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 int ext4_ext_space_block(struct inode
*inode
)
236 size
= (inode
->i_sb
->s_blocksize
- sizeof(struct ext4_extent_header
))
237 / sizeof(struct ext4_extent
);
238 #ifdef AGGRESSIVE_TEST
245 static int ext4_ext_space_block_idx(struct inode
*inode
)
249 size
= (inode
->i_sb
->s_blocksize
- sizeof(struct ext4_extent_header
))
250 / sizeof(struct ext4_extent_idx
);
251 #ifdef AGGRESSIVE_TEST
258 static int ext4_ext_space_root(struct inode
*inode
)
262 size
= sizeof(EXT4_I(inode
)->i_data
);
263 size
-= sizeof(struct ext4_extent_header
);
264 size
/= sizeof(struct ext4_extent
);
265 #ifdef AGGRESSIVE_TEST
272 static int ext4_ext_space_root_idx(struct inode
*inode
)
276 size
= sizeof(EXT4_I(inode
)->i_data
);
277 size
-= sizeof(struct ext4_extent_header
);
278 size
/= sizeof(struct ext4_extent_idx
);
279 #ifdef AGGRESSIVE_TEST
287 * Calculate the number of metadata blocks needed
288 * to allocate @blocks
289 * Worse case is one block per extent
291 int ext4_ext_calc_metadata_amount(struct inode
*inode
, int blocks
)
293 int lcap
, icap
, rcap
, leafs
, idxs
, num
;
294 int newextents
= blocks
;
296 rcap
= ext4_ext_space_root_idx(inode
);
297 lcap
= ext4_ext_space_block(inode
);
298 icap
= ext4_ext_space_block_idx(inode
);
300 /* number of new leaf blocks needed */
301 num
= leafs
= (newextents
+ lcap
- 1) / lcap
;
304 * Worse case, we need separate index block(s)
305 * to link all new leaf blocks
307 idxs
= (leafs
+ icap
- 1) / icap
;
310 idxs
= (idxs
+ icap
- 1) / icap
;
311 } while (idxs
> rcap
);
317 ext4_ext_max_entries(struct inode
*inode
, int depth
)
321 if (depth
== ext_depth(inode
)) {
323 max
= ext4_ext_space_root(inode
);
325 max
= ext4_ext_space_root_idx(inode
);
328 max
= ext4_ext_space_block(inode
);
330 max
= ext4_ext_space_block_idx(inode
);
336 static int ext4_valid_extent(struct inode
*inode
, struct ext4_extent
*ext
)
338 ext4_fsblk_t block
= ext_pblock(ext
);
339 int len
= ext4_ext_get_actual_len(ext
);
341 return ext4_data_block_valid(EXT4_SB(inode
->i_sb
), block
, len
);
344 static int ext4_valid_extent_idx(struct inode
*inode
,
345 struct ext4_extent_idx
*ext_idx
)
347 ext4_fsblk_t block
= idx_pblock(ext_idx
);
349 return ext4_data_block_valid(EXT4_SB(inode
->i_sb
), block
, 1);
352 static int ext4_valid_extent_entries(struct inode
*inode
,
353 struct ext4_extent_header
*eh
,
356 struct ext4_extent
*ext
;
357 struct ext4_extent_idx
*ext_idx
;
358 unsigned short entries
;
359 if (eh
->eh_entries
== 0)
362 entries
= le16_to_cpu(eh
->eh_entries
);
366 ext
= EXT_FIRST_EXTENT(eh
);
368 if (!ext4_valid_extent(inode
, ext
))
374 ext_idx
= EXT_FIRST_INDEX(eh
);
376 if (!ext4_valid_extent_idx(inode
, ext_idx
))
385 static int __ext4_ext_check(const char *function
, struct inode
*inode
,
386 struct ext4_extent_header
*eh
,
389 const char *error_msg
;
392 if (unlikely(eh
->eh_magic
!= EXT4_EXT_MAGIC
)) {
393 error_msg
= "invalid magic";
396 if (unlikely(le16_to_cpu(eh
->eh_depth
) != depth
)) {
397 error_msg
= "unexpected eh_depth";
400 if (unlikely(eh
->eh_max
== 0)) {
401 error_msg
= "invalid eh_max";
404 max
= ext4_ext_max_entries(inode
, depth
);
405 if (unlikely(le16_to_cpu(eh
->eh_max
) > max
)) {
406 error_msg
= "too large eh_max";
409 if (unlikely(le16_to_cpu(eh
->eh_entries
) > le16_to_cpu(eh
->eh_max
))) {
410 error_msg
= "invalid eh_entries";
413 if (!ext4_valid_extent_entries(inode
, eh
, depth
)) {
414 error_msg
= "invalid extent entries";
420 ext4_error(inode
->i_sb
, function
,
421 "bad header/extent in inode #%lu: %s - magic %x, "
422 "entries %u, max %u(%u), depth %u(%u)",
423 inode
->i_ino
, error_msg
, le16_to_cpu(eh
->eh_magic
),
424 le16_to_cpu(eh
->eh_entries
), le16_to_cpu(eh
->eh_max
),
425 max
, le16_to_cpu(eh
->eh_depth
), depth
);
430 #define ext4_ext_check(inode, eh, depth) \
431 __ext4_ext_check(__func__, inode, eh, depth)
433 int ext4_ext_check_inode(struct inode
*inode
)
435 return ext4_ext_check(inode
, ext_inode_hdr(inode
), ext_depth(inode
));
439 static void ext4_ext_show_path(struct inode
*inode
, struct ext4_ext_path
*path
)
441 int k
, l
= path
->p_depth
;
444 for (k
= 0; k
<= l
; k
++, path
++) {
446 ext_debug(" %d->%llu", le32_to_cpu(path
->p_idx
->ei_block
),
447 idx_pblock(path
->p_idx
));
448 } else if (path
->p_ext
) {
449 ext_debug(" %d:[%d]%d:%llu ",
450 le32_to_cpu(path
->p_ext
->ee_block
),
451 ext4_ext_is_uninitialized(path
->p_ext
),
452 ext4_ext_get_actual_len(path
->p_ext
),
453 ext_pblock(path
->p_ext
));
460 static void ext4_ext_show_leaf(struct inode
*inode
, struct ext4_ext_path
*path
)
462 int depth
= ext_depth(inode
);
463 struct ext4_extent_header
*eh
;
464 struct ext4_extent
*ex
;
470 eh
= path
[depth
].p_hdr
;
471 ex
= EXT_FIRST_EXTENT(eh
);
473 ext_debug("Displaying leaf extents for inode %lu\n", inode
->i_ino
);
475 for (i
= 0; i
< le16_to_cpu(eh
->eh_entries
); i
++, ex
++) {
476 ext_debug("%d:[%d]%d:%llu ", le32_to_cpu(ex
->ee_block
),
477 ext4_ext_is_uninitialized(ex
),
478 ext4_ext_get_actual_len(ex
), ext_pblock(ex
));
483 #define ext4_ext_show_path(inode, path)
484 #define ext4_ext_show_leaf(inode, path)
487 void ext4_ext_drop_refs(struct ext4_ext_path
*path
)
489 int depth
= path
->p_depth
;
492 for (i
= 0; i
<= depth
; i
++, path
++)
500 * ext4_ext_binsearch_idx:
501 * binary search for the closest index of the given block
502 * the header must be checked before calling this
505 ext4_ext_binsearch_idx(struct inode
*inode
,
506 struct ext4_ext_path
*path
, ext4_lblk_t block
)
508 struct ext4_extent_header
*eh
= path
->p_hdr
;
509 struct ext4_extent_idx
*r
, *l
, *m
;
512 ext_debug("binsearch for %u(idx): ", block
);
514 l
= EXT_FIRST_INDEX(eh
) + 1;
515 r
= EXT_LAST_INDEX(eh
);
518 if (block
< le32_to_cpu(m
->ei_block
))
522 ext_debug("%p(%u):%p(%u):%p(%u) ", l
, le32_to_cpu(l
->ei_block
),
523 m
, le32_to_cpu(m
->ei_block
),
524 r
, le32_to_cpu(r
->ei_block
));
528 ext_debug(" -> %d->%lld ", le32_to_cpu(path
->p_idx
->ei_block
),
529 idx_pblock(path
->p_idx
));
531 #ifdef CHECK_BINSEARCH
533 struct ext4_extent_idx
*chix
, *ix
;
536 chix
= ix
= EXT_FIRST_INDEX(eh
);
537 for (k
= 0; k
< le16_to_cpu(eh
->eh_entries
); k
++, ix
++) {
539 le32_to_cpu(ix
->ei_block
) <= le32_to_cpu(ix
[-1].ei_block
)) {
540 printk(KERN_DEBUG
"k=%d, ix=0x%p, "
542 ix
, EXT_FIRST_INDEX(eh
));
543 printk(KERN_DEBUG
"%u <= %u\n",
544 le32_to_cpu(ix
->ei_block
),
545 le32_to_cpu(ix
[-1].ei_block
));
547 BUG_ON(k
&& le32_to_cpu(ix
->ei_block
)
548 <= le32_to_cpu(ix
[-1].ei_block
));
549 if (block
< le32_to_cpu(ix
->ei_block
))
553 BUG_ON(chix
!= path
->p_idx
);
560 * ext4_ext_binsearch:
561 * binary search for closest extent of the given block
562 * the header must be checked before calling this
565 ext4_ext_binsearch(struct inode
*inode
,
566 struct ext4_ext_path
*path
, ext4_lblk_t block
)
568 struct ext4_extent_header
*eh
= path
->p_hdr
;
569 struct ext4_extent
*r
, *l
, *m
;
571 if (eh
->eh_entries
== 0) {
573 * this leaf is empty:
574 * we get such a leaf in split/add case
579 ext_debug("binsearch for %u: ", block
);
581 l
= EXT_FIRST_EXTENT(eh
) + 1;
582 r
= EXT_LAST_EXTENT(eh
);
586 if (block
< le32_to_cpu(m
->ee_block
))
590 ext_debug("%p(%u):%p(%u):%p(%u) ", l
, le32_to_cpu(l
->ee_block
),
591 m
, le32_to_cpu(m
->ee_block
),
592 r
, le32_to_cpu(r
->ee_block
));
596 ext_debug(" -> %d:%llu:[%d]%d ",
597 le32_to_cpu(path
->p_ext
->ee_block
),
598 ext_pblock(path
->p_ext
),
599 ext4_ext_is_uninitialized(path
->p_ext
),
600 ext4_ext_get_actual_len(path
->p_ext
));
602 #ifdef CHECK_BINSEARCH
604 struct ext4_extent
*chex
, *ex
;
607 chex
= ex
= EXT_FIRST_EXTENT(eh
);
608 for (k
= 0; k
< le16_to_cpu(eh
->eh_entries
); k
++, ex
++) {
609 BUG_ON(k
&& le32_to_cpu(ex
->ee_block
)
610 <= le32_to_cpu(ex
[-1].ee_block
));
611 if (block
< le32_to_cpu(ex
->ee_block
))
615 BUG_ON(chex
!= path
->p_ext
);
621 int ext4_ext_tree_init(handle_t
*handle
, struct inode
*inode
)
623 struct ext4_extent_header
*eh
;
625 eh
= ext_inode_hdr(inode
);
628 eh
->eh_magic
= EXT4_EXT_MAGIC
;
629 eh
->eh_max
= cpu_to_le16(ext4_ext_space_root(inode
));
630 ext4_mark_inode_dirty(handle
, inode
);
631 ext4_ext_invalidate_cache(inode
);
635 struct ext4_ext_path
*
636 ext4_ext_find_extent(struct inode
*inode
, ext4_lblk_t block
,
637 struct ext4_ext_path
*path
)
639 struct ext4_extent_header
*eh
;
640 struct buffer_head
*bh
;
641 short int depth
, i
, ppos
= 0, alloc
= 0;
643 eh
= ext_inode_hdr(inode
);
644 depth
= ext_depth(inode
);
646 /* account possible depth increase */
648 path
= kzalloc(sizeof(struct ext4_ext_path
) * (depth
+ 2),
651 return ERR_PTR(-ENOMEM
);
658 /* walk through the tree */
660 int need_to_validate
= 0;
662 ext_debug("depth %d: num %d, max %d\n",
663 ppos
, le16_to_cpu(eh
->eh_entries
), le16_to_cpu(eh
->eh_max
));
665 ext4_ext_binsearch_idx(inode
, path
+ ppos
, block
);
666 path
[ppos
].p_block
= idx_pblock(path
[ppos
].p_idx
);
667 path
[ppos
].p_depth
= i
;
668 path
[ppos
].p_ext
= NULL
;
670 bh
= sb_getblk(inode
->i_sb
, path
[ppos
].p_block
);
673 if (!bh_uptodate_or_lock(bh
)) {
674 if (bh_submit_read(bh
) < 0) {
678 /* validate the extent entries */
679 need_to_validate
= 1;
681 eh
= ext_block_hdr(bh
);
683 BUG_ON(ppos
> depth
);
684 path
[ppos
].p_bh
= bh
;
685 path
[ppos
].p_hdr
= eh
;
688 if (need_to_validate
&& ext4_ext_check(inode
, eh
, i
))
692 path
[ppos
].p_depth
= i
;
693 path
[ppos
].p_ext
= NULL
;
694 path
[ppos
].p_idx
= NULL
;
697 ext4_ext_binsearch(inode
, path
+ ppos
, block
);
