2 * Copyright (c) 2003-2006, Cluster File Systems, Inc, info@clusterfs.com
3 * Written by Alex Tomas <alex@clusterfs.com>
5 * Architecture independence:
6 * Copyright (c) 2005, Bull S.A.
7 * Written by Pierre Peiffer <pierre.peiffer@bull.net>
9 * This program is free software; you can redistribute it and/or modify
10 * it under the terms of the GNU General Public License version 2 as
11 * published by the Free Software Foundation.
13 * This program is distributed in the hope that it will be useful,
14 * but WITHOUT ANY WARRANTY; without even the implied warranty of
15 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
16 * GNU General Public License for more details.
18 * You should have received a copy of the GNU General Public Licens
19 * along with this program; if not, write to the Free Software
20 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-
24 * Extents support for EXT4
27 * - ext4*_error() should be used in some situations
28 * - analyze all BUG()/BUG_ON(), use -EIO where appropriate
29 * - smart tree reduction
32 #include <linux/module.h>
34 #include <linux/time.h>
35 #include <linux/jbd2.h>
36 #include <linux/highuid.h>
37 #include <linux/pagemap.h>
38 #include <linux/quotaops.h>
39 #include <linux/string.h>
40 #include <linux/slab.h>
41 #include <linux/falloc.h>
42 #include <asm/uaccess.h>
43 #include <linux/fiemap.h>
44 #include "ext4_jbd2.h"
45 #include "ext4_extents.h"
50 * combine low and high parts of physical block number into ext4_fsblk_t
52 ext4_fsblk_t
ext_pblock(struct ext4_extent
*ex
)
56 block
= le32_to_cpu(ex
->ee_start_lo
);
57 block
|= ((ext4_fsblk_t
) le16_to_cpu(ex
->ee_start_hi
) << 31) << 1;
63 * combine low and high parts of a leaf physical block number into ext4_fsblk_t
65 ext4_fsblk_t
idx_pblock(struct ext4_extent_idx
*ix
)
69 block
= le32_to_cpu(ix
->ei_leaf_lo
);
70 block
|= ((ext4_fsblk_t
) le16_to_cpu(ix
->ei_leaf_hi
) << 31) << 1;
75 * ext4_ext_store_pblock:
76 * stores a large physical block number into an extent struct,
77 * breaking it into parts
79 void ext4_ext_store_pblock(struct ext4_extent
*ex
, ext4_fsblk_t pb
)
81 ex
->ee_start_lo
= cpu_to_le32((unsigned long) (pb
& 0xffffffff));
82 ex
->ee_start_hi
= cpu_to_le16((unsigned long) ((pb
>> 31) >> 1) & 0xffff);
86 * ext4_idx_store_pblock:
87 * stores a large physical block number into an index struct,
88 * breaking it into parts
90 static void ext4_idx_store_pblock(struct ext4_extent_idx
*ix
, ext4_fsblk_t pb
)
92 ix
->ei_leaf_lo
= cpu_to_le32((unsigned long) (pb
& 0xffffffff));
93 ix
->ei_leaf_hi
= cpu_to_le16((unsigned long) ((pb
>> 31) >> 1) & 0xffff);
96 static int ext4_ext_truncate_extend_restart(handle_t
*handle
,
102 if (!ext4_handle_valid(handle
))
104 if (handle
->h_buffer_credits
> needed
)
106 err
= ext4_journal_extend(handle
, needed
);
109 err
= ext4_truncate_restart_trans(handle
, inode
, needed
);
111 * We have dropped i_data_sem so someone might have cached again
112 * an extent we are going to truncate.
114 ext4_ext_invalidate_cache(inode
);
124 static int ext4_ext_get_access(handle_t
*handle
, struct inode
*inode
,
125 struct ext4_ext_path
*path
)
128 /* path points to block */
129 return ext4_journal_get_write_access(handle
, path
->p_bh
);
131 /* path points to leaf/index in inode body */
132 /* we use in-core data, no need to protect them */
142 static int ext4_ext_dirty(handle_t
*handle
, struct inode
*inode
,
143 struct ext4_ext_path
*path
)
147 /* path points to block */
148 err
= ext4_handle_dirty_metadata(handle
, inode
, path
->p_bh
);
150 /* path points to leaf/index in inode body */
151 err
= ext4_mark_inode_dirty(handle
, inode
);
156 static ext4_fsblk_t
ext4_ext_find_goal(struct inode
*inode
,
157 struct ext4_ext_path
*path
,
160 struct ext4_inode_info
*ei
= EXT4_I(inode
);
161 ext4_fsblk_t bg_start
;
162 ext4_fsblk_t last_block
;
163 ext4_grpblk_t colour
;
164 ext4_group_t block_group
;
165 int flex_size
= ext4_flex_bg_size(EXT4_SB(inode
->i_sb
));
169 struct ext4_extent
*ex
;
170 depth
= path
->p_depth
;
172 /* try to predict block placement */
173 ex
= path
[depth
].p_ext
;
175 return ext_pblock(ex
)+(block
-le32_to_cpu(ex
->ee_block
));
177 /* it looks like index is empty;
178 * try to find starting block from index itself */
179 if (path
[depth
].p_bh
)
180 return path
[depth
].p_bh
->b_blocknr
;
183 /* OK. use inode's group */
184 block_group
= ei
->i_block_group
;
185 if (flex_size
>= EXT4_FLEX_SIZE_DIR_ALLOC_SCHEME
) {
187 * If there are at least EXT4_FLEX_SIZE_DIR_ALLOC_SCHEME
188 * block groups per flexgroup, reserve the first block
189 * group for directories and special files. Regular
190 * files will start at the second block group. This
191 * tends to speed up directory access and improves
194 block_group
&= ~(flex_size
-1);
195 if (S_ISREG(inode
->i_mode
))
198 bg_start
= (block_group
* EXT4_BLOCKS_PER_GROUP(inode
->i_sb
)) +
199 le32_to_cpu(EXT4_SB(inode
->i_sb
)->s_es
->s_first_data_block
);
200 last_block
= ext4_blocks_count(EXT4_SB(inode
->i_sb
)->s_es
) - 1;
203 * If we are doing delayed allocation, we don't need take
204 * colour into account.
206 if (test_opt(inode
->i_sb
, DELALLOC
))
209 if (bg_start
+ EXT4_BLOCKS_PER_GROUP(inode
->i_sb
) <= last_block
)
210 colour
= (current
->pid
% 16) *
211 (EXT4_BLOCKS_PER_GROUP(inode
->i_sb
) / 16);
213 colour
= (current
->pid
% 16) * ((last_block
- bg_start
) / 16);
214 return bg_start
+ colour
+ block
;
218 * Allocation for a meta data block
221 ext4_ext_new_meta_block(handle_t
*handle
, struct inode
*inode
,
222 struct ext4_ext_path
*path
,
223 struct ext4_extent
*ex
, int *err
)
225 ext4_fsblk_t goal
, newblock
;
227 goal
= ext4_ext_find_goal(inode
, path
, le32_to_cpu(ex
->ee_block
));
228 newblock
= ext4_new_meta_blocks(handle
, inode
, goal
, NULL
, err
);
232 static inline int ext4_ext_space_block(struct inode
*inode
, int check
)
236 size
= (inode
->i_sb
->s_blocksize
- sizeof(struct ext4_extent_header
))
237 / sizeof(struct ext4_extent
);
239 #ifdef AGGRESSIVE_TEST
247 static inline int ext4_ext_space_block_idx(struct inode
*inode
, int check
)
251 size
= (inode
->i_sb
->s_blocksize
- sizeof(struct ext4_extent_header
))
252 / sizeof(struct ext4_extent_idx
);
254 #ifdef AGGRESSIVE_TEST
262 static inline int ext4_ext_space_root(struct inode
*inode
, int check
)
266 size
= sizeof(EXT4_I(inode
)->i_data
);
267 size
-= sizeof(struct ext4_extent_header
);
268 size
/= sizeof(struct ext4_extent
);
270 #ifdef AGGRESSIVE_TEST
278 static inline int ext4_ext_space_root_idx(struct inode
*inode
, int check
)
282 size
= sizeof(EXT4_I(inode
)->i_data
);
283 size
-= sizeof(struct ext4_extent_header
);
284 size
/= sizeof(struct ext4_extent_idx
);
286 #ifdef AGGRESSIVE_TEST
295 * Calculate the number of metadata blocks needed
296 * to allocate @blocks
297 * Worse case is one block per extent
299 int ext4_ext_calc_metadata_amount(struct inode
*inode
, int blocks
)
301 int lcap
, icap
, rcap
, leafs
, idxs
, num
;
302 int newextents
= blocks
;
304 rcap
= ext4_ext_space_root_idx(inode
, 0);
305 lcap
= ext4_ext_space_block(inode
, 0);
306 icap
= ext4_ext_space_block_idx(inode
, 0);
308 /* number of new leaf blocks needed */
309 num
= leafs
= (newextents
+ lcap
- 1) / lcap
;
312 * Worse case, we need separate index block(s)
313 * to link all new leaf blocks
315 idxs
= (leafs
+ icap
- 1) / icap
;
318 idxs
= (idxs
+ icap
- 1) / icap
;
319 } while (idxs
> rcap
);
325 ext4_ext_max_entries(struct inode
*inode
, int depth
)
329 if (depth
== ext_depth(inode
)) {
331 max
= ext4_ext_space_root(inode
, 1);
333 max
= ext4_ext_space_root_idx(inode
, 1);
336 max
= ext4_ext_space_block(inode
, 1);
338 max
= ext4_ext_space_block_idx(inode
, 1);
344 static int ext4_valid_extent(struct inode
*inode
, struct ext4_extent
*ext
)
346 ext4_fsblk_t block
= ext_pblock(ext
);
347 int len
= ext4_ext_get_actual_len(ext
);
349 return ext4_data_block_valid(EXT4_SB(inode
->i_sb
), block
, len
);
352 static int ext4_valid_extent_idx(struct inode
*inode
,
353 struct ext4_extent_idx
*ext_idx
)
355 ext4_fsblk_t block
= idx_pblock(ext_idx
);
357 return ext4_data_block_valid(EXT4_SB(inode
->i_sb
), block
, 1);
360 static int ext4_valid_extent_entries(struct inode
*inode
,
361 struct ext4_extent_header
*eh
,
364 struct ext4_extent
*ext
;
365 struct ext4_extent_idx
*ext_idx
;
366 unsigned short entries
;
367 if (eh
->eh_entries
== 0)
370 entries
= le16_to_cpu(eh
->eh_entries
);
374 ext
= EXT_FIRST_EXTENT(eh
);
376 if (!ext4_valid_extent(inode
, ext
))
382 ext_idx
= EXT_FIRST_INDEX(eh
);
384 if (!ext4_valid_extent_idx(inode
, ext_idx
))
393 static int __ext4_ext_check(const char *function
, struct inode
*inode
,
394 struct ext4_extent_header
*eh
,
397 const char *error_msg
;
400 if (unlikely(eh
->eh_magic
!= EXT4_EXT_MAGIC
)) {
401 error_msg
= "invalid magic";
404 if (unlikely(le16_to_cpu(eh
->eh_depth
) != depth
)) {
405 error_msg
= "unexpected eh_depth";
408 if (unlikely(eh
->eh_max
== 0)) {
409 error_msg
= "invalid eh_max";
412 max
= ext4_ext_max_entries(inode
, depth
);
413 if (unlikely(le16_to_cpu(eh
->eh_max
) > max
)) {
414 error_msg
= "too large eh_max";
417 if (unlikely(le16_to_cpu(eh
->eh_entries
) > le16_to_cpu(eh
->eh_max
))) {
418 error_msg
= "invalid eh_entries";
421 if (!ext4_valid_extent_entries(inode
, eh
, depth
)) {
422 error_msg
= "invalid extent entries";
428 ext4_error(inode
->i_sb
, function
,
429 "bad header/extent in inode #%lu: %s - magic %x, "
430 "entries %u, max %u(%u), depth %u(%u)",
431 inode
->i_ino
, error_msg
, le16_to_cpu(eh
->eh_magic
),
432 le16_to_cpu(eh
->eh_entries
), le16_to_cpu(eh
->eh_max
),
433 max
, le16_to_cpu(eh
->eh_depth
), depth
);
438 #define ext4_ext_check(inode, eh, depth) \
439 __ext4_ext_check(__func__, inode, eh, depth)
441 int ext4_ext_check_inode(struct inode
*inode
)
443 return ext4_ext_check(inode
, ext_inode_hdr(inode
), ext_depth(inode
));
447 static void ext4_ext_show_path(struct inode
*inode
, struct ext4_ext_path
*path
)
449 int k
, l
= path
->p_depth
;
452 for (k
= 0; k
<= l
; k
++, path
++) {
454 ext_debug(" %d->%llu", le32_to_cpu(path
->p_idx
->ei_block
),
455 idx_pblock(path
->p_idx
));
456 } else if (path
->p_ext
) {
457 ext_debug(" %d:[%d]%d:%llu ",
458 le32_to_cpu(path
->p_ext
->ee_block
),
459 ext4_ext_is_uninitialized(path
->p_ext
),
460 ext4_ext_get_actual_len(path
->p_ext
),
461 ext_pblock(path
->p_ext
));
468 static void ext4_ext_show_leaf(struct inode
*inode
, struct ext4_ext_path
*path
)
470 int depth
= ext_depth(inode
);
471 struct ext4_extent_header
*eh
;
472 struct ext4_extent
*ex
;
478 eh
= path
[depth
].p_hdr
;
479 ex
= EXT_FIRST_EXTENT(eh
);
481 ext_debug("Displaying leaf extents for inode %lu\n", inode
->i_ino
);
483 for (i
= 0; i
< le16_to_cpu(eh
->eh_entries
); i
++, ex
++) {
484 ext_debug("%d:[%d]%d:%llu ", le32_to_cpu(ex
->ee_block
),
485 ext4_ext_is_uninitialized(ex
),
486 ext4_ext_get_actual_len(ex
), ext_pblock(ex
));
491 #define ext4_ext_show_path(inode, path)
492 #define ext4_ext_show_leaf(inode, path)
495 void ext4_ext_drop_refs(struct ext4_ext_path
*path
)
497 int depth
= path
->p_depth
;
500 for (i
= 0; i
<= depth
; i
++, path
++)
508 * ext4_ext_binsearch_idx:
509 * binary search for the closest index of the given block
510 * the header must be checked before calling this
513 ext4_ext_binsearch_idx(struct inode
*inode
,
514 struct ext4_ext_path
*path
, ext4_lblk_t block
)
516 struct ext4_extent_header
*eh
= path
->p_hdr
;
517 struct ext4_extent_idx
*r
, *l
, *m
;
520 ext_debug("binsearch for %u(idx): ", block
);
522 l
= EXT_FIRST_INDEX(eh
) + 1;
523 r
= EXT_LAST_INDEX(eh
);
526 if (block
< le32_to_cpu(m
->ei_block
))
530 ext_debug("%p(%u):%p(%u):%p(%u) ", l
, le32_to_cpu(l
->ei_block
),
531 m
, le32_to_cpu(m
->ei_block
),
532 r
, le32_to_cpu(r
->ei_block
));
536 ext_debug(" -> %d->%lld ", le32_to_cpu(path
->p_idx
->ei_block
),
537 idx_pblock(path
->p_idx
));
539 #ifdef CHECK_BINSEARCH
541 struct ext4_extent_idx
*chix
, *ix
;
544 chix
= ix
= EXT_FIRST_INDEX(eh
);
545 for (k
= 0; k
< le16_to_cpu(eh
->eh_entries
); k
++, ix
++) {
547 le32_to_cpu(ix
->ei_block
) <= le32_to_cpu(ix
[-1].ei_block
)) {
548 printk(KERN_DEBUG
"k=%d, ix=0x%p, "
550 ix
, EXT_FIRST_INDEX(eh
));
551 printk(KERN_DEBUG
"%u <= %u\n",
552 le32_to_cpu(ix
->ei_block
),
553 le32_to_cpu(ix
[-1].ei_block
));
555 BUG_ON(k
&& le32_to_cpu(ix
->ei_block
)
556 <= le32_to_cpu(ix
[-1].ei_block
));
557 if (block
< le32_to_cpu(ix
->ei_block
))
561 BUG_ON(chix
!= path
->p_idx
);
