2 * Copyright (c) 2003-2006, Cluster File Systems, Inc, info@clusterfs.com
3 * Written by Alex Tomas <alex@clusterfs.com>
5 * Architecture independence:
6 * Copyright (c) 2005, Bull S.A.
7 * Written by Pierre Peiffer <pierre.peiffer@bull.net>
9 * This program is free software; you can redistribute it and/or modify
10 * it under the terms of the GNU General Public License version 2 as
11 * published by the Free Software Foundation.
13 * This program is distributed in the hope that it will be useful,
14 * but WITHOUT ANY WARRANTY; without even the implied warranty of
15 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
16 * GNU General Public License for more details.
18 * You should have received a copy of the GNU General Public Licens
19 * along with this program; if not, write to the Free Software
20 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-
24 * Extents support for EXT4
27 * - ext4*_error() should be used in some situations
28 * - analyze all BUG()/BUG_ON(), use -EIO where appropriate
29 * - smart tree reduction
32 #include <linux/module.h>
34 #include <linux/time.h>
35 #include <linux/jbd2.h>
36 #include <linux/highuid.h>
37 #include <linux/pagemap.h>
38 #include <linux/quotaops.h>
39 #include <linux/string.h>
40 #include <linux/slab.h>
41 #include <linux/falloc.h>
42 #include <asm/uaccess.h>
43 #include <linux/fiemap.h>
44 #include "ext4_jbd2.h"
45 #include "ext4_extents.h"
50 * combine low and high parts of physical block number into ext4_fsblk_t
52 static ext4_fsblk_t
ext_pblock(struct ext4_extent
*ex
)
56 block
= le32_to_cpu(ex
->ee_start_lo
);
57 block
|= ((ext4_fsblk_t
) le16_to_cpu(ex
->ee_start_hi
) << 31) << 1;
63 * combine low and high parts of a leaf physical block number into ext4_fsblk_t
65 ext4_fsblk_t
idx_pblock(struct ext4_extent_idx
*ix
)
69 block
= le32_to_cpu(ix
->ei_leaf_lo
);
70 block
|= ((ext4_fsblk_t
) le16_to_cpu(ix
->ei_leaf_hi
) << 31) << 1;
75 * ext4_ext_store_pblock:
76 * stores a large physical block number into an extent struct,
77 * breaking it into parts
79 void ext4_ext_store_pblock(struct ext4_extent
*ex
, ext4_fsblk_t pb
)
81 ex
->ee_start_lo
= cpu_to_le32((unsigned long) (pb
& 0xffffffff));
82 ex
->ee_start_hi
= cpu_to_le16((unsigned long) ((pb
>> 31) >> 1) & 0xffff);
86 * ext4_idx_store_pblock:
87 * stores a large physical block number into an index struct,
88 * breaking it into parts
90 static void ext4_idx_store_pblock(struct ext4_extent_idx
*ix
, ext4_fsblk_t pb
)
92 ix
->ei_leaf_lo
= cpu_to_le32((unsigned long) (pb
& 0xffffffff));
93 ix
->ei_leaf_hi
= cpu_to_le16((unsigned long) ((pb
>> 31) >> 1) & 0xffff);
96 static int ext4_ext_journal_restart(handle_t
*handle
, int needed
)
100 if (!ext4_handle_valid(handle
))
102 if (handle
->h_buffer_credits
> needed
)
104 err
= ext4_journal_extend(handle
, needed
);
107 return ext4_journal_restart(handle
, needed
);
115 static int ext4_ext_get_access(handle_t
*handle
, struct inode
*inode
,
