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 "ext4_jbd2.h"
44 #include "ext4_extents.h"
49 * combine low and high parts of physical block number into ext4_fsblk_t
51 static ext4_fsblk_t
ext_pblock(struct ext4_extent
*ex
)
55 block
= le32_to_cpu(ex
->ee_start_lo
);
56 block
|= ((ext4_fsblk_t
) le16_to_cpu(ex
->ee_start_hi
) << 31) << 1;
62 * combine low and high parts of a leaf physical block number into ext4_fsblk_t
64 ext4_fsblk_t
idx_pblock(struct ext4_extent_idx
*ix
)
68 block
= le32_to_cpu(ix
->ei_leaf_lo
);
69 block
|= ((ext4_fsblk_t
) le16_to_cpu(ix
->ei_leaf_hi
) << 31) << 1;
74 * ext4_ext_store_pblock:
75 * stores a large physical block number into an extent struct,
76 * breaking it into parts
78 void ext4_ext_store_pblock(struct ext4_extent
*ex
, ext4_fsblk_t pb
)
80 ex
->ee_start_lo
= cpu_to_le32((unsigned long) (pb
& 0xffffffff));
81 ex
->ee_start_hi
= cpu_to_le16((unsigned long) ((pb
>> 31) >> 1) & 0xffff);
85 * ext4_idx_store_pblock:
86 * stores a large physical block number into an index struct,
87 * breaking it into parts
89 static void ext4_idx_store_pblock(struct ext4_extent_idx
*ix
, ext4_fsblk_t pb
)
91 ix
->ei_leaf_lo
= cpu_to_le32((unsigned long) (pb
& 0xffffffff));
92 ix
->ei_leaf_hi
= cpu_to_le16((unsigned long) ((pb
>> 31) >> 1) & 0xffff);
95 static int ext4_ext_journal_restart(handle_t
*handle
, int needed
)
99 if (handle
->h_buffer_credits
> needed
)
101 err
= ext4_journal_extend(handle
, needed
);
104 return ext4_journal_restart(handle
, needed
);
112 static int ext4_ext_get_access(handle_t
*handle
, struct inode
*inode
,
113 struct ext4_ext_path
*path
)
116 /* path points to block */
117 return ext4_journal_get_write_access(handle
, path
->p_bh
);
119 /* path points to leaf/index in inode body */
120 /* we use in-core data, no need to protect them */
130 static int ext4_ext_dirty(handle_t
*handle
, struct inode
*inode
,
131 struct ext4_ext_path
*path
)
135 /* path points to block */
136 err
= ext4_journal_dirty_metadata(handle
, path
->p_bh
);
138 /* path points to leaf/index in inode body */
139 err
= ext4_mark_inode_dirty(handle
, inode
);
144 static ext4_fsblk_t
ext4_ext_find_goal(struct inode
*inode
,
145 struct ext4_ext_path
*path
,
148 struct ext4_inode_info
*ei
= EXT4_I(inode
);
149 ext4_fsblk_t bg_start
;
150 ext4_fsblk_t last_block
;
151 ext4_grpblk_t colour
;
155 struct ext4_extent
*ex
;
156 depth
= path
->p_depth
;
158 /* try to predict block placement */
159 ex
= path
[depth
].p_ext
;
161 return ext_pblock(ex
)+(block
-le32_to_cpu(ex
->ee_block
));
163 /* it looks like index is empty;
164 * try to find starting block from index itself */
165 if (path
[depth
].p_bh
)
166 return path
[depth
].p_bh
->b_blocknr
;
169 /* OK. use inode's group */
170 bg_start
= (ei
->i_block_group
* EXT4_BLOCKS_PER_GROUP(inode
->i_sb
)) +
171 le32_to_cpu(EXT4_SB(inode
->i_sb
)->s_es
->s_first_data_block
);
172 last_block
= ext4_blocks_count(EXT4_SB(inode
->i_sb
)->s_es
) - 1;
174 if (bg_start
+ EXT4_BLOCKS_PER_GROUP(inode
->i_sb
) <= last_block
)
175 colour
= (current
->pid
% 16) *
176 (EXT4_BLOCKS_PER_GROUP(inode
->i_sb
) / 16);
178 colour
= (current
->pid
% 16) * ((last_block
- bg_start
) / 16);
179 return bg_start
+ colour
+ block
;
183 ext4_ext_new_block(handle_t
*handle
, struct inode
*inode
,
184 struct ext4_ext_path
*path
,
185 struct ext4_extent
*ex
, int *err
)
187 ext4_fsblk_t goal
, newblock
;
189 goal
= ext4_ext_find_goal(inode
, path
, le32_to_cpu(ex
->ee_block
));
190 newblock
= ext4_new_meta_block(handle
, inode
, goal
, err
);
194 static int ext4_ext_space_block(struct inode
*inode
)
198 size
= (inode
->i_sb
->s_blocksize
- sizeof(struct ext4_extent_header
))
199 / sizeof(struct ext4_extent
);
200 #ifdef AGGRESSIVE_TEST
207 static int ext4_ext_space_block_idx(struct inode
*inode
)
211 size
= (inode
->i_sb
->s_blocksize
- sizeof(struct ext4_extent_header
))
212 / sizeof(struct ext4_extent_idx
);
213 #ifdef AGGRESSIVE_TEST
220 static int ext4_ext_space_root(struct inode
*inode
)
224 size
= sizeof(EXT4_I(inode
)->i_data
);
225 size
-= sizeof(struct ext4_extent_header
);
226 size
/= sizeof(struct ext4_extent
);
227 #ifdef AGGRESSIVE_TEST
234 static int ext4_ext_space_root_idx(struct inode
*inode
)
238 size
= sizeof(EXT4_I(inode
)->i_data
);
239 size
-= sizeof(struct ext4_extent_header
);
240 size
/= sizeof(struct ext4_extent_idx
);
241 #ifdef AGGRESSIVE_TEST
249 ext4_ext_max_entries(struct inode
*inode
, int depth
)
253 if (depth
== ext_depth(inode
)) {
255 max
= ext4_ext_space_root(inode
);
257 max
= ext4_ext_space_root_idx(inode
);
260 max
= ext4_ext_space_block(inode
);
262 max
= ext4_ext_space_block_idx(inode
);
268 static int __ext4_ext_check_header(const char *function
, struct inode
*inode
,
269 struct ext4_extent_header
*eh
,
272 const char *error_msg
;
275 if (unlikely(eh
->eh_magic
!= EXT4_EXT_MAGIC
)) {
276 error_msg
= "invalid magic";
279 if (unlikely(le16_to_cpu(eh
->eh_depth
) != depth
)) {
280 error_msg
= "unexpected eh_depth";
283 if (unlikely(eh
->eh_max
== 0)) {
284 error_msg
= "invalid eh_max";
287 max
= ext4_ext_max_entries(inode
, depth
);
288 if (unlikely(le16_to_cpu(eh
->eh_max
) > max
)) {
289 error_msg
= "too large eh_max";
292 if (unlikely(le16_to_cpu(eh
->eh_entries
) > le16_to_cpu(eh
->eh_max
))) {
293 error_msg
= "invalid eh_entries";
299 ext4_error(inode
->i_sb
, function
,
300 "bad header in inode #%lu: %s - magic %x, "
301 "entries %u, max %u(%u), depth %u(%u)",
302 inode
->i_ino
, error_msg
, le16_to_cpu(eh
->eh_magic
),
303 le16_to_cpu(eh
->eh_entries
), le16_to_cpu(eh
->eh_max
),
304 max
, le16_to_cpu(eh
->eh_depth
), depth
);
309 #define ext4_ext_check_header(inode, eh, depth) \
310 __ext4_ext_check_header(__func__, inode, eh, depth)
313 static void ext4_ext_show_path(struct inode
*inode
, struct ext4_ext_path
*path
)
315 int k
, l
= path
->p_depth
;
318 for (k
= 0; k
<= l
; k
++, path
++) {
320 ext_debug(" %d->%llu", le32_to_cpu(path
->p_idx
->ei_block
),
321 idx_pblock(path
->p_idx
));
322 } else if (path
->p_ext
) {
323 ext_debug(" %d:%d:%llu ",
324 le32_to_cpu(path
->p_ext
->ee_block
),
325 ext4_ext_get_actual_len(path
->p_ext
),
326 ext_pblock(path
->p_ext
));
333 static void ext4_ext_show_leaf(struct inode
*inode
, struct ext4_ext_path
*path
)
335 int depth
= ext_depth(inode
);
336 struct ext4_extent_header
*eh
;
337 struct ext4_extent
*ex
;
343 eh
= path
[depth
].p_hdr
;
344 ex
= EXT_FIRST_EXTENT(eh
);
346 for (i
= 0; i
< le16_to_cpu(eh
->eh_entries
); i
++, ex
++) {
347 ext_debug("%d:%d:%llu ", le32_to_cpu(ex
->ee_block
),
348 ext4_ext_get_actual_len(ex
), ext_pblock(ex
));
353 #define ext4_ext_show_path(inode,path)
354 #define ext4_ext_show_leaf(inode,path)
357 void ext4_ext_drop_refs(struct ext4_ext_path
*path
)
359 int depth
= path
->p_depth
;
362 for (i
= 0; i
<= depth
; i
++, path
++)
370 * ext4_ext_binsearch_idx:
371 * binary search for the closest index of the given block
372 * the header must be checked before calling this
375 ext4_ext_binsearch_idx(struct inode
*inode
,
376 struct ext4_ext_path
*path
, ext4_lblk_t block
)
378 struct ext4_extent_header
*eh
= path
->p_hdr
;
379 struct ext4_extent_idx
*r
, *l
, *m
;
382 ext_debug("binsearch for %u(idx): ", block
);
384 l
= EXT_FIRST_INDEX(eh
) + 1;
385 r
= EXT_LAST_INDEX(eh
);
388 if (block
< le32_to_cpu(m
->ei_block
))
392 ext_debug("%p(%u):%p(%u):%p(%u) ", l
, le32_to_cpu(l
->ei_block
),
393 m
, le32_to_cpu(m
->ei_block
),
394 r
, le32_to_cpu(r
->ei_block
));
398 ext_debug(" -> %d->%lld ", le32_to_cpu(path
->p_idx
->ei_block
),
399 idx_pblock(path
->p_idx
));
401 #ifdef CHECK_BINSEARCH
403 struct ext4_extent_idx
*chix
, *ix
;
406 chix
= ix
= EXT_FIRST_INDEX(eh
);
407 for (k
= 0; k
< le16_to_cpu(eh
->eh_entries
); k
++, ix
++) {
409 le32_to_cpu(ix
->ei_block
) <= le32_to_cpu(ix
[-1].ei_block
)) {
410 printk("k=%d, ix=0x%p, first=0x%p\n", k
,
411 ix
, EXT_FIRST_INDEX(eh
));
413 le32_to_cpu(ix
->ei_block
),
414 le32_to_cpu(ix
[-1].ei_block
));
416 BUG_ON(k
&& le32_to_cpu(ix
->ei_block
)
417 <= le32_to_cpu(ix
[-1].ei_block
));
418 if (block
< le32_to_cpu(ix
->ei_block
))
422 BUG_ON(chix
!= path
->p_idx
);
429 * ext4_ext_binsearch:
430 * binary search for closest extent of the given block
431 * the header must be checked before calling this
434 ext4_ext_binsearch(struct inode
*inode
,
435 struct ext4_ext_path
*path
, ext4_lblk_t block
)
437 struct ext4_extent_header
*eh
= path
->p_hdr
;
438 struct ext4_extent
*r
, *l
, *m
;
440 if (eh
->eh_entries
== 0) {
442 * this leaf is empty:
443 * we get such a leaf in split/add case
448 ext_debug("binsearch for %u: ", block
);
450 l
= EXT_FIRST_EXTENT(eh
) + 1;
451 r
= EXT_LAST_EXTENT(eh
);
455 if (block
< le32_to_cpu(m
->ee_block
))
459 ext_debug("%p(%u):%p(%u):%p(%u) ", l
, le32_to_cpu(l
->ee_block
),
460 m
, le32_to_cpu(m
->ee_block
),
461 r
, le32_to_cpu(r
->ee_block
));
465 ext_debug(" -> %d:%llu:%d ",
