2 * linux/fs/ext4/inode.c
4 * Copyright (C) 1992, 1993, 1994, 1995
5 * Remy Card (card@masi.ibp.fr)
6 * Laboratoire MASI - Institut Blaise Pascal
7 * Universite Pierre et Marie Curie (Paris VI)
11 * linux/fs/minix/inode.c
13 * Copyright (C) 1991, 1992 Linus Torvalds
15 * Goal-directed block allocation by Stephen Tweedie
16 * (sct@redhat.com), 1993, 1998
17 * Big-endian to little-endian byte-swapping/bitmaps by
18 * David S. Miller (davem@caip.rutgers.edu), 1995
19 * 64-bit file support on 64-bit platforms by Jakub Jelinek
20 * (jj@sunsite.ms.mff.cuni.cz)
22 * Assorted race fixes, rewrite of ext4_get_block() by Al Viro, 2000
25 #include <linux/module.h>
27 #include <linux/time.h>
28 #include <linux/jbd2.h>
29 #include <linux/highuid.h>
30 #include <linux/pagemap.h>
31 #include <linux/quotaops.h>
32 #include <linux/string.h>
33 #include <linux/buffer_head.h>
34 #include <linux/writeback.h>
35 #include <linux/pagevec.h>
36 #include <linux/mpage.h>
37 #include <linux/namei.h>
38 #include <linux/uio.h>
39 #include <linux/bio.h>
40 #include <linux/workqueue.h>
42 #include "ext4_jbd2.h"
45 #include "ext4_extents.h"
47 #include <trace/events/ext4.h>
49 #define MPAGE_DA_EXTENT_TAIL 0x01
51 static inline int ext4_begin_ordered_truncate(struct inode
*inode
,
54 return jbd2_journal_begin_ordered_truncate(
55 EXT4_SB(inode
->i_sb
)->s_journal
,
56 &EXT4_I(inode
)->jinode
,
60 static void ext4_invalidatepage(struct page
*page
, unsigned long offset
);
63 * Test whether an inode is a fast symlink.
65 static int ext4_inode_is_fast_symlink(struct inode
*inode
)
67 int ea_blocks
= EXT4_I(inode
)->i_file_acl
?
68 (inode
->i_sb
->s_blocksize
>> 9) : 0;
70 return (S_ISLNK(inode
->i_mode
) && inode
->i_blocks
- ea_blocks
== 0);
74 * The ext4 forget function must perform a revoke if we are freeing data
75 * which has been journaled. Metadata (eg. indirect blocks) must be
76 * revoked in all cases.
78 * "bh" may be NULL: a metadata block may have been freed from memory
79 * but there may still be a record of it in the journal, and that record
80 * still needs to be revoked.
82 * If the handle isn't valid we're not journaling, but we still need to
83 * call into ext4_journal_revoke() to put the buffer head.
85 int ext4_forget(handle_t
*handle
, int is_metadata
, struct inode
*inode
,
86 struct buffer_head
*bh
, ext4_fsblk_t blocknr
)
92 BUFFER_TRACE(bh
, "enter");
94 jbd_debug(4, "forgetting bh %p: is_metadata = %d, mode %o, "
96 bh
, is_metadata
, inode
->i_mode
,
97 test_opt(inode
->i_sb
, DATA_FLAGS
));
99 /* Never use the revoke function if we are doing full data
100 * journaling: there is no need to, and a V1 superblock won't
101 * support it. Otherwise, only skip the revoke on un-journaled
104 if (test_opt(inode
->i_sb
, DATA_FLAGS
) == EXT4_MOUNT_JOURNAL_DATA
||
105 (!is_metadata
&& !ext4_should_journal_data(inode
))) {
107 BUFFER_TRACE(bh
, "call jbd2_journal_forget");
108 return ext4_journal_forget(handle
, bh
);
114 * data!=journal && (is_metadata || should_journal_data(inode))
116 BUFFER_TRACE(bh
, "call ext4_journal_revoke");
117 err
= ext4_journal_revoke(handle
, blocknr
, bh
);
119 ext4_abort(inode
->i_sb
, __func__
,
120 "error %d when attempting revoke", err
);
121 BUFFER_TRACE(bh
, "exit");
126 * Work out how many blocks we need to proceed with the next chunk of a
127 * truncate transaction.
129 static unsigned long blocks_for_truncate(struct inode
*inode
)
133 needed
= inode
->i_blocks
>> (inode
->i_sb
->s_blocksize_bits
- 9);
135 /* Give ourselves just enough room to cope with inodes in which
136 * i_blocks is corrupt: we've seen disk corruptions in the past
137 * which resulted in random data in an inode which looked enough
138 * like a regular file for ext4 to try to delete it. Things
139 * will go a bit crazy if that happens, but at least we should
140 * try not to panic the whole kernel. */
144 /* But we need to bound the transaction so we don't overflow the
146 if (needed
> EXT4_MAX_TRANS_DATA
)
147 needed
= EXT4_MAX_TRANS_DATA
;
149 return EXT4_DATA_TRANS_BLOCKS(inode
->i_sb
) + needed
;
153 * Truncate transactions can be complex and absolutely huge. So we need to
154 * be able to restart the transaction at a conventient checkpoint to make
155 * sure we don't overflow the journal.
157 * start_transaction gets us a new handle for a truncate transaction,
158 * and extend_transaction tries to extend the existing one a bit. If
159 * extend fails, we need to propagate the failure up and restart the
160 * transaction in the top-level truncate loop. --sct
162 static handle_t
*start_transaction(struct inode
*inode
)
166 result
= ext4_journal_start(inode
, blocks_for_truncate(inode
));
170 ext4_std_error(inode
->i_sb
, PTR_ERR(result
));
175 * Try to extend this transaction for the purposes of truncation.
177 * Returns 0 if we managed to create more room. If we can't create more
178 * room, and the transaction must be restarted we return 1.
180 static int try_to_extend_transaction(handle_t
*handle
, struct inode
*inode
)
182 if (!ext4_handle_valid(handle
))
184 if (ext4_handle_has_enough_credits(handle
, EXT4_RESERVE_TRANS_BLOCKS
+1))
186 if (!ext4_journal_extend(handle
, blocks_for_truncate(inode
)))
192 * Restart the transaction associated with *handle. This does a commit,
193 * so before we call here everything must be consistently dirtied against
196 int ext4_truncate_restart_trans(handle_t
*handle
, struct inode
*inode
,
202 * Drop i_data_sem to avoid deadlock with ext4_get_blocks At this
203 * moment, get_block can be called only for blocks inside i_size since
204 * page cache has been already dropped and writes are blocked by
205 * i_mutex. So we can safely drop the i_data_sem here.
207 BUG_ON(EXT4_JOURNAL(inode
) == NULL
);
208 jbd_debug(2, "restarting handle %p\n", handle
);
209 up_write(&EXT4_I(inode
)->i_data_sem
);
210 ret
= ext4_journal_restart(handle
, blocks_for_truncate(inode
));
211 down_write(&EXT4_I(inode
)->i_data_sem
);
212 ext4_discard_preallocations(inode
);
218 * Called at the last iput() if i_nlink is zero.
220 void ext4_delete_inode(struct inode
*inode
)
225 if (ext4_should_order_data(inode
))
226 ext4_begin_ordered_truncate(inode
, 0);
227 truncate_inode_pages(&inode
->i_data
, 0);
229 if (is_bad_inode(inode
))
232 handle
= ext4_journal_start(inode
, blocks_for_truncate(inode
)+3);
233 if (IS_ERR(handle
)) {
234 ext4_std_error(inode
->i_sb
, PTR_ERR(handle
));
236 * If we're going to skip the normal cleanup, we still need to
237 * make sure that the in-core orphan linked list is properly
240 ext4_orphan_del(NULL
, inode
);
245 ext4_handle_sync(handle
);
247 err
= ext4_mark_inode_dirty(handle
, inode
);
249 ext4_warning(inode
->i_sb
, __func__
,
250 "couldn't mark inode dirty (err %d)", err
);
254 ext4_truncate(inode
);
257 * ext4_ext_truncate() doesn't reserve any slop when it
258 * restarts journal transactions; therefore there may not be
259 * enough credits left in the handle to remove the inode from
260 * the orphan list and set the dtime field.
262 if (!ext4_handle_has_enough_credits(handle
, 3)) {
263 err
= ext4_journal_extend(handle
, 3);
265 err
= ext4_journal_restart(handle
, 3);
267 ext4_warning(inode
->i_sb
, __func__
,
268 "couldn't extend journal (err %d)", err
);
270 ext4_journal_stop(handle
);
276 * Kill off the orphan record which ext4_truncate created.
277 * AKPM: I think this can be inside the above `if'.
278 * Note that ext4_orphan_del() has to be able to cope with the
279 * deletion of a non-existent orphan - this is because we don't
280 * know if ext4_truncate() actually created an orphan record.
281 * (Well, we could do this if we need to, but heck - it works)
283 ext4_orphan_del(handle
, inode
);
284 EXT4_I(inode
)->i_dtime
= get_seconds();
287 * One subtle ordering requirement: if anything has gone wrong
288 * (transaction abort, IO errors, whatever), then we can still
289 * do these next steps (the fs will already have been marked as
290 * having errors), but we can't free the inode if the mark_dirty
293 if (ext4_mark_inode_dirty(handle
, inode
))
294 /* If that failed, just do the required in-core inode clear. */
297 ext4_free_inode(handle
, inode
);
298 ext4_journal_stop(handle
);
301 clear_inode(inode
); /* We must guarantee clearing of inode... */
307 struct buffer_head
*bh
;
310 static inline void add_chain(Indirect
*p
, struct buffer_head
*bh
, __le32
*v
)
312 p
->key
= *(p
->p
= v
);
317 * ext4_block_to_path - parse the block number into array of offsets
318 * @inode: inode in question (we are only interested in its superblock)
319 * @i_block: block number to be parsed
320 * @offsets: array to store the offsets in
321 * @boundary: set this non-zero if the referred-to block is likely to be
322 * followed (on disk) by an indirect block.
324 * To store the locations of file's data ext4 uses a data structure common
325 * for UNIX filesystems - tree of pointers anchored in the inode, with
326 * data blocks at leaves and indirect blocks in intermediate nodes.
327 * This function translates the block number into path in that tree -
328 * return value is the path length and @offsets[n] is the offset of
329 * pointer to (n+1)th node in the nth one. If @block is out of range
330 * (negative or too large) warning is printed and zero returned.
332 * Note: function doesn't find node addresses, so no IO is needed. All
333 * we need to know is the capacity of indirect blocks (taken from the
338 * Portability note: the last comparison (check that we fit into triple
339 * indirect block) is spelled differently, because otherwise on an
340 * architecture with 32-bit longs and 8Kb pages we might get into trouble
341 * if our filesystem had 8Kb blocks. We might use long long, but that would
342 * kill us on x86. Oh, well, at least the sign propagation does not matter -
343 * i_block would have to be negative in the very beginning, so we would not
347 static int ext4_block_to_path(struct inode
*inode
,
349 ext4_lblk_t offsets
[4], int *boundary
)
351 int ptrs
= EXT4_ADDR_PER_BLOCK(inode
->i_sb
);
352 int ptrs_bits
= EXT4_ADDR_PER_BLOCK_BITS(inode
->i_sb
);
353 const long direct_blocks
= EXT4_NDIR_BLOCKS
,
354 indirect_blocks
= ptrs
,
355 double_blocks
= (1 << (ptrs_bits
* 2));
359 if (i_block
< direct_blocks
) {
360 offsets
[n
++] = i_block
;
361 final
= direct_blocks
;
362 } else if ((i_block
-= direct_blocks
) < indirect_blocks
) {
363 offsets
[n
++] = EXT4_IND_BLOCK
;
364 offsets
[n
++] = i_block
;
366 } else if ((i_block
-= indirect_blocks
) < double_blocks
) {
367 offsets
[n
++] = EXT4_DIND_BLOCK
;
368 offsets
[n
++] = i_block
>> ptrs_bits
;
369 offsets
[n
++] = i_block
& (ptrs
- 1);
371 } else if (((i_block
-= double_blocks
) >> (ptrs_bits
* 2)) < ptrs
) {
372 offsets
[n
++] = EXT4_TIND_BLOCK
;
373 offsets
[n
++] = i_block
>> (ptrs_bits
* 2);
374 offsets
[n
++] = (i_block
>> ptrs_bits
) & (ptrs
- 1);
375 offsets
[n
++] = i_block
& (ptrs
- 1);
378 ext4_warning(inode
->i_sb
, "ext4_block_to_path",
379 "block %lu > max in inode %lu",
380 i_block
+ direct_blocks
+
381 indirect_blocks
+ double_blocks
, inode
->i_ino
);
384 *boundary
= final
- 1 - (i_block
& (ptrs
- 1));
388 static int __ext4_check_blockref(const char *function
, struct inode
*inode
,
389 __le32
*p
, unsigned int max
)
394 while (bref
< p
+max
) {
395 blk
= le32_to_cpu(*bref
++);
397 unlikely(!ext4_data_block_valid(EXT4_SB(inode
->i_sb
),
399 ext4_error(inode
->i_sb
, function
,
400 "invalid block reference %u "
401 "in inode #%lu", blk
, inode
->i_ino
);
409 #define ext4_check_indirect_blockref(inode, bh) \
410 __ext4_check_blockref(__func__, inode, (__le32 *)(bh)->b_data, \
411 EXT4_ADDR_PER_BLOCK((inode)->i_sb))
413 #define ext4_check_inode_blockref(inode) \
414 __ext4_check_blockref(__func__, inode, EXT4_I(inode)->i_data, \
418 * ext4_get_branch - read the chain of indirect blocks leading to data
419 * @inode: inode in question
420 * @depth: depth of the chain (1 - direct pointer, etc.)
421 * @offsets: offsets of pointers in inode/indirect blocks
422 * @chain: place to store the result
423 * @err: here we store the error value
425 * Function fills the array of triples <key, p, bh> and returns %NULL
426 * if everything went OK or the pointer to the last filled triple
427 * (incomplete one) otherwise. Upon the return chain[i].key contains
428 * the number of (i+1)-th block in the chain (as it is stored in memory,
429 * i.e. little-endian 32-bit), chain[i].p contains the address of that
430 * number (it points into struct inode for i==0 and into the bh->b_data
431 * for i>0) and chain[i].bh points to the buffer_head of i-th indirect
432 * block for i>0 and NULL for i==0. In other words, it holds the block
433 * numbers of the chain, addresses they were taken from (and where we can
434 * verify that chain did not change) and buffer_heads hosting these
437 * Function stops when it stumbles upon zero pointer (absent block)
438 * (pointer to last triple returned, *@err == 0)
439 * or when it gets an IO error reading an indirect block
440 * (ditto, *@err == -EIO)
441 * or when it reads all @depth-1 indirect blocks successfully and finds
442 * the whole chain, all way to the data (returns %NULL, *err == 0).
444 * Need to be called with
445 * down_read(&EXT4_I(inode)->i_data_sem)
447 static Indirect
*ext4_get_branch(struct inode
*inode
, int depth
,
448 ext4_lblk_t
*offsets
,
449 Indirect chain
[4], int *err
)
451 struct super_block
*sb
= inode
->i_sb
;
453 struct buffer_head
*bh
;
456 /* i_data is not going away, no lock needed */
457 add_chain(chain
, NULL
, EXT4_I(inode
)->i_data
+ *offsets
);
461 bh
= sb_getblk(sb
, le32_to_cpu(p
->key
));
465 if (!bh_uptodate_or_lock(bh
)) {
466 if (bh_submit_read(bh
) < 0) {
470 /* validate block references */
471 if (ext4_check_indirect_blockref(inode
, bh
)) {
477 add_chain(++p
, bh
, (__le32
*)bh
->b_data
+ *++offsets
);
491 * ext4_find_near - find a place for allocation with sufficient locality
493 * @ind: descriptor of indirect block.
495 * This function returns the preferred place for block allocation.
496 * It is used when heuristic for sequential allocation fails.
498 * + if there is a block to the left of our position - allocate near it.
499 * + if pointer will live in indirect block - allocate near that block.
500 * + if pointer will live in inode - allocate in the same
503 * In the latter case we colour the starting block by the callers PID to
504 * prevent it from clashing with concurrent allocations for a different inode
505 * in the same block group. The PID is used here so that functionally related
506 * files will be close-by on-disk.
508 * Caller must make sure that @ind is valid and will stay that way.
510 static ext4_fsblk_t
ext4_find_near(struct inode
*inode
, Indirect
*ind
)
512 struct ext4_inode_info
*ei
= EXT4_I(inode
);
513 __le32
*start
= ind
->bh
? (__le32
*) ind
->bh
->b_data
: ei
->i_data
;
515 ext4_fsblk_t bg_start
;
516 ext4_fsblk_t last_block
;
517 ext4_grpblk_t colour
;
518 ext4_group_t block_group
;
519 int flex_size
= ext4_flex_bg_size(EXT4_SB(inode
->i_sb
));
521 /* Try to find previous block */
522 for (p
= ind
->p
- 1; p
>= start
; p
--) {
524 return le32_to_cpu(*p
);
527 /* No such thing, so let's try location of indirect block */
529 return ind
->bh
->b_blocknr
;
532 * It is going to be referred to from the inode itself? OK, just put it
533 * into the same cylinder group then.
535 block_group
= ei
->i_block_group
;
536 if (flex_size
>= EXT4_FLEX_SIZE_DIR_ALLOC_SCHEME
) {
537 block_group
&= ~(flex_size
-1);
538 if (S_ISREG(inode
->i_mode
))
541 bg_start
= ext4_group_first_block_no(inode
->i_sb
, block_group
);
542 last_block
= ext4_blocks_count(EXT4_SB(inode
->i_sb
)->s_es
) - 1;
545 * If we are doing delayed allocation, we don't need take
546 * colour into account.
548 if (test_opt(inode
->i_sb
, DELALLOC
))
551 if (bg_start
+ EXT4_BLOCKS_PER_GROUP(inode
->i_sb
) <= last_block
)
552 colour
= (current
->pid
% 16) *
553 (EXT4_BLOCKS_PER_GROUP(inode
->i_sb
) / 16);
555 colour
= (current
->pid
% 16) * ((last_block
- bg_start
) / 16);
556 return bg_start
+ colour
;
560 * ext4_find_goal - find a preferred place for allocation.
562 * @block: block we want
563 * @partial: pointer to the last triple within a chain
565 * Normally this function find the preferred place for block allocation,
567 * Because this is only used for non-extent files, we limit the block nr
570 static ext4_fsblk_t
ext4_find_goal(struct inode
*inode
, ext4_lblk_t block
,
576 * XXX need to get goal block from mballoc's data structures
579 goal
= ext4_find_near(inode
, partial
);
580 goal
= goal
& EXT4_MAX_BLOCK_FILE_PHYS
;
585 * ext4_blks_to_allocate: Look up the block map and count the number
586 * of direct blocks need to be allocated for the given branch.
588 * @branch: chain of indirect blocks
589 * @k: number of blocks need for indirect blocks
590 * @blks: number of data blocks to be mapped.
591 * @blocks_to_boundary: the offset in the indirect block
593 * return the total number of blocks to be allocate, including the
594 * direct and indirect blocks.
596 static int ext4_blks_to_allocate(Indirect
*branch
, int k
, unsigned int blks
,
597 int blocks_to_boundary
)
599 unsigned int count
= 0;
602 * Simple case, [t,d]Indirect block(s) has not allocated yet
603 * then it's clear blocks on that path have not allocated
606 /* right now we don't handle cross boundary allocation */
607 if (blks
< blocks_to_boundary
+ 1)
610 count
+= blocks_to_boundary
+ 1;
615 while (count
< blks
&& count
<= blocks_to_boundary
&&
616 le32_to_cpu(*(branch
[0].p
+ count
)) == 0) {
623 * ext4_alloc_blocks: multiple allocate blocks needed for a branch
624 * @indirect_blks: the number of blocks need to allocate for indirect
627 * @new_blocks: on return it will store the new block numbers for
628 * the indirect blocks(if needed) and the first direct block,
629 * @blks: on return it will store the total number of allocated
632 static int ext4_alloc_blocks(handle_t
*handle
, struct inode
*inode
,
633 ext4_lblk_t iblock
, ext4_fsblk_t goal
,
634 int indirect_blks
, int blks
,
635 ext4_fsblk_t new_blocks
[4], int *err
)
637 struct ext4_allocation_request ar
;
639 unsigned long count
= 0, blk_allocated
= 0;
641 ext4_fsblk_t current_block
= 0;
645 * Here we try to allocate the requested multiple blocks at once,
646 * on a best-effort basis.
647 * To build a branch, we should allocate blocks for
648 * the indirect blocks(if not allocated yet), and at least
649 * the first direct block of this branch. That's the
650 * minimum number of blocks need to allocate(required)
652 /* first we try to allocate the indirect blocks */
653 target
= indirect_blks
;
656 /* allocating blocks for indirect blocks and direct blocks */
657 current_block
= ext4_new_meta_blocks(handle
, inode
,
662 BUG_ON(current_block
+ count
> EXT4_MAX_BLOCK_FILE_PHYS
);
665 /* allocate blocks for indirect blocks */
666 while (index
< indirect_blks
&& count
) {
667 new_blocks
[index
++] = current_block
++;
672 * save the new block number
673 * for the first direct block
675 new_blocks
[index
] = current_block
;
676 printk(KERN_INFO
"%s returned more blocks than "
677 "requested\n", __func__
);
683 target
= blks
- count
;
684 blk_allocated
= count
;
687 /* Now allocate data blocks */
688 memset(&ar
, 0, sizeof(ar
));
693 if (S_ISREG(inode
->i_mode
))
694 /* enable in-core preallocation only for regular files */
695 ar
.flags
= EXT4_MB_HINT_DATA
;
697 current_block
= ext4_mb_new_blocks(handle
, &ar
, err
);
698 BUG_ON(current_block
+ ar
.len
> EXT4_MAX_BLOCK_FILE_PHYS
);
700 if (*err
&& (target
== blks
)) {
702 * if the allocation failed and we didn't allocate
708 if (target
== blks
) {
710 * save the new block number
711 * for the first direct block
713 new_blocks
[index
] = current_block
;
715 blk_allocated
+= ar
.len
;
718 /* total number of blocks allocated for direct blocks */
723 for (i
= 0; i
< index
; i
++)
724 ext4_free_blocks(handle
, inode
, new_blocks
[i
], 1, 0);
729 * ext4_alloc_branch - allocate and set up a chain of blocks.
731 * @indirect_blks: number of allocated indirect blocks
732 * @blks: number of allocated direct blocks
733 * @offsets: offsets (in the blocks) to store the pointers to next.
734 * @branch: place to store the chain in.
736 * This function allocates blocks, zeroes out all but the last one,
737 * links them into chain and (if we are synchronous) writes them to disk.
738 * In other words, it prepares a branch that can be spliced onto the
739 * inode. It stores the information about that chain in the branch[], in
740 * the same format as ext4_get_branch() would do. We are calling it after
741 * we had read the existing part of chain and partial points to the last
742 * triple of that (one with zero ->key). Upon the exit we have the same
743 * picture as after the successful ext4_get_block(), except that in one
744 * place chain is disconnected - *branch->p is still zero (we did not
745 * set the last link), but branch->key contains the number that should
746 * be placed into *branch->p to fill that gap.
748 * If allocation fails we free all blocks we've allocated (and forget
749 * their buffer_heads) and return the error value the from failed
750 * ext4_alloc_block() (normally -ENOSPC). Otherwise we set the chain
751 * as described above and return 0.
753 static int ext4_alloc_branch(handle_t
*handle
, struct inode
*inode
,
754 ext4_lblk_t iblock
, int indirect_blks
,
755 int *blks
, ext4_fsblk_t goal
,
756 ext4_lblk_t
*offsets
, Indirect
*branch
)
758 int blocksize
= inode
->i_sb
->s_blocksize
;
761 struct buffer_head
*bh
;
763 ext4_fsblk_t new_blocks
[4];
764 ext4_fsblk_t current_block
;
766 num
= ext4_alloc_blocks(handle
, inode
, iblock
, goal
, indirect_blks
,
767 *blks
, new_blocks
, &err
);
771 branch
[0].key
= cpu_to_le32(new_blocks
[0]);
773 * metadata blocks and data blocks are allocated.
