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>
41 #include "ext4_jbd2.h"
44 #include "ext4_extents.h"
46 #include <trace/events/ext4.h>
48 #define MPAGE_DA_EXTENT_TAIL 0x01
50 static inline int ext4_begin_ordered_truncate(struct inode
*inode
,
53 return jbd2_journal_begin_ordered_truncate(
54 EXT4_SB(inode
->i_sb
)->s_journal
,
55 &EXT4_I(inode
)->jinode
,
59 static void ext4_invalidatepage(struct page
*page
, unsigned long offset
);
62 * Test whether an inode is a fast symlink.
64 static int ext4_inode_is_fast_symlink(struct inode
*inode
)
66 int ea_blocks
= EXT4_I(inode
)->i_file_acl
?
67 (inode
->i_sb
->s_blocksize
>> 9) : 0;
69 return (S_ISLNK(inode
->i_mode
) && inode
->i_blocks
- ea_blocks
== 0);
73 * The ext4 forget function must perform a revoke if we are freeing data
74 * which has been journaled. Metadata (eg. indirect blocks) must be
75 * revoked in all cases.
77 * "bh" may be NULL: a metadata block may have been freed from memory
78 * but there may still be a record of it in the journal, and that record
79 * still needs to be revoked.
81 * If the handle isn't valid we're not journaling, but we still need to
82 * call into ext4_journal_revoke() to put the buffer head.
84 int ext4_forget(handle_t
*handle
, int is_metadata
, struct inode
*inode
,
85 struct buffer_head
*bh
, ext4_fsblk_t blocknr
)
91 BUFFER_TRACE(bh
, "enter");
93 jbd_debug(4, "forgetting bh %p: is_metadata = %d, mode %o, "
95 bh
, is_metadata
, inode
->i_mode
,
96 test_opt(inode
->i_sb
, DATA_FLAGS
));
98 /* Never use the revoke function if we are doing full data
99 * journaling: there is no need to, and a V1 superblock won't
100 * support it. Otherwise, only skip the revoke on un-journaled
103 if (test_opt(inode
->i_sb
, DATA_FLAGS
) == EXT4_MOUNT_JOURNAL_DATA
||
104 (!is_metadata
&& !ext4_should_journal_data(inode
))) {
106 BUFFER_TRACE(bh
, "call jbd2_journal_forget");
107 return ext4_journal_forget(handle
, bh
);
113 * data!=journal && (is_metadata || should_journal_data(inode))
115 BUFFER_TRACE(bh
, "call ext4_journal_revoke");
116 err
= ext4_journal_revoke(handle
, blocknr
, bh
);
118 ext4_abort(inode
->i_sb
, __func__
,
119 "error %d when attempting revoke", err
);
120 BUFFER_TRACE(bh
, "exit");
125 * Work out how many blocks we need to proceed with the next chunk of a
126 * truncate transaction.
128 static unsigned long blocks_for_truncate(struct inode
*inode
)
132 needed
= inode
->i_blocks
>> (inode
->i_sb
->s_blocksize_bits
- 9);
134 /* Give ourselves just enough room to cope with inodes in which
135 * i_blocks is corrupt: we've seen disk corruptions in the past
136 * which resulted in random data in an inode which looked enough
137 * like a regular file for ext4 to try to delete it. Things
138 * will go a bit crazy if that happens, but at least we should
139 * try not to panic the whole kernel. */
143 /* But we need to bound the transaction so we don't overflow the
145 if (needed
> EXT4_MAX_TRANS_DATA
)
146 needed
= EXT4_MAX_TRANS_DATA
;
148 return EXT4_DATA_TRANS_BLOCKS(inode
->i_sb
) + needed
;
152 * Truncate transactions can be complex and absolutely huge. So we need to
153 * be able to restart the transaction at a conventient checkpoint to make
154 * sure we don't overflow the journal.
156 * start_transaction gets us a new handle for a truncate transaction,
157 * and extend_transaction tries to extend the existing one a bit. If
158 * extend fails, we need to propagate the failure up and restart the
159 * transaction in the top-level truncate loop. --sct
161 static handle_t
*start_transaction(struct inode
*inode
)
165 result
= ext4_journal_start(inode
, blocks_for_truncate(inode
));
169 ext4_std_error(inode
->i_sb
, PTR_ERR(result
));
174 * Try to extend this transaction for the purposes of truncation.
176 * Returns 0 if we managed to create more room. If we can't create more
177 * room, and the transaction must be restarted we return 1.
179 static int try_to_extend_transaction(handle_t
*handle
, struct inode
*inode
)
181 if (!ext4_handle_valid(handle
))
183 if (ext4_handle_has_enough_credits(handle
, EXT4_RESERVE_TRANS_BLOCKS
+1))
185 if (!ext4_journal_extend(handle
, blocks_for_truncate(inode
)))
191 * Restart the transaction associated with *handle. This does a commit,
192 * so before we call here everything must be consistently dirtied against
195 static int ext4_journal_test_restart(handle_t
*handle
, struct inode
*inode
)
197 BUG_ON(EXT4_JOURNAL(inode
) == NULL
);
198 jbd_debug(2, "restarting handle %p\n", handle
);
199 return ext4_journal_restart(handle
, blocks_for_truncate(inode
));
203 * Called at the last iput() if i_nlink is zero.
205 void ext4_delete_inode(struct inode
*inode
)
210 if (ext4_should_order_data(inode
))
211 ext4_begin_ordered_truncate(inode
, 0);
212 truncate_inode_pages(&inode
->i_data
, 0);
214 if (is_bad_inode(inode
))
217 handle
= ext4_journal_start(inode
, blocks_for_truncate(inode
)+3);
218 if (IS_ERR(handle
)) {
219 ext4_std_error(inode
->i_sb
, PTR_ERR(handle
));
221 * If we're going to skip the normal cleanup, we still need to
222 * make sure that the in-core orphan linked list is properly
225 ext4_orphan_del(NULL
, inode
);
230 ext4_handle_sync(handle
);
232 err
= ext4_mark_inode_dirty(handle
, inode
);
234 ext4_warning(inode
->i_sb
, __func__
,
235 "couldn't mark inode dirty (err %d)", err
);
239 ext4_truncate(inode
);
242 * ext4_ext_truncate() doesn't reserve any slop when it
243 * restarts journal transactions; therefore there may not be
244 * enough credits left in the handle to remove the inode from
245 * the orphan list and set the dtime field.
247 if (!ext4_handle_has_enough_credits(handle
, 3)) {
248 err
= ext4_journal_extend(handle
, 3);
250 err
= ext4_journal_restart(handle
, 3);
252 ext4_warning(inode
->i_sb
, __func__
,
253 "couldn't extend journal (err %d)", err
);
255 ext4_journal_stop(handle
);
261 * Kill off the orphan record which ext4_truncate created.
262 * AKPM: I think this can be inside the above `if'.
263 * Note that ext4_orphan_del() has to be able to cope with the
264 * deletion of a non-existent orphan - this is because we don't
265 * know if ext4_truncate() actually created an orphan record.
266 * (Well, we could do this if we need to, but heck - it works)
268 ext4_orphan_del(handle
, inode
);
269 EXT4_I(inode
)->i_dtime
= get_seconds();
272 * One subtle ordering requirement: if anything has gone wrong
273 * (transaction abort, IO errors, whatever), then we can still
274 * do these next steps (the fs will already have been marked as
275 * having errors), but we can't free the inode if the mark_dirty
278 if (ext4_mark_inode_dirty(handle
, inode
))
279 /* If that failed, just do the required in-core inode clear. */
282 ext4_free_inode(handle
, inode
);
283 ext4_journal_stop(handle
);
286 clear_inode(inode
); /* We must guarantee clearing of inode... */
292 struct buffer_head
*bh
;
295 static inline void add_chain(Indirect
*p
, struct buffer_head
*bh
, __le32
*v
)
297 p
->key
= *(p
->p
= v
);
302 * ext4_block_to_path - parse the block number into array of offsets
303 * @inode: inode in question (we are only interested in its superblock)
304 * @i_block: block number to be parsed
305 * @offsets: array to store the offsets in
306 * @boundary: set this non-zero if the referred-to block is likely to be
307 * followed (on disk) by an indirect block.
309 * To store the locations of file's data ext4 uses a data structure common
310 * for UNIX filesystems - tree of pointers anchored in the inode, with
311 * data blocks at leaves and indirect blocks in intermediate nodes.
312 * This function translates the block number into path in that tree -
313 * return value is the path length and @offsets[n] is the offset of
314 * pointer to (n+1)th node in the nth one. If @block is out of range
315 * (negative or too large) warning is printed and zero returned.
317 * Note: function doesn't find node addresses, so no IO is needed. All
318 * we need to know is the capacity of indirect blocks (taken from the
323 * Portability note: the last comparison (check that we fit into triple
324 * indirect block) is spelled differently, because otherwise on an
325 * architecture with 32-bit longs and 8Kb pages we might get into trouble
326 * if our filesystem had 8Kb blocks. We might use long long, but that would
327 * kill us on x86. Oh, well, at least the sign propagation does not matter -
328 * i_block would have to be negative in the very beginning, so we would not
332 static int ext4_block_to_path(struct inode
*inode
,
334 ext4_lblk_t offsets
[4], int *boundary
)
336 int ptrs
= EXT4_ADDR_PER_BLOCK(inode
->i_sb
);
337 int ptrs_bits
= EXT4_ADDR_PER_BLOCK_BITS(inode
->i_sb
);
338 const long direct_blocks
= EXT4_NDIR_BLOCKS
,
339 indirect_blocks
= ptrs
,
340 double_blocks
= (1 << (ptrs_bits
* 2));
344 if (i_block
< direct_blocks
) {
345 offsets
[n
++] = i_block
;
346 final
= direct_blocks
;
347 } else if ((i_block
-= direct_blocks
) < indirect_blocks
) {
348 offsets
[n
++] = EXT4_IND_BLOCK
;
349 offsets
[n
++] = i_block
;
351 } else if ((i_block
-= indirect_blocks
) < double_blocks
) {
352 offsets
[n
++] = EXT4_DIND_BLOCK
;
353 offsets
[n
++] = i_block
>> ptrs_bits
;
354 offsets
[n
++] = i_block
& (ptrs
- 1);
356 } else if (((i_block
-= double_blocks
) >> (ptrs_bits
* 2)) < ptrs
) {
357 offsets
[n
++] = EXT4_TIND_BLOCK
;
358 offsets
[n
++] = i_block
>> (ptrs_bits
* 2);
359 offsets
[n
++] = (i_block
>> ptrs_bits
) & (ptrs
- 1);
360 offsets
[n
++] = i_block
& (ptrs
- 1);
363 ext4_warning(inode
->i_sb
, "ext4_block_to_path",
364 "block %lu > max in inode %lu",
365 i_block
+ direct_blocks
+
366 indirect_blocks
+ double_blocks
, inode
->i_ino
);
369 *boundary
= final
- 1 - (i_block
& (ptrs
- 1));
373 static int __ext4_check_blockref(const char *function
, struct inode
*inode
,
374 __le32
*p
, unsigned int max
)
379 while (bref
< p
+max
) {
380 blk
= le32_to_cpu(*bref
++);
382 unlikely(!ext4_data_block_valid(EXT4_SB(inode
->i_sb
),
384 ext4_error(inode
->i_sb
, function
,
385 "invalid block reference %u "
386 "in inode #%lu", blk
, inode
->i_ino
);
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
);
763 /* Don't brelse(bh) here; it's done in
764 * ext4_journal_forget() below */
769 memset(bh
->b_data
, 0, blocksize
);
770 branch
[n
].p
= (__le32
*) bh
->b_data
+ offsets
[n
];
771 branch
[n
].key
= cpu_to_le32(new_blocks
[n
]);
772 *branch
[n
].p
= branch
[n
].key
;
773 if (n
== indirect_blks
) {
774 current_block
= new_blocks
[n
];
776 * End of chain, update the last new metablock of
777 * the chain to point to the new allocated
778 * data blocks numbers
780 for (i
= 1; i
< num
; i
++)
781 *(branch
[n
].p
+ i
) = cpu_to_le32(++current_block
);
783 BUFFER_TRACE(bh
, "marking uptodate");
784 set_buffer_uptodate(bh
);
787 BUFFER_TRACE(bh
, "call ext4_handle_dirty_metadata");
788 err
= ext4_handle_dirty_metadata(handle
, inode
, bh
);
795 /* Allocation failed, free what we already allocated */
796 for (i
= 1; i
<= n
; i
++) {
797 BUFFER_TRACE(branch
[i
].bh
, "call jbd2_journal_forget");
798 ext4_journal_forget(handle
, branch
[i
].bh
);
800 for (i
= 0; i
< indirect_blks
; i
++)
801 ext4_free_blocks(handle
, inode
, new_blocks
[i
], 1, 0);
803 ext4_free_blocks(handle
, inode
, new_blocks
[i
], num
, 0);
809 * ext4_splice_branch - splice the allocated branch onto inode.
811 * @block: (logical) number of block we are adding
812 * @chain: chain of indirect blocks (with a missing link - see
814 * @where: location of missing link
815 * @num: number of indirect blocks we are adding
816 * @blks: number of direct blocks we are adding
818 * This function fills the missing link and does all housekeeping needed in
819 * inode (->i_blocks, etc.). In case of success we end up with the full
820 * chain to new block and return 0.
822 static int ext4_splice_branch(handle_t
*handle
, struct inode
*inode
,
823 ext4_lblk_t block
, Indirect
*where
, int num
,
828 ext4_fsblk_t current_block
;
831 * If we're splicing into a [td]indirect block (as opposed to the
832 * inode) then we need to get write access to the [td]indirect block
836 BUFFER_TRACE(where
->bh
, "get_write_access");
837 err
= ext4_journal_get_write_access(handle
, where
->bh
);
843 *where
->p
= where
->key
;
846 * Update the host buffer_head or inode to point to more just allocated
847 * direct blocks blocks
849 if (num
== 0 && blks
> 1) {
850 current_block
= le32_to_cpu(where
->key
) + 1;
851 for (i
= 1; i
< blks
; i
++)
852 *(where
->p
+ i
) = cpu_to_le32(current_block
++);
855 /* We are done with atomic stuff, now do the rest of housekeeping */
856 /* had we spliced it onto indirect block? */
859 * If we spliced it onto an indirect block, we haven't
860 * altered the inode. Note however that if it is being spliced
861 * onto an indirect block at the very end of the file (the
862 * file is growing) then we *will* alter the inode to reflect
863 * the new i_size. But that is not done here - it is done in
864 * generic_commit_write->__mark_inode_dirty->ext4_dirty_inode.
866 jbd_debug(5, "splicing indirect only\n");
867 BUFFER_TRACE(where
->bh
, "call ext4_handle_dirty_metadata");
868 err
= ext4_handle_dirty_metadata(handle
, inode
, where
->bh
);
873 * OK, we spliced it into the inode itself on a direct block.
875 ext4_mark_inode_dirty(handle
, inode
);
876 jbd_debug(5, "splicing direct\n");
881 for (i
= 1; i
<= num
; i
++) {
882 BUFFER_TRACE(where
[i
].bh
, "call jbd2_journal_forget");
883 ext4_journal_forget(handle
, where
[i
].bh
);
884 ext4_free_blocks(handle
, inode
,
885 le32_to_cpu(where
[i
-1].key
), 1, 0);
887 ext4_free_blocks(handle
, inode
, le32_to_cpu(where
[num
].key
), blks
, 0);
893 * The ext4_ind_get_blocks() function handles non-extents inodes
894 * (i.e., using the traditional indirect/double-indirect i_blocks
895 * scheme) for ext4_get_blocks().
897 * Allocation strategy is simple: if we have to allocate something, we will
898 * have to go the whole way to leaf. So let's do it before attaching anything
899 * to tree, set linkage between the newborn blocks, write them if sync is
900 * required, recheck the path, free and repeat if check fails, otherwise
901 * set the last missing link (that will protect us from any truncate-generated
902 * removals - all blocks on the path are immune now) and possibly force the
903 * write on the parent block.
904 * That has a nice additional property: no special recovery from the failed
905 * allocations is needed - we simply release blocks and do not touch anything
906 * reachable from inode.
908 * `handle' can be NULL if create == 0.
910 * return > 0, # of blocks mapped or allocated.
911 * return = 0, if plain lookup failed.
912 * return < 0, error case.
914 * The ext4_ind_get_blocks() function should be called with
915 * down_write(&EXT4_I(inode)->i_data_sem) if allocating filesystem
916 * blocks (i.e., flags has EXT4_GET_BLOCKS_CREATE set) or
917 * down_read(&EXT4_I(inode)->i_data_sem) if not allocating file system
920 static int ext4_ind_get_blocks(handle_t
*handle
, struct inode
*inode
,
921 ext4_lblk_t iblock
, unsigned int maxblocks
,
922 struct buffer_head
*bh_result
,
926 ext4_lblk_t offsets
[4];
931 int blocks_to_boundary
= 0;
934 ext4_fsblk_t first_block
= 0;
936 J_ASSERT(!(EXT4_I(inode
)->i_flags
& EXT4_EXTENTS_FL
));
937 J_ASSERT(handle
!= NULL
|| (flags
& EXT4_GET_BLOCKS_CREATE
) == 0);
938 depth
= ext4_block_to_path(inode
, iblock
, offsets
,
939 &blocks_to_boundary
);
944 partial
= ext4_get_branch(inode
, depth
, offsets
, chain
, &err
);
946 /* Simplest case - block found, no allocation needed */
948 first_block
= le32_to_cpu(chain
[depth
- 1].key
);
949 clear_buffer_new(bh_result
);
952 while (count
< maxblocks
&& count
<= blocks_to_boundary
) {
955 blk
= le32_to_cpu(*(chain
[depth
-1].p
+ count
));
957 if (blk
== first_block
+ count
)
965 /* Next simple case - plain lookup or failed read of indirect block */
966 if ((flags
& EXT4_GET_BLOCKS_CREATE
) == 0 || err
== -EIO
)
970 * Okay, we need to do block allocation.
972 goal
= ext4_find_goal(inode
, iblock
, partial
);
974 /* the number of blocks need to allocate for [d,t]indirect blocks */
975 indirect_blks
= (chain
+ depth
) - partial
- 1;
978 * Next look up the indirect map to count the totoal number of
979 * direct blocks to allocate for this branch.
981 count
= ext4_blks_to_allocate(partial
, indirect_blks
,
982 maxblocks
, blocks_to_boundary
);
984 * Block out ext4_truncate while we alter the tree
986 err
= ext4_alloc_branch(handle
, inode
, iblock
, indirect_blks
,
988 offsets
+ (partial
- chain
), partial
);
991 * The ext4_splice_branch call will free and forget any buffers
992 * on the new chain if there is a failure, but that risks using
993 * up transaction credits, especially for bitmaps where the
994 * credits cannot be returned. Can we handle this somehow? We
995 * may need to return -EAGAIN upwards in the worst case. --sct
998 err
= ext4_splice_branch(handle
, inode
, iblock
,
999 partial
, indirect_blks
, count
);
1003 set_buffer_new(bh_result
);
1005 map_bh(bh_result
, inode
->i_sb
, le32_to_cpu(chain
[depth
-1].key
));
1006 if (count
> blocks_to_boundary
)
1007 set_buffer_boundary(bh_result
);
1009 /* Clean up and exit */
1010 partial
= chain
+ depth
- 1; /* the whole chain */
1012 while (partial
> chain
) {
1013 BUFFER_TRACE(partial
->bh
, "call brelse");
1014 brelse(partial
->bh
);
1017 BUFFER_TRACE(bh_result
, "returned");
1022 qsize_t
ext4_get_reserved_space(struct inode
*inode
)
1024 unsigned long long total
;
1026 spin_lock(&EXT4_I(inode
)->i_block_reservation_lock
);
1027 total
= EXT4_I(inode
)->i_reserved_data_blocks
+
1028 EXT4_I(inode
)->i_reserved_meta_blocks
;
1029 spin_unlock(&EXT4_I(inode
)->i_block_reservation_lock
);
1034 * Calculate the number of metadata blocks need to reserve
1035 * to allocate @blocks for non extent file based file
1037 static int ext4_indirect_calc_metadata_amount(struct inode
*inode
, int blocks
)
1039 int icap
= EXT4_ADDR_PER_BLOCK(inode
->i_sb
);
1040 int ind_blks
, dind_blks
, tind_blks
;
1042 /* number of new indirect blocks needed */
1043 ind_blks
= (blocks
+ icap
- 1) / icap
;
1045 dind_blks
= (ind_blks
+ icap
- 1) / icap
;
1049 return ind_blks
+ dind_blks
+ tind_blks
;
1053 * Calculate the number of metadata blocks need to reserve
1054 * to allocate given number of blocks
1056 static int ext4_calc_metadata_amount(struct inode
*inode
, int blocks
)
1061 if (EXT4_I(inode
)->i_flags
& EXT4_EXTENTS_FL
)
1062 return ext4_ext_calc_metadata_amount(inode
, blocks
);
1064 return ext4_indirect_calc_metadata_amount(inode
, blocks
);
1067 static void ext4_da_update_reserve_space(struct inode
*inode
, int used
)
1069 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
1070 int total
, mdb
, mdb_free
;
1072 spin_lock(&EXT4_I(inode
)->i_block_reservation_lock
);
1073 /* recalculate the number of metablocks still need to be reserved */
1074 total
= EXT4_I(inode
)->i_reserved_data_blocks
- used
;
1075 mdb
= ext4_calc_metadata_amount(inode
, total
);
1077 /* figure out how many metablocks to release */
1078 BUG_ON(mdb
> EXT4_I(inode
)->i_reserved_meta_blocks
);
1079 mdb_free
= EXT4_I(inode
)->i_reserved_meta_blocks
- mdb
;
1082 /* Account for allocated meta_blocks */
1083 mdb_free
-= EXT4_I(inode
)->i_allocated_meta_blocks
;
1085 /* update fs dirty blocks counter */
1086 percpu_counter_sub(&sbi
->s_dirtyblocks_counter
, mdb_free
);
1087 EXT4_I(inode
)->i_allocated_meta_blocks
= 0;
1088 EXT4_I(inode
)->i_reserved_meta_blocks
= mdb
;
1091 /* update per-inode reservations */
1092 BUG_ON(used
> EXT4_I(inode
)->i_reserved_data_blocks
);
1093 EXT4_I(inode
)->i_reserved_data_blocks
-= used
;
1094 spin_unlock(&EXT4_I(inode
)->i_block_reservation_lock
);
1097 * free those over-booking quota for metadata blocks
1100 vfs_dq_release_reservation_block(inode
, mdb_free
);
1103 * If we have done all the pending block allocations and if
1104 * there aren't any writers on the inode, we can discard the
1105 * inode's preallocations.
