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>
41 #include <linux/kernel.h>
42 #include <linux/printk.h>
43 #include <linux/slab.h>
44 #include <linux/ratelimit.h>
46 #include "ext4_jbd2.h"
49 #include "ext4_extents.h"
51 #include <trace/events/ext4.h>
53 #define MPAGE_DA_EXTENT_TAIL 0x01
55 static inline int ext4_begin_ordered_truncate(struct inode
*inode
,
58 trace_ext4_begin_ordered_truncate(inode
, new_size
);
60 * If jinode is zero, then we never opened the file for
61 * writing, so there's no need to call
62 * jbd2_journal_begin_ordered_truncate() since there's no
63 * outstanding writes we need to flush.
65 if (!EXT4_I(inode
)->jinode
)
67 return jbd2_journal_begin_ordered_truncate(EXT4_JOURNAL(inode
),
68 EXT4_I(inode
)->jinode
,
72 static void ext4_invalidatepage(struct page
*page
, unsigned long offset
);
73 static int noalloc_get_block_write(struct inode
*inode
, sector_t iblock
,
74 struct buffer_head
*bh_result
, int create
);
75 static int ext4_set_bh_endio(struct buffer_head
*bh
, struct inode
*inode
);
76 static void ext4_end_io_buffer_write(struct buffer_head
*bh
, int uptodate
);
77 static int __ext4_journalled_writepage(struct page
*page
, unsigned int len
);
78 static int ext4_bh_delay_or_unwritten(handle_t
*handle
, struct buffer_head
*bh
);
81 * Test whether an inode is a fast symlink.
83 static int ext4_inode_is_fast_symlink(struct inode
*inode
)
85 int ea_blocks
= EXT4_I(inode
)->i_file_acl
?
86 (inode
->i_sb
->s_blocksize
>> 9) : 0;
88 return (S_ISLNK(inode
->i_mode
) && inode
->i_blocks
- ea_blocks
== 0);
92 * Work out how many blocks we need to proceed with the next chunk of a
93 * truncate transaction.
95 static unsigned long blocks_for_truncate(struct inode
*inode
)
99 needed
= inode
->i_blocks
>> (inode
->i_sb
->s_blocksize_bits
- 9);
101 /* Give ourselves just enough room to cope with inodes in which
102 * i_blocks is corrupt: we've seen disk corruptions in the past
103 * which resulted in random data in an inode which looked enough
104 * like a regular file for ext4 to try to delete it. Things
105 * will go a bit crazy if that happens, but at least we should
106 * try not to panic the whole kernel. */
110 /* But we need to bound the transaction so we don't overflow the
112 if (needed
> EXT4_MAX_TRANS_DATA
)
113 needed
= EXT4_MAX_TRANS_DATA
;
115 return EXT4_DATA_TRANS_BLOCKS(inode
->i_sb
) + needed
;
119 * Truncate transactions can be complex and absolutely huge. So we need to
120 * be able to restart the transaction at a conventient checkpoint to make
121 * sure we don't overflow the journal.
123 * start_transaction gets us a new handle for a truncate transaction,
124 * and extend_transaction tries to extend the existing one a bit. If
125 * extend fails, we need to propagate the failure up and restart the
126 * transaction in the top-level truncate loop. --sct
128 static handle_t
*start_transaction(struct inode
*inode
)
132 result
= ext4_journal_start(inode
, blocks_for_truncate(inode
));
136 ext4_std_error(inode
->i_sb
, PTR_ERR(result
));
141 * Try to extend this transaction for the purposes of truncation.
143 * Returns 0 if we managed to create more room. If we can't create more
144 * room, and the transaction must be restarted we return 1.
146 static int try_to_extend_transaction(handle_t
*handle
, struct inode
*inode
)
148 if (!ext4_handle_valid(handle
))
150 if (ext4_handle_has_enough_credits(handle
, EXT4_RESERVE_TRANS_BLOCKS
+1))
152 if (!ext4_journal_extend(handle
, blocks_for_truncate(inode
)))
158 * Restart the transaction associated with *handle. This does a commit,
159 * so before we call here everything must be consistently dirtied against
162 int ext4_truncate_restart_trans(handle_t
*handle
, struct inode
*inode
,
168 * Drop i_data_sem to avoid deadlock with ext4_map_blocks. At this
169 * moment, get_block can be called only for blocks inside i_size since
170 * page cache has been already dropped and writes are blocked by
171 * i_mutex. So we can safely drop the i_data_sem here.
173 BUG_ON(EXT4_JOURNAL(inode
) == NULL
);
174 jbd_debug(2, "restarting handle %p\n", handle
);
175 up_write(&EXT4_I(inode
)->i_data_sem
);
176 ret
= ext4_journal_restart(handle
, blocks_for_truncate(inode
));
177 down_write(&EXT4_I(inode
)->i_data_sem
);
178 ext4_discard_preallocations(inode
);
184 * Called at the last iput() if i_nlink is zero.
186 void ext4_evict_inode(struct inode
*inode
)
191 trace_ext4_evict_inode(inode
);
192 if (inode
->i_nlink
) {
193 truncate_inode_pages(&inode
->i_data
, 0);
197 if (!is_bad_inode(inode
))
198 dquot_initialize(inode
);
200 if (ext4_should_order_data(inode
))
201 ext4_begin_ordered_truncate(inode
, 0);
202 truncate_inode_pages(&inode
->i_data
, 0);
204 if (is_bad_inode(inode
))
207 handle
= ext4_journal_start(inode
, blocks_for_truncate(inode
)+3);
208 if (IS_ERR(handle
)) {
209 ext4_std_error(inode
->i_sb
, PTR_ERR(handle
));
211 * If we're going to skip the normal cleanup, we still need to
212 * make sure that the in-core orphan linked list is properly
215 ext4_orphan_del(NULL
, inode
);
220 ext4_handle_sync(handle
);
222 err
= ext4_mark_inode_dirty(handle
, inode
);
224 ext4_warning(inode
->i_sb
,
225 "couldn't mark inode dirty (err %d)", err
);
229 ext4_truncate(inode
);
232 * ext4_ext_truncate() doesn't reserve any slop when it
233 * restarts journal transactions; therefore there may not be
234 * enough credits left in the handle to remove the inode from
235 * the orphan list and set the dtime field.
237 if (!ext4_handle_has_enough_credits(handle
, 3)) {
238 err
= ext4_journal_extend(handle
, 3);
240 err
= ext4_journal_restart(handle
, 3);
242 ext4_warning(inode
->i_sb
,
243 "couldn't extend journal (err %d)", err
);
245 ext4_journal_stop(handle
);
246 ext4_orphan_del(NULL
, inode
);
252 * Kill off the orphan record which ext4_truncate created.
253 * AKPM: I think this can be inside the above `if'.
254 * Note that ext4_orphan_del() has to be able to cope with the
255 * deletion of a non-existent orphan - this is because we don't
256 * know if ext4_truncate() actually created an orphan record.
257 * (Well, we could do this if we need to, but heck - it works)
259 ext4_orphan_del(handle
, inode
);
260 EXT4_I(inode
)->i_dtime
= get_seconds();
263 * One subtle ordering requirement: if anything has gone wrong
264 * (transaction abort, IO errors, whatever), then we can still
265 * do these next steps (the fs will already have been marked as
266 * having errors), but we can't free the inode if the mark_dirty
269 if (ext4_mark_inode_dirty(handle
, inode
))
270 /* If that failed, just do the required in-core inode clear. */
271 ext4_clear_inode(inode
);
273 ext4_free_inode(handle
, inode
);
274 ext4_journal_stop(handle
);
277 ext4_clear_inode(inode
); /* We must guarantee clearing of inode... */
283 struct buffer_head
*bh
;
286 static inline void add_chain(Indirect
*p
, struct buffer_head
*bh
, __le32
*v
)
288 p
->key
= *(p
->p
= v
);
293 * ext4_block_to_path - parse the block number into array of offsets
294 * @inode: inode in question (we are only interested in its superblock)
295 * @i_block: block number to be parsed
296 * @offsets: array to store the offsets in
297 * @boundary: set this non-zero if the referred-to block is likely to be
298 * followed (on disk) by an indirect block.
300 * To store the locations of file's data ext4 uses a data structure common
301 * for UNIX filesystems - tree of pointers anchored in the inode, with
302 * data blocks at leaves and indirect blocks in intermediate nodes.
303 * This function translates the block number into path in that tree -
304 * return value is the path length and @offsets[n] is the offset of
305 * pointer to (n+1)th node in the nth one. If @block is out of range
306 * (negative or too large) warning is printed and zero returned.
308 * Note: function doesn't find node addresses, so no IO is needed. All
309 * we need to know is the capacity of indirect blocks (taken from the
314 * Portability note: the last comparison (check that we fit into triple
315 * indirect block) is spelled differently, because otherwise on an
316 * architecture with 32-bit longs and 8Kb pages we might get into trouble
317 * if our filesystem had 8Kb blocks. We might use long long, but that would
318 * kill us on x86. Oh, well, at least the sign propagation does not matter -
319 * i_block would have to be negative in the very beginning, so we would not
323 static int ext4_block_to_path(struct inode
*inode
,
325 ext4_lblk_t offsets
[4], int *boundary
)
327 int ptrs
= EXT4_ADDR_PER_BLOCK(inode
->i_sb
);
328 int ptrs_bits
= EXT4_ADDR_PER_BLOCK_BITS(inode
->i_sb
);
329 const long direct_blocks
= EXT4_NDIR_BLOCKS
,
330 indirect_blocks
= ptrs
,
331 double_blocks
= (1 << (ptrs_bits
* 2));
335 if (i_block
< direct_blocks
) {
336 offsets
[n
++] = i_block
;
337 final
= direct_blocks
;
338 } else if ((i_block
-= direct_blocks
) < indirect_blocks
) {
339 offsets
[n
++] = EXT4_IND_BLOCK
;
340 offsets
[n
++] = i_block
;
342 } else if ((i_block
-= indirect_blocks
) < double_blocks
) {
343 offsets
[n
++] = EXT4_DIND_BLOCK
;
344 offsets
[n
++] = i_block
>> ptrs_bits
;
345 offsets
[n
++] = i_block
& (ptrs
- 1);
347 } else if (((i_block
-= double_blocks
) >> (ptrs_bits
* 2)) < ptrs
) {
348 offsets
[n
++] = EXT4_TIND_BLOCK
;
349 offsets
[n
++] = i_block
>> (ptrs_bits
* 2);
350 offsets
[n
++] = (i_block
>> ptrs_bits
) & (ptrs
- 1);
351 offsets
[n
++] = i_block
& (ptrs
- 1);
354 ext4_warning(inode
->i_sb
, "block %lu > max in inode %lu",
355 i_block
+ direct_blocks
+
356 indirect_blocks
+ double_blocks
, inode
->i_ino
);
359 *boundary
= final
- 1 - (i_block
& (ptrs
- 1));
363 static int __ext4_check_blockref(const char *function
, unsigned int line
,
365 __le32
*p
, unsigned int max
)
367 struct ext4_super_block
*es
= EXT4_SB(inode
->i_sb
)->s_es
;
371 while (bref
< p
+max
) {
372 blk
= le32_to_cpu(*bref
++);
374 unlikely(!ext4_data_block_valid(EXT4_SB(inode
->i_sb
),
376 es
->s_last_error_block
= cpu_to_le64(blk
);
377 ext4_error_inode(inode
, function
, line
, blk
,
386 #define ext4_check_indirect_blockref(inode, bh) \
387 __ext4_check_blockref(__func__, __LINE__, inode, \
388 (__le32 *)(bh)->b_data, \
389 EXT4_ADDR_PER_BLOCK((inode)->i_sb))
391 #define ext4_check_inode_blockref(inode) \
392 __ext4_check_blockref(__func__, __LINE__, inode, \
393 EXT4_I(inode)->i_data, \
397 * ext4_get_branch - read the chain of indirect blocks leading to data
398 * @inode: inode in question
399 * @depth: depth of the chain (1 - direct pointer, etc.)
400 * @offsets: offsets of pointers in inode/indirect blocks
401 * @chain: place to store the result
402 * @err: here we store the error value
404 * Function fills the array of triples <key, p, bh> and returns %NULL
405 * if everything went OK or the pointer to the last filled triple
406 * (incomplete one) otherwise. Upon the return chain[i].key contains
407 * the number of (i+1)-th block in the chain (as it is stored in memory,
408 * i.e. little-endian 32-bit), chain[i].p contains the address of that
409 * number (it points into struct inode for i==0 and into the bh->b_data
410 * for i>0) and chain[i].bh points to the buffer_head of i-th indirect
411 * block for i>0 and NULL for i==0. In other words, it holds the block
412 * numbers of the chain, addresses they were taken from (and where we can
413 * verify that chain did not change) and buffer_heads hosting these
416 * Function stops when it stumbles upon zero pointer (absent block)
417 * (pointer to last triple returned, *@err == 0)
418 * or when it gets an IO error reading an indirect block
419 * (ditto, *@err == -EIO)
420 * or when it reads all @depth-1 indirect blocks successfully and finds
421 * the whole chain, all way to the data (returns %NULL, *err == 0).
423 * Need to be called with
424 * down_read(&EXT4_I(inode)->i_data_sem)
426 static Indirect
*ext4_get_branch(struct inode
*inode
, int depth
,
427 ext4_lblk_t
*offsets
,
428 Indirect chain
[4], int *err
)
430 struct super_block
*sb
= inode
->i_sb
;
432 struct buffer_head
*bh
;
435 /* i_data is not going away, no lock needed */
436 add_chain(chain
, NULL
, EXT4_I(inode
)->i_data
+ *offsets
);
440 bh
= sb_getblk(sb
, le32_to_cpu(p
->key
));
444 if (!bh_uptodate_or_lock(bh
)) {
445 if (bh_submit_read(bh
) < 0) {
449 /* validate block references */
450 if (ext4_check_indirect_blockref(inode
, bh
)) {
456 add_chain(++p
, bh
, (__le32
*)bh
->b_data
+ *++offsets
);
470 * ext4_find_near - find a place for allocation with sufficient locality
472 * @ind: descriptor of indirect block.
474 * This function returns the preferred place for block allocation.
475 * It is used when heuristic for sequential allocation fails.
477 * + if there is a block to the left of our position - allocate near it.
478 * + if pointer will live in indirect block - allocate near that block.
479 * + if pointer will live in inode - allocate in the same
482 * In the latter case we colour the starting block by the callers PID to
483 * prevent it from clashing with concurrent allocations for a different inode
484 * in the same block group. The PID is used here so that functionally related
485 * files will be close-by on-disk.
487 * Caller must make sure that @ind is valid and will stay that way.
489 static ext4_fsblk_t
ext4_find_near(struct inode
*inode
, Indirect
*ind
)
491 struct ext4_inode_info
*ei
= EXT4_I(inode
);
492 __le32
*start
= ind
->bh
? (__le32
*) ind
->bh
->b_data
: ei
->i_data
;
494 ext4_fsblk_t bg_start
;
495 ext4_fsblk_t last_block
;
496 ext4_grpblk_t colour
;
497 ext4_group_t block_group
;
498 int flex_size
= ext4_flex_bg_size(EXT4_SB(inode
->i_sb
));
500 /* Try to find previous block */
501 for (p
= ind
->p
- 1; p
>= start
; p
--) {
503 return le32_to_cpu(*p
);
506 /* No such thing, so let's try location of indirect block */
508 return ind
->bh
->b_blocknr
;
511 * It is going to be referred to from the inode itself? OK, just put it
512 * into the same cylinder group then.
514 block_group
= ei
->i_block_group
;
515 if (flex_size
>= EXT4_FLEX_SIZE_DIR_ALLOC_SCHEME
) {
516 block_group
&= ~(flex_size
-1);
517 if (S_ISREG(inode
->i_mode
))
520 bg_start
= ext4_group_first_block_no(inode
->i_sb
, block_group
);
521 last_block
= ext4_blocks_count(EXT4_SB(inode
->i_sb
)->s_es
) - 1;
524 * If we are doing delayed allocation, we don't need take
525 * colour into account.
527 if (test_opt(inode
->i_sb
, DELALLOC
))
530 if (bg_start
+ EXT4_BLOCKS_PER_GROUP(inode
->i_sb
) <= last_block
)
531 colour
= (current
->pid
% 16) *
532 (EXT4_BLOCKS_PER_GROUP(inode
->i_sb
) / 16);
534 colour
= (current
->pid
% 16) * ((last_block
- bg_start
) / 16);
535 return bg_start
+ colour
;
539 * ext4_find_goal - find a preferred place for allocation.
541 * @block: block we want
542 * @partial: pointer to the last triple within a chain
544 * Normally this function find the preferred place for block allocation,
546 * Because this is only used for non-extent files, we limit the block nr
549 static ext4_fsblk_t
ext4_find_goal(struct inode
*inode
, ext4_lblk_t block
,
555 * XXX need to get goal block from mballoc's data structures
558 goal
= ext4_find_near(inode
, partial
);
559 goal
= goal
& EXT4_MAX_BLOCK_FILE_PHYS
;
564 * ext4_blks_to_allocate - Look up the block map and count the number
565 * of direct blocks need to be allocated for the given branch.
567 * @branch: chain of indirect blocks
568 * @k: number of blocks need for indirect blocks
569 * @blks: number of data blocks to be mapped.
570 * @blocks_to_boundary: the offset in the indirect block
572 * return the total number of blocks to be allocate, including the
573 * direct and indirect blocks.
575 static int ext4_blks_to_allocate(Indirect
*branch
, int k
, unsigned int blks
,
576 int blocks_to_boundary
)
578 unsigned int count
= 0;
581 * Simple case, [t,d]Indirect block(s) has not allocated yet
582 * then it's clear blocks on that path have not allocated
585 /* right now we don't handle cross boundary allocation */
586 if (blks
< blocks_to_boundary
+ 1)
589 count
+= blocks_to_boundary
+ 1;
594 while (count
< blks
&& count
<= blocks_to_boundary
&&
595 le32_to_cpu(*(branch
[0].p
+ count
)) == 0) {
602 * ext4_alloc_blocks: multiple allocate blocks needed for a branch
603 * @handle: handle for this transaction
604 * @inode: inode which needs allocated blocks
605 * @iblock: the logical block to start allocated at
606 * @goal: preferred physical block of allocation
607 * @indirect_blks: the number of blocks need to allocate for indirect
609 * @blks: number of desired blocks
610 * @new_blocks: on return it will store the new block numbers for
611 * the indirect blocks(if needed) and the first direct block,
612 * @err: on return it will store the error code
614 * This function will return the number of blocks allocated as
615 * requested by the passed-in parameters.
617 static int ext4_alloc_blocks(handle_t
*handle
, struct inode
*inode
,
618 ext4_lblk_t iblock
, ext4_fsblk_t goal
,
619 int indirect_blks
, int blks
,
620 ext4_fsblk_t new_blocks
[4], int *err
)
622 struct ext4_allocation_request ar
;
624 unsigned long count
= 0, blk_allocated
= 0;
626 ext4_fsblk_t current_block
= 0;
630 * Here we try to allocate the requested multiple blocks at once,
631 * on a best-effort basis.
632 * To build a branch, we should allocate blocks for
633 * the indirect blocks(if not allocated yet), and at least
634 * the first direct block of this branch. That's the
635 * minimum number of blocks need to allocate(required)
637 /* first we try to allocate the indirect blocks */
638 target
= indirect_blks
;
641 /* allocating blocks for indirect blocks and direct blocks */
642 current_block
= ext4_new_meta_blocks(handle
, inode
,
647 if (unlikely(current_block
+ count
> EXT4_MAX_BLOCK_FILE_PHYS
)) {
648 EXT4_ERROR_INODE(inode
,
649 "current_block %llu + count %lu > %d!",
650 current_block
, count
,
651 EXT4_MAX_BLOCK_FILE_PHYS
);
657 /* allocate blocks for indirect blocks */
658 while (index
< indirect_blks
&& count
) {
659 new_blocks
[index
++] = current_block
++;
664 * save the new block number
665 * for the first direct block
667 new_blocks
[index
] = current_block
;
668 printk(KERN_INFO
"%s returned more blocks than "
669 "requested\n", __func__
);
675 target
= blks
- count
;
676 blk_allocated
= count
;
679 /* Now allocate data blocks */
680 memset(&ar
, 0, sizeof(ar
));
685 if (S_ISREG(inode
->i_mode
))
686 /* enable in-core preallocation only for regular files */
687 ar
.flags
= EXT4_MB_HINT_DATA
;
689 current_block
= ext4_mb_new_blocks(handle
, &ar
, err
);
690 if (unlikely(current_block
+ ar
.len
> EXT4_MAX_BLOCK_FILE_PHYS
)) {
691 EXT4_ERROR_INODE(inode
,
692 "current_block %llu + ar.len %d > %d!",
693 current_block
, ar
.len
,
694 EXT4_MAX_BLOCK_FILE_PHYS
);
699 if (*err
&& (target
== blks
)) {
701 * if the allocation failed and we didn't allocate
707 if (target
== blks
) {
709 * save the new block number
710 * for the first direct block
712 new_blocks
[index
] = current_block
;
714 blk_allocated
+= ar
.len
;
717 /* total number of blocks allocated for direct blocks */
722 for (i
= 0; i
< index
; i
++)
723 ext4_free_blocks(handle
, inode
, NULL
, new_blocks
[i
], 1, 0);
728 * ext4_alloc_branch - allocate and set up a chain of blocks.
729 * @handle: handle for this transaction
731 * @indirect_blks: number of allocated indirect blocks
732 * @blks: number of allocated direct blocks
733 * @goal: preferred place for allocation
734 * @offsets: offsets (in the blocks) to store the pointers to next.
735 * @branch: place to store the chain in.
737 * This function allocates blocks, zeroes out all but the last one,
738 * links them into chain and (if we are synchronous) writes them to disk.
739 * In other words, it prepares a branch that can be spliced onto the
740 * inode. It stores the information about that chain in the branch[], in
741 * the same format as ext4_get_branch() would do. We are calling it after
742 * we had read the existing part of chain and partial points to the last
743 * triple of that (one with zero ->key). Upon the exit we have the same
744 * picture as after the successful ext4_get_block(), except that in one
745 * place chain is disconnected - *branch->p is still zero (we did not
746 * set the last link), but branch->key contains the number that should
747 * be placed into *branch->p to fill that gap.
749 * If allocation fails we free all blocks we've allocated (and forget
750 * their buffer_heads) and return the error value the from failed
751 * ext4_alloc_block() (normally -ENOSPC). Otherwise we set the chain
752 * as described above and return 0.
754 static int ext4_alloc_branch(handle_t
*handle
, struct inode
*inode
,
755 ext4_lblk_t iblock
, int indirect_blks
,
756 int *blks
, ext4_fsblk_t goal
,
757 ext4_lblk_t
*offsets
, Indirect
*branch
)
759 int blocksize
= inode
->i_sb
->s_blocksize
;
762 struct buffer_head
*bh
;
764 ext4_fsblk_t new_blocks
[4];
765 ext4_fsblk_t current_block
;
767 num
= ext4_alloc_blocks(handle
, inode
, iblock
, goal
, indirect_blks
,
768 *blks
, new_blocks
, &err
);
772 branch
[0].key
= cpu_to_le32(new_blocks
[0]);
774 * metadata blocks and data blocks are allocated.
776 for (n
= 1; n
<= indirect_blks
; n
++) {
778 * Get buffer_head for parent block, zero it out
779 * and set the pointer to new one, then send
782 bh
= sb_getblk(inode
->i_sb
, new_blocks
[n
-1]);
790 BUFFER_TRACE(bh
, "call get_create_access");
791 err
= ext4_journal_get_create_access(handle
, bh
);
793 /* Don't brelse(bh) here; it's done in
794 * ext4_journal_forget() below */
799 memset(bh
->b_data
, 0, blocksize
);
800 branch
[n
].p
= (__le32
*) bh
->b_data
+ offsets
[n
];
801 branch
[n
].key
= cpu_to_le32(new_blocks
[n
]);
802 *branch
[n
].p
= branch
[n
].key
;
803 if (n
== indirect_blks
) {
804 current_block
= new_blocks
[n
];
806 * End of chain, update the last new metablock of
807 * the chain to point to the new allocated
808 * data blocks numbers
810 for (i
= 1; i
< num
; i
++)
811 *(branch
[n
].p
+ i
) = cpu_to_le32(++current_block
);
813 BUFFER_TRACE(bh
, "marking uptodate");
814 set_buffer_uptodate(bh
);
817 BUFFER_TRACE(bh
, "call ext4_handle_dirty_metadata");
818 err
= ext4_handle_dirty_metadata(handle
, inode
, bh
);
825 /* Allocation failed, free what we already allocated */
826 ext4_free_blocks(handle
, inode
, NULL
, new_blocks
[0], 1, 0);
827 for (i
= 1; i
<= n
; i
++) {
829 * branch[i].bh is newly allocated, so there is no
830 * need to revoke the block, which is why we don't
831 * need to set EXT4_FREE_BLOCKS_METADATA.
