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, skip_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 * skip the page and move on.
2091 if (!page_has_buffers(page
)) {
2092 if (__block_write_begin(page
, 0, len
,
2093 noalloc_get_block_write
)) {
2101 bh
= page_bufs
= page_buffers(page
);
2106 if (map
&& (cur_logical
>= map
->m_lblk
) &&
2107 (cur_logical
<= (map
->m_lblk
+
2108 (map
->m_len
- 1)))) {
2109 if (buffer_delay(bh
)) {
2110 clear_buffer_delay(bh
);
2111 bh
->b_blocknr
= pblock
;
2113 if (buffer_unwritten(bh
) ||
2115 BUG_ON(bh
->b_blocknr
!= pblock
);
2116 if (map
->m_flags
& EXT4_MAP_UNINIT
)
2117 set_buffer_uninit(bh
);
2118 clear_buffer_unwritten(bh
);
2121 /* skip page if block allocation undone */
2122 if (buffer_delay(bh
) || buffer_unwritten(bh
))
2124 bh
= bh
->b_this_page
;
2125 block_start
+= bh
->b_size
;
2128 } while (bh
!= page_bufs
);
2134 /* mark the buffer_heads as dirty & uptodate */
2135 block_commit_write(page
, 0, len
);
2137 clear_page_dirty_for_io(page
);
2139 * Delalloc doesn't support data journalling,
2140 * but eventually maybe we'll lift this
2143 if (unlikely(journal_data
&& PageChecked(page
)))
2144 err
= __ext4_journalled_writepage(page
, len
);
2145 else if (test_opt(inode
->i_sb
, MBLK_IO_SUBMIT
))
2146 err
= ext4_bio_write_page(&io_submit
, page
,
2149 err
= block_write_full_page(page
,
2150 noalloc_get_block_write
, mpd
->wbc
);
2153 mpd
->pages_written
++;
2155 * In error case, we have to continue because
2156 * remaining pages are still locked
2161 pagevec_release(&pvec
);
2163 ext4_io_submit(&io_submit
);
2167 static void ext4_da_block_invalidatepages(struct mpage_da_data
*mpd
)
2171 struct pagevec pvec
;
2172 struct inode
*inode
= mpd
->inode
;
2173 struct address_space
*mapping
= inode
->i_mapping
;
2175 index
= mpd
->first_page
;
2176 end
= mpd
->next_page
- 1;
2177 while (index
<= end
) {
2178 nr_pages
= pagevec_lookup(&pvec
, mapping
, index
, PAGEVEC_SIZE
);
2181 for (i
= 0; i
< nr_pages
; i
++) {
2182 struct page
*page
= pvec
.pages
[i
];
2183 if (page
->index
> end
)
2185 BUG_ON(!PageLocked(page
));
2186 BUG_ON(PageWriteback(page
));
2187 block_invalidatepage(page
, 0);
2188 ClearPageUptodate(page
);
2191 index
= pvec
.pages
[nr_pages
- 1]->index
+ 1;
2192 pagevec_release(&pvec
);
2197 static void ext4_print_free_blocks(struct inode
*inode
)
2199 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
2200 printk(KERN_CRIT
"Total free blocks count %lld\n",
2201 ext4_count_free_blocks(inode
->i_sb
));
2202 printk(KERN_CRIT
"Free/Dirty block details\n");
2203 printk(KERN_CRIT
"free_blocks=%lld\n",
2204 (long long) percpu_counter_sum(&sbi
->s_freeblocks_counter
));
2205 printk(KERN_CRIT
"dirty_blocks=%lld\n",
2206 (long long) percpu_counter_sum(&sbi
->s_dirtyblocks_counter
));
2207 printk(KERN_CRIT
"Block reservation details\n");
2208 printk(KERN_CRIT
"i_reserved_data_blocks=%u\n",
2209 EXT4_I(inode
)->i_reserved_data_blocks
);
2210 printk(KERN_CRIT
"i_reserved_meta_blocks=%u\n",
2211 EXT4_I(inode
)->i_reserved_meta_blocks
);
2216 * mpage_da_map_and_submit - go through given space, map them
2217 * if necessary, and then submit them for I/O
2219 * @mpd - bh describing space
2221 * The function skips space we know is already mapped to disk blocks.
2224 static void mpage_da_map_and_submit(struct mpage_da_data
*mpd
)
2226 int err
, blks
, get_blocks_flags
;
2227 struct ext4_map_blocks map
, *mapp
= NULL
;
2228 sector_t next
= mpd
->b_blocknr
;
2229 unsigned max_blocks
= mpd
->b_size
>> mpd
->inode
->i_blkbits
;
2230 loff_t disksize
= EXT4_I(mpd
->inode
)->i_disksize
;
2231 handle_t
*handle
= NULL
;
2234 * If the blocks are mapped already, or we couldn't accumulate
2235 * any blocks, then proceed immediately to the submission stage.
2237 if ((mpd
->b_size
== 0) ||
2238 ((mpd
->b_state
& (1 << BH_Mapped
)) &&
2239 !(mpd
->b_state
& (1 << BH_Delay
)) &&
2240 !(mpd
->b_state
& (1 << BH_Unwritten
))))
2243 handle
= ext4_journal_current_handle();
2247 * Call ext4_map_blocks() to allocate any delayed allocation
2248 * blocks, or to convert an uninitialized extent to be
2249 * initialized (in the case where we have written into
2250 * one or more preallocated blocks).
2252 * We pass in the magic EXT4_GET_BLOCKS_DELALLOC_RESERVE to
2253 * indicate that we are on the delayed allocation path. This
2254 * affects functions in many different parts of the allocation
2255 * call path. This flag exists primarily because we don't
2256 * want to change *many* call functions, so ext4_map_blocks()
2257 * will set the EXT4_STATE_DELALLOC_RESERVED flag once the
2258 * inode's allocation semaphore is taken.
2260 * If the blocks in questions were delalloc blocks, set
2261 * EXT4_GET_BLOCKS_DELALLOC_RESERVE so the delalloc accounting
2262 * variables are updated after the blocks have been allocated.
2265 map
.m_len
= max_blocks
;
2266 get_blocks_flags
= EXT4_GET_BLOCKS_CREATE
;
2267 if (ext4_should_dioread_nolock(mpd
->inode
))
2268 get_blocks_flags
|= EXT4_GET_BLOCKS_IO_CREATE_EXT
;
2269 if (mpd
->b_state
& (1 << BH_Delay
))
2270 get_blocks_flags
|= EXT4_GET_BLOCKS_DELALLOC_RESERVE
;
2272 blks
= ext4_map_blocks(handle
, mpd
->inode
, &map
, get_blocks_flags
);
2274 struct super_block
*sb
= mpd
->inode
->i_sb
;
2278 * If get block returns EAGAIN or ENOSPC and there
2279 * appears to be free blocks we will just let
2280 * mpage_da_submit_io() unlock all of the pages.
2285 if (err
== -ENOSPC
&&
2286 ext4_count_free_blocks(sb
)) {
2292 * get block failure will cause us to loop in
2293 * writepages, because a_ops->writepage won't be able
2294 * to make progress. The page will be redirtied by
2295 * writepage and writepages will again try to write
2298 if (!(EXT4_SB(sb
)->s_mount_flags
& EXT4_MF_FS_ABORTED
)) {
2299 ext4_msg(sb
, KERN_CRIT
,
2300 "delayed block allocation failed for inode %lu "
2301 "at logical offset %llu with max blocks %zd "
2302 "with error %d", mpd
->inode
->i_ino
,
2303 (unsigned long long) next
,
2304 mpd
->b_size
>> mpd
->inode
->i_blkbits
, err
);
2305 ext4_msg(sb
, KERN_CRIT
,
2306 "This should not happen!! Data will be lost\n");
2308 ext4_print_free_blocks(mpd
->inode
);
2310 /* invalidate all the pages */
2311 ext4_da_block_invalidatepages(mpd
);
2313 /* Mark this page range as having been completed */
2320 if (map
.m_flags
& EXT4_MAP_NEW
) {
2321 struct block_device
*bdev
= mpd
->inode
->i_sb
->s_bdev
;
2324 for (i
= 0; i
< map
.m_len
; i
++)
2325 unmap_underlying_metadata(bdev
, map
.m_pblk
+ i
);
2328 if (ext4_should_order_data(mpd
->inode
)) {
2329 err
= ext4_jbd2_file_inode(handle
, mpd
->inode
);
2331 /* This only happens if the journal is aborted */
2336 * Update on-disk size along with block allocation.
2338 disksize
= ((loff_t
) next
+ blks
) << mpd
->inode
->i_blkbits
;
2339 if (disksize
> i_size_read(mpd
->inode
))
2340 disksize
= i_size_read(mpd
->inode
);
2341 if (disksize
> EXT4_I(mpd
->inode
)->i_disksize
) {
2342 ext4_update_i_disksize(mpd
->inode
, disksize
);
2343 err
= ext4_mark_inode_dirty(handle
, mpd
->inode
);
2345 ext4_error(mpd
->inode
->i_sb
,
2346 "Failed to mark inode %lu dirty",
2351 mpage_da_submit_io(mpd
, mapp
);
2355 #define BH_FLAGS ((1 << BH_Uptodate) | (1 << BH_Mapped) | \
2356 (1 << BH_Delay) | (1 << BH_Unwritten))
2359 * mpage_add_bh_to_extent - try to add one more block to extent of blocks
2361 * @mpd->lbh - extent of blocks
2362 * @logical - logical number of the block in the file
2363 * @bh - bh of the block (used to access block's state)
2365 * the function is used to collect contig. blocks in same state
2367 static void mpage_add_bh_to_extent(struct mpage_da_data
*mpd
,
2368 sector_t logical
, size_t b_size
,
2369 unsigned long b_state
)
2372 int nrblocks
= mpd
->b_size
>> mpd
->inode
->i_blkbits
;
2375 * XXX Don't go larger than mballoc is willing to allocate
2376 * This is a stopgap solution. We eventually need to fold
2377 * mpage_da_submit_io() into this function and then call
2378 * ext4_map_blocks() multiple times in a loop
2380 if (nrblocks
>= 8*1024*1024/mpd
->inode
->i_sb
->s_blocksize
)
2383 /* check if thereserved journal credits might overflow */
2384 if (!(ext4_test_inode_flag(mpd
->inode
, EXT4_INODE_EXTENTS
))) {
2385 if (nrblocks
>= EXT4_MAX_TRANS_DATA
) {
2387 * With non-extent format we are limited by the journal
2388 * credit available. Total credit needed to insert
2389 * nrblocks contiguous blocks is dependent on the
2390 * nrblocks. So limit nrblocks.
2393 } else if ((nrblocks
+ (b_size
>> mpd
->inode
->i_blkbits
)) >
2394 EXT4_MAX_TRANS_DATA
) {
2396 * Adding the new buffer_head would make it cross the
2397 * allowed limit for which we have journal credit
2398 * reserved. So limit the new bh->b_size
2400 b_size
= (EXT4_MAX_TRANS_DATA
- nrblocks
) <<
2401 mpd
->inode
->i_blkbits
;
2402 /* we will do mpage_da_submit_io in the next loop */
2406 * First block in the extent
2408 if (mpd
->b_size
== 0) {
2409 mpd
->b_blocknr
= logical
;
2410 mpd
->b_size
= b_size
;
2411 mpd
->b_state
= b_state
& BH_FLAGS
;
2415 next
= mpd
->b_blocknr
+ nrblocks
;
2417 * Can we merge the block to our big extent?
2419 if (logical
== next
&& (b_state
& BH_FLAGS
) == mpd
->b_state
) {
2420 mpd
->b_size
+= b_size
;
2426 * We couldn't merge the block to our extent, so we
2427 * need to flush current extent and start new one
2429 mpage_da_map_and_submit(mpd
);
2433 static int ext4_bh_delay_or_unwritten(handle_t
*handle
, struct buffer_head
*bh
)
2435 return (buffer_delay(bh
) || buffer_unwritten(bh
)) && buffer_dirty(bh
);
2439 * This is a special get_blocks_t callback which is used by
2440 * ext4_da_write_begin(). It will either return mapped block or
2441 * reserve space for a single block.
2443 * For delayed buffer_head we have BH_Mapped, BH_New, BH_Delay set.
2444 * We also have b_blocknr = -1 and b_bdev initialized properly
2446 * For unwritten buffer_head we have BH_Mapped, BH_New, BH_Unwritten set.
2447 * We also have b_blocknr = physicalblock mapping unwritten extent and b_bdev
2448 * initialized properly.
2450 static int ext4_da_get_block_prep(struct inode
*inode
, sector_t iblock
,
2451 struct buffer_head
*bh
, int create
)
2453 struct ext4_map_blocks map
;
2455 sector_t invalid_block
= ~((sector_t
) 0xffff);
2457 if (invalid_block
< ext4_blocks_count(EXT4_SB(inode
->i_sb
)->s_es
))
2460 BUG_ON(create
== 0);
2461 BUG_ON(bh
->b_size
!= inode
->i_sb
->s_blocksize
);
2463 map
.m_lblk
= iblock
;
2467 * first, we need to know whether the block is allocated already
2468 * preallocated blocks are unmapped but should treated
2469 * the same as allocated blocks.
2471 ret
= ext4_map_blocks(NULL
, inode
, &map
, 0);
2475 if (buffer_delay(bh
))
2476 return 0; /* Not sure this could or should happen */
2478 * XXX: __block_write_begin() unmaps passed block, is it OK?
2480 ret
= ext4_da_reserve_space(inode
, iblock
);
2482 /* not enough space to reserve */
2485 map_bh(bh
, inode
->i_sb
, invalid_block
);
2487 set_buffer_delay(bh
);
2491 map_bh(bh
, inode
->i_sb
, map
.m_pblk
);
2492 bh
->b_state
= (bh
->b_state
& ~EXT4_MAP_FLAGS
) | map
.m_flags
;
2494 if (buffer_unwritten(bh
)) {
2495 /* A delayed write to unwritten bh should be marked
2496 * new and mapped. Mapped ensures that we don't do
2497 * get_block multiple times when we write to the same
2498 * offset and new ensures that we do proper zero out
2499 * for partial write.
2502 set_buffer_mapped(bh
);
2508 * This function is used as a standard get_block_t calback function
2509 * when there is no desire to allocate any blocks. It is used as a
2510 * callback function for block_write_begin() and block_write_full_page().
2511 * These functions should only try to map a single block at a time.
2513 * Since this function doesn't do block allocations even if the caller
2514 * requests it by passing in create=1, it is critically important that
2515 * any caller checks to make sure that any buffer heads are returned
2516 * by this function are either all already mapped or marked for
2517 * delayed allocation before calling block_write_full_page(). Otherwise,
2518 * b_blocknr could be left unitialized, and the page write functions will
2519 * be taken by surprise.
2521 static int noalloc_get_block_write(struct inode
*inode
, sector_t iblock
,
2522 struct buffer_head
*bh_result
, int create
)
2524 BUG_ON(bh_result
->b_size
!= inode
->i_sb
->s_blocksize
);
2525 return _ext4_get_block(inode
, iblock
, bh_result
, 0);
2528 static int bget_one(handle_t
*handle
, struct buffer_head
*bh
)
2534 static int bput_one(handle_t
*handle
, struct buffer_head
*bh
)
2540 static int __ext4_journalled_writepage(struct page
*page
,
2543 struct address_space
*mapping
= page
->mapping
;
2544 struct inode
*inode
= mapping
->host
;
2545 struct buffer_head
*page_bufs
;
2546 handle_t
*handle
= NULL
;
2550 ClearPageChecked(page
);
2551 page_bufs
= page_buffers(page
);
2553 walk_page_buffers(handle
, page_bufs
, 0, len
, NULL
, bget_one
);
2554 /* As soon as we unlock the page, it can go away, but we have
2555 * references to buffers so we are safe */
2558 handle
= ext4_journal_start(inode
, ext4_writepage_trans_blocks(inode
));
2559 if (IS_ERR(handle
)) {
2560 ret
= PTR_ERR(handle
);
2564 ret
= walk_page_buffers(handle
, page_bufs
, 0, len
, NULL
,
2565 do_journal_get_write_access
);
2567 err
= walk_page_buffers(handle
, page_bufs
, 0, len
, NULL
,
2571 err
= ext4_journal_stop(handle
);
2575 walk_page_buffers(handle
, page_bufs
, 0, len
, NULL
, bput_one
);
2576 ext4_set_inode_state(inode
, EXT4_STATE_JDATA
);
2581 static int ext4_set_bh_endio(struct buffer_head
*bh
, struct inode
*inode
);
2582 static void ext4_end_io_buffer_write(struct buffer_head
*bh
, int uptodate
);
2585 * Note that we don't need to start a transaction unless we're journaling data
2586 * because we should have holes filled from ext4_page_mkwrite(). We even don't
2587 * need to file the inode to the transaction's list in ordered mode because if
2588 * we are writing back data added by write(), the inode is already there and if
2589 * we are writing back data modified via mmap(), noone guarantees in which
2590 * transaction the data will hit the disk. In case we are journaling data, we
2591 * cannot start transaction directly because transaction start ranks above page
2592 * lock so we have to do some magic.
