2 * Copyright (c) 2003-2006, Cluster File Systems, Inc, info@clusterfs.com
3 * Written by Alex Tomas <alex@clusterfs.com>
5 * Architecture independence:
6 * Copyright (c) 2005, Bull S.A.
7 * Written by Pierre Peiffer <pierre.peiffer@bull.net>
9 * This program is free software; you can redistribute it and/or modify
10 * it under the terms of the GNU General Public License version 2 as
11 * published by the Free Software Foundation.
13 * This program is distributed in the hope that it will be useful,
14 * but WITHOUT ANY WARRANTY; without even the implied warranty of
15 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
16 * GNU General Public License for more details.
18 * You should have received a copy of the GNU General Public Licens
19 * along with this program; if not, write to the Free Software
20 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-
24 * Extents support for EXT4
27 * - ext4*_error() should be used in some situations
28 * - analyze all BUG()/BUG_ON(), use -EIO where appropriate
29 * - smart tree reduction
32 #include <linux/module.h>
34 #include <linux/time.h>
35 #include <linux/jbd2.h>
36 #include <linux/highuid.h>
37 #include <linux/pagemap.h>
38 #include <linux/quotaops.h>
39 #include <linux/string.h>
40 #include <linux/slab.h>
41 #include <linux/falloc.h>
42 #include <asm/uaccess.h>
43 #include <linux/fiemap.h>
44 #include "ext4_jbd2.h"
46 #include <trace/events/ext4.h>
48 static int ext4_split_extent(handle_t
*handle
,
50 struct ext4_ext_path
*path
,
51 struct ext4_map_blocks
*map
,
55 static int ext4_ext_truncate_extend_restart(handle_t
*handle
,
61 if (!ext4_handle_valid(handle
))
63 if (handle
->h_buffer_credits
> needed
)
65 err
= ext4_journal_extend(handle
, needed
);
68 err
= ext4_truncate_restart_trans(handle
, inode
, needed
);
80 static int ext4_ext_get_access(handle_t
*handle
, struct inode
*inode
,
81 struct ext4_ext_path
*path
)
84 /* path points to block */
85 return ext4_journal_get_write_access(handle
, path
->p_bh
);
87 /* path points to leaf/index in inode body */
88 /* we use in-core data, no need to protect them */
98 #define ext4_ext_dirty(handle, inode, path) \
99 __ext4_ext_dirty(__func__, __LINE__, (handle), (inode), (path))
100 static int __ext4_ext_dirty(const char *where
, unsigned int line
,
101 handle_t
*handle
, struct inode
*inode
,
102 struct ext4_ext_path
*path
)
106 /* path points to block */
107 err
= __ext4_handle_dirty_metadata(where
, line
, handle
,
110 /* path points to leaf/index in inode body */
111 err
= ext4_mark_inode_dirty(handle
, inode
);
116 static ext4_fsblk_t
ext4_ext_find_goal(struct inode
*inode
,
117 struct ext4_ext_path
*path
,
123 struct ext4_extent
*ex
;
124 depth
= path
->p_depth
;
127 * Try to predict block placement assuming that we are
128 * filling in a file which will eventually be
129 * non-sparse --- i.e., in the case of libbfd writing
130 * an ELF object sections out-of-order but in a way
131 * the eventually results in a contiguous object or
132 * executable file, or some database extending a table
133 * space file. However, this is actually somewhat
134 * non-ideal if we are writing a sparse file such as
135 * qemu or KVM writing a raw image file that is going
136 * to stay fairly sparse, since it will end up
137 * fragmenting the file system's free space. Maybe we
138 * should have some hueristics or some way to allow
139 * userspace to pass a hint to file system,
140 * especially if the latter case turns out to be
143 ex
= path
[depth
].p_ext
;
145 ext4_fsblk_t ext_pblk
= ext4_ext_pblock(ex
);
146 ext4_lblk_t ext_block
= le32_to_cpu(ex
->ee_block
);
148 if (block
> ext_block
)
149 return ext_pblk
+ (block
- ext_block
);
151 return ext_pblk
- (ext_block
- block
);
154 /* it looks like index is empty;
155 * try to find starting block from index itself */
156 if (path
[depth
].p_bh
)
157 return path
[depth
].p_bh
->b_blocknr
;
160 /* OK. use inode's group */
161 return ext4_inode_to_goal_block(inode
);
165 * Allocation for a meta data block
168 ext4_ext_new_meta_block(handle_t
*handle
, struct inode
*inode
,
169 struct ext4_ext_path
*path
,
170 struct ext4_extent
*ex
, int *err
, unsigned int flags
)
172 ext4_fsblk_t goal
, newblock
;
174 goal
= ext4_ext_find_goal(inode
, path
, le32_to_cpu(ex
->ee_block
));
175 newblock
= ext4_new_meta_blocks(handle
, inode
, goal
, flags
,
180 static inline int ext4_ext_space_block(struct inode
*inode
, int check
)
184 size
= (inode
->i_sb
->s_blocksize
- sizeof(struct ext4_extent_header
))
185 / sizeof(struct ext4_extent
);
187 #ifdef AGGRESSIVE_TEST
195 static inline int ext4_ext_space_block_idx(struct inode
*inode
, int check
)
199 size
= (inode
->i_sb
->s_blocksize
- sizeof(struct ext4_extent_header
))
200 / sizeof(struct ext4_extent_idx
);
202 #ifdef AGGRESSIVE_TEST
210 static inline int ext4_ext_space_root(struct inode
*inode
, int check
)
214 size
= sizeof(EXT4_I(inode
)->i_data
);
215 size
-= sizeof(struct ext4_extent_header
);
216 size
/= sizeof(struct ext4_extent
);
218 #ifdef AGGRESSIVE_TEST
226 static inline int ext4_ext_space_root_idx(struct inode
*inode
, int check
)
230 size
= sizeof(EXT4_I(inode
)->i_data
);
231 size
-= sizeof(struct ext4_extent_header
);
232 size
/= sizeof(struct ext4_extent_idx
);
234 #ifdef AGGRESSIVE_TEST
243 * Calculate the number of metadata blocks needed
244 * to allocate @blocks
245 * Worse case is one block per extent
247 int ext4_ext_calc_metadata_amount(struct inode
*inode
, ext4_lblk_t lblock
)
249 struct ext4_inode_info
*ei
= EXT4_I(inode
);
252 idxs
= ((inode
->i_sb
->s_blocksize
- sizeof(struct ext4_extent_header
))
253 / sizeof(struct ext4_extent_idx
));
256 * If the new delayed allocation block is contiguous with the
257 * previous da block, it can share index blocks with the
258 * previous block, so we only need to allocate a new index
259 * block every idxs leaf blocks. At ldxs**2 blocks, we need
260 * an additional index block, and at ldxs**3 blocks, yet
261 * another index blocks.
263 if (ei
->i_da_metadata_calc_len
&&
264 ei
->i_da_metadata_calc_last_lblock
+1 == lblock
) {
265 if ((ei
->i_da_metadata_calc_len
% idxs
) == 0)
267 if ((ei
->i_da_metadata_calc_len
% (idxs
*idxs
)) == 0)
269 if ((ei
->i_da_metadata_calc_len
% (idxs
*idxs
*idxs
)) == 0) {
271 ei
->i_da_metadata_calc_len
= 0;
273 ei
->i_da_metadata_calc_len
++;
274 ei
->i_da_metadata_calc_last_lblock
++;
279 * In the worst case we need a new set of index blocks at
280 * every level of the inode's extent tree.
282 ei
->i_da_metadata_calc_len
= 1;
283 ei
->i_da_metadata_calc_last_lblock
= lblock
;
284 return ext_depth(inode
) + 1;
288 ext4_ext_max_entries(struct inode
*inode
, int depth
)
292 if (depth
== ext_depth(inode
)) {
294 max
= ext4_ext_space_root(inode
, 1);
296 max
= ext4_ext_space_root_idx(inode
, 1);
299 max
= ext4_ext_space_block(inode
, 1);
301 max
= ext4_ext_space_block_idx(inode
, 1);
307 static int ext4_valid_extent(struct inode
*inode
, struct ext4_extent
*ext
)
309 ext4_fsblk_t block
= ext4_ext_pblock(ext
);
310 int len
= ext4_ext_get_actual_len(ext
);
312 return ext4_data_block_valid(EXT4_SB(inode
->i_sb
), block
, len
);
315 static int ext4_valid_extent_idx(struct inode
*inode
,
316 struct ext4_extent_idx
*ext_idx
)
318 ext4_fsblk_t block
= ext4_idx_pblock(ext_idx
);
320 return ext4_data_block_valid(EXT4_SB(inode
->i_sb
), block
, 1);
323 static int ext4_valid_extent_entries(struct inode
*inode
,
324 struct ext4_extent_header
*eh
,
327 struct ext4_extent
*ext
;
328 struct ext4_extent_idx
*ext_idx
;
329 unsigned short entries
;
330 if (eh
->eh_entries
== 0)
333 entries
= le16_to_cpu(eh
->eh_entries
);
337 ext
= EXT_FIRST_EXTENT(eh
);
339 if (!ext4_valid_extent(inode
, ext
))
345 ext_idx
= EXT_FIRST_INDEX(eh
);
347 if (!ext4_valid_extent_idx(inode
, ext_idx
))
356 static int __ext4_ext_check(const char *function
, unsigned int line
,
357 struct inode
*inode
, struct ext4_extent_header
*eh
,
360 const char *error_msg
;
363 if (unlikely(eh
->eh_magic
!= EXT4_EXT_MAGIC
)) {
364 error_msg
= "invalid magic";
367 if (unlikely(le16_to_cpu(eh
->eh_depth
) != depth
)) {
368 error_msg
= "unexpected eh_depth";
371 if (unlikely(eh
->eh_max
== 0)) {
372 error_msg
= "invalid eh_max";
375 max
= ext4_ext_max_entries(inode
, depth
);
376 if (unlikely(le16_to_cpu(eh
->eh_max
) > max
)) {
377 error_msg
= "too large eh_max";
380 if (unlikely(le16_to_cpu(eh
->eh_entries
) > le16_to_cpu(eh
->eh_max
))) {
381 error_msg
= "invalid eh_entries";
384 if (!ext4_valid_extent_entries(inode
, eh
, depth
)) {
385 error_msg
= "invalid extent entries";
391 ext4_error_inode(inode
, function
, line
, 0,
392 "bad header/extent: %s - magic %x, "
393 "entries %u, max %u(%u), depth %u(%u)",
394 error_msg
, le16_to_cpu(eh
->eh_magic
),
395 le16_to_cpu(eh
->eh_entries
), le16_to_cpu(eh
->eh_max
),
396 max
, le16_to_cpu(eh
->eh_depth
), depth
);
401 #define ext4_ext_check(inode, eh, depth) \
402 __ext4_ext_check(__func__, __LINE__, inode, eh, depth)
404 int ext4_ext_check_inode(struct inode
*inode
)
406 return ext4_ext_check(inode
, ext_inode_hdr(inode
), ext_depth(inode
));
410 static void ext4_ext_show_path(struct inode
*inode
, struct ext4_ext_path
*path
)
412 int k
, l
= path
->p_depth
;
415 for (k
= 0; k
<= l
; k
++, path
++) {
417 ext_debug(" %d->%llu", le32_to_cpu(path
->p_idx
->ei_block
),
418 ext4_idx_pblock(path
->p_idx
));
419 } else if (path
->p_ext
) {
420 ext_debug(" %d:[%d]%d:%llu ",
421 le32_to_cpu(path
->p_ext
->ee_block
),
422 ext4_ext_is_uninitialized(path
->p_ext
),
423 ext4_ext_get_actual_len(path
->p_ext
),
424 ext4_ext_pblock(path
->p_ext
));
431 static void ext4_ext_show_leaf(struct inode
*inode
, struct ext4_ext_path
*path
)
433 int depth
= ext_depth(inode
);
434 struct ext4_extent_header
*eh
;
435 struct ext4_extent
*ex
;
441 eh
= path
[depth
].p_hdr
;
442 ex
= EXT_FIRST_EXTENT(eh
);
444 ext_debug("Displaying leaf extents for inode %lu\n", inode
->i_ino
);
446 for (i
= 0; i
< le16_to_cpu(eh
->eh_entries
); i
++, ex
++) {
447 ext_debug("%d:[%d]%d:%llu ", le32_to_cpu(ex
->ee_block
),
448 ext4_ext_is_uninitialized(ex
),
449 ext4_ext_get_actual_len(ex
), ext4_ext_pblock(ex
));
454 static void ext4_ext_show_move(struct inode
*inode
, struct ext4_ext_path
*path
,
455 ext4_fsblk_t newblock
, int level
)
457 int depth
= ext_depth(inode
);
458 struct ext4_extent
*ex
;
460 if (depth
!= level
) {
461 struct ext4_extent_idx
*idx
;
462 idx
= path
[level
].p_idx
;
463 while (idx
<= EXT_MAX_INDEX(path
[level
].p_hdr
)) {
464 ext_debug("%d: move %d:%llu in new index %llu\n", level
,
465 le32_to_cpu(idx
->ei_block
),
466 ext4_idx_pblock(idx
),
474 ex
= path
[depth
].p_ext
;
475 while (ex
<= EXT_MAX_EXTENT(path
[depth
].p_hdr
)) {
476 ext_debug("move %d:%llu:[%d]%d in new leaf %llu\n",
477 le32_to_cpu(ex
->ee_block
),
479 ext4_ext_is_uninitialized(ex
),
480 ext4_ext_get_actual_len(ex
),
487 #define ext4_ext_show_path(inode, path)
488 #define ext4_ext_show_leaf(inode, path)
489 #define ext4_ext_show_move(inode, path, newblock, level)
492 void ext4_ext_drop_refs(struct ext4_ext_path
*path
)
494 int depth
= path
->p_depth
;
497 for (i
= 0; i
<= depth
; i
++, path
++)
505 * ext4_ext_binsearch_idx:
506 * binary search for the closest index of the given block
507 * the header must be checked before calling this
510 ext4_ext_binsearch_idx(struct inode
*inode
,
511 struct ext4_ext_path
*path
, ext4_lblk_t block
)
513 struct ext4_extent_header
*eh
= path
->p_hdr
;
514 struct ext4_extent_idx
*r
, *l
, *m
;
517 ext_debug("binsearch for %u(idx): ", block
);
519 l
= EXT_FIRST_INDEX(eh
) + 1;
520 r
= EXT_LAST_INDEX(eh
);
523 if (block
< le32_to_cpu(m
->ei_block
))
527 ext_debug("%p(%u):%p(%u):%p(%u) ", l
, le32_to_cpu(l
->ei_block
),
528 m
, le32_to_cpu(m
->ei_block
),
529 r
, le32_to_cpu(r
->ei_block
));
533 ext_debug(" -> %d->%lld ", le32_to_cpu(path
->p_idx
->ei_block
),
534 ext4_idx_pblock(path
->p_idx
));
536 #ifdef CHECK_BINSEARCH
538 struct ext4_extent_idx
*chix
, *ix
;
541 chix
= ix
= EXT_FIRST_INDEX(eh
);
542 for (k
= 0; k
< le16_to_cpu(eh
->eh_entries
); k
++, ix
++) {
544 le32_to_cpu(ix
->ei_block
) <= le32_to_cpu(ix
[-1].ei_block
)) {
545 printk(KERN_DEBUG
"k=%d, ix=0x%p, "
547 ix
, EXT_FIRST_INDEX(eh
));
548 printk(KERN_DEBUG
"%u <= %u\n",
549 le32_to_cpu(ix
->ei_block
),
550 le32_to_cpu(ix
[-1].ei_block
));
552 BUG_ON(k
&& le32_to_cpu(ix
->ei_block
)
553 <= le32_to_cpu(ix
[-1].ei_block
));
554 if (block
< le32_to_cpu(ix
->ei_block
))
558 BUG_ON(chix
!= path
->p_idx
);
565 * ext4_ext_binsearch:
566 * binary search for closest extent of the given block
567 * the header must be checked before calling this
570 ext4_ext_binsearch(struct inode
*inode
,
571 struct ext4_ext_path
*path
, ext4_lblk_t block
)
573 struct ext4_extent_header
*eh
= path
->p_hdr
;
574 struct ext4_extent
*r
, *l
, *m
;
576 if (eh
->eh_entries
== 0) {
578 * this leaf is empty:
579 * we get such a leaf in split/add case
584 ext_debug("binsearch for %u: ", block
);
586 l
= EXT_FIRST_EXTENT(eh
) + 1;
587 r
= EXT_LAST_EXTENT(eh
);
591 if (block
< le32_to_cpu(m
->ee_block
))
595 ext_debug("%p(%u):%p(%u):%p(%u) ", l
, le32_to_cpu(l
->ee_block
),
596 m
, le32_to_cpu(m
->ee_block
),
597 r
, le32_to_cpu(r
->ee_block
));
601 ext_debug(" -> %d:%llu:[%d]%d ",
602 le32_to_cpu(path
->p_ext
->ee_block
),
603 ext4_ext_pblock(path
->p_ext
),
604 ext4_ext_is_uninitialized(path
->p_ext
),
605 ext4_ext_get_actual_len(path
->p_ext
));
607 #ifdef CHECK_BINSEARCH
609 struct ext4_extent
*chex
, *ex
;
612 chex
= ex
= EXT_FIRST_EXTENT(eh
);
613 for (k
= 0; k
< le16_to_cpu(eh
->eh_entries
); k
++, ex
++) {
614 BUG_ON(k
&& le32_to_cpu(ex
->ee_block
)
615 <= le32_to_cpu(ex
[-1].ee_block
));
616 if (block
< le32_to_cpu(ex
->ee_block
))
620 BUG_ON(chex
!= path
->p_ext
);
626 int ext4_ext_tree_init(handle_t
*handle
, struct inode
*inode
)
628 struct ext4_extent_header
*eh
;
630 eh
= ext_inode_hdr(inode
);
633 eh
->eh_magic
= EXT4_EXT_MAGIC
;
634 eh
->eh_max
= cpu_to_le16(ext4_ext_space_root(inode
, 0));
635 ext4_mark_inode_dirty(handle
, inode
);
636 ext4_ext_invalidate_cache(inode
);
640 struct ext4_ext_path
*
641 ext4_ext_find_extent(struct inode
*inode
, ext4_lblk_t block
,
642 struct ext4_ext_path
*path
)
644 struct ext4_extent_header
*eh
;
645 struct buffer_head
*bh
;
646 short int depth
, i
, ppos
= 0, alloc
= 0;
648 eh
= ext_inode_hdr(inode
);
649 depth
= ext_depth(inode
);
651 /* account possible depth increase */
653 path
= kzalloc(sizeof(struct ext4_ext_path
) * (depth
+ 2),
656 return ERR_PTR(-ENOMEM
);
663 /* walk through the tree */
665 int need_to_validate
= 0;
667 ext_debug("depth %d: num %d, max %d\n",
668 ppos
, le16_to_cpu(eh
->eh_entries
), le16_to_cpu(eh
->eh_max
));