568 * ext4_ext_binsearch:
569 * binary search for closest extent of the given block
570 * the header must be checked before calling this
573 ext4_ext_binsearch(struct inode
*inode
,
574 struct ext4_ext_path
*path
, ext4_lblk_t block
)
576 struct ext4_extent_header
*eh
= path
->p_hdr
;
577 struct ext4_extent
*r
, *l
, *m
;
579 if (eh
->eh_entries
== 0) {
581 * this leaf is empty:
582 * we get such a leaf in split/add case
587 ext_debug("binsearch for %u: ", block
);
589 l
= EXT_FIRST_EXTENT(eh
) + 1;
590 r
= EXT_LAST_EXTENT(eh
);
594 if (block
< le32_to_cpu(m
->ee_block
))
598 ext_debug("%p(%u):%p(%u):%p(%u) ", l
, le32_to_cpu(l
->ee_block
),
599 m
, le32_to_cpu(m
->ee_block
),
600 r
, le32_to_cpu(r
->ee_block
));
604 ext_debug(" -> %d:%llu:[%d]%d ",
605 le32_to_cpu(path
->p_ext
->ee_block
),
606 ext_pblock(path
->p_ext
),
607 ext4_ext_is_uninitialized(path
->p_ext
),
608 ext4_ext_get_actual_len(path
->p_ext
));
610 #ifdef CHECK_BINSEARCH
612 struct ext4_extent
*chex
, *ex
;
615 chex
= ex
= EXT_FIRST_EXTENT(eh
);
616 for (k
= 0; k
< le16_to_cpu(eh
->eh_entries
); k
++, ex
++) {
617 BUG_ON(k
&& le32_to_cpu(ex
->ee_block
)
618 <= le32_to_cpu(ex
[-1].ee_block
));
619 if (block
< le32_to_cpu(ex
->ee_block
))
623 BUG_ON(chex
!= path
->p_ext
);
629 int ext4_ext_tree_init(handle_t
*handle
, struct inode
*inode
)
631 struct ext4_extent_header
*eh
;
633 eh
= ext_inode_hdr(inode
);
636 eh
->eh_magic
= EXT4_EXT_MAGIC
;
637 eh
->eh_max
= cpu_to_le16(ext4_ext_space_root(inode
, 0));
638 ext4_mark_inode_dirty(handle
, inode
);
639 ext4_ext_invalidate_cache(inode
);
643 struct ext4_ext_path
*
644 ext4_ext_find_extent(struct inode
*inode
, ext4_lblk_t block
,
645 struct ext4_ext_path
*path
)
647 struct ext4_extent_header
*eh
;
648 struct buffer_head
*bh
;
649 short int depth
, i
, ppos
= 0, alloc
= 0;
651 eh
= ext_inode_hdr(inode
);
652 depth
= ext_depth(inode
);
654 /* account possible depth increase */
656 path
= kzalloc(sizeof(struct ext4_ext_path
) * (depth
+ 2),
659 return ERR_PTR(-ENOMEM
);
666 /* walk through the tree */
668 int need_to_validate
= 0;
670 ext_debug("depth %d: num %d, max %d\n",
671 ppos
, le16_to_cpu(eh
->eh_entries
), le16_to_cpu(eh
->eh_max
));
673 ext4_ext_binsearch_idx(inode
, path
+ ppos
, block
);
674 path
[ppos
].p_block
= idx_pblock(path
[ppos
].p_idx
);
675 path
[ppos
].p_depth
= i
;
676 path
[ppos
].p_ext
= NULL
;
678 bh
= sb_getblk(inode
->i_sb
, path
[ppos
].p_block
);
681 if (!bh_uptodate_or_lock(bh
)) {
682 if (bh_submit_read(bh
) < 0) {
686 /* validate the extent entries */
687 need_to_validate
= 1;
689 eh
= ext_block_hdr(bh
);
691 BUG_ON(ppos
> depth
);
692 path
[ppos
].p_bh
= bh
;
693 path
[ppos
].p_hdr
= eh
;
696 if (need_to_validate
&& ext4_ext_check(inode
, eh
, i
))
700 path
[ppos
].p_depth
= i
;
701 path
[ppos
].p_ext
= NULL
;
702 path
[ppos
].p_idx
= NULL
;
705 ext4_ext_binsearch(inode
, path
+ ppos
, block
);
706 /* if not an empty leaf */
707 if (path
[ppos
].p_ext
)
708 path
[ppos
].p_block
= ext_pblock(path
[ppos
].p_ext
);
710 ext4_ext_show_path(inode
, path
);
715 ext4_ext_drop_refs(path
);
718 return ERR_PTR(-EIO
);
722 * ext4_ext_insert_index:
723 * insert new index [@logical;@ptr] into the block at @curp;
724 * check where to insert: before @curp or after @curp
726 int ext4_ext_insert_index(handle_t
*handle
, struct inode
*inode
,
727 struct ext4_ext_path
*curp
,
728 int logical
, ext4_fsblk_t ptr
)
730 struct ext4_extent_idx
*ix
;
733 err
= ext4_ext_get_access(handle
, inode
, curp
);
737 BUG_ON(logical
== le32_to_cpu(curp
->p_idx
->ei_block
));
738 len
= EXT_MAX_INDEX(curp
->p_hdr
) - curp
->p_idx
;
739 if (logical
> le32_to_cpu(curp
->p_idx
->ei_block
)) {
741 if (curp
->p_idx
!= EXT_LAST_INDEX(curp
->p_hdr
)) {
742 len
= (len
- 1) * sizeof(struct ext4_extent_idx
);
743 len
= len
< 0 ? 0 : len
;
744 ext_debug("insert new index %d after: %llu. "
745 "move %d from 0x%p to 0x%p\n",
747 (curp
->p_idx
+ 1), (curp
->p_idx
+ 2));
748 memmove(curp
->p_idx
+ 2, curp
->p_idx
+ 1, len
);
750 ix
= curp
->p_idx
+ 1;
753 len
= len
* sizeof(struct ext4_extent_idx
);
754 len
= len
< 0 ? 0 : len
;
755 ext_debug("insert new index %d before: %llu. "
756 "move %d from 0x%p to 0x%p\n",
758 curp
->p_idx
, (curp
->p_idx
+ 1));
759 memmove(curp
->p_idx
+ 1, curp
->p_idx
, len
);
763 ix
->ei_block
= cpu_to_le32(logical
);
764 ext4_idx_store_pblock(ix
, ptr
);
765 le16_add_cpu(&curp
->p_hdr
->eh_entries
, 1);
767 BUG_ON(le16_to_cpu(curp
->p_hdr
->eh_entries
)
768 > le16_to_cpu(curp
->p_hdr
->eh_max
));
769 BUG_ON(ix
> EXT_LAST_INDEX(curp
->p_hdr
));
771 err
= ext4_ext_dirty(handle
, inode
, curp
);
772 ext4_std_error(inode
->i_sb
, err
);
779 * inserts new subtree into the path, using free index entry
781 * - allocates all needed blocks (new leaf and all intermediate index blocks)
782 * - makes decision where to split
783 * - moves remaining extents and index entries (right to the split point)
784 * into the newly allocated blocks
785 * - initializes subtree
787 static int ext4_ext_split(handle_t
*handle
, struct inode
*inode
,
788 struct ext4_ext_path
*path
,
789 struct ext4_extent
*newext
, int at
)
791 struct buffer_head
*bh
= NULL
;
792 int depth
= ext_depth(inode
);
793 struct ext4_extent_header
*neh
;
794 struct ext4_extent_idx
*fidx
;
795 struct ext4_extent
*ex
;
797 ext4_fsblk_t newblock
, oldblock
;
799 ext4_fsblk_t
*ablocks
= NULL
; /* array of allocated blocks */
802 /* make decision: where to split? */
803 /* FIXME: now decision is simplest: at current extent */
805 /* if current leaf will be split, then we should use
806 * border from split point */
807 BUG_ON(path
[depth
].p_ext
> EXT_MAX_EXTENT(path
[depth
].p_hdr
));
808 if (path
[depth
].p_ext
!= EXT_MAX_EXTENT(path
[depth
].p_hdr
)) {
809 border
= path
[depth
].p_ext
[1].ee_block
;
810 ext_debug("leaf will be split."
811 " next leaf starts at %d\n",
812 le32_to_cpu(border
));
814 border
= newext
->ee_block
;
815 ext_debug("leaf will be added."
816 " next leaf starts at %d\n",
817 le32_to_cpu(border
));
821 * If error occurs, then we break processing
822 * and mark filesystem read-only. index won't
823 * be inserted and tree will be in consistent
824 * state. Next mount will repair buffers too.
828 * Get array to track all allocated blocks.
829 * We need this to handle errors and free blocks
832 ablocks
= kzalloc(sizeof(ext4_fsblk_t
) * depth
, GFP_NOFS
);
836 /* allocate all needed blocks */
837 ext_debug("allocate %d blocks for indexes/leaf\n", depth
- at
);
838 for (a
= 0; a
< depth
- at
; a
++) {
839 newblock
= ext4_ext_new_meta_block(handle
, inode
, path
,
843 ablocks
[a
] = newblock
;
846 /* initialize new leaf */
847 newblock
= ablocks
[--a
];
848 BUG_ON(newblock
== 0);
849 bh
= sb_getblk(inode
->i_sb
, newblock
);
856 err
= ext4_journal_get_create_access(handle
, bh
);
860 neh
= ext_block_hdr(bh
);
862 neh
->eh_max
= cpu_to_le16(ext4_ext_space_block(inode
, 0));
863 neh
->eh_magic
= EXT4_EXT_MAGIC
;
865 ex
= EXT_FIRST_EXTENT(neh
);
867 /* move remainder of path[depth] to the new leaf */
868 BUG_ON(path
[depth
].p_hdr
->eh_entries
!= path
[depth
].p_hdr
->eh_max
);
869 /* start copy from next extent */
870 /* TODO: we could do it by single memmove */
873 while (path
[depth
].p_ext
<=
874 EXT_MAX_EXTENT(path
[depth
].p_hdr
)) {
875 ext_debug("move %d:%llu:[%d]%d in new leaf %llu\n",
876 le32_to_cpu(path
[depth
].p_ext
->ee_block
),
877 ext_pblock(path
[depth
].p_ext
),
878 ext4_ext_is_uninitialized(path
[depth
].p_ext
),
879 ext4_ext_get_actual_len(path
[depth
].p_ext
),
881 /*memmove(ex++, path[depth].p_ext++,
882 sizeof(struct ext4_extent));
888 memmove(ex
, path
[depth
].p_ext
-m
, sizeof(struct ext4_extent
)*m
);
889 le16_add_cpu(&neh
->eh_entries
, m
);
892 set_buffer_uptodate(bh
);
895 err
= ext4_handle_dirty_metadata(handle
, inode
, bh
);
901 /* correct old leaf */
903 err
= ext4_ext_get_access(handle
, inode
, path
+ depth
);
906 le16_add_cpu(&path
[depth
].p_hdr
->eh_entries
, -m
);
907 err
= ext4_ext_dirty(handle
, inode
, path
+ depth
);
913 /* create intermediate indexes */
917 ext_debug("create %d intermediate indices\n", k
);
918 /* insert new index into current index block */
919 /* current depth stored in i var */
923 newblock
= ablocks
[--a
];
924 bh
= sb_getblk(inode
->i_sb
, newblock
);
931 err
= ext4_journal_get_create_access(handle
, bh
);
935 neh
= ext_block_hdr(bh
);
936 neh
->eh_entries
= cpu_to_le16(1);
937 neh
->eh_magic
= EXT4_EXT_MAGIC
;
938 neh
->eh_max
= cpu_to_le16(ext4_ext_space_block_idx(inode
, 0));
939 neh
->eh_depth
= cpu_to_le16(depth
- i
);
940 fidx
= EXT_FIRST_INDEX(neh
);
941 fidx
->ei_block
= border
;
942 ext4_idx_store_pblock(fidx
, oldblock
);
944 ext_debug("int.index at %d (block %llu): %u -> %llu\n",
945 i
, newblock
, le32_to_cpu(border
), oldblock
);
950 ext_debug("cur 0x%p, last 0x%p\n", path
[i
].p_idx
,
951 EXT_MAX_INDEX(path
[i
].p_hdr
));
952 BUG_ON(EXT_MAX_INDEX(path
[i
].p_hdr
) !=
953 EXT_LAST_INDEX(path
[i
].p_hdr
));
954 while (path
[i
].p_idx
<= EXT_MAX_INDEX(path
[i
].p_hdr
)) {
955 ext_debug("%d: move %d:%llu in new index %llu\n", i
,
956 le32_to_cpu(path
[i
].p_idx
->ei_block
),
957 idx_pblock(path
[i
].p_idx
),
959 /*memmove(++fidx, path[i].p_idx++,
960 sizeof(struct ext4_extent_idx));
962 BUG_ON(neh->eh_entries > neh->eh_max);*/
967 memmove(++fidx
, path
[i
].p_idx
- m
,
968 sizeof(struct ext4_extent_idx
) * m
);
969 le16_add_cpu(&neh
->eh_entries
, m
);
971 set_buffer_uptodate(bh
);
974 err
= ext4_handle_dirty_metadata(handle
, inode
, bh
);
980 /* correct old index */
982 err
= ext4_ext_get_access(handle
, inode
, path
+ i
);
985 le16_add_cpu(&path
[i
].p_hdr
->eh_entries
, -m
);
986 err
= ext4_ext_dirty(handle
, inode
, path
+ i
);
994 /* insert new index */
995 err
= ext4_ext_insert_index(handle
, inode
, path
+ at
,
996 le32_to_cpu(border
), newblock
);
1000 if (buffer_locked(bh
))
1006 /* free all allocated blocks in error case */
1007 for (i
= 0; i
< depth
; i
++) {
1010 ext4_free_blocks(handle
, inode
, 0, ablocks
[i
], 1,
1011 EXT4_FREE_BLOCKS_METADATA
);
1020 * ext4_ext_grow_indepth:
1021 * implements tree growing procedure:
1022 * - allocates new block
1023 * - moves top-level data (index block or leaf) into the new block
1024 * - initializes new top-level, creating index that points to the
1025 * just created block
1027 static int ext4_ext_grow_indepth(handle_t
*handle
, struct inode
*inode
,
1028 struct ext4_ext_path
*path
,
1029 struct ext4_extent
*newext
)
1031 struct ext4_ext_path
*curp
= path
;
1032 struct ext4_extent_header
*neh
;
1033 struct ext4_extent_idx
*fidx
;
1034 struct buffer_head
*bh
;
1035 ext4_fsblk_t newblock
;
1038 newblock
= ext4_ext_new_meta_block(handle
, inode
, path
, newext
, &err
);
1042 bh
= sb_getblk(inode
->i_sb
, newblock
);
1045 ext4_std_error(inode
->i_sb
, err
);
1050 err
= ext4_journal_get_create_access(handle
, bh
);
1056 /* move top-level index/leaf into new block */
1057 memmove(bh
->b_data
, curp
->p_hdr
, sizeof(EXT4_I(inode
)->i_data
));
1059 /* set size of new block */
1060 neh
= ext_block_hdr(bh
);
1061 /* old root could have indexes or leaves
1062 * so calculate e_max right way */
1063 if (ext_depth(inode
))
1064 neh
->eh_max
= cpu_to_le16(ext4_ext_space_block_idx(inode
, 0));
1066 neh
->eh_max
= cpu_to_le16(ext4_ext_space_block(inode
, 0));
1067 neh
->eh_magic
= EXT4_EXT_MAGIC
;
1068 set_buffer_uptodate(bh
);
1071 err
= ext4_handle_dirty_metadata(handle
, inode
, bh
);
1075 /* create index in new top-level index: num,max,pointer */
1076 err
= ext4_ext_get_access(handle
, inode
, curp
);
1080 curp
->p_hdr
->eh_magic
= EXT4_EXT_MAGIC
;
1081 curp
->p_hdr
->eh_max
= cpu_to_le16(ext4_ext_space_root_idx(inode
, 0));
1082 curp
->p_hdr
->eh_entries
= cpu_to_le16(1);
1083 curp
->p_idx
= EXT_FIRST_INDEX(curp
->p_hdr
);
1085 if (path
[0].p_hdr
->eh_depth
)
1086 curp
->p_idx
->ei_block
=
1087 EXT_FIRST_INDEX(path
[0].p_hdr
)->ei_block
;
1089 curp
->p_idx
->ei_block
=
1090 EXT_FIRST_EXTENT(path
[0].p_hdr
)->ee_block
;
1091 ext4_idx_store_pblock(curp
->p_idx
, newblock
);
1093 neh
= ext_inode_hdr(inode
);
1094 fidx
= EXT_FIRST_INDEX(neh
);
1095 ext_debug("new root: num %d(%d), lblock %d, ptr %llu\n",
1096 le16_to_cpu(neh
->eh_entries
), le16_to_cpu(neh
->eh_max
),
1097 le32_to_cpu(fidx
->ei_block
), idx_pblock(fidx
));
1099 neh
->eh_depth
= cpu_to_le16(path
->p_depth
+ 1);
1100 err
= ext4_ext_dirty(handle
, inode
, curp
);
1108 * ext4_ext_create_new_leaf:
1109 * finds empty index and adds new leaf.