775 for (n
= 1; n
<= indirect_blks
; n
++) {
777 * Get buffer_head for parent block, zero it out
778 * and set the pointer to new one, then send
781 bh
= sb_getblk(inode
->i_sb
, new_blocks
[n
-1]);
784 BUFFER_TRACE(bh
, "call get_create_access");
785 err
= ext4_journal_get_create_access(handle
, bh
);
787 /* Don't brelse(bh) here; it's done in
788 * ext4_journal_forget() below */
793 memset(bh
->b_data
, 0, blocksize
);
794 branch
[n
].p
= (__le32
*) bh
->b_data
+ offsets
[n
];
795 branch
[n
].key
= cpu_to_le32(new_blocks
[n
]);
796 *branch
[n
].p
= branch
[n
].key
;
797 if (n
== indirect_blks
) {
798 current_block
= new_blocks
[n
];
800 * End of chain, update the last new metablock of
801 * the chain to point to the new allocated
802 * data blocks numbers
804 for (i
= 1; i
< num
; i
++)
805 *(branch
[n
].p
+ i
) = cpu_to_le32(++current_block
);
807 BUFFER_TRACE(bh
, "marking uptodate");
808 set_buffer_uptodate(bh
);
811 BUFFER_TRACE(bh
, "call ext4_handle_dirty_metadata");
812 err
= ext4_handle_dirty_metadata(handle
, inode
, bh
);
819 /* Allocation failed, free what we already allocated */
820 for (i
= 1; i
<= n
; i
++) {
821 BUFFER_TRACE(branch
[i
].bh
, "call jbd2_journal_forget");
822 ext4_journal_forget(handle
, branch
[i
].bh
);
824 for (i
= 0; i
< indirect_blks
; i
++)
825 ext4_free_blocks(handle
, inode
, new_blocks
[i
], 1, 0);
827 ext4_free_blocks(handle
, inode
, new_blocks
[i
], num
, 0);
833 * ext4_splice_branch - splice the allocated branch onto inode.
835 * @block: (logical) number of block we are adding
836 * @chain: chain of indirect blocks (with a missing link - see
838 * @where: location of missing link
839 * @num: number of indirect blocks we are adding
840 * @blks: number of direct blocks we are adding
842 * This function fills the missing link and does all housekeeping needed in
843 * inode (->i_blocks, etc.). In case of success we end up with the full
844 * chain to new block and return 0.
846 static int ext4_splice_branch(handle_t
*handle
, struct inode
*inode
,
847 ext4_lblk_t block
, Indirect
*where
, int num
,
852 ext4_fsblk_t current_block
;
855 * If we're splicing into a [td]indirect block (as opposed to the
856 * inode) then we need to get write access to the [td]indirect block
860 BUFFER_TRACE(where
->bh
, "get_write_access");
861 err
= ext4_journal_get_write_access(handle
, where
->bh
);
867 *where
->p
= where
->key
;
870 * Update the host buffer_head or inode to point to more just allocated
871 * direct blocks blocks
873 if (num
== 0 && blks
> 1) {
874 current_block
= le32_to_cpu(where
->key
) + 1;
875 for (i
= 1; i
< blks
; i
++)
876 *(where
->p
+ i
) = cpu_to_le32(current_block
++);
879 /* We are done with atomic stuff, now do the rest of housekeeping */
880 /* had we spliced it onto indirect block? */
883 * If we spliced it onto an indirect block, we haven't
884 * altered the inode. Note however that if it is being spliced
885 * onto an indirect block at the very end of the file (the
886 * file is growing) then we *will* alter the inode to reflect
887 * the new i_size. But that is not done here - it is done in
888 * generic_commit_write->__mark_inode_dirty->ext4_dirty_inode.
890 jbd_debug(5, "splicing indirect only\n");
891 BUFFER_TRACE(where
->bh
, "call ext4_handle_dirty_metadata");
892 err
= ext4_handle_dirty_metadata(handle
, inode
, where
->bh
);
897 * OK, we spliced it into the inode itself on a direct block.
899 ext4_mark_inode_dirty(handle
, inode
);
900 jbd_debug(5, "splicing direct\n");
905 for (i
= 1; i
<= num
; i
++) {
906 BUFFER_TRACE(where
[i
].bh
, "call jbd2_journal_forget");
907 ext4_journal_forget(handle
, where
[i
].bh
);
908 ext4_free_blocks(handle
, inode
,
909 le32_to_cpu(where
[i
-1].key
), 1, 0);
911 ext4_free_blocks(handle
, inode
, le32_to_cpu(where
[num
].key
), blks
, 0);
917 * The ext4_ind_get_blocks() function handles non-extents inodes
918 * (i.e., using the traditional indirect/double-indirect i_blocks
919 * scheme) for ext4_get_blocks().
921 * Allocation strategy is simple: if we have to allocate something, we will
922 * have to go the whole way to leaf. So let's do it before attaching anything
923 * to tree, set linkage between the newborn blocks, write them if sync is
924 * required, recheck the path, free and repeat if check fails, otherwise
925 * set the last missing link (that will protect us from any truncate-generated
926 * removals - all blocks on the path are immune now) and possibly force the
927 * write on the parent block.
928 * That has a nice additional property: no special recovery from the failed
929 * allocations is needed - we simply release blocks and do not touch anything
930 * reachable from inode.
932 * `handle' can be NULL if create == 0.
934 * return > 0, # of blocks mapped or allocated.
935 * return = 0, if plain lookup failed.
936 * return < 0, error case.
938 * The ext4_ind_get_blocks() function should be called with
939 * down_write(&EXT4_I(inode)->i_data_sem) if allocating filesystem
940 * blocks (i.e., flags has EXT4_GET_BLOCKS_CREATE set) or
941 * down_read(&EXT4_I(inode)->i_data_sem) if not allocating file system
944 static int ext4_ind_get_blocks(handle_t
*handle
, struct inode
*inode
,
945 ext4_lblk_t iblock
, unsigned int maxblocks
,
946 struct buffer_head
*bh_result
,
950 ext4_lblk_t offsets
[4];
955 int blocks_to_boundary
= 0;
958 ext4_fsblk_t first_block
= 0;
960 J_ASSERT(!(EXT4_I(inode
)->i_flags
& EXT4_EXTENTS_FL
));
961 J_ASSERT(handle
!= NULL
|| (flags
& EXT4_GET_BLOCKS_CREATE
) == 0);
962 depth
= ext4_block_to_path(inode
, iblock
, offsets
,
963 &blocks_to_boundary
);
968 partial
= ext4_get_branch(inode
, depth
, offsets
, chain
, &err
);
970 /* Simplest case - block found, no allocation needed */
972 first_block
= le32_to_cpu(chain
[depth
- 1].key
);
973 clear_buffer_new(bh_result
);
976 while (count
< maxblocks
&& count
<= blocks_to_boundary
) {
979 blk
= le32_to_cpu(*(chain
[depth
-1].p
+ count
));
981 if (blk
== first_block
+ count
)
989 /* Next simple case - plain lookup or failed read of indirect block */
990 if ((flags
& EXT4_GET_BLOCKS_CREATE
) == 0 || err
== -EIO
)
994 * Okay, we need to do block allocation.
996 goal
= ext4_find_goal(inode
, iblock
, partial
);
998 /* the number of blocks need to allocate for [d,t]indirect blocks */
999 indirect_blks
= (chain
+ depth
) - partial
- 1;
1002 * Next look up the indirect map to count the totoal number of
1003 * direct blocks to allocate for this branch.
1005 count
= ext4_blks_to_allocate(partial
, indirect_blks
,
1006 maxblocks
, blocks_to_boundary
);
1008 * Block out ext4_truncate while we alter the tree
1010 err
= ext4_alloc_branch(handle
, inode
, iblock
, indirect_blks
,
1012 offsets
+ (partial
- chain
), partial
);
1015 * The ext4_splice_branch call will free and forget any buffers
1016 * on the new chain if there is a failure, but that risks using
1017 * up transaction credits, especially for bitmaps where the
1018 * credits cannot be returned. Can we handle this somehow? We
1019 * may need to return -EAGAIN upwards in the worst case. --sct
1022 err
= ext4_splice_branch(handle
, inode
, iblock
,
1023 partial
, indirect_blks
, count
);
1027 set_buffer_new(bh_result
);
1029 ext4_update_inode_fsync_trans(handle
, inode
, 1);
1031 map_bh(bh_result
, inode
->i_sb
, le32_to_cpu(chain
[depth
-1].key
));
1032 if (count
> blocks_to_boundary
)
1033 set_buffer_boundary(bh_result
);
1035 /* Clean up and exit */
1036 partial
= chain
+ depth
- 1; /* the whole chain */
1038 while (partial
> chain
) {
1039 BUFFER_TRACE(partial
->bh
, "call brelse");
1040 brelse(partial
->bh
);
1043 BUFFER_TRACE(bh_result
, "returned");
1049 qsize_t
*ext4_get_reserved_space(struct inode
*inode
)
1051 return &EXT4_I(inode
)->i_reserved_quota
;
1056 * Calculate the number of metadata blocks need to reserve
1057 * to allocate a new block at @lblocks for non extent file based file
1059 static int ext4_indirect_calc_metadata_amount(struct inode
*inode
,
1062 struct ext4_inode_info
*ei
= EXT4_I(inode
);
1063 int dind_mask
= EXT4_ADDR_PER_BLOCK(inode
->i_sb
) - 1;
1066 if (lblock
< EXT4_NDIR_BLOCKS
)
1069 lblock
-= EXT4_NDIR_BLOCKS
;
1071 if (ei
->i_da_metadata_calc_len
&&
1072 (lblock
& dind_mask
) == ei
->i_da_metadata_calc_last_lblock
) {
1073 ei
->i_da_metadata_calc_len
++;
1076 ei
->i_da_metadata_calc_last_lblock
= lblock
& dind_mask
;
1077 ei
->i_da_metadata_calc_len
= 1;
1078 blk_bits
= roundup_pow_of_two(lblock
+ 1);
1079 return (blk_bits
/ EXT4_ADDR_PER_BLOCK_BITS(inode
->i_sb
)) + 1;
1083 * Calculate the number of metadata blocks need to reserve
1084 * to allocate a block located at @lblock
1086 static int ext4_calc_metadata_amount(struct inode
*inode
, sector_t lblock
)
1088 if (EXT4_I(inode
)->i_flags
& EXT4_EXTENTS_FL
)
1089 return ext4_ext_calc_metadata_amount(inode
, lblock
);
1091 return ext4_indirect_calc_metadata_amount(inode
, lblock
);
1095 * Called with i_data_sem down, which is important since we can call
1096 * ext4_discard_preallocations() from here.
1098 void ext4_da_update_reserve_space(struct inode
*inode
,
1099 int used
, int quota_claim
)
1101 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
1102 struct ext4_inode_info
*ei
= EXT4_I(inode
);
1103 int mdb_free
= 0, allocated_meta_blocks
= 0;
1105 spin_lock(&ei
->i_block_reservation_lock
);
1106 if (unlikely(used
> ei
->i_reserved_data_blocks
)) {
1107 ext4_msg(inode
->i_sb
, KERN_NOTICE
, "%s: ino %lu, used %d "
1108 "with only %d reserved data blocks\n",
1109 __func__
, inode
->i_ino
, used
,
1110 ei
->i_reserved_data_blocks
);
1112 used
= ei
->i_reserved_data_blocks
;
1115 /* Update per-inode reservations */
1116 ei
->i_reserved_data_blocks
-= used
;
1117 used
+= ei
->i_allocated_meta_blocks
;
1118 ei
->i_reserved_meta_blocks
-= ei
->i_allocated_meta_blocks
;
1119 allocated_meta_blocks
= ei
->i_allocated_meta_blocks
;
1120 ei
->i_allocated_meta_blocks
= 0;
1121 percpu_counter_sub(&sbi
->s_dirtyblocks_counter
, used
);
1123 if (ei
->i_reserved_data_blocks
== 0) {
1125 * We can release all of the reserved metadata blocks
1126 * only when we have written all of the delayed
1127 * allocation blocks.
1129 mdb_free
= ei
->i_reserved_meta_blocks
;
1130 ei
->i_reserved_meta_blocks
= 0;
1131 ei
->i_da_metadata_calc_len
= 0;
1132 percpu_counter_sub(&sbi
->s_dirtyblocks_counter
, mdb_free
);
1134 spin_unlock(&EXT4_I(inode
)->i_block_reservation_lock
);
1136 /* Update quota subsystem */
1138 vfs_dq_claim_block(inode
, used
);
1140 vfs_dq_release_reservation_block(inode
, mdb_free
);
1143 * We did fallocate with an offset that is already delayed
1144 * allocated. So on delayed allocated writeback we should
1145 * not update the quota for allocated blocks. But then
1146 * converting an fallocate region to initialized region would
1147 * have caused a metadata allocation. So claim quota for
1150 if (allocated_meta_blocks
)
1151 vfs_dq_claim_block(inode
, allocated_meta_blocks
);
1152 vfs_dq_release_reservation_block(inode
, mdb_free
+ used
);
1156 * If we have done all the pending block allocations and if
1157 * there aren't any writers on the inode, we can discard the
1158 * inode's preallocations.
1160 if ((ei
->i_reserved_data_blocks
== 0) &&
1161 (atomic_read(&inode
->i_writecount
) == 0))
1162 ext4_discard_preallocations(inode
);
1165 static int check_block_validity(struct inode
*inode
, const char *msg
,
1166 sector_t logical
, sector_t phys
, int len
)
1168 if (!ext4_data_block_valid(EXT4_SB(inode
->i_sb
), phys
, len
)) {
1169 ext4_error(inode
->i_sb
, msg
,
1170 "inode #%lu logical block %llu mapped to %llu "
1171 "(size %d)", inode
->i_ino
,
1172 (unsigned long long) logical
,
1173 (unsigned long long) phys
, len
);
1180 * Return the number of contiguous dirty pages in a given inode
1181 * starting at page frame idx.
1183 static pgoff_t
ext4_num_dirty_pages(struct inode
*inode
, pgoff_t idx
,
1184 unsigned int max_pages
)
1186 struct address_space
*mapping
= inode
->i_mapping
;
1188 struct pagevec pvec
;
1190 int i
, nr_pages
, done
= 0;
1194 pagevec_init(&pvec
, 0);
1197 nr_pages
= pagevec_lookup_tag(&pvec
, mapping
, &index
,
1198 PAGECACHE_TAG_DIRTY
,
1199 (pgoff_t
)PAGEVEC_SIZE
);
1202 for (i
= 0; i
< nr_pages
; i
++) {
1203 struct page
*page
= pvec
.pages
[i
];
1204 struct buffer_head
*bh
, *head
;
1207 if (unlikely(page
->mapping
!= mapping
) ||
1209 PageWriteback(page
) ||
1210 page
->index
!= idx
) {
1215 if (page_has_buffers(page
)) {
1216 bh
= head
= page_buffers(page
);
1218 if (!buffer_delay(bh
) &&
1219 !buffer_unwritten(bh
))
1221 bh
= bh
->b_this_page
;
1222 } while (!done
&& (bh
!= head
));
1229 if (num
>= max_pages
)
1232 pagevec_release(&pvec
);
1238 * The ext4_get_blocks() function tries to look up the requested blocks,
1239 * and returns if the blocks are already mapped.
1241 * Otherwise it takes the write lock of the i_data_sem and allocate blocks
1242 * and store the allocated blocks in the result buffer head and mark it
1245 * If file type is extents based, it will call ext4_ext_get_blocks(),
1246 * Otherwise, call with ext4_ind_get_blocks() to handle indirect mapping
1249 * On success, it returns the number of blocks being mapped or allocate.
1250 * if create==0 and the blocks are pre-allocated and uninitialized block,
1251 * the result buffer head is unmapped. If the create ==1, it will make sure
1252 * the buffer head is mapped.
1254 * It returns 0 if plain look up failed (blocks have not been allocated), in
1255 * that casem, buffer head is unmapped
1257 * It returns the error in case of allocation failure.
1259 int ext4_get_blocks(handle_t
*handle
, struct inode
*inode
, sector_t block
,
1260 unsigned int max_blocks
, struct buffer_head
*bh
,
1265 clear_buffer_mapped(bh
);
1266 clear_buffer_unwritten(bh
);
1268 ext_debug("ext4_get_blocks(): inode %lu, flag %d, max_blocks %u,"
1269 "logical block %lu\n", inode
->i_ino
, flags
, max_blocks
,
1270 (unsigned long)block
);
1272 * Try to see if we can get the block without requesting a new
1273 * file system block.
1275 down_read((&EXT4_I(inode
)->i_data_sem
));
1276 if (EXT4_I(inode
)->i_flags
& EXT4_EXTENTS_FL
) {
1277 retval
= ext4_ext_get_blocks(handle
, inode
, block
, max_blocks
,
1280 retval
= ext4_ind_get_blocks(handle
, inode
, block
, max_blocks
,
1283 up_read((&EXT4_I(inode
)->i_data_sem
));
1285 if (retval
> 0 && buffer_mapped(bh
)) {
1286 int ret
= check_block_validity(inode
, "file system corruption",
1287 block
, bh
->b_blocknr
, retval
);
1292 /* If it is only a block(s) look up */
1293 if ((flags
& EXT4_GET_BLOCKS_CREATE
) == 0)
1297 * Returns if the blocks have already allocated
1299 * Note that if blocks have been preallocated
1300 * ext4_ext_get_block() returns th create = 0
1301 * with buffer head unmapped.
1303 if (retval
> 0 && buffer_mapped(bh
))
1307 * When we call get_blocks without the create flag, the
1308 * BH_Unwritten flag could have gotten set if the blocks
1309 * requested were part of a uninitialized extent. We need to
1310 * clear this flag now that we are committed to convert all or
1311 * part of the uninitialized extent to be an initialized
1312 * extent. This is because we need to avoid the combination
1313 * of BH_Unwritten and BH_Mapped flags being simultaneously
1314 * set on the buffer_head.
1316 clear_buffer_unwritten(bh
);
1319 * New blocks allocate and/or writing to uninitialized extent
1320 * will possibly result in updating i_data, so we take
1321 * the write lock of i_data_sem, and call get_blocks()
1322 * with create == 1 flag.
1324 down_write((&EXT4_I(inode
)->i_data_sem
));
1327 * if the caller is from delayed allocation writeout path
1328 * we have already reserved fs blocks for allocation
1329 * let the underlying get_block() function know to
1330 * avoid double accounting
1332 if (flags
& EXT4_GET_BLOCKS_DELALLOC_RESERVE
)
1333 EXT4_I(inode
)->i_delalloc_reserved_flag
= 1;
1335 * We need to check for EXT4 here because migrate
1336 * could have changed the inode type in between
1338 if (EXT4_I(inode
)->i_flags
& EXT4_EXTENTS_FL
) {
1339 retval
= ext4_ext_get_blocks(handle
, inode
, block
, max_blocks
,
1342 retval
= ext4_ind_get_blocks(handle
, inode
, block
,
1343 max_blocks
, bh
, flags
);
1345 if (retval
> 0 && buffer_new(bh
)) {
1347 * We allocated new blocks which will result in
1348 * i_data's format changing. Force the migrate
1349 * to fail by clearing migrate flags
1351 ext4_clear_inode_state(inode
, EXT4_STATE_EXT_MIGRATE
);
1355 * Update reserved blocks/metadata blocks after successful
1356 * block allocation which had been deferred till now. We don't
1357 * support fallocate for non extent files. So we can update
1358 * reserve space here.
1361 (flags
& EXT4_GET_BLOCKS_DELALLOC_RESERVE
))
1362 ext4_da_update_reserve_space(inode
, retval
, 1);
1364 if (flags
& EXT4_GET_BLOCKS_DELALLOC_RESERVE
)
1365 EXT4_I(inode
)->i_delalloc_reserved_flag
= 0;
1367 up_write((&EXT4_I(inode
)->i_data_sem
));
1368 if (retval
> 0 && buffer_mapped(bh
)) {
1369 int ret
= check_block_validity(inode
, "file system "
1370 "corruption after allocation",
1371 block
, bh
->b_blocknr
, retval
);
1378 /* Maximum number of blocks we map for direct IO at once. */
1379 #define DIO_MAX_BLOCKS 4096
1381 int ext4_get_block(struct inode
*inode
, sector_t iblock
,
1382 struct buffer_head
*bh_result
, int create
)
1384 handle_t
*handle
= ext4_journal_current_handle();
1385 int ret
= 0, started
= 0;
1386 unsigned max_blocks
= bh_result
->b_size
>> inode
->i_blkbits
;
1389 if (create
&& !handle
) {
1390 /* Direct IO write... */
1391 if (max_blocks
> DIO_MAX_BLOCKS
)
1392 max_blocks
= DIO_MAX_BLOCKS
;
1393 dio_credits
= ext4_chunk_trans_blocks(inode
, max_blocks
);
1394 handle
= ext4_journal_start(inode
, dio_credits
);
1395 if (IS_ERR(handle
)) {
1396 ret
= PTR_ERR(handle
);
1402 ret
= ext4_get_blocks(handle
, inode
, iblock
, max_blocks
, bh_result
,
1403 create
? EXT4_GET_BLOCKS_CREATE
: 0);
1405 bh_result
->b_size
= (ret
<< inode
->i_blkbits
);
1409 ext4_journal_stop(handle
);
1415 * `handle' can be NULL if create is zero
1417 struct buffer_head
*ext4_getblk(handle_t
*handle
, struct inode
*inode
,
1418 ext4_lblk_t block
, int create
, int *errp
)
1420 struct buffer_head dummy
;
1424 J_ASSERT(handle
!= NULL
|| create
== 0);
1427 dummy
.b_blocknr
= -1000;
1428 buffer_trace_init(&dummy
.b_history
);
1430 flags
|= EXT4_GET_BLOCKS_CREATE
;
1431 err
= ext4_get_blocks(handle
, inode
, block
, 1, &dummy
, flags
);
1433 * ext4_get_blocks() returns number of blocks mapped. 0 in
1442 if (!err
&& buffer_mapped(&dummy
)) {
1443 struct buffer_head
*bh
;
1444 bh
= sb_getblk(inode
->i_sb
, dummy
.b_blocknr
);
1449 if (buffer_new(&dummy
)) {
1450 J_ASSERT(create
!= 0);
1451 J_ASSERT(handle
!= NULL
);
1454 * Now that we do not always journal data, we should
1455 * keep in mind whether this should always journal the
1456 * new buffer as metadata. For now, regular file
1457 * writes use ext4_get_block instead, so it's not a
1461 BUFFER_TRACE(bh
, "call get_create_access");
1462 fatal
= ext4_journal_get_create_access(handle
, bh
);
1463 if (!fatal
&& !buffer_uptodate(bh
)) {
1464 memset(bh
->b_data
, 0, inode
->i_sb
->s_blocksize
);
1465 set_buffer_uptodate(bh
);
1468 BUFFER_TRACE(bh
, "call ext4_handle_dirty_metadata");
1469 err
= ext4_handle_dirty_metadata(handle
, inode
, bh
);
1473 BUFFER_TRACE(bh
, "not a new buffer");
1486 struct buffer_head
*ext4_bread(handle_t
*handle
, struct inode
*inode
,
1487 ext4_lblk_t block
, int create
, int *err
)
1489 struct buffer_head
*bh
;
1491 bh
= ext4_getblk(handle
, inode
, block
, create
, err
);
1494 if (buffer_uptodate(bh
))
1496 ll_rw_block(READ_META
, 1, &bh
);
1498 if (buffer_uptodate(bh
))
1505 static int walk_page_buffers(handle_t
*handle
,
1506 struct buffer_head
*head
,
1510 int (*fn
)(handle_t
*handle
,
1511 struct buffer_head
*bh
))
1513 struct buffer_head
*bh
;
1514 unsigned block_start
, block_end
;
1515 unsigned blocksize
= head
->b_size
;
1517 struct buffer_head
*next
;
1519 for (bh
= head
, block_start
= 0;
1520 ret
== 0 && (bh
!= head
|| !block_start
);
1521 block_start
= block_end
, bh
= next
) {
1522 next
= bh
->b_this_page
;
1523 block_end
= block_start
+ blocksize
;
1524 if (block_end
<= from
|| block_start
>= to
) {
1525 if (partial
&& !buffer_uptodate(bh
))
1529 err
= (*fn
)(handle
, bh
);
1537 * To preserve ordering, it is essential that the hole instantiation and
1538 * the data write be encapsulated in a single transaction. We cannot
1539 * close off a transaction and start a new one between the ext4_get_block()
1540 * and the commit_write(). So doing the jbd2_journal_start at the start of
1541 * prepare_write() is the right place.