670 ext4_ext_binsearch_idx(inode
, path
+ ppos
, block
);
671 path
[ppos
].p_block
= ext4_idx_pblock(path
[ppos
].p_idx
);
672 path
[ppos
].p_depth
= i
;
673 path
[ppos
].p_ext
= NULL
;
675 bh
= sb_getblk(inode
->i_sb
, path
[ppos
].p_block
);
678 if (!bh_uptodate_or_lock(bh
)) {
679 trace_ext4_ext_load_extent(inode
, block
,
681 if (bh_submit_read(bh
) < 0) {
685 /* validate the extent entries */
686 need_to_validate
= 1;
688 eh
= ext_block_hdr(bh
);
690 if (unlikely(ppos
> depth
)) {
692 EXT4_ERROR_INODE(inode
,
693 "ppos %d > depth %d", ppos
, depth
);
696 path
[ppos
].p_bh
= bh
;
697 path
[ppos
].p_hdr
= eh
;
700 if (need_to_validate
&& ext4_ext_check(inode
, eh
, i
))
704 path
[ppos
].p_depth
= i
;
705 path
[ppos
].p_ext
= NULL
;
706 path
[ppos
].p_idx
= NULL
;
709 ext4_ext_binsearch(inode
, path
+ ppos
, block
);
710 /* if not an empty leaf */
711 if (path
[ppos
].p_ext
)
712 path
[ppos
].p_block
= ext4_ext_pblock(path
[ppos
].p_ext
);
714 ext4_ext_show_path(inode
, path
);
719 ext4_ext_drop_refs(path
);
722 return ERR_PTR(-EIO
);
726 * ext4_ext_insert_index:
727 * insert new index [@logical;@ptr] into the block at @curp;
728 * check where to insert: before @curp or after @curp
730 static int ext4_ext_insert_index(handle_t
*handle
, struct inode
*inode
,
731 struct ext4_ext_path
*curp
,
732 int logical
, ext4_fsblk_t ptr
)
734 struct ext4_extent_idx
*ix
;
737 err
= ext4_ext_get_access(handle
, inode
, curp
);
741 if (unlikely(logical
== le32_to_cpu(curp
->p_idx
->ei_block
))) {
742 EXT4_ERROR_INODE(inode
,
743 "logical %d == ei_block %d!",
744 logical
, le32_to_cpu(curp
->p_idx
->ei_block
));
748 if (unlikely(le16_to_cpu(curp
->p_hdr
->eh_entries
)
749 >= le16_to_cpu(curp
->p_hdr
->eh_max
))) {
750 EXT4_ERROR_INODE(inode
,
751 "eh_entries %d >= eh_max %d!",
752 le16_to_cpu(curp
->p_hdr
->eh_entries
),
753 le16_to_cpu(curp
->p_hdr
->eh_max
));
757 len
= EXT_MAX_INDEX(curp
->p_hdr
) - curp
->p_idx
;
758 if (logical
> le32_to_cpu(curp
->p_idx
->ei_block
)) {
760 if (curp
->p_idx
!= EXT_LAST_INDEX(curp
->p_hdr
)) {
761 len
= (len
- 1) * sizeof(struct ext4_extent_idx
);
762 len
= len
< 0 ? 0 : len
;
763 ext_debug("insert new index %d after: %llu. "
764 "move %d from 0x%p to 0x%p\n",
766 (curp
->p_idx
+ 1), (curp
->p_idx
+ 2));
767 memmove(curp
->p_idx
+ 2, curp
->p_idx
+ 1, len
);
769 ix
= curp
->p_idx
+ 1;
772 len
= len
* sizeof(struct ext4_extent_idx
);
773 len
= len
< 0 ? 0 : len
;
774 ext_debug("insert new index %d before: %llu. "
775 "move %d from 0x%p to 0x%p\n",
777 curp
->p_idx
, (curp
->p_idx
+ 1));
778 memmove(curp
->p_idx
+ 1, curp
->p_idx
, len
);
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_ext_path
*path
,
1054 struct ext4_extent
*newext
)
1056 struct ext4_ext_path
*curp
= path
;
1057 struct ext4_extent_header
*neh
;
1058 struct buffer_head
*bh
;
1059 ext4_fsblk_t newblock
;
1062 newblock
= ext4_ext_new_meta_block(handle
, inode
, path
,
1063 newext
, &err
, flags
);
1067 bh
= sb_getblk(inode
->i_sb
, newblock
);
1070 ext4_std_error(inode
->i_sb
, err
);
1075 err
= ext4_journal_get_create_access(handle
, bh
);
1081 /* move top-level index/leaf into new block */
1082 memmove(bh
->b_data
, curp
->p_hdr
, sizeof(EXT4_I(inode
)->i_data
));
1084 /* set size of new block */
1085 neh
= ext_block_hdr(bh
);
1086 /* old root could have indexes or leaves
1087 * so calculate e_max right way */
1088 if (ext_depth(inode
))
1089 neh
->eh_max
= cpu_to_le16(ext4_ext_space_block_idx(inode
, 0));
1091 neh
->eh_max
= cpu_to_le16(ext4_ext_space_block(inode
, 0));
1092 neh
->eh_magic
= EXT4_EXT_MAGIC
;
1093 set_buffer_uptodate(bh
);
1096 err
= ext4_handle_dirty_metadata(handle
, inode
, bh
);
1100 /* create index in new top-level index: num,max,pointer */
1101 err
= ext4_ext_get_access(handle
, inode
, curp
);
1105 curp
->p_hdr
->eh_magic
= EXT4_EXT_MAGIC
;
1106 curp
->p_hdr
->eh_max
= cpu_to_le16(ext4_ext_space_root_idx(inode
, 0));
1107 curp
->p_hdr
->eh_entries
= cpu_to_le16(1);
1108 curp
->p_idx
= EXT_FIRST_INDEX(curp
->p_hdr
);
1110 if (path
[0].p_hdr
->eh_depth
)
1111 curp
->p_idx
->ei_block
=
1112 EXT_FIRST_INDEX(path
[0].p_hdr
)->ei_block
;
1114 curp
->p_idx
->ei_block
=
1115 EXT_FIRST_EXTENT(path
[0].p_hdr
)->ee_block
;
1116 ext4_idx_store_pblock(curp
->p_idx
, newblock
);
1118 neh
= ext_inode_hdr(inode
);
1119 ext_debug("new root: num %d(%d), lblock %d, ptr %llu\n",
1120 le16_to_cpu(neh
->eh_entries
), le16_to_cpu(neh
->eh_max
),
1121 le32_to_cpu(EXT_FIRST_INDEX(neh
)->ei_block
),
1122 ext4_idx_pblock(EXT_FIRST_INDEX(neh
)));
1124 neh
->eh_depth
= cpu_to_le16(path
->p_depth
+ 1);
1125 err
= ext4_ext_dirty(handle
, inode
, curp
);
1133 * ext4_ext_create_new_leaf:
1134 * finds empty index and adds new leaf.
1135 * if no free index is found, then it requests in-depth growing.
1137 static int ext4_ext_create_new_leaf(handle_t
*handle
, struct inode
*inode
,
1139 struct ext4_ext_path
*path
,
1140 struct ext4_extent
*newext
)
1142 struct ext4_ext_path
*curp
;
1143 int depth
, i
, err
= 0;
1146 i
= depth
= ext_depth(inode
);
1148 /* walk up to the tree and look for free index entry */
1149 curp
= path
+ depth
;
1150 while (i
> 0 && !EXT_HAS_FREE_INDEX(curp
)) {
1155 /* we use already allocated block for index block,
1156 * so subsequent data blocks should be contiguous */
1157 if (EXT_HAS_FREE_INDEX(curp
)) {
1158 /* if we found index with free entry, then use that
1159 * entry: create all needed subtree and add new leaf */
1160 err
= ext4_ext_split(handle
, inode
, flags
, path
, newext
, i
);
1165 ext4_ext_drop_refs(path
);
1166 path
= ext4_ext_find_extent(inode
,
1167 (ext4_lblk_t
)le32_to_cpu(newext
->ee_block
),
1170 err
= PTR_ERR(path
);
1172 /* tree is full, time to grow in depth */
1173 err
= ext4_ext_grow_indepth(handle
, inode
, flags
,
1179 ext4_ext_drop_refs(path
);
1180 path
= ext4_ext_find_extent(inode
,
1181 (ext4_lblk_t
)le32_to_cpu(newext
->ee_block
),
1184 err
= PTR_ERR(path
);
1189 * only first (depth 0 -> 1) produces free space;
1190 * in all other cases we have to split the grown tree
1192 depth
= ext_depth(inode
);
1193 if (path
[depth
].p_hdr
->eh_entries
== path
[depth
].p_hdr
->eh_max
) {
1194 /* now we need to split */
1204 * search the closest allocated block to the left for *logical
1205 * and returns it at @logical + it's physical address at @phys
1206 * if *logical is the smallest allocated block, the function
1207 * returns 0 at @phys
1208 * return value contains 0 (success) or error code
1210 static int ext4_ext_search_left(struct inode
*inode
,
1211 struct ext4_ext_path
*path
,
1212 ext4_lblk_t
*logical
, ext4_fsblk_t
*phys
)
1214 struct ext4_extent_idx
*ix
;
1215 struct ext4_extent
*ex
;
1218 if (unlikely(path
== NULL
)) {
1219 EXT4_ERROR_INODE(inode
, "path == NULL *logical %d!", *logical
);
1222 depth
= path
->p_depth
;
1225 if (depth
== 0 && path
->p_ext
== NULL
)
1228 /* usually extent in the path covers blocks smaller
1229 * then *logical, but it can be that extent is the
1230 * first one in the file */
1232 ex
= path
[depth
].p_ext
;
1233 ee_len
= ext4_ext_get_actual_len(ex
);
1234 if (*logical
< le32_to_cpu(ex
->ee_block
)) {
1235 if (unlikely(EXT_FIRST_EXTENT(path
[depth
].p_hdr
) != ex
)) {
1236 EXT4_ERROR_INODE(inode
,
1237 "EXT_FIRST_EXTENT != ex *logical %d ee_block %d!",
1238 *logical
, le32_to_cpu(ex
->ee_block
));
1241 while (--depth
>= 0) {
1242 ix
= path
[depth
].p_idx
;
1243 if (unlikely(ix
!= EXT_FIRST_INDEX(path
[depth
].p_hdr
))) {
1244 EXT4_ERROR_INODE(inode
,
1245 "ix (%d) != EXT_FIRST_INDEX (%d) (depth %d)!",
1246 ix
!= NULL
? le32_to_cpu(ix
->ei_block
) : 0,
1247 EXT_FIRST_INDEX(path
[depth
].p_hdr
) != NULL
?
1248 le32_to_cpu(EXT_FIRST_INDEX(path
[depth
].p_hdr
)->ei_block
) : 0,
1256 if (unlikely(*logical
< (le32_to_cpu(ex
->ee_block
) + ee_len
))) {
1257 EXT4_ERROR_INODE(inode
,
1258 "logical %d < ee_block %d + ee_len %d!",
1259 *logical
, le32_to_cpu(ex
->ee_block
), ee_len
);
1263 *logical
= le32_to_cpu(ex
->ee_block
) + ee_len
- 1;
1264 *phys
= ext4_ext_pblock(ex
) + ee_len
- 1;
1269 * search the closest allocated block to the right for *logical
1270 * and returns it at @logical + it's physical address at @phys
1271 * if *logical is the largest allocated block, the function
1272 * returns 0 at @phys
1273 * return value contains 0 (success) or error code
1275 static int ext4_ext_search_right(struct inode
*inode
,
1276 struct ext4_ext_path
*path
,
1277 ext4_lblk_t
*logical
, ext4_fsblk_t
*phys
,
1278 struct ext4_extent
**ret_ex
)
1280 struct buffer_head
*bh
= NULL
;
1281 struct ext4_extent_header
*eh
;
1282 struct ext4_extent_idx
*ix
;
1283 struct ext4_extent
*ex
;
1285 int depth
; /* Note, NOT eh_depth; depth from top of tree */
1288 if (unlikely(path
== NULL
)) {
1289 EXT4_ERROR_INODE(inode
, "path == NULL *logical %d!", *logical
);
1292 depth
= path
->p_depth
;
1295 if (depth
== 0 && path
->p_ext
== NULL
)
1298 /* usually extent in the path covers blocks smaller
1299 * then *logical, but it can be that extent is the
1300 * first one in the file */
1302 ex
= path
[depth
].p_ext
;
1303 ee_len
= ext4_ext_get_actual_len(ex
);
1304 if (*logical
< le32_to_cpu(ex
->ee_block
)) {
1305 if (unlikely(EXT_FIRST_EXTENT(path
[depth
].p_hdr
) != ex
)) {
1306 EXT4_ERROR_INODE(inode
,
1307 "first_extent(path[%d].p_hdr) != ex",
1311 while (--depth
>= 0) {
1312 ix
= path
[depth
].p_idx
;
1313 if (unlikely(ix
!= EXT_FIRST_INDEX(path
[depth
].p_hdr
))) {
1314 EXT4_ERROR_INODE(inode
,
1315 "ix != EXT_FIRST_INDEX *logical %d!",
1323 if (unlikely(*logical
< (le32_to_cpu(ex
->ee_block
) + ee_len
))) {
1324 EXT4_ERROR_INODE(inode
,
1325 "logical %d < ee_block %d + ee_len %d!",
1326 *logical
, le32_to_cpu(ex
->ee_block
), ee_len
);
1330 if (ex
!= EXT_LAST_EXTENT(path
[depth
].p_hdr
)) {
1331 /* next allocated block in this leaf */
1336 /* go up and search for index to the right */
1337 while (--depth
>= 0) {
1338 ix
= path
[depth
].p_idx
;
1339 if (ix
!= EXT_LAST_INDEX(path
[depth
].p_hdr
))
1343 /* we've gone up to the root and found no index to the right */
1347 /* we've found index to the right, let's
1348 * follow it and find the closest allocated
1349 * block to the right */
1351 block
= ext4_idx_pblock(ix
);
1352 while (++depth
< path
->p_depth
) {
1353 bh
= sb_bread(inode
->i_sb
, block
);
1356 eh
= ext_block_hdr(bh
);
1357 /* subtract from p_depth to get proper eh_depth */
1358 if (ext4_ext_check(inode
, eh
, path
->p_depth
- depth
)) {
1362 ix
= EXT_FIRST_INDEX(eh
);
1363 block
= ext4_idx_pblock(ix
);
1367 bh
= sb_bread(inode
->i_sb
, block
);
1370 eh
= ext_block_hdr(bh
);
1371 if (ext4_ext_check(inode
, eh
, path
->p_depth
- depth
)) {
1375 ex
= EXT_FIRST_EXTENT(eh
);
1377 *logical
= le32_to_cpu(ex
->ee_block
);
1378 *phys
= ext4_ext_pblock(ex
);
1386 * ext4_ext_next_allocated_block:
1387 * returns allocated block in subsequent extent or EXT_MAX_BLOCKS.
1388 * NOTE: it considers block number from index entry as
1389 * allocated block. Thus, index entries have to be consistent
1393 ext4_ext_next_allocated_block(struct ext4_ext_path
*path
)
1397 BUG_ON(path
== NULL
);
1398 depth
= path
->p_depth
;
1400 if (depth
== 0 && path
->p_ext
== NULL
)
1401 return EXT_MAX_BLOCKS
;
1403 while (depth
>= 0) {
1404 if (depth
== path
->p_depth
) {
1406 if (path
[depth
].p_ext
!=
1407 EXT_LAST_EXTENT(path
[depth
].p_hdr
))
1408 return le32_to_cpu(path
[depth
].p_ext
[1].ee_block
);
1411 if (path
[depth
].p_idx
!=
1412 EXT_LAST_INDEX(path
[depth
].p_hdr
))
1413 return le32_to_cpu(path
[depth
].p_idx
[1].ei_block
);
1418 return EXT_MAX_BLOCKS
;
1422 * ext4_ext_next_leaf_block:
1423 * returns first allocated block from next leaf or EXT_MAX_BLOCKS
1425 static ext4_lblk_t
ext4_ext_next_leaf_block(struct ext4_ext_path
*path
)
1429 BUG_ON(path
== NULL
);
1430 depth
= path
->p_depth
;
1432 /* zero-tree has no leaf blocks at all */
1434 return EXT_MAX_BLOCKS
;
1436 /* go to index block */
1439 while (depth
>= 0) {
1440 if (path
[depth
].p_idx
!=
1441 EXT_LAST_INDEX(path
[depth
].p_hdr
))
1442 return (ext4_lblk_t
)
1443 le32_to_cpu(path
[depth
].p_idx
[1].ei_block
);
1447 return EXT_MAX_BLOCKS
;
1451 * ext4_ext_correct_indexes:
1452 * if leaf gets modified and modified extent is first in the leaf,
1453 * then we have to correct all indexes above.
1454 * TODO: do we need to correct tree in all cases?
1456 static int ext4_ext_correct_indexes(handle_t
*handle
, struct inode
*inode
,
1457 struct ext4_ext_path
*path
)
1459 struct ext4_extent_header
*eh
;
1460 int depth
= ext_depth(inode
);
1461 struct ext4_extent
*ex
;
1465 eh
= path
[depth
].p_hdr
;
1466 ex
= path
[depth
].p_ext
;
1468 if (unlikely(ex
== NULL
|| eh
== NULL
)) {
1469 EXT4_ERROR_INODE(inode
,
1470 "ex %p == NULL or eh %p == NULL", ex
, eh
);
1475 /* there is no tree at all */
1479 if (ex
!= EXT_FIRST_EXTENT(eh
)) {
1480 /* we correct tree if first leaf got modified only */
1485 * TODO: we need correction if border is smaller than current one
1488 border
= path
[depth
].p_ext
->ee_block
;
1489 err
= ext4_ext_get_access(handle
, inode
, path
+ k
);
1492 path
[k
].p_idx
->ei_block
= border
;
1493 err
= ext4_ext_dirty(handle
, inode
, path
+ k
);
1498 /* change all left-side indexes */
1499 if (path
[k
+1].p_idx
!= EXT_FIRST_INDEX(path
[k
+1].p_hdr
))
1501 err
= ext4_ext_get_access(handle
, inode
, path
+ k
);
1504 path
[k
].p_idx
->ei_block
= border
;
1505 err
= ext4_ext_dirty(handle
, inode
, path
+ k
);
1514 ext4_can_extents_be_merged(struct inode
*inode
, struct ext4_extent
*ex1
,
1515 struct ext4_extent
*ex2
)
1517 unsigned short ext1_ee_len
, ext2_ee_len
, max_len
;
1520 * Make sure that either both extents are uninitialized, or
1523 if (ext4_ext_is_uninitialized(ex1
) ^ ext4_ext_is_uninitialized(ex2
))
1526 if (ext4_ext_is_uninitialized(ex1
))
1527 max_len
= EXT_UNINIT_MAX_LEN
;
1529 max_len
= EXT_INIT_MAX_LEN
;
1531 ext1_ee_len
= ext4_ext_get_actual_len(ex1
);
1532 ext2_ee_len
= ext4_ext_get_actual_len(ex2
);
1534 if (le32_to_cpu(ex1
->ee_block
) + ext1_ee_len
!=
1535 le32_to_cpu(ex2
->ee_block
))
1539 * To allow future support for preallocated extents to be added
1540 * as an RO_COMPAT feature, refuse to merge to extents if
1541 * this can result in the top bit of ee_len being set.
1543 if (ext1_ee_len
+ ext2_ee_len
> max_len
)
1545 #ifdef AGGRESSIVE_TEST
1546 if (ext1_ee_len
>= 4)
1550 if (ext4_ext_pblock(ex1
) + ext1_ee_len
== ext4_ext_pblock(ex2
))
1556 * This function tries to merge the "ex" extent to the next extent in the tree.
1557 * It always tries to merge towards right. If you want to merge towards
1558 * left, pass "ex - 1" as argument instead of "ex".
1559 * Returns 0 if the extents (ex and ex+1) were _not_ merged and returns
1560 * 1 if they got merged.