833 ext4_free_blocks(handle
, inode
, NULL
, new_blocks
[i
], 1,
834 EXT4_FREE_BLOCKS_FORGET
);
836 for (i
= n
+1; i
< indirect_blks
; i
++)
837 ext4_free_blocks(handle
, inode
, NULL
, new_blocks
[i
], 1, 0);
839 ext4_free_blocks(handle
, inode
, NULL
, new_blocks
[i
], num
, 0);
845 * ext4_splice_branch - splice the allocated branch onto inode.
846 * @handle: handle for this transaction
848 * @block: (logical) number of block we are adding
849 * @chain: chain of indirect blocks (with a missing link - see
851 * @where: location of missing link
852 * @num: number of indirect blocks we are adding
853 * @blks: number of direct blocks we are adding
855 * This function fills the missing link and does all housekeeping needed in
856 * inode (->i_blocks, etc.). In case of success we end up with the full
857 * chain to new block and return 0.
859 static int ext4_splice_branch(handle_t
*handle
, struct inode
*inode
,
860 ext4_lblk_t block
, Indirect
*where
, int num
,
865 ext4_fsblk_t current_block
;
868 * If we're splicing into a [td]indirect block (as opposed to the
869 * inode) then we need to get write access to the [td]indirect block
873 BUFFER_TRACE(where
->bh
, "get_write_access");
874 err
= ext4_journal_get_write_access(handle
, where
->bh
);
880 *where
->p
= where
->key
;
883 * Update the host buffer_head or inode to point to more just allocated
884 * direct blocks blocks
886 if (num
== 0 && blks
> 1) {
887 current_block
= le32_to_cpu(where
->key
) + 1;
888 for (i
= 1; i
< blks
; i
++)
889 *(where
->p
+ i
) = cpu_to_le32(current_block
++);
892 /* We are done with atomic stuff, now do the rest of housekeeping */
893 /* had we spliced it onto indirect block? */
896 * If we spliced it onto an indirect block, we haven't
897 * altered the inode. Note however that if it is being spliced
898 * onto an indirect block at the very end of the file (the
899 * file is growing) then we *will* alter the inode to reflect
900 * the new i_size. But that is not done here - it is done in
901 * generic_commit_write->__mark_inode_dirty->ext4_dirty_inode.
903 jbd_debug(5, "splicing indirect only\n");
904 BUFFER_TRACE(where
->bh
, "call ext4_handle_dirty_metadata");
905 err
= ext4_handle_dirty_metadata(handle
, inode
, where
->bh
);
910 * OK, we spliced it into the inode itself on a direct block.
912 ext4_mark_inode_dirty(handle
, inode
);
913 jbd_debug(5, "splicing direct\n");
918 for (i
= 1; i
<= num
; i
++) {
920 * branch[i].bh is newly allocated, so there is no
921 * need to revoke the block, which is why we don't
922 * need to set EXT4_FREE_BLOCKS_METADATA.
924 ext4_free_blocks(handle
, inode
, where
[i
].bh
, 0, 1,
925 EXT4_FREE_BLOCKS_FORGET
);
927 ext4_free_blocks(handle
, inode
, NULL
, le32_to_cpu(where
[num
].key
),
934 * The ext4_ind_map_blocks() function handles non-extents inodes
935 * (i.e., using the traditional indirect/double-indirect i_blocks
936 * scheme) for ext4_map_blocks().
938 * Allocation strategy is simple: if we have to allocate something, we will
939 * have to go the whole way to leaf. So let's do it before attaching anything
940 * to tree, set linkage between the newborn blocks, write them if sync is
941 * required, recheck the path, free and repeat if check fails, otherwise
942 * set the last missing link (that will protect us from any truncate-generated
943 * removals - all blocks on the path are immune now) and possibly force the
944 * write on the parent block.
945 * That has a nice additional property: no special recovery from the failed
946 * allocations is needed - we simply release blocks and do not touch anything
947 * reachable from inode.
949 * `handle' can be NULL if create == 0.
951 * return > 0, # of blocks mapped or allocated.
952 * return = 0, if plain lookup failed.
953 * return < 0, error case.
955 * The ext4_ind_get_blocks() function should be called with
956 * down_write(&EXT4_I(inode)->i_data_sem) if allocating filesystem
957 * blocks (i.e., flags has EXT4_GET_BLOCKS_CREATE set) or
958 * down_read(&EXT4_I(inode)->i_data_sem) if not allocating file system
961 static int ext4_ind_map_blocks(handle_t
*handle
, struct inode
*inode
,
962 struct ext4_map_blocks
*map
,
966 ext4_lblk_t offsets
[4];
971 int blocks_to_boundary
= 0;
974 ext4_fsblk_t first_block
= 0;
976 J_ASSERT(!(ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
)));
977 J_ASSERT(handle
!= NULL
|| (flags
& EXT4_GET_BLOCKS_CREATE
) == 0);
978 depth
= ext4_block_to_path(inode
, map
->m_lblk
, offsets
,
979 &blocks_to_boundary
);
984 partial
= ext4_get_branch(inode
, depth
, offsets
, chain
, &err
);
986 /* Simplest case - block found, no allocation needed */
988 first_block
= le32_to_cpu(chain
[depth
- 1].key
);
991 while (count
< map
->m_len
&& count
<= blocks_to_boundary
) {
994 blk
= le32_to_cpu(*(chain
[depth
-1].p
+ count
));
996 if (blk
== first_block
+ count
)
1004 /* Next simple case - plain lookup or failed read of indirect block */
1005 if ((flags
& EXT4_GET_BLOCKS_CREATE
) == 0 || err
== -EIO
)
1009 * Okay, we need to do block allocation.
1011 goal
= ext4_find_goal(inode
, map
->m_lblk
, partial
);
1013 /* the number of blocks need to allocate for [d,t]indirect blocks */
1014 indirect_blks
= (chain
+ depth
) - partial
- 1;
1017 * Next look up the indirect map to count the totoal number of
1018 * direct blocks to allocate for this branch.
1020 count
= ext4_blks_to_allocate(partial
, indirect_blks
,
1021 map
->m_len
, blocks_to_boundary
);
1023 * Block out ext4_truncate while we alter the tree
1025 err
= ext4_alloc_branch(handle
, inode
, map
->m_lblk
, indirect_blks
,
1027 offsets
+ (partial
- chain
), partial
);
1030 * The ext4_splice_branch call will free and forget any buffers
1031 * on the new chain if there is a failure, but that risks using
1032 * up transaction credits, especially for bitmaps where the
1033 * credits cannot be returned. Can we handle this somehow? We
1034 * may need to return -EAGAIN upwards in the worst case. --sct
1037 err
= ext4_splice_branch(handle
, inode
, map
->m_lblk
,
1038 partial
, indirect_blks
, count
);
1042 map
->m_flags
|= EXT4_MAP_NEW
;
1044 ext4_update_inode_fsync_trans(handle
, inode
, 1);
1046 map
->m_flags
|= EXT4_MAP_MAPPED
;
1047 map
->m_pblk
= le32_to_cpu(chain
[depth
-1].key
);
1049 if (count
> blocks_to_boundary
)
1050 map
->m_flags
|= EXT4_MAP_BOUNDARY
;
1052 /* Clean up and exit */
1053 partial
= chain
+ depth
- 1; /* the whole chain */
1055 while (partial
> chain
) {
1056 BUFFER_TRACE(partial
->bh
, "call brelse");
1057 brelse(partial
->bh
);
1065 qsize_t
*ext4_get_reserved_space(struct inode
*inode
)
1067 return &EXT4_I(inode
)->i_reserved_quota
;
1072 * Calculate the number of metadata blocks need to reserve
1073 * to allocate a new block at @lblocks for non extent file based file
1075 static int ext4_indirect_calc_metadata_amount(struct inode
*inode
,
1078 struct ext4_inode_info
*ei
= EXT4_I(inode
);
1079 sector_t dind_mask
= ~((sector_t
)EXT4_ADDR_PER_BLOCK(inode
->i_sb
) - 1);
1082 if (lblock
< EXT4_NDIR_BLOCKS
)
1085 lblock
-= EXT4_NDIR_BLOCKS
;
1087 if (ei
->i_da_metadata_calc_len
&&
1088 (lblock
& dind_mask
) == ei
->i_da_metadata_calc_last_lblock
) {
1089 ei
->i_da_metadata_calc_len
++;
1092 ei
->i_da_metadata_calc_last_lblock
= lblock
& dind_mask
;
1093 ei
->i_da_metadata_calc_len
= 1;
1094 blk_bits
= order_base_2(lblock
);
1095 return (blk_bits
/ EXT4_ADDR_PER_BLOCK_BITS(inode
->i_sb
)) + 1;
1099 * Calculate the number of metadata blocks need to reserve
1100 * to allocate a block located at @lblock
1102 static int ext4_calc_metadata_amount(struct inode
*inode
, ext4_lblk_t lblock
)
1104 if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
))
1105 return ext4_ext_calc_metadata_amount(inode
, lblock
);
1107 return ext4_indirect_calc_metadata_amount(inode
, lblock
);
1111 * Called with i_data_sem down, which is important since we can call
1112 * ext4_discard_preallocations() from here.
1114 void ext4_da_update_reserve_space(struct inode
*inode
,
1115 int used
, int quota_claim
)
1117 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
1118 struct ext4_inode_info
*ei
= EXT4_I(inode
);
1120 spin_lock(&ei
->i_block_reservation_lock
);
1121 trace_ext4_da_update_reserve_space(inode
, used
);
1122 if (unlikely(used
> ei
->i_reserved_data_blocks
)) {
1123 ext4_msg(inode
->i_sb
, KERN_NOTICE
, "%s: ino %lu, used %d "
1124 "with only %d reserved data blocks\n",
1125 __func__
, inode
->i_ino
, used
,
1126 ei
->i_reserved_data_blocks
);
1128 used
= ei
->i_reserved_data_blocks
;
1131 /* Update per-inode reservations */
1132 ei
->i_reserved_data_blocks
-= used
;
1133 ei
->i_reserved_meta_blocks
-= ei
->i_allocated_meta_blocks
;
1134 percpu_counter_sub(&sbi
->s_dirtyblocks_counter
,
1135 used
+ ei
->i_allocated_meta_blocks
);
1136 ei
->i_allocated_meta_blocks
= 0;
1138 if (ei
->i_reserved_data_blocks
== 0) {
1140 * We can release all of the reserved metadata blocks
1141 * only when we have written all of the delayed
1142 * allocation blocks.
1144 percpu_counter_sub(&sbi
->s_dirtyblocks_counter
,
1145 ei
->i_reserved_meta_blocks
);
1146 ei
->i_reserved_meta_blocks
= 0;
1147 ei
->i_da_metadata_calc_len
= 0;
1149 spin_unlock(&EXT4_I(inode
)->i_block_reservation_lock
);
1151 /* Update quota subsystem for data blocks */
1153 dquot_claim_block(inode
, used
);
1156 * We did fallocate with an offset that is already delayed
1157 * allocated. So on delayed allocated writeback we should
1158 * not re-claim the quota for fallocated blocks.
1160 dquot_release_reservation_block(inode
, used
);
1164 * If we have done all the pending block allocations and if
1165 * there aren't any writers on the inode, we can discard the
1166 * inode's preallocations.
1168 if ((ei
->i_reserved_data_blocks
== 0) &&
1169 (atomic_read(&inode
->i_writecount
) == 0))
1170 ext4_discard_preallocations(inode
);
1173 static int __check_block_validity(struct inode
*inode
, const char *func
,
1175 struct ext4_map_blocks
*map
)
1177 if (!ext4_data_block_valid(EXT4_SB(inode
->i_sb
), map
->m_pblk
,
1179 ext4_error_inode(inode
, func
, line
, map
->m_pblk
,
1180 "lblock %lu mapped to illegal pblock "
1181 "(length %d)", (unsigned long) map
->m_lblk
,
1188 #define check_block_validity(inode, map) \
1189 __check_block_validity((inode), __func__, __LINE__, (map))
1192 * Return the number of contiguous dirty pages in a given inode
1193 * starting at page frame idx.
1195 static pgoff_t
ext4_num_dirty_pages(struct inode
*inode
, pgoff_t idx
,
1196 unsigned int max_pages
)
1198 struct address_space
*mapping
= inode
->i_mapping
;
1200 struct pagevec pvec
;
1202 int i
, nr_pages
, done
= 0;
1206 pagevec_init(&pvec
, 0);
1209 nr_pages
= pagevec_lookup_tag(&pvec
, mapping
, &index
,
1210 PAGECACHE_TAG_DIRTY
,
1211 (pgoff_t
)PAGEVEC_SIZE
);
1214 for (i
= 0; i
< nr_pages
; i
++) {
1215 struct page
*page
= pvec
.pages
[i
];
1216 struct buffer_head
*bh
, *head
;
1219 if (unlikely(page
->mapping
!= mapping
) ||
1221 PageWriteback(page
) ||
1222 page
->index
!= idx
) {
1227 if (page_has_buffers(page
)) {
1228 bh
= head
= page_buffers(page
);
1230 if (!buffer_delay(bh
) &&
1231 !buffer_unwritten(bh
))
1233 bh
= bh
->b_this_page
;
1234 } while (!done
&& (bh
!= head
));
1241 if (num
>= max_pages
) {
1246 pagevec_release(&pvec
);
1252 * The ext4_map_blocks() function tries to look up the requested blocks,
1253 * and returns if the blocks are already mapped.
1255 * Otherwise it takes the write lock of the i_data_sem and allocate blocks
1256 * and store the allocated blocks in the result buffer head and mark it
1259 * If file type is extents based, it will call ext4_ext_map_blocks(),
1260 * Otherwise, call with ext4_ind_map_blocks() to handle indirect mapping
1263 * On success, it returns the number of blocks being mapped or allocate.
1264 * if create==0 and the blocks are pre-allocated and uninitialized block,
1265 * the result buffer head is unmapped. If the create ==1, it will make sure
1266 * the buffer head is mapped.
1268 * It returns 0 if plain look up failed (blocks have not been allocated), in
1269 * that casem, buffer head is unmapped
1271 * It returns the error in case of allocation failure.
1273 int ext4_map_blocks(handle_t
*handle
, struct inode
*inode
,
1274 struct ext4_map_blocks
*map
, int flags
)
1279 ext_debug("ext4_map_blocks(): inode %lu, flag %d, max_blocks %u,"
1280 "logical block %lu\n", inode
->i_ino
, flags
, map
->m_len
,
1281 (unsigned long) map
->m_lblk
);
1283 * Try to see if we can get the block without requesting a new
1284 * file system block.
1286 down_read((&EXT4_I(inode
)->i_data_sem
));
1287 if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
)) {
1288 retval
= ext4_ext_map_blocks(handle
, inode
, map
, 0);
1290 retval
= ext4_ind_map_blocks(handle
, inode
, map
, 0);
1292 up_read((&EXT4_I(inode
)->i_data_sem
));
1294 if (retval
> 0 && map
->m_flags
& EXT4_MAP_MAPPED
) {
1295 int ret
= check_block_validity(inode
, map
);
1300 /* If it is only a block(s) look up */
1301 if ((flags
& EXT4_GET_BLOCKS_CREATE
) == 0)
1305 * Returns if the blocks have already allocated
1307 * Note that if blocks have been preallocated
1308 * ext4_ext_get_block() returns th create = 0
1309 * with buffer head unmapped.
1311 if (retval
> 0 && map
->m_flags
& EXT4_MAP_MAPPED
)
1315 * When we call get_blocks without the create flag, the
1316 * BH_Unwritten flag could have gotten set if the blocks
1317 * requested were part of a uninitialized extent. We need to
1318 * clear this flag now that we are committed to convert all or
1319 * part of the uninitialized extent to be an initialized
1320 * extent. This is because we need to avoid the combination
1321 * of BH_Unwritten and BH_Mapped flags being simultaneously
1322 * set on the buffer_head.
1324 map
->m_flags
&= ~EXT4_MAP_UNWRITTEN
;
1327 * New blocks allocate and/or writing to uninitialized extent
1328 * will possibly result in updating i_data, so we take
1329 * the write lock of i_data_sem, and call get_blocks()
1330 * with create == 1 flag.
1332 down_write((&EXT4_I(inode
)->i_data_sem
));
1335 * if the caller is from delayed allocation writeout path
1336 * we have already reserved fs blocks for allocation
1337 * let the underlying get_block() function know to
1338 * avoid double accounting
1340 if (flags
& EXT4_GET_BLOCKS_DELALLOC_RESERVE
)
1341 ext4_set_inode_state(inode
, EXT4_STATE_DELALLOC_RESERVED
);
1343 * We need to check for EXT4 here because migrate
1344 * could have changed the inode type in between
1346 if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
)) {
1347 retval
= ext4_ext_map_blocks(handle
, inode
, map
, flags
);
1349 retval
= ext4_ind_map_blocks(handle
, inode
, map
, flags
);
1351 if (retval
> 0 && map
->m_flags
& EXT4_MAP_NEW
) {
1353 * We allocated new blocks which will result in
1354 * i_data's format changing. Force the migrate
1355 * to fail by clearing migrate flags
1357 ext4_clear_inode_state(inode
, EXT4_STATE_EXT_MIGRATE
);
1361 * Update reserved blocks/metadata blocks after successful
1362 * block allocation which had been deferred till now. We don't
1363 * support fallocate for non extent files. So we can update
1364 * reserve space here.
1367 (flags
& EXT4_GET_BLOCKS_DELALLOC_RESERVE
))
1368 ext4_da_update_reserve_space(inode
, retval
, 1);
1370 if (flags
& EXT4_GET_BLOCKS_DELALLOC_RESERVE
)
1371 ext4_clear_inode_state(inode
, EXT4_STATE_DELALLOC_RESERVED
);
1373 up_write((&EXT4_I(inode
)->i_data_sem
));
1374 if (retval
> 0 && map
->m_flags
& EXT4_MAP_MAPPED
) {
1375 int ret
= check_block_validity(inode
, map
);
1382 /* Maximum number of blocks we map for direct IO at once. */
1383 #define DIO_MAX_BLOCKS 4096
1385 static int _ext4_get_block(struct inode
*inode
, sector_t iblock
,
1386 struct buffer_head
*bh
, int flags
)
1388 handle_t
*handle
= ext4_journal_current_handle();
1389 struct ext4_map_blocks map
;
1390 int ret
= 0, started
= 0;
1393 map
.m_lblk
= iblock
;
1394 map
.m_len
= bh
->b_size
>> inode
->i_blkbits
;
1396 if (flags
&& !handle
) {
1397 /* Direct IO write... */
1398 if (map
.m_len
> DIO_MAX_BLOCKS
)
1399 map
.m_len
= DIO_MAX_BLOCKS
;
1400 dio_credits
= ext4_chunk_trans_blocks(inode
, map
.m_len
);
1401 handle
= ext4_journal_start(inode
, dio_credits
);
1402 if (IS_ERR(handle
)) {
1403 ret
= PTR_ERR(handle
);
1409 ret
= ext4_map_blocks(handle
, inode
, &map
, flags
);
1411 map_bh(bh
, inode
->i_sb
, map
.m_pblk
);
1412 bh
->b_state
= (bh
->b_state
& ~EXT4_MAP_FLAGS
) | map
.m_flags
;
1413 bh
->b_size
= inode
->i_sb
->s_blocksize
* map
.m_len
;
1417 ext4_journal_stop(handle
);
1421 int ext4_get_block(struct inode
*inode
, sector_t iblock
,
1422 struct buffer_head
*bh
, int create
)
1424 return _ext4_get_block(inode
, iblock
, bh
,
1425 create
? EXT4_GET_BLOCKS_CREATE
: 0);
1429 * `handle' can be NULL if create is zero
1431 struct buffer_head
*ext4_getblk(handle_t
*handle
, struct inode
*inode
,
1432 ext4_lblk_t block
, int create
, int *errp
)
1434 struct ext4_map_blocks map
;
1435 struct buffer_head
*bh
;
1438 J_ASSERT(handle
!= NULL
|| create
== 0);
1442 err
= ext4_map_blocks(handle
, inode
, &map
,
1443 create
? EXT4_GET_BLOCKS_CREATE
: 0);
1451 bh
= sb_getblk(inode
->i_sb
, map
.m_pblk
);
1456 if (map
.m_flags
& EXT4_MAP_NEW
) {
1457 J_ASSERT(create
!= 0);
1458 J_ASSERT(handle
!= NULL
);
1461 * Now that we do not always journal data, we should
1462 * keep in mind whether this should always journal the
1463 * new buffer as metadata. For now, regular file
1464 * writes use ext4_get_block instead, so it's not a
1468 BUFFER_TRACE(bh
, "call get_create_access");
1469 fatal
= ext4_journal_get_create_access(handle
, bh
);
1470 if (!fatal
&& !buffer_uptodate(bh
)) {
1471 memset(bh
->b_data
, 0, inode
->i_sb
->s_blocksize
);
1472 set_buffer_uptodate(bh
);
1475 BUFFER_TRACE(bh
, "call ext4_handle_dirty_metadata");
1476 err
= ext4_handle_dirty_metadata(handle
, inode
, bh
);
1480 BUFFER_TRACE(bh
, "not a new buffer");
1490 struct buffer_head
*ext4_bread(handle_t
*handle
, struct inode
*inode
,
1491 ext4_lblk_t block
, int create
, int *err
)
1493 struct buffer_head
*bh
;
1495 bh
= ext4_getblk(handle
, inode
, block
, create
, err
);
1498 if (buffer_uptodate(bh
))
1500 ll_rw_block(READ_META
, 1, &bh
);
1502 if (buffer_uptodate(bh
))
1509 static int walk_page_buffers(handle_t
*handle
,
1510 struct buffer_head
*head
,
1514 int (*fn
)(handle_t
*handle
,
1515 struct buffer_head
*bh
))
1517 struct buffer_head
*bh
;
1518 unsigned block_start
, block_end
;
1519 unsigned blocksize
= head
->b_size
;
1521 struct buffer_head
*next
;
1523 for (bh
= head
, block_start
= 0;
1524 ret
== 0 && (bh
!= head
|| !block_start
);
1525 block_start
= block_end
, bh
= next
) {
1526 next
= bh
->b_this_page
;
1527 block_end
= block_start
+ blocksize
;
1528 if (block_end
<= from
|| block_start
>= to
) {
1529 if (partial
&& !buffer_uptodate(bh
))
1533 err
= (*fn
)(handle
, bh
);
1541 * To preserve ordering, it is essential that the hole instantiation and
1542 * the data write be encapsulated in a single transaction. We cannot
1543 * close off a transaction and start a new one between the ext4_get_block()
1544 * and the commit_write(). So doing the jbd2_journal_start at the start of
1545 * prepare_write() is the right place.
1547 * Also, this function can nest inside ext4_writepage() ->
1548 * block_write_full_page(). In that case, we *know* that ext4_writepage()
1549 * has generated enough buffer credits to do the whole page. So we won't
1550 * block on the journal in that case, which is good, because the caller may
1553 * By accident, ext4 can be reentered when a transaction is open via
1554 * quota file writes. If we were to commit the transaction while thus
1555 * reentered, there can be a deadlock - we would be holding a quota
1556 * lock, and the commit would never complete if another thread had a
1557 * transaction open and was blocking on the quota lock - a ranking
1560 * So what we do is to rely on the fact that jbd2_journal_stop/journal_start
1561 * will _not_ run commit under these circumstances because handle->h_ref
1562 * is elevated. We'll still have enough credits for the tiny quotafile
1565 static int do_journal_get_write_access(handle_t
*handle
,
1566 struct buffer_head
*bh
)
1568 int dirty
= buffer_dirty(bh
);
1571 if (!buffer_mapped(bh
) || buffer_freed(bh
))
1574 * __block_write_begin() could have dirtied some buffers. Clean
1575 * the dirty bit as jbd2_journal_get_write_access() could complain
1576 * otherwise about fs integrity issues. Setting of the dirty bit
1577 * by __block_write_begin() isn't a real problem here as we clear
1578 * the bit before releasing a page lock and thus writeback cannot
1579 * ever write the buffer.
