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
= ext4_group_first_block_no(inode
->i_sb
, block_group
);
199 last_block
= ext4_blocks_count(EXT4_SB(inode
->i_sb
)->s_es
) - 1;
202 * If we are doing delayed allocation, we don't need take
203 * colour into account.
205 if (test_opt(inode
->i_sb
, DELALLOC
))
208 if (bg_start
+ EXT4_BLOCKS_PER_GROUP(inode
->i_sb
) <= last_block
)
209 colour
= (current
->pid
% 16) *
210 (EXT4_BLOCKS_PER_GROUP(inode
->i_sb
) / 16);
212 colour
= (current
->pid
% 16) * ((last_block
- bg_start
) / 16);
213 return bg_start
+ colour
+ block
;
217 * Allocation for a meta data block
220 ext4_ext_new_meta_block(handle_t
*handle
, struct inode
*inode
,
221 struct ext4_ext_path
*path
,
222 struct ext4_extent
*ex
, int *err
)
224 ext4_fsblk_t goal
, newblock
;
226 goal
= ext4_ext_find_goal(inode
, path
, le32_to_cpu(ex
->ee_block
));
227 newblock
= ext4_new_meta_blocks(handle
, inode
, goal
, NULL
, err
);
231 static inline int ext4_ext_space_block(struct inode
*inode
, int check
)
235 size
= (inode
->i_sb
->s_blocksize
- sizeof(struct ext4_extent_header
))
236 / sizeof(struct ext4_extent
);
238 #ifdef AGGRESSIVE_TEST
246 static inline int ext4_ext_space_block_idx(struct inode
*inode
, int check
)
250 size
= (inode
->i_sb
->s_blocksize
- sizeof(struct ext4_extent_header
))
251 / sizeof(struct ext4_extent_idx
);
253 #ifdef AGGRESSIVE_TEST
261 static inline int ext4_ext_space_root(struct inode
*inode
, int check
)
265 size
= sizeof(EXT4_I(inode
)->i_data
);
266 size
-= sizeof(struct ext4_extent_header
);
267 size
/= sizeof(struct ext4_extent
);
269 #ifdef AGGRESSIVE_TEST
277 static inline int ext4_ext_space_root_idx(struct inode
*inode
, int check
)
281 size
= sizeof(EXT4_I(inode
)->i_data
);
282 size
-= sizeof(struct ext4_extent_header
);
283 size
/= sizeof(struct ext4_extent_idx
);
285 #ifdef AGGRESSIVE_TEST
294 * Calculate the number of metadata blocks needed
295 * to allocate @blocks
296 * Worse case is one block per extent
298 int ext4_ext_calc_metadata_amount(struct inode
*inode
, sector_t lblock
)
300 struct ext4_inode_info
*ei
= EXT4_I(inode
);
303 idxs
= ((inode
->i_sb
->s_blocksize
- sizeof(struct ext4_extent_header
))
304 / sizeof(struct ext4_extent_idx
));
307 * If the new delayed allocation block is contiguous with the
308 * previous da block, it can share index blocks with the
309 * previous block, so we only need to allocate a new index
310 * block every idxs leaf blocks. At ldxs**2 blocks, we need
311 * an additional index block, and at ldxs**3 blocks, yet
312 * another index blocks.
314 if (ei
->i_da_metadata_calc_len
&&
315 ei
->i_da_metadata_calc_last_lblock
+1 == lblock
) {
316 if ((ei
->i_da_metadata_calc_len
% idxs
) == 0)
318 if ((ei
->i_da_metadata_calc_len
% (idxs
*idxs
)) == 0)
320 if ((ei
->i_da_metadata_calc_len
% (idxs
*idxs
*idxs
)) == 0) {
322 ei
->i_da_metadata_calc_len
= 0;
324 ei
->i_da_metadata_calc_len
++;
325 ei
->i_da_metadata_calc_last_lblock
++;
330 * In the worst case we need a new set of index blocks at
331 * every level of the inode's extent tree.
333 ei
->i_da_metadata_calc_len
= 1;
334 ei
->i_da_metadata_calc_last_lblock
= lblock
;
335 return ext_depth(inode
) + 1;
339 ext4_ext_max_entries(struct inode
*inode
, int depth
)
343 if (depth
== ext_depth(inode
)) {
345 max
= ext4_ext_space_root(inode
, 1);
347 max
= ext4_ext_space_root_idx(inode
, 1);
350 max
= ext4_ext_space_block(inode
, 1);
352 max
= ext4_ext_space_block_idx(inode
, 1);
358 static int ext4_valid_extent(struct inode
*inode
, struct ext4_extent
*ext
)
360 ext4_fsblk_t block
= ext_pblock(ext
);
361 int len
= ext4_ext_get_actual_len(ext
);
363 return ext4_data_block_valid(EXT4_SB(inode
->i_sb
), block
, len
);
366 static int ext4_valid_extent_idx(struct inode
*inode
,
367 struct ext4_extent_idx
*ext_idx
)
369 ext4_fsblk_t block
= idx_pblock(ext_idx
);
371 return ext4_data_block_valid(EXT4_SB(inode
->i_sb
), block
, 1);
374 static int ext4_valid_extent_entries(struct inode
*inode
,
375 struct ext4_extent_header
*eh
,
378 struct ext4_extent
*ext
;
379 struct ext4_extent_idx
*ext_idx
;
380 unsigned short entries
;
381 if (eh
->eh_entries
== 0)
384 entries
= le16_to_cpu(eh
->eh_entries
);
388 ext
= EXT_FIRST_EXTENT(eh
);
390 if (!ext4_valid_extent(inode
, ext
))
396 ext_idx
= EXT_FIRST_INDEX(eh
);
398 if (!ext4_valid_extent_idx(inode
, ext_idx
))
407 static int __ext4_ext_check(const char *function
, struct inode
*inode
,
408 struct ext4_extent_header
*eh
,
411 const char *error_msg
;
414 if (unlikely(eh
->eh_magic
!= EXT4_EXT_MAGIC
)) {
415 error_msg
= "invalid magic";
418 if (unlikely(le16_to_cpu(eh
->eh_depth
) != depth
)) {
419 error_msg
= "unexpected eh_depth";
422 if (unlikely(eh
->eh_max
== 0)) {
423 error_msg
= "invalid eh_max";
426 max
= ext4_ext_max_entries(inode
, depth
);
427 if (unlikely(le16_to_cpu(eh
->eh_max
) > max
)) {
428 error_msg
= "too large eh_max";
431 if (unlikely(le16_to_cpu(eh
->eh_entries
) > le16_to_cpu(eh
->eh_max
))) {
432 error_msg
= "invalid eh_entries";
435 if (!ext4_valid_extent_entries(inode
, eh
, depth
)) {
436 error_msg
= "invalid extent entries";
442 __ext4_error(inode
->i_sb
, function
,
443 "bad header/extent in inode #%lu: %s - magic %x, "
444 "entries %u, max %u(%u), depth %u(%u)",
445 inode
->i_ino
, error_msg
, le16_to_cpu(eh
->eh_magic
),
446 le16_to_cpu(eh
->eh_entries
), le16_to_cpu(eh
->eh_max
),
447 max
, le16_to_cpu(eh
->eh_depth
), depth
);
452 #define ext4_ext_check(inode, eh, depth) \
453 __ext4_ext_check(__func__, inode, eh, depth)
455 int ext4_ext_check_inode(struct inode
*inode
)
457 return ext4_ext_check(inode
, ext_inode_hdr(inode
), ext_depth(inode
));
461 static void ext4_ext_show_path(struct inode
*inode
, struct ext4_ext_path
*path
)
463 int k
, l
= path
->p_depth
;
466 for (k
= 0; k
<= l
; k
++, path
++) {
468 ext_debug(" %d->%llu", le32_to_cpu(path
->p_idx
->ei_block
),
469 idx_pblock(path
->p_idx
));
470 } else if (path
->p_ext
) {
471 ext_debug(" %d:[%d]%d:%llu ",
472 le32_to_cpu(path
->p_ext
->ee_block
),
473 ext4_ext_is_uninitialized(path
->p_ext
),
474 ext4_ext_get_actual_len(path
->p_ext
),
475 ext_pblock(path
->p_ext
));
482 static void ext4_ext_show_leaf(struct inode
*inode
, struct ext4_ext_path
*path
)
484 int depth
= ext_depth(inode
);
485 struct ext4_extent_header
*eh
;
486 struct ext4_extent
*ex
;
492 eh
= path
[depth
].p_hdr
;
493 ex
= EXT_FIRST_EXTENT(eh
);
495 ext_debug("Displaying leaf extents for inode %lu\n", inode
->i_ino
);
497 for (i
= 0; i
< le16_to_cpu(eh
->eh_entries
); i
++, ex
++) {
498 ext_debug("%d:[%d]%d:%llu ", le32_to_cpu(ex
->ee_block
),
499 ext4_ext_is_uninitialized(ex
),
500 ext4_ext_get_actual_len(ex
), ext_pblock(ex
));
505 #define ext4_ext_show_path(inode, path)
506 #define ext4_ext_show_leaf(inode, path)
509 void ext4_ext_drop_refs(struct ext4_ext_path
*path
)
511 int depth
= path
->p_depth
;
514 for (i
= 0; i
<= depth
; i
++, path
++)
522 * ext4_ext_binsearch_idx:
523 * binary search for the closest index of the given block
524 * the header must be checked before calling this
527 ext4_ext_binsearch_idx(struct inode
*inode
,
528 struct ext4_ext_path
*path
, ext4_lblk_t block
)
530 struct ext4_extent_header
*eh
= path
->p_hdr
;
531 struct ext4_extent_idx
*r
, *l
, *m
;
534 ext_debug("binsearch for %u(idx): ", block
);
536 l
= EXT_FIRST_INDEX(eh
) + 1;
537 r
= EXT_LAST_INDEX(eh
);
540 if (block
< le32_to_cpu(m
->ei_block
))
544 ext_debug("%p(%u):%p(%u):%p(%u) ", l
, le32_to_cpu(l
->ei_block
),
545 m
, le32_to_cpu(m
->ei_block
),
546 r
, le32_to_cpu(r
->ei_block
));
550 ext_debug(" -> %d->%lld ", le32_to_cpu(path
->p_idx
->ei_block
),
551 idx_pblock(path
->p_idx
));
553 #ifdef CHECK_BINSEARCH
555 struct ext4_extent_idx
*chix
, *ix
;
558 chix
= ix
= EXT_FIRST_INDEX(eh
);
559 for (k
= 0; k
< le16_to_cpu(eh
->eh_entries
); k
++, ix
++) {
561 le32_to_cpu(ix
->ei_block
) <= le32_to_cpu(ix
[-1].ei_block
)) {
562 printk(KERN_DEBUG
"k=%d, ix=0x%p, "
564 ix
, EXT_FIRST_INDEX(eh
));
565 printk(KERN_DEBUG
"%u <= %u\n",
566 le32_to_cpu(ix
->ei_block
),
567 le32_to_cpu(ix
[-1].ei_block
));
569 BUG_ON(k
&& le32_to_cpu(ix
->ei_block
)
570 <= le32_to_cpu(ix
[-1].ei_block
));
571 if (block
< le32_to_cpu(ix
->ei_block
))
575 BUG_ON(chix
!= path
->p_idx
);
582 * ext4_ext_binsearch:
583 * binary search for closest extent of the given block
584 * the header must be checked before calling this
587 ext4_ext_binsearch(struct inode
*inode
,
588 struct ext4_ext_path
*path
, ext4_lblk_t block
)
590 struct ext4_extent_header
*eh
= path
->p_hdr
;
591 struct ext4_extent
*r
, *l
, *m
;
593 if (eh
->eh_entries
== 0) {
595 * this leaf is empty:
596 * we get such a leaf in split/add case
601 ext_debug("binsearch for %u: ", block
);
603 l
= EXT_FIRST_EXTENT(eh
) + 1;
604 r
= EXT_LAST_EXTENT(eh
);
608 if (block
< le32_to_cpu(m
->ee_block
))
612 ext_debug("%p(%u):%p(%u):%p(%u) ", l
, le32_to_cpu(l
->ee_block
),
613 m
, le32_to_cpu(m
->ee_block
),
614 r
, le32_to_cpu(r
->ee_block
));
618 ext_debug(" -> %d:%llu:[%d]%d ",
619 le32_to_cpu(path
->p_ext
->ee_block
),
620 ext_pblock(path
->p_ext
),
621 ext4_ext_is_uninitialized(path
->p_ext
),
622 ext4_ext_get_actual_len(path
->p_ext
));
624 #ifdef CHECK_BINSEARCH
626 struct ext4_extent
*chex
, *ex
;
629 chex
= ex
= EXT_FIRST_EXTENT(eh
);
630 for (k
= 0; k
< le16_to_cpu(eh
->eh_entries
); k
++, ex
++) {
631 BUG_ON(k
&& le32_to_cpu(ex
->ee_block
)
632 <= le32_to_cpu(ex
[-1].ee_block
));
633 if (block
< le32_to_cpu(ex
->ee_block
))
637 BUG_ON(chex
!= path
->p_ext
);
643 int ext4_ext_tree_init(handle_t
*handle
, struct inode
*inode
)
645 struct ext4_extent_header
*eh
;
647 eh
= ext_inode_hdr(inode
);
650 eh
->eh_magic
= EXT4_EXT_MAGIC
;
651 eh
->eh_max
= cpu_to_le16(ext4_ext_space_root(inode
, 0));
652 ext4_mark_inode_dirty(handle
, inode
);
653 ext4_ext_invalidate_cache(inode
);
657 struct ext4_ext_path
*
658 ext4_ext_find_extent(struct inode
*inode
, ext4_lblk_t block
,
659 struct ext4_ext_path
*path
)
661 struct ext4_extent_header
*eh
;
662 struct buffer_head
*bh
;
663 short int depth
, i
, ppos
= 0, alloc
= 0;
665 eh
= ext_inode_hdr(inode
);
666 depth
= ext_depth(inode
);
668 /* account possible depth increase */
670 path
= kzalloc(sizeof(struct ext4_ext_path
) * (depth
+ 2),
673 return ERR_PTR(-ENOMEM
);
680 /* walk through the tree */
682 int need_to_validate
= 0;
684 ext_debug("depth %d: num %d, max %d\n",
685 ppos
, le16_to_cpu(eh
->eh_entries
), le16_to_cpu(eh
->eh_max
));
687 ext4_ext_binsearch_idx(inode
, path
+ ppos
, block
);
688 path
[ppos
].p_block
= idx_pblock(path
[ppos
].p_idx
);
689 path
[ppos
].p_depth
= i
;
690 path
[ppos
].p_ext
= NULL
;
692 bh
= sb_getblk(inode
->i_sb
, path
[ppos
].p_block
);
695 if (!bh_uptodate_or_lock(bh
)) {
696 if (bh_submit_read(bh
) < 0) {
700 /* validate the extent entries */
701 need_to_validate
= 1;
703 eh
= ext_block_hdr(bh
);
705 if (unlikely(ppos
> depth
)) {
707 EXT4_ERROR_INODE(inode
,
708 "ppos %d > depth %d", ppos
, depth
);
711 path
[ppos
].p_bh
= bh
;
712 path
[ppos
].p_hdr
= eh
;
715 if (need_to_validate
&& ext4_ext_check(inode
, eh
, i
))
719 path
[ppos
].p_depth
= i
;
720 path
[ppos
].p_ext
= NULL
;
721 path
[ppos
].p_idx
= NULL
;
724 ext4_ext_binsearch(inode
, path
+ ppos
, block
);
725 /* if not an empty leaf */
726 if (path
[ppos
].p_ext
)
727 path
[ppos
].p_block
= ext_pblock(path
[ppos
].p_ext
);
729 ext4_ext_show_path(inode
, path
);
734 ext4_ext_drop_refs(path
);
737 return ERR_PTR(-EIO
);
741 * ext4_ext_insert_index:
742 * insert new index [@logical;@ptr] into the block at @curp;
743 * check where to insert: before @curp or after @curp
745 int ext4_ext_insert_index(handle_t
*handle
, struct inode
*inode
,
746 struct ext4_ext_path
*curp
,
747 int logical
, ext4_fsblk_t ptr
)
749 struct ext4_extent_idx
*ix
;
752 err
= ext4_ext_get_access(handle
, inode
, curp
);
756 if (unlikely(logical
== le32_to_cpu(curp
->p_idx
->ei_block
))) {
757 EXT4_ERROR_INODE(inode
,
758 "logical %d == ei_block %d!",
759 logical
, le32_to_cpu(curp
->p_idx
->ei_block
));
762 len
= EXT_MAX_INDEX(curp
->p_hdr
) - curp
->p_idx
;
763 if (logical
> le32_to_cpu(curp
->p_idx
->ei_block
)) {
765 if (curp
->p_idx
!= EXT_LAST_INDEX(curp
->p_hdr
)) {
766 len
= (len
- 1) * sizeof(struct ext4_extent_idx
);
767 len
= len
< 0 ? 0 : len
;
768 ext_debug("insert new index %d after: %llu. "
769 "move %d from 0x%p to 0x%p\n",
771 (curp
->p_idx
+ 1), (curp
->p_idx
+ 2));
772 memmove(curp
->p_idx
+ 2, curp
->p_idx
+ 1, len
);
774 ix
= curp
->p_idx
+ 1;
777 len
= len
* sizeof(struct ext4_extent_idx
);
778 len
= len
< 0 ? 0 : len
;
779 ext_debug("insert new index %d before: %llu. "
780 "move %d from 0x%p to 0x%p\n",
782 curp
->p_idx
, (curp
->p_idx
+ 1));
783 memmove(curp
->p_idx
+ 1, curp
->p_idx
, len
);
787 ix
->ei_block
= cpu_to_le32(logical
);
788 ext4_idx_store_pblock(ix
, ptr
);
789 le16_add_cpu(&curp
->p_hdr
->eh_entries
, 1);
791 if (unlikely(le16_to_cpu(curp
->p_hdr
->eh_entries
)
792 > le16_to_cpu(curp
->p_hdr
->eh_max
))) {
793 EXT4_ERROR_INODE(inode
,
794 "logical %d == ei_block %d!",
795 logical
, le32_to_cpu(curp
->p_idx
->ei_block
));
798 if (unlikely(ix
> EXT_LAST_INDEX(curp
->p_hdr
))) {
799 EXT4_ERROR_INODE(inode
, "ix > EXT_LAST_INDEX!");
803 err
= ext4_ext_dirty(handle
, inode
, curp
);
804 ext4_std_error(inode
->i_sb
, err
);
811 * inserts new subtree into the path, using free index entry
813 * - allocates all needed blocks (new leaf and all intermediate index blocks)
814 * - makes decision where to split
815 * - moves remaining extents and index entries (right to the split point)
816 * into the newly allocated blocks
817 * - initializes subtree
819 static int ext4_ext_split(handle_t
*handle
, struct inode
*inode
,
820 struct ext4_ext_path
*path
,
821 struct ext4_extent
*newext
, int at
)
823 struct buffer_head
*bh
= NULL
;
824 int depth
= ext_depth(inode
);
825 struct ext4_extent_header
*neh
;
826 struct ext4_extent_idx
*fidx
;
827 struct ext4_extent
*ex
;
829 ext4_fsblk_t newblock
, oldblock
;
831 ext4_fsblk_t
*ablocks
= NULL
; /* array of allocated blocks */
834 /* make decision: where to split? */
835 /* FIXME: now decision is simplest: at current extent */
837 /* if current leaf will be split, then we should use
838 * border from split point */
839 if (unlikely(path
[depth
].p_ext
> EXT_MAX_EXTENT(path
[depth
].p_hdr
))) {
840 EXT4_ERROR_INODE(inode
, "p_ext > EXT_MAX_EXTENT!");
843 if (path
[depth
].p_ext
!= EXT_MAX_EXTENT(path
[depth
].p_hdr
)) {
844 border
= path
[depth
].p_ext
[1].ee_block
;
845 ext_debug("leaf will be split."