1107 if (!total
&& (atomic_read(&inode
->i_writecount
) == 0))
1108 ext4_discard_preallocations(inode
);
1111 static int check_block_validity(struct inode
*inode
, sector_t logical
,
1112 sector_t phys
, int len
)
1114 if (!ext4_data_block_valid(EXT4_SB(inode
->i_sb
), phys
, len
)) {
1115 ext4_error(inode
->i_sb
, "check_block_validity",
1116 "inode #%lu logical block %llu mapped to %llu "
1117 "(size %d)", inode
->i_ino
,
1118 (unsigned long long) logical
,
1119 (unsigned long long) phys
, len
);
1127 * The ext4_get_blocks() function tries to look up the requested blocks,
1128 * and returns if the blocks are already mapped.
1130 * Otherwise it takes the write lock of the i_data_sem and allocate blocks
1131 * and store the allocated blocks in the result buffer head and mark it
1134 * If file type is extents based, it will call ext4_ext_get_blocks(),
1135 * Otherwise, call with ext4_ind_get_blocks() to handle indirect mapping
1138 * On success, it returns the number of blocks being mapped or allocate.
1139 * if create==0 and the blocks are pre-allocated and uninitialized block,
1140 * the result buffer head is unmapped. If the create ==1, it will make sure
1141 * the buffer head is mapped.
1143 * It returns 0 if plain look up failed (blocks have not been allocated), in
1144 * that casem, buffer head is unmapped
1146 * It returns the error in case of allocation failure.
1148 int ext4_get_blocks(handle_t
*handle
, struct inode
*inode
, sector_t block
,
1149 unsigned int max_blocks
, struct buffer_head
*bh
,
1154 clear_buffer_mapped(bh
);
1155 clear_buffer_unwritten(bh
);
1158 * Try to see if we can get the block without requesting a new
1159 * file system block.
1161 down_read((&EXT4_I(inode
)->i_data_sem
));
1162 if (EXT4_I(inode
)->i_flags
& EXT4_EXTENTS_FL
) {
1163 retval
= ext4_ext_get_blocks(handle
, inode
, block
, max_blocks
,
1166 retval
= ext4_ind_get_blocks(handle
, inode
, block
, max_blocks
,
1169 up_read((&EXT4_I(inode
)->i_data_sem
));
1171 if (retval
> 0 && buffer_mapped(bh
)) {
1172 int ret
= check_block_validity(inode
, block
,
1173 bh
->b_blocknr
, retval
);
1178 /* If it is only a block(s) look up */
1179 if ((flags
& EXT4_GET_BLOCKS_CREATE
) == 0)
1183 * Returns if the blocks have already allocated
1185 * Note that if blocks have been preallocated
1186 * ext4_ext_get_block() returns th create = 0
1187 * with buffer head unmapped.
1189 if (retval
> 0 && buffer_mapped(bh
))
1193 * When we call get_blocks without the create flag, the
1194 * BH_Unwritten flag could have gotten set if the blocks
1195 * requested were part of a uninitialized extent. We need to
1196 * clear this flag now that we are committed to convert all or
1197 * part of the uninitialized extent to be an initialized
1198 * extent. This is because we need to avoid the combination
1199 * of BH_Unwritten and BH_Mapped flags being simultaneously
1200 * set on the buffer_head.
1202 clear_buffer_unwritten(bh
);
1205 * New blocks allocate and/or writing to uninitialized extent
1206 * will possibly result in updating i_data, so we take
1207 * the write lock of i_data_sem, and call get_blocks()
1208 * with create == 1 flag.
1210 down_write((&EXT4_I(inode
)->i_data_sem
));
1213 * if the caller is from delayed allocation writeout path
1214 * we have already reserved fs blocks for allocation
1215 * let the underlying get_block() function know to
1216 * avoid double accounting
1218 if (flags
& EXT4_GET_BLOCKS_DELALLOC_RESERVE
)
1219 EXT4_I(inode
)->i_delalloc_reserved_flag
= 1;
1221 * We need to check for EXT4 here because migrate
1222 * could have changed the inode type in between
1224 if (EXT4_I(inode
)->i_flags
& EXT4_EXTENTS_FL
) {
1225 retval
= ext4_ext_get_blocks(handle
, inode
, block
, max_blocks
,
1228 retval
= ext4_ind_get_blocks(handle
, inode
, block
,
1229 max_blocks
, bh
, flags
);
1231 if (retval
> 0 && buffer_new(bh
)) {
1233 * We allocated new blocks which will result in
1234 * i_data's format changing. Force the migrate
1235 * to fail by clearing migrate flags
1237 EXT4_I(inode
)->i_flags
= EXT4_I(inode
)->i_flags
&
1242 if (flags
& EXT4_GET_BLOCKS_DELALLOC_RESERVE
)
1243 EXT4_I(inode
)->i_delalloc_reserved_flag
= 0;
1246 * Update reserved blocks/metadata blocks after successful
1247 * block allocation which had been deferred till now.
1249 if ((retval
> 0) && (flags
& EXT4_GET_BLOCKS_UPDATE_RESERVE_SPACE
))
1250 ext4_da_update_reserve_space(inode
, retval
);
1252 up_write((&EXT4_I(inode
)->i_data_sem
));
1253 if (retval
> 0 && buffer_mapped(bh
)) {
1254 int ret
= check_block_validity(inode
, block
,
1255 bh
->b_blocknr
, retval
);
1262 /* Maximum number of blocks we map for direct IO at once. */
1263 #define DIO_MAX_BLOCKS 4096
1265 int ext4_get_block(struct inode
*inode
, sector_t iblock
,
1266 struct buffer_head
*bh_result
, int create
)
1268 handle_t
*handle
= ext4_journal_current_handle();
1269 int ret
= 0, started
= 0;
1270 unsigned max_blocks
= bh_result
->b_size
>> inode
->i_blkbits
;
1273 if (create
&& !handle
) {
1274 /* Direct IO write... */
1275 if (max_blocks
> DIO_MAX_BLOCKS
)
1276 max_blocks
= DIO_MAX_BLOCKS
;
1277 dio_credits
= ext4_chunk_trans_blocks(inode
, max_blocks
);
1278 handle
= ext4_journal_start(inode
, dio_credits
);
1279 if (IS_ERR(handle
)) {
1280 ret
= PTR_ERR(handle
);
1286 ret
= ext4_get_blocks(handle
, inode
, iblock
, max_blocks
, bh_result
,
1287 create
? EXT4_GET_BLOCKS_CREATE
: 0);
1289 bh_result
->b_size
= (ret
<< inode
->i_blkbits
);
1293 ext4_journal_stop(handle
);
1299 * `handle' can be NULL if create is zero
1301 struct buffer_head
*ext4_getblk(handle_t
*handle
, struct inode
*inode
,
1302 ext4_lblk_t block
, int create
, int *errp
)
1304 struct buffer_head dummy
;
1308 J_ASSERT(handle
!= NULL
|| create
== 0);
1311 dummy
.b_blocknr
= -1000;
1312 buffer_trace_init(&dummy
.b_history
);
1314 flags
|= EXT4_GET_BLOCKS_CREATE
;
1315 err
= ext4_get_blocks(handle
, inode
, block
, 1, &dummy
, flags
);
1317 * ext4_get_blocks() returns number of blocks mapped. 0 in
1326 if (!err
&& buffer_mapped(&dummy
)) {
1327 struct buffer_head
*bh
;
1328 bh
= sb_getblk(inode
->i_sb
, dummy
.b_blocknr
);
1333 if (buffer_new(&dummy
)) {
1334 J_ASSERT(create
!= 0);
1335 J_ASSERT(handle
!= NULL
);
1338 * Now that we do not always journal data, we should
1339 * keep in mind whether this should always journal the
1340 * new buffer as metadata. For now, regular file
1341 * writes use ext4_get_block instead, so it's not a
1345 BUFFER_TRACE(bh
, "call get_create_access");
1346 fatal
= ext4_journal_get_create_access(handle
, bh
);
1347 if (!fatal
&& !buffer_uptodate(bh
)) {
1348 memset(bh
->b_data
, 0, inode
->i_sb
->s_blocksize
);
1349 set_buffer_uptodate(bh
);
1352 BUFFER_TRACE(bh
, "call ext4_handle_dirty_metadata");
1353 err
= ext4_handle_dirty_metadata(handle
, inode
, bh
);
1357 BUFFER_TRACE(bh
, "not a new buffer");
1370 struct buffer_head
*ext4_bread(handle_t
*handle
, struct inode
*inode
,
1371 ext4_lblk_t block
, int create
, int *err
)
1373 struct buffer_head
*bh
;
1375 bh
= ext4_getblk(handle
, inode
, block
, create
, err
);
1378 if (buffer_uptodate(bh
))
1380 ll_rw_block(READ_META
, 1, &bh
);
1382 if (buffer_uptodate(bh
))
1389 static int walk_page_buffers(handle_t
*handle
,
1390 struct buffer_head
*head
,
1394 int (*fn
)(handle_t
*handle
,
1395 struct buffer_head
*bh
))
1397 struct buffer_head
*bh
;
1398 unsigned block_start
, block_end
;
1399 unsigned blocksize
= head
->b_size
;
1401 struct buffer_head
*next
;
1403 for (bh
= head
, block_start
= 0;
1404 ret
== 0 && (bh
!= head
|| !block_start
);
1405 block_start
= block_end
, bh
= next
) {
1406 next
= bh
->b_this_page
;
1407 block_end
= block_start
+ blocksize
;
1408 if (block_end
<= from
|| block_start
>= to
) {
1409 if (partial
&& !buffer_uptodate(bh
))
1413 err
= (*fn
)(handle
, bh
);
1421 * To preserve ordering, it is essential that the hole instantiation and
1422 * the data write be encapsulated in a single transaction. We cannot
1423 * close off a transaction and start a new one between the ext4_get_block()
1424 * and the commit_write(). So doing the jbd2_journal_start at the start of
1425 * prepare_write() is the right place.
1427 * Also, this function can nest inside ext4_writepage() ->
1428 * block_write_full_page(). In that case, we *know* that ext4_writepage()
1429 * has generated enough buffer credits to do the whole page. So we won't
1430 * block on the journal in that case, which is good, because the caller may
1433 * By accident, ext4 can be reentered when a transaction is open via
1434 * quota file writes. If we were to commit the transaction while thus
1435 * reentered, there can be a deadlock - we would be holding a quota
1436 * lock, and the commit would never complete if another thread had a
1437 * transaction open and was blocking on the quota lock - a ranking
1440 * So what we do is to rely on the fact that jbd2_journal_stop/journal_start
1441 * will _not_ run commit under these circumstances because handle->h_ref
1442 * is elevated. We'll still have enough credits for the tiny quotafile
1445 static int do_journal_get_write_access(handle_t
*handle
,
1446 struct buffer_head
*bh
)
1448 if (!buffer_mapped(bh
) || buffer_freed(bh
))
1450 return ext4_journal_get_write_access(handle
, bh
);
1453 static int ext4_write_begin(struct file
*file
, struct address_space
*mapping
,
1454 loff_t pos
, unsigned len
, unsigned flags
,
1455 struct page
**pagep
, void **fsdata
)
1457 struct inode
*inode
= mapping
->host
;
1458 int ret
, needed_blocks
;
1465 trace_ext4_write_begin(inode
, pos
, len
, flags
);
1467 * Reserve one block more for addition to orphan list in case
1468 * we allocate blocks but write fails for some reason
1470 needed_blocks
= ext4_writepage_trans_blocks(inode
) + 1;
1471 index
= pos
>> PAGE_CACHE_SHIFT
;
1472 from
= pos
& (PAGE_CACHE_SIZE
- 1);
1476 handle
= ext4_journal_start(inode
, needed_blocks
);
1477 if (IS_ERR(handle
)) {
1478 ret
= PTR_ERR(handle
);
1482 /* We cannot recurse into the filesystem as the transaction is already
1484 flags
|= AOP_FLAG_NOFS
;
1486 page
= grab_cache_page_write_begin(mapping
, index
, flags
);
1488 ext4_journal_stop(handle
);
1494 ret
= block_write_begin(file
, mapping
, pos
, len
, flags
, pagep
, fsdata
,
1497 if (!ret
&& ext4_should_journal_data(inode
)) {
1498 ret
= walk_page_buffers(handle
, page_buffers(page
),
1499 from
, to
, NULL
, do_journal_get_write_access
);
1504 page_cache_release(page
);
1506 * block_write_begin may have instantiated a few blocks
1507 * outside i_size. Trim these off again. Don't need
1508 * i_size_read because we hold i_mutex.
1510 * Add inode to orphan list in case we crash before
1513 if (pos
+ len
> inode
->i_size
&& ext4_can_truncate(inode
))
1514 ext4_orphan_add(handle
, inode
);
1516 ext4_journal_stop(handle
);
1517 if (pos
+ len
> inode
->i_size
) {
1518 ext4_truncate(inode
);
1520 * If truncate failed early the inode might
1521 * still be on the orphan list; we need to
1522 * make sure the inode is removed from the
1523 * orphan list in that case.
1526 ext4_orphan_del(NULL
, inode
);
1530 if (ret
== -ENOSPC
&& ext4_should_retry_alloc(inode
->i_sb
, &retries
))
1536 /* For write_end() in data=journal mode */
1537 static int write_end_fn(handle_t
*handle
, struct buffer_head
*bh
)
1539 if (!buffer_mapped(bh
) || buffer_freed(bh
))
1541 set_buffer_uptodate(bh
);
1542 return ext4_handle_dirty_metadata(handle
, NULL
, bh
);
1545 static int ext4_generic_write_end(struct file
*file
,
1546 struct address_space
*mapping
,
1547 loff_t pos
, unsigned len
, unsigned copied
,
1548 struct page
*page
, void *fsdata
)
1550 int i_size_changed
= 0;
1551 struct inode
*inode
= mapping
->host
;
1552 handle_t
*handle
= ext4_journal_current_handle();
1554 copied
= block_write_end(file
, mapping
, pos
, len
, copied
, page
, fsdata
);
1557 * No need to use i_size_read() here, the i_size
1558 * cannot change under us because we hold i_mutex.
1560 * But it's important to update i_size while still holding page lock:
1561 * page writeout could otherwise come in and zero beyond i_size.
1563 if (pos
+ copied
> inode
->i_size
) {
1564 i_size_write(inode
, pos
+ copied
);
1568 if (pos
+ copied
> EXT4_I(inode
)->i_disksize
) {
1569 /* We need to mark inode dirty even if
1570 * new_i_size is less that inode->i_size
1571 * bu greater than i_disksize.(hint delalloc)
1573 ext4_update_i_disksize(inode
, (pos
+ copied
));
1577 page_cache_release(page
);
1580 * Don't mark the inode dirty under page lock. First, it unnecessarily
1581 * makes the holding time of page lock longer. Second, it forces lock
1582 * ordering of page lock and transaction start for journaling
1586 ext4_mark_inode_dirty(handle
, inode
);
1592 * We need to pick up the new inode size which generic_commit_write gave us
1593 * `file' can be NULL - eg, when called from page_symlink().
1595 * ext4 never places buffers on inode->i_mapping->private_list. metadata
1596 * buffers are managed internally.
1598 static int ext4_ordered_write_end(struct file
*file
,
1599 struct address_space
*mapping
,
1600 loff_t pos
, unsigned len
, unsigned copied
,
1601 struct page
*page
, void *fsdata
)
1603 handle_t
*handle
= ext4_journal_current_handle();
1604 struct inode
*inode
= mapping
->host
;
1607 trace_ext4_ordered_write_end(inode
, pos
, len
, copied
);
1608 ret
= ext4_jbd2_file_inode(handle
, inode
);
1611 ret2
= ext4_generic_write_end(file
, mapping
, pos
, len
, copied
,
1614 if (pos
+ len
> inode
->i_size
&& ext4_can_truncate(inode
))
1615 /* if we have allocated more blocks and copied
1616 * less. We will have blocks allocated outside
1617 * inode->i_size. So truncate them
1619 ext4_orphan_add(handle
, inode
);
1623 ret2
= ext4_journal_stop(handle
);
1627 if (pos
+ len
> inode
->i_size
) {
1628 ext4_truncate(inode
);
1630 * If truncate failed early the inode might still be
1631 * on the orphan list; we need to make sure the inode
1632 * is removed from the orphan list in that case.
1635 ext4_orphan_del(NULL
, inode
);
1639 return ret
? ret
: copied
;
1642 static int ext4_writeback_write_end(struct file
*file
,
1643 struct address_space
*mapping
,
1644 loff_t pos
, unsigned len
, unsigned copied
,
1645 struct page
*page
, void *fsdata
)
1647 handle_t
*handle
= ext4_journal_current_handle();
1648 struct inode
*inode
= mapping
->host
;
1651 trace_ext4_writeback_write_end(inode
, pos
, len
, copied
);
1652 ret2
= ext4_generic_write_end(file
, mapping
, pos
, len
, copied
,
1655 if (pos
+ len
> inode
->i_size
&& ext4_can_truncate(inode
))
1656 /* if we have allocated more blocks and copied
1657 * less. We will have blocks allocated outside
1658 * inode->i_size. So truncate them
1660 ext4_orphan_add(handle
, inode
);
1665 ret2
= ext4_journal_stop(handle
);
1669 if (pos
+ len
> inode
->i_size
) {
1670 ext4_truncate(inode
);
1672 * If truncate failed early the inode might still be
1673 * on the orphan list; we need to make sure the inode
1674 * is removed from the orphan list in that case.
1677 ext4_orphan_del(NULL
, inode
);
1680 return ret
? ret
: copied
;
1683 static int ext4_journalled_write_end(struct file
*file
,
1684 struct address_space
*mapping
,
1685 loff_t pos
, unsigned len
, unsigned copied
,
1686 struct page
*page
, void *fsdata
)
1688 handle_t
*handle
= ext4_journal_current_handle();
1689 struct inode
*inode
= mapping
->host
;
1695 trace_ext4_journalled_write_end(inode
, pos
, len
, copied
);
1696 from
= pos
& (PAGE_CACHE_SIZE
- 1);
1700 if (!PageUptodate(page
))
1702 page_zero_new_buffers(page
, from
+copied
, to
);
1705 ret
= walk_page_buffers(handle
, page_buffers(page
), from
,
1706 to
, &partial
, write_end_fn
);
1708 SetPageUptodate(page
);
1709 new_i_size
= pos
+ copied
;
1710 if (new_i_size
> inode
->i_size
)
1711 i_size_write(inode
, pos
+copied
);
1712 EXT4_I(inode
)->i_state
|= EXT4_STATE_JDATA
;
1713 if (new_i_size
> EXT4_I(inode
)->i_disksize
) {
1714 ext4_update_i_disksize(inode
, new_i_size
);
1715 ret2
= ext4_mark_inode_dirty(handle
, inode
);
1721 page_cache_release(page
);
1722 if (pos
+ len
> inode
->i_size
&& ext4_can_truncate(inode
))
1723 /* if we have allocated more blocks and copied
1724 * less. We will have blocks allocated outside
1725 * inode->i_size. So truncate them
1727 ext4_orphan_add(handle
, inode
);
1729 ret2
= ext4_journal_stop(handle
);
1732 if (pos
+ len
> inode
->i_size
) {
1733 ext4_truncate(inode
);
1735 * If truncate failed early the inode might still be
1736 * on the orphan list; we need to make sure the inode
1737 * is removed from the orphan list in that case.