1110 * if no free index is found, then it requests in-depth growing.
1112 static int ext4_ext_create_new_leaf(handle_t
*handle
, struct inode
*inode
,
1113 struct ext4_ext_path
*path
,
1114 struct ext4_extent
*newext
)
1116 struct ext4_ext_path
*curp
;
1117 int depth
, i
, err
= 0;
1120 i
= depth
= ext_depth(inode
);
1122 /* walk up to the tree and look for free index entry */
1123 curp
= path
+ depth
;
1124 while (i
> 0 && !EXT_HAS_FREE_INDEX(curp
)) {
1129 /* we use already allocated block for index block,
1130 * so subsequent data blocks should be contiguous */
1131 if (EXT_HAS_FREE_INDEX(curp
)) {
1132 /* if we found index with free entry, then use that
1133 * entry: create all needed subtree and add new leaf */
1134 err
= ext4_ext_split(handle
, inode
, path
, newext
, i
);
1139 ext4_ext_drop_refs(path
);
1140 path
= ext4_ext_find_extent(inode
,
1141 (ext4_lblk_t
)le32_to_cpu(newext
->ee_block
),
1144 err
= PTR_ERR(path
);
1146 /* tree is full, time to grow in depth */
1147 err
= ext4_ext_grow_indepth(handle
, inode
, path
, newext
);
1152 ext4_ext_drop_refs(path
);
1153 path
= ext4_ext_find_extent(inode
,
1154 (ext4_lblk_t
)le32_to_cpu(newext
->ee_block
),
1157 err
= PTR_ERR(path
);
1162 * only first (depth 0 -> 1) produces free space;
1163 * in all other cases we have to split the grown tree
1165 depth
= ext_depth(inode
);
1166 if (path
[depth
].p_hdr
->eh_entries
== path
[depth
].p_hdr
->eh_max
) {
1167 /* now we need to split */
1177 * search the closest allocated block to the left for *logical
1178 * and returns it at @logical + it's physical address at @phys
1179 * if *logical is the smallest allocated block, the function
1180 * returns 0 at @phys
1181 * return value contains 0 (success) or error code
1184 ext4_ext_search_left(struct inode
*inode
, struct ext4_ext_path
*path
,
1185 ext4_lblk_t
*logical
, ext4_fsblk_t
*phys
)
1187 struct ext4_extent_idx
*ix
;
1188 struct ext4_extent
*ex
;
1191 BUG_ON(path
== NULL
);
1192 depth
= path
->p_depth
;
1195 if (depth
== 0 && path
->p_ext
== NULL
)
1198 /* usually extent in the path covers blocks smaller
1199 * then *logical, but it can be that extent is the
1200 * first one in the file */
1202 ex
= path
[depth
].p_ext
;
1203 ee_len
= ext4_ext_get_actual_len(ex
);
1204 if (*logical
< le32_to_cpu(ex
->ee_block
)) {
1205 BUG_ON(EXT_FIRST_EXTENT(path
[depth
].p_hdr
) != ex
);
1206 while (--depth
>= 0) {
1207 ix
= path
[depth
].p_idx
;
1208 BUG_ON(ix
!= EXT_FIRST_INDEX(path
[depth
].p_hdr
));
1213 BUG_ON(*logical
< (le32_to_cpu(ex
->ee_block
) + ee_len
));
1215 *logical
= le32_to_cpu(ex
->ee_block
) + ee_len
- 1;
1216 *phys
= ext_pblock(ex
) + ee_len
- 1;
1221 * search the closest allocated block to the right for *logical
1222 * and returns it at @logical + it's physical address at @phys
1223 * if *logical is the smallest allocated block, the function
1224 * returns 0 at @phys
1225 * return value contains 0 (success) or error code
1228 ext4_ext_search_right(struct inode
*inode
, struct ext4_ext_path
*path
,
1229 ext4_lblk_t
*logical
, ext4_fsblk_t
*phys
)
1231 struct buffer_head
*bh
= NULL
;
1232 struct ext4_extent_header
*eh
;
1233 struct ext4_extent_idx
*ix
;
1234 struct ext4_extent
*ex
;
1236 int depth
; /* Note, NOT eh_depth; depth from top of tree */
1239 BUG_ON(path
== NULL
);
1240 depth
= path
->p_depth
;
1243 if (depth
== 0 && path
->p_ext
== NULL
)
1246 /* usually extent in the path covers blocks smaller
1247 * then *logical, but it can be that extent is the
1248 * first one in the file */
1250 ex
= path
[depth
].p_ext
;
1251 ee_len
= ext4_ext_get_actual_len(ex
);
1252 if (*logical
< le32_to_cpu(ex
->ee_block
)) {
1253 BUG_ON(EXT_FIRST_EXTENT(path
[depth
].p_hdr
) != ex
);
1254 while (--depth
>= 0) {
1255 ix
= path
[depth
].p_idx
;
1256 BUG_ON(ix
!= EXT_FIRST_INDEX(path
[depth
].p_hdr
));
1258 *logical
= le32_to_cpu(ex
->ee_block
);
1259 *phys
= ext_pblock(ex
);
1263 BUG_ON(*logical
< (le32_to_cpu(ex
->ee_block
) + ee_len
));
1265 if (ex
!= EXT_LAST_EXTENT(path
[depth
].p_hdr
)) {
1266 /* next allocated block in this leaf */
1268 *logical
= le32_to_cpu(ex
->ee_block
);
1269 *phys
= ext_pblock(ex
);
1273 /* go up and search for index to the right */
1274 while (--depth
>= 0) {
1275 ix
= path
[depth
].p_idx
;
1276 if (ix
!= EXT_LAST_INDEX(path
[depth
].p_hdr
))
1280 /* we've gone up to the root and found no index to the right */
1284 /* we've found index to the right, let's
1285 * follow it and find the closest allocated
1286 * block to the right */
1288 block
= idx_pblock(ix
);
1289 while (++depth
< path
->p_depth
) {
1290 bh
= sb_bread(inode
->i_sb
, block
);
1293 eh
= ext_block_hdr(bh
);
1294 /* subtract from p_depth to get proper eh_depth */
1295 if (ext4_ext_check(inode
, eh
, path
->p_depth
- depth
)) {
1299 ix
= EXT_FIRST_INDEX(eh
);
1300 block
= idx_pblock(ix
);
1304 bh
= sb_bread(inode
->i_sb
, block
);
1307 eh
= ext_block_hdr(bh
);
1308 if (ext4_ext_check(inode
, eh
, path
->p_depth
- depth
)) {
1312 ex
= EXT_FIRST_EXTENT(eh
);
1313 *logical
= le32_to_cpu(ex
->ee_block
);
1314 *phys
= ext_pblock(ex
);
1320 * ext4_ext_next_allocated_block:
1321 * returns allocated block in subsequent extent or EXT_MAX_BLOCK.
1322 * NOTE: it considers block number from index entry as
1323 * allocated block. Thus, index entries have to be consistent
1327 ext4_ext_next_allocated_block(struct ext4_ext_path
*path
)
1331 BUG_ON(path
== NULL
);
1332 depth
= path
->p_depth
;
1334 if (depth
== 0 && path
->p_ext
== NULL
)
1335 return EXT_MAX_BLOCK
;
1337 while (depth
>= 0) {
1338 if (depth
== path
->p_depth
) {
1340 if (path
[depth
].p_ext
!=
1341 EXT_LAST_EXTENT(path
[depth
].p_hdr
))
1342 return le32_to_cpu(path
[depth
].p_ext
[1].ee_block
);
1345 if (path
[depth
].p_idx
!=
1346 EXT_LAST_INDEX(path
[depth
].p_hdr
))
1347 return le32_to_cpu(path
[depth
].p_idx
[1].ei_block
);
1352 return EXT_MAX_BLOCK
;
1356 * ext4_ext_next_leaf_block:
1357 * returns first allocated block from next leaf or EXT_MAX_BLOCK
1359 static ext4_lblk_t
ext4_ext_next_leaf_block(struct inode
*inode
,
1360 struct ext4_ext_path
*path
)
1364 BUG_ON(path
== NULL
);
1365 depth
= path
->p_depth
;
1367 /* zero-tree has no leaf blocks at all */
1369 return EXT_MAX_BLOCK
;
1371 /* go to index block */
1374 while (depth
>= 0) {
1375 if (path
[depth
].p_idx
!=
1376 EXT_LAST_INDEX(path
[depth
].p_hdr
))
1377 return (ext4_lblk_t
)
1378 le32_to_cpu(path
[depth
].p_idx
[1].ei_block
);
1382 return EXT_MAX_BLOCK
;
1386 * ext4_ext_correct_indexes:
1387 * if leaf gets modified and modified extent is first in the leaf,
1388 * then we have to correct all indexes above.
1389 * TODO: do we need to correct tree in all cases?
1391 static int ext4_ext_correct_indexes(handle_t
*handle
, struct inode
*inode
,
1392 struct ext4_ext_path
*path
)
1394 struct ext4_extent_header
*eh
;
1395 int depth
= ext_depth(inode
);
1396 struct ext4_extent
*ex
;
1400 eh
= path
[depth
].p_hdr
;
1401 ex
= path
[depth
].p_ext
;
1406 /* there is no tree at all */
1410 if (ex
!= EXT_FIRST_EXTENT(eh
)) {
1411 /* we correct tree if first leaf got modified only */
1416 * TODO: we need correction if border is smaller than current one
1419 border
= path
[depth
].p_ext
->ee_block
;
1420 err
= ext4_ext_get_access(handle
, inode
, path
+ k
);
1423 path
[k
].p_idx
->ei_block
= border
;
1424 err
= ext4_ext_dirty(handle
, inode
, path
+ k
);
1429 /* change all left-side indexes */
1430 if (path
[k
+1].p_idx
!= EXT_FIRST_INDEX(path
[k
+1].p_hdr
))
1432 err
= ext4_ext_get_access(handle
, inode
, path
+ k
);
1435 path
[k
].p_idx
->ei_block
= border
;
1436 err
= ext4_ext_dirty(handle
, inode
, path
+ k
);
1445 ext4_can_extents_be_merged(struct inode
*inode
, struct ext4_extent
*ex1
,
1446 struct ext4_extent
*ex2
)
1448 unsigned short ext1_ee_len
, ext2_ee_len
, max_len
;
1451 * Make sure that either both extents are uninitialized, or
1454 if (ext4_ext_is_uninitialized(ex1
) ^ ext4_ext_is_uninitialized(ex2
))
1457 if (ext4_ext_is_uninitialized(ex1
))
1458 max_len
= EXT_UNINIT_MAX_LEN
;
1460 max_len
= EXT_INIT_MAX_LEN
;
1462 ext1_ee_len
= ext4_ext_get_actual_len(ex1
);
1463 ext2_ee_len
= ext4_ext_get_actual_len(ex2
);
1465 if (le32_to_cpu(ex1
->ee_block
) + ext1_ee_len
!=
1466 le32_to_cpu(ex2
->ee_block
))
1470 * To allow future support for preallocated extents to be added
1471 * as an RO_COMPAT feature, refuse to merge to extents if
1472 * this can result in the top bit of ee_len being set.
1474 if (ext1_ee_len
+ ext2_ee_len
> max_len
)
1476 #ifdef AGGRESSIVE_TEST
1477 if (ext1_ee_len
>= 4)
1481 if (ext_pblock(ex1
) + ext1_ee_len
== ext_pblock(ex2
))
1487 * This function tries to merge the "ex" extent to the next extent in the tree.
1488 * It always tries to merge towards right. If you want to merge towards
1489 * left, pass "ex - 1" as argument instead of "ex".
1490 * Returns 0 if the extents (ex and ex+1) were _not_ merged and returns
1491 * 1 if they got merged.
1493 int ext4_ext_try_to_merge(struct inode
*inode
,
1494 struct ext4_ext_path
*path
,
1495 struct ext4_extent
*ex
)
1497 struct ext4_extent_header
*eh
;
1498 unsigned int depth
, len
;
1500 int uninitialized
= 0;
1502 depth
= ext_depth(inode
);
1503 BUG_ON(path
[depth
].p_hdr
== NULL
);
1504 eh
= path
[depth
].p_hdr
;
1506 while (ex
< EXT_LAST_EXTENT(eh
)) {
1507 if (!ext4_can_extents_be_merged(inode
, ex
, ex
+ 1))
1509 /* merge with next extent! */
1510 if (ext4_ext_is_uninitialized(ex
))
1512 ex
->ee_len
= cpu_to_le16(ext4_ext_get_actual_len(ex
)
1513 + ext4_ext_get_actual_len(ex
+ 1));
1515 ext4_ext_mark_uninitialized(ex
);
1517 if (ex
+ 1 < EXT_LAST_EXTENT(eh
)) {
1518 len
= (EXT_LAST_EXTENT(eh
) - ex
- 1)
1519 * sizeof(struct ext4_extent
);
1520 memmove(ex
+ 1, ex
+ 2, len
);
1522 le16_add_cpu(&eh
->eh_entries
, -1);
1524 WARN_ON(eh
->eh_entries
== 0);
1525 if (!eh
->eh_entries
)
1526 ext4_error(inode
->i_sb
, "ext4_ext_try_to_merge",
1527 "inode#%lu, eh->eh_entries = 0!", inode
->i_ino
);
1534 * check if a portion of the "newext" extent overlaps with an
1537 * If there is an overlap discovered, it updates the length of the newext
1538 * such that there will be no overlap, and then returns 1.