698 /* if not an empty leaf */
699 if (path
[ppos
].p_ext
)
700 path
[ppos
].p_block
= ext_pblock(path
[ppos
].p_ext
);
702 ext4_ext_show_path(inode
, path
);
707 ext4_ext_drop_refs(path
);
710 return ERR_PTR(-EIO
);
714 * ext4_ext_insert_index:
715 * insert new index [@logical;@ptr] into the block at @curp;
716 * check where to insert: before @curp or after @curp
718 static int ext4_ext_insert_index(handle_t
*handle
, struct inode
*inode
,
719 struct ext4_ext_path
*curp
,
720 int logical
, ext4_fsblk_t ptr
)
722 struct ext4_extent_idx
*ix
;
725 err
= ext4_ext_get_access(handle
, inode
, curp
);
729 BUG_ON(logical
== le32_to_cpu(curp
->p_idx
->ei_block
));
730 len
= EXT_MAX_INDEX(curp
->p_hdr
) - curp
->p_idx
;
731 if (logical
> le32_to_cpu(curp
->p_idx
->ei_block
)) {
733 if (curp
->p_idx
!= EXT_LAST_INDEX(curp
->p_hdr
)) {
734 len
= (len
- 1) * sizeof(struct ext4_extent_idx
);
735 len
= len
< 0 ? 0 : len
;
736 ext_debug("insert new index %d after: %llu. "
737 "move %d from 0x%p to 0x%p\n",
739 (curp
->p_idx
+ 1), (curp
->p_idx
+ 2));
740 memmove(curp
->p_idx
+ 2, curp
->p_idx
+ 1, len
);
742 ix
= curp
->p_idx
+ 1;
745 len
= len
* sizeof(struct ext4_extent_idx
);
746 len
= len
< 0 ? 0 : len
;
747 ext_debug("insert new index %d before: %llu. "
748 "move %d from 0x%p to 0x%p\n",
750 curp
->p_idx
, (curp
->p_idx
+ 1));
751 memmove(curp
->p_idx
+ 1, curp
->p_idx
, len
);
755 ix
->ei_block
= cpu_to_le32(logical
);
756 ext4_idx_store_pblock(ix
, ptr
);
757 le16_add_cpu(&curp
->p_hdr
->eh_entries
, 1);
759 BUG_ON(le16_to_cpu(curp
->p_hdr
->eh_entries
)
760 > le16_to_cpu(curp
->p_hdr
->eh_max
));
761 BUG_ON(ix
> EXT_LAST_INDEX(curp
->p_hdr
));
763 err
= ext4_ext_dirty(handle
, inode
, curp
);
764 ext4_std_error(inode
->i_sb
, err
);
771 * inserts new subtree into the path, using free index entry
773 * - allocates all needed blocks (new leaf and all intermediate index blocks)
774 * - makes decision where to split
775 * - moves remaining extents and index entries (right to the split point)
776 * into the newly allocated blocks
777 * - initializes subtree
779 static int ext4_ext_split(handle_t
*handle
, struct inode
*inode
,
780 struct ext4_ext_path
*path
,
781 struct ext4_extent
*newext
, int at
)
783 struct buffer_head
*bh
= NULL
;
784 int depth
= ext_depth(inode
);
785 struct ext4_extent_header
*neh
;
786 struct ext4_extent_idx
*fidx
;
787 struct ext4_extent
*ex
;
789 ext4_fsblk_t newblock
, oldblock
;
791 ext4_fsblk_t
*ablocks
= NULL
; /* array of allocated blocks */
794 /* make decision: where to split? */
795 /* FIXME: now decision is simplest: at current extent */
797 /* if current leaf will be split, then we should use
798 * border from split point */
799 BUG_ON(path
[depth
].p_ext
> EXT_MAX_EXTENT(path
[depth
].p_hdr
));
800 if (path
[depth
].p_ext
!= EXT_MAX_EXTENT(path
[depth
].p_hdr
)) {
801 border
= path
[depth
].p_ext
[1].ee_block
;
802 ext_debug("leaf will be split."
803 " next leaf starts at %d\n",
804 le32_to_cpu(border
));
806 border
= newext
->ee_block
;
807 ext_debug("leaf will be added."
808 " next leaf starts at %d\n",
809 le32_to_cpu(border
));
813 * If error occurs, then we break processing
814 * and mark filesystem read-only. index won't
815 * be inserted and tree will be in consistent
816 * state. Next mount will repair buffers too.
820 * Get array to track all allocated blocks.
821 * We need this to handle errors and free blocks
824 ablocks
= kzalloc(sizeof(ext4_fsblk_t
) * depth
, GFP_NOFS
);
828 /* allocate all needed blocks */
829 ext_debug("allocate %d blocks for indexes/leaf\n", depth
- at
);
830 for (a
= 0; a
< depth
- at
; a
++) {
831 newblock
= ext4_ext_new_meta_block(handle
, inode
, path
,
835 ablocks
[a
] = newblock
;
838 /* initialize new leaf */
839 newblock
= ablocks
[--a
];
840 BUG_ON(newblock
== 0);
841 bh
= sb_getblk(inode
->i_sb
, newblock
);
848 err
= ext4_journal_get_create_access(handle
, bh
);
852 neh
= ext_block_hdr(bh
);
854 neh
->eh_max
= cpu_to_le16(ext4_ext_space_block(inode
));
855 neh
->eh_magic
= EXT4_EXT_MAGIC
;
857 ex
= EXT_FIRST_EXTENT(neh
);
859 /* move remainder of path[depth] to the new leaf */
860 BUG_ON(path
[depth
].p_hdr
->eh_entries
!= path
[depth
].p_hdr
->eh_max
);
861 /* start copy from next extent */
862 /* TODO: we could do it by single memmove */
865 while (path
[depth
].p_ext
<=
866 EXT_MAX_EXTENT(path
[depth
].p_hdr
)) {
867 ext_debug("move %d:%llu:[%d]%d in new leaf %llu\n",
868 le32_to_cpu(path
[depth
].p_ext
->ee_block
),
869 ext_pblock(path
[depth
].p_ext
),
870 ext4_ext_is_uninitialized(path
[depth
].p_ext
),
871 ext4_ext_get_actual_len(path
[depth
].p_ext
),
873 /*memmove(ex++, path[depth].p_ext++,
874 sizeof(struct ext4_extent));
880 memmove(ex
, path
[depth
].p_ext
-m
, sizeof(struct ext4_extent
)*m
);
881 le16_add_cpu(&neh
->eh_entries
, m
);
884 set_buffer_uptodate(bh
);
887 err
= ext4_handle_dirty_metadata(handle
, inode
, bh
);
893 /* correct old leaf */
895 err
= ext4_ext_get_access(handle
, inode
, path
+ depth
);
898 le16_add_cpu(&path
[depth
].p_hdr
->eh_entries
, -m
);
899 err
= ext4_ext_dirty(handle
, inode
, path
+ depth
);
905 /* create intermediate indexes */
909 ext_debug("create %d intermediate indices\n", k
);
910 /* insert new index into current index block */
911 /* current depth stored in i var */
915 newblock
= ablocks
[--a
];
916 bh
= sb_getblk(inode
->i_sb
, newblock
);
923 err
= ext4_journal_get_create_access(handle
, bh
);
927 neh
= ext_block_hdr(bh
);
928 neh
->eh_entries
= cpu_to_le16(1);
929 neh
->eh_magic
= EXT4_EXT_MAGIC
;
930 neh
->eh_max
= cpu_to_le16(ext4_ext_space_block_idx(inode
));
931 neh
->eh_depth
= cpu_to_le16(depth
- i
);
932 fidx
= EXT_FIRST_INDEX(neh
);
933 fidx
->ei_block
= border
;
934 ext4_idx_store_pblock(fidx
, oldblock
);
936 ext_debug("int.index at %d (block %llu): %u -> %llu\n",
937 i
, newblock
, le32_to_cpu(border
), oldblock
);
942 ext_debug("cur 0x%p, last 0x%p\n", path
[i
].p_idx
,
943 EXT_MAX_INDEX(path
[i
].p_hdr
));
944 BUG_ON(EXT_MAX_INDEX(path
[i
].p_hdr
) !=
945 EXT_LAST_INDEX(path
[i
].p_hdr
));
946 while (path
[i
].p_idx
<= EXT_MAX_INDEX(path
[i
].p_hdr
)) {
947 ext_debug("%d: move %d:%llu in new index %llu\n", i
,
948 le32_to_cpu(path
[i
].p_idx
->ei_block
),
949 idx_pblock(path
[i
].p_idx
),
951 /*memmove(++fidx, path[i].p_idx++,
952 sizeof(struct ext4_extent_idx));
954 BUG_ON(neh->eh_entries > neh->eh_max);*/
959 memmove(++fidx
, path
[i
].p_idx
- m
,
960 sizeof(struct ext4_extent_idx
) * m
);
961 le16_add_cpu(&neh
->eh_entries
, m
);
963 set_buffer_uptodate(bh
);
966 err
= ext4_handle_dirty_metadata(handle
, inode
, bh
);
972 /* correct old index */
974 err
= ext4_ext_get_access(handle
, inode
, path
+ i
);
977 le16_add_cpu(&path
[i
].p_hdr
->eh_entries
, -m
);
978 err
= ext4_ext_dirty(handle
, inode
, path
+ i
);
986 /* insert new index */
987 err
= ext4_ext_insert_index(handle
, inode
, path
+ at
,
988 le32_to_cpu(border
), newblock
);
992 if (buffer_locked(bh
))
998 /* free all allocated blocks in error case */
999 for (i
= 0; i
< depth
; i
++) {
1002 ext4_free_blocks(handle
, inode
, ablocks
[i
], 1, 1);
1011 * ext4_ext_grow_indepth:
1012 * implements tree growing procedure:
1013 * - allocates new block
1014 * - moves top-level data (index block or leaf) into the new block
1015 * - initializes new top-level, creating index that points to the
1016 * just created block
1018 static int ext4_ext_grow_indepth(handle_t
*handle
, struct inode
*inode
,
1019 struct ext4_ext_path
*path
,
1020 struct ext4_extent
*newext
)
1022 struct ext4_ext_path
*curp
= path
;
1023 struct ext4_extent_header
*neh
;
1024 struct ext4_extent_idx
*fidx
;
1025 struct buffer_head
*bh
;
1026 ext4_fsblk_t newblock
;
1029 newblock
= ext4_ext_new_meta_block(handle
, inode
, path
, newext
, &err
);
1033 bh
= sb_getblk(inode
->i_sb
, newblock
);
1036 ext4_std_error(inode
->i_sb
, err
);
1041 err
= ext4_journal_get_create_access(handle
, bh
);
1047 /* move top-level index/leaf into new block */
1048 memmove(bh
->b_data
, curp
->p_hdr
, sizeof(EXT4_I(inode
)->i_data
));
1050 /* set size of new block */
1051 neh
= ext_block_hdr(bh
);
1052 /* old root could have indexes or leaves
1053 * so calculate e_max right way */
1054 if (ext_depth(inode
))
1055 neh
->eh_max
= cpu_to_le16(ext4_ext_space_block_idx(inode
));
1057 neh
->eh_max
= cpu_to_le16(ext4_ext_space_block(inode
));
1058 neh
->eh_magic
= EXT4_EXT_MAGIC
;
1059 set_buffer_uptodate(bh
);
1062 err
= ext4_handle_dirty_metadata(handle
, inode
, bh
);
1066 /* create index in new top-level index: num,max,pointer */
1067 err
= ext4_ext_get_access(handle
, inode
, curp
);
1071 curp
->p_hdr
->eh_magic
= EXT4_EXT_MAGIC
;
1072 curp
->p_hdr
->eh_max
= cpu_to_le16(ext4_ext_space_root_idx(inode
));
1073 curp
->p_hdr
->eh_entries
= cpu_to_le16(1);
1074 curp
->p_idx
= EXT_FIRST_INDEX(curp
->p_hdr
);
1076 if (path
[0].p_hdr
->eh_depth
)
1077 curp
->p_idx
->ei_block
=
1078 EXT_FIRST_INDEX(path
[0].p_hdr
)->ei_block
;
1080 curp
->p_idx
->ei_block
=
1081 EXT_FIRST_EXTENT(path
[0].p_hdr
)->ee_block
;
1082 ext4_idx_store_pblock(curp
->p_idx
, newblock
);
1084 neh
= ext_inode_hdr(inode
);
1085 fidx
= EXT_FIRST_INDEX(neh
);
1086 ext_debug("new root: num %d(%d), lblock %d, ptr %llu\n",
1087 le16_to_cpu(neh
->eh_entries
), le16_to_cpu(neh
->eh_max
),
1088 le32_to_cpu(fidx
->ei_block
), idx_pblock(fidx
));
1090 neh
->eh_depth
= cpu_to_le16(path
->p_depth
+ 1);
1091 err
= ext4_ext_dirty(handle
, inode
, curp
);
1099 * ext4_ext_create_new_leaf:
1100 * finds empty index and adds new leaf.