846 " next leaf starts at %d\n",
847 le32_to_cpu(border
));
849 border
= newext
->ee_block
;
850 ext_debug("leaf will be added."
851 " next leaf starts at %d\n",
852 le32_to_cpu(border
));
856 * If error occurs, then we break processing
857 * and mark filesystem read-only. index won't
858 * be inserted and tree will be in consistent
859 * state. Next mount will repair buffers too.
863 * Get array to track all allocated blocks.
864 * We need this to handle errors and free blocks
867 ablocks
= kzalloc(sizeof(ext4_fsblk_t
) * depth
, GFP_NOFS
);
871 /* allocate all needed blocks */
872 ext_debug("allocate %d blocks for indexes/leaf\n", depth
- at
);
873 for (a
= 0; a
< depth
- at
; a
++) {
874 newblock
= ext4_ext_new_meta_block(handle
, inode
, path
,
878 ablocks
[a
] = newblock
;
881 /* initialize new leaf */
882 newblock
= ablocks
[--a
];
883 if (unlikely(newblock
== 0)) {
884 EXT4_ERROR_INODE(inode
, "newblock == 0!");
888 bh
= sb_getblk(inode
->i_sb
, newblock
);
895 err
= ext4_journal_get_create_access(handle
, bh
);
899 neh
= ext_block_hdr(bh
);
901 neh
->eh_max
= cpu_to_le16(ext4_ext_space_block(inode
, 0));
902 neh
->eh_magic
= EXT4_EXT_MAGIC
;
904 ex
= EXT_FIRST_EXTENT(neh
);
906 /* move remainder of path[depth] to the new leaf */
907 if (unlikely(path
[depth
].p_hdr
->eh_entries
!=
908 path
[depth
].p_hdr
->eh_max
)) {
909 EXT4_ERROR_INODE(inode
, "eh_entries %d != eh_max %d!",
910 path
[depth
].p_hdr
->eh_entries
,
911 path
[depth
].p_hdr
->eh_max
);
915 /* start copy from next extent */
916 /* TODO: we could do it by single memmove */
919 while (path
[depth
].p_ext
<=
920 EXT_MAX_EXTENT(path
[depth
].p_hdr
)) {
921 ext_debug("move %d:%llu:[%d]%d in new leaf %llu\n",
922 le32_to_cpu(path
[depth
].p_ext
->ee_block
),
923 ext_pblock(path
[depth
].p_ext
),
924 ext4_ext_is_uninitialized(path
[depth
].p_ext
),
925 ext4_ext_get_actual_len(path
[depth
].p_ext
),
927 /*memmove(ex++, path[depth].p_ext++,
928 sizeof(struct ext4_extent));
934 memmove(ex
, path
[depth
].p_ext
-m
, sizeof(struct ext4_extent
)*m
);
935 le16_add_cpu(&neh
->eh_entries
, m
);
938 set_buffer_uptodate(bh
);
941 err
= ext4_handle_dirty_metadata(handle
, inode
, bh
);
947 /* correct old leaf */
949 err
= ext4_ext_get_access(handle
, inode
, path
+ depth
);
952 le16_add_cpu(&path
[depth
].p_hdr
->eh_entries
, -m
);
953 err
= ext4_ext_dirty(handle
, inode
, path
+ depth
);
959 /* create intermediate indexes */
961 if (unlikely(k
< 0)) {
962 EXT4_ERROR_INODE(inode
, "k %d < 0!", k
);
967 ext_debug("create %d intermediate indices\n", k
);
968 /* insert new index into current index block */
969 /* current depth stored in i var */
973 newblock
= ablocks
[--a
];
974 bh
= sb_getblk(inode
->i_sb
, newblock
);
981 err
= ext4_journal_get_create_access(handle
, bh
);
985 neh
= ext_block_hdr(bh
);
986 neh
->eh_entries
= cpu_to_le16(1);
987 neh
->eh_magic
= EXT4_EXT_MAGIC
;
988 neh
->eh_max
= cpu_to_le16(ext4_ext_space_block_idx(inode
, 0));
989 neh
->eh_depth
= cpu_to_le16(depth
- i
);
990 fidx
= EXT_FIRST_INDEX(neh
);
991 fidx
->ei_block
= border
;
992 ext4_idx_store_pblock(fidx
, oldblock
);
994 ext_debug("int.index at %d (block %llu): %u -> %llu\n",
995 i
, newblock
, le32_to_cpu(border
), oldblock
);
1000 ext_debug("cur 0x%p, last 0x%p\n", path
[i
].p_idx
,
1001 EXT_MAX_INDEX(path
[i
].p_hdr
));
1002 if (unlikely(EXT_MAX_INDEX(path
[i
].p_hdr
) !=
1003 EXT_LAST_INDEX(path
[i
].p_hdr
))) {
1004 EXT4_ERROR_INODE(inode
,
1005 "EXT_MAX_INDEX != EXT_LAST_INDEX ee_block %d!",
1006 le32_to_cpu(path
[i
].p_ext
->ee_block
));
1010 while (path
[i
].p_idx
<= EXT_MAX_INDEX(path
[i
].p_hdr
)) {
1011 ext_debug("%d: move %d:%llu in new index %llu\n", i
,
1012 le32_to_cpu(path
[i
].p_idx
->ei_block
),
1013 idx_pblock(path
[i
].p_idx
),
1015 /*memmove(++fidx, path[i].p_idx++,
1016 sizeof(struct ext4_extent_idx));
1018 BUG_ON(neh->eh_entries > neh->eh_max);*/
1023 memmove(++fidx
, path
[i
].p_idx
- m
,
1024 sizeof(struct ext4_extent_idx
) * m
);
1025 le16_add_cpu(&neh
->eh_entries
, m
);
1027 set_buffer_uptodate(bh
);
1030 err
= ext4_handle_dirty_metadata(handle
, inode
, bh
);
1036 /* correct old index */
1038 err
= ext4_ext_get_access(handle
, inode
, path
+ i
);
1041 le16_add_cpu(&path
[i
].p_hdr
->eh_entries
, -m
);
1042 err
= ext4_ext_dirty(handle
, inode
, path
+ i
);
1050 /* insert new index */
1051 err
= ext4_ext_insert_index(handle
, inode
, path
+ at
,
1052 le32_to_cpu(border
), newblock
);
1056 if (buffer_locked(bh
))
1062 /* free all allocated blocks in error case */
1063 for (i
= 0; i
< depth
; i
++) {
1066 ext4_free_blocks(handle
, inode
, 0, ablocks
[i
], 1,
1067 EXT4_FREE_BLOCKS_METADATA
);
1076 * ext4_ext_grow_indepth:
1077 * implements tree growing procedure:
1078 * - allocates new block
1079 * - moves top-level data (index block or leaf) into the new block
1080 * - initializes new top-level, creating index that points to the
1081 * just created block
1083 static int ext4_ext_grow_indepth(handle_t
*handle
, struct inode
*inode
,
1084 struct ext4_ext_path
*path
,
1085 struct ext4_extent
*newext
)
1087 struct ext4_ext_path
*curp
= path
;
1088 struct ext4_extent_header
*neh
;
1089 struct ext4_extent_idx
*fidx
;
1090 struct buffer_head
*bh
;
1091 ext4_fsblk_t newblock
;
1094 newblock
= ext4_ext_new_meta_block(handle
, inode
, path
, newext
, &err
);
1098 bh
= sb_getblk(inode
->i_sb
, newblock
);
1101 ext4_std_error(inode
->i_sb
, err
);
1106 err
= ext4_journal_get_create_access(handle
, bh
);
1112 /* move top-level index/leaf into new block */
1113 memmove(bh
->b_data
, curp
->p_hdr
, sizeof(EXT4_I(inode
)->i_data
));
1115 /* set size of new block */
1116 neh
= ext_block_hdr(bh
);
1117 /* old root could have indexes or leaves
1118 * so calculate e_max right way */
1119 if (ext_depth(inode
))
1120 neh
->eh_max
= cpu_to_le16(ext4_ext_space_block_idx(inode
, 0));
1122 neh
->eh_max
= cpu_to_le16(ext4_ext_space_block(inode
, 0));
1123 neh
->eh_magic
= EXT4_EXT_MAGIC
;
1124 set_buffer_uptodate(bh
);
1127 err
= ext4_handle_dirty_metadata(handle
, inode
, bh
);
1131 /* create index in new top-level index: num,max,pointer */
1132 err
= ext4_ext_get_access(handle
, inode
, curp
);
1136 curp
->p_hdr
->eh_magic
= EXT4_EXT_MAGIC
;
1137 curp
->p_hdr
->eh_max
= cpu_to_le16(ext4_ext_space_root_idx(inode
, 0));
1138 curp
->p_hdr
->eh_entries
= cpu_to_le16(1);
1139 curp
->p_idx
= EXT_FIRST_INDEX(curp
->p_hdr
);
1141 if (path
[0].p_hdr
->eh_depth
)
1142 curp
->p_idx
->ei_block
=
1143 EXT_FIRST_INDEX(path
[0].p_hdr
)->ei_block
;
1145 curp
->p_idx
->ei_block
=
1146 EXT_FIRST_EXTENT(path
[0].p_hdr
)->ee_block
;
1147 ext4_idx_store_pblock(curp
->p_idx
, newblock
);
1149 neh
= ext_inode_hdr(inode
);
1150 fidx
= EXT_FIRST_INDEX(neh
);
1151 ext_debug("new root: num %d(%d), lblock %d, ptr %llu\n",
1152 le16_to_cpu(neh
->eh_entries
), le16_to_cpu(neh
->eh_max
),
1153 le32_to_cpu(fidx
->ei_block
), idx_pblock(fidx
));
1155 neh
->eh_depth
= cpu_to_le16(path
->p_depth
+ 1);
1156 err
= ext4_ext_dirty(handle
, inode
, curp
);
1164 * ext4_ext_create_new_leaf:
1165 * finds empty index and adds new leaf.
1166 * if no free index is found, then it requests in-depth growing.
1168 static int ext4_ext_create_new_leaf(handle_t
*handle
, struct inode
*inode
,
1169 struct ext4_ext_path
*path
,
1170 struct ext4_extent
*newext
)
1172 struct ext4_ext_path
*curp
;
1173 int depth
, i
, err
= 0;
1176 i
= depth
= ext_depth(inode
);
1178 /* walk up to the tree and look for free index entry */
1179 curp
= path
+ depth
;
1180 while (i
> 0 && !EXT_HAS_FREE_INDEX(curp
)) {
1185 /* we use already allocated block for index block,
1186 * so subsequent data blocks should be contiguous */
1187 if (EXT_HAS_FREE_INDEX(curp
)) {
1188 /* if we found index with free entry, then use that
1189 * entry: create all needed subtree and add new leaf */
1190 err
= ext4_ext_split(handle
, inode
, path
, newext
, i
);
1195 ext4_ext_drop_refs(path
);
1196 path
= ext4_ext_find_extent(inode
,
1197 (ext4_lblk_t
)le32_to_cpu(newext
->ee_block
),
1200 err
= PTR_ERR(path
);
1202 /* tree is full, time to grow in depth */
1203 err
= ext4_ext_grow_indepth(handle
, inode
, path
, newext
);
1208 ext4_ext_drop_refs(path
);
1209 path
= ext4_ext_find_extent(inode
,
1210 (ext4_lblk_t
)le32_to_cpu(newext
->ee_block
),
1213 err
= PTR_ERR(path
);
1218 * only first (depth 0 -> 1) produces free space;
1219 * in all other cases we have to split the grown tree
1221 depth
= ext_depth(inode
);
1222 if (path
[depth
].p_hdr
->eh_entries
== path
[depth
].p_hdr
->eh_max
) {
1223 /* now we need to split */
1233 * search the closest allocated block to the left for *logical
1234 * and returns it at @logical + it's physical address at @phys
1235 * if *logical is the smallest allocated block, the function
1236 * returns 0 at @phys
1237 * return value contains 0 (success) or error code
1240 ext4_ext_search_left(struct inode
*inode
, struct ext4_ext_path
*path
,
1241 ext4_lblk_t
*logical
, ext4_fsblk_t
*phys
)
1243 struct ext4_extent_idx
*ix
;
1244 struct ext4_extent
*ex
;
1247 if (unlikely(path
== NULL
)) {
1248 EXT4_ERROR_INODE(inode
, "path == NULL *logical %d!", *logical
);
1251 depth
= path
->p_depth
;
1254 if (depth
== 0 && path
->p_ext
== NULL
)
1257 /* usually extent in the path covers blocks smaller
1258 * then *logical, but it can be that extent is the
1259 * first one in the file */
1261 ex
= path
[depth
].p_ext
;
1262 ee_len
= ext4_ext_get_actual_len(ex
);
1263 if (*logical
< le32_to_cpu(ex
->ee_block
)) {
1264 if (unlikely(EXT_FIRST_EXTENT(path
[depth
].p_hdr
) != ex
)) {
1265 EXT4_ERROR_INODE(inode
,
1266 "EXT_FIRST_EXTENT != ex *logical %d ee_block %d!",
1267 *logical
, le32_to_cpu(ex
->ee_block
));
1270 while (--depth
>= 0) {
1271 ix
= path
[depth
].p_idx
;
1272 if (unlikely(ix
!= EXT_FIRST_INDEX(path
[depth
].p_hdr
))) {
1273 EXT4_ERROR_INODE(inode
,
1274 "ix (%d) != EXT_FIRST_INDEX (%d) (depth %d)!",
1275 ix
!= NULL
? ix
->ei_block
: 0,
1276 EXT_FIRST_INDEX(path
[depth
].p_hdr
) != NULL
?
