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
33 #include <linux/time.h>
34 #include <linux/jbd2.h>
35 #include <linux/highuid.h>
36 #include <linux/pagemap.h>
37 #include <linux/quotaops.h>
38 #include <linux/string.h>
39 #include <linux/slab.h>
40 #include <linux/falloc.h>
41 #include <asm/uaccess.h>
42 #include <linux/fiemap.h>
43 #include "ext4_jbd2.h"
45 #include <trace/events/ext4.h>
48 * used by extent splitting.
50 #define EXT4_EXT_MAY_ZEROOUT 0x1 /* safe to zeroout if split fails \
52 #define EXT4_EXT_MARK_UNINIT1 0x2 /* mark first half uninitialized */
53 #define EXT4_EXT_MARK_UNINIT2 0x4 /* mark second half uninitialized */
55 static int ext4_split_extent(handle_t
*handle
,
57 struct ext4_ext_path
*path
,
58 struct ext4_map_blocks
*map
,
62 static int ext4_split_extent_at(handle_t
*handle
,
64 struct ext4_ext_path
*path
,
69 static int ext4_ext_truncate_extend_restart(handle_t
*handle
,
75 if (!ext4_handle_valid(handle
))
77 if (handle
->h_buffer_credits
> needed
)
79 err
= ext4_journal_extend(handle
, needed
);
82 err
= ext4_truncate_restart_trans(handle
, inode
, needed
);
94 static int ext4_ext_get_access(handle_t
*handle
, struct inode
*inode
,
95 struct ext4_ext_path
*path
)
98 /* path points to block */
99 return ext4_journal_get_write_access(handle
, path
->p_bh
);
101 /* path points to leaf/index in inode body */
102 /* we use in-core data, no need to protect them */
112 #define ext4_ext_dirty(handle, inode, path) \
113 __ext4_ext_dirty(__func__, __LINE__, (handle), (inode), (path))
114 static int __ext4_ext_dirty(const char *where
, unsigned int line
,
115 handle_t
*handle
, struct inode
*inode
,
116 struct ext4_ext_path
*path
)
120 /* path points to block */
121 err
= __ext4_handle_dirty_metadata(where
, line
, handle
,
124 /* path points to leaf/index in inode body */
125 err
= ext4_mark_inode_dirty(handle
, inode
);
130 static ext4_fsblk_t
ext4_ext_find_goal(struct inode
*inode
,
131 struct ext4_ext_path
*path
,
135 int depth
= path
->p_depth
;
136 struct ext4_extent
*ex
;
139 * Try to predict block placement assuming that we are
140 * filling in a file which will eventually be
141 * non-sparse --- i.e., in the case of libbfd writing
142 * an ELF object sections out-of-order but in a way
143 * the eventually results in a contiguous object or
144 * executable file, or some database extending a table
145 * space file. However, this is actually somewhat
146 * non-ideal if we are writing a sparse file such as
147 * qemu or KVM writing a raw image file that is going
148 * to stay fairly sparse, since it will end up
149 * fragmenting the file system's free space. Maybe we
150 * should have some hueristics or some way to allow
151 * userspace to pass a hint to file system,
152 * especially if the latter case turns out to be
155 ex
= path
[depth
].p_ext
;
157 ext4_fsblk_t ext_pblk
= ext4_ext_pblock(ex
);
158 ext4_lblk_t ext_block
= le32_to_cpu(ex
->ee_block
);
160 if (block
> ext_block
)
161 return ext_pblk
+ (block
- ext_block
);
163 return ext_pblk
- (ext_block
- block
);
166 /* it looks like index is empty;
167 * try to find starting block from index itself */
168 if (path
[depth
].p_bh
)
169 return path
[depth
].p_bh
->b_blocknr
;
172 /* OK. use inode's group */
173 return ext4_inode_to_goal_block(inode
);
177 * Allocation for a meta data block
180 ext4_ext_new_meta_block(handle_t
*handle
, struct inode
*inode
,
181 struct ext4_ext_path
*path
,
182 struct ext4_extent
*ex
, int *err
, unsigned int flags
)
184 ext4_fsblk_t goal
, newblock
;
186 goal
= ext4_ext_find_goal(inode
, path
, le32_to_cpu(ex
->ee_block
));
187 newblock
= ext4_new_meta_blocks(handle
, inode
, goal
, flags
,
192 static inline int ext4_ext_space_block(struct inode
*inode
, int check
)
196 size
= (inode
->i_sb
->s_blocksize
- sizeof(struct ext4_extent_header
))
197 / sizeof(struct ext4_extent
);
198 #ifdef AGGRESSIVE_TEST
199 if (!check
&& size
> 6)
205 static inline int ext4_ext_space_block_idx(struct inode
*inode
, int check
)
209 size
= (inode
->i_sb
->s_blocksize
- sizeof(struct ext4_extent_header
))
210 / sizeof(struct ext4_extent_idx
);
211 #ifdef AGGRESSIVE_TEST
212 if (!check
&& size
> 5)
218 static inline int ext4_ext_space_root(struct inode
*inode
, int check
)
222 size
= sizeof(EXT4_I(inode
)->i_data
);
223 size
-= sizeof(struct ext4_extent_header
);
224 size
/= sizeof(struct ext4_extent
);
225 #ifdef AGGRESSIVE_TEST
226 if (!check
&& size
> 3)
232 static inline int ext4_ext_space_root_idx(struct inode
*inode
, int check
)
236 size
= sizeof(EXT4_I(inode
)->i_data
);
237 size
-= sizeof(struct ext4_extent_header
);
238 size
/= sizeof(struct ext4_extent_idx
);
239 #ifdef AGGRESSIVE_TEST
240 if (!check
&& size
> 4)
247 * Calculate the number of metadata blocks needed
248 * to allocate @blocks
249 * Worse case is one block per extent
251 int ext4_ext_calc_metadata_amount(struct inode
*inode
, ext4_lblk_t lblock
)
253 struct ext4_inode_info
*ei
= EXT4_I(inode
);
256 idxs
= ((inode
->i_sb
->s_blocksize
- sizeof(struct ext4_extent_header
))
257 / sizeof(struct ext4_extent_idx
));
260 * If the new delayed allocation block is contiguous with the
261 * previous da block, it can share index blocks with the
262 * previous block, so we only need to allocate a new index
263 * block every idxs leaf blocks. At ldxs**2 blocks, we need
264 * an additional index block, and at ldxs**3 blocks, yet
265 * another index blocks.
267 if (ei
->i_da_metadata_calc_len
&&
268 ei
->i_da_metadata_calc_last_lblock
+1 == lblock
) {
271 if ((ei
->i_da_metadata_calc_len
% idxs
) == 0)
273 if ((ei
->i_da_metadata_calc_len
% (idxs
*idxs
)) == 0)
275 if ((ei
->i_da_metadata_calc_len
% (idxs
*idxs
*idxs
)) == 0) {
277 ei
->i_da_metadata_calc_len
= 0;
279 ei
->i_da_metadata_calc_len
++;
280 ei
->i_da_metadata_calc_last_lblock
++;
285 * In the worst case we need a new set of index blocks at
286 * every level of the inode's extent tree.
288 ei
->i_da_metadata_calc_len
= 1;
289 ei
->i_da_metadata_calc_last_lblock
= lblock
;
290 return ext_depth(inode
) + 1;
294 ext4_ext_max_entries(struct inode
*inode
, int depth
)
298 if (depth
== ext_depth(inode
)) {
300 max
= ext4_ext_space_root(inode
, 1);
302 max
= ext4_ext_space_root_idx(inode
, 1);
305 max
= ext4_ext_space_block(inode
, 1);
307 max
= ext4_ext_space_block_idx(inode
, 1);
313 static int ext4_valid_extent(struct inode
*inode
, struct ext4_extent
*ext
)
315 ext4_fsblk_t block
= ext4_ext_pblock(ext
);
316 int len
= ext4_ext_get_actual_len(ext
);
320 return ext4_data_block_valid(EXT4_SB(inode
->i_sb
), block
, len
);
323 static int ext4_valid_extent_idx(struct inode
*inode
,
324 struct ext4_extent_idx
*ext_idx
)
326 ext4_fsblk_t block
= ext4_idx_pblock(ext_idx
);
328 return ext4_data_block_valid(EXT4_SB(inode
->i_sb
), block
, 1);
331 static int ext4_valid_extent_entries(struct inode
*inode
,
332 struct ext4_extent_header
*eh
,
335 unsigned short entries
;
336 if (eh
->eh_entries
== 0)
339 entries
= le16_to_cpu(eh
->eh_entries
);
343 struct ext4_extent
*ext
= EXT_FIRST_EXTENT(eh
);
345 if (!ext4_valid_extent(inode
, ext
))
351 struct ext4_extent_idx
*ext_idx
= EXT_FIRST_INDEX(eh
);
353 if (!ext4_valid_extent_idx(inode
, ext_idx
))
362 static int __ext4_ext_check(const char *function
, unsigned int line
,
363 struct inode
*inode
, struct ext4_extent_header
*eh
,
366 const char *error_msg
;
369 if (unlikely(eh
->eh_magic
!= EXT4_EXT_MAGIC
)) {
370 error_msg
= "invalid magic";
373 if (unlikely(le16_to_cpu(eh
->eh_depth
) != depth
)) {
374 error_msg
= "unexpected eh_depth";
377 if (unlikely(eh
->eh_max
== 0)) {
378 error_msg
= "invalid eh_max";
381 max
= ext4_ext_max_entries(inode
, depth
);
382 if (unlikely(le16_to_cpu(eh
->eh_max
) > max
)) {
383 error_msg
= "too large eh_max";
386 if (unlikely(le16_to_cpu(eh
->eh_entries
) > le16_to_cpu(eh
->eh_max
))) {
387 error_msg
= "invalid eh_entries";
390 if (!ext4_valid_extent_entries(inode
, eh
, depth
)) {
391 error_msg
= "invalid extent entries";
397 ext4_error_inode(inode
, function
, line
, 0,
398 "bad header/extent: %s - magic %x, "
399 "entries %u, max %u(%u), depth %u(%u)",
400 error_msg
, le16_to_cpu(eh
->eh_magic
),
401 le16_to_cpu(eh
->eh_entries
), le16_to_cpu(eh
->eh_max
),
402 max
, le16_to_cpu(eh
->eh_depth
), depth
);
407 #define ext4_ext_check(inode, eh, depth) \
408 __ext4_ext_check(__func__, __LINE__, inode, eh, depth)
410 int ext4_ext_check_inode(struct inode
*inode
)
412 return ext4_ext_check(inode
, ext_inode_hdr(inode
), ext_depth(inode
));
416 static void ext4_ext_show_path(struct inode
*inode
, struct ext4_ext_path
*path
)
418 int k
, l
= path
->p_depth
;
421 for (k
= 0; k
<= l
; k
++, path
++) {
423 ext_debug(" %d->%llu", le32_to_cpu(path
->p_idx
->ei_block
),
424 ext4_idx_pblock(path
->p_idx
));
425 } else if (path
->p_ext
) {
426 ext_debug(" %d:[%d]%d:%llu ",
427 le32_to_cpu(path
->p_ext
->ee_block
),
428 ext4_ext_is_uninitialized(path
->p_ext
),
429 ext4_ext_get_actual_len(path
->p_ext
),
430 ext4_ext_pblock(path
->p_ext
));
437 static void ext4_ext_show_leaf(struct inode
*inode
, struct ext4_ext_path
*path
)
439 int depth
= ext_depth(inode
);
440 struct ext4_extent_header
*eh
;
441 struct ext4_extent
*ex
;
447 eh
= path
[depth
].p_hdr
;
448 ex
= EXT_FIRST_EXTENT(eh
);
450 ext_debug("Displaying leaf extents for inode %lu\n", inode
->i_ino
);
452 for (i
= 0; i
< le16_to_cpu(eh
->eh_entries
); i
++, ex
++) {
453 ext_debug("%d:[%d]%d:%llu ", le32_to_cpu(ex
->ee_block
),
454 ext4_ext_is_uninitialized(ex
),
455 ext4_ext_get_actual_len(ex
), ext4_ext_pblock(ex
));
460 static void ext4_ext_show_move(struct inode
*inode
, struct ext4_ext_path
*path
,
461 ext4_fsblk_t newblock
, int level
)
463 int depth
= ext_depth(inode
);
464 struct ext4_extent
*ex
;
466 if (depth
!= level
) {
467 struct ext4_extent_idx
*idx
;
468 idx
= path
[level
].p_idx
;
469 while (idx
<= EXT_MAX_INDEX(path
[level
].p_hdr
)) {
470 ext_debug("%d: move %d:%llu in new index %llu\n", level
,
471 le32_to_cpu(idx
->ei_block
),
472 ext4_idx_pblock(idx
),
480 ex
= path
[depth
].p_ext
;
481 while (ex
<= EXT_MAX_EXTENT(path
[depth
].p_hdr
)) {
482 ext_debug("move %d:%llu:[%d]%d in new leaf %llu\n",
483 le32_to_cpu(ex
->ee_block
),
485 ext4_ext_is_uninitialized(ex
),
486 ext4_ext_get_actual_len(ex
),
493 #define ext4_ext_show_path(inode, path)
494 #define ext4_ext_show_leaf(inode, path)
495 #define ext4_ext_show_move(inode, path, newblock, level)
498 void ext4_ext_drop_refs(struct ext4_ext_path
*path
)
500 int depth
= path
->p_depth
;
503 for (i
= 0; i
<= depth
; i
++, path
++)
511 * ext4_ext_binsearch_idx:
512 * binary search for the closest index of the given block
513 * the header must be checked before calling this
516 ext4_ext_binsearch_idx(struct inode
*inode
,
517 struct ext4_ext_path
*path
, ext4_lblk_t block
)
519 struct ext4_extent_header
*eh
= path
->p_hdr
;
520 struct ext4_extent_idx
*r
, *l
, *m
;
523 ext_debug("binsearch for %u(idx): ", block
);
525 l
= EXT_FIRST_INDEX(eh
) + 1;
526 r
= EXT_LAST_INDEX(eh
);
529 if (block
< le32_to_cpu(m
->ei_block
))
533 ext_debug("%p(%u):%p(%u):%p(%u) ", l
, le32_to_cpu(l
->ei_block
),
534 m
, le32_to_cpu(m
->ei_block
),
535 r
, le32_to_cpu(r
->ei_block
));
539 ext_debug(" -> %d->%lld ", le32_to_cpu(path
->p_idx
->ei_block
),
540 ext4_idx_pblock(path
->p_idx
));
542 #ifdef CHECK_BINSEARCH
544 struct ext4_extent_idx
*chix
, *ix
;
547 chix
= ix
= EXT_FIRST_INDEX(eh
);
548 for (k
= 0; k
< le16_to_cpu(eh
->eh_entries
); k
++, ix
++) {
550 le32_to_cpu(ix
->ei_block
) <= le32_to_cpu(ix
[-1].ei_block
)) {
551 printk(KERN_DEBUG
"k=%d, ix=0x%p, "
553 ix
, EXT_FIRST_INDEX(eh
));
554 printk(KERN_DEBUG
"%u <= %u\n",
555 le32_to_cpu(ix
->ei_block
),
556 le32_to_cpu(ix
[-1].ei_block
));
558 BUG_ON(k
&& le32_to_cpu(ix
->ei_block
)
559 <= le32_to_cpu(ix
[-1].ei_block
));
560 if (block
< le32_to_cpu(ix
->ei_block
))
564 BUG_ON(chix
!= path
->p_idx
);
571 * ext4_ext_binsearch:
572 * binary search for closest extent of the given block
573 * the header must be checked before calling this
576 ext4_ext_binsearch(struct inode
*inode
,
577 struct ext4_ext_path
*path
, ext4_lblk_t block
)
579 struct ext4_extent_header
*eh
= path
->p_hdr
;
580 struct ext4_extent
*r
, *l
, *m
;
582 if (eh
->eh_entries
== 0) {
584 * this leaf is empty:
585 * we get such a leaf in split/add case
590 ext_debug("binsearch for %u: ", block
);
592 l
= EXT_FIRST_EXTENT(eh
) + 1;
593 r
= EXT_LAST_EXTENT(eh
);
597 if (block
< le32_to_cpu(m
->ee_block
))
601 ext_debug("%p(%u):%p(%u):%p(%u) ", l
, le32_to_cpu(l
->ee_block
),
602 m
, le32_to_cpu(m
->ee_block
),
603 r
, le32_to_cpu(r
->ee_block
));
607 ext_debug(" -> %d:%llu:[%d]%d ",
608 le32_to_cpu(path
->p_ext
->ee_block
),
609 ext4_ext_pblock(path
->p_ext
),
610 ext4_ext_is_uninitialized(path
->p_ext
),
611 ext4_ext_get_actual_len(path
->p_ext
));
613 #ifdef CHECK_BINSEARCH
615 struct ext4_extent
*chex
, *ex
;
618 chex
= ex
= EXT_FIRST_EXTENT(eh
);
619 for (k
= 0; k
< le16_to_cpu(eh
->eh_entries
); k
++, ex
++) {
620 BUG_ON(k
&& le32_to_cpu(ex
->ee_block
)
621 <= le32_to_cpu(ex
[-1].ee_block
));
622 if (block
< le32_to_cpu(ex
->ee_block
))
626 BUG_ON(chex
!= path
->p_ext
);
632 int ext4_ext_tree_init(handle_t
*handle
, struct inode
*inode
)
634 struct ext4_extent_header
*eh
;
636 eh
= ext_inode_hdr(inode
);
639 eh
->eh_magic
= EXT4_EXT_MAGIC
;
640 eh
->eh_max
= cpu_to_le16(ext4_ext_space_root(inode
, 0));
641 ext4_mark_inode_dirty(handle
, inode
);
642 ext4_ext_invalidate_cache(inode
);
646 struct ext4_ext_path
*
647 ext4_ext_find_extent(struct inode
*inode
, ext4_lblk_t block
,
648 struct ext4_ext_path
*path
)
650 struct ext4_extent_header
*eh
;
651 struct buffer_head
*bh
;
652 short int depth
, i
, ppos
= 0, alloc
= 0;
654 eh
= ext_inode_hdr(inode
);
655 depth
= ext_depth(inode
);
657 /* account possible depth increase */
659 path
= kzalloc(sizeof(struct ext4_ext_path
) * (depth
+ 2),
662 return ERR_PTR(-ENOMEM
);
669 /* walk through the tree */
671 int need_to_validate
= 0;
673 ext_debug("depth %d: num %d, max %d\n",
674 ppos
, le16_to_cpu(eh
->eh_entries
), le16_to_cpu(eh
->eh_max
));
676 ext4_ext_binsearch_idx(inode
, path
+ ppos
, block
);
677 path
[ppos
].p_block
= ext4_idx_pblock(path
[ppos
].p_idx
);
678 path
[ppos
].p_depth
= i
;
679 path
[ppos
].p_ext
= NULL
;
681 bh
= sb_getblk(inode
->i_sb
, path
[ppos
].p_block
);
684 if (!bh_uptodate_or_lock(bh
)) {
685 trace_ext4_ext_load_extent(inode
, block
,
687 if (bh_submit_read(bh
) < 0) {
691 /* validate the extent entries */
692 need_to_validate
= 1;
694 eh
= ext_block_hdr(bh
);
696 if (unlikely(ppos
> depth
)) {
698 EXT4_ERROR_INODE(inode
,
699 "ppos %d > depth %d", ppos
, depth
);
702 path
[ppos
].p_bh
= bh
;
703 path
[ppos
].p_hdr
= eh
;
706 if (need_to_validate
&& ext4_ext_check(inode
, eh
, i
))
710 path
[ppos
].p_depth
= i
;
711 path
[ppos
].p_ext
= NULL
;
712 path
[ppos
].p_idx
= NULL
;
715 ext4_ext_binsearch(inode
, path
+ ppos
, block
);
716 /* if not an empty leaf */
717 if (path
[ppos
].p_ext
)
718 path
[ppos
].p_block
= ext4_ext_pblock(path
[ppos
].p_ext
);
720 ext4_ext_show_path(inode
, path
);
725 ext4_ext_drop_refs(path
);
728 return ERR_PTR(-EIO
);
732 * ext4_ext_insert_index:
733 * insert new index [@logical;@ptr] into the block at @curp;
734 * check where to insert: before @curp or after @curp
736 static int ext4_ext_insert_index(handle_t
*handle
, struct inode
*inode
,
737 struct ext4_ext_path
*curp
,
738 int logical
, ext4_fsblk_t ptr
)
740 struct ext4_extent_idx
*ix
;
743 err
= ext4_ext_get_access(handle
, inode
, curp
);
747 if (unlikely(logical
== le32_to_cpu(curp
->p_idx
->ei_block
))) {
748 EXT4_ERROR_INODE(inode
,
749 "logical %d == ei_block %d!",
750 logical
, le32_to_cpu(curp
->p_idx
->ei_block
));
754 if (unlikely(le16_to_cpu(curp
->p_hdr
->eh_entries
)
755 >= le16_to_cpu(curp
->p_hdr
->eh_max
))) {
756 EXT4_ERROR_INODE(inode
,
757 "eh_entries %d >= eh_max %d!",
758 le16_to_cpu(curp
->p_hdr
->eh_entries
),
759 le16_to_cpu(curp
->p_hdr
->eh_max
));
763 if (logical
> le32_to_cpu(curp
->p_idx
->ei_block
)) {
765 ext_debug("insert new index %d after: %llu\n", logical
, ptr
);
766 ix
= curp
->p_idx
+ 1;
769 ext_debug("insert new index %d before: %llu\n", logical
, ptr
);
773 len
= EXT_LAST_INDEX(curp
->p_hdr
) - ix
+ 1;
776 ext_debug("insert new index %d: "
777 "move %d indices from 0x%p to 0x%p\n",
778 logical
, len
, ix
, ix
+ 1);
779 memmove(ix
+ 1, ix
, len
* sizeof(struct ext4_extent_idx
));
782 if (unlikely(ix
> EXT_MAX_INDEX(curp
->p_hdr
))) {
783 EXT4_ERROR_INODE(inode
, "ix > EXT_MAX_INDEX!");
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(ix
> EXT_LAST_INDEX(curp
->p_hdr
))) {
792 EXT4_ERROR_INODE(inode
, "ix > EXT_LAST_INDEX!");
796 err
= ext4_ext_dirty(handle
, inode
, curp
);
797 ext4_std_error(inode
->i_sb
, err
);
804 * inserts new subtree into the path, using free index entry
806 * - allocates all needed blocks (new leaf and all intermediate index blocks)
807 * - makes decision where to split
808 * - moves remaining extents and index entries (right to the split point)
809 * into the newly allocated blocks
810 * - initializes subtree
812 static int ext4_ext_split(handle_t
*handle
, struct inode
*inode
,
814 struct ext4_ext_path
*path
,
815 struct ext4_extent
*newext
, int at
)
817 struct buffer_head
*bh
= NULL
;
818 int depth
= ext_depth(inode
);
819 struct ext4_extent_header
*neh
;
820 struct ext4_extent_idx
*fidx
;
822 ext4_fsblk_t newblock
, oldblock
;
824 ext4_fsblk_t
*ablocks
= NULL
; /* array of allocated blocks */
827 /* make decision: where to split? */
828 /* FIXME: now decision is simplest: at current extent */
830 /* if current leaf will be split, then we should use
831 * border from split point */
832 if (unlikely(path
[depth
].p_ext
> EXT_MAX_EXTENT(path
[depth
].p_hdr
))) {
833 EXT4_ERROR_INODE(inode
, "p_ext > EXT_MAX_EXTENT!");
836 if (path
[depth
].p_ext
!= EXT_MAX_EXTENT(path
[depth
].p_hdr
)) {
837 border
= path
[depth
].p_ext
[1].ee_block
;
838 ext_debug("leaf will be split."