116 struct ext4_ext_path
*path
)
119 /* path points to block */
120 return ext4_journal_get_write_access(handle
, path
->p_bh
);
122 /* path points to leaf/index in inode body */
123 /* we use in-core data, no need to protect them */
133 static int ext4_ext_dirty(handle_t
*handle
, struct inode
*inode
,
134 struct ext4_ext_path
*path
)
138 /* path points to block */
139 err
= ext4_handle_dirty_metadata(handle
, inode
, path
->p_bh
);
141 /* path points to leaf/index in inode body */
142 err
= ext4_mark_inode_dirty(handle
, inode
);
147 static ext4_fsblk_t
ext4_ext_find_goal(struct inode
*inode
,
148 struct ext4_ext_path
*path
,
151 struct ext4_inode_info
*ei
= EXT4_I(inode
);
152 ext4_fsblk_t bg_start
;
153 ext4_fsblk_t last_block
;
154 ext4_grpblk_t colour
;
158 struct ext4_extent
*ex
;
159 depth
= path
->p_depth
;
161 /* try to predict block placement */
162 ex
= path
[depth
].p_ext
;
164 return ext_pblock(ex
)+(block
-le32_to_cpu(ex
->ee_block
));
166 /* it looks like index is empty;
167 * try to find starting block from index itself */
168 if (path
[depth
].p_bh
)
169 return path
[depth
].p_bh
->b_blocknr
;
172 /* OK. use inode's group */
173 bg_start
= (ei
->i_block_group
* EXT4_BLOCKS_PER_GROUP(inode
->i_sb
)) +
174 le32_to_cpu(EXT4_SB(inode
->i_sb
)->s_es
->s_first_data_block
);
175 last_block
= ext4_blocks_count(EXT4_SB(inode
->i_sb
)->s_es
) - 1;
177 if (bg_start
+ EXT4_BLOCKS_PER_GROUP(inode
->i_sb
) <= last_block
)
178 colour
= (current
->pid
% 16) *
179 (EXT4_BLOCKS_PER_GROUP(inode
->i_sb
) / 16);
181 colour
= (current
->pid
% 16) * ((last_block
- bg_start
) / 16);
182 return bg_start
+ colour
+ block
;
186 * Allocation for a meta data block
189 ext4_ext_new_meta_block(handle_t
*handle
, struct inode
*inode
,
190 struct ext4_ext_path
*path
,
191 struct ext4_extent
*ex
, int *err
)
193 ext4_fsblk_t goal
, newblock
;
195 goal
= ext4_ext_find_goal(inode
, path
, le32_to_cpu(ex
->ee_block
));
196 newblock
= ext4_new_meta_blocks(handle
, inode
, goal
, NULL
, err
);
200 static int ext4_ext_space_block(struct inode
*inode
)
204 size
= (inode
->i_sb
->s_blocksize
- sizeof(struct ext4_extent_header
))
205 / sizeof(struct ext4_extent
);
206 #ifdef AGGRESSIVE_TEST
213 static int ext4_ext_space_block_idx(struct inode
*inode
)
217 size
= (inode
->i_sb
->s_blocksize
- sizeof(struct ext4_extent_header
))
218 / sizeof(struct ext4_extent_idx
);
219 #ifdef AGGRESSIVE_TEST
226 static int ext4_ext_space_root(struct inode
*inode
)
230 size
= sizeof(EXT4_I(inode
)->i_data
);
231 size
-= sizeof(struct ext4_extent_header
);
232 size
/= sizeof(struct ext4_extent
);
233 #ifdef AGGRESSIVE_TEST
240 static int ext4_ext_space_root_idx(struct inode
*inode
)
244 size
= sizeof(EXT4_I(inode
)->i_data
);
245 size
-= sizeof(struct ext4_extent_header
);
246 size
/= sizeof(struct ext4_extent_idx
);
247 #ifdef AGGRESSIVE_TEST
255 * Calculate the number of metadata blocks needed
256 * to allocate @blocks
257 * Worse case is one block per extent