1101 * if no free index is found, then it requests in-depth growing.
1103 static int ext4_ext_create_new_leaf(handle_t
*handle
, struct inode
*inode
,
1104 struct ext4_ext_path
*path
,
1105 struct ext4_extent
*newext
)
1107 struct ext4_ext_path
*curp
;
1108 int depth
, i
, err
= 0;
1111 i
= depth
= ext_depth(inode
);
1113 /* walk up to the tree and look for free index entry */
1114 curp
= path
+ depth
;
1115 while (i
> 0 && !EXT_HAS_FREE_INDEX(curp
)) {
1120 /* we use already allocated block for index block,
1121 * so subsequent data blocks should be contiguous */
1122 if (EXT_HAS_FREE_INDEX(curp
)) {
1123 /* if we found index with free entry, then use that
1124 * entry: create all needed subtree and add new leaf */
1125 err
= ext4_ext_split(handle
, inode
, path
, newext
, i
);
1130 ext4_ext_drop_refs(path
);
1131 path
= ext4_ext_find_extent(inode
,
1132 (ext4_lblk_t
)le32_to_cpu(newext
->ee_block
),
1135 err
= PTR_ERR(path
);
1137 /* tree is full, time to grow in depth */
1138 err
= ext4_ext_grow_indepth(handle
, inode
, path
, newext
);
1143 ext4_ext_drop_refs(path
);
1144 path
= ext4_ext_find_extent(inode
,
1145 (ext4_lblk_t
)le32_to_cpu(newext
->ee_block
),
1148 err
= PTR_ERR(path
);
1153 * only first (depth 0 -> 1) produces free space;
1154 * in all other cases we have to split the grown tree
1156 depth
= ext_depth(inode
);
1157 if (path
[depth
].p_hdr
->eh_entries
== path
[depth
].p_hdr
->eh_max
) {
1158 /* now we need to split */
1168 * search the closest allocated block to the left for *logical
1169 * and returns it at @logical + it's physical address at @phys
1170 * if *logical is the smallest allocated block, the function
1171 * returns 0 at @phys
1172 * return value contains 0 (success) or error code
1175 ext4_ext_search_left(struct inode
*inode
, struct ext4_ext_path
*path
,
1176 ext4_lblk_t
*logical
, ext4_fsblk_t
*phys
)
1178 struct ext4_extent_idx
*ix
;
1179 struct ext4_extent
*ex
;
1182 BUG_ON(path
== NULL
);
1183 depth
= path
->p_depth
;
1186 if (depth
== 0 && path
->p_ext
== NULL
)
1189 /* usually extent in the path covers blocks smaller
1190 * then *logical, but it can be that extent is the
1191 * first one in the file */
1193 ex
= path
[depth
].p_ext
;
1194 ee_len
= ext4_ext_get_actual_len(ex
);
1195 if (*logical
< le32_to_cpu(ex
->ee_block
)) {
1196 BUG_ON(EXT_FIRST_EXTENT(path
[depth
].p_hdr
) != ex
);
1197 while (--depth
>= 0) {
1198 ix
= path
[depth
].p_idx
;
1199 BUG_ON(ix
!= EXT_FIRST_INDEX(path
[depth
].p_hdr
));
1204 BUG_ON(*logical
< (le32_to_cpu(ex
->ee_block
) + ee_len
));
1206 *logical
= le32_to_cpu(ex
->ee_block
) + ee_len
- 1;
1207 *phys
= ext_pblock(ex
) + ee_len
- 1;
1212 * search the closest allocated block to the right for *logical
1213 * and returns it at @logical + it's physical address at @phys
1214 * if *logical is the smallest allocated block, the function
1215 * returns 0 at @phys
1216 * return value contains 0 (success) or error code
1219 ext4_ext_search_right(struct inode
*inode
, struct ext4_ext_path
*path
,
1220 ext4_lblk_t
*logical
, ext4_fsblk_t
*phys
)
1222 struct buffer_head
*bh
= NULL
;
1223 struct ext4_extent_header
*eh
;
1224 struct ext4_extent_idx
*ix
;
1225 struct ext4_extent
*ex
;
1227 int depth
; /* Note, NOT eh_depth; depth from top of tree */
1230 BUG_ON(path
== NULL
);
1231 depth
= path
->p_depth
;
1234 if (depth
== 0 && path
->p_ext
== NULL
)
1237 /* usually extent in the path covers blocks smaller
1238 * then *logical, but it can be that extent is the
1239 * first one in the file */
1241 ex
= path
[depth
].p_ext
;
1242 ee_len
= ext4_ext_get_actual_len(ex
);
1243 if (*logical
< le32_to_cpu(ex
->ee_block
)) {
1244 BUG_ON(EXT_FIRST_EXTENT(path
[depth
].p_hdr
) != ex
);
1245 while (--depth
>= 0) {
1246 ix
= path
[depth
].p_idx
;
1247 BUG_ON(ix
!= EXT_FIRST_INDEX(path
[depth
].p_hdr
));
1249 *logical
= le32_to_cpu(ex
->ee_block
);
1250 *phys
= ext_pblock(ex
);
1254 BUG_ON(*logical
< (le32_to_cpu(ex
->ee_block
) + ee_len
));
1256 if (ex
!= EXT_LAST_EXTENT(path
[depth
].p_hdr
)) {
1257 /* next allocated block in this leaf */
1259 *logical
= le32_to_cpu(ex
->ee_block
);
1260 *phys
= ext_pblock(ex
);
1264 /* go up and search for index to the right */
1265 while (--depth
>= 0) {
1266 ix
= path
[depth
].p_idx
;
1267 if (ix
!= EXT_LAST_INDEX(path
[depth
].p_hdr
))
1271 /* we've gone up to the root and found no index to the right */
1275 /* we've found index to the right, let's
1276 * follow it and find the closest allocated
1277 * block to the right */
1279 block
= idx_pblock(ix
);
1280 while (++depth
< path
->p_depth
) {
1281 bh
= sb_bread(inode
->i_sb
, block
);
1284 eh
= ext_block_hdr(bh
);
1285 /* subtract from p_depth to get proper eh_depth */
1286 if (ext4_ext_check(inode
, eh
, path
->p_depth
- depth
)) {
1290 ix
= EXT_FIRST_INDEX(eh
);
1291 block
= idx_pblock(ix
);
1295 bh
= sb_bread(inode
->i_sb
, block
);
1298 eh
= ext_block_hdr(bh
);
1299 if (ext4_ext_check(inode
, eh
, path
->p_depth
- depth
)) {
1303 ex
= EXT_FIRST_EXTENT(eh
);
1304 *logical
= le32_to_cpu(ex
->ee_block
);
1305 *phys
= ext_pblock(ex
);
1311 * ext4_ext_next_allocated_block:
1312 * returns allocated block in subsequent extent or EXT_MAX_BLOCK.
1313 * NOTE: it considers block number from index entry as
1314 * allocated block. Thus, index entries have to be consistent
1318 ext4_ext_next_allocated_block(struct ext4_ext_path
*path
)
1322 BUG_ON(path
== NULL
);
1323 depth
= path
->p_depth
;
1325 if (depth
== 0 && path
->p_ext
== NULL
)
1326 return EXT_MAX_BLOCK
;
1328 while (depth
>= 0) {
1329 if (depth
== path
->p_depth
) {
1331 if (path
[depth
].p_ext
!=
1332 EXT_LAST_EXTENT(path
[depth
].p_hdr
))
1333 return le32_to_cpu(path
[depth
].p_ext
[1].ee_block
);
1336 if (path
[depth
].p_idx
!=
1337 EXT_LAST_INDEX(path
[depth
].p_hdr
))
1338 return le32_to_cpu(path
[depth
].p_idx
[1].ei_block
);
1343 return EXT_MAX_BLOCK
;
1347 * ext4_ext_next_leaf_block:
1348 * returns first allocated block from next leaf or EXT_MAX_BLOCK
1350 static ext4_lblk_t
ext4_ext_next_leaf_block(struct inode
*inode
,
1351 struct ext4_ext_path
*path
)
1355 BUG_ON(path
== NULL
);
1356 depth
= path
->p_depth
;
1358 /* zero-tree has no leaf blocks at all */
1360 return EXT_MAX_BLOCK
;
1362 /* go to index block */
1365 while (depth
>= 0) {
1366 if (path
[depth
].p_idx
!=
1367 EXT_LAST_INDEX(path
[depth
].p_hdr
))
1368 return (ext4_lblk_t
)
1369 le32_to_cpu(path
[depth
].p_idx
[1].ei_block
);
1373 return EXT_MAX_BLOCK
;
1377 * ext4_ext_correct_indexes:
1378 * if leaf gets modified and modified extent is first in the leaf,
1379 * then we have to correct all indexes above.
1380 * TODO: do we need to correct tree in all cases?
1382 static int ext4_ext_correct_indexes(handle_t
*handle
, struct inode
*inode
,
1383 struct ext4_ext_path
*path
)
1385 struct ext4_extent_header
*eh
;
1386 int depth
= ext_depth(inode
);
1387 struct ext4_extent
*ex
;
1391 eh
= path
[depth
].p_hdr
;
1392 ex
= path
[depth
].p_ext
;
1397 /* there is no tree at all */
1401 if (ex
!= EXT_FIRST_EXTENT(eh
)) {
1402 /* we correct tree if first leaf got modified only */
1407 * TODO: we need correction if border is smaller than current one
1410 border
= path
[depth
].p_ext
->ee_block
;
1411 err
= ext4_ext_get_access(handle
, inode
, path
+ k
);
1414 path
[k
].p_idx
->ei_block
= border
;
1415 err
= ext4_ext_dirty(handle
, inode
, path
+ k
);
1420 /* change all left-side indexes */
1421 if (path
[k
+1].p_idx
!= EXT_FIRST_INDEX(path
[k
+1].p_hdr
))
1423 err
= ext4_ext_get_access(handle
, inode
, path
+ k
);
1426 path
[k
].p_idx
->ei_block
= border
;
1427 err
= ext4_ext_dirty(handle
, inode
, path
+ k
);
1436 ext4_can_extents_be_merged(struct inode
*inode
, struct ext4_extent
*ex1
,
1437 struct ext4_extent
*ex2
)
1439 unsigned short ext1_ee_len
, ext2_ee_len
, max_len
;
1442 * Make sure that either both extents are uninitialized, or
1445 if (ext4_ext_is_uninitialized(ex1
) ^ ext4_ext_is_uninitialized(ex2
))
1448 if (ext4_ext_is_uninitialized(ex1
))
1449 max_len
= EXT_UNINIT_MAX_LEN
;
1451 max_len
= EXT_INIT_MAX_LEN
;
1453 ext1_ee_len
= ext4_ext_get_actual_len(ex1
);
1454 ext2_ee_len
= ext4_ext_get_actual_len(ex2
);
1456 if (le32_to_cpu(ex1
->ee_block
) + ext1_ee_len
!=
1457 le32_to_cpu(ex2
->ee_block
))
1461 * To allow future support for preallocated extents to be added
1462 * as an RO_COMPAT feature, refuse to merge to extents if
1463 * this can result in the top bit of ee_len being set.
1465 if (ext1_ee_len
+ ext2_ee_len
> max_len
)
1467 #ifdef AGGRESSIVE_TEST
1468 if (ext1_ee_len
>= 4)
1472 if (ext_pblock(ex1
) + ext1_ee_len
== ext_pblock(ex2
))
1478 * This function tries to merge the "ex" extent to the next extent in the tree.
1479 * It always tries to merge towards right. If you want to merge towards
1480 * left, pass "ex - 1" as argument instead of "ex".
1481 * Returns 0 if the extents (ex and ex+1) were _not_ merged and returns
1482 * 1 if they got merged.