1582 clear_buffer_dirty(bh
);
1583 ret
= ext4_journal_get_write_access(handle
, bh
);
1585 ret
= ext4_handle_dirty_metadata(handle
, NULL
, bh
);
1590 * Truncate blocks that were not used by write. We have to truncate the
1591 * pagecache as well so that corresponding buffers get properly unmapped.
1593 static void ext4_truncate_failed_write(struct inode
*inode
)
1595 truncate_inode_pages(inode
->i_mapping
, inode
->i_size
);
1596 ext4_truncate(inode
);
1599 static int ext4_get_block_write(struct inode
*inode
, sector_t iblock
,
1600 struct buffer_head
*bh_result
, int create
);
1601 static int ext4_write_begin(struct file
*file
, struct address_space
*mapping
,
1602 loff_t pos
, unsigned len
, unsigned flags
,
1603 struct page
**pagep
, void **fsdata
)
1605 struct inode
*inode
= mapping
->host
;
1606 int ret
, needed_blocks
;
1613 trace_ext4_write_begin(inode
, pos
, len
, flags
);
1615 * Reserve one block more for addition to orphan list in case
1616 * we allocate blocks but write fails for some reason
1618 needed_blocks
= ext4_writepage_trans_blocks(inode
) + 1;
1619 index
= pos
>> PAGE_CACHE_SHIFT
;
1620 from
= pos
& (PAGE_CACHE_SIZE
- 1);
1624 handle
= ext4_journal_start(inode
, needed_blocks
);
1625 if (IS_ERR(handle
)) {
1626 ret
= PTR_ERR(handle
);
1630 /* We cannot recurse into the filesystem as the transaction is already
1632 flags
|= AOP_FLAG_NOFS
;
1634 page
= grab_cache_page_write_begin(mapping
, index
, flags
);
1636 ext4_journal_stop(handle
);
1642 if (ext4_should_dioread_nolock(inode
))
1643 ret
= __block_write_begin(page
, pos
, len
, ext4_get_block_write
);
1645 ret
= __block_write_begin(page
, pos
, len
, ext4_get_block
);
1647 if (!ret
&& ext4_should_journal_data(inode
)) {
1648 ret
= walk_page_buffers(handle
, page_buffers(page
),
1649 from
, to
, NULL
, do_journal_get_write_access
);
1654 page_cache_release(page
);
1656 * __block_write_begin may have instantiated a few blocks
1657 * outside i_size. Trim these off again. Don't need
1658 * i_size_read because we hold i_mutex.
1660 * Add inode to orphan list in case we crash before
1663 if (pos
+ len
> inode
->i_size
&& ext4_can_truncate(inode
))
1664 ext4_orphan_add(handle
, inode
);
1666 ext4_journal_stop(handle
);
1667 if (pos
+ len
> inode
->i_size
) {
1668 ext4_truncate_failed_write(inode
);
1670 * If truncate failed early the inode might
1671 * still be on the orphan list; we need to
1672 * make sure the inode is removed from the
1673 * orphan list in that case.
1676 ext4_orphan_del(NULL
, inode
);
1680 if (ret
== -ENOSPC
&& ext4_should_retry_alloc(inode
->i_sb
, &retries
))
1686 /* For write_end() in data=journal mode */
1687 static int write_end_fn(handle_t
*handle
, struct buffer_head
*bh
)
1689 if (!buffer_mapped(bh
) || buffer_freed(bh
))
1691 set_buffer_uptodate(bh
);
1692 return ext4_handle_dirty_metadata(handle
, NULL
, bh
);
1695 static int ext4_generic_write_end(struct file
*file
,
1696 struct address_space
*mapping
,
1697 loff_t pos
, unsigned len
, unsigned copied
,
1698 struct page
*page
, void *fsdata
)
1700 int i_size_changed
= 0;
1701 struct inode
*inode
= mapping
->host
;
1702 handle_t
*handle
= ext4_journal_current_handle();
1704 copied
= block_write_end(file
, mapping
, pos
, len
, copied
, page
, fsdata
);
1707 * No need to use i_size_read() here, the i_size
1708 * cannot change under us because we hold i_mutex.
1710 * But it's important to update i_size while still holding page lock:
1711 * page writeout could otherwise come in and zero beyond i_size.
1713 if (pos
+ copied
> inode
->i_size
) {
1714 i_size_write(inode
, pos
+ copied
);
1718 if (pos
+ copied
> EXT4_I(inode
)->i_disksize
) {
1719 /* We need to mark inode dirty even if
1720 * new_i_size is less that inode->i_size
1721 * bu greater than i_disksize.(hint delalloc)
1723 ext4_update_i_disksize(inode
, (pos
+ copied
));
1727 page_cache_release(page
);
1730 * Don't mark the inode dirty under page lock. First, it unnecessarily
1731 * makes the holding time of page lock longer. Second, it forces lock
1732 * ordering of page lock and transaction start for journaling
1736 ext4_mark_inode_dirty(handle
, inode
);
1742 * We need to pick up the new inode size which generic_commit_write gave us
1743 * `file' can be NULL - eg, when called from page_symlink().
1745 * ext4 never places buffers on inode->i_mapping->private_list. metadata
1746 * buffers are managed internally.
1748 static int ext4_ordered_write_end(struct file
*file
,
1749 struct address_space
*mapping
,
1750 loff_t pos
, unsigned len
, unsigned copied
,
1751 struct page
*page
, void *fsdata
)
1753 handle_t
*handle
= ext4_journal_current_handle();
1754 struct inode
*inode
= mapping
->host
;
1757 trace_ext4_ordered_write_end(inode
, pos
, len
, copied
);
1758 ret
= ext4_jbd2_file_inode(handle
, inode
);
1761 ret2
= ext4_generic_write_end(file
, mapping
, pos
, len
, copied
,
1764 if (pos
+ len
> inode
->i_size
&& ext4_can_truncate(inode
))
1765 /* if we have allocated more blocks and copied
1766 * less. We will have blocks allocated outside
1767 * inode->i_size. So truncate them
1769 ext4_orphan_add(handle
, inode
);
1773 ret2
= ext4_journal_stop(handle
);
1777 if (pos
+ len
> inode
->i_size
) {
1778 ext4_truncate_failed_write(inode
);
1780 * If truncate failed early the inode might still be
1781 * on the orphan list; we need to make sure the inode
1782 * is removed from the orphan list in that case.
1785 ext4_orphan_del(NULL
, inode
);
1789 return ret
? ret
: copied
;
1792 static int ext4_writeback_write_end(struct file
*file
,
1793 struct address_space
*mapping
,
1794 loff_t pos
, unsigned len
, unsigned copied
,
1795 struct page
*page
, void *fsdata
)
1797 handle_t
*handle
= ext4_journal_current_handle();
1798 struct inode
*inode
= mapping
->host
;
1801 trace_ext4_writeback_write_end(inode
, pos
, len
, copied
);
1802 ret2
= ext4_generic_write_end(file
, mapping
, pos
, len
, copied
,
1805 if (pos
+ len
> inode
->i_size
&& ext4_can_truncate(inode
))
1806 /* if we have allocated more blocks and copied
1807 * less. We will have blocks allocated outside
1808 * inode->i_size. So truncate them
1810 ext4_orphan_add(handle
, inode
);
1815 ret2
= ext4_journal_stop(handle
);
1819 if (pos
+ len
> inode
->i_size
) {
1820 ext4_truncate_failed_write(inode
);
1822 * If truncate failed early the inode might still be
1823 * on the orphan list; we need to make sure the inode
1824 * is removed from the orphan list in that case.
1827 ext4_orphan_del(NULL
, inode
);
1830 return ret
? ret
: copied
;
1833 static int ext4_journalled_write_end(struct file
*file
,
1834 struct address_space
*mapping
,
1835 loff_t pos
, unsigned len
, unsigned copied
,
1836 struct page
*page
, void *fsdata
)
1838 handle_t
*handle
= ext4_journal_current_handle();
1839 struct inode
*inode
= mapping
->host
;
1845 trace_ext4_journalled_write_end(inode
, pos
, len
, copied
);
1846 from
= pos
& (PAGE_CACHE_SIZE
- 1);
1850 if (!PageUptodate(page
))
1852 page_zero_new_buffers(page
, from
+copied
, to
);
1855 ret
= walk_page_buffers(handle
, page_buffers(page
), from
,
1856 to
, &partial
, write_end_fn
);
1858 SetPageUptodate(page
);
1859 new_i_size
= pos
+ copied
;
1860 if (new_i_size
> inode
->i_size
)
1861 i_size_write(inode
, pos
+copied
);
1862 ext4_set_inode_state(inode
, EXT4_STATE_JDATA
);
1863 if (new_i_size
> EXT4_I(inode
)->i_disksize
) {
1864 ext4_update_i_disksize(inode
, new_i_size
);
1865 ret2
= ext4_mark_inode_dirty(handle
, inode
);
1871 page_cache_release(page
);
1872 if (pos
+ len
> inode
->i_size
&& ext4_can_truncate(inode
))
1873 /* if we have allocated more blocks and copied
1874 * less. We will have blocks allocated outside
1875 * inode->i_size. So truncate them
1877 ext4_orphan_add(handle
, inode
);
1879 ret2
= ext4_journal_stop(handle
);
1882 if (pos
+ len
> inode
->i_size
) {
1883 ext4_truncate_failed_write(inode
);
1885 * If truncate failed early the inode might still be
1886 * on the orphan list; we need to make sure the inode
1887 * is removed from the orphan list in that case.
1890 ext4_orphan_del(NULL
, inode
);
1893 return ret
? ret
: copied
;
1897 * Reserve a single block located at lblock
1899 static int ext4_da_reserve_space(struct inode
*inode
, ext4_lblk_t lblock
)
1902 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
1903 struct ext4_inode_info
*ei
= EXT4_I(inode
);
1904 unsigned long md_needed
;
1908 * recalculate the amount of metadata blocks to reserve
1909 * in order to allocate nrblocks
1910 * worse case is one extent per block
1913 spin_lock(&ei
->i_block_reservation_lock
);
1914 md_needed
= ext4_calc_metadata_amount(inode
, lblock
);
1915 trace_ext4_da_reserve_space(inode
, md_needed
);
1916 spin_unlock(&ei
->i_block_reservation_lock
);
1919 * We will charge metadata quota at writeout time; this saves
1920 * us from metadata over-estimation, though we may go over by
1921 * a small amount in the end. Here we just reserve for data.
1923 ret
= dquot_reserve_block(inode
, 1);
1927 * We do still charge estimated metadata to the sb though;
1928 * we cannot afford to run out of free blocks.
1930 if (ext4_claim_free_blocks(sbi
, md_needed
+ 1)) {
1931 dquot_release_reservation_block(inode
, 1);
1932 if (ext4_should_retry_alloc(inode
->i_sb
, &retries
)) {
1938 spin_lock(&ei
->i_block_reservation_lock
);
1939 ei
->i_reserved_data_blocks
++;
1940 ei
->i_reserved_meta_blocks
+= md_needed
;
1941 spin_unlock(&ei
->i_block_reservation_lock
);
1943 return 0; /* success */
1946 static void ext4_da_release_space(struct inode
*inode
, int to_free
)
1948 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
1949 struct ext4_inode_info
*ei
= EXT4_I(inode
);
1952 return; /* Nothing to release, exit */
1954 spin_lock(&EXT4_I(inode
)->i_block_reservation_lock
);
1956 trace_ext4_da_release_space(inode
, to_free
);
1957 if (unlikely(to_free
> ei
->i_reserved_data_blocks
)) {
1959 * if there aren't enough reserved blocks, then the
1960 * counter is messed up somewhere. Since this
1961 * function is called from invalidate page, it's
1962 * harmless to return without any action.
1964 ext4_msg(inode
->i_sb
, KERN_NOTICE
, "ext4_da_release_space: "
1965 "ino %lu, to_free %d with only %d reserved "
1966 "data blocks\n", inode
->i_ino
, to_free
,
1967 ei
->i_reserved_data_blocks
);
1969 to_free
= ei
->i_reserved_data_blocks
;
1971 ei
->i_reserved_data_blocks
-= to_free
;
1973 if (ei
->i_reserved_data_blocks
== 0) {
1975 * We can release all of the reserved metadata blocks
1976 * only when we have written all of the delayed
1977 * allocation blocks.
1979 percpu_counter_sub(&sbi
->s_dirtyblocks_counter
,
1980 ei
->i_reserved_meta_blocks
);
1981 ei
->i_reserved_meta_blocks
= 0;
1982 ei
->i_da_metadata_calc_len
= 0;
1985 /* update fs dirty data blocks counter */
1986 percpu_counter_sub(&sbi
->s_dirtyblocks_counter
, to_free
);
1988 spin_unlock(&EXT4_I(inode
)->i_block_reservation_lock
);
1990 dquot_release_reservation_block(inode
, to_free
);
1993 static void ext4_da_page_release_reservation(struct page
*page
,
1994 unsigned long offset
)
1997 struct buffer_head
*head
, *bh
;
1998 unsigned int curr_off
= 0;
2000 head
= page_buffers(page
);
2003 unsigned int next_off
= curr_off
+ bh
->b_size
;
2005 if ((offset
<= curr_off
) && (buffer_delay(bh
))) {
2007 clear_buffer_delay(bh
);
2009 curr_off
= next_off
;
2010 } while ((bh
= bh
->b_this_page
) != head
);
2011 ext4_da_release_space(page
->mapping
->host
, to_release
);
2015 * Delayed allocation stuff
2019 * mpage_da_submit_io - walks through extent of pages and try to write
2020 * them with writepage() call back
2022 * @mpd->inode: inode
2023 * @mpd->first_page: first page of the extent
2024 * @mpd->next_page: page after the last page of the extent
2026 * By the time mpage_da_submit_io() is called we expect all blocks
2027 * to be allocated. this may be wrong if allocation failed.
2029 * As pages are already locked by write_cache_pages(), we can't use it
2031 static int mpage_da_submit_io(struct mpage_da_data
*mpd
,
2032 struct ext4_map_blocks
*map
)
2034 struct pagevec pvec
;
2035 unsigned long index
, end
;
2036 int ret
= 0, err
, nr_pages
, i
;
2037 struct inode
*inode
= mpd
->inode
;
2038 struct address_space
*mapping
= inode
->i_mapping
;
2039 loff_t size
= i_size_read(inode
);
2040 unsigned int len
, block_start
;
2041 struct buffer_head
*bh
, *page_bufs
= NULL
;
2042 int journal_data
= ext4_should_journal_data(inode
);
2043 sector_t pblock
= 0, cur_logical
= 0;
2044 struct ext4_io_submit io_submit
;
2046 BUG_ON(mpd
->next_page
<= mpd
->first_page
);
2047 memset(&io_submit
, 0, sizeof(io_submit
));
2049 * We need to start from the first_page to the next_page - 1
2050 * to make sure we also write the mapped dirty buffer_heads.
2051 * If we look at mpd->b_blocknr we would only be looking
2052 * at the currently mapped buffer_heads.
2054 index
= mpd
->first_page
;
2055 end
= mpd
->next_page
- 1;
2057 pagevec_init(&pvec
, 0);
2058 while (index
<= end
) {
2059 nr_pages
= pagevec_lookup(&pvec
, mapping
, index
, PAGEVEC_SIZE
);
2062 for (i
= 0; i
< nr_pages
; i
++) {
2063 int commit_write
= 0, redirty_page
= 0;
2064 struct page
*page
= pvec
.pages
[i
];
2066 index
= page
->index
;
2070 if (index
== size
>> PAGE_CACHE_SHIFT
)
2071 len
= size
& ~PAGE_CACHE_MASK
;
2073 len
= PAGE_CACHE_SIZE
;
2075 cur_logical
= index
<< (PAGE_CACHE_SHIFT
-
2077 pblock
= map
->m_pblk
+ (cur_logical
-
2082 BUG_ON(!PageLocked(page
));
2083 BUG_ON(PageWriteback(page
));
2086 * If the page does not have buffers (for
2087 * whatever reason), try to create them using
2088 * __block_write_begin. If this fails,
2089 * redirty the page and move on.
2091 if (!page_has_buffers(page
)) {
2092 if (__block_write_begin(page
, 0, len
,
2093 noalloc_get_block_write
)) {
2095 redirty_page_for_writepage(mpd
->wbc
,
2103 bh
= page_bufs
= page_buffers(page
);
2108 if (map
&& (cur_logical
>= map
->m_lblk
) &&
2109 (cur_logical
<= (map
->m_lblk
+
2110 (map
->m_len
- 1)))) {
2111 if (buffer_delay(bh
)) {
2112 clear_buffer_delay(bh
);
2113 bh
->b_blocknr
= pblock
;
2115 if (buffer_unwritten(bh
) ||
2117 BUG_ON(bh
->b_blocknr
!= pblock
);
2118 if (map
->m_flags
& EXT4_MAP_UNINIT
)
2119 set_buffer_uninit(bh
);
2120 clear_buffer_unwritten(bh
);
2123 /* redirty page if block allocation undone */
2124 if (buffer_delay(bh
) || buffer_unwritten(bh
))
2126 bh
= bh
->b_this_page
;
2127 block_start
+= bh
->b_size
;
2130 } while (bh
!= page_bufs
);
2136 /* mark the buffer_heads as dirty & uptodate */
2137 block_commit_write(page
, 0, len
);
2140 * Delalloc doesn't support data journalling,
2141 * but eventually maybe we'll lift this
2144 if (unlikely(journal_data
&& PageChecked(page
)))
2145 err
= __ext4_journalled_writepage(page
, len
);
2146 else if (test_opt(inode
->i_sb
, MBLK_IO_SUBMIT
))
2147 err
= ext4_bio_write_page(&io_submit
, page
,
2150 err
= block_write_full_page(page
,
2151 noalloc_get_block_write
, mpd
->wbc
);
2154 mpd
->pages_written
++;
2156 * In error case, we have to continue because
2157 * remaining pages are still locked
2162 pagevec_release(&pvec
);
2164 ext4_io_submit(&io_submit
);
2168 static void ext4_da_block_invalidatepages(struct mpage_da_data
*mpd
)
2172 struct pagevec pvec
;
2173 struct inode
*inode
= mpd
->inode
;
2174 struct address_space
*mapping
= inode
->i_mapping
;
2176 index
= mpd
->first_page
;
2177 end
= mpd
->next_page
- 1;
2178 while (index
<= end
) {
2179 nr_pages
= pagevec_lookup(&pvec
, mapping
, index
, PAGEVEC_SIZE
);
2182 for (i
= 0; i
< nr_pages
; i
++) {
2183 struct page
*page
= pvec
.pages
[i
];
2184 if (page
->index
> end
)
2186 BUG_ON(!PageLocked(page
));
2187 BUG_ON(PageWriteback(page
));
2188 block_invalidatepage(page
, 0);
2189 ClearPageUptodate(page
);
2192 index
= pvec
.pages
[nr_pages
- 1]->index
+ 1;
2193 pagevec_release(&pvec
);
2198 static void ext4_print_free_blocks(struct inode
*inode
)
2200 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
2201 printk(KERN_CRIT
"Total free blocks count %lld\n",
2202 ext4_count_free_blocks(inode
->i_sb
));
2203 printk(KERN_CRIT
"Free/Dirty block details\n");
2204 printk(KERN_CRIT
"free_blocks=%lld\n",
2205 (long long) percpu_counter_sum(&sbi
->s_freeblocks_counter
));
2206 printk(KERN_CRIT
"dirty_blocks=%lld\n",
2207 (long long) percpu_counter_sum(&sbi
->s_dirtyblocks_counter
));
2208 printk(KERN_CRIT
"Block reservation details\n");
2209 printk(KERN_CRIT
"i_reserved_data_blocks=%u\n",
2210 EXT4_I(inode
)->i_reserved_data_blocks
);
2211 printk(KERN_CRIT
"i_reserved_meta_blocks=%u\n",
2212 EXT4_I(inode
)->i_reserved_meta_blocks
);
2217 * mpage_da_map_and_submit - go through given space, map them
2218 * if necessary, and then submit them for I/O
2220 * @mpd - bh describing space
2222 * The function skips space we know is already mapped to disk blocks.
2225 static void mpage_da_map_and_submit(struct mpage_da_data
*mpd
)
2227 int err
, blks
, get_blocks_flags
;
2228 struct ext4_map_blocks map
, *mapp
= NULL
;
2229 sector_t next
= mpd
->b_blocknr
;
2230 unsigned max_blocks
= mpd
->b_size
>> mpd
->inode
->i_blkbits
;
2231 loff_t disksize
= EXT4_I(mpd
->inode
)->i_disksize
;
2232 handle_t
*handle
= NULL
;
2235 * If the blocks are mapped already, or we couldn't accumulate
2236 * any blocks, then proceed immediately to the submission stage.
2238 if ((mpd
->b_size
== 0) ||
2239 ((mpd
->b_state
& (1 << BH_Mapped
)) &&
2240 !(mpd
->b_state
& (1 << BH_Delay
)) &&
2241 !(mpd
->b_state
& (1 << BH_Unwritten
))))
2244 handle
= ext4_journal_current_handle();
2248 * Call ext4_map_blocks() to allocate any delayed allocation
2249 * blocks, or to convert an uninitialized extent to be
2250 * initialized (in the case where we have written into
2251 * one or more preallocated blocks).
2253 * We pass in the magic EXT4_GET_BLOCKS_DELALLOC_RESERVE to
2254 * indicate that we are on the delayed allocation path. This
2255 * affects functions in many different parts of the allocation
2256 * call path. This flag exists primarily because we don't
2257 * want to change *many* call functions, so ext4_map_blocks()
2258 * will set the EXT4_STATE_DELALLOC_RESERVED flag once the
2259 * inode's allocation semaphore is taken.
2261 * If the blocks in questions were delalloc blocks, set
2262 * EXT4_GET_BLOCKS_DELALLOC_RESERVE so the delalloc accounting
2263 * variables are updated after the blocks have been allocated.
2266 map
.m_len
= max_blocks
;
2267 get_blocks_flags
= EXT4_GET_BLOCKS_CREATE
;
2268 if (ext4_should_dioread_nolock(mpd
->inode
))
2269 get_blocks_flags
|= EXT4_GET_BLOCKS_IO_CREATE_EXT
;
2270 if (mpd
->b_state
& (1 << BH_Delay
))
2271 get_blocks_flags
|= EXT4_GET_BLOCKS_DELALLOC_RESERVE
;
2273 blks
= ext4_map_blocks(handle
, mpd
->inode
, &map
, get_blocks_flags
);
2275 struct super_block
*sb
= mpd
->inode
->i_sb
;
2279 * If get block returns EAGAIN or ENOSPC and there
2280 * appears to be free blocks we will call
2281 * ext4_writepage() for all of the pages which will
2282 * just redirty the pages.
2287 if (err
== -ENOSPC
&&
2288 ext4_count_free_blocks(sb
)) {
2294 * get block failure will cause us to loop in
2295 * writepages, because a_ops->writepage won't be able
2296 * to make progress. The page will be redirtied by
2297 * writepage and writepages will again try to write
2300 if (!(EXT4_SB(sb
)->s_mount_flags
& EXT4_MF_FS_ABORTED
)) {
2301 ext4_msg(sb
, KERN_CRIT
,
2302 "delayed block allocation failed for inode %lu "
2303 "at logical offset %llu with max blocks %zd "
2304 "with error %d", mpd
->inode
->i_ino
,
2305 (unsigned long long) next
,
2306 mpd
->b_size
>> mpd
->inode
->i_blkbits
, err
);
2307 ext4_msg(sb
, KERN_CRIT
,
2308 "This should not happen!! Data will be lost\n");
2310 ext4_print_free_blocks(mpd
->inode
);
2312 /* invalidate all the pages */
2313 ext4_da_block_invalidatepages(mpd
);
2315 /* Mark this page range as having been completed */
2322 if (map
.m_flags
& EXT4_MAP_NEW
) {
2323 struct block_device
*bdev
= mpd
->inode
->i_sb
->s_bdev
;
2326 for (i
= 0; i
< map
.m_len
; i
++)
2327 unmap_underlying_metadata(bdev
, map
.m_pblk
+ i
);
2330 if (ext4_should_order_data(mpd
->inode
)) {
2331 err
= ext4_jbd2_file_inode(handle
, mpd
->inode
);
2333 /* This only happens if the journal is aborted */
2338 * Update on-disk size along with block allocation.