259 int ext4_ext_calc_metadata_amount(struct inode
*inode
, int blocks
)
261 int lcap
, icap
, rcap
, leafs
, idxs
, num
;
262 int newextents
= blocks
;
264 rcap
= ext4_ext_space_root_idx(inode
);
265 lcap
= ext4_ext_space_block(inode
);
266 icap
= ext4_ext_space_block_idx(inode
);
268 /* number of new leaf blocks needed */
269 num
= leafs
= (newextents
+ lcap
- 1) / lcap
;
272 * Worse case, we need separate index block(s)
273 * to link all new leaf blocks
275 idxs
= (leafs
+ icap
- 1) / icap
;
278 idxs
= (idxs
+ icap
- 1) / icap
;
279 } while (idxs
> rcap
);
285 ext4_ext_max_entries(struct inode
*inode
, int depth
)
289 if (depth
== ext_depth(inode
)) {
291 max
= ext4_ext_space_root(inode
);
293 max
= ext4_ext_space_root_idx(inode
);
296 max
= ext4_ext_space_block(inode
);
298 max
= ext4_ext_space_block_idx(inode
);
304 static int __ext4_ext_check_header(const char *function
, struct inode
*inode
,
305 struct ext4_extent_header
*eh
,
308 const char *error_msg
;
311 if (unlikely(eh
->eh_magic
!= EXT4_EXT_MAGIC
)) {
312 error_msg
= "invalid magic";
315 if (unlikely(le16_to_cpu(eh
->eh_depth
) != depth
)) {
316 error_msg
= "unexpected eh_depth";
319 if (unlikely(eh
->eh_max
== 0)) {
320 error_msg
= "invalid eh_max";
323 max
= ext4_ext_max_entries(inode
, depth
);
324 if (unlikely(le16_to_cpu(eh
->eh_max
) > max
)) {
325 error_msg
= "too large eh_max";
328 if (unlikely(le16_to_cpu(eh
->eh_entries
) > le16_to_cpu(eh
->eh_max
))) {
329 error_msg
= "invalid eh_entries";
335 ext4_error(inode
->i_sb
, function
,
336 "bad header in inode #%lu: %s - magic %x, "
337 "entries %u, max %u(%u), depth %u(%u)",
338 inode
->i_ino
, error_msg
, le16_to_cpu(eh
->eh_magic
),
339 le16_to_cpu(eh
->eh_entries
), le16_to_cpu(eh
->eh_max
),
340 max
, le16_to_cpu(eh
->eh_depth
), depth
);
345 #define ext4_ext_check_header(inode, eh, depth) \
346 __ext4_ext_check_header(__func__, inode, eh, depth)
349 static void ext4_ext_show_path(struct inode
*inode
, struct ext4_ext_path
*path
)
351 int k
, l
= path
->p_depth
;
354 for (k
= 0; k
<= l
; k
++, path
++) {
356 ext_debug(" %d->%llu", le32_to_cpu(path
->p_idx
->ei_block
),
357 idx_pblock(path
->p_idx
));
358 } else if (path
->p_ext
) {
359 ext_debug(" %d:%d:%llu ",
360 le32_to_cpu(path
->p_ext
->ee_block
),
361 ext4_ext_get_actual_len(path
->p_ext
),
362 ext_pblock(path
->p_ext
));
369 static void ext4_ext_show_leaf(struct inode
*inode
, struct ext4_ext_path
*path
)
371 int depth
= ext_depth(inode
);
372 struct ext4_extent_header
*eh
;
373 struct ext4_extent
*ex
;
379 eh
= path
[depth
].p_hdr
;
380 ex
= EXT_FIRST_EXTENT(eh
);
382 for (i
= 0; i
< le16_to_cpu(eh
->eh_entries
); i
++, ex
++) {
383 ext_debug("%d:%d:%llu ", le32_to_cpu(ex
->ee_block
),
384 ext4_ext_get_actual_len(ex
), ext_pblock(ex
));
389 #define ext4_ext_show_path(inode, path)
390 #define ext4_ext_show_leaf(inode, path)
393 void ext4_ext_drop_refs(struct ext4_ext_path
*path
)
395 int depth
= path