1484 int ext4_ext_try_to_merge(struct inode
*inode
,
1485 struct ext4_ext_path
*path
,
1486 struct ext4_extent
*ex
)
1488 struct ext4_extent_header
*eh
;
1489 unsigned int depth
, len
;
1491 int uninitialized
= 0;
1493 depth
= ext_depth(inode
);
1494 BUG_ON(path
[depth
].p_hdr
== NULL
);
1495 eh
= path
[depth
].p_hdr
;
1497 while (ex
< EXT_LAST_EXTENT(eh
)) {
1498 if (!ext4_can_extents_be_merged(inode
, ex
, ex
+ 1))
1500 /* merge with next extent! */
1501 if (ext4_ext_is_uninitialized(ex
))
1503 ex
->ee_len
= cpu_to_le16(ext4_ext_get_actual_len(ex
)
1504 + ext4_ext_get_actual_len(ex
+ 1));
1506 ext4_ext_mark_uninitialized(ex
);
1508 if (ex
+ 1 < EXT_LAST_EXTENT(eh
)) {
1509 len
= (EXT_LAST_EXTENT(eh
) - ex
- 1)
1510 * sizeof(struct ext4_extent
);
1511 memmove(ex
+ 1, ex
+ 2, len
);
1513 le16_add_cpu(&eh
->eh_entries
, -1);
1515 WARN_ON(eh
->eh_entries
== 0);
1516 if (!eh
->eh_entries
)
1517 ext4_error(inode
->i_sb
, "ext4_ext_try_to_merge",
1518 "inode#%lu, eh->eh_entries = 0!", inode
->i_ino
);
1525 * check if a portion of the "newext" extent overlaps with an
1528 * If there is an overlap discovered, it updates the length of the newext
1529 * such that there will be no overlap, and then returns 1.
1530 * If there is no overlap found, it returns 0.
1532 unsigned int ext4_ext_check_overlap(struct inode
*inode
,
1533 struct ext4_extent
*newext
,
1534 struct ext4_ext_path
*path
)
1537 unsigned int depth
, len1
;
1538 unsigned int ret
= 0;
1540 b1
= le32_to_cpu(newext
->ee_block
);
1541 len1
= ext4_ext_get_actual_len(newext
);
1542 depth
= ext_depth(inode
);
1543 if (!path
[depth
].p_ext
)
1545 b2
= le32_to_cpu(path
[depth
].p_ext
->ee_block
);
1548 * get the next allocated block if the extent in the path
1549 * is before the requested block(s)
1552 b2
= ext4_ext_next_allocated_block(path
);
1553 if (b2
== EXT_MAX_BLOCK
)
1557 /* check for wrap through zero on extent logical start block*/
1558 if (b1
+ len1
< b1
) {
1559 len1
= EXT_MAX_BLOCK
- b1
;
1560 newext
->ee_len
= cpu_to_le16(len1
);
1564 /* check for overlap */
1565 if (b1
+ len1
> b2
) {
1566 newext
->ee_len
= cpu_to_le16(b2
- b1
);
1574 * ext4_ext_insert_extent:
1575 * tries to merge requsted extent into the existing extent or
1576 * inserts requested extent as new one into the tree,
1577 * creating new leaf in the no-space case.
1579 int ext4_ext_insert_extent(handle_t
*handle
, struct inode
*inode
,
1580 struct ext4_ext_path
*path
,
1581 struct ext4_extent
*newext
)
1583 struct ext4_extent_header
*eh
;
1584 struct ext4_extent
*ex
, *fex
;
1585 struct ext4_extent
*nearex
; /* nearest extent */
1586 struct ext4_ext_path
*npath
= NULL
;
1587 int depth
, len
, err
;
1589 unsigned uninitialized
= 0;
1591 BUG_ON(ext4_ext_get_actual_len(newext
) == 0);
1592 depth
= ext_depth(inode
);
1593 ex
= path
[depth
].p_ext
;
1594 BUG_ON(path
[depth
].p_hdr
== NULL
);
1596 /* try to insert block into found extent and return */
1597 if (ex
&& ext4_can_extents_be_merged(inode
, ex
, newext
)) {
1598 ext_debug("append [%d]%d block to %d:[%d]%d (from %llu)\n",
1599 ext4_ext_is_uninitialized(newext
),
1600 ext4_ext_get_actual_len(newext
),
1601 le32_to_cpu(ex
->ee_block
),
1602 ext4_ext_is_uninitialized(ex
),
1603 ext4_ext_get_actual_len(ex
), ext_pblock(ex
));
1604 err
= ext4_ext_get_access(handle
, inode
, path
+ depth
);
1609 * ext4_can_extents_be_merged should have checked that either
1610 * both extents are uninitialized, or both aren't. Thus we
1611 * need to check only one of them here.
1613 if (ext4_ext_is_uninitialized(ex
))
1615 ex
->ee_len
= cpu_to_le16(ext4_ext_get_actual_len(ex
)
1616 + ext4_ext_get_actual_len(newext
));
1618 ext4_ext_mark_uninitialized(ex
);
1619 eh
= path
[depth
].p_hdr
;
1625 depth
= ext_depth(inode
);
1626 eh
= path
[depth
].p_hdr
;
1627 if (le16_to_cpu(eh
->eh_entries
) < le16_to_cpu(eh
->eh_max
))
1630 /* probably next leaf has space for us? */
1631 fex
= EXT_LAST_EXTENT(eh
);
1632 next
= ext4_ext_next_leaf_block(inode
, path
);
1633 if (le32_to_cpu(newext
->ee_block
) > le32_to_cpu(fex
->ee_block
)
1634 && next
!= EXT_MAX_BLOCK
) {
1635 ext_debug("next leaf block - %d\n", next
);
1636 BUG_ON(npath
!= NULL
);
1637 npath
= ext4_ext_find_extent(inode
, next
, NULL
);
1639 return PTR_ERR(npath
);
1640 BUG_ON(npath
->p_depth
!= path
->p_depth
);
1641 eh
= npath
[depth
].p_hdr
;
1642 if (le16_to_cpu(eh
->eh_entries
) < le16_to_cpu(eh
->eh_max
)) {
1643 ext_debug("next leaf isnt full(%d)\n",
1644 le16_to_cpu(eh
->eh_entries
));
1648 ext_debug("next leaf has no free space(%d,%d)\n",
1649 le16_to_cpu(eh
->eh_entries
), le16_to_cpu(eh
->eh_max
));
1653 * There is no free space in the found leaf.
1654 * We're gonna add a new leaf in the tree.
1656 err
= ext4_ext_create_new_leaf(handle
, inode
, path
, newext
);
1659 depth
= ext_depth(inode
);
1660 eh
= path
[depth
].p_hdr
;
1663 nearex
= path
[depth
].p_ext
;
1665 err
= ext4_ext_get_access(handle
, inode
, path
+ depth
);
1670 /* there is no extent in this leaf, create first one */
1671 ext_debug("first extent in the leaf: %d:%llu:[%d]%d\n",
1672 le32_to_cpu(newext
->ee_block
),
1674 ext4_ext_is_uninitialized(newext
),
1675 ext4_ext_get_actual_len(newext
));
1676 path
[depth
].p_ext
= EXT_FIRST_EXTENT(eh
);
1677 } else if (le32_to_cpu(newext
->ee_block
)
1678 > le32_to_cpu(nearex
->ee_block
)) {
1679 /* BUG_ON(newext->ee_block == nearex->ee_block); */
1680 if (nearex
!= EXT_LAST_EXTENT(eh
)) {
1681 len
= EXT_MAX_EXTENT(eh
) - nearex
;
1682 len
= (len
- 1) * sizeof(struct ext4_extent
);
1683 len
= len
< 0 ? 0 : len
;
1684 ext_debug("insert %d:%llu:[%d]%d after: nearest 0x%p, "
1685 "move %d from 0x%p to 0x%p\n",
1686 le32_to_cpu(newext
->ee_block
),
1688 ext4_ext_is_uninitialized(newext
),
1689 ext4_ext_get_actual_len(newext
),
1690 nearex
, len
, nearex
+ 1, nearex
+ 2);
1691 memmove(nearex
+ 2, nearex
+ 1, len
);
1693 path
[depth
].p_ext
= nearex
+ 1;
1695 BUG_ON(newext
->ee_block
== nearex
->ee_block
);
1696 len
= (EXT_MAX_EXTENT(eh
) - nearex
) * sizeof(struct ext4_extent
);
1697 len
= len
< 0 ? 0 : len
;
1698 ext_debug("insert %d:%llu:[%d]%d before: nearest 0x%p, "
1699 "move %d from 0x%p to 0x%p\n",
1700 le32_to_cpu(newext
->ee_block
),
1702 ext4_ext_is_uninitialized(newext
),
1703 ext4_ext_get_actual_len(newext
),
1704 nearex
, len
, nearex
+ 1, nearex
+ 2);
1705 memmove(nearex
+ 1, nearex
, len
);
1706 path
[depth
].p_ext
= nearex
;
1709 le16_add_cpu(&eh
->eh_entries
, 1);
1710 nearex
= path
[depth
].p_ext
;
1711 nearex
->ee_block
= newext
->ee_block
;
1712 ext4_ext_store_pblock(nearex
, ext_pblock(newext
));
1713 nearex
->ee_len
= newext
->ee_len
;
1716 /* try to merge extents to the right */
1717 ext4_ext_try_to_merge(inode
, path
, nearex
);
1719 /* try to merge extents to the left */
1721 /* time to correct all indexes above */
1722 err
= ext4_ext_correct_indexes(handle
, inode
, path
);
1726 err
= ext4_ext_dirty(handle
, inode
, path
+ depth
);
1730 ext4_ext_drop_refs(npath
);
1733 ext4_ext_invalidate_cache(inode
);
1737 int ext4_ext_walk_space(struct inode
*inode
, ext4_lblk_t block
,
1738 ext4_lblk_t num
, ext_prepare_callback func
,
1741 struct ext4_ext_path
*path
= NULL
;
1742 struct ext4_ext_cache cbex
;
1743 struct ext4_extent
*ex
;
1744 ext4_lblk_t next
, start
= 0, end
= 0;
1745 ext4_lblk_t last
= block
+ num
;
1746 int depth
, exists
, err
= 0;
1748 BUG_ON(func
== NULL
);
1749 BUG_ON(inode
== NULL
);
1751 while (block
< last
&& block
!= EXT_MAX_BLOCK
) {
1753 /* find extent for this block */
1754 path
= ext4_ext_find_extent(inode
, block
, path
);
1756 err
= PTR_ERR(path
);
1761 depth
= ext_depth(inode
);
1762 BUG_ON(path
[depth
].p_hdr
== NULL
);
1763 ex
= path
[depth
].p_ext
;
1764 next
= ext4_ext_next_allocated_block(path
);
1768 /* there is no extent yet, so try to allocate
1769 * all requested space */
1772 } else if (le32_to_cpu(ex
->ee_block
) > block
) {
1773 /* need to allocate space before found extent */
1775 end
= le32_to_cpu(ex
->ee_block
);
1776 if (block
+ num
< end
)
1778 } else if (block
>= le32_to_cpu(ex
->ee_block
)
1779 + ext4_ext_get_actual_len(ex
)) {
1780 /* need to allocate space after found extent */
1785 } else if (block
>= le32_to_cpu(ex
->ee_block
)) {
1787 * some part of requested space is covered
1791 end
= le32_to_cpu(ex
->ee_block
)
1792 + ext4_ext_get_actual_len(ex
);
1793 if (block
+ num
< end
)
1799 BUG_ON(end
<= start
);
1802 cbex
.ec_block
= start
;
1803 cbex
.ec_len
= end
- start
;
1805 cbex
.ec_type
= EXT4_EXT_CACHE_GAP
;
1807 cbex
.ec_block
= le32_to_cpu(ex
->ee_block
);
1808 cbex
.ec_len
= ext4_ext_get_actual_len(ex
);
1809 cbex
.ec_start
= ext_pblock(ex
);
1810 cbex
.ec_type
= EXT4_EXT_CACHE_EXTENT
;
1813 BUG_ON(cbex
.ec_len
== 0);
1814 err
= func(inode
, path
, &cbex
, ex
, cbdata
);
1815 ext4_ext_drop_refs(path
);
1820 if (err
== EXT_REPEAT
)
1822 else if (err
== EXT_BREAK
) {
1827 if (ext_depth(inode
) != depth
) {
1828 /* depth was changed. we have to realloc path */
1833 block
= cbex
.ec_block
+ cbex
.ec_len
;
1837 ext4_ext_drop_refs(path
);
1845 ext4_ext_put_in_cache(struct inode
*inode
, ext4_lblk_t block
,
1846 __u32 len
, ext4_fsblk_t start
, int type
)
1848 struct ext4_ext_cache
*cex
;
1850 spin_lock(&EXT4_I(inode
)->i_block_reservation_lock
);
1851 cex
= &EXT4_I(inode
)->i_cached_extent
;
1852 cex
->ec_type
= type
;
1853 cex
->ec_block
= block
;
1855 cex
->ec_start
= start
;
1856 spin_unlock(&EXT4_I(inode
)->i_block_reservation_lock
);
1860 * ext4_ext_put_gap_in_cache:
1861 * calculate boundaries of the gap that the requested block fits into
1862 * and cache this gap
1865 ext4_ext_put_gap_in_cache(struct inode
*inode
, struct ext4_ext_path
*path
,
1868 int depth
= ext_depth(inode
);
1871 struct ext4_extent
*ex
;
1873 ex
= path
[depth
].p_ext
;
1875 /* there is no extent yet, so gap is [0;-] */
1877 len
= EXT_MAX_BLOCK
;
1878 ext_debug("cache gap(whole file):");
1879 } else if (block
< le32_to_cpu(ex
->ee_block
)) {
1881 len
= le32_to_cpu(ex
->ee_block
) - block
;
1882 ext_debug("cache gap(before): %u [%u:%u]",
1884 le32_to_cpu(ex
->ee_block
),
1885 ext4_ext_get_actual_len(ex
));
1886 } else if (block
>= le32_to_cpu(ex
->ee_block
)
1887 + ext4_ext_get_actual_len(ex
)) {
1889 lblock
= le32_to_cpu(ex
->ee_block
)
1890 + ext4_ext_get_actual_len(ex
);
1892 next
= ext4_ext_next_allocated_block(path
);
1893 ext_debug("cache gap(after): [%u:%u] %u",
1894 le32_to_cpu(ex
->ee_block
),
1895 ext4_ext_get_actual_len(ex
),
1897 BUG_ON(next
== lblock
);
1898 len
= next
- lblock
;
1904 ext_debug(" -> %u:%lu\n", lblock
, len
);
1905 ext4_ext_put_in_cache(inode
, lblock
, len
, 0, EXT4_EXT_CACHE_GAP
);
1909 ext4_ext_in_cache(struct inode
*inode
, ext4_lblk_t block
,
1910 struct ext4_extent
*ex
)
1912 struct ext4_ext_cache
*cex
;
1913 int ret
= EXT4_EXT_CACHE_NO
;
1916 * We borrow i_block_reservation_lock to protect i_cached_extent
1918 spin_lock(&EXT4_I(inode
)->i_block_reservation_lock
);
1919 cex
= &EXT4_I(inode
)->i_cached_extent
;
1921 /* has cache valid data? */
1922 if (cex
->ec_type
== EXT4_EXT_CACHE_NO
)
1925 BUG_ON(cex
->ec_type
!= EXT4_EXT_CACHE_GAP
&&
1926 cex
->ec_type
!= EXT4_EXT_CACHE_EXTENT
);
1927 if (block
>= cex
->ec_block
&& block
< cex
->ec_block
+ cex
->ec_len
) {
1928 ex
->ee_block
= cpu_to_le32(cex
->ec_block
);
1929 ext4_ext_store_pblock(ex
, cex
->ec_start
);
1930 ex
->ee_len
= cpu_to_le16(cex
->ec_len
);
1931 ext_debug("%u cached by %u:%u:%llu\n",
1933 cex
->ec_block
, cex
->ec_len
, cex
->ec_start
);
1937 spin_unlock(&EXT4_I(inode
)->i_block_reservation_lock
);
1943 * removes index from the index block.