1543 * Also, this function can nest inside ext4_writepage() ->
1544 * block_write_full_page(). In that case, we *know* that ext4_writepage()
1545 * has generated enough buffer credits to do the whole page. So we won't
1546 * block on the journal in that case, which is good, because the caller may
1549 * By accident, ext4 can be reentered when a transaction is open via
1550 * quota file writes. If we were to commit the transaction while thus
1551 * reentered, there can be a deadlock - we would be holding a quota
1552 * lock, and the commit would never complete if another thread had a
1553 * transaction open and was blocking on the quota lock - a ranking
1556 * So what we do is to rely on the fact that jbd2_journal_stop/journal_start
1557 * will _not_ run commit under these circumstances because handle->h_ref
1558 * is elevated. We'll still have enough credits for the tiny quotafile
1561 static int do_journal_get_write_access(handle_t
*handle
,
1562 struct buffer_head
*bh
)
1564 if (!buffer_mapped(bh
) || buffer_freed(bh
))
1566 return ext4_journal_get_write_access(handle
, bh
);
1570 * Truncate blocks that were not used by write. We have to truncate the
1571 * pagecache as well so that corresponding buffers get properly unmapped.
1573 static void ext4_truncate_failed_write(struct inode
*inode
)
1575 truncate_inode_pages(inode
->i_mapping
, inode
->i_size
);
1576 ext4_truncate(inode
);
1579 static int ext4_write_begin(struct file
*file
, struct address_space
*mapping
,
1580 loff_t pos
, unsigned len
, unsigned flags
,
1581 struct page
**pagep
, void **fsdata
)
1583 struct inode
*inode
= mapping
->host
;
1584 int ret
, needed_blocks
;
1591 trace_ext4_write_begin(inode
, pos
, len
, flags
);
1593 * Reserve one block more for addition to orphan list in case
1594 * we allocate blocks but write fails for some reason
1596 needed_blocks
= ext4_writepage_trans_blocks(inode
) + 1;
1597 index
= pos
>> PAGE_CACHE_SHIFT
;
1598 from
= pos
& (PAGE_CACHE_SIZE
- 1);
1602 handle
= ext4_journal_start(inode
, needed_blocks
);
1603 if (IS_ERR(handle
)) {
1604 ret
= PTR_ERR(handle
);
1608 /* We cannot recurse into the filesystem as the transaction is already
1610 flags
|= AOP_FLAG_NOFS
;
1612 page
= grab_cache_page_write_begin(mapping
, index
, flags
);
1614 ext4_journal_stop(handle
);
1620 ret
= block_write_begin(file
, mapping
, pos
, len
, flags
, pagep
, fsdata
,
1623 if (!ret
&& ext4_should_journal_data(inode
)) {
1624 ret
= walk_page_buffers(handle
, page_buffers(page
),
1625 from
, to
, NULL
, do_journal_get_write_access
);
1630 page_cache_release(page
);
1632 * block_write_begin may have instantiated a few blocks
1633 * outside i_size. Trim these off again. Don't need
1634 * i_size_read because we hold i_mutex.
1636 * Add inode to orphan list in case we crash before
1639 if (pos
+ len
> inode
->i_size
&& ext4_can_truncate(inode
))
1640 ext4_orphan_add(handle
, inode
);
1642 ext4_journal_stop(handle
);
1643 if (pos
+ len
> inode
->i_size
) {
1644 ext4_truncate_failed_write(inode
);
1646 * If truncate failed early the inode might
1647 * still be on the orphan list; we need to
1648 * make sure the inode is removed from the
1649 * orphan list in that case.
1652 ext4_orphan_del(NULL
, inode
);
1656 if (ret
== -ENOSPC
&& ext4_should_retry_alloc(inode
->i_sb
, &retries
))
1662 /* For write_end() in data=journal mode */
1663 static int write_end_fn(handle_t
*handle
, struct buffer_head
*bh
)
1665 if (!buffer_mapped(bh
) || buffer_freed(bh
))
1667 set_buffer_uptodate(bh
);
1668 return ext4_handle_dirty_metadata(handle
, NULL
, bh
);
1671 static int ext4_generic_write_end(struct file
*file
,
1672 struct address_space
*mapping
,
1673 loff_t pos
, unsigned len
, unsigned copied
,
1674 struct page
*page
, void *fsdata
)
1676 int i_size_changed
= 0;
1677 struct inode
*inode
= mapping
->host
;
1678 handle_t
*handle
= ext4_journal_current_handle();
1680 copied
= block_write_end(file
, mapping
, pos
, len
, copied
, page
, fsdata
);
1683 * No need to use i_size_read() here, the i_size
1684 * cannot change under us because we hold i_mutex.
1686 * But it's important to update i_size while still holding page lock:
1687 * page writeout could otherwise come in and zero beyond i_size.
1689 if (pos
+ copied
> inode
->i_size
) {
1690 i_size_write(inode
, pos
+ copied
);
1694 if (pos
+ copied
> EXT4_I(inode
)->i_disksize
) {
1695 /* We need to mark inode dirty even if
1696 * new_i_size is less that inode->i_size
1697 * bu greater than i_disksize.(hint delalloc)
1699 ext4_update_i_disksize(inode
, (pos
+ copied
));
1703 page_cache_release(page
);
1706 * Don't mark the inode dirty under page lock. First, it unnecessarily
1707 * makes the holding time of page lock longer. Second, it forces lock
1708 * ordering of page lock and transaction start for journaling
1712 ext4_mark_inode_dirty(handle
, inode
);
1718 * We need to pick up the new inode size which generic_commit_write gave us
1719 * `file' can be NULL - eg, when called from page_symlink().
1721 * ext4 never places buffers on inode->i_mapping->private_list. metadata
1722 * buffers are managed internally.
1724 static int ext4_ordered_write_end(struct file
*file
,
1725 struct address_space
*mapping
,
1726 loff_t pos
, unsigned len
, unsigned copied
,
1727 struct page
*page
, void *fsdata
)
1729 handle_t
*handle
= ext4_journal_current_handle();
1730 struct inode
*inode
= mapping
->host
;
1733 trace_ext4_ordered_write_end(inode
, pos
, len
, copied
);
1734 ret
= ext4_jbd2_file_inode(handle
, inode
);
1737 ret2
= ext4_generic_write_end(file
, mapping
, pos
, len
, copied
,
1740 if (pos
+ len
> inode
->i_size
&& ext4_can_truncate(inode
))
1741 /* if we have allocated more blocks and copied
1742 * less. We will have blocks allocated outside
1743 * inode->i_size. So truncate them
1745 ext4_orphan_add(handle
, inode
);
1749 ret2
= ext4_journal_stop(handle
);
1753 if (pos
+ len
> inode
->i_size
) {
1754 ext4_truncate_failed_write(inode
);
1756 * If truncate failed early the inode might still be
1757 * on the orphan list; we need to make sure the inode
1758 * is removed from the orphan list in that case.
1761 ext4_orphan_del(NULL
, inode
);
1765 return ret
? ret
: copied
;
1768 static int ext4_writeback_write_end(struct file
*file
,
1769 struct address_space
*mapping
,
1770 loff_t pos
, unsigned len
, unsigned copied
,
1771 struct page
*page
, void *fsdata
)
1773 handle_t
*handle
= ext4_journal_current_handle();
1774 struct inode
*inode
= mapping
->host
;
1777 trace_ext4_writeback_write_end(inode
, pos
, len
, copied
);
1778 ret2
= ext4_generic_write_end(file
, mapping
, pos
, len
, copied
,
1781 if (pos
+ len
> inode
->i_size
&& ext4_can_truncate(inode
))
1782 /* if we have allocated more blocks and copied
1783 * less. We will have blocks allocated outside
1784 * inode->i_size. So truncate them
1786 ext4_orphan_add(handle
, inode
);
1791 ret2
= ext4_journal_stop(handle
);
1795 if (pos
+ len
> inode
->i_size
) {
1796 ext4_truncate_failed_write(inode
);
1798 * If truncate failed early the inode might still be
1799 * on the orphan list; we need to make sure the inode
1800 * is removed from the orphan list in that case.
1803 ext4_orphan_del(NULL
, inode
);
1806 return ret
? ret
: copied
;
1809 static int ext4_journalled_write_end(struct file
*file
,
1810 struct address_space
*mapping
,
1811 loff_t pos
, unsigned len
, unsigned copied
,
1812 struct page
*page
, void *fsdata
)
1814 handle_t
*handle
= ext4_journal_current_handle();
1815 struct inode
*inode
= mapping
->host
;
1821 trace_ext4_journalled_write_end(inode
, pos
, len
, copied
);
1822 from
= pos
& (PAGE_CACHE_SIZE
- 1);
1826 if (!PageUptodate(page
))
1828 page_zero_new_buffers(page
, from
+copied
, to
);
1831 ret
= walk_page_buffers(handle
, page_buffers(page
), from
,
1832 to
, &partial
, write_end_fn
);
1834 SetPageUptodate(page
);
1835 new_i_size
= pos
+ copied
;
1836 if (new_i_size
> inode
->i_size
)
1837 i_size_write(inode
, pos
+copied
);
1838 ext4_set_inode_state(inode
, EXT4_STATE_JDATA
);
1839 if (new_i_size
> EXT4_I(inode
)->i_disksize
) {
1840 ext4_update_i_disksize(inode
, new_i_size
);
1841 ret2
= ext4_mark_inode_dirty(handle
, inode
);
1847 page_cache_release(page
);
1848 if (pos
+ len
> inode
->i_size
&& ext4_can_truncate(inode
))
1849 /* if we have allocated more blocks and copied
1850 * less. We will have blocks allocated outside
1851 * inode->i_size. So truncate them
1853 ext4_orphan_add(handle
, inode
);
1855 ret2
= ext4_journal_stop(handle
);
1858 if (pos
+ len
> inode
->i_size
) {
1859 ext4_truncate_failed_write(inode
);
1861 * If truncate failed early the inode might still be
1862 * on the orphan list; we need to make sure the inode
1863 * is removed from the orphan list in that case.
1866 ext4_orphan_del(NULL
, inode
);
1869 return ret
? ret
: copied
;
1873 * Reserve a single block located at lblock
1875 static int ext4_da_reserve_space(struct inode
*inode
, sector_t lblock
)
1878 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
1879 struct ext4_inode_info
*ei
= EXT4_I(inode
);
1880 unsigned long md_needed
, md_reserved
;
1883 * recalculate the amount of metadata blocks to reserve
1884 * in order to allocate nrblocks
1885 * worse case is one extent per block
1888 spin_lock(&ei
->i_block_reservation_lock
);
1889 md_reserved
= ei
->i_reserved_meta_blocks
;
1890 md_needed
= ext4_calc_metadata_amount(inode
, lblock
);
1891 spin_unlock(&ei
->i_block_reservation_lock
);
1894 * Make quota reservation here to prevent quota overflow
1895 * later. Real quota accounting is done at pages writeout
1898 if (vfs_dq_reserve_block(inode
, md_needed
+ 1))
1901 if (ext4_claim_free_blocks(sbi
, md_needed
+ 1)) {
1902 vfs_dq_release_reservation_block(inode
, md_needed
+ 1);
1903 if (ext4_should_retry_alloc(inode
->i_sb
, &retries
)) {
1909 spin_lock(&ei
->i_block_reservation_lock
);
1910 ei
->i_reserved_data_blocks
++;
1911 ei
->i_reserved_meta_blocks
+= md_needed
;
1912 spin_unlock(&ei
->i_block_reservation_lock
);
1914 return 0; /* success */
1917 static void ext4_da_release_space(struct inode
*inode
, int to_free
)
1919 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
1920 struct ext4_inode_info
*ei
= EXT4_I(inode
);
1923 return; /* Nothing to release, exit */
1925 spin_lock(&EXT4_I(inode
)->i_block_reservation_lock
);
1927 if (unlikely(to_free
> ei
->i_reserved_data_blocks
)) {
1929 * if there aren't enough reserved blocks, then the
1930 * counter is messed up somewhere. Since this
1931 * function is called from invalidate page, it's
1932 * harmless to return without any action.
1934 ext4_msg(inode
->i_sb
, KERN_NOTICE
, "ext4_da_release_space: "
1935 "ino %lu, to_free %d with only %d reserved "
1936 "data blocks\n", inode
->i_ino
, to_free
,
1937 ei
->i_reserved_data_blocks
);
1939 to_free
= ei
->i_reserved_data_blocks
;
1941 ei
->i_reserved_data_blocks
-= to_free
;
1943 if (ei
->i_reserved_data_blocks
== 0) {
1945 * We can release all of the reserved metadata blocks
1946 * only when we have written all of the delayed
1947 * allocation blocks.
1949 to_free
+= ei
->i_reserved_meta_blocks
;
1950 ei
->i_reserved_meta_blocks
= 0;
1951 ei
->i_da_metadata_calc_len
= 0;
1954 /* update fs dirty blocks counter */
1955 percpu_counter_sub(&sbi
->s_dirtyblocks_counter
, to_free
);
1957 spin_unlock(&EXT4_I(inode
)->i_block_reservation_lock
);
1959 vfs_dq_release_reservation_block(inode
, to_free
);
1962 static void ext4_da_page_release_reservation(struct page
*page
,
1963 unsigned long offset
)
1966 struct buffer_head
*head
, *bh
;
1967 unsigned int curr_off
= 0;
1969 head
= page_buffers(page
);
1972 unsigned int next_off
= curr_off
+ bh
->b_size
;
1974 if ((offset
<= curr_off
) && (buffer_delay(bh
))) {
1976 clear_buffer_delay(bh
);
1978 curr_off
= next_off
;
1979 } while ((bh
= bh
->b_this_page
) != head
);
1980 ext4_da_release_space(page
->mapping
->host
, to_release
);
1984 * Delayed allocation stuff
1988 * mpage_da_submit_io - walks through extent of pages and try to write
1989 * them with writepage() call back
1991 * @mpd->inode: inode
1992 * @mpd->first_page: first page of the extent
1993 * @mpd->next_page: page after the last page of the extent
1995 * By the time mpage_da_submit_io() is called we expect all blocks
1996 * to be allocated. this may be wrong if allocation failed.
1998 * As pages are already locked by write_cache_pages(), we can't use it
2000 static int mpage_da_submit_io(struct mpage_da_data
*mpd
)
2003 struct pagevec pvec
;
2004 unsigned long index
, end
;
2005 int ret
= 0, err
, nr_pages
, i
;
2006 struct inode
*inode
= mpd
->inode
;
2007 struct address_space
*mapping
= inode
->i_mapping
;
2009 BUG_ON(mpd
->next_page
<= mpd
->first_page
);
2011 * We need to start from the first_page to the next_page - 1
2012 * to make sure we also write the mapped dirty buffer_heads.
2013 * If we look at mpd->b_blocknr we would only be looking
2014 * at the currently mapped buffer_heads.
2016 index
= mpd
->first_page
;
2017 end
= mpd
->next_page
- 1;
2019 pagevec_init(&pvec
, 0);
2020 while (index
<= end
) {
2021 nr_pages
= pagevec_lookup(&pvec
, mapping
, index
, PAGEVEC_SIZE
);
2024 for (i
= 0; i
< nr_pages
; i
++) {
2025 struct page
*page
= pvec
.pages
[i
];
2027 index
= page
->index
;
2032 BUG_ON(!PageLocked(page
));
2033 BUG_ON(PageWriteback(page
));
2035 pages_skipped
= mpd
->wbc
->pages_skipped
;
2036 err
= mapping
->a_ops
->writepage(page
, mpd
->wbc
);
2037 if (!err
&& (pages_skipped
== mpd
->wbc
->pages_skipped
))
2039 * have successfully written the page
2040 * without skipping the same
2042 mpd
->pages_written
++;
2044 * In error case, we have to continue because
2045 * remaining pages are still locked
2046 * XXX: unlock and re-dirty them?
2051 pagevec_release(&pvec
);
2057 * mpage_put_bnr_to_bhs - walk blocks and assign them actual numbers
2059 * @mpd->inode - inode to walk through
2060 * @exbh->b_blocknr - first block on a disk
2061 * @exbh->b_size - amount of space in bytes
2062 * @logical - first logical block to start assignment with
2064 * the function goes through all passed space and put actual disk
2065 * block numbers into buffer heads, dropping BH_Delay and BH_Unwritten
2067 static void mpage_put_bnr_to_bhs(struct mpage_da_data
*mpd
, sector_t logical
,
2068 struct buffer_head
*exbh
)
2070 struct inode
*inode
= mpd
->inode
;
2071 struct address_space
*mapping
= inode
->i_mapping
;
2072 int blocks
= exbh
->b_size
>> inode
->i_blkbits
;
2073 sector_t pblock
= exbh
->b_blocknr
, cur_logical
;
2074 struct buffer_head
*head
, *bh
;
2076 struct pagevec pvec
;
2079 index
= logical
>> (PAGE_CACHE_SHIFT
- inode
->i_blkbits
);
2080 end
= (logical
+ blocks
- 1) >> (PAGE_CACHE_SHIFT
- inode
->i_blkbits
);
2081 cur_logical
= index
<< (PAGE_CACHE_SHIFT
- inode
->i_blkbits
);
2083 pagevec_init(&pvec
, 0);
2085 while (index
<= end
) {
2086 /* XXX: optimize tail */
2087 nr_pages
= pagevec_lookup(&pvec
, mapping
, index
, PAGEVEC_SIZE
);
2090 for (i
= 0; i
< nr_pages
; i
++) {
2091 struct page
*page
= pvec
.pages
[i
];
2093 index
= page
->index
;
2098 BUG_ON(!PageLocked(page
));
2099 BUG_ON(PageWriteback(page
));
2100 BUG_ON(!page_has_buffers(page
));
2102 bh
= page_buffers(page
);
2105 /* skip blocks out of the range */
2107 if (cur_logical
>= logical
)
2110 } while ((bh
= bh
->b_this_page
) != head
);
2113 if (cur_logical
>= logical
+ blocks
)
2116 if (buffer_delay(bh
) ||
2117 buffer_unwritten(bh
)) {
2119 BUG_ON(bh
->b_bdev
!= inode
->i_sb
->s_bdev
);
2121 if (buffer_delay(bh
)) {
2122 clear_buffer_delay(bh
);
2123 bh
->b_blocknr
= pblock
;
2126 * unwritten already should have
2127 * blocknr assigned. Verify that
2129 clear_buffer_unwritten(bh
);
2130 BUG_ON(bh
->b_blocknr
!= pblock
);
2133 } else if (buffer_mapped(bh
))
2134 BUG_ON(bh
->b_blocknr
!= pblock
);
2138 } while ((bh
= bh
->b_this_page
) != head
);
2140 pagevec_release(&pvec
);
2146 * __unmap_underlying_blocks - just a helper function to unmap
2147 * set of blocks described by @bh
2149 static inline void __unmap_underlying_blocks(struct inode
*inode
,
2150 struct buffer_head
*bh
)
2152 struct block_device
*bdev
= inode
->i_sb
->s_bdev
;
2155 blocks
= bh
->b_size
>> inode
->i_blkbits
;
2156 for (i
= 0; i
< blocks
; i
++)
2157 unmap_underlying_metadata(bdev
, bh
->b_blocknr
+ i
);
2160 static void ext4_da_block_invalidatepages(struct mpage_da_data
*mpd
,
2161 sector_t logical
, long blk_cnt
)
2165 struct pagevec pvec
;
2166 struct inode
*inode
= mpd
->inode
;
2167 struct address_space
*mapping
= inode
->i_mapping
;
2169 index
= logical
>> (PAGE_CACHE_SHIFT
- inode
->i_blkbits
);
2170 end
= (logical
+ blk_cnt
- 1) >>
2171 (PAGE_CACHE_SHIFT
- inode
->i_blkbits
);
2172 while (index
<= end
) {
2173 nr_pages
= pagevec_lookup(&pvec
, mapping
, index
, PAGEVEC_SIZE
);
2176 for (i
= 0; i
< nr_pages
; i
++) {
2177 struct page
*page
= pvec
.pages
[i
];
2178 index
= page
->index
;
2183 BUG_ON(!PageLocked(page
));
2184 BUG_ON(PageWriteback(page
));
2185 block_invalidatepage(page
, 0);
2186 ClearPageUptodate(page
);
2193 static void ext4_print_free_blocks(struct inode
*inode
)
2195 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
2196 printk(KERN_CRIT
"Total free blocks count %lld\n",
2197 ext4_count_free_blocks(inode
->i_sb
));
2198 printk(KERN_CRIT
"Free/Dirty block details\n");
2199 printk(KERN_CRIT
"free_blocks=%lld\n",
2200 (long long) percpu_counter_sum(&sbi
->s_freeblocks_counter
));
2201 printk(KERN_CRIT
"dirty_blocks=%lld\n",
2202 (long long) percpu_counter_sum(&sbi
->s_dirtyblocks_counter
));
2203 printk(KERN_CRIT
"Block reservation details\n");
2204 printk(KERN_CRIT
"i_reserved_data_blocks=%u\n",
2205 EXT4_I(inode
)->i_reserved_data_blocks
);
2206 printk(KERN_CRIT
"i_reserved_meta_blocks=%u\n",
2207 EXT4_I(inode
)->i_reserved_meta_blocks
);
2212 * mpage_da_map_blocks - go through given space
2214 * @mpd - bh describing space
2216 * The function skips space we know is already mapped to disk blocks.
2219 static int mpage_da_map_blocks(struct mpage_da_data
*mpd
)
2221 int err
, blks
, get_blocks_flags
;
2222 struct buffer_head
new;
2223 sector_t next
= mpd
->b_blocknr
;
2224 unsigned max_blocks
= mpd
->b_size
>> mpd
->inode
->i_blkbits
;
2225 loff_t disksize
= EXT4_I(mpd
->inode
)->i_disksize
;
2226 handle_t
*handle
= NULL
;
2229 * We consider only non-mapped and non-allocated blocks
2231 if ((mpd
->b_state
& (1 << BH_Mapped
)) &&
2232 !(mpd
->b_state
& (1 << BH_Delay
)) &&
2233 !(mpd
->b_state
& (1 << BH_Unwritten
)))
2237 * If we didn't accumulate anything to write simply return
2242 handle
= ext4_journal_current_handle();
2246 * Call ext4_get_blocks() to allocate any delayed allocation
2247 * blocks, or to convert an uninitialized extent to be
2248 * initialized (in the case where we have written into
2249 * one or more preallocated blocks).
2251 * We pass in the magic EXT4_GET_BLOCKS_DELALLOC_RESERVE to
2252 * indicate that we are on the delayed allocation path. This
2253 * affects functions in many different parts of the allocation
2254 * call path. This flag exists primarily because we don't
2255 * want to change *many* call functions, so ext4_get_blocks()
2256 * will set the magic i_delalloc_reserved_flag once the
2257 * inode's allocation semaphore is taken.
2259 * If the blocks in questions were delalloc blocks, set
2260 * EXT4_GET_BLOCKS_DELALLOC_RESERVE so the delalloc accounting
2261 * variables are updated after the blocks have been allocated.
2264 get_blocks_flags
= EXT4_GET_BLOCKS_CREATE
;
2265 if (mpd
->b_state
& (1 << BH_Delay
))
2266 get_blocks_flags
|= EXT4_GET_BLOCKS_DELALLOC_RESERVE
;
2268 blks
= ext4_get_blocks(handle
, mpd
->inode
, next
, max_blocks
,
2269 &new, get_blocks_flags
);
2273 * If get block returns with error we simply
2274 * return. Later writepage will redirty the page and
2275 * writepages will find the dirty page again
2280 if (err
== -ENOSPC
&&
2281 ext4_count_free_blocks(mpd
->inode
->i_sb
)) {
2287 * get block failure will cause us to loop in
2288 * writepages, because a_ops->writepage won't be able
2289 * to make progress. The page will be redirtied by
2290 * writepage and writepages will again try to write
2293 ext4_msg(mpd
->inode
->i_sb
, KERN_CRIT
,
2294 "delayed block allocation failed for inode %lu at "
2295 "logical offset %llu with max blocks %zd with "
2296 "error %d\n", mpd
->inode
->i_ino
,
2297 (unsigned long long) next
,
2298 mpd
->b_size
>> mpd
->inode
->i_blkbits
, err
);
2299 printk(KERN_CRIT
"This should not happen!! "
2300 "Data will be lost\n");
2301 if (err
== -ENOSPC
) {
2302 ext4_print_free_blocks(mpd
->inode
);
2304 /* invalidate all the pages */
2305 ext4_da_block_invalidatepages(mpd
, next
,
2306 mpd
->b_size
>> mpd
->inode
->i_blkbits
);
2311 new.b_size
= (blks
<< mpd
->inode
->i_blkbits
);
2313 if (buffer_new(&new))
2314 __unmap_underlying_blocks(mpd
->inode
, &new);
2317 * If blocks are delayed marked, we need to
2318 * put actual blocknr and drop delayed bit
2320 if ((mpd
->b_state
& (1 << BH_Delay
)) ||
2321 (mpd
->b_state
& (1 << BH_Unwritten
)))
2322 mpage_put_bnr_to_bhs(mpd
, next
, &new);
2324 if (ext4_should_order_data(mpd
->inode
)) {
2325 err
= ext4_jbd2_file_inode(handle
, mpd
->inode
);
2331 * Update on-disk size along with block allocation.