->p_depth
;
398 for (i
= 0; i
<= depth
; i
++, path
++)
406 * ext4_ext_binsearch_idx:
407 * binary search for the closest index of the given block
408 * the header must be checked before calling this
411 ext4_ext_binsearch_idx(struct inode
*inode
,
412 struct ext4_ext_path
*path
, ext4_lblk_t block
)
414 struct ext4_extent_header
*eh
= path
->p_hdr
;
415 struct ext4_extent_idx
*r
, *l
, *m
;
418 ext_debug("binsearch for %u(idx): ", block
);
420 l
= EXT_FIRST_INDEX(eh
) + 1;
421 r
= EXT_LAST_INDEX(eh
);
424 if (block
< le32_to_cpu(m
->ei_block
))
428 ext_debug("%p(%u):%p(%u):%p(%u) ", l
, le32_to_cpu(l
->ei_block
),
429 m
, le32_to_cpu(m
->ei_block
),
430 r
, le32_to_cpu(r
->ei_block
));
434 ext_debug(" -> %d->%lld ", le32_to_cpu(path
->p_idx
->ei_block
),
435 idx_pblock(path
->p_idx
));
437 #ifdef CHECK_BINSEARCH
439 struct ext4_extent_idx
*chix
, *ix
;
442 chix
= ix
= EXT_FIRST_INDEX(eh
);
443 for (k
= 0; k
< le16_to_cpu(eh
->eh_entries
); k
++, ix
++) {
445 le32_to_cpu(ix
->ei_block
) <= le32_to_cpu(ix
[-1].ei_block
)) {
446 printk(KERN_DEBUG
"k=%d, ix=0x%p, "
448 ix
, EXT_FIRST_INDEX(eh
));
449 printk(KERN_DEBUG
"%u <= %u\n",
450 le32_to_cpu(ix
->ei_block
),
451 le32_to_cpu(ix
[-1].ei_block
));
453 BUG_ON(k
&& le32_to_cpu(ix
->ei_block
)
454 <= le32_to_cpu(ix
[-1].ei_block
));
455 if (block
< le32_to_cpu(ix
->ei_block
))
459 BUG_ON(chix
!= path
->p_idx
);
466 * ext4_ext_binsearch:
467 * binary search for closest extent of the given block
468 * the header must be checked before calling this
471 ext4_ext_binsearch(struct inode
*inode
,
472 struct ext4_ext_path
*path
, ext4_lblk_t block
)
474 struct ext4_extent_header
*eh
= path
->p_hdr
;
475 struct ext4_extent
*r
, *l
, *m
;
477 if (eh
->eh_entries
== 0) {
479 * this leaf is empty:
480 * we get such a leaf in split/add case
485 ext_debug("binsearch for %u: ", block
);
487 l
= EXT_FIRST_EXTENT(eh
) + 1;
488 r
= EXT_LAST_EXTENT(eh
);
492 if (block
< le32_to_cpu(m
->ee_block
))
496 ext_debug("%p(%u):%p(%u):%p(%u) ", l
, le32_to_cpu(l
->ee_block
),
497 m
, le32_to_cpu(m
->ee_block
),
498 r
, le32_to_cpu(r
->ee_block
));
502 ext_debug(" -> %d:%llu:%d ",
503 le32_to_cpu(path
->p_ext
->ee_block
),
504 ext_pblock(path
->p_ext
),
505 ext4_ext_get_actual_len(path
->p_ext
));
507 #ifdef CHECK_BINSEARCH
509 struct ext4_extent
*chex
, *ex
;
512 chex
= ex
= EXT_FIRST_EXTENT(eh
);
513 for (k
= 0; k
< le16_to_cpu(eh
->eh_entries
); k
++, ex
++) {
514 BUG_ON(k
&& le32_to_cpu(ex
->ee_block
)
515 <= le32_to_cpu(ex
[-1].ee_block
));
516 if (block
< le32_to_cpu(ex
->ee_block
))
520 BUG_ON(chex
!= path
->p_ext
);
526 int ext4_ext_tree_init(handle_t
*handle
, struct inode
*inode
)
528 struct ext4_extent_header
*eh
;
530 eh
= ext_inode_hdr(inode
);
533 eh
->eh_magic
= EXT4_EXT_MAGIC
;
534 eh
->eh_max
= cpu_to_le16(ext4_ext_space_root(inode
));
535 ext4_mark_inode_dirty(handle
, inode
);
536 ext4_ext_invalidate_cache(inode
);
540 struct ext4_ext_path
*