1539 * If there is no overlap found, it returns 0.
1541 unsigned int ext4_ext_check_overlap(struct inode
*inode
,
1542 struct ext4_extent
*newext
,
1543 struct ext4_ext_path
*path
)
1546 unsigned int depth
, len1
;
1547 unsigned int ret
= 0;
1549 b1
= le32_to_cpu(newext
->ee_block
);
1550 len1
= ext4_ext_get_actual_len(newext
);
1551 depth
= ext_depth(inode
);
1552 if (!path
[depth
].p_ext
)
1554 b2
= le32_to_cpu(path
[depth
].p_ext
->ee_block
);
1557 * get the next allocated block if the extent in the path
1558 * is before the requested block(s)
1561 b2
= ext4_ext_next_allocated_block(path
);
1562 if (b2
== EXT_MAX_BLOCK
)
1566 /* check for wrap through zero on extent logical start block*/
1567 if (b1
+ len1
< b1
) {
1568 len1
= EXT_MAX_BLOCK
- b1
;
1569 newext
->ee_len
= cpu_to_le16(len1
);
1573 /* check for overlap */
1574 if (b1
+ len1
> b2
) {
1575 newext
->ee_len
= cpu_to_le16(b2
- b1
);
1583 * ext4_ext_insert_extent:
1584 * tries to merge requsted extent into the existing extent or
1585 * inserts requested extent as new one into the tree,
1586 * creating new leaf in the no-space case.
1588 int ext4_ext_insert_extent(handle_t
*handle
, struct inode
*inode
,
1589 struct ext4_ext_path
*path
,
1590 struct ext4_extent
*newext
, int flag
)
1592 struct ext4_extent_header
*eh
;
1593 struct ext4_extent
*ex
, *fex
;
1594 struct ext4_extent
*nearex
; /* nearest extent */
1595 struct ext4_ext_path
*npath
= NULL
;
1596 int depth
, len
, err
;
1598 unsigned uninitialized
= 0;
1600 BUG_ON(ext4_ext_get_actual_len(newext
) == 0);
1601 depth
= ext_depth(inode
);
1602 ex
= path
[depth
].p_ext
;
1603 BUG_ON(path
[depth
].p_hdr
== NULL
);
1605 /* try to insert block into found extent and return */
1606 if (ex
&& (flag
!= EXT4_GET_BLOCKS_DIO_CREATE_EXT
)
1607 && ext4_can_extents_be_merged(inode
, ex
, newext
)) {
1608 ext_debug("append [%d]%d block to %d:[%d]%d (from %llu)\n",
1609 ext4_ext_is_uninitialized(newext
),
1610 ext4_ext_get_actual_len(newext
),
1611 le32_to_cpu(ex
->ee_block
),
1612 ext4_ext_is_uninitialized(ex
),
1613 ext4_ext_get_actual_len(ex
), ext_pblock(ex
));
1614 err
= ext4_ext_get_access(handle
, inode
, path
+ depth
);
1619 * ext4_can_extents_be_merged should have checked that either
1620 * both extents are uninitialized, or both aren't. Thus we
1621 * need to check only one of them here.
1623 if (ext4_ext_is_uninitialized(ex
))
1625 ex
->ee_len
= cpu_to_le16(ext4_ext_get_actual_len(ex
)
1626 + ext4_ext_get_actual_len(newext
));
1628 ext4_ext_mark_uninitialized(ex
);
1629 eh
= path
[depth
].p_hdr
;
1635 depth
= ext_depth(inode
);
1636 eh
= path
[depth
].p_hdr
;
1637 if (le16_to_cpu(eh
->eh_entries
) < le16_to_cpu(eh
->eh_max
))
1640 /* probably next leaf has space for us? */
1641 fex
= EXT_LAST_EXTENT(eh
);
1642 next
= ext4_ext_next_leaf_block(inode
, path
);
1643 if (le32_to_cpu(newext
->ee_block
) > le32_to_cpu(fex
->ee_block
)
1644 && next
!= EXT_MAX_BLOCK
) {
1645 ext_debug("next leaf block - %d\n", next
);
1646 BUG_ON(npath
!= NULL
);
1647 npath
= ext4_ext_find_extent(inode
, next
, NULL
);
1649 return PTR_ERR(npath
);
1650 BUG_ON(npath
->p_depth
!= path
->p_depth
);
1651 eh
= npath
[depth
].p_hdr
;
1652 if (le16_to_cpu(eh
->eh_entries
) < le16_to_cpu(eh
->eh_max
)) {
1653 ext_debug("next leaf isnt full(%d)\n",
1654 le16_to_cpu(eh
->eh_entries
));
1658 ext_debug("next leaf has no free space(%d,%d)\n",
1659 le16_to_cpu(eh
->eh_entries
), le16_to_cpu(eh
->eh_max
));
1663 * There is no free space in the found leaf.
1664 * We're gonna add a new leaf in the tree.
1666 err
= ext4_ext_create_new_leaf(handle
, inode
, path
, newext
);
1669 depth
= ext_depth(inode
);
1670 eh
= path
[depth
].p_hdr
;
1673 nearex
= path
[depth
].p_ext
;
1675 err
= ext4_ext_get_access(handle
, inode
, path
+ depth
);
1680 /* there is no extent in this leaf, create first one */
1681 ext_debug("first extent in the leaf: %d:%llu:[%d]%d\n",
1682 le32_to_cpu(newext
->ee_block
),
1684 ext4_ext_is_uninitialized(newext
),
1685 ext4_ext_get_actual_len(newext
));
1686 path
[depth
].p_ext
= EXT_FIRST_EXTENT(eh
);
1687 } else if (le32_to_cpu(newext
->ee_block
)
1688 > le32_to_cpu(nearex
->ee_block
)) {
1689 /* BUG_ON(newext->ee_block == nearex->ee_block); */
1690 if (nearex
!= EXT_LAST_EXTENT(eh
)) {
1691 len
= EXT_MAX_EXTENT(eh
) - nearex
;
1692 len
= (len
- 1) * sizeof(struct ext4_extent
);
1693 len
= len
< 0 ? 0 : len
;
1694 ext_debug("insert %d:%llu:[%d]%d after: nearest 0x%p, "
1695 "move %d from 0x%p to 0x%p\n",
1696 le32_to_cpu(newext
->ee_block
),
1698 ext4_ext_is_uninitialized(newext
),
1699 ext4_ext_get_actual_len(newext
),
1700 nearex
, len
, nearex
+ 1, nearex
+ 2);
1701 memmove(nearex
+ 2, nearex
+ 1, len
);
1703 path
[depth
].p_ext
= nearex
+ 1;
1705 BUG_ON(newext
->ee_block
== nearex
->ee_block
);
1706 len
= (EXT_MAX_EXTENT(eh
) - nearex
) * sizeof(struct ext4_extent
);
1707 len
= len
< 0 ? 0 : len
;
1708 ext_debug("insert %d:%llu:[%d]%d before: nearest 0x%p, "
1709 "move %d from 0x%p to 0x%p\n",
1710 le32_to_cpu(newext
->ee_block
),
1712 ext4_ext_is_uninitialized(newext
),
1713 ext4_ext_get_actual_len(newext
),
1714 nearex
, len
, nearex
+ 1, nearex
+ 2);
1715 memmove(nearex
+ 1, nearex
, len
);
1716 path
[depth
].p_ext
= nearex
;
1719 le16_add_cpu(&eh
->eh_entries
, 1);
1720 nearex
= path
[depth
].p_ext
;
1721 nearex
->ee_block
= newext
->ee_block
;
1722 ext4_ext_store_pblock(nearex
, ext_pblock(newext
));
1723 nearex
->ee_len
= newext
->ee_len
;
1726 /* try to merge extents to the right */
1727 if (flag
!= EXT4_GET_BLOCKS_DIO_CREATE_EXT
)
1728 ext4_ext_try_to_merge(inode
, path
, nearex
);
1730 /* try to merge extents to the left */
1732 /* time to correct all indexes above */
1733 err
= ext4_ext_correct_indexes(handle
, inode
, path
);
1737 err
= ext4_ext_dirty(handle
, inode
, path
+ depth
);
1741 ext4_ext_drop_refs(npath
);
1744 ext4_ext_invalidate_cache(inode
);
1748 int ext4_ext_walk_space(struct inode
*inode
, ext4_lblk_t block
,
1749 ext4_lblk_t num
, ext_prepare_callback func
,
1752 struct ext4_ext_path
*path
= NULL
;
1753 struct ext4_ext_cache cbex
;
1754 struct ext4_extent
*ex
;
1755 ext4_lblk_t next
, start
= 0, end
= 0;
1756 ext4_lblk_t last
= block
+ num
;
1757 int depth
, exists
, err
= 0;
1759 BUG_ON(func
== NULL
);
1760 BUG_ON(inode
== NULL
);
1762 while (block
< last
&& block
!= EXT_MAX_BLOCK
) {
1764 /* find extent for this block */
1765 path
= ext4_ext_find_extent(inode
, block
, path
);
1767 err
= PTR_ERR(path
);
1772 depth
= ext_depth(inode
);
1773 BUG_ON(path
[depth
].p_hdr
== NULL
);
1774 ex
= path
[depth
].p_ext
;
1775 next
= ext4_ext_next_allocated_block(path
);
1779 /* there is no extent yet, so try to allocate
1780 * all requested space */
1783 } else if (le32_to_cpu(ex
->ee_block
) > block
) {
1784 /* need to allocate space before found extent */
1786 end
= le32_to_cpu(ex
->ee_block
);
1787 if (block
+ num
< end
)
1789 } else if (block
>= le32_to_cpu(ex
->ee_block
)
1790 + ext4_ext_get_actual_len(ex
)) {
1791 /* need to allocate space after found extent */
1796 } else if (block
>= le32_to_cpu(ex
->ee_block
)) {
1798 * some part of requested space is covered
1802 end
= le32_to_cpu(ex
->ee_block
)
1803 + ext4_ext_get_actual_len(ex
);
1804 if (block
+ num
< end
)
1810 BUG_ON(end
<= start
);
1813 cbex
.ec_block
= start
;
1814 cbex
.ec_len
= end
- start
;
1816 cbex
.ec_type
= EXT4_EXT_CACHE_GAP
;
1818 cbex
.ec_block
= le32_to_cpu(ex
->ee_block
);
1819 cbex
.ec_len
= ext4_ext_get_actual_len(ex
);
1820 cbex
.ec_start
= ext_pblock(ex
);
1821 cbex
.ec_type
= EXT4_EXT_CACHE_EXTENT
;
1824 BUG_ON(cbex
.ec_len
== 0);
1825 err
= func(inode
, path
, &cbex
, ex
, cbdata
);
1826 ext4_ext_drop_refs(path
);
1831 if (err
== EXT_REPEAT
)
1833 else if (err
== EXT_BREAK
) {
1838 if (ext_depth(inode
) != depth
) {
1839 /* depth was changed. we have to realloc path */
1844 block
= cbex
.ec_block
+ cbex
.ec_len
;
1848 ext4_ext_drop_refs(path
);
1856 ext4_ext_put_in_cache(struct inode
*inode
, ext4_lblk_t block
,
1857 __u32 len
, ext4_fsblk_t start
, int type
)
1859 struct ext4_ext_cache
*cex
;
1861 spin_lock(&EXT4_I(inode
)->i_block_reservation_lock
);
1862 cex
= &EXT4_I(inode
)->i_cached_extent
;
1863 cex
->ec_type
= type
;
1864 cex
->ec_block
= block
;
1866 cex
->ec_start
= start
;
1867 spin_unlock(&EXT4_I(inode
)->i_block_reservation_lock
);
1871 * ext4_ext_put_gap_in_cache:
1872 * calculate boundaries of the gap that the requested block fits into
1873 * and cache this gap
1876 ext4_ext_put_gap_in_cache(struct inode
*inode
, struct ext4_ext_path
*path
,
1879 int depth
= ext_depth(inode
);
1882 struct ext4_extent
*ex
;
1884 ex
= path
[depth
].p_ext
;
1886 /* there is no extent yet, so gap is [0;-] */
1888 len
= EXT_MAX_BLOCK
;
1889 ext_debug("cache gap(whole file):");
1890 } else if (block
< le32_to_cpu(ex
->ee_block
)) {
1892 len
= le32_to_cpu(ex
->ee_block
) - block
;
1893 ext_debug("cache gap(before): %u [%u:%u]",
1895 le32_to_cpu(ex
->ee_block
),
1896 ext4_ext_get_actual_len(ex
));
1897 } else if (block
>= le32_to_cpu(ex
->ee_block
)
1898 + ext4_ext_get_actual_len(ex
)) {
1900 lblock
= le32_to_cpu(ex
->ee_block
)
1901 + ext4_ext_get_actual_len(ex
);
1903 next
= ext4_ext_next_allocated_block(path
);
1904 ext_debug("cache gap(after): [%u:%u] %u",
1905 le32_to_cpu(ex
->ee_block
),
1906 ext4_ext_get_actual_len(ex
),
1908 BUG_ON(next
== lblock
);
1909 len
= next
- lblock
;
1915 ext_debug(" -> %u:%lu\n", lblock
, len
);
1916 ext4_ext_put_in_cache(inode
, lblock
, len
, 0, EXT4_EXT_CACHE_GAP
);
1920 ext4_ext_in_cache(struct inode
*inode
, ext4_lblk_t block
,
1921 struct ext4_extent
*ex
)
1923 struct ext4_ext_cache
*cex
;
1924 int ret
= EXT4_EXT_CACHE_NO
;
1927 * We borrow i_block_reservation_lock to protect i_cached_extent
1929 spin_lock(&EXT4_I(inode
)->i_block_reservation_lock
);
1930 cex
= &EXT4_I(inode
)->i_cached_extent
;
1932 /* has cache valid data? */
1933 if (cex
->ec_type
== EXT4_EXT_CACHE_NO
)
1936 BUG_ON(cex
->ec_type
!= EXT4_EXT_CACHE_GAP
&&
1937 cex
->ec_type
!= EXT4_EXT_CACHE_EXTENT
);
1938 if (block
>= cex
->ec_block
&& block
< cex
->ec_block
+ cex
->ec_len
) {
1939 ex
->ee_block
= cpu_to_le32(cex
->ec_block
);
1940 ext4_ext_store_pblock(ex
, cex
->ec_start
);
1941 ex
->ee_len
= cpu_to_le16(cex
->ec_len
);
1942 ext_debug("%u cached by %u:%u:%llu\n",
1944 cex
->ec_block
, cex
->ec_len
, cex
->ec_start
);
1948 spin_unlock(&EXT4_I(inode
)->i_block_reservation_lock
);
1954 * removes index from the index block.
1955 * It's used in truncate case only, thus all requests are for
1956 * last index in the block only.