839 " next leaf starts at %d\n",
840 le32_to_cpu(border
));
842 border
= newext
->ee_block
;
843 ext_debug("leaf will be added."
844 " next leaf starts at %d\n",
845 le32_to_cpu(border
));
849 * If error occurs, then we break processing
850 * and mark filesystem read-only. index won't
851 * be inserted and tree will be in consistent
852 * state. Next mount will repair buffers too.
856 * Get array to track all allocated blocks.
857 * We need this to handle errors and free blocks
860 ablocks
= kzalloc(sizeof(ext4_fsblk_t
) * depth
, GFP_NOFS
);
864 /* allocate all needed blocks */
865 ext_debug("allocate %d blocks for indexes/leaf\n", depth
- at
);
866 for (a
= 0; a
< depth
- at
; a
++) {
867 newblock
= ext4_ext_new_meta_block(handle
, inode
, path
,
868 newext
, &err
, flags
);
871 ablocks
[a
] = newblock
;
874 /* initialize new leaf */
875 newblock
= ablocks
[--a
];
876 if (unlikely(newblock
== 0)) {
877 EXT4_ERROR_INODE(inode
, "newblock == 0!");
881 bh
= sb_getblk(inode
->i_sb
, newblock
);
888 err
= ext4_journal_get_create_access(handle
, bh
);
892 neh
= ext_block_hdr(bh
);
894 neh
->eh_max
= cpu_to_le16(ext4_ext_space_block(inode
, 0));
895 neh
->eh_magic
= EXT4_EXT_MAGIC
;
898 /* move remainder of path[depth] to the new leaf */
899 if (unlikely(path
[depth
].p_hdr
->eh_entries
!=
900 path
[depth
].p_hdr
->eh_max
)) {
901 EXT4_ERROR_INODE(inode
, "eh_entries %d != eh_max %d!",
902 path
[depth
].p_hdr
->eh_entries
,
903 path
[depth
].p_hdr
->eh_max
);
907 /* start copy from next extent */
908 m
= EXT_MAX_EXTENT(path
[depth
].p_hdr
) - path
[depth
].p_ext
++;
909 ext4_ext_show_move(inode
, path
, newblock
, depth
);
911 struct ext4_extent
*ex
;
912 ex
= EXT_FIRST_EXTENT(neh
);
913 memmove(ex
, path
[depth
].p_ext
, sizeof(struct ext4_extent
) * m
);
914 le16_add_cpu(&neh
->eh_entries
, m
);
917 set_buffer_uptodate(bh
);
920 err
= ext4_handle_dirty_metadata(handle
, inode
, bh
);
926 /* correct old leaf */
928 err
= ext4_ext_get_access(handle
, inode
, path
+ depth
);
931 le16_add_cpu(&path
[depth
].p_hdr
->eh_entries
, -m
);
932 err
= ext4_ext_dirty(handle
, inode
, path
+ depth
);
938 /* create intermediate indexes */
940 if (unlikely(k
< 0)) {
941 EXT4_ERROR_INODE(inode
, "k %d < 0!", k
);
946 ext_debug("create %d intermediate indices\n", k
);
947 /* insert new index into current index block */
948 /* current depth stored in i var */
952 newblock
= ablocks
[--a
];
953 bh
= sb_getblk(inode
->i_sb
, newblock
);
960 err
= ext4_journal_get_create_access(handle
, bh
);
964 neh
= ext_block_hdr(bh
);
965 neh
->eh_entries
= cpu_to_le16(1);
966 neh
->eh_magic
= EXT4_EXT_MAGIC
;
967 neh
->eh_max
= cpu_to_le16(ext4_ext_space_block_idx(inode
, 0));
968 neh
->eh_depth
= cpu_to_le16(depth
- i
);
969 fidx
= EXT_FIRST_INDEX(neh
);
970 fidx
->ei_block
= border
;
971 ext4_idx_store_pblock(fidx
, oldblock
);
973 ext_debug("int.index at %d (block %llu): %u -> %llu\n",
974 i
, newblock
, le32_to_cpu(border
), oldblock
);
976 /* move remainder of path[i] to the new index block */
977 if (unlikely(EXT_MAX_INDEX(path
[i
].p_hdr
) !=
978 EXT_LAST_INDEX(path
[i
].p_hdr
))) {
979 EXT4_ERROR_INODE(inode
,
980 "EXT_MAX_INDEX != EXT_LAST_INDEX ee_block %d!",
981 le32_to_cpu(path
[i
].p_ext
->ee_block
));
985 /* start copy indexes */
986 m
= EXT_MAX_INDEX(path
[i
].p_hdr
) - path
[i
].p_idx
++;
987 ext_debug("cur 0x%p, last 0x%p\n", path
[i
].p_idx
,
988 EXT_MAX_INDEX(path
[i
].p_hdr
));
989 ext4_ext_show_move(inode
, path
, newblock
, i
);
991 memmove(++fidx
, path
[i
].p_idx
,
992 sizeof(struct ext4_extent_idx
) * m
);
993 le16_add_cpu(&neh
->eh_entries
, m
);
995 set_buffer_uptodate(bh
);
998 err
= ext4_handle_dirty_metadata(handle
, inode
, bh
);
1004 /* correct old index */
1006 err
= ext4_ext_get_access(handle
, inode
, path
+ i
);
1009 le16_add_cpu(&path
[i
].p_hdr
->eh_entries
, -m
);
1010 err
= ext4_ext_dirty(handle
, inode
, path
+ i
);
1018 /* insert new index */
1019 err
= ext4_ext_insert_index(handle
, inode
, path
+ at
,
1020 le32_to_cpu(border
), newblock
);
1024 if (buffer_locked(bh
))
1030 /* free all allocated blocks in error case */
1031 for (i
= 0; i
< depth
; i
++) {
1034 ext4_free_blocks(handle
, inode
, NULL
, ablocks
[i
], 1,
1035 EXT4_FREE_BLOCKS_METADATA
);
1044 * ext4_ext_grow_indepth:
1045 * implements tree growing procedure:
1046 * - allocates new block
1047 * - moves top-level data (index block or leaf) into the new block
1048 * - initializes new top-level, creating index that points to the
1049 * just created block
1051 static int ext4_ext_grow_indepth(handle_t
*handle
, struct inode
*inode
,
1053 struct ext4_extent
*newext
)
1055 struct ext4_extent_header
*neh
;
1056 struct buffer_head
*bh
;
1057 ext4_fsblk_t newblock
;
1060 newblock
= ext4_ext_new_meta_block(handle
, inode
, NULL
,
1061 newext
, &err
, flags
);
1065 bh
= sb_getblk(inode
->i_sb
, newblock
);
1068 ext4_std_error(inode
->i_sb
, err
);
1073 err
= ext4_journal_get_create_access(handle
, bh
);
1079 /* move top-level index/leaf into new block */
1080 memmove(bh
->b_data
, EXT4_I(inode
)->i_data
,
1081 sizeof(EXT4_I(inode
)->i_data
));
1083 /* set size of new block */
1084 neh
= ext_block_hdr(bh
);
1085 /* old root could have indexes or leaves
1086 * so calculate e_max right way */
1087 if (ext_depth(inode
))
1088 neh
->eh_max
= cpu_to_le16(ext4_ext_space_block_idx(inode
, 0));
1090 neh
->eh_max
= cpu_to_le16(ext4_ext_space_block(inode
, 0));
1091 neh
->eh_magic
= EXT4_EXT_MAGIC
;
1092 set_buffer_uptodate(bh
);
1095 err
= ext4_handle_dirty_metadata(handle
, inode
, bh
);
1099 /* Update top-level index: num,max,pointer */
1100 neh
= ext_inode_hdr(inode
);
1101 neh
->eh_entries
= cpu_to_le16(1);
1102 ext4_idx_store_pblock(EXT_FIRST_INDEX(neh
), newblock
);
1103 if (neh
->eh_depth
== 0) {
1104 /* Root extent block becomes index block */
1105 neh
->eh_max
= cpu_to_le16(ext4_ext_space_root_idx(inode
, 0));
1106 EXT_FIRST_INDEX(neh
)->ei_block
=
1107 EXT_FIRST_EXTENT(neh
)->ee_block
;
1109 ext_debug("new root: num %d(%d), lblock %d, ptr %llu\n",
1110 le16_to_cpu(neh
->eh_entries
), le16_to_cpu(neh
->eh_max
),
1111 le32_to_cpu(EXT_FIRST_INDEX(neh
)->ei_block
),
1112 ext4_idx_pblock(EXT_FIRST_INDEX(neh
)));
1114 neh
->eh_depth
= cpu_to_le16(le16_to_cpu(neh
->eh_depth
) + 1);
1115 ext4_mark_inode_dirty(handle
, inode
);
1123 * ext4_ext_create_new_leaf:
1124 * finds empty index and adds new leaf.
1125 * if no free index is found, then it requests in-depth growing.
1127 static int ext4_ext_create_new_leaf(handle_t
*handle
, struct inode
*inode
,
1129 struct ext4_ext_path
*path
,
1130 struct ext4_extent
*newext
)
1132 struct ext4_ext_path
*curp
;
1133 int depth
, i
, err
= 0;
1136 i
= depth
= ext_depth(inode
);
1138 /* walk up to the tree and look for free index entry */
1139 curp
= path
+ depth
;
1140 while (i
> 0 && !EXT_HAS_FREE_INDEX(curp
)) {
1145 /* we use already allocated block for index block,
1146 * so subsequent data blocks should be contiguous */
1147 if (EXT_HAS_FREE_INDEX(curp
)) {
1148 /* if we found index with free entry, then use that
1149 * entry: create all needed subtree and add new leaf */
1150 err
= ext4_ext_split(handle
, inode
, flags
, path
, newext
, i
);
1155 ext4_ext_drop_refs(path
);
1156 path
= ext4_ext_find_extent(inode
,
1157 (ext4_lblk_t
)le32_to_cpu(newext
->ee_block
),
1160 err
= PTR_ERR(path
);
1162 /* tree is full, time to grow in depth */
1163 err
= ext4_ext_grow_indepth(handle
, inode
, flags
, newext
);
1168 ext4_ext_drop_refs(path
);
1169 path
= ext4_ext_find_extent(inode
,
1170 (ext4_lblk_t
)le32_to_cpu(newext
->ee_block
),
1173 err
= PTR_ERR(path
);
1178 * only first (depth 0 -> 1) produces free space;
1179 * in all other cases we have to split the grown tree
1181 depth
= ext_depth(inode
);
1182 if (path
[depth
].p_hdr
->eh_entries
== path
[depth
].p_hdr
->eh_max
) {
1183 /* now we need to split */
1193 * search the closest allocated block to the left for *logical
1194 * and returns it at @logical + it's physical address at @phys
1195 * if *logical is the smallest allocated block, the function
1196 * returns 0 at @phys
1197 * return value contains 0 (success) or error code
1199 static int ext4_ext_search_left(struct inode
*inode
,
1200 struct ext4_ext_path
*path
,
1201 ext4_lblk_t
*logical
, ext4_fsblk_t
*phys
)
1203 struct ext4_extent_idx
*ix
;
1204 struct ext4_extent
*ex
;
1207 if (unlikely(path
== NULL
)) {
1208 EXT4_ERROR_INODE(inode
, "path == NULL *logical %d!", *logical
);
1211 depth
= path
->p_depth
;
1214 if (depth
== 0 && path
->p_ext
== NULL
)
1217 /* usually extent in the path covers blocks smaller
1218 * then *logical, but it can be that extent is the
1219 * first one in the file */
1221 ex
= path
[depth
].p_ext
;
1222 ee_len
= ext4_ext_get_actual_len(ex
);
1223 if (*logical
< le32_to_cpu(ex
->ee_block
)) {
1224 if (unlikely(EXT_FIRST_EXTENT(path
[depth
].p_hdr
) != ex
)) {
1225 EXT4_ERROR_INODE(inode
,
1226 "EXT_FIRST_EXTENT != ex *logical %d ee_block %d!",
1227 *logical
, le32_to_cpu(ex
->ee_block
));
1230 while (--depth
>= 0) {
1231 ix
= path
[depth
].p_idx
;
1232 if (unlikely(ix
!= EXT_FIRST_INDEX(path
[depth
].p_hdr
))) {
1233 EXT4_ERROR_INODE(inode
,
1234 "ix (%d) != EXT_FIRST_INDEX (%d) (depth %d)!",
1235 ix
!= NULL
? le32_to_cpu(ix
->ei_block
) : 0,
1236 EXT_FIRST_INDEX(path
[depth
].p_hdr
) != NULL
?
1237 le32_to_cpu(EXT_FIRST_INDEX(path
[depth
].p_hdr
)->ei_block
) : 0,
1245 if (unlikely(*logical
< (le32_to_cpu(ex
->ee_block
) + ee_len
))) {
1246 EXT4_ERROR_INODE(inode
,
1247 "logical %d < ee_block %d + ee_len %d!",
1248 *logical
, le32_to_cpu(ex
->ee_block
), ee_len
);
1252 *logical
= le32_to_cpu(ex
->ee_block
) + ee_len
- 1;
1253 *phys
= ext4_ext_pblock(ex
) + ee_len
- 1;
1258 * search the closest allocated block to the right for *logical
1259 * and returns it at @logical + it's physical address at @phys
1260 * if *logical is the largest allocated block, the function
1261 * returns 0 at @phys
1262 * return value contains 0 (success) or error code
1264 static int ext4_ext_search_right(struct inode
*inode
,
1265 struct ext4_ext_path
*path
,
1266 ext4_lblk_t
*logical
, ext4_fsblk_t
*phys
,
1267 struct ext4_extent
**ret_ex
)
1269 struct buffer_head
*bh
= NULL
;
1270 struct ext4_extent_header
*eh
;
1271 struct ext4_extent_idx
*ix
;
1272 struct ext4_extent
*ex
;
1274 int depth
; /* Note, NOT eh_depth; depth from top of tree */
1277 if (unlikely(path
== NULL
)) {
1278 EXT4_ERROR_INODE(inode
, "path == NULL *logical %d!", *logical
);
1281 depth
= path
->p_depth
;
1284 if (depth
== 0 && path
->p_ext
== NULL
)
1287 /* usually extent in the path covers blocks smaller
1288 * then *logical, but it can be that extent is the
1289 * first one in the file */
1291 ex
= path
[depth
].p_ext
;
1292 ee_len
= ext4_ext_get_actual_len(ex
);
1293 if (*logical
< le32_to_cpu(ex
->ee_block
)) {
1294 if (unlikely(EXT_FIRST_EXTENT(path
[depth
].p_hdr
) != ex
)) {
1295 EXT4_ERROR_INODE(inode
,
1296 "first_extent(path[%d].p_hdr) != ex",
1300 while (--depth
>= 0) {
1301 ix
= path
[depth
].p_idx
;
1302 if (unlikely(ix
!= EXT_FIRST_INDEX(path
[depth
].p_hdr
))) {
1303 EXT4_ERROR_INODE(inode
,
1304 "ix != EXT_FIRST_INDEX *logical %d!",
1312 if (unlikely(*logical
< (le32_to_cpu(ex
->ee_block
) + ee_len
))) {
1313 EXT4_ERROR_INODE(inode
,
1314 "logical %d < ee_block %d + ee_len %d!",
1315 *logical
, le32_to_cpu(ex
->ee_block
), ee_len
);
1319 if (ex
!= EXT_LAST_EXTENT(path
[depth
].p_hdr
)) {
1320 /* next allocated block in this leaf */
1325 /* go up and search for index to the right */
1326 while (--depth
>= 0) {
1327 ix
= path
[depth
].p_idx
;
1328 if (ix
!= EXT_LAST_INDEX(path
[depth
].p_hdr
))
1332 /* we've gone up to the root and found no index to the right */
1336 /* we've found index to the right, let's
1337 * follow it and find the closest allocated
1338 * block to the right */
1340 block
= ext4_idx_pblock(ix
);
1341 while (++depth
< path
->p_depth
) {
1342 bh
= sb_bread(inode
->i_sb
, block
);
1345 eh
= ext_block_hdr(bh
);
1346 /* subtract from p_depth to get proper eh_depth */
1347 if (ext4_ext_check(inode
, eh
, path
->p_depth
- depth
)) {
1351 ix
= EXT_FIRST_INDEX(eh
);
1352 block
= ext4_idx_pblock(ix
);
1356 bh
= sb_bread(inode
->i_sb
, block
);
1359 eh
= ext_block_hdr(bh
);
1360 if (ext4_ext_check(inode
, eh
, path
->p_depth
- depth
)) {
1364 ex
= EXT_FIRST_EXTENT(eh
);
1366 *logical
= le32_to_cpu(ex
->ee_block
);
1367 *phys
= ext4_ext_pblock(ex
);
1375 * ext4_ext_next_allocated_block:
1376 * returns allocated block in subsequent extent or EXT_MAX_BLOCKS.
