * credit available. Total credit needed to insert
2196 * nrblocks contiguous blocks is dependent on the
2197 * nrblocks. So limit nrblocks.
2200 } else if ((nrblocks
+ (b_size
>> mpd
->inode
->i_blkbits
)) >
2201 EXT4_MAX_TRANS_DATA
) {
2203 * Adding the new buffer_head would make it cross the
2204 * allowed limit for which we have journal credit
2205 * reserved. So limit the new bh->b_size
2207 b_size
= (EXT4_MAX_TRANS_DATA
- nrblocks
) <<
2208 mpd
->inode
->i_blkbits
;
2209 /* we will do mpage_da_submit_io in the next loop */
2213 * First block in the extent
2215 if (mpd
->b_size
== 0) {
2216 mpd
->b_blocknr
= logical
;
2217 mpd
->b_size
= b_size
;
2218 mpd
->b_state
= b_state
& BH_FLAGS
;
2222 next
= mpd
->b_blocknr
+ nrblocks
;
2224 * Can we merge the block to our big extent?
2226 if (logical
== next
&& (b_state
& BH_FLAGS
) == mpd
->b_state
) {
2227 mpd
->b_size
+= b_size
;
2233 * We couldn't merge the block to our extent, so we
2234 * need to flush current extent and start new one
2236 if (mpage_da_map_blocks(mpd
) == 0)
2237 mpage_da_submit_io(mpd
);
2242 static int ext4_bh_unmapped_or_delay(handle_t
*handle
, struct buffer_head
*bh
)
2245 * unmapped buffer is possible for holes.
2246 * delay buffer is possible with delayed allocation.
2247 * We also need to consider unwritten buffer as unmapped.
2249 return (!buffer_mapped(bh
) || buffer_delay(bh
) ||
2250 buffer_unwritten(bh
)) && buffer_dirty(bh
);
2254 * __mpage_da_writepage - finds extent of pages and blocks
2256 * @page: page to consider
2257 * @wbc: not used, we just follow rules
2260 * The function finds extents of pages and scan them for all blocks.
2262 static int __mpage_da_writepage(struct page
*page
,
2263 struct writeback_control
*wbc
, void *data
)
2265 struct mpage_da_data
*mpd
= data
;
2266 struct inode
*inode
= mpd
->inode
;
2267 struct buffer_head
*bh
, *head
;
2272 * Rest of the page in the page_vec
2273 * redirty then and skip then. We will
2274 * try to to write them again after
2275 * starting a new transaction
2277 redirty_page_for_writepage(wbc
, page
);
2279 return MPAGE_DA_EXTENT_TAIL
;
2282 * Can we merge this page to current extent?
2284 if (mpd
->next_page
!= page
->index
) {
2286 * Nope, we can't. So, we map non-allocated blocks
2287 * and start IO on them using writepage()
2289 if (mpd
->next_page
!= mpd
->first_page
) {
2290 if (mpage_da_map_blocks(mpd
) == 0)
2291 mpage_da_submit_io(mpd
);
2293 * skip rest of the page in the page_vec
2296 redirty_page_for_writepage(wbc
, page
);
2298 return MPAGE_DA_EXTENT_TAIL
;
2302 * Start next extent of pages ...
2304 mpd
->first_page
= page
->index
;
2314 mpd
->next_page
= page
->index
+ 1;
2315 logical
= (sector_t
) page
->index
<<
2316 (PAGE_CACHE_SHIFT
- inode
->i_blkbits
);
2318 if (!page_has_buffers(page
)) {
2319 mpage_add_bh_to_extent(mpd
, logical
, PAGE_CACHE_SIZE
,
2320 (1 << BH_Dirty
) | (1 << BH_Uptodate
));
2322 return MPAGE_DA_EXTENT_TAIL
;
2325 * Page with regular buffer heads, just add all dirty ones
2327 head
= page_buffers(page
);
2330 BUG_ON(buffer_locked(bh
));
2332 * We need to try to allocate
2333 * unmapped blocks in the same page.
2334 * Otherwise we won't make progress
2335 * with the page in ext4_da_writepage
2337 if (ext4_bh_unmapped_or_delay(NULL
, bh
)) {
2338 mpage_add_bh_to_extent(mpd
, logical
,
2342 return MPAGE_DA_EXTENT_TAIL
;
2343 } else if (buffer_dirty(bh
) && (buffer_mapped(bh
))) {
2345 * mapped dirty buffer. We need to update
2346 * the b_state because we look at
2347 * b_state in mpage_da_map_blocks. We don't
2348 * update b_size because if we find an
2349 * unmapped buffer_head later we need to
2350 * use the b_state flag of that buffer_head.
2352 if (mpd
->b_size
== 0)
2353 mpd
->b_state
= bh
->b_state
& BH_FLAGS
;
2356 } while ((bh
= bh
->b_this_page
) != head
);
2363 * This is a special get_blocks_t callback which is used by
2364 * ext4_da_write_begin(). It will either return mapped block or
2365 * reserve space for a single block.
2367 * For delayed buffer_head we have BH_Mapped, BH_New, BH_Delay set.
2368 * We also have b_blocknr = -1 and b_bdev initialized properly
2370 * For unwritten buffer_head we have BH_Mapped, BH_New, BH_Unwritten set.
2371 * We also have b_blocknr = physicalblock mapping unwritten extent and b_bdev
2372 * initialized properly.
2374 static int ext4_da_get_block_prep(struct inode
*inode
, sector_t iblock
,
2375 struct buffer_head
*bh_result
, int create
)
2378 sector_t invalid_block
= ~((sector_t
) 0xffff);
2380 if (invalid_block
< ext4_blocks_count(EXT4_SB(inode
->i_sb
)->s_es
))
2383 BUG_ON(create
== 0);
2384 BUG_ON(bh_result
->b_size
!= inode
->i_sb
->s_blocksize
);
2387 * first, we need to know whether the block is allocated already
2388 * preallocated blocks are unmapped but should treated
2389 * the same as allocated blocks.
2391 ret
= ext4_get_blocks(NULL
, inode
, iblock
, 1, bh_result
, 0);
2392 if ((ret
== 0) && !buffer_delay(bh_result
)) {
2393 /* the block isn't (pre)allocated yet, let's reserve space */
2395 * XXX: __block_prepare_write() unmaps passed block,
2398 ret
= ext4_da_reserve_space(inode
, 1);
2400 /* not enough space to reserve */
2403 map_bh(bh_result
, inode
->i_sb
, invalid_block
);
2404 set_buffer_new(bh_result
);
2405 set_buffer_delay(bh_result
);
2406 } else if (ret
> 0) {
2407 bh_result
->b_size
= (ret
<< inode
->i_blkbits
);
2408 if (buffer_unwritten(bh_result
)) {
2409 /* A delayed write to unwritten bh should
2410 * be marked new and mapped. Mapped ensures
2411 * that we don't do get_block multiple times
2412 * when we write to the same offset and new
2413 * ensures that we do proper zero out for
2416 set_buffer_new(bh_result
);
2417 set_buffer_mapped(bh_result
);
2426 * This function is used as a standard get_block_t calback function
2427 * when there is no desire to allocate any blocks. It is used as a
2428 * callback function for block_prepare_write(), nobh_writepage(), and
2429 * block_write_full_page(). These functions should only try to map a
2430 * single block at a time.
2432 * Since this function doesn't do block allocations even if the caller
2433 * requests it by passing in create=1, it is critically important that
2434 * any caller checks to make sure that any buffer heads are returned
2435 * by this function are either all already mapped or marked for
2436 * delayed allocation before calling nobh_writepage() or
2437 * block_write_full_page(). Otherwise, b_blocknr could be left
2438 * unitialized, and the page write functions will be taken by
2441 static int noalloc_get_block_write(struct inode
*inode
, sector_t iblock
,
2442 struct buffer_head
*bh_result
, int create
)
2445 unsigned max_blocks
= bh_result
->b_size
>> inode
->i_blkbits
;
2447 BUG_ON(bh_result
->b_size
!= inode
->i_sb
->s_blocksize
);
2450 * we don't want to do block allocation in writepage
2451 * so call get_block_wrap with create = 0
2453 ret
= ext4_get_blocks(NULL
, inode
, iblock
, max_blocks
, bh_result
, 0);
2454 BUG_ON(create
&& ret
== 0);
2456 bh_result
->b_size
= (ret
<< inode
->i_blkbits
);
2463 * This function can get called via...
2464 * - ext4_da_writepages after taking page lock (have journal handle)
2465 * - journal_submit_inode_data_buffers (no journal handle)
2466 * - shrink_page_list via pdflush (no journal handle)
2467 * - grab_page_cache when doing write_begin (have journal handle)
2469 static int ext4_da_writepage(struct page
*page
,
2470 struct writeback_control
*wbc
)
2475 struct buffer_head
*page_bufs
;
2476 struct inode
*inode
= page
->mapping
->host
;
2478 trace_mark(ext4_da_writepage
,
2479 "dev %s ino %lu page_index %lu",
2480 inode
->i_sb
->s_id
, inode
->i_ino
, page
->index
);
2481 size
= i_size_read(inode
);
2482 if (page
->index
== size
>> PAGE_CACHE_SHIFT
)
2483 len
= size
& ~PAGE_CACHE_MASK
;
2485 len
= PAGE_CACHE_SIZE
;
2487 if (page_has_buffers(page
)) {
2488 page_bufs
= page_buffers(page
);
2489 if (walk_page_buffers(NULL
, page_bufs
, 0, len
, NULL
,
2490 ext4_bh_unmapped_or_delay
)) {
2492 * We don't want to do block allocation
2493 * So redirty the page and return
2494 * We may reach here when we do a journal commit
2495 * via journal_submit_inode_data_buffers.
2496 * If we don't have mapping block we just ignore
2497 * them. We can also reach here via shrink_page_list
2499 redirty_page_for_writepage(wbc
, page
);
2505 * The test for page_has_buffers() is subtle:
2506 * We know the page is dirty but it lost buffers. That means
2507 * that at some moment in time after write_begin()/write_end()
2508 * has been called all buffers have been clean and thus they
2509 * must have been written at least once. So they are all
2510 * mapped and we can happily proceed with mapping them
2511 * and writing the page.
2513 * Try to initialize the buffer_heads and check whether
2514 * all are mapped and non delay. We don't want to
2515 * do block allocation here.
2517 ret
= block_prepare_write(page
, 0, PAGE_CACHE_SIZE
,
2518 noalloc_get_block_write
);
2520 page_bufs
= page_buffers(page
);
2521 /* check whether all are mapped and non delay */
2522 if (walk_page_buffers(NULL
, page_bufs
, 0, len
, NULL
,
2523 ext4_bh_unmapped_or_delay
)) {
2524 redirty_page_for_writepage(wbc
, page
);
2530 * We can't do block allocation here
2531 * so just redity the page and unlock
2534 redirty_page_for_writepage(wbc
, page
);
2538 /* now mark the buffer_heads as dirty and uptodate */
2539 block_commit_write(page
, 0, PAGE_CACHE_SIZE
);
2542 if (test_opt(inode
->i_sb
, NOBH
) && ext4_should_writeback_data(inode
))
2543 ret
= nobh_writepage(page
, noalloc_get_block_write
, wbc
);
2545 ret
= block_write_full_page(page
, noalloc_get_block_write
,
2552 * This is called via ext4_da_writepages() to
2553 * calulate the total number of credits to reserve to fit
2554 * a single extent allocation into a single transaction,
2555 * ext4_da_writpeages() will loop calling this before
2556 * the block allocation.
2559 static int ext4_da_writepages_trans_blocks(struct inode
*inode
)
2561 int max_blocks
= EXT4_I(inode
)->i_reserved_data_blocks
;
2564 * With non-extent format the journal credit needed to
2565 * insert nrblocks contiguous block is dependent on
2566 * number of contiguous block. So we will limit
2567 * number of contiguous block to a sane value
2569 if (!(inode
->i_flags
& EXT4_EXTENTS_FL
) &&
2570 (max_blocks
> EXT4_MAX_TRANS_DATA
))
2571 max_blocks
= EXT4_MAX_TRANS_DATA
;
2573 return ext4_chunk_trans_blocks(inode
, max_blocks
);
2576 static int ext4_da_writepages(struct address_space
*mapping
,
2577 struct writeback_control
*wbc
)
2580 int range_whole
= 0;
2581 handle_t
*handle
= NULL
;
2582 struct mpage_da_data mpd
;
2583 struct inode
*inode
= mapping
->host
;
2584 int no_nrwrite_index_update
;
2585 int pages_written
= 0;
2587 int range_cyclic
, cycled
= 1, io_done
= 0;
2588 int needed_blocks
, ret
= 0, nr_to_writebump
= 0;
2589 struct ext4_sb_info
*sbi
= EXT4_SB(mapping
->host
->i_sb
);
2591 trace_mark(ext4_da_writepages
,
2592 "dev %s ino %lu nr_t_write %ld "
2593 "pages_skipped %ld range_start %llu "
2594 "range_end %llu nonblocking %d "
2595 "for_kupdate %d for_reclaim %d "
2596 "for_writepages %d range_cyclic %d",
2597 inode
->i_sb
->s_id
, inode
->i_ino
,
2598 wbc
->nr_to_write
, wbc
->pages_skipped
,
2599 (unsigned long long) wbc
->range_start
,
2600 (unsigned long long) wbc
->range_end
,
2601 wbc
->nonblocking
, wbc
->for_kupdate
,
2602 wbc
->for_reclaim
, wbc
->for_writepages
,
2606 * No pages to write? This is mainly a kludge to avoid starting
2607 * a transaction for special inodes like journal inode on last iput()
2608 * because that could violate lock ordering on umount
2610 if (!mapping
->nrpages
|| !mapping_tagged(mapping
, PAGECACHE_TAG_DIRTY
))
2614 * If the filesystem has aborted, it is read-only, so return
2615 * right away instead of dumping stack traces later on that
2616 * will obscure the real source of the problem. We test
2617 * EXT4_MOUNT_ABORT instead of sb->s_flag's MS_RDONLY because
2618 * the latter could be true if the filesystem is mounted
2619 * read-only, and in that case, ext4_da_writepages should
2620 * *never* be called, so if that ever happens, we would want
2623 if (unlikely(sbi
->s_mount_opt
& EXT4_MOUNT_ABORT
))
2627 * Make sure nr_to_write is >= sbi->s_mb_stream_request
2628 * This make sure small files blocks are allocated in
2629 * single attempt. This ensure that small files
2630 * get less fragmented.