2340 disksize
= ((loff_t
) next
+ blks
) << mpd
->inode
->i_blkbits
;
2341 if (disksize
> i_size_read(mpd
->inode
))
2342 disksize
= i_size_read(mpd
->inode
);
2343 if (disksize
> EXT4_I(mpd
->inode
)->i_disksize
) {
2344 ext4_update_i_disksize(mpd
->inode
, disksize
);
2345 err
= ext4_mark_inode_dirty(handle
, mpd
->inode
);
2347 ext4_error(mpd
->inode
->i_sb
,
2348 "Failed to mark inode %lu dirty",
2353 mpage_da_submit_io(mpd
, mapp
);
2357 #define BH_FLAGS ((1 << BH_Uptodate) | (1 << BH_Mapped) | \
2358 (1 << BH_Delay) | (1 << BH_Unwritten))
2361 * mpage_add_bh_to_extent - try to add one more block to extent of blocks
2363 * @mpd->lbh - extent of blocks
2364 * @logical - logical number of the block in the file
2365 * @bh - bh of the block (used to access block's state)
2367 * the function is used to collect contig. blocks in same state
2369 static void mpage_add_bh_to_extent(struct mpage_da_data
*mpd
,
2370 sector_t logical
, size_t b_size
,
2371 unsigned long b_state
)
2374 int nrblocks
= mpd
->b_size
>> mpd
->inode
->i_blkbits
;
2377 * XXX Don't go larger than mballoc is willing to allocate
2378 * This is a stopgap solution. We eventually need to fold
2379 * mpage_da_submit_io() into this function and then call
2380 * ext4_map_blocks() multiple times in a loop
2382 if (nrblocks
>= 8*1024*1024/mpd
->inode
->i_sb
->s_blocksize
)
2385 /* check if thereserved journal credits might overflow */
2386 if (!(ext4_test_inode_flag(mpd
->inode
, EXT4_INODE_EXTENTS
))) {
2387 if (nrblocks
>= EXT4_MAX_TRANS_DATA
) {
2389 * With non-extent format we are limited by the journal
2390 * credit available. Total credit needed to insert
2391 * nrblocks contiguous blocks is dependent on the
2392 * nrblocks. So limit nrblocks.
2395 } else if ((nrblocks
+ (b_size
>> mpd
->inode
->i_blkbits
)) >
2396 EXT4_MAX_TRANS_DATA
) {
2398 * Adding the new buffer_head would make it cross the
2399 * allowed limit for which we have journal credit
2400 * reserved. So limit the new bh->b_size
2402 b_size
= (EXT4_MAX_TRANS_DATA
- nrblocks
) <<
2403 mpd
->inode
->i_blkbits
;
2404 /* we will do mpage_da_submit_io in the next loop */
2408 * First block in the extent
2410 if (mpd
->b_size
== 0) {
2411 mpd
->b_blocknr
= logical
;
2412 mpd
->b_size
= b_size
;
2413 mpd
->b_state
= b_state
& BH_FLAGS
;
2417 next
= mpd
->b_blocknr
+ nrblocks
;
2419 * Can we merge the block to our big extent?
2421 if (logical
== next
&& (b_state
& BH_FLAGS
) == mpd
->b_state
) {
2422 mpd
->b_size
+= b_size
;
2428 * We couldn't merge the block to our extent, so we
2429 * need to flush current extent and start new one
2431 mpage_da_map_and_submit(mpd
);
2435 static int ext4_bh_delay_or_unwritten(handle_t
*handle
, struct buffer_head
*bh
)
2437 return (buffer_delay(bh
) || buffer_unwritten(bh
)) && buffer_dirty(bh
);
2441 * This is a special get_blocks_t callback which is used by
2442 * ext4_da_write_begin(). It will either return mapped block or
2443 * reserve space for a single block.
2445 * For delayed buffer_head we have BH_Mapped, BH_New, BH_Delay set.
2446 * We also have b_blocknr = -1 and b_bdev initialized properly
2448 * For unwritten buffer_head we have BH_Mapped, BH_New, BH_Unwritten set.
2449 * We also have b_blocknr = physicalblock mapping unwritten extent and b_bdev
2450 * initialized properly.
2452 static int ext4_da_get_block_prep(struct inode
*inode
, sector_t iblock
,
2453 struct buffer_head
*bh
, int create
)
2455 struct ext4_map_blocks map
;
2457 sector_t invalid_block
= ~((sector_t
) 0xffff);
2459 if (invalid_block
< ext4_blocks_count(EXT4_SB(inode
->i_sb
)->s_es
))
2462 BUG_ON(create
== 0);
2463 BUG_ON(bh
->b_size
!= inode
->i_sb
->s_blocksize
);
2465 map
.m_lblk
= iblock
;
2469 * first, we need to know whether the block is allocated already
2470 * preallocated blocks are unmapped but should treated
2471 * the same as allocated blocks.
2473 ret
= ext4_map_blocks(NULL
, inode
, &map
, 0);
2477 if (buffer_delay(bh
))
2478 return 0; /* Not sure this could or should happen */
2480 * XXX: __block_write_begin() unmaps passed block, is it OK?
2482 ret
= ext4_da_reserve_space(inode
, iblock
);
2484 /* not enough space to reserve */
2487 map_bh(bh
, inode
->i_sb
, invalid_block
);
2489 set_buffer_delay(bh
);
2493 map_bh(bh
, inode
->i_sb
, map
.m_pblk
);
2494 bh
->b_state
= (bh
->b_state
& ~EXT4_MAP_FLAGS
) | map
.m_flags
;
2496 if (buffer_unwritten(bh
)) {
2497 /* A delayed write to unwritten bh should be marked
2498 * new and mapped. Mapped ensures that we don't do
2499 * get_block multiple times when we write to the same
2500 * offset and new ensures that we do proper zero out
2501 * for partial write.
2504 set_buffer_mapped(bh
);
2510 * This function is used as a standard get_block_t calback function
2511 * when there is no desire to allocate any blocks. It is used as a
2512 * callback function for block_write_begin() and block_write_full_page().
2513 * These functions should only try to map a single block at a time.
2515 * Since this function doesn't do block allocations even if the caller
2516 * requests it by passing in create=1, it is critically important that
2517 * any caller checks to make sure that any buffer heads are returned
2518 * by this function are either all already mapped or marked for
2519 * delayed allocation before calling block_write_full_page(). Otherwise,
2520 * b_blocknr could be left unitialized, and the page write functions will
2521 * be taken by surprise.
2523 static int noalloc_get_block_write(struct inode
*inode
, sector_t iblock
,
2524 struct buffer_head
*bh_result
, int create
)
2526 BUG_ON(bh_result
->b_size
!= inode
->i_sb
->s_blocksize
);
2527 return _ext4_get_block(inode
, iblock
, bh_result
, 0);
2530 static int bget_one(handle_t
*handle
, struct buffer_head
*bh
)
2536 static int bput_one(handle_t
*handle
, struct buffer_head
*bh
)
2542 static int __ext4_journalled_writepage(struct page
*page
,
2545 struct address_space
*mapping
= page
->mapping
;
2546 struct inode
*inode
= mapping
->host
;
2547 struct buffer_head
*page_bufs
;
2548 handle_t
*handle
= NULL
;
2552 ClearPageChecked(page
);
2553 page_bufs
= page_buffers(page
);
2555 walk_page_buffers(handle
, page_bufs
, 0, len
, NULL
, bget_one
);
2556 /* As soon as we unlock the page, it can go away, but we have
2557 * references to buffers so we are safe */
2560 handle
= ext4_journal_start(inode
, ext4_writepage_trans_blocks(inode
));
2561 if (IS_ERR(handle
)) {
2562 ret
= PTR_ERR(handle
);
2566 ret
= walk_page_buffers(handle
, page_bufs
, 0, len
, NULL
,
2567 do_journal_get_write_access
);
2569 err
= walk_page_buffers(handle
, page_bufs
, 0, len
, NULL
,
2573 err
= ext4_journal_stop(handle
);
2577 walk_page_buffers(handle
, page_bufs
, 0, len
, NULL
, bput_one
);
2578 ext4_set_inode_state(inode
, EXT4_STATE_JDATA
);
2583 static int ext4_set_bh_endio(struct buffer_head
*bh
, struct inode
*inode
);
2584 static void ext4_end_io_buffer_write(struct buffer_head
*bh
, int uptodate
);
2587 * Note that we don't need to start a transaction unless we're journaling data
2588 * because we should have holes filled from ext4_page_mkwrite(). We even don't
2589 * need to file the inode to the transaction's list in ordered mode because if
2590 * we are writing back data added by write(), the inode is already there and if
2591 * we are writing back data modified via mmap(), noone guarantees in which
2592 * transaction the data will hit the disk. In case we are journaling data, we
2593 * cannot start transaction directly because transaction start ranks above page
2594 * lock so we have to do some magic.
2596 * This function can get called via...
2597 * - ext4_da_writepages after taking page lock (have journal handle)
2598 * - journal_submit_inode_data_buffers (no journal handle)
2599 * - shrink_page_list via pdflush (no journal handle)
2600 * - grab_page_cache when doing write_begin (have journal handle)
2602 * We don't do any block allocation in this function. If we have page with
2603 * multiple blocks we need to write those buffer_heads that are mapped. This
2604 * is important for mmaped based write. So if we do with blocksize 1K
2605 * truncate(f, 1024);
2606 * a = mmap(f, 0, 4096);
2608 * truncate(f, 4096);
2609 * we have in the page first buffer_head mapped via page_mkwrite call back
2610 * but other bufer_heads would be unmapped but dirty(dirty done via the
2611 * do_wp_page). So writepage should write the first block. If we modify
2612 * the mmap area beyond 1024 we will again get a page_fault and the
2613 * page_mkwrite callback will do the block allocation and mark the
2614 * buffer_heads mapped.
2616 * We redirty the page if we have any buffer_heads that is either delay or
2617 * unwritten in the page.
2619 * We can get recursively called as show below.
2621 * ext4_writepage() -> kmalloc() -> __alloc_pages() -> page_launder() ->
2624 * But since we don't do any block allocation we should not deadlock.
2625 * Page also have the dirty flag cleared so we don't get recurive page_lock.
2627 static int ext4_writepage(struct page
*page
,
2628 struct writeback_control
*wbc
)
2630 int ret
= 0, commit_write
= 0;
2633 struct buffer_head
*page_bufs
= NULL
;
2634 struct inode
*inode
= page
->mapping
->host
;
2636 trace_ext4_writepage(inode
, page
);
2637 size
= i_size_read(inode
);
2638 if (page
->index
== size
>> PAGE_CACHE_SHIFT
)
2639 len
= size
& ~PAGE_CACHE_MASK
;
2641 len
= PAGE_CACHE_SIZE
;
2644 * If the page does not have buffers (for whatever reason),
2645 * try to create them using __block_write_begin. If this
2646 * fails, redirty the page and move on.
2648 if (!page_has_buffers(page
)) {
2649 if (__block_write_begin(page
, 0, len
,
2650 noalloc_get_block_write
)) {
2652 redirty_page_for_writepage(wbc
, page
);
2658 page_bufs
= page_buffers(page
);
2659 if (walk_page_buffers(NULL
, page_bufs
, 0, len
, NULL
,
2660 ext4_bh_delay_or_unwritten
)) {
2662 * We don't want to do block allocation, so redirty
2663 * the page and return. We may reach here when we do
2664 * a journal commit via journal_submit_inode_data_buffers.
2665 * We can also reach here via shrink_page_list
2670 /* now mark the buffer_heads as dirty and uptodate */
2671 block_commit_write(page
, 0, len
);
2673 if (PageChecked(page
) && ext4_should_journal_data(inode
))
2675 * It's mmapped pagecache. Add buffers and journal it. There
2676 * doesn't seem much point in redirtying the page here.
2678 return __ext4_journalled_writepage(page
, len
);
2680 if (buffer_uninit(page_bufs
)) {
2681 ext4_set_bh_endio(page_bufs
, inode
);
2682 ret
= block_write_full_page_endio(page
, noalloc_get_block_write
,
2683 wbc
, ext4_end_io_buffer_write
);
2685 ret
= block_write_full_page(page
, noalloc_get_block_write
,
2692 * This is called via ext4_da_writepages() to
2693 * calulate the total number of credits to reserve to fit
2694 * a single extent allocation into a single transaction,
2695 * ext4_da_writpeages() will loop calling this before
2696 * the block allocation.
2699 static int ext4_da_writepages_trans_blocks(struct inode
*inode
)
2701 int max_blocks
= EXT4_I(inode
)->i_reserved_data_blocks
;
2704 * With non-extent format the journal credit needed to
2705 * insert nrblocks contiguous block is dependent on
2706 * number of contiguous block. So we will limit
2707 * number of contiguous block to a sane value
2709 if (!(ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
)) &&
2710 (max_blocks
> EXT4_MAX_TRANS_DATA
))
2711 max_blocks
= EXT4_MAX_TRANS_DATA
;
2713 return ext4_chunk_trans_blocks(inode
, max_blocks
);
2717 * write_cache_pages_da - walk the list of dirty pages of the given
2718 * address space and accumulate pages that need writing, and call
2719 * mpage_da_map_and_submit to map the pages and then write them.
2721 static int write_cache_pages_da(struct address_space
*mapping
,
2722 struct writeback_control
*wbc
,
2723 struct mpage_da_data
*mpd
,
2724 pgoff_t
*done_index
)
2726 struct inode
*inode
= mpd
->inode
;
2727 struct buffer_head
*bh
, *head
;
2731 struct pagevec pvec
;
2734 pgoff_t end
; /* Inclusive */
2735 long nr_to_write
= wbc
->nr_to_write
;
2738 pagevec_init(&pvec
, 0);
2739 index
= wbc
->range_start
>> PAGE_CACHE_SHIFT
;
2740 end
= wbc
->range_end
>> PAGE_CACHE_SHIFT
;
2742 if (wbc
->sync_mode
== WB_SYNC_ALL
)
2743 tag
= PAGECACHE_TAG_TOWRITE
;
2745 tag
= PAGECACHE_TAG_DIRTY
;
2747 *done_index
= index
;
2748 while (!done
&& (index
<= end
)) {
2751 nr_pages
= pagevec_lookup_tag(&pvec
, mapping
, &index
, tag
,
2752 min(end
- index
, (pgoff_t
)PAGEVEC_SIZE
-1) + 1);
2756 for (i
= 0; i
< nr_pages
; i
++) {
2757 struct page
*page
= pvec
.pages
[i
];
2760 * At this point, the page may be truncated or
2761 * invalidated (changing page->mapping to NULL), or
2762 * even swizzled back from swapper_space to tmpfs file
2763 * mapping. However, page->index will not change
2764 * because we have a reference on the page.
2766 if (page
->index
> end
) {
2771 *done_index
= page
->index
+ 1;
2776 * Page truncated or invalidated. We can freely skip it
2777 * then, even for data integrity operations: the page
2778 * has disappeared concurrently, so there could be no
2779 * real expectation of this data interity operation
2780 * even if there is now a new, dirty page at the same
2781 * pagecache address.
2783 if (unlikely(page
->mapping
!= mapping
)) {
2789 if (!PageDirty(page
)) {
2790 /* someone wrote it for us */
2791 goto continue_unlock
;
2794 if (PageWriteback(page
)) {
2795 if (wbc
->sync_mode
!= WB_SYNC_NONE
)
2796 wait_on_page_writeback(page
);
2798 goto continue_unlock
;
2801 BUG_ON(PageWriteback(page
));
2802 if (!clear_page_dirty_for_io(page
))
2803 goto continue_unlock
;
2805 /* BEGIN __mpage_da_writepage */
2808 * Can we merge this page to current extent?
2810 if (mpd
->next_page
!= page
->index
) {
2812 * Nope, we can't. So, we map
2813 * non-allocated blocks and start IO
2816 if (mpd
->next_page
!= mpd
->first_page
) {
2817 mpage_da_map_and_submit(mpd
);
2819 * skip rest of the page in the page_vec
2821 redirty_page_for_writepage(wbc
, page
);
2823 ret
= MPAGE_DA_EXTENT_TAIL
;
2828 * Start next extent of pages and blocks
2830 mpd
->first_page
= page
->index
;
2836 mpd
->next_page
= page
->index
+ 1;
2837 logical
= (sector_t
) page
->index
<<
2838 (PAGE_CACHE_SHIFT
- inode
->i_blkbits
);
2840 if (!page_has_buffers(page
)) {
2841 mpage_add_bh_to_extent(mpd
, logical
, PAGE_CACHE_SIZE
,
2842 (1 << BH_Dirty
) | (1 << BH_Uptodate
));
2844 ret
= MPAGE_DA_EXTENT_TAIL
;
2849 * Page with regular buffer heads, just add all dirty ones
2851 head
= page_buffers(page
);
2854 BUG_ON(buffer_locked(bh
));
2856 * We need to try to allocate
2857 * unmapped blocks in the same page.
2858 * Otherwise we won't make progress
2859 * with the page in ext4_writepage
2861 if (ext4_bh_delay_or_unwritten(NULL
, bh
)) {
2862 mpage_add_bh_to_extent(mpd
, logical
,
2866 ret
= MPAGE_DA_EXTENT_TAIL
;
2869 } else if (buffer_dirty(bh
) && (buffer_mapped(bh
))) {
2871 * mapped dirty buffer. We need to update
2872 * the b_state because we look at
2873 * b_state in mpage_da_map_blocks. We don't
2874 * update b_size because if we find an
2875 * unmapped buffer_head later we need to
2876 * use the b_state flag of that buffer_head.
2878 if (mpd
->b_size
== 0)
2879 mpd
->b_state
= bh
->b_state
& BH_FLAGS
;
2882 } while ((bh
= bh
->b_this_page
) != head
);
2887 /* END __mpage_da_writepage */
2889 if (nr_to_write
> 0) {
2891 if (nr_to_write
== 0 &&
2892 wbc
->sync_mode
== WB_SYNC_NONE
) {
2894 * We stop writing back only if we are
2895 * not doing integrity sync. In case of
2896 * integrity sync we have to keep going
2897 * because someone may be concurrently
2898 * dirtying pages, and we might have
2899 * synced a lot of newly appeared dirty
2900 * pages, but have not synced all of the
2908 pagevec_release(&pvec
);
2913 pagevec_release(&pvec
);
2919 static int ext4_da_writepages(struct address_space
*mapping
,
2920 struct writeback_control
*wbc
)
2923 int range_whole
= 0;
2924 handle_t
*handle
= NULL
;
2925 struct mpage_da_data mpd
;
2926 struct inode
*inode
= mapping
->host
;
2927 int pages_written
= 0;
2929 unsigned int max_pages
;
2930 int range_cyclic
, cycled
= 1, io_done
= 0;
2931 int needed_blocks
, ret
= 0;
2932 long desired_nr_to_write
, nr_to_writebump
= 0;
2933 loff_t range_start
= wbc
->range_start
;
2934 struct ext4_sb_info
*sbi
= EXT4_SB(mapping
->host
->i_sb
);
2935 pgoff_t done_index
= 0;
2938 trace_ext4_da_writepages(inode
, wbc
);
2941 * No pages to write? This is mainly a kludge to avoid starting
2942 * a transaction for special inodes like journal inode on last iput()
2943 * because that could violate lock ordering on umount
2945 if (!mapping
->nrpages
|| !mapping_tagged(mapping
, PAGECACHE_TAG_DIRTY
))
2949 * If the filesystem has aborted, it is read-only, so return
2950 * right away instead of dumping stack traces later on that
2951 * will obscure the real source of the problem. We test
2952 * EXT4_MF_FS_ABORTED instead of sb->s_flag's MS_RDONLY because
2953 * the latter could be true if the filesystem is mounted
2954 * read-only, and in that case, ext4_da_writepages should
2955 * *never* be called, so if that ever happens, we would want
2958 if (unlikely(sbi
->s_mount_flags
& EXT4_MF_FS_ABORTED
))
2961 if (wbc
->range_start
== 0 && wbc
->range_end
== LLONG_MAX
)
2964 range_cyclic
= wbc
->range_cyclic
;
2965 if (wbc
->range_cyclic
) {
2966 index
= mapping
->writeback_index
;
2969 wbc
->range_start
= index
<< PAGE_CACHE_SHIFT
;
2970 wbc
->range_end
= LLONG_MAX
;
2971 wbc
->range_cyclic
= 0;
2974 index
= wbc
->range_start
>> PAGE_CACHE_SHIFT
;
2975 end
= wbc
->range_end
>> PAGE_CACHE_SHIFT
;
2979 * This works around two forms of stupidity. The first is in
2980 * the writeback code, which caps the maximum number of pages
2981 * written to be 1024 pages. This is wrong on multiple
2982 * levels; different architectues have a different page size,
2983 * which changes the maximum amount of data which gets
2984 * written. Secondly, 4 megabytes is way too small. XFS
2985 * forces this value to be 16 megabytes by multiplying
2986 * nr_to_write parameter by four, and then relies on its
2987 * allocator to allocate larger extents to make them
2988 * contiguous. Unfortunately this brings us to the second
2989 * stupidity, which is that ext4's mballoc code only allocates
2990 * at most 2048 blocks. So we force contiguous writes up to
2991 * the number of dirty blocks in the inode, or
2992 * sbi->max_writeback_mb_bump whichever is smaller.
2994 max_pages
= sbi
->s_max_writeback_mb_bump
<< (20 - PAGE_CACHE_SHIFT
);
2995 if (!range_cyclic
&& range_whole
) {
2996 if (wbc
->nr_to_write
== LONG_MAX
)
2997 desired_nr_to_write
= wbc
->nr_to_write
;
2999 desired_nr_to_write
= wbc
->nr_to_write
* 8;
3001 desired_nr_to_write
= ext4_num_dirty_pages(inode
, index
,
3003 if (desired_nr_to_write
> max_pages
)
3004 desired_nr_to_write
= max_pages
;
3006 if (wbc
->nr_to_write
< desired_nr_to_write
) {
3007 nr_to_writebump
= desired_nr_to_write
- wbc
->nr_to_write
;
3008 wbc
->nr_to_write
= desired_nr_to_write
;
3012 mpd
.inode
= mapping
->host
;
3014 pages_skipped
= wbc
->pages_skipped
;
3017 if (wbc
->sync_mode
== WB_SYNC_ALL
)
3018 tag_pages_for_writeback(mapping
, index
, end
);
3020 while (!ret
&& wbc
->nr_to_write
> 0) {
3023 * we insert one extent at a time. So we need
3024 * credit needed for single extent allocation.
3025 * journalled mode is currently not supported
3028 BUG_ON(ext4_should_journal_data(inode
));
3029 needed_blocks
= ext4_da_writepages_trans_blocks(inode
);
3031 /* start a new transaction*/
3032 handle
= ext4_journal_start(inode
, needed_blocks
);
3033 if (IS_ERR(handle
)) {
3034 ret
= PTR_ERR(handle
);
3035 ext4_msg(inode
->i_sb
, KERN_CRIT
, "%s: jbd2_start: "
3036 "%ld pages, ino %lu; err %d", __func__
,
3037 wbc
->nr_to_write
, inode
->i_ino
, ret
);
3038 goto out_writepages
;
3042 * Now call write_cache_pages_da() to find the next
3043 * contiguous region of logical blocks that need
3044 * blocks to be allocated by ext4 and submit them.