541 ext4_ext_find_extent(struct inode
*inode
, ext4_lblk_t block
,
542 struct ext4_ext_path
*path
)
544 struct ext4_extent_header
*eh
;
545 struct buffer_head
*bh
;
546 short int depth
, i
, ppos
= 0, alloc
= 0;
548 eh
= ext_inode_hdr(inode
);
549 depth
= ext_depth(inode
);
550 if (ext4_ext_check_header(inode
, eh
, depth
))
551 return ERR_PTR(-EIO
);
554 /* account possible depth increase */
556 path
= kzalloc(sizeof(struct ext4_ext_path
) * (depth
+ 2),
559 return ERR_PTR(-ENOMEM
);
566 /* walk through the tree */
568 ext_debug("depth %d: num %d, max %d\n",
569 ppos
, le16_to_cpu(eh
->eh_entries
), le16_to_cpu(eh
->eh_max
));
571 ext4_ext_binsearch_idx(inode
, path
+ ppos
, block
);
572 path
[ppos
].p_block
= idx_pblock(path
[ppos
].p_idx
);
573 path
[ppos
].p_depth
= i
;
574 path
[ppos
].p_ext
= NULL
;
576 bh
= sb_bread(inode
->i_sb
, path
[ppos
].p_block
);
580 eh
= ext_block_hdr(bh
);
582 BUG_ON(ppos
> depth
);
583 path
[ppos
].p_bh
= bh
;
584 path
[ppos
].p_hdr
= eh
;
587 if (ext4_ext_check_header(inode
, eh
, i
))
591 path
[ppos
].p_depth
= i
;
592 path
[ppos
].p_ext
= NULL
;
593 path
[ppos
].p_idx
= NULL
;
596 ext4_ext_binsearch(inode
, path
+ ppos
, block
);
597 /* if not an empty leaf */
598 if (path
[ppos
].p_ext
)
599 path
[ppos
].p_block
= ext_pblock(path
[ppos
].p_ext
);
601 ext4_ext_show_path(inode
, path
);
606 ext4_ext_drop_refs(path
);
609 return ERR_PTR(-EIO
);
613 * ext4_ext_insert_index:
614 * insert new index [@logical;@ptr] into the block at @curp;
615 * check where to insert: before @curp or after @curp
617 static int ext4_ext_insert_index(handle_t
*handle
, struct inode
*inode
,
618 struct ext4_ext_path
*curp
,
619 int logical
, ext4_fsblk_t ptr
)
621 struct ext4_extent_idx
*ix
;
624 err
= ext4_ext_get_access(handle
, inode
, curp
);
628 BUG_ON(logical
== le32_to_cpu(curp
->p_idx
->ei_block
));
629 len
= EXT_MAX_INDEX(curp
->p_hdr
) - curp
->p_idx
;
630 if (logical
> le32_to_cpu(curp
->p_idx
->ei_block
)) {
632 if (curp
->p_idx
!= EXT_LAST_INDEX(curp
->p_hdr
)) {
633 len
= (len
- 1) * sizeof(struct ext4_extent_idx
);
634 len
= len
< 0 ? 0 : len
;
635 ext_debug("insert new index %d after: %llu. "
636 "move %d from 0x%p to 0x%p\n",
638 (curp
->p_idx
+ 1), (curp
->p_idx
+ 2));
639 memmove(curp
->p_idx
+ 2, curp
->p_idx
+ 1, len
);
641 ix
= curp
->p_idx
+ 1;
644 len
= len
* sizeof(struct ext4_extent_idx
);
645 len
= len
< 0 ? 0 : len
;
646 ext_debug("insert new index %d before: %llu. "
647 "move %d from 0x%p to 0x%p\n",
649 curp
->p_idx
, (curp
->p_idx
+ 1));
650 memmove(curp
->p_idx
+ 1, curp
->p_idx
, len
);
654 ix
->ei_block
= cpu_to_le32(logical
);
655 ext4_idx_store_pblock(ix
, ptr
);
656 le16_add_cpu(&curp
->p_hdr
->eh_entries
, 1);
658 BUG_ON(le16_to_cpu(curp
->p_hdr
->eh_entries
)
659 > le16_to_cpu(curp
->p_hdr
->eh_max
));
660 BUG_ON(ix
> EXT_LAST_INDEX(curp
->p_hdr
));
662 err
= ext4_ext_dirty(handle
, inode
, curp
);
663 ext4_std_error(inode
->i_sb
, err
);