2632 if (wbc
->nr_to_write
< sbi
->s_mb_stream_request
) {
2633 nr_to_writebump
= sbi
->s_mb_stream_request
- wbc
->nr_to_write
;
2634 wbc
->nr_to_write
= sbi
->s_mb_stream_request
;
2636 if (wbc
->range_start
== 0 && wbc
->range_end
== LLONG_MAX
)
2639 range_cyclic
= wbc
->range_cyclic
;
2640 if (wbc
->range_cyclic
) {
2641 index
= mapping
->writeback_index
;
2644 wbc
->range_start
= index
<< PAGE_CACHE_SHIFT
;
2645 wbc
->range_end
= LLONG_MAX
;
2646 wbc
->range_cyclic
= 0;
2648 index
= wbc
->range_start
>> PAGE_CACHE_SHIFT
;
2651 mpd
.inode
= mapping
->host
;
2654 * we don't want write_cache_pages to update
2655 * nr_to_write and writeback_index
2657 no_nrwrite_index_update
= wbc
->no_nrwrite_index_update
;
2658 wbc
->no_nrwrite_index_update
= 1;
2659 pages_skipped
= wbc
->pages_skipped
;
2662 while (!ret
&& wbc
->nr_to_write
> 0) {
2665 * we insert one extent at a time. So we need
2666 * credit needed for single extent allocation.
2667 * journalled mode is currently not supported
2670 BUG_ON(ext4_should_journal_data(inode
));
2671 needed_blocks
= ext4_da_writepages_trans_blocks(inode
);
2673 /* start a new transaction*/
2674 handle
= ext4_journal_start(inode
, needed_blocks
);
2675 if (IS_ERR(handle
)) {
2676 ret
= PTR_ERR(handle
);
2677 printk(KERN_CRIT
"%s: jbd2_start: "
2678 "%ld pages, ino %lu; err %d\n", __func__
,
2679 wbc
->nr_to_write
, inode
->i_ino
, ret
);
2681 goto out_writepages
;
2685 * Now call __mpage_da_writepage to find the next
2686 * contiguous region of logical blocks that need
2687 * blocks to be allocated by ext4. We don't actually
2688 * submit the blocks for I/O here, even though
2689 * write_cache_pages thinks it will, and will set the
2690 * pages as clean for write before calling
2691 * __mpage_da_writepage().
2699 mpd
.pages_written
= 0;
2701 ret
= write_cache_pages(mapping
, wbc
, __mpage_da_writepage
,
2704 * If we have a contigous extent of pages and we
2705 * haven't done the I/O yet, map the blocks and submit
2708 if (!mpd
.io_done
&& mpd
.next_page
!= mpd
.first_page
) {
2709 if (mpage_da_map_blocks(&mpd
) == 0)
2710 mpage_da_submit_io(&mpd
);
2712 ret
= MPAGE_DA_EXTENT_TAIL
;
2714 wbc
->nr_to_write
-= mpd
.pages_written
;
2716 ext4_journal_stop(handle
);
2718 if ((mpd
.retval
== -ENOSPC
) && sbi
->s_journal
) {
2719 /* commit the transaction which would
2720 * free blocks released in the transaction
2723 jbd2_journal_force_commit_nested(sbi
->s_journal
);
2724 wbc
->pages_skipped
= pages_skipped
;
2726 } else if (ret
== MPAGE_DA_EXTENT_TAIL
) {
2728 * got one extent now try with
2731 pages_written
+= mpd
.pages_written
;
2732 wbc
->pages_skipped
= pages_skipped
;
2735 } else if (wbc
->nr_to_write
)
2737 * There is no more writeout needed
2738 * or we requested for a noblocking writeout
2739 * and we found the device congested
2743 if (!io_done
&& !cycled
) {
2746 wbc
->range_start
= index
<< PAGE_CACHE_SHIFT
;
2747 wbc
->range_end
= mapping
->writeback_index
- 1;
2750 if (pages_skipped
!= wbc
->pages_skipped
)
2751 printk(KERN_EMERG
"This should not happen leaving %s "
2752 "with nr_to_write = %ld ret = %d\n",
2753 __func__
, wbc
->nr_to_write
, ret
);
2756 index
+= pages_written
;
2757 wbc
->range_cyclic
= range_cyclic
;
2758 if (wbc
->range_cyclic
|| (range_whole
&& wbc
->nr_to_write
> 0))
2760 * set the writeback_index so that range_cyclic
2761 * mode will write it back later
2763 mapping
->writeback_index
= index
;
2766 if (!no_nrwrite_index_update
)
2767 wbc
->no_nrwrite_index_update
= 0;
2768 wbc
->nr_to_write
-= nr_to_writebump
;
2769 trace_mark(ext4_da_writepage_result
,
2770 "dev %s ino %lu ret %d pages_written %d "
2771 "pages_skipped %ld congestion %d "
2772 "more_io %d no_nrwrite_index_update %d",
2773 inode
->i_sb
->s_id
, inode
->i_ino
, ret
,
2774 pages_written
, wbc
->pages_skipped
,
2775 wbc
->encountered_congestion
, wbc
->more_io
,
2776 wbc
->no_nrwrite_index_update
);
2780 #define FALL_BACK_TO_NONDELALLOC 1
2781 static int ext4_nonda_switch(struct super_block
*sb
)
2783 s64 free_blocks
, dirty_blocks
;
2784 struct ext4_sb_info
*sbi
= EXT4_SB(sb
);
2787 * switch to non delalloc mode if we are running low
2788 * on free block. The free block accounting via percpu
2789 * counters can get slightly wrong with percpu_counter_batch getting
2790 * accumulated on each CPU without updating global counters
2791 * Delalloc need an accurate free block accounting. So switch
2792 * to non delalloc when we are near to error range.
2794 free_blocks
= percpu_counter_read_positive(&sbi
->s_freeblocks_counter
);
2795 dirty_blocks
= percpu_counter_read_positive(&sbi
->s_dirtyblocks_counter
);
2796 if (2 * free_blocks
< 3 * dirty_blocks
||
2797 free_blocks
< (dirty_blocks
+ EXT4_FREEBLOCKS_WATERMARK
)) {
2799 * free block count is less that 150% of dirty blocks
2800 * or free blocks is less that watermark
2807 static int ext4_da_write_begin(struct file
*file
, struct address_space
*mapping
,
2808 loff_t pos
, unsigned len
, unsigned flags
,
2809 struct page
**pagep
, void **fsdata
)
2811 int ret
, retries
= 0;
2815 struct inode
*inode
= mapping
->host
;
2818 index
= pos
>> PAGE_CACHE_SHIFT
;
2819 from
= pos
& (PAGE_CACHE_SIZE
- 1);
2822 if (ext4_nonda_switch(inode
->i_sb
)) {
2823 *fsdata
= (void *)FALL_BACK_TO_NONDELALLOC
;
2824 return ext4_write_begin(file
, mapping
, pos
,
2825 len
, flags
, pagep
, fsdata
);
2827 *fsdata
= (void *)0;
2829 trace_mark(ext4_da_write_begin
,
2830 "dev %s ino %lu pos %llu len %u flags %u",
2831 inode
->i_sb
->s_id
, inode
->i_ino
,
2832 (unsigned long long) pos
, len
, flags
);
2835 * With delayed allocation, we don't log the i_disksize update
2836 * if there is delayed block allocation. But we still need
2837 * to journalling the i_disksize update if writes to the end
2838 * of file which has an already mapped buffer.
2840 handle
= ext4_journal_start(inode
, 1);
2841 if (IS_ERR(handle
)) {
2842 ret
= PTR_ERR(handle
);
2845 /* We cannot recurse into the filesystem as the transaction is already
2847 flags
|= AOP_FLAG_NOFS
;
2849 page
= grab_cache_page_write_begin(mapping
, index
, flags
);
2851 ext4_journal_stop(handle
);
2857 ret
= block_write_begin(file
, mapping
, pos
, len
, flags
, pagep
, fsdata
,
2858 ext4_da_get_block_prep
);
2861 ext4_journal_stop(handle
);
2862 page_cache_release(page
);
2864 * block_write_begin may have instantiated a few blocks
2865 * outside i_size. Trim these off again. Don't need
2866 * i_size_read because we hold i_mutex.
2868 if (pos
+ len
> inode
->i_size
)
2869 vmtruncate(inode
, inode
->i_size
);
2872 if (ret
== -ENOSPC
&& ext4_should_retry_alloc(inode
->i_sb
, &retries
))
2879 * Check if we should update i_disksize
2880 * when write to the end of file but not require block allocation
2882 static int ext4_da_should_update_i_disksize(struct page
*page
,
2883 unsigned long offset
)
2885 struct buffer_head
*bh
;
2886 struct inode
*inode
= page
->mapping
->host
;
2890 bh
= page_buffers(page
);
2891 idx
= offset
>> inode
->i_blkbits
;
2893 for (i
= 0; i
< idx
; i
++)
2894 bh
= bh
->b_this_page
;
2896 if (!buffer_mapped(bh
) || (buffer_delay(bh
)) || buffer_unwritten(bh
))
2901 static int ext4_da_write_end(struct file
*file
,
2902 struct address_space
*mapping
,
2903 loff_t pos
, unsigned len
, unsigned copied
,
2904 struct page
*page
, void *fsdata
)
2906 struct inode
*inode
= mapping
->host
;
2908 handle_t
*handle
= ext4_journal_current_handle();
2910 unsigned long start
, end
;
2911 int write_mode
= (int)(unsigned long)fsdata
;
2913 if (write_mode
== FALL_BACK_TO_NONDELALLOC
) {
2914 if (ext4_should_order_data(inode
)) {
2915 return ext4_ordered_write_end(file
, mapping
, pos
,
2916 len
, copied
, page
, fsdata
);
2917 } else if (ext4_should_writeback_data(inode
)) {
2918 return ext4_writeback_write_end(file
, mapping
, pos
,
2919 len
, copied
, page
, fsdata
);
2925 trace_mark(ext4_da_write_end
,
2926 "dev %s ino %lu pos %llu len %u copied %u",
2927 inode
->i_sb
->s_id
, inode
->i_ino
,
2928 (unsigned long long) pos
, len
, copied
);
2929 start
= pos
& (PAGE_CACHE_SIZE
- 1);
2930 end
= start
+ copied
- 1;
2933 * generic_write_end() will run mark_inode_dirty() if i_size
2934 * changes. So let's piggyback the i_disksize mark_inode_dirty
2938 new_i_size
= pos
+ copied
;
2939 if (new_i_size
> EXT4_I(inode
)->i_disksize
) {
2940 if (ext4_da_should_update_i_disksize(page
, end
)) {
2941 down_write(&EXT4_I(inode
)->i_data_sem
);
2942 if (new_i_size
> EXT4_I(inode
)->i_disksize
) {
2944 * Updating i_disksize when extending file
2945 * without needing block allocation
2947 if (ext4_should_order_data(inode
))
2948 ret
= ext4_jbd2_file_inode(handle
,
2951 EXT4_I(inode
)->i_disksize
= new_i_size
;
2953 up_write(&EXT4_I(inode
)->i_data_sem
);
2954 /* We need to mark inode dirty even if
2955 * new_i_size is less that inode->i_size
2956 * bu greater than i_disksize.(hint delalloc)
2958 ext4_mark_inode_dirty(handle
, inode
);
2961 ret2
= generic_write_end(file
, mapping
, pos
, len
, copied
,
2966 ret2
= ext4_journal_stop(handle
);
2970 return ret
? ret
: copied
;
2973 static void ext4_da_invalidatepage(struct page
*page
, unsigned long offset
)
2976 * Drop reserved blocks
2978 BUG_ON(!PageLocked(page
));
2979 if (!page_has_buffers(page
))
2982 ext4_da_page_release_reservation(page
, offset
);
2985 ext4_invalidatepage(page
, offset
);
2991 * Force all delayed allocation blocks to be allocated for a given inode.
2993 int ext4_alloc_da_blocks(struct inode
*inode
)
2995 if (!EXT4_I(inode
)->i_reserved_data_blocks
&&
2996 !EXT4_I(inode
)->i_reserved_meta_blocks
)
3000 * We do something simple for now. The filemap_flush() will
3001 * also start triggering a write of the data blocks, which is
3002 * not strictly speaking necessary (and for users of
3003 * laptop_mode, not even desirable). However, to do otherwise
3004 * would require replicating code paths in:
3006 * ext4_da_writepages() ->
3007 * write_cache_pages() ---> (via passed in callback function)
3008 * __mpage_da_writepage() -->
3009 * mpage_add_bh_to_extent()
3010 * mpage_da_map_blocks()
3012 * The problem is that write_cache_pages(), located in
3013 * mm/page-writeback.c, marks pages clean in preparation for
3014 * doing I/O, which is not desirable if we're not planning on
3017 * We could call write_cache_pages(), and then redirty all of
3018 * the pages by calling redirty_page_for_writeback() but that
3019 * would be ugly in the extreme. So instead we would need to
3020 * replicate parts of the code in the above functions,
3021 * simplifying them becuase we wouldn't actually intend to
3022 * write out the pages, but rather only collect contiguous
3023 * logical block extents, call the multi-block allocator, and
3024 * then update the buffer heads with the block allocations.
3026 * For now, though, we'll cheat by calling filemap_flush(),
3027 * which will map the blocks, and start the I/O, but not
3028 * actually wait for the I/O to complete.
3030 return filemap_flush(inode
->i_mapping
);
3034 * bmap() is special. It gets used by applications such as lilo and by
3035 * the swapper to find the on-disk block of a specific piece of data.