2594 * This function can get called via...
2595 * - ext4_da_writepages after taking page lock (have journal handle)
2596 * - journal_submit_inode_data_buffers (no journal handle)
2597 * - shrink_page_list via pdflush (no journal handle)
2598 * - grab_page_cache when doing write_begin (have journal handle)
2600 * We don't do any block allocation in this function. If we have page with
2601 * multiple blocks we need to write those buffer_heads that are mapped. This
2602 * is important for mmaped based write. So if we do with blocksize 1K
2603 * truncate(f, 1024);
2604 * a = mmap(f, 0, 4096);
2606 * truncate(f, 4096);
2607 * we have in the page first buffer_head mapped via page_mkwrite call back
2608 * but other bufer_heads would be unmapped but dirty(dirty done via the
2609 * do_wp_page). So writepage should write the first block. If we modify
2610 * the mmap area beyond 1024 we will again get a page_fault and the
2611 * page_mkwrite callback will do the block allocation and mark the
2612 * buffer_heads mapped.
2614 * We redirty the page if we have any buffer_heads that is either delay or
2615 * unwritten in the page.
2617 * We can get recursively called as show below.
2619 * ext4_writepage() -> kmalloc() -> __alloc_pages() -> page_launder() ->
2622 * But since we don't do any block allocation we should not deadlock.
2623 * Page also have the dirty flag cleared so we don't get recurive page_lock.
2625 static int ext4_writepage(struct page
*page
,
2626 struct writeback_control
*wbc
)
2628 int ret
= 0, commit_write
= 0;
2631 struct buffer_head
*page_bufs
= NULL
;
2632 struct inode
*inode
= page
->mapping
->host
;
2634 trace_ext4_writepage(inode
, page
);
2635 size
= i_size_read(inode
);
2636 if (page
->index
== size
>> PAGE_CACHE_SHIFT
)
2637 len
= size
& ~PAGE_CACHE_MASK
;
2639 len
= PAGE_CACHE_SIZE
;
2642 * If the page does not have buffers (for whatever reason),
2643 * try to create them using __block_write_begin. If this
2644 * fails, redirty the page and move on.
2646 if (!page_has_buffers(page
)) {
2647 if (__block_write_begin(page
, 0, len
,
2648 noalloc_get_block_write
)) {
2650 redirty_page_for_writepage(wbc
, page
);
2656 page_bufs
= page_buffers(page
);
2657 if (walk_page_buffers(NULL
, page_bufs
, 0, len
, NULL
,
2658 ext4_bh_delay_or_unwritten
)) {
2660 * We don't want to do block allocation, so redirty
2661 * the page and return. We may reach here when we do
2662 * a journal commit via journal_submit_inode_data_buffers.
2663 * We can also reach here via shrink_page_list
2668 /* now mark the buffer_heads as dirty and uptodate */
2669 block_commit_write(page
, 0, len
);
2671 if (PageChecked(page
) && ext4_should_journal_data(inode
))
2673 * It's mmapped pagecache. Add buffers and journal it. There
2674 * doesn't seem much point in redirtying the page here.
2676 return __ext4_journalled_writepage(page
, len
);
2678 if (buffer_uninit(page_bufs
)) {
2679 ext4_set_bh_endio(page_bufs
, inode
);
2680 ret
= block_write_full_page_endio(page
, noalloc_get_block_write
,
2681 wbc
, ext4_end_io_buffer_write
);
2683 ret
= block_write_full_page(page
, noalloc_get_block_write
,
2690 * This is called via ext4_da_writepages() to
2691 * calulate the total number of credits to reserve to fit
2692 * a single extent allocation into a single transaction,
2693 * ext4_da_writpeages() will loop calling this before
2694 * the block allocation.
2697 static int ext4_da_writepages_trans_blocks(struct inode
*inode
)
2699 int max_blocks
= EXT4_I(inode
)->i_reserved_data_blocks
;
2702 * With non-extent format the journal credit needed to
2703 * insert nrblocks contiguous block is dependent on
2704 * number of contiguous block. So we will limit
2705 * number of contiguous block to a sane value
2707 if (!(ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
)) &&
2708 (max_blocks
> EXT4_MAX_TRANS_DATA
))
2709 max_blocks
= EXT4_MAX_TRANS_DATA
;
2711 return ext4_chunk_trans_blocks(inode
, max_blocks
);
2715 * write_cache_pages_da - walk the list of dirty pages of the given
2716 * address space and accumulate pages that need writing, and call
2717 * mpage_da_map_and_submit to map the pages and then write them.
2719 static int write_cache_pages_da(struct address_space
*mapping
,
2720 struct writeback_control
*wbc
,
2721 struct mpage_da_data
*mpd
,
2722 pgoff_t
*done_index
)
2724 struct buffer_head
*bh
, *head
;
2725 struct inode
*inode
= mpd
->inode
;
2726 struct pagevec pvec
;
2727 unsigned int nr_pages
;
2730 long nr_to_write
= wbc
->nr_to_write
;
2731 int i
, tag
, ret
= 0;
2733 pagevec_init(&pvec
, 0);
2734 index
= wbc
->range_start
>> PAGE_CACHE_SHIFT
;
2735 end
= wbc
->range_end
>> PAGE_CACHE_SHIFT
;
2737 if (wbc
->sync_mode
== WB_SYNC_ALL
)
2738 tag
= PAGECACHE_TAG_TOWRITE
;
2740 tag
= PAGECACHE_TAG_DIRTY
;
2742 *done_index
= index
;
2743 while (index
<= end
) {
2744 nr_pages
= pagevec_lookup_tag(&pvec
, mapping
, &index
, tag
,
2745 min(end
- index
, (pgoff_t
)PAGEVEC_SIZE
-1) + 1);
2749 for (i
= 0; i
< nr_pages
; i
++) {
2750 struct page
*page
= pvec
.pages
[i
];
2753 * At this point, the page may be truncated or
2754 * invalidated (changing page->mapping to NULL), or
2755 * even swizzled back from swapper_space to tmpfs file
2756 * mapping. However, page->index will not change
2757 * because we have a reference on the page.
2759 if (page
->index
> end
)
2762 *done_index
= page
->index
+ 1;
2765 * If we can't merge this page, and we have
2766 * accumulated an contiguous region, write it
2768 if ((mpd
->next_page
!= page
->index
) &&
2769 (mpd
->next_page
!= mpd
->first_page
)) {
2770 mpage_da_map_and_submit(mpd
);
2771 goto ret_extent_tail
;
2777 * If the page is no longer dirty, or its
2778 * mapping no longer corresponds to inode we
2779 * are writing (which means it has been
2780 * truncated or invalidated), or the page is
2781 * already under writeback and we are not
2782 * doing a data integrity writeback, skip the page
2784 if (!PageDirty(page
) ||
2785 (PageWriteback(page
) &&
2786 (wbc
->sync_mode
== WB_SYNC_NONE
)) ||
2787 unlikely(page
->mapping
!= mapping
)) {
2792 if (PageWriteback(page
))
2793 wait_on_page_writeback(page
);
2795 BUG_ON(PageWriteback(page
));
2797 if (mpd
->next_page
!= page
->index
) {
2799 * Start next extent of pages and blocks
2801 mpd
->first_page
= page
->index
;
2807 mpd
->next_page
= page
->index
+ 1;
2808 logical
= (sector_t
) page
->index
<<
2809 (PAGE_CACHE_SHIFT
- inode
->i_blkbits
);
2811 if (!page_has_buffers(page
)) {
2812 mpage_add_bh_to_extent(mpd
, logical
,
2814 (1 << BH_Dirty
) | (1 << BH_Uptodate
));
2816 goto ret_extent_tail
;
2819 * Page with regular buffer heads,
2820 * just add all dirty ones
2822 head
= page_buffers(page
);
2825 BUG_ON(buffer_locked(bh
));
2827 * We need to try to allocate
2828 * unmapped blocks in the same page.
2829 * Otherwise we won't make progress
2830 * with the page in ext4_writepage
2832 if (ext4_bh_delay_or_unwritten(NULL
, bh
)) {
2833 mpage_add_bh_to_extent(mpd
, logical
,
2837 goto ret_extent_tail
;
2838 } else if (buffer_dirty(bh
) && (buffer_mapped(bh
))) {
2840 * mapped dirty buffer. We need
2841 * to update the b_state
2842 * because we look at b_state
2843 * in mpage_da_map_blocks. We
2844 * don't update b_size because
2845 * if we find an unmapped
2846 * buffer_head later we need to
2847 * use the b_state flag of that
2850 if (mpd
->b_size
== 0)
2851 mpd
->b_state
= bh
->b_state
& BH_FLAGS
;
2854 } while ((bh
= bh
->b_this_page
) != head
);
2857 if (nr_to_write
> 0) {
2859 if (nr_to_write
== 0 &&
2860 wbc
->sync_mode
== WB_SYNC_NONE
)
2862 * We stop writing back only if we are
2863 * not doing integrity sync. In case of
2864 * integrity sync we have to keep going
2865 * because someone may be concurrently
2866 * dirtying pages, and we might have
2867 * synced a lot of newly appeared dirty
2868 * pages, but have not synced all of the
2874 pagevec_release(&pvec
);
2879 ret
= MPAGE_DA_EXTENT_TAIL
;
2881 pagevec_release(&pvec
);
2887 static int ext4_da_writepages(struct address_space
*mapping
,
2888 struct writeback_control
*wbc
)
2891 int range_whole
= 0;
2892 handle_t
*handle
= NULL
;
2893 struct mpage_da_data mpd
;
2894 struct inode
*inode
= mapping
->host
;
2895 int pages_written
= 0;
2896 unsigned int max_pages
;
2897 int range_cyclic
, cycled
= 1, io_done
= 0;
2898 int needed_blocks
, ret
= 0;
2899 long desired_nr_to_write
, nr_to_writebump
= 0;
2900 loff_t range_start
= wbc
->range_start
;
2901 struct ext4_sb_info
*sbi
= EXT4_SB(mapping
->host
->i_sb
);
2902 pgoff_t done_index
= 0;
2905 trace_ext4_da_writepages(inode
, wbc
);
2908 * No pages to write? This is mainly a kludge to avoid starting
2909 * a transaction for special inodes like journal inode on last iput()
2910 * because that could violate lock ordering on umount
2912 if (!mapping
->nrpages
|| !mapping_tagged(mapping
, PAGECACHE_TAG_DIRTY
))
2916 * If the filesystem has aborted, it is read-only, so return
2917 * right away instead of dumping stack traces later on that
2918 * will obscure the real source of the problem. We test
2919 * EXT4_MF_FS_ABORTED instead of sb->s_flag's MS_RDONLY because
2920 * the latter could be true if the filesystem is mounted
2921 * read-only, and in that case, ext4_da_writepages should
2922 * *never* be called, so if that ever happens, we would want
2925 if (unlikely(sbi
->s_mount_flags
& EXT4_MF_FS_ABORTED
))
2928 if (wbc
->range_start
== 0 && wbc
->range_end
== LLONG_MAX
)
2931 range_cyclic
= wbc
->range_cyclic
;
2932 if (wbc
->range_cyclic
) {
2933 index
= mapping
->writeback_index
;
2936 wbc
->range_start
= index
<< PAGE_CACHE_SHIFT
;
2937 wbc
->range_end
= LLONG_MAX
;
2938 wbc
->range_cyclic
= 0;
2941 index
= wbc
->range_start
>> PAGE_CACHE_SHIFT
;
2942 end
= wbc
->range_end
>> PAGE_CACHE_SHIFT
;
2946 * This works around two forms of stupidity. The first is in
2947 * the writeback code, which caps the maximum number of pages
2948 * written to be 1024 pages. This is wrong on multiple
2949 * levels; different architectues have a different page size,
2950 * which changes the maximum amount of data which gets
2951 * written. Secondly, 4 megabytes is way too small. XFS
2952 * forces this value to be 16 megabytes by multiplying
2953 * nr_to_write parameter by four, and then relies on its
2954 * allocator to allocate larger extents to make them
2955 * contiguous. Unfortunately this brings us to the second
2956 * stupidity, which is that ext4's mballoc code only allocates
2957 * at most 2048 blocks. So we force contiguous writes up to
2958 * the number of dirty blocks in the inode, or
2959 * sbi->max_writeback_mb_bump whichever is smaller.
2961 max_pages
= sbi
->s_max_writeback_mb_bump
<< (20 - PAGE_CACHE_SHIFT
);
2962 if (!range_cyclic
&& range_whole
) {
2963 if (wbc
->nr_to_write
== LONG_MAX
)
2964 desired_nr_to_write
= wbc
->nr_to_write
;
2966 desired_nr_to_write
= wbc
->nr_to_write
* 8;
2968 desired_nr_to_write
= ext4_num_dirty_pages(inode
, index
,
2970 if (desired_nr_to_write
> max_pages
)
2971 desired_nr_to_write
= max_pages
;
2973 if (wbc
->nr_to_write
< desired_nr_to_write
) {
2974 nr_to_writebump
= desired_nr_to_write
- wbc
->nr_to_write
;
2975 wbc
->nr_to_write
= desired_nr_to_write
;
2979 mpd
.inode
= mapping
->host
;
2982 if (wbc
->sync_mode
== WB_SYNC_ALL
)
2983 tag_pages_for_writeback(mapping
, index
, end
);
2985 while (!ret
&& wbc
->nr_to_write
> 0) {
2988 * we insert one extent at a time. So we need
2989 * credit needed for single extent allocation.
2990 * journalled mode is currently not supported
2993 BUG_ON(ext4_should_journal_data(inode
));
2994 needed_blocks
= ext4_da_writepages_trans_blocks(inode
);
2996 /* start a new transaction*/
2997 handle
= ext4_journal_start(inode
, needed_blocks
);
2998 if (IS_ERR(handle
)) {
2999 ret
= PTR_ERR(handle
);
3000 ext4_msg(inode
->i_sb
, KERN_CRIT
, "%s: jbd2_start: "
3001 "%ld pages, ino %lu; err %d", __func__
,
3002 wbc
->nr_to_write
, inode
->i_ino
, ret
);
3003 goto out_writepages
;
3007 * Now call write_cache_pages_da() to find the next
3008 * contiguous region of logical blocks that need
3009 * blocks to be allocated by ext4 and submit them.
3017 mpd
.pages_written
= 0;
3019 ret
= write_cache_pages_da(mapping
, wbc
, &mpd
, &done_index
);
3021 * If we have a contiguous extent of pages and we
3022 * haven't done the I/O yet, map the blocks and submit
3025 if (!mpd
.io_done
&& mpd
.next_page
!= mpd
.first_page
) {
3026 mpage_da_map_and_submit(&mpd
);
3027 ret
= MPAGE_DA_EXTENT_TAIL
;
3029 trace_ext4_da_write_pages(inode
, &mpd
);
3030 wbc
->nr_to_write
-= mpd
.pages_written
;
3032 ext4_journal_stop(handle
);
3034 if ((mpd
.retval
== -ENOSPC
) && sbi
->s_journal
) {
3035 /* commit the transaction which would
3036 * free blocks released in the transaction
3039 jbd2_journal_force_commit_nested(sbi
->s_journal
);
3041 } else if (ret
== MPAGE_DA_EXTENT_TAIL
) {
3043 * got one extent now try with
3046 pages_written
+= mpd
.pages_written
;
3049 } else if (wbc
->nr_to_write
)
3051 * There is no more writeout needed
3052 * or we requested for a noblocking writeout
3053 * and we found the device congested
3057 if (!io_done
&& !cycled
) {
3060 wbc
->range_start
= index
<< PAGE_CACHE_SHIFT
;
3061 wbc
->range_end
= mapping
->writeback_index
- 1;
3066 wbc
->range_cyclic
= range_cyclic
;
3067 if (wbc
->range_cyclic
|| (range_whole
&& wbc
->nr_to_write
> 0))
3069 * set the writeback_index so that range_cyclic
3070 * mode will write it back later
3072 mapping
->writeback_index
= done_index
;
3075 wbc
->nr_to_write
-= nr_to_writebump
;
3076 wbc
->range_start
= range_start
;
3077 trace_ext4_da_writepages_result(inode
, wbc
, ret
, pages_written
);
3081 #define FALL_BACK_TO_NONDELALLOC 1
3082 static int ext4_nonda_switch(struct super_block
*sb
)
3084 s64 free_blocks
, dirty_blocks
;
3085 struct ext4_sb_info
*sbi
= EXT4_SB(sb
);
3088 * switch to non delalloc mode if we are running low
3089 * on free block. The free block accounting via percpu
3090 * counters can get slightly wrong with percpu_counter_batch getting
3091 * accumulated on each CPU without updating global counters
3092 * Delalloc need an accurate free block accounting. So switch
3093 * to non delalloc when we are near to error range.