3052 mpd
.pages_written
= 0;
3054 ret
= write_cache_pages_da(mapping
, wbc
, &mpd
, &done_index
);
3056 * If we have a contiguous extent of pages and we
3057 * haven't done the I/O yet, map the blocks and submit
3060 if (!mpd
.io_done
&& mpd
.next_page
!= mpd
.first_page
) {
3061 mpage_da_map_and_submit(&mpd
);
3062 ret
= MPAGE_DA_EXTENT_TAIL
;
3064 trace_ext4_da_write_pages(inode
, &mpd
);
3065 wbc
->nr_to_write
-= mpd
.pages_written
;
3067 ext4_journal_stop(handle
);
3069 if ((mpd
.retval
== -ENOSPC
) && sbi
->s_journal
) {
3070 /* commit the transaction which would
3071 * free blocks released in the transaction
3074 jbd2_journal_force_commit_nested(sbi
->s_journal
);
3075 wbc
->pages_skipped
= pages_skipped
;
3077 } else if (ret
== MPAGE_DA_EXTENT_TAIL
) {
3079 * got one extent now try with
3082 pages_written
+= mpd
.pages_written
;
3083 wbc
->pages_skipped
= pages_skipped
;
3086 } else if (wbc
->nr_to_write
)
3088 * There is no more writeout needed
3089 * or we requested for a noblocking writeout
3090 * and we found the device congested
3094 if (!io_done
&& !cycled
) {
3097 wbc
->range_start
= index
<< PAGE_CACHE_SHIFT
;
3098 wbc
->range_end
= mapping
->writeback_index
- 1;
3101 if (pages_skipped
!= wbc
->pages_skipped
)
3102 ext4_msg(inode
->i_sb
, KERN_CRIT
,
3103 "This should not happen leaving %s "
3104 "with nr_to_write = %ld ret = %d",
3105 __func__
, wbc
->nr_to_write
, ret
);
3108 wbc
->range_cyclic
= range_cyclic
;
3109 if (wbc
->range_cyclic
|| (range_whole
&& wbc
->nr_to_write
> 0))
3111 * set the writeback_index so that range_cyclic
3112 * mode will write it back later
3114 mapping
->writeback_index
= done_index
;
3117 wbc
->nr_to_write
-= nr_to_writebump
;
3118 wbc
->range_start
= range_start
;
3119 trace_ext4_da_writepages_result(inode
, wbc
, ret
, pages_written
);
3123 #define FALL_BACK_TO_NONDELALLOC 1
3124 static int ext4_nonda_switch(struct super_block
*sb
)
3126 s64 free_blocks
, dirty_blocks
;
3127 struct ext4_sb_info
*sbi
= EXT4_SB(sb
);
3130 * switch to non delalloc mode if we are running low
3131 * on free block. The free block accounting via percpu
3132 * counters can get slightly wrong with percpu_counter_batch getting
3133 * accumulated on each CPU without updating global counters
3134 * Delalloc need an accurate free block accounting. So switch
3135 * to non delalloc when we are near to error range.
3137 free_blocks
= percpu_counter_read_positive(&sbi
->s_freeblocks_counter
);
3138 dirty_blocks
= percpu_counter_read_positive(&sbi
->s_dirtyblocks_counter
);
3139 if (2 * free_blocks
< 3 * dirty_blocks
||
3140 free_blocks
< (dirty_blocks
+ EXT4_FREEBLOCKS_WATERMARK
)) {
3142 * free block count is less than 150% of dirty blocks
3143 * or free blocks is less than watermark
3148 * Even if we don't switch but are nearing capacity,
3149 * start pushing delalloc when 1/2 of free blocks are dirty.
3151 if (free_blocks
< 2 * dirty_blocks
)
3152 writeback_inodes_sb_if_idle(sb
);
3157 static int ext4_da_write_begin(struct file
*file
, struct address_space
*mapping
,
3158 loff_t pos
, unsigned len
, unsigned flags
,
3159 struct page
**pagep
, void **fsdata
)
3161 int ret
, retries
= 0;
3164 struct inode
*inode
= mapping
->host
;
3167 index
= pos
>> PAGE_CACHE_SHIFT
;
3169 if (ext4_nonda_switch(inode
->i_sb
)) {
3170 *fsdata
= (void *)FALL_BACK_TO_NONDELALLOC
;
3171 return ext4_write_begin(file
, mapping
, pos
,
3172 len
, flags
, pagep
, fsdata
);
3174 *fsdata
= (void *)0;
3175 trace_ext4_da_write_begin(inode
, pos
, len
, flags
);
3178 * With delayed allocation, we don't log the i_disksize update
3179 * if there is delayed block allocation. But we still need
3180 * to journalling the i_disksize update if writes to the end
3181 * of file which has an already mapped buffer.
3183 handle
= ext4_journal_start(inode
, 1);
3184 if (IS_ERR(handle
)) {
3185 ret
= PTR_ERR(handle
);
3188 /* We cannot recurse into the filesystem as the transaction is already
3190 flags
|= AOP_FLAG_NOFS
;
3192 page
= grab_cache_page_write_begin(mapping
, index
, flags
);
3194 ext4_journal_stop(handle
);
3200 ret
= __block_write_begin(page
, pos
, len
, ext4_da_get_block_prep
);
3203 ext4_journal_stop(handle
);
3204 page_cache_release(page
);
3206 * block_write_begin may have instantiated a few blocks
3207 * outside i_size. Trim these off again. Don't need
3208 * i_size_read because we hold i_mutex.
3210 if (pos
+ len
> inode
->i_size
)
3211 ext4_truncate_failed_write(inode
);
3214 if (ret
== -ENOSPC
&& ext4_should_retry_alloc(inode
->i_sb
, &retries
))
3221 * Check if we should update i_disksize
3222 * when write to the end of file but not require block allocation
3224 static int ext4_da_should_update_i_disksize(struct page
*page
,
3225 unsigned long offset
)
3227 struct buffer_head
*bh
;
3228 struct inode
*inode
= page
->mapping
->host
;
3232 bh
= page_buffers(page
);
3233 idx
= offset
>> inode
->i_blkbits
;
3235 for (i
= 0; i
< idx
; i
++)
3236 bh
= bh
->b_this_page
;
3238 if (!buffer_mapped(bh
) || (buffer_delay(bh
)) || buffer_unwritten(bh
))
3243 static int ext4_da_write_end(struct file
*file
,
3244 struct address_space
*mapping
,
3245 loff_t pos
, unsigned len
, unsigned copied
,
3246 struct page
*page
, void *fsdata
)
3248 struct inode
*inode
= mapping
->host
;
3250 handle_t
*handle
= ext4_journal_current_handle();
3252 unsigned long start
, end
;
3253 int write_mode
= (int)(unsigned long)fsdata
;
3255 if (write_mode
== FALL_BACK_TO_NONDELALLOC
) {
3256 if (ext4_should_order_data(inode
)) {
3257 return ext4_ordered_write_end(file
, mapping
, pos
,
3258 len
, copied
, page
, fsdata
);
3259 } else if (ext4_should_writeback_data(inode
)) {
3260 return ext4_writeback_write_end(file
, mapping
, pos
,
3261 len
, copied
, page
, fsdata
);
3267 trace_ext4_da_write_end(inode
, pos
, len
, copied
);
3268 start
= pos
& (PAGE_CACHE_SIZE
- 1);
3269 end
= start
+ copied
- 1;
3272 * generic_write_end() will run mark_inode_dirty() if i_size
3273 * changes. So let's piggyback the i_disksize mark_inode_dirty
3277 new_i_size
= pos
+ copied
;
3278 if (new_i_size
> EXT4_I(inode
)->i_disksize
) {
3279 if (ext4_da_should_update_i_disksize(page
, end
)) {
3280 down_write(&EXT4_I(inode
)->i_data_sem
);
3281 if (new_i_size
> EXT4_I(inode
)->i_disksize
) {
3283 * Updating i_disksize when extending file
3284 * without needing block allocation
3286 if (ext4_should_order_data(inode
))
3287 ret
= ext4_jbd2_file_inode(handle
,
3290 EXT4_I(inode
)->i_disksize
= new_i_size
;
3292 up_write(&EXT4_I(inode
)->i_data_sem
);
3293 /* We need to mark inode dirty even if
3294 * new_i_size is less that inode->i_size
3295 * bu greater than i_disksize.(hint delalloc)
3297 ext4_mark_inode_dirty(handle
, inode
);
3300 ret2
= generic_write_end(file
, mapping
, pos
, len
, copied
,
3305 ret2
= ext4_journal_stop(handle
);
3309 return ret
? ret
: copied
;
3312 static void ext4_da_invalidatepage(struct page
*page
, unsigned long offset
)
3315 * Drop reserved blocks
3317 BUG_ON(!PageLocked(page
));
3318 if (!page_has_buffers(page
))
3321 ext4_da_page_release_reservation(page
, offset
);
3324 ext4_invalidatepage(page
, offset
);
3330 * Force all delayed allocation blocks to be allocated for a given inode.
3332 int ext4_alloc_da_blocks(struct inode
*inode
)
3334 trace_ext4_alloc_da_blocks(inode
);
3336 if (!EXT4_I(inode
)->i_reserved_data_blocks
&&
3337 !EXT4_I(inode
)->i_reserved_meta_blocks
)
3341 * We do something simple for now. The filemap_flush() will
3342 * also start triggering a write of the data blocks, which is
3343 * not strictly speaking necessary (and for users of
3344 * laptop_mode, not even desirable). However, to do otherwise
3345 * would require replicating code paths in:
3347 * ext4_da_writepages() ->
3348 * write_cache_pages() ---> (via passed in callback function)
3349 * __mpage_da_writepage() -->
3350 * mpage_add_bh_to_extent()
3351 * mpage_da_map_blocks()
3353 * The problem is that write_cache_pages(), located in
3354 * mm/page-writeback.c, marks pages clean in preparation for
3355 * doing I/O, which is not desirable if we're not planning on
3358 * We could call write_cache_pages(), and then redirty all of
3359 * the pages by calling redirty_page_for_writepage() but that
3360 * would be ugly in the extreme. So instead we would need to
3361 * replicate parts of the code in the above functions,
3362 * simplifying them becuase we wouldn't actually intend to
3363 * write out the pages, but rather only collect contiguous
3364 * logical block extents, call the multi-block allocator, and
3365 * then update the buffer heads with the block allocations.
3367 * For now, though, we'll cheat by calling filemap_flush(),
3368 * which will map the blocks, and start the I/O, but not
3369 * actually wait for the I/O to complete.
3371 return filemap_flush(inode
->i_mapping
);
3375 * bmap() is special. It gets used by applications such as lilo and by
3376 * the swapper to find the on-disk block of a specific piece of data.
3378 * Naturally, this is dangerous if the block concerned is still in the
3379 * journal. If somebody makes a swapfile on an ext4 data-journaling
3380 * filesystem and enables swap, then they may get a nasty shock when the
3381 * data getting swapped to that swapfile suddenly gets overwritten by
3382 * the original zero's written out previously to the journal and
3383 * awaiting writeback in the kernel's buffer cache.
3385 * So, if we see any bmap calls here on a modified, data-journaled file,
3386 * take extra steps to flush any blocks which might be in the cache.
3388 static sector_t
ext4_bmap(struct address_space
*mapping
, sector_t block
)
3390 struct inode
*inode
= mapping
->host
;
3394 if (mapping_tagged(mapping
, PAGECACHE_TAG_DIRTY
) &&
3395 test_opt(inode
->i_sb
, DELALLOC
)) {
3397 * With delalloc we want to sync the file
3398 * so that we can make sure we allocate
3401 filemap_write_and_wait(mapping
);
3404 if (EXT4_JOURNAL(inode
) &&
3405 ext4_test_inode_state(inode
, EXT4_STATE_JDATA
)) {
3407 * This is a REALLY heavyweight approach, but the use of
3408 * bmap on dirty files is expected to be extremely rare:
3409 * only if we run lilo or swapon on a freshly made file
3410 * do we expect this to happen.
3412 * (bmap requires CAP_SYS_RAWIO so this does not
3413 * represent an unprivileged user DOS attack --- we'd be
3414 * in trouble if mortal users could trigger this path at
3417 * NB. EXT4_STATE_JDATA is not set on files other than
3418 * regular files. If somebody wants to bmap a directory
3419 * or symlink and gets confused because the buffer
3420 * hasn't yet been flushed to disk, they deserve
3421 * everything they get.
3424 ext4_clear_inode_state(inode
, EXT4_STATE_JDATA
);
3425 journal
= EXT4_JOURNAL(inode
);
3426 jbd2_journal_lock_updates(journal
);
3427 err
= jbd2_journal_flush(journal
);
3428 jbd2_journal_unlock_updates(journal
);
3434 return generic_block_bmap(mapping
, block
, ext4_get_block
);
3437 static int ext4_readpage(struct file
*file
, struct page
*page
)
3439 return mpage_readpage(page
, ext4_get_block
);
3443 ext4_readpages(struct file
*file
, struct address_space
*mapping
,
3444 struct list_head
*pages
, unsigned nr_pages
)
3446 return mpage_readpages(mapping
, pages
, nr_pages
, ext4_get_block
);
3449 static void ext4_invalidatepage_free_endio(struct page
*page
, unsigned long offset
)
3451 struct buffer_head
*head
, *bh
;
3452 unsigned int curr_off
= 0;
3454 if (!page_has_buffers(page
))
3456 head
= bh
= page_buffers(page
);
3458 if (offset
<= curr_off
&& test_clear_buffer_uninit(bh
)
3460 ext4_free_io_end(bh
->b_private
);
3461 bh
->b_private
= NULL
;
3462 bh
->b_end_io
= NULL
;
3464 curr_off
= curr_off
+ bh
->b_size
;
3465 bh
= bh
->b_this_page
;
3466 } while (bh
!= head
);
3469 static void ext4_invalidatepage(struct page
*page
, unsigned long offset
)
3471 journal_t
*journal
= EXT4_JOURNAL(page
->mapping
->host
);
3474 * free any io_end structure allocated for buffers to be discarded
3476 if (ext4_should_dioread_nolock(page
->mapping
->host
))
3477 ext4_invalidatepage_free_endio(page
, offset
);
3479 * If it's a full truncate we just forget about the pending dirtying
3482 ClearPageChecked(page
);
3485 jbd2_journal_invalidatepage(journal
, page
, offset
);
3487 block_invalidatepage(page
, offset
);
3490 static int ext4_releasepage(struct page
*page
, gfp_t wait
)
3492 journal_t
*journal
= EXT4_JOURNAL(page
->mapping
->host
);
3494 WARN_ON(PageChecked(page
));
3495 if (!page_has_buffers(page
))
3498 return jbd2_journal_try_to_free_buffers(journal
, page
, wait
);
3500 return try_to_free_buffers(page
);
3504 * O_DIRECT for ext3 (or indirect map) based files
3506 * If the O_DIRECT write will extend the file then add this inode to the
3507 * orphan list. So recovery will truncate it back to the original size
3508 * if the machine crashes during the write.
3510 * If the O_DIRECT write is intantiating holes inside i_size and the machine
3511 * crashes then stale disk data _may_ be exposed inside the file. But current
3512 * VFS code falls back into buffered path in that case so we are safe.
3514 static ssize_t
ext4_ind_direct_IO(int rw
, struct kiocb
*iocb
,
3515 const struct iovec
*iov
, loff_t offset
,
3516 unsigned long nr_segs
)
3518 struct file
*file
= iocb
->ki_filp
;
3519 struct inode
*inode
= file
->f_mapping
->host
;
3520 struct ext4_inode_info
*ei
= EXT4_I(inode
);
3524 size_t count
= iov_length(iov
, nr_segs
);
3528 loff_t final_size
= offset
+ count
;
3530 if (final_size
> inode
->i_size
) {
3531 /* Credits for sb + inode write */
3532 handle
= ext4_journal_start(inode
, 2);
3533 if (IS_ERR(handle
)) {
3534 ret
= PTR_ERR(handle
);
3537 ret
= ext4_orphan_add(handle
, inode
);
3539 ext4_journal_stop(handle
);
3543 ei
->i_disksize
= inode
->i_size
;
3544 ext4_journal_stop(handle
);
3549 if (rw
== READ
&& ext4_should_dioread_nolock(inode
))
3550 ret
= __blockdev_direct_IO(rw
, iocb
, inode
,
3551 inode
->i_sb
->s_bdev
, iov
,
3553 ext4_get_block
, NULL
, NULL
, 0);
3555 ret
= blockdev_direct_IO(rw
, iocb
, inode
,
3556 inode
->i_sb
->s_bdev
, iov
,
3558 ext4_get_block
, NULL
);
3560 if (unlikely((rw
& WRITE
) && ret
< 0)) {
3561 loff_t isize
= i_size_read(inode
);
3562 loff_t end
= offset
+ iov_length(iov
, nr_segs
);
3565 vmtruncate(inode
, isize
);
3568 if (ret
== -ENOSPC
&& ext4_should_retry_alloc(inode
->i_sb
, &retries
))
3574 /* Credits for sb + inode write */
3575 handle
= ext4_journal_start(inode
, 2);
3576 if (IS_ERR(handle
)) {
3577 /* This is really bad luck. We've written the data
3578 * but cannot extend i_size. Bail out and pretend
3579 * the write failed... */
3580 ret
= PTR_ERR(handle
);
3582 ext4_orphan_del(NULL
, inode
);
3587 ext4_orphan_del(handle
, inode
);
3589 loff_t end
= offset
+ ret
;
3590 if (end
> inode
->i_size
) {
3591 ei
->i_disksize
= end
;
3592 i_size_write(inode
, end
);
3594 * We're going to return a positive `ret'
3595 * here due to non-zero-length I/O, so there's
3596 * no way of reporting error returns from
3597 * ext4_mark_inode_dirty() to userspace. So
3600 ext4_mark_inode_dirty(handle
, inode
);
3603 err
= ext4_journal_stop(handle
);
3612 * ext4_get_block used when preparing for a DIO write or buffer write.
3613 * We allocate an uinitialized extent if blocks haven't been allocated.
3614 * The extent will be converted to initialized after the IO is complete.
3616 static int ext4_get_block_write(struct inode
*inode
, sector_t iblock
,
3617 struct buffer_head
*bh_result
, int create
)
3619 ext4_debug("ext4_get_block_write: inode %lu, create flag %d\n",
3620 inode
->i_ino
, create
);
3621 return _ext4_get_block(inode
, iblock
, bh_result
,
3622 EXT4_GET_BLOCKS_IO_CREATE_EXT
);
3625 static void ext4_end_io_dio(struct kiocb
*iocb
, loff_t offset
,
3626 ssize_t size
, void *private, int ret
,
3629 ext4_io_end_t
*io_end
= iocb
->private;
3630 struct workqueue_struct
*wq
;
3631 unsigned long flags
;
3632 struct ext4_inode_info
*ei
;
3634 /* if not async direct IO or dio with 0 bytes write, just return */
3635 if (!io_end
|| !size
)
3638 ext_debug("ext4_end_io_dio(): io_end 0x%p"
3639 "for inode %lu, iocb 0x%p, offset %llu, size %llu\n",
3640 iocb
->private, io_end
->inode
->i_ino
, iocb
, offset
,
3643 /* if not aio dio with unwritten extents, just free io and return */
3644 if (!(io_end
->flag
& EXT4_IO_END_UNWRITTEN
)) {
3645 ext4_free_io_end(io_end
);
3646 iocb
->private = NULL
;
3649 aio_complete(iocb
, ret
, 0);
3653 io_end
->offset
= offset
;
3654 io_end
->size
= size
;
3656 io_end
->iocb
= iocb
;
3657 io_end
->result
= ret
;
3659 wq
= EXT4_SB(io_end
->inode
->i_sb
)->dio_unwritten_wq
;
3661 /* Add the io_end to per-inode completed aio dio list*/
3662 ei
= EXT4_I(io_end
->inode
);
3663 spin_lock_irqsave(&ei
->i_completed_io_lock
, flags
);
3664 list_add_tail(&io_end
->list
, &ei
->i_completed_io_list
);
3665 spin_unlock_irqrestore(&ei
->i_completed_io_lock
, flags
);
3667 /* queue the work to convert unwritten extents to written */
3668 queue_work(wq
, &io_end
->work
);
3669 iocb
->private = NULL
;
3672 static void ext4_end_io_buffer_write(struct buffer_head
*bh
, int uptodate
)
3674 ext4_io_end_t
*io_end
= bh
->b_private
;
3675 struct workqueue_struct
*wq
;
3676 struct inode
*inode
;
3677 unsigned long flags
;
3679 if (!test_clear_buffer_uninit(bh
) || !io_end
)
3682 if (!(io_end
->inode
->i_sb
->s_flags
& MS_ACTIVE
)) {
3683 printk("sb umounted, discard end_io request for inode %lu\n",
3684 io_end
->inode
->i_ino
);
3685 ext4_free_io_end(io_end
);
3689 io_end
->flag
= EXT4_IO_END_UNWRITTEN
;
3690 inode
= io_end
->inode
;
3692 /* Add the io_end to per-inode completed io list*/
3693 spin_lock_irqsave(&EXT4_I(inode
)->i_completed_io_lock
, flags
);
3694 list_add_tail(&io_end
->list
, &EXT4_I(inode
)->i_completed_io_list
);
3695 spin_unlock_irqrestore(&EXT4_I(inode
)->i_completed_io_lock
, flags
);
3697 wq
= EXT4_SB(inode
->i_sb
)->dio_unwritten_wq
;
3698 /* queue the work to convert unwritten extents to written */
3699 queue_work(wq
, &io_end
->work
);
3701 bh
->b_private
= NULL
;
3702 bh
->b_end_io
= NULL
;
3703 clear_buffer_uninit(bh
);
3704 end_buffer_async_write(bh
, uptodate
);
3707 static int ext4_set_bh_endio(struct buffer_head
*bh
, struct inode
*inode
)
3709 ext4_io_end_t
*io_end
;
3710 struct page
*page
= bh
->b_page
;
3711 loff_t offset
= (sector_t
)page
->index
<< PAGE_CACHE_SHIFT
;
3712 size_t size
= bh
->b_size
;
3715 io_end
= ext4_init_io_end(inode
, GFP_ATOMIC
);
3717 pr_warn_ratelimited("%s: allocation fail\n", __func__
);
3721 io_end
->offset
= offset
;
3722 io_end
->size
= size
;
3724 * We need to hold a reference to the page to make sure it
3725 * doesn't get evicted before ext4_end_io_work() has a chance
3726 * to convert the extent from written to unwritten.
3728 io_end
->page
= page
;
3729 get_page(io_end
->page
);
3731 bh
->b_private
= io_end
;
3732 bh
->b_end_io
= ext4_end_io_buffer_write
;
3737 * For ext4 extent files, ext4 will do direct-io write to holes,
3738 * preallocated extents, and those write extend the file, no need to
3739 * fall back to buffered IO.
3741 * For holes, we fallocate those blocks, mark them as uninitialized
3742 * If those blocks were preallocated, we mark sure they are splited, but
3743 * still keep the range to write as uninitialized.
3745 * The unwrritten extents will be converted to written when DIO is completed.
3746 * For async direct IO, since the IO may still pending when return, we
3747 * set up an end_io call back function, which will do the convertion
3748 * when async direct IO completed.
3750 * If the O_DIRECT write will extend the file then add this inode to the
3751 * orphan list. So recovery will truncate it back to the original size
3752 * if the machine crashes during the write.
3755 static ssize_t
ext4_ext_direct_IO(int rw
, struct kiocb
*iocb
,
3756 const struct iovec
*iov
, loff_t offset
,
3757 unsigned long nr_segs
)
3759 struct file
*file
= iocb
->ki_filp
;
3760 struct inode
*inode
= file
->f_mapping
->host
;
3762 size_t count
= iov_length(iov
, nr_segs
);
3764 loff_t final_size
= offset
+ count
;
3765 if (rw
== WRITE
&& final_size
<= inode
->i_size
) {
3767 * We could direct write to holes and fallocate.
3769 * Allocated blocks to fill the hole are marked as uninitialized
3770 * to prevent paralel buffered read to expose the stale data
3771 * before DIO complete the data IO.
3773 * As to previously fallocated extents, ext4 get_block
3774 * will just simply mark the buffer mapped but still
3775 * keep the extents uninitialized.
3777 * for non AIO case, we will convert those unwritten extents
3778 * to written after return back from blockdev_direct_IO.
3780 * for async DIO, the conversion needs to be defered when
3781 * the IO is completed. The ext4 end_io callback function
3782 * will be called to take care of the conversion work.
3783 * Here for async case, we allocate an io_end structure to
3786 iocb
->private = NULL
;
3787 EXT4_I(inode
)->cur_aio_dio
= NULL
;
3788 if (!is_sync_kiocb(iocb
)) {
3789 iocb
->private = ext4_init_io_end(inode
, GFP_NOFS
);
3793 * we save the io structure for current async
3794 * direct IO, so that later ext4_map_blocks()
3795 * could flag the io structure whether there
3796 * is a unwritten extents needs to be converted
3797 * when IO is completed.
3799 EXT4_I(inode
)->cur_aio_dio
= iocb
->private;
3802 ret
= blockdev_direct_IO(rw
, iocb
, inode
,
3803 inode
->i_sb
->s_bdev
, iov
,
3805 ext4_get_block_write
,
3808 EXT4_I(inode
)->cur_aio_dio
= NULL
;
3810 * The io_end structure takes a reference to the inode,
3811 * that structure needs to be destroyed and the
3812 * reference to the inode need to be dropped, when IO is
3813 * complete, even with 0 byte write, or failed.
3815 * In the successful AIO DIO case, the io_end structure will be
3816 * desctroyed and the reference to the inode will be dropped
3817 * after the end_io call back function is called.