3037 * Naturally, this is dangerous if the block concerned is still in the
3038 * journal. If somebody makes a swapfile on an ext4 data-journaling
3039 * filesystem and enables swap, then they may get a nasty shock when the
3040 * data getting swapped to that swapfile suddenly gets overwritten by
3041 * the original zero's written out previously to the journal and
3042 * awaiting writeback in the kernel's buffer cache.
3044 * So, if we see any bmap calls here on a modified, data-journaled file,
3045 * take extra steps to flush any blocks which might be in the cache.
3047 static sector_t
ext4_bmap(struct address_space
*mapping
, sector_t block
)
3049 struct inode
*inode
= mapping
->host
;
3053 if (mapping_tagged(mapping
, PAGECACHE_TAG_DIRTY
) &&
3054 test_opt(inode
->i_sb
, DELALLOC
)) {
3056 * With delalloc we want to sync the file
3057 * so that we can make sure we allocate
3060 filemap_write_and_wait(mapping
);
3063 if (EXT4_JOURNAL(inode
) && EXT4_I(inode
)->i_state
& EXT4_STATE_JDATA
) {
3065 * This is a REALLY heavyweight approach, but the use of
3066 * bmap on dirty files is expected to be extremely rare:
3067 * only if we run lilo or swapon on a freshly made file
3068 * do we expect this to happen.
3070 * (bmap requires CAP_SYS_RAWIO so this does not
3071 * represent an unprivileged user DOS attack --- we'd be
3072 * in trouble if mortal users could trigger this path at
3075 * NB. EXT4_STATE_JDATA is not set on files other than
3076 * regular files. If somebody wants to bmap a directory
3077 * or symlink and gets confused because the buffer
3078 * hasn't yet been flushed to disk, they deserve
3079 * everything they get.
3082 EXT4_I(inode
)->i_state
&= ~EXT4_STATE_JDATA
;
3083 journal
= EXT4_JOURNAL(inode
);
3084 jbd2_journal_lock_updates(journal
);
3085 err
= jbd2_journal_flush(journal
);
3086 jbd2_journal_unlock_updates(journal
);
3092 return generic_block_bmap(mapping
, block
, ext4_get_block
);
3095 static int bget_one(handle_t
*handle
, struct buffer_head
*bh
)
3101 static int bput_one(handle_t
*handle
, struct buffer_head
*bh
)
3108 * Note that we don't need to start a transaction unless we're journaling data
3109 * because we should have holes filled from ext4_page_mkwrite(). We even don't
3110 * need to file the inode to the transaction's list in ordered mode because if
3111 * we are writing back data added by write(), the inode is already there and if
3112 * we are writing back data modified via mmap(), noone guarantees in which
3113 * transaction the data will hit the disk. In case we are journaling data, we
3114 * cannot start transaction directly because transaction start ranks above page
3115 * lock so we have to do some magic.
3117 * In all journaling modes block_write_full_page() will start the I/O.
3121 * ext4_writepage() -> kmalloc() -> __alloc_pages() -> page_launder() ->
3126 * ext4_file_write() -> generic_file_write() -> __alloc_pages() -> ...
3128 * Same applies to ext4_get_block(). We will deadlock on various things like
3129 * lock_journal and i_data_sem
3131 * Setting PF_MEMALLOC here doesn't work - too many internal memory
3134 * 16May01: If we're reentered then journal_current_handle() will be
3135 * non-zero. We simply *return*.
3137 * 1 July 2001: @@@ FIXME:
3138 * In journalled data mode, a data buffer may be metadata against the
3139 * current transaction. But the same file is part of a shared mapping
3140 * and someone does a writepage() on it.
3142 * We will move the buffer onto the async_data list, but *after* it has
3143 * been dirtied. So there's a small window where we have dirty data on
3146 * Note that this only applies to the last partial page in the file. The
3147 * bit which block_write_full_page() uses prepare/commit for. (That's
3148 * broken code anyway: it's wrong for msync()).
3150 * It's a rare case: affects the final partial page, for journalled data
3151 * where the file is subject to bith write() and writepage() in the same
3152 * transction. To fix it we'll need a custom block_write_full_page().
3153 * We'll probably need that anyway for journalling writepage() output.
3155 * We don't honour synchronous mounts for writepage(). That would be
3156 * disastrous. Any write() or metadata operation will sync the fs for
3160 static int __ext4_normal_writepage(struct page
*page
,
3161 struct writeback_control
*wbc
)
3163 struct inode
*inode
= page
->mapping
->host
;
3165 if (test_opt(inode
->i_sb
, NOBH
))
3166 return nobh_writepage(page
, noalloc_get_block_write
, wbc
);
3168 return block_write_full_page(page
, noalloc_get_block_write
,
3172 static int ext4_normal_writepage(struct page
*page
,
3173 struct writeback_control
*wbc
)
3175 struct inode
*inode
= page
->mapping
->host
;
3176 loff_t size
= i_size_read(inode
);
3179 trace_mark(ext4_normal_writepage
,
3180 "dev %s ino %lu page_index %lu",
3181 inode
->i_sb
->s_id
, inode
->i_ino
, page
->index
);
3182 J_ASSERT(PageLocked(page
));
3183 if (page
->index
== size
>> PAGE_CACHE_SHIFT
)
3184 len
= size
& ~PAGE_CACHE_MASK
;
3186 len
= PAGE_CACHE_SIZE
;
3188 if (page_has_buffers(page
)) {
3189 /* if page has buffers it should all be mapped
3190 * and allocated. If there are not buffers attached
3191 * to the page we know the page is dirty but it lost
3192 * buffers. That means that at some moment in time
3193 * after write_begin() / write_end() has been called
3194 * all buffers have been clean and thus they must have been
3195 * written at least once. So they are all mapped and we can
3196 * happily proceed with mapping them and writing the page.
3198 BUG_ON(walk_page_buffers(NULL
, page_buffers(page
), 0, len
, NULL
,
3199 ext4_bh_unmapped_or_delay
));
3202 if (!ext4_journal_current_handle())
3203 return __ext4_normal_writepage(page
, wbc
);
3205 redirty_page_for_writepage(wbc
, page
);
3210 static int __ext4_journalled_writepage(struct page
*page
,
3211 struct writeback_control
*wbc
)
3213 struct address_space
*mapping
= page
->mapping
;
3214 struct inode
*inode
= mapping
->host
;
3215 struct buffer_head
*page_bufs
;
3216 handle_t
*handle
= NULL
;
3220 ret
= block_prepare_write(page
, 0, PAGE_CACHE_SIZE
,
3221 noalloc_get_block_write
);
3225 page_bufs
= page_buffers(page
);
3226 walk_page_buffers(handle
, page_bufs
, 0, PAGE_CACHE_SIZE
, NULL
,
3228 /* As soon as we unlock the page, it can go away, but we have
3229 * references to buffers so we are safe */
3232 handle
= ext4_journal_start(inode
, ext4_writepage_trans_blocks(inode
));
3233 if (IS_ERR(handle
)) {
3234 ret
= PTR_ERR(handle
);
3238 ret
= walk_page_buffers(handle
, page_bufs
, 0,
3239 PAGE_CACHE_SIZE
, NULL
, do_journal_get_write_access
);
3241 err
= walk_page_buffers(handle
, page_bufs
, 0,
3242 PAGE_CACHE_SIZE
, NULL
, write_end_fn
);
3245 err
= ext4_journal_stop(handle
);
3249 walk_page_buffers(handle
, page_bufs
, 0,
3250 PAGE_CACHE_SIZE
, NULL
, bput_one
);
3251 EXT4_I(inode
)->i_state
|= EXT4_STATE_JDATA
;
3260 static int ext4_journalled_writepage(struct page
*page
,
3261 struct writeback_control
*wbc
)
3263 struct inode
*inode
= page
->mapping
->host
;
3264 loff_t size
= i_size_read(inode
);
3267 trace_mark(ext4_journalled_writepage
,
3268 "dev %s ino %lu page_index %lu",
3269 inode
->i_sb
->s_id
, inode
->i_ino
, page
->index
);
3270 J_ASSERT(PageLocked(page
));
3271 if (page
->index
== size
>> PAGE_CACHE_SHIFT
)
3272 len
= size
& ~PAGE_CACHE_MASK
;
3274 len
= PAGE_CACHE_SIZE
;
3276 if (page_has_buffers(page
)) {
3277 /* if page has buffers it should all be mapped
3278 * and allocated. If there are not buffers attached
3279 * to the page we know the page is dirty but it lost
3280 * buffers. That means that at some moment in time
3281 * after write_begin() / write_end() has been called
3282 * all buffers have been clean and thus they must have been
3283 * written at least once. So they are all mapped and we can
3284 * happily proceed with mapping them and writing the page.
3286 BUG_ON(walk_page_buffers(NULL
, page_buffers(page
), 0, len
, NULL
,
3287 ext4_bh_unmapped_or_delay
));
3290 if (ext4_journal_current_handle())
3293 if (PageChecked(page
)) {
3295 * It's mmapped pagecache. Add buffers and journal it. There
3296 * doesn't seem much point in redirtying the page here.
3298 ClearPageChecked(page
);
3299 return __ext4_journalled_writepage(page
, wbc
);
3302 * It may be a page full of checkpoint-mode buffers. We don't
3303 * really know unless we go poke around in the buffer_heads.
3304 * But block_write_full_page will do the right thing.
3306 return block_write_full_page(page
, noalloc_get_block_write
,
3310 redirty_page_for_writepage(wbc
, page
);
3315 static int ext4_readpage(struct file
*file
, struct page
*page
)
3317 return mpage_readpage(page
, ext4_get_block
);
3321 ext4_readpages(struct file
*file
, struct address_space
*mapping
,
3322 struct list_head
*pages
, unsigned nr_pages
)
3324 return mpage_readpages(mapping
, pages
, nr_pages
, ext4_get_block
);
3327 static void ext4_invalidatepage(struct page
*page
, unsigned long offset
)
3329 journal_t
*journal
= EXT4_JOURNAL(page
->mapping
->host
);
3332 * If it's a full truncate we just forget about the pending dirtying
3335 ClearPageChecked(page
);
3338 jbd2_journal_invalidatepage(journal
, page
, offset
);
3340 block_invalidatepage(page
, offset
);
3343 static int ext4_releasepage(struct page
*page
, gfp_t wait
)
3345 journal_t
*journal
= EXT4_JOURNAL(page
->mapping
->host
);
3347 WARN_ON(PageChecked(page
));
3348 if (!page_has_buffers(page
))
3351 return jbd2_journal_try_to_free_buffers(journal
, page
, wait
);
3353 return try_to_free_buffers(page
);
3357 * If the O_DIRECT write will extend the file then add this inode to the
3358 * orphan list. So recovery will truncate it back to the original size
3359 * if the machine crashes during the write.
3361 * If the O_DIRECT write is intantiating holes inside i_size and the machine
3362 * crashes then stale disk data _may_ be exposed inside the file. But current
3363 * VFS code falls back into buffered path in that case so we are safe.
3365 static ssize_t
ext4_direct_IO(int rw
, struct kiocb
*iocb
,
3366 const struct iovec
*iov
, loff_t offset
,
3367 unsigned long nr_segs
)
3369 struct file
*file
= iocb
->ki_filp
;
3370 struct inode
*inode
= file
->f_mapping
->host
;
3371 struct ext4_inode_info
*ei
= EXT4_I(inode
);
3375 size_t count
= iov_length(iov
, nr_segs
);
3378 loff_t final_size
= offset
+ count
;
3380 if (final_size
> inode
->i_size
) {
3381 /* Credits for sb + inode write */
3382 handle
= ext4_journal_start(inode
, 2);
3383 if (IS_ERR(handle
)) {
3384 ret
= PTR_ERR(handle
);
3387 ret
= ext4_orphan_add(handle
, inode
);
3389 ext4_journal_stop(handle
);
3393 ei
->i_disksize
= inode
->i_size
;
3394 ext4_journal_stop(handle
);
3398 ret
= blockdev_direct_IO(rw
, iocb
, inode
, inode
->i_sb
->s_bdev
, iov
,
3400 ext4_get_block
, NULL
);
3405 /* Credits for sb + inode write */
3406 handle
= ext4_journal_start(inode
, 2);
3407 if (IS_ERR(handle
)) {
3408 /* This is really bad luck. We've written the data
3409 * but cannot extend i_size. Bail out and pretend
3410 * the write failed... */
3411 ret
= PTR_ERR(handle
);
3415 ext4_orphan_del(handle
, inode
);
3417 loff_t end
= offset
+ ret
;
3418 if (end
> inode
->i_size
) {
3419 ei
->i_disksize
= end
;
3420 i_size_write(inode
, end
);
3422 * We're going to return a positive `ret'
3423 * here due to non-zero-length I/O, so there's
3424 * no way of reporting error returns from
3425 * ext4_mark_inode_dirty() to userspace. So
3428 ext4_mark_inode_dirty(handle
, inode
);
3431 err
= ext4_journal_stop(handle
);
3440 * Pages can be marked dirty completely asynchronously from ext4's journalling
3441 * activity. By filemap_sync_pte(), try_to_unmap_one(), etc. We cannot do
3442 * much here because ->set_page_dirty is called under VFS locks. The page is
3443 * not necessarily locked.
3445 * We cannot just dirty the page and leave attached buffers clean, because the
3446 * buffers' dirty state is "definitive". We cannot just set the buffers dirty
3447 * or jbddirty because all the journalling code will explode.
3449 * So what we do is to mark the page "pending dirty" and next time writepage
3450 * is called, propagate that into the buffers appropriately.