1377 * NOTE: it considers block number from index entry as
1378 * allocated block. Thus, index entries have to be consistent
1382 ext4_ext_next_allocated_block(struct ext4_ext_path
*path
)
1386 BUG_ON(path
== NULL
);
1387 depth
= path
->p_depth
;
1389 if (depth
== 0 && path
->p_ext
== NULL
)
1390 return EXT_MAX_BLOCKS
;
1392 while (depth
>= 0) {
1393 if (depth
== path
->p_depth
) {
1395 if (path
[depth
].p_ext
&&
1396 path
[depth
].p_ext
!=
1397 EXT_LAST_EXTENT(path
[depth
].p_hdr
))
1398 return le32_to_cpu(path
[depth
].p_ext
[1].ee_block
);
1401 if (path
[depth
].p_idx
!=
1402 EXT_LAST_INDEX(path
[depth
].p_hdr
))
1403 return le32_to_cpu(path
[depth
].p_idx
[1].ei_block
);
1408 return EXT_MAX_BLOCKS
;
1412 * ext4_ext_next_leaf_block:
1413 * returns first allocated block from next leaf or EXT_MAX_BLOCKS
1415 static ext4_lblk_t
ext4_ext_next_leaf_block(struct ext4_ext_path
*path
)
1419 BUG_ON(path
== NULL
);
1420 depth
= path
->p_depth
;
1422 /* zero-tree has no leaf blocks at all */
1424 return EXT_MAX_BLOCKS
;
1426 /* go to index block */
1429 while (depth
>= 0) {
1430 if (path
[depth
].p_idx
!=
1431 EXT_LAST_INDEX(path
[depth
].p_hdr
))
1432 return (ext4_lblk_t
)
1433 le32_to_cpu(path
[depth
].p_idx
[1].ei_block
);
1437 return EXT_MAX_BLOCKS
;
1441 * ext4_ext_correct_indexes:
1442 * if leaf gets modified and modified extent is first in the leaf,
1443 * then we have to correct all indexes above.
1444 * TODO: do we need to correct tree in all cases?
1446 static int ext4_ext_correct_indexes(handle_t
*handle
, struct inode
*inode
,
1447 struct ext4_ext_path
*path
)
1449 struct ext4_extent_header
*eh
;
1450 int depth
= ext_depth(inode
);
1451 struct ext4_extent
*ex
;
1455 eh
= path
[depth
].p_hdr
;
1456 ex
= path
[depth
].p_ext
;
1458 if (unlikely(ex
== NULL
|| eh
== NULL
)) {
1459 EXT4_ERROR_INODE(inode
,
1460 "ex %p == NULL or eh %p == NULL", ex
, eh
);
1465 /* there is no tree at all */
1469 if (ex
!= EXT_FIRST_EXTENT(eh
)) {
1470 /* we correct tree if first leaf got modified only */
1475 * TODO: we need correction if border is smaller than current one
1478 border
= path
[depth
].p_ext
->ee_block
;
1479 err
= ext4_ext_get_access(handle
, inode
, path
+ k
);
1482 path
[k
].p_idx
->ei_block
= border
;
1483 err
= ext4_ext_dirty(handle
, inode
, path
+ k
);
1488 /* change all left-side indexes */
1489 if (path
[k
+1].p_idx
!= EXT_FIRST_INDEX(path
[k
+1].p_hdr
))
1491 err
= ext4_ext_get_access(handle
, inode
, path
+ k
);
1494 path
[k
].p_idx
->ei_block
= border
;
1495 err
= ext4_ext_dirty(handle
, inode
, path
+ k
);
1504 ext4_can_extents_be_merged(struct inode
*inode
, struct ext4_extent
*ex1
,
1505 struct ext4_extent
*ex2
)
1507 unsigned short ext1_ee_len
, ext2_ee_len
, max_len
;
1510 * Make sure that either both extents are uninitialized, or
1513 if (ext4_ext_is_uninitialized(ex1
) ^ ext4_ext_is_uninitialized(ex2
))
1516 if (ext4_ext_is_uninitialized(ex1
))
1517 max_len
= EXT_UNINIT_MAX_LEN
;
1519 max_len
= EXT_INIT_MAX_LEN
;
1521 ext1_ee_len
= ext4_ext_get_actual_len(ex1
);
1522 ext2_ee_len
= ext4_ext_get_actual_len(ex2
);
1524 if (le32_to_cpu(ex1
->ee_block
) + ext1_ee_len
!=
1525 le32_to_cpu(ex2
->ee_block
))
1529 * To allow future support for preallocated extents to be added
1530 * as an RO_COMPAT feature, refuse to merge to extents if
1531 * this can result in the top bit of ee_len being set.
1533 if (ext1_ee_len
+ ext2_ee_len
> max_len
)
1535 #ifdef AGGRESSIVE_TEST
1536 if (ext1_ee_len
>= 4)
1540 if (ext4_ext_pblock(ex1
) + ext1_ee_len
== ext4_ext_pblock(ex2
))
1546 * This function tries to merge the "ex" extent to the next extent in the tree.
1547 * It always tries to merge towards right. If you want to merge towards
1548 * left, pass "ex - 1" as argument instead of "ex".
1549 * Returns 0 if the extents (ex and ex+1) were _not_ merged and returns
1550 * 1 if they got merged.
1552 static int ext4_ext_try_to_merge_right(struct inode
*inode
,
1553 struct ext4_ext_path
*path
,
1554 struct ext4_extent
*ex
)
1556 struct ext4_extent_header
*eh
;
1557 unsigned int depth
, len
;
1559 int uninitialized
= 0;
1561 depth
= ext_depth(inode
);
1562 BUG_ON(path
[depth
].p_hdr
== NULL
);
1563 eh
= path
[depth
].p_hdr
;
1565 while (ex
< EXT_LAST_EXTENT(eh
)) {
1566 if (!ext4_can_extents_be_merged(inode
, ex
, ex
+ 1))
1568 /* merge with next extent! */
1569 if (ext4_ext_is_uninitialized(ex
))
1571 ex
->ee_len
= cpu_to_le16(ext4_ext_get_actual_len(ex
)
1572 + ext4_ext_get_actual_len(ex
+ 1));
1574 ext4_ext_mark_uninitialized(ex
);
1576 if (ex
+ 1 < EXT_LAST_EXTENT(eh
)) {
1577 len
= (EXT_LAST_EXTENT(eh
) - ex
- 1)
1578 * sizeof(struct ext4_extent
);
1579 memmove(ex
+ 1, ex
+ 2, len
);
1581 le16_add_cpu(&eh
->eh_entries
, -1);
1583 WARN_ON(eh
->eh_entries
== 0);
1584 if (!eh
->eh_entries
)
1585 EXT4_ERROR_INODE(inode
, "eh->eh_entries = 0!");
1592 * This function tries to merge the @ex extent to neighbours in the tree.
1593 * return 1 if merge left else 0.
1595 static int ext4_ext_try_to_merge(struct inode
*inode
,
1596 struct ext4_ext_path
*path
,
1597 struct ext4_extent
*ex
) {
1598 struct ext4_extent_header
*eh
;
1603 depth
= ext_depth(inode
);
1604 BUG_ON(path
[depth
].p_hdr
== NULL
);
1605 eh
= path
[depth
].p_hdr
;
1607 if (ex
> EXT_FIRST_EXTENT(eh
))
1608 merge_done
= ext4_ext_try_to_merge_right(inode
, path
, ex
- 1);
1611 ret
= ext4_ext_try_to_merge_right(inode
, path
, ex
);
1617 * check if a portion of the "newext" extent overlaps with an
1620 * If there is an overlap discovered, it updates the length of the newext
1621 * such that there will be no overlap, and then returns 1.
1622 * If there is no overlap found, it returns 0.
1624 static unsigned int ext4_ext_check_overlap(struct ext4_sb_info
*sbi
,
1625 struct inode
*inode
,
1626 struct ext4_extent
*newext
,
1627 struct ext4_ext_path
*path
)
1630 unsigned int depth
, len1
;
1631 unsigned int ret
= 0;
1633 b1
= le32_to_cpu(newext
->ee_block
);
1634 len1
= ext4_ext_get_actual_len(newext
);
1635 depth
= ext_depth(inode
);
1636 if (!path
[depth
].p_ext
)
1638 b2
= le32_to_cpu(path
[depth
].p_ext
->ee_block
);
1639 b2
&= ~(sbi
->s_cluster_ratio
- 1);
1642 * get the next allocated block if the extent in the path
1643 * is before the requested block(s)
1646 b2
= ext4_ext_next_allocated_block(path
);
1647 if (b2
== EXT_MAX_BLOCKS
)
1649 b2
&= ~(sbi
->s_cluster_ratio
- 1);
1652 /* check for wrap through zero on extent logical start block*/
1653 if (b1
+ len1
< b1
) {
1654 len1
= EXT_MAX_BLOCKS
- b1
;
1655 newext
->ee_len
= cpu_to_le16(len1
);
1659 /* check for overlap */
1660 if (b1
+ len1
> b2
) {
1661 newext
->ee_len
= cpu_to_le16(b2
- b1
);
1669 * ext4_ext_insert_extent:
1670 * tries to merge requsted extent into the existing extent or
1671 * inserts requested extent as new one into the tree,
1672 * creating new leaf in the no-space case.
1674 int ext4_ext_insert_extent(handle_t
*handle
, struct inode
*inode
,
1675 struct ext4_ext_path
*path
,
1676 struct ext4_extent
*newext
, int flag
)
1678 struct ext4_extent_header
*eh
;
1679 struct ext4_extent
*ex
, *fex
;
1680 struct ext4_extent
*nearex
; /* nearest extent */
1681 struct ext4_ext_path
*npath
= NULL
;
1682 int depth
, len
, err
;
1684 unsigned uninitialized
= 0;
1687 if (unlikely(ext4_ext_get_actual_len(newext
) == 0)) {
1688 EXT4_ERROR_INODE(inode
, "ext4_ext_get_actual_len(newext) == 0");
1691 depth
= ext_depth(inode
);
1692 ex
= path
[depth
].p_ext
;
1693 if (unlikely(path
[depth
].p_hdr
== NULL
)) {
1694 EXT4_ERROR_INODE(inode
, "path[%d].p_hdr == NULL", depth
);
1698 /* try to insert block into found extent and return */
1699 if (ex
&& !(flag
& EXT4_GET_BLOCKS_PRE_IO
)
1700 && ext4_can_extents_be_merged(inode
, ex
, newext
)) {
1701 ext_debug("append [%d]%d block to %u:[%d]%d (from %llu)\n",
1702 ext4_ext_is_uninitialized(newext
),
1703 ext4_ext_get_actual_len(newext
),
1704 le32_to_cpu(ex
->ee_block
),
1705 ext4_ext_is_uninitialized(ex
),
1706 ext4_ext_get_actual_len(ex
),
1707 ext4_ext_pblock(ex
));
1708 err
= ext4_ext_get_access(handle
, inode
, path
+ depth
);
1713 * ext4_can_extents_be_merged should have checked that either
1714 * both extents are uninitialized, or both aren't. Thus we
1715 * need to check only one of them here.
1717 if (ext4_ext_is_uninitialized(ex
))
1719 ex
->ee_len
= cpu_to_le16(ext4_ext_get_actual_len(ex
)
1720 + ext4_ext_get_actual_len(newext
));
1722 ext4_ext_mark_uninitialized(ex
);
1723 eh
= path
[depth
].p_hdr
;
1728 depth
= ext_depth(inode
);
1729 eh
= path
[depth
].p_hdr
;
1730 if (le16_to_cpu(eh
->eh_entries
) < le16_to_cpu(eh
->eh_max
))
1733 /* probably next leaf has space for us? */
1734 fex
= EXT_LAST_EXTENT(eh
);
1735 next
= EXT_MAX_BLOCKS
;
1736 if (le32_to_cpu(newext
->ee_block
) > le32_to_cpu(fex
->ee_block
))
1737 next
= ext4_ext_next_leaf_block(path
);
1738 if (next
!= EXT_MAX_BLOCKS
) {
1739 ext_debug("next leaf block - %u\n", next
);
1740 BUG_ON(npath
!= NULL
);
1741 npath
= ext4_ext_find_extent(inode
, next
, NULL
);
1743 return PTR_ERR(npath
);
1744 BUG_ON(npath
->p_depth
!= path
->p_depth
);
1745 eh
= npath
[depth
].p_hdr
;
1746 if (le16_to_cpu(eh
->eh_entries
) < le16_to_cpu(eh
->eh_max
)) {
1747 ext_debug("next leaf isn't full(%d)\n",
1748 le16_to_cpu(eh
->eh_entries
));
1752 ext_debug("next leaf has no free space(%d,%d)\n",
1753 le16_to_cpu(eh
->eh_entries
), le16_to_cpu(eh
->eh_max
));
1757 * There is no free space in the found leaf.
1758 * We're gonna add a new leaf in the tree.
1760 if (flag
& EXT4_GET_BLOCKS_PUNCH_OUT_EXT
)
1761 flags
= EXT4_MB_USE_ROOT_BLOCKS
;
1762 err
= ext4_ext_create_new_leaf(handle
, inode
, flags
, path
, newext
);
1765 depth
= ext_depth(inode
);
1766 eh
= path
[depth
].p_hdr
;
1769 nearex
= path
[depth
].p_ext
;
1771 err
= ext4_ext_get_access(handle
, inode
, path
+ depth
);
1776 /* there is no extent in this leaf, create first one */
1777 ext_debug("first extent in the leaf: %u:%llu:[%d]%d\n",
1778 le32_to_cpu(newext
->ee_block
),
1779 ext4_ext_pblock(newext
),
1780 ext4_ext_is_uninitialized(newext
),
1781 ext4_ext_get_actual_len(newext
));
1782 nearex
= EXT_FIRST_EXTENT(eh
);
1784 if (le32_to_cpu(newext
->ee_block
)
1785 > le32_to_cpu(nearex
->ee_block
)) {
1787 ext_debug("insert %u:%llu:[%d]%d before: "
1789 le32_to_cpu(newext
->ee_block
),
1790 ext4_ext_pblock(newext
),
1791 ext4_ext_is_uninitialized(newext
),
1792 ext4_ext_get_actual_len(newext
),
1797 BUG_ON(newext
->ee_block
== nearex
->ee_block
);
1798 ext_debug("insert %u:%llu:[%d]%d after: "
1800 le32_to_cpu(newext
->ee_block
),
1801 ext4_ext_pblock(newext
),
1802 ext4_ext_is_uninitialized(newext
),
1803 ext4_ext_get_actual_len(newext
),
1806 len
= EXT_LAST_EXTENT(eh
) - nearex
+ 1;
1808 ext_debug("insert %u:%llu:[%d]%d: "
1809 "move %d extents from 0x%p to 0x%p\n",
1810 le32_to_cpu(newext
->ee_block
),
1811 ext4_ext_pblock(newext
),
1812 ext4_ext_is_uninitialized(newext
),
1813 ext4_ext_get_actual_len(newext
),
1814 len
, nearex
, nearex
+ 1);
1815 memmove(nearex
+ 1, nearex
,
1816 len
* sizeof(struct ext4_extent
));
1820 le16_add_cpu(&eh
->eh_entries
, 1);
1821 path
[depth
].p_ext
= nearex
;
1822 nearex
->ee_block
= newext
->ee_block
;
1823 ext4_ext_store_pblock(nearex
, ext4_ext_pblock(newext
));
1824 nearex
->ee_len
= newext
->ee_len
;
1827 /* try to merge extents to the right */
1828 if (!(flag
& EXT4_GET_BLOCKS_PRE_IO
))
1829 ext4_ext_try_to_merge(inode
, path
, nearex
);
1831 /* try to merge extents to the left */
1833 /* time to correct all indexes above */
1834 err
= ext4_ext_correct_indexes(handle
, inode
, path
);
1838 err
= ext4_ext_dirty(handle
, inode
, path
+ depth
);
1842 ext4_ext_drop_refs(npath
);
1845 ext4_ext_invalidate_cache(inode
);
1849 static int ext4_ext_walk_space(struct inode
*inode
, ext4_lblk_t block
,
1850 ext4_lblk_t num
, ext_prepare_callback func
,
1853 struct ext4_ext_path
*path
= NULL
;
1854 struct ext4_ext_cache cbex
;
1855 struct ext4_extent
*ex
;
1856 ext4_lblk_t next
, start
= 0, end
= 0;
1857 ext4_lblk_t last
= block
+ num
;
1858 int depth
, exists
, err
= 0;
1860 BUG_ON(func
== NULL
);
1861 BUG_ON(inode
== NULL
);
1863 while (block
< last
&& block
!= EXT_MAX_BLOCKS
) {
1865 /* find extent for this block */
1866 down_read(&EXT4_I(inode
)->i_data_sem
);
1867 path
= ext4_ext_find_extent(inode
, block
, path
);
1868 up_read(&EXT4_I(inode
)->i_data_sem
);
1870 err
= PTR_ERR(path
);
1875 depth
= ext_depth(inode
);
1876 if (unlikely(path
[depth
].p_hdr
== NULL
)) {
1877 EXT4_ERROR_INODE(inode
, "path[%d].p_hdr == NULL", depth
);
1881 ex
= path
[depth
].p_ext
;
1882 next
= ext4_ext_next_allocated_block(path
);
1886 /* there is no extent yet, so try to allocate
1887 * all requested space */
1890 } else if (le32_to_cpu(ex
->ee_block
) > block
) {
1891 /* need to allocate space before found extent */
1893 end
= le32_to_cpu(ex
->ee_block
);
1894 if (block
+ num
< end
)
1896 } else if (block
>= le32_to_cpu(ex
->ee_block
)
1897 + ext4_ext_get_actual_len(ex
)) {
1898 /* need to allocate space after found extent */
1903 } else if (block
>= le32_to_cpu(ex
->ee_block
)) {
1905 * some part of requested space is covered
1909 end
= le32_to_cpu(ex
->ee_block
)
1910 + ext4_ext_get_actual_len(ex
);
1911 if (block
+ num
< end
)
1917 BUG_ON(end
<= start
);
1920 cbex
.ec_block
= start
;
1921 cbex
.ec_len
= end
- start
;
1924 cbex
.ec_block
= le32_to_cpu(ex
->ee_block
);
1925 cbex
.ec_len
= ext4_ext_get_actual_len(ex
);
1926 cbex
.ec_start
= ext4_ext_pblock(ex
);
1929 if (unlikely(cbex
.ec_len
== 0)) {
1930 EXT4_ERROR_INODE(inode
, "cbex.ec_len == 0");
1934 err
= func(inode
, next
, &cbex
, ex
, cbdata
);
1935 ext4_ext_drop_refs(path
);
1940 if (err
== EXT_REPEAT
)
1942 else if (err
== EXT_BREAK
) {
1947 if (ext_depth(inode
) != depth
) {
1948 /* depth was changed. we have to realloc path */
1953 block
= cbex
.ec_block
+ cbex
.ec_len
;
1957 ext4_ext_drop_refs(path
);
1965 ext4_ext_put_in_cache(struct inode
*inode
, ext4_lblk_t block
,
1966 __u32 len
, ext4_fsblk_t start
)
1968 struct ext4_ext_cache
*cex
;
1970 spin_lock(&EXT4_I(inode
)->i_block_reservation_lock
);
1971 trace_ext4_ext_put_in_cache(inode
, block
, len
, start
);
1972 cex
= &EXT4_I(inode
)->i_cached_extent
;
1973 cex
->ec_block
= block
;
1975 cex
->ec_start
= start
;
1976 spin_unlock(&EXT4_I(inode
)->i_block_reservation_lock
);
1980 * ext4_ext_put_gap_in_cache:
1981 * calculate boundaries of the gap that the requested block fits into
1982 * and cache this gap
1985 ext4_ext_put_gap_in_cache(struct inode
*inode
, struct ext4_ext_path
*path
,
1988 int depth
= ext_depth(inode
);
1991 struct ext4_extent
*ex
;
1993 ex
= path
[depth
].p_ext
;
1995 /* there is no extent yet, so gap is [0;-] */
1997 len
= EXT_MAX_BLOCKS
;
1998 ext_debug("cache gap(whole file):");
1999 } else if (block
< le32_to_cpu(ex
->ee_block
)) {
2001 len
= le32_to_cpu(ex
->ee_block
) - block
;
2002 ext_debug("cache gap(before): %u [%u:%u]",
2004 le32_to_cpu(ex
->ee_block
),
2005 ext4_ext_get_actual_len(ex
));
2006 } else if (block
>= le32_to_cpu(ex
->ee_block
)
2007 + ext4_ext_get_actual_len(ex
)) {
2009 lblock
= le32_to_cpu(ex
->ee_block
)
2010 + ext4_ext_get_actual_len(ex
);
2012 next
= ext4_ext_next_allocated_block(path
);
2013 ext_debug("cache gap(after): [%u:%u] %u",
2014 le32_to_cpu(ex
->ee_block
),
2015 ext4_ext_get_actual_len(ex
),
2017 BUG_ON(next
== lblock
);
2018 len
= next
- lblock
;
2024 ext_debug(" -> %u:%lu\n", lblock
, len
);
2025 ext4_ext_put_in_cache(inode
, lblock
, len
, 0);
2029 * ext4_ext_check_cache()
2030 * Checks to see if the given block is in the cache.