3095 free_blocks
= percpu_counter_read_positive(&sbi
->s_freeblocks_counter
);
3096 dirty_blocks
= percpu_counter_read_positive(&sbi
->s_dirtyblocks_counter
);
3097 if (2 * free_blocks
< 3 * dirty_blocks
||
3098 free_blocks
< (dirty_blocks
+ EXT4_FREEBLOCKS_WATERMARK
)) {
3100 * free block count is less than 150% of dirty blocks
3101 * or free blocks is less than watermark
3106 * Even if we don't switch but are nearing capacity,
3107 * start pushing delalloc when 1/2 of free blocks are dirty.
3109 if (free_blocks
< 2 * dirty_blocks
)
3110 writeback_inodes_sb_if_idle(sb
);
3115 static int ext4_da_write_begin(struct file
*file
, struct address_space
*mapping
,
3116 loff_t pos
, unsigned len
, unsigned flags
,
3117 struct page
**pagep
, void **fsdata
)
3119 int ret
, retries
= 0;
3122 struct inode
*inode
= mapping
->host
;
3125 index
= pos
>> PAGE_CACHE_SHIFT
;
3127 if (ext4_nonda_switch(inode
->i_sb
)) {
3128 *fsdata
= (void *)FALL_BACK_TO_NONDELALLOC
;
3129 return ext4_write_begin(file
, mapping
, pos
,
3130 len
, flags
, pagep
, fsdata
);
3132 *fsdata
= (void *)0;
3133 trace_ext4_da_write_begin(inode
, pos
, len
, flags
);
3136 * With delayed allocation, we don't log the i_disksize update
3137 * if there is delayed block allocation. But we still need
3138 * to journalling the i_disksize update if writes to the end
3139 * of file which has an already mapped buffer.
3141 handle
= ext4_journal_start(inode
, 1);
3142 if (IS_ERR(handle
)) {
3143 ret
= PTR_ERR(handle
);
3146 /* We cannot recurse into the filesystem as the transaction is already
3148 flags
|= AOP_FLAG_NOFS
;
3150 page
= grab_cache_page_write_begin(mapping
, index
, flags
);
3152 ext4_journal_stop(handle
);
3158 ret
= __block_write_begin(page
, pos
, len
, ext4_da_get_block_prep
);
3161 ext4_journal_stop(handle
);
3162 page_cache_release(page
);
3164 * block_write_begin may have instantiated a few blocks
3165 * outside i_size. Trim these off again. Don't need
3166 * i_size_read because we hold i_mutex.
3168 if (pos
+ len
> inode
->i_size
)
3169 ext4_truncate_failed_write(inode
);
3172 if (ret
== -ENOSPC
&& ext4_should_retry_alloc(inode
->i_sb
, &retries
))
3179 * Check if we should update i_disksize
3180 * when write to the end of file but not require block allocation
3182 static int ext4_da_should_update_i_disksize(struct page
*page
,
3183 unsigned long offset
)
3185 struct buffer_head
*bh
;
3186 struct inode
*inode
= page
->mapping
->host
;
3190 bh
= page_buffers(page
);
3191 idx
= offset
>> inode
->i_blkbits
;
3193 for (i
= 0; i
< idx
; i
++)
3194 bh
= bh
->b_this_page
;
3196 if (!buffer_mapped(bh
) || (buffer_delay(bh
)) || buffer_unwritten(bh
))
3201 static int ext4_da_write_end(struct file
*file
,
3202 struct address_space
*mapping
,
3203 loff_t pos
, unsigned len
, unsigned copied
,
3204 struct page
*page
, void *fsdata
)
3206 struct inode
*inode
= mapping
->host
;
3208 handle_t
*handle
= ext4_journal_current_handle();
3210 unsigned long start
, end
;
3211 int write_mode
= (int)(unsigned long)fsdata
;
3213 if (write_mode
== FALL_BACK_TO_NONDELALLOC
) {
3214 if (ext4_should_order_data(inode
)) {
3215 return ext4_ordered_write_end(file
, mapping
, pos
,
3216 len
, copied
, page
, fsdata
);
3217 } else if (ext4_should_writeback_data(inode
)) {
3218 return ext4_writeback_write_end(file
, mapping
, pos
,
3219 len
, copied
, page
, fsdata
);
3225 trace_ext4_da_write_end(inode
, pos
, len
, copied
);
3226 start
= pos
& (PAGE_CACHE_SIZE
- 1);
3227 end
= start
+ copied
- 1;
3230 * generic_write_end() will run mark_inode_dirty() if i_size
3231 * changes. So let's piggyback the i_disksize mark_inode_dirty
3235 new_i_size
= pos
+ copied
;
3236 if (new_i_size
> EXT4_I(inode
)->i_disksize
) {
3237 if (ext4_da_should_update_i_disksize(page
, end
)) {
3238 down_write(&EXT4_I(inode
)->i_data_sem
);
3239 if (new_i_size
> EXT4_I(inode
)->i_disksize
) {
3241 * Updating i_disksize when extending file
3242 * without needing block allocation
3244 if (ext4_should_order_data(inode
))
3245 ret
= ext4_jbd2_file_inode(handle
,
3248 EXT4_I(inode
)->i_disksize
= new_i_size
;
3250 up_write(&EXT4_I(inode
)->i_data_sem
);
3251 /* We need to mark inode dirty even if
3252 * new_i_size is less that inode->i_size
3253 * bu greater than i_disksize.(hint delalloc)
3255 ext4_mark_inode_dirty(handle
, inode
);
3258 ret2
= generic_write_end(file
, mapping
, pos
, len
, copied
,
3263 ret2
= ext4_journal_stop(handle
);
3267 return ret
? ret
: copied
;
3270 static void ext4_da_invalidatepage(struct page
*page
, unsigned long offset
)
3273 * Drop reserved blocks
3275 BUG_ON(!PageLocked(page
));
3276 if (!page_has_buffers(page
))
3279 ext4_da_page_release_reservation(page
, offset
);
3282 ext4_invalidatepage(page
, offset
);
3288 * Force all delayed allocation blocks to be allocated for a given inode.
3290 int ext4_alloc_da_blocks(struct inode
*inode
)
3292 trace_ext4_alloc_da_blocks(inode
);
3294 if (!EXT4_I(inode
)->i_reserved_data_blocks
&&
3295 !EXT4_I(inode
)->i_reserved_meta_blocks
)
3299 * We do something simple for now. The filemap_flush() will
3300 * also start triggering a write of the data blocks, which is
3301 * not strictly speaking necessary (and for users of
3302 * laptop_mode, not even desirable). However, to do otherwise
3303 * would require replicating code paths in:
3305 * ext4_da_writepages() ->
3306 * write_cache_pages() ---> (via passed in callback function)
3307 * __mpage_da_writepage() -->
3308 * mpage_add_bh_to_extent()
3309 * mpage_da_map_blocks()
3311 * The problem is that write_cache_pages(), located in
3312 * mm/page-writeback.c, marks pages clean in preparation for
3313 * doing I/O, which is not desirable if we're not planning on
3316 * We could call write_cache_pages(), and then redirty all of
3317 * the pages by calling redirty_page_for_writepage() but that
3318 * would be ugly in the extreme. So instead we would need to
3319 * replicate parts of the code in the above functions,
3320 * simplifying them becuase we wouldn't actually intend to
3321 * write out the pages, but rather only collect contiguous
3322 * logical block extents, call the multi-block allocator, and
3323 * then update the buffer heads with the block allocations.
3325 * For now, though, we'll cheat by calling filemap_flush(),
3326 * which will map the blocks, and start the I/O, but not
3327 * actually wait for the I/O to complete.
3329 return filemap_flush(inode
->i_mapping
);
3333 * bmap() is special. It gets used by applications such as lilo and by
3334 * the swapper to find the on-disk block of a specific piece of data.
3336 * Naturally, this is dangerous if the block concerned is still in the
3337 * journal. If somebody makes a swapfile on an ext4 data-journaling
3338 * filesystem and enables swap, then they may get a nasty shock when the
3339 * data getting swapped to that swapfile suddenly gets overwritten by
3340 * the original zero's written out previously to the journal and
3341 * awaiting writeback in the kernel's buffer cache.
3343 * So, if we see any bmap calls here on a modified, data-journaled file,
3344 * take extra steps to flush any blocks which might be in the cache.
3346 static sector_t
ext4_bmap(struct address_space
*mapping
, sector_t block
)
3348 struct inode
*inode
= mapping
->host
;
3352 if (mapping_tagged(mapping
, PAGECACHE_TAG_DIRTY
) &&
3353 test_opt(inode
->i_sb
, DELALLOC
)) {
3355 * With delalloc we want to sync the file
3356 * so that we can make sure we allocate
3359 filemap_write_and_wait(mapping
);
3362 if (EXT4_JOURNAL(inode
) &&
3363 ext4_test_inode_state(inode
, EXT4_STATE_JDATA
)) {
3365 * This is a REALLY heavyweight approach, but the use of
3366 * bmap on dirty files is expected to be extremely rare:
3367 * only if we run lilo or swapon on a freshly made file
3368 * do we expect this to happen.
3370 * (bmap requires CAP_SYS_RAWIO so this does not
3371 * represent an unprivileged user DOS attack --- we'd be
3372 * in trouble if mortal users could trigger this path at
3375 * NB. EXT4_STATE_JDATA is not set on files other than
3376 * regular files. If somebody wants to bmap a directory
3377 * or symlink and gets confused because the buffer
3378 * hasn't yet been flushed to disk, they deserve
3379 * everything they get.
3382 ext4_clear_inode_state(inode
, EXT4_STATE_JDATA
);
3383 journal
= EXT4_JOURNAL(inode
);
3384 jbd2_journal_lock_updates(journal
);
3385 err
= jbd2_journal_flush(journal
);
3386 jbd2_journal_unlock_updates(journal
);
3392 return generic_block_bmap(mapping
, block
, ext4_get_block
);
3395 static int ext4_readpage(struct file
*file
, struct page
*page
)
3397 return mpage_readpage(page
, ext4_get_block
);
3401 ext4_readpages(struct file
*file
, struct address_space
*mapping
,
3402 struct list_head
*pages
, unsigned nr_pages
)
3404 return mpage_readpages(mapping
, pages
, nr_pages
, ext4_get_block
);
3407 static void ext4_invalidatepage_free_endio(struct page
*page
, unsigned long offset
)
3409 struct buffer_head
*head
, *bh
;
3410 unsigned int curr_off
= 0;
3412 if (!page_has_buffers(page
))
3414 head
= bh
= page_buffers(page
);
3416 if (offset
<= curr_off
&& test_clear_buffer_uninit(bh
)
3418 ext4_free_io_end(bh
->b_private
);
3419 bh
->b_private
= NULL
;
3420 bh
->b_end_io
= NULL
;
3422 curr_off
= curr_off
+ bh
->b_size
;
3423 bh
= bh
->b_this_page
;
3424 } while (bh
!= head
);
3427 static void ext4_invalidatepage(struct page
*page
, unsigned long offset
)
3429 journal_t
*journal
= EXT4_JOURNAL(page
->mapping
->host
);
3432 * free any io_end structure allocated for buffers to be discarded
3434 if (ext4_should_dioread_nolock(page
->mapping
->host
))
3435 ext4_invalidatepage_free_endio(page
, offset
);
3437 * If it's a full truncate we just forget about the pending dirtying
3440 ClearPageChecked(page
);
3443 jbd2_journal_invalidatepage(journal
, page
, offset
);
3445 block_invalidatepage(page
, offset
);
3448 static int ext4_releasepage(struct page
*page
, gfp_t wait
)
3450 journal_t
*journal
= EXT4_JOURNAL(page
->mapping
->host
);
3452 WARN_ON(PageChecked(page
));
3453 if (!page_has_buffers(page
))
3456 return jbd2_journal_try_to_free_buffers(journal
, page
, wait
);
3458 return try_to_free_buffers(page
);
3462 * O_DIRECT for ext3 (or indirect map) based files
3464 * If the O_DIRECT write will extend the file then add this inode to the
3465 * orphan list. So recovery will truncate it back to the original size
3466 * if the machine crashes during the write.
3468 * If the O_DIRECT write is intantiating holes inside i_size and the machine
3469 * crashes then stale disk data _may_ be exposed inside the file. But current
3470 * VFS code falls back into buffered path in that case so we are safe.
3472 static ssize_t
ext4_ind_direct_IO(int rw
, struct kiocb
*iocb
,
3473 const struct iovec
*iov
, loff_t offset
,
3474 unsigned long nr_segs
)
3476 struct file
*file
= iocb
->ki_filp
;
3477 struct inode
*inode
= file
->f_mapping
->host
;
3478 struct ext4_inode_info
*ei
= EXT4_I(inode
);
3482 size_t count
= iov_length(iov
, nr_segs
);
3486 loff_t final_size
= offset
+ count
;
3488 if (final_size
> inode
->i_size
) {
3489 /* Credits for sb + inode write */
3490 handle
= ext4_journal_start(inode
, 2);
3491 if (IS_ERR(handle
)) {
3492 ret
= PTR_ERR(handle
);
3495 ret
= ext4_orphan_add(handle
, inode
);
3497 ext4_journal_stop(handle
);
3501 ei
->i_disksize
= inode
->i_size
;
3502 ext4_journal_stop(handle
);
3507 if (rw
== READ
&& ext4_should_dioread_nolock(inode
))
3508 ret
= __blockdev_direct_IO(rw
, iocb
, inode
,
3509 inode
->i_sb
->s_bdev
, iov
,
3511 ext4_get_block
, NULL
, NULL
, 0);
3513 ret
= blockdev_direct_IO(rw
, iocb
, inode
,
3514 inode
->i_sb
->s_bdev
, iov
,
3516 ext4_get_block
, NULL
);
3518 if (unlikely((rw
& WRITE
) && ret
< 0)) {
3519 loff_t isize
= i_size_read(inode
);
3520 loff_t end
= offset
+ iov_length(iov
, nr_segs
);
3523 vmtruncate(inode
, isize
);
3526 if (ret
== -ENOSPC
&& ext4_should_retry_alloc(inode
->i_sb
, &retries
))
3532 /* Credits for sb + inode write */
3533 handle
= ext4_journal_start(inode
, 2);
3534 if (IS_ERR(handle
)) {
3535 /* This is really bad luck. We've written the data
3536 * but cannot extend i_size. Bail out and pretend
3537 * the write failed... */
3538 ret
= PTR_ERR(handle
);
3540 ext4_orphan_del(NULL
, inode
);
3545 ext4_orphan_del(handle
, inode
);
3547 loff_t end
= offset
+ ret
;
3548 if (end
> inode
->i_size
) {
3549 ei
->i_disksize
= end
;
3550 i_size_write(inode
, end
);
3552 * We're going to return a positive `ret'
3553 * here due to non-zero-length I/O, so there's
3554 * no way of reporting error returns from
3555 * ext4_mark_inode_dirty() to userspace. So
3558 ext4_mark_inode_dirty(handle
, inode
);
3561 err
= ext4_journal_stop(handle
);
3570 * ext4_get_block used when preparing for a DIO write or buffer write.
3571 * We allocate an uinitialized extent if blocks haven't been allocated.