3452 static int ext4_journalled_set_page_dirty(struct page
*page
)
3454 SetPageChecked(page
);
3455 return __set_page_dirty_nobuffers(page
);
3458 static const struct address_space_operations ext4_ordered_aops
= {
3459 .readpage
= ext4_readpage
,
3460 .readpages
= ext4_readpages
,
3461 .writepage
= ext4_normal_writepage
,
3462 .sync_page
= block_sync_page
,
3463 .write_begin
= ext4_write_begin
,
3464 .write_end
= ext4_ordered_write_end
,
3466 .invalidatepage
= ext4_invalidatepage
,
3467 .releasepage
= ext4_releasepage
,
3468 .direct_IO
= ext4_direct_IO
,
3469 .migratepage
= buffer_migrate_page
,
3470 .is_partially_uptodate
= block_is_partially_uptodate
,
3473 static const struct address_space_operations ext4_writeback_aops
= {
3474 .readpage
= ext4_readpage
,
3475 .readpages
= ext4_readpages
,
3476 .writepage
= ext4_normal_writepage
,
3477 .sync_page
= block_sync_page
,
3478 .write_begin
= ext4_write_begin
,
3479 .write_end
= ext4_writeback_write_end
,
3481 .invalidatepage
= ext4_invalidatepage
,
3482 .releasepage
= ext4_releasepage
,
3483 .direct_IO
= ext4_direct_IO
,
3484 .migratepage
= buffer_migrate_page
,
3485 .is_partially_uptodate
= block_is_partially_uptodate
,
3488 static const struct address_space_operations ext4_journalled_aops
= {
3489 .readpage
= ext4_readpage
,
3490 .readpages
= ext4_readpages
,
3491 .writepage
= ext4_journalled_writepage
,
3492 .sync_page
= block_sync_page
,
3493 .write_begin
= ext4_write_begin
,
3494 .write_end
= ext4_journalled_write_end
,
3495 .set_page_dirty
= ext4_journalled_set_page_dirty
,
3497 .invalidatepage
= ext4_invalidatepage
,
3498 .releasepage
= ext4_releasepage
,
3499 .is_partially_uptodate
= block_is_partially_uptodate
,
3502 static const struct address_space_operations ext4_da_aops
= {
3503 .readpage
= ext4_readpage
,
3504 .readpages
= ext4_readpages
,
3505 .writepage
= ext4_da_writepage
,
3506 .writepages
= ext4_da_writepages
,
3507 .sync_page
= block_sync_page
,
3508 .write_begin
= ext4_da_write_begin
,
3509 .write_end
= ext4_da_write_end
,
3511 .invalidatepage
= ext4_da_invalidatepage
,
3512 .releasepage
= ext4_releasepage
,
3513 .direct_IO
= ext4_direct_IO
,
3514 .migratepage
= buffer_migrate_page
,
3515 .is_partially_uptodate
= block_is_partially_uptodate
,
3518 void ext4_set_aops(struct inode
*inode
)
3520 if (ext4_should_order_data(inode
) &&
3521 test_opt(inode
->i_sb
, DELALLOC
))
3522 inode
->i_mapping
->a_ops
= &ext4_da_aops
;
3523 else if (ext4_should_order_data(inode
))
3524 inode
->i_mapping
->a_ops
= &ext4_ordered_aops
;
3525 else if (ext4_should_writeback_data(inode
) &&
3526 test_opt(inode
->i_sb
, DELALLOC
))
3527 inode
->i_mapping
->a_ops
= &ext4_da_aops
;
3528 else if (ext4_should_writeback_data(inode
))
3529 inode
->i_mapping
->a_ops
= &ext4_writeback_aops
;
3531 inode
->i_mapping
->a_ops
= &ext4_journalled_aops
;
3535 * ext4_block_truncate_page() zeroes out a mapping from file offset `from'
3536 * up to the end of the block which corresponds to `from'.
3537 * This required during truncate. We need to physically zero the tail end
3538 * of that block so it doesn't yield old data if the file is later grown.
3540 int ext4_block_truncate_page(handle_t
*handle
,
3541 struct address_space
*mapping
, loff_t from
)
3543 ext4_fsblk_t index
= from
>> PAGE_CACHE_SHIFT
;
3544 unsigned offset
= from
& (PAGE_CACHE_SIZE
-1);
3545 unsigned blocksize
, length
, pos
;
3547 struct inode
*inode
= mapping
->host
;
3548 struct buffer_head
*bh
;
3552 page
= grab_cache_page(mapping
, from
>> PAGE_CACHE_SHIFT
);
3556 blocksize
= inode
->i_sb
->s_blocksize
;
3557 length
= blocksize
- (offset
& (blocksize
- 1));
3558 iblock
= index
<< (PAGE_CACHE_SHIFT
- inode
->i_sb
->s_blocksize_bits
);
3561 * For "nobh" option, we can only work if we don't need to
3562 * read-in the page - otherwise we create buffers to do the IO.
3564 if (!page_has_buffers(page
) && test_opt(inode
->i_sb
, NOBH
) &&
3565 ext4_should_writeback_data(inode
) && PageUptodate(page
)) {
3566 zero_user(page
, offset
, length
);
3567 set_page_dirty(page
);
3571 if (!page_has_buffers(page
))
3572 create_empty_buffers(page
, blocksize
, 0);
3574 /* Find the buffer that contains "offset" */
3575 bh
= page_buffers(page
);
3577 while (offset
>= pos
) {
3578 bh
= bh
->b_this_page
;
3584 if (buffer_freed(bh
)) {
3585 BUFFER_TRACE(bh
, "freed: skip");
3589 if (!buffer_mapped(bh
)) {
3590 BUFFER_TRACE(bh
, "unmapped");
3591 ext4_get_block(inode
, iblock
, bh
, 0);
3592 /* unmapped? It's a hole - nothing to do */
3593 if (!buffer_mapped(bh
)) {
3594 BUFFER_TRACE(bh
, "still unmapped");
3599 /* Ok, it's mapped. Make sure it's up-to-date */
3600 if (PageUptodate(page
))
3601 set_buffer_uptodate(bh
);
3603 if (!buffer_uptodate(bh
)) {
3605 ll_rw_block(READ
, 1, &bh
);
3607 /* Uhhuh. Read error. Complain and punt. */
3608 if (!buffer_uptodate(bh
))
3612 if (ext4_should_journal_data(inode
)) {
3613 BUFFER_TRACE(bh
, "get write access");
3614 err
= ext4_journal_get_write_access(handle
, bh
);
3619 zero_user(page
, offset
, length
);
3621 BUFFER_TRACE(bh
, "zeroed end of block");
3624 if (ext4_should_journal_data(inode
)) {
3625 err
= ext4_handle_dirty_metadata(handle
, inode
, bh
);
3627 if (ext4_should_order_data(inode
))
3628 err
= ext4_jbd2_file_inode(handle
, inode
);
3629 mark_buffer_dirty(bh
);
3634 page_cache_release(page
);
3639 * Probably it should be a library function... search for first non-zero word
3640 * or memcmp with zero_page, whatever is better for particular architecture.
3643 static inline int all_zeroes(__le32
*p
, __le32
*q
)
3652 * ext4_find_shared - find the indirect blocks for partial truncation.
3653 * @inode: inode in question
3654 * @depth: depth of the affected branch
3655 * @offsets: offsets of pointers in that branch (see ext4_block_to_path)
3656 * @chain: place to store the pointers to partial indirect blocks
3657 * @top: place to the (detached) top of branch
3659 * This is a helper function used by ext4_truncate().
3661 * When we do truncate() we may have to clean the ends of several
3662 * indirect blocks but leave the blocks themselves alive. Block is
3663 * partially truncated if some data below the new i_size is refered
3664 * from it (and it is on the path to the first completely truncated
3665 * data block, indeed). We have to free the top of that path along
3666 * with everything to the right of the path. Since no allocation
3667 * past the truncation point is possible until ext4_truncate()
3668 * finishes, we may safely do the latter, but top of branch may
3669 * require special attention - pageout below the truncation point
3670 * might try to populate it.
3672 * We atomically detach the top of branch from the tree, store the
3673 * block number of its root in *@top, pointers to buffer_heads of
3674 * partially truncated blocks - in @chain[].bh and pointers to
3675 * their last elements that should not be removed - in
3676 * @chain[].p. Return value is the pointer to last filled element
3679 * The work left to caller to do the actual freeing of subtrees:
3680 * a) free the subtree starting from *@top
3681 * b) free the subtrees whose roots are stored in
3682 * (@chain[i].p+1 .. end of @chain[i].bh->b_data)
3683 * c) free the subtrees growing from the inode past the @chain[0].
3684 * (no partially truncated stuff there). */
3686 static Indirect
*ext4_find_shared(struct inode
*inode
, int depth
,
3687 ext4_lblk_t offsets
[4], Indirect chain
[4], __le32
*top
)
3689 Indirect
*partial
, *p
;
3693 /* Make k index the deepest non-null offest + 1 */
3694 for (k
= depth
; k
> 1 && !offsets
[k
-1]; k
--)
3696 partial
= ext4_get_branch(inode
, k
, offsets
, chain
, &err
);
3697 /* Writer: pointers */
3699 partial
= chain
+ k
-1;
3701 * If the branch acquired continuation since we've looked at it -
3702 * fine, it should all survive and (new) top doesn't belong to us.
3704 if (!partial
->key
&& *partial
->p
)
3707 for (p
= partial
; (p
> chain
) && all_zeroes((__le32
*) p
->bh
->b_data
, p
->p
); p
--)
3710 * OK, we've found the last block that must survive. The rest of our
3711 * branch should be detached before unlocking. However, if that rest
3712 * of branch is all ours and does not grow immediately from the inode
3713 * it's easier to cheat and just decrement partial->p.
3715 if (p
== chain
+ k
- 1 && p
> chain
) {
3719 /* Nope, don't do this in ext4. Must leave the tree intact */
3726 while (partial
> p
) {
3727 brelse(partial
->bh
);
3735 * Zero a number of block pointers in either an inode or an indirect block.
3736 * If we restart the transaction we must again get write access to the
3737 * indirect block for further modification.
3739 * We release `count' blocks on disk, but (last - first) may be greater
3740 * than `count' because there can be holes in there.
3742 static void ext4_clear_blocks(handle_t
*handle
, struct inode
*inode
,
3743 struct buffer_head
*bh
, ext4_fsblk_t block_to_free
,
3744 unsigned long count
, __le32
*first
, __le32
*last
)
3747 if (try_to_extend_transaction(handle
, inode
)) {
3749 BUFFER_TRACE(bh
, "call ext4_handle_dirty_metadata");
3750 ext4_handle_dirty_metadata(handle
, inode
, bh
);
3752 ext4_mark_inode_dirty(handle
, inode
);
3753 ext4_journal_test_restart(handle
, inode
);
3755 BUFFER_TRACE(bh
, "retaking write access");
3756 ext4_journal_get_write_access(handle
, bh
);
3761 * Any buffers which are on the journal will be in memory. We find
3762 * them on the hash table so jbd2_journal_revoke() will run jbd2_journal_forget()
3763 * on them. We've already detached each block from the file, so
3764 * bforget() in jbd2_journal_forget() should be safe.
3766 * AKPM: turn on bforget in jbd2_journal_forget()!!!
3768 for (p
= first
; p
< last
; p
++) {
3769 u32 nr
= le32_to_cpu(*p
);
3771 struct buffer_head
*tbh
;
3774 tbh
= sb_find_get_block(inode
->i_sb
, nr
);
3775 ext4_forget(handle
, 0, inode
, tbh
, nr
);
3779 ext4_free_blocks(handle
, inode
, block_to_free
, count
, 0);
3783 * ext4_free_data - free a list of data blocks
3784 * @handle: handle for this transaction
3785 * @inode: inode we are dealing with
3786 * @this_bh: indirect buffer_head which contains *@first and *@last
3787 * @first: array of block numbers
3788 * @last: points immediately past the end of array
3790 * We are freeing all blocks refered from that array (numbers are stored as
3791 * little-endian 32-bit) and updating @inode->i_blocks appropriately.
3793 * We accumulate contiguous runs of blocks to free. Conveniently, if these
3794 * blocks are contiguous then releasing them at one time will only affect one
3795 * or two bitmap blocks (+ group descriptor(s) and superblock) and we won't
3796 * actually use a lot of journal space.
3798 * @this_bh will be %NULL if @first and @last point into the inode's direct
3801 static void ext4_free_data(handle_t
*handle
, struct inode
*inode
,
3802 struct buffer_head
*this_bh
,
3803 __le32
*first
, __le32
*last
)
3805 ext4_fsblk_t block_to_free
= 0; /* Starting block # of a run */
3806 unsigned long count
= 0; /* Number of blocks in the run */
3807 __le32
*block_to_free_p
= NULL
; /* Pointer into inode/ind
3810 ext4_fsblk_t nr
; /* Current block # */
3811 __le32
*p
; /* Pointer into inode/ind
3812 for current block */
3815 if (this_bh
) { /* For indirect block */
3816 BUFFER_TRACE(this_bh
, "get_write_access");
3817 err
= ext4_journal_get_write_access(handle
, this_bh
);
3818 /* Important: if we can't update the indirect pointers
3819 * to the blocks, we can't free them. */
3824 for (p
= first
; p
< last
; p
++) {
3825 nr
= le32_to_cpu(*p
);
3827 /* accumulate blocks to free if they're contiguous */
3830 block_to_free_p
= p
;
3832 } else if (nr
== block_to_free
+ count
) {
3835 ext4_clear_blocks(handle
, inode
, this_bh
,
3837 count
, block_to_free_p
, p
);
3839 block_to_free_p
= p
;
3846 ext4_clear_blocks(handle
, inode
, this_bh
, block_to_free
,
3847 count
, block_to_free_p
, p
);
3850 BUFFER_TRACE(this_bh
, "call ext4_handle_dirty_metadata");
3853 * The buffer head should have an attached journal head at this
3854 * point. However, if the data is corrupted and an indirect
3855 * block pointed to itself, it would have been detached when
3856 * the block was cleared. Check for this instead of OOPSing.