466 le32_to_cpu(path
->p_ext
->ee_block
),
467 ext_pblock(path
->p_ext
),
468 ext4_ext_get_actual_len(path
->p_ext
));
470 #ifdef CHECK_BINSEARCH
472 struct ext4_extent
*chex
, *ex
;
475 chex
= ex
= EXT_FIRST_EXTENT(eh
);
476 for (k
= 0; k
< le16_to_cpu(eh
->eh_entries
); k
++, ex
++) {
477 BUG_ON(k
&& le32_to_cpu(ex
->ee_block
)
478 <= le32_to_cpu(ex
[-1].ee_block
));
479 if (block
< le32_to_cpu(ex
->ee_block
))
483 BUG_ON(chex
!= path
->p_ext
);
489 int ext4_ext_tree_init(handle_t
*handle
, struct inode
*inode
)
491 struct ext4_extent_header
*eh
;
493 eh
= ext_inode_hdr(inode
);
496 eh
->eh_magic
= EXT4_EXT_MAGIC
;
497 eh
->eh_max
= cpu_to_le16(ext4_ext_space_root(inode
));
498 ext4_mark_inode_dirty(handle
, inode
);
499 ext4_ext_invalidate_cache(inode
);
503 struct ext4_ext_path
*
504 ext4_ext_find_extent(struct inode
*inode
, ext4_lblk_t block
,
505 struct ext4_ext_path
*path
)
507 struct ext4_extent_header
*eh
;
508 struct buffer_head
*bh
;
509 short int depth
, i
, ppos
= 0, alloc
= 0;
511 eh
= ext_inode_hdr(inode
);
512 depth
= ext_depth(inode
);
513 if (ext4_ext_check_header(inode
, eh
, depth
))
514 return ERR_PTR(-EIO
);
517 /* account possible depth increase */
519 path
= kzalloc(sizeof(struct ext4_ext_path
) * (depth
+ 2),
522 return ERR_PTR(-ENOMEM
);
529 /* walk through the tree */
531 ext_debug("depth %d: num %d, max %d\n",
532 ppos
, le16_to_cpu(eh
->eh_entries
), le16_to_cpu(eh
->eh_max
));
534 ext4_ext_binsearch_idx(inode
, path
+ ppos
, block
);
535 path
[ppos
].p_block
= idx_pblock(path
[ppos
].p_idx
);
536 path
[ppos
].p_depth
= i
;
537 path
[ppos
].p_ext
= NULL
;
539 bh
= sb_bread(inode
->i_sb
, path
[ppos
].p_block
);
543 eh
= ext_block_hdr(bh
);
545 BUG_ON(ppos
> depth
);
546 path
[ppos
].p_bh
= bh
;
547 path
[ppos
].p_hdr
= eh
;
550 if (ext4_ext_check_header(inode
, eh
, i
))
554 path
[ppos
].p_depth
= i
;
555 path
[ppos
].p_ext
= NULL
;
556 path
[ppos
].p_idx
= NULL
;
559 ext4_ext_binsearch(inode
, path
+ ppos
, block
);
560 /* if not an empty leaf */
561 if (path
[ppos
].p_ext
)
562 path
[ppos
].p_block
= ext_pblock(path
[ppos
].p_ext
);
564 ext4_ext_show_path(inode
, path
);
569 ext4_ext_drop_refs(path
);
572 return ERR_PTR(-EIO
);
576 * ext4_ext_insert_index:
577 * insert new index [@logical;@ptr] into the block at @curp;
578 * check where to insert: before @curp or after @curp
580 static int ext4_ext_insert_index(handle_t
*handle
, struct inode
*inode
,
581 struct ext4_ext_path
*curp
,
582 int logical
, ext4_fsblk_t ptr
)
584 struct ext4_extent_idx
*ix
;
587 err
= ext4_ext_get_access(handle
, inode
, curp
);
591 BUG_ON(logical
== le32_to_cpu(curp
->p_idx
->ei_block
));
592 len
= EXT_MAX_INDEX(curp
->p_hdr
) - curp
->p_idx
;
593 if (logical
> le32_to_cpu(curp
->p_idx
->ei_block
)) {
595 if (curp
->p_idx
!= EXT_LAST_INDEX(curp
->p_hdr
)) {
596 len
= (len
- 1) * sizeof(struct ext4_extent_idx
);
597 len
= len
< 0 ? 0 : len
;
598 ext_debug("insert new index %d after: %llu. "
599 "move %d from 0x%p to 0x%p\n",
601 (curp
->p_idx
+ 1), (curp
->p_idx
+ 2));
602 memmove(curp
->p_idx
+ 2, curp
->p_idx
+ 1, len
);
604 ix
= curp
->p_idx
+ 1;
607 len
= len
* sizeof(struct ext4_extent_idx
);
608 len
= len
< 0 ? 0 : len
;
609 ext_debug("insert new index %d before: %llu. "
610 "move %d from 0x%p to 0x%p\n",
612 curp
->p_idx
, (curp
->p_idx
+ 1));
613 memmove(curp
->p_idx
+ 1, curp
->p_idx
, len
);
617 ix
->ei_block
= cpu_to_le32(logical
);
618 ext4_idx_store_pblock(ix
, ptr
);
619 le16_add_cpu(&curp
->p_hdr
->eh_entries
, 1);
621 BUG_ON(le16_to_cpu(curp
->p_hdr
->eh_entries
)
622 > le16_to_cpu(curp
->p_hdr
->eh_max
));
623 BUG_ON(ix
> EXT_LAST_INDEX(curp
->p_hdr
));
625 err
= ext4_ext_dirty(handle
, inode
, curp
);
626 ext4_std_error(inode
->i_sb
, err
);
633 * inserts new subtree into the path, using free index entry
635 * - allocates all needed blocks (new leaf and all intermediate index blocks)
636 * - makes decision where to split
637 * - moves remaining extents and index entries (right to the split point)
638 * into the newly allocated blocks
639 * - initializes subtree
641 static int ext4_ext_split(handle_t
*handle
, struct inode
*inode
,
642 struct ext4_ext_path
*path
,
643 struct ext4_extent
*newext
, int at
)
645 struct buffer_head
*bh
= NULL
;
646 int depth
= ext_depth(inode
);
647 struct ext4_extent_header
*neh
;
648 struct ext4_extent_idx
*fidx
;
649 struct ext4_extent
*ex
;
651 ext4_fsblk_t newblock
, oldblock
;
653 ext4_fsblk_t
*ablocks
= NULL
; /* array of allocated blocks */
656 /* make decision: where to split? */
657 /* FIXME: now decision is simplest: at current extent */
659 /* if current leaf will be split, then we should use
660 * border from split point */
661 BUG_ON(path
[depth
].p_ext
> EXT_MAX_EXTENT(path
[depth
].p_hdr
));
662 if (path
[depth
].p_ext
!= EXT_MAX_EXTENT(path
[depth
].p_hdr
)) {
663 border
= path
[depth
].p_ext
[1].ee_block
;
664 ext_debug("leaf will be split."
665 " next leaf starts at %d\n",
666 le32_to_cpu(border
));
668 border
= newext
->ee_block
;
669 ext_debug("leaf will be added."
670 " next leaf starts at %d\n",
671 le32_to_cpu(border
));
675 * If error occurs, then we break processing
676 * and mark filesystem read-only. index won't
677 * be inserted and tree will be in consistent
678 * state. Next mount will repair buffers too.
682 * Get array to track all allocated blocks.
683 * We need this to handle errors and free blocks
686 ablocks
= kzalloc(sizeof(ext4_fsblk_t
) * depth
, GFP_NOFS
);
690 /* allocate all needed blocks */
691 ext_debug("allocate %d blocks for indexes/leaf\n", depth
- at
);
692 for (a
= 0; a
< depth
- at
; a
++) {
693 newblock
= ext4_ext_new_block(handle
, inode
, path
, newext
, &err
);
696 ablocks
[a
] = newblock
;
699 /* initialize new leaf */
700 newblock
= ablocks
[--a
];
701 BUG_ON(newblock
== 0);
702 bh
= sb_getblk(inode
->i_sb
, newblock
);
709 err
= ext4_journal_get_create_access(handle
, bh
);
713 neh
= ext_block_hdr(bh
);
715 neh
->eh_max
= cpu_to_le16(ext4_ext_space_block(inode
));
716 neh
->eh_magic
= EXT4_EXT_MAGIC
;
718 ex
= EXT_FIRST_EXTENT(neh
);
720 /* move remainder of path[depth] to the new leaf */
721 BUG_ON(path
[depth
].p_hdr
->eh_entries
!= path
[depth
].p_hdr
->eh_max
);
722 /* start copy from next extent */
723 /* TODO: we could do it by single memmove */
726 while (path
[depth
].p_ext
<=
727 EXT_MAX_EXTENT(path
[depth
].p_hdr
)) {
728 ext_debug("move %d:%llu:%d in new leaf %llu\n",
729 le32_to_cpu(path
[depth
].p_ext
->ee_block
),
730 ext_pblock(path
[depth
].p_ext
),
731 ext4_ext_get_actual_len(path
[depth
].p_ext
),
733 /*memmove(ex++, path[depth].p_ext++,
734 sizeof(struct ext4_extent));
740 memmove(ex
, path
[depth
].p_ext
-m
, sizeof(struct ext4_extent
)*m
);
741 le16_add_cpu(&neh
->eh_entries
, m
);
744 set_buffer_uptodate(bh
);
747 err
= ext4_journal_dirty_metadata(handle
, bh
);
753 /* correct old leaf */
755 err
= ext4_ext_get_access(handle
, inode
, path
+ depth
);
758 le16_add_cpu(&path
[depth
].p_hdr
->eh_entries
, -m
);
759 err
= ext4_ext_dirty(handle
, inode
, path
+ depth
);
765 /* create intermediate indexes */
769 ext_debug("create %d intermediate indices\n", k
);
770 /* insert new index into current index block */
771 /* current depth stored in i var */
775 newblock
= ablocks
[--a
];
776 bh
= sb_getblk(inode
->i_sb
, newblock
);
783 err
= ext4_journal_get_create_access(handle
, bh
);
787 neh
= ext_block_hdr(bh
);
788 neh
->eh_entries
= cpu_to_le16(1);
789 neh
->eh_magic
= EXT4_EXT_MAGIC
;
790 neh
->eh_max
= cpu_to_le16(ext4_ext_space_block_idx(inode
));
791 neh
->eh_depth
= cpu_to_le16(depth
- i
);
792 fidx
= EXT_FIRST_INDEX(neh
);
793 fidx
->ei_block
= border
;
794 ext4_idx_store_pblock(fidx
, oldblock
);
796 ext_debug("int.index at %d (block %llu): %u -> %llu\n",
797 i
, newblock
, le32_to_cpu(border
), oldblock
);
802 ext_debug("cur 0x%p, last 0x%p\n", path
[i
].p_idx
,
803 EXT_MAX_INDEX(path
[i
].p_hdr
));
804 BUG_ON(EXT_MAX_INDEX(path
[i
].p_hdr
) !=
805 EXT_LAST_INDEX(path
[i
].p_hdr
));
806 while (path
[i
].p_idx
<= EXT_MAX_INDEX(path
[i
].p_hdr
)) {
807 ext_debug("%d: move %d:%llu in new index %llu\n", i
,
808 le32_to_cpu(path
[i
].p_idx
->ei_block
),
809 idx_pblock(path
[i
].p_idx
),
811 /*memmove(++fidx, path[i].p_idx++,
812 sizeof(struct ext4_extent_idx));
814 BUG_ON(neh->eh_entries > neh->eh_max);*/
819 memmove(++fidx
, path
[i
].p_idx
- m
,
820 sizeof(struct ext4_extent_idx
) * m
);
821 le16_add_cpu(&neh
->eh_entries
, m
);
823 set_buffer_uptodate(bh
);
826 err
= ext4_journal_dirty_metadata(handle
, bh
);
832 /* correct old index */
834 err
= ext4_ext_get_access(handle
, inode
, path
+ i
);
837 le16_add_cpu(&path
[i
].p_hdr
->eh_entries
, -m
);
838 err
= ext4_ext_dirty(handle
, inode
, path
+ i
);
846 /* insert new index */
847 err
= ext4_ext_insert_index(handle
, inode
, path
+ at
,
848 le32_to_cpu(border
), newblock
);
852 if (buffer_locked(bh
))