1740 ext4_orphan_del(NULL
, inode
);
1743 return ret
? ret
: copied
;
1746 static int ext4_da_reserve_space(struct inode
*inode
, int nrblocks
)
1749 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
1750 unsigned long md_needed
, mdblocks
, total
= 0;
1753 * recalculate the amount of metadata blocks to reserve
1754 * in order to allocate nrblocks
1755 * worse case is one extent per block
1758 spin_lock(&EXT4_I(inode
)->i_block_reservation_lock
);
1759 total
= EXT4_I(inode
)->i_reserved_data_blocks
+ nrblocks
;
1760 mdblocks
= ext4_calc_metadata_amount(inode
, total
);
1761 BUG_ON(mdblocks
< EXT4_I(inode
)->i_reserved_meta_blocks
);
1763 md_needed
= mdblocks
- EXT4_I(inode
)->i_reserved_meta_blocks
;
1764 total
= md_needed
+ nrblocks
;
1767 * Make quota reservation here to prevent quota overflow
1768 * later. Real quota accounting is done at pages writeout
1771 if (vfs_dq_reserve_block(inode
, total
)) {
1772 spin_unlock(&EXT4_I(inode
)->i_block_reservation_lock
);
1776 if (ext4_claim_free_blocks(sbi
, total
)) {
1777 spin_unlock(&EXT4_I(inode
)->i_block_reservation_lock
);
1778 if (ext4_should_retry_alloc(inode
->i_sb
, &retries
)) {
1782 vfs_dq_release_reservation_block(inode
, total
);
1785 EXT4_I(inode
)->i_reserved_data_blocks
+= nrblocks
;
1786 EXT4_I(inode
)->i_reserved_meta_blocks
= mdblocks
;
1788 spin_unlock(&EXT4_I(inode
)->i_block_reservation_lock
);
1789 return 0; /* success */
1792 static void ext4_da_release_space(struct inode
*inode
, int to_free
)
1794 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
1795 int total
, mdb
, mdb_free
, release
;
1798 return; /* Nothing to release, exit */
1800 spin_lock(&EXT4_I(inode
)->i_block_reservation_lock
);
1802 if (!EXT4_I(inode
)->i_reserved_data_blocks
) {
1804 * if there is no reserved blocks, but we try to free some
1805 * then the counter is messed up somewhere.
1806 * but since this function is called from invalidate
1807 * page, it's harmless to return without any action
1809 printk(KERN_INFO
"ext4 delalloc try to release %d reserved "
1810 "blocks for inode %lu, but there is no reserved "
1811 "data blocks\n", to_free
, inode
->i_ino
);
1812 spin_unlock(&EXT4_I(inode
)->i_block_reservation_lock
);
1816 /* recalculate the number of metablocks still need to be reserved */
1817 total
= EXT4_I(inode
)->i_reserved_data_blocks
- to_free
;
1818 mdb
= ext4_calc_metadata_amount(inode
, total
);
1820 /* figure out how many metablocks to release */
1821 BUG_ON(mdb
> EXT4_I(inode
)->i_reserved_meta_blocks
);
1822 mdb_free
= EXT4_I(inode
)->i_reserved_meta_blocks
- mdb
;
1824 release
= to_free
+ mdb_free
;
1826 /* update fs dirty blocks counter for truncate case */
1827 percpu_counter_sub(&sbi
->s_dirtyblocks_counter
, release
);
1829 /* update per-inode reservations */
1830 BUG_ON(to_free
> EXT4_I(inode
)->i_reserved_data_blocks
);
1831 EXT4_I(inode
)->i_reserved_data_blocks
-= to_free
;
1833 BUG_ON(mdb
> EXT4_I(inode
)->i_reserved_meta_blocks
);
1834 EXT4_I(inode
)->i_reserved_meta_blocks
= mdb
;
1835 spin_unlock(&EXT4_I(inode
)->i_block_reservation_lock
);
1837 vfs_dq_release_reservation_block(inode
, release
);
1840 static void ext4_da_page_release_reservation(struct page
*page
,
1841 unsigned long offset
)
1844 struct buffer_head
*head
, *bh
;
1845 unsigned int curr_off
= 0;
1847 head
= page_buffers(page
);
1850 unsigned int next_off
= curr_off
+ bh
->b_size
;
1852 if ((offset
<= curr_off
) && (buffer_delay(bh
))) {
1854 clear_buffer_delay(bh
);
1856 curr_off
= next_off
;
1857 } while ((bh
= bh
->b_this_page
) != head
);
1858 ext4_da_release_space(page
->mapping
->host
, to_release
);
1862 * Delayed allocation stuff
1865 struct mpage_da_data
{
1866 struct inode
*inode
;
1867 sector_t b_blocknr
; /* start block number of extent */
1868 size_t b_size
; /* size of extent */
1869 unsigned long b_state
; /* state of the extent */
1870 unsigned long first_page
, next_page
; /* extent of pages */
1871 struct writeback_control
*wbc
;
1878 * mpage_da_submit_io - walks through extent of pages and try to write
1879 * them with writepage() call back
1881 * @mpd->inode: inode
1882 * @mpd->first_page: first page of the extent
1883 * @mpd->next_page: page after the last page of the extent
1885 * By the time mpage_da_submit_io() is called we expect all blocks
1886 * to be allocated. this may be wrong if allocation failed.
1888 * As pages are already locked by write_cache_pages(), we can't use it
1890 static int mpage_da_submit_io(struct mpage_da_data
*mpd
)
1893 struct pagevec pvec
;
1894 unsigned long index
, end
;
1895 int ret
= 0, err
, nr_pages
, i
;
1896 struct inode
*inode
= mpd
->inode
;
1897 struct address_space
*mapping
= inode
->i_mapping
;
1899 BUG_ON(mpd
->next_page
<= mpd
->first_page
);
1901 * We need to start from the first_page to the next_page - 1
1902 * to make sure we also write the mapped dirty buffer_heads.
1903 * If we look at mpd->b_blocknr we would only be looking
1904 * at the currently mapped buffer_heads.
1906 index
= mpd
->first_page
;
1907 end
= mpd
->next_page
- 1;
1909 pagevec_init(&pvec
, 0);
1910 while (index
<= end
) {
1911 nr_pages
= pagevec_lookup(&pvec
, mapping
, index
, PAGEVEC_SIZE
);
1914 for (i
= 0; i
< nr_pages
; i
++) {
1915 struct page
*page
= pvec
.pages
[i
];
1917 index
= page
->index
;
1922 BUG_ON(!PageLocked(page
));
1923 BUG_ON(PageWriteback(page
));
1925 pages_skipped
= mpd
->wbc
->pages_skipped
;
1926 err
= mapping
->a_ops
->writepage(page
, mpd
->wbc
);
1927 if (!err
&& (pages_skipped
== mpd
->wbc
->pages_skipped
))
1929 * have successfully written the page
1930 * without skipping the same
1932 mpd
->pages_written
++;
1934 * In error case, we have to continue because
1935 * remaining pages are still locked
1936 * XXX: unlock and re-dirty them?
1941 pagevec_release(&pvec
);
1947 * mpage_put_bnr_to_bhs - walk blocks and assign them actual numbers
1949 * @mpd->inode - inode to walk through
1950 * @exbh->b_blocknr - first block on a disk
1951 * @exbh->b_size - amount of space in bytes
1952 * @logical - first logical block to start assignment with
1954 * the function goes through all passed space and put actual disk
1955 * block numbers into buffer heads, dropping BH_Delay and BH_Unwritten
1957 static void mpage_put_bnr_to_bhs(struct mpage_da_data
*mpd
, sector_t logical
,
1958 struct buffer_head
*exbh
)
1960 struct inode
*inode
= mpd
->inode
;
1961 struct address_space
*mapping
= inode
->i_mapping
;
1962 int blocks
= exbh
->b_size
>> inode
->i_blkbits
;
1963 sector_t pblock
= exbh
->b_blocknr
, cur_logical
;
1964 struct buffer_head
*head
, *bh
;
1966 struct pagevec pvec
;
1969 index
= logical
>> (PAGE_CACHE_SHIFT
- inode
->i_blkbits
);
1970 end
= (logical
+ blocks
- 1) >> (PAGE_CACHE_SHIFT
- inode
->i_blkbits
);
1971 cur_logical
= index
<< (PAGE_CACHE_SHIFT
- inode
->i_blkbits
);
1973 pagevec_init(&pvec
, 0);
1975 while (index
<= end
) {
1976 /* XXX: optimize tail */
1977 nr_pages
= pagevec_lookup(&pvec
, mapping
, index
, PAGEVEC_SIZE
);
1980 for (i
= 0; i
< nr_pages
; i
++) {
1981 struct page
*page
= pvec
.pages
[i
];
1983 index
= page
->index
;
1988 BUG_ON(!PageLocked(page
));
1989 BUG_ON(PageWriteback(page
));
1990 BUG_ON(!page_has_buffers(page
));
1992 bh
= page_buffers(page
);
1995 /* skip blocks out of the range */
1997 if (cur_logical
>= logical
)
2000 } while ((bh
= bh
->b_this_page
) != head
);
2003 if (cur_logical
>= logical
+ blocks
)
2006 if (buffer_delay(bh
) ||
2007 buffer_unwritten(bh
)) {
2009 BUG_ON(bh
->b_bdev
!= inode
->i_sb
->s_bdev
);
2011 if (buffer_delay(bh
)) {
2012 clear_buffer_delay(bh
);
2013 bh
->b_blocknr
= pblock
;
2016 * unwritten already should have
2017 * blocknr assigned. Verify that
2019 clear_buffer_unwritten(bh
);
2020 BUG_ON(bh
->b_blocknr
!= pblock
);
2023 } else if (buffer_mapped(bh
))
2024 BUG_ON(bh
->b_blocknr
!= pblock
);
2028 } while ((bh
= bh
->b_this_page
) != head
);
2030 pagevec_release(&pvec
);
2036 * __unmap_underlying_blocks - just a helper function to unmap
2037 * set of blocks described by @bh
2039 static inline void __unmap_underlying_blocks(struct inode
*inode
,
2040 struct buffer_head
*bh
)
2042 struct block_device
*bdev
= inode
->i_sb
->s_bdev
;
2045 blocks
= bh
->b_size
>> inode
->i_blkbits
;
2046 for (i
= 0; i
< blocks
; i
++)
2047 unmap_underlying_metadata(bdev
, bh
->b_blocknr
+ i
);
2050 static void ext4_da_block_invalidatepages(struct mpage_da_data
*mpd
,
2051 sector_t logical
, long blk_cnt
)
2055 struct pagevec pvec
;
2056 struct inode
*inode
= mpd
->inode
;
2057 struct address_space
*mapping
= inode
->i_mapping
;
2059 index
= logical
>> (PAGE_CACHE_SHIFT
- inode
->i_blkbits
);
2060 end
= (logical
+ blk_cnt
- 1) >>
2061 (PAGE_CACHE_SHIFT
- inode
->i_blkbits
);
2062 while (index
<= end
) {
2063 nr_pages
= pagevec_lookup(&pvec
, mapping
, index
, PAGEVEC_SIZE
);
2066 for (i
= 0; i
< nr_pages
; i
++) {
2067 struct page
*page
= pvec
.pages
[i
];
2068 index
= page
->index
;
2073 BUG_ON(!PageLocked(page
));
2074 BUG_ON(PageWriteback(page
));
2075 block_invalidatepage(page
, 0);
2076 ClearPageUptodate(page
);
2083 static void ext4_print_free_blocks(struct inode
*inode
)
2085 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
2086 printk(KERN_EMERG
"Total free blocks count %lld\n",
2087 ext4_count_free_blocks(inode
->i_sb
));
2088 printk(KERN_EMERG
"Free/Dirty block details\n");
2089 printk(KERN_EMERG
"free_blocks=%lld\n",
2090 (long long)percpu_counter_sum(&sbi
->s_freeblocks_counter
));
2091 printk(KERN_EMERG
"dirty_blocks=%lld\n",
2092 (long long)percpu_counter_sum(&sbi
->s_dirtyblocks_counter
));
2093 printk(KERN_EMERG
"Block reservation details\n");
2094 printk(KERN_EMERG
"i_reserved_data_blocks=%u\n",
2095 EXT4_I(inode
)->i_reserved_data_blocks
);
2096 printk(KERN_EMERG
"i_reserved_meta_blocks=%u\n",
2097 EXT4_I(inode
)->i_reserved_meta_blocks
);
2102 * mpage_da_map_blocks - go through given space
2104 * @mpd - bh describing space
2106 * The function skips space we know is already mapped to disk blocks.
2109 static int mpage_da_map_blocks(struct mpage_da_data
*mpd
)
2111 int err
, blks
, get_blocks_flags
;
2112 struct buffer_head
new;
2113 sector_t next
= mpd
->b_blocknr
;
2114 unsigned max_blocks
= mpd
->b_size
>> mpd
->inode
->i_blkbits
;
2115 loff_t disksize
= EXT4_I(mpd
->inode
)->i_disksize
;
2116 handle_t
*handle
= NULL
;
2119 * We consider only non-mapped and non-allocated blocks
2121 if ((mpd
->b_state
& (1 << BH_Mapped
)) &&
2122 !(mpd
->b_state
& (1 << BH_Delay
)) &&
2123 !(mpd
->b_state
& (1 << BH_Unwritten
)))
2127 * If we didn't accumulate anything to write simply return
2132 handle
= ext4_journal_current_handle();
2136 * Call ext4_get_blocks() to allocate any delayed allocation
2137 * blocks, or to convert an uninitialized extent to be
2138 * initialized (in the case where we have written into
2139 * one or more preallocated blocks).
2141 * We pass in the magic EXT4_GET_BLOCKS_DELALLOC_RESERVE to
2142 * indicate that we are on the delayed allocation path. This
2143 * affects functions in many different parts of the allocation
2144 * call path. This flag exists primarily because we don't
2145 * want to change *many* call functions, so ext4_get_blocks()
2146 * will set the magic i_delalloc_reserved_flag once the
2147 * inode's allocation semaphore is taken.
2149 * If the blocks in questions were delalloc blocks, set
2150 * EXT4_GET_BLOCKS_DELALLOC_RESERVE so the delalloc accounting
2151 * variables are updated after the blocks have been allocated.
2154 get_blocks_flags
= (EXT4_GET_BLOCKS_CREATE
|
2155 EXT4_GET_BLOCKS_DELALLOC_RESERVE
);
2156 if (mpd
->b_state
& (1 << BH_Delay
))
2157 get_blocks_flags
|= EXT4_GET_BLOCKS_UPDATE_RESERVE_SPACE
;
2158 blks
= ext4_get_blocks(handle
, mpd
->inode
, next
, max_blocks
,
2159 &new, get_blocks_flags
);
2163 * If get block returns with error we simply
2164 * return. Later writepage will redirty the page and
2165 * writepages will find the dirty page again
2170 if (err
== -ENOSPC
&&
2171 ext4_count_free_blocks(mpd
->inode
->i_sb
)) {
2177 * get block failure will cause us to loop in
2178 * writepages, because a_ops->writepage won't be able
2179 * to make progress. The page will be redirtied by
2180 * writepage and writepages will again try to write
2183 printk(KERN_EMERG
"%s block allocation failed for inode %lu "
2184 "at logical offset %llu with max blocks "
2185 "%zd with error %d\n",
2186 __func__
, mpd
->inode
->i_ino
,
2187 (unsigned long long)next
,
2188 mpd
->b_size
>> mpd
->inode
->i_blkbits
, err
);
2189 printk(KERN_EMERG
"This should not happen.!! "
2190 "Data will be lost\n");
2191 if (err
== -ENOSPC
) {
2192 ext4_print_free_blocks(mpd
->inode
);
2194 /* invalidate all the pages */
2195 ext4_da_block_invalidatepages(mpd
, next
,
2196 mpd
->b_size
>> mpd
->inode
->i_blkbits
);
2201 new.b_size
= (blks
<< mpd
->inode
->i_blkbits
);
2203 if (buffer_new(&new))
2204 __unmap_underlying_blocks(mpd
->inode
, &new);
2207 * If blocks are delayed marked, we need to
2208 * put actual blocknr and drop delayed bit
2210 if ((mpd
->b_state
& (1 << BH_Delay
)) ||
2211 (mpd
->b_state
& (1 << BH_Unwritten
)))
2212 mpage_put_bnr_to_bhs(mpd
, next
, &new);
2214 if (ext4_should_order_data(mpd
->inode
)) {
2215 err
= ext4_jbd2_file_inode(handle
, mpd
->inode
);
2221 * Update on-disk size along with block allocation.
2223 disksize
= ((loff_t
) next
+ blks
) << mpd
->inode
->i_blkbits
;
2224 if (disksize
> i_size_read(mpd
->inode
))
2225 disksize
= i_size_read(mpd
->inode
);
2226 if (disksize
> EXT4_I(mpd
->inode
)->i_disksize
) {
2227 ext4_update_i_disksize(mpd
->inode
, disksize
);
2228 return ext4_mark_inode_dirty(handle
, mpd
->inode
);
2234 #define BH_FLAGS ((1 << BH_Uptodate) | (1 << BH_Mapped) | \
2235 (1 << BH_Delay) | (1 << BH_Unwritten))
2238 * mpage_add_bh_to_extent - try to add one more block to extent of blocks
2240 * @mpd->lbh - extent of blocks
2241 * @logical - logical number of the block in the file
2242 * @bh - bh of the block (used to access block's state)
2244 * the function is used to collect contig. blocks in same state
2246 static void mpage_add_bh_to_extent(struct mpage_da_data
*mpd
,
2247 sector_t logical
, size_t b_size
,
2248 unsigned long b_state
)
2251 int nrblocks
= mpd
->b_size
>> mpd
->inode
->i_blkbits
;
2253 /* check if thereserved journal credits might overflow */
2254 if (!(EXT4_I(mpd
->inode
)->i_flags
& EXT4_EXTENTS_FL
)) {
2255 if (nrblocks
>= EXT4_MAX_TRANS_DATA
) {
2257 * With non-extent format we are limited by the journal
2258 * credit available. Total credit needed to insert
2259 * nrblocks contiguous blocks is dependent on the
2260 * nrblocks. So limit nrblocks.
2263 } else if ((nrblocks
+ (b_size
>> mpd
->inode
->i_blkbits
)) >
2264 EXT4_MAX_TRANS_DATA
) {
2266 * Adding the new buffer_head would make it cross the
2267 * allowed limit for which we have journal credit
2268 * reserved. So limit the new bh->b_size
2270 b_size
= (EXT4_MAX_TRANS_DATA
- nrblocks
) <<
2271 mpd
->inode
->i_blkbits
;
2272 /* we will do mpage_da_submit_io in the next loop */
2276 * First block in the extent
2278 if (mpd
->b_size
== 0) {
2279 mpd
->b_blocknr
= logical
;
2280 mpd
->b_size
= b_size
;
2281 mpd
->b_state
= b_state
& BH_FLAGS
;
2285 next
= mpd
->b_blocknr
+ nrblocks
;
2287 * Can we merge the block to our big extent?
2289 if (logical
== next
&& (b_state
& BH_FLAGS
) == mpd
->b_state
) {
2290 mpd
->b_size
+= b_size
;
2296 * We couldn't merge the block to our extent, so we
2297 * need to flush current extent and start new one
2299 if (mpage_da_map_blocks(mpd
) == 0)
2300 mpage_da_submit_io(mpd
);
2305 static int ext4_bh_delay_or_unwritten(handle_t
*handle
, struct buffer_head
*bh
)
2307 return (buffer_delay(bh
) || buffer_unwritten(bh
)) && buffer_dirty(bh
);
2311 * __mpage_da_writepage - finds extent of pages and blocks
2313 * @page: page to consider
2314 * @wbc: not used, we just follow rules
2317 * The function finds extents of pages and scan them for all blocks.