3858 if ((EXT4_JOURNAL(inode
) == NULL
) || bh2jh(this_bh
))
3859 ext4_handle_dirty_metadata(handle
, inode
, this_bh
);
3861 ext4_error(inode
->i_sb
, __func__
,
3862 "circular indirect block detected, "
3863 "inode=%lu, block=%llu",
3865 (unsigned long long) this_bh
->b_blocknr
);
3870 * ext4_free_branches - free an array of branches
3871 * @handle: JBD handle for this transaction
3872 * @inode: inode we are dealing with
3873 * @parent_bh: the buffer_head which contains *@first and *@last
3874 * @first: array of block numbers
3875 * @last: pointer immediately past the end of array
3876 * @depth: depth of the branches to free
3878 * We are freeing all blocks refered from these branches (numbers are
3879 * stored as little-endian 32-bit) and updating @inode->i_blocks
3882 static void ext4_free_branches(handle_t
*handle
, struct inode
*inode
,
3883 struct buffer_head
*parent_bh
,
3884 __le32
*first
, __le32
*last
, int depth
)
3889 if (ext4_handle_is_aborted(handle
))
3893 struct buffer_head
*bh
;
3894 int addr_per_block
= EXT4_ADDR_PER_BLOCK(inode
->i_sb
);
3896 while (--p
>= first
) {
3897 nr
= le32_to_cpu(*p
);
3899 continue; /* A hole */
3901 /* Go read the buffer for the next level down */
3902 bh
= sb_bread(inode
->i_sb
, nr
);
3905 * A read failure? Report error and clear slot
3909 ext4_error(inode
->i_sb
, "ext4_free_branches",
3910 "Read failure, inode=%lu, block=%llu",
3915 /* This zaps the entire block. Bottom up. */
3916 BUFFER_TRACE(bh
, "free child branches");
3917 ext4_free_branches(handle
, inode
, bh
,
3918 (__le32
*) bh
->b_data
,
3919 (__le32
*) bh
->b_data
+ addr_per_block
,
3923 * We've probably journalled the indirect block several
3924 * times during the truncate. But it's no longer
3925 * needed and we now drop it from the transaction via
3926 * jbd2_journal_revoke().
3928 * That's easy if it's exclusively part of this
3929 * transaction. But if it's part of the committing
3930 * transaction then jbd2_journal_forget() will simply
3931 * brelse() it. That means that if the underlying
3932 * block is reallocated in ext4_get_block(),
3933 * unmap_underlying_metadata() will find this block
3934 * and will try to get rid of it. damn, damn.
3936 * If this block has already been committed to the
3937 * journal, a revoke record will be written. And
3938 * revoke records must be emitted *before* clearing
3939 * this block's bit in the bitmaps.
3941 ext4_forget(handle
, 1, inode
, bh
, bh
->b_blocknr
);
3944 * Everything below this this pointer has been
3945 * released. Now let this top-of-subtree go.
3947 * We want the freeing of this indirect block to be
3948 * atomic in the journal with the updating of the
3949 * bitmap block which owns it. So make some room in
3952 * We zero the parent pointer *after* freeing its
3953 * pointee in the bitmaps, so if extend_transaction()
3954 * for some reason fails to put the bitmap changes and
3955 * the release into the same transaction, recovery
3956 * will merely complain about releasing a free block,
3957 * rather than leaking blocks.
3959 if (ext4_handle_is_aborted(handle
))
3961 if (try_to_extend_transaction(handle
, inode
)) {
3962 ext4_mark_inode_dirty(handle
, inode
);
3963 ext4_journal_test_restart(handle
, inode
);
3966 ext4_free_blocks(handle
, inode
, nr
, 1, 1);
3970 * The block which we have just freed is
3971 * pointed to by an indirect block: journal it
3973 BUFFER_TRACE(parent_bh
, "get_write_access");
3974 if (!ext4_journal_get_write_access(handle
,
3977 BUFFER_TRACE(parent_bh
,
3978 "call ext4_handle_dirty_metadata");
3979 ext4_handle_dirty_metadata(handle
,
3986 /* We have reached the bottom of the tree. */
3987 BUFFER_TRACE(parent_bh
, "free data blocks");
3988 ext4_free_data(handle
, inode
, parent_bh
, first
, last
);
3992 int ext4_can_truncate(struct inode
*inode
)
3994 if (IS_APPEND(inode
) || IS_IMMUTABLE(inode
))
3996 if (S_ISREG(inode
->i_mode
))
3998 if (S_ISDIR(inode
->i_mode
))
4000 if (S_ISLNK(inode
->i_mode
))
4001 return !ext4_inode_is_fast_symlink(inode
);
4008 * We block out ext4_get_block() block instantiations across the entire
4009 * transaction, and VFS/VM ensures that ext4_truncate() cannot run
4010 * simultaneously on behalf of the same inode.
4012 * As we work through the truncate and commmit bits of it to the journal there
4013 * is one core, guiding principle: the file's tree must always be consistent on
4014 * disk. We must be able to restart the truncate after a crash.
4016 * The file's tree may be transiently inconsistent in memory (although it
4017 * probably isn't), but whenever we close off and commit a journal transaction,
4018 * the contents of (the filesystem + the journal) must be consistent and
4019 * restartable. It's pretty simple, really: bottom up, right to left (although
4020 * left-to-right works OK too).
4022 * Note that at recovery time, journal replay occurs *before* the restart of
4023 * truncate against the orphan inode list.
4025 * The committed inode has the new, desired i_size (which is the same as
4026 * i_disksize in this case). After a crash, ext4_orphan_cleanup() will see
4027 * that this inode's truncate did not complete and it will again call
4028 * ext4_truncate() to have another go. So there will be instantiated blocks
4029 * to the right of the truncation point in a crashed ext4 filesystem. But
4030 * that's fine - as long as they are linked from the inode, the post-crash
4031 * ext4_truncate() run will find them and release them.
4033 void ext4_truncate(struct inode
*inode
)
4036 struct ext4_inode_info
*ei
= EXT4_I(inode
);
4037 __le32
*i_data
= ei
->i_data
;
4038 int addr_per_block
= EXT4_ADDR_PER_BLOCK(inode
->i_sb
);
4039 struct address_space
*mapping
= inode
->i_mapping
;
4040 ext4_lblk_t offsets
[4];
4045 ext4_lblk_t last_block
;
4046 unsigned blocksize
= inode
->i_sb
->s_blocksize
;
4048 if (!ext4_can_truncate(inode
))
4051 if (inode
->i_size
== 0 && !test_opt(inode
->i_sb
, NO_AUTO_DA_ALLOC
))
4052 ei
->i_state
|= EXT4_STATE_DA_ALLOC_CLOSE
;
4054 if (EXT4_I(inode
)->i_flags
& EXT4_EXTENTS_FL
) {
4055 ext4_ext_truncate(inode
);
4059 handle
= start_transaction(inode
);
4061 return; /* AKPM: return what? */
4063 last_block
= (inode
->i_size
+ blocksize
-1)
4064 >> EXT4_BLOCK_SIZE_BITS(inode
->i_sb
);
4066 if (inode
->i_size
& (blocksize
- 1))
4067 if (ext4_block_truncate_page(handle
, mapping
, inode
->i_size
))
4070 n
= ext4_block_to_path(inode
, last_block
, offsets
, NULL
);
4072 goto out_stop
; /* error */
4075 * OK. This truncate is going to happen. We add the inode to the
4076 * orphan list, so that if this truncate spans multiple transactions,
4077 * and we crash, we will resume the truncate when the filesystem
4078 * recovers. It also marks the inode dirty, to catch the new size.
4080 * Implication: the file must always be in a sane, consistent
4081 * truncatable state while each transaction commits.
4083 if (ext4_orphan_add(handle
, inode
))
4087 * From here we block out all ext4_get_block() callers who want to
4088 * modify the block allocation tree.
4090 down_write(&ei
->i_data_sem
);
4092 ext4_discard_preallocations(inode
);
4095 * The orphan list entry will now protect us from any crash which
4096 * occurs before the truncate completes, so it is now safe to propagate
4097 * the new, shorter inode size (held for now in i_size) into the
4098 * on-disk inode. We do this via i_disksize, which is the value which
4099 * ext4 *really* writes onto the disk inode.
4101 ei
->i_disksize
= inode
->i_size
;
4103 if (n
== 1) { /* direct blocks */
4104 ext4_free_data(handle
, inode
, NULL
, i_data
+offsets
[0],
4105 i_data
+ EXT4_NDIR_BLOCKS
);
4109 partial
= ext4_find_shared(inode
, n
, offsets
, chain
, &nr
);
4110 /* Kill the top of shared branch (not detached) */
4112 if (partial
== chain
) {
4113 /* Shared branch grows from the inode */
4114 ext4_free_branches(handle
, inode
, NULL
,
4115 &nr
, &nr
+1, (chain
+n
-1) - partial
);
4118 * We mark the inode dirty prior to restart,
4119 * and prior to stop. No need for it here.
4122 /* Shared branch grows from an indirect block */
4123 BUFFER_TRACE(partial
->bh
, "get_write_access");
4124 ext4_free_branches(handle
, inode
, partial
->bh
,
4126 partial
->p
+1, (chain
+n
-1) - partial
);
4129 /* Clear the ends of indirect blocks on the shared branch */
4130 while (partial
> chain
) {
4131 ext4_free_branches(handle
, inode
, partial
->bh
, partial
->p
+ 1,
4132 (__le32
*)partial
->bh
->b_data
+addr_per_block
,
4133 (chain
+n
-1) - partial
);
4134 BUFFER_TRACE(partial
->bh
, "call brelse");
4135 brelse (partial
->bh
);
4139 /* Kill the remaining (whole) subtrees */
4140 switch (offsets
[0]) {
4142 nr
= i_data
[EXT4_IND_BLOCK
];
4144 ext4_free_branches(handle
, inode
, NULL
, &nr
, &nr
+1, 1);
4145 i_data
[EXT4_IND_BLOCK
] = 0;
4147 case EXT4_IND_BLOCK
:
4148 nr
= i_data
[EXT4_DIND_BLOCK
];
4150 ext4_free_branches(handle
, inode
, NULL
, &nr
, &nr
+1, 2);
4151 i_data
[EXT4_DIND_BLOCK
] = 0;
4153 case EXT4_DIND_BLOCK
:
4154 nr
= i_data
[EXT4_TIND_BLOCK
];
4156 ext4_free_branches(handle
, inode
, NULL
, &nr
, &nr
+1, 3);
4157 i_data
[EXT4_TIND_BLOCK
] = 0;
4159 case EXT4_TIND_BLOCK
:
4163 up_write(&ei
->i_data_sem
);
4164 inode
->i_mtime
= inode
->i_ctime
= ext4_current_time(inode
);
4165 ext4_mark_inode_dirty(handle
, inode
);
4168 * In a multi-transaction truncate, we only make the final transaction
4172 ext4_handle_sync(handle
);
4175 * If this was a simple ftruncate(), and the file will remain alive
4176 * then we need to clear up the orphan record which we created above.
4177 * However, if this was a real unlink then we were called by
4178 * ext4_delete_inode(), and we allow that function to clean up the
4179 * orphan info for us.
4182 ext4_orphan_del(handle
, inode
);
4184 ext4_journal_stop(handle
);
4188 * ext4_get_inode_loc returns with an extra refcount against the inode's
4189 * underlying buffer_head on success. If 'in_mem' is true, we have all
4190 * data in memory that is needed to recreate the on-disk version of this
4193 static int __ext4_get_inode_loc(struct inode
*inode
,
4194 struct ext4_iloc
*iloc
, int in_mem
)
4196 struct ext4_group_desc
*gdp
;
4197 struct buffer_head
*bh
;
4198 struct super_block
*sb
= inode
->i_sb
;
4200 int inodes_per_block
, inode_offset
;
4203 if (!ext4_valid_inum(sb
, inode
->i_ino
))
4206 iloc
->block_group
= (inode
->i_ino
- 1) / EXT4_INODES_PER_GROUP(sb
);
4207 gdp
= ext4_get_group_desc(sb
, iloc
->block_group
, NULL
);
4212 * Figure out the offset within the block group inode table
4214 inodes_per_block
= (EXT4_BLOCK_SIZE(sb
) / EXT4_INODE_SIZE(sb
));
4215 inode_offset
= ((inode
->i_ino
- 1) %
4216 EXT4_INODES_PER_GROUP(sb
));
4217 block
= ext4_inode_table(sb
, gdp
) + (inode_offset
/ inodes_per_block
);
4218 iloc
->offset
= (inode_offset
% inodes_per_block
) * EXT4_INODE_SIZE(sb
);
4220 bh
= sb_getblk(sb
, block
);
4222 ext4_error(sb
, "ext4_get_inode_loc", "unable to read "
4223 "inode block - inode=%lu, block=%llu",
4224 inode
->i_ino
, block
);
4227 if (!buffer_uptodate(bh
)) {
4231 * If the buffer has the write error flag, we have failed
4232 * to write out another inode in the same block. In this
4233 * case, we don't have to read the block because we may
4234 * read the old inode data successfully.
4236 if (buffer_write_io_error(bh
) && !buffer_uptodate(bh
))
4237 set_buffer_uptodate(bh
);
4239 if (buffer_uptodate(bh
)) {
4240 /* someone brought it uptodate while we waited */
4246 * If we have all information of the inode in memory and this
4247 * is the only valid inode in the block, we need not read the
4251 struct buffer_head
*bitmap_bh
;
4254 start
= inode_offset
& ~(inodes_per_block
- 1);
4256 /* Is the inode bitmap in cache? */
4257 bitmap_bh
= sb_getblk(sb
, ext4_inode_bitmap(sb
, gdp
));
4262 * If the inode bitmap isn't in cache then the
4263 * optimisation may end up performing two reads instead
4264 * of one, so skip it.
4266 if (!buffer_uptodate(bitmap_bh
)) {
4270 for (i
= start
; i
< start
+ inodes_per_block
; i
++) {
4271 if (i
== inode_offset
)
4273 if (ext4_test_bit(i
, bitmap_bh
->b_data
))
4277 if (i
== start
+ inodes_per_block
) {
4278 /* all other inodes are free, so skip I/O */
4279 memset(bh
->b_data
, 0, bh
->b_size
);
4280 set_buffer_uptodate(bh
);
4288 * If we need to do any I/O, try to pre-readahead extra
4289 * blocks from the inode table.