1277 EXT_FIRST_INDEX(path
[depth
].p_hdr
)->ei_block
: 0,
1285 if (unlikely(*logical
< (le32_to_cpu(ex
->ee_block
) + ee_len
))) {
1286 EXT4_ERROR_INODE(inode
,
1287 "logical %d < ee_block %d + ee_len %d!",
1288 *logical
, le32_to_cpu(ex
->ee_block
), ee_len
);
1292 *logical
= le32_to_cpu(ex
->ee_block
) + ee_len
- 1;
1293 *phys
= ext_pblock(ex
) + ee_len
- 1;
1298 * search the closest allocated block to the right for *logical
1299 * and returns it at @logical + it's physical address at @phys
1300 * if *logical is the smallest allocated block, the function
1301 * returns 0 at @phys
1302 * return value contains 0 (success) or error code
1305 ext4_ext_search_right(struct inode
*inode
, struct ext4_ext_path
*path
,
1306 ext4_lblk_t
*logical
, ext4_fsblk_t
*phys
)
1308 struct buffer_head
*bh
= NULL
;
1309 struct ext4_extent_header
*eh
;
1310 struct ext4_extent_idx
*ix
;
1311 struct ext4_extent
*ex
;
1313 int depth
; /* Note, NOT eh_depth; depth from top of tree */
1316 if (unlikely(path
== NULL
)) {
1317 EXT4_ERROR_INODE(inode
, "path == NULL *logical %d!", *logical
);
1320 depth
= path
->p_depth
;
1323 if (depth
== 0 && path
->p_ext
== NULL
)
1326 /* usually extent in the path covers blocks smaller
1327 * then *logical, but it can be that extent is the
1328 * first one in the file */
1330 ex
= path
[depth
].p_ext
;
1331 ee_len
= ext4_ext_get_actual_len(ex
);
1332 if (*logical
< le32_to_cpu(ex
->ee_block
)) {
1333 if (unlikely(EXT_FIRST_EXTENT(path
[depth
].p_hdr
) != ex
)) {
1334 EXT4_ERROR_INODE(inode
,
1335 "first_extent(path[%d].p_hdr) != ex",
1339 while (--depth
>= 0) {
1340 ix
= path
[depth
].p_idx
;
1341 if (unlikely(ix
!= EXT_FIRST_INDEX(path
[depth
].p_hdr
))) {
1342 EXT4_ERROR_INODE(inode
,
1343 "ix != EXT_FIRST_INDEX *logical %d!",
1348 *logical
= le32_to_cpu(ex
->ee_block
);
1349 *phys
= ext_pblock(ex
);
1353 if (unlikely(*logical
< (le32_to_cpu(ex
->ee_block
) + ee_len
))) {
1354 EXT4_ERROR_INODE(inode
,
1355 "logical %d < ee_block %d + ee_len %d!",
1356 *logical
, le32_to_cpu(ex
->ee_block
), ee_len
);
1360 if (ex
!= EXT_LAST_EXTENT(path
[depth
].p_hdr
)) {
1361 /* next allocated block in this leaf */
1363 *logical
= le32_to_cpu(ex
->ee_block
);
1364 *phys
= ext_pblock(ex
);
1368 /* go up and search for index to the right */
1369 while (--depth
>= 0) {
1370 ix
= path
[depth
].p_idx
;
1371 if (ix
!= EXT_LAST_INDEX(path
[depth
].p_hdr
))
1375 /* we've gone up to the root and found no index to the right */
1379 /* we've found index to the right, let's
1380 * follow it and find the closest allocated
1381 * block to the right */
1383 block
= idx_pblock(ix
);
1384 while (++depth
< path
->p_depth
) {
1385 bh
= sb_bread(inode
->i_sb
, block
);
1388 eh
= ext_block_hdr(bh
);
1389 /* subtract from p_depth to get proper eh_depth */
1390 if (ext4_ext_check(inode
, eh
, path
->p_depth
- depth
)) {
1394 ix
= EXT_FIRST_INDEX(eh
);
1395 block
= idx_pblock(ix
);
1399 bh
= sb_bread(inode
->i_sb
, block
);
1402 eh
= ext_block_hdr(bh
);
1403 if (ext4_ext_check(inode
, eh
, path
->p_depth
- depth
)) {
1407 ex
= EXT_FIRST_EXTENT(eh
);
1408 *logical
= le32_to_cpu(ex
->ee_block
);
1409 *phys
= ext_pblock(ex
);
1415 * ext4_ext_next_allocated_block:
1416 * returns allocated block in subsequent extent or EXT_MAX_BLOCK.
1417 * NOTE: it considers block number from index entry as
1418 * allocated block. Thus, index entries have to be consistent
1422 ext4_ext_next_allocated_block(struct ext4_ext_path
*path
)
1426 BUG_ON(path
== NULL
);
1427 depth
= path
->p_depth
;
1429 if (depth
== 0 && path
->p_ext
== NULL
)
1430 return EXT_MAX_BLOCK
;
1432 while (depth
>= 0) {
1433 if (depth
== path
->p_depth
) {
1435 if (path
[depth
].p_ext
!=
1436 EXT_LAST_EXTENT(path
[depth
].p_hdr
))
1437 return le32_to_cpu(path
[depth
].p_ext
[1].ee_block
);
1440 if (path
[depth
].p_idx
!=
1441 EXT_LAST_INDEX(path
[depth
].p_hdr
))
1442 return le32_to_cpu(path
[depth
].p_idx
[1].ei_block
);
1447 return EXT_MAX_BLOCK
;
1451 * ext4_ext_next_leaf_block:
1452 * returns first allocated block from next leaf or EXT_MAX_BLOCK
1454 static ext4_lblk_t
ext4_ext_next_leaf_block(struct inode
*inode
,
1455 struct ext4_ext_path
*path
)
1459 BUG_ON(path
== NULL
);
1460 depth
= path
->p_depth
;
1462 /* zero-tree has no leaf blocks at all */
1464 return EXT_MAX_BLOCK
;
1466 /* go to index block */
1469 while (depth
>= 0) {
1470 if (path
[depth
].p_idx
!=
1471 EXT_LAST_INDEX(path
[depth
].p_hdr
))
1472 return (ext4_lblk_t
)
1473 le32_to_cpu(path
[depth
].p_idx
[1].ei_block
);
1477 return EXT_MAX_BLOCK
;
1481 * ext4_ext_correct_indexes:
1482 * if leaf gets modified and modified extent is first in the leaf,
1483 * then we have to correct all indexes above.
1484 * TODO: do we need to correct tree in all cases?
1486 static int ext4_ext_correct_indexes(handle_t
*handle
, struct inode
*inode
,
1487 struct ext4_ext_path
*path
)
1489 struct ext4_extent_header
*eh
;
1490 int depth
= ext_depth(inode
);
1491 struct ext4_extent
*ex
;
1495 eh
= path
[depth
].p_hdr
;
1496 ex
= path
[depth
].p_ext
;
1498 if (unlikely(ex
== NULL
|| eh
== NULL
)) {
1499 EXT4_ERROR_INODE(inode
,
1500 "ex %p == NULL or eh %p == NULL", ex
, eh
);
1505 /* there is no tree at all */
1509 if (ex
!= EXT_FIRST_EXTENT(eh
)) {
1510 /* we correct tree if first leaf got modified only */
1515 * TODO: we need correction if border is smaller than current one
1518 border
= path
[depth
].p_ext
->ee_block
;
1519 err
= ext4_ext_get_access(handle
, inode
, path
+ k
);
1522 path
[k
].p_idx
->ei_block
= border
;
1523 err
= ext4_ext_dirty(handle
, inode
, path
+ k
);
1528 /* change all left-side indexes */
1529 if (path
[k
+1].p_idx
!= EXT_FIRST_INDEX(path
[k
+1].p_hdr
))
1531 err
= ext4_ext_get_access(handle
, inode
, path
+ k
);
1534 path
[k
].p_idx
->ei_block
= border
;
1535 err
= ext4_ext_dirty(handle
, inode
, path
+ k
);
1544 ext4_can_extents_be_merged(struct inode
*inode
, struct ext4_extent
*ex1
,
1545 struct ext4_extent
*ex2
)
1547 unsigned short ext1_ee_len
, ext2_ee_len
, max_len
;
1550 * Make sure that either both extents are uninitialized, or
1553 if (ext4_ext_is_uninitialized(ex1
) ^ ext4_ext_is_uninitialized(ex2
))
1556 if (ext4_ext_is_uninitialized(ex1
))
1557 max_len
= EXT_UNINIT_MAX_LEN
;
1559 max_len
= EXT_INIT_MAX_LEN
;
1561 ext1_ee_len
= ext4_ext_get_actual_len(ex1
);
1562 ext2_ee_len
= ext4_ext_get_actual_len(ex2
);
1564 if (le32_to_cpu(ex1
->ee_block
) + ext1_ee_len
!=
1565 le32_to_cpu(ex2
->ee_block
))
1569 * To allow future support for preallocated extents to be added
1570 * as an RO_COMPAT feature, refuse to merge to extents if
1571 * this can result in the top bit of ee_len being set.
1573 if (ext1_ee_len
+ ext2_ee_len
> max_len
)
1575 #ifdef AGGRESSIVE_TEST
1576 if (ext1_ee_len
>= 4)
1580 if (ext_pblock(ex1
) + ext1_ee_len
== ext_pblock(ex2
))
1586 * This function tries to merge the "ex" extent to the next extent in the tree.
1587 * It always tries to merge towards right. If you want to merge towards
1588 * left, pass "ex - 1" as argument instead of "ex".
1589 * Returns 0 if the extents (ex and ex+1) were _not_ merged and returns
1590 * 1 if they got merged.
1592 int ext4_ext_try_to_merge(struct inode
*inode
,
1593 struct ext4_ext_path
*path
,
1594 struct ext4_extent
*ex
)
1596 struct ext4_extent_header
*eh
;
1597 unsigned int depth
, len
;
1599 int uninitialized
= 0;
1601 depth
= ext_depth(inode
);
1602 BUG_ON(path
[depth
].p_hdr
== NULL
);
1603 eh
= path
[depth
].p_hdr
;
1605 while (ex
< EXT_LAST_EXTENT(eh
)) {
1606 if (!ext4_can_extents_be_merged(inode
, ex
, ex
+ 1))
1608 /* merge with next extent! */
1609 if (ext4_ext_is_uninitialized(ex
))
1611 ex
->ee_len
= cpu_to_le16(ext4_ext_get_actual_len(ex
)
1612 + ext4_ext_get_actual_len(ex
+ 1));
1614 ext4_ext_mark_uninitialized(ex
);
1616 if (ex
+ 1 < EXT_LAST_EXTENT(eh
)) {
1617 len
= (EXT_LAST_EXTENT(eh
) - ex
- 1)
1618 * sizeof(struct ext4_extent
);
1619 memmove(ex
+ 1, ex
+ 2, len
);
1621 le16_add_cpu(&eh
->eh_entries
, -1);
1623 WARN_ON(eh
->eh_entries
== 0);
1624 if (!eh
->eh_entries
)
1625 ext4_error(inode
->i_sb
,
1626 "inode#%lu, eh->eh_entries = 0!",
1634 * check if a portion of the "newext" extent overlaps with an
1637 * If there is an overlap discovered, it updates the length of the newext
1638 * such that there will be no overlap, and then returns 1.
1639 * If there is no overlap found, it returns 0.
1641 unsigned int ext4_ext_check_overlap(struct inode
*inode
,
1642 struct ext4_extent
*newext
,
1643 struct ext4_ext_path
*path
)
1646 unsigned int depth
, len1
;
1647 unsigned int ret
= 0;
1649 b1
= le32_to_cpu(newext
->ee_block
);
1650 len1
= ext4_ext_get_actual_len(newext
);
1651 depth
= ext_depth(inode
);
1652 if (!path
[depth
].p_ext
)
1654 b2
= le32_to_cpu(path
[depth
].p_ext
->ee_block
);
1657 * get the next allocated block if the extent in the path
1658 * is before the requested block(s)
1661 b2
= ext4_ext_next_allocated_block(path
);
1662 if (b2
== EXT_MAX_BLOCK
)
1666 /* check for wrap through zero on extent logical start block*/
1667 if (b1
+ len1
< b1
) {
1668 len1
= EXT_MAX_BLOCK
- b1
;
1669 newext
->ee_len
= cpu_to_le16(len1
);
1673 /* check for overlap */
1674 if (b1
+ len1
> b2
) {
1675 newext
->ee_len
= cpu_to_le16(b2
- b1
);
1683 * ext4_ext_insert_extent:
1684 * tries to merge requsted extent into the existing extent or
1685 * inserts requested extent as new one into the tree,
1686 * creating new leaf in the no-space case.
1688 int ext4_ext_insert_extent(handle_t
*handle
, struct inode
*inode
,
1689 struct ext4_ext_path
*path
,
1690 struct ext4_extent
*newext
, int flag
)
1692 struct ext4_extent_header
*eh
;
1693 struct ext4_extent
*ex
, *fex
;
1694 struct ext4_extent
*nearex
; /* nearest extent */
1695 struct ext4_ext_path
*npath
= NULL
;
1696 int depth
, len
, err
;
1698 unsigned uninitialized
= 0;
1700 if (unlikely(ext4_ext_get_actual_len(newext
) == 0)) {
1701 EXT4_ERROR_INODE(inode
, "ext4_ext_get_actual_len(newext) == 0");
1704 depth
= ext_depth(inode
);
1705 ex
= path
[depth
].p_ext
;
1706 if (unlikely(path
[depth
].p_hdr
== NULL
)) {
1707 EXT4_ERROR_INODE(inode
, "path[%d].p_hdr == NULL", depth
);
1711 /* try to insert block into found extent and return */
1712 if (ex
&& !(flag
& EXT4_GET_BLOCKS_PRE_IO
)
1713 && ext4_can_extents_be_merged(inode
, ex
, newext
)) {
1714 ext_debug("append [%d]%d block to %d:[%d]%d (from %llu)\n",
1715 ext4_ext_is_uninitialized(newext
),
1716 ext4_ext_get_actual_len(newext
),
1717 le32_to_cpu(ex
->ee_block
),
1718 ext4_ext_is_uninitialized(ex
),
1719 ext4_ext_get_actual_len(ex
), ext_pblock(ex
));
1720 err
= ext4_ext_get_access(handle
, inode
, path
+ depth
);
1725 * ext4_can_extents_be_merged should have checked that either
1726 * both extents are uninitialized, or both aren't. Thus we
1727 * need to check only one of them here.