1562 static int ext4_ext_try_to_merge_right(struct inode
*inode
,
1563 struct ext4_ext_path
*path
,
1564 struct ext4_extent
*ex
)
1566 struct ext4_extent_header
*eh
;
1567 unsigned int depth
, len
;
1569 int uninitialized
= 0;
1571 depth
= ext_depth(inode
);
1572 BUG_ON(path
[depth
].p_hdr
== NULL
);
1573 eh
= path
[depth
].p_hdr
;
1575 while (ex
< EXT_LAST_EXTENT(eh
)) {
1576 if (!ext4_can_extents_be_merged(inode
, ex
, ex
+ 1))
1578 /* merge with next extent! */
1579 if (ext4_ext_is_uninitialized(ex
))
1581 ex
->ee_len
= cpu_to_le16(ext4_ext_get_actual_len(ex
)
1582 + ext4_ext_get_actual_len(ex
+ 1));
1584 ext4_ext_mark_uninitialized(ex
);
1586 if (ex
+ 1 < EXT_LAST_EXTENT(eh
)) {
1587 len
= (EXT_LAST_EXTENT(eh
) - ex
- 1)
1588 * sizeof(struct ext4_extent
);
1589 memmove(ex
+ 1, ex
+ 2, len
);
1591 le16_add_cpu(&eh
->eh_entries
, -1);
1593 WARN_ON(eh
->eh_entries
== 0);
1594 if (!eh
->eh_entries
)
1595 EXT4_ERROR_INODE(inode
, "eh->eh_entries = 0!");
1602 * This function tries to merge the @ex extent to neighbours in the tree.
1603 * return 1 if merge left else 0.
1605 static int ext4_ext_try_to_merge(struct inode
*inode
,
1606 struct ext4_ext_path
*path
,
1607 struct ext4_extent
*ex
) {
1608 struct ext4_extent_header
*eh
;
1613 depth
= ext_depth(inode
);
1614 BUG_ON(path
[depth
].p_hdr
== NULL
);
1615 eh
= path
[depth
].p_hdr
;
1617 if (ex
> EXT_FIRST_EXTENT(eh
))
1618 merge_done
= ext4_ext_try_to_merge_right(inode
, path
, ex
- 1);
1621 ret
= ext4_ext_try_to_merge_right(inode
, path
, ex
);
1627 * check if a portion of the "newext" extent overlaps with an
1630 * If there is an overlap discovered, it updates the length of the newext
1631 * such that there will be no overlap, and then returns 1.
1632 * If there is no overlap found, it returns 0.
1634 static unsigned int ext4_ext_check_overlap(struct ext4_sb_info
*sbi
,
1635 struct inode
*inode
,
1636 struct ext4_extent
*newext
,
1637 struct ext4_ext_path
*path
)
1640 unsigned int depth
, len1
;
1641 unsigned int ret
= 0;
1643 b1
= le32_to_cpu(newext
->ee_block
);
1644 len1
= ext4_ext_get_actual_len(newext
);
1645 depth
= ext_depth(inode
);
1646 if (!path
[depth
].p_ext
)
1648 b2
= le32_to_cpu(path
[depth
].p_ext
->ee_block
);
1649 b2
&= ~(sbi
->s_cluster_ratio
- 1);
1652 * get the next allocated block if the extent in the path
1653 * is before the requested block(s)
1656 b2
= ext4_ext_next_allocated_block(path
);
1657 if (b2
== EXT_MAX_BLOCKS
)
1659 b2
&= ~(sbi
->s_cluster_ratio
- 1);
1662 /* check for wrap through zero on extent logical start block*/
1663 if (b1
+ len1
< b1
) {
1664 len1
= EXT_MAX_BLOCKS
- b1
;
1665 newext
->ee_len
= cpu_to_le16(len1
);
1669 /* check for overlap */
1670 if (b1
+ len1
> b2
) {
1671 newext
->ee_len
= cpu_to_le16(b2
- b1
);
1679 * ext4_ext_insert_extent:
1680 * tries to merge requsted extent into the existing extent or
1681 * inserts requested extent as new one into the tree,
1682 * creating new leaf in the no-space case.
1684 int ext4_ext_insert_extent(handle_t
*handle
, struct inode
*inode
,
1685 struct ext4_ext_path
*path
,
1686 struct ext4_extent
*newext
, int flag
)
1688 struct ext4_extent_header
*eh
;
1689 struct ext4_extent
*ex
, *fex
;
1690 struct ext4_extent
*nearex
; /* nearest extent */
1691 struct ext4_ext_path
*npath
= NULL
;
1692 int depth
, len
, err
;
1694 unsigned uninitialized
= 0;
1697 if (unlikely(ext4_ext_get_actual_len(newext
) == 0)) {
1698 EXT4_ERROR_INODE(inode
, "ext4_ext_get_actual_len(newext) == 0");
1701 depth
= ext_depth(inode
);
1702 ex
= path
[depth
].p_ext
;
1703 if (unlikely(path
[depth
].p_hdr
== NULL
)) {
1704 EXT4_ERROR_INODE(inode
, "path[%d].p_hdr == NULL", depth
);
1708 /* try to insert block into found extent and return */
1709 if (ex
&& !(flag
& EXT4_GET_BLOCKS_PRE_IO
)
1710 && ext4_can_extents_be_merged(inode
, ex
, newext
)) {
1711 ext_debug("append [%d]%d block to %d:[%d]%d (from %llu)\n",
1712 ext4_ext_is_uninitialized(newext
),
1713 ext4_ext_get_actual_len(newext
),
1714 le32_to_cpu(ex
->ee_block
),
1715 ext4_ext_is_uninitialized(ex
),
1716 ext4_ext_get_actual_len(ex
),
1717 ext4_ext_pblock(ex
));
1718 err
= ext4_ext_get_access(handle
, inode
, path
+ depth
);
1723 * ext4_can_extents_be_merged should have checked that either
1724 * both extents are uninitialized, or both aren't. Thus we
1725 * need to check only one of them here.
1727 if (ext4_ext_is_uninitialized(ex
))
1729 ex
->ee_len
= cpu_to_le16(ext4_ext_get_actual_len(ex
)
1730 + ext4_ext_get_actual_len(newext
));
1732 ext4_ext_mark_uninitialized(ex
);
1733 eh
= path
[depth
].p_hdr
;
1738 depth
= ext_depth(inode
);
1739 eh
= path
[depth
].p_hdr
;
1740 if (le16_to_cpu(eh
->eh_entries
) < le16_to_cpu(eh
->eh_max
))
1743 /* probably next leaf has space for us? */
1744 fex
= EXT_LAST_EXTENT(eh
);
1745 next
= EXT_MAX_BLOCKS
;
1746 if (le32_to_cpu(newext
->ee_block
) > le32_to_cpu(fex
->ee_block
))
1747 next
= ext4_ext_next_leaf_block(path
);
1748 if (next
!= EXT_MAX_BLOCKS
) {
1749 ext_debug("next leaf block - %d\n", next
);
1750 BUG_ON(npath
!= NULL
);
1751 npath
= ext4_ext_find_extent(inode
, next
, NULL
);
1753 return PTR_ERR(npath
);
1754 BUG_ON(npath
->p_depth
!= path
->p_depth
);
1755 eh
= npath
[depth
].p_hdr
;
1756 if (le16_to_cpu(eh
->eh_entries
) < le16_to_cpu(eh
->eh_max
)) {
1757 ext_debug("next leaf isn't full(%d)\n",
1758 le16_to_cpu(eh
->eh_entries
));
1762 ext_debug("next leaf has no free space(%d,%d)\n",
1763 le16_to_cpu(eh
->eh_entries
), le16_to_cpu(eh
->eh_max
));
1767 * There is no free space in the found leaf.
1768 * We're gonna add a new leaf in the tree.
1770 if (flag
& EXT4_GET_BLOCKS_PUNCH_OUT_EXT
)
1771 flags
= EXT4_MB_USE_ROOT_BLOCKS
;
1772 err
= ext4_ext_create_new_leaf(handle
, inode
, flags
, path
, newext
);
1775 depth
= ext_depth(inode
);
1776 eh
= path
[depth
].p_hdr
;
1779 nearex
= path
[depth
].p_ext
;
1781 err
= ext4_ext_get_access(handle
, inode
, path
+ depth
);
1786 /* there is no extent in this leaf, create first one */
1787 ext_debug("first extent in the leaf: %d:%llu:[%d]%d\n",
1788 le32_to_cpu(newext
->ee_block
),
1789 ext4_ext_pblock(newext
),
1790 ext4_ext_is_uninitialized(newext
),
1791 ext4_ext_get_actual_len(newext
));
1792 path
[depth
].p_ext
= EXT_FIRST_EXTENT(eh
);
1793 } else if (le32_to_cpu(newext
->ee_block
)
1794 > le32_to_cpu(nearex
->ee_block
)) {
1795 /* BUG_ON(newext->ee_block == nearex->ee_block); */
1796 if (nearex
!= EXT_LAST_EXTENT(eh
)) {
1797 len
= EXT_MAX_EXTENT(eh
) - nearex
;
1798 len
= (len
- 1) * sizeof(struct ext4_extent
);
1799 len
= len
< 0 ? 0 : len
;
1800 ext_debug("insert %d:%llu:[%d]%d after: nearest 0x%p, "
1801 "move %d from 0x%p to 0x%p\n",
1802 le32_to_cpu(newext
->ee_block
),
1803 ext4_ext_pblock(newext
),
1804 ext4_ext_is_uninitialized(newext
),
1805 ext4_ext_get_actual_len(newext
),
1806 nearex
, len
, nearex
+ 1, nearex
+ 2);
1807 memmove(nearex
+ 2, nearex
+ 1, len
);
1809 path
[depth
].p_ext
= nearex
+ 1;
1811 BUG_ON(newext
->ee_block
== nearex
->ee_block
);
1812 len
= (EXT_MAX_EXTENT(eh
) - nearex
) * sizeof(struct ext4_extent
);
1813 len
= len
< 0 ? 0 : len
;
1814 ext_debug("insert %d:%llu:[%d]%d before: nearest 0x%p, "
1815 "move %d from 0x%p to 0x%p\n",
1816 le32_to_cpu(newext
->ee_block
),
1817 ext4_ext_pblock(newext
),
1818 ext4_ext_is_uninitialized(newext
),
1819 ext4_ext_get_actual_len(newext
),
1820 nearex
, len
, nearex
, nearex
+ 1);
1821 memmove(nearex
+ 1, nearex
, len
);
1822 path
[depth
].p_ext
= nearex
;
1825 le16_add_cpu(&eh
->eh_entries
, 1);
1826 nearex
= path
[depth
].p_ext
;
1827 nearex
->ee_block
= newext
->ee_block
;
1828 ext4_ext_store_pblock(nearex
, ext4_ext_pblock(newext
));
1829 nearex
->ee_len
= newext
->ee_len
;
1832 /* try to merge extents to the right */
1833 if (!(flag
& EXT4_GET_BLOCKS_PRE_IO
))
1834 ext4_ext_try_to_merge(inode
, path
, nearex
);
1836 /* try to merge extents to the left */
1838 /* time to correct all indexes above */
1839 err
= ext4_ext_correct_indexes(handle
, inode
, path
);
1843 err
= ext4_ext_dirty(handle
, inode
, path
+ depth
);
1847 ext4_ext_drop_refs(npath
);
1850 ext4_ext_invalidate_cache(inode
);
1854 static int ext4_ext_walk_space(struct inode
*inode
, ext4_lblk_t block
,
1855 ext4_lblk_t num
, ext_prepare_callback func
,
1858 struct ext4_ext_path
*path
= NULL
;
1859 struct ext4_ext_cache cbex
;
1860 struct ext4_extent
*ex
;
1861 ext4_lblk_t next
, start
= 0, end
= 0;
1862 ext4_lblk_t last
= block
+ num
;
1863 int depth
, exists
, err
= 0;
1865 BUG_ON(func
== NULL
);
1866 BUG_ON(inode
== NULL
);
1868 while (block
< last
&& block
!= EXT_MAX_BLOCKS
) {
1870 /* find extent for this block */
1871 down_read(&EXT4_I(inode
)->i_data_sem
);
1872 path
= ext4_ext_find_extent(inode
, block
, path
);
1873 up_read(&EXT4_I(inode
)->i_data_sem
);
1875 err
= PTR_ERR(path
);
1880 depth
= ext_depth(inode
);
1881 if (unlikely(path
[depth
].p_hdr
== NULL
)) {
1882 EXT4_ERROR_INODE(inode
, "path[%d].p_hdr == NULL", depth
);
1886 ex
= path
[depth
].p_ext
;
1887 next
= ext4_ext_next_allocated_block(path
);
1891 /* there is no extent yet, so try to allocate
1892 * all requested space */
1895 } else if (le32_to_cpu(ex
->ee_block
) > block
) {
1896 /* need to allocate space before found extent */
1898 end
= le32_to_cpu(ex
->ee_block
);
1899 if (block
+ num
< end
)
1901 } else if (block
>= le32_to_cpu(ex
->ee_block
)
1902 + ext4_ext_get_actual_len(ex
)) {
1903 /* need to allocate space after found extent */
1908 } else if (block
>= le32_to_cpu(ex
->ee_block
)) {
1910 * some part of requested space is covered
1914 end
= le32_to_cpu(ex
->ee_block
)
1915 + ext4_ext_get_actual_len(ex
);
1916 if (block
+ num
< end
)
1922 BUG_ON(end
<= start
);
1925 cbex
.ec_block
= start
;
1926 cbex
.ec_len
= end
- start
;
1929 cbex
.ec_block
= le32_to_cpu(ex
->ee_block
);
1930 cbex
.ec_len
= ext4_ext_get_actual_len(ex
);
1931 cbex
.ec_start
= ext4_ext_pblock(ex
);
1934 if (unlikely(cbex
.ec_len
== 0)) {
1935 EXT4_ERROR_INODE(inode
, "cbex.ec_len == 0");
1939 err
= func(inode
, next
, &cbex
, ex
, cbdata
);
1940 ext4_ext_drop_refs(path
);
1945 if (err
== EXT_REPEAT
)
1947 else if (err
== EXT_BREAK
) {
1952 if (ext_depth(inode
) != depth
) {
1953 /* depth was changed. we have to realloc path */
1958 block
= cbex
.ec_block
+ cbex
.ec_len
;
1962 ext4_ext_drop_refs(path
);
1970 ext4_ext_put_in_cache(struct inode
*inode
, ext4_lblk_t block
,
1971 __u32 len
, ext4_fsblk_t start
)
1973 struct ext4_ext_cache
*cex
;
1975 spin_lock(&EXT4_I(inode
)->i_block_reservation_lock
);
1976 trace_ext4_ext_put_in_cache(inode
, block
, len
, start
);
1977 cex
= &EXT4_I(inode
)->i_cached_extent
;
1978 cex
->ec_block
= block
;
1980 cex
->ec_start
= start
;
1981 spin_unlock(&EXT4_I(inode
)->i_block_reservation_lock
);
1985 * ext4_ext_put_gap_in_cache:
1986 * calculate boundaries of the gap that the requested block fits into
1987 * and cache this gap
1990 ext4_ext_put_gap_in_cache(struct inode
*inode
, struct ext4_ext_path
*path
,
1993 int depth
= ext_depth(inode
);
1996 struct ext4_extent
*ex
;
1998 ex
= path
[depth
].p_ext
;
2000 /* there is no extent yet, so gap is [0;-] */
2002 len
= EXT_MAX_BLOCKS
;
2003 ext_debug("cache gap(whole file):");
2004 } else if (block
< le32_to_cpu(ex
->ee_block
)) {
2006 len
= le32_to_cpu(ex
->ee_block
) - block
;
2007 ext_debug("cache gap(before): %u [%u:%u]",
2009 le32_to_cpu(ex
->ee_block
),
2010 ext4_ext_get_actual_len(ex
));
2011 } else if (block
>= le32_to_cpu(ex
->ee_block
)
2012 + ext4_ext_get_actual_len(ex
)) {
2014 lblock
= le32_to_cpu(ex
->ee_block
)
2015 + ext4_ext_get_actual_len(ex
);
2017 next
= ext4_ext_next_allocated_block(path
);
2018 ext_debug("cache gap(after): [%u:%u] %u",
2019 le32_to_cpu(ex
->ee_block
),
2020 ext4_ext_get_actual_len(ex
),
2022 BUG_ON(next
== lblock
);
2023 len
= next
- lblock
;
2029 ext_debug(" -> %u:%lu\n", lblock
, len
);
2030 ext4_ext_put_in_cache(inode
, lblock
, len
, 0);
2034 * ext4_ext_check_cache()
2035 * Checks to see if the given block is in the cache.
2036 * If it is, the cached extent is stored in the given
2037 * cache extent pointer. If the cached extent is a hole,
2038 * this routine should be used instead of
2039 * ext4_ext_in_cache if the calling function needs to
2040 * know the size of the hole.
2042 * @inode: The files inode
2043 * @block: The block to look for in the cache
2044 * @ex: Pointer where the cached extent will be stored
2045 * if it contains block
2047 * Return 0 if cache is invalid; 1 if the cache is valid
2049 static int ext4_ext_check_cache(struct inode
*inode
, ext4_lblk_t block
,
2050 struct ext4_ext_cache
*ex
){
2051 struct ext4_ext_cache
*cex
;
2052 struct ext4_sb_info
*sbi
;
2056 * We borrow i_block_reservation_lock to protect i_cached_extent
2058 spin_lock(&EXT4_I(inode
)->i_block_reservation_lock
);
2059 cex
= &EXT4_I(inode
)->i_cached_extent
;
2060 sbi
= EXT4_SB(inode
->i_sb
);
2062 /* has cache valid data? */
2063 if (cex
->ec_len
== 0)
2066 if (in_range(block
, cex
->ec_block
, cex
->ec_len
)) {
2067 memcpy(ex
, cex
, sizeof(struct ext4_ext_cache
));
2068 ext_debug("%u cached by %u:%u:%llu\n",
2070 cex
->ec_block
, cex
->ec_len
, cex
->ec_start
);
2075 sbi
->extent_cache_misses
++;
2077 sbi
->extent_cache_hits
++;
2078 trace_ext4_ext_in_cache(inode
, block
, ret
);
2079 spin_unlock(&EXT4_I(inode
)->i_block_reservation_lock
);
2084 * ext4_ext_in_cache()
2085 * Checks to see if the given block is in the cache.
2086 * If it is, the cached extent is stored in the given
2089 * @inode: The files inode
2090 * @block: The block to look for in the cache
2091 * @ex: Pointer where the cached extent will be stored
2092 * if it contains block
2094 * Return 0 if cache is invalid; 1 if the cache is valid
2097 ext4_ext_in_cache(struct inode
*inode
, ext4_lblk_t block
,
2098 struct ext4_extent
*ex
)
2100 struct ext4_ext_cache cex
;
2103 if (ext4_ext_check_cache(inode
, block
, &cex
)) {
2104 ex
->ee_block
= cpu_to_le32(cex
.ec_block
);
2105 ext4_ext_store_pblock(ex
, cex
.ec_start
);
2106 ex
->ee_len
= cpu_to_le16(cex
.ec_len
);
2116 * removes index from the index block.
2118 static int ext4_ext_rm_idx(handle_t
*handle
, struct inode
*inode
,
2119 struct ext4_ext_path
*path
)
2124 /* free index block */
2126 leaf
= ext4_idx_pblock(path
->p_idx
);
2127 if (unlikely(path
->p_hdr
->eh_entries
== 0)) {
2128 EXT4_ERROR_INODE(inode
, "path->p_hdr->eh_entries == 0");
2131 err
= ext4_ext_get_access(handle
, inode
, path
);
2135 if (path
->p_idx
!= EXT_LAST_INDEX(path
->p_hdr
)) {
2136 int len
= EXT_LAST_INDEX(path
->p_hdr
) - path
->p_idx
;
2137 len
*= sizeof(struct ext4_extent_idx
);
2138 memmove(path
->p_idx
, path
->p_idx
+ 1, len
);
2141 le16_add_cpu(&path
->p_hdr
->eh_entries
, -1);
2142 err
= ext4_ext_dirty(handle
, inode
, path
);
2145 ext_debug("index is empty, remove it, free block %llu\n", leaf
);
2146 trace_ext4_ext_rm_idx(inode
, leaf
);
2148 ext4_free_blocks(handle
, inode
, NULL
, leaf
, 1,
2149 EXT4_FREE_BLOCKS_METADATA
| EXT4_FREE_BLOCKS_FORGET
);
2154 * ext4_ext_calc_credits_for_single_extent:
2155 * This routine returns max. credits that needed to insert an extent
2156 * to the extent tree.