4291 if (EXT4_SB(sb
)->s_inode_readahead_blks
) {
4292 ext4_fsblk_t b
, end
, table
;
4295 table
= ext4_inode_table(sb
, gdp
);
4296 /* s_inode_readahead_blks is always a power of 2 */
4297 b
= block
& ~(EXT4_SB(sb
)->s_inode_readahead_blks
-1);
4300 end
= b
+ EXT4_SB(sb
)->s_inode_readahead_blks
;
4301 num
= EXT4_INODES_PER_GROUP(sb
);
4302 if (EXT4_HAS_RO_COMPAT_FEATURE(sb
,
4303 EXT4_FEATURE_RO_COMPAT_GDT_CSUM
))
4304 num
-= ext4_itable_unused_count(sb
, gdp
);
4305 table
+= num
/ inodes_per_block
;
4309 sb_breadahead(sb
, b
++);
4313 * There are other valid inodes in the buffer, this inode
4314 * has in-inode xattrs, or we don't have this inode in memory.
4315 * Read the block from disk.
4318 bh
->b_end_io
= end_buffer_read_sync
;
4319 submit_bh(READ_META
, bh
);
4321 if (!buffer_uptodate(bh
)) {
4322 ext4_error(sb
, __func__
,
4323 "unable to read inode block - inode=%lu, "
4324 "block=%llu", inode
->i_ino
, block
);
4334 int ext4_get_inode_loc(struct inode
*inode
, struct ext4_iloc
*iloc
)
4336 /* We have all inode data except xattrs in memory here. */
4337 return __ext4_get_inode_loc(inode
, iloc
,
4338 !(EXT4_I(inode
)->i_state
& EXT4_STATE_XATTR
));
4341 void ext4_set_inode_flags(struct inode
*inode
)
4343 unsigned int flags
= EXT4_I(inode
)->i_flags
;
4345 inode
->i_flags
&= ~(S_SYNC
|S_APPEND
|S_IMMUTABLE
|S_NOATIME
|S_DIRSYNC
);
4346 if (flags
& EXT4_SYNC_FL
)
4347 inode
->i_flags
|= S_SYNC
;
4348 if (flags
& EXT4_APPEND_FL
)
4349 inode
->i_flags
|= S_APPEND
;
4350 if (flags
& EXT4_IMMUTABLE_FL
)
4351 inode
->i_flags
|= S_IMMUTABLE
;
4352 if (flags
& EXT4_NOATIME_FL
)
4353 inode
->i_flags
|= S_NOATIME
;
4354 if (flags
& EXT4_DIRSYNC_FL
)
4355 inode
->i_flags
|= S_DIRSYNC
;
4358 /* Propagate flags from i_flags to EXT4_I(inode)->i_flags */
4359 void ext4_get_inode_flags(struct ext4_inode_info
*ei
)
4361 unsigned int flags
= ei
->vfs_inode
.i_flags
;
4363 ei
->i_flags
&= ~(EXT4_SYNC_FL
|EXT4_APPEND_FL
|
4364 EXT4_IMMUTABLE_FL
|EXT4_NOATIME_FL
|EXT4_DIRSYNC_FL
);
4366 ei
->i_flags
|= EXT4_SYNC_FL
;
4367 if (flags
& S_APPEND
)
4368 ei
->i_flags
|= EXT4_APPEND_FL
;
4369 if (flags
& S_IMMUTABLE
)
4370 ei
->i_flags
|= EXT4_IMMUTABLE_FL
;
4371 if (flags
& S_NOATIME
)
4372 ei
->i_flags
|= EXT4_NOATIME_FL
;
4373 if (flags
& S_DIRSYNC
)
4374 ei
->i_flags
|= EXT4_DIRSYNC_FL
;
4376 static blkcnt_t
ext4_inode_blocks(struct ext4_inode
*raw_inode
,
4377 struct ext4_inode_info
*ei
)
4380 struct inode
*inode
= &(ei
->vfs_inode
);
4381 struct super_block
*sb
= inode
->i_sb
;
4383 if (EXT4_HAS_RO_COMPAT_FEATURE(sb
,
4384 EXT4_FEATURE_RO_COMPAT_HUGE_FILE
)) {
4385 /* we are using combined 48 bit field */
4386 i_blocks
= ((u64
)le16_to_cpu(raw_inode
->i_blocks_high
)) << 32 |
4387 le32_to_cpu(raw_inode
->i_blocks_lo
);
4388 if (ei
->i_flags
& EXT4_HUGE_FILE_FL
) {
4389 /* i_blocks represent file system block size */
4390 return i_blocks
<< (inode
->i_blkbits
- 9);
4395 return le32_to_cpu(raw_inode
->i_blocks_lo
);
4399 struct inode
*ext4_iget(struct super_block
*sb
, unsigned long ino
)
4401 struct ext4_iloc iloc
;
4402 struct ext4_inode
*raw_inode
;
4403 struct ext4_inode_info
*ei
;
4404 struct buffer_head
*bh
;
4405 struct inode
*inode
;
4409 inode
= iget_locked(sb
, ino
);
4411 return ERR_PTR(-ENOMEM
);
4412 if (!(inode
->i_state
& I_NEW
))
4416 #ifdef CONFIG_EXT4_FS_POSIX_ACL
4417 ei
->i_acl
= EXT4_ACL_NOT_CACHED
;
4418 ei
->i_default_acl
= EXT4_ACL_NOT_CACHED
;
4421 ret
= __ext4_get_inode_loc(inode
, &iloc
, 0);
4425 raw_inode
= ext4_raw_inode(&iloc
);
4426 inode
->i_mode
= le16_to_cpu(raw_inode
->i_mode
);
4427 inode
->i_uid
= (uid_t
)le16_to_cpu(raw_inode
->i_uid_low
);
4428 inode
->i_gid
= (gid_t
)le16_to_cpu(raw_inode
->i_gid_low
);
4429 if (!(test_opt(inode
->i_sb
, NO_UID32
))) {
4430 inode
->i_uid
|= le16_to_cpu(raw_inode
->i_uid_high
) << 16;
4431 inode
->i_gid
|= le16_to_cpu(raw_inode
->i_gid_high
) << 16;
4433 inode
->i_nlink
= le16_to_cpu(raw_inode
->i_links_count
);
4436 ei
->i_dir_start_lookup
= 0;
4437 ei
->i_dtime
= le32_to_cpu(raw_inode
->i_dtime
);
4438 /* We now have enough fields to check if the inode was active or not.
4439 * This is needed because nfsd might try to access dead inodes
4440 * the test is that same one that e2fsck uses
4441 * NeilBrown 1999oct15
4443 if (inode
->i_nlink
== 0) {
4444 if (inode
->i_mode
== 0 ||
4445 !(EXT4_SB(inode
->i_sb
)->s_mount_state
& EXT4_ORPHAN_FS
)) {
4446 /* this inode is deleted */
4451 /* The only unlinked inodes we let through here have
4452 * valid i_mode and are being read by the orphan
4453 * recovery code: that's fine, we're about to complete
4454 * the process of deleting those. */
4456 ei
->i_flags
= le32_to_cpu(raw_inode
->i_flags
);
4457 inode
->i_blocks
= ext4_inode_blocks(raw_inode
, ei
);
4458 ei
->i_file_acl
= le32_to_cpu(raw_inode
->i_file_acl_lo
);
4459 if (EXT4_HAS_INCOMPAT_FEATURE(sb
, EXT4_FEATURE_INCOMPAT_64BIT
))
4461 ((__u64
)le16_to_cpu(raw_inode
->i_file_acl_high
)) << 32;
4462 inode
->i_size
= ext4_isize(raw_inode
);
4463 ei
->i_disksize
= inode
->i_size
;
4464 inode
->i_generation
= le32_to_cpu(raw_inode
->i_generation
);
4465 ei
->i_block_group
= iloc
.block_group
;
4466 ei
->i_last_alloc_group
= ~0;
4468 * NOTE! The in-memory inode i_data array is in little-endian order
4469 * even on big-endian machines: we do NOT byteswap the block numbers!
4471 for (block
= 0; block
< EXT4_N_BLOCKS
; block
++)
4472 ei
->i_data
[block
] = raw_inode
->i_block
[block
];
4473 INIT_LIST_HEAD(&ei
->i_orphan
);
4475 if (EXT4_INODE_SIZE(inode
->i_sb
) > EXT4_GOOD_OLD_INODE_SIZE
) {
4476 ei
->i_extra_isize
= le16_to_cpu(raw_inode
->i_extra_isize
);
4477 if (EXT4_GOOD_OLD_INODE_SIZE
+ ei
->i_extra_isize
>
4478 EXT4_INODE_SIZE(inode
->i_sb
)) {
4483 if (ei
->i_extra_isize
== 0) {
4484 /* The extra space is currently unused. Use it. */
4485 ei
->i_extra_isize
= sizeof(struct ext4_inode
) -
4486 EXT4_GOOD_OLD_INODE_SIZE
;
4488 __le32
*magic
= (void *)raw_inode
+
4489 EXT4_GOOD_OLD_INODE_SIZE
+
4491 if (*magic
== cpu_to_le32(EXT4_XATTR_MAGIC
))
4492 ei
->i_state
|= EXT4_STATE_XATTR
;
4495 ei
->i_extra_isize
= 0;
4497 EXT4_INODE_GET_XTIME(i_ctime
, inode
, raw_inode
);
4498 EXT4_INODE_GET_XTIME(i_mtime
, inode
, raw_inode
);
4499 EXT4_INODE_GET_XTIME(i_atime
, inode
, raw_inode
);
4500 EXT4_EINODE_GET_XTIME(i_crtime
, ei
, raw_inode
);
4502 inode
->i_version
= le32_to_cpu(raw_inode
->i_disk_version
);
4503 if (EXT4_INODE_SIZE(inode
->i_sb
) > EXT4_GOOD_OLD_INODE_SIZE
) {
4504 if (EXT4_FITS_IN_INODE(raw_inode
, ei
, i_version_hi
))
4506 (__u64
)(le32_to_cpu(raw_inode
->i_version_hi
)) << 32;
4510 if (ei
->i_file_acl
&&
4512 (le32_to_cpu(EXT4_SB(sb
)->s_es
->s_first_data_block
) +
4513 EXT4_SB(sb
)->s_gdb_count
)) ||
4514 (ei
->i_file_acl
>= ext4_blocks_count(EXT4_SB(sb
)->s_es
)))) {
4515 ext4_error(sb
, __func__
,
4516 "bad extended attribute block %llu in inode #%lu",
4517 ei
->i_file_acl
, inode
->i_ino
);
4520 } else if (ei
->i_flags
& EXT4_EXTENTS_FL
) {
4521 if (S_ISREG(inode
->i_mode
) || S_ISDIR(inode
->i_mode
) ||
4522 (S_ISLNK(inode
->i_mode
) &&
4523 !ext4_inode_is_fast_symlink(inode
)))
4524 /* Validate extent which is part of inode */
4525 ret
= ext4_ext_check_inode(inode
);
4526 } else if (S_ISREG(inode
->i_mode
) || S_ISDIR(inode
->i_mode
) ||
4527 (S_ISLNK(inode
->i_mode
) &&
4528 !ext4_inode_is_fast_symlink(inode
))) {
4529 /* Validate block references which are part of inode */
4530 ret
= ext4_check_inode_blockref(inode
);
4537 if (S_ISREG(inode
->i_mode
)) {
4538 inode
->i_op
= &ext4_file_inode_operations
;
4539 inode
->i_fop
= &ext4_file_operations
;
4540 ext4_set_aops(inode
);
4541 } else if (S_ISDIR(inode
->i_mode
)) {
4542 inode
->i_op
= &ext4_dir_inode_operations
;
4543 inode
->i_fop
= &ext4_dir_operations
;
4544 } else if (S_ISLNK(inode
->i_mode
)) {
4545 if (ext4_inode_is_fast_symlink(inode
)) {
4546 inode
->i_op
= &ext4_fast_symlink_inode_operations
;
4547 nd_terminate_link(ei
->i_data
, inode
->i_size
,
4548 sizeof(ei
->i_data
) - 1);
4550 inode
->i_op
= &ext4_symlink_inode_operations
;
4551 ext4_set_aops(inode
);
4553 } else if (S_ISCHR(inode
->i_mode
) || S_ISBLK(inode
->i_mode
) ||
4554 S_ISFIFO(inode
->i_mode
) || S_ISSOCK(inode
->i_mode
)) {
4555 inode
->i_op
= &ext4_special_inode_operations
;
4556 if (raw_inode
->i_block
[0])
4557 init_special_inode(inode
, inode
->i_mode
,
4558 old_decode_dev(le32_to_cpu(raw_inode
->i_block
[0])));
4560 init_special_inode(inode
, inode
->i_mode
,
4561 new_decode_dev(le32_to_cpu(raw_inode
->i_block
[1])));
4565 ext4_error(inode
->i_sb
, __func__
,
4566 "bogus i_mode (%o) for inode=%lu",
4567 inode
->i_mode
, inode
->i_ino
);
4571 ext4_set_inode_flags(inode
);
4572 unlock_new_inode(inode
);
4577 return ERR_PTR(ret
);
4580 static int ext4_inode_blocks_set(handle_t
*handle
,
4581 struct ext4_inode
*raw_inode
,
4582 struct ext4_inode_info
*ei
)
4584 struct inode
*inode
= &(ei
->vfs_inode
);
4585 u64 i_blocks
= inode
->i_blocks
;
4586 struct super_block
*sb
= inode
->i_sb
;
4588 if (i_blocks
<= ~0U) {
4590 * i_blocks can be represnted in a 32 bit variable
4591 * as multiple of 512 bytes
4593 raw_inode
->i_blocks_lo
= cpu_to_le32(i_blocks
);
4594 raw_inode
->i_blocks_high
= 0;
4595 ei
->i_flags
&= ~EXT4_HUGE_FILE_FL
;
4598 if (!EXT4_HAS_RO_COMPAT_FEATURE(sb
, EXT4_FEATURE_RO_COMPAT_HUGE_FILE
))
4601 if (i_blocks
<= 0xffffffffffffULL
) {
4603 * i_blocks can be represented in a 48 bit variable
4604 * as multiple of 512 bytes
4606 raw_inode
->i_blocks_lo
= cpu_to_le32(i_blocks
);
4607 raw_inode
->i_blocks_high
= cpu_to_le16(i_blocks
>> 32);
4608 ei
->i_flags
&= ~EXT4_HUGE_FILE_FL
;
4610 ei
->i_flags
|= EXT4_HUGE_FILE_FL
;
4611 /* i_block is stored in file system block size */
4612 i_blocks
= i_blocks
>> (inode
->i_blkbits
- 9);
4613 raw_inode
->i_blocks_lo
= cpu_to_le32(i_blocks
);
4614 raw_inode
->i_blocks_high
= cpu_to_le16(i_blocks
>> 32);
4620 * Post the struct inode info into an on-disk inode location in the
4621 * buffer-cache. This gobbles the caller's reference to the
4622 * buffer_head in the inode location struct.