3572 * The extent will be converted to initialized after the IO is complete.
3574 static int ext4_get_block_write(struct inode
*inode
, sector_t iblock
,
3575 struct buffer_head
*bh_result
, int create
)
3577 ext4_debug("ext4_get_block_write: inode %lu, create flag %d\n",
3578 inode
->i_ino
, create
);
3579 return _ext4_get_block(inode
, iblock
, bh_result
,
3580 EXT4_GET_BLOCKS_IO_CREATE_EXT
);
3583 static void ext4_end_io_dio(struct kiocb
*iocb
, loff_t offset
,
3584 ssize_t size
, void *private, int ret
,
3587 ext4_io_end_t
*io_end
= iocb
->private;
3588 struct workqueue_struct
*wq
;
3589 unsigned long flags
;
3590 struct ext4_inode_info
*ei
;
3592 /* if not async direct IO or dio with 0 bytes write, just return */
3593 if (!io_end
|| !size
)
3596 ext_debug("ext4_end_io_dio(): io_end 0x%p"
3597 "for inode %lu, iocb 0x%p, offset %llu, size %llu\n",
3598 iocb
->private, io_end
->inode
->i_ino
, iocb
, offset
,
3601 /* if not aio dio with unwritten extents, just free io and return */
3602 if (!(io_end
->flag
& EXT4_IO_END_UNWRITTEN
)) {
3603 ext4_free_io_end(io_end
);
3604 iocb
->private = NULL
;
3607 aio_complete(iocb
, ret
, 0);
3611 io_end
->offset
= offset
;
3612 io_end
->size
= size
;
3614 io_end
->iocb
= iocb
;
3615 io_end
->result
= ret
;
3617 wq
= EXT4_SB(io_end
->inode
->i_sb
)->dio_unwritten_wq
;
3619 /* Add the io_end to per-inode completed aio dio list*/
3620 ei
= EXT4_I(io_end
->inode
);
3621 spin_lock_irqsave(&ei
->i_completed_io_lock
, flags
);
3622 list_add_tail(&io_end
->list
, &ei
->i_completed_io_list
);
3623 spin_unlock_irqrestore(&ei
->i_completed_io_lock
, flags
);
3625 /* queue the work to convert unwritten extents to written */
3626 queue_work(wq
, &io_end
->work
);
3627 iocb
->private = NULL
;
3630 static void ext4_end_io_buffer_write(struct buffer_head
*bh
, int uptodate
)
3632 ext4_io_end_t
*io_end
= bh
->b_private
;
3633 struct workqueue_struct
*wq
;
3634 struct inode
*inode
;
3635 unsigned long flags
;
3637 if (!test_clear_buffer_uninit(bh
) || !io_end
)
3640 if (!(io_end
->inode
->i_sb
->s_flags
& MS_ACTIVE
)) {
3641 printk("sb umounted, discard end_io request for inode %lu\n",
3642 io_end
->inode
->i_ino
);
3643 ext4_free_io_end(io_end
);
3647 io_end
->flag
= EXT4_IO_END_UNWRITTEN
;
3648 inode
= io_end
->inode
;
3650 /* Add the io_end to per-inode completed io list*/
3651 spin_lock_irqsave(&EXT4_I(inode
)->i_completed_io_lock
, flags
);
3652 list_add_tail(&io_end
->list
, &EXT4_I(inode
)->i_completed_io_list
);
3653 spin_unlock_irqrestore(&EXT4_I(inode
)->i_completed_io_lock
, flags
);
3655 wq
= EXT4_SB(inode
->i_sb
)->dio_unwritten_wq
;
3656 /* queue the work to convert unwritten extents to written */
3657 queue_work(wq
, &io_end
->work
);
3659 bh
->b_private
= NULL
;
3660 bh
->b_end_io
= NULL
;
3661 clear_buffer_uninit(bh
);
3662 end_buffer_async_write(bh
, uptodate
);
3665 static int ext4_set_bh_endio(struct buffer_head
*bh
, struct inode
*inode
)
3667 ext4_io_end_t
*io_end
;
3668 struct page
*page
= bh
->b_page
;
3669 loff_t offset
= (sector_t
)page
->index
<< PAGE_CACHE_SHIFT
;
3670 size_t size
= bh
->b_size
;
3673 io_end
= ext4_init_io_end(inode
, GFP_ATOMIC
);
3675 pr_warn_ratelimited("%s: allocation fail\n", __func__
);
3679 io_end
->offset
= offset
;
3680 io_end
->size
= size
;
3682 * We need to hold a reference to the page to make sure it
3683 * doesn't get evicted before ext4_end_io_work() has a chance
3684 * to convert the extent from written to unwritten.
3686 io_end
->page
= page
;
3687 get_page(io_end
->page
);
3689 bh
->b_private
= io_end
;
3690 bh
->b_end_io
= ext4_end_io_buffer_write
;
3695 * For ext4 extent files, ext4 will do direct-io write to holes,
3696 * preallocated extents, and those write extend the file, no need to
3697 * fall back to buffered IO.
3699 * For holes, we fallocate those blocks, mark them as uninitialized
3700 * If those blocks were preallocated, we mark sure they are splited, but
3701 * still keep the range to write as uninitialized.
3703 * The unwrritten extents will be converted to written when DIO is completed.
3704 * For async direct IO, since the IO may still pending when return, we
3705 * set up an end_io call back function, which will do the convertion
3706 * when async direct IO completed.
3708 * If the O_DIRECT write will extend the file then add this inode to the
3709 * orphan list. So recovery will truncate it back to the original size
3710 * if the machine crashes during the write.
3713 static ssize_t
ext4_ext_direct_IO(int rw
, struct kiocb
*iocb
,
3714 const struct iovec
*iov
, loff_t offset
,
3715 unsigned long nr_segs
)
3717 struct file
*file
= iocb
->ki_filp
;
3718 struct inode
*inode
= file
->f_mapping
->host
;
3720 size_t count
= iov_length(iov
, nr_segs
);
3722 loff_t final_size
= offset
+ count
;
3723 if (rw
== WRITE
&& final_size
<= inode
->i_size
) {
3725 * We could direct write to holes and fallocate.
3727 * Allocated blocks to fill the hole are marked as uninitialized
3728 * to prevent paralel buffered read to expose the stale data
3729 * before DIO complete the data IO.
3731 * As to previously fallocated extents, ext4 get_block
3732 * will just simply mark the buffer mapped but still
3733 * keep the extents uninitialized.
3735 * for non AIO case, we will convert those unwritten extents
3736 * to written after return back from blockdev_direct_IO.
3738 * for async DIO, the conversion needs to be defered when
3739 * the IO is completed. The ext4 end_io callback function
3740 * will be called to take care of the conversion work.
3741 * Here for async case, we allocate an io_end structure to
3744 iocb
->private = NULL
;
3745 EXT4_I(inode
)->cur_aio_dio
= NULL
;
3746 if (!is_sync_kiocb(iocb
)) {
3747 iocb
->private = ext4_init_io_end(inode
, GFP_NOFS
);
3751 * we save the io structure for current async
3752 * direct IO, so that later ext4_map_blocks()
3753 * could flag the io structure whether there
3754 * is a unwritten extents needs to be converted
3755 * when IO is completed.
3757 EXT4_I(inode
)->cur_aio_dio
= iocb
->private;
3760 ret
= blockdev_direct_IO(rw
, iocb
, inode
,
3761 inode
->i_sb
->s_bdev
, iov
,
3763 ext4_get_block_write
,
3766 EXT4_I(inode
)->cur_aio_dio
= NULL
;
3768 * The io_end structure takes a reference to the inode,
3769 * that structure needs to be destroyed and the
3770 * reference to the inode need to be dropped, when IO is
3771 * complete, even with 0 byte write, or failed.
3773 * In the successful AIO DIO case, the io_end structure will be
3774 * desctroyed and the reference to the inode will be dropped
3775 * after the end_io call back function is called.
3777 * In the case there is 0 byte write, or error case, since
3778 * VFS direct IO won't invoke the end_io call back function,
3779 * we need to free the end_io structure here.
3781 if (ret
!= -EIOCBQUEUED
&& ret
<= 0 && iocb
->private) {
3782 ext4_free_io_end(iocb
->private);
3783 iocb
->private = NULL
;
3784 } else if (ret
> 0 && ext4_test_inode_state(inode
,
3785 EXT4_STATE_DIO_UNWRITTEN
)) {
3788 * for non AIO case, since the IO is already
3789 * completed, we could do the convertion right here
3791 err
= ext4_convert_unwritten_extents(inode
,
3795 ext4_clear_inode_state(inode
, EXT4_STATE_DIO_UNWRITTEN
);
3800 /* for write the the end of file case, we fall back to old way */
3801 return ext4_ind_direct_IO(rw
, iocb
, iov
, offset
, nr_segs
);
3804 static ssize_t
ext4_direct_IO(int rw
, struct kiocb
*iocb
,
3805 const struct iovec
*iov
, loff_t offset
,
3806 unsigned long nr_segs
)
3808 struct file
*file
= iocb
->ki_filp
;
3809 struct inode
*inode
= file
->f_mapping
->host
;
3811 if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
))
3812 return ext4_ext_direct_IO(rw
, iocb
, iov
, offset
, nr_segs
);
3814 return ext4_ind_direct_IO(rw
, iocb
, iov
, offset
, nr_segs
);
3818 * Pages can be marked dirty completely asynchronously from ext4's journalling
3819 * activity. By filemap_sync_pte(), try_to_unmap_one(), etc. We cannot do
3820 * much here because ->set_page_dirty is called under VFS locks. The page is
3821 * not necessarily locked.
3823 * We cannot just dirty the page and leave attached buffers clean, because the
3824 * buffers' dirty state is "definitive". We cannot just set the buffers dirty
3825 * or jbddirty because all the journalling code will explode.
3827 * So what we do is to mark the page "pending dirty" and next time writepage
3828 * is called, propagate that into the buffers appropriately.
3830 static int ext4_journalled_set_page_dirty(struct page
*page
)
3832 SetPageChecked(page
);
3833 return __set_page_dirty_nobuffers(page
);
3836 static const struct address_space_operations ext4_ordered_aops
= {
3837 .readpage
= ext4_readpage
,
3838 .readpages
= ext4_readpages
,
3839 .writepage
= ext4_writepage
,
3840 .sync_page
= block_sync_page
,
3841 .write_begin
= ext4_write_begin
,
3842 .write_end
= ext4_ordered_write_end
,
3844 .invalidatepage
= ext4_invalidatepage
,
3845 .releasepage
= ext4_releasepage
,
3846 .direct_IO
= ext4_direct_IO
,
3847 .migratepage
= buffer_migrate_page
,
3848 .is_partially_uptodate
= block_is_partially_uptodate
,
3849 .error_remove_page
= generic_error_remove_page
,
3852 static const struct address_space_operations ext4_writeback_aops
= {
3853 .readpage
= ext4_readpage
,
3854 .readpages
= ext4_readpages
,
3855 .writepage
= ext4_writepage
,
3856 .sync_page
= block_sync_page
,
3857 .write_begin
= ext4_write_begin
,
3858 .write_end
= ext4_writeback_write_end
,
3860 .invalidatepage
= ext4_invalidatepage
,
3861 .releasepage
= ext4_releasepage
,
3862 .direct_IO
= ext4_direct_IO
,
3863 .migratepage
= buffer_migrate_page
,
3864 .is_partially_uptodate
= block_is_partially_uptodate
,
3865 .error_remove_page
= generic_error_remove_page
,
3868 static const struct address_space_operations ext4_journalled_aops
= {
3869 .readpage
= ext4_readpage
,
3870 .readpages
= ext4_readpages
,
3871 .writepage
= ext4_writepage
,
3872 .sync_page
= block_sync_page
,
3873 .write_begin
= ext4_write_begin
,
3874 .write_end
= ext4_journalled_write_end
,
3875 .set_page_dirty
= ext4_journalled_set_page_dirty
,
3877 .invalidatepage
= ext4_invalidatepage
,
3878 .releasepage
= ext4_releasepage
,
3879 .is_partially_uptodate
= block_is_partially_uptodate
,
3880 .error_remove_page
= generic_error_remove_page
,
3883 static const struct address_space_operations ext4_da_aops
= {
3884 .readpage
= ext4_readpage
,
3885 .readpages
= ext4_readpages
,
3886 .writepage
= ext4_writepage
,
3887 .writepages
= ext4_da_writepages
,
3888 .sync_page
= block_sync_page
,
3889 .write_begin
= ext4_da_write_begin
,
3890 .write_end
= ext4_da_write_end
,
3892 .invalidatepage
= ext4_da_invalidatepage
,
3893 .releasepage
= ext4_releasepage
,
3894 .direct_IO
= ext4_direct_IO
,
3895 .migratepage
= buffer_migrate_page
,
3896 .is_partially_uptodate
= block_is_partially_uptodate
,
3897 .error_remove_page
= generic_error_remove_page
,
3900 void ext4_set_aops(struct inode
*inode
)
3902 if (ext4_should_order_data(inode
) &&
3903 test_opt(inode
->i_sb
, DELALLOC
))
3904 inode
->i_mapping
->a_ops
= &ext4_da_aops
;
3905 else if (ext4_should_order_data(inode
))
3906 inode
->i_mapping
->a_ops
= &ext4_ordered_aops
;
3907 else if (ext4_should_writeback_data(inode
) &&
3908 test_opt(inode
->i_sb
, DELALLOC
))
3909 inode
->i_mapping
->a_ops
= &ext4_da_aops
;
3910 else if (ext4_should_writeback_data(inode
))
3911 inode
->i_mapping
->a_ops
= &ext4_writeback_aops
;
3913 inode
->i_mapping
->a_ops
= &ext4_journalled_aops
;
3917 * ext4_block_truncate_page() zeroes out a mapping from file offset `from'
3918 * up to the end of the block which corresponds to `from'.
3919 * This required during truncate. We need to physically zero the tail end
3920 * of that block so it doesn't yield old data if the file is later grown.
3922 int ext4_block_truncate_page(handle_t
*handle
,
3923 struct address_space
*mapping
, loff_t from
)
3925 ext4_fsblk_t index
= from
>> PAGE_CACHE_SHIFT
;
3926 unsigned offset
= from
& (PAGE_CACHE_SIZE
-1);
3927 unsigned blocksize
, length
, pos
;
3929 struct inode
*inode
= mapping
->host
;
3930 struct buffer_head
*bh
;
3934 page
= find_or_create_page(mapping
, from
>> PAGE_CACHE_SHIFT
,
3935 mapping_gfp_mask(mapping
) & ~__GFP_FS
);
3939 blocksize
= inode
->i_sb
->s_blocksize
;
3940 length
= blocksize
- (offset
& (blocksize
- 1));
3941 iblock
= index
<< (PAGE_CACHE_SHIFT
- inode
->i_sb
->s_blocksize_bits
);
3943 if (!page_has_buffers(page
))
3944 create_empty_buffers(page
, blocksize
, 0);
3946 /* Find the buffer that contains "offset" */
3947 bh
= page_buffers(page
);
3949 while (offset
>= pos
) {
3950 bh
= bh
->b_this_page
;
3956 if (buffer_freed(bh
)) {
3957 BUFFER_TRACE(bh
, "freed: skip");
3961 if (!buffer_mapped(bh
)) {
3962 BUFFER_TRACE(bh
, "unmapped");
3963 ext4_get_block(inode
, iblock
, bh
, 0);
3964 /* unmapped? It's a hole - nothing to do */
3965 if (!buffer_mapped(bh
)) {
3966 BUFFER_TRACE(bh
, "still unmapped");
3971 /* Ok, it's mapped. Make sure it's up-to-date */
3972 if (PageUptodate(page
))
3973 set_buffer_uptodate(bh
);
3975 if (!buffer_uptodate(bh
)) {
3977 ll_rw_block(READ
, 1, &bh
);
3979 /* Uhhuh. Read error. Complain and punt. */
3980 if (!buffer_uptodate(bh
))
3984 if (ext4_should_journal_data(inode
)) {
3985 BUFFER_TRACE(bh
, "get write access");
3986 err
= ext4_journal_get_write_access(handle
, bh
);
3991 zero_user(page
, offset
, length
);
3993 BUFFER_TRACE(bh
, "zeroed end of block");
3996 if (ext4_should_journal_data(inode
)) {
3997 err
= ext4_handle_dirty_metadata(handle
, inode
, bh
);
3999 if (ext4_should_order_data(inode
) && EXT4_I(inode
)->jinode
)
4000 err
= ext4_jbd2_file_inode(handle
, inode
);
4001 mark_buffer_dirty(bh
);
4006 page_cache_release(page
);
4011 * Probably it should be a library function... search for first non-zero word
4012 * or memcmp with zero_page, whatever is better for particular architecture.