2157 * When pass the actual path, the caller should calculate credits
2160 int ext4_ext_calc_credits_for_single_extent(struct inode
*inode
, int nrblocks
,
2161 struct ext4_ext_path
*path
)
2164 int depth
= ext_depth(inode
);
2167 /* probably there is space in leaf? */
2168 if (le16_to_cpu(path
[depth
].p_hdr
->eh_entries
)
2169 < le16_to_cpu(path
[depth
].p_hdr
->eh_max
)) {
2172 * There are some space in the leaf tree, no
2173 * need to account for leaf block credit
2175 * bitmaps and block group descriptor blocks
2176 * and other metadata blocks still need to be
2179 /* 1 bitmap, 1 block group descriptor */
2180 ret
= 2 + EXT4_META_TRANS_BLOCKS(inode
->i_sb
);
2185 return ext4_chunk_trans_blocks(inode
, nrblocks
);
2189 * How many index/leaf blocks need to change/allocate to modify nrblocks?
2191 * if nrblocks are fit in a single extent (chunk flag is 1), then
2192 * in the worse case, each tree level index/leaf need to be changed
2193 * if the tree split due to insert a new extent, then the old tree
2194 * index/leaf need to be updated too
2196 * If the nrblocks are discontiguous, they could cause
2197 * the whole tree split more than once, but this is really rare.
2199 int ext4_ext_index_trans_blocks(struct inode
*inode
, int nrblocks
, int chunk
)
2202 int depth
= ext_depth(inode
);
2212 static int ext4_remove_blocks(handle_t
*handle
, struct inode
*inode
,
2213 struct ext4_extent
*ex
,
2214 ext4_fsblk_t
*partial_cluster
,
2215 ext4_lblk_t from
, ext4_lblk_t to
)
2217 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
2218 unsigned short ee_len
= ext4_ext_get_actual_len(ex
);
2220 int flags
= EXT4_FREE_BLOCKS_FORGET
;
2222 if (S_ISDIR(inode
->i_mode
) || S_ISLNK(inode
->i_mode
))
2223 flags
|= EXT4_FREE_BLOCKS_METADATA
;
2225 * For bigalloc file systems, we never free a partial cluster
2226 * at the beginning of the extent. Instead, we make a note
2227 * that we tried freeing the cluster, and check to see if we
2228 * need to free it on a subsequent call to ext4_remove_blocks,
2229 * or at the end of the ext4_truncate() operation.
2231 flags
|= EXT4_FREE_BLOCKS_NOFREE_FIRST_CLUSTER
;
2233 trace_ext4_remove_blocks(inode
, ex
, from
, to
, *partial_cluster
);
2235 * If we have a partial cluster, and it's different from the
2236 * cluster of the last block, we need to explicitly free the
2237 * partial cluster here.
2239 pblk
= ext4_ext_pblock(ex
) + ee_len
- 1;
2240 if (*partial_cluster
&& (EXT4_B2C(sbi
, pblk
) != *partial_cluster
)) {
2241 ext4_free_blocks(handle
, inode
, NULL
,
2242 EXT4_C2B(sbi
, *partial_cluster
),
2243 sbi
->s_cluster_ratio
, flags
);
2244 *partial_cluster
= 0;
2247 #ifdef EXTENTS_STATS
2249 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
2250 spin_lock(&sbi
->s_ext_stats_lock
);
2251 sbi
->s_ext_blocks
+= ee_len
;
2252 sbi
->s_ext_extents
++;
2253 if (ee_len
< sbi
->s_ext_min
)
2254 sbi
->s_ext_min
= ee_len
;
2255 if (ee_len
> sbi
->s_ext_max
)
2256 sbi
->s_ext_max
= ee_len
;
2257 if (ext_depth(inode
) > sbi
->s_depth_max
)
2258 sbi
->s_depth_max
= ext_depth(inode
);
2259 spin_unlock(&sbi
->s_ext_stats_lock
);
2262 if (from
>= le32_to_cpu(ex
->ee_block
)
2263 && to
== le32_to_cpu(ex
->ee_block
) + ee_len
- 1) {
2267 num
= le32_to_cpu(ex
->ee_block
) + ee_len
- from
;
2268 pblk
= ext4_ext_pblock(ex
) + ee_len
- num
;
2269 ext_debug("free last %u blocks starting %llu\n", num
, pblk
);
2270 ext4_free_blocks(handle
, inode
, NULL
, pblk
, num
, flags
);
2272 * If the block range to be freed didn't start at the
2273 * beginning of a cluster, and we removed the entire
2274 * extent, save the partial cluster here, since we
2275 * might need to delete if we determine that the
2276 * truncate operation has removed all of the blocks in
2279 if (pblk
& (sbi
->s_cluster_ratio
- 1) &&
2281 *partial_cluster
= EXT4_B2C(sbi
, pblk
);
2283 *partial_cluster
= 0;
2284 } else if (from
== le32_to_cpu(ex
->ee_block
)
2285 && to
<= le32_to_cpu(ex
->ee_block
) + ee_len
- 1) {
2291 start
= ext4_ext_pblock(ex
);
2293 ext_debug("free first %u blocks starting %llu\n", num
, start
);
2294 ext4_free_blocks(handle
, inode
, NULL
, start
, num
, flags
);
2297 printk(KERN_INFO
"strange request: removal(2) "
2298 "%u-%u from %u:%u\n",
2299 from
, to
, le32_to_cpu(ex
->ee_block
), ee_len
);
2306 * ext4_ext_rm_leaf() Removes the extents associated with the
2307 * blocks appearing between "start" and "end", and splits the extents
2308 * if "start" and "end" appear in the same extent
2310 * @handle: The journal handle
2311 * @inode: The files inode
2312 * @path: The path to the leaf
2313 * @start: The first block to remove
2314 * @end: The last block to remove
2317 ext4_ext_rm_leaf(handle_t
*handle
, struct inode
*inode
,
2318 struct ext4_ext_path
*path
, ext4_fsblk_t
*partial_cluster
,
2319 ext4_lblk_t start
, ext4_lblk_t end
)
2321 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
2322 int err
= 0, correct_index
= 0;
2323 int depth
= ext_depth(inode
), credits
;
2324 struct ext4_extent_header
*eh
;
2325 ext4_lblk_t a
, b
, block
;
2327 ext4_lblk_t ex_ee_block
;
2328 unsigned short ex_ee_len
;
2329 unsigned uninitialized
= 0;
2330 struct ext4_extent
*ex
;
2331 struct ext4_map_blocks map
;
2333 /* the header must be checked already in ext4_ext_remove_space() */
2334 ext_debug("truncate since %u in leaf\n", start
);
2335 if (!path
[depth
].p_hdr
)
2336 path
[depth
].p_hdr
= ext_block_hdr(path
[depth
].p_bh
);
2337 eh
= path
[depth
].p_hdr
;
2338 if (unlikely(path
[depth
].p_hdr
== NULL
)) {
2339 EXT4_ERROR_INODE(inode
, "path[%d].p_hdr == NULL", depth
);
2342 /* find where to start removing */
2343 ex
= EXT_LAST_EXTENT(eh
);
2345 ex_ee_block
= le32_to_cpu(ex
->ee_block
);
2346 ex_ee_len
= ext4_ext_get_actual_len(ex
);
2348 trace_ext4_ext_rm_leaf(inode
, start
, ex
, *partial_cluster
);
2350 while (ex
>= EXT_FIRST_EXTENT(eh
) &&
2351 ex_ee_block
+ ex_ee_len
> start
) {
2353 if (ext4_ext_is_uninitialized(ex
))
2358 ext_debug("remove ext %u:[%d]%d\n", ex_ee_block
,
2359 uninitialized
, ex_ee_len
);
2360 path
[depth
].p_ext
= ex
;
2362 a
= ex_ee_block
> start
? ex_ee_block
: start
;
2363 b
= ex_ee_block
+ex_ee_len
- 1 < end
?
2364 ex_ee_block
+ex_ee_len
- 1 : end
;
2366 ext_debug(" border %u:%u\n", a
, b
);
2368 /* If this extent is beyond the end of the hole, skip it */
2369 if (end
<= ex_ee_block
) {
2371 ex_ee_block
= le32_to_cpu(ex
->ee_block
);
2372 ex_ee_len
= ext4_ext_get_actual_len(ex
);
2374 } else if (a
!= ex_ee_block
&&
2375 b
!= ex_ee_block
+ ex_ee_len
- 1) {
2377 * If this is a truncate, then this condition should
2378 * never happen because at least one of the end points
2379 * needs to be on the edge of the extent.
2381 if (end
== EXT_MAX_BLOCKS
- 1) {
2382 ext_debug(" bad truncate %u:%u\n",
2390 * else this is a hole punch, so the extent needs to
2391 * be split since neither edge of the hole is on the
2395 map
.m_pblk
= ext4_ext_pblock(ex
);
2396 map
.m_lblk
= ex_ee_block
;
2397 map
.m_len
= b
- ex_ee_block
;
2399 err
= ext4_split_extent(handle
,
2400 inode
, path
, &map
, 0,
2401 EXT4_GET_BLOCKS_PUNCH_OUT_EXT
|
2402 EXT4_GET_BLOCKS_PRE_IO
);
2407 ex_ee_len
= ext4_ext_get_actual_len(ex
);
2409 b
= ex_ee_block
+ex_ee_len
- 1 < end
?
2410 ex_ee_block
+ex_ee_len
- 1 : end
;
2412 /* Then remove tail of this extent */
2413 block
= ex_ee_block
;
2416 } else if (a
!= ex_ee_block
) {
2417 /* remove tail of the extent */
2418 block
= ex_ee_block
;
2420 } else if (b
!= ex_ee_block
+ ex_ee_len
- 1) {
2421 /* remove head of the extent */
2423 num
= ex_ee_block
+ ex_ee_len
- b
;
2426 * If this is a truncate, this condition
2427 * should never happen
2429 if (end
== EXT_MAX_BLOCKS
- 1) {
2430 ext_debug(" bad truncate %u:%u\n",
2436 /* remove whole extent: excellent! */
2437 block
= ex_ee_block
;
2439 if (a
!= ex_ee_block
) {
2440 ext_debug(" bad truncate %u:%u\n",
2446 if (b
!= ex_ee_block
+ ex_ee_len
- 1) {
2447 ext_debug(" bad truncate %u:%u\n",
2455 * 3 for leaf, sb, and inode plus 2 (bmap and group
2456 * descriptor) for each block group; assume two block
2457 * groups plus ex_ee_len/blocks_per_block_group for
2460 credits
= 7 + 2*(ex_ee_len
/EXT4_BLOCKS_PER_GROUP(inode
->i_sb
));
2461 if (ex
== EXT_FIRST_EXTENT(eh
)) {
2463 credits
+= (ext_depth(inode
)) + 1;
2465 credits
+= EXT4_MAXQUOTAS_TRANS_BLOCKS(inode
->i_sb
);
2467 err
= ext4_ext_truncate_extend_restart(handle
, inode
, credits
);
2471 err
= ext4_ext_get_access(handle
, inode
, path
+ depth
);
2475 err
= ext4_remove_blocks(handle
, inode
, ex
, partial_cluster
,
2481 /* this extent is removed; mark slot entirely unused */
2482 ext4_ext_store_pblock(ex
, 0);
2483 } else if (block
!= ex_ee_block
) {
2485 * If this was a head removal, then we need to update
2486 * the physical block since it is now at a different
2489 ext4_ext_store_pblock(ex
, ext4_ext_pblock(ex
) + (b
-a
));
2492 ex
->ee_block
= cpu_to_le32(block
);
2493 ex
->ee_len
= cpu_to_le16(num
);
2495 * Do not mark uninitialized if all the blocks in the
2496 * extent have been removed.
2498 if (uninitialized
&& num
)
2499 ext4_ext_mark_uninitialized(ex
);
2501 err
= ext4_ext_dirty(handle
, inode
, path
+ depth
);
2506 * If the extent was completely released,
2507 * we need to remove it from the leaf
2510 if (end
!= EXT_MAX_BLOCKS
- 1) {
2512 * For hole punching, we need to scoot all the
2513 * extents up when an extent is removed so that
2514 * we dont have blank extents in the middle
2516 memmove(ex
, ex
+1, (EXT_LAST_EXTENT(eh
) - ex
) *
2517 sizeof(struct ext4_extent
));
2519 /* Now get rid of the one at the end */
2520 memset(EXT_LAST_EXTENT(eh
), 0,
2521 sizeof(struct ext4_extent
));
2523 le16_add_cpu(&eh
->eh_entries
, -1);
2525 *partial_cluster
= 0;
2527 ext_debug("new extent: %u:%u:%llu\n", block
, num
,
2528 ext4_ext_pblock(ex
));
2530 ex_ee_block
= le32_to_cpu(ex
->ee_block
);
2531 ex_ee_len
= ext4_ext_get_actual_len(ex
);
2534 if (correct_index
&& eh
->eh_entries
)
2535 err
= ext4_ext_correct_indexes(handle
, inode
, path
);
2538 * If there is still a entry in the leaf node, check to see if
2539 * it references the partial cluster. This is the only place
2540 * where it could; if it doesn't, we can free the cluster.
2542 if (*partial_cluster
&& ex
>= EXT_FIRST_EXTENT(eh
) &&
2543 (EXT4_B2C(sbi
, ext4_ext_pblock(ex
) + ex_ee_len
- 1) !=
2544 *partial_cluster
)) {
2545 int flags
= EXT4_FREE_BLOCKS_FORGET
;
2547 if (S_ISDIR(inode
->i_mode
) || S_ISLNK(inode
->i_mode
))
2548 flags
|= EXT4_FREE_BLOCKS_METADATA
;
2550 ext4_free_blocks(handle
, inode
, NULL
,
2551 EXT4_C2B(sbi
, *partial_cluster
),
2552 sbi
->s_cluster_ratio
, flags
);
2553 *partial_cluster
= 0;
2556 /* if this leaf is free, then we should
2557 * remove it from index block above */
2558 if (err
== 0 && eh
->eh_entries
== 0 && path
[depth
].p_bh
!= NULL
)
2559 err
= ext4_ext_rm_idx(handle
, inode
, path
+ depth
);
2566 * ext4_ext_more_to_rm:
2567 * returns 1 if current index has to be freed (even partial)
2570 ext4_ext_more_to_rm(struct ext4_ext_path
*path
)
2572 BUG_ON(path
->p_idx
== NULL
);
2574 if (path
->p_idx
< EXT_FIRST_INDEX(path
->p_hdr
))
2578 * if truncate on deeper level happened, it wasn't partial,
2579 * so we have to consider current index for truncation
2581 if (le16_to_cpu(path
->p_hdr
->eh_entries
) == path
->p_block
)
2586 static int ext4_ext_remove_space(struct inode
*inode
, ext4_lblk_t start
)
2588 struct super_block
*sb
= inode
->i_sb
;
2589 int depth
= ext_depth(inode
);
2590 struct ext4_ext_path
*path
;
2591 ext4_fsblk_t partial_cluster
= 0;
2595 ext_debug("truncate since %u\n", start
);
2597 /* probably first extent we're gonna free will be last in block */
2598 handle
= ext4_journal_start(inode
, depth
+ 1);
2600 return PTR_ERR(handle
);
2603 ext4_ext_invalidate_cache(inode
);
2605 trace_ext4_ext_remove_space(inode
, start
, depth
);
2608 * We start scanning from right side, freeing all the blocks
2609 * after i_size and walking into the tree depth-wise.
2611 depth
= ext_depth(inode
);
2612 path
= kzalloc(sizeof(struct ext4_ext_path
) * (depth
+ 1), GFP_NOFS
);
2614 ext4_journal_stop(handle
);
2617 path
[0].p_depth
= depth
;
2618 path
[0].p_hdr
= ext_inode_hdr(inode
);
2619 if (ext4_ext_check(inode
, path
[0].p_hdr
, depth
)) {
2625 while (i
>= 0 && err
== 0) {
2627 /* this is leaf block */
2628 err
= ext4_ext_rm_leaf(handle
, inode
, path
,
2629 &partial_cluster
, start
,
2630 EXT_MAX_BLOCKS
- 1);
2631 /* root level has p_bh == NULL, brelse() eats this */
2632 brelse(path
[i
].p_bh
);
2633 path
[i
].p_bh
= NULL
;
2638 /* this is index block */
2639 if (!path
[i
].p_hdr
) {
2640 ext_debug("initialize header\n");
2641 path
[i
].p_hdr
= ext_block_hdr(path
[i
].p_bh
);
2644 if (!path
[i
].p_idx
) {
2645 /* this level hasn't been touched yet */
2646 path
[i
].p_idx
= EXT_LAST_INDEX(path
[i
].p_hdr
);
2647 path
[i
].p_block
= le16_to_cpu(path
[i
].p_hdr
->eh_entries
)+1;
2648 ext_debug("init index ptr: hdr 0x%p, num %d\n",
2650 le16_to_cpu(path
[i
].p_hdr
->eh_entries
));
2652 /* we were already here, see at next index */
2656 ext_debug("level %d - index, first 0x%p, cur 0x%p\n",
2657 i
, EXT_FIRST_INDEX(path
[i
].p_hdr
),
2659 if (ext4_ext_more_to_rm(path
+ i
)) {
2660 struct buffer_head
*bh
;
2661 /* go to the next level */
2662 ext_debug("move to level %d (block %llu)\n",
2663 i
+ 1, ext4_idx_pblock(path
[i
].p_idx
));
2664 memset(path
+ i
+ 1, 0, sizeof(*path
));
2665 bh
= sb_bread(sb
, ext4_idx_pblock(path
[i
].p_idx
));
2667 /* should we reset i_size? */
2671 if (WARN_ON(i
+ 1 > depth
)) {
2675 if (ext4_ext_check(inode
, ext_block_hdr(bh
),
2680 path
[i
+ 1].p_bh
= bh
;
2682 /* save actual number of indexes since this
2683 * number is changed at the next iteration */
2684 path
[i
].p_block
= le16_to_cpu(path
[i
].p_hdr
->eh_entries
);
2687 /* we finished processing this index, go up */
2688 if (path
[i
].p_hdr
->eh_entries
== 0 && i
> 0) {
2689 /* index is empty, remove it;
2690 * handle must be already prepared by the
2691 * truncatei_leaf() */
2692 err
= ext4_ext_rm_idx(handle
, inode
, path
+ i
);
2694 /* root level has p_bh == NULL, brelse() eats this */
2695 brelse(path
[i
].p_bh
);
2696 path
[i
].p_bh
= NULL
;
2698 ext_debug("return to level %d\n", i
);
2702 trace_ext4_ext_remove_space_done(inode
, start
, depth
, partial_cluster
,
2703 path
->p_hdr
->eh_entries
);
2705 /* If we still have something in the partial cluster and we have removed
2706 * even the first extent, then we should free the blocks in the partial
2707 * cluster as well. */
2708 if (partial_cluster
&& path
->p_hdr
->eh_entries
== 0) {
2709 int flags
= EXT4_FREE_BLOCKS_FORGET
;
2711 if (S_ISDIR(inode
->i_mode
) || S_ISLNK(inode
->i_mode
))
2712 flags
|= EXT4_FREE_BLOCKS_METADATA
;
2714 ext4_free_blocks(handle
, inode
, NULL
,
2715 EXT4_C2B(EXT4_SB(sb
), partial_cluster
),
2716 EXT4_SB(sb
)->s_cluster_ratio
, flags
);
2717 partial_cluster
= 0;
2720 /* TODO: flexible tree reduction should be here */
2721 if (path
->p_hdr
->eh_entries
== 0) {
2723 * truncate to zero freed all the tree,
2724 * so we need to correct eh_depth
2726 err
= ext4_ext_get_access(handle
, inode
, path
);
2728 ext_inode_hdr(inode
)->eh_depth
= 0;
2729 ext_inode_hdr(inode
)->eh_max
=
2730 cpu_to_le16(ext4_ext_space_root(inode
, 0));
2731 err
= ext4_ext_dirty(handle
, inode
, path
);
2735 ext4_ext_drop_refs(path
);
2739 ext4_journal_stop(handle
);
2745 * called at mount time
2747 void ext4_ext_init(struct super_block
*sb
)
2750 * possible initialization would be here
2753 if (EXT4_HAS_INCOMPAT_FEATURE(sb
, EXT4_FEATURE_INCOMPAT_EXTENTS
)) {
2754 #if defined(AGGRESSIVE_TEST) || defined(CHECK_BINSEARCH) || defined(EXTENTS_STATS)
2755 printk(KERN_INFO
"EXT4-fs: file extents enabled");
2756 #ifdef AGGRESSIVE_TEST
2757 printk(", aggressive tests");
2759 #ifdef CHECK_BINSEARCH
2760 printk(", check binsearch");
2762 #ifdef EXTENTS_STATS
2767 #ifdef EXTENTS_STATS
2768 spin_lock_init(&EXT4_SB(sb
)->s_ext_stats_lock
);
2769 EXT4_SB(sb
)->s_ext_min
= 1 << 30;
2770 EXT4_SB(sb
)->s_ext_max
= 0;
2776 * called at umount time
2778 void ext4_ext_release(struct super_block
*sb
)
2780 if (!EXT4_HAS_INCOMPAT_FEATURE(sb
, EXT4_FEATURE_INCOMPAT_EXTENTS
))
2783 #ifdef EXTENTS_STATS
2784 if (EXT4_SB(sb
)->s_ext_blocks
&& EXT4_SB(sb
)->s_ext_extents
) {
2785 struct ext4_sb_info
*sbi
= EXT4_SB(sb
);
2786 printk(KERN_ERR
"EXT4-fs: %lu blocks in %lu extents (%lu ave)\n",
2787 sbi
->s_ext_blocks
, sbi
->s_ext_extents
,
2788 sbi
->s_ext_blocks
/ sbi
->s_ext_extents
);
2789 printk(KERN_ERR
"EXT4-fs: extents: %lu min, %lu max, max depth %lu\n",
2790 sbi
->s_ext_min
, sbi
->s_ext_max
, sbi
->s_depth_max
);
2795 /* FIXME!! we need to try to merge to left or right after zero-out */
2796 static int ext4_ext_zeroout(struct inode
*inode
, struct ext4_extent
*ex
)
2798 ext4_fsblk_t ee_pblock
;
2799 unsigned int ee_len
;
2802 ee_len
= ext4_ext_get_actual_len(ex
);
2803 ee_pblock
= ext4_ext_pblock(ex
);
2805 ret
= sb_issue_zeroout(inode
->i_sb
, ee_pblock
, ee_len
, GFP_NOFS
);
2813 * used by extent splitting.