4624 * The caller must have write access to iloc->bh.
4626 static int ext4_do_update_inode(handle_t
*handle
,
4627 struct inode
*inode
,
4628 struct ext4_iloc
*iloc
)
4630 struct ext4_inode
*raw_inode
= ext4_raw_inode(iloc
);
4631 struct ext4_inode_info
*ei
= EXT4_I(inode
);
4632 struct buffer_head
*bh
= iloc
->bh
;
4633 int err
= 0, rc
, block
;
4635 /* For fields not not tracking in the in-memory inode,
4636 * initialise them to zero for new inodes. */
4637 if (ei
->i_state
& EXT4_STATE_NEW
)
4638 memset(raw_inode
, 0, EXT4_SB(inode
->i_sb
)->s_inode_size
);
4640 ext4_get_inode_flags(ei
);
4641 raw_inode
->i_mode
= cpu_to_le16(inode
->i_mode
);
4642 if (!(test_opt(inode
->i_sb
, NO_UID32
))) {
4643 raw_inode
->i_uid_low
= cpu_to_le16(low_16_bits(inode
->i_uid
));
4644 raw_inode
->i_gid_low
= cpu_to_le16(low_16_bits(inode
->i_gid
));
4646 * Fix up interoperability with old kernels. Otherwise, old inodes get
4647 * re-used with the upper 16 bits of the uid/gid intact
4650 raw_inode
->i_uid_high
=
4651 cpu_to_le16(high_16_bits(inode
->i_uid
));
4652 raw_inode
->i_gid_high
=
4653 cpu_to_le16(high_16_bits(inode
->i_gid
));
4655 raw_inode
->i_uid_high
= 0;
4656 raw_inode
->i_gid_high
= 0;
4659 raw_inode
->i_uid_low
=
4660 cpu_to_le16(fs_high2lowuid(inode
->i_uid
));
4661 raw_inode
->i_gid_low
=
4662 cpu_to_le16(fs_high2lowgid(inode
->i_gid
));
4663 raw_inode
->i_uid_high
= 0;
4664 raw_inode
->i_gid_high
= 0;
4666 raw_inode
->i_links_count
= cpu_to_le16(inode
->i_nlink
);
4668 EXT4_INODE_SET_XTIME(i_ctime
, inode
, raw_inode
);
4669 EXT4_INODE_SET_XTIME(i_mtime
, inode
, raw_inode
);
4670 EXT4_INODE_SET_XTIME(i_atime
, inode
, raw_inode
);
4671 EXT4_EINODE_SET_XTIME(i_crtime
, ei
, raw_inode
);
4673 if (ext4_inode_blocks_set(handle
, raw_inode
, ei
))
4675 raw_inode
->i_dtime
= cpu_to_le32(ei
->i_dtime
);
4676 /* clear the migrate flag in the raw_inode */
4677 raw_inode
->i_flags
= cpu_to_le32(ei
->i_flags
& ~EXT4_EXT_MIGRATE
);
4678 if (EXT4_SB(inode
->i_sb
)->s_es
->s_creator_os
!=
4679 cpu_to_le32(EXT4_OS_HURD
))
4680 raw_inode
->i_file_acl_high
=
4681 cpu_to_le16(ei
->i_file_acl
>> 32);
4682 raw_inode
->i_file_acl_lo
= cpu_to_le32(ei
->i_file_acl
);
4683 ext4_isize_set(raw_inode
, ei
->i_disksize
);
4684 if (ei
->i_disksize
> 0x7fffffffULL
) {
4685 struct super_block
*sb
= inode
->i_sb
;
4686 if (!EXT4_HAS_RO_COMPAT_FEATURE(sb
,
4687 EXT4_FEATURE_RO_COMPAT_LARGE_FILE
) ||
4688 EXT4_SB(sb
)->s_es
->s_rev_level
==
4689 cpu_to_le32(EXT4_GOOD_OLD_REV
)) {
4690 /* If this is the first large file
4691 * created, add a flag to the superblock.
4693 err
= ext4_journal_get_write_access(handle
,
4694 EXT4_SB(sb
)->s_sbh
);
4697 ext4_update_dynamic_rev(sb
);
4698 EXT4_SET_RO_COMPAT_FEATURE(sb
,
4699 EXT4_FEATURE_RO_COMPAT_LARGE_FILE
);
4701 ext4_handle_sync(handle
);
4702 err
= ext4_handle_dirty_metadata(handle
, inode
,
4703 EXT4_SB(sb
)->s_sbh
);
4706 raw_inode
->i_generation
= cpu_to_le32(inode
->i_generation
);
4707 if (S_ISCHR(inode
->i_mode
) || S_ISBLK(inode
->i_mode
)) {
4708 if (old_valid_dev(inode
->i_rdev
)) {
4709 raw_inode
->i_block
[0] =
4710 cpu_to_le32(old_encode_dev(inode
->i_rdev
));
4711 raw_inode
->i_block
[1] = 0;
4713 raw_inode
->i_block
[0] = 0;
4714 raw_inode
->i_block
[1] =
4715 cpu_to_le32(new_encode_dev(inode
->i_rdev
));
4716 raw_inode
->i_block
[2] = 0;
4718 } else for (block
= 0; block
< EXT4_N_BLOCKS
; block
++)
4719 raw_inode
->i_block
[block
] = ei
->i_data
[block
];
4721 raw_inode
->i_disk_version
= cpu_to_le32(inode
->i_version
);
4722 if (ei
->i_extra_isize
) {
4723 if (EXT4_FITS_IN_INODE(raw_inode
, ei
, i_version_hi
))
4724 raw_inode
->i_version_hi
=
4725 cpu_to_le32(inode
->i_version
>> 32);
4726 raw_inode
->i_extra_isize
= cpu_to_le16(ei
->i_extra_isize
);
4729 BUFFER_TRACE(bh
, "call ext4_handle_dirty_metadata");
4730 rc
= ext4_handle_dirty_metadata(handle
, inode
, bh
);
4733 ei
->i_state
&= ~EXT4_STATE_NEW
;
4737 ext4_std_error(inode
->i_sb
, err
);
4742 * ext4_write_inode()
4744 * We are called from a few places:
4746 * - Within generic_file_write() for O_SYNC files.
4747 * Here, there will be no transaction running. We wait for any running
4748 * trasnaction to commit.
4750 * - Within sys_sync(), kupdate and such.
4751 * We wait on commit, if tol to.
4753 * - Within prune_icache() (PF_MEMALLOC == true)
4754 * Here we simply return. We can't afford to block kswapd on the
4757 * In all cases it is actually safe for us to return without doing anything,
4758 * because the inode has been copied into a raw inode buffer in
4759 * ext4_mark_inode_dirty(). This is a correctness thing for O_SYNC and for
4762 * Note that we are absolutely dependent upon all inode dirtiers doing the
4763 * right thing: they *must* call mark_inode_dirty() after dirtying info in
4764 * which we are interested.
4766 * It would be a bug for them to not do this. The code:
4768 * mark_inode_dirty(inode)
4770 * inode->i_size = expr;
4772 * is in error because a kswapd-driven write_inode() could occur while
4773 * `stuff()' is running, and the new i_size will be lost. Plus the inode
4774 * will no longer be on the superblock's dirty inode list.
4776 int ext4_write_inode(struct inode
*inode
, int wait
)
4778 if (current
->flags
& PF_MEMALLOC
)
4781 if (ext4_journal_current_handle()) {
4782 jbd_debug(1, "called recursively, non-PF_MEMALLOC!\n");
4790 return ext4_force_commit(inode
->i_sb
);
4796 * Called from notify_change.
4798 * We want to trap VFS attempts to truncate the file as soon as
4799 * possible. In particular, we want to make sure that when the VFS
4800 * shrinks i_size, we put the inode on the orphan list and modify
4801 * i_disksize immediately, so that during the subsequent flushing of
4802 * dirty pages and freeing of disk blocks, we can guarantee that any
4803 * commit will leave the blocks being flushed in an unused state on
4804 * disk. (On recovery, the inode will get truncated and the blocks will
4805 * be freed, so we have a strong guarantee that no future commit will
4806 * leave these blocks visible to the user.)
4808 * Another thing we have to assure is that if we are in ordered mode
4809 * and inode is still attached to the committing transaction, we must
4810 * we start writeout of all the dirty pages which are being truncated.
4811 * This way we are sure that all the data written in the previous
4812 * transaction are already on disk (truncate waits for pages under
4815 * Called with inode->i_mutex down.
4817 int ext4_setattr(struct dentry
*dentry
, struct iattr
*attr
)
4819 struct inode
*inode
= dentry
->d_inode
;
4821 const unsigned int ia_valid
= attr
->ia_valid
;
4823 error
= inode_change_ok(inode
, attr
);
4827 if ((ia_valid
& ATTR_UID
&& attr
->ia_uid
!= inode
->i_uid
) ||
4828 (ia_valid
& ATTR_GID
&& attr
->ia_gid
!= inode
->i_gid
)) {
4831 /* (user+group)*(old+new) structure, inode write (sb,
4832 * inode block, ? - but truncate inode update has it) */
4833 handle
= ext4_journal_start(inode
, 2*(EXT4_QUOTA_INIT_BLOCKS(inode
->i_sb
)+
4834 EXT4_QUOTA_DEL_BLOCKS(inode
->i_sb
))+3);
4835 if (IS_ERR(handle
)) {
4836 error
= PTR_ERR(handle
);
4839 error
= vfs_dq_transfer(inode
, attr
) ? -EDQUOT
: 0;
4841 ext4_journal_stop(handle
);
4844 /* Update corresponding info in inode so that everything is in
4845 * one transaction */
4846 if (attr
->ia_valid
& ATTR_UID
)
4847 inode
->i_uid
= attr
->ia_uid
;
4848 if (attr
->ia_valid
& ATTR_GID
)
4849 inode
->i_gid
= attr
->ia_gid
;
4850 error
= ext4_mark_inode_dirty(handle
, inode
);
4851 ext4_journal_stop(handle
);
4854 if (attr
->ia_valid
& ATTR_SIZE
) {
4855 if (!(EXT4_I(inode
)->i_flags
& EXT4_EXTENTS_FL
)) {
4856 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
4858 if (attr
->ia_size
> sbi
->s_bitmap_maxbytes
) {
4865 if (S_ISREG(inode
->i_mode
) &&
4866 attr
->ia_valid
& ATTR_SIZE
&& attr
->ia_size
< inode
->i_size
) {
4869 handle
= ext4_journal_start(inode
, 3);
4870 if (IS_ERR(handle
)) {
4871 error
= PTR_ERR(handle
);
4875 error
= ext4_orphan_add(handle
, inode
);
4876 EXT4_I(inode
)->i_disksize
= attr
->ia_size
;
4877 rc
= ext4_mark_inode_dirty(handle
, inode
);
4880 ext4_journal_stop(handle
);
4882 if (ext4_should_order_data(inode
)) {
4883 error
= ext4_begin_ordered_truncate(inode
,
4886 /* Do as much error cleanup as possible */
4887 handle
= ext4_journal_start(inode
, 3);
4888 if (IS_ERR(handle
)) {
4889 ext4_orphan_del(NULL
, inode
);
4892 ext4_orphan_del(handle
, inode
);
4893 ext4_journal_stop(handle
);
4899 rc
= inode_setattr(inode
, attr
);
4901 /* If inode_setattr's call to ext4_truncate failed to get a
4902 * transaction handle at all, we need to clean up the in-core
4903 * orphan list manually. */
4905 ext4_orphan_del(NULL
, inode
);
4907 if (!rc
&& (ia_valid
& ATTR_MODE
))
4908 rc
= ext4_acl_chmod(inode
);
4911 ext4_std_error(inode
->i_sb
, error
);
4917 int ext4_getattr(struct vfsmount
*mnt
, struct dentry
*dentry
,
4920 struct inode
*inode
;
4921 unsigned long delalloc_blocks
;
4923 inode
= dentry
->d_inode
;
4924 generic_fillattr(inode
, stat
);
4927 * We can't update i_blocks if the block allocation is delayed
4928 * otherwise in the case of system crash before the real block
4929 * allocation is done, we will have i_blocks inconsistent with
4930 * on-disk file blocks.
4931 * We always keep i_blocks updated together with real
4932 * allocation. But to not confuse with user, stat
4933 * will return the blocks that include the delayed allocation
4934 * blocks for this file.