1958 static int ext4_ext_rm_idx(handle_t
*handle
, struct inode
*inode
,
1959 struct ext4_ext_path
*path
)
1964 /* free index block */
1966 leaf
= idx_pblock(path
->p_idx
);
1967 BUG_ON(path
->p_hdr
->eh_entries
== 0);
1968 err
= ext4_ext_get_access(handle
, inode
, path
);
1971 le16_add_cpu(&path
->p_hdr
->eh_entries
, -1);
1972 err
= ext4_ext_dirty(handle
, inode
, path
);
1975 ext_debug("index is empty, remove it, free block %llu\n", leaf
);
1976 ext4_free_blocks(handle
, inode
, 0, leaf
, 1,
1977 EXT4_FREE_BLOCKS_METADATA
| EXT4_FREE_BLOCKS_FORGET
);
1982 * ext4_ext_calc_credits_for_single_extent:
1983 * This routine returns max. credits that needed to insert an extent
1984 * to the extent tree.
1985 * When pass the actual path, the caller should calculate credits
1988 int ext4_ext_calc_credits_for_single_extent(struct inode
*inode
, int nrblocks
,
1989 struct ext4_ext_path
*path
)
1992 int depth
= ext_depth(inode
);
1995 /* probably there is space in leaf? */
1996 if (le16_to_cpu(path
[depth
].p_hdr
->eh_entries
)
1997 < le16_to_cpu(path
[depth
].p_hdr
->eh_max
)) {
2000 * There are some space in the leaf tree, no
2001 * need to account for leaf block credit
2003 * bitmaps and block group descriptor blocks
2004 * and other metadat blocks still need to be
2007 /* 1 bitmap, 1 block group descriptor */
2008 ret
= 2 + EXT4_META_TRANS_BLOCKS(inode
->i_sb
);
2013 return ext4_chunk_trans_blocks(inode
, nrblocks
);
2017 * How many index/leaf blocks need to change/allocate to modify nrblocks?
2019 * if nrblocks are fit in a single extent (chunk flag is 1), then
2020 * in the worse case, each tree level index/leaf need to be changed
2021 * if the tree split due to insert a new extent, then the old tree
2022 * index/leaf need to be updated too
2024 * If the nrblocks are discontiguous, they could cause
2025 * the whole tree split more than once, but this is really rare.
2027 int ext4_ext_index_trans_blocks(struct inode
*inode
, int nrblocks
, int chunk
)
2030 int depth
= ext_depth(inode
);
2040 static int ext4_remove_blocks(handle_t
*handle
, struct inode
*inode
,
2041 struct ext4_extent
*ex
,
2042 ext4_lblk_t from
, ext4_lblk_t to
)
2044 unsigned short ee_len
= ext4_ext_get_actual_len(ex
);
2045 int flags
= EXT4_FREE_BLOCKS_FORGET
;
2047 if (S_ISDIR(inode
->i_mode
) || S_ISLNK(inode
->i_mode
))
2048 flags
|= EXT4_FREE_BLOCKS_METADATA
;
2049 #ifdef EXTENTS_STATS
2051 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
2052 spin_lock(&sbi
->s_ext_stats_lock
);
2053 sbi
->s_ext_blocks
+= ee_len
;
2054 sbi
->s_ext_extents
++;
2055 if (ee_len
< sbi
->s_ext_min
)
2056 sbi
->s_ext_min
= ee_len
;
2057 if (ee_len
> sbi
->s_ext_max
)
2058 sbi
->s_ext_max
= ee_len
;
2059 if (ext_depth(inode
) > sbi
->s_depth_max
)
2060 sbi
->s_depth_max
= ext_depth(inode
);
2061 spin_unlock(&sbi
->s_ext_stats_lock
);
2064 if (from
>= le32_to_cpu(ex
->ee_block
)
2065 && to
== le32_to_cpu(ex
->ee_block
) + ee_len
- 1) {
2070 num
= le32_to_cpu(ex
->ee_block
) + ee_len
- from
;
2071 start
= ext_pblock(ex
) + ee_len
- num
;
2072 ext_debug("free last %u blocks starting %llu\n", num
, start
);
2073 ext4_free_blocks(handle
, inode
, 0, start
, num
, flags
);
2074 } else if (from
== le32_to_cpu(ex
->ee_block
)
2075 && to
<= le32_to_cpu(ex
->ee_block
) + ee_len
- 1) {
2076 printk(KERN_INFO
"strange request: removal %u-%u from %u:%u\n",
2077 from
, to
, le32_to_cpu(ex
->ee_block
), ee_len
);
2079 printk(KERN_INFO
"strange request: removal(2) "
2080 "%u-%u from %u:%u\n",
2081 from
, to
, le32_to_cpu(ex
->ee_block
), ee_len
);
2087 ext4_ext_rm_leaf(handle_t
*handle
, struct inode
*inode
,
2088 struct ext4_ext_path
*path
, ext4_lblk_t start
)
2090 int err
= 0, correct_index
= 0;
2091 int depth
= ext_depth(inode
), credits
;
2092 struct ext4_extent_header
*eh
;
2093 ext4_lblk_t a
, b
, block
;
2095 ext4_lblk_t ex_ee_block
;
2096 unsigned short ex_ee_len
;
2097 unsigned uninitialized
= 0;
2098 struct ext4_extent
*ex
;
2100 /* the header must be checked already in ext4_ext_remove_space() */
2101 ext_debug("truncate since %u in leaf\n", start
);
2102 if (!path
[depth
].p_hdr
)
2103 path
[depth
].p_hdr
= ext_block_hdr(path
[depth
].p_bh
);
2104 eh
= path
[depth
].p_hdr
;
2107 /* find where to start removing */
2108 ex
= EXT_LAST_EXTENT(eh
);
2110 ex_ee_block
= le32_to_cpu(ex
->ee_block
);
2111 ex_ee_len
= ext4_ext_get_actual_len(ex
);
2113 while (ex
>= EXT_FIRST_EXTENT(eh
) &&
2114 ex_ee_block
+ ex_ee_len
> start
) {
2116 if (ext4_ext_is_uninitialized(ex
))
2121 ext_debug("remove ext %u:[%d]%d\n", ex_ee_block
,
2122 uninitialized
, ex_ee_len
);
2123 path
[depth
].p_ext
= ex
;
2125 a
= ex_ee_block
> start
? ex_ee_block
: start
;
2126 b
= ex_ee_block
+ ex_ee_len
- 1 < EXT_MAX_BLOCK
?
2127 ex_ee_block
+ ex_ee_len
- 1 : EXT_MAX_BLOCK
;
2129 ext_debug(" border %u:%u\n", a
, b
);
2131 if (a
!= ex_ee_block
&& b
!= ex_ee_block
+ ex_ee_len
- 1) {
2135 } else if (a
!= ex_ee_block
) {
2136 /* remove tail of the extent */
2137 block
= ex_ee_block
;
2139 } else if (b
!= ex_ee_block
+ ex_ee_len
- 1) {
2140 /* remove head of the extent */
2143 /* there is no "make a hole" API yet */
2146 /* remove whole extent: excellent! */
2147 block
= ex_ee_block
;
2149 BUG_ON(a
!= ex_ee_block
);
2150 BUG_ON(b
!= ex_ee_block
+ ex_ee_len
- 1);
2154 * 3 for leaf, sb, and inode plus 2 (bmap and group
2155 * descriptor) for each block group; assume two block
2156 * groups plus ex_ee_len/blocks_per_block_group for
2159 credits
= 7 + 2*(ex_ee_len
/EXT4_BLOCKS_PER_GROUP(inode
->i_sb
));
2160 if (ex
== EXT_FIRST_EXTENT(eh
)) {
2162 credits
+= (ext_depth(inode
)) + 1;
2164 credits
+= EXT4_MAXQUOTAS_TRANS_BLOCKS(inode
->i_sb
);
2166 err
= ext4_ext_truncate_extend_restart(handle
, inode
, credits
);
2170 err
= ext4_ext_get_access(handle
, inode
, path
+ depth
);
2174 err
= ext4_remove_blocks(handle
, inode
, ex
, a
, b
);
2179 /* this extent is removed; mark slot entirely unused */
2180 ext4_ext_store_pblock(ex
, 0);
2181 le16_add_cpu(&eh
->eh_entries
, -1);
2184 ex
->ee_block
= cpu_to_le32(block
);
2185 ex
->ee_len
= cpu_to_le16(num
);
2187 * Do not mark uninitialized if all the blocks in the
2188 * extent have been removed.
2190 if (uninitialized
&& num
)
2191 ext4_ext_mark_uninitialized(ex
);
2193 err
= ext4_ext_dirty(handle
, inode
, path
+ depth
);
2197 ext_debug("new extent: %u:%u:%llu\n", block
, num
,
2200 ex_ee_block
= le32_to_cpu(ex
->ee_block
);
2201 ex_ee_len
= ext4_ext_get_actual_len(ex
);
2204 if (correct_index
&& eh
->eh_entries
)
2205 err
= ext4_ext_correct_indexes(handle
, inode
, path
);
2207 /* if this leaf is free, then we should
2208 * remove it from index block above */
2209 if (err
== 0 && eh
->eh_entries
== 0 && path
[depth
].p_bh
!= NULL
)
2210 err
= ext4_ext_rm_idx(handle
, inode
, path
+ depth
);
2217 * ext4_ext_more_to_rm:
2218 * returns 1 if current index has to be freed (even partial)
2221 ext4_ext_more_to_rm(struct ext4_ext_path
*path
)
2223 BUG_ON(path
->p_idx
== NULL
);
2225 if (path
->p_idx
< EXT_FIRST_INDEX(path
->p_hdr
))
2229 * if truncate on deeper level happened, it wasn't partial,
2230 * so we have to consider current index for truncation
2232 if (le16_to_cpu(path
->p_hdr
->eh_entries
) == path
->p_block
)
2237 static int ext4_ext_remove_space(struct inode
*inode
, ext4_lblk_t start
)
2239 struct super_block
*sb
= inode
->i_sb
;
2240 int depth
= ext_depth(inode
);
2241 struct ext4_ext_path
*path
;
2245 ext_debug("truncate since %u\n", start
);
2247 /* probably first extent we're gonna free will be last in block */
2248 handle
= ext4_journal_start(inode
, depth
+ 1);
2250 return PTR_ERR(handle
);
2252 ext4_ext_invalidate_cache(inode
);
2255 * We start scanning from right side, freeing all the blocks
2256 * after i_size and walking into the tree depth-wise.
2258 path
= kzalloc(sizeof(struct ext4_ext_path
) * (depth
+ 1), GFP_NOFS
);
2260 ext4_journal_stop(handle
);
2263 path
[0].p_hdr
= ext_inode_hdr(inode
);
2264 if (ext4_ext_check(inode
, path
[0].p_hdr
, depth
)) {
2268 path
[0].p_depth
= depth
;
2270 while (i
>= 0 && err
== 0) {
2272 /* this is leaf block */
2273 err
= ext4_ext_rm_leaf(handle
, inode
, path
, start
);
2274 /* root level has p_bh == NULL, brelse() eats this */
2275 brelse(path
[i
].p_bh
);
2276 path
[i
].p_bh
= NULL
;
2281 /* this is index block */
2282 if (!path
[i
].p_hdr
) {
2283 ext_debug("initialize header\n");
2284 path
[i
].p_hdr
= ext_block_hdr(path
[i
].p_bh
);
2287 if (!path
[i
].p_idx
) {
2288 /* this level hasn't been touched yet */
2289 path
[i
].p_idx
= EXT_LAST_INDEX(path
[i
].p_hdr
);
2290 path
[i
].p_block
= le16_to_cpu(path
[i
].p_hdr
->eh_entries
)+1;
2291 ext_debug("init index ptr: hdr 0x%p, num %d\n",
2293 le16_to_cpu(path
[i
].p_hdr
->eh_entries
));
2295 /* we were already here, see at next index */
2299 ext_debug("level %d - index, first 0x%p, cur 0x%p\n",
2300 i
, EXT_FIRST_INDEX(path
[i
].p_hdr
),
2302 if (ext4_ext_more_to_rm(path
+ i
)) {
2303 struct buffer_head
*bh
;
2304 /* go to the next level */
2305 ext_debug("move to level %d (block %llu)\n",
2306 i
+ 1, idx_pblock(path
[i
].p_idx
));
2307 memset(path
+ i
+ 1, 0, sizeof(*path
));
2308 bh
= sb_bread(sb
, idx_pblock(path
[i
].p_idx
));
2310 /* should we reset i_size? */
2314 if (WARN_ON(i
+ 1 > depth
)) {
2318 if (ext4_ext_check(inode
, ext_block_hdr(bh
),
2323 path
[i
+ 1].p_bh
= bh
;
2325 /* save actual number of indexes since this
2326 * number is changed at the next iteration */
2327 path
[i
].p_block
= le16_to_cpu(path
[i
].p_hdr
->eh_entries
);
2330 /* we finished processing this index, go up */
2331 if (path
[i
].p_hdr
->eh_entries
== 0 && i
> 0) {
2332 /* index is empty, remove it;
2333 * handle must be already prepared by the
2334 * truncatei_leaf() */
2335 err
= ext4_ext_rm_idx(handle
, inode
, path
+ i
);
2337 /* root level has p_bh == NULL, brelse() eats this */
2338 brelse(path
[i
].p_bh
);
2339 path
[i
].p_bh
= NULL
;
2341 ext_debug("return to level %d\n", i
);
2345 /* TODO: flexible tree reduction should be here */
2346 if (path
->p_hdr
->eh_entries
== 0) {
2348 * truncate to zero freed all the tree,
2349 * so we need to correct eh_depth
2351 err
= ext4_ext_get_access(handle
, inode
, path
);
2353 ext_inode_hdr(inode
)->eh_depth
= 0;
2354 ext_inode_hdr(inode
)->eh_max
=
2355 cpu_to_le16(ext4_ext_space_root(inode
, 0));
2356 err
= ext4_ext_dirty(handle
, inode
, path
);
2360 ext4_ext_drop_refs(path
);
2362 ext4_journal_stop(handle
);
2368 * called at mount time
2370 void ext4_ext_init(struct super_block
*sb
)
2373 * possible initialization would be here
2376 if (EXT4_HAS_INCOMPAT_FEATURE(sb
, EXT4_FEATURE_INCOMPAT_EXTENTS
)) {
2377 #if defined(AGGRESSIVE_TEST) || defined(CHECK_BINSEARCH) || defined(EXTENTS_STATS)
2378 printk(KERN_INFO
"EXT4-fs: file extents enabled");
2379 #ifdef AGGRESSIVE_TEST
2380 printk(", aggressive tests");
2382 #ifdef CHECK_BINSEARCH
2383 printk(", check binsearch");
2385 #ifdef EXTENTS_STATS
2390 #ifdef EXTENTS_STATS
2391 spin_lock_init(&EXT4_SB(sb
)->s_ext_stats_lock
);
2392 EXT4_SB(sb
)->s_ext_min
= 1 << 30;
2393 EXT4_SB(sb
)->s_ext_max
= 0;
2399 * called at umount time
2401 void ext4_ext_release(struct super_block
*sb
)
2403 if (!EXT4_HAS_INCOMPAT_FEATURE(sb
, EXT4_FEATURE_INCOMPAT_EXTENTS
))
2406 #ifdef EXTENTS_STATS
2407 if (EXT4_SB(sb
)->s_ext_blocks
&& EXT4_SB(sb
)->s_ext_extents
) {
2408 struct ext4_sb_info
*sbi
= EXT4_SB(sb
);
2409 printk(KERN_ERR
"EXT4-fs: %lu blocks in %lu extents (%lu ave)\n",
2410 sbi
->s_ext_blocks
, sbi
->s_ext_extents
,
2411 sbi
->s_ext_blocks
/ sbi
->s_ext_extents
);
2412 printk(KERN_ERR
"EXT4-fs: extents: %lu min, %lu max, max depth %lu\n",
2413 sbi
->s_ext_min
, sbi
->s_ext_max
, sbi
->s_depth_max
);
2418 static void bi_complete(struct bio
*bio
, int error
)
2420 complete((struct completion
*)bio
->bi_private
);
2423 /* FIXME!! we need to try to merge to left or right after zero-out */
2424 static int ext4_ext_zeroout(struct inode
*inode
, struct ext4_extent
*ex
)
2428 int blkbits
, blocksize
;
2430 struct completion event
;
2431 unsigned int ee_len
, len
, done
, offset
;
2434 blkbits
= inode
->i_blkbits
;
2435 blocksize
= inode
->i_sb
->s_blocksize
;
2436 ee_len
= ext4_ext_get_actual_len(ex
);
2437 ee_pblock
= ext_pblock(ex
);
2439 /* convert ee_pblock to 512 byte sectors */
2440 ee_pblock
= ee_pblock
<< (blkbits
- 9);
2442 while (ee_len
> 0) {
2444 if (ee_len
> BIO_MAX_PAGES
)
2445 len
= BIO_MAX_PAGES
;
2449 bio
= bio_alloc(GFP_NOIO
, len
);
2450 bio
->bi_sector
= ee_pblock
;
2451 bio
->bi_bdev
= inode
->i_sb
->s_bdev
;
2455 while (done
< len
) {
2456 ret
= bio_add_page(bio
, ZERO_PAGE(0),
2458 if (ret
!= blocksize
) {
2460 * We can't add any more pages because of
2461 * hardware limitations. Start a new bio.
