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
) {
377 unsigned int maxblocks
= ext4_blocks_count(EXT4_SB(inode
->i_sb
)->s_es
);
379 while (bref
< p
+max
) {
380 if (unlikely(le32_to_cpu(*bref
) >= maxblocks
)) {
381 ext4_error(inode
->i_sb
, function
,
382 "block reference %u >= max (%u) "
383 "in inode #%lu, offset=%d",
384 le32_to_cpu(*bref
), maxblocks
,
385 inode
->i_ino
, (int)(bref
-p
));
394 #define ext4_check_indirect_blockref(inode, bh) \
395 __ext4_check_blockref(__func__, inode, (__le32 *)(bh)->b_data, \
396 EXT4_ADDR_PER_BLOCK((inode)->i_sb))
398 #define ext4_check_inode_blockref(inode) \
399 __ext4_check_blockref(__func__, inode, EXT4_I(inode)->i_data, \
403 * ext4_get_branch - read the chain of indirect blocks leading to data
404 * @inode: inode in question
405 * @depth: depth of the chain (1 - direct pointer, etc.)
406 * @offsets: offsets of pointers in inode/indirect blocks
407 * @chain: place to store the result
408 * @err: here we store the error value
410 * Function fills the array of triples <key, p, bh> and returns %NULL
411 * if everything went OK or the pointer to the last filled triple
412 * (incomplete one) otherwise. Upon the return chain[i].key contains
413 * the number of (i+1)-th block in the chain (as it is stored in memory,
414 * i.e. little-endian 32-bit), chain[i].p contains the address of that
415 * number (it points into struct inode for i==0 and into the bh->b_data
416 * for i>0) and chain[i].bh points to the buffer_head of i-th indirect
417 * block for i>0 and NULL for i==0. In other words, it holds the block
418 * numbers of the chain, addresses they were taken from (and where we can
419 * verify that chain did not change) and buffer_heads hosting these
422 * Function stops when it stumbles upon zero pointer (absent block)
423 * (pointer to last triple returned, *@err == 0)
424 * or when it gets an IO error reading an indirect block
425 * (ditto, *@err == -EIO)
426 * or when it reads all @depth-1 indirect blocks successfully and finds
427 * the whole chain, all way to the data (returns %NULL, *err == 0).
429 * Need to be called with
430 * down_read(&EXT4_I(inode)->i_data_sem)
432 static Indirect
*ext4_get_branch(struct inode
*inode
, int depth
,
433 ext4_lblk_t
*offsets
,
434 Indirect chain
[4], int *err
)
436 struct super_block
*sb
= inode
->i_sb
;
438 struct buffer_head
*bh
;
441 /* i_data is not going away, no lock needed */
442 add_chain(chain
, NULL
, EXT4_I(inode
)->i_data
+ *offsets
);
446 bh
= sb_getblk(sb
, le32_to_cpu(p
->key
));
450 if (!bh_uptodate_or_lock(bh
)) {
451 if (bh_submit_read(bh
) < 0) {
455 /* validate block references */
456 if (ext4_check_indirect_blockref(inode
, bh
)) {
462 add_chain(++p
, bh
, (__le32
*)bh
->b_data
+ *++offsets
);
476 * ext4_find_near - find a place for allocation with sufficient locality
478 * @ind: descriptor of indirect block.
480 * This function returns the preferred place for block allocation.
481 * It is used when heuristic for sequential allocation fails.
483 * + if there is a block to the left of our position - allocate near it.
484 * + if pointer will live in indirect block - allocate near that block.
485 * + if pointer will live in inode - allocate in the same
488 * In the latter case we colour the starting block by the callers PID to
489 * prevent it from clashing with concurrent allocations for a different inode
490 * in the same block group. The PID is used here so that functionally related
491 * files will be close-by on-disk.
493 * Caller must make sure that @ind is valid and will stay that way.
495 static ext4_fsblk_t
ext4_find_near(struct inode
*inode
, Indirect
*ind
)
497 struct ext4_inode_info
*ei
= EXT4_I(inode
);
498 __le32
*start
= ind
->bh
? (__le32
*) ind
->bh
->b_data
: ei
->i_data
;
500 ext4_fsblk_t bg_start
;
501 ext4_fsblk_t last_block
;
502 ext4_grpblk_t colour
;
503 ext4_group_t block_group
;
504 int flex_size
= ext4_flex_bg_size(EXT4_SB(inode
->i_sb
));
506 /* Try to find previous block */
507 for (p
= ind
->p
- 1; p
>= start
; p
--) {
509 return le32_to_cpu(*p
);
512 /* No such thing, so let's try location of indirect block */
514 return ind
->bh
->b_blocknr
;
517 * It is going to be referred to from the inode itself? OK, just put it
518 * into the same cylinder group then.
520 block_group
= ei
->i_block_group
;
521 if (flex_size
>= EXT4_FLEX_SIZE_DIR_ALLOC_SCHEME
) {
522 block_group
&= ~(flex_size
-1);
523 if (S_ISREG(inode
->i_mode
))
526 bg_start
= ext4_group_first_block_no(inode
->i_sb
, block_group
);
527 last_block
= ext4_blocks_count(EXT4_SB(inode
->i_sb
)->s_es
) - 1;
530 * If we are doing delayed allocation, we don't need take
531 * colour into account.
533 if (test_opt(inode
->i_sb
, DELALLOC
))
536 if (bg_start
+ EXT4_BLOCKS_PER_GROUP(inode
->i_sb
) <= last_block
)
537 colour
= (current
->pid
% 16) *
538 (EXT4_BLOCKS_PER_GROUP(inode
->i_sb
) / 16);
540 colour
= (current
->pid
% 16) * ((last_block
- bg_start
) / 16);
541 return bg_start
+ colour
;
545 * ext4_find_goal - find a preferred place for allocation.
547 * @block: block we want
548 * @partial: pointer to the last triple within a chain
550 * Normally this function find the preferred place for block allocation,
553 static ext4_fsblk_t
ext4_find_goal(struct inode
*inode
, ext4_lblk_t block
,
557 * XXX need to get goal block from mballoc's data structures
560 return ext4_find_near(inode
, partial
);
564 * ext4_blks_to_allocate: Look up the block map and count the number
565 * of direct blocks need to be allocated for the given branch.
567 * @branch: chain of indirect blocks
568 * @k: number of blocks need for indirect blocks
569 * @blks: number of data blocks to be mapped.
570 * @blocks_to_boundary: the offset in the indirect block
572 * return the total number of blocks to be allocate, including the
573 * direct and indirect blocks.
575 static int ext4_blks_to_allocate(Indirect
*branch
, int k
, unsigned int blks
,
576 int blocks_to_boundary
)
578 unsigned int count
= 0;
581 * Simple case, [t,d]Indirect block(s) has not allocated yet
582 * then it's clear blocks on that path have not allocated
585 /* right now we don't handle cross boundary allocation */
586 if (blks
< blocks_to_boundary
+ 1)
589 count
+= blocks_to_boundary
+ 1;
594 while (count
< blks
&& count
<= blocks_to_boundary
&&
595 le32_to_cpu(*(branch
[0].p
+ count
)) == 0) {
602 * ext4_alloc_blocks: multiple allocate blocks needed for a branch
603 * @indirect_blks: the number of blocks need to allocate for indirect
606 * @new_blocks: on return it will store the new block numbers for
607 * the indirect blocks(if needed) and the first direct block,
608 * @blks: on return it will store the total number of allocated
611 static int ext4_alloc_blocks(handle_t
*handle
, struct inode
*inode
,
612 ext4_lblk_t iblock
, ext4_fsblk_t goal
,
613 int indirect_blks
, int blks
,
614 ext4_fsblk_t new_blocks
[4], int *err
)
616 struct ext4_allocation_request ar
;
618 unsigned long count
= 0, blk_allocated
= 0;
620 ext4_fsblk_t current_block
= 0;
624 * Here we try to allocate the requested multiple blocks at once,
625 * on a best-effort basis.
626 * To build a branch, we should allocate blocks for
627 * the indirect blocks(if not allocated yet), and at least
628 * the first direct block of this branch. That's the
629 * minimum number of blocks need to allocate(required)
631 /* first we try to allocate the indirect blocks */
632 target
= indirect_blks
;
635 /* allocating blocks for indirect blocks and direct blocks */
636 current_block
= ext4_new_meta_blocks(handle
, inode
,
642 /* allocate blocks for indirect blocks */
643 while (index
< indirect_blks
&& count
) {
644 new_blocks
[index
++] = current_block
++;
649 * save the new block number
650 * for the first direct block
652 new_blocks
[index
] = current_block
;
653 printk(KERN_INFO
"%s returned more blocks than "
654 "requested\n", __func__
);
660 target
= blks
- count
;
661 blk_allocated
= count
;
664 /* Now allocate data blocks */
665 memset(&ar
, 0, sizeof(ar
));
670 if (S_ISREG(inode
->i_mode
))
671 /* enable in-core preallocation only for regular files */
672 ar
.flags
= EXT4_MB_HINT_DATA
;
674 current_block
= ext4_mb_new_blocks(handle
, &ar
, err
);
676 if (*err
&& (target
== blks
)) {
678 * if the allocation failed and we didn't allocate
684 if (target
== blks
) {
686 * save the new block number
687 * for the first direct block
689 new_blocks
[index
] = current_block
;
691 blk_allocated
+= ar
.len
;
694 /* total number of blocks allocated for direct blocks */
699 for (i
= 0; i
< index
; i
++)
700 ext4_free_blocks(handle
, inode
, new_blocks
[i
], 1, 0);
705 * ext4_alloc_branch - allocate and set up a chain of blocks.
707 * @indirect_blks: number of allocated indirect blocks
708 * @blks: number of allocated direct blocks
709 * @offsets: offsets (in the blocks) to store the pointers to next.
710 * @branch: place to store the chain in.
712 * This function allocates blocks, zeroes out all but the last one,
713 * links them into chain and (if we are synchronous) writes them to disk.
714 * In other words, it prepares a branch that can be spliced onto the
715 * inode. It stores the information about that chain in the branch[], in
716 * the same format as ext4_get_branch() would do. We are calling it after
717 * we had read the existing part of chain and partial points to the last
718 * triple of that (one with zero ->key). Upon the exit we have the same
719 * picture as after the successful ext4_get_block(), except that in one
720 * place chain is disconnected - *branch->p is still zero (we did not
721 * set the last link), but branch->key contains the number that should
722 * be placed into *branch->p to fill that gap.
724 * If allocation fails we free all blocks we've allocated (and forget
725 * their buffer_heads) and return the error value the from failed
726 * ext4_alloc_block() (normally -ENOSPC). Otherwise we set the chain
727 * as described above and return 0.
729 static int ext4_alloc_branch(handle_t
*handle
, struct inode
*inode
,
730 ext4_lblk_t iblock
, int indirect_blks
,
731 int *blks
, ext4_fsblk_t goal
,
732 ext4_lblk_t
*offsets
, Indirect
*branch
)
734 int blocksize
= inode
->i_sb
->s_blocksize
;
737 struct buffer_head
*bh
;
739 ext4_fsblk_t new_blocks
[4];
740 ext4_fsblk_t current_block
;
742 num
= ext4_alloc_blocks(handle
, inode
, iblock
, goal
, indirect_blks
,
743 *blks
, new_blocks
, &err
);
747 branch
[0].key
= cpu_to_le32(new_blocks
[0]);
749 * metadata blocks and data blocks are allocated.
751 for (n
= 1; n
<= indirect_blks
; n
++) {
753 * Get buffer_head for parent block, zero it out
754 * and set the pointer to new one, then send
757 bh
= sb_getblk(inode
->i_sb
, new_blocks
[n
-1]);
760 BUFFER_TRACE(bh
, "call get_create_access");
761 err
= ext4_journal_get_create_access(handle
, bh
);
768 memset(bh
->b_data
, 0, blocksize
);
769 branch
[n
].p
= (__le32
*) bh
->b_data
+ offsets
[n
];
770 branch
[n
].key
= cpu_to_le32(new_blocks
[n
]);
771 *branch
[n
].p
= branch
[n
].key
;
772 if (n
== indirect_blks
) {
773 current_block
= new_blocks
[n
];
775 * End of chain, update the last new metablock of
776 * the chain to point to the new allocated
777 * data blocks numbers
779 for (i
=1; i
< num
; i
++)
780 *(branch
[n
].p
+ i
) = cpu_to_le32(++current_block
);
782 BUFFER_TRACE(bh
, "marking uptodate");
783 set_buffer_uptodate(bh
);
786 BUFFER_TRACE(bh
, "call ext4_handle_dirty_metadata");
787 err
= ext4_handle_dirty_metadata(handle
, inode
, bh
);
794 /* Allocation failed, free what we already allocated */
795 for (i
= 1; i
<= n
; i
++) {
796 BUFFER_TRACE(branch
[i
].bh
, "call jbd2_journal_forget");
797 ext4_journal_forget(handle
, branch
[i
].bh
);
799 for (i
= 0; i
< indirect_blks
; i
++)
800 ext4_free_blocks(handle
, inode
, new_blocks
[i
], 1, 0);
802 ext4_free_blocks(handle
, inode
, new_blocks
[i
], num
, 0);
808 * ext4_splice_branch - splice the allocated branch onto inode.
810 * @block: (logical) number of block we are adding
811 * @chain: chain of indirect blocks (with a missing link - see
813 * @where: location of missing link
814 * @num: number of indirect blocks we are adding
815 * @blks: number of direct blocks we are adding
817 * This function fills the missing link and does all housekeeping needed in
818 * inode (->i_blocks, etc.). In case of success we end up with the full
819 * chain to new block and return 0.
821 static int ext4_splice_branch(handle_t
*handle
, struct inode
*inode
,
822 ext4_lblk_t block
, Indirect
*where
, int num
, int blks
)
826 ext4_fsblk_t current_block
;
829 * If we're splicing into a [td]indirect block (as opposed to the
830 * inode) then we need to get write access to the [td]indirect block
834 BUFFER_TRACE(where
->bh
, "get_write_access");
835 err
= ext4_journal_get_write_access(handle
, where
->bh
);
841 *where
->p
= where
->key
;
844 * Update the host buffer_head or inode to point to more just allocated
845 * direct blocks blocks
847 if (num
== 0 && blks
> 1) {
848 current_block
= le32_to_cpu(where
->key
) + 1;
849 for (i
= 1; i
< blks
; i
++)
850 *(where
->p
+ i
) = cpu_to_le32(current_block
++);
853 /* We are done with atomic stuff, now do the rest of housekeeping */
855 inode
->i_ctime
= ext4_current_time(inode
);
856 ext4_mark_inode_dirty(handle
, inode
);
858 /* had we spliced it onto indirect block? */
861 * If we spliced it onto an indirect block, we haven't
862 * altered the inode. Note however that if it is being spliced
863 * onto an indirect block at the very end of the file (the
864 * file is growing) then we *will* alter the inode to reflect
865 * the new i_size. But that is not done here - it is done in
866 * generic_commit_write->__mark_inode_dirty->ext4_dirty_inode.
868 jbd_debug(5, "splicing indirect only\n");
869 BUFFER_TRACE(where
->bh
, "call ext4_handle_dirty_metadata");
870 err
= ext4_handle_dirty_metadata(handle
, inode
, where
->bh
);
875 * OK, we spliced it into the inode itself on a direct block.
876 * Inode was dirtied above.
878 jbd_debug(5, "splicing direct\n");
883 for (i
= 1; i
<= num
; i
++) {
884 BUFFER_TRACE(where
[i
].bh
, "call jbd2_journal_forget");
885 ext4_journal_forget(handle
, where
[i
].bh
);
886 ext4_free_blocks(handle
, inode
,
887 le32_to_cpu(where
[i
-1].key
), 1, 0);
889 ext4_free_blocks(handle
, inode
, le32_to_cpu(where
[num
].key
), blks
, 0);
895 * The ext4_ind_get_blocks() function handles non-extents inodes
896 * (i.e., using the traditional indirect/double-indirect i_blocks
897 * scheme) for ext4_get_blocks().
899 * Allocation strategy is simple: if we have to allocate something, we will
900 * have to go the whole way to leaf. So let's do it before attaching anything
901 * to tree, set linkage between the newborn blocks, write them if sync is
902 * required, recheck the path, free and repeat if check fails, otherwise
903 * set the last missing link (that will protect us from any truncate-generated
904 * removals - all blocks on the path are immune now) and possibly force the
905 * write on the parent block.
906 * That has a nice additional property: no special recovery from the failed
907 * allocations is needed - we simply release blocks and do not touch anything
908 * reachable from inode.
910 * `handle' can be NULL if create == 0.
912 * return > 0, # of blocks mapped or allocated.
913 * return = 0, if plain lookup failed.
914 * return < 0, error case.
916 * The ext4_ind_get_blocks() function should be called with
917 * down_write(&EXT4_I(inode)->i_data_sem) if allocating filesystem
918 * blocks (i.e., flags has EXT4_GET_BLOCKS_CREATE set) or
919 * down_read(&EXT4_I(inode)->i_data_sem) if not allocating file system
922 static int ext4_ind_get_blocks(handle_t
*handle
, struct inode
*inode
,
923 ext4_lblk_t iblock
, unsigned int maxblocks
,
924 struct buffer_head
*bh_result
,
928 ext4_lblk_t offsets
[4];
933 int blocks_to_boundary
= 0;
935 struct ext4_inode_info
*ei
= EXT4_I(inode
);
937 ext4_fsblk_t first_block
= 0;
941 J_ASSERT(!(EXT4_I(inode
)->i_flags
& EXT4_EXTENTS_FL
));
942 J_ASSERT(handle
!= NULL
|| (flags
& EXT4_GET_BLOCKS_CREATE
) == 0);
943 depth
= ext4_block_to_path(inode
, iblock
, offsets
,
944 &blocks_to_boundary
);
949 partial
= ext4_get_branch(inode
, depth
, offsets
, chain
, &err
);
951 /* Simplest case - block found, no allocation needed */
953 first_block
= le32_to_cpu(chain
[depth
- 1].key
);
954 clear_buffer_new(bh_result
);
957 while (count
< maxblocks
&& count
<= blocks_to_boundary
) {
960 blk
= le32_to_cpu(*(chain
[depth
-1].p
+ count
));
962 if (blk
== first_block
+ count
)
970 /* Next simple case - plain lookup or failed read of indirect block */
971 if ((flags
& EXT4_GET_BLOCKS_CREATE
) == 0 || err
== -EIO
)
975 * Okay, we need to do block allocation.
977 goal
= ext4_find_goal(inode
, iblock
, partial
);
979 /* the number of blocks need to allocate for [d,t]indirect blocks */
980 indirect_blks
= (chain
+ depth
) - partial
- 1;
983 * Next look up the indirect map to count the totoal number of
984 * direct blocks to allocate for this branch.