4936 spin_lock(&EXT4_I(inode
)->i_block_reservation_lock
);
4937 delalloc_blocks
= EXT4_I(inode
)->i_reserved_data_blocks
;
4938 spin_unlock(&EXT4_I(inode
)->i_block_reservation_lock
);
4940 stat
->blocks
+= (delalloc_blocks
<< inode
->i_sb
->s_blocksize_bits
)>>9;
4944 static int ext4_indirect_trans_blocks(struct inode
*inode
, int nrblocks
,
4949 /* if nrblocks are contiguous */
4952 * With N contiguous data blocks, it need at most
4953 * N/EXT4_ADDR_PER_BLOCK(inode->i_sb) indirect blocks
4954 * 2 dindirect blocks
4957 indirects
= nrblocks
/ EXT4_ADDR_PER_BLOCK(inode
->i_sb
);
4958 return indirects
+ 3;
4961 * if nrblocks are not contiguous, worse case, each block touch
4962 * a indirect block, and each indirect block touch a double indirect
4963 * block, plus a triple indirect block
4965 indirects
= nrblocks
* 2 + 1;
4969 static int ext4_index_trans_blocks(struct inode
*inode
, int nrblocks
, int chunk
)
4971 if (!(EXT4_I(inode
)->i_flags
& EXT4_EXTENTS_FL
))
4972 return ext4_indirect_trans_blocks(inode
, nrblocks
, chunk
);
4973 return ext4_ext_index_trans_blocks(inode
, nrblocks
, chunk
);
4977 * Account for index blocks, block groups bitmaps and block group
4978 * descriptor blocks if modify datablocks and index blocks
4979 * worse case, the indexs blocks spread over different block groups
4981 * If datablocks are discontiguous, they are possible to spread over
4982 * different block groups too. If they are contiugous, with flexbg,
4983 * they could still across block group boundary.
4985 * Also account for superblock, inode, quota and xattr blocks
4987 int ext4_meta_trans_blocks(struct inode
*inode
, int nrblocks
, int chunk
)
4989 ext4_group_t groups
, ngroups
= ext4_get_groups_count(inode
->i_sb
);
4995 * How many index blocks need to touch to modify nrblocks?
4996 * The "Chunk" flag indicating whether the nrblocks is
4997 * physically contiguous on disk
4999 * For Direct IO and fallocate, they calls get_block to allocate
5000 * one single extent at a time, so they could set the "Chunk" flag
5002 idxblocks
= ext4_index_trans_blocks(inode
, nrblocks
, chunk
);
5007 * Now let's see how many group bitmaps and group descriptors need
5017 if (groups
> ngroups
)
5019 if (groups
> EXT4_SB(inode
->i_sb
)->s_gdb_count
)
5020 gdpblocks
= EXT4_SB(inode
->i_sb
)->s_gdb_count
;
5022 /* bitmaps and block group descriptor blocks */
5023 ret
+= groups
+ gdpblocks
;
5025 /* Blocks for super block, inode, quota and xattr blocks */
5026 ret
+= EXT4_META_TRANS_BLOCKS(inode
->i_sb
);
5032 * Calulate the total number of credits to reserve to fit
5033 * the modification of a single pages into a single transaction,
5034 * which may include multiple chunks of block allocations.
5036 * This could be called via ext4_write_begin()
5038 * We need to consider the worse case, when
5039 * one new block per extent.
5041 int ext4_writepage_trans_blocks(struct inode
*inode
)
5043 int bpp
= ext4_journal_blocks_per_page(inode
);
5046 ret
= ext4_meta_trans_blocks(inode
, bpp
, 0);
5048 /* Account for data blocks for journalled mode */
5049 if (ext4_should_journal_data(inode
))
5055 * Calculate the journal credits for a chunk of data modification.
5057 * This is called from DIO, fallocate or whoever calling
5058 * ext4_get_blocks() to map/allocate a chunk of contigous disk blocks.
5060 * journal buffers for data blocks are not included here, as DIO
5061 * and fallocate do no need to journal data buffers.
5063 int ext4_chunk_trans_blocks(struct inode
*inode
, int nrblocks
)
5065 return ext4_meta_trans_blocks(inode
, nrblocks
, 1);
5069 * The caller must have previously called ext4_reserve_inode_write().
5070 * Give this, we know that the caller already has write access to iloc->bh.
5072 int ext4_mark_iloc_dirty(handle_t
*handle
,
5073 struct inode
*inode
, struct ext4_iloc
*iloc
)
5077 if (test_opt(inode
->i_sb
, I_VERSION
))
5078 inode_inc_iversion(inode
);
5080 /* the do_update_inode consumes one bh->b_count */
5083 /* ext4_do_update_inode() does jbd2_journal_dirty_metadata */
5084 err
= ext4_do_update_inode(handle
, inode
, iloc
);
5090 * On success, We end up with an outstanding reference count against
5091 * iloc->bh. This _must_ be cleaned up later.
5095 ext4_reserve_inode_write(handle_t
*handle
, struct inode
*inode
,
5096 struct ext4_iloc
*iloc
)
5100 err
= ext4_get_inode_loc(inode
, iloc
);
5102 BUFFER_TRACE(iloc
->bh
, "get_write_access");
5103 err
= ext4_journal_get_write_access(handle
, iloc
->bh
);
5109 ext4_std_error(inode
->i_sb
, err
);
5114 * Expand an inode by new_extra_isize bytes.
5115 * Returns 0 on success or negative error number on failure.
5117 static int ext4_expand_extra_isize(struct inode
*inode
,
5118 unsigned int new_extra_isize
,
5119 struct ext4_iloc iloc
,
5122 struct ext4_inode
*raw_inode
;
5123 struct ext4_xattr_ibody_header
*header
;
5124 struct ext4_xattr_entry
*entry
;
5126 if (EXT4_I(inode
)->i_extra_isize
>= new_extra_isize
)
5129 raw_inode
= ext4_raw_inode(&iloc
);
5131 header
= IHDR(inode
, raw_inode
);
5132 entry
= IFIRST(header
);
5134 /* No extended attributes present */
5135 if (!(EXT4_I(inode
)->i_state
& EXT4_STATE_XATTR
) ||
5136 header
->h_magic
!= cpu_to_le32(EXT4_XATTR_MAGIC
)) {
5137 memset((void *)raw_inode
+ EXT4_GOOD_OLD_INODE_SIZE
, 0,
5139 EXT4_I(inode
)->i_extra_isize
= new_extra_isize
;
5143 /* try to expand with EAs present */
5144 return ext4_expand_extra_isize_ea(inode
, new_extra_isize
,
5149 * What we do here is to mark the in-core inode as clean with respect to inode
5150 * dirtiness (it may still be data-dirty).
5151 * This means that the in-core inode may be reaped by prune_icache
5152 * without having to perform any I/O. This is a very good thing,
5153 * because *any* task may call prune_icache - even ones which
5154 * have a transaction open against a different journal.
5156 * Is this cheating? Not really. Sure, we haven't written the
5157 * inode out, but prune_icache isn't a user-visible syncing function.
5158 * Whenever the user wants stuff synced (sys_sync, sys_msync, sys_fsync)
5159 * we start and wait on commits.
5161 * Is this efficient/effective? Well, we're being nice to the system
5162 * by cleaning up our inodes proactively so they can be reaped
5163 * without I/O. But we are potentially leaving up to five seconds'
5164 * worth of inodes floating about which prune_icache wants us to
5165 * write out. One way to fix that would be to get prune_icache()
5166 * to do a write_super() to free up some memory. It has the desired
5169 int ext4_mark_inode_dirty(handle_t
*handle
, struct inode
*inode
)
5171 struct ext4_iloc iloc
;
5172 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
5173 static unsigned int mnt_count
;
5177 err
= ext4_reserve_inode_write(handle
, inode
, &iloc
);
5178 if (ext4_handle_valid(handle
) &&
5179 EXT4_I(inode
)->i_extra_isize
< sbi
->s_want_extra_isize
&&
5180 !(EXT4_I(inode
)->i_state
& EXT4_STATE_NO_EXPAND
)) {
5182 * We need extra buffer credits since we may write into EA block
5183 * with this same handle. If journal_extend fails, then it will
5184 * only result in a minor loss of functionality for that inode.
5185 * If this is felt to be critical, then e2fsck should be run to
5186 * force a large enough s_min_extra_isize.
5188 if ((jbd2_journal_extend(handle
,
5189 EXT4_DATA_TRANS_BLOCKS(inode
->i_sb
))) == 0) {
5190 ret
= ext4_expand_extra_isize(inode
,
5191 sbi
->s_want_extra_isize
,
5194 EXT4_I(inode
)->i_state
|= EXT4_STATE_NO_EXPAND
;
5196 le16_to_cpu(sbi
->s_es
->s_mnt_count
)) {
5197 ext4_warning(inode
->i_sb
, __func__
,
5198 "Unable to expand inode %lu. Delete"
5199 " some EAs or run e2fsck.",
5202 le16_to_cpu(sbi
->s_es
->s_mnt_count
);
5208 err
= ext4_mark_iloc_dirty(handle
, inode
, &iloc
);
5213 * ext4_dirty_inode() is called from __mark_inode_dirty()
5215 * We're really interested in the case where a file is being extended.
5216 * i_size has been changed by generic_commit_write() and we thus need
5217 * to include the updated inode in the current transaction.
5219 * Also, vfs_dq_alloc_block() will always dirty the inode when blocks
5220 * are allocated to the file.
5222 * If the inode is marked synchronous, we don't honour that here - doing
5223 * so would cause a commit on atime updates, which we don't bother doing.
5224 * We handle synchronous inodes at the highest possible level.
5226 void ext4_dirty_inode(struct inode
*inode
)
5228 handle_t
*current_handle
= ext4_journal_current_handle();
5231 if (!ext4_handle_valid(current_handle
)) {
5232 ext4_mark_inode_dirty(current_handle
, inode
);
5236 handle
= ext4_journal_start(inode
, 2);
5239 if (current_handle
&&
5240 current_handle
->h_transaction
!= handle
->h_transaction
) {
5241 /* This task has a transaction open against a different fs */
5242 printk(KERN_EMERG
"%s: transactions do not match!\n",
5245 jbd_debug(5, "marking dirty. outer handle=%p\n",
5247 ext4_mark_inode_dirty(handle
, inode
);
5249 ext4_journal_stop(handle
);
5256 * Bind an inode's backing buffer_head into this transaction, to prevent
5257 * it from being flushed to disk early. Unlike
5258 * ext4_reserve_inode_write, this leaves behind no bh reference and
5259 * returns no iloc structure, so the caller needs to repeat the iloc
5260 * lookup to mark the inode dirty later.
5262 static int ext4_pin_inode(handle_t
*handle
, struct inode
*inode
)
5264 struct ext4_iloc iloc
;
5268 err
= ext4_get_inode_loc(inode
, &iloc
);
5270 BUFFER_TRACE(iloc
.bh
, "get_write_access");
5271 err
= jbd2_journal_get_write_access(handle
, iloc
.bh
);
5273 err
= ext4_handle_dirty_metadata(handle
,
5279 ext4_std_error(inode
->i_sb
, err
);
5284 int ext4_change_inode_journal_flag(struct inode
*inode
, int val
)
5291 * We have to be very careful here: changing a data block's
5292 * journaling status dynamically is dangerous. If we write a
5293 * data block to the journal, change the status and then delete
5294 * that block, we risk forgetting to revoke the old log record
5295 * from the journal and so a subsequent replay can corrupt data.
5296 * So, first we make sure that the journal is empty and that
5297 * nobody is changing anything.
5300 journal
= EXT4_JOURNAL(inode
);
5303 if (is_journal_aborted(journal
))
5306 jbd2_journal_lock_updates(journal
);
5307 jbd2_journal_flush(journal
);
5310 * OK, there are no updates running now, and all cached data is
5311 * synced to disk. We are now in a completely consistent state
5312 * which doesn't have anything in the journal, and we know that
5313 * no filesystem updates are running, so it is safe to modify
5314 * the inode's in-core data-journaling state flag now.
5318 EXT4_I(inode
)->i_flags
|= EXT4_JOURNAL_DATA_FL
;
5320 EXT4_I(inode
)->i_flags
&= ~EXT4_JOURNAL_DATA_FL
;
5321 ext4_set_aops(inode
);
5323 jbd2_journal_unlock_updates(journal
);
5325 /* Finally we can mark the inode as dirty. */
5327 handle
= ext4_journal_start(inode
, 1);
5329 return PTR_ERR(handle
);
5331 err
= ext4_mark_inode_dirty(handle
, inode
);
5332 ext4_handle_sync(handle
);
5333 ext4_journal_stop(handle
);
5334 ext4_std_error(inode
->i_sb
, err
);
5339 static int ext4_bh_unmapped(handle_t
*handle
, struct buffer_head
*bh
)
5341 return !buffer_mapped(bh
);
5344 int ext4_page_mkwrite(struct vm_area_struct
*vma
, struct vm_fault
*vmf
)
5346 struct page
*page
= vmf
->page
;
5351 struct file
*file
= vma
->vm_file
;
5352 struct inode
*inode
= file
->f_path
.dentry
->d_inode
;
5353 struct address_space
*mapping
= inode
->i_mapping
;
5356 * Get i_alloc_sem to stop truncates messing with the inode. We cannot
5357 * get i_mutex because we are already holding mmap_sem.
5359 down_read(&inode
->i_alloc_sem
);
5360 size
= i_size_read(inode
);
5361 if (page
->mapping
!= mapping
|| size
<= page_offset(page
)
5362 || !PageUptodate(page
)) {
5363 /* page got truncated from under us? */
5367 if (PageMappedToDisk(page
))
5370 if (page
->index
== size
>> PAGE_CACHE_SHIFT
)
5371 len
= size
& ~PAGE_CACHE_MASK
;
5373 len
= PAGE_CACHE_SIZE
;
5375 if (page_has_buffers(page
)) {
5376 /* return if we have all the buffers mapped */
5377 if (!walk_page_buffers(NULL
, page_buffers(page
), 0, len
, NULL
,
5382 * OK, we need to fill the hole... Do write_begin write_end
5383 * to do block allocation/reservation.We are not holding
5384 * inode.i__mutex here. That allow * parallel write_begin,
5385 * write_end call. lock_page prevent this from happening
5386 * on the same page though
5388 ret
= mapping
->a_ops
->write_begin(file
, mapping
, page_offset(page
),
5389 len
, AOP_FLAG_UNINTERRUPTIBLE
, &page
, &fsdata
);
5392 ret
= mapping
->a_ops
->write_end(file
, mapping
, page_offset(page
),
5393 len
, len
, page
, fsdata
);
5399 ret
= VM_FAULT_SIGBUS
;
5400 up_read(&inode
->i_alloc_sem
);