858 /* free all allocated blocks in error case */
859 for (i
= 0; i
< depth
; i
++) {
862 ext4_free_blocks(handle
, inode
, ablocks
[i
], 1, 1);
871 * ext4_ext_grow_indepth:
872 * implements tree growing procedure:
873 * - allocates new block
874 * - moves top-level data (index block or leaf) into the new block
875 * - initializes new top-level, creating index that points to the
878 static int ext4_ext_grow_indepth(handle_t
*handle
, struct inode
*inode
,
879 struct ext4_ext_path
*path
,
880 struct ext4_extent
*newext
)
882 struct ext4_ext_path
*curp
= path
;
883 struct ext4_extent_header
*neh
;
884 struct ext4_extent_idx
*fidx
;
885 struct buffer_head
*bh
;
886 ext4_fsblk_t newblock
;
889 newblock
= ext4_ext_new_block(handle
, inode
, path
, newext
, &err
);
893 bh
= sb_getblk(inode
->i_sb
, newblock
);
896 ext4_std_error(inode
->i_sb
, err
);
901 err
= ext4_journal_get_create_access(handle
, bh
);
907 /* move top-level index/leaf into new block */
908 memmove(bh
->b_data
, curp
->p_hdr
, sizeof(EXT4_I(inode
)->i_data
));
910 /* set size of new block */
911 neh
= ext_block_hdr(bh
);
912 /* old root could have indexes or leaves
913 * so calculate e_max right way */
914 if (ext_depth(inode
))
915 neh
->eh_max
= cpu_to_le16(ext4_ext_space_block_idx(inode
));
917 neh
->eh_max
= cpu_to_le16(ext4_ext_space_block(inode
));
918 neh
->eh_magic
= EXT4_EXT_MAGIC
;
919 set_buffer_uptodate(bh
);
922 err
= ext4_journal_dirty_metadata(handle
, bh
);
926 /* create index in new top-level index: num,max,pointer */
927 err
= ext4_ext_get_access(handle
, inode
, curp
);
931 curp
->p_hdr
->eh_magic
= EXT4_EXT_MAGIC
;
932 curp
->p_hdr
->eh_max
= cpu_to_le16(ext4_ext_space_root_idx(inode
));
933 curp
->p_hdr
->eh_entries
= cpu_to_le16(1);
934 curp
->p_idx
= EXT_FIRST_INDEX(curp
->p_hdr
);
936 if (path
[0].p_hdr
->eh_depth
)
937 curp
->p_idx
->ei_block
=
938 EXT_FIRST_INDEX(path
[0].p_hdr
)->ei_block
;
940 curp
->p_idx
->ei_block
=
941 EXT_FIRST_EXTENT(path
[0].p_hdr
)->ee_block
;
942 ext4_idx_store_pblock(curp
->p_idx
, newblock
);
944 neh
= ext_inode_hdr(inode
);
945 fidx
= EXT_FIRST_INDEX(neh
);
946 ext_debug("new root: num %d(%d), lblock %d, ptr %llu\n",
947 le16_to_cpu(neh
->eh_entries
), le16_to_cpu(neh
->eh_max
),
948 le32_to_cpu(fidx
->ei_block
), idx_pblock(fidx
));
950 neh
->eh_depth
= cpu_to_le16(path
->p_depth
+ 1);
951 err
= ext4_ext_dirty(handle
, inode
, curp
);
959 * ext4_ext_create_new_leaf:
960 * finds empty index and adds new leaf.
961 * if no free index is found, then it requests in-depth growing.
963 static int ext4_ext_create_new_leaf(handle_t
*handle
, struct inode
*inode
,
964 struct ext4_ext_path
*path
,
965 struct ext4_extent
*newext
)
967 struct ext4_ext_path
*curp
;
968 int depth
, i
, err
= 0;
971 i
= depth
= ext_depth(inode
);
973 /* walk up to the tree and look for free index entry */
975 while (i
> 0 && !EXT_HAS_FREE_INDEX(curp
)) {
980 /* we use already allocated block for index block,
981 * so subsequent data blocks should be contiguous */
982 if (EXT_HAS_FREE_INDEX(curp
)) {
983 /* if we found index with free entry, then use that
984 * entry: create all needed subtree and add new leaf */
985 err
= ext4_ext_split(handle
, inode
, path
, newext
, i
);
990 ext4_ext_drop_refs(path
);
991 path
= ext4_ext_find_extent(inode
,
992 (ext4_lblk_t
)le32_to_cpu(newext
->ee_block
),
997 /* tree is full, time to grow in depth */
998 err
= ext4_ext_grow_indepth(handle
, inode
, path
, newext
);
1003 ext4_ext_drop_refs(path
);
1004 path
= ext4_ext_find_extent(inode
,
1005 (ext4_lblk_t
)le32_to_cpu(newext
->ee_block
),
1008 err
= PTR_ERR(path
);
1013 * only first (depth 0 -> 1) produces free space;
1014 * in all other cases we have to split the grown tree
1016 depth
= ext_depth(inode
);
1017 if (path
[depth
].p_hdr
->eh_entries
== path
[depth
].p_hdr
->eh_max
) {
1018 /* now we need to split */
1028 * search the closest allocated block to the left for *logical
1029 * and returns it at @logical + it's physical address at @phys
1030 * if *logical is the smallest allocated block, the function
1031 * returns 0 at @phys
1032 * return value contains 0 (success) or error code
1035 ext4_ext_search_left(struct inode
*inode
, struct ext4_ext_path
*path
,
1036 ext4_lblk_t
*logical
, ext4_fsblk_t
*phys
)
1038 struct ext4_extent_idx
*ix
;
1039 struct ext4_extent
*ex
;
1042 BUG_ON(path
== NULL
);
1043 depth
= path
->p_depth
;
1046 if (depth
== 0 && path
->p_ext
== NULL
)
1049 /* usually extent in the path covers blocks smaller
1050 * then *logical, but it can be that extent is the
1051 * first one in the file */
1053 ex
= path
[depth
].p_ext
;
1054 ee_len
= ext4_ext_get_actual_len(ex
);
1055 if (*logical
< le32_to_cpu(ex
->ee_block
)) {
1056 BUG_ON(EXT_FIRST_EXTENT(path
[depth
].p_hdr
) != ex
);
1057 while (--depth
>= 0) {
1058 ix
= path
[depth
].p_idx
;
1059 BUG_ON(ix
!= EXT_FIRST_INDEX(path
[depth
].p_hdr
));
1064 BUG_ON(*logical
< (le32_to_cpu(ex
->ee_block
) + ee_len
));
1066 *logical
= le32_to_cpu(ex
->ee_block
) + ee_len
- 1;
1067 *phys
= ext_pblock(ex
) + ee_len
- 1;
1072 * search the closest allocated block to the right for *logical
1073 * and returns it at @logical + it's physical address at @phys
1074 * if *logical is the smallest allocated block, the function
1075 * returns 0 at @phys
1076 * return value contains 0 (success) or error code
1079 ext4_ext_search_right(struct inode
*inode
, struct ext4_ext_path
*path
,
1080 ext4_lblk_t
*logical
, ext4_fsblk_t
*phys
)
1082 struct buffer_head
*bh
= NULL
;
1083 struct ext4_extent_header
*eh
;
1084 struct ext4_extent_idx
*ix
;
1085 struct ext4_extent
*ex
;
1089 BUG_ON(path
== NULL
);
1090 depth
= path
->p_depth
;
1093 if (depth
== 0 && path
->p_ext
== NULL
)
1096 /* usually extent in the path covers blocks smaller
1097 * then *logical, but it can be that extent is the
1098 * first one in the file */
1100 ex
= path
[depth
].p_ext
;
1101 ee_len
= ext4_ext_get_actual_len(ex
);
1102 if (*logical
< le32_to_cpu(ex
->ee_block
)) {
1103 BUG_ON(EXT_FIRST_EXTENT(path
[depth
].p_hdr
) != ex
);
1104 while (--depth
>= 0) {
1105 ix
= path
[depth
].p_idx
;
1106 BUG_ON(ix
!= EXT_FIRST_INDEX(path
[depth
].p_hdr
));
1108 *logical
= le32_to_cpu(ex
->ee_block
);
1109 *phys
= ext_pblock(ex
);
1113 BUG_ON(*logical
< (le32_to_cpu(ex
->ee_block
) + ee_len
));
1115 if (ex
!= EXT_LAST_EXTENT(path
[depth
].p_hdr
)) {
1116 /* next allocated block in this leaf */
1118 *logical
= le32_to_cpu(ex
->ee_block
);
1119 *phys
= ext_pblock(ex
);
1123 /* go up and search for index to the right */
1124 while (--depth
>= 0) {
1125 ix
= path
[depth
].p_idx
;
1126 if (ix
!= EXT_LAST_INDEX(path
[depth
].p_hdr
))
1131 /* we've gone up to the root and
1132 * found no index to the right */
1136 /* we've found index to the right, let's
1137 * follow it and find the closest allocated
1138 * block to the right */
1140 block
= idx_pblock(ix
);
1141 while (++depth
< path
->p_depth
) {
1142 bh
= sb_bread(inode
->i_sb
, block
);
1145 eh
= ext_block_hdr(bh
);
1146 if (ext4_ext_check_header(inode
, eh
, depth
)) {
1150 ix
= EXT_FIRST_INDEX(eh
);
1151 block
= idx_pblock(ix
);
1155 bh
= sb_bread(inode
->i_sb
, block
);
1158 eh
= ext_block_hdr(bh
);
1159 if (ext4_ext_check_header(inode
, eh
, path
->p_depth
- depth
)) {
1163 ex
= EXT_FIRST_EXTENT(eh
);
1164 *logical
= le32_to_cpu(ex
->ee_block
);
1165 *phys
= ext_pblock(ex
);
1172 * ext4_ext_next_allocated_block:
1173 * returns allocated block in subsequent extent or EXT_MAX_BLOCK.
1174 * NOTE: it considers block number from index entry as
1175 * allocated block. Thus, index entries have to be consistent
1179 ext4_ext_next_allocated_block(struct ext4_ext_path
*path
)
1183 BUG_ON(path
== NULL
);
1184 depth
= path
->p_depth
;
1186 if (depth
== 0 && path
->p_ext
== NULL
)
1187 return EXT_MAX_BLOCK
;
1189 while (depth
>= 0) {
1190 if (depth
== path
->p_depth
) {
1192 if (path
[depth
].p_ext
!=
1193 EXT_LAST_EXTENT(path
[depth
].p_hdr
))
1194 return le32_to_cpu(path
[depth
].p_ext
[1].ee_block
);
1197 if (path
[depth
].p_idx
!=
1198 EXT_LAST_INDEX(path
[depth
].p_hdr
))
1199 return le32_to_cpu(path
[depth
].p_idx
[1].ei_block
);
1204 return EXT_MAX_BLOCK
;
1208 * ext4_ext_next_leaf_block:
1209 * returns first allocated block from next leaf or EXT_MAX_BLOCK
1211 static ext4_lblk_t
ext4_ext_next_leaf_block(struct inode
*inode
,
1212 struct ext4_ext_path
*path
)
1216 BUG_ON(path
== NULL
);
1217 depth
= path
->p_depth
;
1219 /* zero-tree has no leaf blocks at all */
1221 return EXT_MAX_BLOCK
;
1223 /* go to index block */
1226 while (depth
>= 0) {
1227 if (path
[depth
].p_idx
!=
1228 EXT_LAST_INDEX(path
[depth
].p_hdr
))
1229 return (ext4_lblk_t
)
1230 le32_to_cpu(path
[depth
].p_idx
[1].ei_block
);
1234 return EXT_MAX_BLOCK
;
1238 * ext4_ext_correct_indexes:
1239 * if leaf gets modified and modified extent is first in the leaf,
1240 * then we have to correct all indexes above.