670 * inserts new subtree into the path, using free index entry
672 * - allocates all needed blocks (new leaf and all intermediate index blocks)
673 * - makes decision where to split
674 * - moves remaining extents and index entries (right to the split point)
675 * into the newly allocated blocks
676 * - initializes subtree
678 static int ext4_ext_split(handle_t
*handle
, struct inode
*inode
,
679 struct ext4_ext_path
*path
,
680 struct ext4_extent
*newext
, int at
)
682 struct buffer_head
*bh
= NULL
;
683 int depth
= ext_depth(inode
);
684 struct ext4_extent_header
*neh
;
685 struct ext4_extent_idx
*fidx
;
686 struct ext4_extent
*ex
;
688 ext4_fsblk_t newblock
, oldblock
;
690 ext4_fsblk_t
*ablocks
= NULL
; /* array of allocated blocks */
693 /* make decision: where to split? */
694 /* FIXME: now decision is simplest: at current extent */
696 /* if current leaf will be split, then we should use
697 * border from split point */
698 BUG_ON(path
[depth
].p_ext
> EXT_MAX_EXTENT(path
[depth
].p_hdr
));
699 if (path
[depth
].p_ext
!= EXT_MAX_EXTENT(path
[depth
].p_hdr
)) {
700 border
= path
[depth
].p_ext
[1].ee_block
;
701 ext_debug("leaf will be split."
702 " next leaf starts at %d\n",
703 le32_to_cpu(border
));
705 border
= newext
->ee_block
;
706 ext_debug("leaf will be added."
707 " next leaf starts at %d\n",
708 le32_to_cpu(border
));
712 * If error occurs, then we break processing
713 * and mark filesystem read-only. index won't
714 * be inserted and tree will be in consistent
715 * state. Next mount will repair buffers too.
719 * Get array to track all allocated blocks.
720 * We need this to handle errors and free blocks
723 ablocks
= kzalloc(sizeof(ext4_fsblk_t
) * depth
, GFP_NOFS
);
727 /* allocate all needed blocks */
728 ext_debug("allocate %d blocks for indexes/leaf\n", depth
- at
);
729 for (a
= 0; a
< depth
- at
; a
++) {
730 newblock
= ext4_ext_new_meta_block(handle
, inode
, path
,
734 ablocks
[a
] = newblock
;
737 /* initialize new leaf */
738 newblock
= ablocks
[--a
];
739 BUG_ON(newblock
== 0);
740 bh
= sb_getblk(inode
->i_sb
, newblock
);
747 err
= ext4_journal_get_create_access(handle
, bh
);
751 neh
= ext_block_hdr(bh
);
753 neh
->eh_max
= cpu_to_le16(ext4_ext_space_block(inode
));
754 neh
->eh_magic
= EXT4_EXT_MAGIC
;
756 ex
= EXT_FIRST_EXTENT(neh
);
758 /* move remainder of path[depth] to the new leaf */
759 BUG_ON(path
[depth
].p_hdr
->eh_entries
!= path
[depth
].p_hdr
->eh_max
);
760 /* start copy from next extent */
761 /* TODO: we could do it by single memmove */
764 while (path
[depth
].p_ext
<=
765 EXT_MAX_EXTENT(path
[depth
].p_hdr
)) {
766 ext_debug("move %d:%llu:%d in new leaf %llu\n",
767 le32_to_cpu(path
[depth
].p_ext
->ee_block
),
768 ext_pblock(path
[depth
].p_ext
),
769 ext4_ext_get_actual_len(path
[depth
].p_ext
),
771 /*memmove(ex++, path[depth].p_ext++,
772 sizeof(struct ext4_extent));
778 memmove(ex
, path
[depth
].p_ext
-m
, sizeof(struct ext4_extent
)*m
);
779 le16_add_cpu(&neh