2333 disksize
= ((loff_t
) next
+ blks
) << mpd
->inode
->i_blkbits
;
2334 if (disksize
> i_size_read(mpd
->inode
))
2335 disksize
= i_size_read(mpd
->inode
);
2336 if (disksize
> EXT4_I(mpd
->inode
)->i_disksize
) {
2337 ext4_update_i_disksize(mpd
->inode
, disksize
);
2338 return ext4_mark_inode_dirty(handle
, mpd
->inode
);
2344 #define BH_FLAGS ((1 << BH_Uptodate) | (1 << BH_Mapped) | \
2345 (1 << BH_Delay) | (1 << BH_Unwritten))
2348 * mpage_add_bh_to_extent - try to add one more block to extent of blocks
2350 * @mpd->lbh - extent of blocks
2351 * @logical - logical number of the block in the file
2352 * @bh - bh of the block (used to access block's state)
2354 * the function is used to collect contig. blocks in same state
2356 static void mpage_add_bh_to_extent(struct mpage_da_data
*mpd
,
2357 sector_t logical
, size_t b_size
,
2358 unsigned long b_state
)
2361 int nrblocks
= mpd
->b_size
>> mpd
->inode
->i_blkbits
;
2363 /* check if thereserved journal credits might overflow */
2364 if (!(EXT4_I(mpd
->inode
)->i_flags
& EXT4_EXTENTS_FL
)) {
2365 if (nrblocks
>= EXT4_MAX_TRANS_DATA
) {
2367 * With non-extent format we are limited by the journal
2368 * credit available. Total credit needed to insert
2369 * nrblocks contiguous blocks is dependent on the
2370 * nrblocks. So limit nrblocks.
2373 } else if ((nrblocks
+ (b_size
>> mpd
->inode
->i_blkbits
)) >
2374 EXT4_MAX_TRANS_DATA
) {
2376 * Adding the new buffer_head would make it cross the
2377 * allowed limit for which we have journal credit
2378 * reserved. So limit the new bh->b_size
2380 b_size
= (EXT4_MAX_TRANS_DATA
- nrblocks
) <<
2381 mpd
->inode
->i_blkbits
;
2382 /* we will do mpage_da_submit_io in the next loop */
2386 * First block in the extent
2388 if (mpd
->b_size
== 0) {
2389 mpd
->b_blocknr
= logical
;
2390 mpd
->b_size
= b_size
;
2391 mpd
->b_state
= b_state
& BH_FLAGS
;
2395 next
= mpd
->b_blocknr
+ nrblocks
;
2397 * Can we merge the block to our big extent?
2399 if (logical
== next
&& (b_state
& BH_FLAGS
) == mpd
->b_state
) {
2400 mpd
->b_size
+= b_size
;
2406 * We couldn't merge the block to our extent, so we
2407 * need to flush current extent and start new one
2409 if (mpage_da_map_blocks(mpd
) == 0)
2410 mpage_da_submit_io(mpd
);
2415 static int ext4_bh_delay_or_unwritten(handle_t
*handle
, struct buffer_head
*bh
)
2417 return (buffer_delay(bh
) || buffer_unwritten(bh
)) && buffer_dirty(bh
);
2421 * __mpage_da_writepage - finds extent of pages and blocks
2423 * @page: page to consider
2424 * @wbc: not used, we just follow rules
2427 * The function finds extents of pages and scan them for all blocks.
2429 static int __mpage_da_writepage(struct page
*page
,
2430 struct writeback_control
*wbc
, void *data
)
2432 struct mpage_da_data
*mpd
= data
;
2433 struct inode
*inode
= mpd
->inode
;
2434 struct buffer_head
*bh
, *head
;
2439 * Rest of the page in the page_vec
2440 * redirty then and skip then. We will
2441 * try to write them again after
2442 * starting a new transaction
2444 redirty_page_for_writepage(wbc
, page
);
2446 return MPAGE_DA_EXTENT_TAIL
;
2449 * Can we merge this page to current extent?
2451 if (mpd
->next_page
!= page
->index
) {
2453 * Nope, we can't. So, we map non-allocated blocks
2454 * and start IO on them using writepage()
2456 if (mpd
->next_page
!= mpd
->first_page
) {
2457 if (mpage_da_map_blocks(mpd
) == 0)
2458 mpage_da_submit_io(mpd
);
2460 * skip rest of the page in the page_vec
2463 redirty_page_for_writepage(wbc
, page
);
2465 return MPAGE_DA_EXTENT_TAIL
;
2469 * Start next extent of pages ...
2471 mpd
->first_page
= page
->index
;
2481 mpd
->next_page
= page
->index
+ 1;
2482 logical
= (sector_t
) page
->index
<<
2483 (PAGE_CACHE_SHIFT
- inode
->i_blkbits
);
2485 if (!page_has_buffers(page
)) {
2486 mpage_add_bh_to_extent(mpd
, logical
, PAGE_CACHE_SIZE
,
2487 (1 << BH_Dirty
) | (1 << BH_Uptodate
));
2489 return MPAGE_DA_EXTENT_TAIL
;
2492 * Page with regular buffer heads, just add all dirty ones
2494 head
= page_buffers(page
);
2497 BUG_ON(buffer_locked(bh
));
2499 * We need to try to allocate
2500 * unmapped blocks in the same page.
2501 * Otherwise we won't make progress
2502 * with the page in ext4_writepage
2504 if (ext4_bh_delay_or_unwritten(NULL
, bh
)) {
2505 mpage_add_bh_to_extent(mpd
, logical
,
2509 return MPAGE_DA_EXTENT_TAIL
;
2510 } else if (buffer_dirty(bh
) && (buffer_mapped(bh
))) {
2512 * mapped dirty buffer. We need to update
2513 * the b_state because we look at
2514 * b_state in mpage_da_map_blocks. We don't
2515 * update b_size because if we find an
2516 * unmapped buffer_head later we need to
2517 * use the b_state flag of that buffer_head.
2519 if (mpd
->b_size
== 0)
2520 mpd
->b_state
= bh
->b_state
& BH_FLAGS
;
2523 } while ((bh
= bh
->b_this_page
) != head
);
2530 * This is a special get_blocks_t callback which is used by
2531 * ext4_da_write_begin(). It will either return mapped block or
2532 * reserve space for a single block.
2534 * For delayed buffer_head we have BH_Mapped, BH_New, BH_Delay set.
2535 * We also have b_blocknr = -1 and b_bdev initialized properly
2537 * For unwritten buffer_head we have BH_Mapped, BH_New, BH_Unwritten set.
2538 * We also have b_blocknr = physicalblock mapping unwritten extent and b_bdev
2539 * initialized properly.
2541 static int ext4_da_get_block_prep(struct inode
*inode
, sector_t iblock
,
2542 struct buffer_head
*bh_result
, int create
)
2545 sector_t invalid_block
= ~((sector_t
) 0xffff);
2547 if (invalid_block
< ext4_blocks_count(EXT4_SB(inode
->i_sb
)->s_es
))
2550 BUG_ON(create
== 0);
2551 BUG_ON(bh_result
->b_size
!= inode
->i_sb
->s_blocksize
);
2554 * first, we need to know whether the block is allocated already
2555 * preallocated blocks are unmapped but should treated
2556 * the same as allocated blocks.
2558 ret
= ext4_get_blocks(NULL
, inode
, iblock
, 1, bh_result
, 0);
2559 if ((ret
== 0) && !buffer_delay(bh_result
)) {
2560 /* the block isn't (pre)allocated yet, let's reserve space */
2562 * XXX: __block_prepare_write() unmaps passed block,
2565 ret
= ext4_da_reserve_space(inode
, iblock
);
2567 /* not enough space to reserve */
2570 map_bh(bh_result
, inode
->i_sb
, invalid_block
);
2571 set_buffer_new(bh_result
);
2572 set_buffer_delay(bh_result
);
2573 } else if (ret
> 0) {
2574 bh_result
->b_size
= (ret
<< inode
->i_blkbits
);
2575 if (buffer_unwritten(bh_result
)) {
2576 /* A delayed write to unwritten bh should
2577 * be marked new and mapped. Mapped ensures
2578 * that we don't do get_block multiple times
2579 * when we write to the same offset and new
2580 * ensures that we do proper zero out for
2583 set_buffer_new(bh_result
);
2584 set_buffer_mapped(bh_result
);
2593 * This function is used as a standard get_block_t calback function
2594 * when there is no desire to allocate any blocks. It is used as a
2595 * callback function for block_prepare_write(), nobh_writepage(), and
2596 * block_write_full_page(). These functions should only try to map a
2597 * single block at a time.
2599 * Since this function doesn't do block allocations even if the caller
2600 * requests it by passing in create=1, it is critically important that
2601 * any caller checks to make sure that any buffer heads are returned
2602 * by this function are either all already mapped or marked for
2603 * delayed allocation before calling nobh_writepage() or
2604 * block_write_full_page(). Otherwise, b_blocknr could be left
2605 * unitialized, and the page write functions will be taken by
2608 static int noalloc_get_block_write(struct inode
*inode
, sector_t iblock
,
2609 struct buffer_head
*bh_result
, int create
)
2612 unsigned max_blocks
= bh_result
->b_size
>> inode
->i_blkbits
;
2614 BUG_ON(bh_result
->b_size
!= inode
->i_sb
->s_blocksize
);
2617 * we don't want to do block allocation in writepage
2618 * so call get_block_wrap with create = 0
2620 ret
= ext4_get_blocks(NULL
, inode
, iblock
, max_blocks
, bh_result
, 0);
2622 bh_result
->b_size
= (ret
<< inode
->i_blkbits
);
2628 static int bget_one(handle_t
*handle
, struct buffer_head
*bh
)
2634 static int bput_one(handle_t
*handle
, struct buffer_head
*bh
)
2640 static int __ext4_journalled_writepage(struct page
*page
,
2641 struct writeback_control
*wbc
,
2644 struct address_space
*mapping
= page
->mapping
;
2645 struct inode
*inode
= mapping
->host
;
2646 struct buffer_head
*page_bufs
;
2647 handle_t
*handle
= NULL
;
2651 page_bufs
= page_buffers(page
);
2653 walk_page_buffers(handle
, page_bufs
, 0, len
, NULL
, bget_one
);
2654 /* As soon as we unlock the page, it can go away, but we have
2655 * references to buffers so we are safe */
2658 handle
= ext4_journal_start(inode
, ext4_writepage_trans_blocks(inode
));
2659 if (IS_ERR(handle
)) {
2660 ret
= PTR_ERR(handle
);
2664 ret
= walk_page_buffers(handle
, page_bufs
, 0, len
, NULL
,
2665 do_journal_get_write_access
);
2667 err
= walk_page_buffers(handle
, page_bufs
, 0, len
, NULL
,
2671 err
= ext4_journal_stop(handle
);
2675 walk_page_buffers(handle
, page_bufs
, 0, len
, NULL
, bput_one
);
2676 ext4_set_inode_state(inode
, EXT4_STATE_JDATA
);
2682 * Note that we don't need to start a transaction unless we're journaling data
2683 * because we should have holes filled from ext4_page_mkwrite(). We even don't
2684 * need to file the inode to the transaction's list in ordered mode because if
2685 * we are writing back data added by write(), the inode is already there and if
2686 * we are writing back data modified via mmap(), noone guarantees in which
2687 * transaction the data will hit the disk. In case we are journaling data, we
2688 * cannot start transaction directly because transaction start ranks above page
2689 * lock so we have to do some magic.
2691 * This function can get called via...
2692 * - ext4_da_writepages after taking page lock (have journal handle)
2693 * - journal_submit_inode_data_buffers (no journal handle)
2694 * - shrink_page_list via pdflush (no journal handle)
2695 * - grab_page_cache when doing write_begin (have journal handle)
2697 * We don't do any block allocation in this function. If we have page with
2698 * multiple blocks we need to write those buffer_heads that are mapped. This
2699 * is important for mmaped based write. So if we do with blocksize 1K
2700 * truncate(f, 1024);
2701 * a = mmap(f, 0, 4096);
2703 * truncate(f, 4096);
2704 * we have in the page first buffer_head mapped via page_mkwrite call back
2705 * but other bufer_heads would be unmapped but dirty(dirty done via the
2706 * do_wp_page). So writepage should write the first block. If we modify
2707 * the mmap area beyond 1024 we will again get a page_fault and the
2708 * page_mkwrite callback will do the block allocation and mark the
2709 * buffer_heads mapped.
2711 * We redirty the page if we have any buffer_heads that is either delay or
2712 * unwritten in the page.
2714 * We can get recursively called as show below.
2716 * ext4_writepage() -> kmalloc() -> __alloc_pages() -> page_launder() ->
2719 * But since we don't do any block allocation we should not deadlock.
2720 * Page also have the dirty flag cleared so we don't get recurive page_lock.
2722 static int ext4_writepage(struct page
*page
,
2723 struct writeback_control
*wbc
)
2728 struct buffer_head
*page_bufs
;
2729 struct inode
*inode
= page
->mapping
->host
;
2731 trace_ext4_writepage(inode
, page
);
2732 size
= i_size_read(inode
);
2733 if (page
->index
== size
>> PAGE_CACHE_SHIFT
)
2734 len
= size
& ~PAGE_CACHE_MASK
;
2736 len
= PAGE_CACHE_SIZE
;
2738 if (page_has_buffers(page
)) {
2739 page_bufs
= page_buffers(page
);
2740 if (walk_page_buffers(NULL
, page_bufs
, 0, len
, NULL
,
2741 ext4_bh_delay_or_unwritten
)) {
2743 * We don't want to do block allocation
2744 * So redirty the page and return
2745 * We may reach here when we do a journal commit
2746 * via journal_submit_inode_data_buffers.
2747 * If we don't have mapping block we just ignore
2748 * them. We can also reach here via shrink_page_list
2750 redirty_page_for_writepage(wbc
, page
);
2756 * The test for page_has_buffers() is subtle:
2757 * We know the page is dirty but it lost buffers. That means
2758 * that at some moment in time after write_begin()/write_end()
2759 * has been called all buffers have been clean and thus they
2760 * must have been written at least once. So they are all
2761 * mapped and we can happily proceed with mapping them
2762 * and writing the page.
2764 * Try to initialize the buffer_heads and check whether
2765 * all are mapped and non delay. We don't want to
2766 * do block allocation here.
2768 ret
= block_prepare_write(page
, 0, len
,
2769 noalloc_get_block_write
);
2771 page_bufs
= page_buffers(page
);
2772 /* check whether all are mapped and non delay */
2773 if (walk_page_buffers(NULL
, page_bufs
, 0, len
, NULL
,
2774 ext4_bh_delay_or_unwritten
)) {
2775 redirty_page_for_writepage(wbc
, page
);
2781 * We can't do block allocation here
2782 * so just redity the page and unlock
2785 redirty_page_for_writepage(wbc
, page
);
2789 /* now mark the buffer_heads as dirty and uptodate */
2790 block_commit_write(page
, 0, len
);
2793 if (PageChecked(page
) && ext4_should_journal_data(inode
)) {
2795 * It's mmapped pagecache. Add buffers and journal it. There
2796 * doesn't seem much point in redirtying the page here.
2798 ClearPageChecked(page
);
2799 return __ext4_journalled_writepage(page
, wbc
, len
);
2802 if (test_opt(inode
->i_sb
, NOBH
) && ext4_should_writeback_data(inode
))
2803 ret
= nobh_writepage(page
, noalloc_get_block_write
, wbc
);
2805 ret
= block_write_full_page(page
, noalloc_get_block_write
,
2812 * This is called via ext4_da_writepages() to
2813 * calulate the total number of credits to reserve to fit
2814 * a single extent allocation into a single transaction,
2815 * ext4_da_writpeages() will loop calling this before
2816 * the block allocation.
2819 static int ext4_da_writepages_trans_blocks(struct inode
*inode
)
2821 int max_blocks
= EXT4_I(inode
)->i_reserved_data_blocks
;
2824 * With non-extent format the journal credit needed to
2825 * insert nrblocks contiguous block is dependent on
2826 * number of contiguous block. So we will limit
2827 * number of contiguous block to a sane value
2829 if (!(EXT4_I(inode
)->i_flags
& EXT4_EXTENTS_FL
) &&
2830 (max_blocks
> EXT4_MAX_TRANS_DATA
))
2831 max_blocks
= EXT4_MAX_TRANS_DATA
;
2833 return ext4_chunk_trans_blocks(inode
, max_blocks
);
2836 static int ext4_da_writepages(struct address_space
*mapping
,
2837 struct writeback_control
*wbc
)
2840 int range_whole
= 0;
2841 handle_t
*handle
= NULL
;
2842 struct mpage_da_data mpd
;
2843 struct inode
*inode
= mapping
->host
;
2844 int no_nrwrite_index_update
;
2845 int pages_written
= 0;
2847 unsigned int max_pages
;
2848 int range_cyclic
, cycled
= 1, io_done
= 0;
2849 int needed_blocks
, ret
= 0;
2850 long desired_nr_to_write
, nr_to_writebump
= 0;
2851 loff_t range_start
= wbc
->range_start
;
2852 struct ext4_sb_info
*sbi
= EXT4_SB(mapping
->host
->i_sb
);
2854 trace_ext4_da_writepages(inode
, wbc
);
2857 * No pages to write? This is mainly a kludge to avoid starting
2858 * a transaction for special inodes like journal inode on last iput()
2859 * because that could violate lock ordering on umount
2861 if (!mapping
->nrpages
|| !mapping_tagged(mapping
, PAGECACHE_TAG_DIRTY
))
2865 * If the filesystem has aborted, it is read-only, so return
2866 * right away instead of dumping stack traces later on that
2867 * will obscure the real source of the problem. We test
2868 * EXT4_MF_FS_ABORTED instead of sb->s_flag's MS_RDONLY because
2869 * the latter could be true if the filesystem is mounted
2870 * read-only, and in that case, ext4_da_writepages should
2871 * *never* be called, so if that ever happens, we would want
2874 if (unlikely(sbi
->s_mount_flags
& EXT4_MF_FS_ABORTED
))
2877 if (wbc
->range_start
== 0 && wbc
->range_end
== LLONG_MAX
)
2880 range_cyclic
= wbc
->range_cyclic
;
2881 if (wbc
->range_cyclic
) {
2882 index
= mapping
->writeback_index
;
2885 wbc
->range_start
= index
<< PAGE_CACHE_SHIFT
;
2886 wbc
->range_end
= LLONG_MAX
;
2887 wbc
->range_cyclic
= 0;
2889 index
= wbc
->range_start
>> PAGE_CACHE_SHIFT
;
2892 * This works around two forms of stupidity. The first is in
2893 * the writeback code, which caps the maximum number of pages
2894 * written to be 1024 pages. This is wrong on multiple
2895 * levels; different architectues have a different page size,
2896 * which changes the maximum amount of data which gets
2897 * written. Secondly, 4 megabytes is way too small. XFS
2898 * forces this value to be 16 megabytes by multiplying
2899 * nr_to_write parameter by four, and then relies on its
2900 * allocator to allocate larger extents to make them
2901 * contiguous. Unfortunately this brings us to the second
2902 * stupidity, which is that ext4's mballoc code only allocates
2903 * at most 2048 blocks. So we force contiguous writes up to
2904 * the number of dirty blocks in the inode, or
2905 * sbi->max_writeback_mb_bump whichever is smaller.
2907 max_pages
= sbi
->s_max_writeback_mb_bump
<< (20 - PAGE_CACHE_SHIFT
);
2908 if (!range_cyclic
&& range_whole
)
2909 desired_nr_to_write
= wbc
->nr_to_write
* 8;
2911 desired_nr_to_write
= ext4_num_dirty_pages(inode
, index
,
2913 if (desired_nr_to_write
> max_pages
)
2914 desired_nr_to_write
= max_pages
;
2916 if (wbc
->nr_to_write
< desired_nr_to_write
) {
2917 nr_to_writebump
= desired_nr_to_write
- wbc
->nr_to_write
;
2918 wbc
->nr_to_write
= desired_nr_to_write
;
2922 mpd
.inode
= mapping
->host
;
2925 * we don't want write_cache_pages to update
2926 * nr_to_write and writeback_index
2928 no_nrwrite_index_update
= wbc
->no_nrwrite_index_update
;
2929 wbc
->no_nrwrite_index_update
= 1;
2930 pages_skipped
= wbc
->pages_skipped
;
2933 while (!ret
&& wbc
->nr_to_write
> 0) {
2936 * we insert one extent at a time. So we need
2937 * credit needed for single extent allocation.
2938 * journalled mode is currently not supported
2941 BUG_ON(ext4_should_journal_data(inode
));
2942 needed_blocks
= ext4_da_writepages_trans_blocks(inode
);
2944 /* start a new transaction*/
2945 handle
= ext4_journal_start(inode
, needed_blocks
);
2946 if (IS_ERR(handle
)) {
2947 ret
= PTR_ERR(handle
);
2948 ext4_msg(inode
->i_sb
, KERN_CRIT
, "%s: jbd2_start: "
2949 "%ld pages, ino %lu; err %d\n", __func__
,
2950 wbc
->nr_to_write
, inode
->i_ino
, ret
);
2951 goto out_writepages
;
2955 * Now call __mpage_da_writepage to find the next
2956 * contiguous region of logical blocks that need
2957 * blocks to be allocated by ext4. We don't actually
2958 * submit the blocks for I/O here, even though
2959 * write_cache_pages thinks it will, and will set the
2960 * pages as clean for write before calling
2961 * __mpage_da_writepage().
2969 mpd
.pages_written
= 0;
2971 ret
= write_cache_pages(mapping
, wbc
, __mpage_da_writepage
,
2974 * If we have a contigous extent of pages and we
2975 * haven't done the I/O yet, map the blocks and submit
2978 if (!mpd
.io_done
&& mpd
.next_page
!= mpd
.first_page
) {
2979 if (mpage_da_map_blocks(&mpd
) == 0)
2980 mpage_da_submit_io(&mpd
);
2982 ret
= MPAGE_DA_EXTENT_TAIL
;
2984 trace_ext4_da_write_pages(inode
, &mpd
);
2985 wbc
->nr_to_write
-= mpd
.pages_written
;
2987 ext4_journal_stop(handle
);
2989 if ((mpd
.retval
== -ENOSPC
) && sbi
->s_journal
) {
2990 /* commit the transaction which would
2991 * free blocks released in the transaction
2994 jbd2_journal_force_commit_nested(sbi
->s_journal
);
2995 wbc
->pages_skipped
= pages_skipped
;
2997 } else if (ret
== MPAGE_DA_EXTENT_TAIL
) {
2999 * got one extent now try with
3002 pages_written
+= mpd
.pages_written
;
3003 wbc
->pages_skipped
= pages_skipped
;
3006 } else if (wbc
->nr_to_write
)
3008 * There is no more writeout needed
3009 * or we requested for a noblocking writeout
3010 * and we found the device congested
3014 if (!io_done
&& !cycled
) {
3017 wbc
->range_start
= index
<< PAGE_CACHE_SHIFT
;
3018 wbc
->range_end
= mapping
->writeback_index
- 1;
3021 if (pages_skipped
!= wbc
->pages_skipped
)
3022 ext4_msg(inode
->i_sb
, KERN_CRIT
,
3023 "This should not happen leaving %s "
3024 "with nr_to_write = %ld ret = %d\n",
3025 __func__
, wbc
->nr_to_write
, ret
);
3028 index
+= pages_written
;
3029 wbc
->range_cyclic
= range_cyclic
;
3030 if (wbc
->range_cyclic
|| (range_whole
&& wbc
->nr_to_write
> 0))
3032 * set the writeback_index so that range_cyclic
3033 * mode will write it back later
3035 mapping
->writeback_index
= index
;
3038 if (!no_nrwrite_index_update
)
3039 wbc
->no_nrwrite_index_update
= 0;
3040 wbc
->nr_to_write
-= nr_to_writebump
;
3041 wbc
->range_start
= range_start
;
3042 trace_ext4_da_writepages_result(inode
, wbc
, ret
, pages_written
);
3046 #define FALL_BACK_TO_NONDELALLOC 1
3047 static int ext4_nonda_switch(struct super_block
*sb
)
3049 s64 free_blocks
, dirty_blocks
;
3050 struct ext4_sb_info
*sbi
= EXT4_SB(sb
);
3053 * switch to non delalloc mode if we are running low
3054 * on free block. The free block accounting via percpu
3055 * counters can get slightly wrong with percpu_counter_batch getting
3056 * accumulated on each CPU without updating global counters
3057 * Delalloc need an accurate free block accounting. So switch
3058 * to non delalloc when we are near to error range.