2031 * If it is, the cached extent is stored in the given
2032 * cache extent pointer. If the cached extent is a hole,
2033 * this routine should be used instead of
2034 * ext4_ext_in_cache if the calling function needs to
2035 * know the size of the hole.
2037 * @inode: The files inode
2038 * @block: The block to look for in the cache
2039 * @ex: Pointer where the cached extent will be stored
2040 * if it contains block
2042 * Return 0 if cache is invalid; 1 if the cache is valid
2044 static int ext4_ext_check_cache(struct inode
*inode
, ext4_lblk_t block
,
2045 struct ext4_ext_cache
*ex
){
2046 struct ext4_ext_cache
*cex
;
2047 struct ext4_sb_info
*sbi
;
2051 * We borrow i_block_reservation_lock to protect i_cached_extent
2053 spin_lock(&EXT4_I(inode
)->i_block_reservation_lock
);
2054 cex
= &EXT4_I(inode
)->i_cached_extent
;
2055 sbi
= EXT4_SB(inode
->i_sb
);
2057 /* has cache valid data? */
2058 if (cex
->ec_len
== 0)
2061 if (in_range(block
, cex
->ec_block
, cex
->ec_len
)) {
2062 memcpy(ex
, cex
, sizeof(struct ext4_ext_cache
));
2063 ext_debug("%u cached by %u:%u:%llu\n",
2065 cex
->ec_block
, cex
->ec_len
, cex
->ec_start
);
2070 sbi
->extent_cache_misses
++;
2072 sbi
->extent_cache_hits
++;
2073 trace_ext4_ext_in_cache(inode
, block
, ret
);
2074 spin_unlock(&EXT4_I(inode
)->i_block_reservation_lock
);
2079 * ext4_ext_in_cache()
2080 * Checks to see if the given block is in the cache.
2081 * If it is, the cached extent is stored in the given
2084 * @inode: The files inode
2085 * @block: The block to look for in the cache
2086 * @ex: Pointer where the cached extent will be stored
2087 * if it contains block
2089 * Return 0 if cache is invalid; 1 if the cache is valid
2092 ext4_ext_in_cache(struct inode
*inode
, ext4_lblk_t block
,
2093 struct ext4_extent
*ex
)
2095 struct ext4_ext_cache cex
;
2098 if (ext4_ext_check_cache(inode
, block
, &cex
)) {
2099 ex
->ee_block
= cpu_to_le32(cex
.ec_block
);
2100 ext4_ext_store_pblock(ex
, cex
.ec_start
);
2101 ex
->ee_len
= cpu_to_le16(cex
.ec_len
);
2111 * removes index from the index block.
2113 static int ext4_ext_rm_idx(handle_t
*handle
, struct inode
*inode
,
2114 struct ext4_ext_path
*path
)
2119 /* free index block */
2121 leaf
= ext4_idx_pblock(path
->p_idx
);
2122 if (unlikely(path
->p_hdr
->eh_entries
== 0)) {
2123 EXT4_ERROR_INODE(inode
, "path->p_hdr->eh_entries == 0");
2126 err
= ext4_ext_get_access(handle
, inode
, path
);
2130 if (path
->p_idx
!= EXT_LAST_INDEX(path
->p_hdr
)) {
2131 int len
= EXT_LAST_INDEX(path
->p_hdr
) - path
->p_idx
;
2132 len
*= sizeof(struct ext4_extent_idx
);
2133 memmove(path
->p_idx
, path
->p_idx
+ 1, len
);
2136 le16_add_cpu(&path
->p_hdr
->eh_entries
, -1);
2137 err
= ext4_ext_dirty(handle
, inode
, path
);
2140 ext_debug("index is empty, remove it, free block %llu\n", leaf
);
2141 trace_ext4_ext_rm_idx(inode
, leaf
);
2143 ext4_free_blocks(handle
, inode
, NULL
, leaf
, 1,
2144 EXT4_FREE_BLOCKS_METADATA
| EXT4_FREE_BLOCKS_FORGET
);
2149 * ext4_ext_calc_credits_for_single_extent:
2150 * This routine returns max. credits that needed to insert an extent
2151 * to the extent tree.
2152 * When pass the actual path, the caller should calculate credits
2155 int ext4_ext_calc_credits_for_single_extent(struct inode
*inode
, int nrblocks
,
2156 struct ext4_ext_path
*path
)
2159 int depth
= ext_depth(inode
);
2162 /* probably there is space in leaf? */
2163 if (le16_to_cpu(path
[depth
].p_hdr
->eh_entries
)
2164 < le16_to_cpu(path
[depth
].p_hdr
->eh_max
)) {
2167 * There are some space in the leaf tree, no
2168 * need to account for leaf block credit
2170 * bitmaps and block group descriptor blocks
2171 * and other metadata blocks still need to be
2174 /* 1 bitmap, 1 block group descriptor */
2175 ret
= 2 + EXT4_META_TRANS_BLOCKS(inode
->i_sb
);
2180 return ext4_chunk_trans_blocks(inode
, nrblocks
);
2184 * How many index/leaf blocks need to change/allocate to modify nrblocks?
2186 * if nrblocks are fit in a single extent (chunk flag is 1), then
2187 * in the worse case, each tree level index/leaf need to be changed
2188 * if the tree split due to insert a new extent, then the old tree
2189 * index/leaf need to be updated too
2191 * If the nrblocks are discontiguous, they could cause
2192 * the whole tree split more than once, but this is really rare.
2194 int ext4_ext_index_trans_blocks(struct inode
*inode
, int nrblocks
, int chunk
)
2197 int depth
= ext_depth(inode
);
2207 static int ext4_remove_blocks(handle_t
*handle
, struct inode
*inode
,
2208 struct ext4_extent
*ex
,
2209 ext4_fsblk_t
*partial_cluster
,
2210 ext4_lblk_t from
, ext4_lblk_t to
)
2212 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
2213 unsigned short ee_len
= ext4_ext_get_actual_len(ex
);
2215 int flags
= EXT4_FREE_BLOCKS_FORGET
;
2217 if (S_ISDIR(inode
->i_mode
) || S_ISLNK(inode
->i_mode
))
2218 flags
|= EXT4_FREE_BLOCKS_METADATA
;
2220 * For bigalloc file systems, we never free a partial cluster
2221 * at the beginning of the extent. Instead, we make a note
2222 * that we tried freeing the cluster, and check to see if we
2223 * need to free it on a subsequent call to ext4_remove_blocks,
2224 * or at the end of the ext4_truncate() operation.
2226 flags
|= EXT4_FREE_BLOCKS_NOFREE_FIRST_CLUSTER
;
2228 trace_ext4_remove_blocks(inode
, ex
, from
, to
, *partial_cluster
);
2230 * If we have a partial cluster, and it's different from the
2231 * cluster of the last block, we need to explicitly free the
2232 * partial cluster here.
2234 pblk
= ext4_ext_pblock(ex
) + ee_len
- 1;
2235 if (*partial_cluster
&& (EXT4_B2C(sbi
, pblk
) != *partial_cluster
)) {
2236 ext4_free_blocks(handle
, inode
, NULL
,
2237 EXT4_C2B(sbi
, *partial_cluster
),
2238 sbi
->s_cluster_ratio
, flags
);
2239 *partial_cluster
= 0;
2242 #ifdef EXTENTS_STATS
2244 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
2245 spin_lock(&sbi
->s_ext_stats_lock
);
2246 sbi
->s_ext_blocks
+= ee_len
;
2247 sbi
->s_ext_extents
++;
2248 if (ee_len
< sbi
->s_ext_min
)
2249 sbi
->s_ext_min
= ee_len
;
2250 if (ee_len
> sbi
->s_ext_max
)
2251 sbi
->s_ext_max
= ee_len
;
2252 if (ext_depth(inode
) > sbi
->s_depth_max
)
2253 sbi
->s_depth_max
= ext_depth(inode
);
2254 spin_unlock(&sbi
->s_ext_stats_lock
);
2257 if (from
>= le32_to_cpu(ex
->ee_block
)
2258 && to
== le32_to_cpu(ex
->ee_block
) + ee_len
- 1) {
2262 num
= le32_to_cpu(ex
->ee_block
) + ee_len
- from
;
2263 pblk
= ext4_ext_pblock(ex
) + ee_len
- num
;
2264 ext_debug("free last %u blocks starting %llu\n", num
, pblk
);
2265 ext4_free_blocks(handle
, inode
, NULL
, pblk
, num
, flags
);
2267 * If the block range to be freed didn't start at the
2268 * beginning of a cluster, and we removed the entire
2269 * extent, save the partial cluster here, since we
2270 * might need to delete if we determine that the
2271 * truncate operation has removed all of the blocks in
2274 if (pblk
& (sbi
->s_cluster_ratio
- 1) &&
2276 *partial_cluster
= EXT4_B2C(sbi
, pblk
);
2278 *partial_cluster
= 0;
2279 } else if (from
== le32_to_cpu(ex
->ee_block
)
2280 && to
<= le32_to_cpu(ex
->ee_block
) + ee_len
- 1) {
2286 start
= ext4_ext_pblock(ex
);
2288 ext_debug("free first %u blocks starting %llu\n", num
, start
);
2289 ext4_free_blocks(handle
, inode
, NULL
, start
, num
, flags
);
2292 printk(KERN_INFO
"strange request: removal(2) "
2293 "%u-%u from %u:%u\n",
2294 from
, to
, le32_to_cpu(ex
->ee_block
), ee_len
);
2301 * ext4_ext_rm_leaf() Removes the extents associated with the
2302 * blocks appearing between "start" and "end", and splits the extents
2303 * if "start" and "end" appear in the same extent
2305 * @handle: The journal handle
2306 * @inode: The files inode
2307 * @path: The path to the leaf
2308 * @start: The first block to remove
2309 * @end: The last block to remove
2312 ext4_ext_rm_leaf(handle_t
*handle
, struct inode
*inode
,
2313 struct ext4_ext_path
*path
, ext4_fsblk_t
*partial_cluster
,
2314 ext4_lblk_t start
, ext4_lblk_t end
)
2316 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
2317 int err
= 0, correct_index
= 0;
2318 int depth
= ext_depth(inode
), credits
;
2319 struct ext4_extent_header
*eh
;
2322 ext4_lblk_t ex_ee_block
;
2323 unsigned short ex_ee_len
;
2324 unsigned uninitialized
= 0;
2325 struct ext4_extent
*ex
;
2327 /* the header must be checked already in ext4_ext_remove_space() */
2328 ext_debug("truncate since %u in leaf to %u\n", start
, end
);
2329 if (!path
[depth
].p_hdr
)
2330 path
[depth
].p_hdr
= ext_block_hdr(path
[depth
].p_bh
);
2331 eh
= path
[depth
].p_hdr
;
2332 if (unlikely(path
[depth
].p_hdr
== NULL
)) {
2333 EXT4_ERROR_INODE(inode
, "path[%d].p_hdr == NULL", depth
);
2336 /* find where to start removing */
2337 ex
= EXT_LAST_EXTENT(eh
);
2339 ex_ee_block
= le32_to_cpu(ex
->ee_block
);
2340 ex_ee_len
= ext4_ext_get_actual_len(ex
);
2342 trace_ext4_ext_rm_leaf(inode
, start
, ex
, *partial_cluster
);
2344 while (ex
>= EXT_FIRST_EXTENT(eh
) &&
2345 ex_ee_block
+ ex_ee_len
> start
) {
2347 if (ext4_ext_is_uninitialized(ex
))
2352 ext_debug("remove ext %u:[%d]%d\n", ex_ee_block
,
2353 uninitialized
, ex_ee_len
);
2354 path
[depth
].p_ext
= ex
;
2356 a
= ex_ee_block
> start
? ex_ee_block
: start
;
2357 b
= ex_ee_block
+ex_ee_len
- 1 < end
?
2358 ex_ee_block
+ex_ee_len
- 1 : end
;
2360 ext_debug(" border %u:%u\n", a
, b
);
2362 /* If this extent is beyond the end of the hole, skip it */
2363 if (end
< ex_ee_block
) {
2365 ex_ee_block
= le32_to_cpu(ex
->ee_block
);
2366 ex_ee_len
= ext4_ext_get_actual_len(ex
);
2368 } else if (b
!= ex_ee_block
+ ex_ee_len
- 1) {
2369 EXT4_ERROR_INODE(inode
,
2370 "can not handle truncate %u:%u "
2372 start
, end
, ex_ee_block
,
2373 ex_ee_block
+ ex_ee_len
- 1);
2376 } else if (a
!= ex_ee_block
) {
2377 /* remove tail of the extent */
2378 num
= a
- ex_ee_block
;
2380 /* remove whole extent: excellent! */
2384 * 3 for leaf, sb, and inode plus 2 (bmap and group
2385 * descriptor) for each block group; assume two block
2386 * groups plus ex_ee_len/blocks_per_block_group for
2389 credits
= 7 + 2*(ex_ee_len
/EXT4_BLOCKS_PER_GROUP(inode
->i_sb
));
2390 if (ex
== EXT_FIRST_EXTENT(eh
)) {
2392 credits
+= (ext_depth(inode
)) + 1;
2394 credits
+= EXT4_MAXQUOTAS_TRANS_BLOCKS(inode
->i_sb
);
2396 err
= ext4_ext_truncate_extend_restart(handle
, inode
, credits
);
2400 err
= ext4_ext_get_access(handle
, inode
, path
+ depth
);
2404 err
= ext4_remove_blocks(handle
, inode
, ex
, partial_cluster
,
2410 /* this extent is removed; mark slot entirely unused */
2411 ext4_ext_store_pblock(ex
, 0);
2413 ex
->ee_len
= cpu_to_le16(num
);
2415 * Do not mark uninitialized if all the blocks in the
2416 * extent have been removed.
2418 if (uninitialized
&& num
)
2419 ext4_ext_mark_uninitialized(ex
);
2421 * If the extent was completely released,
2422 * we need to remove it from the leaf
2425 if (end
!= EXT_MAX_BLOCKS
- 1) {
2427 * For hole punching, we need to scoot all the
2428 * extents up when an extent is removed so that
2429 * we dont have blank extents in the middle
2431 memmove(ex
, ex
+1, (EXT_LAST_EXTENT(eh
) - ex
) *
2432 sizeof(struct ext4_extent
));
2434 /* Now get rid of the one at the end */
2435 memset(EXT_LAST_EXTENT(eh
), 0,
2436 sizeof(struct ext4_extent
));
2438 le16_add_cpu(&eh
->eh_entries
, -1);
2440 *partial_cluster
= 0;
2442 err
= ext4_ext_dirty(handle
, inode
, path
+ depth
);
2446 ext_debug("new extent: %u:%u:%llu\n", ex_ee_block
, num
,
2447 ext4_ext_pblock(ex
));
2449 ex_ee_block
= le32_to_cpu(ex
->ee_block
);
2450 ex_ee_len
= ext4_ext_get_actual_len(ex
);
2453 if (correct_index
&& eh
->eh_entries
)
2454 err
= ext4_ext_correct_indexes(handle
, inode
, path
);
2457 * If there is still a entry in the leaf node, check to see if
2458 * it references the partial cluster. This is the only place
2459 * where it could; if it doesn't, we can free the cluster.
2461 if (*partial_cluster
&& ex
>= EXT_FIRST_EXTENT(eh
) &&
2462 (EXT4_B2C(sbi
, ext4_ext_pblock(ex
) + ex_ee_len
- 1) !=
2463 *partial_cluster
)) {
2464 int flags
= EXT4_FREE_BLOCKS_FORGET
;
2466 if (S_ISDIR(inode
->i_mode
) || S_ISLNK(inode
->i_mode
))
2467 flags
|= EXT4_FREE_BLOCKS_METADATA
;
2469 ext4_free_blocks(handle
, inode
, NULL
,
2470 EXT4_C2B(sbi
, *partial_cluster
),
2471 sbi
->s_cluster_ratio
, flags
);
2472 *partial_cluster
= 0;
2475 /* if this leaf is free, then we should
2476 * remove it from index block above */
2477 if (err
== 0 && eh
->eh_entries
== 0 && path
[depth
].p_bh
!= NULL
)
2478 err
= ext4_ext_rm_idx(handle
, inode
, path
+ depth
);
2485 * ext4_ext_more_to_rm:
2486 * returns 1 if current index has to be freed (even partial)
2489 ext4_ext_more_to_rm(struct ext4_ext_path
*path
)
2491 BUG_ON(path
->p_idx
== NULL
);
2493 if (path
->p_idx
< EXT_FIRST_INDEX(path
->p_hdr
))
2497 * if truncate on deeper level happened, it wasn't partial,
2498 * so we have to consider current index for truncation
2500 if (le16_to_cpu(path
->p_hdr
->eh_entries
) == path
->p_block
)
2505 static int ext4_ext_remove_space(struct inode
*inode
, ext4_lblk_t start
,
2508 struct super_block
*sb
= inode
->i_sb
;
2509 int depth
= ext_depth(inode
);
2510 struct ext4_ext_path
*path
;
2511 ext4_fsblk_t partial_cluster
= 0;
2515 ext_debug("truncate since %u to %u\n", start
, end
);
2517 /* probably first extent we're gonna free will be last in block */
2518 handle
= ext4_journal_start(inode
, depth
+ 1);
2520 return PTR_ERR(handle
);
2523 ext4_ext_invalidate_cache(inode
);
2525 trace_ext4_ext_remove_space(inode
, start
, depth
);
2528 * Check if we are removing extents inside the extent tree. If that
2529 * is the case, we are going to punch a hole inside the extent tree
2530 * so we have to check whether we need to split the extent covering
2531 * the last block to remove so we can easily remove the part of it
2532 * in ext4_ext_rm_leaf().
2534 if (end
< EXT_MAX_BLOCKS
- 1) {
2535 struct ext4_extent
*ex
;
2536 ext4_lblk_t ee_block
;
2538 /* find extent for this block */
2539 path
= ext4_ext_find_extent(inode
, end
, NULL
);
2541 ext4_journal_stop(handle
);
2542 return PTR_ERR(path
);
2544 depth
= ext_depth(inode
);
2545 ex
= path
[depth
].p_ext
;
2549 ee_block
= le32_to_cpu(ex
->ee_block
);
2552 * See if the last block is inside the extent, if so split
2553 * the extent at 'end' block so we can easily remove the
2554 * tail of the first part of the split extent in
2555 * ext4_ext_rm_leaf().
2557 if (end
>= ee_block
&&
2558 end
< ee_block
+ ext4_ext_get_actual_len(ex
) - 1) {
2561 if (ext4_ext_is_uninitialized(ex
))
2562 split_flag
= EXT4_EXT_MARK_UNINIT1
|
2563 EXT4_EXT_MARK_UNINIT2
;
2566 * Split the extent in two so that 'end' is the last
2567 * block in the first new extent
2569 err
= ext4_split_extent_at(handle
, inode
, path
,
2570 end
+ 1, split_flag
,
2571 EXT4_GET_BLOCKS_PRE_IO
|
2572 EXT4_GET_BLOCKS_PUNCH_OUT_EXT
);
2577 ext4_ext_drop_refs(path
);
2583 * We start scanning from right side, freeing all the blocks
2584 * after i_size and walking into the tree depth-wise.