986 count
= ext4_blks_to_allocate(partial
, indirect_blks
,
987 maxblocks
, blocks_to_boundary
);
989 * Block out ext4_truncate while we alter the tree
991 err
= ext4_alloc_branch(handle
, inode
, iblock
, indirect_blks
,
993 offsets
+ (partial
- chain
), partial
);
996 * The ext4_splice_branch call will free and forget any buffers
997 * on the new chain if there is a failure, but that risks using
998 * up transaction credits, especially for bitmaps where the
999 * credits cannot be returned. Can we handle this somehow? We
1000 * may need to return -EAGAIN upwards in the worst case. --sct
1003 err
= ext4_splice_branch(handle
, inode
, iblock
,
1004 partial
, indirect_blks
, count
);
1006 * i_disksize growing is protected by i_data_sem. Don't forget to
1007 * protect it if you're about to implement concurrent
1008 * ext4_get_block() -bzzz
1010 if (!err
&& (flags
& EXT4_GET_BLOCKS_EXTEND_DISKSIZE
)) {
1011 disksize
= ((loff_t
) iblock
+ count
) << inode
->i_blkbits
;
1012 if (disksize
> i_size_read(inode
))
1013 disksize
= i_size_read(inode
);
1014 if (disksize
> ei
->i_disksize
)
1015 ei
->i_disksize
= disksize
;
1020 set_buffer_new(bh_result
);
1022 map_bh(bh_result
, inode
->i_sb
, le32_to_cpu(chain
[depth
-1].key
));
1023 if (count
> blocks_to_boundary
)
1024 set_buffer_boundary(bh_result
);
1026 /* Clean up and exit */
1027 partial
= chain
+ depth
- 1; /* the whole chain */
1029 while (partial
> chain
) {
1030 BUFFER_TRACE(partial
->bh
, "call brelse");
1031 brelse(partial
->bh
);
1034 BUFFER_TRACE(bh_result
, "returned");
1039 qsize_t
ext4_get_reserved_space(struct inode
*inode
)
1041 unsigned long long total
;
1043 spin_lock(&EXT4_I(inode
)->i_block_reservation_lock
);
1044 total
= EXT4_I(inode
)->i_reserved_data_blocks
+
1045 EXT4_I(inode
)->i_reserved_meta_blocks
;
1046 spin_unlock(&EXT4_I(inode
)->i_block_reservation_lock
);
1051 * Calculate the number of metadata blocks need to reserve
1052 * to allocate @blocks for non extent file based file
1054 static int ext4_indirect_calc_metadata_amount(struct inode
*inode
, int blocks
)
1056 int icap
= EXT4_ADDR_PER_BLOCK(inode
->i_sb
);
1057 int ind_blks
, dind_blks
, tind_blks
;
1059 /* number of new indirect blocks needed */
1060 ind_blks
= (blocks
+ icap
- 1) / icap
;
1062 dind_blks
= (ind_blks
+ icap
- 1) / icap
;
1066 return ind_blks
+ dind_blks
+ tind_blks
;
1070 * Calculate the number of metadata blocks need to reserve
1071 * to allocate given number of blocks
1073 static int ext4_calc_metadata_amount(struct inode
*inode
, int blocks
)
1078 if (EXT4_I(inode
)->i_flags
& EXT4_EXTENTS_FL
)
1079 return ext4_ext_calc_metadata_amount(inode
, blocks
);
1081 return ext4_indirect_calc_metadata_amount(inode
, blocks
);
1084 static void ext4_da_update_reserve_space(struct inode
*inode
, int used
)
1086 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
1087 int total
, mdb
, mdb_free
;
1089 spin_lock(&EXT4_I(inode
)->i_block_reservation_lock
);
1090 /* recalculate the number of metablocks still need to be reserved */
1091 total
= EXT4_I(inode
)->i_reserved_data_blocks
- used
;
1092 mdb
= ext4_calc_metadata_amount(inode
, total
);
1094 /* figure out how many metablocks to release */
1095 BUG_ON(mdb
> EXT4_I(inode
)->i_reserved_meta_blocks
);
1096 mdb_free
= EXT4_I(inode
)->i_reserved_meta_blocks
- mdb
;
1099 /* Account for allocated meta_blocks */
1100 mdb_free
-= EXT4_I(inode
)->i_allocated_meta_blocks
;
1102 /* update fs dirty blocks counter */
1103 percpu_counter_sub(&sbi
->s_dirtyblocks_counter
, mdb_free
);
1104 EXT4_I(inode
)->i_allocated_meta_blocks
= 0;
1105 EXT4_I(inode
)->i_reserved_meta_blocks
= mdb
;
1108 /* update per-inode reservations */
1109 BUG_ON(used
> EXT4_I(inode
)->i_reserved_data_blocks
);
1110 EXT4_I(inode
)->i_reserved_data_blocks
-= used
;
1111 spin_unlock(&EXT4_I(inode
)->i_block_reservation_lock
);
1114 * free those over-booking quota for metadata blocks
1117 vfs_dq_release_reservation_block(inode
, mdb_free
);
1120 * If we have done all the pending block allocations and if
1121 * there aren't any writers on the inode, we can discard the
1122 * inode's preallocations.
1124 if (!total
&& (atomic_read(&inode
->i_writecount
) == 0))
1125 ext4_discard_preallocations(inode
);
1129 * The ext4_get_blocks() function tries to look up the requested blocks,
1130 * and returns if the blocks are already mapped.
1132 * Otherwise it takes the write lock of the i_data_sem and allocate blocks
1133 * and store the allocated blocks in the result buffer head and mark it
1136 * If file type is extents based, it will call ext4_ext_get_blocks(),
1137 * Otherwise, call with ext4_ind_get_blocks() to handle indirect mapping
1140 * On success, it returns the number of blocks being mapped or allocate.
1141 * if create==0 and the blocks are pre-allocated and uninitialized block,
1142 * the result buffer head is unmapped. If the create ==1, it will make sure
1143 * the buffer head is mapped.
1145 * It returns 0 if plain look up failed (blocks have not been allocated), in
1146 * that casem, buffer head is unmapped
1148 * It returns the error in case of allocation failure.
1150 int ext4_get_blocks(handle_t
*handle
, struct inode
*inode
, sector_t block
,
1151 unsigned int max_blocks
, struct buffer_head
*bh
,
1156 clear_buffer_mapped(bh
);
1157 clear_buffer_unwritten(bh
);
1160 * Try to see if we can get the block without requesting a new
1161 * file system block.
1163 down_read((&EXT4_I(inode
)->i_data_sem
));
1164 if (EXT4_I(inode
)->i_flags
& EXT4_EXTENTS_FL
) {
1165 retval
= ext4_ext_get_blocks(handle
, inode
, block
, max_blocks
,
1168 retval
= ext4_ind_get_blocks(handle
, inode
, block
, max_blocks
,
1171 up_read((&EXT4_I(inode
)->i_data_sem
));
1173 /* If it is only a block(s) look up */
1174 if ((flags
& EXT4_GET_BLOCKS_CREATE
) == 0)
1178 * Returns if the blocks have already allocated
1180 * Note that if blocks have been preallocated
1181 * ext4_ext_get_block() returns th create = 0
1182 * with buffer head unmapped.
1184 if (retval
> 0 && buffer_mapped(bh
))
1188 * When we call get_blocks without the create flag, the
1189 * BH_Unwritten flag could have gotten set if the blocks
1190 * requested were part of a uninitialized extent. We need to
1191 * clear this flag now that we are committed to convert all or
1192 * part of the uninitialized extent to be an initialized
1193 * extent. This is because we need to avoid the combination
1194 * of BH_Unwritten and BH_Mapped flags being simultaneously
1195 * set on the buffer_head.
1197 clear_buffer_unwritten(bh
);
1200 * New blocks allocate and/or writing to uninitialized extent
1201 * will possibly result in updating i_data, so we take
1202 * the write lock of i_data_sem, and call get_blocks()
1203 * with create == 1 flag.
1205 down_write((&EXT4_I(inode
)->i_data_sem
));
1208 * if the caller is from delayed allocation writeout path
1209 * we have already reserved fs blocks for allocation
1210 * let the underlying get_block() function know to
1211 * avoid double accounting
1213 if (flags
& EXT4_GET_BLOCKS_DELALLOC_RESERVE
)
1214 EXT4_I(inode
)->i_delalloc_reserved_flag
= 1;
1216 * We need to check for EXT4 here because migrate
1217 * could have changed the inode type in between
1219 if (EXT4_I(inode
)->i_flags
& EXT4_EXTENTS_FL
) {
1220 retval
= ext4_ext_get_blocks(handle
, inode
, block
, max_blocks
,
1223 retval
= ext4_ind_get_blocks(handle
, inode
, block
,
1224 max_blocks
, bh
, flags
);
1226 if (retval
> 0 && buffer_new(bh
)) {
1228 * We allocated new blocks which will result in
1229 * i_data's format changing. Force the migrate
1230 * to fail by clearing migrate flags
1232 EXT4_I(inode
)->i_flags
= EXT4_I(inode
)->i_flags
&
1237 if (flags
& EXT4_GET_BLOCKS_DELALLOC_RESERVE
) {
1238 EXT4_I(inode
)->i_delalloc_reserved_flag
= 0;
1240 * Update reserved blocks/metadata blocks
1241 * after successful block allocation
1242 * which were deferred till now
1244 if ((retval
> 0) && buffer_delay(bh
))
1245 ext4_da_update_reserve_space(inode
, retval
);
1248 up_write((&EXT4_I(inode
)->i_data_sem
));
1252 /* Maximum number of blocks we map for direct IO at once. */
1253 #define DIO_MAX_BLOCKS 4096
1255 int ext4_get_block(struct inode
*inode
, sector_t iblock
,
1256 struct buffer_head
*bh_result
, int create
)
1258 handle_t
*handle
= ext4_journal_current_handle();
1259 int ret
= 0, started
= 0;
1260 unsigned max_blocks
= bh_result
->b_size
>> inode
->i_blkbits
;
1263 if (create
&& !handle
) {
1264 /* Direct IO write... */
1265 if (max_blocks
> DIO_MAX_BLOCKS
)
1266 max_blocks
= DIO_MAX_BLOCKS
;
1267 dio_credits
= ext4_chunk_trans_blocks(inode
, max_blocks
);
1268 handle
= ext4_journal_start(inode
, dio_credits
);
1269 if (IS_ERR(handle
)) {
1270 ret
= PTR_ERR(handle
);
1276 ret
= ext4_get_blocks(handle
, inode
, iblock
, max_blocks
, bh_result
,
1277 create
? EXT4_GET_BLOCKS_CREATE
: 0);
1279 bh_result
->b_size
= (ret
<< inode
->i_blkbits
);
1283 ext4_journal_stop(handle
);
1289 * `handle' can be NULL if create is zero
1291 struct buffer_head
*ext4_getblk(handle_t
*handle
, struct inode
*inode
,
1292 ext4_lblk_t block
, int create
, int *errp
)
1294 struct buffer_head dummy
;
1296 int flags
= EXT4_GET_BLOCKS_EXTEND_DISKSIZE
;
1298 J_ASSERT(handle
!= NULL
|| create
== 0);
1301 dummy
.b_blocknr
= -1000;
1302 buffer_trace_init(&dummy
.b_history
);
1304 flags
|= EXT4_GET_BLOCKS_CREATE
;
1305 err
= ext4_get_blocks(handle
, inode
, block
, 1, &dummy
, flags
);
1307 * ext4_get_blocks() returns number of blocks mapped. 0 in
1316 if (!err
&& buffer_mapped(&dummy
)) {
1317 struct buffer_head
*bh
;
1318 bh
= sb_getblk(inode
->i_sb
, dummy
.b_blocknr
);
1323 if (buffer_new(&dummy
)) {
1324 J_ASSERT(create
!= 0);
1325 J_ASSERT(handle
!= NULL
);
1328 * Now that we do not always journal data, we should
1329 * keep in mind whether this should always journal the
1330 * new buffer as metadata. For now, regular file
1331 * writes use ext4_get_block instead, so it's not a
1335 BUFFER_TRACE(bh
, "call get_create_access");
1336 fatal
= ext4_journal_get_create_access(handle
, bh
);
1337 if (!fatal
&& !buffer_uptodate(bh
)) {
1338 memset(bh
->b_data
, 0, inode
->i_sb
->s_blocksize
);
1339 set_buffer_uptodate(bh
);
1342 BUFFER_TRACE(bh
, "call ext4_handle_dirty_metadata");
1343 err
= ext4_handle_dirty_metadata(handle
, inode
, bh
);
1347 BUFFER_TRACE(bh
, "not a new buffer");
1360 struct buffer_head
*ext4_bread(handle_t
*handle
, struct inode
*inode
,
1361 ext4_lblk_t block
, int create
, int *err
)
1363 struct buffer_head
*bh
;
1365 bh
= ext4_getblk(handle
, inode
, block
, create
, err
);
1368 if (buffer_uptodate(bh
))
1370 ll_rw_block(READ_META
, 1, &bh
);
1372 if (buffer_uptodate(bh
))
1379 static int walk_page_buffers(handle_t
*handle
,
1380 struct buffer_head
*head
,
1384 int (*fn
)(handle_t
*handle
,
1385 struct buffer_head
*bh
))
1387 struct buffer_head
*bh
;
1388 unsigned block_start
, block_end
;
1389 unsigned blocksize
= head
->b_size
;
1391 struct buffer_head
*next
;
1393 for (bh
= head
, block_start
= 0;
1394 ret
== 0 && (bh
!= head
|| !block_start
);
1395 block_start
= block_end
, bh
= next
)
1397 next
= bh
->b_this_page
;
1398 block_end
= block_start
+ blocksize
;
1399 if (block_end
<= from
|| block_start
>= to
) {
1400 if (partial
&& !buffer_uptodate(bh
))
1404 err
= (*fn
)(handle
, bh
);
1412 * To preserve ordering, it is essential that the hole instantiation and
1413 * the data write be encapsulated in a single transaction. We cannot
1414 * close off a transaction and start a new one between the ext4_get_block()
1415 * and the commit_write(). So doing the jbd2_journal_start at the start of
1416 * prepare_write() is the right place.
1418 * Also, this function can nest inside ext4_writepage() ->
1419 * block_write_full_page(). In that case, we *know* that ext4_writepage()
1420 * has generated enough buffer credits to do the whole page. So we won't
1421 * block on the journal in that case, which is good, because the caller may
1424 * By accident, ext4 can be reentered when a transaction is open via
1425 * quota file writes. If we were to commit the transaction while thus
1426 * reentered, there can be a deadlock - we would be holding a quota
1427 * lock, and the commit would never complete if another thread had a
1428 * transaction open and was blocking on the quota lock - a ranking
1431 * So what we do is to rely on the fact that jbd2_journal_stop/journal_start
1432 * will _not_ run commit under these circumstances because handle->h_ref
1433 * is elevated. We'll still have enough credits for the tiny quotafile
1436 static int do_journal_get_write_access(handle_t
*handle
,
1437 struct buffer_head
*bh
)
1439 if (!buffer_mapped(bh
) || buffer_freed(bh
))
1441 return ext4_journal_get_write_access(handle
, bh
);
1444 static int ext4_write_begin(struct file
*file
, struct address_space
*mapping
,
1445 loff_t pos
, unsigned len
, unsigned flags
,
1446 struct page
**pagep
, void **fsdata
)
1448 struct inode
*inode
= mapping
->host
;
1449 int ret
, needed_blocks
= ext4_writepage_trans_blocks(inode
);
1456 trace_mark(ext4_write_begin
,
1457 "dev %s ino %lu pos %llu len %u flags %u",
1458 inode
->i_sb
->s_id
, inode
->i_ino
,
1459 (unsigned long long) pos
, len
, flags
);
1460 index
= pos
>> PAGE_CACHE_SHIFT
;
1461 from
= pos
& (PAGE_CACHE_SIZE
- 1);
1465 handle
= ext4_journal_start(inode
, needed_blocks
);
1466 if (IS_ERR(handle
)) {
1467 ret
= PTR_ERR(handle
);
1471 /* We cannot recurse into the filesystem as the transaction is already
1473 flags
|= AOP_FLAG_NOFS
;
1475 page
= grab_cache_page_write_begin(mapping
, index
, flags
);
1477 ext4_journal_stop(handle
);
1483 ret
= block_write_begin(file
, mapping
, pos
, len
, flags
, pagep
, fsdata
,
1486 if (!ret
&& ext4_should_journal_data(inode
)) {
1487 ret
= walk_page_buffers(handle
, page_buffers(page
),
1488 from
, to
, NULL
, do_journal_get_write_access
);
1493 ext4_journal_stop(handle
);
1494 page_cache_release(page
);
1496 * block_write_begin may have instantiated a few blocks
1497 * outside i_size. Trim these off again. Don't need
1498 * i_size_read because we hold i_mutex.
1500 if (pos
+ len
> inode
->i_size
)
1501 vmtruncate(inode
, inode
->i_size
);
1504 if (ret
== -ENOSPC
&& ext4_should_retry_alloc(inode
->i_sb
, &retries
))
1510 /* For write_end() in data=journal mode */
1511 static int write_end_fn(handle_t
*handle
, struct buffer_head
*bh
)
1513 if (!buffer_mapped(bh
) || buffer_freed(bh
))
1515 set_buffer_uptodate(bh
);
1516 return ext4_handle_dirty_metadata(handle
, NULL
, bh
);
1520 * We need to pick up the new inode size which generic_commit_write gave us
1521 * `file' can be NULL - eg, when called from page_symlink().
1523 * ext4 never places buffers on inode->i_mapping->private_list. metadata
1524 * buffers are managed internally.
1526 static int ext4_ordered_write_end(struct file
*file
,
1527 struct address_space
*mapping
,
1528 loff_t pos
, unsigned len
, unsigned copied
,
1529 struct page
*page
, void *fsdata
)
1531 handle_t
*handle
= ext4_journal_current_handle();
1532 struct inode
*inode
= mapping
->host
;
1535 trace_mark(ext4_ordered_write_end
,
1536 "dev %s ino %lu pos %llu len %u copied %u",
1537 inode
->i_sb
->s_id
, inode
->i_ino
,
1538 (unsigned long long) pos
, len
, copied
);
1539 ret
= ext4_jbd2_file_inode(handle
, inode
);
1544 new_i_size
= pos
+ copied
;
1545 if (new_i_size
> EXT4_I(inode
)->i_disksize
) {
1546 ext4_update_i_disksize(inode
, new_i_size
);
1547 /* We need to mark inode dirty even if
1548 * new_i_size is less that inode->i_size
1549 * bu greater than i_disksize.(hint delalloc)
1551 ext4_mark_inode_dirty(handle
, inode
);
1554 ret2
= generic_write_end(file
, mapping
, pos
, len
, copied
,
1560 ret2
= ext4_journal_stop(handle
);
1564 return ret
? ret
: copied
;
1567 static int ext4_writeback_write_end(struct file
*file
,
1568 struct address_space
*mapping
,
1569 loff_t pos
, unsigned len
, unsigned copied
,
1570 struct page
*page
, void *fsdata
)
1572 handle_t
*handle
= ext4_journal_current_handle();
1573 struct inode
*inode
= mapping
->host
;
1577 trace_mark(ext4_writeback_write_end
,
1578 "dev %s ino %lu pos %llu len %u copied %u",
1579 inode
->i_sb
->s_id
, inode
->i_ino
,
1580 (unsigned long long) pos
, len
, copied
);
1581 new_i_size
= pos
+ copied
;
1582 if (new_i_size
> EXT4_I(inode
)->i_disksize
) {
1583 ext4_update_i_disksize(inode
, new_i_size
);
1584 /* We need to mark inode dirty even if
1585 * new_i_size is less that inode->i_size
1586 * bu greater than i_disksize.(hint delalloc)
1588 ext4_mark_inode_dirty(handle
, inode
);
1591 ret2
= generic_write_end(file
, mapping
, pos
, len
, copied
,
1597 ret2
= ext4_journal_stop(handle
);
1601 return ret
? ret
: copied
;
1604 static int ext4_journalled_write_end(struct file
*file
,
1605 struct address_space
*mapping
,
1606 loff_t pos
, unsigned len
, unsigned copied
,
1607 struct page
*page
, void *fsdata
)
1609 handle_t
*handle
= ext4_journal_current_handle();
1610 struct inode
*inode
= mapping
->host
;
1616 trace_mark(ext4_journalled_write_end
,
1617 "dev %s ino %lu pos %llu len %u copied %u",
1618 inode
->i_sb
->s_id
, inode
->i_ino
,
1619 (unsigned long long) pos
, len
, copied
);
1620 from
= pos
& (PAGE_CACHE_SIZE
- 1);
1624 if (!PageUptodate(page
))
1626 page_zero_new_buffers(page
, from
+copied
, to
);
1629 ret
= walk_page_buffers(handle
, page_buffers(page
), from
,
1630 to
, &partial
, write_end_fn
);
1632 SetPageUptodate(page
);
1633 new_i_size
= pos
+ copied
;
1634 if (new_i_size
> inode
->i_size
)
1635 i_size_write(inode
, pos
+copied
);
1636 EXT4_I(inode
)->i_state
|= EXT4_STATE_JDATA
;
1637 if (new_i_size
> EXT4_I(inode
)->i_disksize
) {
1638 ext4_update_i_disksize(inode
, new_i_size
);
1639 ret2
= ext4_mark_inode_dirty(handle
, inode
);
1645 ret2
= ext4_journal_stop(handle
);
1648 page_cache_release(page
);
1650 return ret
? ret
: copied
;
1653 static int ext4_da_reserve_space(struct inode
*inode
, int nrblocks
)
1656 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
1657 unsigned long md_needed
, mdblocks
, total
= 0;
1660 * recalculate the amount of metadata blocks to reserve
1661 * in order to allocate nrblocks
1662 * worse case is one extent per block
1665 spin_lock(&EXT4_I(inode
)->i_block_reservation_lock
);
1666 total
= EXT4_I(inode
)->i_reserved_data_blocks
+ nrblocks
;
1667 mdblocks
= ext4_calc_metadata_amount(inode
, total
);
1668 BUG_ON(mdblocks
< EXT4_I(inode
)->i_reserved_meta_blocks
);
1670 md_needed
= mdblocks
- EXT4_I(inode
)->i_reserved_meta_blocks
;
1671 total
= md_needed
+ nrblocks
;
1674 * Make quota reservation here to prevent quota overflow
1675 * later. Real quota accounting is done at pages writeout
1678 if (vfs_dq_reserve_block(inode
, total
)) {
1679 spin_unlock(&EXT4_I(inode
)->i_block_reservation_lock
);
1683 if (ext4_claim_free_blocks(sbi
, total
)) {
1684 spin_unlock(&EXT4_I(inode
)->i_block_reservation_lock
);
1685 if (ext4_should_retry_alloc(inode
->i_sb
, &retries
)) {
1689 vfs_dq_release_reservation_block(inode
, total
);
1692 EXT4_I(inode
)->i_reserved_data_blocks
+= nrblocks
;
1693 EXT4_I(inode
)->i_reserved_meta_blocks
= mdblocks
;
1695 spin_unlock(&EXT4_I(inode
)->i_block_reservation_lock
);
1696 return 0; /* success */
1699 static void ext4_da_release_space(struct inode
*inode
, int to_free
)
1701 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
1702 int total
, mdb
, mdb_free
, release
;
1705 return; /* Nothing to release, exit */
1707 spin_lock(&EXT4_I(inode
)->i_block_reservation_lock
);
1709 if (!EXT4_I(inode
)->i_reserved_data_blocks
) {
1711 * if there is no reserved blocks, but we try to free some
1712 * then the counter is messed up somewhere.