3060 free_blocks
= percpu_counter_read_positive(&sbi
->s_freeblocks_counter
);
3061 dirty_blocks
= percpu_counter_read_positive(&sbi
->s_dirtyblocks_counter
);
3062 if (2 * free_blocks
< 3 * dirty_blocks
||
3063 free_blocks
< (dirty_blocks
+ EXT4_FREEBLOCKS_WATERMARK
)) {
3065 * free block count is less than 150% of dirty blocks
3066 * or free blocks is less than watermark
3071 * Even if we don't switch but are nearing capacity,
3072 * start pushing delalloc when 1/2 of free blocks are dirty.
3074 if (free_blocks
< 2 * dirty_blocks
)
3075 writeback_inodes_sb_if_idle(sb
);
3080 static int ext4_da_write_begin(struct file
*file
, struct address_space
*mapping
,
3081 loff_t pos
, unsigned len
, unsigned flags
,
3082 struct page
**pagep
, void **fsdata
)
3084 int ret
, retries
= 0, quota_retries
= 0;
3088 struct inode
*inode
= mapping
->host
;
3091 index
= pos
>> PAGE_CACHE_SHIFT
;
3092 from
= pos
& (PAGE_CACHE_SIZE
- 1);
3095 if (ext4_nonda_switch(inode
->i_sb
)) {
3096 *fsdata
= (void *)FALL_BACK_TO_NONDELALLOC
;
3097 return ext4_write_begin(file
, mapping
, pos
,
3098 len
, flags
, pagep
, fsdata
);
3100 *fsdata
= (void *)0;
3101 trace_ext4_da_write_begin(inode
, pos
, len
, flags
);
3104 * With delayed allocation, we don't log the i_disksize update
3105 * if there is delayed block allocation. But we still need
3106 * to journalling the i_disksize update if writes to the end
3107 * of file which has an already mapped buffer.
3109 handle
= ext4_journal_start(inode
, 1);
3110 if (IS_ERR(handle
)) {
3111 ret
= PTR_ERR(handle
);
3114 /* We cannot recurse into the filesystem as the transaction is already
3116 flags
|= AOP_FLAG_NOFS
;
3118 page
= grab_cache_page_write_begin(mapping
, index
, flags
);
3120 ext4_journal_stop(handle
);
3126 ret
= block_write_begin(file
, mapping
, pos
, len
, flags
, pagep
, fsdata
,
3127 ext4_da_get_block_prep
);
3130 ext4_journal_stop(handle
);
3131 page_cache_release(page
);
3133 * block_write_begin may have instantiated a few blocks
3134 * outside i_size. Trim these off again. Don't need
3135 * i_size_read because we hold i_mutex.
3137 if (pos
+ len
> inode
->i_size
)
3138 ext4_truncate_failed_write(inode
);
3141 if (ret
== -ENOSPC
&& ext4_should_retry_alloc(inode
->i_sb
, &retries
))
3144 if ((ret
== -EDQUOT
) &&
3145 EXT4_I(inode
)->i_reserved_meta_blocks
&&
3146 (quota_retries
++ < 3)) {
3148 * Since we often over-estimate the number of meta
3149 * data blocks required, we may sometimes get a
3150 * spurios out of quota error even though there would
3151 * be enough space once we write the data blocks and
3152 * find out how many meta data blocks were _really_
3153 * required. So try forcing the inode write to see if
3156 write_inode_now(inode
, (quota_retries
== 3));
3164 * Check if we should update i_disksize
3165 * when write to the end of file but not require block allocation
3167 static int ext4_da_should_update_i_disksize(struct page
*page
,
3168 unsigned long offset
)
3170 struct buffer_head
*bh
;
3171 struct inode
*inode
= page
->mapping
->host
;
3175 bh
= page_buffers(page
);
3176 idx
= offset
>> inode
->i_blkbits
;
3178 for (i
= 0; i
< idx
; i
++)
3179 bh
= bh
->b_this_page
;
3181 if (!buffer_mapped(bh
) || (buffer_delay(bh
)) || buffer_unwritten(bh
))
3186 static int ext4_da_write_end(struct file
*file
,
3187 struct address_space
*mapping
,
3188 loff_t pos
, unsigned len
, unsigned copied
,
3189 struct page
*page
, void *fsdata
)
3191 struct inode
*inode
= mapping
->host
;
3193 handle_t
*handle
= ext4_journal_current_handle();
3195 unsigned long start
, end
;
3196 int write_mode
= (int)(unsigned long)fsdata
;
3198 if (write_mode
== FALL_BACK_TO_NONDELALLOC
) {
3199 if (ext4_should_order_data(inode
)) {
3200 return ext4_ordered_write_end(file
, mapping
, pos
,
3201 len
, copied
, page
, fsdata
);
3202 } else if (ext4_should_writeback_data(inode
)) {
3203 return ext4_writeback_write_end(file
, mapping
, pos
,
3204 len
, copied
, page
, fsdata
);
3210 trace_ext4_da_write_end(inode
, pos
, len
, copied
);
3211 start
= pos
& (PAGE_CACHE_SIZE
- 1);
3212 end
= start
+ copied
- 1;
3215 * generic_write_end() will run mark_inode_dirty() if i_size
3216 * changes. So let's piggyback the i_disksize mark_inode_dirty
3220 new_i_size
= pos
+ copied
;
3221 if (new_i_size
> EXT4_I(inode
)->i_disksize
) {
3222 if (ext4_da_should_update_i_disksize(page
, end
)) {
3223 down_write(&EXT4_I(inode
)->i_data_sem
);
3224 if (new_i_size
> EXT4_I(inode
)->i_disksize
) {
3226 * Updating i_disksize when extending file
3227 * without needing block allocation
3229 if (ext4_should_order_data(inode
))
3230 ret
= ext4_jbd2_file_inode(handle
,
3233 EXT4_I(inode
)->i_disksize
= new_i_size
;
3235 up_write(&EXT4_I(inode
)->i_data_sem
);
3236 /* We need to mark inode dirty even if
3237 * new_i_size is less that inode->i_size
3238 * bu greater than i_disksize.(hint delalloc)
3240 ext4_mark_inode_dirty(handle
, inode
);
3243 ret2
= generic_write_end(file
, mapping
, pos
, len
, copied
,
3248 ret2
= ext4_journal_stop(handle
);
3252 return ret
? ret
: copied
;
3255 static void ext4_da_invalidatepage(struct page
*page
, unsigned long offset
)
3258 * Drop reserved blocks
3260 BUG_ON(!PageLocked(page
));
3261 if (!page_has_buffers(page
))
3264 ext4_da_page_release_reservation(page
, offset
);
3267 ext4_invalidatepage(page
, offset
);
3273 * Force all delayed allocation blocks to be allocated for a given inode.
3275 int ext4_alloc_da_blocks(struct inode
*inode
)
3277 trace_ext4_alloc_da_blocks(inode
);
3279 if (!EXT4_I(inode
)->i_reserved_data_blocks
&&
3280 !EXT4_I(inode
)->i_reserved_meta_blocks
)
3284 * We do something simple for now. The filemap_flush() will
3285 * also start triggering a write of the data blocks, which is
3286 * not strictly speaking necessary (and for users of
3287 * laptop_mode, not even desirable). However, to do otherwise
3288 * would require replicating code paths in:
3290 * ext4_da_writepages() ->
3291 * write_cache_pages() ---> (via passed in callback function)
3292 * __mpage_da_writepage() -->
3293 * mpage_add_bh_to_extent()
3294 * mpage_da_map_blocks()
3296 * The problem is that write_cache_pages(), located in
3297 * mm/page-writeback.c, marks pages clean in preparation for
3298 * doing I/O, which is not desirable if we're not planning on
3301 * We could call write_cache_pages(), and then redirty all of
3302 * the pages by calling redirty_page_for_writeback() but that
3303 * would be ugly in the extreme. So instead we would need to
3304 * replicate parts of the code in the above functions,
3305 * simplifying them becuase we wouldn't actually intend to
3306 * write out the pages, but rather only collect contiguous
3307 * logical block extents, call the multi-block allocator, and
3308 * then update the buffer heads with the block allocations.
3310 * For now, though, we'll cheat by calling filemap_flush(),
3311 * which will map the blocks, and start the I/O, but not
3312 * actually wait for the I/O to complete.
3314 return filemap_flush(inode
->i_mapping
);
3318 * bmap() is special. It gets used by applications such as lilo and by
3319 * the swapper to find the on-disk block of a specific piece of data.
3321 * Naturally, this is dangerous if the block concerned is still in the
3322 * journal. If somebody makes a swapfile on an ext4 data-journaling
3323 * filesystem and enables swap, then they may get a nasty shock when the
3324 * data getting swapped to that swapfile suddenly gets overwritten by
3325 * the original zero's written out previously to the journal and
3326 * awaiting writeback in the kernel's buffer cache.
3328 * So, if we see any bmap calls here on a modified, data-journaled file,
3329 * take extra steps to flush any blocks which might be in the cache.
3331 static sector_t
ext4_bmap(struct address_space
*mapping
, sector_t block
)
3333 struct inode
*inode
= mapping
->host
;
3337 if (mapping_tagged(mapping
, PAGECACHE_TAG_DIRTY
) &&
3338 test_opt(inode
->i_sb
, DELALLOC
)) {
3340 * With delalloc we want to sync the file
3341 * so that we can make sure we allocate
3344 filemap_write_and_wait(mapping
);
3347 if (EXT4_JOURNAL(inode
) &&
3348 ext4_test_inode_state(inode
, EXT4_STATE_JDATA
)) {
3350 * This is a REALLY heavyweight approach, but the use of
3351 * bmap on dirty files is expected to be extremely rare:
3352 * only if we run lilo or swapon on a freshly made file
3353 * do we expect this to happen.
3355 * (bmap requires CAP_SYS_RAWIO so this does not
3356 * represent an unprivileged user DOS attack --- we'd be
3357 * in trouble if mortal users could trigger this path at
3360 * NB. EXT4_STATE_JDATA is not set on files other than
3361 * regular files. If somebody wants to bmap a directory
3362 * or symlink and gets confused because the buffer
3363 * hasn't yet been flushed to disk, they deserve
3364 * everything they get.
3367 ext4_clear_inode_state(inode
, EXT4_STATE_JDATA
);
3368 journal
= EXT4_JOURNAL(inode
);
3369 jbd2_journal_lock_updates(journal
);
3370 err
= jbd2_journal_flush(journal
);
3371 jbd2_journal_unlock_updates(journal
);
3377 return generic_block_bmap(mapping
, block
, ext4_get_block
);
3380 static int ext4_readpage(struct file
*file
, struct page
*page
)
3382 return mpage_readpage(page
, ext4_get_block
);
3386 ext4_readpages(struct file
*file
, struct address_space
*mapping
,
3387 struct list_head
*pages
, unsigned nr_pages
)
3389 return mpage_readpages(mapping
, pages
, nr_pages
, ext4_get_block
);
3392 static void ext4_invalidatepage(struct page
*page
, unsigned long offset
)
3394 journal_t
*journal
= EXT4_JOURNAL(page
->mapping
->host
);
3397 * If it's a full truncate we just forget about the pending dirtying
3400 ClearPageChecked(page
);
3403 jbd2_journal_invalidatepage(journal
, page
, offset
);
3405 block_invalidatepage(page
, offset
);
3408 static int ext4_releasepage(struct page
*page
, gfp_t wait
)
3410 journal_t
*journal
= EXT4_JOURNAL(page
->mapping
->host
);
3412 WARN_ON(PageChecked(page
));
3413 if (!page_has_buffers(page
))
3416 return jbd2_journal_try_to_free_buffers(journal
, page
, wait
);
3418 return try_to_free_buffers(page
);
3422 * O_DIRECT for ext3 (or indirect map) based files
3424 * If the O_DIRECT write will extend the file then add this inode to the
3425 * orphan list. So recovery will truncate it back to the original size
3426 * if the machine crashes during the write.
3428 * If the O_DIRECT write is intantiating holes inside i_size and the machine
3429 * crashes then stale disk data _may_ be exposed inside the file. But current
3430 * VFS code falls back into buffered path in that case so we are safe.
3432 static ssize_t
ext4_ind_direct_IO(int rw
, struct kiocb
*iocb
,
3433 const struct iovec
*iov
, loff_t offset
,
3434 unsigned long nr_segs
)
3436 struct file
*file
= iocb
->ki_filp
;
3437 struct inode
*inode
= file
->f_mapping
->host
;
3438 struct ext4_inode_info
*ei
= EXT4_I(inode
);
3442 size_t count
= iov_length(iov
, nr_segs
);
3446 loff_t final_size
= offset
+ count
;
3448 if (final_size
> inode
->i_size
) {
3449 /* Credits for sb + inode write */
3450 handle
= ext4_journal_start(inode
, 2);
3451 if (IS_ERR(handle
)) {
3452 ret
= PTR_ERR(handle
);
3455 ret
= ext4_orphan_add(handle
, inode
);
3457 ext4_journal_stop(handle
);
3461 ei
->i_disksize
= inode
->i_size
;
3462 ext4_journal_stop(handle
);
3467 ret
= blockdev_direct_IO(rw
, iocb
, inode
, inode
->i_sb
->s_bdev
, iov
,
3469 ext4_get_block
, NULL
);
3470 if (ret
== -ENOSPC
&& ext4_should_retry_alloc(inode
->i_sb
, &retries
))
3476 /* Credits for sb + inode write */
3477 handle
= ext4_journal_start(inode
, 2);
3478 if (IS_ERR(handle
)) {
3479 /* This is really bad luck. We've written the data
3480 * but cannot extend i_size. Bail out and pretend
3481 * the write failed... */
3482 ret
= PTR_ERR(handle
);
3486 ext4_orphan_del(handle
, inode
);
3488 loff_t end
= offset
+ ret
;
3489 if (end
> inode
->i_size
) {
3490 ei
->i_disksize
= end
;
3491 i_size_write(inode
, end
);
3493 * We're going to return a positive `ret'
3494 * here due to non-zero-length I/O, so there's
3495 * no way of reporting error returns from
3496 * ext4_mark_inode_dirty() to userspace. So
3499 ext4_mark_inode_dirty(handle
, inode
);
3502 err
= ext4_journal_stop(handle
);
3510 static int ext4_get_block_dio_write(struct inode
*inode
, sector_t iblock
,
3511 struct buffer_head
*bh_result
, int create
)
3513 handle_t
*handle
= NULL
;
3515 unsigned max_blocks
= bh_result
->b_size
>> inode
->i_blkbits
;
3518 ext4_debug("ext4_get_block_dio_write: inode %lu, create flag %d\n",
3519 inode
->i_ino
, create
);
3521 * DIO VFS code passes create = 0 flag for write to
3522 * the middle of file. It does this to avoid block
3523 * allocation for holes, to prevent expose stale data
3524 * out when there is parallel buffered read (which does
3525 * not hold the i_mutex lock) while direct IO write has
3526 * not completed. DIO request on holes finally falls back
3527 * to buffered IO for this reason.
3529 * For ext4 extent based file, since we support fallocate,
3530 * new allocated extent as uninitialized, for holes, we
3531 * could fallocate blocks for holes, thus parallel
3532 * buffered IO read will zero out the page when read on
3533 * a hole while parallel DIO write to the hole has not completed.
3535 * when we come here, we know it's a direct IO write to
3536 * to the middle of file (<i_size)
3537 * so it's safe to override the create flag from VFS.
3539 create
= EXT4_GET_BLOCKS_DIO_CREATE_EXT
;
3541 if (max_blocks
> DIO_MAX_BLOCKS
)
3542 max_blocks
= DIO_MAX_BLOCKS
;
3543 dio_credits
= ext4_chunk_trans_blocks(inode
, max_blocks
);
3544 handle
= ext4_journal_start(inode
, dio_credits
);
3545 if (IS_ERR(handle
)) {
3546 ret
= PTR_ERR(handle
);
3549 ret
= ext4_get_blocks(handle
, inode
, iblock
, max_blocks
, bh_result
,
3552 bh_result
->b_size
= (ret
<< inode
->i_blkbits
);
3555 ext4_journal_stop(handle
);
3560 static void ext4_free_io_end(ext4_io_end_t
*io
)
3566 static void dump_aio_dio_list(struct inode
* inode
)
3569 struct list_head
*cur
, *before
, *after
;
3570 ext4_io_end_t
*io
, *io0
, *io1
;
3572 if (list_empty(&EXT4_I(inode
)->i_aio_dio_complete_list
)){
3573 ext4_debug("inode %lu aio dio list is empty\n", inode
->i_ino
);
3577 ext4_debug("Dump inode %lu aio_dio_completed_IO list \n", inode
->i_ino
);
3578 list_for_each_entry(io
, &EXT4_I(inode
)->i_aio_dio_complete_list
, list
){
3581 io0
= container_of(before
, ext4_io_end_t
, list
);
3583 io1
= container_of(after
, ext4_io_end_t
, list
);
3585 ext4_debug("io 0x%p from inode %lu,prev 0x%p,next 0x%p\n",
3586 io
, inode
->i_ino
, io0
, io1
);
3592 * check a range of space and convert unwritten extents to written.
3594 static int ext4_end_aio_dio_nolock(ext4_io_end_t
*io
)
3596 struct inode
*inode
= io
->inode
;
3597 loff_t offset
= io
->offset
;
3598 ssize_t size
= io
->size
;
3601 ext4_debug("end_aio_dio_onlock: io 0x%p from inode %lu,list->next 0x%p,"
3602 "list->prev 0x%p\n",
3603 io
, inode
->i_ino
, io
->list
.next
, io
->list
.prev
);
3605 if (list_empty(&io
->list
))
3608 if (io
->flag
!= DIO_AIO_UNWRITTEN
)
3611 if (offset
+ size
<= i_size_read(inode
))
3612 ret
= ext4_convert_unwritten_extents(inode
, offset
, size
);
3615 printk(KERN_EMERG
"%s: failed to convert unwritten"
3616 "extents to written extents, error is %d"
3617 " io is still on inode %lu aio dio list\n",
3618 __func__
, ret
, inode
->i_ino
);
3622 /* clear the DIO AIO unwritten flag */
3627 * work on completed aio dio IO, to convert unwritten extents to extents
3629 static void ext4_end_aio_dio_work(struct work_struct
*work
)
3631 ext4_io_end_t
*io
= container_of(work
, ext4_io_end_t
, work
);
3632 struct inode
*inode
= io
->inode
;
3635 mutex_lock(&inode
->i_mutex
);
3636 ret
= ext4_end_aio_dio_nolock(io
);
3638 if (!list_empty(&io
->list
))
3639 list_del_init(&io
->list
);
3640 ext4_free_io_end(io
);
3642 mutex_unlock(&inode
->i_mutex
);
3645 * This function is called from ext4_sync_file().
3647 * When AIO DIO IO is completed, the work to convert unwritten
3648 * extents to written is queued on workqueue but may not get immediately
3649 * scheduled. When fsync is called, we need to ensure the
3650 * conversion is complete before fsync returns.
3651 * The inode keeps track of a list of completed AIO from DIO path
3652 * that might needs to do the conversion. This function walks through
3653 * the list and convert the related unwritten extents to written.
3655 int flush_aio_dio_completed_IO(struct inode
*inode
)
3661 if (list_empty(&EXT4_I(inode
)->i_aio_dio_complete_list
))
3664 dump_aio_dio_list(inode
);
3665 while (!list_empty(&EXT4_I(inode
)->i_aio_dio_complete_list
)){
3666 io
= list_entry(EXT4_I(inode
)->i_aio_dio_complete_list
.next
,
3667 ext4_io_end_t
, list
);
3669 * Calling ext4_end_aio_dio_nolock() to convert completed
3672 * When ext4_sync_file() is called, run_queue() may already
3673 * about to flush the work corresponding to this io structure.
3674 * It will be upset if it founds the io structure related
3675 * to the work-to-be schedule is freed.
3677 * Thus we need to keep the io structure still valid here after
3678 * convertion finished. The io structure has a flag to
3679 * avoid double converting from both fsync and background work
3682 ret
= ext4_end_aio_dio_nolock(io
);
3686 list_del_init(&io
->list
);
3688 return (ret2
< 0) ? ret2
: 0;
3691 static ext4_io_end_t
*ext4_init_io_end (struct inode
*inode
)
3693 ext4_io_end_t
*io
= NULL
;
3695 io
= kmalloc(sizeof(*io
), GFP_NOFS
);
3704 INIT_WORK(&io
->work
, ext4_end_aio_dio_work
);
3705 INIT_LIST_HEAD(&io
->list
);
3711 static void ext4_end_io_dio(struct kiocb
*iocb
, loff_t offset
,
3712 ssize_t size
, void *private)
3714 ext4_io_end_t
*io_end
= iocb
->private;
3715 struct workqueue_struct
*wq
;
3717 /* if not async direct IO or dio with 0 bytes write, just return */
3718 if (!io_end
|| !size
)
3721 ext_debug("ext4_end_io_dio(): io_end 0x%p"
3722 "for inode %lu, iocb 0x%p, offset %llu, size %llu\n",
3723 iocb
->private, io_end
->inode
->i_ino
, iocb
, offset
,
3726 /* if not aio dio with unwritten extents, just free io and return */
3727 if (io_end
->flag
!= DIO_AIO_UNWRITTEN
){
3728 ext4_free_io_end(io_end
);
3729 iocb
->private = NULL
;
3733 io_end
->offset
= offset
;
3734 io_end
->size
= size
;
3735 wq
= EXT4_SB(io_end
->inode
->i_sb
)->dio_unwritten_wq
;
3737 /* queue the work to convert unwritten extents to written */
3738 queue_work(wq
, &io_end
->work
);
3740 /* Add the io_end to per-inode completed aio dio list*/
3741 list_add_tail(&io_end
->list
,
3742 &EXT4_I(io_end
->inode
)->i_aio_dio_complete_list
);
3743 iocb
->private = NULL
;
3746 * For ext4 extent files, ext4 will do direct-io write to holes,
3747 * preallocated extents, and those write extend the file, no need to
3748 * fall back to buffered IO.
3750 * For holes, we fallocate those blocks, mark them as unintialized
3751 * If those blocks were preallocated, we mark sure they are splited, but
3752 * still keep the range to write as unintialized.
3754 * The unwrritten extents will be converted to written when DIO is completed.
3755 * For async direct IO, since the IO may still pending when return, we
3756 * set up an end_io call back function, which will do the convertion
3757 * when async direct IO completed.
3759 * If the O_DIRECT write will extend the file then add this inode to the
3760 * orphan list. So recovery will truncate it back to the original size
3761 * if the machine crashes during the write.
3764 static ssize_t
ext4_ext_direct_IO(int rw
, struct kiocb
*iocb
,
3765 const struct iovec
*iov
, loff_t offset
,
3766 unsigned long nr_segs
)
3768 struct file
*file
= iocb
->ki_filp
;
3769 struct inode
*inode
= file
->f_mapping
->host
;
3771 size_t count
= iov_length(iov
, nr_segs
);
3773 loff_t final_size
= offset
+ count
;
3774 if (rw
== WRITE
&& final_size
<= inode
->i_size
) {
3776 * We could direct write to holes and fallocate.
3778 * Allocated blocks to fill the hole are marked as uninitialized
3779 * to prevent paralel buffered read to expose the stale data
3780 * before DIO complete the data IO.
3782 * As to previously fallocated extents, ext4 get_block
3783 * will just simply mark the buffer mapped but still
3784 * keep the extents uninitialized.
3786 * for non AIO case, we will convert those unwritten extents
3787 * to written after return back from blockdev_direct_IO.
3789 * for async DIO, the conversion needs to be defered when
3790 * the IO is completed. The ext4 end_io callback function
3791 * will be called to take care of the conversion work.
3792 * Here for async case, we allocate an io_end structure to
3795 iocb
->private = NULL
;
3796 EXT4_I(inode
)->cur_aio_dio
= NULL
;
3797 if (!is_sync_kiocb(iocb
)) {
3798 iocb
->private = ext4_init_io_end(inode
);
3802 * we save the io structure for current async
3803 * direct IO, so that later ext4_get_blocks()
3804 * could flag the io structure whether there
3805 * is a unwritten extents needs to be converted
3806 * when IO is completed.