1241 * TODO: do we need to correct tree in all cases?
1243 static int ext4_ext_correct_indexes(handle_t
*handle
, struct inode
*inode
,
1244 struct ext4_ext_path
*path
)
1246 struct ext4_extent_header
*eh
;
1247 int depth
= ext_depth(inode
);
1248 struct ext4_extent
*ex
;
1252 eh
= path
[depth
].p_hdr
;
1253 ex
= path
[depth
].p_ext
;
1258 /* there is no tree at all */
1262 if (ex
!= EXT_FIRST_EXTENT(eh
)) {
1263 /* we correct tree if first leaf got modified only */
1268 * TODO: we need correction if border is smaller than current one
1271 border
= path
[depth
].p_ext
->ee_block
;
1272 err
= ext4_ext_get_access(handle
, inode
, path
+ k
);
1275 path
[k
].p_idx
->ei_block
= border
;
1276 err
= ext4_ext_dirty(handle
, inode
, path
+ k
);
1281 /* change all left-side indexes */
1282 if (path
[k
+1].p_idx
!= EXT_FIRST_INDEX(path
[k
+1].p_hdr
))
1284 err
= ext4_ext_get_access(handle
, inode
, path
+ k
);
1287 path
[k
].p_idx
->ei_block
= border
;
1288 err
= ext4_ext_dirty(handle
, inode
, path
+ k
);
1297 ext4_can_extents_be_merged(struct inode
*inode
, struct ext4_extent
*ex1
,
1298 struct ext4_extent
*ex2
)
1300 unsigned short ext1_ee_len
, ext2_ee_len
, max_len
;
1303 * Make sure that either both extents are uninitialized, or
1306 if (ext4_ext_is_uninitialized(ex1
) ^ ext4_ext_is_uninitialized(ex2
))
1309 if (ext4_ext_is_uninitialized(ex1
))
1310 max_len
= EXT_UNINIT_MAX_LEN
;
1312 max_len
= EXT_INIT_MAX_LEN
;
1314 ext1_ee_len
= ext4_ext_get_actual_len(ex1
);
1315 ext2_ee_len
= ext4_ext_get_actual_len(ex2
);
1317 if (le32_to_cpu(ex1
->ee_block
) + ext1_ee_len
!=
1318 le32_to_cpu(ex2
->ee_block
))
1322 * To allow future support for preallocated extents to be added
1323 * as an RO_COMPAT feature, refuse to merge to extents if
1324 * this can result in the top bit of ee_len being set.
1326 if (ext1_ee_len
+ ext2_ee_len
> max_len
)
1328 #ifdef AGGRESSIVE_TEST
1329 if (ext1_ee_len
>= 4)
1333 if (ext_pblock(ex1
) + ext1_ee_len
== ext_pblock(ex2
))
1339 * This function tries to merge the "ex" extent to the next extent in the tree.
1340 * It always tries to merge towards right. If you want to merge towards
1341 * left, pass "ex - 1" as argument instead of "ex".
1342 * Returns 0 if the extents (ex and ex+1) were _not_ merged and returns
1343 * 1 if they got merged.
1345 int ext4_ext_try_to_merge(struct inode
*inode
,
1346 struct ext4_ext_path
*path
,
1347 struct ext4_extent
*ex
)
1349 struct ext4_extent_header
*eh
;
1350 unsigned int depth
, len
;
1352 int uninitialized
= 0;
1354 depth
= ext_depth(inode
);
1355 BUG_ON(path
[depth
].p_hdr
== NULL
);
1356 eh
= path
[depth
].p_hdr
;
1358 while (ex
< EXT_LAST_EXTENT(eh
)) {
1359 if (!ext4_can_extents_be_merged(inode
, ex
, ex
+ 1))
1361 /* merge with next extent! */
1362 if (ext4_ext_is_uninitialized(ex
))
1364 ex
->ee_len
= cpu_to_le16(ext4_ext_get_actual_len(ex
)
1365 + ext4_ext_get_actual_len(ex
+ 1));
1367 ext4_ext_mark_uninitialized(ex
);
1369 if (ex
+ 1 < EXT_LAST_EXTENT(eh
)) {
1370 len
= (EXT_LAST_EXTENT(eh
) - ex
- 1)
1371 * sizeof(struct ext4_extent
);
1372 memmove(ex
+ 1, ex
+ 2, len
);
1374 le16_add_cpu(&eh
->eh_entries
, -1);
1376 WARN_ON(eh
->eh_entries
== 0);
1377 if (!eh
->eh_entries
)
1378 ext4_error(inode
->i_sb
, "ext4_ext_try_to_merge",
1379 "inode#%lu, eh->eh_entries = 0!", inode
->i_ino
);
1386 * check if a portion of the "newext" extent overlaps with an
1389 * If there is an overlap discovered, it updates the length of the newext
1390 * such that there will be no overlap, and then returns 1.
1391 * If there is no overlap found, it returns 0.
1393 unsigned int ext4_ext_check_overlap(struct inode
*inode
,
1394 struct ext4_extent
*newext
,
1395 struct ext4_ext_path
*path
)
1398 unsigned int depth
, len1
;
1399 unsigned int ret
= 0;
1401 b1
= le32_to_cpu(newext
->ee_block
);
1402 len1
= ext4_ext_get_actual_len(newext
);
1403 depth
= ext_depth(inode
);
1404 if (!path
[depth
].p_ext
)
1406 b2
= le32_to_cpu(path
[depth
].p_ext
->ee_block
);
1409 * get the next allocated block if the extent in the path
1410 * is before the requested block(s)
1413 b2
= ext4_ext_next_allocated_block(path
);
1414 if (b2
== EXT_MAX_BLOCK
)
1418 /* check for wrap through zero on extent logical start block*/
1419 if (b1
+ len1
< b1
) {
1420 len1
= EXT_MAX_BLOCK
- b1
;
1421 newext
->ee_len
= cpu_to_le16(len1
);
1425 /* check for overlap */
1426 if (b1
+ len1
> b2
) {
1427 newext
->ee_len
= cpu_to_le16(b2
- b1
);
1435 * ext4_ext_insert_extent:
1436 * tries to merge requsted extent into the existing extent or
1437 * inserts requested extent as new one into the tree,
1438 * creating new leaf in the no-space case.
1440 int ext4_ext_insert_extent(handle_t
*handle
, struct inode
*inode
,
1441 struct ext4_ext_path
*path
,
1442 struct ext4_extent
*newext
)
1444 struct ext4_extent_header
* eh
;
1445 struct ext4_extent
*ex
, *fex
;
1446 struct ext4_extent
*nearex
; /* nearest extent */
1447 struct ext4_ext_path
*npath
= NULL
;
1448 int depth
, len
, err
;
1450 unsigned uninitialized
= 0;
1452 BUG_ON(ext4_ext_get_actual_len(newext
) == 0);
1453 depth
= ext_depth(inode
);
1454 ex
= path
[depth
].p_ext
;
1455 BUG_ON(path
[depth
].p_hdr
== NULL
);
1457 /* try to insert block into found extent and return */
1458 if (ex
&& ext4_can_extents_be_merged(inode
, ex
, newext
)) {
1459 ext_debug("append %d block to %d:%d (from %llu)\n",
1460 ext4_ext_get_actual_len(newext
),
1461 le32_to_cpu(ex
->ee_block
),
1462 ext4_ext_get_actual_len(ex
), ext_pblock(ex
));
1463 err
= ext4_ext_get_access(handle
, inode
, path
+ depth
);
1468 * ext4_can_extents_be_merged should have checked that either
1469 * both extents are uninitialized, or both aren't. Thus we
1470 * need to check only one of them here.
1472 if (ext4_ext_is_uninitialized(ex
))
1474 ex
->ee_len
= cpu_to_le16(ext4_ext_get_actual_len(ex
)
1475 + ext4_ext_get_actual_len(newext
));
1477 ext4_ext_mark_uninitialized(ex
);
1478 eh
= path
[depth
].p_hdr
;
1484 depth
= ext_depth(inode
);
1485 eh
= path
[depth
].p_hdr
;
1486 if (le16_to_cpu(eh
->eh_entries
) < le16_to_cpu(eh
->eh_max
))
1489 /* probably next leaf has space for us? */
1490 fex
= EXT_LAST_EXTENT(eh
);
1491 next
= ext4_ext_next_leaf_block(inode
, path
);
1492 if (le32_to_cpu(newext
->ee_block
) > le32_to_cpu(fex
->ee_block
)
1493 && next
!= EXT_MAX_BLOCK
) {
1494 ext_debug("next leaf block - %d\n", next
);
1495 BUG_ON(npath
!= NULL
);
1496 npath
= ext4_ext_find_extent(inode
, next
, NULL
);
1498 return PTR_ERR(npath
);
1499 BUG_ON(npath
->p_depth
!= path
->p_depth
);
1500 eh
= npath
[depth
].p_hdr
;
1501 if (le16_to_cpu(eh
->eh_entries
) < le16_to_cpu(eh
->eh_max
)) {
1502 ext_debug("next leaf isnt full(%d)\n",
1503 le16_to_cpu(eh
->eh_entries
));
1507 ext_debug("next leaf has no free space(%d,%d)\n",
1508 le16_to_cpu(eh
->eh_entries
), le16_to_cpu(eh
->eh_max
));
1512 * There is no free space in the found leaf.
1513 * We're gonna add a new leaf in the tree.