2319 static int __mpage_da_writepage(struct page
*page
,
2320 struct writeback_control
*wbc
, void *data
)
2322 struct mpage_da_data
*mpd
= data
;
2323 struct inode
*inode
= mpd
->inode
;
2324 struct buffer_head
*bh
, *head
;
2329 * Rest of the page in the page_vec
2330 * redirty then and skip then. We will
2331 * try to to write them again after
2332 * starting a new transaction
2334 redirty_page_for_writepage(wbc
, page
);
2336 return MPAGE_DA_EXTENT_TAIL
;
2339 * Can we merge this page to current extent?
2341 if (mpd
->next_page
!= page
->index
) {
2343 * Nope, we can't. So, we map non-allocated blocks
2344 * and start IO on them using writepage()
2346 if (mpd
->next_page
!= mpd
->first_page
) {
2347 if (mpage_da_map_blocks(mpd
) == 0)
2348 mpage_da_submit_io(mpd
);
2350 * skip rest of the page in the page_vec
2353 redirty_page_for_writepage(wbc
, page
);
2355 return MPAGE_DA_EXTENT_TAIL
;
2359 * Start next extent of pages ...
2361 mpd
->first_page
= page
->index
;
2371 mpd
->next_page
= page
->index
+ 1;
2372 logical
= (sector_t
) page
->index
<<
2373 (PAGE_CACHE_SHIFT
- inode
->i_blkbits
);
2375 if (!page_has_buffers(page
)) {
2376 mpage_add_bh_to_extent(mpd
, logical
, PAGE_CACHE_SIZE
,
2377 (1 << BH_Dirty
) | (1 << BH_Uptodate
));
2379 return MPAGE_DA_EXTENT_TAIL
;
2382 * Page with regular buffer heads, just add all dirty ones
2384 head
= page_buffers(page
);
2387 BUG_ON(buffer_locked(bh
));
2389 * We need to try to allocate
2390 * unmapped blocks in the same page.
2391 * Otherwise we won't make progress
2392 * with the page in ext4_writepage
2394 if (ext4_bh_delay_or_unwritten(NULL
, bh
)) {
2395 mpage_add_bh_to_extent(mpd
, logical
,
2399 return MPAGE_DA_EXTENT_TAIL
;
2400 } else if (buffer_dirty(bh
) && (buffer_mapped(bh
))) {
2402 * mapped dirty buffer. We need to update
2403 * the b_state because we look at
2404 * b_state in mpage_da_map_blocks. We don't
2405 * update b_size because if we find an
2406 * unmapped buffer_head later we need to
2407 * use the b_state flag of that buffer_head.
2409 if (mpd
->b_size
== 0)
2410 mpd
->b_state
= bh
->b_state
& BH_FLAGS
;
2413 } while ((bh
= bh
->b_this_page
) != head
);
2420 * This is a special get_blocks_t callback which is used by
2421 * ext4_da_write_begin(). It will either return mapped block or
2422 * reserve space for a single block.
2424 * For delayed buffer_head we have BH_Mapped, BH_New, BH_Delay set.
2425 * We also have b_blocknr = -1 and b_bdev initialized properly
2427 * For unwritten buffer_head we have BH_Mapped, BH_New, BH_Unwritten set.
2428 * We also have b_blocknr = physicalblock mapping unwritten extent and b_bdev
2429 * initialized properly.
2431 static int ext4_da_get_block_prep(struct inode
*inode
, sector_t iblock
,
2432 struct buffer_head
*bh_result
, int create
)
2435 sector_t invalid_block
= ~((sector_t
) 0xffff);
2437 if (invalid_block
< ext4_blocks_count(EXT4_SB(inode
->i_sb
)->s_es
))
2440 BUG_ON(create
== 0);
2441 BUG_ON(bh_result
->b_size
!= inode
->i_sb
->s_blocksize
);
2444 * first, we need to know whether the block is allocated already
2445 * preallocated blocks are unmapped but should treated
2446 * the same as allocated blocks.
2448 ret
= ext4_get_blocks(NULL
, inode
, iblock
, 1, bh_result
, 0);
2449 if ((ret
== 0) && !buffer_delay(bh_result
)) {
2450 /* the block isn't (pre)allocated yet, let's reserve space */
2452 * XXX: __block_prepare_write() unmaps passed block,
2455 ret
= ext4_da_reserve_space(inode
, 1);
2457 /* not enough space to reserve */
2460 map_bh(bh_result
, inode
->i_sb
, invalid_block
);
2461 set_buffer_new(bh_result
);
2462 set_buffer_delay(bh_result
);
2463 } else if (ret
> 0) {
2464 bh_result
->b_size
= (ret
<< inode
->i_blkbits
);
2465 if (buffer_unwritten(bh_result
)) {
2466 /* A delayed write to unwritten bh should
2467 * be marked new and mapped. Mapped ensures
2468 * that we don't do get_block multiple times
2469 * when we write to the same offset and new
2470 * ensures that we do proper zero out for
2473 set_buffer_new(bh_result
);
2474 set_buffer_mapped(bh_result
);
2483 * This function is used as a standard get_block_t calback function
2484 * when there is no desire to allocate any blocks. It is used as a
2485 * callback function for block_prepare_write(), nobh_writepage(), and
2486 * block_write_full_page(). These functions should only try to map a
2487 * single block at a time.
2489 * Since this function doesn't do block allocations even if the caller
2490 * requests it by passing in create=1, it is critically important that
2491 * any caller checks to make sure that any buffer heads are returned
2492 * by this function are either all already mapped or marked for
2493 * delayed allocation before calling nobh_writepage() or
2494 * block_write_full_page(). Otherwise, b_blocknr could be left
2495 * unitialized, and the page write functions will be taken by
2498 static int noalloc_get_block_write(struct inode
*inode
, sector_t iblock
,
2499 struct buffer_head
*bh_result
, int create
)
2502 unsigned max_blocks
= bh_result
->b_size
>> inode
->i_blkbits
;
2504 BUG_ON(bh_result
->b_size
!= inode
->i_sb
->s_blocksize
);
2507 * we don't want to do block allocation in writepage
2508 * so call get_block_wrap with create = 0
2510 ret
= ext4_get_blocks(NULL
, inode
, iblock
, max_blocks
, bh_result
, 0);
2512 bh_result
->b_size
= (ret
<< inode
->i_blkbits
);
2518 static int bget_one(handle_t
*handle
, struct buffer_head
*bh
)
2524 static int bput_one(handle_t
*handle
, struct buffer_head
*bh
)
2530 static int __ext4_journalled_writepage(struct page
*page
,
2531 struct writeback_control
*wbc
,
2534 struct address_space
*mapping
= page
->mapping
;
2535 struct inode
*inode
= mapping
->host
;
2536 struct buffer_head
*page_bufs
;
2537 handle_t
*handle
= NULL
;
2541 page_bufs
= page_buffers(page
);
2543 walk_page_buffers(handle
, page_bufs
, 0, len
, NULL
, bget_one
);
2544 /* As soon as we unlock the page, it can go away, but we have
2545 * references to buffers so we are safe */
2548 handle
= ext4_journal_start(inode
, ext4_writepage_trans_blocks(inode
));
2549 if (IS_ERR(handle
)) {
2550 ret
= PTR_ERR(handle
);
2554 ret
= walk_page_buffers(handle
, page_bufs
, 0, len
, NULL
,
2555 do_journal_get_write_access
);
2557 err
= walk_page_buffers(handle
, page_bufs
, 0, len
, NULL
,
2561 err
= ext4_journal_stop(handle
);
2565 walk_page_buffers(handle
, page_bufs
, 0, len
, NULL
, bput_one
);
2566 EXT4_I(inode
)->i_state
|= EXT4_STATE_JDATA
;
2572 * Note that we don't need to start a transaction unless we're journaling data
2573 * because we should have holes filled from ext4_page_mkwrite(). We even don't
2574 * need to file the inode to the transaction's list in ordered mode because if
2575 * we are writing back data added by write(), the inode is already there and if
2576 * we are writing back data modified via mmap(), noone guarantees in which
2577 * transaction the data will hit the disk. In case we are journaling data, we
2578 * cannot start transaction directly because transaction start ranks above page
2579 * lock so we have to do some magic.
2581 * This function can get called via...
2582 * - ext4_da_writepages after taking page lock (have journal handle)
2583 * - journal_submit_inode_data_buffers (no journal handle)
2584 * - shrink_page_list via pdflush (no journal handle)
2585 * - grab_page_cache when doing write_begin (have journal handle)
2587 * We don't do any block allocation in this function. If we have page with
2588 * multiple blocks we need to write those buffer_heads that are mapped. This
2589 * is important for mmaped based write. So if we do with blocksize 1K
2590 * truncate(f, 1024);
2591 * a = mmap(f, 0, 4096);
2593 * truncate(f, 4096);
2594 * we have in the page first buffer_head mapped via page_mkwrite call back
2595 * but other bufer_heads would be unmapped but dirty(dirty done via the
2596 * do_wp_page). So writepage should write the first block. If we modify
2597 * the mmap area beyond 1024 we will again get a page_fault and the
2598 * page_mkwrite callback will do the block allocation and mark the
2599 * buffer_heads mapped.
2601 * We redirty the page if we have any buffer_heads that is either delay or
2602 * unwritten in the page.
2604 * We can get recursively called as show below.
2606 * ext4_writepage() -> kmalloc() -> __alloc_pages() -> page_launder() ->
2609 * But since we don't do any block allocation we should not deadlock.
2610 * Page also have the dirty flag cleared so we don't get recurive page_lock.
2612 static int ext4_writepage(struct page
*page
,
2613 struct writeback_control
*wbc
)
2618 struct buffer_head
*page_bufs
;
2619 struct inode
*inode
= page
->mapping
->host
;
2621 trace_ext4_writepage(inode
, page
);
2622 size
= i_size_read(inode
);
2623 if (page
->index
== size
>> PAGE_CACHE_SHIFT
)
2624 len
= size
& ~PAGE_CACHE_MASK
;
2626 len
= PAGE_CACHE_SIZE
;
2628 if (page_has_buffers(page
)) {
2629 page_bufs
= page_buffers(page
);
2630 if (walk_page_buffers(NULL
, page_bufs
, 0, len
, NULL
,
2631 ext4_bh_delay_or_unwritten
)) {
2633 * We don't want to do block allocation
2634 * So redirty the page and return
2635 * We may reach here when we do a journal commit
2636 * via journal_submit_inode_data_buffers.
2637 * If we don't have mapping block we just ignore
2638 * them. We can also reach here via shrink_page_list
2640 redirty_page_for_writepage(wbc
, page
);
2646 * The test for page_has_buffers() is subtle:
2647 * We know the page is dirty but it lost buffers. That means
2648 * that at some moment in time after write_begin()/write_end()
2649 * has been called all buffers have been clean and thus they
2650 * must have been written at least once. So they are all
2651 * mapped and we can happily proceed with mapping them
2652 * and writing the page.
2654 * Try to initialize the buffer_heads and check whether
2655 * all are mapped and non delay. We don't want to
2656 * do block allocation here.
2658 ret
= block_prepare_write(page
, 0, len
,
2659 noalloc_get_block_write
);
2661 page_bufs
= page_buffers(page
);
2662 /* check whether all are mapped and non delay */
2663 if (walk_page_buffers(NULL
, page_bufs
, 0, len
, NULL
,
2664 ext4_bh_delay_or_unwritten
)) {
2665 redirty_page_for_writepage(wbc
, page
);
2671 * We can't do block allocation here
2672 * so just redity the page and unlock
2675 redirty_page_for_writepage(wbc
, page
);
2679 /* now mark the buffer_heads as dirty and uptodate */
2680 block_commit_write(page
, 0, len
);
2683 if (PageChecked(page
) && ext4_should_journal_data(inode
)) {
2685 * It's mmapped pagecache. Add buffers and journal it. There
2686 * doesn't seem much point in redirtying the page here.
2688 ClearPageChecked(page
);
2689 return __ext4_journalled_writepage(page
, wbc
, len
);
2692 if (test_opt(inode
->i_sb
, NOBH
) && ext4_should_writeback_data(inode
))
2693 ret
= nobh_writepage(page
, noalloc_get_block_write
, wbc
);
2695 ret
= block_write_full_page(page
, noalloc_get_block_write
,
2702 * This is called via ext4_da_writepages() to
2703 * calulate the total number of credits to reserve to fit
2704 * a single extent allocation into a single transaction,
2705 * ext4_da_writpeages() will loop calling this before
2706 * the block allocation.
2709 static int ext4_da_writepages_trans_blocks(struct inode
*inode
)
2711 int max_blocks
= EXT4_I(inode
)->i_reserved_data_blocks
;
2714 * With non-extent format the journal credit needed to
2715 * insert nrblocks contiguous block is dependent on
2716 * number of contiguous block. So we will limit
2717 * number of contiguous block to a sane value
2719 if (!(inode
->i_flags
& EXT4_EXTENTS_FL
) &&
2720 (max_blocks
> EXT4_MAX_TRANS_DATA
))
2721 max_blocks
= EXT4_MAX_TRANS_DATA
;
2723 return ext4_chunk_trans_blocks(inode
, max_blocks
);
2726 static int ext4_da_writepages(struct address_space
*mapping
,
2727 struct writeback_control
*wbc
)
2730 int range_whole
= 0;
2731 handle_t
*handle
= NULL
;
2732 struct mpage_da_data mpd
;
2733 struct inode
*inode
= mapping
->host
;
2734 int no_nrwrite_index_update
;
2735 int pages_written
= 0;
2737 int range_cyclic
, cycled
= 1, io_done
= 0;
2738 int needed_blocks
, ret
= 0, nr_to_writebump
= 0;
2739 struct ext4_sb_info
*sbi
= EXT4_SB(mapping
->host
->i_sb
);
2741 trace_ext4_da_writepages(inode
, wbc
);
2744 * No pages to write? This is mainly a kludge to avoid starting
2745 * a transaction for special inodes like journal inode on last iput()
2746 * because that could violate lock ordering on umount
2748 if (!mapping
->nrpages
|| !mapping_tagged(mapping
, PAGECACHE_TAG_DIRTY
))
2752 * If the filesystem has aborted, it is read-only, so return
2753 * right away instead of dumping stack traces later on that
2754 * will obscure the real source of the problem. We test
2755 * EXT4_MF_FS_ABORTED instead of sb->s_flag's MS_RDONLY because
2756 * the latter could be true if the filesystem is mounted
2757 * read-only, and in that case, ext4_da_writepages should
2758 * *never* be called, so if that ever happens, we would want
2761 if (unlikely(sbi
->s_mount_flags
& EXT4_MF_FS_ABORTED
))
2765 * Make sure nr_to_write is >= sbi->s_mb_stream_request
2766 * This make sure small files blocks are allocated in
2767 * single attempt. This ensure that small files
2768 * get less fragmented.
2770 if (wbc
->nr_to_write
< sbi
->s_mb_stream_request
) {
2771 nr_to_writebump
= sbi
->s_mb_stream_request
- wbc
->nr_to_write
;
2772 wbc
->nr_to_write
= sbi
->s_mb_stream_request
;
2774 if (wbc
->range_start
== 0 && wbc
->range_end
== LLONG_MAX
)
2777 range_cyclic
= wbc
->range_cyclic
;
2778 if (wbc
->range_cyclic
) {
2779 index
= mapping
->writeback_index
;
2782 wbc
->range_start
= index
<< PAGE_CACHE_SHIFT
;
2783 wbc
->range_end
= LLONG_MAX
;
2784 wbc
->range_cyclic
= 0;
2786 index
= wbc
->range_start
>> PAGE_CACHE_SHIFT
;
2789 mpd
.inode
= mapping
->host
;
2792 * we don't want write_cache_pages to update
2793 * nr_to_write and writeback_index
2795 no_nrwrite_index_update
= wbc
->no_nrwrite_index_update
;
2796 wbc
->no_nrwrite_index_update
= 1;
2797 pages_skipped
= wbc
->pages_skipped
;
2800 while (!ret
&& wbc
->nr_to_write
> 0) {
2803 * we insert one extent at a time. So we need
2804 * credit needed for single extent allocation.
2805 * journalled mode is currently not supported
2808 BUG_ON(ext4_should_journal_data(inode
));
2809 needed_blocks
= ext4_da_writepages_trans_blocks(inode
);
2811 /* start a new transaction*/
2812 handle
= ext4_journal_start(inode
, needed_blocks
);
2813 if (IS_ERR(handle
)) {
2814 ret
= PTR_ERR(handle
);
2815 printk(KERN_CRIT
"%s: jbd2_start: "
2816 "%ld pages, ino %lu; err %d\n", __func__
,
2817 wbc
->nr_to_write
, inode
->i_ino
, ret
);
2819 goto out_writepages
;
2823 * Now call __mpage_da_writepage to find the next
2824 * contiguous region of logical blocks that need
2825 * blocks to be allocated by ext4. We don't actually
2826 * submit the blocks for I/O here, even though
2827 * write_cache_pages thinks it will, and will set the
2828 * pages as clean for write before calling
2829 * __mpage_da_writepage().
2837 mpd
.pages_written
= 0;
2839 ret
= write_cache_pages(mapping
, wbc
, __mpage_da_writepage
,
2842 * If we have a contigous extent of pages and we
2843 * haven't done the I/O yet, map the blocks and submit
2846 if (!mpd
.io_done
&& mpd
.next_page
!= mpd
.first_page
) {
2847 if (mpage_da_map_blocks(&mpd
) == 0)
2848 mpage_da_submit_io(&mpd
);
2850 ret
= MPAGE_DA_EXTENT_TAIL
;
2852 wbc
->nr_to_write
-= mpd
.pages_written
;
2854 ext4_journal_stop(handle
);
2856 if ((mpd
.retval
== -ENOSPC
) && sbi
->s_journal
) {
2857 /* commit the transaction which would
2858 * free blocks released in the transaction
2861 jbd2_journal_force_commit_nested(sbi
->s_journal
);
2862 wbc
->pages_skipped
= pages_skipped
;
2864 } else if (ret
== MPAGE_DA_EXTENT_TAIL
) {
2866 * got one extent now try with
2869 pages_written
+= mpd
.pages_written
;
2870 wbc
->pages_skipped
= pages_skipped
;
2873 } else if (wbc
->nr_to_write
)
2875 * There is no more writeout needed
2876 * or we requested for a noblocking writeout
2877 * and we found the device congested
2881 if (!io_done
&& !cycled
) {
2884 wbc
->range_start
= index
<< PAGE_CACHE_SHIFT
;
2885 wbc
->range_end
= mapping
->writeback_index
- 1;
2888 if (pages_skipped
!= wbc
->pages_skipped
)
2889 printk(KERN_EMERG
"This should not happen leaving %s "
2890 "with nr_to_write = %ld ret = %d\n",
2891 __func__
, wbc
->nr_to_write
, ret
);
2894 index
+= pages_written
;
2895 wbc
->range_cyclic
= range_cyclic
;
2896 if (wbc
->range_cyclic
|| (range_whole
&& wbc
->nr_to_write
> 0))
2898 * set the writeback_index so that range_cyclic
2899 * mode will write it back later
2901 mapping
->writeback_index
= index
;
2904 if (!no_nrwrite_index_update
)
2905 wbc
->no_nrwrite_index_update
= 0;
2906 wbc
->nr_to_write
-= nr_to_writebump
;
2907 trace_ext4_da_writepages_result(inode
, wbc
, ret
, pages_written
);
2911 #define FALL_BACK_TO_NONDELALLOC 1
2912 static int ext4_nonda_switch(struct super_block
*sb
)
2914 s64 free_blocks
, dirty_blocks
;
2915 struct ext4_sb_info
*sbi
= EXT4_SB(sb
);
2918 * switch to non delalloc mode if we are running low
2919 * on free block. The free block accounting via percpu
2920 * counters can get slightly wrong with percpu_counter_batch getting
2921 * accumulated on each CPU without updating global counters
2922 * Delalloc need an accurate free block accounting. So switch
2923 * to non delalloc when we are near to error range.
2925 free_blocks
= percpu_counter_read_positive(&sbi
->s_freeblocks_counter
);
2926 dirty_blocks
= percpu_counter_read_positive(&sbi
->s_dirtyblocks_counter
);
2927 if (2 * free_blocks
< 3 * dirty_blocks
||
2928 free_blocks
< (dirty_blocks
+ EXT4_FREEBLOCKS_WATERMARK
)) {
2930 * free block count is less that 150% of dirty blocks
2931 * or free blocks is less that watermark
2938 static int ext4_da_write_begin(struct file
*file
, struct address_space
*mapping
,
2939 loff_t pos
, unsigned len
, unsigned flags
,
2940 struct page
**pagep
, void **fsdata
)
2942 int ret
, retries
= 0;
2946 struct inode
*inode
= mapping
->host
;
2949 index
= pos
>> PAGE_CACHE_SHIFT
;
2950 from
= pos
& (PAGE_CACHE_SIZE
- 1);
2953 if (ext4_nonda_switch(inode
->i_sb
)) {
2954 *fsdata
= (void *)FALL_BACK_TO_NONDELALLOC
;
2955 return ext4_write_begin(file
, mapping
, pos
,
2956 len
, flags
, pagep
, fsdata
);
2958 *fsdata
= (void *)0;
2959 trace_ext4_da_write_begin(inode
, pos
, len
, flags
);
2962 * With delayed allocation, we don't log the i_disksize update
2963 * if there is delayed block allocation. But we still need
2964 * to journalling the i_disksize update if writes to the end
2965 * of file which has an already mapped buffer.