1944 * It's used in truncate case only, thus all requests are for
1945 * last index in the block only.
1947 static int ext4_ext_rm_idx(handle_t
*handle
, struct inode
*inode
,
1948 struct ext4_ext_path
*path
)
1950 struct buffer_head
*bh
;
1954 /* free index block */
1956 leaf
= idx_pblock(path
->p_idx
);
1957 BUG_ON(path
->p_hdr
->eh_entries
== 0);
1958 err
= ext4_ext_get_access(handle
, inode
, path
);
1961 le16_add_cpu(&path
->p_hdr
->eh_entries
, -1);
1962 err
= ext4_ext_dirty(handle
, inode
, path
);
1965 ext_debug("index is empty, remove it, free block %llu\n", leaf
);
1966 bh
= sb_find_get_block(inode
->i_sb
, leaf
);
1967 ext4_forget(handle
, 1, inode
, bh
, leaf
);
1968 ext4_free_blocks(handle
, inode
, leaf
, 1, 1);
1973 * ext4_ext_calc_credits_for_single_extent:
1974 * This routine returns max. credits that needed to insert an extent
1975 * to the extent tree.
1976 * When pass the actual path, the caller should calculate credits
1979 int ext4_ext_calc_credits_for_single_extent(struct inode
*inode
, int nrblocks
,
1980 struct ext4_ext_path
*path
)
1983 int depth
= ext_depth(inode
);
1986 /* probably there is space in leaf? */
1987 if (le16_to_cpu(path
[depth
].p_hdr
->eh_entries
)
1988 < le16_to_cpu(path
[depth
].p_hdr
->eh_max
)) {
1991 * There are some space in the leaf tree, no
1992 * need to account for leaf block credit
1994 * bitmaps and block group descriptor blocks
1995 * and other metadat blocks still need to be
1998 /* 1 bitmap, 1 block group descriptor */
1999 ret
= 2 + EXT4_META_TRANS_BLOCKS(inode
->i_sb
);
2004 return ext4_chunk_trans_blocks(inode
, nrblocks
);
2008 * How many index/leaf blocks need to change/allocate to modify nrblocks?
2010 * if nrblocks are fit in a single extent (chunk flag is 1), then
2011 * in the worse case, each tree level index/leaf need to be changed
2012 * if the tree split due to insert a new extent, then the old tree
2013 * index/leaf need to be updated too
2015 * If the nrblocks are discontiguous, they could cause
2016 * the whole tree split more than once, but this is really rare.
2018 int ext4_ext_index_trans_blocks(struct inode
*inode
, int nrblocks
, int chunk
)
2021 int depth
= ext_depth(inode
);
2031 static int ext4_remove_blocks(handle_t
*handle
, struct inode
*inode
,
2032 struct ext4_extent
*ex
,
2033 ext4_lblk_t from
, ext4_lblk_t to
)
2035 struct buffer_head
*bh
;
2036 unsigned short ee_len
= ext4_ext_get_actual_len(ex
);
2037 int i
, metadata
= 0;
2039 if (S_ISDIR(inode
->i_mode
) || S_ISLNK(inode
->i_mode
))
2041 #ifdef EXTENTS_STATS
2043 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
2044 spin_lock(&sbi
->s_ext_stats_lock
);
2045 sbi
->s_ext_blocks
+= ee_len
;
2046 sbi
->s_ext_extents
++;
2047 if (ee_len
< sbi
->s_ext_min
)
2048 sbi
->s_ext_min
= ee_len
;
2049 if (ee_len
> sbi
->s_ext_max
)
2050 sbi
->s_ext_max
= ee_len
;
2051 if (ext_depth(inode
) > sbi
->s_depth_max
)
2052 sbi
->s_depth_max
= ext_depth(inode
);
2053 spin_unlock(&sbi
->s_ext_stats_lock
);
2056 if (from
>= le32_to_cpu(ex
->ee_block
)
2057 && to
== le32_to_cpu(ex
->ee_block
) + ee_len
- 1) {
2062 num
= le32_to_cpu(ex
->ee_block
) + ee_len
- from
;
2063 start
= ext_pblock(ex
) + ee_len
- num
;
2064 ext_debug("free last %u blocks starting %llu\n", num
, start
);
2065 for (i
= 0; i
< num
; i
++) {
2066 bh
= sb_find_get_block(inode
->i_sb
, start
+ i
);
2067 ext4_forget(handle
, 0, inode
, bh
, start
+ i
);
2069 ext4_free_blocks(handle
, inode
, start
, num
, metadata
);
2070 } else if (from
== le32_to_cpu(ex
->ee_block
)
2071 && to
<= le32_to_cpu(ex
->ee_block
) + ee_len
- 1) {
2072 printk(KERN_INFO
"strange request: removal %u-%u from %u:%u\n",
2073 from
, to
, le32_to_cpu(ex
->ee_block
), ee_len
);
2075 printk(KERN_INFO
"strange request: removal(2) "
2076 "%u-%u from %u:%u\n",
2077 from
, to
, le32_to_cpu(ex
->ee_block
), ee_len
);
2083 ext4_ext_rm_leaf(handle_t
*handle
, struct inode
*inode
,
2084 struct ext4_ext_path
*path
, ext4_lblk_t start
)
2086 int err
= 0, correct_index
= 0;
2087 int depth
= ext_depth(inode
), credits
;
2088 struct ext4_extent_header
*eh
;
2089 ext4_lblk_t a
, b
, block
;
2091 ext4_lblk_t ex_ee_block
;
2092 unsigned short ex_ee_len
;
2093 unsigned uninitialized
= 0;
2094 struct ext4_extent
*ex
;
2096 /* the header must be checked already in ext4_ext_remove_space() */
2097 ext_debug("truncate since %u in leaf\n", start
);
2098 if (!path
[depth
].p_hdr
)
2099 path
[depth
].p_hdr
= ext_block_hdr(path
[depth
].p_bh
);
2100 eh
= path
[depth
].p_hdr
;
2103 /* find where to start removing */
2104 ex
= EXT_LAST_EXTENT(eh
);
2106 ex_ee_block
= le32_to_cpu(ex
->ee_block
);
2107 ex_ee_len
= ext4_ext_get_actual_len(ex
);
2109 while (ex
>= EXT_FIRST_EXTENT(eh
) &&
2110 ex_ee_block
+ ex_ee_len
> start
) {
2112 if (ext4_ext_is_uninitialized(ex
))
2117 ext_debug("remove ext %u:[%d]%d\n", ex_ee_block
,
2118 uninitialized
, ex_ee_len
);
2119 path
[depth
].p_ext
= ex
;
2121 a
= ex_ee_block
> start
? ex_ee_block
: start
;
2122 b
= ex_ee_block
+ ex_ee_len
- 1 < EXT_MAX_BLOCK
?
2123 ex_ee_block
+ ex_ee_len
- 1 : EXT_MAX_BLOCK
;
2125 ext_debug(" border %u:%u\n", a
, b
);
2127 if (a
!= ex_ee_block
&& b
!= ex_ee_block
+ ex_ee_len
- 1) {
2131 } else if (a
!= ex_ee_block
) {
2132 /* remove tail of the extent */
2133 block
= ex_ee_block
;
2135 } else if (b
!= ex_ee_block
+ ex_ee_len
- 1) {
2136 /* remove head of the extent */
2139 /* there is no "make a hole" API yet */
2142 /* remove whole extent: excellent! */
2143 block
= ex_ee_block
;
2145 BUG_ON(a
!= ex_ee_block
);
2146 BUG_ON(b
!= ex_ee_block
+ ex_ee_len
- 1);
2150 * 3 for leaf, sb, and inode plus 2 (bmap and group
2151 * descriptor) for each block group; assume two block
2152 * groups plus ex_ee_len/blocks_per_block_group for
2155 credits
= 7 + 2*(ex_ee_len
/EXT4_BLOCKS_PER_GROUP(inode
->i_sb
));
2156 if (ex
== EXT_FIRST_EXTENT(eh
)) {
2158 credits
+= (ext_depth(inode
)) + 1;
2160 credits
+= 2 * EXT4_QUOTA_TRANS_BLOCKS(inode
->i_sb
);
2162 err
= ext4_ext_truncate_extend_restart(handle
, inode
, credits
);
2166 err
= ext4_ext_get_access(handle
, inode
, path
+ depth
);
2170 err
= ext4_remove_blocks(handle
, inode
, ex
, a
, b
);
2175 /* this extent is removed; mark slot entirely unused */
2176 ext4_ext_store_pblock(ex
, 0);
2177 le16_add_cpu(&eh
->eh_entries
, -1);
2180 ex
->ee_block
= cpu_to_le32(block
);
2181 ex
->ee_len
= cpu_to_le16(num
);
2183 * Do not mark uninitialized if all the blocks in the
2184 * extent have been removed.
2186 if (uninitialized
&& num
)
2187 ext4_ext_mark_uninitialized(ex
);
2189 err
= ext4_ext_dirty(handle
, inode
, path
+ depth
);
2193 ext_debug("new extent: %u:%u:%llu\n", block
, num
,
2196 ex_ee_block
= le32_to_cpu(ex
->ee_block
);
2197 ex_ee_len
= ext4_ext_get_actual_len(ex
);
2200 if (correct_index
&& eh
->eh_entries
)
2201 err
= ext4_ext_correct_indexes(handle
, inode
, path
);
2203 /* if this leaf is free, then we should
2204 * remove it from index block above */
2205 if (err
== 0 && eh
->eh_entries
== 0 && path
[depth
].p_bh
!= NULL
)
2206 err
= ext4_ext_rm_idx(handle
, inode
, path
+ depth
);
2213 * ext4_ext_more_to_rm:
2214 * returns 1 if current index has to be freed (even partial)
2217 ext4_ext_more_to_rm(struct ext4_ext_path
*path
)
2219 BUG_ON(path
->p_idx
== NULL
);
2221 if (path
->p_idx
< EXT_FIRST_INDEX(path
->p_hdr
))
2225 * if truncate on deeper level happened, it wasn't partial,
2226 * so we have to consider current index for truncation
2228 if (le16_to_cpu(path
->p_hdr
->eh_entries
) == path
->p_block
)
2233 static int ext4_ext_remove_space(struct inode
*inode
, ext4_lblk_t start
)
2235 struct super_block
*sb
= inode
->i_sb
;
2236 int depth
= ext_depth(inode
);
2237 struct ext4_ext_path
*path
;
2241 ext_debug("truncate since %u\n", start
);
2243 /* probably first extent we're gonna free will be last in block */
2244 handle
= ext4_journal_start(inode
, depth
+ 1);
2246 return PTR_ERR(handle
);
2248 ext4_ext_invalidate_cache(inode
);
2251 * We start scanning from right side, freeing all the blocks
2252 * after i_size and walking into the tree depth-wise.