1729 if (ext4_ext_is_uninitialized(ex
))
1731 ex
->ee_len
= cpu_to_le16(ext4_ext_get_actual_len(ex
)
1732 + ext4_ext_get_actual_len(newext
));
1734 ext4_ext_mark_uninitialized(ex
);
1735 eh
= path
[depth
].p_hdr
;
1741 depth
= ext_depth(inode
);
1742 eh
= path
[depth
].p_hdr
;
1743 if (le16_to_cpu(eh
->eh_entries
) < le16_to_cpu(eh
->eh_max
))
1746 /* probably next leaf has space for us? */
1747 fex
= EXT_LAST_EXTENT(eh
);
1748 next
= ext4_ext_next_leaf_block(inode
, path
);
1749 if (le32_to_cpu(newext
->ee_block
) > le32_to_cpu(fex
->ee_block
)
1750 && next
!= EXT_MAX_BLOCK
) {
1751 ext_debug("next leaf block - %d\n", next
);
1752 BUG_ON(npath
!= NULL
);
1753 npath
= ext4_ext_find_extent(inode
, next
, NULL
);
1755 return PTR_ERR(npath
);
1756 BUG_ON(npath
->p_depth
!= path
->p_depth
);
1757 eh
= npath
[depth
].p_hdr
;
1758 if (le16_to_cpu(eh
->eh_entries
) < le16_to_cpu(eh
->eh_max
)) {
1759 ext_debug("next leaf isnt full(%d)\n",
1760 le16_to_cpu(eh
->eh_entries
));
1764 ext_debug("next leaf has no free space(%d,%d)\n",
1765 le16_to_cpu(eh
->eh_entries
), le16_to_cpu(eh
->eh_max
));
1769 * There is no free space in the found leaf.
1770 * We're gonna add a new leaf in the tree.
1772 err
= ext4_ext_create_new_leaf(handle
, inode
, path
, newext
);
1775 depth
= ext_depth(inode
);
1776 eh
= path
[depth
].p_hdr
;
1779 nearex
= path
[depth
].p_ext
;
1781 err
= ext4_ext_get_access(handle
, inode
, path
+ depth
);
1786 /* there is no extent in this leaf, create first one */
1787 ext_debug("first extent in the leaf: %d:%llu:[%d]%d\n",
1788 le32_to_cpu(newext
->ee_block
),
1790 ext4_ext_is_uninitialized(newext
),
1791 ext4_ext_get_actual_len(newext
));
1792 path
[depth
].p_ext
= EXT_FIRST_EXTENT(eh
);
1793 } else if (le32_to_cpu(newext
->ee_block
)
1794 > le32_to_cpu(nearex
->ee_block
)) {
1795 /* BUG_ON(newext->ee_block == nearex->ee_block); */
1796 if (nearex
!= EXT_LAST_EXTENT(eh
)) {
1797 len
= EXT_MAX_EXTENT(eh
) - nearex
;
1798 len
= (len
- 1) * sizeof(struct ext4_extent
);
1799 len
= len
< 0 ? 0 : len
;
1800 ext_debug("insert %d:%llu:[%d]%d after: nearest 0x%p, "
1801 "move %d from 0x%p to 0x%p\n",
1802 le32_to_cpu(newext
->ee_block
),
1804 ext4_ext_is_uninitialized(newext
),
1805 ext4_ext_get_actual_len(newext
),
1806 nearex
, len
, nearex
+ 1, nearex
+ 2);
1807 memmove(nearex
+ 2, nearex
+ 1, len
);
1809 path
[depth
].p_ext
= nearex
+ 1;
1811 BUG_ON(newext
->ee_block
== nearex
->ee_block
);
1812 len
= (EXT_MAX_EXTENT(eh
) - nearex
) * sizeof(struct ext4_extent
);
1813 len
= len
< 0 ? 0 : len
;
1814 ext_debug("insert %d:%llu:[%d]%d before: nearest 0x%p, "
1815 "move %d from 0x%p to 0x%p\n",
1816 le32_to_cpu(newext
->ee_block
),
1818 ext4_ext_is_uninitialized(newext
),
1819 ext4_ext_get_actual_len(newext
),
1820 nearex
, len
, nearex
+ 1, nearex
+ 2);
1821 memmove(nearex
+ 1, nearex
, len
);
1822 path
[depth
].p_ext
= nearex
;
1825 le16_add_cpu(&eh
->eh_entries
, 1);
1826 nearex
= path
[depth
].p_ext
;
1827 nearex
->ee_block
= newext
->ee_block
;
1828 ext4_ext_store_pblock(nearex
, ext_pblock(newext
));
1829 nearex
->ee_len
= newext
->ee_len
;
1832 /* try to merge extents to the right */
1833 if (!(flag
& EXT4_GET_BLOCKS_PRE_IO
))
1834 ext4_ext_try_to_merge(inode
, path
, nearex
);
1836 /* try to merge extents to the left */
1838 /* time to correct all indexes above */
1839 err
= ext4_ext_correct_indexes(handle
, inode
, path
);
1843 err
= ext4_ext_dirty(handle
, inode
, path
+ depth
);
1847 ext4_ext_drop_refs(npath
);
1850 ext4_ext_invalidate_cache(inode
);
1854 int ext4_ext_walk_space(struct inode
*inode
, ext4_lblk_t block
,
1855 ext4_lblk_t num
, ext_prepare_callback func
,
1858 struct ext4_ext_path
*path
= NULL
;
1859 struct ext4_ext_cache cbex
;
1860 struct ext4_extent
*ex
;
1861 ext4_lblk_t next
, start
= 0, end
= 0;
1862 ext4_lblk_t last
= block
+ num
;
1863 int depth
, exists
, err
= 0;
1865 BUG_ON(func
== NULL
);
1866 BUG_ON(inode
== NULL
);
1868 while (block
< last
&& block
!= EXT_MAX_BLOCK
) {
1870 /* find extent for this block */
1871 down_read(&EXT4_I(inode
)->i_data_sem
);
1872 path
= ext4_ext_find_extent(inode
, block
, path
);
1873 up_read(&EXT4_I(inode
)->i_data_sem
);
1875 err
= PTR_ERR(path
);
1880 depth
= ext_depth(inode
);
1881 if (unlikely(path
[depth
].p_hdr
== NULL
)) {
1882 EXT4_ERROR_INODE(inode
, "path[%d].p_hdr == NULL", depth
);
1886 ex
= path
[depth
].p_ext
;
1887 next
= ext4_ext_next_allocated_block(path
);
1891 /* there is no extent yet, so try to allocate
1892 * all requested space */
1895 } else if (le32_to_cpu(ex
->ee_block
) > block
) {
1896 /* need to allocate space before found extent */
1898 end
= le32_to_cpu(ex
->ee_block
);
1899 if (block
+ num
< end
)
1901 } else if (block
>= le32_to_cpu(ex
->ee_block
)
1902 + ext4_ext_get_actual_len(ex
)) {
1903 /* need to allocate space after found extent */
1908 } else if (block
>= le32_to_cpu(ex
->ee_block
)) {
1910 * some part of requested space is covered
1914 end
= le32_to_cpu(ex
->ee_block
)
1915 + ext4_ext_get_actual_len(ex
);
1916 if (block
+ num
< end
)
1922 BUG_ON(end
<= start
);
1925 cbex
.ec_block
= start
;
1926 cbex
.ec_len
= end
- start
;
1928 cbex
.ec_type
= EXT4_EXT_CACHE_GAP
;
1930 cbex
.ec_block
= le32_to_cpu(ex
->ee_block
);
1931 cbex
.ec_len
= ext4_ext_get_actual_len(ex
);
1932 cbex
.ec_start
= ext_pblock(ex
);
1933 cbex
.ec_type
= EXT4_EXT_CACHE_EXTENT
;
1936 if (unlikely(cbex
.ec_len
== 0)) {
1937 EXT4_ERROR_INODE(inode
, "cbex.ec_len == 0");
1941 err
= func(inode
, path
, &cbex
, ex
, cbdata
);
1942 ext4_ext_drop_refs(path
);
1947 if (err
== EXT_REPEAT
)
1949 else if (err
== EXT_BREAK
) {
1954 if (ext_depth(inode
) != depth
) {
1955 /* depth was changed. we have to realloc path */
1960 block
= cbex
.ec_block
+ cbex
.ec_len
;
1964 ext4_ext_drop_refs(path
);
1972 ext4_ext_put_in_cache(struct inode
*inode
, ext4_lblk_t block
,
1973 __u32 len
, ext4_fsblk_t start
, int type
)
1975 struct ext4_ext_cache
*cex
;
1977 spin_lock(&EXT4_I(inode
)->i_block_reservation_lock
);
1978 cex
= &EXT4_I(inode
)->i_cached_extent
;
1979 cex
->ec_type
= type
;
1980 cex
->ec_block
= block
;
1982 cex
->ec_start
= start
;
1983 spin_unlock(&EXT4_I(inode
)->i_block_reservation_lock
);
1987 * ext4_ext_put_gap_in_cache:
1988 * calculate boundaries of the gap that the requested block fits into
1989 * and cache this gap
1992 ext4_ext_put_gap_in_cache(struct inode
*inode
, struct ext4_ext_path
*path
,
1995 int depth
= ext_depth(inode
);
1998 struct ext4_extent
*ex
;
2000 ex
= path
[depth
].p_ext
;
2002 /* there is no extent yet, so gap is [0;-] */
2004 len
= EXT_MAX_BLOCK
;
2005 ext_debug("cache gap(whole file):");
2006 } else if (block
< le32_to_cpu(ex
->ee_block
)) {
2008 len
= le32_to_cpu(ex
->ee_block
) - block
;
2009 ext_debug("cache gap(before): %u [%u:%u]",
2011 le32_to_cpu(ex
->ee_block
),
2012 ext4_ext_get_actual_len(ex
));
2013 } else if (block
>= le32_to_cpu(ex
->ee_block
)
2014 + ext4_ext_get_actual_len(ex
)) {
2016 lblock
= le32_to_cpu(ex
->ee_block
)
2017 + ext4_ext_get_actual_len(ex
);
2019 next
= ext4_ext_next_allocated_block(path
);
2020 ext_debug("cache gap(after): [%u:%u] %u",
2021 le32_to_cpu(ex
->ee_block
),
2022 ext4_ext_get_actual_len(ex
),
2024 BUG_ON(next
== lblock
);
2025 len
= next
- lblock
;
2031 ext_debug(" -> %u:%lu\n", lblock
, len
);
2032 ext4_ext_put_in_cache(inode
, lblock
, len
, 0, EXT4_EXT_CACHE_GAP
);
2036 ext4_ext_in_cache(struct inode
*inode
, ext4_lblk_t block
,
2037 struct ext4_extent
*ex
)
2039 struct ext4_ext_cache
*cex
;
2040 int ret
= EXT4_EXT_CACHE_NO
;
2043 * We borrow i_block_reservation_lock to protect i_cached_extent
2045 spin_lock(&EXT4_I(inode
)->i_block_reservation_lock
);
2046 cex
= &EXT4_I(inode
)->i_cached_extent
;
2048 /* has cache valid data? */
2049 if (cex
->ec_type
== EXT4_EXT_CACHE_NO
)
2052 BUG_ON(cex
->ec_type
!= EXT4_EXT_CACHE_GAP
&&
2053 cex
->ec_type
!= EXT4_EXT_CACHE_EXTENT
);
2054 if (in_range(block
, cex
->ec_block
, cex
->ec_len
)) {
2055 ex
->ee_block
= cpu_to_le32(cex
->ec_block
);
2056 ext4_ext_store_pblock(ex
, cex
->ec_start
);
2057 ex
->ee_len
= cpu_to_le16(cex
->ec_len
);
2058 ext_debug("%u cached by %u:%u:%llu\n",
2060 cex
->ec_block
, cex
->ec_len
, cex
->ec_start
);
2064 spin_unlock(&EXT4_I(inode
)->i_block_reservation_lock
);
2070 * removes index from the index block.
2071 * It's used in truncate case only, thus all requests are for
2072 * last index in the block only.
2074 static int ext4_ext_rm_idx(handle_t
*handle
, struct inode
*inode
,
2075 struct ext4_ext_path
*path
)
2080 /* free index block */
2082 leaf
= idx_pblock(path
->p_idx
);
2083 if (unlikely(path
->p_hdr
->eh_entries
== 0)) {
2084 EXT4_ERROR_INODE(inode
, "path->p_hdr->eh_entries == 0");
2087 err
= ext4_ext_get_access(handle
, inode
, path
);
2090 le16_add_cpu(&path
->p_hdr
->eh_entries
, -1);
2091 err
= ext4_ext_dirty(handle
, inode
, path
);
2094 ext_debug("index is empty, remove it, free block %llu\n", leaf
);
2095 ext4_free_blocks(handle
, inode
, 0, leaf
, 1,
2096 EXT4_FREE_BLOCKS_METADATA
| EXT4_FREE_BLOCKS_FORGET
);
2101 * ext4_ext_calc_credits_for_single_extent:
2102 * This routine returns max. credits that needed to insert an extent
2103 * to the extent tree.
2104 * When pass the actual path, the caller should calculate credits
2107 int ext4_ext_calc_credits_for_single_extent(struct inode
*inode
, int nrblocks
,
2108 struct ext4_ext_path
*path
)
2111 int depth
= ext_depth(inode
);
2114 /* probably there is space in leaf? */
2115 if (le16_to_cpu(path
[depth
].p_hdr
->eh_entries
)
2116 < le16_to_cpu(path
[depth
].p_hdr
->eh_max
)) {
2119 * There are some space in the leaf tree, no
2120 * need to account for leaf block credit
2122 * bitmaps and block group descriptor blocks
2123 * and other metadat blocks still need to be
2126 /* 1 bitmap, 1 block group descriptor */
2127 ret
= 2 + EXT4_META_TRANS_BLOCKS(inode
->i_sb
);
2132 return ext4_chunk_trans_blocks(inode
, nrblocks
);
2136 * How many index/leaf blocks need to change/allocate to modify nrblocks?
2138 * if nrblocks are fit in a single extent (chunk flag is 1), then
2139 * in the worse case, each tree level index/leaf need to be changed
2140 * if the tree split due to insert a new extent, then the old tree
2141 * index/leaf need to be updated too
2143 * If the nrblocks are discontiguous, they could cause
2144 * the whole tree split more than once, but this is really rare.