2466 offset
+= blocksize
;
2467 if (offset
>= PAGE_CACHE_SIZE
)
2471 init_completion(&event
);
2472 bio
->bi_private
= &event
;
2473 bio
->bi_end_io
= bi_complete
;
2474 submit_bio(WRITE
, bio
);
2475 wait_for_completion(&event
);
2477 if (test_bit(BIO_UPTODATE
, &bio
->bi_flags
))
2485 ee_pblock
+= done
<< (blkbits
- 9);
2490 #define EXT4_EXT_ZERO_LEN 7
2492 * This function is called by ext4_ext_get_blocks() if someone tries to write
2493 * to an uninitialized extent. It may result in splitting the uninitialized
2494 * extent into multiple extents (upto three - one initialized and two
2496 * There are three possibilities:
2497 * a> There is no split required: Entire extent should be initialized
2498 * b> Splits in two extents: Write is happening at either end of the extent
2499 * c> Splits in three extents: Somone is writing in middle of the extent
2501 static int ext4_ext_convert_to_initialized(handle_t
*handle
,
2502 struct inode
*inode
,
2503 struct ext4_ext_path
*path
,
2505 unsigned int max_blocks
)
2507 struct ext4_extent
*ex
, newex
, orig_ex
;
2508 struct ext4_extent
*ex1
= NULL
;
2509 struct ext4_extent
*ex2
= NULL
;
2510 struct ext4_extent
*ex3
= NULL
;
2511 struct ext4_extent_header
*eh
;
2512 ext4_lblk_t ee_block
;
2513 unsigned int allocated
, ee_len
, depth
;
2514 ext4_fsblk_t newblock
;
2518 depth
= ext_depth(inode
);
2519 eh
= path
[depth
].p_hdr
;
2520 ex
= path
[depth
].p_ext
;
2521 ee_block
= le32_to_cpu(ex
->ee_block
);
2522 ee_len
= ext4_ext_get_actual_len(ex
);
2523 allocated
= ee_len
- (iblock
- ee_block
);
2524 newblock
= iblock
- ee_block
+ ext_pblock(ex
);
2526 orig_ex
.ee_block
= ex
->ee_block
;
2527 orig_ex
.ee_len
= cpu_to_le16(ee_len
);
2528 ext4_ext_store_pblock(&orig_ex
, ext_pblock(ex
));
2530 err
= ext4_ext_get_access(handle
, inode
, path
+ depth
);
2533 /* If extent has less than 2*EXT4_EXT_ZERO_LEN zerout directly */
2534 if (ee_len
<= 2*EXT4_EXT_ZERO_LEN
) {
2535 err
= ext4_ext_zeroout(inode
, &orig_ex
);
2537 goto fix_extent_len
;
2538 /* update the extent length and mark as initialized */
2539 ex
->ee_block
= orig_ex
.ee_block
;
2540 ex
->ee_len
= orig_ex
.ee_len
;
2541 ext4_ext_store_pblock(ex
, ext_pblock(&orig_ex
));
2542 ext4_ext_dirty(handle
, inode
, path
+ depth
);
2543 /* zeroed the full extent */
2547 /* ex1: ee_block to iblock - 1 : uninitialized */
2548 if (iblock
> ee_block
) {
2550 ex1
->ee_len
= cpu_to_le16(iblock
- ee_block
);
2551 ext4_ext_mark_uninitialized(ex1
);
2555 * for sanity, update the length of the ex2 extent before
2556 * we insert ex3, if ex1 is NULL. This is to avoid temporary
2557 * overlap of blocks.
2559 if (!ex1
&& allocated
> max_blocks
)
2560 ex2
->ee_len
= cpu_to_le16(max_blocks
);
2561 /* ex3: to ee_block + ee_len : uninitialised */
2562 if (allocated
> max_blocks
) {
2563 unsigned int newdepth
;
2564 /* If extent has less than EXT4_EXT_ZERO_LEN zerout directly */
2565 if (allocated
<= EXT4_EXT_ZERO_LEN
) {
2567 * iblock == ee_block is handled by the zerouout
2569 * Mark first half uninitialized.
2570 * Mark second half initialized and zero out the
2571 * initialized extent
2573 ex
->ee_block
= orig_ex
.ee_block
;
2574 ex
->ee_len
= cpu_to_le16(ee_len
- allocated
);
2575 ext4_ext_mark_uninitialized(ex
);
2576 ext4_ext_store_pblock(ex
, ext_pblock(&orig_ex
));
2577 ext4_ext_dirty(handle
, inode
, path
+ depth
);
2580 ex3
->ee_block
= cpu_to_le32(iblock
);
2581 ext4_ext_store_pblock(ex3
, newblock
);
2582 ex3
->ee_len
= cpu_to_le16(allocated
);
2583 err
= ext4_ext_insert_extent(handle
, inode
, path
,
2585 if (err
== -ENOSPC
) {
2586 err
= ext4_ext_zeroout(inode
, &orig_ex
);
2588 goto fix_extent_len
;
2589 ex
->ee_block
= orig_ex
.ee_block
;
2590 ex
->ee_len
= orig_ex
.ee_len
;
2591 ext4_ext_store_pblock(ex
, ext_pblock(&orig_ex
));
2592 ext4_ext_dirty(handle
, inode
, path
+ depth
);
2593 /* blocks available from iblock */
2597 goto fix_extent_len
;
2600 * We need to zero out the second half because
2601 * an fallocate request can update file size and
2602 * converting the second half to initialized extent
2603 * implies that we can leak some junk data to user
2606 err
= ext4_ext_zeroout(inode
, ex3
);
2609 * We should actually mark the
2610 * second half as uninit and return error
2611 * Insert would have changed the extent
2613 depth
= ext_depth(inode
);
2614 ext4_ext_drop_refs(path
);
2615 path
= ext4_ext_find_extent(inode
,
2618 err
= PTR_ERR(path
);
2621 /* get the second half extent details */
2622 ex
= path
[depth
].p_ext
;
2623 err
= ext4_ext_get_access(handle
, inode
,
2627 ext4_ext_mark_uninitialized(ex
);
2628 ext4_ext_dirty(handle
, inode
, path
+ depth
);
2632 /* zeroed the second half */
2636 ex3
->ee_block
= cpu_to_le32(iblock
+ max_blocks
);
2637 ext4_ext_store_pblock(ex3
, newblock
+ max_blocks
);
2638 ex3
->ee_len
= cpu_to_le16(allocated
- max_blocks
);
2639 ext4_ext_mark_uninitialized(ex3
);
2640 err
= ext4_ext_insert_extent(handle
, inode
, path
, ex3
, 0);
2641 if (err
== -ENOSPC
) {
2642 err
= ext4_ext_zeroout(inode
, &orig_ex
);
2644 goto fix_extent_len
;
2645 /* update the extent length and mark as initialized */
2646 ex
->ee_block
= orig_ex
.ee_block
;
2647 ex
->ee_len
= orig_ex
.ee_len
;
2648 ext4_ext_store_pblock(ex
, ext_pblock(&orig_ex
));
2649 ext4_ext_dirty(handle
, inode
, path
+ depth
);
2650 /* zeroed the full extent */
2651 /* blocks available from iblock */
2655 goto fix_extent_len
;
2657 * The depth, and hence eh & ex might change
2658 * as part of the insert above.
2660 newdepth
= ext_depth(inode
);
2662 * update the extent length after successful insert of the
2665 orig_ex
.ee_len
= cpu_to_le16(ee_len
-
2666 ext4_ext_get_actual_len(ex3
));
2668 ext4_ext_drop_refs(path
);
2669 path
= ext4_ext_find_extent(inode
, iblock
, path
);
2671 err
= PTR_ERR(path
);
2674 eh
= path
[depth
].p_hdr
;
2675 ex
= path
[depth
].p_ext
;
2679 err
= ext4_ext_get_access(handle
, inode
, path
+ depth
);
2683 allocated
= max_blocks
;
2685 /* If extent has less than EXT4_EXT_ZERO_LEN and we are trying
2686 * to insert a extent in the middle zerout directly
2687 * otherwise give the extent a chance to merge to left
2689 if (le16_to_cpu(orig_ex
.ee_len
) <= EXT4_EXT_ZERO_LEN
&&
2690 iblock
!= ee_block
) {
2691 err
= ext4_ext_zeroout(inode
, &orig_ex
);
2693 goto fix_extent_len
;
2694 /* update the extent length and mark as initialized */
2695 ex
->ee_block
= orig_ex
.ee_block
;
2696 ex
->ee_len
= orig_ex
.ee_len
;
2697 ext4_ext_store_pblock(ex
, ext_pblock(&orig_ex
));
2698 ext4_ext_dirty(handle
, inode
, path
+ depth
);
2699 /* zero out the first half */
2700 /* blocks available from iblock */
2705 * If there was a change of depth as part of the
2706 * insertion of ex3 above, we need to update the length
2707 * of the ex1 extent again here
2709 if (ex1
&& ex1
!= ex
) {
2711 ex1
->ee_len
= cpu_to_le16(iblock
- ee_block
);
2712 ext4_ext_mark_uninitialized(ex1
);
2715 /* ex2: iblock to iblock + maxblocks-1 : initialised */
2716 ex2
->ee_block
= cpu_to_le32(iblock
);
2717 ext4_ext_store_pblock(ex2
, newblock
);
2718 ex2
->ee_len
= cpu_to_le16(allocated
);
2722 * New (initialized) extent starts from the first block
2723 * in the current extent. i.e., ex2 == ex
2724 * We have to see if it can be merged with the extent
2727 if (ex2
> EXT_FIRST_EXTENT(eh
)) {
2729 * To merge left, pass "ex2 - 1" to try_to_merge(),
2730 * since it merges towards right _only_.
2732 ret
= ext4_ext_try_to_merge(inode
, path
, ex2
- 1);
2734 err
= ext4_ext_correct_indexes(handle
, inode
, path
);
2737 depth
= ext_depth(inode
);
2742 * Try to Merge towards right. This might be required
2743 * only when the whole extent is being written to.
2744 * i.e. ex2 == ex and ex3 == NULL.
2747 ret
= ext4_ext_try_to_merge(inode
, path
, ex2
);
2749 err
= ext4_ext_correct_indexes(handle
, inode
, path
);
2754 /* Mark modified extent as dirty */
2755 err
= ext4_ext_dirty(handle
, inode
, path
+ depth
);
2758 err
= ext4_ext_insert_extent(handle
, inode
, path
, &newex
, 0);
2759 if (err
== -ENOSPC
) {
2760 err
= ext4_ext_zeroout(inode
, &orig_ex
);
2762 goto fix_extent_len
;
2763 /* update the extent length and mark as initialized */
2764 ex
->ee_block
= orig_ex
.ee_block
;
2765 ex
->ee_len
= orig_ex
.ee_len
;
2766 ext4_ext_store_pblock(ex
, ext_pblock(&orig_ex
));
2767 ext4_ext_dirty(handle
, inode
, path
+ depth
);
2768 /* zero out the first half */
2771 goto fix_extent_len
;
2773 ext4_ext_show_leaf(inode
, path
);
2774 return err
? err
: allocated
;
2777 ex
->ee_block
= orig_ex
.ee_block
;
2778 ex
->ee_len
= orig_ex
.ee_len
;
2779 ext4_ext_store_pblock(ex
, ext_pblock(&orig_ex
));
2780 ext4_ext_mark_uninitialized(ex
);
2781 ext4_ext_dirty(handle
, inode
, path
+ depth
);
2786 * This function is called by ext4_ext_get_blocks() from
2787 * ext4_get_blocks_dio_write() when DIO to write
2788 * to an uninitialized extent.
2790 * Writing to an uninitized extent may result in splitting the uninitialized
2791 * extent into multiple /intialized unintialized extents (up to three)
2792 * There are three possibilities:
2793 * a> There is no split required: Entire extent should be uninitialized
2794 * b> Splits in two extents: Write is happening at either end of the extent
2795 * c> Splits in three extents: Somone is writing in middle of the extent
2797 * One of more index blocks maybe needed if the extent tree grow after
2798 * the unintialized extent split. To prevent ENOSPC occur at the IO
2799 * complete, we need to split the uninitialized extent before DIO submit
2800 * the IO. The uninitilized extent called at this time will be split
2801 * into three uninitialized extent(at most). After IO complete, the part
2802 * being filled will be convert to initialized by the end_io callback function
2803 * via ext4_convert_unwritten_extents().
2805 * Returns the size of uninitialized extent to be written on success.
2807 static int ext4_split_unwritten_extents(handle_t
*handle
,
2808 struct inode
*inode
,
2809 struct ext4_ext_path
*path
,
2811 unsigned int max_blocks
,
2814 struct ext4_extent
*ex
, newex
, orig_ex
;
2815 struct ext4_extent
*ex1
= NULL
;
2816 struct ext4_extent
*ex2
= NULL
;
2817 struct ext4_extent
*ex3
= NULL
;
2818 struct ext4_extent_header
*eh
;
2819 ext4_lblk_t ee_block
;
2820 unsigned int allocated
, ee_len
, depth
;
2821 ext4_fsblk_t newblock
;
2824 ext_debug("ext4_split_unwritten_extents: inode %lu,"
2825 "iblock %llu, max_blocks %u\n", inode
->i_ino
,
2826 (unsigned long long)iblock
, max_blocks
);
2827 depth
= ext_depth(inode
);
2828 eh
= path
[depth
].p_hdr
;
2829 ex
= path
[depth
].p_ext
;
2830 ee_block
= le32_to_cpu(ex
->ee_block
);
2831 ee_len
= ext4_ext_get_actual_len(ex
);
2832 allocated
= ee_len
- (iblock
- ee_block
);
2833 newblock
= iblock
- ee_block
+ ext_pblock(ex
);
2835 orig_ex
.ee_block
= ex
->ee_block
;
2836 orig_ex
.ee_len
= cpu_to_le16(ee_len
);
2837 ext4_ext_store_pblock(&orig_ex
, ext_pblock(ex
));
2840 * If the uninitialized extent begins at the same logical
2841 * block where the write begins, and the write completely
2842 * covers the extent, then we don't need to split it.