2254 path
= kzalloc(sizeof(struct ext4_ext_path
) * (depth
+ 1), GFP_NOFS
);
2256 ext4_journal_stop(handle
);
2259 path
[0].p_hdr
= ext_inode_hdr(inode
);
2260 if (ext4_ext_check(inode
, path
[0].p_hdr
, depth
)) {
2264 path
[0].p_depth
= depth
;
2266 while (i
>= 0 && err
== 0) {
2268 /* this is leaf block */
2269 err
= ext4_ext_rm_leaf(handle
, inode
, path
, start
);
2270 /* root level has p_bh == NULL, brelse() eats this */
2271 brelse(path
[i
].p_bh
);
2272 path
[i
].p_bh
= NULL
;
2277 /* this is index block */
2278 if (!path
[i
].p_hdr
) {
2279 ext_debug("initialize header\n");
2280 path
[i
].p_hdr
= ext_block_hdr(path
[i
].p_bh
);
2283 if (!path
[i
].p_idx
) {
2284 /* this level hasn't been touched yet */
2285 path
[i
].p_idx
= EXT_LAST_INDEX(path
[i
].p_hdr
);
2286 path
[i
].p_block
= le16_to_cpu(path
[i
].p_hdr
->eh_entries
)+1;
2287 ext_debug("init index ptr: hdr 0x%p, num %d\n",
2289 le16_to_cpu(path
[i
].p_hdr
->eh_entries
));
2291 /* we were already here, see at next index */
2295 ext_debug("level %d - index, first 0x%p, cur 0x%p\n",
2296 i
, EXT_FIRST_INDEX(path
[i
].p_hdr
),
2298 if (ext4_ext_more_to_rm(path
+ i
)) {
2299 struct buffer_head
*bh
;
2300 /* go to the next level */
2301 ext_debug("move to level %d (block %llu)\n",
2302 i
+ 1, idx_pblock(path
[i
].p_idx
));
2303 memset(path
+ i
+ 1, 0, sizeof(*path
));
2304 bh
= sb_bread(sb
, idx_pblock(path
[i
].p_idx
));
2306 /* should we reset i_size? */
2310 if (WARN_ON(i
+ 1 > depth
)) {
2314 if (ext4_ext_check(inode
, ext_block_hdr(bh
),
2319 path
[i
+ 1].p_bh
= bh
;
2321 /* save actual number of indexes since this
2322 * number is changed at the next iteration */
2323 path
[i
].p_block
= le16_to_cpu(path
[i
].p_hdr
->eh_entries
);
2326 /* we finished processing this index, go up */
2327 if (path
[i
].p_hdr
->eh_entries
== 0 && i
> 0) {
2328 /* index is empty, remove it;
2329 * handle must be already prepared by the
2330 * truncatei_leaf() */
2331 err
= ext4_ext_rm_idx(handle
, inode
, path
+ i
);
2333 /* root level has p_bh == NULL, brelse() eats this */
2334 brelse(path
[i
].p_bh
);
2335 path
[i
].p_bh
= NULL
;
2337 ext_debug("return to level %d\n", i
);
2341 /* TODO: flexible tree reduction should be here */
2342 if (path
->p_hdr
->eh_entries
== 0) {
2344 * truncate to zero freed all the tree,
2345 * so we need to correct eh_depth
2347 err
= ext4_ext_get_access(handle
, inode
, path
);
2349 ext_inode_hdr(inode
)->eh_depth
= 0;
2350 ext_inode_hdr(inode
)->eh_max
=
2351 cpu_to_le16(ext4_ext_space_root(inode
));
2352 err
= ext4_ext_dirty(handle
, inode
, path
);
2356 ext4_ext_drop_refs(path
);
2358 ext4_journal_stop(handle
);
2364 * called at mount time
2366 void ext4_ext_init(struct super_block
*sb
)
2369 * possible initialization would be here
2372 if (EXT4_HAS_INCOMPAT_FEATURE(sb
, EXT4_FEATURE_INCOMPAT_EXTENTS
)) {
2373 printk(KERN_INFO
"EXT4-fs: file extents enabled");
2374 #ifdef AGGRESSIVE_TEST
2375 printk(", aggressive tests");
2377 #ifdef CHECK_BINSEARCH
2378 printk(", check binsearch");
2380 #ifdef EXTENTS_STATS
2384 #ifdef EXTENTS_STATS
2385 spin_lock_init(&EXT4_SB(sb
)->s_ext_stats_lock
);
2386 EXT4_SB(sb
)->s_ext_min
= 1 << 30;
2387 EXT4_SB(sb
)->s_ext_max
= 0;
2393 * called at umount time
2395 void ext4_ext_release(struct super_block
*sb
)
2397 if (!EXT4_HAS_INCOMPAT_FEATURE(sb
, EXT4_FEATURE_INCOMPAT_EXTENTS
))
2400 #ifdef EXTENTS_STATS
2401 if (EXT4_SB(sb
)->s_ext_blocks
&& EXT4_SB(sb
)->s_ext_extents
) {
2402 struct ext4_sb_info
*sbi
= EXT4_SB(sb
);
2403 printk(KERN_ERR
"EXT4-fs: %lu blocks in %lu extents (%lu ave)\n",
2404 sbi
->s_ext_blocks
, sbi
->s_ext_extents
,
2405 sbi
->s_ext_blocks
/ sbi
->s_ext_extents
);
2406 printk(KERN_ERR
"EXT4-fs: extents: %lu min, %lu max, max depth %lu\n",
2407 sbi
->s_ext_min
, sbi
->s_ext_max
, sbi
->s_depth_max
);
2412 static void bi_complete(struct bio
*bio
, int error
)
2414 complete((struct completion
*)bio
->bi_private
);
2417 /* FIXME!! we need to try to merge to left or right after zero-out */
2418 static int ext4_ext_zeroout(struct inode
*inode
, struct ext4_extent
*ex
)
2422 int blkbits
, blocksize
;
2424 struct completion event
;
2425 unsigned int ee_len
, len
, done
, offset
;
2428 blkbits
= inode
->i_blkbits
;
2429 blocksize
= inode
->i_sb
->s_blocksize
;
2430 ee_len
= ext4_ext_get_actual_len(ex
);
2431 ee_pblock
= ext_pblock(ex
);
2433 /* convert ee_pblock to 512 byte sectors */
2434 ee_pblock
= ee_pblock
<< (blkbits
- 9);
2436 while (ee_len
> 0) {
2438 if (ee_len
> BIO_MAX_PAGES
)
2439 len
= BIO_MAX_PAGES
;
2443 bio
= bio_alloc(GFP_NOIO
, len
);
2444 bio
->bi_sector
= ee_pblock
;
2445 bio
->bi_bdev
= inode
->i_sb
->s_bdev
;
2449 while (done
< len
) {
2450 ret
= bio_add_page(bio
, ZERO_PAGE(0),
2452 if (ret
!= blocksize
) {
2454 * We can't add any more pages because of
2455 * hardware limitations. Start a new bio.
2460 offset
+= blocksize
;
2461 if (offset
>= PAGE_CACHE_SIZE
)
2465 init_completion(&event
);
2466 bio
->bi_private
= &event
;
2467 bio
->bi_end_io
= bi_complete
;
2468 submit_bio(WRITE
, bio
);
2469 wait_for_completion(&event
);
2471 if (test_bit(BIO_UPTODATE
, &bio
->bi_flags
))
2479 ee_pblock
+= done
<< (blkbits
- 9);
2484 #define EXT4_EXT_ZERO_LEN 7
2487 * This function is called by ext4_ext_get_blocks() if someone tries to write
2488 * to an uninitialized extent. It may result in splitting the uninitialized
2489 * extent into multiple extents (upto three - one initialized and two
2491 * There are three possibilities:
2492 * a> There is no split required: Entire extent should be initialized
2493 * b> Splits in two extents: Write is happening at either end of the extent
2494 * c> Splits in three extents: Somone is writing in middle of the extent
2496 static int ext4_ext_convert_to_initialized(handle_t
*handle
,
2497 struct inode
*inode
,
2498 struct ext4_ext_path
*path
,
2500 unsigned int max_blocks
)
2502 struct ext4_extent
*ex
, newex
, orig_ex
;
2503 struct ext4_extent
*ex1
= NULL
;
2504 struct ext4_extent
*ex2
= NULL
;
2505 struct ext4_extent
*ex3
= NULL
;
2506 struct ext4_extent_header
*eh
;
2507 ext4_lblk_t ee_block
;
2508 unsigned int allocated
, ee_len
, depth
;
2509 ext4_fsblk_t newblock
;
2513 depth
= ext_depth(inode
);
2514 eh
= path
[depth
].p_hdr
;
2515 ex
= path
[depth
].p_ext
;
2516 ee_block
= le32_to_cpu(ex
->ee_block
);
2517 ee_len
= ext4_ext_get_actual_len(ex
);
2518 allocated
= ee_len
- (iblock
- ee_block
);
2519 newblock
= iblock
- ee_block
+ ext_pblock(ex
);
2521 orig_ex
.ee_block
= ex
->ee_block
;
2522 orig_ex
.ee_len
= cpu_to_le16(ee_len
);
2523 ext4_ext_store_pblock(&orig_ex
, ext_pblock(ex
));
2525 err
= ext4_ext_get_access(handle
, inode
, path
+ depth
);
2528 /* If extent has less than 2*EXT4_EXT_ZERO_LEN zerout directly */
2529 if (ee_len
<= 2*EXT4_EXT_ZERO_LEN
) {
2530 err
= ext4_ext_zeroout(inode
, &orig_ex
);
2532 goto fix_extent_len
;
2533 /* update the extent length and mark as initialized */
2534 ex
->ee_block
= orig_ex
.ee_block
;
2535 ex
->ee_len
= orig_ex
.ee_len
;
2536 ext4_ext_store_pblock(ex
, ext_pblock(&orig_ex
));
2537 ext4_ext_dirty(handle
, inode
, path
+ depth
);
2538 /* zeroed the full extent */
2542 /* ex1: ee_block to iblock - 1 : uninitialized */
2543 if (iblock
> ee_block
) {
2545 ex1
->ee_len
= cpu_to_le16(iblock
- ee_block
);
2546 ext4_ext_mark_uninitialized(ex1
);
2550 * for sanity, update the length of the ex2 extent before
2551 * we insert ex3, if ex1 is NULL. This is to avoid temporary
2552 * overlap of blocks.
2554 if (!ex1
&& allocated
> max_blocks
)
2555 ex2
->ee_len
= cpu_to_le16(max_blocks
);
2556 /* ex3: to ee_block + ee_len : uninitialised */
2557 if (allocated
> max_blocks
) {
2558 unsigned int newdepth
;
2559 /* If extent has less than EXT4_EXT_ZERO_LEN zerout directly */
2560 if (allocated
<= EXT4_EXT_ZERO_LEN
) {
2562 * iblock == ee_block is handled by the zerouout
2564 * Mark first half uninitialized.