2146 int ext4_ext_index_trans_blocks(struct inode
*inode
, int nrblocks
, int chunk
)
2149 int depth
= ext_depth(inode
);
2159 static int ext4_remove_blocks(handle_t
*handle
, struct inode
*inode
,
2160 struct ext4_extent
*ex
,
2161 ext4_lblk_t from
, ext4_lblk_t to
)
2163 unsigned short ee_len
= ext4_ext_get_actual_len(ex
);
2164 int flags
= EXT4_FREE_BLOCKS_FORGET
;
2166 if (S_ISDIR(inode
->i_mode
) || S_ISLNK(inode
->i_mode
))
2167 flags
|= EXT4_FREE_BLOCKS_METADATA
;
2168 #ifdef EXTENTS_STATS
2170 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
2171 spin_lock(&sbi
->s_ext_stats_lock
);
2172 sbi
->s_ext_blocks
+= ee_len
;
2173 sbi
->s_ext_extents
++;
2174 if (ee_len
< sbi
->s_ext_min
)
2175 sbi
->s_ext_min
= ee_len
;
2176 if (ee_len
> sbi
->s_ext_max
)
2177 sbi
->s_ext_max
= ee_len
;
2178 if (ext_depth(inode
) > sbi
->s_depth_max
)
2179 sbi
->s_depth_max
= ext_depth(inode
);
2180 spin_unlock(&sbi
->s_ext_stats_lock
);
2183 if (from
>= le32_to_cpu(ex
->ee_block
)
2184 && to
== le32_to_cpu(ex
->ee_block
) + ee_len
- 1) {
2189 num
= le32_to_cpu(ex
->ee_block
) + ee_len
- from
;
2190 start
= ext_pblock(ex
) + ee_len
- num
;
2191 ext_debug("free last %u blocks starting %llu\n", num
, start
);
2192 ext4_free_blocks(handle
, inode
, 0, start
, num
, flags
);
2193 } else if (from
== le32_to_cpu(ex
->ee_block
)
2194 && to
<= le32_to_cpu(ex
->ee_block
) + ee_len
- 1) {
2195 printk(KERN_INFO
"strange request: removal %u-%u from %u:%u\n",
2196 from
, to
, le32_to_cpu(ex
->ee_block
), ee_len
);
2198 printk(KERN_INFO
"strange request: removal(2) "
2199 "%u-%u from %u:%u\n",
2200 from
, to
, le32_to_cpu(ex
->ee_block
), ee_len
);
2206 ext4_ext_rm_leaf(handle_t
*handle
, struct inode
*inode
,
2207 struct ext4_ext_path
*path
, ext4_lblk_t start
)
2209 int err
= 0, correct_index
= 0;
2210 int depth
= ext_depth(inode
), credits
;
2211 struct ext4_extent_header
*eh
;
2212 ext4_lblk_t a
, b
, block
;
2214 ext4_lblk_t ex_ee_block
;
2215 unsigned short ex_ee_len
;
2216 unsigned uninitialized
= 0;
2217 struct ext4_extent
*ex
;
2219 /* the header must be checked already in ext4_ext_remove_space() */
2220 ext_debug("truncate since %u in leaf\n", start
);
2221 if (!path
[depth
].p_hdr
)
2222 path
[depth
].p_hdr
= ext_block_hdr(path
[depth
].p_bh
);
2223 eh
= path
[depth
].p_hdr
;
2224 if (unlikely(path
[depth
].p_hdr
== NULL
)) {
2225 EXT4_ERROR_INODE(inode
, "path[%d].p_hdr == NULL", depth
);
2228 /* find where to start removing */
2229 ex
= EXT_LAST_EXTENT(eh
);
2231 ex_ee_block
= le32_to_cpu(ex
->ee_block
);
2232 ex_ee_len
= ext4_ext_get_actual_len(ex
);
2234 while (ex
>= EXT_FIRST_EXTENT(eh
) &&
2235 ex_ee_block
+ ex_ee_len
> start
) {
2237 if (ext4_ext_is_uninitialized(ex
))
2242 ext_debug("remove ext %u:[%d]%d\n", ex_ee_block
,
2243 uninitialized
, ex_ee_len
);
2244 path
[depth
].p_ext
= ex
;
2246 a
= ex_ee_block
> start
? ex_ee_block
: start
;
2247 b
= ex_ee_block
+ ex_ee_len
- 1 < EXT_MAX_BLOCK
?
2248 ex_ee_block
+ ex_ee_len
- 1 : EXT_MAX_BLOCK
;
2250 ext_debug(" border %u:%u\n", a
, b
);
2252 if (a
!= ex_ee_block
&& b
!= ex_ee_block
+ ex_ee_len
- 1) {
2256 } else if (a
!= ex_ee_block
) {
2257 /* remove tail of the extent */
2258 block
= ex_ee_block
;
2260 } else if (b
!= ex_ee_block
+ ex_ee_len
- 1) {
2261 /* remove head of the extent */
2264 /* there is no "make a hole" API yet */
2267 /* remove whole extent: excellent! */
2268 block
= ex_ee_block
;
2270 BUG_ON(a
!= ex_ee_block
);
2271 BUG_ON(b
!= ex_ee_block
+ ex_ee_len
- 1);
2275 * 3 for leaf, sb, and inode plus 2 (bmap and group
2276 * descriptor) for each block group; assume two block
2277 * groups plus ex_ee_len/blocks_per_block_group for
2280 credits
= 7 + 2*(ex_ee_len
/EXT4_BLOCKS_PER_GROUP(inode
->i_sb
));
2281 if (ex
== EXT_FIRST_EXTENT(eh
)) {
2283 credits
+= (ext_depth(inode
)) + 1;
2285 credits
+= EXT4_MAXQUOTAS_TRANS_BLOCKS(inode
->i_sb
);
2287 err
= ext4_ext_truncate_extend_restart(handle
, inode
, credits
);
2291 err
= ext4_ext_get_access(handle
, inode
, path
+ depth
);
2295 err
= ext4_remove_blocks(handle
, inode
, ex
, a
, b
);
2300 /* this extent is removed; mark slot entirely unused */
2301 ext4_ext_store_pblock(ex
, 0);
2302 le16_add_cpu(&eh
->eh_entries
, -1);
2305 ex
->ee_block
= cpu_to_le32(block
);
2306 ex
->ee_len
= cpu_to_le16(num
);
2308 * Do not mark uninitialized if all the blocks in the
2309 * extent have been removed.
2311 if (uninitialized
&& num
)
2312 ext4_ext_mark_uninitialized(ex
);
2314 err
= ext4_ext_dirty(handle
, inode
, path
+ depth
);
2318 ext_debug("new extent: %u:%u:%llu\n", block
, num
,
2321 ex_ee_block
= le32_to_cpu(ex
->ee_block
);
2322 ex_ee_len
= ext4_ext_get_actual_len(ex
);
2325 if (correct_index
&& eh
->eh_entries
)
2326 err
= ext4_ext_correct_indexes(handle
, inode
, path
);
2328 /* if this leaf is free, then we should
2329 * remove it from index block above */
2330 if (err
== 0 && eh
->eh_entries
== 0 && path
[depth
].p_bh
!= NULL
)
2331 err
= ext4_ext_rm_idx(handle
, inode
, path
+ depth
);
2338 * ext4_ext_more_to_rm:
2339 * returns 1 if current index has to be freed (even partial)
2342 ext4_ext_more_to_rm(struct ext4_ext_path
*path
)
2344 BUG_ON(path
->p_idx
== NULL
);
2346 if (path
->p_idx
< EXT_FIRST_INDEX(path
->p_hdr
))
2350 * if truncate on deeper level happened, it wasn't partial,
2351 * so we have to consider current index for truncation
2353 if (le16_to_cpu(path
->p_hdr
->eh_entries
) == path
->p_block
)
2358 static int ext4_ext_remove_space(struct inode
*inode
, ext4_lblk_t start
)
2360 struct super_block
*sb
= inode
->i_sb
;
2361 int depth
= ext_depth(inode
);
2362 struct ext4_ext_path
*path
;
2366 ext_debug("truncate since %u\n", start
);
2368 /* probably first extent we're gonna free will be last in block */
2369 handle
= ext4_journal_start(inode
, depth
+ 1);
2371 return PTR_ERR(handle
);
2373 ext4_ext_invalidate_cache(inode
);
2376 * We start scanning from right side, freeing all the blocks
2377 * after i_size and walking into the tree depth-wise.
2379 path
= kzalloc(sizeof(struct ext4_ext_path
) * (depth
+ 1), GFP_NOFS
);
2381 ext4_journal_stop(handle
);
2384 path
[0].p_hdr
= ext_inode_hdr(inode
);
2385 if (ext4_ext_check(inode
, path
[0].p_hdr
, depth
)) {
2389 path
[0].p_depth
= depth
;
2391 while (i
>= 0 && err
== 0) {
2393 /* this is leaf block */
2394 err
= ext4_ext_rm_leaf(handle
, inode
, path
, start
);
2395 /* root level has p_bh == NULL, brelse() eats this */
2396 brelse(path
[i
].p_bh
);
2397 path
[i
].p_bh
= NULL
;
2402 /* this is index block */
2403 if (!path
[i
].p_hdr
) {
2404 ext_debug("initialize header\n");
2405 path
[i
].p_hdr
= ext_block_hdr(path
[i
].p_bh
);
2408 if (!path
[i
].p_idx
) {
2409 /* this level hasn't been touched yet */
2410 path
[i
].p_idx
= EXT_LAST_INDEX(path
[i
].p_hdr
);
2411 path
[i
].p_block
= le16_to_cpu(path
[i
].p_hdr
->eh_entries
)+1;
2412 ext_debug("init index ptr: hdr 0x%p, num %d\n",
2414 le16_to_cpu(path
[i
].p_hdr
->eh_entries
));
2416 /* we were already here, see at next index */
2420 ext_debug("level %d - index, first 0x%p, cur 0x%p\n",
2421 i
, EXT_FIRST_INDEX(path
[i
].p_hdr
),
2423 if (ext4_ext_more_to_rm(path
+ i
)) {
2424 struct buffer_head
*bh
;
2425 /* go to the next level */
2426 ext_debug("move to level %d (block %llu)\n",
2427 i
+ 1, idx_pblock(path
[i
].p_idx
));
2428 memset(path
+ i
+ 1, 0, sizeof(*path
));
2429 bh
= sb_bread(sb
, idx_pblock(path
[i
].p_idx
));
2431 /* should we reset i_size? */
2435 if (WARN_ON(i
+ 1 > depth
)) {
2439 if (ext4_ext_check(inode
, ext_block_hdr(bh
),
2444 path
[i
+ 1].p_bh
= bh
;
2446 /* save actual number of indexes since this
2447 * number is changed at the next iteration */
2448 path
[i
].p_block
= le16_to_cpu(path
[i
].p_hdr
->eh_entries
);
2451 /* we finished processing this index, go up */
2452 if (path
[i
].p_hdr
->eh_entries
== 0 && i
> 0) {
2453 /* index is empty, remove it;
2454 * handle must be already prepared by the
2455 * truncatei_leaf() */
2456 err
= ext4_ext_rm_idx(handle
, inode
, path
+ i
);
2458 /* root level has p_bh == NULL, brelse() eats this */
2459 brelse(path
[i
].p_bh
);
2460 path
[i
].p_bh
= NULL
;
2462 ext_debug("return to level %d\n", i
);
2466 /* TODO: flexible tree reduction should be here */
2467 if (path
->p_hdr
->eh_entries
== 0) {
2469 * truncate to zero freed all the tree,
2470 * so we need to correct eh_depth
2472 err
= ext4_ext_get_access(handle
, inode
, path
);
2474 ext_inode_hdr(inode
)->eh_depth
= 0;
2475 ext_inode_hdr(inode
)->eh_max
=
2476 cpu_to_le16(ext4_ext_space_root(inode
, 0));
2477 err
= ext4_ext_dirty(handle
, inode
, path
);
2481 ext4_ext_drop_refs(path
);
2483 ext4_journal_stop(handle
);
2489 * called at mount time
2491 void ext4_ext_init(struct super_block
*sb
)
2494 * possible initialization would be here
2497 if (EXT4_HAS_INCOMPAT_FEATURE(sb
, EXT4_FEATURE_INCOMPAT_EXTENTS
)) {
2498 #if defined(AGGRESSIVE_TEST) || defined(CHECK_BINSEARCH) || defined(EXTENTS_STATS)
2499 printk(KERN_INFO
"EXT4-fs: file extents enabled");
2500 #ifdef AGGRESSIVE_TEST
2501 printk(", aggressive tests");
2503 #ifdef CHECK_BINSEARCH
2504 printk(", check binsearch");
2506 #ifdef EXTENTS_STATS
2511 #ifdef EXTENTS_STATS
2512 spin_lock_init(&EXT4_SB(sb
)->s_ext_stats_lock
);
2513 EXT4_SB(sb
)->s_ext_min
= 1 << 30;
2514 EXT4_SB(sb
)->s_ext_max
= 0;
2520 * called at umount time
2522 void ext4_ext_release(struct super_block
*sb
)
2524 if (!EXT4_HAS_INCOMPAT_FEATURE(sb
, EXT4_FEATURE_INCOMPAT_EXTENTS
))
2527 #ifdef EXTENTS_STATS
2528 if (EXT4_SB(sb
)->s_ext_blocks
&& EXT4_SB(sb
)->s_ext_extents
) {
2529 struct ext4_sb_info
*sbi
= EXT4_SB(sb
);
2530 printk(KERN_ERR
"EXT4-fs: %lu blocks in %lu extents (%lu ave)\n",
2531 sbi
->s_ext_blocks
, sbi
->s_ext_extents
,
2532 sbi
->s_ext_blocks
/ sbi
->s_ext_extents
);
2533 printk(KERN_ERR
"EXT4-fs: extents: %lu min, %lu max, max depth %lu\n",
2534 sbi
->s_ext_min
, sbi
->s_ext_max
, sbi
->s_depth_max
);
2539 static void bi_complete(struct bio
*bio
, int error
)
2541 complete((struct completion
*)bio
->bi_private
);
2544 /* FIXME!! we need to try to merge to left or right after zero-out */
2545 static int ext4_ext_zeroout(struct inode
*inode
, struct ext4_extent
*ex
)
2549 int blkbits
, blocksize
;
2551 struct completion event
;
2552 unsigned int ee_len
, len
, done
, offset
;
2555 blkbits
= inode
->i_blkbits
;
2556 blocksize
= inode
->i_sb
->s_blocksize
;
2557 ee_len
= ext4_ext_get_actual_len(ex
);
2558 ee_pblock
= ext_pblock(ex
);
2560 /* convert ee_pblock to 512 byte sectors */
2561 ee_pblock
= ee_pblock
<< (blkbits
- 9);
2563 while (ee_len
> 0) {
2565 if (ee_len
> BIO_MAX_PAGES
)
2566 len
= BIO_MAX_PAGES
;
2570 bio
= bio_alloc(GFP_NOIO
, len
);
2574 bio
->bi_sector
= ee_pblock
;
2575 bio
->bi_bdev
= inode
->i_sb
->s_bdev
;
2579 while (done
< len
) {
2580 ret
= bio_add_page(bio
, ZERO_PAGE(0),
2582 if (ret
!= blocksize
) {
2584 * We can't add any more pages because of
2585 * hardware limitations. Start a new bio.
2590 offset
+= blocksize
;
2591 if (offset
>= PAGE_CACHE_SIZE
)
2595 init_completion(&event
);
2596 bio
->bi_private
= &event
;
2597 bio
->bi_end_io
= bi_complete
;
2598 submit_bio(WRITE
, bio
);
2599 wait_for_completion(&event
);
2601 if (!test_bit(BIO_UPTODATE
, &bio
->bi_flags
)) {
2607 ee_pblock
+= done
<< (blkbits
- 9);
2612 #define EXT4_EXT_ZERO_LEN 7
2614 * This function is called by ext4_ext_map_blocks() if someone tries to write
2615 * to an uninitialized extent. It may result in splitting the uninitialized
2616 * extent into multiple extents (upto three - one initialized and two
2618 * There are three possibilities:
2619 * a> There is no split required: Entire extent should be initialized
2620 * b> Splits in two extents: Write is happening at either end of the extent
2621 * c> Splits in three extents: Somone is writing in middle of the extent
2623 static int ext4_ext_convert_to_initialized(handle_t
*handle
,
2624 struct inode
*inode
,
2625 struct ext4_map_blocks
*map
,
2626 struct ext4_ext_path
*path
)
2628 struct ext4_extent
*ex
, newex
, orig_ex
;
2629 struct ext4_extent
*ex1
= NULL
;
2630 struct ext4_extent
*ex2
= NULL
;
2631 struct ext4_extent
*ex3
= NULL
;
2632 struct ext4_extent_header
*eh
;
2633 ext4_lblk_t ee_block
, eof_block
;
2634 unsigned int allocated
, ee_len
, depth
;
2635 ext4_fsblk_t newblock
;
2640 ext_debug("ext4_ext_convert_to_initialized: inode %lu, logical"
2641 "block %llu, max_blocks %u\n", inode
->i_ino
,
2642 (unsigned long long)map
->m_lblk
, map
->m_len
);
2644 eof_block
= (inode
->i_size
+ inode
->i_sb
->s_blocksize
- 1) >>
2645 inode
->i_sb
->s_blocksize_bits
;
2646 if (eof_block
< map
->m_lblk
+ map
->m_len
)
2647 eof_block
= map
->m_lblk
+ map
->m_len
;
2649 depth
= ext_depth(inode
);
2650 eh
= path
[depth
].p_hdr
;
2651 ex
= path
[depth
].p_ext
;
2652 ee_block
= le32_to_cpu(ex
->ee_block
);
2653 ee_len
= ext4_ext_get_actual_len(ex
);
2654 allocated
= ee_len
- (map
->m_lblk
- ee_block
);
2655 newblock
= map
->m_lblk
- ee_block
+ ext_pblock(ex
);
2658 orig_ex
.ee_block
= ex
->ee_block
;
2659 orig_ex
.ee_len
= cpu_to_le16(ee_len
);
2660 ext4_ext_store_pblock(&orig_ex
, ext_pblock(ex
));
2663 * It is safe to convert extent to initialized via explicit
2664 * zeroout only if extent is fully insde i_size or new_size.