2967 handle
= ext4_journal_start(inode
, 1);
2968 if (IS_ERR(handle
)) {
2969 ret
= PTR_ERR(handle
);
2972 /* We cannot recurse into the filesystem as the transaction is already
2974 flags
|= AOP_FLAG_NOFS
;
2976 page
= grab_cache_page_write_begin(mapping
, index
, flags
);
2978 ext4_journal_stop(handle
);
2984 ret
= block_write_begin(file
, mapping
, pos
, len
, flags
, pagep
, fsdata
,
2985 ext4_da_get_block_prep
);
2988 ext4_journal_stop(handle
);
2989 page_cache_release(page
);
2991 * block_write_begin may have instantiated a few blocks
2992 * outside i_size. Trim these off again. Don't need
2993 * i_size_read because we hold i_mutex.
2995 if (pos
+ len
> inode
->i_size
)
2996 ext4_truncate(inode
);
2999 if (ret
== -ENOSPC
&& ext4_should_retry_alloc(inode
->i_sb
, &retries
))
3006 * Check if we should update i_disksize
3007 * when write to the end of file but not require block allocation
3009 static int ext4_da_should_update_i_disksize(struct page
*page
,
3010 unsigned long offset
)
3012 struct buffer_head
*bh
;
3013 struct inode
*inode
= page
->mapping
->host
;
3017 bh
= page_buffers(page
);
3018 idx
= offset
>> inode
->i_blkbits
;
3020 for (i
= 0; i
< idx
; i
++)
3021 bh
= bh
->b_this_page
;
3023 if (!buffer_mapped(bh
) || (buffer_delay(bh
)) || buffer_unwritten(bh
))
3028 static int ext4_da_write_end(struct file
*file
,
3029 struct address_space
*mapping
,
3030 loff_t pos
, unsigned len
, unsigned copied
,
3031 struct page
*page
, void *fsdata
)
3033 struct inode
*inode
= mapping
->host
;
3035 handle_t
*handle
= ext4_journal_current_handle();
3037 unsigned long start
, end
;
3038 int write_mode
= (int)(unsigned long)fsdata
;
3040 if (write_mode
== FALL_BACK_TO_NONDELALLOC
) {
3041 if (ext4_should_order_data(inode
)) {
3042 return ext4_ordered_write_end(file
, mapping
, pos
,
3043 len
, copied
, page
, fsdata
);
3044 } else if (ext4_should_writeback_data(inode
)) {
3045 return ext4_writeback_write_end(file
, mapping
, pos
,
3046 len
, copied
, page
, fsdata
);
3052 trace_ext4_da_write_end(inode
, pos
, len
, copied
);
3053 start
= pos
& (PAGE_CACHE_SIZE
- 1);
3054 end
= start
+ copied
- 1;
3057 * generic_write_end() will run mark_inode_dirty() if i_size
3058 * changes. So let's piggyback the i_disksize mark_inode_dirty
3062 new_i_size
= pos
+ copied
;
3063 if (new_i_size
> EXT4_I(inode
)->i_disksize
) {
3064 if (ext4_da_should_update_i_disksize(page
, end
)) {
3065 down_write(&EXT4_I(inode
)->i_data_sem
);
3066 if (new_i_size
> EXT4_I(inode
)->i_disksize
) {
3068 * Updating i_disksize when extending file
3069 * without needing block allocation
3071 if (ext4_should_order_data(inode
))
3072 ret
= ext4_jbd2_file_inode(handle
,
3075 EXT4_I(inode
)->i_disksize
= new_i_size
;
3077 up_write(&EXT4_I(inode
)->i_data_sem
);
3078 /* We need to mark inode dirty even if
3079 * new_i_size is less that inode->i_size
3080 * bu greater than i_disksize.(hint delalloc)
3082 ext4_mark_inode_dirty(handle
, inode
);
3085 ret2
= generic_write_end(file
, mapping
, pos
, len
, copied
,
3090 ret2
= ext4_journal_stop(handle
);
3094 return ret
? ret
: copied
;
3097 static void ext4_da_invalidatepage(struct page
*page
, unsigned long offset
)
3100 * Drop reserved blocks
3102 BUG_ON(!PageLocked(page
));
3103 if (!page_has_buffers(page
))
3106 ext4_da_page_release_reservation(page
, offset
);
3109 ext4_invalidatepage(page
, offset
);
3115 * Force all delayed allocation blocks to be allocated for a given inode.
3117 int ext4_alloc_da_blocks(struct inode
*inode
)
3119 if (!EXT4_I(inode
)->i_reserved_data_blocks
&&
3120 !EXT4_I(inode
)->i_reserved_meta_blocks
)
3124 * We do something simple for now. The filemap_flush() will
3125 * also start triggering a write of the data blocks, which is
3126 * not strictly speaking necessary (and for users of
3127 * laptop_mode, not even desirable). However, to do otherwise
3128 * would require replicating code paths in:
3130 * ext4_da_writepages() ->
3131 * write_cache_pages() ---> (via passed in callback function)
3132 * __mpage_da_writepage() -->
3133 * mpage_add_bh_to_extent()
3134 * mpage_da_map_blocks()
3136 * The problem is that write_cache_pages(), located in
3137 * mm/page-writeback.c, marks pages clean in preparation for
3138 * doing I/O, which is not desirable if we're not planning on
3141 * We could call write_cache_pages(), and then redirty all of
3142 * the pages by calling redirty_page_for_writeback() but that
3143 * would be ugly in the extreme. So instead we would need to
3144 * replicate parts of the code in the above functions,
3145 * simplifying them becuase we wouldn't actually intend to
3146 * write out the pages, but rather only collect contiguous
3147 * logical block extents, call the multi-block allocator, and
3148 * then update the buffer heads with the block allocations.
3150 * For now, though, we'll cheat by calling filemap_flush(),
3151 * which will map the blocks, and start the I/O, but not
3152 * actually wait for the I/O to complete.
3154 return filemap_flush(inode
->i_mapping
);
3158 * bmap() is special. It gets used by applications such as lilo and by
3159 * the swapper to find the on-disk block of a specific piece of data.
3161 * Naturally, this is dangerous if the block concerned is still in the
3162 * journal. If somebody makes a swapfile on an ext4 data-journaling
3163 * filesystem and enables swap, then they may get a nasty shock when the
3164 * data getting swapped to that swapfile suddenly gets overwritten by
3165 * the original zero's written out previously to the journal and
3166 * awaiting writeback in the kernel's buffer cache.
3168 * So, if we see any bmap calls here on a modified, data-journaled file,
3169 * take extra steps to flush any blocks which might be in the cache.
3171 static sector_t
ext4_bmap(struct address_space
*mapping
, sector_t block
)
3173 struct inode
*inode
= mapping
->host
;
3177 if (mapping_tagged(mapping
, PAGECACHE_TAG_DIRTY
) &&
3178 test_opt(inode
->i_sb
, DELALLOC
)) {
3180 * With delalloc we want to sync the file
3181 * so that we can make sure we allocate
3184 filemap_write_and_wait(mapping
);
3187 if (EXT4_JOURNAL(inode
) && EXT4_I(inode
)->i_state
& EXT4_STATE_JDATA
) {
3189 * This is a REALLY heavyweight approach, but the use of
3190 * bmap on dirty files is expected to be extremely rare:
3191 * only if we run lilo or swapon on a freshly made file
3192 * do we expect this to happen.
3194 * (bmap requires CAP_SYS_RAWIO so this does not
3195 * represent an unprivileged user DOS attack --- we'd be
3196 * in trouble if mortal users could trigger this path at
3199 * NB. EXT4_STATE_JDATA is not set on files other than
3200 * regular files. If somebody wants to bmap a directory
3201 * or symlink and gets confused because the buffer
3202 * hasn't yet been flushed to disk, they deserve
3203 * everything they get.
3206 EXT4_I(inode
)->i_state
&= ~EXT4_STATE_JDATA
;
3207 journal
= EXT4_JOURNAL(inode
);
3208 jbd2_journal_lock_updates(journal
);
3209 err
= jbd2_journal_flush(journal
);
3210 jbd2_journal_unlock_updates(journal
);
3216 return generic_block_bmap(mapping
, block
, ext4_get_block
);
3219 static int ext4_readpage(struct file
*file
, struct page
*page
)
3221 return mpage_readpage(page
, ext4_get_block
);
3225 ext4_readpages(struct file
*file
, struct address_space
*mapping
,
3226 struct list_head
*pages
, unsigned nr_pages
)
3228 return mpage_readpages(mapping
, pages
, nr_pages
, ext4_get_block
);
3231 static void ext4_invalidatepage(struct page
*page
, unsigned long offset
)
3233 journal_t
*journal
= EXT4_JOURNAL(page
->mapping
->host
);
3236 * If it's a full truncate we just forget about the pending dirtying
3239 ClearPageChecked(page
);
3242 jbd2_journal_invalidatepage(journal
, page
, offset
);
3244 block_invalidatepage(page
, offset
);
3247 static int ext4_releasepage(struct page
*page
, gfp_t wait
)
3249 journal_t
*journal
= EXT4_JOURNAL(page
->mapping
->host
);
3251 WARN_ON(PageChecked(page
));
3252 if (!page_has_buffers(page
))
3255 return jbd2_journal_try_to_free_buffers(journal
, page
, wait
);
3257 return try_to_free_buffers(page
);
3261 * If the O_DIRECT write will extend the file then add this inode to the
3262 * orphan list. So recovery will truncate it back to the original size
3263 * if the machine crashes during the write.
3265 * If the O_DIRECT write is intantiating holes inside i_size and the machine
3266 * crashes then stale disk data _may_ be exposed inside the file. But current
3267 * VFS code falls back into buffered path in that case so we are safe.
3269 static ssize_t
ext4_direct_IO(int rw
, struct kiocb
*iocb
,
3270 const struct iovec
*iov
, loff_t offset
,
3271 unsigned long nr_segs
)
3273 struct file
*file
= iocb
->ki_filp
;
3274 struct inode
*inode
= file
->f_mapping
->host
;
3275 struct ext4_inode_info
*ei
= EXT4_I(inode
);
3279 size_t count
= iov_length(iov
, nr_segs
);
3282 loff_t final_size
= offset
+ count
;
3284 if (final_size
> inode
->i_size
) {
3285 /* Credits for sb + inode write */
3286 handle
= ext4_journal_start(inode
, 2);
3287 if (IS_ERR(handle
)) {
3288 ret
= PTR_ERR(handle
);
3291 ret
= ext4_orphan_add(handle
, inode
);
3293 ext4_journal_stop(handle
);
3297 ei
->i_disksize
= inode
->i_size
;
3298 ext4_journal_stop(handle
);
3302 ret
= blockdev_direct_IO(rw
, iocb
, inode
, inode
->i_sb
->s_bdev
, iov
,
3304 ext4_get_block
, NULL
);
3309 /* Credits for sb + inode write */
3310 handle
= ext4_journal_start(inode
, 2);
3311 if (IS_ERR(handle
)) {
3312 /* This is really bad luck. We've written the data
3313 * but cannot extend i_size. Bail out and pretend
3314 * the write failed... */
3315 ret
= PTR_ERR(handle
);
3319 ext4_orphan_del(handle
, inode
);
3321 loff_t end
= offset
+ ret
;
3322 if (end
> inode
->i_size
) {
3323 ei
->i_disksize
= end
;
3324 i_size_write(inode
, end
);
3326 * We're going to return a positive `ret'
3327 * here due to non-zero-length I/O, so there's
3328 * no way of reporting error returns from
3329 * ext4_mark_inode_dirty() to userspace. So
3332 ext4_mark_inode_dirty(handle
, inode
);
3335 err
= ext4_journal_stop(handle
);
3344 * Pages can be marked dirty completely asynchronously from ext4's journalling
3345 * activity. By filemap_sync_pte(), try_to_unmap_one(), etc. We cannot do
3346 * much here because ->set_page_dirty is called under VFS locks. The page is
3347 * not necessarily locked.
3349 * We cannot just dirty the page and leave attached buffers clean, because the
3350 * buffers' dirty state is "definitive". We cannot just set the buffers dirty
3351 * or jbddirty because all the journalling code will explode.
3353 * So what we do is to mark the page "pending dirty" and next time writepage
3354 * is called, propagate that into the buffers appropriately.
3356 static int ext4_journalled_set_page_dirty(struct page
*page
)
3358 SetPageChecked(page
);
3359 return __set_page_dirty_nobuffers(page
);
3362 static const struct address_space_operations ext4_ordered_aops
= {
3363 .readpage
= ext4_readpage
,
3364 .readpages
= ext4_readpages
,
3365 .writepage
= ext4_writepage
,
3366 .sync_page
= block_sync_page
,
3367 .write_begin
= ext4_write_begin
,
3368 .write_end
= ext4_ordered_write_end
,
3370 .invalidatepage
= ext4_invalidatepage
,
3371 .releasepage
= ext4_releasepage
,
3372 .direct_IO
= ext4_direct_IO
,
3373 .migratepage
= buffer_migrate_page
,
3374 .is_partially_uptodate
= block_is_partially_uptodate
,
3377 static const struct address_space_operations ext4_writeback_aops
= {
3378 .readpage
= ext4_readpage
,
3379 .readpages
= ext4_readpages
,
3380 .writepage
= ext4_writepage
,
3381 .sync_page
= block_sync_page
,
3382 .write_begin
= ext4_write_begin
,
3383 .write_end
= ext4_writeback_write_end
,
3385 .invalidatepage
= ext4_invalidatepage
,
3386 .releasepage
= ext4_releasepage
,
3387 .direct_IO
= ext4_direct_IO
,
3388 .migratepage
= buffer_migrate_page
,
3389 .is_partially_uptodate
= block_is_partially_uptodate
,
3392 static const struct address_space_operations ext4_journalled_aops
= {
3393 .readpage
= ext4_readpage
,
3394 .readpages
= ext4_readpages
,
3395 .writepage
= ext4_writepage
,
3396 .sync_page
= block_sync_page
,
3397 .write_begin
= ext4_write_begin
,
3398 .write_end
= ext4_journalled_write_end
,
3399 .set_page_dirty
= ext4_journalled_set_page_dirty
,
3401 .invalidatepage
= ext4_invalidatepage
,
3402 .releasepage
= ext4_releasepage
,
3403 .is_partially_uptodate
= block_is_partially_uptodate
,
3406 static const struct address_space_operations ext4_da_aops
= {
3407 .readpage
= ext4_readpage
,
3408 .readpages
= ext4_readpages
,
3409 .writepage
= ext4_writepage
,
3410 .writepages
= ext4_da_writepages
,
3411 .sync_page
= block_sync_page
,
3412 .write_begin
= ext4_da_write_begin
,
3413 .write_end
= ext4_da_write_end
,
3415 .invalidatepage
= ext4_da_invalidatepage
,
3416 .releasepage
= ext4_releasepage
,
3417 .direct_IO
= ext4_direct_IO
,
3418 .migratepage
= buffer_migrate_page
,
3419 .is_partially_uptodate
= block_is_partially_uptodate
,
3422 void ext4_set_aops(struct inode
*inode
)
3424 if (ext4_should_order_data(inode
) &&
3425 test_opt(inode
->i_sb
, DELALLOC
))
3426 inode
->i_mapping
->a_ops
= &ext4_da_aops
;
3427 else if (ext4_should_order_data(inode
))
3428 inode
->i_mapping
->a_ops
= &ext4_ordered_aops
;
3429 else if (ext4_should_writeback_data(inode
) &&
3430 test_opt(inode
->i_sb
, DELALLOC
))
3431 inode
->i_mapping
->a_ops
= &ext4_da_aops
;
3432 else if (ext4_should_writeback_data(inode
))
3433 inode
->i_mapping
->a_ops
= &ext4_writeback_aops
;
3435 inode
->i_mapping
->a_ops
= &ext4_journalled_aops
;
3439 * ext4_block_truncate_page() zeroes out a mapping from file offset `from'
3440 * up to the end of the block which corresponds to `from'.
3441 * This required during truncate. We need to physically zero the tail end
3442 * of that block so it doesn't yield old data if the file is later grown.
3444 int ext4_block_truncate_page(handle_t
*handle
,
3445 struct address_space
*mapping
, loff_t from
)
3447 ext4_fsblk_t index
= from
>> PAGE_CACHE_SHIFT
;
3448 unsigned offset
= from
& (PAGE_CACHE_SIZE
-1);
3449 unsigned blocksize
, length
, pos
;
3451 struct inode
*inode
= mapping
->host
;
3452 struct buffer_head
*bh
;
3456 page
= find_or_create_page(mapping
, from
>> PAGE_CACHE_SHIFT
,
3457 mapping_gfp_mask(mapping
) & ~__GFP_FS
);
3461 blocksize
= inode
->i_sb
->s_blocksize
;
3462 length
= blocksize
- (offset
& (blocksize
- 1));
3463 iblock
= index
<< (PAGE_CACHE_SHIFT
- inode
->i_sb
->s_blocksize_bits
);
3466 * For "nobh" option, we can only work if we don't need to
3467 * read-in the page - otherwise we create buffers to do the IO.
3469 if (!page_has_buffers(page
) && test_opt(inode
->i_sb
, NOBH
) &&
3470 ext4_should_writeback_data(inode
) && PageUptodate(page
)) {
3471 zero_user(page
, offset
, length
);
3472 set_page_dirty(page
);
3476 if (!page_has_buffers(page
))
3477 create_empty_buffers(page
, blocksize
, 0);
3479 /* Find the buffer that contains "offset" */
3480 bh
= page_buffers(page
);
3482 while (offset
>= pos
) {
3483 bh
= bh
->b_this_page
;
3489 if (buffer_freed(bh
)) {
3490 BUFFER_TRACE(bh
, "freed: skip");
3494 if (!buffer_mapped(bh
)) {
3495 BUFFER_TRACE(bh
, "unmapped");
3496 ext4_get_block(inode
, iblock
, bh
, 0);
3497 /* unmapped? It's a hole - nothing to do */
3498 if (!buffer_mapped(bh
)) {
3499 BUFFER_TRACE(bh
, "still unmapped");
3504 /* Ok, it's mapped. Make sure it's up-to-date */
3505 if (PageUptodate(page
))
3506 set_buffer_uptodate(bh
);
3508 if (!buffer_uptodate(bh
)) {
3510 ll_rw_block(READ
, 1, &bh
);
3512 /* Uhhuh. Read error. Complain and punt. */
3513 if (!buffer_uptodate(bh
))
3517 if (ext4_should_journal_data(inode
)) {
3518 BUFFER_TRACE(bh
, "get write access");
3519 err
= ext4_journal_get_write_access(handle
, bh
);
3524 zero_user(page
, offset
, length
);
3526 BUFFER_TRACE(bh
, "zeroed end of block");
3529 if (ext4_should_journal_data(inode
)) {
3530 err
= ext4_handle_dirty_metadata(handle
, inode
, bh
);
3532 if (ext4_should_order_data(inode
))
3533 err
= ext4_jbd2_file_inode(handle
, inode
);
3534 mark_buffer_dirty(bh
);
3539 page_cache_release(page
);
3544 * Probably it should be a library function... search for first non-zero word
3545 * or memcmp with zero_page, whatever is better for particular architecture.
3548 static inline int all_zeroes(__le32
*p
, __le32
*q
)
3557 * ext4_find_shared - find the indirect blocks for partial truncation.
3558 * @inode: inode in question
3559 * @depth: depth of the affected branch
3560 * @offsets: offsets of pointers in that branch (see ext4_block_to_path)
3561 * @chain: place to store the pointers to partial indirect blocks
3562 * @top: place to the (detached) top of branch
3564 * This is a helper function used by ext4_truncate().
3566 * When we do truncate() we may have to clean the ends of several
3567 * indirect blocks but leave the blocks themselves alive. Block is
3568 * partially truncated if some data below the new i_size is refered
3569 * from it (and it is on the path to the first completely truncated
3570 * data block, indeed). We have to free the top of that path along
3571 * with everything to the right of the path. Since no allocation
3572 * past the truncation point is possible until ext4_truncate()
3573 * finishes, we may safely do the latter, but top of branch may
3574 * require special attention - pageout below the truncation point
3575 * might try to populate it.