2815 #define EXT4_EXT_MAY_ZEROOUT 0x1 /* safe to zeroout if split fails \
2817 #define EXT4_EXT_MARK_UNINIT1 0x2 /* mark first half uninitialized */
2818 #define EXT4_EXT_MARK_UNINIT2 0x4 /* mark second half uninitialized */
2821 * ext4_split_extent_at() splits an extent at given block.
2823 * @handle: the journal handle
2824 * @inode: the file inode
2825 * @path: the path to the extent
2826 * @split: the logical block where the extent is splitted.
2827 * @split_flags: indicates if the extent could be zeroout if split fails, and
2828 * the states(init or uninit) of new extents.
2829 * @flags: flags used to insert new extent to extent tree.
2832 * Splits extent [a, b] into two extents [a, @split) and [@split, b], states
2833 * of which are deterimined by split_flag.
2835 * There are two cases:
2836 * a> the extent are splitted into two extent.
2837 * b> split is not needed, and just mark the extent.
2839 * return 0 on success.
2841 static int ext4_split_extent_at(handle_t
*handle
,
2842 struct inode
*inode
,
2843 struct ext4_ext_path
*path
,
2848 ext4_fsblk_t newblock
;
2849 ext4_lblk_t ee_block
;
2850 struct ext4_extent
*ex
, newex
, orig_ex
;
2851 struct ext4_extent
*ex2
= NULL
;
2852 unsigned int ee_len
, depth
;
2855 ext_debug("ext4_split_extents_at: inode %lu, logical"
2856 "block %llu\n", inode
->i_ino
, (unsigned long long)split
);
2858 ext4_ext_show_leaf(inode
, path
);
2860 depth
= ext_depth(inode
);
2861 ex
= path
[depth
].p_ext
;
2862 ee_block
= le32_to_cpu(ex
->ee_block
);
2863 ee_len
= ext4_ext_get_actual_len(ex
);
2864 newblock
= split
- ee_block
+ ext4_ext_pblock(ex
);
2866 BUG_ON(split
< ee_block
|| split
>= (ee_block
+ ee_len
));
2868 err
= ext4_ext_get_access(handle
, inode
, path
+ depth
);
2872 if (split
== ee_block
) {
2874 * case b: block @split is the block that the extent begins with
2875 * then we just change the state of the extent, and splitting
2878 if (split_flag
& EXT4_EXT_MARK_UNINIT2
)
2879 ext4_ext_mark_uninitialized(ex
);
2881 ext4_ext_mark_initialized(ex
);
2883 if (!(flags
& EXT4_GET_BLOCKS_PRE_IO
))
2884 ext4_ext_try_to_merge(inode
, path
, ex
);
2886 err
= ext4_ext_dirty(handle
, inode
, path
+ depth
);
2891 memcpy(&orig_ex
, ex
, sizeof(orig_ex
));
2892 ex
->ee_len
= cpu_to_le16(split
- ee_block
);
2893 if (split_flag
& EXT4_EXT_MARK_UNINIT1
)
2894 ext4_ext_mark_uninitialized(ex
);
2897 * path may lead to new leaf, not to original leaf any more
2898 * after ext4_ext_insert_extent() returns,
2900 err
= ext4_ext_dirty(handle
, inode
, path
+ depth
);
2902 goto fix_extent_len
;
2905 ex2
->ee_block
= cpu_to_le32(split
);
2906 ex2
->ee_len
= cpu_to_le16(ee_len
- (split
- ee_block
));
2907 ext4_ext_store_pblock(ex2
, newblock
);
2908 if (split_flag
& EXT4_EXT_MARK_UNINIT2
)
2909 ext4_ext_mark_uninitialized(ex2
);
2911 err
= ext4_ext_insert_extent(handle
, inode
, path
, &newex
, flags
);
2912 if (err
== -ENOSPC
&& (EXT4_EXT_MAY_ZEROOUT
& split_flag
)) {
2913 err
= ext4_ext_zeroout(inode
, &orig_ex
);
2915 goto fix_extent_len
;
2916 /* update the extent length and mark as initialized */
2917 ex
->ee_len
= cpu_to_le32(ee_len
);
2918 ext4_ext_try_to_merge(inode
, path
, ex
);
2919 err
= ext4_ext_dirty(handle
, inode
, path
+ depth
);
2922 goto fix_extent_len
;
2925 ext4_ext_show_leaf(inode
, path
);
2929 ex
->ee_len
= orig_ex
.ee_len
;
2930 ext4_ext_dirty(handle
, inode
, path
+ depth
);
2935 * ext4_split_extents() splits an extent and mark extent which is covered
2936 * by @map as split_flags indicates
2938 * It may result in splitting the extent into multiple extents (upto three)
2939 * There are three possibilities:
2940 * a> There is no split required
2941 * b> Splits in two extents: Split is happening at either end of the extent
2942 * c> Splits in three extents: Somone is splitting in middle of the extent
2945 static int ext4_split_extent(handle_t
*handle
,
2946 struct inode
*inode
,
2947 struct ext4_ext_path
*path
,
2948 struct ext4_map_blocks
*map
,
2952 ext4_lblk_t ee_block
;
2953 struct ext4_extent
*ex
;
2954 unsigned int ee_len
, depth
;
2957 int split_flag1
, flags1
;
2959 depth
= ext_depth(inode
);
2960 ex
= path
[depth
].p_ext
;
2961 ee_block
= le32_to_cpu(ex
->ee_block
);
2962 ee_len
= ext4_ext_get_actual_len(ex
);
2963 uninitialized
= ext4_ext_is_uninitialized(ex
);
2965 if (map
->m_lblk
+ map
->m_len
< ee_block
+ ee_len
) {
2966 split_flag1
= split_flag
& EXT4_EXT_MAY_ZEROOUT
?
2967 EXT4_EXT_MAY_ZEROOUT
: 0;
2968 flags1
= flags
| EXT4_GET_BLOCKS_PRE_IO
;
2970 split_flag1
|= EXT4_EXT_MARK_UNINIT1
|
2971 EXT4_EXT_MARK_UNINIT2
;
2972 err
= ext4_split_extent_at(handle
, inode
, path
,
2973 map
->m_lblk
+ map
->m_len
, split_flag1
, flags1
);
2978 ext4_ext_drop_refs(path
);
2979 path
= ext4_ext_find_extent(inode
, map
->m_lblk
, path
);
2981 return PTR_ERR(path
);
2983 if (map
->m_lblk
>= ee_block
) {
2984 split_flag1
= split_flag
& EXT4_EXT_MAY_ZEROOUT
?
2985 EXT4_EXT_MAY_ZEROOUT
: 0;
2987 split_flag1
|= EXT4_EXT_MARK_UNINIT1
;
2988 if (split_flag
& EXT4_EXT_MARK_UNINIT2
)
2989 split_flag1
|= EXT4_EXT_MARK_UNINIT2
;
2990 err
= ext4_split_extent_at(handle
, inode
, path
,
2991 map
->m_lblk
, split_flag1
, flags
);
2996 ext4_ext_show_leaf(inode
, path
);
2998 return err
? err
: map
->m_len
;
3001 #define EXT4_EXT_ZERO_LEN 7
3003 * This function is called by ext4_ext_map_blocks() if someone tries to write
3004 * to an uninitialized extent. It may result in splitting the uninitialized
3005 * extent into multiple extents (up to three - one initialized and two
3007 * There are three possibilities:
3008 * a> There is no split required: Entire extent should be initialized
3009 * b> Splits in two extents: Write is happening at either end of the extent
3010 * c> Splits in three extents: Somone is writing in middle of the extent
3012 static int ext4_ext_convert_to_initialized(handle_t
*handle
,
3013 struct inode
*inode
,
3014 struct ext4_map_blocks
*map
,
3015 struct ext4_ext_path
*path
)
3017 struct ext4_map_blocks split_map
;
3018 struct ext4_extent zero_ex
;
3019 struct ext4_extent
*ex
;
3020 ext4_lblk_t ee_block
, eof_block
;
3021 unsigned int allocated
, ee_len
, depth
;
3025 ext_debug("ext4_ext_convert_to_initialized: inode %lu, logical"
3026 "block %llu, max_blocks %u\n", inode
->i_ino
,
3027 (unsigned long long)map
->m_lblk
, map
->m_len
);
3029 eof_block
= (inode
->i_size
+ inode
->i_sb
->s_blocksize
- 1) >>
3030 inode
->i_sb
->s_blocksize_bits
;
3031 if (eof_block
< map
->m_lblk
+ map
->m_len
)
3032 eof_block
= map
->m_lblk
+ map
->m_len
;
3034 depth
= ext_depth(inode
);
3035 ex
= path
[depth
].p_ext
;
3036 ee_block
= le32_to_cpu(ex
->ee_block
);
3037 ee_len
= ext4_ext_get_actual_len(ex
);
3038 allocated
= ee_len
- (map
->m_lblk
- ee_block
);
3040 WARN_ON(map
->m_lblk
< ee_block
);
3042 * It is safe to convert extent to initialized via explicit
3043 * zeroout only if extent is fully insde i_size or new_size.
3045 split_flag
|= ee_block
+ ee_len
<= eof_block
? EXT4_EXT_MAY_ZEROOUT
: 0;
3047 /* If extent has less than 2*EXT4_EXT_ZERO_LEN zerout directly */
3048 if (ee_len
<= 2*EXT4_EXT_ZERO_LEN
&&
3049 (EXT4_EXT_MAY_ZEROOUT
& split_flag
)) {
3050 err
= ext4_ext_zeroout(inode
, ex
);
3054 err
= ext4_ext_get_access(handle
, inode
, path
+ depth
);
3057 ext4_ext_mark_initialized(ex
);
3058 ext4_ext_try_to_merge(inode
, path
, ex
);
3059 err
= ext4_ext_dirty(handle
, inode
, path
+ depth
);
3065 * 1. split the extent into three extents.
3066 * 2. split the extent into two extents, zeroout the first half.
3067 * 3. split the extent into two extents, zeroout the second half.
3068 * 4. split the extent into two extents with out zeroout.
3070 split_map
.m_lblk
= map
->m_lblk
;
3071 split_map
.m_len
= map
->m_len
;
3073 if (allocated
> map
->m_len
) {
3074 if (allocated
<= EXT4_EXT_ZERO_LEN
&&
3075 (EXT4_EXT_MAY_ZEROOUT
& split_flag
)) {
3078 cpu_to_le32(map
->m_lblk
);
3079 zero_ex
.ee_len
= cpu_to_le16(allocated
);
3080 ext4_ext_store_pblock(&zero_ex
,
3081 ext4_ext_pblock(ex
) + map
->m_lblk
- ee_block
);
3082 err
= ext4_ext_zeroout(inode
, &zero_ex
);
3085 split_map
.m_lblk
= map
->m_lblk
;
3086 split_map
.m_len
= allocated
;
3087 } else if ((map
->m_lblk
- ee_block
+ map
->m_len
<
3088 EXT4_EXT_ZERO_LEN
) &&
3089 (EXT4_EXT_MAY_ZEROOUT
& split_flag
)) {
3091 if (map
->m_lblk
!= ee_block
) {
3092 zero_ex
.ee_block
= ex
->ee_block
;
3093 zero_ex
.ee_len
= cpu_to_le16(map
->m_lblk
-
3095 ext4_ext_store_pblock(&zero_ex
,
3096 ext4_ext_pblock(ex
));
3097 err
= ext4_ext_zeroout(inode
, &zero_ex
);
3102 split_map
.m_lblk
= ee_block
;
3103 split_map
.m_len
= map
->m_lblk
- ee_block
+ map
->m_len
;
3104 allocated
= map
->m_len
;
3108 allocated
= ext4_split_extent(handle
, inode
, path
,
3109 &split_map
, split_flag
, 0);
3114 return err
? err
: allocated
;
3118 * This function is called by ext4_ext_map_blocks() from
3119 * ext4_get_blocks_dio_write() when DIO to write
3120 * to an uninitialized extent.
3122 * Writing to an uninitialized extent may result in splitting the uninitialized
3123 * extent into multiple /initialized uninitialized extents (up to three)
3124 * There are three possibilities:
3125 * a> There is no split required: Entire extent should be uninitialized
3126 * b> Splits in two extents: Write is happening at either end of the extent
3127 * c> Splits in three extents: Somone is writing in middle of the extent
3129 * One of more index blocks maybe needed if the extent tree grow after
3130 * the uninitialized extent split. To prevent ENOSPC occur at the IO
3131 * complete, we need to split the uninitialized extent before DIO submit
3132 * the IO. The uninitialized extent called at this time will be split
3133 * into three uninitialized extent(at most). After IO complete, the part
3134 * being filled will be convert to initialized by the end_io callback function
3135 * via ext4_convert_unwritten_extents().
3137 * Returns the size of uninitialized extent to be written on success.
3139 static int ext4_split_unwritten_extents(handle_t
*handle
,
3140 struct inode
*inode
,
3141 struct ext4_map_blocks
*map
,
3142 struct ext4_ext_path
*path
,
3145 ext4_lblk_t eof_block
;
3146 ext4_lblk_t ee_block
;
3147 struct ext4_extent
*ex
;
3148 unsigned int ee_len
;
3149 int split_flag
= 0, depth
;
3151 ext_debug("ext4_split_unwritten_extents: inode %lu, logical"
3152 "block %llu, max_blocks %u\n", inode
->i_ino
,
3153 (unsigned long long)map
->m_lblk
, map
->m_len
);
3155 eof_block
= (inode
->i_size
+ inode
->i_sb
->s_blocksize
- 1) >>
3156 inode
->i_sb
->s_blocksize_bits
;
3157 if (eof_block
< map
->m_lblk
+ map
->m_len
)
3158 eof_block
= map
->m_lblk
+ map
->m_len
;
3160 * It is safe to convert extent to initialized via explicit
3161 * zeroout only if extent is fully insde i_size or new_size.
3163 depth
= ext_depth(inode
);
3164 ex
= path
[depth
].p_ext
;
3165 ee_block
= le32_to_cpu(ex
->ee_block
);
3166 ee_len
= ext4_ext_get_actual_len(ex
);
3168 split_flag
|= ee_block
+ ee_len
<= eof_block
? EXT4_EXT_MAY_ZEROOUT
: 0;
3169 split_flag
|= EXT4_EXT_MARK_UNINIT2
;
3171 flags
|= EXT4_GET_BLOCKS_PRE_IO
;
3172 return ext4_split_extent(handle
, inode
, path
, map
, split_flag
, flags
);
3175 static int ext4_convert_unwritten_extents_endio(handle_t
*handle
,
3176 struct inode
*inode
,
3177 struct ext4_ext_path
*path
)
3179 struct ext4_extent
*ex
;
3183 depth
= ext_depth(inode
);
3184 ex
= path
[depth
].p_ext
;
3186 ext_debug("ext4_convert_unwritten_extents_endio: inode %lu, logical"
3187 "block %llu, max_blocks %u\n", inode
->i_ino
,
3188 (unsigned long long)le32_to_cpu(ex
->ee_block
),
3189 ext4_ext_get_actual_len(ex
));
3191 err
= ext4_ext_get_access(handle
, inode
, path
+ depth
);
3194 /* first mark the extent as initialized */
3195 ext4_ext_mark_initialized(ex
);
3197 /* note: ext4_ext_correct_indexes() isn't needed here because
3198 * borders are not changed
3200 ext4_ext_try_to_merge(inode
, path
, ex
);
3202 /* Mark modified extent as dirty */
3203 err
= ext4_ext_dirty(handle
, inode
, path
+ depth
);
3205 ext4_ext_show_leaf(inode
, path
);
3209 static void unmap_underlying_metadata_blocks(struct block_device
*bdev
,
3210 sector_t block
, int count
)
3213 for (i
= 0; i
< count
; i
++)
3214 unmap_underlying_metadata(bdev
, block
+ i
);
3218 * Handle EOFBLOCKS_FL flag, clearing it if necessary
3220 static int check_eofblocks_fl(handle_t
*handle
, struct inode
*inode
,
3222 struct ext4_ext_path
*path
,
3226 struct ext4_extent_header
*eh
;
3227 struct ext4_extent
*last_ex
;
3229 if (!ext4_test_inode_flag(inode
, EXT4_INODE_EOFBLOCKS
))
3232 depth
= ext_depth(inode
);
3233 eh
= path
[depth
].p_hdr
;
3235 if (unlikely(!eh
->eh_entries
)) {
3236 EXT4_ERROR_INODE(inode
, "eh->eh_entries == 0 and "
3237 "EOFBLOCKS_FL set");
3240 last_ex
= EXT_LAST_EXTENT(eh
);
3242 * We should clear the EOFBLOCKS_FL flag if we are writing the
3243 * last block in the last extent in the file. We test this by
3244 * first checking to see if the caller to
3245 * ext4_ext_get_blocks() was interested in the last block (or
3246 * a block beyond the last block) in the current extent. If
3247 * this turns out to be false, we can bail out from this
3248 * function immediately.