1713 * but since this function is called from invalidate
1714 * page, it's harmless to return without any action
1716 printk(KERN_INFO
"ext4 delalloc try to release %d reserved "
1717 "blocks for inode %lu, but there is no reserved "
1718 "data blocks\n", to_free
, inode
->i_ino
);
1719 spin_unlock(&EXT4_I(inode
)->i_block_reservation_lock
);
1723 /* recalculate the number of metablocks still need to be reserved */
1724 total
= EXT4_I(inode
)->i_reserved_data_blocks
- to_free
;
1725 mdb
= ext4_calc_metadata_amount(inode
, total
);
1727 /* figure out how many metablocks to release */
1728 BUG_ON(mdb
> EXT4_I(inode
)->i_reserved_meta_blocks
);
1729 mdb_free
= EXT4_I(inode
)->i_reserved_meta_blocks
- mdb
;
1731 release
= to_free
+ mdb_free
;
1733 /* update fs dirty blocks counter for truncate case */
1734 percpu_counter_sub(&sbi
->s_dirtyblocks_counter
, release
);
1736 /* update per-inode reservations */
1737 BUG_ON(to_free
> EXT4_I(inode
)->i_reserved_data_blocks
);
1738 EXT4_I(inode
)->i_reserved_data_blocks
-= to_free
;
1740 BUG_ON(mdb
> EXT4_I(inode
)->i_reserved_meta_blocks
);
1741 EXT4_I(inode
)->i_reserved_meta_blocks
= mdb
;
1742 spin_unlock(&EXT4_I(inode
)->i_block_reservation_lock
);
1744 vfs_dq_release_reservation_block(inode
, release
);
1747 static void ext4_da_page_release_reservation(struct page
*page
,
1748 unsigned long offset
)
1751 struct buffer_head
*head
, *bh
;
1752 unsigned int curr_off
= 0;
1754 head
= page_buffers(page
);
1757 unsigned int next_off
= curr_off
+ bh
->b_size
;
1759 if ((offset
<= curr_off
) && (buffer_delay(bh
))) {
1761 clear_buffer_delay(bh
);
1763 curr_off
= next_off
;
1764 } while ((bh
= bh
->b_this_page
) != head
);
1765 ext4_da_release_space(page
->mapping
->host
, to_release
);
1769 * Delayed allocation stuff
1772 struct mpage_da_data
{
1773 struct inode
*inode
;
1774 sector_t b_blocknr
; /* start block number of extent */
1775 size_t b_size
; /* size of extent */
1776 unsigned long b_state
; /* state of the extent */
1777 unsigned long first_page
, next_page
; /* extent of pages */
1778 struct writeback_control
*wbc
;
1785 * mpage_da_submit_io - walks through extent of pages and try to write
1786 * them with writepage() call back
1788 * @mpd->inode: inode
1789 * @mpd->first_page: first page of the extent
1790 * @mpd->next_page: page after the last page of the extent
1792 * By the time mpage_da_submit_io() is called we expect all blocks
1793 * to be allocated. this may be wrong if allocation failed.
1795 * As pages are already locked by write_cache_pages(), we can't use it
1797 static int mpage_da_submit_io(struct mpage_da_data
*mpd
)
1800 struct pagevec pvec
;
1801 unsigned long index
, end
;
1802 int ret
= 0, err
, nr_pages
, i
;
1803 struct inode
*inode
= mpd
->inode
;
1804 struct address_space
*mapping
= inode
->i_mapping
;
1806 BUG_ON(mpd
->next_page
<= mpd
->first_page
);
1808 * We need to start from the first_page to the next_page - 1
1809 * to make sure we also write the mapped dirty buffer_heads.
1810 * If we look at mpd->b_blocknr we would only be looking
1811 * at the currently mapped buffer_heads.
1813 index
= mpd
->first_page
;
1814 end
= mpd
->next_page
- 1;
1816 pagevec_init(&pvec
, 0);
1817 while (index
<= end
) {
1818 nr_pages
= pagevec_lookup(&pvec
, mapping
, index
, PAGEVEC_SIZE
);
1821 for (i
= 0; i
< nr_pages
; i
++) {
1822 struct page
*page
= pvec
.pages
[i
];
1824 index
= page
->index
;
1829 BUG_ON(!PageLocked(page
));
1830 BUG_ON(PageWriteback(page
));
1832 pages_skipped
= mpd
->wbc
->pages_skipped
;
1833 err
= mapping
->a_ops
->writepage(page
, mpd
->wbc
);
1834 if (!err
&& (pages_skipped
== mpd
->wbc
->pages_skipped
))
1836 * have successfully written the page
1837 * without skipping the same
1839 mpd
->pages_written
++;
1841 * In error case, we have to continue because
1842 * remaining pages are still locked
1843 * XXX: unlock and re-dirty them?
1848 pagevec_release(&pvec
);
1854 * mpage_put_bnr_to_bhs - walk blocks and assign them actual numbers
1856 * @mpd->inode - inode to walk through
1857 * @exbh->b_blocknr - first block on a disk
1858 * @exbh->b_size - amount of space in bytes
1859 * @logical - first logical block to start assignment with
1861 * the function goes through all passed space and put actual disk
1862 * block numbers into buffer heads, dropping BH_Delay and BH_Unwritten
1864 static void mpage_put_bnr_to_bhs(struct mpage_da_data
*mpd
, sector_t logical
,
1865 struct buffer_head
*exbh
)
1867 struct inode
*inode
= mpd
->inode
;
1868 struct address_space
*mapping
= inode
->i_mapping
;
1869 int blocks
= exbh
->b_size
>> inode
->i_blkbits
;
1870 sector_t pblock
= exbh
->b_blocknr
, cur_logical
;
1871 struct buffer_head
*head
, *bh
;
1873 struct pagevec pvec
;
1876 index
= logical
>> (PAGE_CACHE_SHIFT
- inode
->i_blkbits
);
1877 end
= (logical
+ blocks
- 1) >> (PAGE_CACHE_SHIFT
- inode
->i_blkbits
);
1878 cur_logical
= index
<< (PAGE_CACHE_SHIFT
- inode
->i_blkbits
);
1880 pagevec_init(&pvec
, 0);
1882 while (index
<= end
) {
1883 /* XXX: optimize tail */
1884 nr_pages
= pagevec_lookup(&pvec
, mapping
, index
, PAGEVEC_SIZE
);
1887 for (i
= 0; i
< nr_pages
; i
++) {
1888 struct page
*page
= pvec
.pages
[i
];
1890 index
= page
->index
;
1895 BUG_ON(!PageLocked(page
));
1896 BUG_ON(PageWriteback(page
));
1897 BUG_ON(!page_has_buffers(page
));
1899 bh
= page_buffers(page
);
1902 /* skip blocks out of the range */
1904 if (cur_logical
>= logical
)
1907 } while ((bh
= bh
->b_this_page
) != head
);
1910 if (cur_logical
>= logical
+ blocks
)
1913 if (buffer_delay(bh
) ||
1914 buffer_unwritten(bh
)) {
1916 BUG_ON(bh
->b_bdev
!= inode
->i_sb
->s_bdev
);
1918 if (buffer_delay(bh
)) {
1919 clear_buffer_delay(bh
);
1920 bh
->b_blocknr
= pblock
;
1923 * unwritten already should have
1924 * blocknr assigned. Verify that
1926 clear_buffer_unwritten(bh
);
1927 BUG_ON(bh
->b_blocknr
!= pblock
);
1930 } else if (buffer_mapped(bh
))
1931 BUG_ON(bh
->b_blocknr
!= pblock
);
1935 } while ((bh
= bh
->b_this_page
) != head
);
1937 pagevec_release(&pvec
);
1943 * __unmap_underlying_blocks - just a helper function to unmap
1944 * set of blocks described by @bh
1946 static inline void __unmap_underlying_blocks(struct inode
*inode
,
1947 struct buffer_head
*bh
)
1949 struct block_device
*bdev
= inode
->i_sb
->s_bdev
;
1952 blocks
= bh
->b_size
>> inode
->i_blkbits
;
1953 for (i
= 0; i
< blocks
; i
++)
1954 unmap_underlying_metadata(bdev
, bh
->b_blocknr
+ i
);
1957 static void ext4_da_block_invalidatepages(struct mpage_da_data
*mpd
,
1958 sector_t logical
, long blk_cnt
)
1962 struct pagevec pvec
;
1963 struct inode
*inode
= mpd
->inode
;
1964 struct address_space
*mapping
= inode
->i_mapping
;
1966 index
= logical
>> (PAGE_CACHE_SHIFT
- inode
->i_blkbits
);
1967 end
= (logical
+ blk_cnt
- 1) >>
1968 (PAGE_CACHE_SHIFT
- inode
->i_blkbits
);
1969 while (index
<= end
) {
1970 nr_pages
= pagevec_lookup(&pvec
, mapping
, index
, PAGEVEC_SIZE
);
1973 for (i
= 0; i
< nr_pages
; i
++) {
1974 struct page
*page
= pvec
.pages
[i
];
1975 index
= page
->index
;
1980 BUG_ON(!PageLocked(page
));
1981 BUG_ON(PageWriteback(page
));
1982 block_invalidatepage(page
, 0);
1983 ClearPageUptodate(page
);
1990 static void ext4_print_free_blocks(struct inode
*inode
)
1992 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
1993 printk(KERN_EMERG
"Total free blocks count %lld\n",
1994 ext4_count_free_blocks(inode
->i_sb
));
1995 printk(KERN_EMERG
"Free/Dirty block details\n");
1996 printk(KERN_EMERG
"free_blocks=%lld\n",
1997 (long long)percpu_counter_sum(&sbi
->s_freeblocks_counter
));
1998 printk(KERN_EMERG
"dirty_blocks=%lld\n",
1999 (long long)percpu_counter_sum(&sbi
->s_dirtyblocks_counter
));
2000 printk(KERN_EMERG
"Block reservation details\n");
2001 printk(KERN_EMERG
"i_reserved_data_blocks=%u\n",
2002 EXT4_I(inode
)->i_reserved_data_blocks
);
2003 printk(KERN_EMERG
"i_reserved_meta_blocks=%u\n",
2004 EXT4_I(inode
)->i_reserved_meta_blocks
);
2009 * mpage_da_map_blocks - go through given space
2011 * @mpd - bh describing space
2013 * The function skips space we know is already mapped to disk blocks.
2016 static int mpage_da_map_blocks(struct mpage_da_data
*mpd
)
2019 struct buffer_head
new;
2020 sector_t next
= mpd
->b_blocknr
;
2021 unsigned max_blocks
= mpd
->b_size
>> mpd
->inode
->i_blkbits
;
2022 loff_t disksize
= EXT4_I(mpd
->inode
)->i_disksize
;
2023 handle_t
*handle
= NULL
;
2026 * We consider only non-mapped and non-allocated blocks
2028 if ((mpd
->b_state
& (1 << BH_Mapped
)) &&
2029 !(mpd
->b_state
& (1 << BH_Delay
)) &&
2030 !(mpd
->b_state
& (1 << BH_Unwritten
)))
2034 * If we didn't accumulate anything to write simply return
2039 handle
= ext4_journal_current_handle();
2043 * We need to make sure the BH_Delay flag is passed down to
2044 * ext4_da_get_block_write(), since it calls ext4_get_blocks()
2045 * with the EXT4_GET_BLOCKS_DELALLOC_RESERVE flag. This flag
2046 * causes ext4_get_blocks() to call
2047 * ext4_da_update_reserve_space() if the passed buffer head
2048 * has the BH_Delay flag set. In the future, once we clean up
2049 * the interfaces to ext4_get_blocks(), we should pass in a
2050 * separate flag which requests that the delayed allocation
2051 * statistics should be updated, instead of depending on the
2052 * state information getting passed down via the map_bh's
2053 * state bitmasks plus the magic
2054 * EXT4_GET_BLOCKS_DELALLOC_RESERVE flag.
2056 new.b_state
= mpd
->b_state
& (1 << BH_Delay
);
2057 blks
= ext4_get_blocks(handle
, mpd
->inode
, next
, max_blocks
,
2058 &new, EXT4_GET_BLOCKS_CREATE
|
2059 EXT4_GET_BLOCKS_DELALLOC_RESERVE
);
2063 * If get block returns with error we simply
2064 * return. Later writepage will redirty the page and
2065 * writepages will find the dirty page again
2070 if (err
== -ENOSPC
&&
2071 ext4_count_free_blocks(mpd
->inode
->i_sb
)) {
2077 * get block failure will cause us to loop in
2078 * writepages, because a_ops->writepage won't be able
2079 * to make progress. The page will be redirtied by
2080 * writepage and writepages will again try to write
2083 printk(KERN_EMERG
"%s block allocation failed for inode %lu "
2084 "at logical offset %llu with max blocks "
2085 "%zd with error %d\n",
2086 __func__
, mpd
->inode
->i_ino
,
2087 (unsigned long long)next
,
2088 mpd
->b_size
>> mpd
->inode
->i_blkbits
, err
);
2089 printk(KERN_EMERG
"This should not happen.!! "
2090 "Data will be lost\n");
2091 if (err
== -ENOSPC
) {
2092 ext4_print_free_blocks(mpd
->inode
);
2094 /* invalidate all the pages */
2095 ext4_da_block_invalidatepages(mpd
, next
,
2096 mpd
->b_size
>> mpd
->inode
->i_blkbits
);
2101 new.b_size
= (blks
<< mpd
->inode
->i_blkbits
);
2103 if (buffer_new(&new))
2104 __unmap_underlying_blocks(mpd
->inode
, &new);
2107 * If blocks are delayed marked, we need to
2108 * put actual blocknr and drop delayed bit
2110 if ((mpd
->b_state
& (1 << BH_Delay
)) ||
2111 (mpd
->b_state
& (1 << BH_Unwritten
)))
2112 mpage_put_bnr_to_bhs(mpd
, next
, &new);
2114 if (ext4_should_order_data(mpd
->inode
)) {
2115 err
= ext4_jbd2_file_inode(handle
, mpd
->inode
);
2121 * Update on-disk size along with block allocation we don't
2122 * use EXT4_GET_BLOCKS_EXTEND_DISKSIZE as size may change
2123 * within already allocated block -bzzz
2125 disksize
= ((loff_t
) next
+ blks
) << mpd
->inode
->i_blkbits
;
2126 if (disksize
> i_size_read(mpd
->inode
))
2127 disksize
= i_size_read(mpd
->inode
);
2128 if (disksize
> EXT4_I(mpd
->inode
)->i_disksize
) {
2129 ext4_update_i_disksize(mpd
->inode
, disksize
);
2130 return ext4_mark_inode_dirty(handle
, mpd
->inode
);
2136 #define BH_FLAGS ((1 << BH_Uptodate) | (1 << BH_Mapped) | \
2137 (1 << BH_Delay) | (1 << BH_Unwritten))
2140 * mpage_add_bh_to_extent - try to add one more block to extent of blocks
2142 * @mpd->lbh - extent of blocks
2143 * @logical - logical number of the block in the file
2144 * @bh - bh of the block (used to access block's state)
2146 * the function is used to collect contig. blocks in same state
2148 static void mpage_add_bh_to_extent(struct mpage_da_data
*mpd
,
2149 sector_t logical
, size_t b_size
,
2150 unsigned long b_state
)
2153 int nrblocks
= mpd
->b_size
>> mpd
->inode
->i_blkbits
;
2155 /* check if thereserved journal credits might overflow */
2156 if (!(EXT4_I(mpd
->inode
)->i_flags
& EXT4_EXTENTS_FL
)) {
2157 if (nrblocks
>= EXT4_MAX_TRANS_DATA
) {
2159 * With non-extent format we are limited by the journal
2160 * credit available. Total credit needed to insert
2161 * nrblocks contiguous blocks is dependent on the
2162 * nrblocks. So limit nrblocks.
2165 } else if ((nrblocks
+ (b_size
>> mpd
->inode
->i_blkbits
)) >
2166 EXT4_MAX_TRANS_DATA
) {
2168 * Adding the new buffer_head would make it cross the
2169 * allowed limit for which we have journal credit
2170 * reserved. So limit the new bh->b_size
2172 b_size
= (EXT4_MAX_TRANS_DATA
- nrblocks
) <<
2173 mpd
->inode
->i_blkbits
;
2174 /* we will do mpage_da_submit_io in the next loop */
2178 * First block in the extent
2180 if (mpd
->b_size
== 0) {
2181 mpd
->b_blocknr
= logical
;
2182 mpd
->b_size
= b_size
;
2183 mpd
->b_state
= b_state
& BH_FLAGS
;
2187 next
= mpd
->b_blocknr
+ nrblocks
;
2189 * Can we merge the block to our big extent?
2191 if (logical
== next
&& (b_state
& BH_FLAGS
) == mpd
->b_state
) {
2192 mpd
->b_size
+= b_size
;
2198 * We couldn't merge the block to our extent, so we
2199 * need to flush current extent and start new one
2201 if (mpage_da_map_blocks(mpd
) == 0)
2202 mpage_da_submit_io(mpd
);
2207 static int ext4_bh_unmapped_or_delay(handle_t
*handle
, struct buffer_head
*bh
)
2210 * unmapped buffer is possible for holes.
2211 * delay buffer is possible with delayed allocation.
2212 * We also need to consider unwritten buffer as unmapped.
2214 return (!buffer_mapped(bh
) || buffer_delay(bh
) ||
2215 buffer_unwritten(bh
)) && buffer_dirty(bh
);
2219 * __mpage_da_writepage - finds extent of pages and blocks
2221 * @page: page to consider
2222 * @wbc: not used, we just follow rules
2225 * The function finds extents of pages and scan them for all blocks.
2227 static int __mpage_da_writepage(struct page
*page
,
2228 struct writeback_control
*wbc
, void *data
)
2230 struct mpage_da_data
*mpd
= data
;
2231 struct inode
*inode
= mpd
->inode
;
2232 struct buffer_head
*bh
, *head
;
2237 * Rest of the page in the page_vec
2238 * redirty then and skip then. We will
2239 * try to to write them again after
2240 * starting a new transaction
2242 redirty_page_for_writepage(wbc
, page
);
2244 return MPAGE_DA_EXTENT_TAIL
;
2247 * Can we merge this page to current extent?
2249 if (mpd
->next_page
!= page
->index
) {
2251 * Nope, we can't. So, we map non-allocated blocks
2252 * and start IO on them using writepage()
2254 if (mpd
->next_page
!= mpd
->first_page
) {
2255 if (mpage_da_map_blocks(mpd
) == 0)
2256 mpage_da_submit_io(mpd
);
2258 * skip rest of the page in the page_vec
2261 redirty_page_for_writepage(wbc
, page
);
2263 return MPAGE_DA_EXTENT_TAIL
;
2267 * Start next extent of pages ...
2269 mpd
->first_page
= page
->index
;
2279 mpd
->next_page
= page
->index
+ 1;
2280 logical
= (sector_t
) page
->index
<<
2281 (PAGE_CACHE_SHIFT
- inode
->i_blkbits
);
2283 if (!page_has_buffers(page
)) {
2284 mpage_add_bh_to_extent(mpd
, logical
, PAGE_CACHE_SIZE
,
2285 (1 << BH_Dirty
) | (1 << BH_Uptodate
));
2287 return MPAGE_DA_EXTENT_TAIL
;
2290 * Page with regular buffer heads, just add all dirty ones
2292 head
= page_buffers(page
);
2295 BUG_ON(buffer_locked(bh
));
2297 * We need to try to allocate
2298 * unmapped blocks in the same page.
2299 * Otherwise we won't make progress
2300 * with the page in ext4_da_writepage
2302 if (ext4_bh_unmapped_or_delay(NULL
, bh
)) {
2303 mpage_add_bh_to_extent(mpd
, logical
,
2307 return MPAGE_DA_EXTENT_TAIL
;
2308 } else if (buffer_dirty(bh
) && (buffer_mapped(bh
))) {
2310 * mapped dirty buffer. We need to update
2311 * the b_state because we look at
2312 * b_state in mpage_da_map_blocks. We don't
2313 * update b_size because if we find an
2314 * unmapped buffer_head later we need to
2315 * use the b_state flag of that buffer_head.
2317 if (mpd
->b_size
== 0)
2318 mpd
->b_state
= bh
->b_state
& BH_FLAGS
;
2321 } while ((bh
= bh
->b_this_page
) != head
);
2328 * This is a special get_blocks_t callback which is used by
2329 * ext4_da_write_begin(). It will either return mapped block or
2330 * reserve space for a single block.
2332 * For delayed buffer_head we have BH_Mapped, BH_New, BH_Delay set.
2333 * We also have b_blocknr = -1 and b_bdev initialized properly
2335 * For unwritten buffer_head we have BH_Mapped, BH_New, BH_Unwritten set.
2336 * We also have b_blocknr = physicalblock mapping unwritten extent and b_bdev
2337 * initialized properly.
2339 static int ext4_da_get_block_prep(struct inode
*inode
, sector_t iblock
,
2340 struct buffer_head
*bh_result
, int create
)
2343 sector_t invalid_block
= ~((sector_t
) 0xffff);
2345 if (invalid_block
< ext4_blocks_count(EXT4_SB(inode
->i_sb
)->s_es
))
2348 BUG_ON(create
== 0);
2349 BUG_ON(bh_result
->b_size
!= inode
->i_sb
->s_blocksize
);
2352 * first, we need to know whether the block is allocated already
2353 * preallocated blocks are unmapped but should treated
2354 * the same as allocated blocks.