3577 * We atomically detach the top of branch from the tree, store the
3578 * block number of its root in *@top, pointers to buffer_heads of
3579 * partially truncated blocks - in @chain[].bh and pointers to
3580 * their last elements that should not be removed - in
3581 * @chain[].p. Return value is the pointer to last filled element
3584 * The work left to caller to do the actual freeing of subtrees:
3585 * a) free the subtree starting from *@top
3586 * b) free the subtrees whose roots are stored in
3587 * (@chain[i].p+1 .. end of @chain[i].bh->b_data)
3588 * c) free the subtrees growing from the inode past the @chain[0].
3589 * (no partially truncated stuff there). */
3591 static Indirect
*ext4_find_shared(struct inode
*inode
, int depth
,
3592 ext4_lblk_t offsets
[4], Indirect chain
[4],
3595 Indirect
*partial
, *p
;
3599 /* Make k index the deepest non-null offest + 1 */
3600 for (k
= depth
; k
> 1 && !offsets
[k
-1]; k
--)
3602 partial
= ext4_get_branch(inode
, k
, offsets
, chain
, &err
);
3603 /* Writer: pointers */
3605 partial
= chain
+ k
-1;
3607 * If the branch acquired continuation since we've looked at it -
3608 * fine, it should all survive and (new) top doesn't belong to us.
3610 if (!partial
->key
&& *partial
->p
)
3613 for (p
= partial
; (p
> chain
) && all_zeroes((__le32
*) p
->bh
->b_data
, p
->p
); p
--)
3616 * OK, we've found the last block that must survive. The rest of our
3617 * branch should be detached before unlocking. However, if that rest
3618 * of branch is all ours and does not grow immediately from the inode
3619 * it's easier to cheat and just decrement partial->p.
3621 if (p
== chain
+ k
- 1 && p
> chain
) {
3625 /* Nope, don't do this in ext4. Must leave the tree intact */
3632 while (partial
> p
) {
3633 brelse(partial
->bh
);
3641 * Zero a number of block pointers in either an inode or an indirect block.
3642 * If we restart the transaction we must again get write access to the
3643 * indirect block for further modification.
3645 * We release `count' blocks on disk, but (last - first) may be greater
3646 * than `count' because there can be holes in there.
3648 static void ext4_clear_blocks(handle_t
*handle
, struct inode
*inode
,
3649 struct buffer_head
*bh
,
3650 ext4_fsblk_t block_to_free
,
3651 unsigned long count
, __le32
*first
,
3655 if (try_to_extend_transaction(handle
, inode
)) {
3657 BUFFER_TRACE(bh
, "call ext4_handle_dirty_metadata");
3658 ext4_handle_dirty_metadata(handle
, inode
, bh
);
3660 ext4_mark_inode_dirty(handle
, inode
);
3661 ext4_journal_test_restart(handle
, inode
);
3663 BUFFER_TRACE(bh
, "retaking write access");
3664 ext4_journal_get_write_access(handle
, bh
);
3669 * Any buffers which are on the journal will be in memory. We
3670 * find them on the hash table so jbd2_journal_revoke() will
3671 * run jbd2_journal_forget() on them. We've already detached
3672 * each block from the file, so bforget() in
3673 * jbd2_journal_forget() should be safe.
3675 * AKPM: turn on bforget in jbd2_journal_forget()!!!
3677 for (p
= first
; p
< last
; p
++) {
3678 u32 nr
= le32_to_cpu(*p
);
3680 struct buffer_head
*tbh
;
3683 tbh
= sb_find_get_block(inode
->i_sb
, nr
);
3684 ext4_forget(handle
, 0, inode
, tbh
, nr
);
3688 ext4_free_blocks(handle
, inode
, block_to_free
, count
, 0);
3692 * ext4_free_data - free a list of data blocks
3693 * @handle: handle for this transaction
3694 * @inode: inode we are dealing with
3695 * @this_bh: indirect buffer_head which contains *@first and *@last
3696 * @first: array of block numbers
3697 * @last: points immediately past the end of array
3699 * We are freeing all blocks refered from that array (numbers are stored as
3700 * little-endian 32-bit) and updating @inode->i_blocks appropriately.
3702 * We accumulate contiguous runs of blocks to free. Conveniently, if these
3703 * blocks are contiguous then releasing them at one time will only affect one
3704 * or two bitmap blocks (+ group descriptor(s) and superblock) and we won't
3705 * actually use a lot of journal space.
3707 * @this_bh will be %NULL if @first and @last point into the inode's direct
3710 static void ext4_free_data(handle_t
*handle
, struct inode
*inode
,
3711 struct buffer_head
*this_bh
,
3712 __le32
*first
, __le32
*last
)
3714 ext4_fsblk_t block_to_free
= 0; /* Starting block # of a run */
3715 unsigned long count
= 0; /* Number of blocks in the run */
3716 __le32
*block_to_free_p
= NULL
; /* Pointer into inode/ind
3719 ext4_fsblk_t nr
; /* Current block # */
3720 __le32
*p
; /* Pointer into inode/ind
3721 for current block */
3724 if (this_bh
) { /* For indirect block */
3725 BUFFER_TRACE(this_bh
, "get_write_access");
3726 err
= ext4_journal_get_write_access(handle
, this_bh
);
3727 /* Important: if we can't update the indirect pointers
3728 * to the blocks, we can't free them. */
3733 for (p
= first
; p
< last
; p
++) {
3734 nr
= le32_to_cpu(*p
);
3736 /* accumulate blocks to free if they're contiguous */
3739 block_to_free_p
= p
;
3741 } else if (nr
== block_to_free
+ count
) {
3744 ext4_clear_blocks(handle
, inode
, this_bh
,
3746 count
, block_to_free_p
, p
);
3748 block_to_free_p
= p
;
3755 ext4_clear_blocks(handle
, inode
, this_bh
, block_to_free
,
3756 count
, block_to_free_p
, p
);
3759 BUFFER_TRACE(this_bh
, "call ext4_handle_dirty_metadata");
3762 * The buffer head should have an attached journal head at this
3763 * point. However, if the data is corrupted and an indirect
3764 * block pointed to itself, it would have been detached when
3765 * the block was cleared. Check for this instead of OOPSing.
3767 if ((EXT4_JOURNAL(inode
) == NULL
) || bh2jh(this_bh
))
3768 ext4_handle_dirty_metadata(handle
, inode
, this_bh
);
3770 ext4_error(inode
->i_sb
, __func__
,
3771 "circular indirect block detected, "
3772 "inode=%lu, block=%llu",
3774 (unsigned long long) this_bh
->b_blocknr
);
3779 * ext4_free_branches - free an array of branches
3780 * @handle: JBD handle for this transaction
3781 * @inode: inode we are dealing with
3782 * @parent_bh: the buffer_head which contains *@first and *@last
3783 * @first: array of block numbers
3784 * @last: pointer immediately past the end of array
3785 * @depth: depth of the branches to free
3787 * We are freeing all blocks refered from these branches (numbers are
3788 * stored as little-endian 32-bit) and updating @inode->i_blocks
3791 static void ext4_free_branches(handle_t
*handle
, struct inode
*inode
,
3792 struct buffer_head
*parent_bh
,
3793 __le32
*first
, __le32
*last
, int depth
)
3798 if (ext4_handle_is_aborted(handle
))
3802 struct buffer_head
*bh
;
3803 int addr_per_block
= EXT4_ADDR_PER_BLOCK(inode
->i_sb
);
3805 while (--p
>= first
) {
3806 nr
= le32_to_cpu(*p
);
3808 continue; /* A hole */
3810 /* Go read the buffer for the next level down */
3811 bh
= sb_bread(inode
->i_sb
, nr
);
3814 * A read failure? Report error and clear slot
3818 ext4_error(inode
->i_sb
, "ext4_free_branches",
3819 "Read failure, inode=%lu, block=%llu",
3824 /* This zaps the entire block. Bottom up. */
3825 BUFFER_TRACE(bh
, "free child branches");
3826 ext4_free_branches(handle
, inode
, bh
,
3827 (__le32
*) bh
->b_data
,
3828 (__le32
*) bh
->b_data
+ addr_per_block
,
3832 * We've probably journalled the indirect block several
3833 * times during the truncate. But it's no longer
3834 * needed and we now drop it from the transaction via
3835 * jbd2_journal_revoke().
3837 * That's easy if it's exclusively part of this
3838 * transaction. But if it's part of the committing
3839 * transaction then jbd2_journal_forget() will simply
3840 * brelse() it. That means that if the underlying
3841 * block is reallocated in ext4_get_block(),
3842 * unmap_underlying_metadata() will find this block
3843 * and will try to get rid of it. damn, damn.
3845 * If this block has already been committed to the
3846 * journal, a revoke record will be written. And
3847 * revoke records must be emitted *before* clearing
3848 * this block's bit in the bitmaps.
3850 ext4_forget(handle
, 1, inode
, bh
, bh
->b_blocknr
);
3853 * Everything below this this pointer has been
3854 * released. Now let this top-of-subtree go.
3856 * We want the freeing of this indirect block to be
3857 * atomic in the journal with the updating of the
3858 * bitmap block which owns it. So make some room in
3861 * We zero the parent pointer *after* freeing its
3862 * pointee in the bitmaps, so if extend_transaction()
3863 * for some reason fails to put the bitmap changes and
3864 * the release into the same transaction, recovery
3865 * will merely complain about releasing a free block,
3866 * rather than leaking blocks.
3868 if (ext4_handle_is_aborted(handle
))
3870 if (try_to_extend_transaction(handle
, inode
)) {
3871 ext4_mark_inode_dirty(handle
, inode
);
3872 ext4_journal_test_restart(handle
, inode
);
3875 ext4_free_blocks(handle
, inode
, nr
, 1, 1);
3879 * The block which we have just freed is
3880 * pointed to by an indirect block: journal it
3882 BUFFER_TRACE(parent_bh
, "get_write_access");
3883 if (!ext4_journal_get_write_access(handle
,
3886 BUFFER_TRACE(parent_bh
,
3887 "call ext4_handle_dirty_metadata");
3888 ext4_handle_dirty_metadata(handle
,
3895 /* We have reached the bottom of the tree. */
3896 BUFFER_TRACE(parent_bh
, "free data blocks");
3897 ext4_free_data(handle
, inode
, parent_bh
, first
, last
);
3901 int ext4_can_truncate(struct inode
*inode
)
3903 if (IS_APPEND(inode
) || IS_IMMUTABLE(inode
))
3905 if (S_ISREG(inode
->i_mode
))
3907 if (S_ISDIR(inode
->i_mode
))
3909 if (S_ISLNK(inode
->i_mode
))
3910 return !ext4_inode_is_fast_symlink(inode
);
3917 * We block out ext4_get_block() block instantiations across the entire
3918 * transaction, and VFS/VM ensures that ext4_truncate() cannot run
3919 * simultaneously on behalf of the same inode.
3921 * As we work through the truncate and commmit bits of it to the journal there
3922 * is one core, guiding principle: the file's tree must always be consistent on
3923 * disk. We must be able to restart the truncate after a crash.
3925 * The file's tree may be transiently inconsistent in memory (although it
3926 * probably isn't), but whenever we close off and commit a journal transaction,
3927 * the contents of (the filesystem + the journal) must be consistent and
3928 * restartable. It's pretty simple, really: bottom up, right to left (although
3929 * left-to-right works OK too).
3931 * Note that at recovery time, journal replay occurs *before* the restart of
3932 * truncate against the orphan inode list.
3934 * The committed inode has the new, desired i_size (which is the same as
3935 * i_disksize in this case). After a crash, ext4_orphan_cleanup() will see
3936 * that this inode's truncate did not complete and it will again call
3937 * ext4_truncate() to have another go. So there will be instantiated blocks
3938 * to the right of the truncation point in a crashed ext4 filesystem. But
3939 * that's fine - as long as they are linked from the inode, the post-crash
3940 * ext4_truncate() run will find them and release them.
3942 void ext4_truncate(struct inode
*inode
)
3945 struct ext4_inode_info
*ei
= EXT4_I(inode
);
3946 __le32
*i_data
= ei
->i_data
;
3947 int addr_per_block
= EXT4_ADDR_PER_BLOCK(inode
->i_sb
);
3948 struct address_space
*mapping
= inode
->i_mapping
;
3949 ext4_lblk_t offsets
[4];
3954 ext4_lblk_t last_block
;
3955 unsigned blocksize
= inode
->i_sb
->s_blocksize
;
3957 if (!ext4_can_truncate(inode
))
3960 if (ei
->i_disksize
&& inode
->i_size
== 0 &&
3961 !test_opt(inode
->i_sb
, NO_AUTO_DA_ALLOC
))
3962 ei
->i_state
|= EXT4_STATE_DA_ALLOC_CLOSE
;
3964 if (EXT4_I(inode
)->i_flags
& EXT4_EXTENTS_FL
) {
3965 ext4_ext_truncate(inode
);
3969 handle
= start_transaction(inode
);
3971 return; /* AKPM: return what? */
3973 last_block
= (inode
->i_size
+ blocksize
-1)
3974 >> EXT4_BLOCK_SIZE_BITS(inode
->i_sb
);
3976 if (inode
->i_size
& (blocksize
- 1))
3977 if (ext4_block_truncate_page(handle
, mapping
, inode
->i_size
))
3980 n
= ext4_block_to_path(inode
, last_block
, offsets
, NULL
);
3982 goto out_stop
; /* error */
3985 * OK. This truncate is going to happen. We add the inode to the
3986 * orphan list, so that if this truncate spans multiple transactions,
3987 * and we crash, we will resume the truncate when the filesystem
3988 * recovers. It also marks the inode dirty, to catch the new size.
3990 * Implication: the file must always be in a sane, consistent
3991 * truncatable state while each transaction commits.
3993 if (ext4_orphan_add(handle
, inode
))
3997 * From here we block out all ext4_get_block() callers who want to
3998 * modify the block allocation tree.
4000 down_write(&ei
->i_data_sem
);
4002 ext4_discard_preallocations(inode
);
4005 * The orphan list entry will now protect us from any crash which
4006 * occurs before the truncate completes, so it is now safe to propagate
4007 * the new, shorter inode size (held for now in i_size) into the
4008 * on-disk inode. We do this via i_disksize, which is the value which
4009 * ext4 *really* writes onto the disk inode.
4011 ei
->i_disksize
= inode
->i_size
;
4013 if (n
== 1) { /* direct blocks */
4014 ext4_free_data(handle
, inode
, NULL
, i_data
+offsets
[0],
4015 i_data
+ EXT4_NDIR_BLOCKS
);
4019 partial
= ext4_find_shared(inode
, n
, offsets
, chain
, &nr
);
4020 /* Kill the top of shared branch (not detached) */
4022 if (partial
== chain
) {
4023 /* Shared branch grows from the inode */
4024 ext4_free_branches(handle
, inode
, NULL
,
4025 &nr
, &nr
+1, (chain
+n
-1) - partial
);
4028 * We mark the inode dirty prior to restart,
4029 * and prior to stop. No need for it here.
4032 /* Shared branch grows from an indirect block */
4033 BUFFER_TRACE(partial
->bh
, "get_write_access");
4034 ext4_free_branches(handle
, inode
, partial
->bh
,
4036 partial
->p
+1, (chain
+n
-1) - partial
);
4039 /* Clear the ends of indirect blocks on the shared branch */
4040 while (partial
> chain
) {
4041 ext4_free_branches(handle
, inode
, partial
->bh
, partial
->p
+ 1,
4042 (__le32
*)partial
->bh
->b_data
+addr_per_block
,
4043 (chain
+n
-1) - partial
);
4044 BUFFER_TRACE(partial
->bh
, "call brelse");
4045 brelse(partial
->bh
);
4049 /* Kill the remaining (whole) subtrees */
4050 switch (offsets
[0]) {
4052 nr
= i_data
[EXT4_IND_BLOCK
];
4054 ext4_free_branches(handle
, inode
, NULL
, &nr
, &nr
+1, 1);
4055 i_data
[EXT4_IND_BLOCK
] = 0;
4057 case EXT4_IND_BLOCK
:
4058 nr
= i_data
[EXT4_DIND_BLOCK
];
4060 ext4_free_branches(handle
, inode
, NULL
, &nr
, &nr
+1, 2);
4061 i_data
[EXT4_DIND_BLOCK
] = 0;
4063 case EXT4_DIND_BLOCK
:
4064 nr
= i_data
[EXT4_TIND_BLOCK
];
4066 ext4_free_branches(handle
, inode
, NULL
, &nr
, &nr
+1, 3);
4067 i_data
[EXT4_TIND_BLOCK
] = 0;
4069 case EXT4_TIND_BLOCK
:
4073 up_write(&ei
->i_data_sem
);
4074 inode
->i_mtime
= inode
->i_ctime
= ext4_current_time(inode
);
4075 ext4_mark_inode_dirty(handle
, inode
);
4078 * In a multi-transaction truncate, we only make the final transaction
4082 ext4_handle_sync(handle
);
4085 * If this was a simple ftruncate(), and the file will remain alive
4086 * then we need to clear up the orphan record which we created above.
4087 * However, if this was a real unlink then we were called by
4088 * ext4_delete_inode(), and we allow that function to clean up the
4089 * orphan info for us.
4092 ext4_orphan_del(handle
, inode
);
4094 ext4_journal_stop(handle
);
4098 * ext4_get_inode_loc returns with an extra refcount against the inode's
4099 * underlying buffer_head on success. If 'in_mem' is true, we have all
4100 * data in memory that is needed to recreate the on-disk version of this
4103 static int __ext4_get_inode_loc(struct inode
*inode
,
4104 struct ext4_iloc
*iloc
, int in_mem
)
4106 struct ext4_group_desc
*gdp
;
4107 struct buffer_head
*bh
;
4108 struct super_block
*sb
= inode
->i_sb
;
4110 int inodes_per_block
, inode_offset
;
4113 if (!ext4_valid_inum(sb
, inode
->i_ino
))
4116 iloc
->block_group
= (inode
->i_ino
- 1) / EXT4_INODES_PER_GROUP(sb
);
4117 gdp
= ext4_get_group_desc(sb
, iloc
->block_group
, NULL
);
4122 * Figure out the offset within the block group inode table
4124 inodes_per_block
= (EXT4_BLOCK_SIZE(sb
) / EXT4_INODE_SIZE(sb
));
4125 inode_offset
= ((inode
->i_ino
- 1) %
4126 EXT4_INODES_PER_GROUP(sb
));
4127 block
= ext4_inode_table(sb
, gdp
) + (inode_offset
/ inodes_per_block
);
4128 iloc
->offset
= (inode_offset
% inodes_per_block
) * EXT4_INODE_SIZE(sb
);
4130 bh
= sb_getblk(sb
, block
);
4132 ext4_error(sb
, "ext4_get_inode_loc", "unable to read "
4133 "inode block - inode=%lu, block=%llu",
4134 inode
->i_ino
, block
);
4137 if (!buffer_uptodate(bh
)) {
4141 * If the buffer has the write error flag, we have failed
4142 * to write out another inode in the same block. In this
4143 * case, we don't have to read the block because we may
4144 * read the old inode data successfully.
4146 if (buffer_write_io_error(bh
) && !buffer_uptodate(bh
))
4147 set_buffer_uptodate(bh
);
4149 if (buffer_uptodate(bh
)) {
4150 /* someone brought it uptodate while we waited */
4156 * If we have all information of the inode in memory and this
4157 * is the only valid inode in the block, we need not read the
4161 struct buffer_head
*bitmap_bh
;
4164 start
= inode_offset
& ~(inodes_per_block
- 1);
4166 /* Is the inode bitmap in cache? */
4167 bitmap_bh
= sb_getblk(sb
, ext4_inode_bitmap(sb
, gdp
));
4172 * If the inode bitmap isn't in cache then the
4173 * optimisation may end up performing two reads instead
4174 * of one, so skip it.
4176 if (!buffer_uptodate(bitmap_bh
)) {
4180 for (i
= start
; i
< start
+ inodes_per_block
; i
++) {
4181 if (i
== inode_offset
)
4183 if (ext4_test_bit(i
, bitmap_bh
->b_data
))
4187 if (i
== start
+ inodes_per_block
) {
4188 /* all other inodes are free, so skip I/O */
4189 memset(bh
->b_data
, 0, bh
->b_size
);
4190 set_buffer_uptodate(bh
);
4198 * If we need to do any I/O, try to pre-readahead extra
4199 * blocks from the inode table.