3250 if (lblk
+ len
< le32_to_cpu(last_ex
->ee_block
) +
3251 ext4_ext_get_actual_len(last_ex
))
3254 * If the caller does appear to be planning to write at or
3255 * beyond the end of the current extent, we then test to see
3256 * if the current extent is the last extent in the file, by
3257 * checking to make sure it was reached via the rightmost node
3258 * at each level of the tree.
3260 for (i
= depth
-1; i
>= 0; i
--)
3261 if (path
[i
].p_idx
!= EXT_LAST_INDEX(path
[i
].p_hdr
))
3263 ext4_clear_inode_flag(inode
, EXT4_INODE_EOFBLOCKS
);
3264 return ext4_mark_inode_dirty(handle
, inode
);
3268 * ext4_find_delalloc_range: find delayed allocated block in the given range.
3270 * Goes through the buffer heads in the range [lblk_start, lblk_end] and returns
3271 * whether there are any buffers marked for delayed allocation. It returns '1'
3272 * on the first delalloc'ed buffer head found. If no buffer head in the given
3273 * range is marked for delalloc, it returns 0.
3274 * lblk_start should always be <= lblk_end.
3275 * search_hint_reverse is to indicate that searching in reverse from lblk_end to
3276 * lblk_start might be more efficient (i.e., we will likely hit the delalloc'ed
3277 * block sooner). This is useful when blocks are truncated sequentially from
3278 * lblk_start towards lblk_end.
3280 static int ext4_find_delalloc_range(struct inode
*inode
,
3281 ext4_lblk_t lblk_start
,
3282 ext4_lblk_t lblk_end
,
3283 int search_hint_reverse
)
3285 struct address_space
*mapping
= inode
->i_mapping
;
3286 struct buffer_head
*head
, *bh
= NULL
;
3288 ext4_lblk_t i
, pg_lblk
;
3291 /* reverse search wont work if fs block size is less than page size */
3292 if (inode
->i_blkbits
< PAGE_CACHE_SHIFT
)
3293 search_hint_reverse
= 0;
3295 if (search_hint_reverse
)
3300 index
= i
>> (PAGE_CACHE_SHIFT
- inode
->i_blkbits
);
3302 while ((i
>= lblk_start
) && (i
<= lblk_end
)) {
3303 page
= find_get_page(mapping
, index
);
3307 if (!page_has_buffers(page
))
3310 head
= page_buffers(page
);
3315 pg_lblk
= index
<< (PAGE_CACHE_SHIFT
-
3318 if (unlikely(pg_lblk
< lblk_start
)) {
3320 * This is possible when fs block size is less
3321 * than page size and our cluster starts/ends in
3322 * middle of the page. So we need to skip the
3323 * initial few blocks till we reach the 'lblk'
3329 /* Check if the buffer is delayed allocated and that it
3330 * is not yet mapped. (when da-buffers are mapped during
3331 * their writeout, their da_mapped bit is set.)
3333 if (buffer_delay(bh
) && !buffer_da_mapped(bh
)) {
3334 page_cache_release(page
);
3335 trace_ext4_find_delalloc_range(inode
,
3336 lblk_start
, lblk_end
,
3337 search_hint_reverse
,
3341 if (search_hint_reverse
)
3345 } while ((i
>= lblk_start
) && (i
<= lblk_end
) &&
3346 ((bh
= bh
->b_this_page
) != head
));
3349 page_cache_release(page
);
3351 * Move to next page. 'i' will be the first lblk in the next
3354 if (search_hint_reverse
)
3358 i
= index
<< (PAGE_CACHE_SHIFT
- inode
->i_blkbits
);
3361 trace_ext4_find_delalloc_range(inode
, lblk_start
, lblk_end
,
3362 search_hint_reverse
, 0, 0);
3366 int ext4_find_delalloc_cluster(struct inode
*inode
, ext4_lblk_t lblk
,
3367 int search_hint_reverse
)
3369 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
3370 ext4_lblk_t lblk_start
, lblk_end
;
3371 lblk_start
= lblk
& (~(sbi
->s_cluster_ratio
- 1));
3372 lblk_end
= lblk_start
+ sbi
->s_cluster_ratio
- 1;
3374 return ext4_find_delalloc_range(inode
, lblk_start
, lblk_end
,
3375 search_hint_reverse
);
3379 * Determines how many complete clusters (out of those specified by the 'map')
3380 * are under delalloc and were reserved quota for.
3381 * This function is called when we are writing out the blocks that were
3382 * originally written with their allocation delayed, but then the space was
3383 * allocated using fallocate() before the delayed allocation could be resolved.
3384 * The cases to look for are:
3385 * ('=' indicated delayed allocated blocks
3386 * '-' indicates non-delayed allocated blocks)
3387 * (a) partial clusters towards beginning and/or end outside of allocated range
3388 * are not delalloc'ed.
3390 * |----c---=|====c====|====c====|===-c----|
3391 * |++++++ allocated ++++++|
3392 * ==> 4 complete clusters in above example
3394 * (b) partial cluster (outside of allocated range) towards either end is
3395 * marked for delayed allocation. In this case, we will exclude that
3398 * |----====c========|========c========|
3399 * |++++++ allocated ++++++|
3400 * ==> 1 complete clusters in above example
3403 * |================c================|
3404 * |++++++ allocated ++++++|
3405 * ==> 0 complete clusters in above example
3407 * The ext4_da_update_reserve_space will be called only if we
3408 * determine here that there were some "entire" clusters that span
3409 * this 'allocated' range.
3410 * In the non-bigalloc case, this function will just end up returning num_blks
3411 * without ever calling ext4_find_delalloc_range.
3414 get_reserved_cluster_alloc(struct inode
*inode
, ext4_lblk_t lblk_start
,
3415 unsigned int num_blks
)
3417 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
3418 ext4_lblk_t alloc_cluster_start
, alloc_cluster_end
;
3419 ext4_lblk_t lblk_from
, lblk_to
, c_offset
;
3420 unsigned int allocated_clusters
= 0;
3422 alloc_cluster_start
= EXT4_B2C(sbi
, lblk_start
);
3423 alloc_cluster_end
= EXT4_B2C(sbi
, lblk_start
+ num_blks
- 1);
3425 /* max possible clusters for this allocation */
3426 allocated_clusters
= alloc_cluster_end
- alloc_cluster_start
+ 1;
3428 trace_ext4_get_reserved_cluster_alloc(inode
, lblk_start
, num_blks
);
3430 /* Check towards left side */
3431 c_offset
= lblk_start
& (sbi
->s_cluster_ratio
- 1);
3433 lblk_from
= lblk_start
& (~(sbi
->s_cluster_ratio
- 1));
3434 lblk_to
= lblk_from
+ c_offset
- 1;
3436 if (ext4_find_delalloc_range(inode
, lblk_from
, lblk_to
, 0))
3437 allocated_clusters
--;
3440 /* Now check towards right. */
3441 c_offset
= (lblk_start
+ num_blks
) & (sbi
->s_cluster_ratio
- 1);
3442 if (allocated_clusters
&& c_offset
) {
3443 lblk_from
= lblk_start
+ num_blks
;
3444 lblk_to
= lblk_from
+ (sbi
->s_cluster_ratio
- c_offset
) - 1;
3446 if (ext4_find_delalloc_range(inode
, lblk_from
, lblk_to
, 0))
3447 allocated_clusters
--;
3450 return allocated_clusters
;
3454 ext4_ext_handle_uninitialized_extents(handle_t
*handle
, struct inode
*inode
,
3455 struct ext4_map_blocks
*map
,
3456 struct ext4_ext_path
*path
, int flags
,
3457 unsigned int allocated
, ext4_fsblk_t newblock
)
3461 ext4_io_end_t
*io
= EXT4_I(inode
)->cur_aio_dio
;
3463 ext_debug("ext4_ext_handle_uninitialized_extents: inode %lu, logical"
3464 "block %llu, max_blocks %u, flags %d, allocated %u",
3465 inode
->i_ino
, (unsigned long long)map
->m_lblk
, map
->m_len
,
3467 ext4_ext_show_leaf(inode
, path
);
3469 trace_ext4_ext_handle_uninitialized_extents(inode
, map
, allocated
,
3472 /* get_block() before submit the IO, split the extent */
3473 if ((flags
& EXT4_GET_BLOCKS_PRE_IO
)) {
3474 ret
= ext4_split_unwritten_extents(handle
, inode
, map
,
3477 * Flag the inode(non aio case) or end_io struct (aio case)
3478 * that this IO needs to conversion to written when IO is
3481 if (io
&& !(io
->flag
& EXT4_IO_END_UNWRITTEN
)) {
3482 io
->flag
= EXT4_IO_END_UNWRITTEN
;
3483 atomic_inc(&EXT4_I(inode
)->i_aiodio_unwritten
);
3485 ext4_set_inode_state(inode
, EXT4_STATE_DIO_UNWRITTEN
);
3486 if (ext4_should_dioread_nolock(inode
))
3487 map
->m_flags
|= EXT4_MAP_UNINIT
;
3490 /* IO end_io complete, convert the filled extent to written */
3491 if ((flags
& EXT4_GET_BLOCKS_CONVERT
)) {
3492 ret
= ext4_convert_unwritten_extents_endio(handle
, inode
,
3495 ext4_update_inode_fsync_trans(handle
, inode
, 1);
3496 err
= check_eofblocks_fl(handle
, inode
, map
->m_lblk
,
3502 /* buffered IO case */
3504 * repeat fallocate creation request
3505 * we already have an unwritten extent
3507 if (flags
& EXT4_GET_BLOCKS_UNINIT_EXT
)
3510 /* buffered READ or buffered write_begin() lookup */
3511 if ((flags
& EXT4_GET_BLOCKS_CREATE
) == 0) {
3513 * We have blocks reserved already. We
3514 * return allocated blocks so that delalloc
3515 * won't do block reservation for us. But
3516 * the buffer head will be unmapped so that
3517 * a read from the block returns 0s.
3519 map
->m_flags
|= EXT4_MAP_UNWRITTEN
;
3523 /* buffered write, writepage time, convert*/
3524 ret
= ext4_ext_convert_to_initialized(handle
, inode
, map
, path
);
3526 ext4_update_inode_fsync_trans(handle
, inode
, 1);
3527 err
= check_eofblocks_fl(handle
, inode
, map
->m_lblk
, path
,
3539 map
->m_flags
|= EXT4_MAP_NEW
;
3541 * if we allocated more blocks than requested
3542 * we need to make sure we unmap the extra block
3543 * allocated. The actual needed block will get
3544 * unmapped later when we find the buffer_head marked
3547 if (allocated
> map
->m_len
) {
3548 unmap_underlying_metadata_blocks(inode
->i_sb
->s_bdev
,
3549 newblock
+ map
->m_len
,
3550 allocated
- map
->m_len
);
3551 allocated
= map
->m_len
;
3555 * If we have done fallocate with the offset that is already
3556 * delayed allocated, we would have block reservation
3557 * and quota reservation done in the delayed write path.
3558 * But fallocate would have already updated quota and block
3559 * count for this offset. So cancel these reservation
3561 if (flags
& EXT4_GET_BLOCKS_DELALLOC_RESERVE
) {
3562 unsigned int reserved_clusters
;
3563 reserved_clusters
= get_reserved_cluster_alloc(inode
,
3564 map
->m_lblk
, map
->m_len
);
3565 if (reserved_clusters
)
3566 ext4_da_update_reserve_space(inode
,
3572 map
->m_flags
|= EXT4_MAP_MAPPED
;
3574 if (allocated
> map
->m_len
)
3575 allocated
= map
->m_len
;
3576 ext4_ext_show_leaf(inode
, path
);
3577 map
->m_pblk
= newblock
;
3578 map
->m_len
= allocated
;
3581 ext4_ext_drop_refs(path
);
3584 return err
? err
: allocated
;
3588 * get_implied_cluster_alloc - check to see if the requested
3589 * allocation (in the map structure) overlaps with a cluster already
3590 * allocated in an extent.
3591 * @sb The filesystem superblock structure
3592 * @map The requested lblk->pblk mapping
3593 * @ex The extent structure which might contain an implied
3594 * cluster allocation
3596 * This function is called by ext4_ext_map_blocks() after we failed to
3597 * find blocks that were already in the inode's extent tree. Hence,
3598 * we know that the beginning of the requested region cannot overlap
3599 * the extent from the inode's extent tree. There are three cases we
3600 * want to catch. The first is this case:
3602 * |--- cluster # N--|
3603 * |--- extent ---| |---- requested region ---|
3606 * The second case that we need to test for is this one:
3608 * |--------- cluster # N ----------------|
3609 * |--- requested region --| |------- extent ----|
3610 * |=======================|
3612 * The third case is when the requested region lies between two extents
3613 * within the same cluster:
3614 * |------------- cluster # N-------------|
3615 * |----- ex -----| |---- ex_right ----|
3616 * |------ requested region ------|
3617 * |================|
3619 * In each of the above cases, we need to set the map->m_pblk and
3620 * map->m_len so it corresponds to the return the extent labelled as
3621 * "|====|" from cluster #N, since it is already in use for data in
3622 * cluster EXT4_B2C(sbi, map->m_lblk). We will then return 1 to
3623 * signal to ext4_ext_map_blocks() that map->m_pblk should be treated
3624 * as a new "allocated" block region. Otherwise, we will return 0 and
3625 * ext4_ext_map_blocks() will then allocate one or more new clusters
3626 * by calling ext4_mb_new_blocks().
3628 static int get_implied_cluster_alloc(struct super_block
*sb
,
3629 struct ext4_map_blocks
*map
,
3630 struct ext4_extent
*ex
,
3631 struct ext4_ext_path
*path
)
3633 struct ext4_sb_info
*sbi
= EXT4_SB(sb
);
3634 ext4_lblk_t c_offset
= map
->m_lblk
& (sbi
->s_cluster_ratio
-1);
3635 ext4_lblk_t ex_cluster_start
, ex_cluster_end
;
3636 ext4_lblk_t rr_cluster_start
, rr_cluster_end
;
3637 ext4_lblk_t ee_block
= le32_to_cpu(ex
->ee_block
);
3638 ext4_fsblk_t ee_start
= ext4_ext_pblock(ex
);
3639 unsigned short ee_len
= ext4_ext_get_actual_len(ex
);
3641 /* The extent passed in that we are trying to match */
3642 ex_cluster_start
= EXT4_B2C(sbi
, ee_block
);
3643 ex_cluster_end
= EXT4_B2C(sbi
, ee_block
+ ee_len
- 1);
3645 /* The requested region passed into ext4_map_blocks() */
3646 rr_cluster_start
= EXT4_B2C(sbi
, map
->m_lblk
);
3647 rr_cluster_end
= EXT4_B2C(sbi
, map
->m_lblk
+ map
->m_len
- 1);
3649 if ((rr_cluster_start
== ex_cluster_end
) ||
3650 (rr_cluster_start
== ex_cluster_start
)) {
3651 if (rr_cluster_start
== ex_cluster_end
)
3652 ee_start
+= ee_len
- 1;
3653 map
->m_pblk
= (ee_start
& ~(sbi
->s_cluster_ratio
- 1)) +
3655 map
->m_len
= min(map
->m_len
,
3656 (unsigned) sbi
->s_cluster_ratio
- c_offset
);
3658 * Check for and handle this case:
3660 * |--------- cluster # N-------------|
3661 * |------- extent ----|
3662 * |--- requested region ---|
3666 if (map
->m_lblk
< ee_block
)
3667 map
->m_len
= min(map
->m_len
, ee_block
- map
->m_lblk
);
3670 * Check for the case where there is already another allocated
3671 * block to the right of 'ex' but before the end of the cluster.
3673 * |------------- cluster # N-------------|
3674 * |----- ex -----| |---- ex_right ----|
3675 * |------ requested region ------|
3676 * |================|
3678 if (map
->m_lblk
> ee_block
) {
3679 ext4_lblk_t next
= ext4_ext_next_allocated_block(path
);
3680 map
->m_len
= min(map
->m_len
, next
- map
->m_lblk
);
3683 trace_ext4_get_implied_cluster_alloc_exit(sb
, map
, 1);
3687 trace_ext4_get_implied_cluster_alloc_exit(sb
, map
, 0);
3693 * Block allocation/map/preallocation routine for extents based files
3696 * Need to be called with
3697 * down_read(&EXT4_I(inode)->i_data_sem) if not allocating file system block
3698 * (ie, create is zero). Otherwise down_write(&EXT4_I(inode)->i_data_sem)
3700 * return > 0, number of of blocks already mapped/allocated
3701 * if create == 0 and these are pre-allocated blocks
3702 * buffer head is unmapped
3703 * otherwise blocks are mapped
3705 * return = 0, if plain look up failed (blocks have not been allocated)
3706 * buffer head is unmapped
3708 * return < 0, error case.
3710 int ext4_ext_map_blocks(handle_t
*handle
, struct inode
*inode
,
3711 struct ext4_map_blocks
*map
, int flags
)
3713 struct ext4_ext_path
*path
= NULL
;
3714 struct ext4_extent newex
, *ex
, *ex2
;
3715 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
3716 ext4_fsblk_t newblock
= 0;
3717 int free_on_err
= 0, err
= 0, depth
, ret
;
3718 unsigned int allocated
= 0, offset
= 0;
3719 unsigned int allocated_clusters
= 0, reserved_clusters
= 0;
3720 unsigned int punched_out
= 0;
3721 unsigned int result
= 0;
3722 struct ext4_allocation_request ar
;
3723 ext4_io_end_t
*io
= EXT4_I(inode
)->cur_aio_dio
;
3724 ext4_lblk_t cluster_offset
;
3725 struct ext4_map_blocks punch_map
;
3727 ext_debug("blocks %u/%u requested for inode %lu\n",
3728 map
->m_lblk
, map
->m_len
, inode
->i_ino
);
3729 trace_ext4_ext_map_blocks_enter(inode
, map
->m_lblk
, map
->m_len
, flags
);
3731 /* check in cache */
3732 if (!(flags
& EXT4_GET_BLOCKS_PUNCH_OUT_EXT
) &&
3733 ext4_ext_in_cache(inode
, map
->m_lblk
, &newex
)) {
3734 if (!newex
.ee_start_lo
&& !newex
.ee_start_hi
) {
3735 if ((sbi
->s_cluster_ratio
> 1) &&
3736 ext4_find_delalloc_cluster(inode
, map
->m_lblk
, 0))
3737 map
->m_flags
|= EXT4_MAP_FROM_CLUSTER
;
3739 if ((flags
& EXT4_GET_BLOCKS_CREATE
) == 0) {
3741 * block isn't allocated yet and
3742 * user doesn't want to allocate it
3746 /* we should allocate requested block */
3748 /* block is already allocated */
3749 if (sbi
->s_cluster_ratio
> 1)
3750 map
->m_flags
|= EXT4_MAP_FROM_CLUSTER
;
3751 newblock
= map
->m_lblk
3752 - le32_to_cpu(newex
.ee_block
)
3753 + ext4_ext_pblock(&newex
);
3754 /* number of remaining blocks in the extent */
3755 allocated
= ext4_ext_get_actual_len(&newex
) -
3756 (map
->m_lblk
- le32_to_cpu(newex
.ee_block
));
3761 /* find extent for this block */
3762 path
= ext4_ext_find_extent(inode
, map
->m_lblk
, NULL
);
3764 err
= PTR_ERR(path
);
3769 depth
= ext_depth(inode
);
3772 * consistent leaf must not be empty;
3773 * this situation is possible, though, _during_ tree modification;
3774 * this is why assert can't be put in ext4_ext_find_extent()
3776 if (unlikely(path
[depth
].p_ext
== NULL
&& depth
!= 0)) {
3777 EXT4_ERROR_INODE(inode
, "bad extent address "
3778 "lblock: %lu, depth: %d pblock %lld",
3779 (unsigned long) map
->m_lblk
, depth
,
3780 path
[depth
].p_block
);
3785 ex
= path
[depth
].p_ext
;
3787 ext4_lblk_t ee_block
= le32_to_cpu(ex
->ee_block
);
3788 ext4_fsblk_t ee_start
= ext4_ext_pblock(ex
);
3789 unsigned short ee_len
;
3792 * Uninitialized extents are treated as holes, except that
3793 * we split out initialized portions during a write.