2356 ret
= ext4_get_blocks(NULL
, inode
, iblock
, 1, bh_result
, 0);
2357 if ((ret
== 0) && !buffer_delay(bh_result
)) {
2358 /* the block isn't (pre)allocated yet, let's reserve space */
2360 * XXX: __block_prepare_write() unmaps passed block,
2363 ret
= ext4_da_reserve_space(inode
, 1);
2365 /* not enough space to reserve */
2368 map_bh(bh_result
, inode
->i_sb
, invalid_block
);
2369 set_buffer_new(bh_result
);
2370 set_buffer_delay(bh_result
);
2371 } else if (ret
> 0) {
2372 bh_result
->b_size
= (ret
<< inode
->i_blkbits
);
2373 if (buffer_unwritten(bh_result
)) {
2374 /* A delayed write to unwritten bh should
2375 * be marked new and mapped. Mapped ensures
2376 * that we don't do get_block multiple times
2377 * when we write to the same offset and new
2378 * ensures that we do proper zero out for
2381 set_buffer_new(bh_result
);
2382 set_buffer_mapped(bh_result
);
2391 * This function is used as a standard get_block_t calback function
2392 * when there is no desire to allocate any blocks. It is used as a
2393 * callback function for block_prepare_write(), nobh_writepage(), and
2394 * block_write_full_page(). These functions should only try to map a
2395 * single block at a time.
2397 * Since this function doesn't do block allocations even if the caller
2398 * requests it by passing in create=1, it is critically important that
2399 * any caller checks to make sure that any buffer heads are returned
2400 * by this function are either all already mapped or marked for
2401 * delayed allocation before calling nobh_writepage() or
2402 * block_write_full_page(). Otherwise, b_blocknr could be left
2403 * unitialized, and the page write functions will be taken by
2406 static int noalloc_get_block_write(struct inode
*inode
, sector_t iblock
,
2407 struct buffer_head
*bh_result
, int create
)
2410 unsigned max_blocks
= bh_result
->b_size
>> inode
->i_blkbits
;
2412 BUG_ON(bh_result
->b_size
!= inode
->i_sb
->s_blocksize
);
2415 * we don't want to do block allocation in writepage
2416 * so call get_block_wrap with create = 0
2418 ret
= ext4_get_blocks(NULL
, inode
, iblock
, max_blocks
, bh_result
, 0);
2419 BUG_ON(create
&& ret
== 0);
2421 bh_result
->b_size
= (ret
<< inode
->i_blkbits
);
2428 * This function can get called via...
2429 * - ext4_da_writepages after taking page lock (have journal handle)
2430 * - journal_submit_inode_data_buffers (no journal handle)
2431 * - shrink_page_list via pdflush (no journal handle)
2432 * - grab_page_cache when doing write_begin (have journal handle)
2434 static int ext4_da_writepage(struct page
*page
,
2435 struct writeback_control
*wbc
)
2440 struct buffer_head
*page_bufs
;
2441 struct inode
*inode
= page
->mapping
->host
;
2443 trace_mark(ext4_da_writepage
,
2444 "dev %s ino %lu page_index %lu",
2445 inode
->i_sb
->s_id
, inode
->i_ino
, page
->index
);
2446 size
= i_size_read(inode
);
2447 if (page
->index
== size
>> PAGE_CACHE_SHIFT
)
2448 len
= size
& ~PAGE_CACHE_MASK
;
2450 len
= PAGE_CACHE_SIZE
;
2452 if (page_has_buffers(page
)) {
2453 page_bufs
= page_buffers(page
);
2454 if (walk_page_buffers(NULL
, page_bufs
, 0, len
, NULL
,
2455 ext4_bh_unmapped_or_delay
)) {
2457 * We don't want to do block allocation
2458 * So redirty the page and return
2459 * We may reach here when we do a journal commit
2460 * via journal_submit_inode_data_buffers.
2461 * If we don't have mapping block we just ignore
2462 * them. We can also reach here via shrink_page_list
2464 redirty_page_for_writepage(wbc
, page
);
2470 * The test for page_has_buffers() is subtle:
2471 * We know the page is dirty but it lost buffers. That means
2472 * that at some moment in time after write_begin()/write_end()
2473 * has been called all buffers have been clean and thus they
2474 * must have been written at least once. So they are all
2475 * mapped and we can happily proceed with mapping them
2476 * and writing the page.
2478 * Try to initialize the buffer_heads and check whether
2479 * all are mapped and non delay. We don't want to
2480 * do block allocation here.
2482 ret
= block_prepare_write(page
, 0, PAGE_CACHE_SIZE
,
2483 noalloc_get_block_write
);
2485 page_bufs
= page_buffers(page
);
2486 /* check whether all are mapped and non delay */
2487 if (walk_page_buffers(NULL
, page_bufs
, 0, len
, NULL
,
2488 ext4_bh_unmapped_or_delay
)) {
2489 redirty_page_for_writepage(wbc
, page
);
2495 * We can't do block allocation here
2496 * so just redity the page and unlock
2499 redirty_page_for_writepage(wbc
, page
);
2503 /* now mark the buffer_heads as dirty and uptodate */
2504 block_commit_write(page
, 0, PAGE_CACHE_SIZE
);
2507 if (test_opt(inode
->i_sb
, NOBH
) && ext4_should_writeback_data(inode
))
2508 ret
= nobh_writepage(page
, noalloc_get_block_write
, wbc
);
2510 ret
= block_write_full_page(page
, noalloc_get_block_write
,
2517 * This is called via ext4_da_writepages() to
2518 * calulate the total number of credits to reserve to fit
2519 * a single extent allocation into a single transaction,
2520 * ext4_da_writpeages() will loop calling this before
2521 * the block allocation.
2524 static int ext4_da_writepages_trans_blocks(struct inode
*inode
)
2526 int max_blocks
= EXT4_I(inode
)->i_reserved_data_blocks
;
2529 * With non-extent format the journal credit needed to
2530 * insert nrblocks contiguous block is dependent on
2531 * number of contiguous block. So we will limit
2532 * number of contiguous block to a sane value
2534 if (!(inode
->i_flags
& EXT4_EXTENTS_FL
) &&
2535 (max_blocks
> EXT4_MAX_TRANS_DATA
))
2536 max_blocks
= EXT4_MAX_TRANS_DATA
;
2538 return ext4_chunk_trans_blocks(inode
, max_blocks
);
2541 static int ext4_da_writepages(struct address_space
*mapping
,
2542 struct writeback_control
*wbc
)
2545 int range_whole
= 0;
2546 handle_t
*handle
= NULL
;
2547 struct mpage_da_data mpd
;
2548 struct inode
*inode
= mapping
->host
;
2549 int no_nrwrite_index_update
;
2550 int pages_written
= 0;
2552 int range_cyclic
, cycled
= 1, io_done
= 0;
2553 int needed_blocks
, ret
= 0, nr_to_writebump
= 0;
2554 struct ext4_sb_info
*sbi
= EXT4_SB(mapping
->host
->i_sb
);
2556 trace_mark(ext4_da_writepages
,
2557 "dev %s ino %lu nr_t_write %ld "
2558 "pages_skipped %ld range_start %llu "
2559 "range_end %llu nonblocking %d "
2560 "for_kupdate %d for_reclaim %d "
2561 "for_writepages %d range_cyclic %d",
2562 inode
->i_sb
->s_id
, inode
->i_ino
,
2563 wbc
->nr_to_write
, wbc
->pages_skipped
,
2564 (unsigned long long) wbc
->range_start
,
2565 (unsigned long long) wbc
->range_end
,
2566 wbc
->nonblocking
, wbc
->for_kupdate
,
2567 wbc
->for_reclaim
, wbc
->for_writepages
,
2571 * No pages to write? This is mainly a kludge to avoid starting
2572 * a transaction for special inodes like journal inode on last iput()
2573 * because that could violate lock ordering on umount
2575 if (!mapping
->nrpages
|| !mapping_tagged(mapping
, PAGECACHE_TAG_DIRTY
))
2579 * If the filesystem has aborted, it is read-only, so return
2580 * right away instead of dumping stack traces later on that
2581 * will obscure the real source of the problem. We test
2582 * EXT4_MOUNT_ABORT instead of sb->s_flag's MS_RDONLY because
2583 * the latter could be true if the filesystem is mounted
2584 * read-only, and in that case, ext4_da_writepages should
2585 * *never* be called, so if that ever happens, we would want
2588 if (unlikely(sbi
->s_mount_opt
& EXT4_MOUNT_ABORT
))
2592 * Make sure nr_to_write is >= sbi->s_mb_stream_request
2593 * This make sure small files blocks are allocated in
2594 * single attempt. This ensure that small files
2595 * get less fragmented.
2597 if (wbc
->nr_to_write
< sbi
->s_mb_stream_request
) {
2598 nr_to_writebump
= sbi
->s_mb_stream_request
- wbc
->nr_to_write
;
2599 wbc
->nr_to_write
= sbi
->s_mb_stream_request
;
2601 if (wbc
->range_start
== 0 && wbc
->range_end
== LLONG_MAX
)
2604 range_cyclic
= wbc
->range_cyclic
;
2605 if (wbc
->range_cyclic
) {
2606 index
= mapping
->writeback_index
;
2609 wbc
->range_start
= index
<< PAGE_CACHE_SHIFT
;
2610 wbc
->range_end
= LLONG_MAX
;
2611 wbc
->range_cyclic
= 0;
2613 index
= wbc
->range_start
>> PAGE_CACHE_SHIFT
;
2616 mpd
.inode
= mapping
->host
;
2619 * we don't want write_cache_pages to update
2620 * nr_to_write and writeback_index
2622 no_nrwrite_index_update
= wbc
->no_nrwrite_index_update
;
2623 wbc
->no_nrwrite_index_update
= 1;
2624 pages_skipped
= wbc
->pages_skipped
;
2627 while (!ret
&& wbc
->nr_to_write
> 0) {
2630 * we insert one extent at a time. So we need
2631 * credit needed for single extent allocation.
2632 * journalled mode is currently not supported
2635 BUG_ON(ext4_should_journal_data(inode
));
2636 needed_blocks
= ext4_da_writepages_trans_blocks(inode
);
2638 /* start a new transaction*/
2639 handle
= ext4_journal_start(inode
, needed_blocks
);
2640 if (IS_ERR(handle
)) {
2641 ret
= PTR_ERR(handle
);
2642 printk(KERN_CRIT
"%s: jbd2_start: "
2643 "%ld pages, ino %lu; err %d\n", __func__
,
2644 wbc
->nr_to_write
, inode
->i_ino
, ret
);
2646 goto out_writepages
;
2650 * Now call __mpage_da_writepage to find the next
2651 * contiguous region of logical blocks that need
2652 * blocks to be allocated by ext4. We don't actually
2653 * submit the blocks for I/O here, even though
2654 * write_cache_pages thinks it will, and will set the
2655 * pages as clean for write before calling
2656 * __mpage_da_writepage().
2664 mpd
.pages_written
= 0;
2666 ret
= write_cache_pages(mapping
, wbc
, __mpage_da_writepage
,
2669 * If we have a contigous extent of pages and we
2670 * haven't done the I/O yet, map the blocks and submit
2673 if (!mpd
.io_done
&& mpd
.next_page
!= mpd
.first_page
) {
2674 if (mpage_da_map_blocks(&mpd
) == 0)
2675 mpage_da_submit_io(&mpd
);
2677 ret
= MPAGE_DA_EXTENT_TAIL
;
2679 wbc
->nr_to_write
-= mpd
.pages_written
;
2681 ext4_journal_stop(handle
);
2683 if ((mpd
.retval
== -ENOSPC
) && sbi
->s_journal
) {
2684 /* commit the transaction which would
2685 * free blocks released in the transaction
2688 jbd2_journal_force_commit_nested(sbi
->s_journal
);
2689 wbc
->pages_skipped
= pages_skipped
;
2691 } else if (ret
== MPAGE_DA_EXTENT_TAIL
) {
2693 * got one extent now try with
2696 pages_written
+= mpd
.pages_written
;
2697 wbc
->pages_skipped
= pages_skipped
;
2700 } else if (wbc
->nr_to_write
)
2702 * There is no more writeout needed
2703 * or we requested for a noblocking writeout
2704 * and we found the device congested
2708 if (!io_done
&& !cycled
) {
2711 wbc
->range_start
= index
<< PAGE_CACHE_SHIFT
;
2712 wbc
->range_end
= mapping
->writeback_index
- 1;
2715 if (pages_skipped
!= wbc
->pages_skipped
)
2716 printk(KERN_EMERG
"This should not happen leaving %s "
2717 "with nr_to_write = %ld ret = %d\n",
2718 __func__
, wbc
->nr_to_write
, ret
);
2721 index
+= pages_written
;
2722 wbc
->range_cyclic
= range_cyclic
;
2723 if (wbc
->range_cyclic
|| (range_whole
&& wbc
->nr_to_write
> 0))
2725 * set the writeback_index so that range_cyclic
2726 * mode will write it back later
2728 mapping
->writeback_index
= index
;
2731 if (!no_nrwrite_index_update
)
2732 wbc
->no_nrwrite_index_update
= 0;
2733 wbc
->nr_to_write
-= nr_to_writebump
;
2734 trace_mark(ext4_da_writepage_result
,
2735 "dev %s ino %lu ret %d pages_written %d "
2736 "pages_skipped %ld congestion %d "
2737 "more_io %d no_nrwrite_index_update %d",
2738 inode
->i_sb
->s_id
, inode
->i_ino
, ret
,
2739 pages_written
, wbc
->pages_skipped
,
2740 wbc
->encountered_congestion
, wbc
->more_io
,
2741 wbc
->no_nrwrite_index_update
);
2745 #define FALL_BACK_TO_NONDELALLOC 1
2746 static int ext4_nonda_switch(struct super_block
*sb
)
2748 s64 free_blocks
, dirty_blocks
;
2749 struct ext4_sb_info
*sbi
= EXT4_SB(sb
);
2752 * switch to non delalloc mode if we are running low
2753 * on free block. The free block accounting via percpu
2754 * counters can get slightly wrong with percpu_counter_batch getting
2755 * accumulated on each CPU without updating global counters
2756 * Delalloc need an accurate free block accounting. So switch
2757 * to non delalloc when we are near to error range.
2759 free_blocks
= percpu_counter_read_positive(&sbi
->s_freeblocks_counter
);
2760 dirty_blocks
= percpu_counter_read_positive(&sbi
->s_dirtyblocks_counter
);
2761 if (2 * free_blocks
< 3 * dirty_blocks
||
2762 free_blocks
< (dirty_blocks
+ EXT4_FREEBLOCKS_WATERMARK
)) {
2764 * free block count is less that 150% of dirty blocks
2765 * or free blocks is less that watermark
2772 static int ext4_da_write_begin(struct file
*file
, struct address_space
*mapping
,
2773 loff_t pos
, unsigned len
, unsigned flags
,
2774 struct page
**pagep
, void **fsdata
)
2776 int ret
, retries
= 0;
2780 struct inode
*inode
= mapping
->host
;
2783 index
= pos
>> PAGE_CACHE_SHIFT
;
2784 from
= pos
& (PAGE_CACHE_SIZE
- 1);
2787 if (ext4_nonda_switch(inode
->i_sb
)) {
2788 *fsdata
= (void *)FALL_BACK_TO_NONDELALLOC
;
2789 return ext4_write_begin(file
, mapping
, pos
,
2790 len
, flags
, pagep
, fsdata
);
2792 *fsdata
= (void *)0;
2794 trace_mark(ext4_da_write_begin
,
2795 "dev %s ino %lu pos %llu len %u flags %u",
2796 inode
->i_sb
->s_id
, inode
->i_ino
,
2797 (unsigned long long) pos
, len
, flags
);
2800 * With delayed allocation, we don't log the i_disksize update
2801 * if there is delayed block allocation. But we still need
2802 * to journalling the i_disksize update if writes to the end
2803 * of file which has an already mapped buffer.
2805 handle
= ext4_journal_start(inode
, 1);
2806 if (IS_ERR(handle
)) {
2807 ret
= PTR_ERR(handle
);
2810 /* We cannot recurse into the filesystem as the transaction is already
2812 flags
|= AOP_FLAG_NOFS
;
2814 page
= grab_cache_page_write_begin(mapping
, index
, flags
);
2816 ext4_journal_stop(handle
);
2822 ret
= block_write_begin(file
, mapping
, pos
, len
, flags
, pagep
, fsdata
,
2823 ext4_da_get_block_prep
);
2826 ext4_journal_stop(handle
);
2827 page_cache_release(page
);
2829 * block_write_begin may have instantiated a few blocks
2830 * outside i_size. Trim these off again. Don't need
2831 * i_size_read because we hold i_mutex.
2833 if (pos
+ len
> inode
->i_size
)
2834 vmtruncate(inode
, inode
->i_size
);
2837 if (ret
== -ENOSPC
&& ext4_should_retry_alloc(inode
->i_sb
, &retries
))
2844 * Check if we should update i_disksize
2845 * when write to the end of file but not require block allocation
2847 static int ext4_da_should_update_i_disksize(struct page
*page
,
2848 unsigned long offset
)
2850 struct buffer_head
*bh
;
2851 struct inode
*inode
= page
->mapping
->host
;
2855 bh
= page_buffers(page
);
2856 idx
= offset
>> inode
->i_blkbits
;
2858 for (i
= 0; i
< idx
; i
++)
2859 bh
= bh
->b_this_page
;
2861 if (!buffer_mapped(bh
) || (buffer_delay(bh
)) || buffer_unwritten(bh
))
2866 static int ext4_da_write_end(struct file
*file
,
2867 struct address_space
*mapping
,
2868 loff_t pos
, unsigned len
, unsigned copied
,
2869 struct page
*page
, void *fsdata
)
2871 struct inode
*inode
= mapping
->host
;
2873 handle_t
*handle
= ext4_journal_current_handle();
2875 unsigned long start
, end
;
2876 int write_mode
= (int)(unsigned long)fsdata
;
2878 if (write_mode
== FALL_BACK_TO_NONDELALLOC
) {
2879 if (ext4_should_order_data(inode
)) {
2880 return ext4_ordered_write_end(file
, mapping
, pos
,
2881 len
, copied
, page
, fsdata
);
2882 } else if (ext4_should_writeback_data(inode
)) {
2883 return ext4_writeback_write_end(file
, mapping
, pos
,
2884 len
, copied
, page
, fsdata
);
2890 trace_mark(ext4_da_write_end
,
2891 "dev %s ino %lu pos %llu len %u copied %u",
2892 inode
->i_sb
->s_id
, inode
->i_ino
,
2893 (unsigned long long) pos
, len
, copied
);
2894 start
= pos
& (PAGE_CACHE_SIZE
- 1);
2895 end
= start
+ copied
- 1;
2898 * generic_write_end() will run mark_inode_dirty() if i_size
2899 * changes. So let's piggyback the i_disksize mark_inode_dirty
2903 new_i_size
= pos
+ copied
;
2904 if (new_i_size
> EXT4_I(inode
)->i_disksize
) {
2905 if (ext4_da_should_update_i_disksize(page
, end
)) {
2906 down_write(&EXT4_I(inode
)->i_data_sem
);
2907 if (new_i_size
> EXT4_I(inode
)->i_disksize
) {
2909 * Updating i_disksize when extending file
2910 * without needing block allocation
2912 if (ext4_should_order_data(inode
))
2913 ret
= ext4_jbd2_file_inode(handle
,
2916 EXT4_I(inode
)->i_disksize
= new_i_size
;
2918 up_write(&EXT4_I(inode
)->i_data_sem
);
2919 /* We need to mark inode dirty even if
2920 * new_i_size is less that inode->i_size
2921 * bu greater than i_disksize.(hint delalloc)
2923 ext4_mark_inode_dirty(handle
, inode
);
2926 ret2
= generic_write_end(file
, mapping
, pos
, len
, copied
,
2931 ret2
= ext4_journal_stop(handle
);
2935 return ret
? ret
: copied
;
2938 static void ext4_da_invalidatepage(struct page
*page
, unsigned long offset
)
2941 * Drop reserved blocks
2943 BUG_ON(!PageLocked(page
));
2944 if (!page_has_buffers(page
))
2947 ext4_da_page_release_reservation(page
, offset
);
2950 ext4_invalidatepage(page
, offset
);
2956 * Force all delayed allocation blocks to be allocated for a given inode.
2958 int ext4_alloc_da_blocks(struct inode
*inode
)
2960 if (!EXT4_I(inode
)->i_reserved_data_blocks
&&
2961 !EXT4_I(inode
)->i_reserved_meta_blocks
)
2965 * We do something simple for now. The filemap_flush() will
2966 * also start triggering a write of the data blocks, which is
2967 * not strictly speaking necessary (and for users of
2968 * laptop_mode, not even desirable). However, to do otherwise
2969 * would require replicating code paths in:
2971 * ext4_da_writepages() ->
2972 * write_cache_pages() ---> (via passed in callback function)
2973 * __mpage_da_writepage() -->
2974 * mpage_add_bh_to_extent()
2975 * mpage_da_map_blocks()
2977 * The problem is that write_cache_pages(), located in
2978 * mm/page-writeback.c, marks pages clean in preparation for
2979 * doing I/O, which is not desirable if we're not planning on
2982 * We could call write_cache_pages(), and then redirty all of
2983 * the pages by calling redirty_page_for_writeback() but that
2984 * would be ugly in the extreme. So instead we would need to
2985 * replicate parts of the code in the above functions,
2986 * simplifying them becuase we wouldn't actually intend to
2987 * write out the pages, but rather only collect contiguous
2988 * logical block extents, call the multi-block allocator, and
2989 * then update the buffer heads with the block allocations.