4015 static inline int all_zeroes(__le32
*p
, __le32
*q
)
4024 * ext4_find_shared - find the indirect blocks for partial truncation.
4025 * @inode: inode in question
4026 * @depth: depth of the affected branch
4027 * @offsets: offsets of pointers in that branch (see ext4_block_to_path)
4028 * @chain: place to store the pointers to partial indirect blocks
4029 * @top: place to the (detached) top of branch
4031 * This is a helper function used by ext4_truncate().
4033 * When we do truncate() we may have to clean the ends of several
4034 * indirect blocks but leave the blocks themselves alive. Block is
4035 * partially truncated if some data below the new i_size is refered
4036 * from it (and it is on the path to the first completely truncated
4037 * data block, indeed). We have to free the top of that path along
4038 * with everything to the right of the path. Since no allocation
4039 * past the truncation point is possible until ext4_truncate()
4040 * finishes, we may safely do the latter, but top of branch may
4041 * require special attention - pageout below the truncation point
4042 * might try to populate it.
4044 * We atomically detach the top of branch from the tree, store the
4045 * block number of its root in *@top, pointers to buffer_heads of
4046 * partially truncated blocks - in @chain[].bh and pointers to
4047 * their last elements that should not be removed - in
4048 * @chain[].p. Return value is the pointer to last filled element
4051 * The work left to caller to do the actual freeing of subtrees:
4052 * a) free the subtree starting from *@top
4053 * b) free the subtrees whose roots are stored in
4054 * (@chain[i].p+1 .. end of @chain[i].bh->b_data)
4055 * c) free the subtrees growing from the inode past the @chain[0].
4056 * (no partially truncated stuff there). */
4058 static Indirect
*ext4_find_shared(struct inode
*inode
, int depth
,
4059 ext4_lblk_t offsets
[4], Indirect chain
[4],
4062 Indirect
*partial
, *p
;
4066 /* Make k index the deepest non-null offset + 1 */
4067 for (k
= depth
; k
> 1 && !offsets
[k
-1]; k
--)
4069 partial
= ext4_get_branch(inode
, k
, offsets
, chain
, &err
);
4070 /* Writer: pointers */
4072 partial
= chain
+ k
-1;
4074 * If the branch acquired continuation since we've looked at it -
4075 * fine, it should all survive and (new) top doesn't belong to us.
4077 if (!partial
->key
&& *partial
->p
)
4080 for (p
= partial
; (p
> chain
) && all_zeroes((__le32
*) p
->bh
->b_data
, p
->p
); p
--)
4083 * OK, we've found the last block that must survive. The rest of our
4084 * branch should be detached before unlocking. However, if that rest
4085 * of branch is all ours and does not grow immediately from the inode
4086 * it's easier to cheat and just decrement partial->p.
4088 if (p
== chain
+ k
- 1 && p
> chain
) {
4092 /* Nope, don't do this in ext4. Must leave the tree intact */
4099 while (partial
> p
) {
4100 brelse(partial
->bh
);
4108 * Zero a number of block pointers in either an inode or an indirect block.
4109 * If we restart the transaction we must again get write access to the
4110 * indirect block for further modification.
4112 * We release `count' blocks on disk, but (last - first) may be greater
4113 * than `count' because there can be holes in there.
4115 static int ext4_clear_blocks(handle_t
*handle
, struct inode
*inode
,
4116 struct buffer_head
*bh
,
4117 ext4_fsblk_t block_to_free
,
4118 unsigned long count
, __le32
*first
,
4122 int flags
= EXT4_FREE_BLOCKS_FORGET
| EXT4_FREE_BLOCKS_VALIDATED
;
4125 if (S_ISDIR(inode
->i_mode
) || S_ISLNK(inode
->i_mode
))
4126 flags
|= EXT4_FREE_BLOCKS_METADATA
;
4128 if (!ext4_data_block_valid(EXT4_SB(inode
->i_sb
), block_to_free
,
4130 EXT4_ERROR_INODE(inode
, "attempt to clear invalid "
4131 "blocks %llu len %lu",
4132 (unsigned long long) block_to_free
, count
);
4136 if (try_to_extend_transaction(handle
, inode
)) {
4138 BUFFER_TRACE(bh
, "call ext4_handle_dirty_metadata");
4139 err
= ext4_handle_dirty_metadata(handle
, inode
, bh
);
4140 if (unlikely(err
)) {
4141 ext4_std_error(inode
->i_sb
, err
);
4145 err
= ext4_mark_inode_dirty(handle
, inode
);
4146 if (unlikely(err
)) {
4147 ext4_std_error(inode
->i_sb
, err
);
4150 err
= ext4_truncate_restart_trans(handle
, inode
,
4151 blocks_for_truncate(inode
));
4152 if (unlikely(err
)) {
4153 ext4_std_error(inode
->i_sb
, err
);
4157 BUFFER_TRACE(bh
, "retaking write access");
4158 ext4_journal_get_write_access(handle
, bh
);
4162 for (p
= first
; p
< last
; p
++)
4165 ext4_free_blocks(handle
, inode
, NULL
, block_to_free
, count
, flags
);
4170 * ext4_free_data - free a list of data blocks
4171 * @handle: handle for this transaction
4172 * @inode: inode we are dealing with
4173 * @this_bh: indirect buffer_head which contains *@first and *@last
4174 * @first: array of block numbers
4175 * @last: points immediately past the end of array
4177 * We are freeing all blocks refered from that array (numbers are stored as
4178 * little-endian 32-bit) and updating @inode->i_blocks appropriately.
4180 * We accumulate contiguous runs of blocks to free. Conveniently, if these
4181 * blocks are contiguous then releasing them at one time will only affect one
4182 * or two bitmap blocks (+ group descriptor(s) and superblock) and we won't
4183 * actually use a lot of journal space.
4185 * @this_bh will be %NULL if @first and @last point into the inode's direct
4188 static void ext4_free_data(handle_t
*handle
, struct inode
*inode
,
4189 struct buffer_head
*this_bh
,
4190 __le32
*first
, __le32
*last
)
4192 ext4_fsblk_t block_to_free
= 0; /* Starting block # of a run */
4193 unsigned long count
= 0; /* Number of blocks in the run */
4194 __le32
*block_to_free_p
= NULL
; /* Pointer into inode/ind
4197 ext4_fsblk_t nr
; /* Current block # */
4198 __le32
*p
; /* Pointer into inode/ind
4199 for current block */
4202 if (this_bh
) { /* For indirect block */
4203 BUFFER_TRACE(this_bh
, "get_write_access");
4204 err
= ext4_journal_get_write_access(handle
, this_bh
);
4205 /* Important: if we can't update the indirect pointers
4206 * to the blocks, we can't free them. */
4211 for (p
= first
; p
< last
; p
++) {
4212 nr
= le32_to_cpu(*p
);
4214 /* accumulate blocks to free if they're contiguous */
4217 block_to_free_p
= p
;
4219 } else if (nr
== block_to_free
+ count
) {
4222 if (ext4_clear_blocks(handle
, inode
, this_bh
,
4223 block_to_free
, count
,
4224 block_to_free_p
, p
))
4227 block_to_free_p
= p
;
4234 ext4_clear_blocks(handle
, inode
, this_bh
, block_to_free
,
4235 count
, block_to_free_p
, p
);
4238 BUFFER_TRACE(this_bh
, "call ext4_handle_dirty_metadata");
4241 * The buffer head should have an attached journal head at this
4242 * point. However, if the data is corrupted and an indirect
4243 * block pointed to itself, it would have been detached when
4244 * the block was cleared. Check for this instead of OOPSing.
4246 if ((EXT4_JOURNAL(inode
) == NULL
) || bh2jh(this_bh
))
4247 ext4_handle_dirty_metadata(handle
, inode
, this_bh
);
4249 EXT4_ERROR_INODE(inode
,
4250 "circular indirect block detected at "
4252 (unsigned long long) this_bh
->b_blocknr
);
4257 * ext4_free_branches - free an array of branches
4258 * @handle: JBD handle for this transaction
4259 * @inode: inode we are dealing with
4260 * @parent_bh: the buffer_head which contains *@first and *@last
4261 * @first: array of block numbers
4262 * @last: pointer immediately past the end of array
4263 * @depth: depth of the branches to free
4265 * We are freeing all blocks refered from these branches (numbers are
4266 * stored as little-endian 32-bit) and updating @inode->i_blocks
4269 static void ext4_free_branches(handle_t
*handle
, struct inode
*inode
,
4270 struct buffer_head
*parent_bh
,
4271 __le32
*first
, __le32
*last
, int depth
)
4276 if (ext4_handle_is_aborted(handle
))
4280 struct buffer_head
*bh
;
4281 int addr_per_block
= EXT4_ADDR_PER_BLOCK(inode
->i_sb
);
4283 while (--p
>= first
) {
4284 nr
= le32_to_cpu(*p
);
4286 continue; /* A hole */
4288 if (!ext4_data_block_valid(EXT4_SB(inode
->i_sb
),
4290 EXT4_ERROR_INODE(inode
,
4291 "invalid indirect mapped "
4292 "block %lu (level %d)",
4293 (unsigned long) nr
, depth
);
4297 /* Go read the buffer for the next level down */
4298 bh
= sb_bread(inode
->i_sb
, nr
);
4301 * A read failure? Report error and clear slot
4305 EXT4_ERROR_INODE_BLOCK(inode
, nr
,
4310 /* This zaps the entire block. Bottom up. */
4311 BUFFER_TRACE(bh
, "free child branches");
4312 ext4_free_branches(handle
, inode
, bh
,
4313 (__le32
*) bh
->b_data
,
4314 (__le32
*) bh
->b_data
+ addr_per_block
,
4319 * Everything below this this pointer has been
4320 * released. Now let this top-of-subtree go.
4322 * We want the freeing of this indirect block to be
4323 * atomic in the journal with the updating of the
4324 * bitmap block which owns it. So make some room in
4327 * We zero the parent pointer *after* freeing its
4328 * pointee in the bitmaps, so if extend_transaction()
4329 * for some reason fails to put the bitmap changes and
4330 * the release into the same transaction, recovery
4331 * will merely complain about releasing a free block,
4332 * rather than leaking blocks.
4334 if (ext4_handle_is_aborted(handle
))
4336 if (try_to_extend_transaction(handle
, inode
)) {
4337 ext4_mark_inode_dirty(handle
, inode
);
4338 ext4_truncate_restart_trans(handle
, inode
,
4339 blocks_for_truncate(inode
));
4343 * The forget flag here is critical because if
4344 * we are journaling (and not doing data
4345 * journaling), we have to make sure a revoke
4346 * record is written to prevent the journal
4347 * replay from overwriting the (former)
4348 * indirect block if it gets reallocated as a
4349 * data block. This must happen in the same
4350 * transaction where the data blocks are
4353 ext4_free_blocks(handle
, inode
, NULL
, nr
, 1,
4354 EXT4_FREE_BLOCKS_METADATA
|
4355 EXT4_FREE_BLOCKS_FORGET
);
4359 * The block which we have just freed is
4360 * pointed to by an indirect block: journal it
4362 BUFFER_TRACE(parent_bh
, "get_write_access");
4363 if (!ext4_journal_get_write_access(handle
,
4366 BUFFER_TRACE(parent_bh
,
4367 "call ext4_handle_dirty_metadata");
4368 ext4_handle_dirty_metadata(handle
,
4375 /* We have reached the bottom of the tree. */
4376 BUFFER_TRACE(parent_bh
, "free data blocks");
4377 ext4_free_data(handle
, inode
, parent_bh
, first
, last
);
4381 int ext4_can_truncate(struct inode
*inode
)
4383 if (IS_APPEND(inode
) || IS_IMMUTABLE(inode
))
4385 if (S_ISREG(inode
->i_mode
))
4387 if (S_ISDIR(inode
->i_mode
))
4389 if (S_ISLNK(inode
->i_mode
))
4390 return !ext4_inode_is_fast_symlink(inode
);
4397 * We block out ext4_get_block() block instantiations across the entire
4398 * transaction, and VFS/VM ensures that ext4_truncate() cannot run
4399 * simultaneously on behalf of the same inode.
4401 * As we work through the truncate and commmit bits of it to the journal there
4402 * is one core, guiding principle: the file's tree must always be consistent on
4403 * disk. We must be able to restart the truncate after a crash.
4405 * The file's tree may be transiently inconsistent in memory (although it
4406 * probably isn't), but whenever we close off and commit a journal transaction,
4407 * the contents of (the filesystem + the journal) must be consistent and
4408 * restartable. It's pretty simple, really: bottom up, right to left (although
4409 * left-to-right works OK too).
4411 * Note that at recovery time, journal replay occurs *before* the restart of
4412 * truncate against the orphan inode list.
4414 * The committed inode has the new, desired i_size (which is the same as
4415 * i_disksize in this case). After a crash, ext4_orphan_cleanup() will see
4416 * that this inode's truncate did not complete and it will again call
4417 * ext4_truncate() to have another go. So there will be instantiated blocks
4418 * to the right of the truncation point in a crashed ext4 filesystem. But
4419 * that's fine - as long as they are linked from the inode, the post-crash
4420 * ext4_truncate() run will find them and release them.
4422 void ext4_truncate(struct inode
*inode
)
4425 struct ext4_inode_info
*ei
= EXT4_I(inode
);
4426 __le32
*i_data
= ei
->i_data
;
4427 int addr_per_block
= EXT4_ADDR_PER_BLOCK(inode
->i_sb
);
4428 struct address_space
*mapping
= inode
->i_mapping
;
4429 ext4_lblk_t offsets
[4];
4434 ext4_lblk_t last_block
;
4435 unsigned blocksize
= inode
->i_sb
->s_blocksize
;
4437 if (!ext4_can_truncate(inode
))
4440 ext4_clear_inode_flag(inode
, EXT4_INODE_EOFBLOCKS
);
4442 if (inode
->i_size
== 0 && !test_opt(inode
->i_sb
, NO_AUTO_DA_ALLOC
))
4443 ext4_set_inode_state(inode
, EXT4_STATE_DA_ALLOC_CLOSE
);
4445 if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
)) {
4446 ext4_ext_truncate(inode
);
4450 handle
= start_transaction(inode
);
4452 return; /* AKPM: return what? */
4454 last_block
= (inode
->i_size
+ blocksize
-1)
4455 >> EXT4_BLOCK_SIZE_BITS(inode
->i_sb
);
4457 if (inode
->i_size
& (blocksize
- 1))
4458 if (ext4_block_truncate_page(handle
, mapping
, inode
->i_size
))
4461 n
= ext4_block_to_path(inode
, last_block
, offsets
, NULL
);
4463 goto out_stop
; /* error */
4466 * OK. This truncate is going to happen. We add the inode to the
4467 * orphan list, so that if this truncate spans multiple transactions,
4468 * and we crash, we will resume the truncate when the filesystem
4469 * recovers. It also marks the inode dirty, to catch the new size.
4471 * Implication: the file must always be in a sane, consistent
4472 * truncatable state while each transaction commits.
4474 if (ext4_orphan_add(handle
, inode
))
4478 * From here we block out all ext4_get_block() callers who want to
4479 * modify the block allocation tree.
4481 down_write(&ei
->i_data_sem
);
4483 ext4_discard_preallocations(inode
);
4486 * The orphan list entry will now protect us from any crash which
4487 * occurs before the truncate completes, so it is now safe to propagate
4488 * the new, shorter inode size (held for now in i_size) into the
4489 * on-disk inode. We do this via i_disksize, which is the value which
4490 * ext4 *really* writes onto the disk inode.
4492 ei
->i_disksize
= inode
->i_size
;
4494 if (n
== 1) { /* direct blocks */
4495 ext4_free_data(handle
, inode
, NULL
, i_data
+offsets
[0],
4496 i_data
+ EXT4_NDIR_BLOCKS
);
4500 partial
= ext4_find_shared(inode
, n
, offsets
, chain
, &nr
);
4501 /* Kill the top of shared branch (not detached) */
4503 if (partial
== chain
) {
4504 /* Shared branch grows from the inode */
4505 ext4_free_branches(handle
, inode
, NULL
,
4506 &nr
, &nr
+1, (chain
+n
-1) - partial
);
4509 * We mark the inode dirty prior to restart,
4510 * and prior to stop. No need for it here.