2586 depth
= ext_depth(inode
);
2587 path
= kzalloc(sizeof(struct ext4_ext_path
) * (depth
+ 1), GFP_NOFS
);
2589 ext4_journal_stop(handle
);
2592 path
[0].p_depth
= depth
;
2593 path
[0].p_hdr
= ext_inode_hdr(inode
);
2595 if (ext4_ext_check(inode
, path
[0].p_hdr
, depth
)) {
2601 while (i
>= 0 && err
== 0) {
2603 /* this is leaf block */
2604 err
= ext4_ext_rm_leaf(handle
, inode
, path
,
2605 &partial_cluster
, start
,
2607 /* root level has p_bh == NULL, brelse() eats this */
2608 brelse(path
[i
].p_bh
);
2609 path
[i
].p_bh
= NULL
;
2614 /* this is index block */
2615 if (!path
[i
].p_hdr
) {
2616 ext_debug("initialize header\n");
2617 path
[i
].p_hdr
= ext_block_hdr(path
[i
].p_bh
);
2620 if (!path
[i
].p_idx
) {
2621 /* this level hasn't been touched yet */
2622 path
[i
].p_idx
= EXT_LAST_INDEX(path
[i
].p_hdr
);
2623 path
[i
].p_block
= le16_to_cpu(path
[i
].p_hdr
->eh_entries
)+1;
2624 ext_debug("init index ptr: hdr 0x%p, num %d\n",
2626 le16_to_cpu(path
[i
].p_hdr
->eh_entries
));
2628 /* we were already here, see at next index */
2632 ext_debug("level %d - index, first 0x%p, cur 0x%p\n",
2633 i
, EXT_FIRST_INDEX(path
[i
].p_hdr
),
2635 if (ext4_ext_more_to_rm(path
+ i
)) {
2636 struct buffer_head
*bh
;
2637 /* go to the next level */
2638 ext_debug("move to level %d (block %llu)\n",
2639 i
+ 1, ext4_idx_pblock(path
[i
].p_idx
));
2640 memset(path
+ i
+ 1, 0, sizeof(*path
));
2641 bh
= sb_bread(sb
, ext4_idx_pblock(path
[i
].p_idx
));
2643 /* should we reset i_size? */
2647 if (WARN_ON(i
+ 1 > depth
)) {
2651 if (ext4_ext_check(inode
, ext_block_hdr(bh
),
2656 path
[i
+ 1].p_bh
= bh
;
2658 /* save actual number of indexes since this
2659 * number is changed at the next iteration */
2660 path
[i
].p_block
= le16_to_cpu(path
[i
].p_hdr
->eh_entries
);
2663 /* we finished processing this index, go up */
2664 if (path
[i
].p_hdr
->eh_entries
== 0 && i
> 0) {
2665 /* index is empty, remove it;
2666 * handle must be already prepared by the
2667 * truncatei_leaf() */
2668 err
= ext4_ext_rm_idx(handle
, inode
, path
+ i
);
2670 /* root level has p_bh == NULL, brelse() eats this */
2671 brelse(path
[i
].p_bh
);
2672 path
[i
].p_bh
= NULL
;
2674 ext_debug("return to level %d\n", i
);
2678 trace_ext4_ext_remove_space_done(inode
, start
, depth
, partial_cluster
,
2679 path
->p_hdr
->eh_entries
);
2681 /* If we still have something in the partial cluster and we have removed
2682 * even the first extent, then we should free the blocks in the partial
2683 * cluster as well. */
2684 if (partial_cluster
&& path
->p_hdr
->eh_entries
== 0) {
2685 int flags
= EXT4_FREE_BLOCKS_FORGET
;
2687 if (S_ISDIR(inode
->i_mode
) || S_ISLNK(inode
->i_mode
))
2688 flags
|= EXT4_FREE_BLOCKS_METADATA
;
2690 ext4_free_blocks(handle
, inode
, NULL
,
2691 EXT4_C2B(EXT4_SB(sb
), partial_cluster
),
2692 EXT4_SB(sb
)->s_cluster_ratio
, flags
);
2693 partial_cluster
= 0;
2696 /* TODO: flexible tree reduction should be here */
2697 if (path
->p_hdr
->eh_entries
== 0) {
2699 * truncate to zero freed all the tree,
2700 * so we need to correct eh_depth
2702 err
= ext4_ext_get_access(handle
, inode
, path
);
2704 ext_inode_hdr(inode
)->eh_depth
= 0;
2705 ext_inode_hdr(inode
)->eh_max
=
2706 cpu_to_le16(ext4_ext_space_root(inode
, 0));
2707 err
= ext4_ext_dirty(handle
, inode
, path
);
2711 ext4_ext_drop_refs(path
);
2715 ext4_journal_stop(handle
);
2721 * called at mount time
2723 void ext4_ext_init(struct super_block
*sb
)
2726 * possible initialization would be here
2729 if (EXT4_HAS_INCOMPAT_FEATURE(sb
, EXT4_FEATURE_INCOMPAT_EXTENTS
)) {
2730 #if defined(AGGRESSIVE_TEST) || defined(CHECK_BINSEARCH) || defined(EXTENTS_STATS)
2731 printk(KERN_INFO
"EXT4-fs: file extents enabled");
2732 #ifdef AGGRESSIVE_TEST
2733 printk(", aggressive tests");
2735 #ifdef CHECK_BINSEARCH
2736 printk(", check binsearch");
2738 #ifdef EXTENTS_STATS
2743 #ifdef EXTENTS_STATS
2744 spin_lock_init(&EXT4_SB(sb
)->s_ext_stats_lock
);
2745 EXT4_SB(sb
)->s_ext_min
= 1 << 30;
2746 EXT4_SB(sb
)->s_ext_max
= 0;
2752 * called at umount time
2754 void ext4_ext_release(struct super_block
*sb
)
2756 if (!EXT4_HAS_INCOMPAT_FEATURE(sb
, EXT4_FEATURE_INCOMPAT_EXTENTS
))
2759 #ifdef EXTENTS_STATS
2760 if (EXT4_SB(sb
)->s_ext_blocks
&& EXT4_SB(sb
)->s_ext_extents
) {
2761 struct ext4_sb_info
*sbi
= EXT4_SB(sb
);
2762 printk(KERN_ERR
"EXT4-fs: %lu blocks in %lu extents (%lu ave)\n",
2763 sbi
->s_ext_blocks
, sbi
->s_ext_extents
,
2764 sbi
->s_ext_blocks
/ sbi
->s_ext_extents
);
2765 printk(KERN_ERR
"EXT4-fs: extents: %lu min, %lu max, max depth %lu\n",
2766 sbi
->s_ext_min
, sbi
->s_ext_max
, sbi
->s_depth_max
);
2771 /* FIXME!! we need to try to merge to left or right after zero-out */
2772 static int ext4_ext_zeroout(struct inode
*inode
, struct ext4_extent
*ex
)
2774 ext4_fsblk_t ee_pblock
;
2775 unsigned int ee_len
;
2778 ee_len
= ext4_ext_get_actual_len(ex
);
2779 ee_pblock
= ext4_ext_pblock(ex
);
2781 ret
= sb_issue_zeroout(inode
->i_sb
, ee_pblock
, ee_len
, GFP_NOFS
);
2789 * ext4_split_extent_at() splits an extent at given block.
2791 * @handle: the journal handle
2792 * @inode: the file inode
2793 * @path: the path to the extent
2794 * @split: the logical block where the extent is splitted.
2795 * @split_flags: indicates if the extent could be zeroout if split fails, and
2796 * the states(init or uninit) of new extents.
2797 * @flags: flags used to insert new extent to extent tree.
2800 * Splits extent [a, b] into two extents [a, @split) and [@split, b], states
2801 * of which are deterimined by split_flag.
2803 * There are two cases:
2804 * a> the extent are splitted into two extent.
2805 * b> split is not needed, and just mark the extent.
2807 * return 0 on success.
2809 static int ext4_split_extent_at(handle_t
*handle
,
2810 struct inode
*inode
,
2811 struct ext4_ext_path
*path
,
2816 ext4_fsblk_t newblock
;
2817 ext4_lblk_t ee_block
;
2818 struct ext4_extent
*ex
, newex
, orig_ex
;
2819 struct ext4_extent
*ex2
= NULL
;
2820 unsigned int ee_len
, depth
;
2823 ext_debug("ext4_split_extents_at: inode %lu, logical"
2824 "block %llu\n", inode
->i_ino
, (unsigned long long)split
);
2826 ext4_ext_show_leaf(inode
, path
);
2828 depth
= ext_depth(inode
);
2829 ex
= path
[depth
].p_ext
;
2830 ee_block
= le32_to_cpu(ex
->ee_block
);
2831 ee_len
= ext4_ext_get_actual_len(ex
);
2832 newblock
= split
- ee_block
+ ext4_ext_pblock(ex
);
2834 BUG_ON(split
< ee_block
|| split
>= (ee_block
+ ee_len
));
2836 err
= ext4_ext_get_access(handle
, inode
, path
+ depth
);
2840 if (split
== ee_block
) {
2842 * case b: block @split is the block that the extent begins with
2843 * then we just change the state of the extent, and splitting
2846 if (split_flag
& EXT4_EXT_MARK_UNINIT2
)
2847 ext4_ext_mark_uninitialized(ex
);
2849 ext4_ext_mark_initialized(ex
);
2851 if (!(flags
& EXT4_GET_BLOCKS_PRE_IO
))
2852 ext4_ext_try_to_merge(inode
, path
, ex
);
2854 err
= ext4_ext_dirty(handle
, inode
, path
+ depth
);
2859 memcpy(&orig_ex
, ex
, sizeof(orig_ex
));
2860 ex
->ee_len
= cpu_to_le16(split
- ee_block
);
2861 if (split_flag
& EXT4_EXT_MARK_UNINIT1
)
2862 ext4_ext_mark_uninitialized(ex
);
2865 * path may lead to new leaf, not to original leaf any more
2866 * after ext4_ext_insert_extent() returns,
2868 err
= ext4_ext_dirty(handle
, inode
, path
+ depth
);
2870 goto fix_extent_len
;
2873 ex2
->ee_block
= cpu_to_le32(split
);
2874 ex2
->ee_len
= cpu_to_le16(ee_len
- (split
- ee_block
));
2875 ext4_ext_store_pblock(ex2
, newblock
);
2876 if (split_flag
& EXT4_EXT_MARK_UNINIT2
)
2877 ext4_ext_mark_uninitialized(ex2
);
2879 err
= ext4_ext_insert_extent(handle
, inode
, path
, &newex
, flags
);
2880 if (err
== -ENOSPC
&& (EXT4_EXT_MAY_ZEROOUT
& split_flag
)) {
2881 err
= ext4_ext_zeroout(inode
, &orig_ex
);
2883 goto fix_extent_len
;
2884 /* update the extent length and mark as initialized */
2885 ex
->ee_len
= cpu_to_le32(ee_len
);
2886 ext4_ext_try_to_merge(inode
, path
, ex
);
2887 err
= ext4_ext_dirty(handle
, inode
, path
+ depth
);
2890 goto fix_extent_len
;
2893 ext4_ext_show_leaf(inode
, path
);
2897 ex
->ee_len
= orig_ex
.ee_len
;
2898 ext4_ext_dirty(handle
, inode
, path
+ depth
);
2903 * ext4_split_extents() splits an extent and mark extent which is covered
2904 * by @map as split_flags indicates
2906 * It may result in splitting the extent into multiple extents (upto three)
2907 * There are three possibilities:
2908 * a> There is no split required
2909 * b> Splits in two extents: Split is happening at either end of the extent
2910 * c> Splits in three extents: Somone is splitting in middle of the extent
2913 static int ext4_split_extent(handle_t
*handle
,
2914 struct inode
*inode
,
2915 struct ext4_ext_path
*path
,
2916 struct ext4_map_blocks
*map
,
2920 ext4_lblk_t ee_block
;
2921 struct ext4_extent
*ex
;
2922 unsigned int ee_len
, depth
;
2925 int split_flag1
, flags1
;
2927 depth
= ext_depth(inode
);
2928 ex
= path
[depth
].p_ext
;
2929 ee_block
= le32_to_cpu(ex
->ee_block
);
2930 ee_len
= ext4_ext_get_actual_len(ex
);
2931 uninitialized
= ext4_ext_is_uninitialized(ex
);
2933 if (map
->m_lblk
+ map
->m_len
< ee_block
+ ee_len
) {
2934 split_flag1
= split_flag
& EXT4_EXT_MAY_ZEROOUT
?
2935 EXT4_EXT_MAY_ZEROOUT
: 0;
2936 flags1
= flags
| EXT4_GET_BLOCKS_PRE_IO
;
2938 split_flag1
|= EXT4_EXT_MARK_UNINIT1
|
2939 EXT4_EXT_MARK_UNINIT2
;
2940 err
= ext4_split_extent_at(handle
, inode
, path
,
2941 map
->m_lblk
+ map
->m_len
, split_flag1
, flags1
);
2946 ext4_ext_drop_refs(path
);
2947 path
= ext4_ext_find_extent(inode
, map
->m_lblk
, path
);
2949 return PTR_ERR(path
);
2951 if (map
->m_lblk
>= ee_block
) {
2952 split_flag1
= split_flag
& EXT4_EXT_MAY_ZEROOUT
?
2953 EXT4_EXT_MAY_ZEROOUT
: 0;
2955 split_flag1
|= EXT4_EXT_MARK_UNINIT1
;
2956 if (split_flag
& EXT4_EXT_MARK_UNINIT2
)
2957 split_flag1
|= EXT4_EXT_MARK_UNINIT2
;
2958 err
= ext4_split_extent_at(handle
, inode
, path
,
2959 map
->m_lblk
, split_flag1
, flags
);
2964 ext4_ext_show_leaf(inode
, path
);
2966 return err
? err
: map
->m_len
;
2969 #define EXT4_EXT_ZERO_LEN 7
2971 * This function is called by ext4_ext_map_blocks() if someone tries to write
2972 * to an uninitialized extent. It may result in splitting the uninitialized
2973 * extent into multiple extents (up to three - one initialized and two
2975 * There are three possibilities:
2976 * a> There is no split required: Entire extent should be initialized
2977 * b> Splits in two extents: Write is happening at either end of the extent
2978 * c> Splits in three extents: Somone is writing in middle of the extent
2981 * - The extent pointed to by 'path' is uninitialized.
2982 * - The extent pointed to by 'path' contains a superset
2983 * of the logical span [map->m_lblk, map->m_lblk + map->m_len).
2985 * Post-conditions on success:
2986 * - the returned value is the number of blocks beyond map->l_lblk
2987 * that are allocated and initialized.
2988 * It is guaranteed to be >= map->m_len.
2990 static int ext4_ext_convert_to_initialized(handle_t
*handle
,
2991 struct inode
*inode
,
2992 struct ext4_map_blocks
*map
,
2993 struct ext4_ext_path
*path
)
2995 struct ext4_extent_header
*eh
;
2996 struct ext4_map_blocks split_map
;
2997 struct ext4_extent zero_ex
;
2998 struct ext4_extent
*ex
;
2999 ext4_lblk_t ee_block
, eof_block
;
3000 unsigned int ee_len
, depth
;
3005 ext_debug("ext4_ext_convert_to_initialized: inode %lu, logical"
3006 "block %llu, max_blocks %u\n", inode
->i_ino
,
3007 (unsigned long long)map
->m_lblk
, map
->m_len
);
3009 eof_block
= (inode
->i_size
+ inode
->i_sb
->s_blocksize
- 1) >>
3010 inode
->i_sb
->s_blocksize_bits
;
3011 if (eof_block
< map
->m_lblk
+ map
->m_len
)
3012 eof_block
= map
->m_lblk
+ map
->m_len
;
3014 depth
= ext_depth(inode
);
3015 eh
= path
[depth
].p_hdr
;
3016 ex
= path
[depth
].p_ext
;
3017 ee_block
= le32_to_cpu(ex
->ee_block
);
3018 ee_len
= ext4_ext_get_actual_len(ex
);
3019 allocated
= ee_len
- (map
->m_lblk
- ee_block
);
3021 trace_ext4_ext_convert_to_initialized_enter(inode
, map
, ex
);
3023 /* Pre-conditions */
3024 BUG_ON(!ext4_ext_is_uninitialized(ex
));
3025 BUG_ON(!in_range(map
->m_lblk
, ee_block
, ee_len
));
3028 * Attempt to transfer newly initialized blocks from the currently
3029 * uninitialized extent to its left neighbor. This is much cheaper
3030 * than an insertion followed by a merge as those involve costly
3031 * memmove() calls. This is the common case in steady state for
3032 * workloads doing fallocate(FALLOC_FL_KEEP_SIZE) followed by append
3035 * Limitations of the current logic:
3036 * - L1: we only deal with writes at the start of the extent.
3037 * The approach could be extended to writes at the end
3038 * of the extent but this scenario was deemed less common.
3039 * - L2: we do not deal with writes covering the whole extent.
3040 * This would require removing the extent if the transfer
3042 * - L3: we only attempt to merge with an extent stored in the
3043 * same extent tree node.
3045 if ((map
->m_lblk
== ee_block
) && /*L1*/
3046 (map
->m_len
< ee_len
) && /*L2*/
3047 (ex
> EXT_FIRST_EXTENT(eh
))) { /*L3*/
3048 struct ext4_extent
*prev_ex
;
3049 ext4_lblk_t prev_lblk
;
3050 ext4_fsblk_t prev_pblk
, ee_pblk
;
3051 unsigned int prev_len
, write_len
;
3054 prev_lblk
= le32_to_cpu(prev_ex
->ee_block
);
3055 prev_len
= ext4_ext_get_actual_len(prev_ex
);
3056 prev_pblk
= ext4_ext_pblock(prev_ex
);
3057 ee_pblk
= ext4_ext_pblock(ex
);
3058 write_len
= map
->m_len
;
3061 * A transfer of blocks from 'ex' to 'prev_ex' is allowed
3062 * upon those conditions:
3063 * - C1: prev_ex is initialized,
3064 * - C2: prev_ex is logically abutting ex,
3065 * - C3: prev_ex is physically abutting ex,
3066 * - C4: prev_ex can receive the additional blocks without
3067 * overflowing the (initialized) length limit.
3069 if ((!ext4_ext_is_uninitialized(prev_ex
)) && /*C1*/
3070 ((prev_lblk
+ prev_len
) == ee_block
) && /*C2*/
3071 ((prev_pblk
+ prev_len
) == ee_pblk
) && /*C3*/
3072 (prev_len
< (EXT_INIT_MAX_LEN
- write_len
))) { /*C4*/
3073 err
= ext4_ext_get_access(handle
, inode
, path
+ depth
);
3077 trace_ext4_ext_convert_to_initialized_fastpath(inode
,
3080 /* Shift the start of ex by 'write_len' blocks */
3081 ex
->ee_block
= cpu_to_le32(ee_block
+ write_len
);
3082 ext4_ext_store_pblock(ex
, ee_pblk
+ write_len
);
3083 ex
->ee_len
= cpu_to_le16(ee_len
- write_len
);
3084 ext4_ext_mark_uninitialized(ex
); /* Restore the flag */
3086 /* Extend prev_ex by 'write_len' blocks */
3087 prev_ex
->ee_len
= cpu_to_le16(prev_len
+ write_len
);
3089 /* Mark the block containing both extents as dirty */
3090 ext4_ext_dirty(handle
, inode
, path
+ depth
);
3092 /* Update path to point to the right extent */
3093 path
[depth
].p_ext
= prev_ex
;
3095 /* Result: number of initialized blocks past m_lblk */
3096 allocated
= write_len
;
3101 WARN_ON(map
->m_lblk
< ee_block
);
3103 * It is safe to convert extent to initialized via explicit
3104 * zeroout only if extent is fully insde i_size or new_size.
3106 split_flag
|= ee_block
+ ee_len
<= eof_block
? EXT4_EXT_MAY_ZEROOUT
: 0;
3108 /* If extent has less than 2*EXT4_EXT_ZERO_LEN zerout directly */
3109 if (ee_len
<= 2*EXT4_EXT_ZERO_LEN
&&
3110 (EXT4_EXT_MAY_ZEROOUT
& split_flag
)) {
3111 err
= ext4_ext_zeroout(inode
, ex
);
3115 err
= ext4_ext_get_access(handle
, inode
, path
+ depth
);
3118 ext4_ext_mark_initialized(ex
);
3119 ext4_ext_try_to_merge(inode
, path
, ex
);
3120 err
= ext4_ext_dirty(handle
, inode
, path
+ depth
);
3126 * 1. split the extent into three extents.
3127 * 2. split the extent into two extents, zeroout the first half.
3128 * 3. split the extent into two extents, zeroout the second half.
3129 * 4. split the extent into two extents with out zeroout.