2991 * For now, though, we'll cheat by calling filemap_flush(),
2992 * which will map the blocks, and start the I/O, but not
2993 * actually wait for the I/O to complete.
2995 return filemap_flush(inode
->i_mapping
);
2999 * bmap() is special. It gets used by applications such as lilo and by
3000 * the swapper to find the on-disk block of a specific piece of data.
3002 * Naturally, this is dangerous if the block concerned is still in the
3003 * journal. If somebody makes a swapfile on an ext4 data-journaling
3004 * filesystem and enables swap, then they may get a nasty shock when the
3005 * data getting swapped to that swapfile suddenly gets overwritten by
3006 * the original zero's written out previously to the journal and
3007 * awaiting writeback in the kernel's buffer cache.
3009 * So, if we see any bmap calls here on a modified, data-journaled file,
3010 * take extra steps to flush any blocks which might be in the cache.
3012 static sector_t
ext4_bmap(struct address_space
*mapping
, sector_t block
)
3014 struct inode
*inode
= mapping
->host
;
3018 if (mapping_tagged(mapping
, PAGECACHE_TAG_DIRTY
) &&
3019 test_opt(inode
->i_sb
, DELALLOC
)) {
3021 * With delalloc we want to sync the file
3022 * so that we can make sure we allocate
3025 filemap_write_and_wait(mapping
);
3028 if (EXT4_JOURNAL(inode
) && EXT4_I(inode
)->i_state
& EXT4_STATE_JDATA
) {
3030 * This is a REALLY heavyweight approach, but the use of
3031 * bmap on dirty files is expected to be extremely rare:
3032 * only if we run lilo or swapon on a freshly made file
3033 * do we expect this to happen.
3035 * (bmap requires CAP_SYS_RAWIO so this does not
3036 * represent an unprivileged user DOS attack --- we'd be
3037 * in trouble if mortal users could trigger this path at
3040 * NB. EXT4_STATE_JDATA is not set on files other than
3041 * regular files. If somebody wants to bmap a directory
3042 * or symlink and gets confused because the buffer
3043 * hasn't yet been flushed to disk, they deserve
3044 * everything they get.
3047 EXT4_I(inode
)->i_state
&= ~EXT4_STATE_JDATA
;
3048 journal
= EXT4_JOURNAL(inode
);
3049 jbd2_journal_lock_updates(journal
);
3050 err
= jbd2_journal_flush(journal
);
3051 jbd2_journal_unlock_updates(journal
);
3057 return generic_block_bmap(mapping
, block
, ext4_get_block
);
3060 static int bget_one(handle_t
*handle
, struct buffer_head
*bh
)
3066 static int bput_one(handle_t
*handle
, struct buffer_head
*bh
)
3073 * Note that we don't need to start a transaction unless we're journaling data
3074 * because we should have holes filled from ext4_page_mkwrite(). We even don't
3075 * need to file the inode to the transaction's list in ordered mode because if
3076 * we are writing back data added by write(), the inode is already there and if
3077 * we are writing back data modified via mmap(), noone guarantees in which
3078 * transaction the data will hit the disk. In case we are journaling data, we
3079 * cannot start transaction directly because transaction start ranks above page
3080 * lock so we have to do some magic.
3082 * In all journaling modes block_write_full_page() will start the I/O.
3086 * ext4_writepage() -> kmalloc() -> __alloc_pages() -> page_launder() ->
3091 * ext4_file_write() -> generic_file_write() -> __alloc_pages() -> ...
3093 * Same applies to ext4_get_block(). We will deadlock on various things like
3094 * lock_journal and i_data_sem
3096 * Setting PF_MEMALLOC here doesn't work - too many internal memory
3099 * 16May01: If we're reentered then journal_current_handle() will be
3100 * non-zero. We simply *return*.
3102 * 1 July 2001: @@@ FIXME:
3103 * In journalled data mode, a data buffer may be metadata against the
3104 * current transaction. But the same file is part of a shared mapping
3105 * and someone does a writepage() on it.
3107 * We will move the buffer onto the async_data list, but *after* it has
3108 * been dirtied. So there's a small window where we have dirty data on
3111 * Note that this only applies to the last partial page in the file. The
3112 * bit which block_write_full_page() uses prepare/commit for. (That's
3113 * broken code anyway: it's wrong for msync()).
3115 * It's a rare case: affects the final partial page, for journalled data
3116 * where the file is subject to bith write() and writepage() in the same
3117 * transction. To fix it we'll need a custom block_write_full_page().
3118 * We'll probably need that anyway for journalling writepage() output.
3120 * We don't honour synchronous mounts for writepage(). That would be
3121 * disastrous. Any write() or metadata operation will sync the fs for
3125 static int __ext4_normal_writepage(struct page
*page
,
3126 struct writeback_control
*wbc
)
3128 struct inode
*inode
= page
->mapping
->host
;
3130 if (test_opt(inode
->i_sb
, NOBH
))
3131 return nobh_writepage(page
, noalloc_get_block_write
, wbc
);
3133 return block_write_full_page(page
, noalloc_get_block_write
,
3137 static int ext4_normal_writepage(struct page
*page
,
3138 struct writeback_control
*wbc
)
3140 struct inode
*inode
= page
->mapping
->host
;
3141 loff_t size
= i_size_read(inode
);
3144 trace_mark(ext4_normal_writepage
,
3145 "dev %s ino %lu page_index %lu",
3146 inode
->i_sb
->s_id
, inode
->i_ino
, page
->index
);
3147 J_ASSERT(PageLocked(page
));
3148 if (page
->index
== size
>> PAGE_CACHE_SHIFT
)
3149 len
= size
& ~PAGE_CACHE_MASK
;
3151 len
= PAGE_CACHE_SIZE
;
3153 if (page_has_buffers(page
)) {
3154 /* if page has buffers it should all be mapped
3155 * and allocated. If there are not buffers attached
3156 * to the page we know the page is dirty but it lost
3157 * buffers. That means that at some moment in time
3158 * after write_begin() / write_end() has been called
3159 * all buffers have been clean and thus they must have been
3160 * written at least once. So they are all mapped and we can
3161 * happily proceed with mapping them and writing the page.
3163 BUG_ON(walk_page_buffers(NULL
, page_buffers(page
), 0, len
, NULL
,
3164 ext4_bh_unmapped_or_delay
));
3167 if (!ext4_journal_current_handle())
3168 return __ext4_normal_writepage(page
, wbc
);
3170 redirty_page_for_writepage(wbc
, page
);
3175 static int __ext4_journalled_writepage(struct page
*page
,
3176 struct writeback_control
*wbc
)
3178 struct address_space
*mapping
= page
->mapping
;
3179 struct inode
*inode
= mapping
->host
;
3180 struct buffer_head
*page_bufs
;
3181 handle_t
*handle
= NULL
;
3185 ret
= block_prepare_write(page
, 0, PAGE_CACHE_SIZE
,
3186 noalloc_get_block_write
);
3190 page_bufs
= page_buffers(page
);
3191 walk_page_buffers(handle
, page_bufs
, 0, PAGE_CACHE_SIZE
, NULL
,
3193 /* As soon as we unlock the page, it can go away, but we have
3194 * references to buffers so we are safe */
3197 handle
= ext4_journal_start(inode
, ext4_writepage_trans_blocks(inode
));
3198 if (IS_ERR(handle
)) {
3199 ret
= PTR_ERR(handle
);
3203 ret
= walk_page_buffers(handle
, page_bufs
, 0,
3204 PAGE_CACHE_SIZE
, NULL
, do_journal_get_write_access
);
3206 err
= walk_page_buffers(handle
, page_bufs
, 0,
3207 PAGE_CACHE_SIZE
, NULL
, write_end_fn
);
3210 err
= ext4_journal_stop(handle
);
3214 walk_page_buffers(handle
, page_bufs
, 0,
3215 PAGE_CACHE_SIZE
, NULL
, bput_one
);
3216 EXT4_I(inode
)->i_state
|= EXT4_STATE_JDATA
;
3225 static int ext4_journalled_writepage(struct page
*page
,
3226 struct writeback_control
*wbc
)
3228 struct inode
*inode
= page
->mapping
->host
;
3229 loff_t size
= i_size_read(inode
);
3232 trace_mark(ext4_journalled_writepage
,
3233 "dev %s ino %lu page_index %lu",
3234 inode
->i_sb
->s_id
, inode
->i_ino
, page
->index
);
3235 J_ASSERT(PageLocked(page
));
3236 if (page
->index
== size
>> PAGE_CACHE_SHIFT
)
3237 len
= size
& ~PAGE_CACHE_MASK
;
3239 len
= PAGE_CACHE_SIZE
;
3241 if (page_has_buffers(page
)) {
3242 /* if page has buffers it should all be mapped
3243 * and allocated. If there are not buffers attached
3244 * to the page we know the page is dirty but it lost
3245 * buffers. That means that at some moment in time
3246 * after write_begin() / write_end() has been called
3247 * all buffers have been clean and thus they must have been
3248 * written at least once. So they are all mapped and we can
3249 * happily proceed with mapping them and writing the page.
3251 BUG_ON(walk_page_buffers(NULL
, page_buffers(page
), 0, len
, NULL
,
3252 ext4_bh_unmapped_or_delay
));
3255 if (ext4_journal_current_handle())
3258 if (PageChecked(page
)) {
3260 * It's mmapped pagecache. Add buffers and journal it. There
3261 * doesn't seem much point in redirtying the page here.
3263 ClearPageChecked(page
);
3264 return __ext4_journalled_writepage(page
, wbc
);
3267 * It may be a page full of checkpoint-mode buffers. We don't
3268 * really know unless we go poke around in the buffer_heads.
3269 * But block_write_full_page will do the right thing.
3271 return block_write_full_page(page
, noalloc_get_block_write
,
3275 redirty_page_for_writepage(wbc
, page
);
3280 static int ext4_readpage(struct file
*file
, struct page
*page
)
3282 return mpage_readpage(page
, ext4_get_block
);
3286 ext4_readpages(struct file
*file
, struct address_space
*mapping
,
3287 struct list_head
*pages
, unsigned nr_pages
)
3289 return mpage_readpages(mapping
, pages
, nr_pages
, ext4_get_block
);
3292 static void ext4_invalidatepage(struct page
*page
, unsigned long offset
)
3294 journal_t
*journal
= EXT4_JOURNAL(page
->mapping
->host
);
3297 * If it's a full truncate we just forget about the pending dirtying
3300 ClearPageChecked(page
);
3303 jbd2_journal_invalidatepage(journal
, page
, offset
);
3305 block_invalidatepage(page
, offset
);
3308 static int ext4_releasepage(struct page
*page
, gfp_t wait
)
3310 journal_t
*journal
= EXT4_JOURNAL(page
->mapping
->host
);
3312 WARN_ON(PageChecked(page
));
3313 if (!page_has_buffers(page
))
3316 return jbd2_journal_try_to_free_buffers(journal
, page
, wait
);
3318 return try_to_free_buffers(page
);
3322 * If the O_DIRECT write will extend the file then add this inode to the
3323 * orphan list. So recovery will truncate it back to the original size
3324 * if the machine crashes during the write.
3326 * If the O_DIRECT write is intantiating holes inside i_size and the machine
3327 * crashes then stale disk data _may_ be exposed inside the file. But current
3328 * VFS code falls back into buffered path in that case so we are safe.
3330 static ssize_t
ext4_direct_IO(int rw
, struct kiocb
*iocb
,
3331 const struct iovec
*iov
, loff_t offset
,
3332 unsigned long nr_segs
)
3334 struct file
*file
= iocb
->ki_filp
;
3335 struct inode
*inode
= file
->f_mapping
->host
;
3336 struct ext4_inode_info
*ei
= EXT4_I(inode
);
3340 size_t count
= iov_length(iov
, nr_segs
);
3343 loff_t final_size
= offset
+ count
;
3345 if (final_size
> inode
->i_size
) {
3346 /* Credits for sb + inode write */
3347 handle
= ext4_journal_start(inode
, 2);
3348 if (IS_ERR(handle
)) {
3349 ret
= PTR_ERR(handle
);
3352 ret
= ext4_orphan_add(handle
, inode
);
3354 ext4_journal_stop(handle
);
3358 ei
->i_disksize
= inode
->i_size
;
3359 ext4_journal_stop(handle
);
3363 ret
= blockdev_direct_IO(rw
, iocb
, inode
, inode
->i_sb
->s_bdev
, iov
,
3365 ext4_get_block
, NULL
);
3370 /* Credits for sb + inode write */
3371 handle
= ext4_journal_start(inode
, 2);
3372 if (IS_ERR(handle
)) {
3373 /* This is really bad luck. We've written the data
3374 * but cannot extend i_size. Bail out and pretend
3375 * the write failed... */
3376 ret
= PTR_ERR(handle
);
3380 ext4_orphan_del(handle
, inode
);
3382 loff_t end
= offset
+ ret
;
3383 if (end
> inode
->i_size
) {
3384 ei
->i_disksize
= end
;
3385 i_size_write(inode
, end
);
3387 * We're going to return a positive `ret'
3388 * here due to non-zero-length I/O, so there's
3389 * no way of reporting error returns from
3390 * ext4_mark_inode_dirty() to userspace. So
3393 ext4_mark_inode_dirty(handle
, inode
);
3396 err
= ext4_journal_stop(handle
);
3405 * Pages can be marked dirty completely asynchronously from ext4's journalling
3406 * activity. By filemap_sync_pte(), try_to_unmap_one(), etc. We cannot do
3407 * much here because ->set_page_dirty is called under VFS locks. The page is
3408 * not necessarily locked.
3410 * We cannot just dirty the page and leave attached buffers clean, because the
3411 * buffers' dirty state is "definitive". We cannot just set the buffers dirty
3412 * or jbddirty because all the journalling code will explode.
3414 * So what we do is to mark the page "pending dirty" and next time writepage
3415 * is called, propagate that into the buffers appropriately.
3417 static int ext4_journalled_set_page_dirty(struct page
*page
)
3419 SetPageChecked(page
);
3420 return __set_page_dirty_nobuffers(page
);
3423 static const struct address_space_operations ext4_ordered_aops
= {
3424 .readpage
= ext4_readpage
,
3425 .readpages
= ext4_readpages
,
3426 .writepage
= ext4_normal_writepage
,
3427 .sync_page
= block_sync_page
,
3428 .write_begin
= ext4_write_begin
,
3429 .write_end
= ext4_ordered_write_end
,
3431 .invalidatepage
= ext4_invalidatepage
,
3432 .releasepage
= ext4_releasepage
,
3433 .direct_IO
= ext4_direct_IO
,
3434 .migratepage
= buffer_migrate_page
,
3435 .is_partially_uptodate
= block_is_partially_uptodate
,
3438 static const struct address_space_operations ext4_writeback_aops
= {
3439 .readpage
= ext4_readpage
,
3440 .readpages
= ext4_readpages
,
3441 .writepage
= ext4_normal_writepage
,
3442 .sync_page
= block_sync_page
,
3443 .write_begin
= ext4_write_begin
,
3444 .write_end
= ext4_writeback_write_end
,
3446 .invalidatepage
= ext4_invalidatepage
,
3447 .releasepage
= ext4_releasepage
,
3448 .direct_IO
= ext4_direct_IO
,
3449 .migratepage
= buffer_migrate_page
,
3450 .is_partially_uptodate
= block_is_partially_uptodate
,
3453 static const struct address_space_operations ext4_journalled_aops
= {
3454 .readpage
= ext4_readpage
,
3455 .readpages
= ext4_readpages
,
3456 .writepage
= ext4_journalled_writepage
,
3457 .sync_page
= block_sync_page
,
3458 .write_begin
= ext4_write_begin
,
3459 .write_end
= ext4_journalled_write_end
,
3460 .set_page_dirty
= ext4_journalled_set_page_dirty
,
3462 .invalidatepage
= ext4_invalidatepage
,
3463 .releasepage
= ext4_releasepage
,
3464 .is_partially_uptodate
= block_is_partially_uptodate
,
3467 static const struct address_space_operations ext4_da_aops
= {
3468 .readpage
= ext4_readpage
,
3469 .readpages
= ext4_readpages
,
3470 .writepage
= ext4_da_writepage
,
3471 .writepages
= ext4_da_writepages
,
3472 .sync_page
= block_sync_page
,
3473 .write_begin
= ext4_da_write_begin
,
3474 .write_end
= ext4_da_write_end
,
3476 .invalidatepage
= ext4_da_invalidatepage
,
3477 .releasepage
= ext4_releasepage
,
3478 .direct_IO
= ext4_direct_IO
,
3479 .migratepage
= buffer_migrate_page
,
3480 .is_partially_uptodate
= block_is_partially_uptodate
,
3483 void ext4_set_aops(struct inode
*inode
)
3485 if (ext4_should_order_data(inode
) &&
3486 test_opt(inode
->i_sb
, DELALLOC
))
3487 inode
->i_mapping
->a_ops
= &ext4_da_aops
;
3488 else if (ext4_should_order_data(inode
))
3489 inode
->i_mapping
->a_ops
= &ext4_ordered_aops
;
3490 else if (ext4_should_writeback_data(inode
) &&
3491 test_opt(inode
->i_sb
, DELALLOC
))
3492 inode
->i_mapping
->a_ops
= &ext4_da_aops
;
3493 else if (ext4_should_writeback_data(inode
))
3494 inode
->i_mapping
->a_ops
= &ext4_writeback_aops
;
3496 inode
->i_mapping
->a_ops
= &ext4_journalled_aops
;
3500 * ext4_block_truncate_page() zeroes out a mapping from file offset `from'
3501 * up to the end of the block which corresponds to `from'.
3502 * This required during truncate. We need to physically zero the tail end
3503 * of that block so it doesn't yield old data if the file is later grown.
3505 int ext4_block_truncate_page(handle_t
*handle
,
3506 struct address_space
*mapping
, loff_t from
)
3508 ext4_fsblk_t index
= from
>> PAGE_CACHE_SHIFT
;
3509 unsigned offset
= from
& (PAGE_CACHE_SIZE
-1);
3510 unsigned blocksize
, length
, pos
;
3512 struct inode
*inode
= mapping
->host
;
3513 struct buffer_head
*bh
;
3517 page
= grab_cache_page(mapping
, from
>> PAGE_CACHE_SHIFT
);
3521 blocksize
= inode
->i_sb
->s_blocksize
;
3522 length
= blocksize
- (offset
& (blocksize
- 1));
3523 iblock
= index
<< (PAGE_CACHE_SHIFT
- inode
->i_sb
->s_blocksize_bits
);
3526 * For "nobh" option, we can only work if we don't need to
3527 * read-in the page - otherwise we create buffers to do the IO.
3529 if (!page_has_buffers(page
) && test_opt(inode
->i_sb
, NOBH
) &&
3530 ext4_should_writeback_data(inode
) && PageUptodate(page
)) {
3531 zero_user(page
, offset
, length
);
3532 set_page_dirty(page
);
3536 if (!page_has_buffers(page
))
3537 create_empty_buffers(page
, blocksize
, 0);
3539 /* Find the buffer that contains "offset" */
3540 bh
= page_buffers(page
);
3542 while (offset
>= pos
) {
3543 bh
= bh
->b_this_page
;
3549 if (buffer_freed(bh
)) {
3550 BUFFER_TRACE(bh
, "freed: skip");
3554 if (!buffer_mapped(bh
)) {
3555 BUFFER_TRACE(bh
, "unmapped");
3556 ext4_get_block(inode
, iblock
, bh
, 0);
3557 /* unmapped? It's a hole - nothing to do */
3558 if (!buffer_mapped(bh
)) {
3559 BUFFER_TRACE(bh
, "still unmapped");
3564 /* Ok, it's mapped. Make sure it's up-to-date */
3565 if (PageUptodate(page
))
3566 set_buffer_uptodate(bh
);
3568 if (!buffer_uptodate(bh
)) {
3570 ll_rw_block(READ
, 1, &bh
);
3572 /* Uhhuh. Read error. Complain and punt. */
3573 if (!buffer_uptodate(bh
))
3577 if (ext4_should_journal_data(inode
)) {
3578 BUFFER_TRACE(bh
, "get write access");
3579 err
= ext4_journal_get_write_access(handle
, bh
);
3584 zero_user(page
, offset
, length
);
3586 BUFFER_TRACE(bh
, "zeroed end of block");
3589 if (ext4_should_journal_data(inode
)) {
3590 err
= ext4_handle_dirty_metadata(handle
, inode
, bh
);
3592 if (ext4_should_order_data(inode
))
3593 err
= ext4_jbd2_file_inode(handle
, inode
);
3594 mark_buffer_dirty(bh
);
3599 page_cache_release(page
);
3604 * Probably it should be a library function... search for first non-zero word
3605 * or memcmp with zero_page, whatever is better for particular architecture.