3795 ee_len
= ext4_ext_get_actual_len(ex
);
3797 trace_ext4_ext_show_extent(inode
, ee_block
, ee_start
, ee_len
);
3799 /* if found extent covers block, simply return it */
3800 if (in_range(map
->m_lblk
, ee_block
, ee_len
)) {
3801 ext4_fsblk_t partial_cluster
= 0;
3803 newblock
= map
->m_lblk
- ee_block
+ ee_start
;
3804 /* number of remaining blocks in the extent */
3805 allocated
= ee_len
- (map
->m_lblk
- ee_block
);
3806 ext_debug("%u fit into %u:%d -> %llu\n", map
->m_lblk
,
3807 ee_block
, ee_len
, newblock
);
3809 if ((flags
& EXT4_GET_BLOCKS_PUNCH_OUT_EXT
) == 0) {
3811 * Do not put uninitialized extent
3814 if (!ext4_ext_is_uninitialized(ex
)) {
3815 ext4_ext_put_in_cache(inode
, ee_block
,
3819 ret
= ext4_ext_handle_uninitialized_extents(
3820 handle
, inode
, map
, path
, flags
,
3821 allocated
, newblock
);
3826 * Punch out the map length, but only to the
3829 punched_out
= allocated
< map
->m_len
?
3830 allocated
: map
->m_len
;
3833 * Sense extents need to be converted to
3834 * uninitialized, they must fit in an
3835 * uninitialized extent
3837 if (punched_out
> EXT_UNINIT_MAX_LEN
)
3838 punched_out
= EXT_UNINIT_MAX_LEN
;
3840 punch_map
.m_lblk
= map
->m_lblk
;
3841 punch_map
.m_pblk
= newblock
;
3842 punch_map
.m_len
= punched_out
;
3843 punch_map
.m_flags
= 0;
3845 /* Check to see if the extent needs to be split */
3846 if (punch_map
.m_len
!= ee_len
||
3847 punch_map
.m_lblk
!= ee_block
) {
3849 ret
= ext4_split_extent(handle
, inode
,
3850 path
, &punch_map
, 0,
3851 EXT4_GET_BLOCKS_PUNCH_OUT_EXT
|
3852 EXT4_GET_BLOCKS_PRE_IO
);
3859 * find extent for the block at
3860 * the start of the hole
3862 ext4_ext_drop_refs(path
);
3865 path
= ext4_ext_find_extent(inode
,
3868 err
= PTR_ERR(path
);
3873 depth
= ext_depth(inode
);
3874 ex
= path
[depth
].p_ext
;
3875 ee_len
= ext4_ext_get_actual_len(ex
);
3876 ee_block
= le32_to_cpu(ex
->ee_block
);
3877 ee_start
= ext4_ext_pblock(ex
);
3881 ext4_ext_mark_uninitialized(ex
);
3883 ext4_ext_invalidate_cache(inode
);
3885 err
= ext4_ext_rm_leaf(handle
, inode
, path
,
3886 &partial_cluster
, map
->m_lblk
,
3887 map
->m_lblk
+ punched_out
);
3889 if (!err
&& path
->p_hdr
->eh_entries
== 0) {
3891 * Punch hole freed all of this sub tree,
3892 * so we need to correct eh_depth
3894 err
= ext4_ext_get_access(handle
, inode
, path
);
3896 ext_inode_hdr(inode
)->eh_depth
= 0;
3897 ext_inode_hdr(inode
)->eh_max
=
3898 cpu_to_le16(ext4_ext_space_root(
3901 err
= ext4_ext_dirty(
3902 handle
, inode
, path
);
3910 if ((sbi
->s_cluster_ratio
> 1) &&
3911 ext4_find_delalloc_cluster(inode
, map
->m_lblk
, 0))
3912 map
->m_flags
|= EXT4_MAP_FROM_CLUSTER
;
3915 * requested block isn't allocated yet;
3916 * we couldn't try to create block if create flag is zero
3918 if ((flags
& EXT4_GET_BLOCKS_CREATE
) == 0) {
3920 * put just found gap into cache to speed up
3921 * subsequent requests
3923 ext4_ext_put_gap_in_cache(inode
, path
, map
->m_lblk
);
3928 * Okay, we need to do block allocation.
3930 map
->m_flags
&= ~EXT4_MAP_FROM_CLUSTER
;
3931 newex
.ee_block
= cpu_to_le32(map
->m_lblk
);
3932 cluster_offset
= map
->m_lblk
& (sbi
->s_cluster_ratio
-1);
3935 * If we are doing bigalloc, check to see if the extent returned
3936 * by ext4_ext_find_extent() implies a cluster we can use.
3938 if (cluster_offset
&& ex
&&
3939 get_implied_cluster_alloc(inode
->i_sb
, map
, ex
, path
)) {
3940 ar
.len
= allocated
= map
->m_len
;
3941 newblock
= map
->m_pblk
;
3942 map
->m_flags
|= EXT4_MAP_FROM_CLUSTER
;
3943 goto got_allocated_blocks
;
3946 /* find neighbour allocated blocks */
3947 ar
.lleft
= map
->m_lblk
;
3948 err
= ext4_ext_search_left(inode
, path
, &ar
.lleft
, &ar
.pleft
);
3951 ar
.lright
= map
->m_lblk
;
3953 err
= ext4_ext_search_right(inode
, path
, &ar
.lright
, &ar
.pright
, &ex2
);
3957 /* Check if the extent after searching to the right implies a
3958 * cluster we can use. */
3959 if ((sbi
->s_cluster_ratio
> 1) && ex2
&&
3960 get_implied_cluster_alloc(inode
->i_sb
, map
, ex2
, path
)) {
3961 ar
.len
= allocated
= map
->m_len
;
3962 newblock
= map
->m_pblk
;
3963 map
->m_flags
|= EXT4_MAP_FROM_CLUSTER
;
3964 goto got_allocated_blocks
;
3968 * See if request is beyond maximum number of blocks we can have in
3969 * a single extent. For an initialized extent this limit is
3970 * EXT_INIT_MAX_LEN and for an uninitialized extent this limit is
3971 * EXT_UNINIT_MAX_LEN.
3973 if (map
->m_len
> EXT_INIT_MAX_LEN
&&
3974 !(flags
& EXT4_GET_BLOCKS_UNINIT_EXT
))
3975 map
->m_len
= EXT_INIT_MAX_LEN
;
3976 else if (map
->m_len
> EXT_UNINIT_MAX_LEN
&&
3977 (flags
& EXT4_GET_BLOCKS_UNINIT_EXT
))
3978 map
->m_len
= EXT_UNINIT_MAX_LEN
;
3980 /* Check if we can really insert (m_lblk)::(m_lblk + m_len) extent */
3981 newex
.ee_len
= cpu_to_le16(map
->m_len
);
3982 err
= ext4_ext_check_overlap(sbi
, inode
, &newex
, path
);
3984 allocated
= ext4_ext_get_actual_len(&newex
);
3986 allocated
= map
->m_len
;
3988 /* allocate new block */
3990 ar
.goal
= ext4_ext_find_goal(inode
, path
, map
->m_lblk
);
3991 ar
.logical
= map
->m_lblk
;
3993 * We calculate the offset from the beginning of the cluster
3994 * for the logical block number, since when we allocate a
3995 * physical cluster, the physical block should start at the
3996 * same offset from the beginning of the cluster. This is
3997 * needed so that future calls to get_implied_cluster_alloc()
4000 offset
= map
->m_lblk
& (sbi
->s_cluster_ratio
- 1);
4001 ar
.len
= EXT4_NUM_B2C(sbi
, offset
+allocated
);
4003 ar
.logical
-= offset
;
4004 if (S_ISREG(inode
->i_mode
))
4005 ar
.flags
= EXT4_MB_HINT_DATA
;
4007 /* disable in-core preallocation for non-regular files */
4009 if (flags
& EXT4_GET_BLOCKS_NO_NORMALIZE
)
4010 ar
.flags
|= EXT4_MB_HINT_NOPREALLOC
;
4011 newblock
= ext4_mb_new_blocks(handle
, &ar
, &err
);
4014 ext_debug("allocate new block: goal %llu, found %llu/%u\n",
4015 ar
.goal
, newblock
, allocated
);
4017 allocated_clusters
= ar
.len
;
4018 ar
.len
= EXT4_C2B(sbi
, ar
.len
) - offset
;
4019 if (ar
.len
> allocated
)
4022 got_allocated_blocks
:
4023 /* try to insert new extent into found leaf and return */
4024 ext4_ext_store_pblock(&newex
, newblock
+ offset
);
4025 newex
.ee_len
= cpu_to_le16(ar
.len
);
4026 /* Mark uninitialized */
4027 if (flags
& EXT4_GET_BLOCKS_UNINIT_EXT
){
4028 ext4_ext_mark_uninitialized(&newex
);
4030 * io_end structure was created for every IO write to an
4031 * uninitialized extent. To avoid unnecessary conversion,
4032 * here we flag the IO that really needs the conversion.
4033 * For non asycn direct IO case, flag the inode state
4034 * that we need to perform conversion when IO is done.
4036 if ((flags
& EXT4_GET_BLOCKS_PRE_IO
)) {
4037 if (io
&& !(io
->flag
& EXT4_IO_END_UNWRITTEN
)) {
4038 io
->flag
= EXT4_IO_END_UNWRITTEN
;
4039 atomic_inc(&EXT4_I(inode
)->i_aiodio_unwritten
);
4041 ext4_set_inode_state(inode
,
4042 EXT4_STATE_DIO_UNWRITTEN
);
4044 if (ext4_should_dioread_nolock(inode
))
4045 map
->m_flags
|= EXT4_MAP_UNINIT
;
4048 err
= check_eofblocks_fl(handle
, inode
, map
->m_lblk
, path
, ar
.len
);
4050 err
= ext4_ext_insert_extent(handle
, inode
, path
,
4052 if (err
&& free_on_err
) {
4053 int fb_flags
= flags
& EXT4_GET_BLOCKS_DELALLOC_RESERVE
?
4054 EXT4_FREE_BLOCKS_NO_QUOT_UPDATE
: 0;
4055 /* free data blocks we just allocated */
4056 /* not a good idea to call discard here directly,
4057 * but otherwise we'd need to call it every free() */
4058 ext4_discard_preallocations(inode
);
4059 ext4_free_blocks(handle
, inode
, NULL
, ext4_ext_pblock(&newex
),
4060 ext4_ext_get_actual_len(&newex
), fb_flags
);
4064 /* previous routine could use block we allocated */
4065 newblock
= ext4_ext_pblock(&newex
);
4066 allocated
= ext4_ext_get_actual_len(&newex
);
4067 if (allocated
> map
->m_len
)
4068 allocated
= map
->m_len
;
4069 map
->m_flags
|= EXT4_MAP_NEW
;
4072 * Update reserved blocks/metadata blocks after successful
4073 * block allocation which had been deferred till now.
4075 if (flags
& EXT4_GET_BLOCKS_DELALLOC_RESERVE
) {
4077 * Check how many clusters we had reserved this allocted range.
4079 reserved_clusters
= get_reserved_cluster_alloc(inode
,
4080 map
->m_lblk
, allocated
);
4081 if (map
->m_flags
& EXT4_MAP_FROM_CLUSTER
) {
4082 if (reserved_clusters
) {
4084 * We have clusters reserved for this range.
4085 * But since we are not doing actual allocation
4086 * and are simply using blocks from previously
4087 * allocated cluster, we should release the
4088 * reservation and not claim quota.
4090 ext4_da_update_reserve_space(inode
,
4091 reserved_clusters
, 0);
4094 BUG_ON(allocated_clusters
< reserved_clusters
);
4095 /* We will claim quota for all newly allocated blocks.*/
4096 ext4_da_update_reserve_space(inode
, allocated_clusters
,
4098 if (reserved_clusters
< allocated_clusters
) {
4099 struct ext4_inode_info
*ei
= EXT4_I(inode
);
4100 int reservation
= allocated_clusters
-
4103 * It seems we claimed few clusters outside of
4104 * the range of this allocation. We should give
4105 * it back to the reservation pool. This can
4106 * happen in the following case:
4108 * * Suppose s_cluster_ratio is 4 (i.e., each
4109 * cluster has 4 blocks. Thus, the clusters
4110 * are [0-3],[4-7],[8-11]...
4111 * * First comes delayed allocation write for
4112 * logical blocks 10 & 11. Since there were no
4113 * previous delayed allocated blocks in the
4114 * range [8-11], we would reserve 1 cluster
4116 * * Next comes write for logical blocks 3 to 8.
4117 * In this case, we will reserve 2 clusters
4118 * (for [0-3] and [4-7]; and not for [8-11] as
4119 * that range has a delayed allocated blocks.
4120 * Thus total reserved clusters now becomes 3.
4121 * * Now, during the delayed allocation writeout
4122 * time, we will first write blocks [3-8] and
4123 * allocate 3 clusters for writing these
4124 * blocks. Also, we would claim all these
4125 * three clusters above.
4126 * * Now when we come here to writeout the
4127 * blocks [10-11], we would expect to claim
4128 * the reservation of 1 cluster we had made
4129 * (and we would claim it since there are no
4130 * more delayed allocated blocks in the range
4131 * [8-11]. But our reserved cluster count had
4132 * already gone to 0.
4134 * Thus, at the step 4 above when we determine
4135 * that there are still some unwritten delayed
4136 * allocated blocks outside of our current
4137 * block range, we should increment the
4138 * reserved clusters count so that when the
4139 * remaining blocks finally gets written, we
4142 dquot_reserve_block(inode
,
4143 EXT4_C2B(sbi
, reservation
));
4144 spin_lock(&ei
->i_block_reservation_lock
);
4145 ei
->i_reserved_data_blocks
+= reservation
;
4146 spin_unlock(&ei
->i_block_reservation_lock
);
4152 * Cache the extent and update transaction to commit on fdatasync only
4153 * when it is _not_ an uninitialized extent.
4155 if ((flags
& EXT4_GET_BLOCKS_UNINIT_EXT
) == 0) {
4156 ext4_ext_put_in_cache(inode
, map
->m_lblk
, allocated
, newblock
);
4157 ext4_update_inode_fsync_trans(handle
, inode
, 1);
4159 ext4_update_inode_fsync_trans(handle
, inode
, 0);
4161 if (allocated
> map
->m_len
)
4162 allocated
= map
->m_len
;
4163 ext4_ext_show_leaf(inode
, path
);
4164 map
->m_flags
|= EXT4_MAP_MAPPED
;
4165 map
->m_pblk
= newblock
;
4166 map
->m_len
= allocated
;
4169 ext4_ext_drop_refs(path
);
4172 trace_ext4_ext_map_blocks_exit(inode
, map
->m_lblk
,
4173 newblock
, map
->m_len
, err
? err
: allocated
);
4175 result
= (flags
& EXT4_GET_BLOCKS_PUNCH_OUT_EXT
) ?
4176 punched_out
: allocated
;
4178 return err
? err
: result
;
4181 void ext4_ext_truncate(struct inode
*inode
)
4183 struct address_space
*mapping
= inode
->i_mapping
;
4184 struct super_block
*sb
= inode
->i_sb
;
4185 ext4_lblk_t last_block
;
4191 * finish any pending end_io work so we won't run the risk of
4192 * converting any truncated blocks to initialized later
4194 ext4_flush_completed_IO(inode
);
4197 * probably first extent we're gonna free will be last in block
4199 err
= ext4_writepage_trans_blocks(inode
);
4200 handle
= ext4_journal_start(inode
, err
);
4204 if (inode
->i_size
% PAGE_CACHE_SIZE
!= 0) {
4205 page_len
= PAGE_CACHE_SIZE
-
4206 (inode
->i_size
& (PAGE_CACHE_SIZE
- 1));
4208 err
= ext4_discard_partial_page_buffers(handle
,
4209 mapping
, inode
->i_size
, page_len
, 0);
4215 if (ext4_orphan_add(handle
, inode
))
4218 down_write(&EXT4_I(inode
)->i_data_sem
);
4219 ext4_ext_invalidate_cache(inode
);
4221 ext4_discard_preallocations(inode
);
4224 * TODO: optimization is possible here.
4225 * Probably we need not scan at all,
4226 * because page truncation is enough.
4229 /* we have to know where to truncate from in crash case */
4230 EXT4_I(inode
)->i_disksize
= inode
->i_size
;
4231 ext4_mark_inode_dirty(handle
, inode
);
4233 last_block
= (inode
->i_size
+ sb
->s_blocksize
- 1)
4234 >> EXT4_BLOCK_SIZE_BITS(sb
);
4235 err
= ext4_ext_remove_space(inode
, last_block
);
4237 /* In a multi-transaction truncate, we only make the final
4238 * transaction synchronous.
4241 ext4_handle_sync(handle
);
4243 up_write(&EXT4_I(inode
)->i_data_sem
);
4247 * If this was a simple ftruncate() and the file will remain alive,
4248 * then we need to clear up the orphan record which we created above.
4249 * However, if this was a real unlink then we were called by
4250 * ext4_delete_inode(), and we allow that function to clean up the
4251 * orphan info for us.
4254 ext4_orphan_del(handle
, inode
);
4256 inode
->i_mtime
= inode
->i_ctime
= ext4_current_time(inode
);
4257 ext4_mark_inode_dirty(handle
, inode
);
4258 ext4_journal_stop(handle
);
4261 static void ext4_falloc_update_inode(struct inode
*inode
,
4262 int mode
, loff_t new_size
, int update_ctime
)
4264 struct timespec now
;
4267 now
= current_fs_time(inode
->i_sb
);
4268 if (!timespec_equal(&inode
->i_ctime
, &now
))
4269 inode
->i_ctime
= now
;
4272 * Update only when preallocation was requested beyond
4275 if (!(mode
& FALLOC_FL_KEEP_SIZE
)) {
4276 if (new_size
> i_size_read(inode
))
4277 i_size_write(inode
, new_size
);
4278 if (new_size
> EXT4_I(inode
)->i_disksize
)
4279 ext4_update_i_disksize(inode
, new_size
);
4282 * Mark that we allocate beyond EOF so the subsequent truncate
4283 * can proceed even if the new size is the same as i_size.
4285 if (new_size
> i_size_read(inode
))
4286 ext4_set_inode_flag(inode
, EXT4_INODE_EOFBLOCKS
);
4292 * preallocate space for a file. This implements ext4's fallocate file
4293 * operation, which gets called from sys_fallocate system call.
4294 * For block-mapped files, posix_fallocate should fall back to the method
4295 * of writing zeroes to the required new blocks (the same behavior which is
4296 * expected for file systems which do not support fallocate() system call).