3808 EXT4_I(inode
)->cur_aio_dio
= iocb
->private;
3811 ret
= blockdev_direct_IO(rw
, iocb
, inode
,
3812 inode
->i_sb
->s_bdev
, iov
,
3814 ext4_get_block_dio_write
,
3817 EXT4_I(inode
)->cur_aio_dio
= NULL
;
3819 * The io_end structure takes a reference to the inode,
3820 * that structure needs to be destroyed and the
3821 * reference to the inode need to be dropped, when IO is
3822 * complete, even with 0 byte write, or failed.
3824 * In the successful AIO DIO case, the io_end structure will be
3825 * desctroyed and the reference to the inode will be dropped
3826 * after the end_io call back function is called.
3828 * In the case there is 0 byte write, or error case, since
3829 * VFS direct IO won't invoke the end_io call back function,
3830 * we need to free the end_io structure here.
3832 if (ret
!= -EIOCBQUEUED
&& ret
<= 0 && iocb
->private) {
3833 ext4_free_io_end(iocb
->private);
3834 iocb
->private = NULL
;
3835 } else if (ret
> 0 && ext4_test_inode_state(inode
,
3836 EXT4_STATE_DIO_UNWRITTEN
)) {
3839 * for non AIO case, since the IO is already
3840 * completed, we could do the convertion right here
3842 err
= ext4_convert_unwritten_extents(inode
,
3846 ext4_clear_inode_state(inode
, EXT4_STATE_DIO_UNWRITTEN
);
3851 /* for write the the end of file case, we fall back to old way */
3852 return ext4_ind_direct_IO(rw
, iocb
, iov
, offset
, nr_segs
);
3855 static ssize_t
ext4_direct_IO(int rw
, struct kiocb
*iocb
,
3856 const struct iovec
*iov
, loff_t offset
,
3857 unsigned long nr_segs
)
3859 struct file
*file
= iocb
->ki_filp
;
3860 struct inode
*inode
= file
->f_mapping
->host
;
3862 if (EXT4_I(inode
)->i_flags
& EXT4_EXTENTS_FL
)
3863 return ext4_ext_direct_IO(rw
, iocb
, iov
, offset
, nr_segs
);
3865 return ext4_ind_direct_IO(rw
, iocb
, iov
, offset
, nr_segs
);
3869 * Pages can be marked dirty completely asynchronously from ext4's journalling
3870 * activity. By filemap_sync_pte(), try_to_unmap_one(), etc. We cannot do
3871 * much here because ->set_page_dirty is called under VFS locks. The page is
3872 * not necessarily locked.
3874 * We cannot just dirty the page and leave attached buffers clean, because the
3875 * buffers' dirty state is "definitive". We cannot just set the buffers dirty
3876 * or jbddirty because all the journalling code will explode.
3878 * So what we do is to mark the page "pending dirty" and next time writepage
3879 * is called, propagate that into the buffers appropriately.
3881 static int ext4_journalled_set_page_dirty(struct page
*page
)
3883 SetPageChecked(page
);
3884 return __set_page_dirty_nobuffers(page
);
3887 static const struct address_space_operations ext4_ordered_aops
= {
3888 .readpage
= ext4_readpage
,
3889 .readpages
= ext4_readpages
,
3890 .writepage
= ext4_writepage
,
3891 .sync_page
= block_sync_page
,
3892 .write_begin
= ext4_write_begin
,
3893 .write_end
= ext4_ordered_write_end
,
3895 .invalidatepage
= ext4_invalidatepage
,
3896 .releasepage
= ext4_releasepage
,
3897 .direct_IO
= ext4_direct_IO
,
3898 .migratepage
= buffer_migrate_page
,
3899 .is_partially_uptodate
= block_is_partially_uptodate
,
3900 .error_remove_page
= generic_error_remove_page
,
3903 static const struct address_space_operations ext4_writeback_aops
= {
3904 .readpage
= ext4_readpage
,
3905 .readpages
= ext4_readpages
,
3906 .writepage
= ext4_writepage
,
3907 .sync_page
= block_sync_page
,
3908 .write_begin
= ext4_write_begin
,
3909 .write_end
= ext4_writeback_write_end
,
3911 .invalidatepage
= ext4_invalidatepage
,
3912 .releasepage
= ext4_releasepage
,
3913 .direct_IO
= ext4_direct_IO
,
3914 .migratepage
= buffer_migrate_page
,
3915 .is_partially_uptodate
= block_is_partially_uptodate
,
3916 .error_remove_page
= generic_error_remove_page
,
3919 static const struct address_space_operations ext4_journalled_aops
= {
3920 .readpage
= ext4_readpage
,
3921 .readpages
= ext4_readpages
,
3922 .writepage
= ext4_writepage
,
3923 .sync_page
= block_sync_page
,
3924 .write_begin
= ext4_write_begin
,
3925 .write_end
= ext4_journalled_write_end
,
3926 .set_page_dirty
= ext4_journalled_set_page_dirty
,
3928 .invalidatepage
= ext4_invalidatepage
,
3929 .releasepage
= ext4_releasepage
,
3930 .is_partially_uptodate
= block_is_partially_uptodate
,
3931 .error_remove_page
= generic_error_remove_page
,
3934 static const struct address_space_operations ext4_da_aops
= {
3935 .readpage
= ext4_readpage
,
3936 .readpages
= ext4_readpages
,
3937 .writepage
= ext4_writepage
,
3938 .writepages
= ext4_da_writepages
,
3939 .sync_page
= block_sync_page
,
3940 .write_begin
= ext4_da_write_begin
,
3941 .write_end
= ext4_da_write_end
,
3943 .invalidatepage
= ext4_da_invalidatepage
,
3944 .releasepage
= ext4_releasepage
,
3945 .direct_IO
= ext4_direct_IO
,
3946 .migratepage
= buffer_migrate_page
,
3947 .is_partially_uptodate
= block_is_partially_uptodate
,
3948 .error_remove_page
= generic_error_remove_page
,
3951 void ext4_set_aops(struct inode
*inode
)
3953 if (ext4_should_order_data(inode
) &&
3954 test_opt(inode
->i_sb
, DELALLOC
))
3955 inode
->i_mapping
->a_ops
= &ext4_da_aops
;
3956 else if (ext4_should_order_data(inode
))
3957 inode
->i_mapping
->a_ops
= &ext4_ordered_aops
;
3958 else if (ext4_should_writeback_data(inode
) &&
3959 test_opt(inode
->i_sb
, DELALLOC
))
3960 inode
->i_mapping
->a_ops
= &ext4_da_aops
;
3961 else if (ext4_should_writeback_data(inode
))
3962 inode
->i_mapping
->a_ops
= &ext4_writeback_aops
;
3964 inode
->i_mapping
->a_ops
= &ext4_journalled_aops
;
3968 * ext4_block_truncate_page() zeroes out a mapping from file offset `from'
3969 * up to the end of the block which corresponds to `from'.
3970 * This required during truncate. We need to physically zero the tail end
3971 * of that block so it doesn't yield old data if the file is later grown.
3973 int ext4_block_truncate_page(handle_t
*handle
,
3974 struct address_space
*mapping
, loff_t from
)
3976 ext4_fsblk_t index
= from
>> PAGE_CACHE_SHIFT
;
3977 unsigned offset
= from
& (PAGE_CACHE_SIZE
-1);
3978 unsigned blocksize
, length
, pos
;
3980 struct inode
*inode
= mapping
->host
;
3981 struct buffer_head
*bh
;
3985 page
= find_or_create_page(mapping
, from
>> PAGE_CACHE_SHIFT
,
3986 mapping_gfp_mask(mapping
) & ~__GFP_FS
);
3990 blocksize
= inode
->i_sb
->s_blocksize
;
3991 length
= blocksize
- (offset
& (blocksize
- 1));
3992 iblock
= index
<< (PAGE_CACHE_SHIFT
- inode
->i_sb
->s_blocksize_bits
);
3995 * For "nobh" option, we can only work if we don't need to
3996 * read-in the page - otherwise we create buffers to do the IO.
3998 if (!page_has_buffers(page
) && test_opt(inode
->i_sb
, NOBH
) &&
3999 ext4_should_writeback_data(inode
) && PageUptodate(page
)) {
4000 zero_user(page
, offset
, length
);
4001 set_page_dirty(page
);
4005 if (!page_has_buffers(page
))
4006 create_empty_buffers(page
, blocksize
, 0);
4008 /* Find the buffer that contains "offset" */
4009 bh
= page_buffers(page
);
4011 while (offset
>= pos
) {
4012 bh
= bh
->b_this_page
;
4018 if (buffer_freed(bh
)) {
4019 BUFFER_TRACE(bh
, "freed: skip");
4023 if (!buffer_mapped(bh
)) {
4024 BUFFER_TRACE(bh
, "unmapped");
4025 ext4_get_block(inode
, iblock
, bh
, 0);
4026 /* unmapped? It's a hole - nothing to do */
4027 if (!buffer_mapped(bh
)) {
4028 BUFFER_TRACE(bh
, "still unmapped");
4033 /* Ok, it's mapped. Make sure it's up-to-date */
4034 if (PageUptodate(page
))
4035 set_buffer_uptodate(bh
);
4037 if (!buffer_uptodate(bh
)) {
4039 ll_rw_block(READ
, 1, &bh
);
4041 /* Uhhuh. Read error. Complain and punt. */
4042 if (!buffer_uptodate(bh
))
4046 if (ext4_should_journal_data(inode
)) {
4047 BUFFER_TRACE(bh
, "get write access");
4048 err
= ext4_journal_get_write_access(handle
, bh
);
4053 zero_user(page
, offset
, length
);
4055 BUFFER_TRACE(bh
, "zeroed end of block");
4058 if (ext4_should_journal_data(inode
)) {
4059 err
= ext4_handle_dirty_metadata(handle
, inode
, bh
);
4061 if (ext4_should_order_data(inode
))
4062 err
= ext4_jbd2_file_inode(handle
, inode
);
4063 mark_buffer_dirty(bh
);
4068 page_cache_release(page
);
4073 * Probably it should be a library function... search for first non-zero word
4074 * or memcmp with zero_page, whatever is better for particular architecture.
4077 static inline int all_zeroes(__le32
*p
, __le32
*q
)
4086 * ext4_find_shared - find the indirect blocks for partial truncation.
4087 * @inode: inode in question
4088 * @depth: depth of the affected branch
4089 * @offsets: offsets of pointers in that branch (see ext4_block_to_path)
4090 * @chain: place to store the pointers to partial indirect blocks
4091 * @top: place to the (detached) top of branch
4093 * This is a helper function used by ext4_truncate().
4095 * When we do truncate() we may have to clean the ends of several
4096 * indirect blocks but leave the blocks themselves alive. Block is
4097 * partially truncated if some data below the new i_size is refered
4098 * from it (and it is on the path to the first completely truncated
4099 * data block, indeed). We have to free the top of that path along
4100 * with everything to the right of the path. Since no allocation
4101 * past the truncation point is possible until ext4_truncate()
4102 * finishes, we may safely do the latter, but top of branch may
4103 * require special attention - pageout below the truncation point
4104 * might try to populate it.
4106 * We atomically detach the top of branch from the tree, store the
4107 * block number of its root in *@top, pointers to buffer_heads of
4108 * partially truncated blocks - in @chain[].bh and pointers to
4109 * their last elements that should not be removed - in
4110 * @chain[].p. Return value is the pointer to last filled element
4113 * The work left to caller to do the actual freeing of subtrees:
4114 * a) free the subtree starting from *@top
4115 * b) free the subtrees whose roots are stored in
4116 * (@chain[i].p+1 .. end of @chain[i].bh->b_data)
4117 * c) free the subtrees growing from the inode past the @chain[0].
4118 * (no partially truncated stuff there). */
4120 static Indirect
*ext4_find_shared(struct inode
*inode
, int depth
,
4121 ext4_lblk_t offsets
[4], Indirect chain
[4],
4124 Indirect
*partial
, *p
;
4128 /* Make k index the deepest non-null offest + 1 */
4129 for (k
= depth
; k
> 1 && !offsets
[k
-1]; k
--)
4131 partial
= ext4_get_branch(inode
, k
, offsets
, chain
, &err
);
4132 /* Writer: pointers */
4134 partial
= chain
+ k
-1;
4136 * If the branch acquired continuation since we've looked at it -
4137 * fine, it should all survive and (new) top doesn't belong to us.
4139 if (!partial
->key
&& *partial
->p
)
4142 for (p
= partial
; (p
> chain
) && all_zeroes((__le32
*) p
->bh
->b_data
, p
->p
); p
--)
4145 * OK, we've found the last block that must survive. The rest of our
4146 * branch should be detached before unlocking. However, if that rest
4147 * of branch is all ours and does not grow immediately from the inode
4148 * it's easier to cheat and just decrement partial->p.
4150 if (p
== chain
+ k
- 1 && p
> chain
) {
4154 /* Nope, don't do this in ext4. Must leave the tree intact */
4161 while (partial
> p
) {
4162 brelse(partial
->bh
);
4170 * Zero a number of block pointers in either an inode or an indirect block.
4171 * If we restart the transaction we must again get write access to the
4172 * indirect block for further modification.
4174 * We release `count' blocks on disk, but (last - first) may be greater
4175 * than `count' because there can be holes in there.
4177 static void ext4_clear_blocks(handle_t
*handle
, struct inode
*inode
,
4178 struct buffer_head
*bh
,
4179 ext4_fsblk_t block_to_free
,
4180 unsigned long count
, __le32
*first
,
4184 int is_metadata
= S_ISDIR(inode
->i_mode
) || S_ISLNK(inode
->i_mode
);
4186 if (try_to_extend_transaction(handle
, inode
)) {
4188 BUFFER_TRACE(bh
, "call ext4_handle_dirty_metadata");
4189 ext4_handle_dirty_metadata(handle
, inode
, bh
);
4191 ext4_mark_inode_dirty(handle
, inode
);
4192 ext4_truncate_restart_trans(handle
, inode
,
4193 blocks_for_truncate(inode
));
4195 BUFFER_TRACE(bh
, "retaking write access");
4196 ext4_journal_get_write_access(handle
, bh
);
4201 * Any buffers which are on the journal will be in memory. We
4202 * find them on the hash table so jbd2_journal_revoke() will
4203 * run jbd2_journal_forget() on them. We've already detached
4204 * each block from the file, so bforget() in
4205 * jbd2_journal_forget() should be safe.
4207 * AKPM: turn on bforget in jbd2_journal_forget()!!!
4209 for (p
= first
; p
< last
; p
++) {
4210 u32 nr
= le32_to_cpu(*p
);
4212 struct buffer_head
*tbh
;
4215 tbh
= sb_find_get_block(inode
->i_sb
, nr
);
4216 ext4_forget(handle
, is_metadata
, inode
, tbh
, nr
);
4220 ext4_free_blocks(handle
, inode
, block_to_free
, count
, is_metadata
);
4224 * ext4_free_data - free a list of data blocks
4225 * @handle: handle for this transaction
4226 * @inode: inode we are dealing with
4227 * @this_bh: indirect buffer_head which contains *@first and *@last
4228 * @first: array of block numbers
4229 * @last: points immediately past the end of array
4231 * We are freeing all blocks refered from that array (numbers are stored as
4232 * little-endian 32-bit) and updating @inode->i_blocks appropriately.
4234 * We accumulate contiguous runs of blocks to free. Conveniently, if these
4235 * blocks are contiguous then releasing them at one time will only affect one
4236 * or two bitmap blocks (+ group descriptor(s) and superblock) and we won't
4237 * actually use a lot of journal space.
4239 * @this_bh will be %NULL if @first and @last point into the inode's direct
4242 static void ext4_free_data(handle_t
*handle
, struct inode
*inode
,
4243 struct buffer_head
*this_bh
,
4244 __le32
*first
, __le32
*last
)
4246 ext4_fsblk_t block_to_free
= 0; /* Starting block # of a run */
4247 unsigned long count
= 0; /* Number of blocks in the run */
4248 __le32
*block_to_free_p
= NULL
; /* Pointer into inode/ind
4251 ext4_fsblk_t nr
; /* Current block # */
4252 __le32
*p
; /* Pointer into inode/ind
4253 for current block */
4256 if (this_bh
) { /* For indirect block */
4257 BUFFER_TRACE(this_bh
, "get_write_access");
4258 err
= ext4_journal_get_write_access(handle
, this_bh
);
4259 /* Important: if we can't update the indirect pointers
4260 * to the blocks, we can't free them. */
4265 for (p
= first
; p
< last
; p
++) {
4266 nr
= le32_to_cpu(*p
);
4268 /* accumulate blocks to free if they're contiguous */
4271 block_to_free_p
= p
;
4273 } else if (nr
== block_to_free
+ count
) {
4276 ext4_clear_blocks(handle
, inode
, this_bh
,
4278 count
, block_to_free_p
, p
);
4280 block_to_free_p
= p
;
4287 ext4_clear_blocks(handle
, inode
, this_bh
, block_to_free
,
4288 count
, block_to_free_p
, p
);
4291 BUFFER_TRACE(this_bh
, "call ext4_handle_dirty_metadata");
4294 * The buffer head should have an attached journal head at this
4295 * point. However, if the data is corrupted and an indirect
4296 * block pointed to itself, it would have been detached when
4297 * the block was cleared. Check for this instead of OOPSing.
4299 if ((EXT4_JOURNAL(inode
) == NULL
) || bh2jh(this_bh
))
4300 ext4_handle_dirty_metadata(handle
, inode
, this_bh
);
4302 ext4_error(inode
->i_sb
, __func__
,
4303 "circular indirect block detected, "
4304 "inode=%lu, block=%llu",
4306 (unsigned long long) this_bh
->b_blocknr
);
4311 * ext4_free_branches - free an array of branches
4312 * @handle: JBD handle for this transaction
4313 * @inode: inode we are dealing with
4314 * @parent_bh: the buffer_head which contains *@first and *@last
4315 * @first: array of block numbers
4316 * @last: pointer immediately past the end of array
4317 * @depth: depth of the branches to free
4319 * We are freeing all blocks refered from these branches (numbers are
4320 * stored as little-endian 32-bit) and updating @inode->i_blocks
4323 static void ext4_free_branches(handle_t
*handle
, struct inode
*inode
,
4324 struct buffer_head
*parent_bh
,
4325 __le32
*first
, __le32
*last
, int depth
)
4330 if (ext4_handle_is_aborted(handle
))
4334 struct buffer_head
*bh
;
4335 int addr_per_block
= EXT4_ADDR_PER_BLOCK(inode
->i_sb
);
4337 while (--p
>= first
) {
4338 nr
= le32_to_cpu(*p
);
4340 continue; /* A hole */
4342 /* Go read the buffer for the next level down */
4343 bh
= sb_bread(inode
->i_sb
, nr
);
4346 * A read failure? Report error and clear slot
4350 ext4_error(inode
->i_sb
, "ext4_free_branches",
4351 "Read failure, inode=%lu, block=%llu",
4356 /* This zaps the entire block. Bottom up. */
4357 BUFFER_TRACE(bh
, "free child branches");
4358 ext4_free_branches(handle
, inode
, bh
,
4359 (__le32
*) bh
->b_data
,
4360 (__le32
*) bh
->b_data
+ addr_per_block
,
4364 * We've probably journalled the indirect block several
4365 * times during the truncate. But it's no longer
4366 * needed and we now drop it from the transaction via
4367 * jbd2_journal_revoke().
4369 * That's easy if it's exclusively part of this
4370 * transaction. But if it's part of the committing
4371 * transaction then jbd2_journal_forget() will simply
4372 * brelse() it. That means that if the underlying
4373 * block is reallocated in ext4_get_block(),
4374 * unmap_underlying_metadata() will find this block
4375 * and will try to get rid of it. damn, damn.
4377 * If this block has already been committed to the
4378 * journal, a revoke record will be written. And
4379 * revoke records must be emitted *before* clearing
4380 * this block's bit in the bitmaps.
4382 ext4_forget(handle
, 1, inode
, bh
, bh
->b_blocknr
);
4385 * Everything below this this pointer has been
4386 * released. Now let this top-of-subtree go.
4388 * We want the freeing of this indirect block to be
4389 * atomic in the journal with the updating of the
4390 * bitmap block which owns it. So make some room in
4393 * We zero the parent pointer *after* freeing its
4394 * pointee in the bitmaps, so if extend_transaction()
4395 * for some reason fails to put the bitmap changes and
4396 * the release into the same transaction, recovery
4397 * will merely complain about releasing a free block,
4398 * rather than leaking blocks.
4400 if (ext4_handle_is_aborted(handle
))
4402 if (try_to_extend_transaction(handle
, inode
)) {
4403 ext4_mark_inode_dirty(handle
, inode
);
4404 ext4_truncate_restart_trans(handle
, inode
,
4405 blocks_for_truncate(inode
));
4408 ext4_free_blocks(handle
, inode
, nr
, 1, 1);
4412 * The block which we have just freed is
4413 * pointed to by an indirect block: journal it
4415 BUFFER_TRACE(parent_bh
, "get_write_access");
4416 if (!ext4_journal_get_write_access(handle
,
4419 BUFFER_TRACE(parent_bh
,
4420 "call ext4_handle_dirty_metadata");
4421 ext4_handle_dirty_metadata(handle
,
4428 /* We have reached the bottom of the tree. */
4429 BUFFER_TRACE(parent_bh
, "free data blocks");
4430 ext4_free_data(handle
, inode
, parent_bh
, first
, last
);
4434 int ext4_can_truncate(struct inode
*inode
)
4436 if (IS_APPEND(inode
) || IS_IMMUTABLE(inode
))
4438 if (S_ISREG(inode
->i_mode
))
4440 if (S_ISDIR(inode
->i_mode
))
4442 if (S_ISLNK(inode
->i_mode
))
4443 return !ext4_inode_is_fast_symlink(inode
);
4450 * We block out ext4_get_block() block instantiations across the entire
4451 * transaction, and VFS/VM ensures that ext4_truncate() cannot run
4452 * simultaneously on behalf of the same inode.
4454 * As we work through the truncate and commmit bits of it to the journal there
4455 * is one core, guiding principle: the file's tree must always be consistent on
4456 * disk. We must be able to restart the truncate after a crash.
4458 * The file's tree may be transiently inconsistent in memory (although it
4459 * probably isn't), but whenever we close off and commit a journal transaction,
4460 * the contents of (the filesystem + the journal) must be consistent and
4461 * restartable. It's pretty simple, really: bottom up, right to left (although
4462 * left-to-right works OK too).
4464 * Note that at recovery time, journal replay occurs *before* the restart of
4465 * truncate against the orphan inode list.
4467 * The committed inode has the new, desired i_size (which is the same as
4468 * i_disksize in this case). After a crash, ext4_orphan_cleanup() will see
4469 * that this inode's truncate did not complete and it will again call
4470 * ext4_truncate() to have another go. So there will be instantiated blocks
4471 * to the right of the truncation point in a crashed ext4 filesystem. But
4472 * that's fine - as long as they are linked from the inode, the post-crash
4473 * ext4_truncate() run will find them and release them.
4475 void ext4_truncate(struct inode
*inode
)
4478 struct ext4_inode_info
*ei
= EXT4_I(inode
);
4479 __le32
*i_data
= ei
->i_data
;
4480 int addr_per_block
= EXT4_ADDR_PER_BLOCK(inode
->i_sb
);
4481 struct address_space
*mapping
= inode
->i_mapping
;
4482 ext4_lblk_t offsets
[4];
4487 ext4_lblk_t last_block
;
4488 unsigned blocksize
= inode
->i_sb
->s_blocksize
;
4490 if (!ext4_can_truncate(inode
))
4493 if (inode
->i_size
== 0 && !test_opt(inode
->i_sb
, NO_AUTO_DA_ALLOC
))
4494 ext4_set_inode_state(inode
, EXT4_STATE_DA_ALLOC_CLOSE
);
4496 if (EXT4_I(inode
)->i_flags
& EXT4_EXTENTS_FL
) {
4497 ext4_ext_truncate(inode
);
4501 handle
= start_transaction(inode
);
4503 return; /* AKPM: return what? */
4505 last_block
= (inode
->i_size
+ blocksize
-1)
4506 >> EXT4_BLOCK_SIZE_BITS(inode
->i_sb
);
4508 if (inode
->i_size
& (blocksize
- 1))
4509 if (ext4_block_truncate_page(handle
, mapping
, inode
->i_size
))
4512 n
= ext4_block_to_path(inode
, last_block
, offsets
, NULL
);
4514 goto out_stop
; /* error */
4517 * OK. This truncate is going to happen. We add the inode to the
4518 * orphan list, so that if this truncate spans multiple transactions,
4519 * and we crash, we will resume the truncate when the filesystem
4520 * recovers. It also marks the inode dirty, to catch the new size.
4522 * Implication: the file must always be in a sane, consistent
4523 * truncatable state while each transaction commits.