4513 /* Shared branch grows from an indirect block */
4514 BUFFER_TRACE(partial
->bh
, "get_write_access");
4515 ext4_free_branches(handle
, inode
, partial
->bh
,
4517 partial
->p
+1, (chain
+n
-1) - partial
);
4520 /* Clear the ends of indirect blocks on the shared branch */
4521 while (partial
> chain
) {
4522 ext4_free_branches(handle
, inode
, partial
->bh
, partial
->p
+ 1,
4523 (__le32
*)partial
->bh
->b_data
+addr_per_block
,
4524 (chain
+n
-1) - partial
);
4525 BUFFER_TRACE(partial
->bh
, "call brelse");
4526 brelse(partial
->bh
);
4530 /* Kill the remaining (whole) subtrees */
4531 switch (offsets
[0]) {
4533 nr
= i_data
[EXT4_IND_BLOCK
];
4535 ext4_free_branches(handle
, inode
, NULL
, &nr
, &nr
+1, 1);
4536 i_data
[EXT4_IND_BLOCK
] = 0;
4538 case EXT4_IND_BLOCK
:
4539 nr
= i_data
[EXT4_DIND_BLOCK
];
4541 ext4_free_branches(handle
, inode
, NULL
, &nr
, &nr
+1, 2);
4542 i_data
[EXT4_DIND_BLOCK
] = 0;
4544 case EXT4_DIND_BLOCK
:
4545 nr
= i_data
[EXT4_TIND_BLOCK
];
4547 ext4_free_branches(handle
, inode
, NULL
, &nr
, &nr
+1, 3);
4548 i_data
[EXT4_TIND_BLOCK
] = 0;
4550 case EXT4_TIND_BLOCK
:
4554 up_write(&ei
->i_data_sem
);
4555 inode
->i_mtime
= inode
->i_ctime
= ext4_current_time(inode
);
4556 ext4_mark_inode_dirty(handle
, inode
);
4559 * In a multi-transaction truncate, we only make the final transaction
4563 ext4_handle_sync(handle
);
4566 * If this was a simple ftruncate(), and the file will remain alive
4567 * then we need to clear up the orphan record which we created above.
4568 * However, if this was a real unlink then we were called by
4569 * ext4_delete_inode(), and we allow that function to clean up the
4570 * orphan info for us.
4573 ext4_orphan_del(handle
, inode
);
4575 ext4_journal_stop(handle
);
4579 * ext4_get_inode_loc returns with an extra refcount against the inode's
4580 * underlying buffer_head on success. If 'in_mem' is true, we have all
4581 * data in memory that is needed to recreate the on-disk version of this
4584 static int __ext4_get_inode_loc(struct inode
*inode
,
4585 struct ext4_iloc
*iloc
, int in_mem
)
4587 struct ext4_group_desc
*gdp
;
4588 struct buffer_head
*bh
;
4589 struct super_block
*sb
= inode
->i_sb
;
4591 int inodes_per_block
, inode_offset
;
4594 if (!ext4_valid_inum(sb
, inode
->i_ino
))
4597 iloc
->block_group
= (inode
->i_ino
- 1) / EXT4_INODES_PER_GROUP(sb
);
4598 gdp
= ext4_get_group_desc(sb
, iloc
->block_group
, NULL
);
4603 * Figure out the offset within the block group inode table
4605 inodes_per_block
= (EXT4_BLOCK_SIZE(sb
) / EXT4_INODE_SIZE(sb
));
4606 inode_offset
= ((inode
->i_ino
- 1) %
4607 EXT4_INODES_PER_GROUP(sb
));
4608 block
= ext4_inode_table(sb
, gdp
) + (inode_offset
/ inodes_per_block
);
4609 iloc
->offset
= (inode_offset
% inodes_per_block
) * EXT4_INODE_SIZE(sb
);
4611 bh
= sb_getblk(sb
, block
);
4613 EXT4_ERROR_INODE_BLOCK(inode
, block
,
4614 "unable to read itable block");
4617 if (!buffer_uptodate(bh
)) {
4621 * If the buffer has the write error flag, we have failed
4622 * to write out another inode in the same block. In this
4623 * case, we don't have to read the block because we may
4624 * read the old inode data successfully.
4626 if (buffer_write_io_error(bh
) && !buffer_uptodate(bh
))
4627 set_buffer_uptodate(bh
);
4629 if (buffer_uptodate(bh
)) {
4630 /* someone brought it uptodate while we waited */
4636 * If we have all information of the inode in memory and this
4637 * is the only valid inode in the block, we need not read the
4641 struct buffer_head
*bitmap_bh
;
4644 start
= inode_offset
& ~(inodes_per_block
- 1);
4646 /* Is the inode bitmap in cache? */
4647 bitmap_bh
= sb_getblk(sb
, ext4_inode_bitmap(sb
, gdp
));
4652 * If the inode bitmap isn't in cache then the
4653 * optimisation may end up performing two reads instead
4654 * of one, so skip it.
4656 if (!buffer_uptodate(bitmap_bh
)) {
4660 for (i
= start
; i
< start
+ inodes_per_block
; i
++) {
4661 if (i
== inode_offset
)
4663 if (ext4_test_bit(i
, bitmap_bh
->b_data
))
4667 if (i
== start
+ inodes_per_block
) {
4668 /* all other inodes are free, so skip I/O */
4669 memset(bh
->b_data
, 0, bh
->b_size
);
4670 set_buffer_uptodate(bh
);
4678 * If we need to do any I/O, try to pre-readahead extra
4679 * blocks from the inode table.
4681 if (EXT4_SB(sb
)->s_inode_readahead_blks
) {
4682 ext4_fsblk_t b
, end
, table
;
4685 table
= ext4_inode_table(sb
, gdp
);
4686 /* s_inode_readahead_blks is always a power of 2 */
4687 b
= block
& ~(EXT4_SB(sb
)->s_inode_readahead_blks
-1);
4690 end
= b
+ EXT4_SB(sb
)->s_inode_readahead_blks
;
4691 num
= EXT4_INODES_PER_GROUP(sb
);
4692 if (EXT4_HAS_RO_COMPAT_FEATURE(sb
,
4693 EXT4_FEATURE_RO_COMPAT_GDT_CSUM
))
4694 num
-= ext4_itable_unused_count(sb
, gdp
);
4695 table
+= num
/ inodes_per_block
;
4699 sb_breadahead(sb
, b
++);
4703 * There are other valid inodes in the buffer, this inode
4704 * has in-inode xattrs, or we don't have this inode in memory.
4705 * Read the block from disk.
4708 bh
->b_end_io
= end_buffer_read_sync
;
4709 submit_bh(READ_META
, bh
);
4711 if (!buffer_uptodate(bh
)) {
4712 EXT4_ERROR_INODE_BLOCK(inode
, block
,
4713 "unable to read itable block");
4723 int ext4_get_inode_loc(struct inode
*inode
, struct ext4_iloc
*iloc
)
4725 /* We have all inode data except xattrs in memory here. */
4726 return __ext4_get_inode_loc(inode
, iloc
,
4727 !ext4_test_inode_state(inode
, EXT4_STATE_XATTR
));
4730 void ext4_set_inode_flags(struct inode
*inode
)
4732 unsigned int flags
= EXT4_I(inode
)->i_flags
;
4734 inode
->i_flags
&= ~(S_SYNC
|S_APPEND
|S_IMMUTABLE
|S_NOATIME
|S_DIRSYNC
);
4735 if (flags
& EXT4_SYNC_FL
)
4736 inode
->i_flags
|= S_SYNC
;
4737 if (flags
& EXT4_APPEND_FL
)
4738 inode
->i_flags
|= S_APPEND
;
4739 if (flags
& EXT4_IMMUTABLE_FL
)
4740 inode
->i_flags
|= S_IMMUTABLE
;
4741 if (flags
& EXT4_NOATIME_FL
)
4742 inode
->i_flags
|= S_NOATIME
;
4743 if (flags
& EXT4_DIRSYNC_FL
)
4744 inode
->i_flags
|= S_DIRSYNC
;
4747 /* Propagate flags from i_flags to EXT4_I(inode)->i_flags */
4748 void ext4_get_inode_flags(struct ext4_inode_info
*ei
)
4750 unsigned int vfs_fl
;
4751 unsigned long old_fl
, new_fl
;
4754 vfs_fl
= ei
->vfs_inode
.i_flags
;
4755 old_fl
= ei
->i_flags
;
4756 new_fl
= old_fl
& ~(EXT4_SYNC_FL
|EXT4_APPEND_FL
|
4757 EXT4_IMMUTABLE_FL
|EXT4_NOATIME_FL
|
4759 if (vfs_fl
& S_SYNC
)
4760 new_fl
|= EXT4_SYNC_FL
;
4761 if (vfs_fl
& S_APPEND
)
4762 new_fl
|= EXT4_APPEND_FL
;
4763 if (vfs_fl
& S_IMMUTABLE
)
4764 new_fl
|= EXT4_IMMUTABLE_FL
;
4765 if (vfs_fl
& S_NOATIME
)
4766 new_fl
|= EXT4_NOATIME_FL
;
4767 if (vfs_fl
& S_DIRSYNC
)
4768 new_fl
|= EXT4_DIRSYNC_FL
;
4769 } while (cmpxchg(&ei
->i_flags
, old_fl
, new_fl
) != old_fl
);
4772 static blkcnt_t
ext4_inode_blocks(struct ext4_inode
*raw_inode
,
4773 struct ext4_inode_info
*ei
)
4776 struct inode
*inode
= &(ei
->vfs_inode
);
4777 struct super_block
*sb
= inode
->i_sb
;
4779 if (EXT4_HAS_RO_COMPAT_FEATURE(sb
,
4780 EXT4_FEATURE_RO_COMPAT_HUGE_FILE
)) {
4781 /* we are using combined 48 bit field */
4782 i_blocks
= ((u64
)le16_to_cpu(raw_inode
->i_blocks_high
)) << 32 |
4783 le32_to_cpu(raw_inode
->i_blocks_lo
);
4784 if (ext4_test_inode_flag(inode
, EXT4_INODE_HUGE_FILE
)) {
4785 /* i_blocks represent file system block size */
4786 return i_blocks
<< (inode
->i_blkbits
- 9);
4791 return le32_to_cpu(raw_inode
->i_blocks_lo
);
4795 struct inode
*ext4_iget(struct super_block
*sb
, unsigned long ino
)
4797 struct ext4_iloc iloc
;
4798 struct ext4_inode
*raw_inode
;
4799 struct ext4_inode_info
*ei
;
4800 struct inode
*inode
;
4801 journal_t
*journal
= EXT4_SB(sb
)->s_journal
;
4805 inode
= iget_locked(sb
, ino
);
4807 return ERR_PTR(-ENOMEM
);
4808 if (!(inode
->i_state
& I_NEW
))
4814 ret
= __ext4_get_inode_loc(inode
, &iloc
, 0);
4817 raw_inode
= ext4_raw_inode(&iloc
);
4818 inode
->i_mode
= le16_to_cpu(raw_inode
->i_mode
);
4819 inode
->i_uid
= (uid_t
)le16_to_cpu(raw_inode
->i_uid_low
);
4820 inode
->i_gid
= (gid_t
)le16_to_cpu(raw_inode
->i_gid_low
);
4821 if (!(test_opt(inode
->i_sb
, NO_UID32
))) {
4822 inode
->i_uid
|= le16_to_cpu(raw_inode
->i_uid_high
) << 16;
4823 inode
->i_gid
|= le16_to_cpu(raw_inode
->i_gid_high
) << 16;
4825 inode
->i_nlink
= le16_to_cpu(raw_inode
->i_links_count
);
4827 ext4_clear_state_flags(ei
); /* Only relevant on 32-bit archs */
4828 ei
->i_dir_start_lookup
= 0;
4829 ei
->i_dtime
= le32_to_cpu(raw_inode
->i_dtime
);
4830 /* We now have enough fields to check if the inode was active or not.
4831 * This is needed because nfsd might try to access dead inodes
4832 * the test is that same one that e2fsck uses
4833 * NeilBrown 1999oct15
4835 if (inode
->i_nlink
== 0) {
4836 if (inode
->i_mode
== 0 ||
4837 !(EXT4_SB(inode
->i_sb
)->s_mount_state
& EXT4_ORPHAN_FS
)) {
4838 /* this inode is deleted */
4842 /* The only unlinked inodes we let through here have
4843 * valid i_mode and are being read by the orphan
4844 * recovery code: that's fine, we're about to complete
4845 * the process of deleting those. */
4847 ei
->i_flags
= le32_to_cpu(raw_inode
->i_flags
);
4848 inode
->i_blocks
= ext4_inode_blocks(raw_inode
, ei
);
4849 ei
->i_file_acl
= le32_to_cpu(raw_inode
->i_file_acl_lo
);
4850 if (EXT4_HAS_INCOMPAT_FEATURE(sb
, EXT4_FEATURE_INCOMPAT_64BIT
))
4852 ((__u64
)le16_to_cpu(raw_inode
->i_file_acl_high
)) << 32;
4853 inode
->i_size
= ext4_isize(raw_inode
);
4854 ei
->i_disksize
= inode
->i_size
;
4856 ei
->i_reserved_quota
= 0;
4858 inode
->i_generation
= le32_to_cpu(raw_inode
->i_generation
);
4859 ei
->i_block_group
= iloc
.block_group
;
4860 ei
->i_last_alloc_group
= ~0;
4862 * NOTE! The in-memory inode i_data array is in little-endian order
4863 * even on big-endian machines: we do NOT byteswap the block numbers!
4865 for (block
= 0; block
< EXT4_N_BLOCKS
; block
++)
4866 ei
->i_data
[block
] = raw_inode
->i_block
[block
];
4867 INIT_LIST_HEAD(&ei
->i_orphan
);
4870 * Set transaction id's of transactions that have to be committed
4871 * to finish f[data]sync. We set them to currently running transaction
4872 * as we cannot be sure that the inode or some of its metadata isn't
4873 * part of the transaction - the inode could have been reclaimed and
4874 * now it is reread from disk.