1515 err
= ext4_ext_create_new_leaf(handle
, inode
, path
, newext
);
1518 depth
= ext_depth(inode
);
1519 eh
= path
[depth
].p_hdr
;
1522 nearex
= path
[depth
].p_ext
;
1524 err
= ext4_ext_get_access(handle
, inode
, path
+ depth
);
1529 /* there is no extent in this leaf, create first one */
1530 ext_debug("first extent in the leaf: %d:%llu:%d\n",
1531 le32_to_cpu(newext
->ee_block
),
1533 ext4_ext_get_actual_len(newext
));
1534 path
[depth
].p_ext
= EXT_FIRST_EXTENT(eh
);
1535 } else if (le32_to_cpu(newext
->ee_block
)
1536 > le32_to_cpu(nearex
->ee_block
)) {
1537 /* BUG_ON(newext->ee_block == nearex->ee_block); */
1538 if (nearex
!= EXT_LAST_EXTENT(eh
)) {
1539 len
= EXT_MAX_EXTENT(eh
) - nearex
;
1540 len
= (len
- 1) * sizeof(struct ext4_extent
);
1541 len
= len
< 0 ? 0 : len
;
1542 ext_debug("insert %d:%llu:%d after: nearest 0x%p, "
1543 "move %d from 0x%p to 0x%p\n",
1544 le32_to_cpu(newext
->ee_block
),
1546 ext4_ext_get_actual_len(newext
),
1547 nearex
, len
, nearex
+ 1, nearex
+ 2);
1548 memmove(nearex
+ 2, nearex
+ 1, len
);
1550 path
[depth
].p_ext
= nearex
+ 1;
1552 BUG_ON(newext
->ee_block
== nearex
->ee_block
);
1553 len
= (EXT_MAX_EXTENT(eh
) - nearex
) * sizeof(struct ext4_extent
);
1554 len
= len
< 0 ? 0 : len
;
1555 ext_debug("insert %d:%llu:%d before: nearest 0x%p, "
1556 "move %d from 0x%p to 0x%p\n",
1557 le32_to_cpu(newext
->ee_block
),
1559 ext4_ext_get_actual_len(newext
),
1560 nearex
, len
, nearex
+ 1, nearex
+ 2);
1561 memmove(nearex
+ 1, nearex
, len
);
1562 path
[depth
].p_ext
= nearex
;
1565 le16_add_cpu(&eh
->eh_entries
, 1);
1566 nearex
= path
[depth
].p_ext
;
1567 nearex
->ee_block
= newext
->ee_block
;
1568 ext4_ext_store_pblock(nearex
, ext_pblock(newext
));
1569 nearex
->ee_len
= newext
->ee_len
;
1572 /* try to merge extents to the right */
1573 ext4_ext_try_to_merge(inode
, path
, nearex
);
1575 /* try to merge extents to the left */
1577 /* time to correct all indexes above */
1578 err
= ext4_ext_correct_indexes(handle
, inode
, path
);
1582 err
= ext4_ext_dirty(handle
, inode
, path
+ depth
);
1586 ext4_ext_drop_refs(npath
);
1589 ext4_ext_tree_changed(inode
);
1590 ext4_ext_invalidate_cache(inode
);
1595 ext4_ext_put_in_cache(struct inode
*inode
, ext4_lblk_t block
,
1596 __u32 len
, ext4_fsblk_t start
, int type
)
1598 struct ext4_ext_cache
*cex
;
1600 cex
= &EXT4_I(inode
)->i_cached_extent
;
1601 cex
->ec_type
= type
;
1602 cex
->ec_block
= block
;
1604 cex
->ec_start
= start
;
1608 * ext4_ext_put_gap_in_cache:
1609 * calculate boundaries of the gap that the requested block fits into
1610 * and cache this gap
1613 ext4_ext_put_gap_in_cache(struct inode
*inode
, struct ext4_ext_path
*path
,
1616 int depth
= ext_depth(inode
);
1619 struct ext4_extent
*ex
;
1621 ex
= path
[depth
].p_ext
;
1623 /* there is no extent yet, so gap is [0;-] */
1625 len
= EXT_MAX_BLOCK
;
1626 ext_debug("cache gap(whole file):");
1627 } else if (block
< le32_to_cpu(ex
->ee_block
)) {
1629 len
= le32_to_cpu(ex
->ee_block
) - block
;
1630 ext_debug("cache gap(before): %u [%u:%u]",
1632 le32_to_cpu(ex
->ee_block
),
1633 ext4_ext_get_actual_len(ex
));
1634 } else if (block
>= le32_to_cpu(ex
->ee_block
)
1635 + ext4_ext_get_actual_len(ex
)) {
1637 lblock
= le32_to_cpu(ex
->ee_block
)
1638 + ext4_ext_get_actual_len(ex
);
1640 next
= ext4_ext_next_allocated_block(path
);
1641 ext_debug("cache gap(after): [%u:%u] %u",
1642 le32_to_cpu(ex
->ee_block
),
1643 ext4_ext_get_actual_len(ex
),
1645 BUG_ON(next
== lblock
);
1646 len
= next
- lblock
;
1652 ext_debug(" -> %u:%lu\n", lblock
, len
);
1653 ext4_ext_put_in_cache(inode
, lblock
, len
, 0, EXT4_EXT_CACHE_GAP
);
1657 ext4_ext_in_cache(struct inode
*inode
, ext4_lblk_t block
,
1658 struct ext4_extent
*ex
)
1660 struct ext4_ext_cache
*cex
;
1662 cex
= &EXT4_I(inode
)->i_cached_extent
;
1664 /* has cache valid data? */
1665 if (cex
->ec_type
== EXT4_EXT_CACHE_NO
)
1666 return EXT4_EXT_CACHE_NO
;
1668 BUG_ON(cex
->ec_type
!= EXT4_EXT_CACHE_GAP
&&
1669 cex
->ec_type
!= EXT4_EXT_CACHE_EXTENT
);
1670 if (block
>= cex
->ec_block
&& block
< cex
->ec_block
+ cex
->ec_len
) {
1671 ex
->ee_block
= cpu_to_le32(cex
->ec_block
);
1672 ext4_ext_store_pblock(ex
, cex
->ec_start
);
1673 ex
->ee_len
= cpu_to_le16(cex
->ec_len
);
1674 ext_debug("%u cached by %u:%u:%llu\n",
1676 cex
->ec_block
, cex
->ec_len
, cex
->ec_start
);
1677 return cex
->ec_type
;
1681 return EXT4_EXT_CACHE_NO
;
1686 * removes index from the index block.
1687 * It's used in truncate case only, thus all requests are for
1688 * last index in the block only.
1690 static int ext4_ext_rm_idx(handle_t
*handle
, struct inode
*inode
,
1691 struct ext4_ext_path
*path
)
1693 struct buffer_head
*bh
;
1697 /* free index block */
1699 leaf
= idx_pblock(path
->p_idx
);
1700 BUG_ON(path
->p_hdr
->eh_entries
== 0);
1701 err
= ext4_ext_get_access(handle
, inode
, path
);
1704 le16_add_cpu(&path
->p_hdr
->eh_entries
, -1);
1705 err
= ext4_ext_dirty(handle
, inode
, path
);
1708 ext_debug("index is empty, remove it, free block %llu\n", leaf
);
1709 bh
= sb_find_get_block(inode
->i_sb
, leaf
);
1710 ext4_forget(handle
, 1, inode
, bh
, leaf
);
1711 ext4_free_blocks(handle
, inode
, leaf
, 1, 1);
1716 * ext4_ext_calc_credits_for_insert:
1717 * This routine returns max. credits that the extent tree can consume.
1718 * It should be OK for low-performance paths like ->writepage()
1719 * To allow many writing processes to fit into a single transaction,
1720 * the caller should calculate credits under i_data_sem and
1721 * pass the actual path.
1723 int ext4_ext_calc_credits_for_insert(struct inode
*inode
,
1724 struct ext4_ext_path
*path
)
1729 /* probably there is space in leaf? */
1730 depth
= ext_depth(inode
);
1731 if (le16_to_cpu(path
[depth
].p_hdr
->eh_entries
)
1732 < le16_to_cpu(path
[depth
].p_hdr
->eh_max
))
1737 * given 32-bit logical block (4294967296 blocks), max. tree
1738 * can be 4 levels in depth -- 4 * 340^4 == 53453440000.
1739 * Let's also add one more level for imbalance.
1743 /* allocation of new data block(s) */
1747 * tree can be full, so it would need to grow in depth:
1748 * we need one credit to modify old root, credits for
1749 * new root will be added in split accounting
1754 * Index split can happen, we would need:
1755 * allocate intermediate indexes (bitmap + group)
1756 * + change two blocks at each level, but root (already included)
1758 needed
+= (depth
* 2) + (depth
* 2);
1760 /* any allocation modifies superblock */
1766 static int ext4_remove_blocks(handle_t
*handle
, struct inode
*inode
,
1767 struct ext4_extent
*ex
,
1768 ext4_lblk_t from
, ext4_lblk_t to
)
1770 struct buffer_head
*bh
;
1771 unsigned short ee_len
= ext4_ext_get_actual_len(ex
);
1772 int i
, metadata
= 0;
1774 if (S_ISDIR(inode
->i_mode
) || S_ISLNK(inode
->i_mode
))
1776 #ifdef EXTENTS_STATS
1778 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
1779 spin_lock(&sbi
->s_ext_stats_lock
);
1780 sbi
->s_ext_blocks
+= ee_len
;
1781 sbi
->s_ext_extents
++;
1782 if (ee_len
< sbi
->s_ext_min
)
1783 sbi
->s_ext_min
= ee_len
;
1784 if (ee_len
> sbi
->s_ext_max
)
1785 sbi
->s_ext_max
= ee_len
;
1786 if (ext_depth(inode
) > sbi
->s_depth_max
)
1787 sbi
->s_depth_max
= ext_depth(inode
);
1788 spin_unlock(&sbi
->s_ext_stats_lock
);
1791 if (from
>= le32_to_cpu(ex
->ee_block
)
1792 && to
== le32_to_cpu(ex
->ee_block
) + ee_len
- 1) {
1797 num
= le32_to_cpu(ex
->ee_block
) + ee_len
- from
;
1798 start
= ext_pblock(ex
) + ee_len
- num
;
1799 ext_debug("free last %u blocks starting %llu\n", num
, start
);
1800 for (i
= 0; i
< num
; i
++) {
1801 bh
= sb_find_get_block(inode
->i_sb
, start
+ i
);
1802 ext4_forget(handle
, 0, inode
, bh
, start
+ i
);
1804 ext4_free_blocks(handle
, inode
, start
, num
, metadata
);
1805 } else if (from
== le32_to_cpu(ex
->ee_block
)
1806 && to
<= le32_to_cpu(ex
->ee_block
) + ee_len
- 1) {
1807 printk(KERN_INFO
"strange request: removal %u-%u from %u:%u\n",
1808 from
, to
, le32_to_cpu(ex
->ee_block
), ee_len
);
1810 printk(KERN_INFO
"strange request: removal(2) "
1811 "%u-%u from %u:%u\n",
1812 from
, to
, le32_to_cpu(ex
->ee_block
), ee_len
);
1818 ext4_ext_rm_leaf(handle_t
*handle
, struct inode
*inode
,
1819 struct ext4_ext_path
*path
, ext4_lblk_t start
)
1821 int err
= 0, correct_index
= 0;
1822 int depth
= ext_depth(inode
), credits
;
1823 struct ext4_extent_header
*eh
;
1824 ext4_lblk_t a
, b
, block
;
1826 ext4_lblk_t ex_ee_block
;
1827 unsigned short ex_ee_len
;
1828 unsigned uninitialized
= 0;
1829 struct ext4_extent
*ex
;
1831 /* the header must be checked already in ext4_ext_remove_space() */
1832 ext_debug("truncate since %u in leaf\n", start
);
1833 if (!path
[depth
].p_hdr
)
1834 path
[depth
].p_hdr
= ext_block_hdr(path
[depth
].p_bh
);
1835 eh
= path
[depth
].p_hdr
;
1838 /* find where to start removing */
1839 ex
= EXT_LAST_EXTENT(eh
);
1841 ex_ee_block
= le32_to_cpu(ex
->ee_block
);
1842 if (ext4_ext_is_uninitialized(ex
))
1844 ex_ee_len
= ext4_ext_get_actual_len(ex
);
1846 while (ex
>= EXT_FIRST_EXTENT(eh
) &&
1847 ex_ee_block
+ ex_ee_len
> start
) {
1848 ext_debug("remove ext %lu:%u\n", ex_ee_block
, ex_ee_len
);
1849 path
[depth
].p_ext
= ex
;
1851 a
= ex_ee_block
> start
? ex_ee_block
: start
;
1852 b
= ex_ee_block
+ ex_ee_len
- 1 < EXT_MAX_BLOCK
?
1853 ex_ee_block
+ ex_ee_len
- 1 : EXT_MAX_BLOCK
;
1855 ext_debug(" border %u:%u\n", a
, b
);
1857 if (a
!= ex_ee_block
&& b
!= ex_ee_block
+ ex_ee_len
- 1) {
1861 } else if (a
!= ex_ee_block
) {
1862 /* remove tail of the extent */
1863 block
= ex_ee_block
;
1865 } else if (b
!= ex_ee_block
+ ex_ee_len
- 1) {
1866 /* remove head of the extent */
1869 /* there is no "make a hole" API yet */
1872 /* remove whole extent: excellent! */
1873 block
= ex_ee_block
;
1875 BUG_ON(a
!= ex_ee_block
);
1876 BUG_ON(b
!= ex_ee_block
+ ex_ee_len
- 1);
1879 /* at present, extent can't cross block group: */
1880 /* leaf + bitmap + group desc + sb + inode */
1882 if (ex
== EXT_FIRST_EXTENT(eh
)) {
1884 credits
+= (ext_depth(inode
)) + 1;
1887 credits
+= 2 * EXT4_QUOTA_TRANS_BLOCKS(inode
->i_sb
);
1890 err
= ext4_ext_journal_restart(handle
, credits
);
1894 err
= ext4_ext_get_access(handle
, inode
, path
+ depth
);
1898 err
= ext4_remove_blocks(handle
, inode
, ex
, a
, b
);
1903 /* this extent is removed; mark slot entirely unused */
1904 ext4_ext_store_pblock(ex
, 0);
1905 le16_add_cpu(&eh
->eh_entries
, -1);
1908 ex
->ee_block
= cpu_to_le32(block
);
1909 ex
->ee_len
= cpu_to_le16(num
);
1911 * Do not mark uninitialized if all the blocks in the
1912 * extent have been removed.