3608 static inline int all_zeroes(__le32
*p
, __le32
*q
)
3617 * ext4_find_shared - find the indirect blocks for partial truncation.
3618 * @inode: inode in question
3619 * @depth: depth of the affected branch
3620 * @offsets: offsets of pointers in that branch (see ext4_block_to_path)
3621 * @chain: place to store the pointers to partial indirect blocks
3622 * @top: place to the (detached) top of branch
3624 * This is a helper function used by ext4_truncate().
3626 * When we do truncate() we may have to clean the ends of several
3627 * indirect blocks but leave the blocks themselves alive. Block is
3628 * partially truncated if some data below the new i_size is refered
3629 * from it (and it is on the path to the first completely truncated
3630 * data block, indeed). We have to free the top of that path along
3631 * with everything to the right of the path. Since no allocation
3632 * past the truncation point is possible until ext4_truncate()
3633 * finishes, we may safely do the latter, but top of branch may
3634 * require special attention - pageout below the truncation point
3635 * might try to populate it.
3637 * We atomically detach the top of branch from the tree, store the
3638 * block number of its root in *@top, pointers to buffer_heads of
3639 * partially truncated blocks - in @chain[].bh and pointers to
3640 * their last elements that should not be removed - in
3641 * @chain[].p. Return value is the pointer to last filled element
3644 * The work left to caller to do the actual freeing of subtrees:
3645 * a) free the subtree starting from *@top
3646 * b) free the subtrees whose roots are stored in
3647 * (@chain[i].p+1 .. end of @chain[i].bh->b_data)
3648 * c) free the subtrees growing from the inode past the @chain[0].
3649 * (no partially truncated stuff there). */
3651 static Indirect
*ext4_find_shared(struct inode
*inode
, int depth
,
3652 ext4_lblk_t offsets
[4], Indirect chain
[4], __le32
*top
)
3654 Indirect
*partial
, *p
;
3658 /* Make k index the deepest non-null offest + 1 */
3659 for (k
= depth
; k
> 1 && !offsets
[k
-1]; k
--)
3661 partial
= ext4_get_branch(inode
, k
, offsets
, chain
, &err
);
3662 /* Writer: pointers */
3664 partial
= chain
+ k
-1;
3666 * If the branch acquired continuation since we've looked at it -
3667 * fine, it should all survive and (new) top doesn't belong to us.
3669 if (!partial
->key
&& *partial
->p
)
3672 for (p
= partial
; (p
> chain
) && all_zeroes((__le32
*) p
->bh
->b_data
, p
->p
); p
--)
3675 * OK, we've found the last block that must survive. The rest of our
3676 * branch should be detached before unlocking. However, if that rest
3677 * of branch is all ours and does not grow immediately from the inode
3678 * it's easier to cheat and just decrement partial->p.
3680 if (p
== chain
+ k
- 1 && p
> chain
) {
3684 /* Nope, don't do this in ext4. Must leave the tree intact */
3691 while (partial
> p
) {
3692 brelse(partial
->bh
);
3700 * Zero a number of block pointers in either an inode or an indirect block.
3701 * If we restart the transaction we must again get write access to the
3702 * indirect block for further modification.
3704 * We release `count' blocks on disk, but (last - first) may be greater
3705 * than `count' because there can be holes in there.
3707 static void ext4_clear_blocks(handle_t
*handle
, struct inode
*inode
,
3708 struct buffer_head
*bh
, ext4_fsblk_t block_to_free
,
3709 unsigned long count
, __le32
*first
, __le32
*last
)
3712 if (try_to_extend_transaction(handle
, inode
)) {
3714 BUFFER_TRACE(bh
, "call ext4_handle_dirty_metadata");
3715 ext4_handle_dirty_metadata(handle
, inode
, bh
);
3717 ext4_mark_inode_dirty(handle
, inode
);
3718 ext4_journal_test_restart(handle
, inode
);
3720 BUFFER_TRACE(bh
, "retaking write access");
3721 ext4_journal_get_write_access(handle
, bh
);
3726 * Any buffers which are on the journal will be in memory. We find
3727 * them on the hash table so jbd2_journal_revoke() will run jbd2_journal_forget()
3728 * on them. We've already detached each block from the file, so
3729 * bforget() in jbd2_journal_forget() should be safe.
3731 * AKPM: turn on bforget in jbd2_journal_forget()!!!
3733 for (p
= first
; p
< last
; p
++) {
3734 u32 nr
= le32_to_cpu(*p
);
3736 struct buffer_head
*tbh
;
3739 tbh
= sb_find_get_block(inode
->i_sb
, nr
);
3740 ext4_forget(handle
, 0, inode
, tbh
, nr
);
3744 ext4_free_blocks(handle
, inode
, block_to_free
, count
, 0);
3748 * ext4_free_data - free a list of data blocks
3749 * @handle: handle for this transaction
3750 * @inode: inode we are dealing with
3751 * @this_bh: indirect buffer_head which contains *@first and *@last
3752 * @first: array of block numbers
3753 * @last: points immediately past the end of array
3755 * We are freeing all blocks refered from that array (numbers are stored as
3756 * little-endian 32-bit) and updating @inode->i_blocks appropriately.
3758 * We accumulate contiguous runs of blocks to free. Conveniently, if these
3759 * blocks are contiguous then releasing them at one time will only affect one
3760 * or two bitmap blocks (+ group descriptor(s) and superblock) and we won't
3761 * actually use a lot of journal space.
3763 * @this_bh will be %NULL if @first and @last point into the inode's direct
3766 static void ext4_free_data(handle_t
*handle
, struct inode
*inode
,
3767 struct buffer_head
*this_bh
,
3768 __le32
*first
, __le32
*last
)
3770 ext4_fsblk_t block_to_free
= 0; /* Starting block # of a run */
3771 unsigned long count
= 0; /* Number of blocks in the run */
3772 __le32
*block_to_free_p
= NULL
; /* Pointer into inode/ind
3775 ext4_fsblk_t nr
; /* Current block # */
3776 __le32
*p
; /* Pointer into inode/ind
3777 for current block */
3780 if (this_bh
) { /* For indirect block */
3781 BUFFER_TRACE(this_bh
, "get_write_access");
3782 err
= ext4_journal_get_write_access(handle
, this_bh
);
3783 /* Important: if we can't update the indirect pointers
3784 * to the blocks, we can't free them. */
3789 for (p
= first
; p
< last
; p
++) {
3790 nr
= le32_to_cpu(*p
);
3792 /* accumulate blocks to free if they're contiguous */
3795 block_to_free_p
= p
;
3797 } else if (nr
== block_to_free
+ count
) {
3800 ext4_clear_blocks(handle
, inode
, this_bh
,
3802 count
, block_to_free_p
, p
);
3804 block_to_free_p
= p
;
3811 ext4_clear_blocks(handle
, inode
, this_bh
, block_to_free
,
3812 count
, block_to_free_p
, p
);
3815 BUFFER_TRACE(this_bh
, "call ext4_handle_dirty_metadata");
3818 * The buffer head should have an attached journal head at this
3819 * point. However, if the data is corrupted and an indirect
3820 * block pointed to itself, it would have been detached when
3821 * the block was cleared. Check for this instead of OOPSing.
3823 if ((EXT4_JOURNAL(inode
) == NULL
) || bh2jh(this_bh
))
3824 ext4_handle_dirty_metadata(handle
, inode
, this_bh
);
3826 ext4_error(inode
->i_sb
, __func__
,
3827 "circular indirect block detected, "
3828 "inode=%lu, block=%llu",
3830 (unsigned long long) this_bh
->b_blocknr
);
3835 * ext4_free_branches - free an array of branches
3836 * @handle: JBD handle for this transaction
3837 * @inode: inode we are dealing with
3838 * @parent_bh: the buffer_head which contains *@first and *@last
3839 * @first: array of block numbers
3840 * @last: pointer immediately past the end of array
3841 * @depth: depth of the branches to free
3843 * We are freeing all blocks refered from these branches (numbers are
3844 * stored as little-endian 32-bit) and updating @inode->i_blocks
3847 static void ext4_free_branches(handle_t
*handle
, struct inode
*inode
,
3848 struct buffer_head
*parent_bh
,
3849 __le32
*first
, __le32
*last
, int depth
)
3854 if (ext4_handle_is_aborted(handle
))
3858 struct buffer_head
*bh
;
3859 int addr_per_block
= EXT4_ADDR_PER_BLOCK(inode
->i_sb
);
3861 while (--p
>= first
) {
3862 nr
= le32_to_cpu(*p
);
3864 continue; /* A hole */
3866 /* Go read the buffer for the next level down */
3867 bh
= sb_bread(inode
->i_sb
, nr
);
3870 * A read failure? Report error and clear slot
3874 ext4_error(inode
->i_sb
, "ext4_free_branches",
3875 "Read failure, inode=%lu, block=%llu",
3880 /* This zaps the entire block. Bottom up. */
3881 BUFFER_TRACE(bh
, "free child branches");
3882 ext4_free_branches(handle
, inode
, bh
,
3883 (__le32
*) bh
->b_data
,
3884 (__le32
*) bh
->b_data
+ addr_per_block
,
3888 * We've probably journalled the indirect block several
3889 * times during the truncate. But it's no longer
3890 * needed and we now drop it from the transaction via
3891 * jbd2_journal_revoke().
3893 * That's easy if it's exclusively part of this
3894 * transaction. But if it's part of the committing
3895 * transaction then jbd2_journal_forget() will simply
3896 * brelse() it. That means that if the underlying
3897 * block is reallocated in ext4_get_block(),
3898 * unmap_underlying_metadata() will find this block
3899 * and will try to get rid of it. damn, damn.
3901 * If this block has already been committed to the
3902 * journal, a revoke record will be written. And
3903 * revoke records must be emitted *before* clearing
3904 * this block's bit in the bitmaps.
3906 ext4_forget(handle
, 1, inode
, bh
, bh
->b_blocknr
);
3909 * Everything below this this pointer has been
3910 * released. Now let this top-of-subtree go.
3912 * We want the freeing of this indirect block to be
3913 * atomic in the journal with the updating of the
3914 * bitmap block which owns it. So make some room in
3917 * We zero the parent pointer *after* freeing its
3918 * pointee in the bitmaps, so if extend_transaction()
3919 * for some reason fails to put the bitmap changes and
3920 * the release into the same transaction, recovery
3921 * will merely complain about releasing a free block,
3922 * rather than leaking blocks.
3924 if (ext4_handle_is_aborted(handle
))
3926 if (try_to_extend_transaction(handle
, inode
)) {
3927 ext4_mark_inode_dirty(handle
, inode
);
3928 ext4_journal_test_restart(handle
, inode
);
3931 ext4_free_blocks(handle
, inode
, nr
, 1, 1);
3935 * The block which we have just freed is
3936 * pointed to by an indirect block: journal it
3938 BUFFER_TRACE(parent_bh
, "get_write_access");
3939 if (!ext4_journal_get_write_access(handle
,
3942 BUFFER_TRACE(parent_bh
,
3943 "call ext4_handle_dirty_metadata");
3944 ext4_handle_dirty_metadata(handle
,
3951 /* We have reached the bottom of the tree. */
3952 BUFFER_TRACE(parent_bh
, "free data blocks");
3953 ext4_free_data(handle
, inode
, parent_bh
, first
, last
);
3957 int ext4_can_truncate(struct inode
*inode
)
3959 if (IS_APPEND(inode
) || IS_IMMUTABLE(inode
))
3961 if (S_ISREG(inode
->i_mode
))
3963 if (S_ISDIR(inode
->i_mode
))
3965 if (S_ISLNK(inode
->i_mode
))
3966 return !ext4_inode_is_fast_symlink(inode
);
3973 * We block out ext4_get_block() block instantiations across the entire
3974 * transaction, and VFS/VM ensures that ext4_truncate() cannot run
3975 * simultaneously on behalf of the same inode.
3977 * As we work through the truncate and commmit bits of it to the journal there
3978 * is one core, guiding principle: the file's tree must always be consistent on
3979 * disk. We must be able to restart the truncate after a crash.
3981 * The file's tree may be transiently inconsistent in memory (although it
3982 * probably isn't), but whenever we close off and commit a journal transaction,
3983 * the contents of (the filesystem + the journal) must be consistent and
3984 * restartable. It's pretty simple, really: bottom up, right to left (although
3985 * left-to-right works OK too).
3987 * Note that at recovery time, journal replay occurs *before* the restart of
3988 * truncate against the orphan inode list.
3990 * The committed inode has the new, desired i_size (which is the same as
3991 * i_disksize in this case). After a crash, ext4_orphan_cleanup() will see
3992 * that this inode's truncate did not complete and it will again call
3993 * ext4_truncate() to have another go. So there will be instantiated blocks
3994 * to the right of the truncation point in a crashed ext4 filesystem. But
3995 * that's fine - as long as they are linked from the inode, the post-crash
3996 * ext4_truncate() run will find them and release them.
3998 void ext4_truncate(struct inode
*inode
)
4001 struct ext4_inode_info
*ei
= EXT4_I(inode
);
4002 __le32
*i_data
= ei
->i_data
;
4003 int addr_per_block
= EXT4_ADDR_PER_BLOCK(inode
->i_sb
);
4004 struct address_space
*mapping
= inode
->i_mapping
;
4005 ext4_lblk_t offsets
[4];
4010 ext4_lblk_t last_block
;
4011 unsigned blocksize
= inode
->i_sb
->s_blocksize
;
4013 if (!ext4_can_truncate(inode
))
4016 if (inode
->i_size
== 0 && !test_opt(inode
->i_sb
, NO_AUTO_DA_ALLOC
))
4017 ei
->i_state
|= EXT4_STATE_DA_ALLOC_CLOSE
;
4019 if (EXT4_I(inode
)->i_flags
& EXT4_EXTENTS_FL
) {
4020 ext4_ext_truncate(inode
);
4024 handle
= start_transaction(inode
);
4026 return; /* AKPM: return what? */
4028 last_block
= (inode
->i_size
+ blocksize
-1)
4029 >> EXT4_BLOCK_SIZE_BITS(inode
->i_sb
);
4031 if (inode
->i_size
& (blocksize
- 1))
4032 if (ext4_block_truncate_page(handle
, mapping
, inode
->i_size
))
4035 n
= ext4_block_to_path(inode
, last_block
, offsets
, NULL
);
4037 goto out_stop
; /* error */
4040 * OK. This truncate is going to happen. We add the inode to the
4041 * orphan list, so that if this truncate spans multiple transactions,
4042 * and we crash, we will resume the truncate when the filesystem
4043 * recovers. It also marks the inode dirty, to catch the new size.
4045 * Implication: the file must always be in a sane, consistent
4046 * truncatable state while each transaction commits.
4048 if (ext4_orphan_add(handle
, inode
))
4052 * From here we block out all ext4_get_block() callers who want to
4053 * modify the block allocation tree.
4055 down_write(&ei
->i_data_sem
);
4057 ext4_discard_preallocations(inode
);
4060 * The orphan list entry will now protect us from any crash which
4061 * occurs before the truncate completes, so it is now safe to propagate
4062 * the new, shorter inode size (held for now in i_size) into the
4063 * on-disk inode. We do this via i_disksize, which is the value which
4064 * ext4 *really* writes onto the disk inode.
4066 ei
->i_disksize
= inode
->i_size
;
4068 if (n
== 1) { /* direct blocks */
4069 ext4_free_data(handle
, inode
, NULL
, i_data
+offsets
[0],
4070 i_data
+ EXT4_NDIR_BLOCKS
);
4074 partial
= ext4_find_shared(inode
, n
, offsets
, chain
, &nr
);
4075 /* Kill the top of shared branch (not detached) */
4077 if (partial
== chain
) {
4078 /* Shared branch grows from the inode */
4079 ext4_free_branches(handle
, inode
, NULL
,
4080 &nr
, &nr
+1, (chain
+n
-1) - partial
);
4083 * We mark the inode dirty prior to restart,
4084 * and prior to stop. No need for it here.
4087 /* Shared branch grows from an indirect block */
4088 BUFFER_TRACE(partial
->bh
, "get_write_access");
4089 ext4_free_branches(handle
, inode
, partial
->bh
,
4091 partial
->p
+1, (chain
+n
-1) - partial
);
4094 /* Clear the ends of indirect blocks on the shared branch */
4095 while (partial
> chain
) {
4096 ext4_free_branches(handle
, inode
, partial
->bh
, partial
->p
+ 1,
4097 (__le32
*)partial
->bh
->b_data
+addr_per_block
,
4098 (chain
+n
-1) - partial
);
4099 BUFFER_TRACE(partial
->bh
, "call brelse");
4100 brelse (partial
->bh
);
4104 /* Kill the remaining (whole) subtrees */
4105 switch (offsets
[0]) {
4107 nr
= i_data
[EXT4_IND_BLOCK
];
4109 ext4_free_branches(handle
, inode
, NULL
, &nr
, &nr
+1, 1);
4110 i_data
[EXT4_IND_BLOCK
] = 0;
4112 case EXT4_IND_BLOCK
:
4113 nr
= i_data
[EXT4_DIND_BLOCK
];
4115 ext4_free_branches(handle
, inode
, NULL
, &nr
, &nr
+1, 2);
4116 i_data
[EXT4_DIND_BLOCK
] = 0;
4118 case EXT4_DIND_BLOCK
:
4119 nr
= i_data
[EXT4_TIND_BLOCK
];
4121 ext4_free_branches(handle
, inode
, NULL
, &nr
, &nr
+1, 3);
4122 i_data
[EXT4_TIND_BLOCK
] = 0;
4124 case EXT4_TIND_BLOCK
:
4128 up_write(&ei
->i_data_sem
);
4129 inode
->i_mtime
= inode
->i_ctime
= ext4_current_time(inode
);
4130 ext4_mark_inode_dirty(handle
, inode
);
4133 * In a multi-transaction truncate, we only make the final transaction
4137 ext4_handle_sync(handle
);
4140 * If this was a simple ftruncate(), and the file will remain alive
4141 * then we need to clear up the orphan record which we created above.
4142 * However, if this was a real unlink then we were called by
4143 * ext4_delete_inode(), and we allow that function to clean up the
4144 * orphan info for us.
4147 ext4_orphan_del(handle
, inode
);
4149 ext4_journal_stop(handle
);
4153 * ext4_get_inode_loc returns with an extra refcount against the inode's
4154 * underlying buffer_head on success. If 'in_mem' is true, we have all
4155 * data in memory that is needed to recreate the on-disk version of this
4158 static int __ext4_get_inode_loc(struct inode
*inode
,
4159 struct ext4_iloc
*iloc
, int in_mem
)
4161 struct ext4_group_desc
*gdp
;
4162 struct buffer_head
*bh
;
4163 struct super_block
*sb
= inode
->i_sb
;
4165 int inodes_per_block
, inode_offset
;
4168 if (!ext4_valid_inum(sb
, inode
->i_ino
))
4171 iloc
->block_group
= (inode
->i_ino
- 1) / EXT4_INODES_PER_GROUP(sb
);
4172 gdp
= ext4_get_group_desc(sb
, iloc
->block_group
, NULL
);
4177 * Figure out the offset within the block group inode table
4179 inodes_per_block
= (EXT4_BLOCK_SIZE(sb
) / EXT4_INODE_SIZE(sb
));
4180 inode_offset
= ((inode
->i_ino
- 1) %
4181 EXT4_INODES_PER_GROUP(sb
));
4182 block
= ext4_inode_table(sb
, gdp
) + (inode_offset
/ inodes_per_block
);
4183 iloc
->offset
= (inode_offset
% inodes_per_block
) * EXT4_INODE_SIZE(sb
);
4185 bh
= sb_getblk(sb
, block
);
4187 ext4_error(sb
, "ext4_get_inode_loc", "unable to read "
4188 "inode block - inode=%lu, block=%llu",
4189 inode
->i_ino
, block
);
4192 if (!buffer_uptodate(bh
)) {
4196 * If the buffer has the write error flag, we have failed
4197 * to write out another inode in the same block. In this
4198 * case, we don't have to read the block because we may
4199 * read the old inode data successfully.
4201 if (buffer_write_io_error(bh
) && !buffer_uptodate(bh
))
4202 set_buffer_uptodate(bh
);
4204 if (buffer_uptodate(bh
)) {
4205 /* someone brought it uptodate while we waited */
4211 * If we have all information of the inode in memory and this
4212 * is the only valid inode in the block, we need not read the
4216 struct buffer_head
*bitmap_bh
;
4219 start
= inode_offset
& ~(inodes_per_block
- 1);
4221 /* Is the inode bitmap in cache? */
4222 bitmap_bh
= sb_getblk(sb
, ext4_inode_bitmap(sb
, gdp
));
4227 * If the inode bitmap isn't in cache then the
4228 * optimisation may end up performing two reads instead
4229 * of one, so skip it.
4231 if (!buffer_uptodate(bitmap_bh
)) {
4235 for (i
= start
; i
< start
+ inodes_per_block
; i
++) {
4236 if (i
== inode_offset
)
4238 if (ext4_test_bit(i
, bitmap_bh
->b_data
))
4242 if (i
== start
+ inodes_per_block
) {
4243 /* all other inodes are free, so skip I/O */
4244 memset(bh
->b_data
, 0, bh
->b_size
);
4245 set_buffer_uptodate(bh
);
4253 * If we need to do any I/O, try to pre-readahead extra
4254 * blocks from the inode table.