4877 transaction_t
*transaction
;
4880 read_lock(&journal
->j_state_lock
);
4881 if (journal
->j_running_transaction
)
4882 transaction
= journal
->j_running_transaction
;
4884 transaction
= journal
->j_committing_transaction
;
4886 tid
= transaction
->t_tid
;
4888 tid
= journal
->j_commit_sequence
;
4889 read_unlock(&journal
->j_state_lock
);
4890 ei
->i_sync_tid
= tid
;
4891 ei
->i_datasync_tid
= tid
;
4894 if (EXT4_INODE_SIZE(inode
->i_sb
) > EXT4_GOOD_OLD_INODE_SIZE
) {
4895 ei
->i_extra_isize
= le16_to_cpu(raw_inode
->i_extra_isize
);
4896 if (EXT4_GOOD_OLD_INODE_SIZE
+ ei
->i_extra_isize
>
4897 EXT4_INODE_SIZE(inode
->i_sb
)) {
4901 if (ei
->i_extra_isize
== 0) {
4902 /* The extra space is currently unused. Use it. */
4903 ei
->i_extra_isize
= sizeof(struct ext4_inode
) -
4904 EXT4_GOOD_OLD_INODE_SIZE
;
4906 __le32
*magic
= (void *)raw_inode
+
4907 EXT4_GOOD_OLD_INODE_SIZE
+
4909 if (*magic
== cpu_to_le32(EXT4_XATTR_MAGIC
))
4910 ext4_set_inode_state(inode
, EXT4_STATE_XATTR
);
4913 ei
->i_extra_isize
= 0;
4915 EXT4_INODE_GET_XTIME(i_ctime
, inode
, raw_inode
);
4916 EXT4_INODE_GET_XTIME(i_mtime
, inode
, raw_inode
);
4917 EXT4_INODE_GET_XTIME(i_atime
, inode
, raw_inode
);
4918 EXT4_EINODE_GET_XTIME(i_crtime
, ei
, raw_inode
);
4920 inode
->i_version
= le32_to_cpu(raw_inode
->i_disk_version
);
4921 if (EXT4_INODE_SIZE(inode
->i_sb
) > EXT4_GOOD_OLD_INODE_SIZE
) {
4922 if (EXT4_FITS_IN_INODE(raw_inode
, ei
, i_version_hi
))
4924 (__u64
)(le32_to_cpu(raw_inode
->i_version_hi
)) << 32;
4928 if (ei
->i_file_acl
&&
4929 !ext4_data_block_valid(EXT4_SB(sb
), ei
->i_file_acl
, 1)) {
4930 EXT4_ERROR_INODE(inode
, "bad extended attribute block %llu",
4934 } else if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
)) {
4935 if (S_ISREG(inode
->i_mode
) || S_ISDIR(inode
->i_mode
) ||
4936 (S_ISLNK(inode
->i_mode
) &&
4937 !ext4_inode_is_fast_symlink(inode
)))
4938 /* Validate extent which is part of inode */
4939 ret
= ext4_ext_check_inode(inode
);
4940 } else if (S_ISREG(inode
->i_mode
) || S_ISDIR(inode
->i_mode
) ||
4941 (S_ISLNK(inode
->i_mode
) &&
4942 !ext4_inode_is_fast_symlink(inode
))) {
4943 /* Validate block references which are part of inode */
4944 ret
= ext4_check_inode_blockref(inode
);
4949 if (S_ISREG(inode
->i_mode
)) {
4950 inode
->i_op
= &ext4_file_inode_operations
;
4951 inode
->i_fop
= &ext4_file_operations
;
4952 ext4_set_aops(inode
);
4953 } else if (S_ISDIR(inode
->i_mode
)) {
4954 inode
->i_op
= &ext4_dir_inode_operations
;
4955 inode
->i_fop
= &ext4_dir_operations
;
4956 } else if (S_ISLNK(inode
->i_mode
)) {
4957 if (ext4_inode_is_fast_symlink(inode
)) {
4958 inode
->i_op
= &ext4_fast_symlink_inode_operations
;
4959 nd_terminate_link(ei
->i_data
, inode
->i_size
,
4960 sizeof(ei
->i_data
) - 1);
4962 inode
->i_op
= &ext4_symlink_inode_operations
;
4963 ext4_set_aops(inode
);
4965 } else if (S_ISCHR(inode
->i_mode
) || S_ISBLK(inode
->i_mode
) ||
4966 S_ISFIFO(inode
->i_mode
) || S_ISSOCK(inode
->i_mode
)) {
4967 inode
->i_op
= &ext4_special_inode_operations
;
4968 if (raw_inode
->i_block
[0])
4969 init_special_inode(inode
, inode
->i_mode
,
4970 old_decode_dev(le32_to_cpu(raw_inode
->i_block
[0])));
4972 init_special_inode(inode
, inode
->i_mode
,
4973 new_decode_dev(le32_to_cpu(raw_inode
->i_block
[1])));
4976 EXT4_ERROR_INODE(inode
, "bogus i_mode (%o)", inode
->i_mode
);
4980 ext4_set_inode_flags(inode
);
4981 unlock_new_inode(inode
);
4987 return ERR_PTR(ret
);
4990 static int ext4_inode_blocks_set(handle_t
*handle
,
4991 struct ext4_inode
*raw_inode
,
4992 struct ext4_inode_info
*ei
)
4994 struct inode
*inode
= &(ei
->vfs_inode
);
4995 u64 i_blocks
= inode
->i_blocks
;
4996 struct super_block
*sb
= inode
->i_sb
;
4998 if (i_blocks
<= ~0U) {
5000 * i_blocks can be represnted in a 32 bit variable
5001 * as multiple of 512 bytes
5003 raw_inode
->i_blocks_lo
= cpu_to_le32(i_blocks
);
5004 raw_inode
->i_blocks_high
= 0;
5005 ext4_clear_inode_flag(inode
, EXT4_INODE_HUGE_FILE
);
5008 if (!EXT4_HAS_RO_COMPAT_FEATURE(sb
, EXT4_FEATURE_RO_COMPAT_HUGE_FILE
))
5011 if (i_blocks
<= 0xffffffffffffULL
) {
5013 * i_blocks can be represented in a 48 bit variable
5014 * as multiple of 512 bytes
5016 raw_inode
->i_blocks_lo
= cpu_to_le32(i_blocks
);
5017 raw_inode
->i_blocks_high
= cpu_to_le16(i_blocks
>> 32);
5018 ext4_clear_inode_flag(inode
, EXT4_INODE_HUGE_FILE
);
5020 ext4_set_inode_flag(inode
, EXT4_INODE_HUGE_FILE
);
5021 /* i_block is stored in file system block size */
5022 i_blocks
= i_blocks
>> (inode
->i_blkbits
- 9);
5023 raw_inode
->i_blocks_lo
= cpu_to_le32(i_blocks
);
5024 raw_inode
->i_blocks_high
= cpu_to_le16(i_blocks
>> 32);
5030 * Post the struct inode info into an on-disk inode location in the
5031 * buffer-cache. This gobbles the caller's reference to the
5032 * buffer_head in the inode location struct.
5034 * The caller must have write access to iloc->bh.
5036 static int ext4_do_update_inode(handle_t
*handle
,
5037 struct inode
*inode
,
5038 struct ext4_iloc
*iloc
)
5040 struct ext4_inode
*raw_inode
= ext4_raw_inode(iloc
);
5041 struct ext4_inode_info
*ei
= EXT4_I(inode
);
5042 struct buffer_head
*bh
= iloc
->bh
;
5043 int err
= 0, rc
, block
;
5045 /* For fields not not tracking in the in-memory inode,
5046 * initialise them to zero for new inodes. */
5047 if (ext4_test_inode_state(inode
, EXT4_STATE_NEW
))
5048 memset(raw_inode
, 0, EXT4_SB(inode
->i_sb
)->s_inode_size
);
5050 ext4_get_inode_flags(ei
);
5051 raw_inode
->i_mode
= cpu_to_le16(inode
->i_mode
);
5052 if (!(test_opt(inode
->i_sb
, NO_UID32
))) {
5053 raw_inode
->i_uid_low
= cpu_to_le16(low_16_bits(inode
->i_uid
));
5054 raw_inode
->i_gid_low
= cpu_to_le16(low_16_bits(inode
->i_gid
));
5056 * Fix up interoperability with old kernels. Otherwise, old inodes get
5057 * re-used with the upper 16 bits of the uid/gid intact
5060 raw_inode
->i_uid_high
=
5061 cpu_to_le16(high_16_bits(inode
->i_uid
));
5062 raw_inode
->i_gid_high
=
5063 cpu_to_le16(high_16_bits(inode
->i_gid
));
5065 raw_inode
->i_uid_high
= 0;
5066 raw_inode
->i_gid_high
= 0;
5069 raw_inode
->i_uid_low
=
5070 cpu_to_le16(fs_high2lowuid(inode
->i_uid
));
5071 raw_inode
->i_gid_low
=
5072 cpu_to_le16(fs_high2lowgid(inode
->i_gid
));
5073 raw_inode
->i_uid_high
= 0;
5074 raw_inode
->i_gid_high
= 0;
5076 raw_inode
->i_links_count
= cpu_to_le16(inode
->i_nlink
);
5078 EXT4_INODE_SET_XTIME(i_ctime
, inode
, raw_inode
);
5079 EXT4_INODE_SET_XTIME(i_mtime
, inode
, raw_inode
);
5080 EXT4_INODE_SET_XTIME(i_atime
, inode
, raw_inode
);
5081 EXT4_EINODE_SET_XTIME(i_crtime
, ei
, raw_inode
);
5083 if (ext4_inode_blocks_set(handle
, raw_inode
, ei
))
5085 raw_inode
->i_dtime
= cpu_to_le32(ei
->i_dtime
);
5086 raw_inode
->i_flags
= cpu_to_le32(ei
->i_flags
& 0xFFFFFFFF);
5087 if (EXT4_SB(inode
->i_sb
)->s_es
->s_creator_os
!=
5088 cpu_to_le32(EXT4_OS_HURD
))
5089 raw_inode
->i_file_acl_high
=
5090 cpu_to_le16(ei
->i_file_acl
>> 32);
5091 raw_inode
->i_file_acl_lo
= cpu_to_le32(ei
->i_file_acl
);
5092 ext4_isize_set(raw_inode
, ei
->i_disksize
);
5093 if (ei
->i_disksize
> 0x7fffffffULL
) {
5094 struct super_block
*sb
= inode
->i_sb
;
5095 if (!EXT4_HAS_RO_COMPAT_FEATURE(sb
,
5096 EXT4_FEATURE_RO_COMPAT_LARGE_FILE
) ||
5097 EXT4_SB(sb
)->s_es
->s_rev_level
==
5098 cpu_to_le32(EXT4_GOOD_OLD_REV
)) {
5099 /* If this is the first large file
5100 * created, add a flag to the superblock.
5102 err
= ext4_journal_get_write_access(handle
,
5103 EXT4_SB(sb
)->s_sbh
);
5106 ext4_update_dynamic_rev(sb
);
5107 EXT4_SET_RO_COMPAT_FEATURE(sb
,
5108 EXT4_FEATURE_RO_COMPAT_LARGE_FILE
);
5110 ext4_handle_sync(handle
);
5111 err
= ext4_handle_dirty_metadata(handle
, NULL
,
5112 EXT4_SB(sb
)->s_sbh
);
5115 raw_inode
->i_generation
= cpu_to_le32(inode
->i_generation
);
5116 if (S_ISCHR(inode
->i_mode
) || S_ISBLK(inode
->i_mode
)) {
5117 if (old_valid_dev(inode
->i_rdev
)) {
5118 raw_inode
->i_block
[0] =
5119 cpu_to_le32(old_encode_dev(inode
->i_rdev
));
5120 raw_inode
->i_block
[1] = 0;
5122 raw_inode
->i_block
[0] = 0;
5123 raw_inode
->i_block
[1] =
5124 cpu_to_le32(new_encode_dev(inode
->i_rdev
));
5125 raw_inode
->i_block
[2] = 0;
5128 for (block
= 0; block
< EXT4_N_BLOCKS
; block
++)
5129 raw_inode
->i_block
[block
] = ei
->i_data
[block
];
5131 raw_inode
->i_disk_version
= cpu_to_le32(inode
->i_version
);
5132 if (ei
->i_extra_isize
) {
5133 if (EXT4_FITS_IN_INODE(raw_inode
, ei
, i_version_hi
))
5134 raw_inode
->i_version_hi
=
5135 cpu_to_le32(inode
->i_version
>> 32);
5136 raw_inode
->i_extra_isize
= cpu_to_le16(ei
->i_extra_isize
);
5139 BUFFER_TRACE(bh
, "call ext4_handle_dirty_metadata");
5140 rc
= ext4_handle_dirty_metadata(handle
, NULL
, bh
);
5143 ext4_clear_inode_state(inode
, EXT4_STATE_NEW
);
5145 ext4_update_inode_fsync_trans(handle
, inode
, 0);
5148 ext4_std_error(inode
->i_sb
, err
);
5153 * ext4_write_inode()
5155 * We are called from a few places:
5157 * - Within generic_file_write() for O_SYNC files.
5158 * Here, there will be no transaction running. We wait for any running
5159 * trasnaction to commit.
5161 * - Within sys_sync(), kupdate and such.
5162 * We wait on commit, if tol to.
5164 * - Within prune_icache() (PF_MEMALLOC == true)
5165 * Here we simply return. We can't afford to block kswapd on the
5168 * In all cases it is actually safe for us to return without doing anything,
5169 * because the inode has been copied into a raw inode buffer in
5170 * ext4_mark_inode_dirty(). This is a correctness thing for O_SYNC and for
5173 * Note that we are absolutely dependent upon all inode dirtiers doing the
5174 * right thing: they *must* call mark_inode_dirty() after dirtying info in
5175 * which we are interested.
5177 * It would be a bug for them to not do this. The code:
5179 * mark_inode_dirty(inode)
5181 * inode->i_size = expr;
5183 * is in error because a kswapd-driven write_inode() could occur while
5184 * `stuff()' is running, and the new i_size will be lost. Plus the inode
5185 * will no longer be on the superblock's dirty inode list.
5187 int ext4_write_inode(struct inode
*inode
, struct writeback_control
*wbc
)
5191 if (current
->flags
& PF_MEMALLOC
)
5194 if (EXT4_SB(inode
->i_sb
)->s_journal
) {
5195 if (ext4_journal_current_handle()) {
5196 jbd_debug(1, "called recursively, non-PF_MEMALLOC!\n");
5201 if (wbc
->sync_mode
!= WB_SYNC_ALL
)
5204 err
= ext4_force_commit(inode
->i_sb
);
5206 struct ext4_iloc iloc
;
5208 err
= __ext4_get_inode_loc(inode
, &iloc
, 0);
5211 if (wbc
->sync_mode
== WB_SYNC_ALL
)
5212 sync_dirty_buffer(iloc
.bh
);
5213 if (buffer_req(iloc
.bh
) && !buffer_uptodate(iloc
.bh
)) {
5214 EXT4_ERROR_INODE_BLOCK(inode
, iloc
.bh
->b_blocknr
,
5215 "IO error syncing inode");
5226 * Called from notify_change.
5228 * We want to trap VFS attempts to truncate the file as soon as
5229 * possible. In particular, we want to make sure that when the VFS
5230 * shrinks i_size, we put the inode on the orphan list and modify
5231 * i_disksize immediately, so that during the subsequent flushing of
5232 * dirty pages and freeing of disk blocks, we can guarantee that any
5233 * commit will leave the blocks being flushed in an unused state on
5234 * disk. (On recovery, the inode will get truncated and the blocks will
5235 * be freed, so we have a strong guarantee that no future commit will
5236 * leave these blocks visible to the user.)
5238 * Another thing we have to assure is that if we are in ordered mode
5239 * and inode is still attached to the committing transaction, we must
5240 * we start writeout of all the dirty pages which are being truncated.
5241 * This way we are sure that all the data written in the previous
5242 * transaction are already on disk (truncate waits for pages under
5245 * Called with inode->i_mutex down.
5247 int ext4_setattr(struct dentry
*dentry
, struct iattr
*attr
)
5249 struct inode
*inode
= dentry
->d_inode
;
5252 const unsigned int ia_valid
= attr
->ia_valid
;
5254 error
= inode_change_ok(inode
, attr
);
5258 if (is_quota_modification(inode
, attr
))
5259 dquot_initialize(inode
);
5260 if ((ia_valid
& ATTR_UID
&& attr
->ia_uid
!= inode
->i_uid
) ||
5261 (ia_valid
& ATTR_GID
&& attr
->ia_gid
!= inode
->i_gid
)) {
5264 /* (user+group)*(old+new) structure, inode write (sb,
5265 * inode block, ? - but truncate inode update has it) */
5266 handle
= ext4_journal_start(inode
, (EXT4_MAXQUOTAS_INIT_BLOCKS(inode
->i_sb
)+
5267 EXT4_MAXQUOTAS_DEL_BLOCKS(inode
->i_sb
))+3);
5268 if (IS_ERR(handle
)) {
5269 error
= PTR_ERR(handle
);
5272 error
= dquot_transfer(inode
, attr
);
5274 ext4_journal_stop(handle
);
5277 /* Update corresponding info in inode so that everything is in
5278 * one transaction */
5279 if (attr
->ia_valid
& ATTR_UID
)
5280 inode
->i_uid
= attr
->ia_uid
;
5281 if (attr
->ia_valid
& ATTR_GID
)
5282 inode
->i_gid
= attr
->ia_gid
;
5283 error
= ext4_mark_inode_dirty(handle
, inode
);
5284 ext4_journal_stop(handle
);
5287 if (attr
->ia_valid
& ATTR_SIZE
) {
5288 if (!(ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
))) {
5289 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
5291 if (attr
->ia_size
> sbi
->s_bitmap_maxbytes
)
5296 if (S_ISREG(inode
->i_mode
) &&
5297 attr
->ia_valid
& ATTR_SIZE
&&
5298 (attr
->ia_size
< inode
->i_size
||
5299 (ext4_test_inode_flag(inode
, EXT4_INODE_EOFBLOCKS
)))) {
5302 handle
= ext4_journal_start(inode
, 3);
5303 if (IS_ERR(handle
)) {
5304 error
= PTR_ERR(handle
);
5307 if (ext4_handle_valid(handle
)) {
5308 error
= ext4_orphan_add(handle
, inode
);
5311 EXT4_I(inode
)->i_disksize
= attr
->ia_size
;
5312 rc
= ext4_mark_inode_dirty(handle
, inode
);
5315 ext4_journal_stop(handle
);
5317 if (ext4_should_order_data(inode
)) {
5318 error
= ext4_begin_ordered_truncate(inode
,
5321 /* Do as much error cleanup as possible */
5322 handle
= ext4_journal_start(inode
, 3);
5323 if (IS_ERR(handle
)) {
5324 ext4_orphan_del(NULL
, inode
);
5327 ext4_orphan_del(handle
, inode
);
5329 ext4_journal_stop(handle
);
5333 /* ext4_truncate will clear the flag */
5334 if ((ext4_test_inode_flag(inode
, EXT4_INODE_EOFBLOCKS
)))
5335 ext4_truncate(inode
);
5338 if ((attr
->ia_valid
& ATTR_SIZE
) &&
5339 attr
->ia_size
!= i_size_read(inode
))
5340 rc
= vmtruncate(inode
, attr
->ia_size
);
5343 setattr_copy(inode
, attr
);
5344 mark_inode_dirty(inode
);
5348 * If the call to ext4_truncate failed to get a transaction handle at
5349 * all, we need to clean up the in-core orphan list manually.