3131 split_map
.m_lblk
= map
->m_lblk
;
3132 split_map
.m_len
= map
->m_len
;
3134 if (allocated
> map
->m_len
) {
3135 if (allocated
<= EXT4_EXT_ZERO_LEN
&&
3136 (EXT4_EXT_MAY_ZEROOUT
& split_flag
)) {
3139 cpu_to_le32(map
->m_lblk
);
3140 zero_ex
.ee_len
= cpu_to_le16(allocated
);
3141 ext4_ext_store_pblock(&zero_ex
,
3142 ext4_ext_pblock(ex
) + map
->m_lblk
- ee_block
);
3143 err
= ext4_ext_zeroout(inode
, &zero_ex
);
3146 split_map
.m_lblk
= map
->m_lblk
;
3147 split_map
.m_len
= allocated
;
3148 } else if ((map
->m_lblk
- ee_block
+ map
->m_len
<
3149 EXT4_EXT_ZERO_LEN
) &&
3150 (EXT4_EXT_MAY_ZEROOUT
& split_flag
)) {
3152 if (map
->m_lblk
!= ee_block
) {
3153 zero_ex
.ee_block
= ex
->ee_block
;
3154 zero_ex
.ee_len
= cpu_to_le16(map
->m_lblk
-
3156 ext4_ext_store_pblock(&zero_ex
,
3157 ext4_ext_pblock(ex
));
3158 err
= ext4_ext_zeroout(inode
, &zero_ex
);
3163 split_map
.m_lblk
= ee_block
;
3164 split_map
.m_len
= map
->m_lblk
- ee_block
+ map
->m_len
;
3165 allocated
= map
->m_len
;
3169 allocated
= ext4_split_extent(handle
, inode
, path
,
3170 &split_map
, split_flag
, 0);
3175 return err
? err
: allocated
;
3179 * This function is called by ext4_ext_map_blocks() from
3180 * ext4_get_blocks_dio_write() when DIO to write
3181 * to an uninitialized extent.
3183 * Writing to an uninitialized extent may result in splitting the uninitialized
3184 * extent into multiple /initialized uninitialized extents (up to three)
3185 * There are three possibilities:
3186 * a> There is no split required: Entire extent should be uninitialized
3187 * b> Splits in two extents: Write is happening at either end of the extent
3188 * c> Splits in three extents: Somone is writing in middle of the extent
3190 * One of more index blocks maybe needed if the extent tree grow after
3191 * the uninitialized extent split. To prevent ENOSPC occur at the IO
3192 * complete, we need to split the uninitialized extent before DIO submit
3193 * the IO. The uninitialized extent called at this time will be split
3194 * into three uninitialized extent(at most). After IO complete, the part
3195 * being filled will be convert to initialized by the end_io callback function
3196 * via ext4_convert_unwritten_extents().
3198 * Returns the size of uninitialized extent to be written on success.
3200 static int ext4_split_unwritten_extents(handle_t
*handle
,
3201 struct inode
*inode
,
3202 struct ext4_map_blocks
*map
,
3203 struct ext4_ext_path
*path
,
3206 ext4_lblk_t eof_block
;
3207 ext4_lblk_t ee_block
;
3208 struct ext4_extent
*ex
;
3209 unsigned int ee_len
;
3210 int split_flag
= 0, depth
;
3212 ext_debug("ext4_split_unwritten_extents: inode %lu, logical"
3213 "block %llu, max_blocks %u\n", inode
->i_ino
,
3214 (unsigned long long)map
->m_lblk
, map
->m_len
);
3216 eof_block
= (inode
->i_size
+ inode
->i_sb
->s_blocksize
- 1) >>
3217 inode
->i_sb
->s_blocksize_bits
;
3218 if (eof_block
< map
->m_lblk
+ map
->m_len
)
3219 eof_block
= map
->m_lblk
+ map
->m_len
;
3221 * It is safe to convert extent to initialized via explicit
3222 * zeroout only if extent is fully insde i_size or new_size.
3224 depth
= ext_depth(inode
);
3225 ex
= path
[depth
].p_ext
;
3226 ee_block
= le32_to_cpu(ex
->ee_block
);
3227 ee_len
= ext4_ext_get_actual_len(ex
);
3229 split_flag
|= ee_block
+ ee_len
<= eof_block
? EXT4_EXT_MAY_ZEROOUT
: 0;
3230 split_flag
|= EXT4_EXT_MARK_UNINIT2
;
3232 flags
|= EXT4_GET_BLOCKS_PRE_IO
;
3233 return ext4_split_extent(handle
, inode
, path
, map
, split_flag
, flags
);
3236 static int ext4_convert_unwritten_extents_endio(handle_t
*handle
,
3237 struct inode
*inode
,
3238 struct ext4_ext_path
*path
)
3240 struct ext4_extent
*ex
;
3244 depth
= ext_depth(inode
);
3245 ex
= path
[depth
].p_ext
;
3247 ext_debug("ext4_convert_unwritten_extents_endio: inode %lu, logical"
3248 "block %llu, max_blocks %u\n", inode
->i_ino
,
3249 (unsigned long long)le32_to_cpu(ex
->ee_block
),
3250 ext4_ext_get_actual_len(ex
));
3252 err
= ext4_ext_get_access(handle
, inode
, path
+ depth
);
3255 /* first mark the extent as initialized */
3256 ext4_ext_mark_initialized(ex
);
3258 /* note: ext4_ext_correct_indexes() isn't needed here because
3259 * borders are not changed
3261 ext4_ext_try_to_merge(inode
, path
, ex
);
3263 /* Mark modified extent as dirty */
3264 err
= ext4_ext_dirty(handle
, inode
, path
+ depth
);
3266 ext4_ext_show_leaf(inode
, path
);
3270 static void unmap_underlying_metadata_blocks(struct block_device
*bdev
,
3271 sector_t block
, int count
)
3274 for (i
= 0; i
< count
; i
++)
3275 unmap_underlying_metadata(bdev
, block
+ i
);
3279 * Handle EOFBLOCKS_FL flag, clearing it if necessary
3281 static int check_eofblocks_fl(handle_t
*handle
, struct inode
*inode
,
3283 struct ext4_ext_path
*path
,
3287 struct ext4_extent_header
*eh
;
3288 struct ext4_extent
*last_ex
;
3290 if (!ext4_test_inode_flag(inode
, EXT4_INODE_EOFBLOCKS
))
3293 depth
= ext_depth(inode
);
3294 eh
= path
[depth
].p_hdr
;
3296 if (unlikely(!eh
->eh_entries
)) {
3297 EXT4_ERROR_INODE(inode
, "eh->eh_entries == 0 and "
3298 "EOFBLOCKS_FL set");
3301 last_ex
= EXT_LAST_EXTENT(eh
);
3303 * We should clear the EOFBLOCKS_FL flag if we are writing the
3304 * last block in the last extent in the file. We test this by
3305 * first checking to see if the caller to
3306 * ext4_ext_get_blocks() was interested in the last block (or
3307 * a block beyond the last block) in the current extent. If
3308 * this turns out to be false, we can bail out from this
3309 * function immediately.
3311 if (lblk
+ len
< le32_to_cpu(last_ex
->ee_block
) +
3312 ext4_ext_get_actual_len(last_ex
))
3315 * If the caller does appear to be planning to write at or
3316 * beyond the end of the current extent, we then test to see
3317 * if the current extent is the last extent in the file, by
3318 * checking to make sure it was reached via the rightmost node
3319 * at each level of the tree.
3321 for (i
= depth
-1; i
>= 0; i
--)
3322 if (path
[i
].p_idx
!= EXT_LAST_INDEX(path
[i
].p_hdr
))
3324 ext4_clear_inode_flag(inode
, EXT4_INODE_EOFBLOCKS
);
3325 return ext4_mark_inode_dirty(handle
, inode
);
3329 * ext4_find_delalloc_range: find delayed allocated block in the given range.
3331 * Goes through the buffer heads in the range [lblk_start, lblk_end] and returns
3332 * whether there are any buffers marked for delayed allocation. It returns '1'
3333 * on the first delalloc'ed buffer head found. If no buffer head in the given
3334 * range is marked for delalloc, it returns 0.
3335 * lblk_start should always be <= lblk_end.
3336 * search_hint_reverse is to indicate that searching in reverse from lblk_end to
3337 * lblk_start might be more efficient (i.e., we will likely hit the delalloc'ed
3338 * block sooner). This is useful when blocks are truncated sequentially from
3339 * lblk_start towards lblk_end.
3341 static int ext4_find_delalloc_range(struct inode
*inode
,
3342 ext4_lblk_t lblk_start
,
3343 ext4_lblk_t lblk_end
,
3344 int search_hint_reverse
)
3346 struct address_space
*mapping
= inode
->i_mapping
;
3347 struct buffer_head
*head
, *bh
= NULL
;
3349 ext4_lblk_t i
, pg_lblk
;
3352 if (!test_opt(inode
->i_sb
, DELALLOC
))
3355 /* reverse search wont work if fs block size is less than page size */
3356 if (inode
->i_blkbits
< PAGE_CACHE_SHIFT
)
3357 search_hint_reverse
= 0;
3359 if (search_hint_reverse
)
3364 index
= i
>> (PAGE_CACHE_SHIFT
- inode
->i_blkbits
);
3366 while ((i
>= lblk_start
) && (i
<= lblk_end
)) {
3367 page
= find_get_page(mapping
, index
);
3371 if (!page_has_buffers(page
))
3374 head
= page_buffers(page
);
3379 pg_lblk
= index
<< (PAGE_CACHE_SHIFT
-
3382 if (unlikely(pg_lblk
< lblk_start
)) {
3384 * This is possible when fs block size is less
3385 * than page size and our cluster starts/ends in
3386 * middle of the page. So we need to skip the
3387 * initial few blocks till we reach the 'lblk'
3393 /* Check if the buffer is delayed allocated and that it
3394 * is not yet mapped. (when da-buffers are mapped during
3395 * their writeout, their da_mapped bit is set.)
3397 if (buffer_delay(bh
) && !buffer_da_mapped(bh
)) {
3398 page_cache_release(page
);
3399 trace_ext4_find_delalloc_range(inode
,
3400 lblk_start
, lblk_end
,
3401 search_hint_reverse
,
3405 if (search_hint_reverse
)
3409 } while ((i
>= lblk_start
) && (i
<= lblk_end
) &&
3410 ((bh
= bh
->b_this_page
) != head
));
3413 page_cache_release(page
);
3415 * Move to next page. 'i' will be the first lblk in the next
3418 if (search_hint_reverse
)
3422 i
= index
<< (PAGE_CACHE_SHIFT
- inode
->i_blkbits
);
3425 trace_ext4_find_delalloc_range(inode
, lblk_start
, lblk_end
,
3426 search_hint_reverse
, 0, 0);
3430 int ext4_find_delalloc_cluster(struct inode
*inode
, ext4_lblk_t lblk
,
3431 int search_hint_reverse
)
3433 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
3434 ext4_lblk_t lblk_start
, lblk_end
;
3435 lblk_start
= lblk
& (~(sbi
->s_cluster_ratio
- 1));
3436 lblk_end
= lblk_start
+ sbi
->s_cluster_ratio
- 1;
3438 return ext4_find_delalloc_range(inode
, lblk_start
, lblk_end
,
3439 search_hint_reverse
);
3443 * Determines how many complete clusters (out of those specified by the 'map')
3444 * are under delalloc and were reserved quota for.
3445 * This function is called when we are writing out the blocks that were
3446 * originally written with their allocation delayed, but then the space was
3447 * allocated using fallocate() before the delayed allocation could be resolved.
3448 * The cases to look for are:
3449 * ('=' indicated delayed allocated blocks
3450 * '-' indicates non-delayed allocated blocks)
3451 * (a) partial clusters towards beginning and/or end outside of allocated range
3452 * are not delalloc'ed.
3454 * |----c---=|====c====|====c====|===-c----|
3455 * |++++++ allocated ++++++|
3456 * ==> 4 complete clusters in above example
3458 * (b) partial cluster (outside of allocated range) towards either end is
3459 * marked for delayed allocation. In this case, we will exclude that
3462 * |----====c========|========c========|
3463 * |++++++ allocated ++++++|
3464 * ==> 1 complete clusters in above example
3467 * |================c================|
3468 * |++++++ allocated ++++++|
3469 * ==> 0 complete clusters in above example
3471 * The ext4_da_update_reserve_space will be called only if we
3472 * determine here that there were some "entire" clusters that span
3473 * this 'allocated' range.
3474 * In the non-bigalloc case, this function will just end up returning num_blks
3475 * without ever calling ext4_find_delalloc_range.
3478 get_reserved_cluster_alloc(struct inode
*inode
, ext4_lblk_t lblk_start
,
3479 unsigned int num_blks
)
3481 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
3482 ext4_lblk_t alloc_cluster_start
, alloc_cluster_end
;
3483 ext4_lblk_t lblk_from
, lblk_to
, c_offset
;
3484 unsigned int allocated_clusters
= 0;
3486 alloc_cluster_start
= EXT4_B2C(sbi
, lblk_start
);
3487 alloc_cluster_end
= EXT4_B2C(sbi
, lblk_start
+ num_blks
- 1);
3489 /* max possible clusters for this allocation */
3490 allocated_clusters
= alloc_cluster_end
- alloc_cluster_start
+ 1;
3492 trace_ext4_get_reserved_cluster_alloc(inode
, lblk_start
, num_blks
);
3494 /* Check towards left side */
3495 c_offset
= lblk_start
& (sbi
->s_cluster_ratio
- 1);
3497 lblk_from
= lblk_start
& (~(sbi
->s_cluster_ratio
- 1));
3498 lblk_to
= lblk_from
+ c_offset
- 1;
3500 if (ext4_find_delalloc_range(inode
, lblk_from
, lblk_to
, 0))
3501 allocated_clusters
--;
3504 /* Now check towards right. */
3505 c_offset
= (lblk_start
+ num_blks
) & (sbi
->s_cluster_ratio
- 1);
3506 if (allocated_clusters
&& c_offset
) {
3507 lblk_from
= lblk_start
+ num_blks
;
3508 lblk_to
= lblk_from
+ (sbi
->s_cluster_ratio
- c_offset
) - 1;
3510 if (ext4_find_delalloc_range(inode
, lblk_from
, lblk_to
, 0))
3511 allocated_clusters
--;
3514 return allocated_clusters
;
3518 ext4_ext_handle_uninitialized_extents(handle_t
*handle
, struct inode
*inode
,
3519 struct ext4_map_blocks
*map
,
3520 struct ext4_ext_path
*path
, int flags
,
3521 unsigned int allocated
, ext4_fsblk_t newblock
)
3525 ext4_io_end_t
*io
= EXT4_I(inode
)->cur_aio_dio
;
3527 ext_debug("ext4_ext_handle_uninitialized_extents: inode %lu, logical "
3528 "block %llu, max_blocks %u, flags %x, allocated %u\n",
3529 inode
->i_ino
, (unsigned long long)map
->m_lblk
, map
->m_len
,
3531 ext4_ext_show_leaf(inode
, path
);
3533 trace_ext4_ext_handle_uninitialized_extents(inode
, map
, allocated
,
3536 /* get_block() before submit the IO, split the extent */
3537 if ((flags
& EXT4_GET_BLOCKS_PRE_IO
)) {
3538 ret
= ext4_split_unwritten_extents(handle
, inode
, map
,
3541 * Flag the inode(non aio case) or end_io struct (aio case)
3542 * that this IO needs to conversion to written when IO is
3546 ext4_set_io_unwritten_flag(inode
, io
);
3548 ext4_set_inode_state(inode
, EXT4_STATE_DIO_UNWRITTEN
);
3549 if (ext4_should_dioread_nolock(inode
))
3550 map
->m_flags
|= EXT4_MAP_UNINIT
;
3553 /* IO end_io complete, convert the filled extent to written */
3554 if ((flags
& EXT4_GET_BLOCKS_CONVERT
)) {
3555 ret
= ext4_convert_unwritten_extents_endio(handle
, inode
,
3558 ext4_update_inode_fsync_trans(handle
, inode
, 1);
3559 err
= check_eofblocks_fl(handle
, inode
, map
->m_lblk
,
3565 /* buffered IO case */
3567 * repeat fallocate creation request
3568 * we already have an unwritten extent
3570 if (flags
& EXT4_GET_BLOCKS_UNINIT_EXT
)
3573 /* buffered READ or buffered write_begin() lookup */
3574 if ((flags
& EXT4_GET_BLOCKS_CREATE
) == 0) {
3576 * We have blocks reserved already. We
3577 * return allocated blocks so that delalloc
3578 * won't do block reservation for us. But
3579 * the buffer head will be unmapped so that
3580 * a read from the block returns 0s.
3582 map
->m_flags
|= EXT4_MAP_UNWRITTEN
;
3586 /* buffered write, writepage time, convert*/
3587 ret
= ext4_ext_convert_to_initialized(handle
, inode
, map
, path
);
3589 ext4_update_inode_fsync_trans(handle
, inode
, 1);
3596 map
->m_flags
|= EXT4_MAP_NEW
;
3598 * if we allocated more blocks than requested
3599 * we need to make sure we unmap the extra block
3600 * allocated. The actual needed block will get
3601 * unmapped later when we find the buffer_head marked
3604 if (allocated
> map
->m_len
) {
3605 unmap_underlying_metadata_blocks(inode
->i_sb
->s_bdev
,
3606 newblock
+ map
->m_len
,
3607 allocated
- map
->m_len
);
3608 allocated
= map
->m_len
;
3612 * If we have done fallocate with the offset that is already
3613 * delayed allocated, we would have block reservation
3614 * and quota reservation done in the delayed write path.
3615 * But fallocate would have already updated quota and block
3616 * count for this offset. So cancel these reservation
3618 if (flags
& EXT4_GET_BLOCKS_DELALLOC_RESERVE
) {
3619 unsigned int reserved_clusters
;
3620 reserved_clusters
= get_reserved_cluster_alloc(inode
,
3621 map
->m_lblk
, map
->m_len
);
3622 if (reserved_clusters
)
3623 ext4_da_update_reserve_space(inode
,
3629 map
->m_flags
|= EXT4_MAP_MAPPED
;
3630 if ((flags
& EXT4_GET_BLOCKS_KEEP_SIZE
) == 0) {
3631 err
= check_eofblocks_fl(handle
, inode
, map
->m_lblk
, path
,
3637 if (allocated
> map
->m_len
)
3638 allocated
= map
->m_len
;
3639 ext4_ext_show_leaf(inode
, path
);
3640 map
->m_pblk
= newblock
;
3641 map
->m_len
= allocated
;
3644 ext4_ext_drop_refs(path
);
3647 return err
? err
: allocated
;
3651 * get_implied_cluster_alloc - check to see if the requested
3652 * allocation (in the map structure) overlaps with a cluster already
3653 * allocated in an extent.
3654 * @sb The filesystem superblock structure
3655 * @map The requested lblk->pblk mapping
3656 * @ex The extent structure which might contain an implied
3657 * cluster allocation
3659 * This function is called by ext4_ext_map_blocks() after we failed to
3660 * find blocks that were already in the inode's extent tree. Hence,
3661 * we know that the beginning of the requested region cannot overlap
3662 * the extent from the inode's extent tree. There are three cases we
3663 * want to catch. The first is this case:
3665 * |--- cluster # N--|
3666 * |--- extent ---| |---- requested region ---|
3669 * The second case that we need to test for is this one:
3671 * |--------- cluster # N ----------------|
3672 * |--- requested region --| |------- extent ----|
3673 * |=======================|
3675 * The third case is when the requested region lies between two extents
3676 * within the same cluster:
3677 * |------------- cluster # N-------------|
3678 * |----- ex -----| |---- ex_right ----|
3679 * |------ requested region ------|
3680 * |================|
3682 * In each of the above cases, we need to set the map->m_pblk and
3683 * map->m_len so it corresponds to the return the extent labelled as
3684 * "|====|" from cluster #N, since it is already in use for data in
3685 * cluster EXT4_B2C(sbi, map->m_lblk). We will then return 1 to
3686 * signal to ext4_ext_map_blocks() that map->m_pblk should be treated
3687 * as a new "allocated" block region. Otherwise, we will return 0 and
3688 * ext4_ext_map_blocks() will then allocate one or more new clusters
3689 * by calling ext4_mb_new_blocks().