2666 may_zeroout
= ee_block
+ ee_len
<= eof_block
;
2668 err
= ext4_ext_get_access(handle
, inode
, path
+ depth
);
2671 /* If extent has less than 2*EXT4_EXT_ZERO_LEN zerout directly */
2672 if (ee_len
<= 2*EXT4_EXT_ZERO_LEN
&& may_zeroout
) {
2673 err
= ext4_ext_zeroout(inode
, &orig_ex
);
2675 goto fix_extent_len
;
2676 /* update the extent length and mark as initialized */
2677 ex
->ee_block
= orig_ex
.ee_block
;
2678 ex
->ee_len
= orig_ex
.ee_len
;
2679 ext4_ext_store_pblock(ex
, ext_pblock(&orig_ex
));
2680 ext4_ext_dirty(handle
, inode
, path
+ depth
);
2681 /* zeroed the full extent */
2685 /* ex1: ee_block to map->m_lblk - 1 : uninitialized */
2686 if (map
->m_lblk
> ee_block
) {
2688 ex1
->ee_len
= cpu_to_le16(map
->m_lblk
- ee_block
);
2689 ext4_ext_mark_uninitialized(ex1
);
2693 * for sanity, update the length of the ex2 extent before
2694 * we insert ex3, if ex1 is NULL. This is to avoid temporary
2695 * overlap of blocks.
2697 if (!ex1
&& allocated
> map
->m_len
)
2698 ex2
->ee_len
= cpu_to_le16(map
->m_len
);
2699 /* ex3: to ee_block + ee_len : uninitialised */
2700 if (allocated
> map
->m_len
) {
2701 unsigned int newdepth
;
2702 /* If extent has less than EXT4_EXT_ZERO_LEN zerout directly */
2703 if (allocated
<= EXT4_EXT_ZERO_LEN
&& may_zeroout
) {
2705 * map->m_lblk == ee_block is handled by the zerouout
2707 * Mark first half uninitialized.
2708 * Mark second half initialized and zero out the
2709 * initialized extent
2711 ex
->ee_block
= orig_ex
.ee_block
;
2712 ex
->ee_len
= cpu_to_le16(ee_len
- allocated
);
2713 ext4_ext_mark_uninitialized(ex
);
2714 ext4_ext_store_pblock(ex
, ext_pblock(&orig_ex
));
2715 ext4_ext_dirty(handle
, inode
, path
+ depth
);
2718 ex3
->ee_block
= cpu_to_le32(map
->m_lblk
);
2719 ext4_ext_store_pblock(ex3
, newblock
);
2720 ex3
->ee_len
= cpu_to_le16(allocated
);
2721 err
= ext4_ext_insert_extent(handle
, inode
, path
,
2723 if (err
== -ENOSPC
) {
2724 err
= ext4_ext_zeroout(inode
, &orig_ex
);
2726 goto fix_extent_len
;
2727 ex
->ee_block
= orig_ex
.ee_block
;
2728 ex
->ee_len
= orig_ex
.ee_len
;
2729 ext4_ext_store_pblock(ex
, ext_pblock(&orig_ex
));
2730 ext4_ext_dirty(handle
, inode
, path
+ depth
);
2731 /* blocks available from map->m_lblk */
2735 goto fix_extent_len
;
2738 * We need to zero out the second half because
2739 * an fallocate request can update file size and
2740 * converting the second half to initialized extent
2741 * implies that we can leak some junk data to user
2744 err
= ext4_ext_zeroout(inode
, ex3
);
2747 * We should actually mark the
2748 * second half as uninit and return error
2749 * Insert would have changed the extent
2751 depth
= ext_depth(inode
);
2752 ext4_ext_drop_refs(path
);
2753 path
= ext4_ext_find_extent(inode
, map
->m_lblk
,
2756 err
= PTR_ERR(path
);
2759 /* get the second half extent details */
2760 ex
= path
[depth
].p_ext
;
2761 err
= ext4_ext_get_access(handle
, inode
,
2765 ext4_ext_mark_uninitialized(ex
);
2766 ext4_ext_dirty(handle
, inode
, path
+ depth
);
2770 /* zeroed the second half */
2774 ex3
->ee_block
= cpu_to_le32(map
->m_lblk
+ map
->m_len
);
2775 ext4_ext_store_pblock(ex3
, newblock
+ map
->m_len
);
2776 ex3
->ee_len
= cpu_to_le16(allocated
- map
->m_len
);
2777 ext4_ext_mark_uninitialized(ex3
);
2778 err
= ext4_ext_insert_extent(handle
, inode
, path
, ex3
, 0);
2779 if (err
== -ENOSPC
&& may_zeroout
) {
2780 err
= ext4_ext_zeroout(inode
, &orig_ex
);
2782 goto fix_extent_len
;
2783 /* update the extent length and mark as initialized */
2784 ex
->ee_block
= orig_ex
.ee_block
;
2785 ex
->ee_len
= orig_ex
.ee_len
;
2786 ext4_ext_store_pblock(ex
, ext_pblock(&orig_ex
));
2787 ext4_ext_dirty(handle
, inode
, path
+ depth
);
2788 /* zeroed the full extent */
2789 /* blocks available from map->m_lblk */
2793 goto fix_extent_len
;
2795 * The depth, and hence eh & ex might change
2796 * as part of the insert above.
2798 newdepth
= ext_depth(inode
);
2800 * update the extent length after successful insert of the
2803 ee_len
-= ext4_ext_get_actual_len(ex3
);
2804 orig_ex
.ee_len
= cpu_to_le16(ee_len
);
2805 may_zeroout
= ee_block
+ ee_len
<= eof_block
;
2808 ext4_ext_drop_refs(path
);
2809 path
= ext4_ext_find_extent(inode
, map
->m_lblk
, path
);
2811 err
= PTR_ERR(path
);
2814 eh
= path
[depth
].p_hdr
;
2815 ex
= path
[depth
].p_ext
;
2819 err
= ext4_ext_get_access(handle
, inode
, path
+ depth
);
2823 allocated
= map
->m_len
;
2825 /* If extent has less than EXT4_EXT_ZERO_LEN and we are trying
2826 * to insert a extent in the middle zerout directly
2827 * otherwise give the extent a chance to merge to left
2829 if (le16_to_cpu(orig_ex
.ee_len
) <= EXT4_EXT_ZERO_LEN
&&
2830 map
->m_lblk
!= ee_block
&& may_zeroout
) {
2831 err
= ext4_ext_zeroout(inode
, &orig_ex
);
2833 goto fix_extent_len
;
2834 /* update the extent length and mark as initialized */
2835 ex
->ee_block
= orig_ex
.ee_block
;
2836 ex
->ee_len
= orig_ex
.ee_len
;
2837 ext4_ext_store_pblock(ex
, ext_pblock(&orig_ex
));
2838 ext4_ext_dirty(handle
, inode
, path
+ depth
);
2839 /* zero out the first half */
2840 /* blocks available from map->m_lblk */
2845 * If there was a change of depth as part of the
2846 * insertion of ex3 above, we need to update the length
2847 * of the ex1 extent again here
2849 if (ex1
&& ex1
!= ex
) {
2851 ex1
->ee_len
= cpu_to_le16(map
->m_lblk
- ee_block
);
2852 ext4_ext_mark_uninitialized(ex1
);
2855 /* ex2: map->m_lblk to map->m_lblk + maxblocks-1 : initialised */
2856 ex2
->ee_block
= cpu_to_le32(map
->m_lblk
);
2857 ext4_ext_store_pblock(ex2
, newblock
);
2858 ex2
->ee_len
= cpu_to_le16(allocated
);
2862 * New (initialized) extent starts from the first block
2863 * in the current extent. i.e., ex2 == ex
2864 * We have to see if it can be merged with the extent
2867 if (ex2
> EXT_FIRST_EXTENT(eh
)) {
2869 * To merge left, pass "ex2 - 1" to try_to_merge(),
2870 * since it merges towards right _only_.
2872 ret
= ext4_ext_try_to_merge(inode
, path
, ex2
- 1);
2874 err
= ext4_ext_correct_indexes(handle
, inode
, path
);
2877 depth
= ext_depth(inode
);
2882 * Try to Merge towards right. This might be required
2883 * only when the whole extent is being written to.
2884 * i.e. ex2 == ex and ex3 == NULL.
2887 ret
= ext4_ext_try_to_merge(inode
, path
, ex2
);
2889 err
= ext4_ext_correct_indexes(handle
, inode
, path
);
2894 /* Mark modified extent as dirty */
2895 err
= ext4_ext_dirty(handle
, inode
, path
+ depth
);
2898 err
= ext4_ext_insert_extent(handle
, inode
, path
, &newex
, 0);
2899 if (err
== -ENOSPC
&& may_zeroout
) {
2900 err
= ext4_ext_zeroout(inode
, &orig_ex
);
2902 goto fix_extent_len
;
2903 /* update the extent length and mark as initialized */
2904 ex
->ee_block
= orig_ex
.ee_block
;
2905 ex
->ee_len
= orig_ex
.ee_len
;
2906 ext4_ext_store_pblock(ex
, ext_pblock(&orig_ex
));
2907 ext4_ext_dirty(handle
, inode
, path
+ depth
);
2908 /* zero out the first half */
2911 goto fix_extent_len
;
2913 ext4_ext_show_leaf(inode
, path
);
2914 return err
? err
: allocated
;
2917 ex
->ee_block
= orig_ex
.ee_block
;
2918 ex
->ee_len
= orig_ex
.ee_len
;
2919 ext4_ext_store_pblock(ex
, ext_pblock(&orig_ex
));
2920 ext4_ext_mark_uninitialized(ex
);
2921 ext4_ext_dirty(handle
, inode
, path
+ depth
);
2926 * This function is called by ext4_ext_map_blocks() from
2927 * ext4_get_blocks_dio_write() when DIO to write
2928 * to an uninitialized extent.
2930 * Writing to an uninitized extent may result in splitting the uninitialized
2931 * extent into multiple /intialized unintialized extents (up to three)
2932 * There are three possibilities:
2933 * a> There is no split required: Entire extent should be uninitialized
2934 * b> Splits in two extents: Write is happening at either end of the extent
2935 * c> Splits in three extents: Somone is writing in middle of the extent
2937 * One of more index blocks maybe needed if the extent tree grow after
2938 * the unintialized extent split. To prevent ENOSPC occur at the IO
2939 * complete, we need to split the uninitialized extent before DIO submit
2940 * the IO. The uninitilized extent called at this time will be split
2941 * into three uninitialized extent(at most). After IO complete, the part
2942 * being filled will be convert to initialized by the end_io callback function
2943 * via ext4_convert_unwritten_extents().
2945 * Returns the size of uninitialized extent to be written on success.
2947 static int ext4_split_unwritten_extents(handle_t
*handle
,
2948 struct inode
*inode
,
2949 struct ext4_map_blocks
*map
,
2950 struct ext4_ext_path
*path
,
2953 struct ext4_extent
*ex
, newex
, orig_ex
;
2954 struct ext4_extent
*ex1
= NULL
;
2955 struct ext4_extent
*ex2
= NULL
;
2956 struct ext4_extent
*ex3
= NULL
;
2957 struct ext4_extent_header
*eh
;
2958 ext4_lblk_t ee_block
, eof_block
;
2959 unsigned int allocated
, ee_len
, depth
;
2960 ext4_fsblk_t newblock
;
2964 ext_debug("ext4_split_unwritten_extents: inode %lu, logical"
2965 "block %llu, max_blocks %u\n", inode
->i_ino
,
2966 (unsigned long long)map
->m_lblk
, map
->m_len
);
2968 eof_block
= (inode
->i_size
+ inode
->i_sb
->s_blocksize
- 1) >>
2969 inode
->i_sb
->s_blocksize_bits
;
2970 if (eof_block
< map
->m_lblk
+ map
->m_len
)
2971 eof_block
= map
->m_lblk
+ map
->m_len
;
2973 depth
= ext_depth(inode
);
2974 eh
= path
[depth
].p_hdr
;
2975 ex
= path
[depth
].p_ext
;
2976 ee_block
= le32_to_cpu(ex
->ee_block
);
2977 ee_len
= ext4_ext_get_actual_len(ex
);
2978 allocated
= ee_len
- (map
->m_lblk
- ee_block
);
2979 newblock
= map
->m_lblk
- ee_block
+ ext_pblock(ex
);
2982 orig_ex
.ee_block
= ex
->ee_block
;
2983 orig_ex
.ee_len
= cpu_to_le16(ee_len
);
2984 ext4_ext_store_pblock(&orig_ex
, ext_pblock(ex
));
2987 * It is safe to convert extent to initialized via explicit
2988 * zeroout only if extent is fully insde i_size or new_size.
2990 may_zeroout
= ee_block
+ ee_len
<= eof_block
;
2993 * If the uninitialized extent begins at the same logical
2994 * block where the write begins, and the write completely
2995 * covers the extent, then we don't need to split it.
2997 if ((map
->m_lblk
== ee_block
) && (allocated
<= map
->m_len
))
3000 err
= ext4_ext_get_access(handle
, inode
, path
+ depth
);
3003 /* ex1: ee_block to map->m_lblk - 1 : uninitialized */
3004 if (map
->m_lblk
> ee_block
) {
3006 ex1
->ee_len
= cpu_to_le16(map
->m_lblk
- ee_block
);
3007 ext4_ext_mark_uninitialized(ex1
);
3011 * for sanity, update the length of the ex2 extent before
3012 * we insert ex3, if ex1 is NULL. This is to avoid temporary
3013 * overlap of blocks.
3015 if (!ex1
&& allocated
> map
->m_len
)
3016 ex2
->ee_len
= cpu_to_le16(map
->m_len
);
3017 /* ex3: to ee_block + ee_len : uninitialised */
3018 if (allocated
> map
->m_len
) {
3019 unsigned int newdepth
;
3021 ex3
->ee_block
= cpu_to_le32(map
->m_lblk
+ map
->m_len
);
3022 ext4_ext_store_pblock(ex3
, newblock
+ map
->m_len
);
3023 ex3
->ee_len
= cpu_to_le16(allocated
- map
->m_len
);
3024 ext4_ext_mark_uninitialized(ex3
);
3025 err
= ext4_ext_insert_extent(handle
, inode
, path
, ex3
, flags
);
3026 if (err
== -ENOSPC
&& may_zeroout
) {
3027 err
= ext4_ext_zeroout(inode
, &orig_ex
);
3029 goto fix_extent_len
;
3030 /* update the extent length and mark as initialized */
3031 ex
->ee_block
= orig_ex
.ee_block
;
3032 ex
->ee_len
= orig_ex
.ee_len
;
3033 ext4_ext_store_pblock(ex
, ext_pblock(&orig_ex
));
3034 ext4_ext_dirty(handle
, inode
, path
+ depth
);
3035 /* zeroed the full extent */
3036 /* blocks available from map->m_lblk */
3040 goto fix_extent_len
;
3042 * The depth, and hence eh & ex might change
3043 * as part of the insert above.