4298 long ext4_fallocate(struct file
*file
, int mode
, loff_t offset
, loff_t len
)
4300 struct inode
*inode
= file
->f_path
.dentry
->d_inode
;
4303 unsigned int max_blocks
;
4307 struct ext4_map_blocks map
;
4308 unsigned int credits
, blkbits
= inode
->i_blkbits
;
4311 * currently supporting (pre)allocate mode for extent-based
4314 if (!(ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
)))
4317 /* Return error if mode is not supported */
4318 if (mode
& ~(FALLOC_FL_KEEP_SIZE
| FALLOC_FL_PUNCH_HOLE
))
4321 if (mode
& FALLOC_FL_PUNCH_HOLE
)
4322 return ext4_punch_hole(file
, offset
, len
);
4324 trace_ext4_fallocate_enter(inode
, offset
, len
, mode
);
4325 map
.m_lblk
= offset
>> blkbits
;
4327 * We can't just convert len to max_blocks because
4328 * If blocksize = 4096 offset = 3072 and len = 2048
4330 max_blocks
= (EXT4_BLOCK_ALIGN(len
+ offset
, blkbits
) >> blkbits
)
4333 * credits to insert 1 extent into extent tree
4335 credits
= ext4_chunk_trans_blocks(inode
, max_blocks
);
4336 mutex_lock(&inode
->i_mutex
);
4337 ret
= inode_newsize_ok(inode
, (len
+ offset
));
4339 mutex_unlock(&inode
->i_mutex
);
4340 trace_ext4_fallocate_exit(inode
, offset
, max_blocks
, ret
);
4344 while (ret
>= 0 && ret
< max_blocks
) {
4345 map
.m_lblk
= map
.m_lblk
+ ret
;
4346 map
.m_len
= max_blocks
= max_blocks
- ret
;
4347 handle
= ext4_journal_start(inode
, credits
);
4348 if (IS_ERR(handle
)) {
4349 ret
= PTR_ERR(handle
);
4352 ret
= ext4_map_blocks(handle
, inode
, &map
,
4353 EXT4_GET_BLOCKS_CREATE_UNINIT_EXT
|
4354 EXT4_GET_BLOCKS_NO_NORMALIZE
);
4358 printk(KERN_ERR
"%s: ext4_ext_map_blocks "
4359 "returned error inode#%lu, block=%u, "
4360 "max_blocks=%u", __func__
,
4361 inode
->i_ino
, map
.m_lblk
, max_blocks
);
4363 ext4_mark_inode_dirty(handle
, inode
);
4364 ret2
= ext4_journal_stop(handle
);
4367 if ((map
.m_lblk
+ ret
) >= (EXT4_BLOCK_ALIGN(offset
+ len
,
4368 blkbits
) >> blkbits
))
4369 new_size
= offset
+ len
;
4371 new_size
= ((loff_t
) map
.m_lblk
+ ret
) << blkbits
;
4373 ext4_falloc_update_inode(inode
, mode
, new_size
,
4374 (map
.m_flags
& EXT4_MAP_NEW
));
4375 ext4_mark_inode_dirty(handle
, inode
);
4376 ret2
= ext4_journal_stop(handle
);
4380 if (ret
== -ENOSPC
&&
4381 ext4_should_retry_alloc(inode
->i_sb
, &retries
)) {
4385 mutex_unlock(&inode
->i_mutex
);
4386 trace_ext4_fallocate_exit(inode
, offset
, max_blocks
,
4387 ret
> 0 ? ret2
: ret
);
4388 return ret
> 0 ? ret2
: ret
;
4392 * This function convert a range of blocks to written extents
4393 * The caller of this function will pass the start offset and the size.
4394 * all unwritten extents within this range will be converted to
4397 * This function is called from the direct IO end io call back
4398 * function, to convert the fallocated extents after IO is completed.
4399 * Returns 0 on success.
4401 int ext4_convert_unwritten_extents(struct inode
*inode
, loff_t offset
,
4405 unsigned int max_blocks
;
4408 struct ext4_map_blocks map
;
4409 unsigned int credits
, blkbits
= inode
->i_blkbits
;
4411 map
.m_lblk
= offset
>> blkbits
;
4413 * We can't just convert len to max_blocks because
4414 * If blocksize = 4096 offset = 3072 and len = 2048
4416 max_blocks
= ((EXT4_BLOCK_ALIGN(len
+ offset
, blkbits
) >> blkbits
) -
4419 * credits to insert 1 extent into extent tree
4421 credits
= ext4_chunk_trans_blocks(inode
, max_blocks
);
4422 while (ret
>= 0 && ret
< max_blocks
) {
4424 map
.m_len
= (max_blocks
-= ret
);
4425 handle
= ext4_journal_start(inode
, credits
);
4426 if (IS_ERR(handle
)) {
4427 ret
= PTR_ERR(handle
);
4430 ret
= ext4_map_blocks(handle
, inode
, &map
,
4431 EXT4_GET_BLOCKS_IO_CONVERT_EXT
);
4434 printk(KERN_ERR
"%s: ext4_ext_map_blocks "
4435 "returned error inode#%lu, block=%u, "
4436 "max_blocks=%u", __func__
,
4437 inode
->i_ino
, map
.m_lblk
, map
.m_len
);
4439 ext4_mark_inode_dirty(handle
, inode
);
4440 ret2
= ext4_journal_stop(handle
);
4441 if (ret
<= 0 || ret2
)
4444 return ret
> 0 ? ret2
: ret
;
4448 * Callback function called for each extent to gather FIEMAP information.
4450 static int ext4_ext_fiemap_cb(struct inode
*inode
, ext4_lblk_t next
,
4451 struct ext4_ext_cache
*newex
, struct ext4_extent
*ex
,
4459 struct fiemap_extent_info
*fieinfo
= data
;
4460 unsigned char blksize_bits
;
4462 blksize_bits
= inode
->i_sb
->s_blocksize_bits
;
4463 logical
= (__u64
)newex
->ec_block
<< blksize_bits
;
4465 if (newex
->ec_start
== 0) {
4467 * No extent in extent-tree contains block @newex->ec_start,
4468 * then the block may stay in 1)a hole or 2)delayed-extent.
4470 * Holes or delayed-extents are processed as follows.
4471 * 1. lookup dirty pages with specified range in pagecache.
4472 * If no page is got, then there is no delayed-extent and
4473 * return with EXT_CONTINUE.
4474 * 2. find the 1st mapped buffer,
4475 * 3. check if the mapped buffer is both in the request range
4476 * and a delayed buffer. If not, there is no delayed-extent,
4478 * 4. a delayed-extent is found, the extent will be collected.
4480 ext4_lblk_t end
= 0;
4481 pgoff_t last_offset
;
4484 pgoff_t start_index
= 0;
4485 struct page
**pages
= NULL
;
4486 struct buffer_head
*bh
= NULL
;
4487 struct buffer_head
*head
= NULL
;
4488 unsigned int nr_pages
= PAGE_SIZE
/ sizeof(struct page
*);
4490 pages
= kmalloc(PAGE_SIZE
, GFP_KERNEL
);
4494 offset
= logical
>> PAGE_SHIFT
;
4496 last_offset
= offset
;
4498 ret
= find_get_pages_tag(inode
->i_mapping
, &offset
,
4499 PAGECACHE_TAG_DIRTY
, nr_pages
, pages
);
4501 if (!(flags
& FIEMAP_EXTENT_DELALLOC
)) {
4502 /* First time, try to find a mapped buffer. */
4505 for (index
= 0; index
< ret
; index
++)
4506 page_cache_release(pages
[index
]);
4509 return EXT_CONTINUE
;
4514 /* Try to find the 1st mapped buffer. */
4515 end
= ((__u64
)pages
[index
]->index
<< PAGE_SHIFT
) >>
4517 if (!page_has_buffers(pages
[index
]))
4519 head
= page_buffers(pages
[index
]);
4526 if (end
>= newex
->ec_block
+
4528 /* The buffer is out of
4529 * the request range.
4533 if (buffer_mapped(bh
) &&
4534 end
>= newex
->ec_block
) {
4535 start_index
= index
- 1;
4536 /* get the 1st mapped buffer. */
4537 goto found_mapped_buffer
;
4540 bh
= bh
->b_this_page
;
4542 } while (bh
!= head
);
4544 /* No mapped buffer in the range found in this page,
4545 * We need to look up next page.
4548 /* There is no page left, but we need to limit
4551 newex
->ec_len
= end
- newex
->ec_block
;
4556 /*Find contiguous delayed buffers. */
4557 if (ret
> 0 && pages
[0]->index
== last_offset
)
4558 head
= page_buffers(pages
[0]);
4564 found_mapped_buffer
:
4565 if (bh
!= NULL
&& buffer_delay(bh
)) {
4566 /* 1st or contiguous delayed buffer found. */
4567 if (!(flags
& FIEMAP_EXTENT_DELALLOC
)) {
4569 * 1st delayed buffer found, record
4570 * the start of extent.
4572 flags
|= FIEMAP_EXTENT_DELALLOC
;
4573 newex
->ec_block
= end
;
4574 logical
= (__u64
)end
<< blksize_bits
;
4576 /* Find contiguous delayed buffers. */
4578 if (!buffer_delay(bh
))
4579 goto found_delayed_extent
;
4580 bh
= bh
->b_this_page
;
4582 } while (bh
!= head
);
4584 for (; index
< ret
; index
++) {
4585 if (!page_has_buffers(pages
[index
])) {
4589 head
= page_buffers(pages
[index
]);
4595 if (pages
[index
]->index
!=
4596 pages
[start_index
]->index
+ index
4598 /* Blocks are not contiguous. */
4604 if (!buffer_delay(bh
))
4605 /* Delayed-extent ends. */
4606 goto found_delayed_extent
;
4607 bh
= bh
->b_this_page
;
4609 } while (bh
!= head
);
4611 } else if (!(flags
& FIEMAP_EXTENT_DELALLOC
))
4615 found_delayed_extent
:
4616 newex
->ec_len
= min(end
- newex
->ec_block
,
4617 (ext4_lblk_t
)EXT_INIT_MAX_LEN
);
4618 if (ret
== nr_pages
&& bh
!= NULL
&&
4619 newex
->ec_len
< EXT_INIT_MAX_LEN
&&
4621 /* Have not collected an extent and continue. */
4622 for (index
= 0; index
< ret
; index
++)
4623 page_cache_release(pages
[index
]);
4627 for (index
= 0; index
< ret
; index
++)
4628 page_cache_release(pages
[index
]);
4632 physical
= (__u64
)newex
->ec_start
<< blksize_bits
;
4633 length
= (__u64
)newex
->ec_len
<< blksize_bits
;
4635 if (ex
&& ext4_ext_is_uninitialized(ex
))
4636 flags
|= FIEMAP_EXTENT_UNWRITTEN
;
4638 if (next
== EXT_MAX_BLOCKS
)
4639 flags
|= FIEMAP_EXTENT_LAST
;
4641 ret
= fiemap_fill_next_extent(fieinfo
, logical
, physical
,
4647 return EXT_CONTINUE
;
4649 /* fiemap flags we can handle specified here */
4650 #define EXT4_FIEMAP_FLAGS (FIEMAP_FLAG_SYNC|FIEMAP_FLAG_XATTR)
4652 static int ext4_xattr_fiemap(struct inode
*inode
,
4653 struct fiemap_extent_info
*fieinfo
)
4657 __u32 flags
= FIEMAP_EXTENT_LAST
;
4658 int blockbits
= inode
->i_sb
->s_blocksize_bits
;
4662 if (ext4_test_inode_state(inode
, EXT4_STATE_XATTR
)) {
4663 struct ext4_iloc iloc
;
4664 int offset
; /* offset of xattr in inode */
4666 error
= ext4_get_inode_loc(inode
, &iloc
);
4669 physical
= iloc
.bh
->b_blocknr
<< blockbits
;
4670 offset
= EXT4_GOOD_OLD_INODE_SIZE
+
4671 EXT4_I(inode
)->i_extra_isize
;
4673 length
= EXT4_SB(inode
->i_sb
)->s_inode_size
- offset
;
4674 flags
|= FIEMAP_EXTENT_DATA_INLINE
;
4676 } else { /* external block */
4677 physical
= EXT4_I(inode
)->i_file_acl
<< blockbits
;
4678 length
= inode
->i_sb
->s_blocksize
;
4682 error
= fiemap_fill_next_extent(fieinfo
, 0, physical
,
4684 return (error
< 0 ? error
: 0);
4688 * ext4_ext_punch_hole
4690 * Punches a hole of "length" bytes in a file starting
4693 * @inode: The inode of the file to punch a hole in
4694 * @offset: The starting byte offset of the hole
4695 * @length: The length of the hole
4697 * Returns the number of blocks removed or negative on err
4699 int ext4_ext_punch_hole(struct file
*file
, loff_t offset
, loff_t length
)
4701 struct inode
*inode
= file
->f_path
.dentry
->d_inode
;
4702 struct super_block
*sb
= inode
->i_sb
;
4703 struct ext4_ext_cache cache_ex
;
4704 ext4_lblk_t first_block
, last_block
, num_blocks
, iblock
, max_blocks
;
4705 struct address_space
*mapping
= inode
->i_mapping
;
4706 struct ext4_map_blocks map
;
4708 loff_t first_page
, last_page
, page_len
;
4709 loff_t first_page_offset
, last_page_offset
;
4710 int ret
, credits
, blocks_released
, err
= 0;
4712 /* No need to punch hole beyond i_size */
4713 if (offset
>= inode
->i_size
)
4717 * If the hole extends beyond i_size, set the hole
4718 * to end after the page that contains i_size
4720 if (offset
+ length
> inode
->i_size
) {
4721 length
= inode
->i_size
+
4722 PAGE_CACHE_SIZE
- (inode
->i_size
& (PAGE_CACHE_SIZE
- 1)) -
4726 first_block
= (offset
+ sb
->s_blocksize
- 1) >>
4727 EXT4_BLOCK_SIZE_BITS(sb
);
4728 last_block
= (offset
+ length
) >> EXT4_BLOCK_SIZE_BITS(sb
);
4730 first_page
= (offset
+ PAGE_CACHE_SIZE
- 1) >> PAGE_CACHE_SHIFT
;
4731 last_page
= (offset
+ length
) >> PAGE_CACHE_SHIFT
;
4733 first_page_offset
= first_page
<< PAGE_CACHE_SHIFT
;
4734 last_page_offset
= last_page
<< PAGE_CACHE_SHIFT
;
4737 * Write out all dirty pages to avoid race conditions
4738 * Then release them.
4740 if (mapping
->nrpages
&& mapping_tagged(mapping
, PAGECACHE_TAG_DIRTY
)) {
4741 err
= filemap_write_and_wait_range(mapping
,
4742 offset
, offset
+ length
- 1);
4748 /* Now release the pages */
4749 if (last_page_offset
> first_page_offset
) {
4750 truncate_inode_pages_range(mapping
, first_page_offset
,
4751 last_page_offset
-1);
4754 /* finish any pending end_io work */
4755 ext4_flush_completed_IO(inode
);
4757 credits
= ext4_writepage_trans_blocks(inode
);
4758 handle
= ext4_journal_start(inode
, credits
);
4760 return PTR_ERR(handle
);
4762 err
= ext4_orphan_add(handle
, inode
);
4767 * Now we need to zero out the non-page-aligned data in the
4768 * pages at the start and tail of the hole, and unmap the buffer
4769 * heads for the block aligned regions of the page that were
4770 * completely zeroed.
4772 if (first_page
> last_page
) {
4774 * If the file space being truncated is contained within a page
4775 * just zero out and unmap the middle of that page
4777 err
= ext4_discard_partial_page_buffers(handle
,
4778 mapping
, offset
, length
, 0);
4784 * zero out and unmap the partial page that contains
4785 * the start of the hole
4787 page_len
= first_page_offset
- offset
;
4789 err
= ext4_discard_partial_page_buffers(handle
, mapping
,
4790 offset
, page_len
, 0);
4796 * zero out and unmap the partial page that contains
4797 * the end of the hole
4799 page_len
= offset
+ length
- last_page_offset
;
4801 err
= ext4_discard_partial_page_buffers(handle
, mapping
,
4802 last_page_offset
, page_len
, 0);
4810 * If i_size is contained in the last page, we need to
4811 * unmap and zero the partial page after i_size
4813 if (inode
->i_size
>> PAGE_CACHE_SHIFT
== last_page
&&
4814 inode
->i_size
% PAGE_CACHE_SIZE
!= 0) {
4816 page_len
= PAGE_CACHE_SIZE
-
4817 (inode
->i_size
& (PAGE_CACHE_SIZE
- 1));
4820 err
= ext4_discard_partial_page_buffers(handle
,
4821 mapping
, inode
->i_size
, page_len
, 0);
4828 /* If there are no blocks to remove, return now */
4829 if (first_block
>= last_block
)
4832 down_write(&EXT4_I(inode
)->i_data_sem
);
4833 ext4_ext_invalidate_cache(inode
);
4834 ext4_discard_preallocations(inode
);
4837 * Loop over all the blocks and identify blocks
4838 * that need to be punched out
4840 iblock
= first_block
;
4841 blocks_released
= 0;
4842 while (iblock
< last_block
) {
4843 max_blocks
= last_block
- iblock
;
4845 memset(&map
, 0, sizeof(map
));
4846 map
.m_lblk
= iblock
;
4847 map
.m_len
= max_blocks
;
4848 ret
= ext4_ext_map_blocks(handle
, inode
, &map
,
4849 EXT4_GET_BLOCKS_PUNCH_OUT_EXT
);
4852 blocks_released
+= ret
;
4854 } else if (ret
== 0) {
4856 * If map blocks could not find the block,
4857 * then it is in a hole. If the hole was
4858 * not already cached, then map blocks should
4859 * put it in the cache. So we can get the hole
4862 memset(&cache_ex
, 0, sizeof(cache_ex
));
4863 if ((ext4_ext_check_cache(inode
, iblock
, &cache_ex
)) &&
4864 !cache_ex
.ec_start
) {
4866 /* The hole is cached */
4867 num_blocks
= cache_ex
.ec_block
+
4868 cache_ex
.ec_len
- iblock
;
4871 /* The block could not be identified */
4876 /* Map blocks error */
4881 if (num_blocks
== 0) {
4882 /* This condition should never happen */
4883 ext_debug("Block lookup failed");
4888 iblock
+= num_blocks
;
4891 if (blocks_released
> 0) {
4892 ext4_ext_invalidate_cache(inode
);
4893 ext4_discard_preallocations(inode
);
4897 ext4_handle_sync(handle
);
4899 up_write(&EXT4_I(inode
)->i_data_sem
);
4902 ext4_orphan_del(handle
, inode
);
4903 inode
->i_mtime
= inode
->i_ctime
= ext4_current_time(inode
);
4904 ext4_mark_inode_dirty(handle
, inode
);
4905 ext4_journal_stop(handle
);
4908 int ext4_fiemap(struct inode
*inode
, struct fiemap_extent_info
*fieinfo
,
4909 __u64 start
, __u64 len
)
4911 ext4_lblk_t start_blk
;
4914 /* fallback to generic here if not in extents fmt */
4915 if (!(ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
)))
4916 return generic_block_fiemap(inode
, fieinfo
, start
, len
,
4919 if (fiemap_check_flags(fieinfo
, EXT4_FIEMAP_FLAGS
))
4922 if (fieinfo
->fi_flags
& FIEMAP_FLAG_XATTR
) {
4923 error
= ext4_xattr_fiemap(inode
, fieinfo
);
4925 ext4_lblk_t len_blks
;
4928 start_blk
= start
>> inode
->i_sb
->s_blocksize_bits
;
4929 last_blk
= (start
+ len
- 1) >> inode
->i_sb
->s_blocksize_bits
;
4930 if (last_blk
>= EXT_MAX_BLOCKS
)
4931 last_blk
= EXT_MAX_BLOCKS
-1;
4932 len_blks
= ((ext4_lblk_t
) last_blk
) - start_blk
+ 1;
4935 * Walk the extent tree gathering extent information.
4936 * ext4_ext_fiemap_cb will push extents back to user.
4938 error
= ext4_ext_walk_space(inode
, start_blk
, len_blks
,
4939 ext4_ext_fiemap_cb
, fieinfo
);