3819 * In the case there is 0 byte write, or error case, since
3820 * VFS direct IO won't invoke the end_io call back function,
3821 * we need to free the end_io structure here.
3823 if (ret
!= -EIOCBQUEUED
&& ret
<= 0 && iocb
->private) {
3824 ext4_free_io_end(iocb
->private);
3825 iocb
->private = NULL
;
3826 } else if (ret
> 0 && ext4_test_inode_state(inode
,
3827 EXT4_STATE_DIO_UNWRITTEN
)) {
3830 * for non AIO case, since the IO is already
3831 * completed, we could do the convertion right here
3833 err
= ext4_convert_unwritten_extents(inode
,
3837 ext4_clear_inode_state(inode
, EXT4_STATE_DIO_UNWRITTEN
);
3842 /* for write the the end of file case, we fall back to old way */
3843 return ext4_ind_direct_IO(rw
, iocb
, iov
, offset
, nr_segs
);
3846 static ssize_t
ext4_direct_IO(int rw
, struct kiocb
*iocb
,
3847 const struct iovec
*iov
, loff_t offset
,
3848 unsigned long nr_segs
)
3850 struct file
*file
= iocb
->ki_filp
;
3851 struct inode
*inode
= file
->f_mapping
->host
;
3853 if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
))
3854 return ext4_ext_direct_IO(rw
, iocb
, iov
, offset
, nr_segs
);
3856 return ext4_ind_direct_IO(rw
, iocb
, iov
, offset
, nr_segs
);
3860 * Pages can be marked dirty completely asynchronously from ext4's journalling
3861 * activity. By filemap_sync_pte(), try_to_unmap_one(), etc. We cannot do
3862 * much here because ->set_page_dirty is called under VFS locks. The page is
3863 * not necessarily locked.
3865 * We cannot just dirty the page and leave attached buffers clean, because the
3866 * buffers' dirty state is "definitive". We cannot just set the buffers dirty
3867 * or jbddirty because all the journalling code will explode.
3869 * So what we do is to mark the page "pending dirty" and next time writepage
3870 * is called, propagate that into the buffers appropriately.
3872 static int ext4_journalled_set_page_dirty(struct page
*page
)
3874 SetPageChecked(page
);
3875 return __set_page_dirty_nobuffers(page
);
3878 static const struct address_space_operations ext4_ordered_aops
= {
3879 .readpage
= ext4_readpage
,
3880 .readpages
= ext4_readpages
,
3881 .writepage
= ext4_writepage
,
3882 .sync_page
= block_sync_page
,
3883 .write_begin
= ext4_write_begin
,
3884 .write_end
= ext4_ordered_write_end
,
3886 .invalidatepage
= ext4_invalidatepage
,
3887 .releasepage
= ext4_releasepage
,
3888 .direct_IO
= ext4_direct_IO
,
3889 .migratepage
= buffer_migrate_page
,
3890 .is_partially_uptodate
= block_is_partially_uptodate
,
3891 .error_remove_page
= generic_error_remove_page
,
3894 static const struct address_space_operations ext4_writeback_aops
= {
3895 .readpage
= ext4_readpage
,
3896 .readpages
= ext4_readpages
,
3897 .writepage
= ext4_writepage
,
3898 .sync_page
= block_sync_page
,
3899 .write_begin
= ext4_write_begin
,
3900 .write_end
= ext4_writeback_write_end
,
3902 .invalidatepage
= ext4_invalidatepage
,
3903 .releasepage
= ext4_releasepage
,
3904 .direct_IO
= ext4_direct_IO
,
3905 .migratepage
= buffer_migrate_page
,
3906 .is_partially_uptodate
= block_is_partially_uptodate
,
3907 .error_remove_page
= generic_error_remove_page
,
3910 static const struct address_space_operations ext4_journalled_aops
= {
3911 .readpage
= ext4_readpage
,
3912 .readpages
= ext4_readpages
,
3913 .writepage
= ext4_writepage
,
3914 .sync_page
= block_sync_page
,
3915 .write_begin
= ext4_write_begin
,
3916 .write_end
= ext4_journalled_write_end
,
3917 .set_page_dirty
= ext4_journalled_set_page_dirty
,
3919 .invalidatepage
= ext4_invalidatepage
,
3920 .releasepage
= ext4_releasepage
,
3921 .is_partially_uptodate
= block_is_partially_uptodate
,
3922 .error_remove_page
= generic_error_remove_page
,
3925 static const struct address_space_operations ext4_da_aops
= {
3926 .readpage
= ext4_readpage
,
3927 .readpages
= ext4_readpages
,
3928 .writepage
= ext4_writepage
,
3929 .writepages
= ext4_da_writepages
,
3930 .sync_page
= block_sync_page
,
3931 .write_begin
= ext4_da_write_begin
,
3932 .write_end
= ext4_da_write_end
,
3934 .invalidatepage
= ext4_da_invalidatepage
,
3935 .releasepage
= ext4_releasepage
,
3936 .direct_IO
= ext4_direct_IO
,
3937 .migratepage
= buffer_migrate_page
,
3938 .is_partially_uptodate
= block_is_partially_uptodate
,
3939 .error_remove_page
= generic_error_remove_page
,
3942 void ext4_set_aops(struct inode
*inode
)
3944 if (ext4_should_order_data(inode
) &&
3945 test_opt(inode
->i_sb
, DELALLOC
))
3946 inode
->i_mapping
->a_ops
= &ext4_da_aops
;
3947 else if (ext4_should_order_data(inode
))
3948 inode
->i_mapping
->a_ops
= &ext4_ordered_aops
;
3949 else if (ext4_should_writeback_data(inode
) &&
3950 test_opt(inode
->i_sb
, DELALLOC
))
3951 inode
->i_mapping
->a_ops
= &ext4_da_aops
;
3952 else if (ext4_should_writeback_data(inode
))
3953 inode
->i_mapping
->a_ops
= &ext4_writeback_aops
;
3955 inode
->i_mapping
->a_ops
= &ext4_journalled_aops
;
3959 * ext4_block_truncate_page() zeroes out a mapping from file offset `from'
3960 * up to the end of the block which corresponds to `from'.
3961 * This required during truncate. We need to physically zero the tail end
3962 * of that block so it doesn't yield old data if the file is later grown.
3964 int ext4_block_truncate_page(handle_t
*handle
,
3965 struct address_space
*mapping
, loff_t from
)
3967 ext4_fsblk_t index
= from
>> PAGE_CACHE_SHIFT
;
3968 unsigned offset
= from
& (PAGE_CACHE_SIZE
-1);
3969 unsigned blocksize
, length
, pos
;
3971 struct inode
*inode
= mapping
->host
;
3972 struct buffer_head
*bh
;
3976 page
= find_or_create_page(mapping
, from
>> PAGE_CACHE_SHIFT
,
3977 mapping_gfp_mask(mapping
) & ~__GFP_FS
);
3981 blocksize
= inode
->i_sb
->s_blocksize
;
3982 length
= blocksize
- (offset
& (blocksize
- 1));
3983 iblock
= index
<< (PAGE_CACHE_SHIFT
- inode
->i_sb
->s_blocksize_bits
);
3985 if (!page_has_buffers(page
))
3986 create_empty_buffers(page
, blocksize
, 0);
3988 /* Find the buffer that contains "offset" */
3989 bh
= page_buffers(page
);
3991 while (offset
>= pos
) {
3992 bh
= bh
->b_this_page
;
3998 if (buffer_freed(bh
)) {
3999 BUFFER_TRACE(bh
, "freed: skip");
4003 if (!buffer_mapped(bh
)) {
4004 BUFFER_TRACE(bh
, "unmapped");
4005 ext4_get_block(inode
, iblock
, bh
, 0);
4006 /* unmapped? It's a hole - nothing to do */
4007 if (!buffer_mapped(bh
)) {
4008 BUFFER_TRACE(bh
, "still unmapped");
4013 /* Ok, it's mapped. Make sure it's up-to-date */
4014 if (PageUptodate(page
))
4015 set_buffer_uptodate(bh
);
4017 if (!buffer_uptodate(bh
)) {
4019 ll_rw_block(READ
, 1, &bh
);
4021 /* Uhhuh. Read error. Complain and punt. */
4022 if (!buffer_uptodate(bh
))
4026 if (ext4_should_journal_data(inode
)) {
4027 BUFFER_TRACE(bh
, "get write access");
4028 err
= ext4_journal_get_write_access(handle
, bh
);
4033 zero_user(page
, offset
, length
);
4035 BUFFER_TRACE(bh
, "zeroed end of block");
4038 if (ext4_should_journal_data(inode
)) {
4039 err
= ext4_handle_dirty_metadata(handle
, inode
, bh
);
4041 if (ext4_should_order_data(inode
) && EXT4_I(inode
)->jinode
)
4042 err
= ext4_jbd2_file_inode(handle
, inode
);
4043 mark_buffer_dirty(bh
);
4048 page_cache_release(page
);
4053 * Probably it should be a library function... search for first non-zero word
4054 * or memcmp with zero_page, whatever is better for particular architecture.
4057 static inline int all_zeroes(__le32
*p
, __le32
*q
)
4066 * ext4_find_shared - find the indirect blocks for partial truncation.
4067 * @inode: inode in question
4068 * @depth: depth of the affected branch
4069 * @offsets: offsets of pointers in that branch (see ext4_block_to_path)
4070 * @chain: place to store the pointers to partial indirect blocks
4071 * @top: place to the (detached) top of branch
4073 * This is a helper function used by ext4_truncate().
4075 * When we do truncate() we may have to clean the ends of several
4076 * indirect blocks but leave the blocks themselves alive. Block is
4077 * partially truncated if some data below the new i_size is refered
4078 * from it (and it is on the path to the first completely truncated
4079 * data block, indeed). We have to free the top of that path along
4080 * with everything to the right of the path. Since no allocation
4081 * past the truncation point is possible until ext4_truncate()
4082 * finishes, we may safely do the latter, but top of branch may
4083 * require special attention - pageout below the truncation point
4084 * might try to populate it.
4086 * We atomically detach the top of branch from the tree, store the
4087 * block number of its root in *@top, pointers to buffer_heads of
4088 * partially truncated blocks - in @chain[].bh and pointers to
4089 * their last elements that should not be removed - in
4090 * @chain[].p. Return value is the pointer to last filled element
4093 * The work left to caller to do the actual freeing of subtrees:
4094 * a) free the subtree starting from *@top
4095 * b) free the subtrees whose roots are stored in
4096 * (@chain[i].p+1 .. end of @chain[i].bh->b_data)
4097 * c) free the subtrees growing from the inode past the @chain[0].
4098 * (no partially truncated stuff there). */
4100 static Indirect
*ext4_find_shared(struct inode
*inode
, int depth
,
4101 ext4_lblk_t offsets
[4], Indirect chain
[4],
4104 Indirect
*partial
, *p
;
4108 /* Make k index the deepest non-null offset + 1 */
4109 for (k
= depth
; k
> 1 && !offsets
[k
-1]; k
--)
4111 partial
= ext4_get_branch(inode
, k
, offsets
, chain
, &err
);
4112 /* Writer: pointers */
4114 partial
= chain
+ k
-1;
4116 * If the branch acquired continuation since we've looked at it -
4117 * fine, it should all survive and (new) top doesn't belong to us.
4119 if (!partial
->key
&& *partial
->p
)
4122 for (p
= partial
; (p
> chain
) && all_zeroes((__le32
*) p
->bh
->b_data
, p
->p
); p
--)
4125 * OK, we've found the last block that must survive. The rest of our
4126 * branch should be detached before unlocking. However, if that rest
4127 * of branch is all ours and does not grow immediately from the inode
4128 * it's easier to cheat and just decrement partial->p.
4130 if (p
== chain
+ k
- 1 && p
> chain
) {
4134 /* Nope, don't do this in ext4. Must leave the tree intact */
4141 while (partial
> p
) {
4142 brelse(partial
->bh
);
4150 * Zero a number of block pointers in either an inode or an indirect block.
4151 * If we restart the transaction we must again get write access to the
4152 * indirect block for further modification.
4154 * We release `count' blocks on disk, but (last - first) may be greater
4155 * than `count' because there can be holes in there.
4157 static int ext4_clear_blocks(handle_t
*handle
, struct inode
*inode
,
4158 struct buffer_head
*bh
,
4159 ext4_fsblk_t block_to_free
,
4160 unsigned long count
, __le32
*first
,
4164 int flags
= EXT4_FREE_BLOCKS_FORGET
| EXT4_FREE_BLOCKS_VALIDATED
;
4167 if (S_ISDIR(inode
->i_mode
) || S_ISLNK(inode
->i_mode
))
4168 flags
|= EXT4_FREE_BLOCKS_METADATA
;
4170 if (!ext4_data_block_valid(EXT4_SB(inode
->i_sb
), block_to_free
,
4172 EXT4_ERROR_INODE(inode
, "attempt to clear invalid "
4173 "blocks %llu len %lu",
4174 (unsigned long long) block_to_free
, count
);
4178 if (try_to_extend_transaction(handle
, inode
)) {
4180 BUFFER_TRACE(bh
, "call ext4_handle_dirty_metadata");
4181 err
= ext4_handle_dirty_metadata(handle
, inode
, bh
);
4182 if (unlikely(err
)) {
4183 ext4_std_error(inode
->i_sb
, err
);
4187 err
= ext4_mark_inode_dirty(handle
, inode
);
4188 if (unlikely(err
)) {
4189 ext4_std_error(inode
->i_sb
, err
);
4192 err
= ext4_truncate_restart_trans(handle
, inode
,
4193 blocks_for_truncate(inode
));
4194 if (unlikely(err
)) {
4195 ext4_std_error(inode
->i_sb
, err
);
4199 BUFFER_TRACE(bh
, "retaking write access");
4200 ext4_journal_get_write_access(handle
, bh
);
4204 for (p
= first
; p
< last
; p
++)
4207 ext4_free_blocks(handle
, inode
, NULL
, block_to_free
, count
, flags
);
4212 * ext4_free_data - free a list of data blocks
4213 * @handle: handle for this transaction
4214 * @inode: inode we are dealing with
4215 * @this_bh: indirect buffer_head which contains *@first and *@last
4216 * @first: array of block numbers
4217 * @last: points immediately past the end of array
4219 * We are freeing all blocks refered from that array (numbers are stored as
4220 * little-endian 32-bit) and updating @inode->i_blocks appropriately.
4222 * We accumulate contiguous runs of blocks to free. Conveniently, if these
4223 * blocks are contiguous then releasing them at one time will only affect one
4224 * or two bitmap blocks (+ group descriptor(s) and superblock) and we won't
4225 * actually use a lot of journal space.
4227 * @this_bh will be %NULL if @first and @last point into the inode's direct
4230 static void ext4_free_data(handle_t
*handle
, struct inode
*inode
,
4231 struct buffer_head
*this_bh
,
4232 __le32
*first
, __le32
*last
)
4234 ext4_fsblk_t block_to_free
= 0; /* Starting block # of a run */
4235 unsigned long count
= 0; /* Number of blocks in the run */
4236 __le32
*block_to_free_p
= NULL
; /* Pointer into inode/ind
4239 ext4_fsblk_t nr
; /* Current block # */
4240 __le32
*p
; /* Pointer into inode/ind
4241 for current block */
4244 if (this_bh
) { /* For indirect block */
4245 BUFFER_TRACE(this_bh
, "get_write_access");
4246 err
= ext4_journal_get_write_access(handle
, this_bh
);
4247 /* Important: if we can't update the indirect pointers
4248 * to the blocks, we can't free them. */
4253 for (p
= first
; p
< last
; p
++) {
4254 nr
= le32_to_cpu(*p
);
4256 /* accumulate blocks to free if they're contiguous */
4259 block_to_free_p
= p
;
4261 } else if (nr
== block_to_free
+ count
) {
4264 if (ext4_clear_blocks(handle
, inode
, this_bh
,
4265 block_to_free
, count
,
4266 block_to_free_p
, p
))
4269 block_to_free_p
= p
;
4276 ext4_clear_blocks(handle
, inode
, this_bh
, block_to_free
,
4277 count
, block_to_free_p
, p
);
4280 BUFFER_TRACE(this_bh
, "call ext4_handle_dirty_metadata");
4283 * The buffer head should have an attached journal head at this
4284 * point. However, if the data is corrupted and an indirect
4285 * block pointed to itself, it would have been detached when
4286 * the block was cleared. Check for this instead of OOPSing.
4288 if ((EXT4_JOURNAL(inode
) == NULL
) || bh2jh(this_bh
))
4289 ext4_handle_dirty_metadata(handle
, inode
, this_bh
);
4291 EXT4_ERROR_INODE(inode
,
4292 "circular indirect block detected at "
4294 (unsigned long long) this_bh
->b_blocknr
);
4299 * ext4_free_branches - free an array of branches
4300 * @handle: JBD handle for this transaction
4301 * @inode: inode we are dealing with
4302 * @parent_bh: the buffer_head which contains *@first and *@last
4303 * @first: array of block numbers
4304 * @last: pointer immediately past the end of array
4305 * @depth: depth of the branches to free
4307 * We are freeing all blocks refered from these branches (numbers are
4308 * stored as little-endian 32-bit) and updating @inode->i_blocks
4311 static void ext4_free_branches(handle_t
*handle
, struct inode
*inode
,
4312 struct buffer_head
*parent_bh
,
4313 __le32
*first
, __le32
*last
, int depth
)
4318 if (ext4_handle_is_aborted(handle
))
4322 struct buffer_head
*bh
;
4323 int addr_per_block
= EXT4_ADDR_PER_BLOCK(inode
->i_sb
);
4325 while (--p
>= first
) {
4326 nr
= le32_to_cpu(*p
);
4328 continue; /* A hole */
4330 if (!ext4_data_block_valid(EXT4_SB(inode
->i_sb
),
4332 EXT4_ERROR_INODE(inode
,
4333 "invalid indirect mapped "
4334 "block %lu (level %d)",
4335 (unsigned long) nr
, depth
);
4339 /* Go read the buffer for the next level down */
4340 bh
= sb_bread(inode
->i_sb
, nr
);
4343 * A read failure? Report error and clear slot
4347 EXT4_ERROR_INODE_BLOCK(inode
, nr
,
4352 /* This zaps the entire block. Bottom up. */
4353 BUFFER_TRACE(bh
, "free child branches");
4354 ext4_free_branches(handle
, inode
, bh
,
4355 (__le32
*) bh
->b_data
,
4356 (__le32
*) bh
->b_data
+ addr_per_block
,
4361 * Everything below this this pointer has been
4362 * released. Now let this top-of-subtree go.
4364 * We want the freeing of this indirect block to be
4365 * atomic in the journal with the updating of the
4366 * bitmap block which owns it. So make some room in
4369 * We zero the parent pointer *after* freeing its
4370 * pointee in the bitmaps, so if extend_transaction()
4371 * for some reason fails to put the bitmap changes and
4372 * the release into the same transaction, recovery
4373 * will merely complain about releasing a free block,
4374 * rather than leaking blocks.
4376 if (ext4_handle_is_aborted(handle
))
4378 if (try_to_extend_transaction(handle
, inode
)) {
4379 ext4_mark_inode_dirty(handle
, inode
);
4380 ext4_truncate_restart_trans(handle
, inode
,
4381 blocks_for_truncate(inode
));
4385 * The forget flag here is critical because if
4386 * we are journaling (and not doing data
4387 * journaling), we have to make sure a revoke
4388 * record is written to prevent the journal
4389 * replay from overwriting the (former)
4390 * indirect block if it gets reallocated as a
4391 * data block. This must happen in the same
4392 * transaction where the data blocks are
4395 ext4_free_blocks(handle
, inode
, NULL
, nr
, 1,
4396 EXT4_FREE_BLOCKS_METADATA
|
4397 EXT4_FREE_BLOCKS_FORGET
);
4401 * The block which we have just freed is
4402 * pointed to by an indirect block: journal it
4404 BUFFER_TRACE(parent_bh
, "get_write_access");
4405 if (!ext4_journal_get_write_access(handle
,
4408 BUFFER_TRACE(parent_bh
,
4409 "call ext4_handle_dirty_metadata");
4410 ext4_handle_dirty_metadata(handle
,
4417 /* We have reached the bottom of the tree. */
4418 BUFFER_TRACE(parent_bh
, "free data blocks");
4419 ext4_free_data(handle
, inode
, parent_bh
, first
, last
);
4423 int ext4_can_truncate(struct inode
*inode
)
4425 if (IS_APPEND(inode
) || IS_IMMUTABLE(inode
))
4427 if (S_ISREG(inode
->i_mode
))
4429 if (S_ISDIR(inode
->i_mode
))
4431 if (S_ISLNK(inode
->i_mode
))
4432 return !ext4_inode_is_fast_symlink(inode
);
4439 * We block out ext4_get_block() block instantiations across the entire
4440 * transaction, and VFS/VM ensures that ext4_truncate() cannot run
4441 * simultaneously on behalf of the same inode.
4443 * As we work through the truncate and commmit bits of it to the journal there
4444 * is one core, guiding principle: the file's tree must always be consistent on
4445 * disk. We must be able to restart the truncate after a crash.
4447 * The file's tree may be transiently inconsistent in memory (although it
4448 * probably isn't), but whenever we close off and commit a journal transaction,
4449 * the contents of (the filesystem + the journal) must be consistent and
4450 * restartable. It's pretty simple, really: bottom up, right to left (although
4451 * left-to-right works OK too).
4453 * Note that at recovery time, journal replay occurs *before* the restart of
4454 * truncate against the orphan inode list.
4456 * The committed inode has the new, desired i_size (which is the same as
4457 * i_disksize in this case). After a crash, ext4_orphan_cleanup() will see
4458 * that this inode's truncate did not complete and it will again call
4459 * ext4_truncate() to have another go. So there will be instantiated blocks
4460 * to the right of the truncation point in a crashed ext4 filesystem. But
4461 * that's fine - as long as they are linked from the inode, the post-crash
4462 * ext4_truncate() run will find them and release them.
4464 void ext4_truncate(struct inode
*inode
)
4467 struct ext4_inode_info
*ei
= EXT4_I(inode
);
4468 __le32
*i_data
= ei
->i_data
;
4469 int addr_per_block
= EXT4_ADDR_PER_BLOCK(inode
->i_sb
);
4470 struct address_space
*mapping
= inode
->i_mapping
;
4471 ext4_lblk_t offsets
[4];
4476 ext4_lblk_t last_block
;
4477 unsigned blocksize
= inode
->i_sb
->s_blocksize
;
4479 if (!ext4_can_truncate(inode
))
4482 ext4_clear_inode_flag(inode
, EXT4_INODE_EOFBLOCKS
);
4484 if (inode
->i_size
== 0 && !test_opt(inode
->i_sb
, NO_AUTO_DA_ALLOC
))
4485 ext4_set_inode_state(inode
, EXT4_STATE_DA_ALLOC_CLOSE
);
4487 if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
)) {
4488 ext4_ext_truncate(inode
);
4492 handle
= start_transaction(inode
);
4494 return; /* AKPM: return what? */
4496 last_block
= (inode
->i_size
+ blocksize
-1)
4497 >> EXT4_BLOCK_SIZE_BITS(inode
->i_sb
);
4499 if (inode
->i_size
& (blocksize
- 1))
4500 if (ext4_block_truncate_page(handle
, mapping
, inode
->i_size
))
4503 n
= ext4_block_to_path(inode
, last_block
, offsets
, NULL
);
4505 goto out_stop
; /* error */
4508 * OK. This truncate is going to happen. We add the inode to the
4509 * orphan list, so that if this truncate spans multiple transactions,
4510 * and we crash, we will resume the truncate when the filesystem
4511 * recovers. It also marks the inode dirty, to catch the new size.
4513 * Implication: the file must always be in a sane, consistent
4514 * truncatable state while each transaction commits.
4516 if (ext4_orphan_add(handle
, inode
))
4520 * From here we block out all ext4_get_block() callers who want to
4521 * modify the block allocation tree.
4523 down_write(&ei
->i_data_sem
);
4525 ext4_discard_preallocations(inode
);
4528 * The orphan list entry will now protect us from any crash which
4529 * occurs before the truncate completes, so it is now safe to propagate
4530 * the new, shorter inode size (held for now in i_size) into the
4531 * on-disk inode. We do this via i_disksize, which is the value which
4532 * ext4 *really* writes onto the disk inode.