4256 if (EXT4_SB(sb
)->s_inode_readahead_blks
) {
4257 ext4_fsblk_t b
, end
, table
;
4260 table
= ext4_inode_table(sb
, gdp
);
4261 /* s_inode_readahead_blks is always a power of 2 */
4262 b
= block
& ~(EXT4_SB(sb
)->s_inode_readahead_blks
-1);
4265 end
= b
+ EXT4_SB(sb
)->s_inode_readahead_blks
;
4266 num
= EXT4_INODES_PER_GROUP(sb
);
4267 if (EXT4_HAS_RO_COMPAT_FEATURE(sb
,
4268 EXT4_FEATURE_RO_COMPAT_GDT_CSUM
))
4269 num
-= ext4_itable_unused_count(sb
, gdp
);
4270 table
+= num
/ inodes_per_block
;
4274 sb_breadahead(sb
, b
++);
4278 * There are other valid inodes in the buffer, this inode
4279 * has in-inode xattrs, or we don't have this inode in memory.
4280 * Read the block from disk.
4283 bh
->b_end_io
= end_buffer_read_sync
;
4284 submit_bh(READ_META
, bh
);
4286 if (!buffer_uptodate(bh
)) {
4287 ext4_error(sb
, __func__
,
4288 "unable to read inode block - inode=%lu, "
4289 "block=%llu", inode
->i_ino
, block
);
4299 int ext4_get_inode_loc(struct inode
*inode
, struct ext4_iloc
*iloc
)
4301 /* We have all inode data except xattrs in memory here. */
4302 return __ext4_get_inode_loc(inode
, iloc
,
4303 !(EXT4_I(inode
)->i_state
& EXT4_STATE_XATTR
));
4306 void ext4_set_inode_flags(struct inode
*inode
)
4308 unsigned int flags
= EXT4_I(inode
)->i_flags
;
4310 inode
->i_flags
&= ~(S_SYNC
|S_APPEND
|S_IMMUTABLE
|S_NOATIME
|S_DIRSYNC
);
4311 if (flags
& EXT4_SYNC_FL
)
4312 inode
->i_flags
|= S_SYNC
;
4313 if (flags
& EXT4_APPEND_FL
)
4314 inode
->i_flags
|= S_APPEND
;
4315 if (flags
& EXT4_IMMUTABLE_FL
)
4316 inode
->i_flags
|= S_IMMUTABLE
;
4317 if (flags
& EXT4_NOATIME_FL
)
4318 inode
->i_flags
|= S_NOATIME
;
4319 if (flags
& EXT4_DIRSYNC_FL
)
4320 inode
->i_flags
|= S_DIRSYNC
;
4323 /* Propagate flags from i_flags to EXT4_I(inode)->i_flags */
4324 void ext4_get_inode_flags(struct ext4_inode_info
*ei
)
4326 unsigned int flags
= ei
->vfs_inode
.i_flags
;
4328 ei
->i_flags
&= ~(EXT4_SYNC_FL
|EXT4_APPEND_FL
|
4329 EXT4_IMMUTABLE_FL
|EXT4_NOATIME_FL
|EXT4_DIRSYNC_FL
);
4331 ei
->i_flags
|= EXT4_SYNC_FL
;
4332 if (flags
& S_APPEND
)
4333 ei
->i_flags
|= EXT4_APPEND_FL
;
4334 if (flags
& S_IMMUTABLE
)
4335 ei
->i_flags
|= EXT4_IMMUTABLE_FL
;
4336 if (flags
& S_NOATIME
)
4337 ei
->i_flags
|= EXT4_NOATIME_FL
;
4338 if (flags
& S_DIRSYNC
)
4339 ei
->i_flags
|= EXT4_DIRSYNC_FL
;
4341 static blkcnt_t
ext4_inode_blocks(struct ext4_inode
*raw_inode
,
4342 struct ext4_inode_info
*ei
)
4345 struct inode
*inode
= &(ei
->vfs_inode
);
4346 struct super_block
*sb
= inode
->i_sb
;
4348 if (EXT4_HAS_RO_COMPAT_FEATURE(sb
,
4349 EXT4_FEATURE_RO_COMPAT_HUGE_FILE
)) {
4350 /* we are using combined 48 bit field */
4351 i_blocks
= ((u64
)le16_to_cpu(raw_inode
->i_blocks_high
)) << 32 |
4352 le32_to_cpu(raw_inode
->i_blocks_lo
);
4353 if (ei
->i_flags
& EXT4_HUGE_FILE_FL
) {
4354 /* i_blocks represent file system block size */
4355 return i_blocks
<< (inode
->i_blkbits
- 9);
4360 return le32_to_cpu(raw_inode
->i_blocks_lo
);
4364 struct inode
*ext4_iget(struct super_block
*sb
, unsigned long ino
)
4366 struct ext4_iloc iloc
;
4367 struct ext4_inode
*raw_inode
;
4368 struct ext4_inode_info
*ei
;
4369 struct buffer_head
*bh
;
4370 struct inode
*inode
;
4374 inode
= iget_locked(sb
, ino
);
4376 return ERR_PTR(-ENOMEM
);
4377 if (!(inode
->i_state
& I_NEW
))
4381 #ifdef CONFIG_EXT4_FS_POSIX_ACL
4382 ei
->i_acl
= EXT4_ACL_NOT_CACHED
;
4383 ei
->i_default_acl
= EXT4_ACL_NOT_CACHED
;
4386 ret
= __ext4_get_inode_loc(inode
, &iloc
, 0);
4390 raw_inode
= ext4_raw_inode(&iloc
);
4391 inode
->i_mode
= le16_to_cpu(raw_inode
->i_mode
);
4392 inode
->i_uid
= (uid_t
)le16_to_cpu(raw_inode
->i_uid_low
);
4393 inode
->i_gid
= (gid_t
)le16_to_cpu(raw_inode
->i_gid_low
);
4394 if (!(test_opt(inode
->i_sb
, NO_UID32
))) {
4395 inode
->i_uid
|= le16_to_cpu(raw_inode
->i_uid_high
) << 16;
4396 inode
->i_gid
|= le16_to_cpu(raw_inode
->i_gid_high
) << 16;
4398 inode
->i_nlink
= le16_to_cpu(raw_inode
->i_links_count
);
4401 ei
->i_dir_start_lookup
= 0;
4402 ei
->i_dtime
= le32_to_cpu(raw_inode
->i_dtime
);
4403 /* We now have enough fields to check if the inode was active or not.
4404 * This is needed because nfsd might try to access dead inodes
4405 * the test is that same one that e2fsck uses
4406 * NeilBrown 1999oct15
4408 if (inode
->i_nlink
== 0) {
4409 if (inode
->i_mode
== 0 ||
4410 !(EXT4_SB(inode
->i_sb
)->s_mount_state
& EXT4_ORPHAN_FS
)) {
4411 /* this inode is deleted */
4416 /* The only unlinked inodes we let through here have
4417 * valid i_mode and are being read by the orphan
4418 * recovery code: that's fine, we're about to complete
4419 * the process of deleting those. */
4421 ei
->i_flags
= le32_to_cpu(raw_inode
->i_flags
);
4422 inode
->i_blocks
= ext4_inode_blocks(raw_inode
, ei
);
4423 ei
->i_file_acl
= le32_to_cpu(raw_inode
->i_file_acl_lo
);
4424 if (EXT4_HAS_INCOMPAT_FEATURE(sb
, EXT4_FEATURE_INCOMPAT_64BIT
))
4426 ((__u64
)le16_to_cpu(raw_inode
->i_file_acl_high
)) << 32;
4427 inode
->i_size
= ext4_isize(raw_inode
);
4428 ei
->i_disksize
= inode
->i_size
;
4429 inode
->i_generation
= le32_to_cpu(raw_inode
->i_generation
);
4430 ei
->i_block_group
= iloc
.block_group
;
4431 ei
->i_last_alloc_group
= ~0;
4433 * NOTE! The in-memory inode i_data array is in little-endian order
4434 * even on big-endian machines: we do NOT byteswap the block numbers!
4436 for (block
= 0; block
< EXT4_N_BLOCKS
; block
++)
4437 ei
->i_data
[block
] = raw_inode
->i_block
[block
];
4438 INIT_LIST_HEAD(&ei
->i_orphan
);
4440 if (EXT4_INODE_SIZE(inode
->i_sb
) > EXT4_GOOD_OLD_INODE_SIZE
) {
4441 ei
->i_extra_isize
= le16_to_cpu(raw_inode
->i_extra_isize
);
4442 if (EXT4_GOOD_OLD_INODE_SIZE
+ ei
->i_extra_isize
>
4443 EXT4_INODE_SIZE(inode
->i_sb
)) {
4448 if (ei
->i_extra_isize
== 0) {
4449 /* The extra space is currently unused. Use it. */
4450 ei
->i_extra_isize
= sizeof(struct ext4_inode
) -
4451 EXT4_GOOD_OLD_INODE_SIZE
;
4453 __le32
*magic
= (void *)raw_inode
+
4454 EXT4_GOOD_OLD_INODE_SIZE
+
4456 if (*magic
== cpu_to_le32(EXT4_XATTR_MAGIC
))
4457 ei
->i_state
|= EXT4_STATE_XATTR
;
4460 ei
->i_extra_isize
= 0;
4462 EXT4_INODE_GET_XTIME(i_ctime
, inode
, raw_inode
);
4463 EXT4_INODE_GET_XTIME(i_mtime
, inode
, raw_inode
);
4464 EXT4_INODE_GET_XTIME(i_atime
, inode
, raw_inode
);
4465 EXT4_EINODE_GET_XTIME(i_crtime
, ei
, raw_inode
);
4467 inode
->i_version
= le32_to_cpu(raw_inode
->i_disk_version
);
4468 if (EXT4_INODE_SIZE(inode
->i_sb
) > EXT4_GOOD_OLD_INODE_SIZE
) {
4469 if (EXT4_FITS_IN_INODE(raw_inode
, ei
, i_version_hi
))
4471 (__u64
)(le32_to_cpu(raw_inode
->i_version_hi
)) << 32;
4475 if (ei
->i_file_acl
&&
4477 (le32_to_cpu(EXT4_SB(sb
)->s_es
->s_first_data_block
) +
4478 EXT4_SB(sb
)->s_gdb_count
)) ||
4479 (ei
->i_file_acl
>= ext4_blocks_count(EXT4_SB(sb
)->s_es
)))) {
4480 ext4_error(sb
, __func__
,
4481 "bad extended attribute block %llu in inode #%lu",
4482 ei
->i_file_acl
, inode
->i_ino
);
4485 } else if (ei
->i_flags
& EXT4_EXTENTS_FL
) {
4486 if (S_ISREG(inode
->i_mode
) || S_ISDIR(inode
->i_mode
) ||
4487 (S_ISLNK(inode
->i_mode
) &&
4488 !ext4_inode_is_fast_symlink(inode
)))
4489 /* Validate extent which is part of inode */
4490 ret
= ext4_ext_check_inode(inode
);
4491 } else if (S_ISREG(inode
->i_mode
) || S_ISDIR(inode
->i_mode
) ||
4492 (S_ISLNK(inode
->i_mode
) &&
4493 !ext4_inode_is_fast_symlink(inode
))) {
4494 /* Validate block references which are part of inode */
4495 ret
= ext4_check_inode_blockref(inode
);
4502 if (S_ISREG(inode
->i_mode
)) {
4503 inode
->i_op
= &ext4_file_inode_operations
;
4504 inode
->i_fop
= &ext4_file_operations
;
4505 ext4_set_aops(inode
);
4506 } else if (S_ISDIR(inode
->i_mode
)) {
4507 inode
->i_op
= &ext4_dir_inode_operations
;
4508 inode
->i_fop
= &ext4_dir_operations
;
4509 } else if (S_ISLNK(inode
->i_mode
)) {
4510 if (ext4_inode_is_fast_symlink(inode
)) {
4511 inode
->i_op
= &ext4_fast_symlink_inode_operations
;
4512 nd_terminate_link(ei
->i_data
, inode
->i_size
,
4513 sizeof(ei
->i_data
) - 1);
4515 inode
->i_op
= &ext4_symlink_inode_operations
;
4516 ext4_set_aops(inode
);
4518 } else if (S_ISCHR(inode
->i_mode
) || S_ISBLK(inode
->i_mode
) ||
4519 S_ISFIFO(inode
->i_mode
) || S_ISSOCK(inode
->i_mode
)) {
4520 inode
->i_op
= &ext4_special_inode_operations
;
4521 if (raw_inode
->i_block
[0])
4522 init_special_inode(inode
, inode
->i_mode
,
4523 old_decode_dev(le32_to_cpu(raw_inode
->i_block
[0])));
4525 init_special_inode(inode
, inode
->i_mode
,
4526 new_decode_dev(le32_to_cpu(raw_inode
->i_block
[1])));
4530 ext4_error(inode
->i_sb
, __func__
,
4531 "bogus i_mode (%o) for inode=%lu",
4532 inode
->i_mode
, inode
->i_ino
);
4536 ext4_set_inode_flags(inode
);
4537 unlock_new_inode(inode
);
4542 return ERR_PTR(ret
);
4545 static int ext4_inode_blocks_set(handle_t
*handle
,
4546 struct ext4_inode
*raw_inode
,
4547 struct ext4_inode_info
*ei
)
4549 struct inode
*inode
= &(ei
->vfs_inode
);
4550 u64 i_blocks
= inode
->i_blocks
;
4551 struct super_block
*sb
= inode
->i_sb
;
4553 if (i_blocks
<= ~0U) {
4555 * i_blocks can be represnted in a 32 bit variable
4556 * as multiple of 512 bytes
4558 raw_inode
->i_blocks_lo
= cpu_to_le32(i_blocks
);
4559 raw_inode
->i_blocks_high
= 0;
4560 ei
->i_flags
&= ~EXT4_HUGE_FILE_FL
;
4563 if (!EXT4_HAS_RO_COMPAT_FEATURE(sb
, EXT4_FEATURE_RO_COMPAT_HUGE_FILE
))
4566 if (i_blocks
<= 0xffffffffffffULL
) {
4568 * i_blocks can be represented in a 48 bit variable
4569 * as multiple of 512 bytes
4571 raw_inode
->i_blocks_lo
= cpu_to_le32(i_blocks
);
4572 raw_inode
->i_blocks_high
= cpu_to_le16(i_blocks
>> 32);
4573 ei
->i_flags
&= ~EXT4_HUGE_FILE_FL
;
4575 ei
->i_flags
|= EXT4_HUGE_FILE_FL
;
4576 /* i_block is stored in file system block size */
4577 i_blocks
= i_blocks
>> (inode
->i_blkbits
- 9);
4578 raw_inode
->i_blocks_lo
= cpu_to_le32(i_blocks
);
4579 raw_inode
->i_blocks_high
= cpu_to_le16(i_blocks
>> 32);
4585 * Post the struct inode info into an on-disk inode location in the
4586 * buffer-cache. This gobbles the caller's reference to the
4587 * buffer_head in the inode location struct.
4589 * The caller must have write access to iloc->bh.
4591 static int ext4_do_update_inode(handle_t
*handle
,
4592 struct inode
*inode
,
4593 struct ext4_iloc
*iloc
)
4595 struct ext4_inode
*raw_inode
= ext4_raw_inode(iloc
);
4596 struct ext4_inode_info
*ei
= EXT4_I(inode
);
4597 struct buffer_head
*bh
= iloc
->bh
;
4598 int err
= 0, rc
, block
;
4600 /* For fields not not tracking in the in-memory inode,
4601 * initialise them to zero for new inodes. */
4602 if (ei
->i_state
& EXT4_STATE_NEW
)
4603 memset(raw_inode
, 0, EXT4_SB(inode
->i_sb
)->s_inode_size
);
4605 ext4_get_inode_flags(ei
);
4606 raw_inode
->i_mode
= cpu_to_le16(inode
->i_mode
);
4607 if (!(test_opt(inode
->i_sb
, NO_UID32
))) {
4608 raw_inode
->i_uid_low
= cpu_to_le16(low_16_bits(inode
->i_uid
));
4609 raw_inode
->i_gid_low
= cpu_to_le16(low_16_bits(inode
->i_gid
));
4611 * Fix up interoperability with old kernels. Otherwise, old inodes get
4612 * re-used with the upper 16 bits of the uid/gid intact
4615 raw_inode
->i_uid_high
=
4616 cpu_to_le16(high_16_bits(inode
->i_uid
));
4617 raw_inode
->i_gid_high
=
4618 cpu_to_le16(high_16_bits(inode
->i_gid
));
4620 raw_inode
->i_uid_high
= 0;
4621 raw_inode
->i_gid_high
= 0;
4624 raw_inode
->i_uid_low
=
4625 cpu_to_le16(fs_high2lowuid(inode
->i_uid
));
4626 raw_inode
->i_gid_low
=
4627 cpu_to_le16(fs_high2lowgid(inode
->i_gid
));
4628 raw_inode
->i_uid_high
= 0;
4629 raw_inode
->i_gid_high
= 0;
4631 raw_inode
->i_links_count
= cpu_to_le16(inode
->i_nlink
);
4633 EXT4_INODE_SET_XTIME(i_ctime
, inode
, raw_inode
);
4634 EXT4_INODE_SET_XTIME(i_mtime
, inode
, raw_inode
);
4635 EXT4_INODE_SET_XTIME(i_atime
, inode
, raw_inode
);
4636 EXT4_EINODE_SET_XTIME(i_crtime
, ei
, raw_inode
);
4638 if (ext4_inode_blocks_set(handle
, raw_inode
, ei
))
4640 raw_inode
->i_dtime
= cpu_to_le32(ei
->i_dtime
);
4641 /* clear the migrate flag in the raw_inode */
4642 raw_inode
->i_flags
= cpu_to_le32(ei
->i_flags
& ~EXT4_EXT_MIGRATE
);
4643 if (EXT4_SB(inode
->i_sb
)->s_es
->s_creator_os
!=
4644 cpu_to_le32(EXT4_OS_HURD
))
4645 raw_inode
->i_file_acl_high
=
4646 cpu_to_le16(ei
->i_file_acl
>> 32);
4647 raw_inode
->i_file_acl_lo
= cpu_to_le32(ei
->i_file_acl
);
4648 ext4_isize_set(raw_inode
, ei
->i_disksize
);
4649 if (ei
->i_disksize
> 0x7fffffffULL
) {
4650 struct super_block
*sb
= inode
->i_sb
;
4651 if (!EXT4_HAS_RO_COMPAT_FEATURE(sb
,
4652 EXT4_FEATURE_RO_COMPAT_LARGE_FILE
) ||
4653 EXT4_SB(sb
)->s_es
->s_rev_level
==
4654 cpu_to_le32(EXT4_GOOD_OLD_REV
)) {
4655 /* If this is the first large file
4656 * created, add a flag to the superblock.
4658 err
= ext4_journal_get_write_access(handle
,
4659 EXT4_SB(sb
)->s_sbh
);
4662 ext4_update_dynamic_rev(sb
);
4663 EXT4_SET_RO_COMPAT_FEATURE(sb
,
4664 EXT4_FEATURE_RO_COMPAT_LARGE_FILE
);
4666 ext4_handle_sync(handle
);
4667 err
= ext4_handle_dirty_metadata(handle
, inode
,
4668 EXT4_SB(sb
)->s_sbh
);
4671 raw_inode
->i_generation
= cpu_to_le32(inode
->i_generation
);
4672 if (S_ISCHR(inode
->i_mode
) || S_ISBLK(inode
->i_mode
)) {
4673 if (old_valid_dev(inode
->i_rdev
)) {
4674 raw_inode
->i_block
[0] =
4675 cpu_to_le32(old_encode_dev(inode
->i_rdev
));
4676 raw_inode
->i_block
[1] = 0;
4678 raw_inode
->i_block
[0] = 0;
4679 raw_inode
->i_block
[1] =
4680 cpu_to_le32(new_encode_dev(inode
->i_rdev
));
4681 raw_inode
->i_block
[2] = 0;
4683 } else for (block
= 0; block
< EXT4_N_BLOCKS
; block
++)
4684 raw_inode
->i_block
[block
] = ei
->i_data
[block
];
4686 raw_inode
->i_disk_version
= cpu_to_le32(inode
->i_version
);
4687 if (ei
->i_extra_isize
) {
4688 if (EXT4_FITS_IN_INODE(raw_inode
, ei
, i_version_hi
))
4689 raw_inode
->i_version_hi
=
4690 cpu_to_le32(inode
->i_version
>> 32);
4691 raw_inode
->i_extra_isize
= cpu_to_le16(ei
->i_extra_isize
);
4694 BUFFER_TRACE(bh
, "call ext4_handle_dirty_metadata");
4695 rc
= ext4_handle_dirty_metadata(handle
, inode
, bh
);
4698 ei
->i_state
&= ~EXT4_STATE_NEW
;
4702 ext4_std_error(inode
->i_sb
, err
);
4707 * ext4_write_inode()
4709 * We are called from a few places:
4711 * - Within generic_file_write() for O_SYNC files.
4712 * Here, there will be no transaction running. We wait for any running
4713 * trasnaction to commit.
4715 * - Within sys_sync(), kupdate and such.