3045 newdepth
= ext_depth(inode
);
3047 * update the extent length after successful insert of the
3050 ee_len
-= ext4_ext_get_actual_len(ex3
);
3051 orig_ex
.ee_len
= cpu_to_le16(ee_len
);
3052 may_zeroout
= ee_block
+ ee_len
<= eof_block
;
3055 ext4_ext_drop_refs(path
);
3056 path
= ext4_ext_find_extent(inode
, map
->m_lblk
, path
);
3058 err
= PTR_ERR(path
);
3061 eh
= path
[depth
].p_hdr
;
3062 ex
= path
[depth
].p_ext
;
3066 err
= ext4_ext_get_access(handle
, inode
, path
+ depth
);
3070 allocated
= map
->m_len
;
3073 * If there was a change of depth as part of the
3074 * insertion of ex3 above, we need to update the length
3075 * of the ex1 extent again here
3077 if (ex1
&& ex1
!= ex
) {
3079 ex1
->ee_len
= cpu_to_le16(map
->m_lblk
- ee_block
);
3080 ext4_ext_mark_uninitialized(ex1
);
3084 * ex2: map->m_lblk to map->m_lblk + map->m_len-1 : to be written
3085 * using direct I/O, uninitialised still.
3087 ex2
->ee_block
= cpu_to_le32(map
->m_lblk
);
3088 ext4_ext_store_pblock(ex2
, newblock
);
3089 ex2
->ee_len
= cpu_to_le16(allocated
);
3090 ext4_ext_mark_uninitialized(ex2
);
3093 /* Mark modified extent as dirty */
3094 err
= ext4_ext_dirty(handle
, inode
, path
+ depth
);
3095 ext_debug("out here\n");
3098 err
= ext4_ext_insert_extent(handle
, inode
, path
, &newex
, flags
);
3099 if (err
== -ENOSPC
&& may_zeroout
) {
3100 err
= ext4_ext_zeroout(inode
, &orig_ex
);
3102 goto fix_extent_len
;
3103 /* update the extent length and mark as initialized */
3104 ex
->ee_block
= orig_ex
.ee_block
;
3105 ex
->ee_len
= orig_ex
.ee_len
;
3106 ext4_ext_store_pblock(ex
, ext_pblock(&orig_ex
));
3107 ext4_ext_dirty(handle
, inode
, path
+ depth
);
3108 /* zero out the first half */
3111 goto fix_extent_len
;
3113 ext4_ext_show_leaf(inode
, path
);
3114 return err
? err
: allocated
;
3117 ex
->ee_block
= orig_ex
.ee_block
;
3118 ex
->ee_len
= orig_ex
.ee_len
;
3119 ext4_ext_store_pblock(ex
, ext_pblock(&orig_ex
));
3120 ext4_ext_mark_uninitialized(ex
);
3121 ext4_ext_dirty(handle
, inode
, path
+ depth
);
3124 static int ext4_convert_unwritten_extents_endio(handle_t
*handle
,
3125 struct inode
*inode
,
3126 struct ext4_ext_path
*path
)
3128 struct ext4_extent
*ex
;
3129 struct ext4_extent_header
*eh
;
3134 depth
= ext_depth(inode
);
3135 eh
= path
[depth
].p_hdr
;
3136 ex
= path
[depth
].p_ext
;
3138 err
= ext4_ext_get_access(handle
, inode
, path
+ depth
);
3141 /* first mark the extent as initialized */
3142 ext4_ext_mark_initialized(ex
);
3145 * We have to see if it can be merged with the extent
3148 if (ex
> EXT_FIRST_EXTENT(eh
)) {
3150 * To merge left, pass "ex - 1" to try_to_merge(),
3151 * since it merges towards right _only_.
3153 ret
= ext4_ext_try_to_merge(inode
, path
, ex
- 1);
3155 err
= ext4_ext_correct_indexes(handle
, inode
, path
);
3158 depth
= ext_depth(inode
);
3163 * Try to Merge towards right.
3165 ret
= ext4_ext_try_to_merge(inode
, path
, ex
);
3167 err
= ext4_ext_correct_indexes(handle
, inode
, path
);
3170 depth
= ext_depth(inode
);
3172 /* Mark modified extent as dirty */
3173 err
= ext4_ext_dirty(handle
, inode
, path
+ depth
);
3175 ext4_ext_show_leaf(inode
, path
);
3179 static void unmap_underlying_metadata_blocks(struct block_device
*bdev
,
3180 sector_t block
, int count
)
3183 for (i
= 0; i
< count
; i
++)
3184 unmap_underlying_metadata(bdev
, block
+ i
);
3188 ext4_ext_handle_uninitialized_extents(handle_t
*handle
, struct inode
*inode
,
3189 struct ext4_map_blocks
*map
,
3190 struct ext4_ext_path
*path
, int flags
,
3191 unsigned int allocated
, ext4_fsblk_t newblock
)
3195 ext4_io_end_t
*io
= EXT4_I(inode
)->cur_aio_dio
;
3197 ext_debug("ext4_ext_handle_uninitialized_extents: inode %lu, logical"
3198 "block %llu, max_blocks %u, flags %d, allocated %u",
3199 inode
->i_ino
, (unsigned long long)map
->m_lblk
, map
->m_len
,
3201 ext4_ext_show_leaf(inode
, path
);
3203 /* get_block() before submit the IO, split the extent */
3204 if ((flags
& EXT4_GET_BLOCKS_PRE_IO
)) {
3205 ret
= ext4_split_unwritten_extents(handle
, inode
, map
,
3208 * Flag the inode(non aio case) or end_io struct (aio case)
3209 * that this IO needs to convertion to written when IO is
3213 io
->flag
= EXT4_IO_UNWRITTEN
;
3215 ext4_set_inode_state(inode
, EXT4_STATE_DIO_UNWRITTEN
);
3216 if (ext4_should_dioread_nolock(inode
))
3217 map
->m_flags
|= EXT4_MAP_UNINIT
;
3220 /* IO end_io complete, convert the filled extent to written */
3221 if ((flags
& EXT4_GET_BLOCKS_CONVERT
)) {
3222 ret
= ext4_convert_unwritten_extents_endio(handle
, inode
,
3225 ext4_update_inode_fsync_trans(handle
, inode
, 1);
3228 /* buffered IO case */
3230 * repeat fallocate creation request
3231 * we already have an unwritten extent
3233 if (flags
& EXT4_GET_BLOCKS_UNINIT_EXT
)
3236 /* buffered READ or buffered write_begin() lookup */
3237 if ((flags
& EXT4_GET_BLOCKS_CREATE
) == 0) {
3239 * We have blocks reserved already. We
3240 * return allocated blocks so that delalloc
3241 * won't do block reservation for us. But
3242 * the buffer head will be unmapped so that
3243 * a read from the block returns 0s.
3245 map
->m_flags
|= EXT4_MAP_UNWRITTEN
;
3249 /* buffered write, writepage time, convert*/
3250 ret
= ext4_ext_convert_to_initialized(handle
, inode
, map
, path
);
3252 ext4_update_inode_fsync_trans(handle
, inode
, 1);
3259 map
->m_flags
|= EXT4_MAP_NEW
;
3261 * if we allocated more blocks than requested
3262 * we need to make sure we unmap the extra block
3263 * allocated. The actual needed block will get
3264 * unmapped later when we find the buffer_head marked
3267 if (allocated
> map
->m_len
) {
3268 unmap_underlying_metadata_blocks(inode
->i_sb
->s_bdev
,
3269 newblock
+ map
->m_len
,
3270 allocated
- map
->m_len
);
3271 allocated
= map
->m_len
;
3275 * If we have done fallocate with the offset that is already
3276 * delayed allocated, we would have block reservation
3277 * and quota reservation done in the delayed write path.
3278 * But fallocate would have already updated quota and block
3279 * count for this offset. So cancel these reservation
3281 if (flags
& EXT4_GET_BLOCKS_DELALLOC_RESERVE
)
3282 ext4_da_update_reserve_space(inode
, allocated
, 0);
3285 map
->m_flags
|= EXT4_MAP_MAPPED
;
3287 if (allocated
> map
->m_len
)
3288 allocated
= map
->m_len
;
3289 ext4_ext_show_leaf(inode
, path
);
3290 map
->m_pblk
= newblock
;
3291 map
->m_len
= allocated
;
3294 ext4_ext_drop_refs(path
);
3297 return err
? err
: allocated
;
3300 * Block allocation/map/preallocation routine for extents based files
3303 * Need to be called with
3304 * down_read(&EXT4_I(inode)->i_data_sem) if not allocating file system block
3305 * (ie, create is zero). Otherwise down_write(&EXT4_I(inode)->i_data_sem)
3307 * return > 0, number of of blocks already mapped/allocated
3308 * if create == 0 and these are pre-allocated blocks
3309 * buffer head is unmapped
3310 * otherwise blocks are mapped
3312 * return = 0, if plain look up failed (blocks have not been allocated)
3313 * buffer head is unmapped
3315 * return < 0, error case.
3317 int ext4_ext_map_blocks(handle_t
*handle
, struct inode
*inode
,
3318 struct ext4_map_blocks
*map
, int flags
)
3320 struct ext4_ext_path
*path
= NULL
;
3321 struct ext4_extent_header
*eh
;
3322 struct ext4_extent newex
, *ex
, *last_ex
;
3323 ext4_fsblk_t newblock
;
3324 int err
= 0, depth
, ret
, cache_type
;
3325 unsigned int allocated
= 0;
3326 struct ext4_allocation_request ar
;
3327 ext4_io_end_t
*io
= EXT4_I(inode
)->cur_aio_dio
;
3329 ext_debug("blocks %u/%u requested for inode %lu\n",
3330 map
->m_lblk
, map
->m_len
, inode
->i_ino
);
3332 /* check in cache */
3333 cache_type
= ext4_ext_in_cache(inode
, map
->m_lblk
, &newex
);
3335 if (cache_type
== EXT4_EXT_CACHE_GAP
) {
3336 if ((flags
& EXT4_GET_BLOCKS_CREATE
) == 0) {
3338 * block isn't allocated yet and
3339 * user doesn't want to allocate it
3343 /* we should allocate requested block */
3344 } else if (cache_type
== EXT4_EXT_CACHE_EXTENT
) {
3345 /* block is already allocated */
3346 newblock
= map
->m_lblk
3347 - le32_to_cpu(newex
.ee_block
)
3348 + ext_pblock(&newex
);
3349 /* number of remaining blocks in the extent */
3350 allocated
= ext4_ext_get_actual_len(&newex
) -
3351 (map
->m_lblk
- le32_to_cpu(newex
.ee_block
));
3358 /* find extent for this block */
3359 path
= ext4_ext_find_extent(inode
, map
->m_lblk
, NULL
);
3361 err
= PTR_ERR(path
);
3366 depth
= ext_depth(inode
);
3369 * consistent leaf must not be empty;
3370 * this situation is possible, though, _during_ tree modification;
3371 * this is why assert can't be put in ext4_ext_find_extent()
3373 if (unlikely(path
[depth
].p_ext
== NULL
&& depth
!= 0)) {
3374 EXT4_ERROR_INODE(inode
, "bad extent address "
3375 "iblock: %d, depth: %d pblock %lld",
3376 map
->m_lblk
, depth
, path
[depth
].p_block
);
3380 eh
= path
[depth
].p_hdr
;
3382 ex
= path
[depth
].p_ext
;
3384 ext4_lblk_t ee_block
= le32_to_cpu(ex
->ee_block
);
3385 ext4_fsblk_t ee_start
= ext_pblock(ex
);
3386 unsigned short ee_len
;
3389 * Uninitialized extents are treated as holes, except that
3390 * we split out initialized portions during a write.
3392 ee_len
= ext4_ext_get_actual_len(ex
);
3393 /* if found extent covers block, simply return it */
3394 if (in_range(map
->m_lblk
, ee_block
, ee_len
)) {
3395 newblock
= map
->m_lblk
- ee_block
+ ee_start
;
3396 /* number of remaining blocks in the extent */
3397 allocated
= ee_len
- (map
->m_lblk
- ee_block
);
3398 ext_debug("%u fit into %u:%d -> %llu\n", map
->m_lblk
,
3399 ee_block
, ee_len
, newblock
);
3401 /* Do not put uninitialized extent in the cache */
3402 if (!ext4_ext_is_uninitialized(ex
)) {
3403 ext4_ext_put_in_cache(inode
, ee_block
,
3405 EXT4_EXT_CACHE_EXTENT
);
3408 ret
= ext4_ext_handle_uninitialized_extents(handle
,
3409 inode
, map
, path
, flags
, allocated
,
3416 * requested block isn't allocated yet;
3417 * we couldn't try to create block if create flag is zero
3419 if ((flags
& EXT4_GET_BLOCKS_CREATE
) == 0) {
3421 * put just found gap into cache to speed up
3422 * subsequent requests
3424 ext4_ext_put_gap_in_cache(inode
, path
, map
->m_lblk
);
3428 * Okay, we need to do block allocation.
3431 /* find neighbour allocated blocks */
3432 ar
.lleft
= map
->m_lblk
;
3433 err
= ext4_ext_search_left(inode
, path
, &ar
.lleft
, &ar
.pleft
);
3436 ar
.lright
= map
->m_lblk
;
3437 err
= ext4_ext_search_right(inode
, path
, &ar
.lright
, &ar
.pright
);
3442 * See if request is beyond maximum number of blocks we can have in
3443 * a single extent. For an initialized extent this limit is
3444 * EXT_INIT_MAX_LEN and for an uninitialized extent this limit is
3445 * EXT_UNINIT_MAX_LEN.
3447 if (map
->m_len
> EXT_INIT_MAX_LEN
&&
3448 !(flags
& EXT4_GET_BLOCKS_UNINIT_EXT
))
3449 map
->m_len
= EXT_INIT_MAX_LEN
;
3450 else if (map
->m_len
> EXT_UNINIT_MAX_LEN
&&
3451 (flags
& EXT4_GET_BLOCKS_UNINIT_EXT
))
3452 map
->m_len
= EXT_UNINIT_MAX_LEN
;
3454 /* Check if we can really insert (m_lblk)::(m_lblk + m_len) extent */
3455 newex
.ee_block
= cpu_to_le32(map
->m_lblk
);
3456 newex
.ee_len
= cpu_to_le16(map
->m_len
);
3457 err
= ext4_ext_check_overlap(inode
, &newex
, path
);
3459 allocated
= ext4_ext_get_actual_len(&newex
);
3461 allocated
= map
->m_len
;
3463 /* allocate new block */
3465 ar
.goal
= ext4_ext_find_goal(inode
, path
, map
->m_lblk
);
3466 ar
.logical
= map
->m_lblk
;
3468 if (S_ISREG(inode
->i_mode
))
3469 ar
.flags
= EXT4_MB_HINT_DATA
;
3471 /* disable in-core preallocation for non-regular files */
3473 newblock
= ext4_mb_new_blocks(handle
, &ar
, &err
);
3476 ext_debug("allocate new block: goal %llu, found %llu/%u\n",
3477 ar
.goal
, newblock
, allocated
);
3479 /* try to insert new extent into found leaf and return */
3480 ext4_ext_store_pblock(&newex
, newblock
);
3481 newex
.ee_len
= cpu_to_le16(ar
.len
);
3482 /* Mark uninitialized */
3483 if (flags
& EXT4_GET_BLOCKS_UNINIT_EXT
){
3484 ext4_ext_mark_uninitialized(&newex
);
3486 * io_end structure was created for every IO write to an
3487 * uninitialized extent. To avoid unecessary conversion,
3488 * here we flag the IO that really needs the conversion.
3489 * For non asycn direct IO case, flag the inode state
3490 * that we need to perform convertion when IO is done.