4525 if (ext4_orphan_add(handle
, inode
))
4529 * From here we block out all ext4_get_block() callers who want to
4530 * modify the block allocation tree.
4532 down_write(&ei
->i_data_sem
);
4534 ext4_discard_preallocations(inode
);
4537 * The orphan list entry will now protect us from any crash which
4538 * occurs before the truncate completes, so it is now safe to propagate
4539 * the new, shorter inode size (held for now in i_size) into the
4540 * on-disk inode. We do this via i_disksize, which is the value which
4541 * ext4 *really* writes onto the disk inode.
4543 ei
->i_disksize
= inode
->i_size
;
4545 if (n
== 1) { /* direct blocks */
4546 ext4_free_data(handle
, inode
, NULL
, i_data
+offsets
[0],
4547 i_data
+ EXT4_NDIR_BLOCKS
);
4551 partial
= ext4_find_shared(inode
, n
, offsets
, chain
, &nr
);
4552 /* Kill the top of shared branch (not detached) */
4554 if (partial
== chain
) {
4555 /* Shared branch grows from the inode */
4556 ext4_free_branches(handle
, inode
, NULL
,
4557 &nr
, &nr
+1, (chain
+n
-1) - partial
);
4560 * We mark the inode dirty prior to restart,
4561 * and prior to stop. No need for it here.
4564 /* Shared branch grows from an indirect block */
4565 BUFFER_TRACE(partial
->bh
, "get_write_access");
4566 ext4_free_branches(handle
, inode
, partial
->bh
,
4568 partial
->p
+1, (chain
+n
-1) - partial
);
4571 /* Clear the ends of indirect blocks on the shared branch */
4572 while (partial
> chain
) {
4573 ext4_free_branches(handle
, inode
, partial
->bh
, partial
->p
+ 1,
4574 (__le32
*)partial
->bh
->b_data
+addr_per_block
,
4575 (chain
+n
-1) - partial
);
4576 BUFFER_TRACE(partial
->bh
, "call brelse");
4577 brelse(partial
->bh
);
4581 /* Kill the remaining (whole) subtrees */
4582 switch (offsets
[0]) {
4584 nr
= i_data
[EXT4_IND_BLOCK
];
4586 ext4_free_branches(handle
, inode
, NULL
, &nr
, &nr
+1, 1);
4587 i_data
[EXT4_IND_BLOCK
] = 0;
4589 case EXT4_IND_BLOCK
:
4590 nr
= i_data
[EXT4_DIND_BLOCK
];
4592 ext4_free_branches(handle
, inode
, NULL
, &nr
, &nr
+1, 2);
4593 i_data
[EXT4_DIND_BLOCK
] = 0;
4595 case EXT4_DIND_BLOCK
:
4596 nr
= i_data
[EXT4_TIND_BLOCK
];
4598 ext4_free_branches(handle
, inode
, NULL
, &nr
, &nr
+1, 3);
4599 i_data
[EXT4_TIND_BLOCK
] = 0;
4601 case EXT4_TIND_BLOCK
:
4605 up_write(&ei
->i_data_sem
);
4606 inode
->i_mtime
= inode
->i_ctime
= ext4_current_time(inode
);
4607 ext4_mark_inode_dirty(handle
, inode
);
4610 * In a multi-transaction truncate, we only make the final transaction
4614 ext4_handle_sync(handle
);
4617 * If this was a simple ftruncate(), and the file will remain alive
4618 * then we need to clear up the orphan record which we created above.
4619 * However, if this was a real unlink then we were called by
4620 * ext4_delete_inode(), and we allow that function to clean up the
4621 * orphan info for us.
4624 ext4_orphan_del(handle
, inode
);
4626 ext4_journal_stop(handle
);
4630 * ext4_get_inode_loc returns with an extra refcount against the inode's
4631 * underlying buffer_head on success. If 'in_mem' is true, we have all
4632 * data in memory that is needed to recreate the on-disk version of this
4635 static int __ext4_get_inode_loc(struct inode
*inode
,
4636 struct ext4_iloc
*iloc
, int in_mem
)
4638 struct ext4_group_desc
*gdp
;
4639 struct buffer_head
*bh
;
4640 struct super_block
*sb
= inode
->i_sb
;
4642 int inodes_per_block
, inode_offset
;
4645 if (!ext4_valid_inum(sb
, inode
->i_ino
))
4648 iloc
->block_group
= (inode
->i_ino
- 1) / EXT4_INODES_PER_GROUP(sb
);
4649 gdp
= ext4_get_group_desc(sb
, iloc
->block_group
, NULL
);
4654 * Figure out the offset within the block group inode table
4656 inodes_per_block
= (EXT4_BLOCK_SIZE(sb
) / EXT4_INODE_SIZE(sb
));
4657 inode_offset
= ((inode
->i_ino
- 1) %
4658 EXT4_INODES_PER_GROUP(sb
));
4659 block
= ext4_inode_table(sb
, gdp
) + (inode_offset
/ inodes_per_block
);
4660 iloc
->offset
= (inode_offset
% inodes_per_block
) * EXT4_INODE_SIZE(sb
);
4662 bh
= sb_getblk(sb
, block
);
4664 ext4_error(sb
, "ext4_get_inode_loc", "unable to read "
4665 "inode block - inode=%lu, block=%llu",
4666 inode
->i_ino
, block
);
4669 if (!buffer_uptodate(bh
)) {
4673 * If the buffer has the write error flag, we have failed
4674 * to write out another inode in the same block. In this
4675 * case, we don't have to read the block because we may
4676 * read the old inode data successfully.
4678 if (buffer_write_io_error(bh
) && !buffer_uptodate(bh
))
4679 set_buffer_uptodate(bh
);
4681 if (buffer_uptodate(bh
)) {
4682 /* someone brought it uptodate while we waited */
4688 * If we have all information of the inode in memory and this
4689 * is the only valid inode in the block, we need not read the
4693 struct buffer_head
*bitmap_bh
;
4696 start
= inode_offset
& ~(inodes_per_block
- 1);
4698 /* Is the inode bitmap in cache? */
4699 bitmap_bh
= sb_getblk(sb
, ext4_inode_bitmap(sb
, gdp
));
4704 * If the inode bitmap isn't in cache then the
4705 * optimisation may end up performing two reads instead
4706 * of one, so skip it.
4708 if (!buffer_uptodate(bitmap_bh
)) {
4712 for (i
= start
; i
< start
+ inodes_per_block
; i
++) {
4713 if (i
== inode_offset
)
4715 if (ext4_test_bit(i
, bitmap_bh
->b_data
))
4719 if (i
== start
+ inodes_per_block
) {
4720 /* all other inodes are free, so skip I/O */
4721 memset(bh
->b_data
, 0, bh
->b_size
);
4722 set_buffer_uptodate(bh
);
4730 * If we need to do any I/O, try to pre-readahead extra
4731 * blocks from the inode table.
4733 if (EXT4_SB(sb
)->s_inode_readahead_blks
) {
4734 ext4_fsblk_t b
, end
, table
;
4737 table
= ext4_inode_table(sb
, gdp
);
4738 /* s_inode_readahead_blks is always a power of 2 */
4739 b
= block
& ~(EXT4_SB(sb
)->s_inode_readahead_blks
-1);
4742 end
= b
+ EXT4_SB(sb
)->s_inode_readahead_blks
;
4743 num
= EXT4_INODES_PER_GROUP(sb
);
4744 if (EXT4_HAS_RO_COMPAT_FEATURE(sb
,
4745 EXT4_FEATURE_RO_COMPAT_GDT_CSUM
))
4746 num
-= ext4_itable_unused_count(sb
, gdp
);
4747 table
+= num
/ inodes_per_block
;
4751 sb_breadahead(sb
, b
++);
4755 * There are other valid inodes in the buffer, this inode
4756 * has in-inode xattrs, or we don't have this inode in memory.
4757 * Read the block from disk.
4760 bh
->b_end_io
= end_buffer_read_sync
;
4761 submit_bh(READ_META
, bh
);
4763 if (!buffer_uptodate(bh
)) {
4764 ext4_error(sb
, __func__
,
4765 "unable to read inode block - inode=%lu, "
4766 "block=%llu", inode
->i_ino
, block
);
4776 int ext4_get_inode_loc(struct inode
*inode
, struct ext4_iloc
*iloc
)
4778 /* We have all inode data except xattrs in memory here. */
4779 return __ext4_get_inode_loc(inode
, iloc
,
4780 !ext4_test_inode_state(inode
, EXT4_STATE_XATTR
));
4783 void ext4_set_inode_flags(struct inode
*inode
)
4785 unsigned int flags
= EXT4_I(inode
)->i_flags
;
4787 inode
->i_flags
&= ~(S_SYNC
|S_APPEND
|S_IMMUTABLE
|S_NOATIME
|S_DIRSYNC
);
4788 if (flags
& EXT4_SYNC_FL
)
4789 inode
->i_flags
|= S_SYNC
;
4790 if (flags
& EXT4_APPEND_FL
)
4791 inode
->i_flags
|= S_APPEND
;
4792 if (flags
& EXT4_IMMUTABLE_FL
)
4793 inode
->i_flags
|= S_IMMUTABLE
;
4794 if (flags
& EXT4_NOATIME_FL
)
4795 inode
->i_flags
|= S_NOATIME
;
4796 if (flags
& EXT4_DIRSYNC_FL
)
4797 inode
->i_flags
|= S_DIRSYNC
;
4800 /* Propagate flags from i_flags to EXT4_I(inode)->i_flags */
4801 void ext4_get_inode_flags(struct ext4_inode_info
*ei
)
4803 unsigned int flags
= ei
->vfs_inode
.i_flags
;
4805 ei
->i_flags
&= ~(EXT4_SYNC_FL
|EXT4_APPEND_FL
|
4806 EXT4_IMMUTABLE_FL
|EXT4_NOATIME_FL
|EXT4_DIRSYNC_FL
);
4808 ei
->i_flags
|= EXT4_SYNC_FL
;
4809 if (flags
& S_APPEND
)
4810 ei
->i_flags
|= EXT4_APPEND_FL
;
4811 if (flags
& S_IMMUTABLE
)
4812 ei
->i_flags
|= EXT4_IMMUTABLE_FL
;
4813 if (flags
& S_NOATIME
)
4814 ei
->i_flags
|= EXT4_NOATIME_FL
;
4815 if (flags
& S_DIRSYNC
)
4816 ei
->i_flags
|= EXT4_DIRSYNC_FL
;
4819 static blkcnt_t
ext4_inode_blocks(struct ext4_inode
*raw_inode
,
4820 struct ext4_inode_info
*ei
)
4823 struct inode
*inode
= &(ei
->vfs_inode
);
4824 struct super_block
*sb
= inode
->i_sb
;
4826 if (EXT4_HAS_RO_COMPAT_FEATURE(sb
,
4827 EXT4_FEATURE_RO_COMPAT_HUGE_FILE
)) {
4828 /* we are using combined 48 bit field */
4829 i_blocks
= ((u64
)le16_to_cpu(raw_inode
->i_blocks_high
)) << 32 |
4830 le32_to_cpu(raw_inode
->i_blocks_lo
);
4831 if (ei
->i_flags
& EXT4_HUGE_FILE_FL
) {
4832 /* i_blocks represent file system block size */
4833 return i_blocks
<< (inode
->i_blkbits
- 9);
4838 return le32_to_cpu(raw_inode
->i_blocks_lo
);
4842 struct inode
*ext4_iget(struct super_block
*sb
, unsigned long ino
)
4844 struct ext4_iloc iloc
;
4845 struct ext4_inode
*raw_inode
;
4846 struct ext4_inode_info
*ei
;
4847 struct inode
*inode
;
4848 journal_t
*journal
= EXT4_SB(sb
)->s_journal
;
4852 inode
= iget_locked(sb
, ino
);
4854 return ERR_PTR(-ENOMEM
);
4855 if (!(inode
->i_state
& I_NEW
))
4861 ret
= __ext4_get_inode_loc(inode
, &iloc
, 0);
4864 raw_inode
= ext4_raw_inode(&iloc
);
4865 inode
->i_mode
= le16_to_cpu(raw_inode
->i_mode
);
4866 inode
->i_uid
= (uid_t
)le16_to_cpu(raw_inode
->i_uid_low
);
4867 inode
->i_gid
= (gid_t
)le16_to_cpu(raw_inode
->i_gid_low
);
4868 if (!(test_opt(inode
->i_sb
, NO_UID32
))) {
4869 inode
->i_uid
|= le16_to_cpu(raw_inode
->i_uid_high
) << 16;
4870 inode
->i_gid
|= le16_to_cpu(raw_inode
->i_gid_high
) << 16;
4872 inode
->i_nlink
= le16_to_cpu(raw_inode
->i_links_count
);
4874 ei
->i_state_flags
= 0;
4875 ei
->i_dir_start_lookup
= 0;
4876 ei
->i_dtime
= le32_to_cpu(raw_inode
->i_dtime
);
4877 /* We now have enough fields to check if the inode was active or not.
4878 * This is needed because nfsd might try to access dead inodes
4879 * the test is that same one that e2fsck uses
4880 * NeilBrown 1999oct15
4882 if (inode
->i_nlink
== 0) {
4883 if (inode
->i_mode
== 0 ||
4884 !(EXT4_SB(inode
->i_sb
)->s_mount_state
& EXT4_ORPHAN_FS
)) {
4885 /* this inode is deleted */
4889 /* The only unlinked inodes we let through here have
4890 * valid i_mode and are being read by the orphan
4891 * recovery code: that's fine, we're about to complete
4892 * the process of deleting those. */
4894 ei
->i_flags
= le32_to_cpu(raw_inode
->i_flags
);
4895 inode
->i_blocks
= ext4_inode_blocks(raw_inode
, ei
);
4896 ei
->i_file_acl
= le32_to_cpu(raw_inode
->i_file_acl_lo
);
4897 if (EXT4_HAS_INCOMPAT_FEATURE(sb
, EXT4_FEATURE_INCOMPAT_64BIT
))
4899 ((__u64
)le16_to_cpu(raw_inode
->i_file_acl_high
)) << 32;
4900 inode
->i_size
= ext4_isize(raw_inode
);
4901 ei
->i_disksize
= inode
->i_size
;
4903 ei
->i_reserved_quota
= 0;
4905 inode
->i_generation
= le32_to_cpu(raw_inode
->i_generation
);
4906 ei
->i_block_group
= iloc
.block_group
;
4907 ei
->i_last_alloc_group
= ~0;
4909 * NOTE! The in-memory inode i_data array is in little-endian order
4910 * even on big-endian machines: we do NOT byteswap the block numbers!
4912 for (block
= 0; block
< EXT4_N_BLOCKS
; block
++)
4913 ei
->i_data
[block
] = raw_inode
->i_block
[block
];
4914 INIT_LIST_HEAD(&ei
->i_orphan
);
4917 * Set transaction id's of transactions that have to be committed
4918 * to finish f[data]sync. We set them to currently running transaction
4919 * as we cannot be sure that the inode or some of its metadata isn't
4920 * part of the transaction - the inode could have been reclaimed and
4921 * now it is reread from disk.
4924 transaction_t
*transaction
;
4927 spin_lock(&journal
->j_state_lock
);
4928 if (journal
->j_running_transaction
)
4929 transaction
= journal
->j_running_transaction
;
4931 transaction
= journal
->j_committing_transaction
;
4933 tid
= transaction
->t_tid
;
4935 tid
= journal
->j_commit_sequence
;
4936 spin_unlock(&journal
->j_state_lock
);
4937 ei
->i_sync_tid
= tid
;
4938 ei
->i_datasync_tid
= tid
;
4941 if (EXT4_INODE_SIZE(inode
->i_sb
) > EXT4_GOOD_OLD_INODE_SIZE
) {
4942 ei
->i_extra_isize
= le16_to_cpu(raw_inode
->i_extra_isize
);
4943 if (EXT4_GOOD_OLD_INODE_SIZE
+ ei
->i_extra_isize
>
4944 EXT4_INODE_SIZE(inode
->i_sb
)) {
4948 if (ei
->i_extra_isize
== 0) {
4949 /* The extra space is currently unused. Use it. */
4950 ei
->i_extra_isize
= sizeof(struct ext4_inode
) -
4951 EXT4_GOOD_OLD_INODE_SIZE
;
4953 __le32
*magic
= (void *)raw_inode
+
4954 EXT4_GOOD_OLD_INODE_SIZE
+
4956 if (*magic
== cpu_to_le32(EXT4_XATTR_MAGIC
))
4957 ext4_set_inode_state(inode
, EXT4_STATE_XATTR
);
4960 ei
->i_extra_isize
= 0;
4962 EXT4_INODE_GET_XTIME(i_ctime
, inode
, raw_inode
);
4963 EXT4_INODE_GET_XTIME(i_mtime
, inode
, raw_inode
);
4964 EXT4_INODE_GET_XTIME(i_atime
, inode
, raw_inode
);
4965 EXT4_EINODE_GET_XTIME(i_crtime
, ei
, raw_inode
);
4967 inode
->i_version
= le32_to_cpu(raw_inode
->i_disk_version
);
4968 if (EXT4_INODE_SIZE(inode
->i_sb
) > EXT4_GOOD_OLD_INODE_SIZE
) {
4969 if (EXT4_FITS_IN_INODE(raw_inode
, ei
, i_version_hi
))
4971 (__u64
)(le32_to_cpu(raw_inode
->i_version_hi
)) << 32;
4975 if (ei
->i_file_acl
&&
4976 !ext4_data_block_valid(EXT4_SB(sb
), ei
->i_file_acl
, 1)) {
4977 ext4_error(sb
, __func__
,
4978 "bad extended attribute block %llu in inode #%lu",
4979 ei
->i_file_acl
, inode
->i_ino
);
4982 } else if (ei
->i_flags
& EXT4_EXTENTS_FL
) {
4983 if (S_ISREG(inode
->i_mode
) || S_ISDIR(inode
->i_mode
) ||
4984 (S_ISLNK(inode
->i_mode
) &&
4985 !ext4_inode_is_fast_symlink(inode
)))
4986 /* Validate extent which is part of inode */
4987 ret
= ext4_ext_check_inode(inode
);
4988 } else if (S_ISREG(inode
->i_mode
) || S_ISDIR(inode
->i_mode
) ||
4989 (S_ISLNK(inode
->i_mode
) &&
4990 !ext4_inode_is_fast_symlink(inode
))) {
4991 /* Validate block references which are part of inode */
4992 ret
= ext4_check_inode_blockref(inode
);
4997 if (S_ISREG(inode
->i_mode
)) {
4998 inode
->i_op
= &ext4_file_inode_operations
;
4999 inode
->i_fop
= &ext4_file_operations
;
5000 ext4_set_aops(inode
);
5001 } else if (S_ISDIR(inode
->i_mode
)) {
5002 inode
->i_op
= &ext4_dir_inode_operations
;
5003 inode
->i_fop
= &ext4_dir_operations
;
5004 } else if (S_ISLNK(inode
->i_mode
)) {
5005 if (ext4_inode_is_fast_symlink(inode
)) {
5006 inode
->i_op
= &ext4_fast_symlink_inode_operations
;
5007 nd_terminate_link(ei
->i_data
, inode
->i_size
,
5008 sizeof(ei
->i_data
) - 1);
5010 inode
->i_op
= &ext4_symlink_inode_operations
;
5011 ext4_set_aops(inode
);
5013 } else if (S_ISCHR(inode
->i_mode
) || S_ISBLK(inode
->i_mode
) ||
5014 S_ISFIFO(inode
->i_mode
) || S_ISSOCK(inode
->i_mode
)) {
5015 inode
->i_op
= &ext4_special_inode_operations
;
5016 if (raw_inode
->i_block
[0])
5017 init_special_inode(inode
, inode
->i_mode
,
5018 old_decode_dev(le32_to_cpu(raw_inode
->i_block
[0])));
5020 init_special_inode(inode
, inode
->i_mode
,
5021 new_decode_dev(le32_to_cpu(raw_inode
->i_block
[1])));
5024 ext4_error(inode
->i_sb
, __func__
,
5025 "bogus i_mode (%o) for inode=%lu",
5026 inode
->i_mode
, inode
->i_ino
);
5030 ext4_set_inode_flags(inode
);
5031 unlock_new_inode(inode
);
5037 return ERR_PTR(ret
);
5040 static int ext4_inode_blocks_set(handle_t
*handle
,
5041 struct ext4_inode
*raw_inode
,
5042 struct ext4_inode_info
*ei
)
5044 struct inode
*inode
= &(ei
->vfs_inode
);
5045 u64 i_blocks
= inode
->i_blocks
;
5046 struct super_block
*sb
= inode
->i_sb
;
5048 if (i_blocks
<= ~0U) {
5050 * i_blocks can be represnted in a 32 bit variable
5051 * as multiple of 512 bytes
5053 raw_inode
->i_blocks_lo
= cpu_to_le32(i_blocks
);
5054 raw_inode
->i_blocks_high
= 0;
5055 ei
->i_flags
&= ~EXT4_HUGE_FILE_FL
;
5058 if (!EXT4_HAS_RO_COMPAT_FEATURE(sb
, EXT4_FEATURE_RO_COMPAT_HUGE_FILE
))
5061 if (i_blocks
<= 0xffffffffffffULL
) {
5063 * i_blocks can be represented in a 48 bit variable
5064 * as multiple of 512 bytes
5066 raw_inode
->i_blocks_lo
= cpu_to_le32(i_blocks
);
5067 raw_inode
->i_blocks_high
= cpu_to_le16(i_blocks
>> 32);
5068 ei
->i_flags
&= ~EXT4_HUGE_FILE_FL
;
5070 ei
->i_flags
|= EXT4_HUGE_FILE_FL
;
5071 /* i_block is stored in file system block size */
5072 i_blocks
= i_blocks
>> (inode
->i_blkbits
- 9);
5073 raw_inode
->i_blocks_lo
= cpu_to_le32(i_blocks
);
5074 raw_inode
->i_blocks_high
= cpu_to_le16(i_blocks
>> 32);
5080 * Post the struct inode info into an on-disk inode location in the
5081 * buffer-cache. This gobbles the caller's reference to the
5082 * buffer_head in the inode location struct.
5084 * The caller must have write access to iloc->bh.
5086 static int ext4_do_update_inode(handle_t
*handle
,
5087 struct inode
*inode
,
5088 struct ext4_iloc
*iloc
)
5090 struct ext4_inode
*raw_inode
= ext4_raw_inode(iloc
);
5091 struct ext4_inode_info
*ei
= EXT4_I(inode
);
5092 struct buffer_head
*bh
= iloc
->bh
;
5093 int err
= 0, rc
, block
;
5095 /* For fields not not tracking in the in-memory inode,
5096 * initialise them to zero for new inodes. */
5097 if (ext4_test_inode_state(inode
, EXT4_STATE_NEW
))
5098 memset(raw_inode
, 0, EXT4_SB(inode
->i_sb
)->s_inode_size
);
5100 ext4_get_inode_flags(ei
);
5101 raw_inode
->i_mode
= cpu_to_le16(inode
->i_mode
);
5102 if (!(test_opt(inode
->i_sb
, NO_UID32
))) {
5103 raw_inode
->i_uid_low
= cpu_to_le16(low_16_bits(inode
->i_uid
));
5104 raw_inode
->i_gid_low
= cpu_to_le16(low_16_bits(inode
->i_gid
));
5106 * Fix up interoperability with old kernels. Otherwise, old inodes get
5107 * re-used with the upper 16 bits of the uid/gid intact
5110 raw_inode
->i_uid_high
=
5111 cpu_to_le16(high_16_bits(inode
->i_uid
));
5112 raw_inode
->i_gid_high
=
5113 cpu_to_le16(high_16_bits(inode
->i_gid
));
5115 raw_inode
->i_uid_high
= 0;
5116 raw_inode
->i_gid_high
= 0;
5119 raw_inode
->i_uid_low
=
5120 cpu_to_le16(fs_high2lowuid(inode
->i_uid
));
5121 raw_inode
->i_gid_low
=
5122 cpu_to_le16(fs_high2lowgid(inode
->i_gid
));
5123 raw_inode
->i_uid_high
= 0;
5124 raw_inode
->i_gid_high
= 0;
5126 raw_inode
->i_links_count
= cpu_to_le16(inode
->i_nlink
);
5128 EXT4_INODE_SET_XTIME(i_ctime
, inode
, raw_inode
);
5129 EXT4_INODE_SET_XTIME(i_mtime
, inode
, raw_inode
);
5130 EXT4_INODE_SET_XTIME(i_atime
, inode
, raw_inode
);
5131 EXT4_EINODE_SET_XTIME(i_crtime
, ei
, raw_inode
);
5133 if (ext4_inode_blocks_set(handle
, raw_inode
, ei
))
5135 raw_inode
->i_dtime
= cpu_to_le32(ei
->i_dtime
);
5136 raw_inode
->i_flags
= cpu_to_le32(ei
->i_flags
);
5137 if (EXT4_SB(inode
->i_sb
)->s_es
->s_creator_os
!=
5138 cpu_to_le32(EXT4_OS_HURD
))
5139 raw_inode
->i_file_acl_high
=
5140 cpu_to_le16(ei
->i_file_acl
>> 32);
5141 raw_inode
->i_file_acl_lo
= cpu_to_le32(ei
->i_file_acl
);
5142 ext4_isize_set(raw_inode
, ei
->i_disksize
);
5143 if (ei
->i_disksize
> 0x7fffffffULL
) {
5144 struct super_block
*sb
= inode
->i_sb
;
5145 if (!EXT4_HAS_RO_COMPAT_FEATURE(sb
,
5146 EXT4_FEATURE_RO_COMPAT_LARGE_FILE
) ||
5147 EXT4_SB(sb
)->s_es
->s_rev_level
==
5148 cpu_to_le32(EXT4_GOOD_OLD_REV
)) {
5149 /* If this is the first large file
5150 * created, add a flag to the superblock.