4716 * We wait on commit, if tol to.
4718 * - Within prune_icache() (PF_MEMALLOC == true)
4719 * Here we simply return. We can't afford to block kswapd on the
4722 * In all cases it is actually safe for us to return without doing anything,
4723 * because the inode has been copied into a raw inode buffer in
4724 * ext4_mark_inode_dirty(). This is a correctness thing for O_SYNC and for
4727 * Note that we are absolutely dependent upon all inode dirtiers doing the
4728 * right thing: they *must* call mark_inode_dirty() after dirtying info in
4729 * which we are interested.
4731 * It would be a bug for them to not do this. The code:
4733 * mark_inode_dirty(inode)
4735 * inode->i_size = expr;
4737 * is in error because a kswapd-driven write_inode() could occur while
4738 * `stuff()' is running, and the new i_size will be lost. Plus the inode
4739 * will no longer be on the superblock's dirty inode list.
4741 int ext4_write_inode(struct inode
*inode
, int wait
)
4743 if (current
->flags
& PF_MEMALLOC
)
4746 if (ext4_journal_current_handle()) {
4747 jbd_debug(1, "called recursively, non-PF_MEMALLOC!\n");
4755 return ext4_force_commit(inode
->i_sb
);
4758 int __ext4_write_dirty_metadata(struct inode
*inode
, struct buffer_head
*bh
)
4762 mark_buffer_dirty(bh
);
4763 if (inode
&& inode_needs_sync(inode
)) {
4764 sync_dirty_buffer(bh
);
4765 if (buffer_req(bh
) && !buffer_uptodate(bh
)) {
4766 ext4_error(inode
->i_sb
, __func__
,
4767 "IO error syncing inode, "
4768 "inode=%lu, block=%llu",
4770 (unsigned long long)bh
->b_blocknr
);
4780 * Called from notify_change.
4782 * We want to trap VFS attempts to truncate the file as soon as
4783 * possible. In particular, we want to make sure that when the VFS
4784 * shrinks i_size, we put the inode on the orphan list and modify
4785 * i_disksize immediately, so that during the subsequent flushing of
4786 * dirty pages and freeing of disk blocks, we can guarantee that any
4787 * commit will leave the blocks being flushed in an unused state on
4788 * disk. (On recovery, the inode will get truncated and the blocks will
4789 * be freed, so we have a strong guarantee that no future commit will
4790 * leave these blocks visible to the user.)
4792 * Another thing we have to assure is that if we are in ordered mode
4793 * and inode is still attached to the committing transaction, we must
4794 * we start writeout of all the dirty pages which are being truncated.
4795 * This way we are sure that all the data written in the previous
4796 * transaction are already on disk (truncate waits for pages under
4799 * Called with inode->i_mutex down.
4801 int ext4_setattr(struct dentry
*dentry
, struct iattr
*attr
)
4803 struct inode
*inode
= dentry
->d_inode
;
4805 const unsigned int ia_valid
= attr
->ia_valid
;
4807 error
= inode_change_ok(inode
, attr
);
4811 if ((ia_valid
& ATTR_UID
&& attr
->ia_uid
!= inode
->i_uid
) ||
4812 (ia_valid
& ATTR_GID
&& attr
->ia_gid
!= inode
->i_gid
)) {
4815 /* (user+group)*(old+new) structure, inode write (sb,
4816 * inode block, ? - but truncate inode update has it) */
4817 handle
= ext4_journal_start(inode
, 2*(EXT4_QUOTA_INIT_BLOCKS(inode
->i_sb
)+
4818 EXT4_QUOTA_DEL_BLOCKS(inode
->i_sb
))+3);
4819 if (IS_ERR(handle
)) {
4820 error
= PTR_ERR(handle
);
4823 error
= vfs_dq_transfer(inode
, attr
) ? -EDQUOT
: 0;
4825 ext4_journal_stop(handle
);
4828 /* Update corresponding info in inode so that everything is in
4829 * one transaction */
4830 if (attr
->ia_valid
& ATTR_UID
)
4831 inode
->i_uid
= attr
->ia_uid
;
4832 if (attr
->ia_valid
& ATTR_GID
)
4833 inode
->i_gid
= attr
->ia_gid
;
4834 error
= ext4_mark_inode_dirty(handle
, inode
);
4835 ext4_journal_stop(handle
);
4838 if (attr
->ia_valid
& ATTR_SIZE
) {
4839 if (!(EXT4_I(inode
)->i_flags
& EXT4_EXTENTS_FL
)) {
4840 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
4842 if (attr
->ia_size
> sbi
->s_bitmap_maxbytes
) {
4849 if (S_ISREG(inode
->i_mode
) &&
4850 attr
->ia_valid
& ATTR_SIZE
&& attr
->ia_size
< inode
->i_size
) {
4853 handle
= ext4_journal_start(inode
, 3);
4854 if (IS_ERR(handle
)) {
4855 error
= PTR_ERR(handle
);
4859 error
= ext4_orphan_add(handle
, inode
);
4860 EXT4_I(inode
)->i_disksize
= attr
->ia_size
;
4861 rc
= ext4_mark_inode_dirty(handle
, inode
);
4864 ext4_journal_stop(handle
);
4866 if (ext4_should_order_data(inode
)) {
4867 error
= ext4_begin_ordered_truncate(inode
,
4870 /* Do as much error cleanup as possible */
4871 handle
= ext4_journal_start(inode
, 3);
4872 if (IS_ERR(handle
)) {
4873 ext4_orphan_del(NULL
, inode
);
4876 ext4_orphan_del(handle
, inode
);
4877 ext4_journal_stop(handle
);
4883 rc
= inode_setattr(inode
, attr
);
4885 /* If inode_setattr's call to ext4_truncate failed to get a
4886 * transaction handle at all, we need to clean up the in-core
4887 * orphan list manually. */
4889 ext4_orphan_del(NULL
, inode
);
4891 if (!rc
&& (ia_valid
& ATTR_MODE
))
4892 rc
= ext4_acl_chmod(inode
);
4895 ext4_std_error(inode
->i_sb
, error
);
4901 int ext4_getattr(struct vfsmount
*mnt
, struct dentry
*dentry
,
4904 struct inode
*inode
;
4905 unsigned long delalloc_blocks
;
4907 inode
= dentry
->d_inode
;
4908 generic_fillattr(inode
, stat
);
4911 * We can't update i_blocks if the block allocation is delayed
4912 * otherwise in the case of system crash before the real block
4913 * allocation is done, we will have i_blocks inconsistent with
4914 * on-disk file blocks.
4915 * We always keep i_blocks updated together with real
4916 * allocation. But to not confuse with user, stat
4917 * will return the blocks that include the delayed allocation
4918 * blocks for this file.
4920 spin_lock(&EXT4_I(inode
)->i_block_reservation_lock
);
4921 delalloc_blocks
= EXT4_I(inode
)->i_reserved_data_blocks
;
4922 spin_unlock(&EXT4_I(inode
)->i_block_reservation_lock
);
4924 stat
->blocks
+= (delalloc_blocks
<< inode
->i_sb
->s_blocksize_bits
)>>9;
4928 static int ext4_indirect_trans_blocks(struct inode
*inode
, int nrblocks
,
4933 /* if nrblocks are contiguous */
4936 * With N contiguous data blocks, it need at most
4937 * N/EXT4_ADDR_PER_BLOCK(inode->i_sb) indirect blocks
4938 * 2 dindirect blocks
4941 indirects
= nrblocks
/ EXT4_ADDR_PER_BLOCK(inode
->i_sb
);
4942 return indirects
+ 3;
4945 * if nrblocks are not contiguous, worse case, each block touch
4946 * a indirect block, and each indirect block touch a double indirect
4947 * block, plus a triple indirect block
4949 indirects
= nrblocks
* 2 + 1;
4953 static int ext4_index_trans_blocks(struct inode
*inode
, int nrblocks
, int chunk
)
4955 if (!(EXT4_I(inode
)->i_flags
& EXT4_EXTENTS_FL
))
4956 return ext4_indirect_trans_blocks(inode
, nrblocks
, chunk
);
4957 return ext4_ext_index_trans_blocks(inode
, nrblocks
, chunk
);
4961 * Account for index blocks, block groups bitmaps and block group
4962 * descriptor blocks if modify datablocks and index blocks
4963 * worse case, the indexs blocks spread over different block groups
4965 * If datablocks are discontiguous, they are possible to spread over
4966 * different block groups too. If they are contiugous, with flexbg,
4967 * they could still across block group boundary.
4969 * Also account for superblock, inode, quota and xattr blocks
4971 int ext4_meta_trans_blocks(struct inode
*inode
, int nrblocks
, int chunk
)
4973 ext4_group_t groups
, ngroups
= ext4_get_groups_count(inode
->i_sb
);
4979 * How many index blocks need to touch to modify nrblocks?
4980 * The "Chunk" flag indicating whether the nrblocks is
4981 * physically contiguous on disk
4983 * For Direct IO and fallocate, they calls get_block to allocate
4984 * one single extent at a time, so they could set the "Chunk" flag
4986 idxblocks
= ext4_index_trans_blocks(inode
, nrblocks
, chunk
);
4991 * Now let's see how many group bitmaps and group descriptors need
5001 if (groups
> ngroups
)
5003 if (groups
> EXT4_SB(inode
->i_sb
)->s_gdb_count
)
5004 gdpblocks
= EXT4_SB(inode
->i_sb
)->s_gdb_count
;
5006 /* bitmaps and block group descriptor blocks */
5007 ret
+= groups
+ gdpblocks
;
5009 /* Blocks for super block, inode, quota and xattr blocks */
5010 ret
+= EXT4_META_TRANS_BLOCKS(inode
->i_sb
);
5016 * Calulate the total number of credits to reserve to fit
5017 * the modification of a single pages into a single transaction,
5018 * which may include multiple chunks of block allocations.
5020 * This could be called via ext4_write_begin()
5022 * We need to consider the worse case, when
5023 * one new block per extent.
5025 int ext4_writepage_trans_blocks(struct inode
*inode
)
5027 int bpp
= ext4_journal_blocks_per_page(inode
);
5030 ret
= ext4_meta_trans_blocks(inode
, bpp
, 0);
5032 /* Account for data blocks for journalled mode */
5033 if (ext4_should_journal_data(inode
))
5039 * Calculate the journal credits for a chunk of data modification.
5041 * This is called from DIO, fallocate or whoever calling
5042 * ext4_get_blocks() to map/allocate a chunk of contigous disk blocks.
5044 * journal buffers for data blocks are not included here, as DIO
5045 * and fallocate do no need to journal data buffers.
5047 int ext4_chunk_trans_blocks(struct inode
*inode
, int nrblocks
)
5049 return ext4_meta_trans_blocks(inode
, nrblocks
, 1);
5053 * The caller must have previously called ext4_reserve_inode_write().
5054 * Give this, we know that the caller already has write access to iloc->bh.
5056 int ext4_mark_iloc_dirty(handle_t
*handle
,
5057 struct inode
*inode
, struct ext4_iloc
*iloc
)
5061 if (test_opt(inode
->i_sb
, I_VERSION
))
5062 inode_inc_iversion(inode
);
5064 /* the do_update_inode consumes one bh->b_count */
5067 /* ext4_do_update_inode() does jbd2_journal_dirty_metadata */
5068 err
= ext4_do_update_inode(handle
, inode
, iloc
);
5074 * On success, We end up with an outstanding reference count against
5075 * iloc->bh. This _must_ be cleaned up later.
5079 ext4_reserve_inode_write(handle_t
*handle
, struct inode
*inode
,
5080 struct ext4_iloc
*iloc
)
5084 err
= ext4_get_inode_loc(inode
, iloc
);
5086 BUFFER_TRACE(iloc
->bh
, "get_write_access");
5087 err
= ext4_journal_get_write_access(handle
, iloc
->bh
);
5093 ext4_std_error(inode
->i_sb
, err
);
5098 * Expand an inode by new_extra_isize bytes.
5099 * Returns 0 on success or negative error number on failure.
5101 static int ext4_expand_extra_isize(struct inode
*inode
,
5102 unsigned int new_extra_isize
,
5103 struct ext4_iloc iloc
,
5106 struct ext4_inode
*raw_inode
;
5107 struct ext4_xattr_ibody_header
*header
;
5108 struct ext4_xattr_entry
*entry
;
5110 if (EXT4_I(inode
)->i_extra_isize
>= new_extra_isize
)
5113 raw_inode
= ext4_raw_inode(&iloc
);
5115 header
= IHDR(inode
, raw_inode
);
5116 entry
= IFIRST(header
);
5118 /* No extended attributes present */
5119 if (!(EXT4_I(inode
)->i_state
& EXT4_STATE_XATTR
) ||
5120 header
->h_magic
!= cpu_to_le32(EXT4_XATTR_MAGIC
)) {
5121 memset((void *)raw_inode
+ EXT4_GOOD_OLD_INODE_SIZE
, 0,
5123 EXT4_I(inode
)->i_extra_isize
= new_extra_isize
;
5127 /* try to expand with EAs present */
5128 return ext4_expand_extra_isize_ea(inode
, new_extra_isize
,
5133 * What we do here is to mark the in-core inode as clean with respect to inode
5134 * dirtiness (it may still be data-dirty).
5135 * This means that the in-core inode may be reaped by prune_icache
5136 * without having to perform any I/O. This is a very good thing,
5137 * because *any* task may call prune_icache - even ones which
5138 * have a transaction open against a different journal.
5140 * Is this cheating? Not really. Sure, we haven't written the
5141 * inode out, but prune_icache isn't a user-visible syncing function.
5142 * Whenever the user wants stuff synced (sys_sync, sys_msync, sys_fsync)
5143 * we start and wait on commits.
5145 * Is this efficient/effective? Well, we're being nice to the system
5146 * by cleaning up our inodes proactively so they can be reaped
5147 * without I/O. But we are potentially leaving up to five seconds'
5148 * worth of inodes floating about which prune_icache wants us to
5149 * write out. One way to fix that would be to get prune_icache()
5150 * to do a write_super() to free up some memory. It has the desired
5153 int ext4_mark_inode_dirty(handle_t
*handle
, struct inode
*inode
)
5155 struct ext4_iloc iloc
;
5156 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
5157 static unsigned int mnt_count
;
5161 err
= ext4_reserve_inode_write(handle
, inode
, &iloc
);
5162 if (ext4_handle_valid(handle
) &&
5163 EXT4_I(inode
)->i_extra_isize
< sbi
->s_want_extra_isize
&&
5164 !(EXT4_I(inode
)->i_state
& EXT4_STATE_NO_EXPAND
)) {
5166 * We need extra buffer credits since we may write into EA block
5167 * with this same handle. If journal_extend fails, then it will
5168 * only result in a minor loss of functionality for that inode.
5169 * If this is felt to be critical, then e2fsck should be run to
5170 * force a large enough s_min_extra_isize.
5172 if ((jbd2_journal_extend(handle
,
5173 EXT4_DATA_TRANS_BLOCKS(inode
->i_sb
))) == 0) {
5174 ret
= ext4_expand_extra_isize(inode
,
5175 sbi
->s_want_extra_isize
,
5178 EXT4_I(inode
)->i_state
|= EXT4_STATE_NO_EXPAND
;
5180 le16_to_cpu(sbi
->s_es
->s_mnt_count
)) {
5181 ext4_warning(inode
->i_sb
, __func__
,
5182 "Unable to expand inode %lu. Delete"
5183 " some EAs or run e2fsck.",
5186 le16_to_cpu(sbi
->s_es
->s_mnt_count
);
5192 err
= ext4_mark_iloc_dirty(handle
, inode
, &iloc
);
5197 * ext4_dirty_inode() is called from __mark_inode_dirty()
5199 * We're really interested in the case where a file is being extended.
5200 * i_size has been changed by generic_commit_write() and we thus need
5201 * to include the updated inode in the current transaction.
5203 * Also, vfs_dq_alloc_block() will always dirty the inode when blocks
5204 * are allocated to the file.
5206 * If the inode is marked synchronous, we don't honour that here - doing
5207 * so would cause a commit on atime updates, which we don't bother doing.
5208 * We handle synchronous inodes at the highest possible level.
5210 void ext4_dirty_inode(struct inode
*inode
)
5212 handle_t
*current_handle
= ext4_journal_current_handle();
5215 if (!ext4_handle_valid(current_handle
)) {
5216 ext4_mark_inode_dirty(current_handle
, inode
);
5220 handle
= ext4_journal_start(inode
, 2);
5223 if (current_handle
&&
5224 current_handle
->h_transaction
!= handle
->h_transaction
) {
5225 /* This task has a transaction open against a different fs */
5226 printk(KERN_EMERG
"%s: transactions do not match!\n",
5229 jbd_debug(5, "marking dirty. outer handle=%p\n",
5231 ext4_mark_inode_dirty(handle
, inode
);
5233 ext4_journal_stop(handle
);
5240 * Bind an inode's backing buffer_head into this transaction, to prevent
5241 * it from being flushed to disk early. Unlike
5242 * ext4_reserve_inode_write, this leaves behind no bh reference and
5243 * returns no iloc structure, so the caller needs to repeat the iloc
5244 * lookup to mark the inode dirty later.
5246 static int ext4_pin_inode(handle_t
*handle
, struct inode
*inode
)
5248 struct ext4_iloc iloc
;
5252 err
= ext4_get_inode_loc(inode
, &iloc
);
5254 BUFFER_TRACE(iloc
.bh
, "get_write_access");
5255 err
= jbd2_journal_get_write_access(handle
, iloc
.bh
);
5257 err
= ext4_handle_dirty_metadata(handle
,
5263 ext4_std_error(inode
->i_sb
, err
);
5268 int ext4_change_inode_journal_flag(struct inode
*inode
, int val
)
5275 * We have to be very careful here: changing a data block's
5276 * journaling status dynamically is dangerous. If we write a
5277 * data block to the journal, change the status and then delete
5278 * that block, we risk forgetting to revoke the old log record
5279 * from the journal and so a subsequent replay can corrupt data.
5280 * So, first we make sure that the journal is empty and that
5281 * nobody is changing anything.
5284 journal
= EXT4_JOURNAL(inode
);
5287 if (is_journal_aborted(journal
))
5290 jbd2_journal_lock_updates(journal
);
5291 jbd2_journal_flush(journal
);
5294 * OK, there are no updates running now, and all cached data is
5295 * synced to disk. We are now in a completely consistent state
5296 * which doesn't have anything in the journal, and we know that
5297 * no filesystem updates are running, so it is safe to modify
5298 * the inode's in-core data-journaling state flag now.
5302 EXT4_I(inode
)->i_flags
|= EXT4_JOURNAL_DATA_FL
;
5304 EXT4_I(inode
)->i_flags
&= ~EXT4_JOURNAL_DATA_FL
;
5305 ext4_set_aops(inode
);
5307 jbd2_journal_unlock_updates(journal
);
5309 /* Finally we can mark the inode as dirty. */
5311 handle
= ext4_journal_start(inode
, 1);
5313 return PTR_ERR(handle
);
5315 err
= ext4_mark_inode_dirty(handle
, inode
);
5316 ext4_handle_sync(handle
);
5317 ext4_journal_stop(handle
);
5318 ext4_std_error(inode
->i_sb
, err
);
5323 static int ext4_bh_unmapped(handle_t
*handle
, struct buffer_head
*bh
)
5325 return !buffer_mapped(bh
);
5328 int ext4_page_mkwrite(struct vm_area_struct
*vma
, struct vm_fault
*vmf
)
5330 struct page
*page
= vmf
->page
;
5335 struct file
*file
= vma
->vm_file
;
5336 struct inode
*inode
= file
->f_path
.dentry
->d_inode
;
5337 struct address_space
*mapping
= inode
->i_mapping
;
5340 * Get i_alloc_sem to stop truncates messing with the inode. We cannot
5341 * get i_mutex because we are already holding mmap_sem.
5343 down_read(&inode
->i_alloc_sem
);
5344 size
= i_size_read(inode
);
5345 if (page
->mapping
!= mapping
|| size
<= page_offset(page
)
5346 || !PageUptodate(page
)) {
5347 /* page got truncated from under us? */
5351 if (PageMappedToDisk(page
))
5354 if (page
->index
== size
>> PAGE_CACHE_SHIFT
)
5355 len
= size
& ~PAGE_CACHE_MASK
;
5357 len
= PAGE_CACHE_SIZE
;
5359 if (page_has_buffers(page
)) {
5360 /* return if we have all the buffers mapped */
5361 if (!walk_page_buffers(NULL
, page_buffers(page
), 0, len
, NULL
,
5366 * OK, we need to fill the hole... Do write_begin write_end
5367 * to do block allocation/reservation.We are not holding
5368 * inode.i__mutex here. That allow * parallel write_begin,
5369 * write_end call. lock_page prevent this from happening
5370 * on the same page though
5372 ret
= mapping
->a_ops
->write_begin(file
, mapping
, page_offset(page
),
5373 len
, AOP_FLAG_UNINTERRUPTIBLE
, &page
, &fsdata
);
5376 ret
= mapping
->a_ops
->write_end(file
, mapping
, page_offset(page
),
5377 len
, len
, page
, fsdata
);
5383 ret
= VM_FAULT_SIGBUS
;
5384 up_read(&inode
->i_alloc_sem
);