4534 ei
->i_disksize
= inode
->i_size
;
4536 if (n
== 1) { /* direct blocks */
4537 ext4_free_data(handle
, inode
, NULL
, i_data
+offsets
[0],
4538 i_data
+ EXT4_NDIR_BLOCKS
);
4542 partial
= ext4_find_shared(inode
, n
, offsets
, chain
, &nr
);
4543 /* Kill the top of shared branch (not detached) */
4545 if (partial
== chain
) {
4546 /* Shared branch grows from the inode */
4547 ext4_free_branches(handle
, inode
, NULL
,
4548 &nr
, &nr
+1, (chain
+n
-1) - partial
);
4551 * We mark the inode dirty prior to restart,
4552 * and prior to stop. No need for it here.
4555 /* Shared branch grows from an indirect block */
4556 BUFFER_TRACE(partial
->bh
, "get_write_access");
4557 ext4_free_branches(handle
, inode
, partial
->bh
,
4559 partial
->p
+1, (chain
+n
-1) - partial
);
4562 /* Clear the ends of indirect blocks on the shared branch */
4563 while (partial
> chain
) {
4564 ext4_free_branches(handle
, inode
, partial
->bh
, partial
->p
+ 1,
4565 (__le32
*)partial
->bh
->b_data
+addr_per_block
,
4566 (chain
+n
-1) - partial
);
4567 BUFFER_TRACE(partial
->bh
, "call brelse");
4568 brelse(partial
->bh
);
4572 /* Kill the remaining (whole) subtrees */
4573 switch (offsets
[0]) {
4575 nr
= i_data
[EXT4_IND_BLOCK
];
4577 ext4_free_branches(handle
, inode
, NULL
, &nr
, &nr
+1, 1);
4578 i_data
[EXT4_IND_BLOCK
] = 0;
4580 case EXT4_IND_BLOCK
:
4581 nr
= i_data
[EXT4_DIND_BLOCK
];
4583 ext4_free_branches(handle
, inode
, NULL
, &nr
, &nr
+1, 2);
4584 i_data
[EXT4_DIND_BLOCK
] = 0;
4586 case EXT4_DIND_BLOCK
:
4587 nr
= i_data
[EXT4_TIND_BLOCK
];
4589 ext4_free_branches(handle
, inode
, NULL
, &nr
, &nr
+1, 3);
4590 i_data
[EXT4_TIND_BLOCK
] = 0;
4592 case EXT4_TIND_BLOCK
:
4596 up_write(&ei
->i_data_sem
);
4597 inode
->i_mtime
= inode
->i_ctime
= ext4_current_time(inode
);
4598 ext4_mark_inode_dirty(handle
, inode
);
4601 * In a multi-transaction truncate, we only make the final transaction
4605 ext4_handle_sync(handle
);
4608 * If this was a simple ftruncate(), and the file will remain alive
4609 * then we need to clear up the orphan record which we created above.
4610 * However, if this was a real unlink then we were called by
4611 * ext4_delete_inode(), and we allow that function to clean up the
4612 * orphan info for us.
4615 ext4_orphan_del(handle
, inode
);
4617 ext4_journal_stop(handle
);
4621 * ext4_get_inode_loc returns with an extra refcount against the inode's
4622 * underlying buffer_head on success. If 'in_mem' is true, we have all
4623 * data in memory that is needed to recreate the on-disk version of this
4626 static int __ext4_get_inode_loc(struct inode
*inode
,
4627 struct ext4_iloc
*iloc
, int in_mem
)
4629 struct ext4_group_desc
*gdp
;
4630 struct buffer_head
*bh
;
4631 struct super_block
*sb
= inode
->i_sb
;
4633 int inodes_per_block
, inode_offset
;
4636 if (!ext4_valid_inum(sb
, inode
->i_ino
))
4639 iloc
->block_group
= (inode
->i_ino
- 1) / EXT4_INODES_PER_GROUP(sb
);
4640 gdp
= ext4_get_group_desc(sb
, iloc
->block_group
, NULL
);
4645 * Figure out the offset within the block group inode table
4647 inodes_per_block
= (EXT4_BLOCK_SIZE(sb
) / EXT4_INODE_SIZE(sb
));
4648 inode_offset
= ((inode
->i_ino
- 1) %
4649 EXT4_INODES_PER_GROUP(sb
));
4650 block
= ext4_inode_table(sb
, gdp
) + (inode_offset
/ inodes_per_block
);
4651 iloc
->offset
= (inode_offset
% inodes_per_block
) * EXT4_INODE_SIZE(sb
);
4653 bh
= sb_getblk(sb
, block
);
4655 EXT4_ERROR_INODE_BLOCK(inode
, block
,
4656 "unable to read itable block");
4659 if (!buffer_uptodate(bh
)) {
4663 * If the buffer has the write error flag, we have failed
4664 * to write out another inode in the same block. In this
4665 * case, we don't have to read the block because we may
4666 * read the old inode data successfully.
4668 if (buffer_write_io_error(bh
) && !buffer_uptodate(bh
))
4669 set_buffer_uptodate(bh
);
4671 if (buffer_uptodate(bh
)) {
4672 /* someone brought it uptodate while we waited */
4678 * If we have all information of the inode in memory and this
4679 * is the only valid inode in the block, we need not read the
4683 struct buffer_head
*bitmap_bh
;
4686 start
= inode_offset
& ~(inodes_per_block
- 1);
4688 /* Is the inode bitmap in cache? */
4689 bitmap_bh
= sb_getblk(sb
, ext4_inode_bitmap(sb
, gdp
));
4694 * If the inode bitmap isn't in cache then the
4695 * optimisation may end up performing two reads instead
4696 * of one, so skip it.
4698 if (!buffer_uptodate(bitmap_bh
)) {
4702 for (i
= start
; i
< start
+ inodes_per_block
; i
++) {
4703 if (i
== inode_offset
)
4705 if (ext4_test_bit(i
, bitmap_bh
->b_data
))
4709 if (i
== start
+ inodes_per_block
) {
4710 /* all other inodes are free, so skip I/O */
4711 memset(bh
->b_data
, 0, bh
->b_size
);
4712 set_buffer_uptodate(bh
);
4720 * If we need to do any I/O, try to pre-readahead extra
4721 * blocks from the inode table.
4723 if (EXT4_SB(sb
)->s_inode_readahead_blks
) {
4724 ext4_fsblk_t b
, end
, table
;
4727 table
= ext4_inode_table(sb
, gdp
);
4728 /* s_inode_readahead_blks is always a power of 2 */
4729 b
= block
& ~(EXT4_SB(sb
)->s_inode_readahead_blks
-1);
4732 end
= b
+ EXT4_SB(sb
)->s_inode_readahead_blks
;
4733 num
= EXT4_INODES_PER_GROUP(sb
);
4734 if (EXT4_HAS_RO_COMPAT_FEATURE(sb
,
4735 EXT4_FEATURE_RO_COMPAT_GDT_CSUM
))
4736 num
-= ext4_itable_unused_count(sb
, gdp
);
4737 table
+= num
/ inodes_per_block
;
4741 sb_breadahead(sb
, b
++);
4745 * There are other valid inodes in the buffer, this inode
4746 * has in-inode xattrs, or we don't have this inode in memory.
4747 * Read the block from disk.
4750 bh
->b_end_io
= end_buffer_read_sync
;
4751 submit_bh(READ_META
, bh
);
4753 if (!buffer_uptodate(bh
)) {
4754 EXT4_ERROR_INODE_BLOCK(inode
, block
,
4755 "unable to read itable block");
4765 int ext4_get_inode_loc(struct inode
*inode
, struct ext4_iloc
*iloc
)
4767 /* We have all inode data except xattrs in memory here. */
4768 return __ext4_get_inode_loc(inode
, iloc
,
4769 !ext4_test_inode_state(inode
, EXT4_STATE_XATTR
));
4772 void ext4_set_inode_flags(struct inode
*inode
)
4774 unsigned int flags
= EXT4_I(inode
)->i_flags
;
4776 inode
->i_flags
&= ~(S_SYNC
|S_APPEND
|S_IMMUTABLE
|S_NOATIME
|S_DIRSYNC
);
4777 if (flags
& EXT4_SYNC_FL
)
4778 inode
->i_flags
|= S_SYNC
;
4779 if (flags
& EXT4_APPEND_FL
)
4780 inode
->i_flags
|= S_APPEND
;
4781 if (flags
& EXT4_IMMUTABLE_FL
)
4782 inode
->i_flags
|= S_IMMUTABLE
;
4783 if (flags
& EXT4_NOATIME_FL
)
4784 inode
->i_flags
|= S_NOATIME
;
4785 if (flags
& EXT4_DIRSYNC_FL
)
4786 inode
->i_flags
|= S_DIRSYNC
;
4789 /* Propagate flags from i_flags to EXT4_I(inode)->i_flags */
4790 void ext4_get_inode_flags(struct ext4_inode_info
*ei
)
4792 unsigned int vfs_fl
;
4793 unsigned long old_fl
, new_fl
;
4796 vfs_fl
= ei
->vfs_inode
.i_flags
;
4797 old_fl
= ei
->i_flags
;
4798 new_fl
= old_fl
& ~(EXT4_SYNC_FL
|EXT4_APPEND_FL
|
4799 EXT4_IMMUTABLE_FL
|EXT4_NOATIME_FL
|
4801 if (vfs_fl
& S_SYNC
)
4802 new_fl
|= EXT4_SYNC_FL
;
4803 if (vfs_fl
& S_APPEND
)
4804 new_fl
|= EXT4_APPEND_FL
;
4805 if (vfs_fl
& S_IMMUTABLE
)
4806 new_fl
|= EXT4_IMMUTABLE_FL
;
4807 if (vfs_fl
& S_NOATIME
)
4808 new_fl
|= EXT4_NOATIME_FL
;
4809 if (vfs_fl
& S_DIRSYNC
)
4810 new_fl
|= EXT4_DIRSYNC_FL
;
4811 } while (cmpxchg(&ei
->i_flags
, old_fl
, new_fl
) != old_fl
);
4814 static blkcnt_t
ext4_inode_blocks(struct ext4_inode
*raw_inode
,
4815 struct ext4_inode_info
*ei
)
4818 struct inode
*inode
= &(ei
->vfs_inode
);
4819 struct super_block
*sb
= inode
->i_sb
;
4821 if (EXT4_HAS_RO_COMPAT_FEATURE(sb
,
4822 EXT4_FEATURE_RO_COMPAT_HUGE_FILE
)) {
4823 /* we are using combined 48 bit field */
4824 i_blocks
= ((u64
)le16_to_cpu(raw_inode
->i_blocks_high
)) << 32 |
4825 le32_to_cpu(raw_inode
->i_blocks_lo
);
4826 if (ext4_test_inode_flag(inode
, EXT4_INODE_HUGE_FILE
)) {
4827 /* i_blocks represent file system block size */
4828 return i_blocks
<< (inode
->i_blkbits
- 9);
4833 return le32_to_cpu(raw_inode
->i_blocks_lo
);
4837 struct inode
*ext4_iget(struct super_block
*sb
, unsigned long ino
)
4839 struct ext4_iloc iloc
;
4840 struct ext4_inode
*raw_inode
;
4841 struct ext4_inode_info
*ei
;
4842 struct inode
*inode
;
4843 journal_t
*journal
= EXT4_SB(sb
)->s_journal
;
4847 inode
= iget_locked(sb
, ino
);
4849 return ERR_PTR(-ENOMEM
);
4850 if (!(inode
->i_state
& I_NEW
))
4856 ret
= __ext4_get_inode_loc(inode
, &iloc
, 0);
4859 raw_inode
= ext4_raw_inode(&iloc
);
4860 inode
->i_mode
= le16_to_cpu(raw_inode
->i_mode
);
4861 inode
->i_uid
= (uid_t
)le16_to_cpu(raw_inode
->i_uid_low
);
4862 inode
->i_gid
= (gid_t
)le16_to_cpu(raw_inode
->i_gid_low
);
4863 if (!(test_opt(inode
->i_sb
, NO_UID32
))) {
4864 inode
->i_uid
|= le16_to_cpu(raw_inode
->i_uid_high
) << 16;
4865 inode
->i_gid
|= le16_to_cpu(raw_inode
->i_gid_high
) << 16;
4867 inode
->i_nlink
= le16_to_cpu(raw_inode
->i_links_count
);
4869 ext4_clear_state_flags(ei
); /* Only relevant on 32-bit archs */
4870 ei
->i_dir_start_lookup
= 0;
4871 ei
->i_dtime
= le32_to_cpu(raw_inode
->i_dtime
);
4872 /* We now have enough fields to check if the inode was active or not.
4873 * This is needed because nfsd might try to access dead inodes
4874 * the test is that same one that e2fsck uses
4875 * NeilBrown 1999oct15
4877 if (inode
->i_nlink
== 0) {
4878 if (inode
->i_mode
== 0 ||
4879 !(EXT4_SB(inode
->i_sb
)->s_mount_state
& EXT4_ORPHAN_FS
)) {
4880 /* this inode is deleted */
4884 /* The only unlinked inodes we let through here have
4885 * valid i_mode and are being read by the orphan
4886 * recovery code: that's fine, we're about to complete
4887 * the process of deleting those. */
4889 ei
->i_flags
= le32_to_cpu(raw_inode
->i_flags
);
4890 inode
->i_blocks
= ext4_inode_blocks(raw_inode
, ei
);
4891 ei
->i_file_acl
= le32_to_cpu(raw_inode
->i_file_acl_lo
);
4892 if (EXT4_HAS_INCOMPAT_FEATURE(sb
, EXT4_FEATURE_INCOMPAT_64BIT
))
4894 ((__u64
)le16_to_cpu(raw_inode
->i_file_acl_high
)) << 32;
4895 inode
->i_size
= ext4_isize(raw_inode
);
4896 ei
->i_disksize
= inode
->i_size
;
4898 ei
->i_reserved_quota
= 0;
4900 inode
->i_generation
= le32_to_cpu(raw_inode
->i_generation
);
4901 ei
->i_block_group
= iloc
.block_group
;
4902 ei
->i_last_alloc_group
= ~0;
4904 * NOTE! The in-memory inode i_data array is in little-endian order
4905 * even on big-endian machines: we do NOT byteswap the block numbers!
4907 for (block
= 0; block
< EXT4_N_BLOCKS
; block
++)
4908 ei
->i_data
[block
] = raw_inode
->i_block
[block
];
4909 INIT_LIST_HEAD(&ei
->i_orphan
);
4912 * Set transaction id's of transactions that have to be committed
4913 * to finish f[data]sync. We set them to currently running transaction
4914 * as we cannot be sure that the inode or some of its metadata isn't
4915 * part of the transaction - the inode could have been reclaimed and
4916 * now it is reread from disk.
4919 transaction_t
*transaction
;
4922 read_lock(&journal
->j_state_lock
);
4923 if (journal
->j_running_transaction
)
4924 transaction
= journal
->j_running_transaction
;
4926 transaction
= journal
->j_committing_transaction
;
4928 tid
= transaction
->t_tid
;
4930 tid
= journal
->j_commit_sequence
;
4931 read_unlock(&journal
->j_state_lock
);
4932 ei
->i_sync_tid
= tid
;
4933 ei
->i_datasync_tid
= tid
;
4936 if (EXT4_INODE_SIZE(inode
->i_sb
) > EXT4_GOOD_OLD_INODE_SIZE
) {
4937 ei
->i_extra_isize
= le16_to_cpu(raw_inode
->i_extra_isize
);
4938 if (EXT4_GOOD_OLD_INODE_SIZE
+ ei
->i_extra_isize
>
4939 EXT4_INODE_SIZE(inode
->i_sb
)) {
4943 if (ei
->i_extra_isize
== 0) {
4944 /* The extra space is currently unused. Use it. */
4945 ei
->i_extra_isize
= sizeof(struct ext4_inode
) -
4946 EXT4_GOOD_OLD_INODE_SIZE
;
4948 __le32
*magic
= (void *)raw_inode
+
4949 EXT4_GOOD_OLD_INODE_SIZE
+
4951 if (*magic
== cpu_to_le32(EXT4_XATTR_MAGIC
))
4952 ext4_set_inode_state(inode
, EXT4_STATE_XATTR
);
4955 ei
->i_extra_isize
= 0;
4957 EXT4_INODE_GET_XTIME(i_ctime
, inode
, raw_inode
);
4958 EXT4_INODE_GET_XTIME(i_mtime
, inode
, raw_inode
);
4959 EXT4_INODE_GET_XTIME(i_atime
, inode
, raw_inode
);
4960 EXT4_EINODE_GET_XTIME(i_crtime
, ei
, raw_inode
);
4962 inode
->i_version
= le32_to_cpu(raw_inode
->i_disk_version
);
4963 if (EXT4_INODE_SIZE(inode
->i_sb
) > EXT4_GOOD_OLD_INODE_SIZE
) {
4964 if (EXT4_FITS_IN_INODE(raw_inode
, ei
, i_version_hi
))
4966 (__u64
)(le32_to_cpu(raw_inode
->i_version_hi
)) << 32;
4970 if (ei
->i_file_acl
&&
4971 !ext4_data_block_valid(EXT4_SB(sb
), ei
->i_file_acl
, 1)) {
4972 EXT4_ERROR_INODE(inode
, "bad extended attribute block %llu",
4976 } else if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
)) {
4977 if (S_ISREG(inode
->i_mode
) || S_ISDIR(inode
->i_mode
) ||
4978 (S_ISLNK(inode
->i_mode
) &&
4979 !ext4_inode_is_fast_symlink(inode
)))
4980 /* Validate extent which is part of inode */
4981 ret
= ext4_ext_check_inode(inode
);
4982 } else if (S_ISREG(inode
->i_mode
) || S_ISDIR(inode
->i_mode
) ||
4983 (S_ISLNK(inode
->i_mode
) &&
4984 !ext4_inode_is_fast_symlink(inode
))) {
4985 /* Validate block references which are part of inode */
4986 ret
= ext4_check_inode_blockref(inode
);
4991 if (S_ISREG(inode
->i_mode
)) {
4992 inode
->i_op
= &ext4_file_inode_operations
;
4993 inode
->i_fop
= &ext4_file_operations
;
4994 ext4_set_aops(inode
);
4995 } else if (S_ISDIR(inode
->i_mode
)) {
4996 inode
->i_op
= &ext4_dir_inode_operations
;
4997 inode
->i_fop
= &ext4_dir_operations
;
4998 } else if (S_ISLNK(inode
->i_mode
)) {
4999 if (ext4_inode_is_fast_symlink(inode
)) {
5000 inode
->i_op
= &ext4_fast_symlink_inode_operations
;
5001 nd_terminate_link(ei
->i_data
, inode
->i_size
,
5002 sizeof(ei
->i_data
) - 1);
5004 inode
->i_op
= &ext4_symlink_inode_operations
;
5005 ext4_set_aops(inode
);
5007 } else if (S_ISCHR(inode
->i_mode
) || S_ISBLK(inode
->i_mode
) ||
5008 S_ISFIFO(inode
->i_mode
) || S_ISSOCK(inode
->i_mode
)) {
5009 inode
->i_op
= &ext4_special_inode_operations
;
5010 if (raw_inode
->i_block
[0])
5011 init_special_inode(inode
, inode
->i_mode
,
5012 old_decode_dev(le32_to_cpu(raw_inode
->i_block
[0])));
5014 init_special_inode(inode
, inode
->i_mode
,
5015 new_decode_dev(le32_to_cpu(raw_inode
->i_block
[1])));
5018 EXT4_ERROR_INODE(inode
, "bogus i_mode (%o)", inode
->i_mode
);
5022 ext4_set_inode_flags(inode
);
5023 unlock_new_inode(inode
);
5029 return ERR_PTR(ret
);
5032 static int ext4_inode_blocks_set(handle_t
*handle
,
5033 struct ext4_inode
*raw_inode
,
5034 struct ext4_inode_info
*ei
)
5036 struct inode
*inode
= &(ei
->vfs_inode
);
5037 u64 i_blocks
= inode
->i_blocks
;
5038 struct super_block
*sb
= inode
->i_sb
;
5040 if (i_blocks
<= ~0U) {
5042 * i_blocks can be represnted in a 32 bit variable
5043 * as multiple of 512 bytes
5045 raw_inode
->i_blocks_lo
= cpu_to_le32(i_blocks
);
5046 raw_inode
->i_blocks_high
= 0;
5047 ext4_clear_inode_flag(inode
, EXT4_INODE_HUGE_FILE
);
5050 if (!EXT4_HAS_RO_COMPAT_FEATURE(sb
, EXT4_FEATURE_RO_COMPAT_HUGE_FILE
))
5053 if (i_blocks
<= 0xffffffffffffULL
) {
5055 * i_blocks can be represented in a 48 bit variable
5056 * as multiple of 512 bytes
5058 raw_inode
->i_blocks_lo
= cpu_to_le32(i_blocks
);
5059 raw_inode
->i_blocks_high
= cpu_to_le16(i_blocks
>> 32);
5060 ext4_clear_inode_flag(inode
, EXT4_INODE_HUGE_FILE
);
5062 ext4_set_inode_flag(inode
, EXT4_INODE_HUGE_FILE
);
5063 /* i_block is stored in file system block size */
5064 i_blocks
= i_blocks
>> (inode
->i_blkbits
- 9);
5065 raw_inode
->i_blocks_lo
= cpu_to_le32(i_blocks
);
5066 raw_inode
->i_blocks_high
= cpu_to_le16(i_blocks
>> 32);
5072 * Post the struct inode info into an on-disk inode location in the
5073 * buffer-cache. This gobbles the caller's reference to the
5074 * buffer_head in the inode location struct.
5076 * The caller must have write access to iloc->bh.
5078 static int ext4_do_update_inode(handle_t
*handle
,
5079 struct inode
*inode
,
5080 struct ext4_iloc
*iloc
)
5082 struct ext4_inode
*raw_inode
= ext4_raw_inode(iloc
);
5083 struct ext4_inode_info
*ei
= EXT4_I(inode
);
5084 struct buffer_head
*bh
= iloc
->bh
;
5085 int err
= 0, rc
, block
;
5087 /* For fields not not tracking in the in-memory inode,
5088 * initialise them to zero for new inodes. */
5089 if (ext4_test_inode_state(inode
, EXT4_STATE_NEW
))
5090 memset(raw_inode
, 0, EXT4_SB(inode
->i_sb
)->s_inode_size
);
5092 ext4_get_inode_flags(ei
);
5093 raw_inode
->i_mode
= cpu_to_le16(inode
->i_mode
);
5094 if (!(test_opt(inode
->i_sb
, NO_UID32
))) {
5095 raw_inode
->i_uid_low
= cpu_to_le16(low_16_bits(inode
->i_uid
));
5096 raw_inode
->i_gid_low
= cpu_to_le16(low_16_bits(inode
->i_gid
));
5098 * Fix up interoperability with old kernels. Otherwise, old inodes get
5099 * re-used with the upper 16 bits of the uid/gid intact
5102 raw_inode
->i_uid_high
=
5103 cpu_to_le16(high_16_bits(inode
->i_uid
));
5104 raw_inode
->i_gid_high
=
5105 cpu_to_le16(high_16_bits(inode
->i_gid
));
5107 raw_inode
->i_uid_high
= 0;
5108 raw_inode
->i_gid_high
= 0;
5111 raw_inode
->i_uid_low
=
5112 cpu_to_le16(fs_high2lowuid(inode
->i_uid
));
5113 raw_inode
->i_gid_low
=
5114 cpu_to_le16(fs_high2lowgid(inode
->i_gid
));
5115 raw_inode
->i_uid_high
= 0;
5116 raw_inode
->i_gid_high
= 0;
5118 raw_inode
->i_links_count
= cpu_to_le16(inode
->i_nlink
);
5120 EXT4_INODE_SET_XTIME(i_ctime
, inode
, raw_inode
);
5121 EXT4_INODE_SET_XTIME(i_mtime
, inode
, raw_inode
);
5122 EXT4_INODE_SET_XTIME(i_atime
, inode
, raw_inode
);
5123 EXT4_EINODE_SET_XTIME(i_crtime
, ei
, raw_inode
);
5125 if (ext4_inode_blocks_set(handle
, raw_inode
, ei
))
5127 raw_inode
->i_dtime
= cpu_to_le32(ei
->i_dtime
);
5128 raw_inode
->i_flags
= cpu_to_le32(ei
->i_flags
& 0xFFFFFFFF);
5129 if (EXT4_SB(inode
->i_sb
)->s_es
->s_creator_os
!=
5130 cpu_to_le32(EXT4_OS_HURD
))
5131 raw_inode
->i_file_acl_high
=
5132 cpu_to_le16(ei
->i_file_acl
>> 32);
5133 raw_inode
->i_file_acl_lo
= cpu_to_le32(ei
->i_file_acl
);
5134 ext4_isize_set(raw_inode
, ei
->i_disksize
);
5135 if (ei
->i_disksize
> 0x7fffffffULL
) {
5136 struct super_block
*sb
= inode
->i_sb
;
5137 if (!EXT4_HAS_RO_COMPAT_FEATURE(sb
,
5138 EXT4_FEATURE_RO_COMPAT_LARGE_FILE
) ||
5139 EXT4_SB(sb
)->s_es
->s_rev_level
==
5140 cpu_to_le32(EXT4_GOOD_OLD_REV
)) {
5141 /* If this is the first large file
5142 * created, add a flag to the superblock.