4201 if (EXT4_SB(sb
)->s_inode_readahead_blks
) {
4202 ext4_fsblk_t b
, end
, table
;
4205 table
= ext4_inode_table(sb
, gdp
);
4206 /* s_inode_readahead_blks is always a power of 2 */
4207 b
= block
& ~(EXT4_SB(sb
)->s_inode_readahead_blks
-1);
4210 end
= b
+ EXT4_SB(sb
)->s_inode_readahead_blks
;
4211 num
= EXT4_INODES_PER_GROUP(sb
);
4212 if (EXT4_HAS_RO_COMPAT_FEATURE(sb
,
4213 EXT4_FEATURE_RO_COMPAT_GDT_CSUM
))
4214 num
-= ext4_itable_unused_count(sb
, gdp
);
4215 table
+= num
/ inodes_per_block
;
4219 sb_breadahead(sb
, b
++);
4223 * There are other valid inodes in the buffer, this inode
4224 * has in-inode xattrs, or we don't have this inode in memory.
4225 * Read the block from disk.
4228 bh
->b_end_io
= end_buffer_read_sync
;
4229 submit_bh(READ_META
, bh
);
4231 if (!buffer_uptodate(bh
)) {
4232 ext4_error(sb
, __func__
,
4233 "unable to read inode block - inode=%lu, "
4234 "block=%llu", inode
->i_ino
, block
);
4244 int ext4_get_inode_loc(struct inode
*inode
, struct ext4_iloc
*iloc
)
4246 /* We have all inode data except xattrs in memory here. */
4247 return __ext4_get_inode_loc(inode
, iloc
,
4248 !(EXT4_I(inode
)->i_state
& EXT4_STATE_XATTR
));
4251 void ext4_set_inode_flags(struct inode
*inode
)
4253 unsigned int flags
= EXT4_I(inode
)->i_flags
;
4255 inode
->i_flags
&= ~(S_SYNC
|S_APPEND
|S_IMMUTABLE
|S_NOATIME
|S_DIRSYNC
);
4256 if (flags
& EXT4_SYNC_FL
)
4257 inode
->i_flags
|= S_SYNC
;
4258 if (flags
& EXT4_APPEND_FL
)
4259 inode
->i_flags
|= S_APPEND
;
4260 if (flags
& EXT4_IMMUTABLE_FL
)
4261 inode
->i_flags
|= S_IMMUTABLE
;
4262 if (flags
& EXT4_NOATIME_FL
)
4263 inode
->i_flags
|= S_NOATIME
;
4264 if (flags
& EXT4_DIRSYNC_FL
)
4265 inode
->i_flags
|= S_DIRSYNC
;
4268 /* Propagate flags from i_flags to EXT4_I(inode)->i_flags */
4269 void ext4_get_inode_flags(struct ext4_inode_info
*ei
)
4271 unsigned int flags
= ei
->vfs_inode
.i_flags
;
4273 ei
->i_flags
&= ~(EXT4_SYNC_FL
|EXT4_APPEND_FL
|
4274 EXT4_IMMUTABLE_FL
|EXT4_NOATIME_FL
|EXT4_DIRSYNC_FL
);
4276 ei
->i_flags
|= EXT4_SYNC_FL
;
4277 if (flags
& S_APPEND
)
4278 ei
->i_flags
|= EXT4_APPEND_FL
;
4279 if (flags
& S_IMMUTABLE
)
4280 ei
->i_flags
|= EXT4_IMMUTABLE_FL
;
4281 if (flags
& S_NOATIME
)
4282 ei
->i_flags
|= EXT4_NOATIME_FL
;
4283 if (flags
& S_DIRSYNC
)
4284 ei
->i_flags
|= EXT4_DIRSYNC_FL
;
4287 static blkcnt_t
ext4_inode_blocks(struct ext4_inode
*raw_inode
,
4288 struct ext4_inode_info
*ei
)
4291 struct inode
*inode
= &(ei
->vfs_inode
);
4292 struct super_block
*sb
= inode
->i_sb
;
4294 if (EXT4_HAS_RO_COMPAT_FEATURE(sb
,
4295 EXT4_FEATURE_RO_COMPAT_HUGE_FILE
)) {
4296 /* we are using combined 48 bit field */
4297 i_blocks
= ((u64
)le16_to_cpu(raw_inode
->i_blocks_high
)) << 32 |
4298 le32_to_cpu(raw_inode
->i_blocks_lo
);
4299 if (ei
->i_flags
& EXT4_HUGE_FILE_FL
) {
4300 /* i_blocks represent file system block size */
4301 return i_blocks
<< (inode
->i_blkbits
- 9);
4306 return le32_to_cpu(raw_inode
->i_blocks_lo
);
4310 struct inode
*ext4_iget(struct super_block
*sb
, unsigned long ino
)
4312 struct ext4_iloc iloc
;
4313 struct ext4_inode
*raw_inode
;
4314 struct ext4_inode_info
*ei
;
4315 struct buffer_head
*bh
;
4316 struct inode
*inode
;
4320 inode
= iget_locked(sb
, ino
);
4322 return ERR_PTR(-ENOMEM
);
4323 if (!(inode
->i_state
& I_NEW
))
4328 ret
= __ext4_get_inode_loc(inode
, &iloc
, 0);
4332 raw_inode
= ext4_raw_inode(&iloc
);
4333 inode
->i_mode
= le16_to_cpu(raw_inode
->i_mode
);
4334 inode
->i_uid
= (uid_t
)le16_to_cpu(raw_inode
->i_uid_low
);
4335 inode
->i_gid
= (gid_t
)le16_to_cpu(raw_inode
->i_gid_low
);
4336 if (!(test_opt(inode
->i_sb
, NO_UID32
))) {
4337 inode
->i_uid
|= le16_to_cpu(raw_inode
->i_uid_high
) << 16;
4338 inode
->i_gid
|= le16_to_cpu(raw_inode
->i_gid_high
) << 16;
4340 inode
->i_nlink
= le16_to_cpu(raw_inode
->i_links_count
);
4343 ei
->i_dir_start_lookup
= 0;
4344 ei
->i_dtime
= le32_to_cpu(raw_inode
->i_dtime
);
4345 /* We now have enough fields to check if the inode was active or not.
4346 * This is needed because nfsd might try to access dead inodes
4347 * the test is that same one that e2fsck uses
4348 * NeilBrown 1999oct15
4350 if (inode
->i_nlink
== 0) {
4351 if (inode
->i_mode
== 0 ||
4352 !(EXT4_SB(inode
->i_sb
)->s_mount_state
& EXT4_ORPHAN_FS
)) {
4353 /* this inode is deleted */
4358 /* The only unlinked inodes we let through here have
4359 * valid i_mode and are being read by the orphan
4360 * recovery code: that's fine, we're about to complete
4361 * the process of deleting those. */
4363 ei
->i_flags
= le32_to_cpu(raw_inode
->i_flags
);
4364 inode
->i_blocks
= ext4_inode_blocks(raw_inode
, ei
);
4365 ei
->i_file_acl
= le32_to_cpu(raw_inode
->i_file_acl_lo
);
4366 if (EXT4_HAS_INCOMPAT_FEATURE(sb
, EXT4_FEATURE_INCOMPAT_64BIT
))
4368 ((__u64
)le16_to_cpu(raw_inode
->i_file_acl_high
)) << 32;
4369 inode
->i_size
= ext4_isize(raw_inode
);
4370 ei
->i_disksize
= inode
->i_size
;
4371 inode
->i_generation
= le32_to_cpu(raw_inode
->i_generation
);
4372 ei
->i_block_group
= iloc
.block_group
;
4373 ei
->i_last_alloc_group
= ~0;
4375 * NOTE! The in-memory inode i_data array is in little-endian order
4376 * even on big-endian machines: we do NOT byteswap the block numbers!
4378 for (block
= 0; block
< EXT4_N_BLOCKS
; block
++)
4379 ei
->i_data
[block
] = raw_inode
->i_block
[block
];
4380 INIT_LIST_HEAD(&ei
->i_orphan
);
4382 if (EXT4_INODE_SIZE(inode
->i_sb
) > EXT4_GOOD_OLD_INODE_SIZE
) {
4383 ei
->i_extra_isize
= le16_to_cpu(raw_inode
->i_extra_isize
);
4384 if (EXT4_GOOD_OLD_INODE_SIZE
+ ei
->i_extra_isize
>
4385 EXT4_INODE_SIZE(inode
->i_sb
)) {
4390 if (ei
->i_extra_isize
== 0) {
4391 /* The extra space is currently unused. Use it. */
4392 ei
->i_extra_isize
= sizeof(struct ext4_inode
) -
4393 EXT4_GOOD_OLD_INODE_SIZE
;
4395 __le32
*magic
= (void *)raw_inode
+
4396 EXT4_GOOD_OLD_INODE_SIZE
+
4398 if (*magic
== cpu_to_le32(EXT4_XATTR_MAGIC
))
4399 ei
->i_state
|= EXT4_STATE_XATTR
;
4402 ei
->i_extra_isize
= 0;
4404 EXT4_INODE_GET_XTIME(i_ctime
, inode
, raw_inode
);
4405 EXT4_INODE_GET_XTIME(i_mtime
, inode
, raw_inode
);
4406 EXT4_INODE_GET_XTIME(i_atime
, inode
, raw_inode
);
4407 EXT4_EINODE_GET_XTIME(i_crtime
, ei
, raw_inode
);
4409 inode
->i_version
= le32_to_cpu(raw_inode
->i_disk_version
);
4410 if (EXT4_INODE_SIZE(inode
->i_sb
) > EXT4_GOOD_OLD_INODE_SIZE
) {
4411 if (EXT4_FITS_IN_INODE(raw_inode
, ei
, i_version_hi
))
4413 (__u64
)(le32_to_cpu(raw_inode
->i_version_hi
)) << 32;
4417 if (ei
->i_file_acl
&&
4419 (le32_to_cpu(EXT4_SB(sb
)->s_es
->s_first_data_block
) +
4420 EXT4_SB(sb
)->s_gdb_count
)) ||
4421 (ei
->i_file_acl
>= ext4_blocks_count(EXT4_SB(sb
)->s_es
)))) {
4422 ext4_error(sb
, __func__
,
4423 "bad extended attribute block %llu in inode #%lu",
4424 ei
->i_file_acl
, inode
->i_ino
);
4427 } else if (ei
->i_flags
& EXT4_EXTENTS_FL
) {
4428 if (S_ISREG(inode
->i_mode
) || S_ISDIR(inode
->i_mode
) ||
4429 (S_ISLNK(inode
->i_mode
) &&
4430 !ext4_inode_is_fast_symlink(inode
)))
4431 /* Validate extent which is part of inode */
4432 ret
= ext4_ext_check_inode(inode
);
4433 } else if (S_ISREG(inode
->i_mode
) || S_ISDIR(inode
->i_mode
) ||
4434 (S_ISLNK(inode
->i_mode
) &&
4435 !ext4_inode_is_fast_symlink(inode
))) {
4436 /* Validate block references which are part of inode */
4437 ret
= ext4_check_inode_blockref(inode
);
4444 if (S_ISREG(inode
->i_mode
)) {
4445 inode
->i_op
= &ext4_file_inode_operations
;
4446 inode
->i_fop
= &ext4_file_operations
;
4447 ext4_set_aops(inode
);
4448 } else if (S_ISDIR(inode
->i_mode
)) {
4449 inode
->i_op
= &ext4_dir_inode_operations
;
4450 inode
->i_fop
= &ext4_dir_operations
;
4451 } else if (S_ISLNK(inode
->i_mode
)) {
4452 if (ext4_inode_is_fast_symlink(inode
)) {
4453 inode
->i_op
= &ext4_fast_symlink_inode_operations
;
4454 nd_terminate_link(ei
->i_data
, inode
->i_size
,
4455 sizeof(ei
->i_data
) - 1);
4457 inode
->i_op
= &ext4_symlink_inode_operations
;
4458 ext4_set_aops(inode
);
4460 } else if (S_ISCHR(inode
->i_mode
) || S_ISBLK(inode
->i_mode
) ||
4461 S_ISFIFO(inode
->i_mode
) || S_ISSOCK(inode
->i_mode
)) {
4462 inode
->i_op
= &ext4_special_inode_operations
;
4463 if (raw_inode
->i_block
[0])
4464 init_special_inode(inode
, inode
->i_mode
,
4465 old_decode_dev(le32_to_cpu(raw_inode
->i_block
[0])));
4467 init_special_inode(inode
, inode
->i_mode
,
4468 new_decode_dev(le32_to_cpu(raw_inode
->i_block
[1])));
4472 ext4_error(inode
->i_sb
, __func__
,
4473 "bogus i_mode (%o) for inode=%lu",
4474 inode
->i_mode
, inode
->i_ino
);
4478 ext4_set_inode_flags(inode
);
4479 unlock_new_inode(inode
);
4484 return ERR_PTR(ret
);
4487 static int ext4_inode_blocks_set(handle_t
*handle
,
4488 struct ext4_inode
*raw_inode
,
4489 struct ext4_inode_info
*ei
)
4491 struct inode
*inode
= &(ei
->vfs_inode
);
4492 u64 i_blocks
= inode
->i_blocks
;
4493 struct super_block
*sb
= inode
->i_sb
;
4495 if (i_blocks
<= ~0U) {
4497 * i_blocks can be represnted in a 32 bit variable
4498 * as multiple of 512 bytes
4500 raw_inode
->i_blocks_lo
= cpu_to_le32(i_blocks
);
4501 raw_inode
->i_blocks_high
= 0;
4502 ei
->i_flags
&= ~EXT4_HUGE_FILE_FL
;
4505 if (!EXT4_HAS_RO_COMPAT_FEATURE(sb
, EXT4_FEATURE_RO_COMPAT_HUGE_FILE
))
4508 if (i_blocks
<= 0xffffffffffffULL
) {
4510 * i_blocks can be represented in a 48 bit variable
4511 * as multiple of 512 bytes
4513 raw_inode
->i_blocks_lo
= cpu_to_le32(i_blocks
);
4514 raw_inode
->i_blocks_high
= cpu_to_le16(i_blocks
>> 32);
4515 ei
->i_flags
&= ~EXT4_HUGE_FILE_FL
;
4517 ei
->i_flags
|= EXT4_HUGE_FILE_FL
;
4518 /* i_block is stored in file system block size */
4519 i_blocks
= i_blocks
>> (inode
->i_blkbits
- 9);
4520 raw_inode
->i_blocks_lo
= cpu_to_le32(i_blocks
);
4521 raw_inode
->i_blocks_high
= cpu_to_le16(i_blocks
>> 32);
4527 * Post the struct inode info into an on-disk inode location in the
4528 * buffer-cache. This gobbles the caller's reference to the
4529 * buffer_head in the inode location struct.
4531 * The caller must have write access to iloc->bh.
4533 static int ext4_do_update_inode(handle_t
*handle
,
4534 struct inode
*inode
,
4535 struct ext4_iloc
*iloc
)
4537 struct ext4_inode
*raw_inode
= ext4_raw_inode(iloc
);
4538 struct ext4_inode_info
*ei
= EXT4_I(inode
);
4539 struct buffer_head
*bh
= iloc
->bh
;
4540 int err
= 0, rc
, block
;
4542 /* For fields not not tracking in the in-memory inode,
4543 * initialise them to zero for new inodes. */
4544 if (ei
->i_state
& EXT4_STATE_NEW
)
4545 memset(raw_inode
, 0, EXT4_SB(inode
->i_sb
)->s_inode_size
);
4547 ext4_get_inode_flags(ei
);
4548 raw_inode
->i_mode
= cpu_to_le16(inode
->i_mode
);
4549 if (!(test_opt(inode
->i_sb
, NO_UID32
))) {
4550 raw_inode
->i_uid_low
= cpu_to_le16(low_16_bits(inode
->i_uid
));
4551 raw_inode
->i_gid_low
= cpu_to_le16(low_16_bits(inode
->i_gid
));
4553 * Fix up interoperability with old kernels. Otherwise, old inodes get
4554 * re-used with the upper 16 bits of the uid/gid intact
4557 raw_inode
->i_uid_high
=
4558 cpu_to_le16(high_16_bits(inode
->i_uid
));
4559 raw_inode
->i_gid_high
=
4560 cpu_to_le16(high_16_bits(inode
->i_gid
));
4562 raw_inode
->i_uid_high
= 0;
4563 raw_inode
->i_gid_high
= 0;
4566 raw_inode
->i_uid_low
=
4567 cpu_to_le16(fs_high2lowuid(inode
->i_uid
));
4568 raw_inode
->i_gid_low
=
4569 cpu_to_le16(fs_high2lowgid(inode
->i_gid
));
4570 raw_inode
->i_uid_high
= 0;
4571 raw_inode
->i_gid_high
= 0;
4573 raw_inode
->i_links_count
= cpu_to_le16(inode
->i_nlink
);
4575 EXT4_INODE_SET_XTIME(i_ctime
, inode
, raw_inode
);
4576 EXT4_INODE_SET_XTIME(i_mtime
, inode
, raw_inode
);
4577 EXT4_INODE_SET_XTIME(i_atime
, inode
, raw_inode
);
4578 EXT4_EINODE_SET_XTIME(i_crtime
, ei
, raw_inode
);
4580 if (ext4_inode_blocks_set(handle
, raw_inode
, ei
))
4582 raw_inode
->i_dtime
= cpu_to_le32(ei
->i_dtime
);
4583 /* clear the migrate flag in the raw_inode */
4584 raw_inode
->i_flags
= cpu_to_le32(ei
->i_flags
& ~EXT4_EXT_MIGRATE
);
4585 if (EXT4_SB(inode
->i_sb
)->s_es
->s_creator_os
!=
4586 cpu_to_le32(EXT4_OS_HURD
))
4587 raw_inode
->i_file_acl_high
=
4588 cpu_to_le16(ei
->i_file_acl
>> 32);
4589 raw_inode
->i_file_acl_lo
= cpu_to_le32(ei
->i_file_acl
);
4590 ext4_isize_set(raw_inode
, ei
->i_disksize
);
4591 if (ei
->i_disksize
> 0x7fffffffULL
) {
4592 struct super_block
*sb
= inode
->i_sb
;
4593 if (!EXT4_HAS_RO_COMPAT_FEATURE(sb
,
4594 EXT4_FEATURE_RO_COMPAT_LARGE_FILE
) ||
4595 EXT4_SB(sb
)->s_es
->s_rev_level
==
4596 cpu_to_le32(EXT4_GOOD_OLD_REV
)) {
4597 /* If this is the first large file
4598 * created, add a flag to the superblock.
4600 err
= ext4_journal_get_write_access(handle
,
4601 EXT4_SB(sb
)->s_sbh
);
4604 ext4_update_dynamic_rev(sb
);
4605 EXT4_SET_RO_COMPAT_FEATURE(sb
,
4606 EXT4_FEATURE_RO_COMPAT_LARGE_FILE
);
4608 ext4_handle_sync(handle
);
4609 err
= ext4_handle_dirty_metadata(handle
, inode
,
4610 EXT4_SB(sb
)->s_sbh
);
4613 raw_inode
->i_generation
= cpu_to_le32(inode
->i_generation
);
4614 if (S_ISCHR(inode
->i_mode
) || S_ISBLK(inode
->i_mode
)) {
4615 if (old_valid_dev(inode
->i_rdev
)) {
4616 raw_inode
->i_block
[0] =
4617 cpu_to_le32(old_encode_dev(inode
->i_rdev
));
4618 raw_inode
->i_block
[1] = 0;
4620 raw_inode
->i_block
[0] = 0;
4621 raw_inode
->i_block
[1] =
4622 cpu_to_le32(new_encode_dev(inode
->i_rdev
));
4623 raw_inode
->i_block
[2] = 0;
4626 for (block
= 0; block
< EXT4_N_BLOCKS
; block
++)
4627 raw_inode
->i_block
[block
] = ei
->i_data
[block
];
4629 raw_inode
->i_disk_version
= cpu_to_le32(inode
->i_version
);
4630 if (ei
->i_extra_isize
) {
4631 if (EXT4_FITS_IN_INODE(raw_inode
, ei
, i_version_hi
))
4632 raw_inode
->i_version_hi
=
4633 cpu_to_le32(inode
->i_version
>> 32);
4634 raw_inode
->i_extra_isize
= cpu_to_le16(ei
->i_extra_isize
);
4637 BUFFER_TRACE(bh
, "call ext4_handle_dirty_metadata");
4638 rc
= ext4_handle_dirty_metadata(handle
, inode
, bh
);
4641 ei
->i_state
&= ~EXT4_STATE_NEW
;
4645 ext4_std_error(inode
->i_sb
, err
);
4650 * ext4_write_inode()
4652 * We are called from a few places:
4654 * - Within generic_file_write() for O_SYNC files.
4655 * Here, there will be no transaction running. We wait for any running
4656 * trasnaction to commit.
4658 * - Within sys_sync(), kupdate and such.