3492 if ((flags
& EXT4_GET_BLOCKS_PRE_IO
)) {
3494 io
->flag
= EXT4_IO_UNWRITTEN
;
3496 ext4_set_inode_state(inode
,
3497 EXT4_STATE_DIO_UNWRITTEN
);
3499 if (ext4_should_dioread_nolock(inode
))
3500 map
->m_flags
|= EXT4_MAP_UNINIT
;
3503 if (unlikely(EXT4_I(inode
)->i_flags
& EXT4_EOFBLOCKS_FL
)) {
3504 if (unlikely(!eh
->eh_entries
)) {
3505 EXT4_ERROR_INODE(inode
,
3506 "eh->eh_entries == 0 ee_block %d",
3511 last_ex
= EXT_LAST_EXTENT(eh
);
3512 if (map
->m_lblk
+ ar
.len
> le32_to_cpu(last_ex
->ee_block
)
3513 + ext4_ext_get_actual_len(last_ex
))
3514 EXT4_I(inode
)->i_flags
&= ~EXT4_EOFBLOCKS_FL
;
3516 err
= ext4_ext_insert_extent(handle
, inode
, path
, &newex
, flags
);
3518 /* free data blocks we just allocated */
3519 /* not a good idea to call discard here directly,
3520 * but otherwise we'd need to call it every free() */
3521 ext4_discard_preallocations(inode
);
3522 ext4_free_blocks(handle
, inode
, 0, ext_pblock(&newex
),
3523 ext4_ext_get_actual_len(&newex
), 0);
3527 /* previous routine could use block we allocated */
3528 newblock
= ext_pblock(&newex
);
3529 allocated
= ext4_ext_get_actual_len(&newex
);
3530 if (allocated
> map
->m_len
)
3531 allocated
= map
->m_len
;
3532 map
->m_flags
|= EXT4_MAP_NEW
;
3535 * Update reserved blocks/metadata blocks after successful
3536 * block allocation which had been deferred till now.
3538 if (flags
& EXT4_GET_BLOCKS_DELALLOC_RESERVE
)
3539 ext4_da_update_reserve_space(inode
, allocated
, 1);
3542 * Cache the extent and update transaction to commit on fdatasync only
3543 * when it is _not_ an uninitialized extent.
3545 if ((flags
& EXT4_GET_BLOCKS_UNINIT_EXT
) == 0) {
3546 ext4_ext_put_in_cache(inode
, map
->m_lblk
, allocated
, newblock
,
3547 EXT4_EXT_CACHE_EXTENT
);
3548 ext4_update_inode_fsync_trans(handle
, inode
, 1);
3550 ext4_update_inode_fsync_trans(handle
, inode
, 0);
3552 if (allocated
> map
->m_len
)
3553 allocated
= map
->m_len
;
3554 ext4_ext_show_leaf(inode
, path
);
3555 map
->m_flags
|= EXT4_MAP_MAPPED
;
3556 map
->m_pblk
= newblock
;
3557 map
->m_len
= allocated
;
3560 ext4_ext_drop_refs(path
);
3563 return err
? err
: allocated
;
3566 void ext4_ext_truncate(struct inode
*inode
)
3568 struct address_space
*mapping
= inode
->i_mapping
;
3569 struct super_block
*sb
= inode
->i_sb
;
3570 ext4_lblk_t last_block
;
3575 * probably first extent we're gonna free will be last in block
3577 err
= ext4_writepage_trans_blocks(inode
);
3578 handle
= ext4_journal_start(inode
, err
);
3582 if (inode
->i_size
& (sb
->s_blocksize
- 1))
3583 ext4_block_truncate_page(handle
, mapping
, inode
->i_size
);
3585 if (ext4_orphan_add(handle
, inode
))
3588 down_write(&EXT4_I(inode
)->i_data_sem
);
3589 ext4_ext_invalidate_cache(inode
);
3591 ext4_discard_preallocations(inode
);
3594 * TODO: optimization is possible here.
3595 * Probably we need not scan at all,
3596 * because page truncation is enough.
3599 /* we have to know where to truncate from in crash case */
3600 EXT4_I(inode
)->i_disksize
= inode
->i_size
;
3601 ext4_mark_inode_dirty(handle
, inode
);
3603 last_block
= (inode
->i_size
+ sb
->s_blocksize
- 1)
3604 >> EXT4_BLOCK_SIZE_BITS(sb
);
3605 err
= ext4_ext_remove_space(inode
, last_block
);
3607 /* In a multi-transaction truncate, we only make the final
3608 * transaction synchronous.
3611 ext4_handle_sync(handle
);
3614 up_write(&EXT4_I(inode
)->i_data_sem
);
3616 * If this was a simple ftruncate() and the file will remain alive,
3617 * then we need to clear up the orphan record which we created above.
3618 * However, if this was a real unlink then we were called by
3619 * ext4_delete_inode(), and we allow that function to clean up the
3620 * orphan info for us.
3623 ext4_orphan_del(handle
, inode
);
3625 inode
->i_mtime
= inode
->i_ctime
= ext4_current_time(inode
);
3626 ext4_mark_inode_dirty(handle
, inode
);
3627 ext4_journal_stop(handle
);
3630 static void ext4_falloc_update_inode(struct inode
*inode
,
3631 int mode
, loff_t new_size
, int update_ctime
)
3633 struct timespec now
;
3636 now
= current_fs_time(inode
->i_sb
);
3637 if (!timespec_equal(&inode
->i_ctime
, &now
))
3638 inode
->i_ctime
= now
;
3641 * Update only when preallocation was requested beyond
3644 if (!(mode
& FALLOC_FL_KEEP_SIZE
)) {
3645 if (new_size
> i_size_read(inode
))
3646 i_size_write(inode
, new_size
);
3647 if (new_size
> EXT4_I(inode
)->i_disksize
)
3648 ext4_update_i_disksize(inode
, new_size
);
3651 * Mark that we allocate beyond EOF so the subsequent truncate
3652 * can proceed even if the new size is the same as i_size.
3654 if (new_size
> i_size_read(inode
))
3655 EXT4_I(inode
)->i_flags
|= EXT4_EOFBLOCKS_FL
;
3661 * preallocate space for a file. This implements ext4's fallocate inode
3662 * operation, which gets called from sys_fallocate system call.
3663 * For block-mapped files, posix_fallocate should fall back to the method
3664 * of writing zeroes to the required new blocks (the same behavior which is
3665 * expected for file systems which do not support fallocate() system call).
3667 long ext4_fallocate(struct inode
*inode
, int mode
, loff_t offset
, loff_t len
)
3671 unsigned int max_blocks
;
3675 struct ext4_map_blocks map
;
3676 unsigned int credits
, blkbits
= inode
->i_blkbits
;
3679 * currently supporting (pre)allocate mode for extent-based
3682 if (!(EXT4_I(inode
)->i_flags
& EXT4_EXTENTS_FL
))
3685 /* preallocation to directories is currently not supported */
3686 if (S_ISDIR(inode
->i_mode
))
3689 map
.m_lblk
= offset
>> blkbits
;
3691 * We can't just convert len to max_blocks because
3692 * If blocksize = 4096 offset = 3072 and len = 2048
3694 max_blocks
= (EXT4_BLOCK_ALIGN(len
+ offset
, blkbits
) >> blkbits
)
3697 * credits to insert 1 extent into extent tree
3699 credits
= ext4_chunk_trans_blocks(inode
, max_blocks
);
3700 mutex_lock(&inode
->i_mutex
);
3701 ret
= inode_newsize_ok(inode
, (len
+ offset
));
3703 mutex_unlock(&inode
->i_mutex
);
3707 while (ret
>= 0 && ret
< max_blocks
) {
3708 map
.m_lblk
= map
.m_lblk
+ ret
;
3709 map
.m_len
= max_blocks
= max_blocks
- ret
;
3710 handle
= ext4_journal_start(inode
, credits
);
3711 if (IS_ERR(handle
)) {
3712 ret
= PTR_ERR(handle
);
3715 ret
= ext4_map_blocks(handle
, inode
, &map
,
3716 EXT4_GET_BLOCKS_CREATE_UNINIT_EXT
);
3720 printk(KERN_ERR
"%s: ext4_ext_map_blocks "
3721 "returned error inode#%lu, block=%u, "
3722 "max_blocks=%u", __func__
,
3723 inode
->i_ino
, block
, max_blocks
);
3725 ext4_mark_inode_dirty(handle
, inode
);
3726 ret2
= ext4_journal_stop(handle
);
3729 if ((map
.m_lblk
+ ret
) >= (EXT4_BLOCK_ALIGN(offset
+ len
,
3730 blkbits
) >> blkbits
))
3731 new_size
= offset
+ len
;
3733 new_size
= (map
.m_lblk
+ ret
) << blkbits
;
3735 ext4_falloc_update_inode(inode
, mode
, new_size
,
3736 (map
.m_flags
& EXT4_MAP_NEW
));
3737 ext4_mark_inode_dirty(handle
, inode
);
3738 ret2
= ext4_journal_stop(handle
);
3742 if (ret
== -ENOSPC
&&
3743 ext4_should_retry_alloc(inode
->i_sb
, &retries
)) {
3747 mutex_unlock(&inode
->i_mutex
);
3748 return ret
> 0 ? ret2
: ret
;
3752 * This function convert a range of blocks to written extents
3753 * The caller of this function will pass the start offset and the size.
3754 * all unwritten extents within this range will be converted to
3757 * This function is called from the direct IO end io call back
3758 * function, to convert the fallocated extents after IO is completed.
3759 * Returns 0 on success.
3761 int ext4_convert_unwritten_extents(struct inode
*inode
, loff_t offset
,
3765 unsigned int max_blocks
;
3768 struct ext4_map_blocks map
;
3769 unsigned int credits
, blkbits
= inode
->i_blkbits
;
3771 map
.m_lblk
= offset
>> blkbits
;
3773 * We can't just convert len to max_blocks because
3774 * If blocksize = 4096 offset = 3072 and len = 2048
3776 max_blocks
= ((EXT4_BLOCK_ALIGN(len
+ offset
, blkbits
) >> blkbits
) -
3779 * credits to insert 1 extent into extent tree
3781 credits
= ext4_chunk_trans_blocks(inode
, max_blocks
);
3782 while (ret
>= 0 && ret
< max_blocks
) {
3784 map
.m_len
= (max_blocks
-= ret
);
3785 handle
= ext4_journal_start(inode
, credits
);
3786 if (IS_ERR(handle
)) {
3787 ret
= PTR_ERR(handle
);
3790 ret
= ext4_map_blocks(handle
, inode
, &map
,
3791 EXT4_GET_BLOCKS_IO_CONVERT_EXT
);
3794 printk(KERN_ERR
"%s: ext4_ext_map_blocks "
3795 "returned error inode#%lu, block=%u, "
3796 "max_blocks=%u", __func__
,
3797 inode
->i_ino
, map
.m_lblk
, map
.m_len
);
3799 ext4_mark_inode_dirty(handle
, inode
);
3800 ret2
= ext4_journal_stop(handle
);
3801 if (ret
<= 0 || ret2
)
3804 return ret
> 0 ? ret2
: ret
;
3807 * Callback function called for each extent to gather FIEMAP information.
3809 static int ext4_ext_fiemap_cb(struct inode
*inode
, struct ext4_ext_path
*path
,
3810 struct ext4_ext_cache
*newex
, struct ext4_extent
*ex
,
3813 struct fiemap_extent_info
*fieinfo
= data
;
3814 unsigned char blksize_bits
= inode
->i_sb
->s_blocksize_bits
;
3821 logical
= (__u64
)newex
->ec_block
<< blksize_bits
;
3823 if (newex
->ec_type
== EXT4_EXT_CACHE_GAP
) {
3826 struct buffer_head
*bh
= NULL
;
3828 offset
= logical
>> PAGE_SHIFT
;
3829 page
= find_get_page(inode
->i_mapping
, offset
);
3830 if (!page
|| !page_has_buffers(page
))
3831 return EXT_CONTINUE
;
3833 bh
= page_buffers(page
);
3836 return EXT_CONTINUE
;
3838 if (buffer_delay(bh
)) {
3839 flags
|= FIEMAP_EXTENT_DELALLOC
;
3840 page_cache_release(page
);
3842 page_cache_release(page
);
3843 return EXT_CONTINUE
;
3847 physical
= (__u64
)newex
->ec_start
<< blksize_bits
;
3848 length
= (__u64
)newex
->ec_len
<< blksize_bits
;
3850 if (ex
&& ext4_ext_is_uninitialized(ex
))
3851 flags
|= FIEMAP_EXTENT_UNWRITTEN
;
3854 * If this extent reaches EXT_MAX_BLOCK, it must be last.
3856 * Or if ext4_ext_next_allocated_block is EXT_MAX_BLOCK,
3857 * this also indicates no more allocated blocks.
3859 * XXX this might miss a single-block extent at EXT_MAX_BLOCK
3861 if (ext4_ext_next_allocated_block(path
) == EXT_MAX_BLOCK
||
3862 newex
->ec_block
+ newex
->ec_len
- 1 == EXT_MAX_BLOCK
) {
3863 loff_t size
= i_size_read(inode
);
3864 loff_t bs
= EXT4_BLOCK_SIZE(inode
->i_sb
);
3866 flags
|= FIEMAP_EXTENT_LAST
;
3867 if ((flags
& FIEMAP_EXTENT_DELALLOC
) &&
3868 logical
+length
> size
)
3869 length
= (size
- logical
+ bs
- 1) & ~(bs
-1);
3872 error
= fiemap_fill_next_extent(fieinfo
, logical
, physical
,
3879 return EXT_CONTINUE
;
3882 /* fiemap flags we can handle specified here */
3883 #define EXT4_FIEMAP_FLAGS (FIEMAP_FLAG_SYNC|FIEMAP_FLAG_XATTR)
3885 static int ext4_xattr_fiemap(struct inode
*inode
,
3886 struct fiemap_extent_info
*fieinfo
)
3890 __u32 flags
= FIEMAP_EXTENT_LAST
;
3891 int blockbits
= inode
->i_sb
->s_blocksize_bits
;
3895 if (ext4_test_inode_state(inode
, EXT4_STATE_XATTR
)) {
3896 struct ext4_iloc iloc
;
3897 int offset
; /* offset of xattr in inode */
3899 error
= ext4_get_inode_loc(inode
, &iloc
);
3902 physical
= iloc
.bh
->b_blocknr
<< blockbits
;
3903 offset
= EXT4_GOOD_OLD_INODE_SIZE
+
3904 EXT4_I(inode
)->i_extra_isize
;
3906 length
= EXT4_SB(inode
->i_sb
)->s_inode_size
- offset
;
3907 flags
|= FIEMAP_EXTENT_DATA_INLINE
;
3909 } else { /* external block */
3910 physical
= EXT4_I(inode
)->i_file_acl
<< blockbits
;
3911 length
= inode
->i_sb
->s_blocksize
;
3915 error
= fiemap_fill_next_extent(fieinfo
, 0, physical
,
3917 return (error
< 0 ? error
: 0);
3920 int ext4_fiemap(struct inode
*inode
, struct fiemap_extent_info
*fieinfo
,
3921 __u64 start
, __u64 len
)
3923 ext4_lblk_t start_blk
;
3926 /* fallback to generic here if not in extents fmt */
3927 if (!(EXT4_I(inode
)->i_flags
& EXT4_EXTENTS_FL
))
3928 return generic_block_fiemap(inode
, fieinfo
, start
, len
,
3931 if (fiemap_check_flags(fieinfo
, EXT4_FIEMAP_FLAGS
))
3934 if (fieinfo
->fi_flags
& FIEMAP_FLAG_XATTR
) {
3935 error
= ext4_xattr_fiemap(inode
, fieinfo
);
3937 ext4_lblk_t len_blks
;
3940 start_blk
= start
>> inode
->i_sb
->s_blocksize_bits
;
3941 last_blk
= (start
+ len
- 1) >> inode
->i_sb
->s_blocksize_bits
;
3942 if (last_blk
>= EXT_MAX_BLOCK
)
3943 last_blk
= EXT_MAX_BLOCK
-1;
3944 len_blks
= ((ext4_lblk_t
) last_blk
) - start_blk
+ 1;
3947 * Walk the extent tree gathering extent information.
3948 * ext4_ext_fiemap_cb will push extents back to user.
3950 error
= ext4_ext_walk_space(inode
, start_blk
, len_blks
,
3951 ext4_ext_fiemap_cb
, fieinfo
);