3691 static int get_implied_cluster_alloc(struct super_block
*sb
,
3692 struct ext4_map_blocks
*map
,
3693 struct ext4_extent
*ex
,
3694 struct ext4_ext_path
*path
)
3696 struct ext4_sb_info
*sbi
= EXT4_SB(sb
);
3697 ext4_lblk_t c_offset
= map
->m_lblk
& (sbi
->s_cluster_ratio
-1);
3698 ext4_lblk_t ex_cluster_start
, ex_cluster_end
;
3699 ext4_lblk_t rr_cluster_start
;
3700 ext4_lblk_t ee_block
= le32_to_cpu(ex
->ee_block
);
3701 ext4_fsblk_t ee_start
= ext4_ext_pblock(ex
);
3702 unsigned short ee_len
= ext4_ext_get_actual_len(ex
);
3704 /* The extent passed in that we are trying to match */
3705 ex_cluster_start
= EXT4_B2C(sbi
, ee_block
);
3706 ex_cluster_end
= EXT4_B2C(sbi
, ee_block
+ ee_len
- 1);
3708 /* The requested region passed into ext4_map_blocks() */
3709 rr_cluster_start
= EXT4_B2C(sbi
, map
->m_lblk
);
3711 if ((rr_cluster_start
== ex_cluster_end
) ||
3712 (rr_cluster_start
== ex_cluster_start
)) {
3713 if (rr_cluster_start
== ex_cluster_end
)
3714 ee_start
+= ee_len
- 1;
3715 map
->m_pblk
= (ee_start
& ~(sbi
->s_cluster_ratio
- 1)) +
3717 map
->m_len
= min(map
->m_len
,
3718 (unsigned) sbi
->s_cluster_ratio
- c_offset
);
3720 * Check for and handle this case:
3722 * |--------- cluster # N-------------|
3723 * |------- extent ----|
3724 * |--- requested region ---|
3728 if (map
->m_lblk
< ee_block
)
3729 map
->m_len
= min(map
->m_len
, ee_block
- map
->m_lblk
);
3732 * Check for the case where there is already another allocated
3733 * block to the right of 'ex' but before the end of the cluster.
3735 * |------------- cluster # N-------------|
3736 * |----- ex -----| |---- ex_right ----|
3737 * |------ requested region ------|
3738 * |================|
3740 if (map
->m_lblk
> ee_block
) {
3741 ext4_lblk_t next
= ext4_ext_next_allocated_block(path
);
3742 map
->m_len
= min(map
->m_len
, next
- map
->m_lblk
);
3745 trace_ext4_get_implied_cluster_alloc_exit(sb
, map
, 1);
3749 trace_ext4_get_implied_cluster_alloc_exit(sb
, map
, 0);
3755 * Block allocation/map/preallocation routine for extents based files
3758 * Need to be called with
3759 * down_read(&EXT4_I(inode)->i_data_sem) if not allocating file system block
3760 * (ie, create is zero). Otherwise down_write(&EXT4_I(inode)->i_data_sem)
3762 * return > 0, number of of blocks already mapped/allocated
3763 * if create == 0 and these are pre-allocated blocks
3764 * buffer head is unmapped
3765 * otherwise blocks are mapped
3767 * return = 0, if plain look up failed (blocks have not been allocated)
3768 * buffer head is unmapped
3770 * return < 0, error case.
3772 int ext4_ext_map_blocks(handle_t
*handle
, struct inode
*inode
,
3773 struct ext4_map_blocks
*map
, int flags
)
3775 struct ext4_ext_path
*path
= NULL
;
3776 struct ext4_extent newex
, *ex
, *ex2
;
3777 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
3778 ext4_fsblk_t newblock
= 0;
3779 int free_on_err
= 0, err
= 0, depth
, ret
;
3780 unsigned int allocated
= 0, offset
= 0;
3781 unsigned int allocated_clusters
= 0;
3782 struct ext4_allocation_request ar
;
3783 ext4_io_end_t
*io
= EXT4_I(inode
)->cur_aio_dio
;
3784 ext4_lblk_t cluster_offset
;
3786 ext_debug("blocks %u/%u requested for inode %lu\n",
3787 map
->m_lblk
, map
->m_len
, inode
->i_ino
);
3788 trace_ext4_ext_map_blocks_enter(inode
, map
->m_lblk
, map
->m_len
, flags
);
3790 /* check in cache */
3791 if (ext4_ext_in_cache(inode
, map
->m_lblk
, &newex
)) {
3792 if (!newex
.ee_start_lo
&& !newex
.ee_start_hi
) {
3793 if ((sbi
->s_cluster_ratio
> 1) &&
3794 ext4_find_delalloc_cluster(inode
, map
->m_lblk
, 0))
3795 map
->m_flags
|= EXT4_MAP_FROM_CLUSTER
;
3797 if ((flags
& EXT4_GET_BLOCKS_CREATE
) == 0) {
3799 * block isn't allocated yet and
3800 * user doesn't want to allocate it
3804 /* we should allocate requested block */
3806 /* block is already allocated */
3807 if (sbi
->s_cluster_ratio
> 1)
3808 map
->m_flags
|= EXT4_MAP_FROM_CLUSTER
;
3809 newblock
= map
->m_lblk
3810 - le32_to_cpu(newex
.ee_block
)
3811 + ext4_ext_pblock(&newex
);
3812 /* number of remaining blocks in the extent */
3813 allocated
= ext4_ext_get_actual_len(&newex
) -
3814 (map
->m_lblk
- le32_to_cpu(newex
.ee_block
));
3819 /* find extent for this block */
3820 path
= ext4_ext_find_extent(inode
, map
->m_lblk
, NULL
);
3822 err
= PTR_ERR(path
);
3827 depth
= ext_depth(inode
);
3830 * consistent leaf must not be empty;
3831 * this situation is possible, though, _during_ tree modification;
3832 * this is why assert can't be put in ext4_ext_find_extent()
3834 if (unlikely(path
[depth
].p_ext
== NULL
&& depth
!= 0)) {
3835 EXT4_ERROR_INODE(inode
, "bad extent address "
3836 "lblock: %lu, depth: %d pblock %lld",
3837 (unsigned long) map
->m_lblk
, depth
,
3838 path
[depth
].p_block
);
3843 ex
= path
[depth
].p_ext
;
3845 ext4_lblk_t ee_block
= le32_to_cpu(ex
->ee_block
);
3846 ext4_fsblk_t ee_start
= ext4_ext_pblock(ex
);
3847 unsigned short ee_len
;
3850 * Uninitialized extents are treated as holes, except that
3851 * we split out initialized portions during a write.
3853 ee_len
= ext4_ext_get_actual_len(ex
);
3855 trace_ext4_ext_show_extent(inode
, ee_block
, ee_start
, ee_len
);
3857 /* if found extent covers block, simply return it */
3858 if (in_range(map
->m_lblk
, ee_block
, ee_len
)) {
3859 newblock
= map
->m_lblk
- ee_block
+ ee_start
;
3860 /* number of remaining blocks in the extent */
3861 allocated
= ee_len
- (map
->m_lblk
- ee_block
);
3862 ext_debug("%u fit into %u:%d -> %llu\n", map
->m_lblk
,
3863 ee_block
, ee_len
, newblock
);
3866 * Do not put uninitialized extent
3869 if (!ext4_ext_is_uninitialized(ex
)) {
3870 ext4_ext_put_in_cache(inode
, ee_block
,
3874 ret
= ext4_ext_handle_uninitialized_extents(
3875 handle
, inode
, map
, path
, flags
,
3876 allocated
, newblock
);
3881 if ((sbi
->s_cluster_ratio
> 1) &&
3882 ext4_find_delalloc_cluster(inode
, map
->m_lblk
, 0))
3883 map
->m_flags
|= EXT4_MAP_FROM_CLUSTER
;
3886 * requested block isn't allocated yet;
3887 * we couldn't try to create block if create flag is zero
3889 if ((flags
& EXT4_GET_BLOCKS_CREATE
) == 0) {
3891 * put just found gap into cache to speed up
3892 * subsequent requests
3894 ext4_ext_put_gap_in_cache(inode
, path
, map
->m_lblk
);
3899 * Okay, we need to do block allocation.
3901 map
->m_flags
&= ~EXT4_MAP_FROM_CLUSTER
;
3902 newex
.ee_block
= cpu_to_le32(map
->m_lblk
);
3903 cluster_offset
= map
->m_lblk
& (sbi
->s_cluster_ratio
-1);
3906 * If we are doing bigalloc, check to see if the extent returned
3907 * by ext4_ext_find_extent() implies a cluster we can use.
3909 if (cluster_offset
&& ex
&&
3910 get_implied_cluster_alloc(inode
->i_sb
, map
, ex
, path
)) {
3911 ar
.len
= allocated
= map
->m_len
;
3912 newblock
= map
->m_pblk
;
3913 map
->m_flags
|= EXT4_MAP_FROM_CLUSTER
;
3914 goto got_allocated_blocks
;
3917 /* find neighbour allocated blocks */
3918 ar
.lleft
= map
->m_lblk
;
3919 err
= ext4_ext_search_left(inode
, path
, &ar
.lleft
, &ar
.pleft
);
3922 ar
.lright
= map
->m_lblk
;
3924 err
= ext4_ext_search_right(inode
, path
, &ar
.lright
, &ar
.pright
, &ex2
);
3928 /* Check if the extent after searching to the right implies a
3929 * cluster we can use. */
3930 if ((sbi
->s_cluster_ratio
> 1) && ex2
&&
3931 get_implied_cluster_alloc(inode
->i_sb
, map
, ex2
, path
)) {
3932 ar
.len
= allocated
= map
->m_len
;
3933 newblock
= map
->m_pblk
;
3934 map
->m_flags
|= EXT4_MAP_FROM_CLUSTER
;
3935 goto got_allocated_blocks
;
3939 * See if request is beyond maximum number of blocks we can have in
3940 * a single extent. For an initialized extent this limit is
3941 * EXT_INIT_MAX_LEN and for an uninitialized extent this limit is
3942 * EXT_UNINIT_MAX_LEN.
3944 if (map
->m_len
> EXT_INIT_MAX_LEN
&&
3945 !(flags
& EXT4_GET_BLOCKS_UNINIT_EXT
))
3946 map
->m_len
= EXT_INIT_MAX_LEN
;
3947 else if (map
->m_len
> EXT_UNINIT_MAX_LEN
&&
3948 (flags
& EXT4_GET_BLOCKS_UNINIT_EXT
))
3949 map
->m_len
= EXT_UNINIT_MAX_LEN
;
3951 /* Check if we can really insert (m_lblk)::(m_lblk + m_len) extent */
3952 newex
.ee_len
= cpu_to_le16(map
->m_len
);
3953 err
= ext4_ext_check_overlap(sbi
, inode
, &newex
, path
);
3955 allocated
= ext4_ext_get_actual_len(&newex
);
3957 allocated
= map
->m_len
;
3959 /* allocate new block */
3961 ar
.goal
= ext4_ext_find_goal(inode
, path
, map
->m_lblk
);
3962 ar
.logical
= map
->m_lblk
;
3964 * We calculate the offset from the beginning of the cluster
3965 * for the logical block number, since when we allocate a
3966 * physical cluster, the physical block should start at the
3967 * same offset from the beginning of the cluster. This is
3968 * needed so that future calls to get_implied_cluster_alloc()
3971 offset
= map
->m_lblk
& (sbi
->s_cluster_ratio
- 1);
3972 ar
.len
= EXT4_NUM_B2C(sbi
, offset
+allocated
);
3974 ar
.logical
-= offset
;
3975 if (S_ISREG(inode
->i_mode
))
3976 ar
.flags
= EXT4_MB_HINT_DATA
;
3978 /* disable in-core preallocation for non-regular files */
3980 if (flags
& EXT4_GET_BLOCKS_NO_NORMALIZE
)
3981 ar
.flags
|= EXT4_MB_HINT_NOPREALLOC
;
3982 newblock
= ext4_mb_new_blocks(handle
, &ar
, &err
);
3985 ext_debug("allocate new block: goal %llu, found %llu/%u\n",
3986 ar
.goal
, newblock
, allocated
);
3988 allocated_clusters
= ar
.len
;
3989 ar
.len
= EXT4_C2B(sbi
, ar
.len
) - offset
;
3990 if (ar
.len
> allocated
)
3993 got_allocated_blocks
:
3994 /* try to insert new extent into found leaf and return */
3995 ext4_ext_store_pblock(&newex
, newblock
+ offset
);
3996 newex
.ee_len
= cpu_to_le16(ar
.len
);
3997 /* Mark uninitialized */
3998 if (flags
& EXT4_GET_BLOCKS_UNINIT_EXT
){
3999 ext4_ext_mark_uninitialized(&newex
);
4001 * io_end structure was created for every IO write to an
4002 * uninitialized extent. To avoid unnecessary conversion,
4003 * here we flag the IO that really needs the conversion.
4004 * For non asycn direct IO case, flag the inode state
4005 * that we need to perform conversion when IO is done.
4007 if ((flags
& EXT4_GET_BLOCKS_PRE_IO
)) {
4009 ext4_set_io_unwritten_flag(inode
, io
);
4011 ext4_set_inode_state(inode
,
4012 EXT4_STATE_DIO_UNWRITTEN
);
4014 if (ext4_should_dioread_nolock(inode
))
4015 map
->m_flags
|= EXT4_MAP_UNINIT
;
4019 if ((flags
& EXT4_GET_BLOCKS_KEEP_SIZE
) == 0)
4020 err
= check_eofblocks_fl(handle
, inode
, map
->m_lblk
,
4023 err
= ext4_ext_insert_extent(handle
, inode
, path
,
4025 if (err
&& free_on_err
) {
4026 int fb_flags
= flags
& EXT4_GET_BLOCKS_DELALLOC_RESERVE
?
4027 EXT4_FREE_BLOCKS_NO_QUOT_UPDATE
: 0;
4028 /* free data blocks we just allocated */
4029 /* not a good idea to call discard here directly,
4030 * but otherwise we'd need to call it every free() */
4031 ext4_discard_preallocations(inode
);
4032 ext4_free_blocks(handle
, inode
, NULL
, ext4_ext_pblock(&newex
),
4033 ext4_ext_get_actual_len(&newex
), fb_flags
);
4037 /* previous routine could use block we allocated */
4038 newblock
= ext4_ext_pblock(&newex
);
4039 allocated
= ext4_ext_get_actual_len(&newex
);
4040 if (allocated
> map
->m_len
)
4041 allocated
= map
->m_len
;
4042 map
->m_flags
|= EXT4_MAP_NEW
;
4045 * Update reserved blocks/metadata blocks after successful
4046 * block allocation which had been deferred till now.
4048 if (flags
& EXT4_GET_BLOCKS_DELALLOC_RESERVE
) {
4049 unsigned int reserved_clusters
;
4051 * Check how many clusters we had reserved this allocated range
4053 reserved_clusters
= get_reserved_cluster_alloc(inode
,
4054 map
->m_lblk
, allocated
);
4055 if (map
->m_flags
& EXT4_MAP_FROM_CLUSTER
) {
4056 if (reserved_clusters
) {
4058 * We have clusters reserved for this range.
4059 * But since we are not doing actual allocation
4060 * and are simply using blocks from previously
4061 * allocated cluster, we should release the
4062 * reservation and not claim quota.
4064 ext4_da_update_reserve_space(inode
,
4065 reserved_clusters
, 0);
4068 BUG_ON(allocated_clusters
< reserved_clusters
);
4069 /* We will claim quota for all newly allocated blocks.*/
4070 ext4_da_update_reserve_space(inode
, allocated_clusters
,
4072 if (reserved_clusters
< allocated_clusters
) {
4073 struct ext4_inode_info
*ei
= EXT4_I(inode
);
4074 int reservation
= allocated_clusters
-
4077 * It seems we claimed few clusters outside of
4078 * the range of this allocation. We should give
4079 * it back to the reservation pool. This can
4080 * happen in the following case:
4082 * * Suppose s_cluster_ratio is 4 (i.e., each
4083 * cluster has 4 blocks. Thus, the clusters
4084 * are [0-3],[4-7],[8-11]...
4085 * * First comes delayed allocation write for
4086 * logical blocks 10 & 11. Since there were no
4087 * previous delayed allocated blocks in the
4088 * range [8-11], we would reserve 1 cluster
4090 * * Next comes write for logical blocks 3 to 8.
4091 * In this case, we will reserve 2 clusters
4092 * (for [0-3] and [4-7]; and not for [8-11] as
4093 * that range has a delayed allocated blocks.
4094 * Thus total reserved clusters now becomes 3.
4095 * * Now, during the delayed allocation writeout
4096 * time, we will first write blocks [3-8] and
4097 * allocate 3 clusters for writing these
4098 * blocks. Also, we would claim all these
4099 * three clusters above.
4100 * * Now when we come here to writeout the
4101 * blocks [10-11], we would expect to claim
4102 * the reservation of 1 cluster we had made
4103 * (and we would claim it since there are no
4104 * more delayed allocated blocks in the range
4105 * [8-11]. But our reserved cluster count had
4106 * already gone to 0.
4108 * Thus, at the step 4 above when we determine
4109 * that there are still some unwritten delayed
4110 * allocated blocks outside of our current
4111 * block range, we should increment the
4112 * reserved clusters count so that when the
4113 * remaining blocks finally gets written, we
4116 dquot_reserve_block(inode
,
4117 EXT4_C2B(sbi
, reservation
));
4118 spin_lock(&ei
->i_block_reservation_lock
);
4119 ei
->i_reserved_data_blocks
+= reservation
;
4120 spin_unlock(&ei
->i_block_reservation_lock
);
4126 * Cache the extent and update transaction to commit on fdatasync only
4127 * when it is _not_ an uninitialized extent.
4129 if ((flags
& EXT4_GET_BLOCKS_UNINIT_EXT
) == 0) {
4130 ext4_ext_put_in_cache(inode
, map
->m_lblk
, allocated
, newblock
);
4131 ext4_update_inode_fsync_trans(handle
, inode
, 1);
4133 ext4_update_inode_fsync_trans(handle
, inode
, 0);
4135 if (allocated
> map
->m_len
)
4136 allocated
= map
->m_len
;
4137 ext4_ext_show_leaf(inode
, path
);
4138 map
->m_flags
|= EXT4_MAP_MAPPED
;
4139 map
->m_pblk
= newblock
;
4140 map
->m_len
= allocated
;
4143 ext4_ext_drop_refs(path
);
4147 trace_ext4_ext_map_blocks_exit(inode
, map
->m_lblk
,
4148 newblock
, map
->m_len
, err
? err
: allocated
);
4150 return err
? err
: allocated
;
4153 void ext4_ext_truncate(struct inode
*inode
)
4155 struct address_space
*mapping
= inode
->i_mapping
;
4156 struct super_block
*sb
= inode
->i_sb
;
4157 ext4_lblk_t last_block
;
4163 * finish any pending end_io work so we won't run the risk of
4164 * converting any truncated blocks to initialized later
4166 ext4_flush_completed_IO(inode
);
4169 * probably first extent we're gonna free will be last in block
4171 err
= ext4_writepage_trans_blocks(inode
);
4172 handle
= ext4_journal_start(inode
, err
);
4176 if (inode
->i_size
% PAGE_CACHE_SIZE
!= 0) {
4177 page_len
= PAGE_CACHE_SIZE
-
4178 (inode
->i_size
& (PAGE_CACHE_SIZE
- 1));
4180 err
= ext4_discard_partial_page_buffers(handle
,
4181 mapping
, inode
->i_size
, page_len
, 0);
4187 if (ext4_orphan_add(handle
, inode
))
4190 down_write(&EXT4_I(inode
)->i_data_sem
);
4191 ext4_ext_invalidate_cache(inode
);
4193 ext4_discard_preallocations(inode
);
4196 * TODO: optimization is possible here.
4197 * Probably we need not scan at all,
4198 * because page truncation is enough.
4201 /* we have to know where to truncate from in crash case */
4202 EXT4_I(inode
)->i_disksize
= inode
->i_size
;
4203 ext4_mark_inode_dirty(handle
, inode
);
4205 last_block
= (inode
->i_size
+ sb
->s_blocksize
- 1)
4206 >> EXT4_BLOCK_SIZE_BITS(sb
);
4207 err
= ext4_ext_remove_space(inode
, last_block
, EXT_MAX_BLOCKS
- 1);
4209 /* In a multi-transaction truncate, we only make the final
4210 * transaction synchronous.
4213 ext4_handle_sync(handle
);
4215 up_write(&EXT4_I(inode
)->i_data_sem
);
4219 * If this was a simple ftruncate() and the file will remain alive,
4220 * then we need to clear up the orphan record which we created above.
4221 * However, if this was a real unlink then we were called by
4222 * ext4_delete_inode(), and we allow that function to clean up the
4223 * orphan info for us.