4659 * We wait on commit, if tol to.
4661 * - Within prune_icache() (PF_MEMALLOC == true)
4662 * Here we simply return. We can't afford to block kswapd on the
4665 * In all cases it is actually safe for us to return without doing anything,
4666 * because the inode has been copied into a raw inode buffer in
4667 * ext4_mark_inode_dirty(). This is a correctness thing for O_SYNC and for
4670 * Note that we are absolutely dependent upon all inode dirtiers doing the
4671 * right thing: they *must* call mark_inode_dirty() after dirtying info in
4672 * which we are interested.
4674 * It would be a bug for them to not do this. The code:
4676 * mark_inode_dirty(inode)
4678 * inode->i_size = expr;
4680 * is in error because a kswapd-driven write_inode() could occur while
4681 * `stuff()' is running, and the new i_size will be lost. Plus the inode
4682 * will no longer be on the superblock's dirty inode list.
4684 int ext4_write_inode(struct inode
*inode
, int wait
)
4686 if (current
->flags
& PF_MEMALLOC
)
4689 if (ext4_journal_current_handle()) {
4690 jbd_debug(1, "called recursively, non-PF_MEMALLOC!\n");
4698 return ext4_force_commit(inode
->i_sb
);
4704 * Called from notify_change.
4706 * We want to trap VFS attempts to truncate the file as soon as
4707 * possible. In particular, we want to make sure that when the VFS
4708 * shrinks i_size, we put the inode on the orphan list and modify
4709 * i_disksize immediately, so that during the subsequent flushing of
4710 * dirty pages and freeing of disk blocks, we can guarantee that any
4711 * commit will leave the blocks being flushed in an unused state on
4712 * disk. (On recovery, the inode will get truncated and the blocks will
4713 * be freed, so we have a strong guarantee that no future commit will
4714 * leave these blocks visible to the user.)
4716 * Another thing we have to assure is that if we are in ordered mode
4717 * and inode is still attached to the committing transaction, we must
4718 * we start writeout of all the dirty pages which are being truncated.
4719 * This way we are sure that all the data written in the previous
4720 * transaction are already on disk (truncate waits for pages under
4723 * Called with inode->i_mutex down.
4725 int ext4_setattr(struct dentry
*dentry
, struct iattr
*attr
)
4727 struct inode
*inode
= dentry
->d_inode
;
4729 const unsigned int ia_valid
= attr
->ia_valid
;
4731 error
= inode_change_ok(inode
, attr
);
4735 if ((ia_valid
& ATTR_UID
&& attr
->ia_uid
!= inode
->i_uid
) ||
4736 (ia_valid
& ATTR_GID
&& attr
->ia_gid
!= inode
->i_gid
)) {
4739 /* (user+group)*(old+new) structure, inode write (sb,
4740 * inode block, ? - but truncate inode update has it) */
4741 handle
= ext4_journal_start(inode
, 2*(EXT4_QUOTA_INIT_BLOCKS(inode
->i_sb
)+
4742 EXT4_QUOTA_DEL_BLOCKS(inode
->i_sb
))+3);
4743 if (IS_ERR(handle
)) {
4744 error
= PTR_ERR(handle
);
4747 error
= vfs_dq_transfer(inode
, attr
) ? -EDQUOT
: 0;
4749 ext4_journal_stop(handle
);
4752 /* Update corresponding info in inode so that everything is in
4753 * one transaction */
4754 if (attr
->ia_valid
& ATTR_UID
)
4755 inode
->i_uid
= attr
->ia_uid
;
4756 if (attr
->ia_valid
& ATTR_GID
)
4757 inode
->i_gid
= attr
->ia_gid
;
4758 error
= ext4_mark_inode_dirty(handle
, inode
);
4759 ext4_journal_stop(handle
);
4762 if (attr
->ia_valid
& ATTR_SIZE
) {
4763 if (!(EXT4_I(inode
)->i_flags
& EXT4_EXTENTS_FL
)) {
4764 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
4766 if (attr
->ia_size
> sbi
->s_bitmap_maxbytes
) {
4773 if (S_ISREG(inode
->i_mode
) &&
4774 attr
->ia_valid
& ATTR_SIZE
&& attr
->ia_size
< inode
->i_size
) {
4777 handle
= ext4_journal_start(inode
, 3);
4778 if (IS_ERR(handle
)) {
4779 error
= PTR_ERR(handle
);
4783 error
= ext4_orphan_add(handle
, inode
);
4784 EXT4_I(inode
)->i_disksize
= attr
->ia_size
;
4785 rc
= ext4_mark_inode_dirty(handle
, inode
);
4788 ext4_journal_stop(handle
);
4790 if (ext4_should_order_data(inode
)) {
4791 error
= ext4_begin_ordered_truncate(inode
,
4794 /* Do as much error cleanup as possible */
4795 handle
= ext4_journal_start(inode
, 3);
4796 if (IS_ERR(handle
)) {
4797 ext4_orphan_del(NULL
, inode
);
4800 ext4_orphan_del(handle
, inode
);
4801 ext4_journal_stop(handle
);
4807 rc
= inode_setattr(inode
, attr
);
4809 /* If inode_setattr's call to ext4_truncate failed to get a
4810 * transaction handle at all, we need to clean up the in-core
4811 * orphan list manually. */
4813 ext4_orphan_del(NULL
, inode
);
4815 if (!rc
&& (ia_valid
& ATTR_MODE
))
4816 rc
= ext4_acl_chmod(inode
);
4819 ext4_std_error(inode
->i_sb
, error
);
4825 int ext4_getattr(struct vfsmount
*mnt
, struct dentry
*dentry
,
4828 struct inode
*inode
;
4829 unsigned long delalloc_blocks
;
4831 inode
= dentry
->d_inode
;
4832 generic_fillattr(inode
, stat
);
4835 * We can't update i_blocks if the block allocation is delayed
4836 * otherwise in the case of system crash before the real block
4837 * allocation is done, we will have i_blocks inconsistent with
4838 * on-disk file blocks.
4839 * We always keep i_blocks updated together with real
4840 * allocation. But to not confuse with user, stat
4841 * will return the blocks that include the delayed allocation
4842 * blocks for this file.
4844 spin_lock(&EXT4_I(inode
)->i_block_reservation_lock
);
4845 delalloc_blocks
= EXT4_I(inode
)->i_reserved_data_blocks
;
4846 spin_unlock(&EXT4_I(inode
)->i_block_reservation_lock
);
4848 stat
->blocks
+= (delalloc_blocks
<< inode
->i_sb
->s_blocksize_bits
)>>9;
4852 static int ext4_indirect_trans_blocks(struct inode
*inode
, int nrblocks
,
4857 /* if nrblocks are contiguous */
4860 * With N contiguous data blocks, it need at most
4861 * N/EXT4_ADDR_PER_BLOCK(inode->i_sb) indirect blocks
4862 * 2 dindirect blocks
4865 indirects
= nrblocks
/ EXT4_ADDR_PER_BLOCK(inode
->i_sb
);
4866 return indirects
+ 3;
4869 * if nrblocks are not contiguous, worse case, each block touch
4870 * a indirect block, and each indirect block touch a double indirect
4871 * block, plus a triple indirect block
4873 indirects
= nrblocks
* 2 + 1;
4877 static int ext4_index_trans_blocks(struct inode
*inode
, int nrblocks
, int chunk
)
4879 if (!(EXT4_I(inode
)->i_flags
& EXT4_EXTENTS_FL
))
4880 return ext4_indirect_trans_blocks(inode
, nrblocks
, chunk
);
4881 return ext4_ext_index_trans_blocks(inode
, nrblocks
, chunk
);
4885 * Account for index blocks, block groups bitmaps and block group
4886 * descriptor blocks if modify datablocks and index blocks
4887 * worse case, the indexs blocks spread over different block groups
4889 * If datablocks are discontiguous, they are possible to spread over
4890 * different block groups too. If they are contiugous, with flexbg,
4891 * they could still across block group boundary.
4893 * Also account for superblock, inode, quota and xattr blocks
4895 int ext4_meta_trans_blocks(struct inode
*inode
, int nrblocks
, int chunk
)
4897 ext4_group_t groups
, ngroups
= ext4_get_groups_count(inode
->i_sb
);
4903 * How many index blocks need to touch to modify nrblocks?
4904 * The "Chunk" flag indicating whether the nrblocks is
4905 * physically contiguous on disk
4907 * For Direct IO and fallocate, they calls get_block to allocate
4908 * one single extent at a time, so they could set the "Chunk" flag
4910 idxblocks
= ext4_index_trans_blocks(inode
, nrblocks
, chunk
);
4915 * Now let's see how many group bitmaps and group descriptors need
4925 if (groups
> ngroups
)
4927 if (groups
> EXT4_SB(inode
->i_sb
)->s_gdb_count
)
4928 gdpblocks
= EXT4_SB(inode
->i_sb
)->s_gdb_count
;
4930 /* bitmaps and block group descriptor blocks */
4931 ret
+= groups
+ gdpblocks
;
4933 /* Blocks for super block, inode, quota and xattr blocks */
4934 ret
+= EXT4_META_TRANS_BLOCKS(inode
->i_sb
);
4940 * Calulate the total number of credits to reserve to fit
4941 * the modification of a single pages into a single transaction,
4942 * which may include multiple chunks of block allocations.
4944 * This could be called via ext4_write_begin()
4946 * We need to consider the worse case, when
4947 * one new block per extent.
4949 int ext4_writepage_trans_blocks(struct inode
*inode
)
4951 int bpp
= ext4_journal_blocks_per_page(inode
);
4954 ret
= ext4_meta_trans_blocks(inode
, bpp
, 0);
4956 /* Account for data blocks for journalled mode */
4957 if (ext4_should_journal_data(inode
))
4963 * Calculate the journal credits for a chunk of data modification.
4965 * This is called from DIO, fallocate or whoever calling
4966 * ext4_get_blocks() to map/allocate a chunk of contigous disk blocks.
4968 * journal buffers for data blocks are not included here, as DIO
4969 * and fallocate do no need to journal data buffers.
4971 int ext4_chunk_trans_blocks(struct inode
*inode
, int nrblocks
)
4973 return ext4_meta_trans_blocks(inode
, nrblocks
, 1);
4977 * The caller must have previously called ext4_reserve_inode_write().
4978 * Give this, we know that the caller already has write access to iloc->bh.
4980 int ext4_mark_iloc_dirty(handle_t
*handle
,
4981 struct inode
*inode
, struct ext4_iloc
*iloc
)
4985 if (test_opt(inode
->i_sb
, I_VERSION
))
4986 inode_inc_iversion(inode
);
4988 /* the do_update_inode consumes one bh->b_count */
4991 /* ext4_do_update_inode() does jbd2_journal_dirty_metadata */
4992 err
= ext4_do_update_inode(handle
, inode
, iloc
);
4998 * On success, We end up with an outstanding reference count against
4999 * iloc->bh. This _must_ be cleaned up later.
5003 ext4_reserve_inode_write(handle_t
*handle
, struct inode
*inode
,
5004 struct ext4_iloc
*iloc
)
5008 err
= ext4_get_inode_loc(inode
, iloc
);
5010 BUFFER_TRACE(iloc
->bh
, "get_write_access");
5011 err
= ext4_journal_get_write_access(handle
, iloc
->bh
);
5017 ext4_std_error(inode
->i_sb
, err
);
5022 * Expand an inode by new_extra_isize bytes.
5023 * Returns 0 on success or negative error number on failure.
5025 static int ext4_expand_extra_isize(struct inode
*inode
,
5026 unsigned int new_extra_isize
,
5027 struct ext4_iloc iloc
,
5030 struct ext4_inode
*raw_inode
;
5031 struct ext4_xattr_ibody_header
*header
;
5032 struct ext4_xattr_entry
*entry
;
5034 if (EXT4_I(inode
)->i_extra_isize
>= new_extra_isize
)
5037 raw_inode
= ext4_raw_inode(&iloc
);
5039 header
= IHDR(inode
, raw_inode
);
5040 entry
= IFIRST(header
);
5042 /* No extended attributes present */
5043 if (!(EXT4_I(inode
)->i_state
& EXT4_STATE_XATTR
) ||
5044 header
->h_magic
!= cpu_to_le32(EXT4_XATTR_MAGIC
)) {
5045 memset((void *)raw_inode
+ EXT4_GOOD_OLD_INODE_SIZE
, 0,
5047 EXT4_I(inode
)->i_extra_isize
= new_extra_isize
;
5051 /* try to expand with EAs present */
5052 return ext4_expand_extra_isize_ea(inode
, new_extra_isize
,
5057 * What we do here is to mark the in-core inode as clean with respect to inode
5058 * dirtiness (it may still be data-dirty).
5059 * This means that the in-core inode may be reaped by prune_icache
5060 * without having to perform any I/O. This is a very good thing,
5061 * because *any* task may call prune_icache - even ones which
5062 * have a transaction open against a different journal.
5064 * Is this cheating? Not really. Sure, we haven't written the
5065 * inode out, but prune_icache isn't a user-visible syncing function.
5066 * Whenever the user wants stuff synced (sys_sync, sys_msync, sys_fsync)
5067 * we start and wait on commits.
5069 * Is this efficient/effective? Well, we're being nice to the system
5070 * by cleaning up our inodes proactively so they can be reaped
5071 * without I/O. But we are potentially leaving up to five seconds'
5072 * worth of inodes floating about which prune_icache wants us to
5073 * write out. One way to fix that would be to get prune_icache()
5074 * to do a write_super() to free up some memory. It has the desired
5077 int ext4_mark_inode_dirty(handle_t
*handle
, struct inode
*inode
)
5079 struct ext4_iloc iloc
;
5080 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
5081 static unsigned int mnt_count
;
5085 err
= ext4_reserve_inode_write(handle
, inode
, &iloc
);
5086 if (ext4_handle_valid(handle
) &&
5087 EXT4_I(inode
)->i_extra_isize
< sbi
->s_want_extra_isize
&&
5088 !(EXT4_I(inode
)->i_state
& EXT4_STATE_NO_EXPAND
)) {
5090 * We need extra buffer credits since we may write into EA block
5091 * with this same handle. If journal_extend fails, then it will
5092 * only result in a minor loss of functionality for that inode.
5093 * If this is felt to be critical, then e2fsck should be run to
5094 * force a large enough s_min_extra_isize.
5096 if ((jbd2_journal_extend(handle
,
5097 EXT4_DATA_TRANS_BLOCKS(inode
->i_sb
))) == 0) {
5098 ret
= ext4_expand_extra_isize(inode
,
5099 sbi
->s_want_extra_isize
,
5102 EXT4_I(inode
)->i_state
|= EXT4_STATE_NO_EXPAND
;
5104 le16_to_cpu(sbi
->s_es
->s_mnt_count
)) {
5105 ext4_warning(inode
->i_sb
, __func__
,
5106 "Unable to expand inode %lu. Delete"
5107 " some EAs or run e2fsck.",
5110 le16_to_cpu(sbi
->s_es
->s_mnt_count
);
5116 err
= ext4_mark_iloc_dirty(handle
, inode
, &iloc
);
5121 * ext4_dirty_inode() is called from __mark_inode_dirty()
5123 * We're really interested in the case where a file is being extended.
5124 * i_size has been changed by generic_commit_write() and we thus need
5125 * to include the updated inode in the current transaction.
5127 * Also, vfs_dq_alloc_block() will always dirty the inode when blocks
5128 * are allocated to the file.
5130 * If the inode is marked synchronous, we don't honour that here - doing
5131 * so would cause a commit on atime updates, which we don't bother doing.
5132 * We handle synchronous inodes at the highest possible level.
5134 void ext4_dirty_inode(struct inode
*inode
)
5136 handle_t
*current_handle
= ext4_journal_current_handle();
5139 if (!ext4_handle_valid(current_handle
)) {
5140 ext4_mark_inode_dirty(current_handle
, inode
);
5144 handle
= ext4_journal_start(inode
, 2);
5147 if (current_handle
&&
5148 current_handle
->h_transaction
!= handle
->h_transaction
) {
5149 /* This task has a transaction open against a different fs */
5150 printk(KERN_EMERG
"%s: transactions do not match!\n",
5153 jbd_debug(5, "marking dirty. outer handle=%p\n",
5155 ext4_mark_inode_dirty(handle
, inode
);
5157 ext4_journal_stop(handle
);
5164 * Bind an inode's backing buffer_head into this transaction, to prevent
5165 * it from being flushed to disk early. Unlike
5166 * ext4_reserve_inode_write, this leaves behind no bh reference and
5167 * returns no iloc structure, so the caller needs to repeat the iloc
5168 * lookup to mark the inode dirty later.
5170 static int ext4_pin_inode(handle_t
*handle
, struct inode
*inode
)
5172 struct ext4_iloc iloc
;
5176 err
= ext4_get_inode_loc(inode
, &iloc
);
5178 BUFFER_TRACE(iloc
.bh
, "get_write_access");
5179 err
= jbd2_journal_get_write_access(handle
, iloc
.bh
);
5181 err
= ext4_handle_dirty_metadata(handle
,
5187 ext4_std_error(inode
->i_sb
, err
);
5192 int ext4_change_inode_journal_flag(struct inode
*inode
, int val
)
5199 * We have to be very careful here: changing a data block's
5200 * journaling status dynamically is dangerous. If we write a
5201 * data block to the journal, change the status and then delete
5202 * that block, we risk forgetting to revoke the old log record
5203 * from the journal and so a subsequent replay can corrupt data.
5204 * So, first we make sure that the journal is empty and that
5205 * nobody is changing anything.
5208 journal
= EXT4_JOURNAL(inode
);
5211 if (is_journal_aborted(journal
))
5214 jbd2_journal_lock_updates(journal
);
5215 jbd2_journal_flush(journal
);
5218 * OK, there are no updates running now, and all cached data is
5219 * synced to disk. We are now in a completely consistent state
5220 * which doesn't have anything in the journal, and we know that
5221 * no filesystem updates are running, so it is safe to modify
5222 * the inode's in-core data-journaling state flag now.
5226 EXT4_I(inode
)->i_flags
|= EXT4_JOURNAL_DATA_FL
;
5228 EXT4_I(inode
)->i_flags
&= ~EXT4_JOURNAL_DATA_FL
;
5229 ext4_set_aops(inode
);
5231 jbd2_journal_unlock_updates(journal
);
5233 /* Finally we can mark the inode as dirty. */
5235 handle
= ext4_journal_start(inode
, 1);
5237 return PTR_ERR(handle
);
5239 err
= ext4_mark_inode_dirty(handle
, inode
);
5240 ext4_handle_sync(handle
);
5241 ext4_journal_stop(handle
);
5242 ext4_std_error(inode
->i_sb
, err
);
5247 static int ext4_bh_unmapped(handle_t
*handle
, struct buffer_head
*bh
)
5249 return !buffer_mapped(bh
);
5252 int ext4_page_mkwrite(struct vm_area_struct
*vma
, struct vm_fault
*vmf
)
5254 struct page
*page
= vmf
->page
;
5259 struct file
*file
= vma
->vm_file
;
5260 struct inode
*inode
= file
->f_path
.dentry
->d_inode
;
5261 struct address_space
*mapping
= inode
->i_mapping
;
5264 * Get i_alloc_sem to stop truncates messing with the inode. We cannot
5265 * get i_mutex because we are already holding mmap_sem.
5267 down_read(&inode
->i_alloc_sem
);
5268 size
= i_size_read(inode
);
5269 if (page
->mapping
!= mapping
|| size
<= page_offset(page
)
5270 || !PageUptodate(page
)) {
5271 /* page got truncated from under us? */
5275 if (PageMappedToDisk(page
))
5278 if (page
->index
== size
>> PAGE_CACHE_SHIFT
)
5279 len
= size
& ~PAGE_CACHE_MASK
;
5281 len
= PAGE_CACHE_SIZE
;
5283 if (page_has_buffers(page
)) {
5284 /* return if we have all the buffers mapped */
5285 if (!walk_page_buffers(NULL
, page_buffers(page
), 0, len
, NULL
,
5290 * OK, we need to fill the hole... Do write_begin write_end
5291 * to do block allocation/reservation.We are not holding
5292 * inode.i__mutex here. That allow * parallel write_begin,
5293 * write_end call. lock_page prevent this from happening
5294 * on the same page though
5296 ret
= mapping
->a_ops
->write_begin(file
, mapping
, page_offset(page
),
5297 len
, AOP_FLAG_UNINTERRUPTIBLE
, &page
, &fsdata
);
5300 ret
= mapping
->a_ops
->write_end(file
, mapping
, page_offset(page
),
5301 len
, len
, page
, fsdata
);
5307 ret
= VM_FAULT_SIGBUS
;
5308 up_read(&inode
->i_alloc_sem
);