1914 if (uninitialized
&& num
)
1915 ext4_ext_mark_uninitialized(ex
);
1917 err
= ext4_ext_dirty(handle
, inode
, path
+ depth
);
1921 ext_debug("new extent: %u:%u:%llu\n", block
, num
,
1924 ex_ee_block
= le32_to_cpu(ex
->ee_block
);
1925 ex_ee_len
= ext4_ext_get_actual_len(ex
);
1928 if (correct_index
&& eh
->eh_entries
)
1929 err
= ext4_ext_correct_indexes(handle
, inode
, path
);
1931 /* if this leaf is free, then we should
1932 * remove it from index block above */
1933 if (err
== 0 && eh
->eh_entries
== 0 && path
[depth
].p_bh
!= NULL
)
1934 err
= ext4_ext_rm_idx(handle
, inode
, path
+ depth
);
1941 * ext4_ext_more_to_rm:
1942 * returns 1 if current index has to be freed (even partial)
1945 ext4_ext_more_to_rm(struct ext4_ext_path
*path
)
1947 BUG_ON(path
->p_idx
== NULL
);
1949 if (path
->p_idx
< EXT_FIRST_INDEX(path
->p_hdr
))
1953 * if truncate on deeper level happened, it wasn't partial,
1954 * so we have to consider current index for truncation
1956 if (le16_to_cpu(path
->p_hdr
->eh_entries
) == path
->p_block
)
1961 static int ext4_ext_remove_space(struct inode
*inode
, ext4_lblk_t start
)
1963 struct super_block
*sb
= inode
->i_sb
;
1964 int depth
= ext_depth(inode
);
1965 struct ext4_ext_path
*path
;
1969 ext_debug("truncate since %u\n", start
);
1971 /* probably first extent we're gonna free will be last in block */
1972 handle
= ext4_journal_start(inode
, depth
+ 1);
1974 return PTR_ERR(handle
);
1976 ext4_ext_invalidate_cache(inode
);
1979 * We start scanning from right side, freeing all the blocks
1980 * after i_size and walking into the tree depth-wise.
1982 path
= kzalloc(sizeof(struct ext4_ext_path
) * (depth
+ 1), GFP_NOFS
);
1984 ext4_journal_stop(handle
);
1987 path
[0].p_hdr
= ext_inode_hdr(inode
);
1988 if (ext4_ext_check_header(inode
, path
[0].p_hdr
, depth
)) {
1992 path
[0].p_depth
= depth
;
1994 while (i
>= 0 && err
== 0) {
1996 /* this is leaf block */
1997 err
= ext4_ext_rm_leaf(handle
, inode
, path
, start
);
1998 /* root level has p_bh == NULL, brelse() eats this */
1999 brelse(path
[i
].p_bh
);
2000 path
[i
].p_bh
= NULL
;
2005 /* this is index block */
2006 if (!path
[i
].p_hdr
) {
2007 ext_debug("initialize header\n");
2008 path
[i
].p_hdr
= ext_block_hdr(path
[i
].p_bh
);
2011 if (!path
[i
].p_idx
) {
2012 /* this level hasn't been touched yet */
2013 path
[i
].p_idx
= EXT_LAST_INDEX(path
[i
].p_hdr
);
2014 path
[i
].p_block
= le16_to_cpu(path
[i
].p_hdr
->eh_entries
)+1;
2015 ext_debug("init index ptr: hdr 0x%p, num %d\n",
2017 le16_to_cpu(path
[i
].p_hdr
->eh_entries
));
2019 /* we were already here, see at next index */
2023 ext_debug("level %d - index, first 0x%p, cur 0x%p\n",
2024 i
, EXT_FIRST_INDEX(path
[i
].p_hdr
),
2026 if (ext4_ext_more_to_rm(path
+ i
)) {
2027 struct buffer_head
*bh
;
2028 /* go to the next level */
2029 ext_debug("move to level %d (block %llu)\n",
2030 i
+ 1, idx_pblock(path
[i
].p_idx
));
2031 memset(path
+ i
+ 1, 0, sizeof(*path
));
2032 bh
= sb_bread(sb
, idx_pblock(path
[i
].p_idx
));
2034 /* should we reset i_size? */
2038 if (WARN_ON(i
+ 1 > depth
)) {
2042 if (ext4_ext_check_header(inode
, ext_block_hdr(bh
),
2047 path
[i
+ 1].p_bh
= bh
;
2049 /* save actual number of indexes since this
2050 * number is changed at the next iteration */
2051 path
[i
].p_block
= le16_to_cpu(path
[i
].p_hdr
->eh_entries
);
2054 /* we finished processing this index, go up */
2055 if (path
[i
].p_hdr
->eh_entries
== 0 && i
> 0) {
2056 /* index is empty, remove it;
2057 * handle must be already prepared by the
2058 * truncatei_leaf() */
2059 err
= ext4_ext_rm_idx(handle
, inode
, path
+ i
);
2061 /* root level has p_bh == NULL, brelse() eats this */
2062 brelse(path
[i
].p_bh
);
2063 path
[i
].p_bh
= NULL
;
2065 ext_debug("return to level %d\n", i
);
2069 /* TODO: flexible tree reduction should be here */
2070 if (path
->p_hdr
->eh_entries
== 0) {
2072 * truncate to zero freed all the tree,
2073 * so we need to correct eh_depth
2075 err
= ext4_ext_get_access(handle
, inode
, path
);
2077 ext_inode_hdr(inode
)->eh_depth
= 0;
2078 ext_inode_hdr(inode
)->eh_max
=
2079 cpu_to_le16(ext4_ext_space_root(inode
));
2080 err
= ext4_ext_dirty(handle
, inode
, path
);
2084 ext4_ext_tree_changed(inode
);
2085 ext4_ext_drop_refs(path
);
2087 ext4_journal_stop(handle
);
2093 * called at mount time
2095 void ext4_ext_init(struct super_block
*sb
)
2098 * possible initialization would be here
2101 if (test_opt(sb
, EXTENTS
)) {
2102 printk("EXT4-fs: file extents enabled");
2103 #ifdef AGGRESSIVE_TEST
2104 printk(", aggressive tests");
2106 #ifdef CHECK_BINSEARCH
2107 printk(", check binsearch");
2109 #ifdef EXTENTS_STATS
2113 #ifdef EXTENTS_STATS
2114 spin_lock_init(&EXT4_SB(sb
)->s_ext_stats_lock
);
2115 EXT4_SB(sb
)->s_ext_min
= 1 << 30;
2116 EXT4_SB(sb
)->s_ext_max
= 0;
2122 * called at umount time
2124 void ext4_ext_release(struct super_block
*sb
)
2126 if (!test_opt(sb
, EXTENTS
))
2129 #ifdef EXTENTS_STATS
2130 if (EXT4_SB(sb
)->s_ext_blocks
&& EXT4_SB(sb
)->s_ext_extents
) {
2131 struct ext4_sb_info
*sbi
= EXT4_SB(sb
);
2132 printk(KERN_ERR
"EXT4-fs: %lu blocks in %lu extents (%lu ave)\n",
2133 sbi
->s_ext_blocks
, sbi
->s_ext_extents
,
2134 sbi
->s_ext_blocks
/ sbi
->s_ext_extents
);
2135 printk(KERN_ERR
"EXT4-fs: extents: %lu min, %lu max, max depth %lu\n",
2136 sbi
->s_ext_min
, sbi
->s_ext_max
, sbi
->s_depth_max
);
2141 static void bi_complete(struct bio
*bio
, int error
)
2143 complete((struct completion
*)bio
->bi_private
);
2146 /* FIXME!! we need to try to merge to left or right after zero-out */
2147 static int ext4_ext_zeroout(struct inode
*inode
, struct ext4_extent
*ex
)
2151 int blkbits
, blocksize
;
2153 struct completion event
;
2154 unsigned int ee_len
, len
, done
, offset
;
2157 blkbits
= inode
->i_blkbits
;
2158 blocksize
= inode
->i_sb
->s_blocksize
;
2159 ee_len
= ext4_ext_get_actual_len(ex
);
2160 ee_pblock
= ext_pblock(ex
);
2162 /* convert ee_pblock to 512 byte sectors */
2163 ee_pblock
= ee_pblock
<< (blkbits
- 9);
2165 while (ee_len
> 0) {
2167 if (ee_len
> BIO_MAX_PAGES
)
2168 len
= BIO_MAX_PAGES
;
2172 bio
= bio_alloc(GFP_NOIO
, len
);
2175 bio
->bi_sector
= ee_pblock
;
2176 bio
->bi_bdev
= inode
->i_sb
->s_bdev
;
2180 while (done
< len
) {
2181 ret
= bio_add_page(bio
, ZERO_PAGE(0),
2183 if (ret
!= blocksize
) {
2185 * We can't add any more pages because of
2186 * hardware limitations. Start a new bio.
2191 offset
+= blocksize
;
2192 if (offset
>= PAGE_CACHE_SIZE
)
2196 init_completion(&event
);
2197 bio
->bi_private
= &event
;
2198 bio
->bi_end_io
= bi_complete
;
2199 submit_bio(WRITE
, bio
);
2200 wait_for_completion(&event
);
2202 if (test_bit(BIO_UPTODATE
, &bio
->bi_flags
))
2210 ee_pblock
+= done
<< (blkbits
- 9);
2215 #define EXT4_EXT_ZERO_LEN 7
2218 * This function is called by ext4_ext_get_blocks() if someone tries to write
2219 * to an uninitialized extent. It may result in splitting the uninitialized
2220 * extent into multiple extents (upto three - one initialized and two
2222 * There are three possibilities:
2223 * a> There is no split required: Entire extent should be initialized
2224 * b> Splits in two extents: Write is happening at either end of the extent
2225 * c> Splits in three extents: Somone is writing in middle of the extent
2227 static int ext4_ext_convert_to_initialized(handle_t
*handle
,
2228 struct inode
*inode
,
2229 struct ext4_ext_path
*path
,
2231 unsigned long max_blocks
)
2233 struct ext4_extent
*ex
, newex
, orig_ex
;
2234 struct ext4_extent
*ex1
= NULL
;
2235 struct ext4_extent
*ex2
= NULL
;
2236 struct ext4_extent
*ex3
= NULL
;
2237 struct ext4_extent_header
*eh
;
2238 ext4_lblk_t ee_block
;
2239 unsigned int allocated
, ee_len
, depth
;
2240 ext4_fsblk_t newblock
;
2244 depth
= ext_depth(inode
);
2245 eh
= path
[depth
].p_hdr
;
2246 ex
= path
[depth
].p_ext
;
2247 ee_block
= le32_to_cpu(ex
->ee_block
);
2248 ee_len
= ext4_ext_get_actual_len(ex
);
2249 allocated
= ee_len
- (iblock
- ee_block
);
2250 newblock
= iblock
- ee_block
+ ext_pblock(ex
);
2252 orig_ex
.ee_block
= ex
->ee_block
;
2253 orig_ex
.ee_len
= cpu_to_le16(ee_len
);
2254 ext4_ext_store_pblock(&orig_ex
, ext_pblock(ex
));
2256 err
= ext4_ext_get_access(handle
, inode
, path
+ depth
);
2259 /* If extent has less than 2*EXT4_EXT_ZERO_LEN zerout directly */
2260 if (ee_len
<= 2*EXT4_EXT_ZERO_LEN
) {
2261 err
= ext4_ext_zeroout(inode
, &orig_ex
);
2263 goto fix_extent_len
;
2264 /* update the extent length and mark as initialized */
2265 ex
->ee_block
= orig_ex
.ee_block
;
2266 ex
->ee_len
= orig_ex
.ee_len
;
2267 ext4_ext_store_pblock(ex
, ext_pblock(&orig_ex
));
2268 ext4_ext_dirty(handle
, inode
, path
+ depth
);
2269 /* zeroed the full extent */
2273 /* ex1: ee_block to iblock - 1 : uninitialized */
2274 if (iblock
> ee_block
) {
2276 ex1
->ee_len
= cpu_to_le16(iblock
- ee_block
);
2277 ext4_ext_mark_uninitialized(ex1
);
2281 * for sanity, update the length of the ex2 extent before
2282 * we insert ex3, if ex1 is NULL. This is to avoid temporary
2283 * overlap of blocks.