2565 * Mark second half initialized and zero out the
2566 * initialized extent
2568 ex
->ee_block
= orig_ex
.ee_block
;
2569 ex
->ee_len
= cpu_to_le16(ee_len
- allocated
);
2570 ext4_ext_mark_uninitialized(ex
);
2571 ext4_ext_store_pblock(ex
, ext_pblock(&orig_ex
));
2572 ext4_ext_dirty(handle
, inode
, path
+ depth
);
2575 ex3
->ee_block
= cpu_to_le32(iblock
);
2576 ext4_ext_store_pblock(ex3
, newblock
);
2577 ex3
->ee_len
= cpu_to_le16(allocated
);
2578 err
= ext4_ext_insert_extent(handle
, inode
, path
, ex3
);
2579 if (err
== -ENOSPC
) {
2580 err
= ext4_ext_zeroout(inode
, &orig_ex
);
2582 goto fix_extent_len
;
2583 ex
->ee_block
= orig_ex
.ee_block
;
2584 ex
->ee_len
= orig_ex
.ee_len
;
2585 ext4_ext_store_pblock(ex
, ext_pblock(&orig_ex
));
2586 ext4_ext_dirty(handle
, inode
, path
+ depth
);
2587 /* blocks available from iblock */
2591 goto fix_extent_len
;
2594 * We need to zero out the second half because
2595 * an fallocate request can update file size and
2596 * converting the second half to initialized extent
2597 * implies that we can leak some junk data to user
2600 err
= ext4_ext_zeroout(inode
, ex3
);
2603 * We should actually mark the
2604 * second half as uninit and return error
2605 * Insert would have changed the extent
2607 depth
= ext_depth(inode
);
2608 ext4_ext_drop_refs(path
);
2609 path
= ext4_ext_find_extent(inode
,
2612 err
= PTR_ERR(path
);
2615 /* get the second half extent details */
2616 ex
= path
[depth
].p_ext
;
2617 err
= ext4_ext_get_access(handle
, inode
,
2621 ext4_ext_mark_uninitialized(ex
);
2622 ext4_ext_dirty(handle
, inode
, path
+ depth
);
2626 /* zeroed the second half */
2630 ex3
->ee_block
= cpu_to_le32(iblock
+ max_blocks
);
2631 ext4_ext_store_pblock(ex3
, newblock
+ max_blocks
);
2632 ex3
->ee_len
= cpu_to_le16(allocated
- max_blocks
);
2633 ext4_ext_mark_uninitialized(ex3
);
2634 err
= ext4_ext_insert_extent(handle
, inode
, path
, ex3
);
2635 if (err
== -ENOSPC
) {
2636 err
= ext4_ext_zeroout(inode
, &orig_ex
);
2638 goto fix_extent_len
;
2639 /* update the extent length and mark as initialized */
2640 ex
->ee_block
= orig_ex
.ee_block
;
2641 ex
->ee_len
= orig_ex
.ee_len
;
2642 ext4_ext_store_pblock(ex
, ext_pblock(&orig_ex
));
2643 ext4_ext_dirty(handle
, inode
, path
+ depth
);
2644 /* zeroed the full extent */
2645 /* blocks available from iblock */
2649 goto fix_extent_len
;
2651 * The depth, and hence eh & ex might change
2652 * as part of the insert above.
2654 newdepth
= ext_depth(inode
);
2656 * update the extent length after successful insert of the
2659 orig_ex
.ee_len
= cpu_to_le16(ee_len
-
2660 ext4_ext_get_actual_len(ex3
));
2662 ext4_ext_drop_refs(path
);
2663 path
= ext4_ext_find_extent(inode
, iblock
, path
);
2665 err
= PTR_ERR(path
);
2668 eh
= path
[depth
].p_hdr
;
2669 ex
= path
[depth
].p_ext
;
2673 err
= ext4_ext_get_access(handle
, inode
, path
+ depth
);
2677 allocated
= max_blocks
;
2679 /* If extent has less than EXT4_EXT_ZERO_LEN and we are trying
2680 * to insert a extent in the middle zerout directly
2681 * otherwise give the extent a chance to merge to left
2683 if (le16_to_cpu(orig_ex
.ee_len
) <= EXT4_EXT_ZERO_LEN
&&
2684 iblock
!= ee_block
) {
2685 err
= ext4_ext_zeroout(inode
, &orig_ex
);
2687 goto fix_extent_len
;
2688 /* update the extent length and mark as initialized */
2689 ex
->ee_block
= orig_ex
.ee_block
;
2690 ex
->ee_len
= orig_ex
.ee_len
;
2691 ext4_ext_store_pblock(ex
, ext_pblock(&orig_ex
));
2692 ext4_ext_dirty(handle
, inode
, path
+ depth
);
2693 /* zero out the first half */
2694 /* blocks available from iblock */
2699 * If there was a change of depth as part of the
2700 * insertion of ex3 above, we need to update the length
2701 * of the ex1 extent again here
2703 if (ex1
&& ex1
!= ex
) {
2705 ex1
->ee_len
= cpu_to_le16(iblock
- ee_block
);
2706 ext4_ext_mark_uninitialized(ex1
);
2709 /* ex2: iblock to iblock + maxblocks-1 : initialised */
2710 ex2
->ee_block
= cpu_to_le32(iblock
);
2711 ext4_ext_store_pblock(ex2
, newblock
);
2712 ex2
->ee_len
= cpu_to_le16(allocated
);
2716 * New (initialized) extent starts from the first block
2717 * in the current extent. i.e., ex2 == ex
2718 * We have to see if it can be merged with the extent
2721 if (ex2
> EXT_FIRST_EXTENT(eh
)) {
2723 * To merge left, pass "ex2 - 1" to try_to_merge(),
2724 * since it merges towards right _only_.
2726 ret
= ext4_ext_try_to_merge(inode
, path
, ex2
- 1);
2728 err
= ext4_ext_correct_indexes(handle
, inode
, path
);
2731 depth
= ext_depth(inode
);
2736 * Try to Merge towards right. This might be required
2737 * only when the whole extent is being written to.
2738 * i.e. ex2 == ex and ex3 == NULL.
2741 ret
= ext4_ext_try_to_merge(inode
, path
, ex2
);
2743 err
= ext4_ext_correct_indexes(handle
, inode
, path
);
2748 /* Mark modified extent as dirty */
2749 err
= ext4_ext_dirty(handle
, inode
, path
+ depth
);
2752 err
= ext4_ext_insert_extent(handle
, inode
, path
, &newex
);
2753 if (err
== -ENOSPC
) {
2754 err
= ext4_ext_zeroout(inode
, &orig_ex
);
2756 goto fix_extent_len
;
2757 /* update the extent length and mark as initialized */
2758 ex
->ee_block
= orig_ex
.ee_block
;
2759 ex
->ee_len
= orig_ex
.ee_len
;
2760 ext4_ext_store_pblock(ex
, ext_pblock(&orig_ex
));
2761 ext4_ext_dirty(handle
, inode
, path
+ depth
);
2762 /* zero out the first half */
2765 goto fix_extent_len
;
2767 ext4_ext_show_leaf(inode
, path
);
2768 return err
? err
: allocated
;
2771 ex
->ee_block
= orig_ex
.ee_block
;
2772 ex
->ee_len
= orig_ex
.ee_len
;
2773 ext4_ext_store_pblock(ex
, ext_pblock(&orig_ex
));
2774 ext4_ext_mark_uninitialized(ex
);
2775 ext4_ext_dirty(handle
, inode
, path
+ depth
);
2780 * Block allocation/map/preallocation routine for extents based files
2783 * Need to be called with
2784 * down_read(&EXT4_I(inode)->i_data_sem) if not allocating file system block
2785 * (ie, create is zero). Otherwise down_write(&EXT4_I(inode)->i_data_sem)
2787 * return > 0, number of of blocks already mapped/allocated
2788 * if create == 0 and these are pre-allocated blocks
2789 * buffer head is unmapped
2790 * otherwise blocks are mapped
2792 * return = 0, if plain look up failed (blocks have not been allocated)
2793 * buffer head is unmapped
2795 * return < 0, error case.
2797 int ext4_ext_get_blocks(handle_t
*handle
, struct inode
*inode
,
2799 unsigned int max_blocks
, struct buffer_head
*bh_result
,
2802 struct ext4_ext_path
*path
= NULL
;
2803 struct ext4_extent_header
*eh
;
2804 struct ext4_extent newex
, *ex
;
2805 ext4_fsblk_t newblock
;
2806 int err
= 0, depth
, ret
, cache_type
;
2807 unsigned int allocated
= 0;
2808 struct ext4_allocation_request ar
;
2810 __clear_bit(BH_New
, &bh_result
->b_state
);
2811 ext_debug("blocks %u/%u requested for inode %lu\n",
2812 iblock
, max_blocks
, inode
->i_ino
);
2814 /* check in cache */
2815 cache_type
= ext4_ext_in_cache(inode
, iblock
, &newex
);
2817 if (cache_type
== EXT4_EXT_CACHE_GAP
) {
2818 if ((flags
& EXT4_GET_BLOCKS_CREATE
) == 0) {
2820 * block isn't allocated yet and
2821 * user doesn't want to allocate it
2825 /* we should allocate requested block */
2826 } else if (cache_type
== EXT4_EXT_CACHE_EXTENT
) {
2827 /* block is already allocated */
2829 - le32_to_cpu(newex
.ee_block
)
2830 + ext_pblock(&newex
);
2831 /* number of remaining blocks in the extent */
2832 allocated
= ext4_ext_get_actual_len(&newex
) -
2833 (iblock
- le32_to_cpu(newex
.ee_block
));
2840 /* find extent for this block */
2841 path
= ext4_ext_find_extent(inode
, iblock
, NULL
);
2843 err
= PTR_ERR(path
);
2848 depth
= ext_depth(inode
);
2851 * consistent leaf must not be empty;
2852 * this situation is possible, though, _during_ tree modification;
2853 * this is why assert can't be put in ext4_ext_find_extent()
2855 BUG_ON(path
[depth
].p_ext
== NULL
&& depth
!= 0);
2856 eh
= path
[depth
].p_hdr
;
2858 ex
= path
[depth
].p_ext
;
2860 ext4_lblk_t ee_block
= le32_to_cpu(ex
->ee_block
);
2861 ext4_fsblk_t ee_start
= ext_pblock(ex
);
2862 unsigned short ee_len
;
2865 * Uninitialized extents are treated as holes, except that
2866 * we split out initialized portions during a write.
2868 ee_len
= ext4_ext_get_actual_len(ex
);
2869 /* if found extent covers block, simply return it */
2870 if (iblock
>= ee_block
&& iblock
< ee_block
+ ee_len
) {
2871 newblock
= iblock
- ee_block
+ ee_start
;
2872 /* number of remaining blocks in the extent */
2873 allocated
= ee_len
- (iblock
- ee_block
);
2874 ext_debug("%u fit into %u:%d -> %llu\n", iblock
,
2875 ee_block
, ee_len
, newblock
);
2877 /* Do not put uninitialized extent in the cache */
2878 if (!ext4_ext_is_uninitialized(ex
)) {
2879 ext4_ext_put_in_cache(inode
, ee_block
,
2881 EXT4_EXT_CACHE_EXTENT
);
2884 if (flags
& EXT4_GET_BLOCKS_UNINIT_EXT
)
2886 if ((flags
& EXT4_GET_BLOCKS_CREATE
) == 0) {
2887 if (allocated
> max_blocks
)
2888 allocated
= max_blocks
;
2890 * We have blocks reserved already. We
2891 * return allocated blocks so that delalloc
2892 * won't do block reservation for us. But
2893 * the buffer head will be unmapped so that
2894 * a read from the block returns 0s.
2896 set_buffer_unwritten(bh_result
);
2897 bh_result
->b_bdev
= inode
->i_sb
->s_bdev
;
2898 bh_result
->b_blocknr
= newblock
;
2902 ret
= ext4_ext_convert_to_initialized(handle
, inode
,
2915 * requested block isn't allocated yet;
2916 * we couldn't try to create block if create flag is zero
2918 if ((flags
& EXT4_GET_BLOCKS_CREATE
) == 0) {
2920 * put just found gap into cache to speed up
2921 * subsequent requests
2923 ext4_ext_put_gap_in_cache(inode
, path
, iblock
);
2927 * Okay, we need to do block allocation.
2930 /* find neighbour allocated blocks */
2932 err
= ext4_ext_search_left(inode
, path
, &ar
.lleft
, &ar
.pleft
);
2936 err
= ext4_ext_search_right(inode
, path
, &ar
.lright
, &ar
.pright
);
2941 * See if request is beyond maximum number of blocks we can have in
2942 * a single extent. For an initialized extent this limit is
2943 * EXT_INIT_MAX_LEN and for an uninitialized extent this limit is
2944 * EXT_UNINIT_MAX_LEN.