5152 err
= ext4_journal_get_write_access(handle
,
5153 EXT4_SB(sb
)->s_sbh
);
5156 ext4_update_dynamic_rev(sb
);
5157 EXT4_SET_RO_COMPAT_FEATURE(sb
,
5158 EXT4_FEATURE_RO_COMPAT_LARGE_FILE
);
5160 ext4_handle_sync(handle
);
5161 err
= ext4_handle_dirty_metadata(handle
, NULL
,
5162 EXT4_SB(sb
)->s_sbh
);
5165 raw_inode
->i_generation
= cpu_to_le32(inode
->i_generation
);
5166 if (S_ISCHR(inode
->i_mode
) || S_ISBLK(inode
->i_mode
)) {
5167 if (old_valid_dev(inode
->i_rdev
)) {
5168 raw_inode
->i_block
[0] =
5169 cpu_to_le32(old_encode_dev(inode
->i_rdev
));
5170 raw_inode
->i_block
[1] = 0;
5172 raw_inode
->i_block
[0] = 0;
5173 raw_inode
->i_block
[1] =
5174 cpu_to_le32(new_encode_dev(inode
->i_rdev
));
5175 raw_inode
->i_block
[2] = 0;
5178 for (block
= 0; block
< EXT4_N_BLOCKS
; block
++)
5179 raw_inode
->i_block
[block
] = ei
->i_data
[block
];
5181 raw_inode
->i_disk_version
= cpu_to_le32(inode
->i_version
);
5182 if (ei
->i_extra_isize
) {
5183 if (EXT4_FITS_IN_INODE(raw_inode
, ei
, i_version_hi
))
5184 raw_inode
->i_version_hi
=
5185 cpu_to_le32(inode
->i_version
>> 32);
5186 raw_inode
->i_extra_isize
= cpu_to_le16(ei
->i_extra_isize
);
5189 BUFFER_TRACE(bh
, "call ext4_handle_dirty_metadata");
5190 rc
= ext4_handle_dirty_metadata(handle
, NULL
, bh
);
5193 ext4_clear_inode_state(inode
, EXT4_STATE_NEW
);
5195 ext4_update_inode_fsync_trans(handle
, inode
, 0);
5198 ext4_std_error(inode
->i_sb
, err
);
5203 * ext4_write_inode()
5205 * We are called from a few places:
5207 * - Within generic_file_write() for O_SYNC files.
5208 * Here, there will be no transaction running. We wait for any running
5209 * trasnaction to commit.
5211 * - Within sys_sync(), kupdate and such.
5212 * We wait on commit, if tol to.
5214 * - Within prune_icache() (PF_MEMALLOC == true)
5215 * Here we simply return. We can't afford to block kswapd on the
5218 * In all cases it is actually safe for us to return without doing anything,
5219 * because the inode has been copied into a raw inode buffer in
5220 * ext4_mark_inode_dirty(). This is a correctness thing for O_SYNC and for
5223 * Note that we are absolutely dependent upon all inode dirtiers doing the
5224 * right thing: they *must* call mark_inode_dirty() after dirtying info in
5225 * which we are interested.
5227 * It would be a bug for them to not do this. The code:
5229 * mark_inode_dirty(inode)
5231 * inode->i_size = expr;
5233 * is in error because a kswapd-driven write_inode() could occur while
5234 * `stuff()' is running, and the new i_size will be lost. Plus the inode
5235 * will no longer be on the superblock's dirty inode list.
5237 int ext4_write_inode(struct inode
*inode
, int wait
)
5241 if (current
->flags
& PF_MEMALLOC
)
5244 if (EXT4_SB(inode
->i_sb
)->s_journal
) {
5245 if (ext4_journal_current_handle()) {
5246 jbd_debug(1, "called recursively, non-PF_MEMALLOC!\n");
5254 err
= ext4_force_commit(inode
->i_sb
);
5256 struct ext4_iloc iloc
;
5258 err
= ext4_get_inode_loc(inode
, &iloc
);
5262 sync_dirty_buffer(iloc
.bh
);
5263 if (buffer_req(iloc
.bh
) && !buffer_uptodate(iloc
.bh
)) {
5264 ext4_error(inode
->i_sb
, __func__
,
5265 "IO error syncing inode, "
5266 "inode=%lu, block=%llu",
5268 (unsigned long long)iloc
.bh
->b_blocknr
);
5278 * Called from notify_change.
5280 * We want to trap VFS attempts to truncate the file as soon as
5281 * possible. In particular, we want to make sure that when the VFS
5282 * shrinks i_size, we put the inode on the orphan list and modify
5283 * i_disksize immediately, so that during the subsequent flushing of
5284 * dirty pages and freeing of disk blocks, we can guarantee that any
5285 * commit will leave the blocks being flushed in an unused state on
5286 * disk. (On recovery, the inode will get truncated and the blocks will
5287 * be freed, so we have a strong guarantee that no future commit will
5288 * leave these blocks visible to the user.)
5290 * Another thing we have to assure is that if we are in ordered mode
5291 * and inode is still attached to the committing transaction, we must
5292 * we start writeout of all the dirty pages which are being truncated.
5293 * This way we are sure that all the data written in the previous
5294 * transaction are already on disk (truncate waits for pages under
5297 * Called with inode->i_mutex down.
5299 int ext4_setattr(struct dentry
*dentry
, struct iattr
*attr
)
5301 struct inode
*inode
= dentry
->d_inode
;
5303 const unsigned int ia_valid
= attr
->ia_valid
;
5305 error
= inode_change_ok(inode
, attr
);
5309 if ((ia_valid
& ATTR_UID
&& attr
->ia_uid
!= inode
->i_uid
) ||
5310 (ia_valid
& ATTR_GID
&& attr
->ia_gid
!= inode
->i_gid
)) {
5313 /* (user+group)*(old+new) structure, inode write (sb,
5314 * inode block, ? - but truncate inode update has it) */
5315 handle
= ext4_journal_start(inode
, (EXT4_MAXQUOTAS_INIT_BLOCKS(inode
->i_sb
)+
5316 EXT4_MAXQUOTAS_DEL_BLOCKS(inode
->i_sb
))+3);
5317 if (IS_ERR(handle
)) {
5318 error
= PTR_ERR(handle
);
5321 error
= vfs_dq_transfer(inode
, attr
) ? -EDQUOT
: 0;
5323 ext4_journal_stop(handle
);
5326 /* Update corresponding info in inode so that everything is in
5327 * one transaction */
5328 if (attr
->ia_valid
& ATTR_UID
)
5329 inode
->i_uid
= attr
->ia_uid
;
5330 if (attr
->ia_valid
& ATTR_GID
)
5331 inode
->i_gid
= attr
->ia_gid
;
5332 error
= ext4_mark_inode_dirty(handle
, inode
);
5333 ext4_journal_stop(handle
);
5336 if (attr
->ia_valid
& ATTR_SIZE
) {
5337 if (!(EXT4_I(inode
)->i_flags
& EXT4_EXTENTS_FL
)) {
5338 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
5340 if (attr
->ia_size
> sbi
->s_bitmap_maxbytes
) {
5347 if (S_ISREG(inode
->i_mode
) &&
5348 attr
->ia_valid
& ATTR_SIZE
&& attr
->ia_size
< inode
->i_size
) {
5351 handle
= ext4_journal_start(inode
, 3);
5352 if (IS_ERR(handle
)) {
5353 error
= PTR_ERR(handle
);
5357 error
= ext4_orphan_add(handle
, inode
);
5358 EXT4_I(inode
)->i_disksize
= attr
->ia_size
;
5359 rc
= ext4_mark_inode_dirty(handle
, inode
);
5362 ext4_journal_stop(handle
);
5364 if (ext4_should_order_data(inode
)) {
5365 error
= ext4_begin_ordered_truncate(inode
,
5368 /* Do as much error cleanup as possible */
5369 handle
= ext4_journal_start(inode
, 3);
5370 if (IS_ERR(handle
)) {
5371 ext4_orphan_del(NULL
, inode
);
5374 ext4_orphan_del(handle
, inode
);
5375 ext4_journal_stop(handle
);
5381 rc
= inode_setattr(inode
, attr
);
5383 /* If inode_setattr's call to ext4_truncate failed to get a
5384 * transaction handle at all, we need to clean up the in-core
5385 * orphan list manually. */
5387 ext4_orphan_del(NULL
, inode
);
5389 if (!rc
&& (ia_valid
& ATTR_MODE
))
5390 rc
= ext4_acl_chmod(inode
);
5393 ext4_std_error(inode
->i_sb
, error
);
5399 int ext4_getattr(struct vfsmount
*mnt
, struct dentry
*dentry
,
5402 struct inode
*inode
;
5403 unsigned long delalloc_blocks
;
5405 inode
= dentry
->d_inode
;
5406 generic_fillattr(inode
, stat
);
5409 * We can't update i_blocks if the block allocation is delayed
5410 * otherwise in the case of system crash before the real block
5411 * allocation is done, we will have i_blocks inconsistent with
5412 * on-disk file blocks.
5413 * We always keep i_blocks updated together with real
5414 * allocation. But to not confuse with user, stat
5415 * will return the blocks that include the delayed allocation
5416 * blocks for this file.
5418 spin_lock(&EXT4_I(inode
)->i_block_reservation_lock
);
5419 delalloc_blocks
= EXT4_I(inode
)->i_reserved_data_blocks
;
5420 spin_unlock(&EXT4_I(inode
)->i_block_reservation_lock
);
5422 stat
->blocks
+= (delalloc_blocks
<< inode
->i_sb
->s_blocksize_bits
)>>9;
5426 static int ext4_indirect_trans_blocks(struct inode
*inode
, int nrblocks
,
5431 /* if nrblocks are contiguous */
5434 * With N contiguous data blocks, it need at most
5435 * N/EXT4_ADDR_PER_BLOCK(inode->i_sb) indirect blocks
5436 * 2 dindirect blocks
5439 indirects
= nrblocks
/ EXT4_ADDR_PER_BLOCK(inode
->i_sb
);
5440 return indirects
+ 3;
5443 * if nrblocks are not contiguous, worse case, each block touch
5444 * a indirect block, and each indirect block touch a double indirect
5445 * block, plus a triple indirect block
5447 indirects
= nrblocks
* 2 + 1;
5451 static int ext4_index_trans_blocks(struct inode
*inode
, int nrblocks
, int chunk
)
5453 if (!(EXT4_I(inode
)->i_flags
& EXT4_EXTENTS_FL
))
5454 return ext4_indirect_trans_blocks(inode
, nrblocks
, chunk
);
5455 return ext4_ext_index_trans_blocks(inode
, nrblocks
, chunk
);
5459 * Account for index blocks, block groups bitmaps and block group
5460 * descriptor blocks if modify datablocks and index blocks
5461 * worse case, the indexs blocks spread over different block groups
5463 * If datablocks are discontiguous, they are possible to spread over
5464 * different block groups too. If they are contiugous, with flexbg,
5465 * they could still across block group boundary.
5467 * Also account for superblock, inode, quota and xattr blocks
5469 int ext4_meta_trans_blocks(struct inode
*inode
, int nrblocks
, int chunk
)
5471 ext4_group_t groups
, ngroups
= ext4_get_groups_count(inode
->i_sb
);
5477 * How many index blocks need to touch to modify nrblocks?
5478 * The "Chunk" flag indicating whether the nrblocks is
5479 * physically contiguous on disk
5481 * For Direct IO and fallocate, they calls get_block to allocate
5482 * one single extent at a time, so they could set the "Chunk" flag
5484 idxblocks
= ext4_index_trans_blocks(inode
, nrblocks
, chunk
);
5489 * Now let's see how many group bitmaps and group descriptors need
5499 if (groups
> ngroups
)
5501 if (groups
> EXT4_SB(inode
->i_sb
)->s_gdb_count
)
5502 gdpblocks
= EXT4_SB(inode
->i_sb
)->s_gdb_count
;
5504 /* bitmaps and block group descriptor blocks */
5505 ret
+= groups
+ gdpblocks
;
5507 /* Blocks for super block, inode, quota and xattr blocks */
5508 ret
+= EXT4_META_TRANS_BLOCKS(inode
->i_sb
);
5514 * Calulate the total number of credits to reserve to fit
5515 * the modification of a single pages into a single transaction,
5516 * which may include multiple chunks of block allocations.
5518 * This could be called via ext4_write_begin()
5520 * We need to consider the worse case, when
5521 * one new block per extent.
5523 int ext4_writepage_trans_blocks(struct inode
*inode
)
5525 int bpp
= ext4_journal_blocks_per_page(inode
);
5528 ret
= ext4_meta_trans_blocks(inode
, bpp
, 0);
5530 /* Account for data blocks for journalled mode */
5531 if (ext4_should_journal_data(inode
))
5537 * Calculate the journal credits for a chunk of data modification.
5539 * This is called from DIO, fallocate or whoever calling
5540 * ext4_get_blocks() to map/allocate a chunk of contigous disk blocks.
5542 * journal buffers for data blocks are not included here, as DIO
5543 * and fallocate do no need to journal data buffers.
5545 int ext4_chunk_trans_blocks(struct inode
*inode
, int nrblocks
)
5547 return ext4_meta_trans_blocks(inode
, nrblocks
, 1);
5551 * The caller must have previously called ext4_reserve_inode_write().
5552 * Give this, we know that the caller already has write access to iloc->bh.
5554 int ext4_mark_iloc_dirty(handle_t
*handle
,
5555 struct inode
*inode
, struct ext4_iloc
*iloc
)
5559 if (test_opt(inode
->i_sb
, I_VERSION
))
5560 inode_inc_iversion(inode
);
5562 /* the do_update_inode consumes one bh->b_count */
5565 /* ext4_do_update_inode() does jbd2_journal_dirty_metadata */
5566 err
= ext4_do_update_inode(handle
, inode
, iloc
);
5572 * On success, We end up with an outstanding reference count against
5573 * iloc->bh. This _must_ be cleaned up later.
5577 ext4_reserve_inode_write(handle_t
*handle
, struct inode
*inode
,
5578 struct ext4_iloc
*iloc
)
5582 err
= ext4_get_inode_loc(inode
, iloc
);
5584 BUFFER_TRACE(iloc
->bh
, "get_write_access");
5585 err
= ext4_journal_get_write_access(handle
, iloc
->bh
);
5591 ext4_std_error(inode
->i_sb
, err
);
5596 * Expand an inode by new_extra_isize bytes.
5597 * Returns 0 on success or negative error number on failure.
5599 static int ext4_expand_extra_isize(struct inode
*inode
,
5600 unsigned int new_extra_isize
,
5601 struct ext4_iloc iloc
,
5604 struct ext4_inode
*raw_inode
;
5605 struct ext4_xattr_ibody_header
*header
;
5606 struct ext4_xattr_entry
*entry
;
5608 if (EXT4_I(inode
)->i_extra_isize
>= new_extra_isize
)
5611 raw_inode
= ext4_raw_inode(&iloc
);
5613 header
= IHDR(inode
, raw_inode
);
5614 entry
= IFIRST(header
);
5616 /* No extended attributes present */
5617 if (!ext4_test_inode_state(inode
, EXT4_STATE_XATTR
) ||
5618 header
->h_magic
!= cpu_to_le32(EXT4_XATTR_MAGIC
)) {
5619 memset((void *)raw_inode
+ EXT4_GOOD_OLD_INODE_SIZE
, 0,
5621 EXT4_I(inode
)->i_extra_isize
= new_extra_isize
;
5625 /* try to expand with EAs present */
5626 return ext4_expand_extra_isize_ea(inode
, new_extra_isize
,
5631 * What we do here is to mark the in-core inode as clean with respect to inode
5632 * dirtiness (it may still be data-dirty).
5633 * This means that the in-core inode may be reaped by prune_icache
5634 * without having to perform any I/O. This is a very good thing,
5635 * because *any* task may call prune_icache - even ones which
5636 * have a transaction open against a different journal.
5638 * Is this cheating? Not really. Sure, we haven't written the
5639 * inode out, but prune_icache isn't a user-visible syncing function.
5640 * Whenever the user wants stuff synced (sys_sync, sys_msync, sys_fsync)
5641 * we start and wait on commits.
5643 * Is this efficient/effective? Well, we're being nice to the system
5644 * by cleaning up our inodes proactively so they can be reaped
5645 * without I/O. But we are potentially leaving up to five seconds'
5646 * worth of inodes floating about which prune_icache wants us to
5647 * write out. One way to fix that would be to get prune_icache()
5648 * to do a write_super() to free up some memory. It has the desired
5651 int ext4_mark_inode_dirty(handle_t
*handle
, struct inode
*inode
)
5653 struct ext4_iloc iloc
;
5654 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
5655 static unsigned int mnt_count
;
5659 err
= ext4_reserve_inode_write(handle
, inode
, &iloc
);
5660 if (ext4_handle_valid(handle
) &&
5661 EXT4_I(inode
)->i_extra_isize
< sbi
->s_want_extra_isize
&&
5662 !ext4_test_inode_state(inode
, EXT4_STATE_NO_EXPAND
)) {
5664 * We need extra buffer credits since we may write into EA block
5665 * with this same handle. If journal_extend fails, then it will
5666 * only result in a minor loss of functionality for that inode.
5667 * If this is felt to be critical, then e2fsck should be run to
5668 * force a large enough s_min_extra_isize.
5670 if ((jbd2_journal_extend(handle
,
5671 EXT4_DATA_TRANS_BLOCKS(inode
->i_sb
))) == 0) {
5672 ret
= ext4_expand_extra_isize(inode
,
5673 sbi
->s_want_extra_isize
,
5676 ext4_set_inode_state(inode
,
5677 EXT4_STATE_NO_EXPAND
);
5679 le16_to_cpu(sbi
->s_es
->s_mnt_count
)) {
5680 ext4_warning(inode
->i_sb
, __func__
,
5681 "Unable to expand inode %lu. Delete"
5682 " some EAs or run e2fsck.",
5685 le16_to_cpu(sbi
->s_es
->s_mnt_count
);
5691 err
= ext4_mark_iloc_dirty(handle
, inode
, &iloc
);
5696 * ext4_dirty_inode() is called from __mark_inode_dirty()
5698 * We're really interested in the case where a file is being extended.
5699 * i_size has been changed by generic_commit_write() and we thus need
5700 * to include the updated inode in the current transaction.
5702 * Also, vfs_dq_alloc_block() will always dirty the inode when blocks
5703 * are allocated to the file.
5705 * If the inode is marked synchronous, we don't honour that here - doing
5706 * so would cause a commit on atime updates, which we don't bother doing.
5707 * We handle synchronous inodes at the highest possible level.
5709 void ext4_dirty_inode(struct inode
*inode
)
5713 handle
= ext4_journal_start(inode
, 2);
5717 ext4_mark_inode_dirty(handle
, inode
);
5719 ext4_journal_stop(handle
);
5726 * Bind an inode's backing buffer_head into this transaction, to prevent
5727 * it from being flushed to disk early. Unlike
5728 * ext4_reserve_inode_write, this leaves behind no bh reference and
5729 * returns no iloc structure, so the caller needs to repeat the iloc
5730 * lookup to mark the inode dirty later.
5732 static int ext4_pin_inode(handle_t
*handle
, struct inode
*inode
)
5734 struct ext4_iloc iloc
;
5738 err
= ext4_get_inode_loc(inode
, &iloc
);
5740 BUFFER_TRACE(iloc
.bh
, "get_write_access");
5741 err
= jbd2_journal_get_write_access(handle
, iloc
.bh
);
5743 err
= ext4_handle_dirty_metadata(handle
,
5749 ext4_std_error(inode
->i_sb
, err
);
5754 int ext4_change_inode_journal_flag(struct inode
*inode
, int val
)
5761 * We have to be very careful here: changing a data block's
5762 * journaling status dynamically is dangerous. If we write a
5763 * data block to the journal, change the status and then delete
5764 * that block, we risk forgetting to revoke the old log record
5765 * from the journal and so a subsequent replay can corrupt data.
5766 * So, first we make sure that the journal is empty and that
5767 * nobody is changing anything.
5770 journal
= EXT4_JOURNAL(inode
);
5773 if (is_journal_aborted(journal
))
5776 jbd2_journal_lock_updates(journal
);
5777 jbd2_journal_flush(journal
);
5780 * OK, there are no updates running now, and all cached data is
5781 * synced to disk. We are now in a completely consistent state
5782 * which doesn't have anything in the journal, and we know that
5783 * no filesystem updates are running, so it is safe to modify
5784 * the inode's in-core data-journaling state flag now.
5788 EXT4_I(inode
)->i_flags
|= EXT4_JOURNAL_DATA_FL
;
5790 EXT4_I(inode
)->i_flags
&= ~EXT4_JOURNAL_DATA_FL
;
5791 ext4_set_aops(inode
);
5793 jbd2_journal_unlock_updates(journal
);
5795 /* Finally we can mark the inode as dirty. */
5797 handle
= ext4_journal_start(inode
, 1);
5799 return PTR_ERR(handle
);
5801 err
= ext4_mark_inode_dirty(handle
, inode
);
5802 ext4_handle_sync(handle
);
5803 ext4_journal_stop(handle
);
5804 ext4_std_error(inode
->i_sb
, err
);
5809 static int ext4_bh_unmapped(handle_t
*handle
, struct buffer_head
*bh
)
5811 return !buffer_mapped(bh
);
5814 int ext4_page_mkwrite(struct vm_area_struct
*vma
, struct vm_fault
*vmf
)
5816 struct page
*page
= vmf
->page
;
5821 struct file
*file
= vma
->vm_file
;
5822 struct inode
*inode
= file
->f_path
.dentry
->d_inode
;
5823 struct address_space
*mapping
= inode
->i_mapping
;
5826 * Get i_alloc_sem to stop truncates messing with the inode. We cannot
5827 * get i_mutex because we are already holding mmap_sem.
5829 down_read(&inode
->i_alloc_sem
);
5830 size
= i_size_read(inode
);
5831 if (page
->mapping
!= mapping
|| size
<= page_offset(page
)
5832 || !PageUptodate(page
)) {
5833 /* page got truncated from under us? */
5837 if (PageMappedToDisk(page
))
5840 if (page
->index
== size
>> PAGE_CACHE_SHIFT
)
5841 len
= size
& ~PAGE_CACHE_MASK
;
5843 len
= PAGE_CACHE_SIZE
;
5847 * return if we have all the buffers mapped. This avoid
5848 * the need to call write_begin/write_end which does a
5849 * journal_start/journal_stop which can block and take
5852 if (page_has_buffers(page
)) {
5853 if (!walk_page_buffers(NULL
, page_buffers(page
), 0, len
, NULL
,
5854 ext4_bh_unmapped
)) {
5861 * OK, we need to fill the hole... Do write_begin write_end
5862 * to do block allocation/reservation.We are not holding
5863 * inode.i__mutex here. That allow * parallel write_begin,
5864 * write_end call. lock_page prevent this from happening
5865 * on the same page though
5867 ret
= mapping
->a_ops
->write_begin(file
, mapping
, page_offset(page
),
5868 len
, AOP_FLAG_UNINTERRUPTIBLE
, &page
, &fsdata
);
5871 ret
= mapping
->a_ops
->write_end(file
, mapping
, page_offset(page
),
5872 len
, len
, page
, fsdata
);
5878 ret
= VM_FAULT_SIGBUS
;
5879 up_read(&inode
->i_alloc_sem
);