2844 if ((iblock
== ee_block
) && (allocated
<= max_blocks
))
2847 err
= ext4_ext_get_access(handle
, inode
, path
+ depth
);
2850 /* ex1: ee_block to iblock - 1 : uninitialized */
2851 if (iblock
> ee_block
) {
2853 ex1
->ee_len
= cpu_to_le16(iblock
- ee_block
);
2854 ext4_ext_mark_uninitialized(ex1
);
2858 * for sanity, update the length of the ex2 extent before
2859 * we insert ex3, if ex1 is NULL. This is to avoid temporary
2860 * overlap of blocks.
2862 if (!ex1
&& allocated
> max_blocks
)
2863 ex2
->ee_len
= cpu_to_le16(max_blocks
);
2864 /* ex3: to ee_block + ee_len : uninitialised */
2865 if (allocated
> max_blocks
) {
2866 unsigned int newdepth
;
2868 ex3
->ee_block
= cpu_to_le32(iblock
+ max_blocks
);
2869 ext4_ext_store_pblock(ex3
, newblock
+ max_blocks
);
2870 ex3
->ee_len
= cpu_to_le16(allocated
- max_blocks
);
2871 ext4_ext_mark_uninitialized(ex3
);
2872 err
= ext4_ext_insert_extent(handle
, inode
, path
, ex3
, flags
);
2873 if (err
== -ENOSPC
) {
2874 err
= ext4_ext_zeroout(inode
, &orig_ex
);
2876 goto fix_extent_len
;
2877 /* update the extent length and mark as initialized */
2878 ex
->ee_block
= orig_ex
.ee_block
;
2879 ex
->ee_len
= orig_ex
.ee_len
;
2880 ext4_ext_store_pblock(ex
, ext_pblock(&orig_ex
));
2881 ext4_ext_dirty(handle
, inode
, path
+ depth
);
2882 /* zeroed the full extent */
2883 /* blocks available from iblock */
2887 goto fix_extent_len
;
2889 * The depth, and hence eh & ex might change
2890 * as part of the insert above.
2892 newdepth
= ext_depth(inode
);
2894 * update the extent length after successful insert of the
2897 orig_ex
.ee_len
= cpu_to_le16(ee_len
-
2898 ext4_ext_get_actual_len(ex3
));
2900 ext4_ext_drop_refs(path
);
2901 path
= ext4_ext_find_extent(inode
, iblock
, path
);
2903 err
= PTR_ERR(path
);
2906 eh
= path
[depth
].p_hdr
;
2907 ex
= path
[depth
].p_ext
;
2911 err
= ext4_ext_get_access(handle
, inode
, path
+ depth
);
2915 allocated
= max_blocks
;
2918 * If there was a change of depth as part of the
2919 * insertion of ex3 above, we need to update the length
2920 * of the ex1 extent again here
2922 if (ex1
&& ex1
!= ex
) {
2924 ex1
->ee_len
= cpu_to_le16(iblock
- ee_block
);
2925 ext4_ext_mark_uninitialized(ex1
);
2929 * ex2: iblock to iblock + maxblocks-1 : to be direct IO written,
2930 * uninitialised still.
2932 ex2
->ee_block
= cpu_to_le32(iblock
);
2933 ext4_ext_store_pblock(ex2
, newblock
);
2934 ex2
->ee_len
= cpu_to_le16(allocated
);
2935 ext4_ext_mark_uninitialized(ex2
);
2938 /* Mark modified extent as dirty */
2939 err
= ext4_ext_dirty(handle
, inode
, path
+ depth
);
2940 ext_debug("out here\n");
2943 err
= ext4_ext_insert_extent(handle
, inode
, path
, &newex
, flags
);
2944 if (err
== -ENOSPC
) {
2945 err
= ext4_ext_zeroout(inode
, &orig_ex
);
2947 goto fix_extent_len
;
2948 /* update the extent length and mark as initialized */
2949 ex
->ee_block
= orig_ex
.ee_block
;
2950 ex
->ee_len
= orig_ex
.ee_len
;
2951 ext4_ext_store_pblock(ex
, ext_pblock(&orig_ex
));
2952 ext4_ext_dirty(handle
, inode
, path
+ depth
);
2953 /* zero out the first half */
2956 goto fix_extent_len
;
2958 ext4_ext_show_leaf(inode
, path
);
2959 return err
? err
: allocated
;
2962 ex
->ee_block
= orig_ex
.ee_block
;
2963 ex
->ee_len
= orig_ex
.ee_len
;
2964 ext4_ext_store_pblock(ex
, ext_pblock(&orig_ex
));
2965 ext4_ext_mark_uninitialized(ex
);
2966 ext4_ext_dirty(handle
, inode
, path
+ depth
);
2969 static int ext4_convert_unwritten_extents_dio(handle_t
*handle
,
2970 struct inode
*inode
,
2971 struct ext4_ext_path
*path
)
2973 struct ext4_extent
*ex
;
2974 struct ext4_extent_header
*eh
;
2979 depth
= ext_depth(inode
);
2980 eh
= path
[depth
].p_hdr
;
2981 ex
= path
[depth
].p_ext
;
2983 err
= ext4_ext_get_access(handle
, inode
, path
+ depth
);
2986 /* first mark the extent as initialized */
2987 ext4_ext_mark_initialized(ex
);
2990 * We have to see if it can be merged with the extent
2993 if (ex
> EXT_FIRST_EXTENT(eh
)) {
2995 * To merge left, pass "ex - 1" to try_to_merge(),
2996 * since it merges towards right _only_.
2998 ret
= ext4_ext_try_to_merge(inode
, path
, ex
- 1);
3000 err
= ext4_ext_correct_indexes(handle
, inode
, path
);
3003 depth
= ext_depth(inode
);
3008 * Try to Merge towards right.
3010 ret
= ext4_ext_try_to_merge(inode
, path
, ex
);
3012 err
= ext4_ext_correct_indexes(handle
, inode
, path
);
3015 depth
= ext_depth(inode
);
3017 /* Mark modified extent as dirty */
3018 err
= ext4_ext_dirty(handle
, inode
, path
+ depth
);
3020 ext4_ext_show_leaf(inode
, path
);
3025 ext4_ext_handle_uninitialized_extents(handle_t
*handle
, struct inode
*inode
,
3026 ext4_lblk_t iblock
, unsigned int max_blocks
,
3027 struct ext4_ext_path
*path
, int flags
,
3028 unsigned int allocated
, struct buffer_head
*bh_result
,
3029 ext4_fsblk_t newblock
)
3033 ext4_io_end_t
*io
= EXT4_I(inode
)->cur_aio_dio
;
3035 ext_debug("ext4_ext_handle_uninitialized_extents: inode %lu, logical"
3036 "block %llu, max_blocks %u, flags %d, allocated %u",
3037 inode
->i_ino
, (unsigned long long)iblock
, max_blocks
,
3039 ext4_ext_show_leaf(inode
, path
);
3041 /* DIO get_block() before submit the IO, split the extent */
3042 if (flags
== EXT4_GET_BLOCKS_DIO_CREATE_EXT
) {
3043 ret
= ext4_split_unwritten_extents(handle
,
3044 inode
, path
, iblock
,
3047 * Flag the inode(non aio case) or end_io struct (aio case)
3048 * that this IO needs to convertion to written when IO is
3052 io
->flag
= DIO_AIO_UNWRITTEN
;
3054 EXT4_I(inode
)->i_state
|= EXT4_STATE_DIO_UNWRITTEN
;
3057 /* async DIO end_io complete, convert the filled extent to written */
3058 if (flags
== EXT4_GET_BLOCKS_DIO_CONVERT_EXT
) {
3059 ret
= ext4_convert_unwritten_extents_dio(handle
, inode
,
3062 ext4_update_inode_fsync_trans(handle
, inode
, 1);
3065 /* buffered IO case */
3067 * repeat fallocate creation request
3068 * we already have an unwritten extent
3070 if (flags
& EXT4_GET_BLOCKS_UNINIT_EXT
)
3073 /* buffered READ or buffered write_begin() lookup */
3074 if ((flags
& EXT4_GET_BLOCKS_CREATE
) == 0) {
3076 * We have blocks reserved already. We
3077 * return allocated blocks so that delalloc
3078 * won't do block reservation for us. But
3079 * the buffer head will be unmapped so that
3080 * a read from the block returns 0s.
3082 set_buffer_unwritten(bh_result
);
3086 /* buffered write, writepage time, convert*/
3087 ret
= ext4_ext_convert_to_initialized(handle
, inode
,
3091 ext4_update_inode_fsync_trans(handle
, inode
, 1);
3098 set_buffer_new(bh_result
);
3100 set_buffer_mapped(bh_result
);
3102 if (allocated
> max_blocks
)
3103 allocated
= max_blocks
;
3104 ext4_ext_show_leaf(inode
, path
);
3105 bh_result
->b_bdev
= inode
->i_sb
->s_bdev
;
3106 bh_result
->b_blocknr
= newblock
;
3109 ext4_ext_drop_refs(path
);
3112 return err
? err
: allocated
;
3115 * Block allocation/map/preallocation routine for extents based files
3118 * Need to be called with
3119 * down_read(&EXT4_I(inode)->i_data_sem) if not allocating file system block
3120 * (ie, create is zero). Otherwise down_write(&EXT4_I(inode)->i_data_sem)
3122 * return > 0, number of of blocks already mapped/allocated
3123 * if create == 0 and these are pre-allocated blocks
3124 * buffer head is unmapped
3125 * otherwise blocks are mapped
3127 * return = 0, if plain look up failed (blocks have not been allocated)
3128 * buffer head is unmapped
3130 * return < 0, error case.
3132 int ext4_ext_get_blocks(handle_t
*handle
, struct inode
*inode
,
3134 unsigned int max_blocks
, struct buffer_head
*bh_result
,
3137 struct ext4_ext_path
*path
= NULL
;
3138 struct ext4_extent_header
*eh
;
3139 struct ext4_extent newex
, *ex
;
3140 ext4_fsblk_t newblock
;
3141 int err
= 0, depth
, ret
, cache_type
;
3142 unsigned int allocated
= 0;
3143 struct ext4_allocation_request ar
;
3144 ext4_io_end_t
*io
= EXT4_I(inode
)->cur_aio_dio
;
3146 __clear_bit(BH_New
, &bh_result
->b_state
);
3147 ext_debug("blocks %u/%u requested for inode %lu\n",
3148 iblock
, max_blocks
, inode
->i_ino
);
3150 /* check in cache */
3151 cache_type
= ext4_ext_in_cache(inode
, iblock
, &newex
);
3153 if (cache_type
== EXT4_EXT_CACHE_GAP
) {
3154 if ((flags
& EXT4_GET_BLOCKS_CREATE
) == 0) {
3156 * block isn't allocated yet and
3157 * user doesn't want to allocate it
3161 /* we should allocate requested block */
3162 } else if (cache_type
== EXT4_EXT_CACHE_EXTENT
) {
3163 /* block is already allocated */
3165 - le32_to_cpu(newex
.ee_block
)
3166 + ext_pblock(&newex
);
3167 /* number of remaining blocks in the extent */
3168 allocated
= ext4_ext_get_actual_len(&newex
) -
3169 (iblock
- le32_to_cpu(newex
.ee_block
));
3176 /* find extent for this block */
3177 path
= ext4_ext_find_extent(inode
, iblock
, NULL
);
3179 err
= PTR_ERR(path
);
3184 depth
= ext_depth(inode
);
3187 * consistent leaf must not be empty;
3188 * this situation is possible, though, _during_ tree modification;
3189 * this is why assert can't be put in ext4_ext_find_extent()
3191 BUG_ON(path
[depth
].p_ext
== NULL
&& depth
!= 0);
3192 eh
= path
[depth
].p_hdr
;
3194 ex
= path
[depth
].p_ext
;
3196 ext4_lblk_t ee_block
= le32_to_cpu(ex
->ee_block
);
3197 ext4_fsblk_t ee_start
= ext_pblock(ex
);
3198 unsigned short ee_len
;
3201 * Uninitialized extents are treated as holes, except that
3202 * we split out initialized portions during a write.
3204 ee_len
= ext4_ext_get_actual_len(ex
);
3205 /* if found extent covers block, simply return it */
3206 if (iblock
>= ee_block
&& iblock
< ee_block
+ ee_len
) {
3207 newblock
= iblock
- ee_block
+ ee_start
;
3208 /* number of remaining blocks in the extent */
3209 allocated
= ee_len
- (iblock
- ee_block
);
3210 ext_debug("%u fit into %u:%d -> %llu\n", iblock
,
3211 ee_block
, ee_len
, newblock
);
3213 /* Do not put uninitialized extent in the cache */
3214 if (!ext4_ext_is_uninitialized(ex
)) {
3215 ext4_ext_put_in_cache(inode
, ee_block
,
3217 EXT4_EXT_CACHE_EXTENT
);
3220 ret
= ext4_ext_handle_uninitialized_extents(handle
,
3221 inode
, iblock
, max_blocks
, path
,
3222 flags
, allocated
, bh_result
, newblock
);
3228 * requested block isn't allocated yet;
3229 * we couldn't try to create block if create flag is zero
3231 if ((flags
& EXT4_GET_BLOCKS_CREATE
) == 0) {
3233 * put just found gap into cache to speed up
3234 * subsequent requests
3236 ext4_ext_put_gap_in_cache(inode
, path
, iblock
);
3240 * Okay, we need to do block allocation.
3243 /* find neighbour allocated blocks */
3245 err
= ext4_ext_search_left(inode
, path
, &ar
.lleft
, &ar
.pleft
);
3249 err
= ext4_ext_search_right(inode
, path
, &ar
.lright
, &ar
.pright
);
3254 * See if request is beyond maximum number of blocks we can have in
3255 * a single extent. For an initialized extent this limit is
3256 * EXT_INIT_MAX_LEN and for an uninitialized extent this limit is
3257 * EXT_UNINIT_MAX_LEN.
3259 if (max_blocks
> EXT_INIT_MAX_LEN
&&
3260 !(flags
& EXT4_GET_BLOCKS_UNINIT_EXT
))
3261 max_blocks
= EXT_INIT_MAX_LEN
;
3262 else if (max_blocks
> EXT_UNINIT_MAX_LEN
&&
3263 (flags
& EXT4_GET_BLOCKS_UNINIT_EXT
))
3264 max_blocks
= EXT_UNINIT_MAX_LEN
;
3266 /* Check if we can really insert (iblock)::(iblock+max_blocks) extent */
3267 newex
.ee_block
= cpu_to_le32(iblock
);
3268 newex
.ee_len
= cpu_to_le16(max_blocks
);
3269 err
= ext4_ext_check_overlap(inode
, &newex
, path
);
3271 allocated
= ext4_ext_get_actual_len(&newex
);
3273 allocated
= max_blocks
;
3275 /* allocate new block */
3277 ar
.goal
= ext4_ext_find_goal(inode
, path
, iblock
);
3278 ar
.logical
= iblock
;
3280 if (S_ISREG(inode
->i_mode
))
3281 ar
.flags
= EXT4_MB_HINT_DATA
;
3283 /* disable in-core preallocation for non-regular files */
3285 newblock
= ext4_mb_new_blocks(handle
, &ar
, &err
);
3288 ext_debug("allocate new block: goal %llu, found %llu/%u\n",
3289 ar
.goal
, newblock
, allocated
);
3291 /* try to insert new extent into found leaf and return */
3292 ext4_ext_store_pblock(&newex
, newblock
);
3293 newex
.ee_len
= cpu_to_le16(ar
.len
);
3294 /* Mark uninitialized */
3295 if (flags
& EXT4_GET_BLOCKS_UNINIT_EXT
){
3296 ext4_ext_mark_uninitialized(&newex
);
3298 * io_end structure was created for every async
3299 * direct IO write to the middle of the file.