2946 if (max_blocks
> EXT_INIT_MAX_LEN
&&
2947 !(flags
& EXT4_GET_BLOCKS_UNINIT_EXT
))
2948 max_blocks
= EXT_INIT_MAX_LEN
;
2949 else if (max_blocks
> EXT_UNINIT_MAX_LEN
&&
2950 (flags
& EXT4_GET_BLOCKS_UNINIT_EXT
))
2951 max_blocks
= EXT_UNINIT_MAX_LEN
;
2953 /* Check if we can really insert (iblock)::(iblock+max_blocks) extent */
2954 newex
.ee_block
= cpu_to_le32(iblock
);
2955 newex
.ee_len
= cpu_to_le16(max_blocks
);
2956 err
= ext4_ext_check_overlap(inode
, &newex
, path
);
2958 allocated
= ext4_ext_get_actual_len(&newex
);
2960 allocated
= max_blocks
;
2962 /* allocate new block */
2964 ar
.goal
= ext4_ext_find_goal(inode
, path
, iblock
);
2965 ar
.logical
= iblock
;
2967 if (S_ISREG(inode
->i_mode
))
2968 ar
.flags
= EXT4_MB_HINT_DATA
;
2970 /* disable in-core preallocation for non-regular files */
2972 newblock
= ext4_mb_new_blocks(handle
, &ar
, &err
);
2975 ext_debug("allocate new block: goal %llu, found %llu/%u\n",
2976 ar
.goal
, newblock
, allocated
);
2978 /* try to insert new extent into found leaf and return */
2979 ext4_ext_store_pblock(&newex
, newblock
);
2980 newex
.ee_len
= cpu_to_le16(ar
.len
);
2981 if (flags
& EXT4_GET_BLOCKS_UNINIT_EXT
) /* Mark uninitialized */
2982 ext4_ext_mark_uninitialized(&newex
);
2983 err
= ext4_ext_insert_extent(handle
, inode
, path
, &newex
);
2985 /* free data blocks we just allocated */
2986 /* not a good idea to call discard here directly,
2987 * but otherwise we'd need to call it every free() */
2988 ext4_discard_preallocations(inode
);
2989 ext4_free_blocks(handle
, inode
, ext_pblock(&newex
),
2990 ext4_ext_get_actual_len(&newex
), 0);
2994 /* previous routine could use block we allocated */
2995 newblock
= ext_pblock(&newex
);
2996 allocated
= ext4_ext_get_actual_len(&newex
);
2998 set_buffer_new(bh_result
);
3000 /* Cache only when it is _not_ an uninitialized extent */
3001 if ((flags
& EXT4_GET_BLOCKS_UNINIT_EXT
) == 0)
3002 ext4_ext_put_in_cache(inode
, iblock
, allocated
, newblock
,
3003 EXT4_EXT_CACHE_EXTENT
);
3005 if (allocated
> max_blocks
)
3006 allocated
= max_blocks
;
3007 ext4_ext_show_leaf(inode
, path
);
3008 set_buffer_mapped(bh_result
);
3009 bh_result
->b_bdev
= inode
->i_sb
->s_bdev
;
3010 bh_result
->b_blocknr
= newblock
;
3013 ext4_ext_drop_refs(path
);
3016 return err
? err
: allocated
;
3019 void ext4_ext_truncate(struct inode
*inode
)
3021 struct address_space
*mapping
= inode
->i_mapping
;
3022 struct super_block
*sb
= inode
->i_sb
;
3023 ext4_lblk_t last_block
;
3028 * probably first extent we're gonna free will be last in block
3030 err
= ext4_writepage_trans_blocks(inode
);
3031 handle
= ext4_journal_start(inode
, err
);
3035 if (inode
->i_size
& (sb
->s_blocksize
- 1))
3036 ext4_block_truncate_page(handle
, mapping
, inode
->i_size
);
3038 if (ext4_orphan_add(handle
, inode
))
3041 down_write(&EXT4_I(inode
)->i_data_sem
);
3042 ext4_ext_invalidate_cache(inode
);
3044 ext4_discard_preallocations(inode
);
3047 * TODO: optimization is possible here.
3048 * Probably we need not scan at all,
3049 * because page truncation is enough.
3052 /* we have to know where to truncate from in crash case */
3053 EXT4_I(inode
)->i_disksize
= inode
->i_size
;
3054 ext4_mark_inode_dirty(handle
, inode
);
3056 last_block
= (inode
->i_size
+ sb
->s_blocksize
- 1)
3057 >> EXT4_BLOCK_SIZE_BITS(sb
);
3058 err
= ext4_ext_remove_space(inode
, last_block
);
3060 /* In a multi-transaction truncate, we only make the final
3061 * transaction synchronous.
3064 ext4_handle_sync(handle
);
3067 up_write(&EXT4_I(inode
)->i_data_sem
);
3069 * If this was a simple ftruncate() and the file will remain alive,
3070 * then we need to clear up the orphan record which we created above.
3071 * However, if this was a real unlink then we were called by
3072 * ext4_delete_inode(), and we allow that function to clean up the
3073 * orphan info for us.
3076 ext4_orphan_del(handle
, inode
);
3078 inode
->i_mtime
= inode
->i_ctime
= ext4_current_time(inode
);
3079 ext4_mark_inode_dirty(handle
, inode
);
3080 ext4_journal_stop(handle
);
3083 static void ext4_falloc_update_inode(struct inode
*inode
,
3084 int mode
, loff_t new_size
, int update_ctime
)
3086 struct timespec now
;
3089 now
= current_fs_time(inode
->i_sb
);
3090 if (!timespec_equal(&inode
->i_ctime
, &now
))
3091 inode
->i_ctime
= now
;
3094 * Update only when preallocation was requested beyond
3097 if (!(mode
& FALLOC_FL_KEEP_SIZE
)) {
3098 if (new_size
> i_size_read(inode
))
3099 i_size_write(inode
, new_size
);
3100 if (new_size
> EXT4_I(inode
)->i_disksize
)
3101 ext4_update_i_disksize(inode
, new_size
);
3107 * preallocate space for a file. This implements ext4's fallocate inode
3108 * operation, which gets called from sys_fallocate system call.
3109 * For block-mapped files, posix_fallocate should fall back to the method
3110 * of writing zeroes to the required new blocks (the same behavior which is
3111 * expected for file systems which do not support fallocate() system call).
3113 long ext4_fallocate(struct inode
*inode
, int mode
, loff_t offset
, loff_t len
)
3118 unsigned int max_blocks
;
3122 struct buffer_head map_bh
;
3123 unsigned int credits
, blkbits
= inode
->i_blkbits
;
3126 * currently supporting (pre)allocate mode for extent-based
3129 if (!(EXT4_I(inode
)->i_flags
& EXT4_EXTENTS_FL
))
3132 /* preallocation to directories is currently not supported */
3133 if (S_ISDIR(inode
->i_mode
))
3136 block
= offset
>> blkbits
;
3138 * We can't just convert len to max_blocks because
3139 * If blocksize = 4096 offset = 3072 and len = 2048
3141 max_blocks
= (EXT4_BLOCK_ALIGN(len
+ offset
, blkbits
) >> blkbits
)
3144 * credits to insert 1 extent into extent tree
3146 credits
= ext4_chunk_trans_blocks(inode
, max_blocks
);
3147 mutex_lock(&inode
->i_mutex
);
3149 while (ret
>= 0 && ret
< max_blocks
) {
3150 block
= block
+ ret
;
3151 max_blocks
= max_blocks
- ret
;
3152 handle
= ext4_journal_start(inode
, credits
);
3153 if (IS_ERR(handle
)) {
3154 ret
= PTR_ERR(handle
);
3158 ret
= ext4_get_blocks(handle
, inode
, block
,
3159 max_blocks
, &map_bh
,
3160 EXT4_GET_BLOCKS_CREATE_UNINIT_EXT
);
3164 printk(KERN_ERR
"%s: ext4_ext_get_blocks "
3165 "returned error inode#%lu, block=%u, "
3166 "max_blocks=%u", __func__
,
3167 inode
->i_ino
, block
, max_blocks
);
3169 ext4_mark_inode_dirty(handle
, inode
);
3170 ret2
= ext4_journal_stop(handle
);
3173 if ((block
+ ret
) >= (EXT4_BLOCK_ALIGN(offset
+ len
,
3174 blkbits
) >> blkbits
))
3175 new_size
= offset
+ len
;
3177 new_size
= (block
+ ret
) << blkbits
;
3179 ext4_falloc_update_inode(inode
, mode
, new_size
,
3180 buffer_new(&map_bh
));
3181 ext4_mark_inode_dirty(handle
, inode
);
3182 ret2
= ext4_journal_stop(handle
);
3186 if (ret
== -ENOSPC
&&
3187 ext4_should_retry_alloc(inode
->i_sb
, &retries
)) {
3191 mutex_unlock(&inode
->i_mutex
);
3192 return ret
> 0 ? ret2
: ret
;
3196 * Callback function called for each extent to gather FIEMAP information.
3198 static int ext4_ext_fiemap_cb(struct inode
*inode
, struct ext4_ext_path
*path
,
3199 struct ext4_ext_cache
*newex
, struct ext4_extent
*ex
,
3202 struct fiemap_extent_info
*fieinfo
= data
;
3203 unsigned char blksize_bits
= inode
->i_sb
->s_blocksize_bits
;
3210 logical
= (__u64
)newex
->ec_block
<< blksize_bits
;
3212 if (newex
->ec_type
== EXT4_EXT_CACHE_GAP
) {
3215 struct buffer_head
*bh
= NULL
;
3217 offset
= logical
>> PAGE_SHIFT
;
3218 page
= find_get_page(inode
->i_mapping
, offset
);
3219 if (!page
|| !page_has_buffers(page
))
3220 return EXT_CONTINUE
;
3222 bh
= page_buffers(page
);
3225 return EXT_CONTINUE
;
3227 if (buffer_delay(bh
)) {
3228 flags
|= FIEMAP_EXTENT_DELALLOC
;
3229 page_cache_release(page
);
3231 page_cache_release(page
);
3232 return EXT_CONTINUE
;
3236 physical
= (__u64
)newex
->ec_start
<< blksize_bits
;
3237 length
= (__u64
)newex
->ec_len
<< blksize_bits
;
3239 if (ex
&& ext4_ext_is_uninitialized(ex
))
3240 flags
|= FIEMAP_EXTENT_UNWRITTEN
;
3243 * If this extent reaches EXT_MAX_BLOCK, it must be last.
3245 * Or if ext4_ext_next_allocated_block is EXT_MAX_BLOCK,
3246 * this also indicates no more allocated blocks.
3248 * XXX this might miss a single-block extent at EXT_MAX_BLOCK
3250 if (ext4_ext_next_allocated_block(path
) == EXT_MAX_BLOCK
||
3251 newex
->ec_block
+ newex
->ec_len
- 1 == EXT_MAX_BLOCK
) {
3252 loff_t size
= i_size_read(inode
);
3253 loff_t bs
= EXT4_BLOCK_SIZE(inode
->i_sb
);
3255 flags
|= FIEMAP_EXTENT_LAST
;
3256 if ((flags
& FIEMAP_EXTENT_DELALLOC
) &&
3257 logical
+length
> size
)
3258 length
= (size
- logical
+ bs
- 1) & ~(bs
-1);
3261 error
= fiemap_fill_next_extent(fieinfo
, logical
, physical
,
3268 return EXT_CONTINUE
;
3271 /* fiemap flags we can handle specified here */
3272 #define EXT4_FIEMAP_FLAGS (FIEMAP_FLAG_SYNC|FIEMAP_FLAG_XATTR)
3274 static int ext4_xattr_fiemap(struct inode
*inode
,
3275 struct fiemap_extent_info
*fieinfo
)
3279 __u32 flags
= FIEMAP_EXTENT_LAST
;
3280 int blockbits
= inode
->i_sb
->s_blocksize_bits
;
3284 if (EXT4_I(inode
)->i_state
& EXT4_STATE_XATTR
) {
3285 struct ext4_iloc iloc
;
3286 int offset
; /* offset of xattr in inode */
3288 error
= ext4_get_inode_loc(inode
, &iloc
);
3291 physical
= iloc
.bh
->b_blocknr
<< blockbits
;
3292 offset
= EXT4_GOOD_OLD_INODE_SIZE
+
3293 EXT4_I(inode
)->i_extra_isize
;
3295 length
= EXT4_SB(inode
->i_sb
)->s_inode_size
- offset
;
3296 flags
|= FIEMAP_EXTENT_DATA_INLINE
;
3297 } else { /* external block */
3298 physical
= EXT4_I(inode
)->i_file_acl
<< blockbits
;
3299 length
= inode
->i_sb
->s_blocksize
;
3303 error
= fiemap_fill_next_extent(fieinfo
, 0, physical
,
3305 return (error
< 0 ? error
: 0);
3308 int ext4_fiemap(struct inode
*inode
, struct fiemap_extent_info
*fieinfo
,
3309 __u64 start
, __u64 len
)
3311 ext4_lblk_t start_blk
;
3312 ext4_lblk_t len_blks
;
3315 /* fallback to generic here if not in extents fmt */
3316 if (!(EXT4_I(inode
)->i_flags
& EXT4_EXTENTS_FL
))
3317 return generic_block_fiemap(inode
, fieinfo
, start
, len
,
3320 if (fiemap_check_flags(fieinfo
, EXT4_FIEMAP_FLAGS
))
3323 if (fieinfo
->fi_flags
& FIEMAP_FLAG_XATTR
) {
3324 error
= ext4_xattr_fiemap(inode
, fieinfo
);
3326 start_blk
= start
>> inode
->i_sb
->s_blocksize_bits
;
3327 len_blks
= len
>> inode
->i_sb
->s_blocksize_bits
;
3330 * Walk the extent tree gathering extent information.
3331 * ext4_ext_fiemap_cb will push extents back to user.
3333 down_read(&EXT4_I(inode
)->i_data_sem
);
3334 error
= ext4_ext_walk_space(inode
, start_blk
, len_blks
,
3335 ext4_ext_fiemap_cb
, fieinfo
);
3336 up_read(&EXT4_I(inode
)->i_data_sem
);