4226 ext4_orphan_del(handle
, inode
);
4228 inode
->i_mtime
= inode
->i_ctime
= ext4_current_time(inode
);
4229 ext4_mark_inode_dirty(handle
, inode
);
4230 ext4_journal_stop(handle
);
4233 static void ext4_falloc_update_inode(struct inode
*inode
,
4234 int mode
, loff_t new_size
, int update_ctime
)
4236 struct timespec now
;
4239 now
= current_fs_time(inode
->i_sb
);
4240 if (!timespec_equal(&inode
->i_ctime
, &now
))
4241 inode
->i_ctime
= now
;
4244 * Update only when preallocation was requested beyond
4247 if (!(mode
& FALLOC_FL_KEEP_SIZE
)) {
4248 if (new_size
> i_size_read(inode
))
4249 i_size_write(inode
, new_size
);
4250 if (new_size
> EXT4_I(inode
)->i_disksize
)
4251 ext4_update_i_disksize(inode
, new_size
);
4254 * Mark that we allocate beyond EOF so the subsequent truncate
4255 * can proceed even if the new size is the same as i_size.
4257 if (new_size
> i_size_read(inode
))
4258 ext4_set_inode_flag(inode
, EXT4_INODE_EOFBLOCKS
);
4264 * preallocate space for a file. This implements ext4's fallocate file
4265 * operation, which gets called from sys_fallocate system call.
4266 * For block-mapped files, posix_fallocate should fall back to the method
4267 * of writing zeroes to the required new blocks (the same behavior which is
4268 * expected for file systems which do not support fallocate() system call).
4270 long ext4_fallocate(struct file
*file
, int mode
, loff_t offset
, loff_t len
)
4272 struct inode
*inode
= file
->f_path
.dentry
->d_inode
;
4275 unsigned int max_blocks
;
4280 struct ext4_map_blocks map
;
4281 unsigned int credits
, blkbits
= inode
->i_blkbits
;
4284 * currently supporting (pre)allocate mode for extent-based
4287 if (!(ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
)))
4290 /* Return error if mode is not supported */
4291 if (mode
& ~(FALLOC_FL_KEEP_SIZE
| FALLOC_FL_PUNCH_HOLE
))
4294 if (mode
& FALLOC_FL_PUNCH_HOLE
)
4295 return ext4_punch_hole(file
, offset
, len
);
4297 trace_ext4_fallocate_enter(inode
, offset
, len
, mode
);
4298 map
.m_lblk
= offset
>> blkbits
;
4300 * We can't just convert len to max_blocks because
4301 * If blocksize = 4096 offset = 3072 and len = 2048
4303 max_blocks
= (EXT4_BLOCK_ALIGN(len
+ offset
, blkbits
) >> blkbits
)
4306 * credits to insert 1 extent into extent tree
4308 credits
= ext4_chunk_trans_blocks(inode
, max_blocks
);
4309 mutex_lock(&inode
->i_mutex
);
4310 ret
= inode_newsize_ok(inode
, (len
+ offset
));
4312 mutex_unlock(&inode
->i_mutex
);
4313 trace_ext4_fallocate_exit(inode
, offset
, max_blocks
, ret
);
4316 flags
= EXT4_GET_BLOCKS_CREATE_UNINIT_EXT
;
4317 if (mode
& FALLOC_FL_KEEP_SIZE
)
4318 flags
|= EXT4_GET_BLOCKS_KEEP_SIZE
;
4320 * Don't normalize the request if it can fit in one extent so
4321 * that it doesn't get unnecessarily split into multiple
4324 if (len
<= EXT_UNINIT_MAX_LEN
<< blkbits
)
4325 flags
|= EXT4_GET_BLOCKS_NO_NORMALIZE
;
4327 while (ret
>= 0 && ret
< max_blocks
) {
4328 map
.m_lblk
= map
.m_lblk
+ ret
;
4329 map
.m_len
= max_blocks
= max_blocks
- ret
;
4330 handle
= ext4_journal_start(inode
, credits
);
4331 if (IS_ERR(handle
)) {
4332 ret
= PTR_ERR(handle
);
4335 ret
= ext4_map_blocks(handle
, inode
, &map
, flags
);
4339 printk(KERN_ERR
"%s: ext4_ext_map_blocks "
4340 "returned error inode#%lu, block=%u, "
4341 "max_blocks=%u", __func__
,
4342 inode
->i_ino
, map
.m_lblk
, max_blocks
);
4344 ext4_mark_inode_dirty(handle
, inode
);
4345 ret2
= ext4_journal_stop(handle
);
4348 if ((map
.m_lblk
+ ret
) >= (EXT4_BLOCK_ALIGN(offset
+ len
,
4349 blkbits
) >> blkbits
))
4350 new_size
= offset
+ len
;
4352 new_size
= ((loff_t
) map
.m_lblk
+ ret
) << blkbits
;
4354 ext4_falloc_update_inode(inode
, mode
, new_size
,
4355 (map
.m_flags
& EXT4_MAP_NEW
));
4356 ext4_mark_inode_dirty(handle
, inode
);
4357 ret2
= ext4_journal_stop(handle
);
4361 if (ret
== -ENOSPC
&&
4362 ext4_should_retry_alloc(inode
->i_sb
, &retries
)) {
4366 mutex_unlock(&inode
->i_mutex
);
4367 trace_ext4_fallocate_exit(inode
, offset
, max_blocks
,
4368 ret
> 0 ? ret2
: ret
);
4369 return ret
> 0 ? ret2
: ret
;
4373 * This function convert a range of blocks to written extents
4374 * The caller of this function will pass the start offset and the size.
4375 * all unwritten extents within this range will be converted to
4378 * This function is called from the direct IO end io call back
4379 * function, to convert the fallocated extents after IO is completed.
4380 * Returns 0 on success.
4382 int ext4_convert_unwritten_extents(struct inode
*inode
, loff_t offset
,
4386 unsigned int max_blocks
;
4389 struct ext4_map_blocks map
;
4390 unsigned int credits
, blkbits
= inode
->i_blkbits
;
4392 map
.m_lblk
= offset
>> blkbits
;
4394 * We can't just convert len to max_blocks because
4395 * If blocksize = 4096 offset = 3072 and len = 2048
4397 max_blocks
= ((EXT4_BLOCK_ALIGN(len
+ offset
, blkbits
) >> blkbits
) -
4400 * credits to insert 1 extent into extent tree
4402 credits
= ext4_chunk_trans_blocks(inode
, max_blocks
);
4403 while (ret
>= 0 && ret
< max_blocks
) {
4405 map
.m_len
= (max_blocks
-= ret
);
4406 handle
= ext4_journal_start(inode
, credits
);
4407 if (IS_ERR(handle
)) {
4408 ret
= PTR_ERR(handle
);
4411 ret
= ext4_map_blocks(handle
, inode
, &map
,
4412 EXT4_GET_BLOCKS_IO_CONVERT_EXT
);
4415 printk(KERN_ERR
"%s: ext4_ext_map_blocks "
4416 "returned error inode#%lu, block=%u, "
4417 "max_blocks=%u", __func__
,
4418 inode
->i_ino
, map
.m_lblk
, map
.m_len
);
4420 ext4_mark_inode_dirty(handle
, inode
);
4421 ret2
= ext4_journal_stop(handle
);
4422 if (ret
<= 0 || ret2
)
4425 return ret
> 0 ? ret2
: ret
;
4429 * Callback function called for each extent to gather FIEMAP information.
4431 static int ext4_ext_fiemap_cb(struct inode
*inode
, ext4_lblk_t next
,
4432 struct ext4_ext_cache
*newex
, struct ext4_extent
*ex
,
4440 struct fiemap_extent_info
*fieinfo
= data
;
4441 unsigned char blksize_bits
;
4443 blksize_bits
= inode
->i_sb
->s_blocksize_bits
;
4444 logical
= (__u64
)newex
->ec_block
<< blksize_bits
;
4446 if (newex
->ec_start
== 0) {
4448 * No extent in extent-tree contains block @newex->ec_start,
4449 * then the block may stay in 1)a hole or 2)delayed-extent.
4451 * Holes or delayed-extents are processed as follows.
4452 * 1. lookup dirty pages with specified range in pagecache.
4453 * If no page is got, then there is no delayed-extent and
4454 * return with EXT_CONTINUE.
4455 * 2. find the 1st mapped buffer,
4456 * 3. check if the mapped buffer is both in the request range
4457 * and a delayed buffer. If not, there is no delayed-extent,
4459 * 4. a delayed-extent is found, the extent will be collected.
4461 ext4_lblk_t end
= 0;
4462 pgoff_t last_offset
;
4465 pgoff_t start_index
= 0;
4466 struct page
**pages
= NULL
;
4467 struct buffer_head
*bh
= NULL
;
4468 struct buffer_head
*head
= NULL
;
4469 unsigned int nr_pages
= PAGE_SIZE
/ sizeof(struct page
*);
4471 pages
= kmalloc(PAGE_SIZE
, GFP_KERNEL
);
4475 offset
= logical
>> PAGE_SHIFT
;
4477 last_offset
= offset
;
4479 ret
= find_get_pages_tag(inode
->i_mapping
, &offset
,
4480 PAGECACHE_TAG_DIRTY
, nr_pages
, pages
);
4482 if (!(flags
& FIEMAP_EXTENT_DELALLOC
)) {
4483 /* First time, try to find a mapped buffer. */
4486 for (index
= 0; index
< ret
; index
++)
4487 page_cache_release(pages
[index
]);
4490 return EXT_CONTINUE
;
4495 /* Try to find the 1st mapped buffer. */
4496 end
= ((__u64
)pages
[index
]->index
<< PAGE_SHIFT
) >>
4498 if (!page_has_buffers(pages
[index
]))
4500 head
= page_buffers(pages
[index
]);
4507 if (end
>= newex
->ec_block
+
4509 /* The buffer is out of
4510 * the request range.
4514 if (buffer_mapped(bh
) &&
4515 end
>= newex
->ec_block
) {
4516 start_index
= index
- 1;
4517 /* get the 1st mapped buffer. */
4518 goto found_mapped_buffer
;
4521 bh
= bh
->b_this_page
;
4523 } while (bh
!= head
);
4525 /* No mapped buffer in the range found in this page,
4526 * We need to look up next page.
4529 /* There is no page left, but we need to limit
4532 newex
->ec_len
= end
- newex
->ec_block
;
4537 /*Find contiguous delayed buffers. */
4538 if (ret
> 0 && pages
[0]->index
== last_offset
)
4539 head
= page_buffers(pages
[0]);
4545 found_mapped_buffer
:
4546 if (bh
!= NULL
&& buffer_delay(bh
)) {
4547 /* 1st or contiguous delayed buffer found. */
4548 if (!(flags
& FIEMAP_EXTENT_DELALLOC
)) {
4550 * 1st delayed buffer found, record
4551 * the start of extent.
4553 flags
|= FIEMAP_EXTENT_DELALLOC
;
4554 newex
->ec_block
= end
;
4555 logical
= (__u64
)end
<< blksize_bits
;
4557 /* Find contiguous delayed buffers. */
4559 if (!buffer_delay(bh
))
4560 goto found_delayed_extent
;
4561 bh
= bh
->b_this_page
;
4563 } while (bh
!= head
);
4565 for (; index
< ret
; index
++) {
4566 if (!page_has_buffers(pages
[index
])) {
4570 head
= page_buffers(pages
[index
]);
4576 if (pages
[index
]->index
!=
4577 pages
[start_index
]->index
+ index
4579 /* Blocks are not contiguous. */
4585 if (!buffer_delay(bh
))
4586 /* Delayed-extent ends. */
4587 goto found_delayed_extent
;
4588 bh
= bh
->b_this_page
;
4590 } while (bh
!= head
);
4592 } else if (!(flags
& FIEMAP_EXTENT_DELALLOC
))
4596 found_delayed_extent
:
4597 newex
->ec_len
= min(end
- newex
->ec_block
,
4598 (ext4_lblk_t
)EXT_INIT_MAX_LEN
);
4599 if (ret
== nr_pages
&& bh
!= NULL
&&
4600 newex
->ec_len
< EXT_INIT_MAX_LEN
&&
4602 /* Have not collected an extent and continue. */
4603 for (index
= 0; index
< ret
; index
++)
4604 page_cache_release(pages
[index
]);
4608 for (index
= 0; index
< ret
; index
++)
4609 page_cache_release(pages
[index
]);
4613 physical
= (__u64
)newex
->ec_start
<< blksize_bits
;
4614 length
= (__u64
)newex
->ec_len
<< blksize_bits
;
4616 if (ex
&& ext4_ext_is_uninitialized(ex
))
4617 flags
|= FIEMAP_EXTENT_UNWRITTEN
;
4619 if (next
== EXT_MAX_BLOCKS
)
4620 flags
|= FIEMAP_EXTENT_LAST
;
4622 ret
= fiemap_fill_next_extent(fieinfo
, logical
, physical
,
4628 return EXT_CONTINUE
;
4630 /* fiemap flags we can handle specified here */
4631 #define EXT4_FIEMAP_FLAGS (FIEMAP_FLAG_SYNC|FIEMAP_FLAG_XATTR)
4633 static int ext4_xattr_fiemap(struct inode
*inode
,
4634 struct fiemap_extent_info
*fieinfo
)
4638 __u32 flags
= FIEMAP_EXTENT_LAST
;
4639 int blockbits
= inode
->i_sb
->s_blocksize_bits
;
4643 if (ext4_test_inode_state(inode
, EXT4_STATE_XATTR
)) {
4644 struct ext4_iloc iloc
;
4645 int offset
; /* offset of xattr in inode */
4647 error
= ext4_get_inode_loc(inode
, &iloc
);
4650 physical
= iloc
.bh
->b_blocknr
<< blockbits
;
4651 offset
= EXT4_GOOD_OLD_INODE_SIZE
+
4652 EXT4_I(inode
)->i_extra_isize
;
4654 length
= EXT4_SB(inode
->i_sb
)->s_inode_size
- offset
;
4655 flags
|= FIEMAP_EXTENT_DATA_INLINE
;
4657 } else { /* external block */
4658 physical
= EXT4_I(inode
)->i_file_acl
<< blockbits
;
4659 length
= inode
->i_sb
->s_blocksize
;
4663 error
= fiemap_fill_next_extent(fieinfo
, 0, physical
,
4665 return (error
< 0 ? error
: 0);
4669 * ext4_ext_punch_hole
4671 * Punches a hole of "length" bytes in a file starting
4674 * @inode: The inode of the file to punch a hole in
4675 * @offset: The starting byte offset of the hole
4676 * @length: The length of the hole
4678 * Returns the number of blocks removed or negative on err
4680 int ext4_ext_punch_hole(struct file
*file
, loff_t offset
, loff_t length
)
4682 struct inode
*inode
= file
->f_path
.dentry
->d_inode
;
4683 struct super_block
*sb
= inode
->i_sb
;
4684 ext4_lblk_t first_block
, stop_block
;
4685 struct address_space
*mapping
= inode
->i_mapping
;
4687 loff_t first_page
, last_page
, page_len
;
4688 loff_t first_page_offset
, last_page_offset
;
4689 int credits
, err
= 0;
4691 /* No need to punch hole beyond i_size */
4692 if (offset
>= inode
->i_size
)
4696 * If the hole extends beyond i_size, set the hole
4697 * to end after the page that contains i_size
4699 if (offset
+ length
> inode
->i_size
) {
4700 length
= inode
->i_size
+
4701 PAGE_CACHE_SIZE
- (inode
->i_size
& (PAGE_CACHE_SIZE
- 1)) -
4705 first_page
= (offset
+ PAGE_CACHE_SIZE
- 1) >> PAGE_CACHE_SHIFT
;
4706 last_page
= (offset
+ length
) >> PAGE_CACHE_SHIFT
;
4708 first_page_offset
= first_page
<< PAGE_CACHE_SHIFT
;
4709 last_page_offset
= last_page
<< PAGE_CACHE_SHIFT
;
4712 * Write out all dirty pages to avoid race conditions
4713 * Then release them.
4715 if (mapping
->nrpages
&& mapping_tagged(mapping
, PAGECACHE_TAG_DIRTY
)) {
4716 err
= filemap_write_and_wait_range(mapping
,
4717 offset
, offset
+ length
- 1);
4723 /* Now release the pages */
4724 if (last_page_offset
> first_page_offset
) {
4725 truncate_inode_pages_range(mapping
, first_page_offset
,
4726 last_page_offset
-1);
4729 /* finish any pending end_io work */
4730 ext4_flush_completed_IO(inode
);
4732 credits
= ext4_writepage_trans_blocks(inode
);
4733 handle
= ext4_journal_start(inode
, credits
);
4735 return PTR_ERR(handle
);
4737 err
= ext4_orphan_add(handle
, inode
);
4742 * Now we need to zero out the non-page-aligned data in the
4743 * pages at the start and tail of the hole, and unmap the buffer
4744 * heads for the block aligned regions of the page that were
4745 * completely zeroed.
4747 if (first_page
> last_page
) {
4749 * If the file space being truncated is contained within a page
4750 * just zero out and unmap the middle of that page
4752 err
= ext4_discard_partial_page_buffers(handle
,
4753 mapping
, offset
, length
, 0);
4759 * zero out and unmap the partial page that contains
4760 * the start of the hole
4762 page_len
= first_page_offset
- offset
;
4764 err
= ext4_discard_partial_page_buffers(handle
, mapping
,
4765 offset
, page_len
, 0);
4771 * zero out and unmap the partial page that contains
4772 * the end of the hole
4774 page_len
= offset
+ length
- last_page_offset
;
4776 err
= ext4_discard_partial_page_buffers(handle
, mapping
,
4777 last_page_offset
, page_len
, 0);
4784 * If i_size is contained in the last page, we need to
4785 * unmap and zero the partial page after i_size
4787 if (inode
->i_size
>> PAGE_CACHE_SHIFT
== last_page
&&
4788 inode
->i_size
% PAGE_CACHE_SIZE
!= 0) {
4790 page_len
= PAGE_CACHE_SIZE
-
4791 (inode
->i_size
& (PAGE_CACHE_SIZE
- 1));
4794 err
= ext4_discard_partial_page_buffers(handle
,
4795 mapping
, inode
->i_size
, page_len
, 0);
4802 first_block
= (offset
+ sb
->s_blocksize
- 1) >>
4803 EXT4_BLOCK_SIZE_BITS(sb
);
4804 stop_block
= (offset
+ length
) >> EXT4_BLOCK_SIZE_BITS(sb
);
4806 /* If there are no blocks to remove, return now */
4807 if (first_block
>= stop_block
)
4810 down_write(&EXT4_I(inode
)->i_data_sem
);
4811 ext4_ext_invalidate_cache(inode
);
4812 ext4_discard_preallocations(inode
);
4814 err
= ext4_ext_remove_space(inode
, first_block
, stop_block
- 1);
4816 ext4_ext_invalidate_cache(inode
);
4817 ext4_discard_preallocations(inode
);
4820 ext4_handle_sync(handle
);
4822 up_write(&EXT4_I(inode
)->i_data_sem
);
4825 ext4_orphan_del(handle
, inode
);
4826 inode
->i_mtime
= inode
->i_ctime
= ext4_current_time(inode
);
4827 ext4_mark_inode_dirty(handle
, inode
);
4828 ext4_journal_stop(handle
);
4831 int ext4_fiemap(struct inode
*inode
, struct fiemap_extent_info
*fieinfo
,
4832 __u64 start
, __u64 len
)
4834 ext4_lblk_t start_blk
;
4837 /* fallback to generic here if not in extents fmt */
4838 if (!(ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
)))
4839 return generic_block_fiemap(inode
, fieinfo
, start
, len
,
4842 if (fiemap_check_flags(fieinfo
, EXT4_FIEMAP_FLAGS
))
4845 if (fieinfo
->fi_flags
& FIEMAP_FLAG_XATTR
) {
4846 error
= ext4_xattr_fiemap(inode
, fieinfo
);
4848 ext4_lblk_t len_blks
;
4851 start_blk
= start
>> inode
->i_sb
->s_blocksize_bits
;
4852 last_blk
= (start
+ len
- 1) >> inode
->i_sb
->s_blocksize_bits
;
4853 if (last_blk
>= EXT_MAX_BLOCKS
)
4854 last_blk
= EXT_MAX_BLOCKS
-1;
4855 len_blks
= ((ext4_lblk_t
) last_blk
) - start_blk
+ 1;
4858 * Walk the extent tree gathering extent information.
4859 * ext4_ext_fiemap_cb will push extents back to user.
4861 error
= ext4_ext_walk_space(inode
, start_blk
, len_blks
,
4862 ext4_ext_fiemap_cb
, fieinfo
);