3300 * To avoid unecessary convertion for every aio dio rewrite
3301 * to the mid of file, here we flag the IO that is really
3302 * need the convertion.
3303 * For non asycn direct IO case, flag the inode state
3304 * that we need to perform convertion when IO is done.
3306 if (flags
== EXT4_GET_BLOCKS_DIO_CREATE_EXT
) {
3308 io
->flag
= DIO_AIO_UNWRITTEN
;
3310 EXT4_I(inode
)->i_state
|=
3311 EXT4_STATE_DIO_UNWRITTEN
;;
3314 err
= ext4_ext_insert_extent(handle
, inode
, path
, &newex
, flags
);
3316 /* free data blocks we just allocated */
3317 /* not a good idea to call discard here directly,
3318 * but otherwise we'd need to call it every free() */
3319 ext4_discard_preallocations(inode
);
3320 ext4_free_blocks(handle
, inode
, 0, ext_pblock(&newex
),
3321 ext4_ext_get_actual_len(&newex
), 0);
3325 /* previous routine could use block we allocated */
3326 newblock
= ext_pblock(&newex
);
3327 allocated
= ext4_ext_get_actual_len(&newex
);
3328 set_buffer_new(bh_result
);
3331 * Cache the extent and update transaction to commit on fdatasync only
3332 * when it is _not_ an uninitialized extent.
3334 if ((flags
& EXT4_GET_BLOCKS_UNINIT_EXT
) == 0) {
3335 ext4_ext_put_in_cache(inode
, iblock
, allocated
, newblock
,
3336 EXT4_EXT_CACHE_EXTENT
);
3337 ext4_update_inode_fsync_trans(handle
, inode
, 1);
3339 ext4_update_inode_fsync_trans(handle
, inode
, 0);
3341 if (allocated
> max_blocks
)
3342 allocated
= max_blocks
;
3343 ext4_ext_show_leaf(inode
, path
);
3344 set_buffer_mapped(bh_result
);
3345 bh_result
->b_bdev
= inode
->i_sb
->s_bdev
;
3346 bh_result
->b_blocknr
= newblock
;
3349 ext4_ext_drop_refs(path
);
3352 return err
? err
: allocated
;
3355 void ext4_ext_truncate(struct inode
*inode
)
3357 struct address_space
*mapping
= inode
->i_mapping
;
3358 struct super_block
*sb
= inode
->i_sb
;
3359 ext4_lblk_t last_block
;
3364 * probably first extent we're gonna free will be last in block
3366 err
= ext4_writepage_trans_blocks(inode
);
3367 handle
= ext4_journal_start(inode
, err
);
3371 if (inode
->i_size
& (sb
->s_blocksize
- 1))
3372 ext4_block_truncate_page(handle
, mapping
, inode
->i_size
);
3374 if (ext4_orphan_add(handle
, inode
))
3377 down_write(&EXT4_I(inode
)->i_data_sem
);
3378 ext4_ext_invalidate_cache(inode
);
3380 ext4_discard_preallocations(inode
);
3383 * TODO: optimization is possible here.
3384 * Probably we need not scan at all,
3385 * because page truncation is enough.
3388 /* we have to know where to truncate from in crash case */
3389 EXT4_I(inode
)->i_disksize
= inode
->i_size
;
3390 ext4_mark_inode_dirty(handle
, inode
);
3392 last_block
= (inode
->i_size
+ sb
->s_blocksize
- 1)
3393 >> EXT4_BLOCK_SIZE_BITS(sb
);
3394 err
= ext4_ext_remove_space(inode
, last_block
);
3396 /* In a multi-transaction truncate, we only make the final
3397 * transaction synchronous.
3400 ext4_handle_sync(handle
);
3403 up_write(&EXT4_I(inode
)->i_data_sem
);
3405 * If this was a simple ftruncate() and the file will remain alive,
3406 * then we need to clear up the orphan record which we created above.
3407 * However, if this was a real unlink then we were called by
3408 * ext4_delete_inode(), and we allow that function to clean up the
3409 * orphan info for us.
3412 ext4_orphan_del(handle
, inode
);
3414 inode
->i_mtime
= inode
->i_ctime
= ext4_current_time(inode
);
3415 ext4_mark_inode_dirty(handle
, inode
);
3416 ext4_journal_stop(handle
);
3419 static void ext4_falloc_update_inode(struct inode
*inode
,
3420 int mode
, loff_t new_size
, int update_ctime
)
3422 struct timespec now
;
3425 now
= current_fs_time(inode
->i_sb
);
3426 if (!timespec_equal(&inode
->i_ctime
, &now
))
3427 inode
->i_ctime
= now
;
3430 * Update only when preallocation was requested beyond
3433 if (!(mode
& FALLOC_FL_KEEP_SIZE
)) {
3434 if (new_size
> i_size_read(inode
))
3435 i_size_write(inode
, new_size
);
3436 if (new_size
> EXT4_I(inode
)->i_disksize
)
3437 ext4_update_i_disksize(inode
, new_size
);
3443 * preallocate space for a file. This implements ext4's fallocate inode
3444 * operation, which gets called from sys_fallocate system call.
3445 * For block-mapped files, posix_fallocate should fall back to the method
3446 * of writing zeroes to the required new blocks (the same behavior which is
3447 * expected for file systems which do not support fallocate() system call).
3449 long ext4_fallocate(struct inode
*inode
, int mode
, loff_t offset
, loff_t len
)
3454 unsigned int max_blocks
;
3458 struct buffer_head map_bh
;
3459 unsigned int credits
, blkbits
= inode
->i_blkbits
;
3462 * currently supporting (pre)allocate mode for extent-based
3465 if (!(EXT4_I(inode
)->i_flags
& EXT4_EXTENTS_FL
))
3468 /* preallocation to directories is currently not supported */
3469 if (S_ISDIR(inode
->i_mode
))
3472 block
= offset
>> blkbits
;
3474 * We can't just convert len to max_blocks because
3475 * If blocksize = 4096 offset = 3072 and len = 2048
3477 max_blocks
= (EXT4_BLOCK_ALIGN(len
+ offset
, blkbits
) >> blkbits
)
3480 * credits to insert 1 extent into extent tree
3482 credits
= ext4_chunk_trans_blocks(inode
, max_blocks
);
3483 mutex_lock(&inode
->i_mutex
);
3485 while (ret
>= 0 && ret
< max_blocks
) {
3486 block
= block
+ ret
;
3487 max_blocks
= max_blocks
- ret
;
3488 handle
= ext4_journal_start(inode
, credits
);
3489 if (IS_ERR(handle
)) {
3490 ret
= PTR_ERR(handle
);
3494 ret
= ext4_get_blocks(handle
, inode
, block
,
3495 max_blocks
, &map_bh
,
3496 EXT4_GET_BLOCKS_CREATE_UNINIT_EXT
);
3500 printk(KERN_ERR
"%s: ext4_ext_get_blocks "
3501 "returned error inode#%lu, block=%u, "
3502 "max_blocks=%u", __func__
,
3503 inode
->i_ino
, block
, max_blocks
);
3505 ext4_mark_inode_dirty(handle
, inode
);
3506 ret2
= ext4_journal_stop(handle
);
3509 if ((block
+ ret
) >= (EXT4_BLOCK_ALIGN(offset
+ len
,
3510 blkbits
) >> blkbits
))
3511 new_size
= offset
+ len
;
3513 new_size
= (block
+ ret
) << blkbits
;
3515 ext4_falloc_update_inode(inode
, mode
, new_size
,
3516 buffer_new(&map_bh
));
3517 ext4_mark_inode_dirty(handle
, inode
);
3518 ret2
= ext4_journal_stop(handle
);
3522 if (ret
== -ENOSPC
&&
3523 ext4_should_retry_alloc(inode
->i_sb
, &retries
)) {
3527 mutex_unlock(&inode
->i_mutex
);
3528 return ret
> 0 ? ret2
: ret
;
3532 * This function convert a range of blocks to written extents
3533 * The caller of this function will pass the start offset and the size.
3534 * all unwritten extents within this range will be converted to
3537 * This function is called from the direct IO end io call back
3538 * function, to convert the fallocated extents after IO is completed.
3539 * Returns 0 on success.
3541 int ext4_convert_unwritten_extents(struct inode
*inode
, loff_t offset
,
3546 unsigned int max_blocks
;
3549 struct buffer_head map_bh
;
3550 unsigned int credits
, blkbits
= inode
->i_blkbits
;
3552 block
= offset
>> blkbits
;
3554 * We can't just convert len to max_blocks because
3555 * If blocksize = 4096 offset = 3072 and len = 2048
3557 max_blocks
= (EXT4_BLOCK_ALIGN(len
+ offset
, blkbits
) >> blkbits
)
3560 * credits to insert 1 extent into extent tree
3562 credits
= ext4_chunk_trans_blocks(inode
, max_blocks
);
3563 while (ret
>= 0 && ret
< max_blocks
) {
3564 block
= block
+ ret
;
3565 max_blocks
= max_blocks
- ret
;
3566 handle
= ext4_journal_start(inode
, credits
);
3567 if (IS_ERR(handle
)) {
3568 ret
= PTR_ERR(handle
);
3572 ret
= ext4_get_blocks(handle
, inode
, block
,
3573 max_blocks
, &map_bh
,
3574 EXT4_GET_BLOCKS_DIO_CONVERT_EXT
);
3577 printk(KERN_ERR
"%s: ext4_ext_get_blocks "
3578 "returned error inode#%lu, block=%u, "
3579 "max_blocks=%u", __func__
,
3580 inode
->i_ino
, block
, max_blocks
);
3582 ext4_mark_inode_dirty(handle
, inode
);
3583 ret2
= ext4_journal_stop(handle
);
3584 if (ret
<= 0 || ret2
)
3587 return ret
> 0 ? ret2
: ret
;
3590 * Callback function called for each extent to gather FIEMAP information.
3592 static int ext4_ext_fiemap_cb(struct inode
*inode
, struct ext4_ext_path
*path
,
3593 struct ext4_ext_cache
*newex
, struct ext4_extent
*ex
,
3596 struct fiemap_extent_info
*fieinfo
= data
;
3597 unsigned char blksize_bits
= inode
->i_sb
->s_blocksize_bits
;
3604 logical
= (__u64
)newex
->ec_block
<< blksize_bits
;
3606 if (newex
->ec_type
== EXT4_EXT_CACHE_GAP
) {
3609 struct buffer_head
*bh
= NULL
;
3611 offset
= logical
>> PAGE_SHIFT
;
3612 page
= find_get_page(inode
->i_mapping
, offset
);
3613 if (!page
|| !page_has_buffers(page
))
3614 return EXT_CONTINUE
;
3616 bh
= page_buffers(page
);
3619 return EXT_CONTINUE
;
3621 if (buffer_delay(bh
)) {
3622 flags
|= FIEMAP_EXTENT_DELALLOC
;
3623 page_cache_release(page
);
3625 page_cache_release(page
);
3626 return EXT_CONTINUE
;
3630 physical
= (__u64
)newex
->ec_start
<< blksize_bits
;
3631 length
= (__u64
)newex
->ec_len
<< blksize_bits
;
3633 if (ex
&& ext4_ext_is_uninitialized(ex
))
3634 flags
|= FIEMAP_EXTENT_UNWRITTEN
;
3637 * If this extent reaches EXT_MAX_BLOCK, it must be last.
3639 * Or if ext4_ext_next_allocated_block is EXT_MAX_BLOCK,
3640 * this also indicates no more allocated blocks.
3642 * XXX this might miss a single-block extent at EXT_MAX_BLOCK
3644 if (ext4_ext_next_allocated_block(path
) == EXT_MAX_BLOCK
||
3645 newex
->ec_block
+ newex
->ec_len
- 1 == EXT_MAX_BLOCK
) {
3646 loff_t size
= i_size_read(inode
);
3647 loff_t bs
= EXT4_BLOCK_SIZE(inode
->i_sb
);
3649 flags
|= FIEMAP_EXTENT_LAST
;
3650 if ((flags
& FIEMAP_EXTENT_DELALLOC
) &&
3651 logical
+length
> size
)
3652 length
= (size
- logical
+ bs
- 1) & ~(bs
-1);
3655 error
= fiemap_fill_next_extent(fieinfo
, logical
, physical
,
3662 return EXT_CONTINUE
;
3665 /* fiemap flags we can handle specified here */
3666 #define EXT4_FIEMAP_FLAGS (FIEMAP_FLAG_SYNC|FIEMAP_FLAG_XATTR)
3668 static int ext4_xattr_fiemap(struct inode
*inode
,
3669 struct fiemap_extent_info
*fieinfo
)
3673 __u32 flags
= FIEMAP_EXTENT_LAST
;
3674 int blockbits
= inode
->i_sb
->s_blocksize_bits
;
3678 if (EXT4_I(inode
)->i_state
& EXT4_STATE_XATTR
) {
3679 struct ext4_iloc iloc
;
3680 int offset
; /* offset of xattr in inode */
3682 error
= ext4_get_inode_loc(inode
, &iloc
);
3685 physical
= iloc
.bh
->b_blocknr
<< blockbits
;
3686 offset
= EXT4_GOOD_OLD_INODE_SIZE
+
3687 EXT4_I(inode
)->i_extra_isize
;
3689 length
= EXT4_SB(inode
->i_sb
)->s_inode_size
- offset
;
3690 flags
|= FIEMAP_EXTENT_DATA_INLINE
;
3691 } else { /* external block */
3692 physical
= EXT4_I(inode
)->i_file_acl
<< blockbits
;
3693 length
= inode
->i_sb
->s_blocksize
;
3697 error
= fiemap_fill_next_extent(fieinfo
, 0, physical
,
3699 return (error
< 0 ? error
: 0);
3702 int ext4_fiemap(struct inode
*inode
, struct fiemap_extent_info
*fieinfo
,
3703 __u64 start
, __u64 len
)
3705 ext4_lblk_t start_blk
;
3706 ext4_lblk_t len_blks
;
3709 /* fallback to generic here if not in extents fmt */
3710 if (!(EXT4_I(inode
)->i_flags
& EXT4_EXTENTS_FL
))
3711 return generic_block_fiemap(inode
, fieinfo
, start
, len
,
3714 if (fiemap_check_flags(fieinfo
, EXT4_FIEMAP_FLAGS
))
3717 if (fieinfo
->fi_flags
& FIEMAP_FLAG_XATTR
) {
3718 error
= ext4_xattr_fiemap(inode
, fieinfo
);
3720 start_blk
= start
>> inode
->i_sb
->s_blocksize_bits
;
3721 len_blks
= len
>> inode
->i_sb
->s_blocksize_bits
;
3724 * Walk the extent tree gathering extent information.
3725 * ext4_ext_fiemap_cb will push extents back to user.
3727 down_read(&EXT4_I(inode
)->i_data_sem
);
3728 error
= ext4_ext_walk_space(inode
, start_blk
, len_blks
,
3729 ext4_ext_fiemap_cb
, fieinfo
);
3730 up_read(&EXT4_I(inode
)->i_data_sem
);