2285 if (!ex1
&& allocated
> max_blocks
)
2286 ex2
->ee_len
= cpu_to_le16(max_blocks
);
2287 /* ex3: to ee_block + ee_len : uninitialised */
2288 if (allocated
> max_blocks
) {
2289 unsigned int newdepth
;
2290 /* If extent has less than EXT4_EXT_ZERO_LEN zerout directly */
2291 if (allocated
<= EXT4_EXT_ZERO_LEN
) {
2292 /* Mark first half uninitialized.
2293 * Mark second half initialized and zero out the
2294 * initialized extent
2296 ex
->ee_block
= orig_ex
.ee_block
;
2297 ex
->ee_len
= cpu_to_le16(ee_len
- allocated
);
2298 ext4_ext_mark_uninitialized(ex
);
2299 ext4_ext_store_pblock(ex
, ext_pblock(&orig_ex
));
2300 ext4_ext_dirty(handle
, inode
, path
+ depth
);
2303 ex3
->ee_block
= cpu_to_le32(iblock
);
2304 ext4_ext_store_pblock(ex3
, newblock
);
2305 ex3
->ee_len
= cpu_to_le16(allocated
);
2306 err
= ext4_ext_insert_extent(handle
, inode
, path
, ex3
);
2307 if (err
== -ENOSPC
) {
2308 err
= ext4_ext_zeroout(inode
, &orig_ex
);
2310 goto fix_extent_len
;
2311 ex
->ee_block
= orig_ex
.ee_block
;
2312 ex
->ee_len
= orig_ex
.ee_len
;
2313 ext4_ext_store_pblock(ex
, ext_pblock(&orig_ex
));
2314 ext4_ext_dirty(handle
, inode
, path
+ depth
);
2315 /* zeroed the full extent */
2319 goto fix_extent_len
;
2322 * We need to zero out the second half because
2323 * an fallocate request can update file size and
2324 * converting the second half to initialized extent
2325 * implies that we can leak some junk data to user
2328 err
= ext4_ext_zeroout(inode
, ex3
);
2331 * We should actually mark the
2332 * second half as uninit and return error
2333 * Insert would have changed the extent
2335 depth
= ext_depth(inode
);
2336 ext4_ext_drop_refs(path
);
2337 path
= ext4_ext_find_extent(inode
,
2340 err
= PTR_ERR(path
);
2343 ex
= path
[depth
].p_ext
;
2344 err
= ext4_ext_get_access(handle
, inode
,
2348 ext4_ext_mark_uninitialized(ex
);
2349 ext4_ext_dirty(handle
, inode
, path
+ depth
);
2353 /* zeroed the second half */
2357 ex3
->ee_block
= cpu_to_le32(iblock
+ max_blocks
);
2358 ext4_ext_store_pblock(ex3
, newblock
+ max_blocks
);
2359 ex3
->ee_len
= cpu_to_le16(allocated
- max_blocks
);
2360 ext4_ext_mark_uninitialized(ex3
);
2361 err
= ext4_ext_insert_extent(handle
, inode
, path
, ex3
);
2362 if (err
== -ENOSPC
) {
2363 err
= ext4_ext_zeroout(inode
, &orig_ex
);
2365 goto fix_extent_len
;
2366 /* update the extent length and mark as initialized */
2367 ex
->ee_block
= orig_ex
.ee_block
;
2368 ex
->ee_len
= orig_ex
.ee_len
;
2369 ext4_ext_store_pblock(ex
, ext_pblock(&orig_ex
));
2370 ext4_ext_dirty(handle
, inode
, path
+ depth
);
2371 /* zeroed the full extent */
2375 goto fix_extent_len
;
2377 * The depth, and hence eh & ex might change
2378 * as part of the insert above.
2380 newdepth
= ext_depth(inode
);
2382 * update the extent length after successfull insert of the
2385 orig_ex
.ee_len
= cpu_to_le16(ee_len
-
2386 ext4_ext_get_actual_len(ex3
));
2387 if (newdepth
!= depth
) {
2389 ext4_ext_drop_refs(path
);
2390 path
= ext4_ext_find_extent(inode
, iblock
, path
);
2392 err
= PTR_ERR(path
);
2395 eh
= path
[depth
].p_hdr
;
2396 ex
= path
[depth
].p_ext
;
2400 err
= ext4_ext_get_access(handle
, inode
, path
+ depth
);
2404 allocated
= max_blocks
;
2406 /* If extent has less than EXT4_EXT_ZERO_LEN and we are trying
2407 * to insert a extent in the middle zerout directly
2408 * otherwise give the extent a chance to merge to left
2410 if (le16_to_cpu(orig_ex
.ee_len
) <= EXT4_EXT_ZERO_LEN
&&
2411 iblock
!= ee_block
) {
2412 err
= ext4_ext_zeroout(inode
, &orig_ex
);
2414 goto fix_extent_len
;
2415 /* update the extent length and mark as initialized */
2416 ex
->ee_block
= orig_ex
.ee_block
;
2417 ex
->ee_len
= orig_ex
.ee_len
;
2418 ext4_ext_store_pblock(ex
, ext_pblock(&orig_ex
));
2419 ext4_ext_dirty(handle
, inode
, path
+ depth
);
2420 /* zero out the first half */
2425 * If there was a change of depth as part of the
2426 * insertion of ex3 above, we need to update the length
2427 * of the ex1 extent again here
2429 if (ex1
&& ex1
!= ex
) {
2431 ex1
->ee_len
= cpu_to_le16(iblock
- ee_block
);
2432 ext4_ext_mark_uninitialized(ex1
);
2435 /* ex2: iblock to iblock + maxblocks-1 : initialised */
2436 ex2
->ee_block
= cpu_to_le32(iblock
);
2437 ext4_ext_store_pblock(ex2
, newblock
);
2438 ex2
->ee_len
= cpu_to_le16(allocated
);
2442 * New (initialized) extent starts from the first block
2443 * in the current extent. i.e., ex2 == ex
2444 * We have to see if it can be merged with the extent
2447 if (ex2
> EXT_FIRST_EXTENT(eh
)) {
2449 * To merge left, pass "ex2 - 1" to try_to_merge(),
2450 * since it merges towards right _only_.
2452 ret
= ext4_ext_try_to_merge(inode
, path
, ex2
- 1);
2454 err
= ext4_ext_correct_indexes(handle
, inode
, path
);
2457 depth
= ext_depth(inode
);
2462 * Try to Merge towards right. This might be required
2463 * only when the whole extent is being written to.
2464 * i.e. ex2 == ex and ex3 == NULL.
2467 ret
= ext4_ext_try_to_merge(inode
, path
, ex2
);
2469 err
= ext4_ext_correct_indexes(handle
, inode
, path
);
2474 /* Mark modified extent as dirty */
2475 err
= ext4_ext_dirty(handle
, inode
, path
+ depth
);
2478 err
= ext4_ext_insert_extent(handle
, inode
, path
, &newex
);
2479 if (err
== -ENOSPC
) {
2480 err
= ext4_ext_zeroout(inode
, &orig_ex
);
2482 goto fix_extent_len
;
2483 /* update the extent length and mark as initialized */
2484 ex
->ee_block
= orig_ex
.ee_block
;
2485 ex
->ee_len
= orig_ex
.ee_len
;
2486 ext4_ext_store_pblock(ex
, ext_pblock(&orig_ex
));
2487 ext4_ext_dirty(handle
, inode
, path
+ depth
);
2488 /* zero out the first half */
2491 goto fix_extent_len
;
2493 return err
? err
: allocated
;
2496 ex
->ee_block
= orig_ex
.ee_block
;
2497 ex
->ee_len
= orig_ex
.ee_len
;
2498 ext4_ext_store_pblock(ex
, ext_pblock(&orig_ex
));
2499 ext4_ext_mark_uninitialized(ex
);
2500 ext4_ext_dirty(handle
, inode
, path
+ depth
);
2505 * Block allocation/map/preallocation routine for extents based files
2508 * Need to be called with
2509 * down_read(&EXT4_I(inode)->i_data_sem) if not allocating file system block
2510 * (ie, create is zero). Otherwise down_write(&EXT4_I(inode)->i_data_sem)
2512 * return > 0, number of of blocks already mapped/allocated
2513 * if create == 0 and these are pre-allocated blocks
2514 * buffer head is unmapped
2515 * otherwise blocks are mapped
2517 * return = 0, if plain look up failed (blocks have not been allocated)
2518 * buffer head is unmapped
2520 * return < 0, error case.
2522 int ext4_ext_get_blocks(handle_t
*handle
, struct inode
*inode
,
2524 unsigned long max_blocks
, struct buffer_head
*bh_result
,
2525 int create
, int extend_disksize
)
2527 struct ext4_ext_path
*path
= NULL
;
2528 struct ext4_extent_header
*eh
;
2529 struct ext4_extent newex
, *ex
;
2530 ext4_fsblk_t goal
, newblock
;
2531 int err
= 0, depth
, ret
;
2532 unsigned long allocated
= 0;
2533 struct ext4_allocation_request ar
;
2535 __clear_bit(BH_New
, &bh_result
->b_state
);
2536 ext_debug("blocks %u/%lu requested for inode %u\n",
2537 iblock
, max_blocks
, inode
->i_ino
);
2539 /* check in cache */
2540 goal
= ext4_ext_in_cache(inode
, iblock
, &newex
);
2542 if (goal
== EXT4_EXT_CACHE_GAP
) {
2545 * block isn't allocated yet and
2546 * user doesn't want to allocate it
2550 /* we should allocate requested block */
2551 } else if (goal
== EXT4_EXT_CACHE_EXTENT
) {
2552 /* block is already allocated */
2554 - le32_to_cpu(newex
.ee_block
)
2555 + ext_pblock(&newex
);
2556 /* number of remaining blocks in the extent */
2557 allocated
= ext4_ext_get_actual_len(&newex
) -
2558 (iblock
- le32_to_cpu(newex
.ee_block
));
2565 /* find extent for this block */
2566 path
= ext4_ext_find_extent(inode
, iblock
, NULL
);
2568 err
= PTR_ERR(path
);
2573 depth
= ext_depth(inode
);
2576 * consistent leaf must not be empty;
2577 * this situation is possible, though, _during_ tree modification;
2578 * this is why assert can't be put in ext4_ext_find_extent()
2580 BUG_ON(path
[depth
].p_ext
== NULL
&& depth
!= 0);
2581 eh
= path
[depth
].p_hdr
;
2583 ex
= path
[depth
].p_ext
;
2585 ext4_lblk_t ee_block
= le32_to_cpu(ex
->ee_block
);
2586 ext4_fsblk_t ee_start
= ext_pblock(ex
);
2587 unsigned short ee_len
;
2590 * Uninitialized extents are treated as holes, except that
2591 * we split out initialized portions during a write.