5144 err
= ext4_journal_get_write_access(handle
,
5145 EXT4_SB(sb
)->s_sbh
);
5148 ext4_update_dynamic_rev(sb
);
5149 EXT4_SET_RO_COMPAT_FEATURE(sb
,
5150 EXT4_FEATURE_RO_COMPAT_LARGE_FILE
);
5152 ext4_handle_sync(handle
);
5153 err
= ext4_handle_dirty_metadata(handle
, NULL
,
5154 EXT4_SB(sb
)->s_sbh
);
5157 raw_inode
->i_generation
= cpu_to_le32(inode
->i_generation
);
5158 if (S_ISCHR(inode
->i_mode
) || S_ISBLK(inode
->i_mode
)) {
5159 if (old_valid_dev(inode
->i_rdev
)) {
5160 raw_inode
->i_block
[0] =
5161 cpu_to_le32(old_encode_dev(inode
->i_rdev
));
5162 raw_inode
->i_block
[1] = 0;
5164 raw_inode
->i_block
[0] = 0;
5165 raw_inode
->i_block
[1] =
5166 cpu_to_le32(new_encode_dev(inode
->i_rdev
));
5167 raw_inode
->i_block
[2] = 0;
5170 for (block
= 0; block
< EXT4_N_BLOCKS
; block
++)
5171 raw_inode
->i_block
[block
] = ei
->i_data
[block
];
5173 raw_inode
->i_disk_version
= cpu_to_le32(inode
->i_version
);
5174 if (ei
->i_extra_isize
) {
5175 if (EXT4_FITS_IN_INODE(raw_inode
, ei
, i_version_hi
))
5176 raw_inode
->i_version_hi
=
5177 cpu_to_le32(inode
->i_version
>> 32);
5178 raw_inode
->i_extra_isize
= cpu_to_le16(ei
->i_extra_isize
);
5181 BUFFER_TRACE(bh
, "call ext4_handle_dirty_metadata");
5182 rc
= ext4_handle_dirty_metadata(handle
, NULL
, bh
);
5185 ext4_clear_inode_state(inode
, EXT4_STATE_NEW
);
5187 ext4_update_inode_fsync_trans(handle
, inode
, 0);
5190 ext4_std_error(inode
->i_sb
, err
);
5195 * ext4_write_inode()
5197 * We are called from a few places:
5199 * - Within generic_file_write() for O_SYNC files.
5200 * Here, there will be no transaction running. We wait for any running
5201 * trasnaction to commit.
5203 * - Within sys_sync(), kupdate and such.
5204 * We wait on commit, if tol to.
5206 * - Within prune_icache() (PF_MEMALLOC == true)
5207 * Here we simply return. We can't afford to block kswapd on the
5210 * In all cases it is actually safe for us to return without doing anything,
5211 * because the inode has been copied into a raw inode buffer in
5212 * ext4_mark_inode_dirty(). This is a correctness thing for O_SYNC and for
5215 * Note that we are absolutely dependent upon all inode dirtiers doing the
5216 * right thing: they *must* call mark_inode_dirty() after dirtying info in
5217 * which we are interested.
5219 * It would be a bug for them to not do this. The code:
5221 * mark_inode_dirty(inode)
5223 * inode->i_size = expr;
5225 * is in error because a kswapd-driven write_inode() could occur while
5226 * `stuff()' is running, and the new i_size will be lost. Plus the inode
5227 * will no longer be on the superblock's dirty inode list.
5229 int ext4_write_inode(struct inode
*inode
, struct writeback_control
*wbc
)
5233 if (current
->flags
& PF_MEMALLOC
)
5236 if (EXT4_SB(inode
->i_sb
)->s_journal
) {
5237 if (ext4_journal_current_handle()) {
5238 jbd_debug(1, "called recursively, non-PF_MEMALLOC!\n");
5243 if (wbc
->sync_mode
!= WB_SYNC_ALL
)
5246 err
= ext4_force_commit(inode
->i_sb
);
5248 struct ext4_iloc iloc
;
5250 err
= __ext4_get_inode_loc(inode
, &iloc
, 0);
5253 if (wbc
->sync_mode
== WB_SYNC_ALL
)
5254 sync_dirty_buffer(iloc
.bh
);
5255 if (buffer_req(iloc
.bh
) && !buffer_uptodate(iloc
.bh
)) {
5256 EXT4_ERROR_INODE_BLOCK(inode
, iloc
.bh
->b_blocknr
,
5257 "IO error syncing inode");
5268 * Called from notify_change.
5270 * We want to trap VFS attempts to truncate the file as soon as
5271 * possible. In particular, we want to make sure that when the VFS
5272 * shrinks i_size, we put the inode on the orphan list and modify
5273 * i_disksize immediately, so that during the subsequent flushing of
5274 * dirty pages and freeing of disk blocks, we can guarantee that any
5275 * commit will leave the blocks being flushed in an unused state on
5276 * disk. (On recovery, the inode will get truncated and the blocks will
5277 * be freed, so we have a strong guarantee that no future commit will
5278 * leave these blocks visible to the user.)
5280 * Another thing we have to assure is that if we are in ordered mode
5281 * and inode is still attached to the committing transaction, we must
5282 * we start writeout of all the dirty pages which are being truncated.
5283 * This way we are sure that all the data written in the previous
5284 * transaction are already on disk (truncate waits for pages under
5287 * Called with inode->i_mutex down.
5289 int ext4_setattr(struct dentry
*dentry
, struct iattr
*attr
)
5291 struct inode
*inode
= dentry
->d_inode
;
5294 const unsigned int ia_valid
= attr
->ia_valid
;
5296 error
= inode_change_ok(inode
, attr
);
5300 if (is_quota_modification(inode
, attr
))
5301 dquot_initialize(inode
);
5302 if ((ia_valid
& ATTR_UID
&& attr
->ia_uid
!= inode
->i_uid
) ||
5303 (ia_valid
& ATTR_GID
&& attr
->ia_gid
!= inode
->i_gid
)) {
5306 /* (user+group)*(old+new) structure, inode write (sb,
5307 * inode block, ? - but truncate inode update has it) */
5308 handle
= ext4_journal_start(inode
, (EXT4_MAXQUOTAS_INIT_BLOCKS(inode
->i_sb
)+
5309 EXT4_MAXQUOTAS_DEL_BLOCKS(inode
->i_sb
))+3);
5310 if (IS_ERR(handle
)) {
5311 error
= PTR_ERR(handle
);
5314 error
= dquot_transfer(inode
, attr
);
5316 ext4_journal_stop(handle
);
5319 /* Update corresponding info in inode so that everything is in
5320 * one transaction */
5321 if (attr
->ia_valid
& ATTR_UID
)
5322 inode
->i_uid
= attr
->ia_uid
;
5323 if (attr
->ia_valid
& ATTR_GID
)
5324 inode
->i_gid
= attr
->ia_gid
;
5325 error
= ext4_mark_inode_dirty(handle
, inode
);
5326 ext4_journal_stop(handle
);
5329 if (attr
->ia_valid
& ATTR_SIZE
) {
5330 if (!(ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
))) {
5331 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
5333 if (attr
->ia_size
> sbi
->s_bitmap_maxbytes
)
5338 if (S_ISREG(inode
->i_mode
) &&
5339 attr
->ia_valid
& ATTR_SIZE
&&
5340 (attr
->ia_size
< inode
->i_size
||
5341 (ext4_test_inode_flag(inode
, EXT4_INODE_EOFBLOCKS
)))) {
5344 handle
= ext4_journal_start(inode
, 3);
5345 if (IS_ERR(handle
)) {
5346 error
= PTR_ERR(handle
);
5349 if (ext4_handle_valid(handle
)) {
5350 error
= ext4_orphan_add(handle
, inode
);
5353 EXT4_I(inode
)->i_disksize
= attr
->ia_size
;
5354 rc
= ext4_mark_inode_dirty(handle
, inode
);
5357 ext4_journal_stop(handle
);
5359 if (ext4_should_order_data(inode
)) {
5360 error
= ext4_begin_ordered_truncate(inode
,
5363 /* Do as much error cleanup as possible */
5364 handle
= ext4_journal_start(inode
, 3);
5365 if (IS_ERR(handle
)) {
5366 ext4_orphan_del(NULL
, inode
);
5369 ext4_orphan_del(handle
, inode
);
5371 ext4_journal_stop(handle
);
5375 /* ext4_truncate will clear the flag */
5376 if ((ext4_test_inode_flag(inode
, EXT4_INODE_EOFBLOCKS
)))
5377 ext4_truncate(inode
);
5380 if ((attr
->ia_valid
& ATTR_SIZE
) &&
5381 attr
->ia_size
!= i_size_read(inode
))
5382 rc
= vmtruncate(inode
, attr
->ia_size
);
5385 setattr_copy(inode
, attr
);
5386 mark_inode_dirty(inode
);
5390 * If the call to ext4_truncate failed to get a transaction handle at
5391 * all, we need to clean up the in-core orphan list manually.
5393 if (orphan
&& inode
->i_nlink
)
5394 ext4_orphan_del(NULL
, inode
);
5396 if (!rc
&& (ia_valid
& ATTR_MODE
))
5397 rc
= ext4_acl_chmod(inode
);
5400 ext4_std_error(inode
->i_sb
, error
);
5406 int ext4_getattr(struct vfsmount
*mnt
, struct dentry
*dentry
,
5409 struct inode
*inode
;
5410 unsigned long delalloc_blocks
;
5412 inode
= dentry
->d_inode
;
5413 generic_fillattr(inode
, stat
);
5416 * We can't update i_blocks if the block allocation is delayed
5417 * otherwise in the case of system crash before the real block
5418 * allocation is done, we will have i_blocks inconsistent with
5419 * on-disk file blocks.
5420 * We always keep i_blocks updated together with real
5421 * allocation. But to not confuse with user, stat
5422 * will return the blocks that include the delayed allocation
5423 * blocks for this file.
5425 delalloc_blocks
= EXT4_I(inode
)->i_reserved_data_blocks
;
5427 stat
->blocks
+= (delalloc_blocks
<< inode
->i_sb
->s_blocksize_bits
)>>9;
5431 static int ext4_indirect_trans_blocks(struct inode
*inode
, int nrblocks
,
5436 /* if nrblocks are contiguous */
5439 * With N contiguous data blocks, it need at most
5440 * N/EXT4_ADDR_PER_BLOCK(inode->i_sb) indirect blocks
5441 * 2 dindirect blocks
5444 indirects
= nrblocks
/ EXT4_ADDR_PER_BLOCK(inode
->i_sb
);
5445 return indirects
+ 3;
5448 * if nrblocks are not contiguous, worse case, each block touch
5449 * a indirect block, and each indirect block touch a double indirect
5450 * block, plus a triple indirect block
5452 indirects
= nrblocks
* 2 + 1;
5456 static int ext4_index_trans_blocks(struct inode
*inode
, int nrblocks
, int chunk
)
5458 if (!(ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
)))
5459 return ext4_indirect_trans_blocks(inode
, nrblocks
, chunk
);
5460 return ext4_ext_index_trans_blocks(inode
, nrblocks
, chunk
);
5464 * Account for index blocks, block groups bitmaps and block group
5465 * descriptor blocks if modify datablocks and index blocks
5466 * worse case, the indexs blocks spread over different block groups
5468 * If datablocks are discontiguous, they are possible to spread over
5469 * different block groups too. If they are contiuguous, with flexbg,
5470 * they could still across block group boundary.
5472 * Also account for superblock, inode, quota and xattr blocks
5474 static int ext4_meta_trans_blocks(struct inode
*inode
, int nrblocks
, int chunk
)
5476 ext4_group_t groups
, ngroups
= ext4_get_groups_count(inode
->i_sb
);
5482 * How many index blocks need to touch to modify nrblocks?
5483 * The "Chunk" flag indicating whether the nrblocks is
5484 * physically contiguous on disk
5486 * For Direct IO and fallocate, they calls get_block to allocate
5487 * one single extent at a time, so they could set the "Chunk" flag
5489 idxblocks
= ext4_index_trans_blocks(inode
, nrblocks
, chunk
);
5494 * Now let's see how many group bitmaps and group descriptors need
5504 if (groups
> ngroups
)
5506 if (groups
> EXT4_SB(inode
->i_sb
)->s_gdb_count
)
5507 gdpblocks
= EXT4_SB(inode
->i_sb
)->s_gdb_count
;
5509 /* bitmaps and block group descriptor blocks */
5510 ret
+= groups
+ gdpblocks
;
5512 /* Blocks for super block, inode, quota and xattr blocks */
5513 ret
+= EXT4_META_TRANS_BLOCKS(inode
->i_sb
);
5519 * Calulate the total number of credits to reserve to fit
5520 * the modification of a single pages into a single transaction,
5521 * which may include multiple chunks of block allocations.
5523 * This could be called via ext4_write_begin()
5525 * We need to consider the worse case, when
5526 * one new block per extent.
5528 int ext4_writepage_trans_blocks(struct inode
*inode
)
5530 int bpp
= ext4_journal_blocks_per_page(inode
);
5533 ret
= ext4_meta_trans_blocks(inode
, bpp
, 0);
5535 /* Account for data blocks for journalled mode */
5536 if (ext4_should_journal_data(inode
))
5542 * Calculate the journal credits for a chunk of data modification.
5544 * This is called from DIO, fallocate or whoever calling
5545 * ext4_map_blocks() to map/allocate a chunk of contiguous disk blocks.
5547 * journal buffers for data blocks are not included here, as DIO
5548 * and fallocate do no need to journal data buffers.
5550 int ext4_chunk_trans_blocks(struct inode
*inode
, int nrblocks
)
5552 return ext4_meta_trans_blocks(inode
, nrblocks
, 1);
5556 * The caller must have previously called ext4_reserve_inode_write().
5557 * Give this, we know that the caller already has write access to iloc->bh.
5559 int ext4_mark_iloc_dirty(handle_t
*handle
,
5560 struct inode
*inode
, struct ext4_iloc
*iloc
)
5564 if (test_opt(inode
->i_sb
, I_VERSION
))
5565 inode_inc_iversion(inode
);
5567 /* the do_update_inode consumes one bh->b_count */
5570 /* ext4_do_update_inode() does jbd2_journal_dirty_metadata */
5571 err
= ext4_do_update_inode(handle
, inode
, iloc
);
5577 * On success, We end up with an outstanding reference count against
5578 * iloc->bh. This _must_ be cleaned up later.
5582 ext4_reserve_inode_write(handle_t
*handle
, struct inode
*inode
,
5583 struct ext4_iloc
*iloc
)
5587 err
= ext4_get_inode_loc(inode
, iloc
);
5589 BUFFER_TRACE(iloc
->bh
, "get_write_access");
5590 err
= ext4_journal_get_write_access(handle
, iloc
->bh
);
5596 ext4_std_error(inode
->i_sb
, err
);
5601 * Expand an inode by new_extra_isize bytes.
5602 * Returns 0 on success or negative error number on failure.
5604 static int ext4_expand_extra_isize(struct inode
*inode
,
5605 unsigned int new_extra_isize
,
5606 struct ext4_iloc iloc
,
5609 struct ext4_inode
*raw_inode
;
5610 struct ext4_xattr_ibody_header
*header
;
5612 if (EXT4_I(inode
)->i_extra_isize
>= new_extra_isize
)
5615 raw_inode
= ext4_raw_inode(&iloc
);
5617 header
= IHDR(inode
, raw_inode
);
5619 /* No extended attributes present */
5620 if (!ext4_test_inode_state(inode
, EXT4_STATE_XATTR
) ||
5621 header
->h_magic
!= cpu_to_le32(EXT4_XATTR_MAGIC
)) {
5622 memset((void *)raw_inode
+ EXT4_GOOD_OLD_INODE_SIZE
, 0,
5624 EXT4_I(inode
)->i_extra_isize
= new_extra_isize
;
5628 /* try to expand with EAs present */
5629 return ext4_expand_extra_isize_ea(inode
, new_extra_isize
,
5634 * What we do here is to mark the in-core inode as clean with respect to inode
5635 * dirtiness (it may still be data-dirty).
5636 * This means that the in-core inode may be reaped by prune_icache
5637 * without having to perform any I/O. This is a very good thing,
5638 * because *any* task may call prune_icache - even ones which
5639 * have a transaction open against a different journal.
5641 * Is this cheating? Not really. Sure, we haven't written the
5642 * inode out, but prune_icache isn't a user-visible syncing function.
5643 * Whenever the user wants stuff synced (sys_sync, sys_msync, sys_fsync)
5644 * we start and wait on commits.
5646 * Is this efficient/effective? Well, we're being nice to the system
5647 * by cleaning up our inodes proactively so they can be reaped
5648 * without I/O. But we are potentially leaving up to five seconds'
5649 * worth of inodes floating about which prune_icache wants us to
5650 * write out. One way to fix that would be to get prune_icache()
5651 * to do a write_super() to free up some memory. It has the desired
5654 int ext4_mark_inode_dirty(handle_t
*handle
, struct inode
*inode
)
5656 struct ext4_iloc iloc
;
5657 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
5658 static unsigned int mnt_count
;
5662 trace_ext4_mark_inode_dirty(inode
, _RET_IP_
);
5663 err
= ext4_reserve_inode_write(handle
, inode
, &iloc
);
5664 if (ext4_handle_valid(handle
) &&
5665 EXT4_I(inode
)->i_extra_isize
< sbi
->s_want_extra_isize
&&
5666 !ext4_test_inode_state(inode
, EXT4_STATE_NO_EXPAND
)) {
5668 * We need extra buffer credits since we may write into EA block
5669 * with this same handle. If journal_extend fails, then it will
5670 * only result in a minor loss of functionality for that inode.
5671 * If this is felt to be critical, then e2fsck should be run to
5672 * force a large enough s_min_extra_isize.
5674 if ((jbd2_journal_extend(handle
,
5675 EXT4_DATA_TRANS_BLOCKS(inode
->i_sb
))) == 0) {
5676 ret
= ext4_expand_extra_isize(inode
,
5677 sbi
->s_want_extra_isize
,
5680 ext4_set_inode_state(inode
,
5681 EXT4_STATE_NO_EXPAND
);
5683 le16_to_cpu(sbi
->s_es
->s_mnt_count
)) {
5684 ext4_warning(inode
->i_sb
,
5685 "Unable to expand inode %lu. Delete"
5686 " some EAs or run e2fsck.",
5689 le16_to_cpu(sbi
->s_es
->s_mnt_count
);
5695 err
= ext4_mark_iloc_dirty(handle
, inode
, &iloc
);
5700 * ext4_dirty_inode() is called from __mark_inode_dirty()
5702 * We're really interested in the case where a file is being extended.
5703 * i_size has been changed by generic_commit_write() and we thus need
5704 * to include the updated inode in the current transaction.
5706 * Also, dquot_alloc_block() will always dirty the inode when blocks
5707 * are allocated to the file.
5709 * If the inode is marked synchronous, we don't honour that here - doing
5710 * so would cause a commit on atime updates, which we don't bother doing.
5711 * We handle synchronous inodes at the highest possible level.
5713 void ext4_dirty_inode(struct inode
*inode
)
5717 handle
= ext4_journal_start(inode
, 2);
5721 ext4_mark_inode_dirty(handle
, inode
);
5723 ext4_journal_stop(handle
);
5730 * Bind an inode's backing buffer_head into this transaction, to prevent
5731 * it from being flushed to disk early. Unlike
5732 * ext4_reserve_inode_write, this leaves behind no bh reference and
5733 * returns no iloc structure, so the caller needs to repeat the iloc
5734 * lookup to mark the inode dirty later.
5736 static int ext4_pin_inode(handle_t
*handle
, struct inode
*inode
)
5738 struct ext4_iloc iloc
;
5742 err
= ext4_get_inode_loc(inode
, &iloc
);
5744 BUFFER_TRACE(iloc
.bh
, "get_write_access");
5745 err
= jbd2_journal_get_write_access(handle
, iloc
.bh
);
5747 err
= ext4_handle_dirty_metadata(handle
,
5753 ext4_std_error(inode
->i_sb
, err
);
5758 int ext4_change_inode_journal_flag(struct inode
*inode
, int val
)
5765 * We have to be very careful here: changing a data block's
5766 * journaling status dynamically is dangerous. If we write a
5767 * data block to the journal, change the status and then delete
5768 * that block, we risk forgetting to revoke the old log record
5769 * from the journal and so a subsequent replay can corrupt data.
5770 * So, first we make sure that the journal is empty and that
5771 * nobody is changing anything.
5774 journal
= EXT4_JOURNAL(inode
);
5777 if (is_journal_aborted(journal
))
5780 jbd2_journal_lock_updates(journal
);
5781 jbd2_journal_flush(journal
);
5784 * OK, there are no updates running now, and all cached data is
5785 * synced to disk. We are now in a completely consistent state
5786 * which doesn't have anything in the journal, and we know that
5787 * no filesystem updates are running, so it is safe to modify
5788 * the inode's in-core data-journaling state flag now.
5792 ext4_set_inode_flag(inode
, EXT4_INODE_JOURNAL_DATA
);
5794 ext4_clear_inode_flag(inode
, EXT4_INODE_JOURNAL_DATA
);
5795 ext4_set_aops(inode
);
5797 jbd2_journal_unlock_updates(journal
);
5799 /* Finally we can mark the inode as dirty. */
5801 handle
= ext4_journal_start(inode
, 1);
5803 return PTR_ERR(handle
);
5805 err
= ext4_mark_inode_dirty(handle
, inode
);
5806 ext4_handle_sync(handle
);
5807 ext4_journal_stop(handle
);
5808 ext4_std_error(inode
->i_sb
, err
);
5813 static int ext4_bh_unmapped(handle_t
*handle
, struct buffer_head
*bh
)
5815 return !buffer_mapped(bh
);
5818 int ext4_page_mkwrite(struct vm_area_struct
*vma
, struct vm_fault
*vmf
)
5820 struct page
*page
= vmf
->page
;
5825 struct file
*file
= vma
->vm_file
;
5826 struct inode
*inode
= file
->f_path
.dentry
->d_inode
;
5827 struct address_space
*mapping
= inode
->i_mapping
;
5830 * Get i_alloc_sem to stop truncates messing with the inode. We cannot
5831 * get i_mutex because we are already holding mmap_sem.
5833 down_read(&inode
->i_alloc_sem
);
5834 size
= i_size_read(inode
);
5835 if (page
->mapping
!= mapping
|| size
<= page_offset(page
)
5836 || !PageUptodate(page
)) {
5837 /* page got truncated from under us? */
5841 if (PageMappedToDisk(page
))
5844 if (page
->index
== size
>> PAGE_CACHE_SHIFT
)
5845 len
= size
& ~PAGE_CACHE_MASK
;
5847 len
= PAGE_CACHE_SIZE
;
5851 * return if we have all the buffers mapped. This avoid
5852 * the need to call write_begin/write_end which does a
5853 * journal_start/journal_stop which can block and take
5856 if (page_has_buffers(page
)) {
5857 if (!walk_page_buffers(NULL
, page_buffers(page
), 0, len
, NULL
,
5858 ext4_bh_unmapped
)) {
5865 * OK, we need to fill the hole... Do write_begin write_end
5866 * to do block allocation/reservation.We are not holding
5867 * inode.i__mutex here. That allow * parallel write_begin,
5868 * write_end call. lock_page prevent this from happening
5869 * on the same page though
5871 ret
= mapping
->a_ops
->write_begin(file
, mapping
, page_offset(page
),
5872 len
, AOP_FLAG_UNINTERRUPTIBLE
, &page
, &fsdata
);
5875 ret
= mapping
->a_ops
->write_end(file
, mapping
, page_offset(page
),
5876 len
, len
, page
, fsdata
);
5882 ret
= VM_FAULT_SIGBUS
;
5883 up_read(&inode
->i_alloc_sem
);