5351 if (orphan
&& inode
->i_nlink
)
5352 ext4_orphan_del(NULL
, inode
);
5354 if (!rc
&& (ia_valid
& ATTR_MODE
))
5355 rc
= ext4_acl_chmod(inode
);
5358 ext4_std_error(inode
->i_sb
, error
);
5364 int ext4_getattr(struct vfsmount
*mnt
, struct dentry
*dentry
,
5367 struct inode
*inode
;
5368 unsigned long delalloc_blocks
;
5370 inode
= dentry
->d_inode
;
5371 generic_fillattr(inode
, stat
);
5374 * We can't update i_blocks if the block allocation is delayed
5375 * otherwise in the case of system crash before the real block
5376 * allocation is done, we will have i_blocks inconsistent with
5377 * on-disk file blocks.
5378 * We always keep i_blocks updated together with real
5379 * allocation. But to not confuse with user, stat
5380 * will return the blocks that include the delayed allocation
5381 * blocks for this file.
5383 delalloc_blocks
= EXT4_I(inode
)->i_reserved_data_blocks
;
5385 stat
->blocks
+= (delalloc_blocks
<< inode
->i_sb
->s_blocksize_bits
)>>9;
5389 static int ext4_indirect_trans_blocks(struct inode
*inode
, int nrblocks
,
5394 /* if nrblocks are contiguous */
5397 * With N contiguous data blocks, it need at most
5398 * N/EXT4_ADDR_PER_BLOCK(inode->i_sb) indirect blocks
5399 * 2 dindirect blocks
5402 indirects
= nrblocks
/ EXT4_ADDR_PER_BLOCK(inode
->i_sb
);
5403 return indirects
+ 3;
5406 * if nrblocks are not contiguous, worse case, each block touch
5407 * a indirect block, and each indirect block touch a double indirect
5408 * block, plus a triple indirect block
5410 indirects
= nrblocks
* 2 + 1;
5414 static int ext4_index_trans_blocks(struct inode
*inode
, int nrblocks
, int chunk
)
5416 if (!(ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
)))
5417 return ext4_indirect_trans_blocks(inode
, nrblocks
, chunk
);
5418 return ext4_ext_index_trans_blocks(inode
, nrblocks
, chunk
);
5422 * Account for index blocks, block groups bitmaps and block group
5423 * descriptor blocks if modify datablocks and index blocks
5424 * worse case, the indexs blocks spread over different block groups
5426 * If datablocks are discontiguous, they are possible to spread over
5427 * different block groups too. If they are contiuguous, with flexbg,
5428 * they could still across block group boundary.
5430 * Also account for superblock, inode, quota and xattr blocks
5432 static int ext4_meta_trans_blocks(struct inode
*inode
, int nrblocks
, int chunk
)
5434 ext4_group_t groups
, ngroups
= ext4_get_groups_count(inode
->i_sb
);
5440 * How many index blocks need to touch to modify nrblocks?
5441 * The "Chunk" flag indicating whether the nrblocks is
5442 * physically contiguous on disk
5444 * For Direct IO and fallocate, they calls get_block to allocate
5445 * one single extent at a time, so they could set the "Chunk" flag
5447 idxblocks
= ext4_index_trans_blocks(inode
, nrblocks
, chunk
);
5452 * Now let's see how many group bitmaps and group descriptors need
5462 if (groups
> ngroups
)
5464 if (groups
> EXT4_SB(inode
->i_sb
)->s_gdb_count
)
5465 gdpblocks
= EXT4_SB(inode
->i_sb
)->s_gdb_count
;
5467 /* bitmaps and block group descriptor blocks */
5468 ret
+= groups
+ gdpblocks
;
5470 /* Blocks for super block, inode, quota and xattr blocks */
5471 ret
+= EXT4_META_TRANS_BLOCKS(inode
->i_sb
);
5477 * Calulate the total number of credits to reserve to fit
5478 * the modification of a single pages into a single transaction,
5479 * which may include multiple chunks of block allocations.
5481 * This could be called via ext4_write_begin()
5483 * We need to consider the worse case, when
5484 * one new block per extent.
5486 int ext4_writepage_trans_blocks(struct inode
*inode
)
5488 int bpp
= ext4_journal_blocks_per_page(inode
);
5491 ret
= ext4_meta_trans_blocks(inode
, bpp
, 0);
5493 /* Account for data blocks for journalled mode */
5494 if (ext4_should_journal_data(inode
))
5500 * Calculate the journal credits for a chunk of data modification.
5502 * This is called from DIO, fallocate or whoever calling
5503 * ext4_map_blocks() to map/allocate a chunk of contiguous disk blocks.
5505 * journal buffers for data blocks are not included here, as DIO
5506 * and fallocate do no need to journal data buffers.
5508 int ext4_chunk_trans_blocks(struct inode
*inode
, int nrblocks
)
5510 return ext4_meta_trans_blocks(inode
, nrblocks
, 1);
5514 * The caller must have previously called ext4_reserve_inode_write().
5515 * Give this, we know that the caller already has write access to iloc->bh.
5517 int ext4_mark_iloc_dirty(handle_t
*handle
,
5518 struct inode
*inode
, struct ext4_iloc
*iloc
)
5522 if (test_opt(inode
->i_sb
, I_VERSION
))
5523 inode_inc_iversion(inode
);
5525 /* the do_update_inode consumes one bh->b_count */
5528 /* ext4_do_update_inode() does jbd2_journal_dirty_metadata */
5529 err
= ext4_do_update_inode(handle
, inode
, iloc
);
5535 * On success, We end up with an outstanding reference count against
5536 * iloc->bh. This _must_ be cleaned up later.
5540 ext4_reserve_inode_write(handle_t
*handle
, struct inode
*inode
,
5541 struct ext4_iloc
*iloc
)
5545 err
= ext4_get_inode_loc(inode
, iloc
);
5547 BUFFER_TRACE(iloc
->bh
, "get_write_access");
5548 err
= ext4_journal_get_write_access(handle
, iloc
->bh
);
5554 ext4_std_error(inode
->i_sb
, err
);
5559 * Expand an inode by new_extra_isize bytes.
5560 * Returns 0 on success or negative error number on failure.
5562 static int ext4_expand_extra_isize(struct inode
*inode
,
5563 unsigned int new_extra_isize
,
5564 struct ext4_iloc iloc
,
5567 struct ext4_inode
*raw_inode
;
5568 struct ext4_xattr_ibody_header
*header
;
5570 if (EXT4_I(inode
)->i_extra_isize
>= new_extra_isize
)
5573 raw_inode
= ext4_raw_inode(&iloc
);
5575 header
= IHDR(inode
, raw_inode
);
5577 /* No extended attributes present */
5578 if (!ext4_test_inode_state(inode
, EXT4_STATE_XATTR
) ||
5579 header
->h_magic
!= cpu_to_le32(EXT4_XATTR_MAGIC
)) {
5580 memset((void *)raw_inode
+ EXT4_GOOD_OLD_INODE_SIZE
, 0,
5582 EXT4_I(inode
)->i_extra_isize
= new_extra_isize
;
5586 /* try to expand with EAs present */
5587 return ext4_expand_extra_isize_ea(inode
, new_extra_isize
,
5592 * What we do here is to mark the in-core inode as clean with respect to inode
5593 * dirtiness (it may still be data-dirty).
5594 * This means that the in-core inode may be reaped by prune_icache
5595 * without having to perform any I/O. This is a very good thing,
5596 * because *any* task may call prune_icache - even ones which
5597 * have a transaction open against a different journal.
5599 * Is this cheating? Not really. Sure, we haven't written the
5600 * inode out, but prune_icache isn't a user-visible syncing function.
5601 * Whenever the user wants stuff synced (sys_sync, sys_msync, sys_fsync)
5602 * we start and wait on commits.
5604 * Is this efficient/effective? Well, we're being nice to the system
5605 * by cleaning up our inodes proactively so they can be reaped
5606 * without I/O. But we are potentially leaving up to five seconds'
5607 * worth of inodes floating about which prune_icache wants us to
5608 * write out. One way to fix that would be to get prune_icache()
5609 * to do a write_super() to free up some memory. It has the desired
5612 int ext4_mark_inode_dirty(handle_t
*handle
, struct inode
*inode
)
5614 struct ext4_iloc iloc
;
5615 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
5616 static unsigned int mnt_count
;
5620 trace_ext4_mark_inode_dirty(inode
, _RET_IP_
);
5621 err
= ext4_reserve_inode_write(handle
, inode
, &iloc
);
5622 if (ext4_handle_valid(handle
) &&
5623 EXT4_I(inode
)->i_extra_isize
< sbi
->s_want_extra_isize
&&
5624 !ext4_test_inode_state(inode
, EXT4_STATE_NO_EXPAND
)) {
5626 * We need extra buffer credits since we may write into EA block
5627 * with this same handle. If journal_extend fails, then it will
5628 * only result in a minor loss of functionality for that inode.
5629 * If this is felt to be critical, then e2fsck should be run to
5630 * force a large enough s_min_extra_isize.
5632 if ((jbd2_journal_extend(handle
,
5633 EXT4_DATA_TRANS_BLOCKS(inode
->i_sb
))) == 0) {
5634 ret
= ext4_expand_extra_isize(inode
,
5635 sbi
->s_want_extra_isize
,
5638 ext4_set_inode_state(inode
,
5639 EXT4_STATE_NO_EXPAND
);
5641 le16_to_cpu(sbi
->s_es
->s_mnt_count
)) {
5642 ext4_warning(inode
->i_sb
,
5643 "Unable to expand inode %lu. Delete"
5644 " some EAs or run e2fsck.",
5647 le16_to_cpu(sbi
->s_es
->s_mnt_count
);
5653 err
= ext4_mark_iloc_dirty(handle
, inode
, &iloc
);
5658 * ext4_dirty_inode() is called from __mark_inode_dirty()
5660 * We're really interested in the case where a file is being extended.
5661 * i_size has been changed by generic_commit_write() and we thus need
5662 * to include the updated inode in the current transaction.
5664 * Also, dquot_alloc_block() will always dirty the inode when blocks
5665 * are allocated to the file.
5667 * If the inode is marked synchronous, we don't honour that here - doing
5668 * so would cause a commit on atime updates, which we don't bother doing.
5669 * We handle synchronous inodes at the highest possible level.
5671 void ext4_dirty_inode(struct inode
*inode
)
5675 handle
= ext4_journal_start(inode
, 2);
5679 ext4_mark_inode_dirty(handle
, inode
);
5681 ext4_journal_stop(handle
);
5688 * Bind an inode's backing buffer_head into this transaction, to prevent
5689 * it from being flushed to disk early. Unlike
5690 * ext4_reserve_inode_write, this leaves behind no bh reference and
5691 * returns no iloc structure, so the caller needs to repeat the iloc
5692 * lookup to mark the inode dirty later.
5694 static int ext4_pin_inode(handle_t
*handle
, struct inode
*inode
)
5696 struct ext4_iloc iloc
;
5700 err
= ext4_get_inode_loc(inode
, &iloc
);
5702 BUFFER_TRACE(iloc
.bh
, "get_write_access");
5703 err
= jbd2_journal_get_write_access(handle
, iloc
.bh
);
5705 err
= ext4_handle_dirty_metadata(handle
,
5711 ext4_std_error(inode
->i_sb
, err
);
5716 int ext4_change_inode_journal_flag(struct inode
*inode
, int val
)
5723 * We have to be very careful here: changing a data block's
5724 * journaling status dynamically is dangerous. If we write a
5725 * data block to the journal, change the status and then delete
5726 * that block, we risk forgetting to revoke the old log record
5727 * from the journal and so a subsequent replay can corrupt data.
5728 * So, first we make sure that the journal is empty and that
5729 * nobody is changing anything.
5732 journal
= EXT4_JOURNAL(inode
);
5735 if (is_journal_aborted(journal
))
5738 jbd2_journal_lock_updates(journal
);
5739 jbd2_journal_flush(journal
);
5742 * OK, there are no updates running now, and all cached data is
5743 * synced to disk. We are now in a completely consistent state
5744 * which doesn't have anything in the journal, and we know that
5745 * no filesystem updates are running, so it is safe to modify
5746 * the inode's in-core data-journaling state flag now.
5750 ext4_set_inode_flag(inode
, EXT4_INODE_JOURNAL_DATA
);
5752 ext4_clear_inode_flag(inode
, EXT4_INODE_JOURNAL_DATA
);
5753 ext4_set_aops(inode
);
5755 jbd2_journal_unlock_updates(journal
);
5757 /* Finally we can mark the inode as dirty. */
5759 handle
= ext4_journal_start(inode
, 1);
5761 return PTR_ERR(handle
);
5763 err
= ext4_mark_inode_dirty(handle
, inode
);
5764 ext4_handle_sync(handle
);
5765 ext4_journal_stop(handle
);
5766 ext4_std_error(inode
->i_sb
, err
);
5771 static int ext4_bh_unmapped(handle_t
*handle
, struct buffer_head
*bh
)
5773 return !buffer_mapped(bh
);
5776 int ext4_page_mkwrite(struct vm_area_struct
*vma
, struct vm_fault
*vmf
)
5778 struct page
*page
= vmf
->page
;
5783 struct file
*file
= vma
->vm_file
;
5784 struct inode
*inode
= file
->f_path
.dentry
->d_inode
;
5785 struct address_space
*mapping
= inode
->i_mapping
;
5788 * Get i_alloc_sem to stop truncates messing with the inode. We cannot
5789 * get i_mutex because we are already holding mmap_sem.
5791 down_read(&inode
->i_alloc_sem
);
5792 size
= i_size_read(inode
);
5793 if (page
->mapping
!= mapping
|| size
<= page_offset(page
)
5794 || !PageUptodate(page
)) {
5795 /* page got truncated from under us? */
5799 if (PageMappedToDisk(page
))
5802 if (page
->index
== size
>> PAGE_CACHE_SHIFT
)
5803 len
= size
& ~PAGE_CACHE_MASK
;
5805 len
= PAGE_CACHE_SIZE
;
5809 * return if we have all the buffers mapped. This avoid
5810 * the need to call write_begin/write_end which does a
5811 * journal_start/journal_stop which can block and take
5814 if (page_has_buffers(page
)) {
5815 if (!walk_page_buffers(NULL
, page_buffers(page
), 0, len
, NULL
,
5816 ext4_bh_unmapped
)) {
5823 * OK, we need to fill the hole... Do write_begin write_end
5824 * to do block allocation/reservation.We are not holding
5825 * inode.i__mutex here. That allow * parallel write_begin,
5826 * write_end call. lock_page prevent this from happening
5827 * on the same page though
5829 ret
= mapping
->a_ops
->write_begin(file
, mapping
, page_offset(page
),
5830 len
, AOP_FLAG_UNINTERRUPTIBLE
, &page
, &fsdata
);
5833 ret
= mapping
->a_ops
->write_end(file
, mapping
, page_offset(page
),
5834 len
, len
, page
, fsdata
);
5840 ret
= VM_FAULT_SIGBUS
;
5841 up_read(&inode
->i_alloc_sem
);