2593 ee_len
= ext4_ext_get_actual_len(ex
);
2594 /* if found extent covers block, simply return it */
2595 if (iblock
>= ee_block
&& iblock
< ee_block
+ ee_len
) {
2596 newblock
= iblock
- ee_block
+ ee_start
;
2597 /* number of remaining blocks in the extent */
2598 allocated
= ee_len
- (iblock
- ee_block
);
2599 ext_debug("%u fit into %lu:%d -> %llu\n", iblock
,
2600 ee_block
, ee_len
, newblock
);
2602 /* Do not put uninitialized extent in the cache */
2603 if (!ext4_ext_is_uninitialized(ex
)) {
2604 ext4_ext_put_in_cache(inode
, ee_block
,
2606 EXT4_EXT_CACHE_EXTENT
);
2609 if (create
== EXT4_CREATE_UNINITIALIZED_EXT
)
2613 * We have blocks reserved already. We
2614 * return allocated blocks so that delalloc
2615 * won't do block reservation for us. But
2616 * the buffer head will be unmapped so that
2617 * a read from the block returns 0s.
2619 if (allocated
> max_blocks
)
2620 allocated
= max_blocks
;
2621 /* mark the buffer unwritten */
2622 __set_bit(BH_Unwritten
, &bh_result
->b_state
);
2626 ret
= ext4_ext_convert_to_initialized(handle
, inode
,
2639 * requested block isn't allocated yet;
2640 * we couldn't try to create block if create flag is zero
2644 * put just found gap into cache to speed up
2645 * subsequent requests
2647 ext4_ext_put_gap_in_cache(inode
, path
, iblock
);
2651 * Okay, we need to do block allocation. Lazily initialize the block
2652 * allocation info here if necessary.
2654 if (S_ISREG(inode
->i_mode
) && (!EXT4_I(inode
)->i_block_alloc_info
))
2655 ext4_init_block_alloc_info(inode
);
2657 /* find neighbour allocated blocks */
2659 err
= ext4_ext_search_left(inode
, path
, &ar
.lleft
, &ar
.pleft
);
2663 err
= ext4_ext_search_right(inode
, path
, &ar
.lright
, &ar
.pright
);
2668 * See if request is beyond maximum number of blocks we can have in
2669 * a single extent. For an initialized extent this limit is
2670 * EXT_INIT_MAX_LEN and for an uninitialized extent this limit is
2671 * EXT_UNINIT_MAX_LEN.
2673 if (max_blocks
> EXT_INIT_MAX_LEN
&&
2674 create
!= EXT4_CREATE_UNINITIALIZED_EXT
)
2675 max_blocks
= EXT_INIT_MAX_LEN
;
2676 else if (max_blocks
> EXT_UNINIT_MAX_LEN
&&
2677 create
== EXT4_CREATE_UNINITIALIZED_EXT
)
2678 max_blocks
= EXT_UNINIT_MAX_LEN
;
2680 /* Check if we can really insert (iblock)::(iblock+max_blocks) extent */
2681 newex
.ee_block
= cpu_to_le32(iblock
);
2682 newex
.ee_len
= cpu_to_le16(max_blocks
);
2683 err
= ext4_ext_check_overlap(inode
, &newex
, path
);
2685 allocated
= ext4_ext_get_actual_len(&newex
);
2687 allocated
= max_blocks
;
2689 /* allocate new block */
2691 ar
.goal
= ext4_ext_find_goal(inode
, path
, iblock
);
2692 ar
.logical
= iblock
;
2694 if (S_ISREG(inode
->i_mode
))
2695 ar
.flags
= EXT4_MB_HINT_DATA
;
2697 /* disable in-core preallocation for non-regular files */
2699 newblock
= ext4_mb_new_blocks(handle
, &ar
, &err
);
2702 ext_debug("allocate new block: goal %llu, found %llu/%lu\n",
2703 goal
, newblock
, allocated
);
2705 /* try to insert new extent into found leaf and return */
2706 ext4_ext_store_pblock(&newex
, newblock
);
2707 newex
.ee_len
= cpu_to_le16(ar
.len
);
2708 if (create
== EXT4_CREATE_UNINITIALIZED_EXT
) /* Mark uninitialized */
2709 ext4_ext_mark_uninitialized(&newex
);
2710 err
= ext4_ext_insert_extent(handle
, inode
, path
, &newex
);
2712 /* free data blocks we just allocated */
2713 /* not a good idea to call discard here directly,
2714 * but otherwise we'd need to call it every free() */
2715 ext4_mb_discard_inode_preallocations(inode
);
2716 ext4_free_blocks(handle
, inode
, ext_pblock(&newex
),
2717 ext4_ext_get_actual_len(&newex
), 0);
2721 /* previous routine could use block we allocated */
2722 newblock
= ext_pblock(&newex
);
2723 allocated
= ext4_ext_get_actual_len(&newex
);
2725 if (extend_disksize
&& inode
->i_size
> EXT4_I(inode
)->i_disksize
)
2726 EXT4_I(inode
)->i_disksize
= inode
->i_size
;
2728 __set_bit(BH_New
, &bh_result
->b_state
);
2730 /* Cache only when it is _not_ an uninitialized extent */
2731 if (create
!= EXT4_CREATE_UNINITIALIZED_EXT
)
2732 ext4_ext_put_in_cache(inode
, iblock
, allocated
, newblock
,
2733 EXT4_EXT_CACHE_EXTENT
);
2735 if (allocated
> max_blocks
)
2736 allocated
= max_blocks
;
2737 ext4_ext_show_leaf(inode
, path
);
2738 __set_bit(BH_Mapped
, &bh_result
->b_state
);
2739 bh_result
->b_bdev
= inode
->i_sb
->s_bdev
;
2740 bh_result
->b_blocknr
= newblock
;
2743 ext4_ext_drop_refs(path
);
2746 return err
? err
: allocated
;
2749 void ext4_ext_truncate(struct inode
* inode
, struct page
*page
)
2751 struct address_space
*mapping
= inode
->i_mapping
;
2752 struct super_block
*sb
= inode
->i_sb
;
2753 ext4_lblk_t last_block
;
2758 * probably first extent we're gonna free will be last in block
2760 err
= ext4_writepage_trans_blocks(inode
) + 3;
2761 handle
= ext4_journal_start(inode
, err
);
2762 if (IS_ERR(handle
)) {
2764 clear_highpage(page
);
2765 flush_dcache_page(page
);
2767 page_cache_release(page
);
2773 ext4_block_truncate_page(handle
, page
, mapping
, inode
->i_size
);
2775 down_write(&EXT4_I(inode
)->i_data_sem
);
2776 ext4_ext_invalidate_cache(inode
);
2778 ext4_mb_discard_inode_preallocations(inode
);
2781 * TODO: optimization is possible here.
2782 * Probably we need not scan at all,
2783 * because page truncation is enough.
2785 if (ext4_orphan_add(handle
, inode
))
2788 /* we have to know where to truncate from in crash case */
2789 EXT4_I(inode
)->i_disksize
= inode
->i_size
;
2790 ext4_mark_inode_dirty(handle
, inode
);
2792 last_block
= (inode
->i_size
+ sb
->s_blocksize
- 1)
2793 >> EXT4_BLOCK_SIZE_BITS(sb
);
2794 err
= ext4_ext_remove_space(inode
, last_block
);
2796 /* In a multi-transaction truncate, we only make the final
2797 * transaction synchronous.
2804 * If this was a simple ftruncate() and the file will remain alive,
2805 * then we need to clear up the orphan record which we created above.
2806 * However, if this was a real unlink then we were called by
2807 * ext4_delete_inode(), and we allow that function to clean up the
2808 * orphan info for us.
2811 ext4_orphan_del(handle
, inode
);
2813 up_write(&EXT4_I(inode
)->i_data_sem
);
2814 inode
->i_mtime
= inode
->i_ctime
= ext4_current_time(inode
);
2815 ext4_mark_inode_dirty(handle
, inode
);
2816 ext4_journal_stop(handle
);
2820 * ext4_ext_writepage_trans_blocks:
2821 * calculate max number of blocks we could modify
2822 * in order to allocate new block for an inode
2824 int ext4_ext_writepage_trans_blocks(struct inode
*inode
, int num
)
2828 needed
= ext4_ext_calc_credits_for_insert(inode
, NULL
);
2830 /* caller wants to allocate num blocks, but note it includes sb */
2831 needed
= needed
* num
- (num
- 1);
2834 needed
+= 2 * EXT4_QUOTA_TRANS_BLOCKS(inode
->i_sb
);
2840 static void ext4_falloc_update_inode(struct inode
*inode
,
2841 int mode
, loff_t new_size
, int update_ctime
)
2843 struct timespec now
;
2846 now
= current_fs_time(inode
->i_sb
);
2847 if (!timespec_equal(&inode
->i_ctime
, &now
))
2848 inode
->i_ctime
= now
;
2851 * Update only when preallocation was requested beyond
2854 if (!(mode
& FALLOC_FL_KEEP_SIZE
) &&
2855 new_size
> i_size_read(inode
)) {
2856 i_size_write(inode
, new_size
);
2857 EXT4_I(inode
)->i_disksize
= new_size
;
2863 * preallocate space for a file. This implements ext4's fallocate inode
2864 * operation, which gets called from sys_fallocate system call.
2865 * For block-mapped files, posix_fallocate should fall back to the method
2866 * of writing zeroes to the required new blocks (the same behavior which is
2867 * expected for file systems which do not support fallocate() system call).
2869 long ext4_fallocate(struct inode
*inode
, int mode
, loff_t offset
, loff_t len
)
2874 unsigned long max_blocks
;
2878 struct buffer_head map_bh
;
2879 unsigned int credits
, blkbits
= inode
->i_blkbits
;
2882 * currently supporting (pre)allocate mode for extent-based
2885 if (!(EXT4_I(inode
)->i_flags
& EXT4_EXTENTS_FL
))
2888 /* preallocation to directories is currently not supported */
2889 if (S_ISDIR(inode
->i_mode
))
2892 block
= offset
>> blkbits
;
2894 * We can't just convert len to max_blocks because
2895 * If blocksize = 4096 offset = 3072 and len = 2048
2897 max_blocks
= (EXT4_BLOCK_ALIGN(len
+ offset
, blkbits
) >> blkbits
)
2900 * credits to insert 1 extent into extent tree + buffers to be able to
2901 * modify 1 super block, 1 block bitmap and 1 group descriptor.
2903 credits
= EXT4_DATA_TRANS_BLOCKS(inode
->i_sb
) + 3;
2904 mutex_lock(&inode
->i_mutex
);
2906 while (ret
>= 0 && ret
< max_blocks
) {
2907 block
= block
+ ret
;
2908 max_blocks
= max_blocks
- ret
;
2909 handle
= ext4_journal_start(inode
, credits
);
2910 if (IS_ERR(handle
)) {
2911 ret
= PTR_ERR(handle
);
2914 ret
= ext4_get_blocks_wrap(handle
, inode
, block
,
2915 max_blocks
, &map_bh
,
2916 EXT4_CREATE_UNINITIALIZED_EXT
, 0);
2920 printk(KERN_ERR
"%s: ext4_ext_get_blocks "
2921 "returned error inode#%lu, block=%u, "
2922 "max_blocks=%lu", __func__
,
2923 inode
->i_ino
, block
, max_blocks
);
2925 ext4_mark_inode_dirty(handle
, inode
);
2926 ret2
= ext4_journal_stop(handle
);
2929 if ((block
+ ret
) >= (EXT4_BLOCK_ALIGN(offset
+ len
,
2930 blkbits
) >> blkbits
))
2931 new_size
= offset
+ len
;
2933 new_size
= (block
+ ret
) << blkbits
;
2935 ext4_falloc_update_inode(inode
, mode
, new_size
,
2936 buffer_new(&map_bh
));
2937 ext4_mark_inode_dirty(handle
, inode
);
2938 ret2
= ext4_journal_stop(handle
);
2942 if (ret
== -ENOSPC
&&
2943 ext4_should_retry_alloc(inode
->i_sb
, &retries
)) {
2947 mutex_unlock(&inode
->i_mutex
);
2948 return ret
> 0 ? ret2
: ret
;