->eh_entries
, m
);
782 set_buffer_uptodate(bh
);
785 err
= ext4_handle_dirty_metadata(handle
, inode
, bh
);
791 /* correct old leaf */
793 err
= ext4_ext_get_access(handle
, inode
, path
+ depth
);
796 le16_add_cpu(&path
[depth
].p_hdr
->eh_entries
, -m
);
797 err
= ext4_ext_dirty(handle
, inode
, path
+ depth
);
803 /* create intermediate indexes */
807 ext_debug("create %d intermediate indices\n", k
);
808 /* insert new index into current index block */
809 /* current depth stored in i var */
813 newblock
= ablocks
[--a
];
814 bh
= sb_getblk(inode
->i_sb
, newblock
);
821 err
= ext4_journal_get_create_access(handle
, bh
);
825 neh
= ext_block_hdr(bh
);
826 neh
->eh_entries
= cpu_to_le16(1);
827 neh
->eh_magic
= EXT4_EXT_MAGIC
;
828 neh
->eh_max
= cpu_to_le16(ext4_ext_space_block_idx(inode
));
829 neh
->eh_depth
= cpu_to_le16(depth
- i
);
830 fidx
= EXT_FIRST_INDEX(neh
);
831 fidx
->ei_block
= border
;
832 ext4_idx_store_pblock(fidx
, oldblock
);
834 ext_debug("int.index at %d (block %llu): %u -> %llu\n",
835 i
, newblock
, le32_to_cpu(border
), oldblock
);
840 ext_debug("cur 0x%p, last 0x%p\n", path
[i
].p_idx
,
841 EXT_MAX_INDEX(path
[i
].p_hdr
));
842 BUG_ON(EXT_MAX_INDEX(path
[i
].p_hdr
) !=
843 EXT_LAST_INDEX(path
[i
].p_hdr
));
844 while (path
[i
].p_idx
<= EXT_MAX_INDEX(path
[i
].p_hdr
)) {
845 ext_debug("%d: move %d:%llu in new index %llu\n", i
,
846 le32_to_cpu(path
[i
].p_idx
->ei_block
),
847 idx_pblock(path
[i
].p_idx
),
849 /*memmove(++fidx, path[i].p_idx++,
850 sizeof(struct ext4_extent_idx));
852 BUG_ON(neh->eh_entries > neh->eh_max);*/
857 memmove(++fidx
, path
[i
].p_idx
- m
,
858 sizeof(struct ext4_extent_idx
) * m
);
859 le16_add_cpu(&neh
->eh_entries
, m
);
861 set_buffer_uptodate(bh
);
864 err
= ext4_handle_dirty_metadata(handle
, inode
, bh
);
870 /* correct old index */
872 err
= ext4_ext_get_access(handle
, inode
, path
+ i
);
875 le16_add_cpu(&path
[i
].p_hdr
->eh_entries
, -m
);
876 err
= ext4_ext_dirty(handle
, inode
, path
+ i
);
884 /* insert new index */
885 err
= ext4_ext_insert_index(handle
, inode
, path
+ at
,
886 le32_to_cpu(border
), newblock
);
890 if (buffer_locked(bh
))
896 /* free all allocated blocks in error case */
897 for (i
= 0; i
< depth
; i
++) {
900 ext4_free_blocks(handle
, inode
, ablocks
[i
], 1, 1);
909 * ext4_ext_grow_indepth:
910 * implements tree growing procedure:
911 * - allocates new block
912 * - moves top-level data (index block or leaf) into the new block
913 * - initializes new top-level, creating index that points to the
916 static int ext4_ext_grow_indepth(handle_t
*handle
, struct inode
*inode
,
917 struct ext4_ext_path
*path
,
918 struct ext4_extent
*newext
)
920 struct ext4_ext_path
*curp
= path
;
921 struct ext4_extent_header
*neh
;
922 struct ext4_extent_idx
*fidx
;
923 struct buffer_head
*bh
;
924 ext4_fsblk_t newblock
;
927 newblock
= ext4_ext_new_meta_block(handle
, inode
, path
, newext
, &err
);
931 bh
= sb_getblk(inode
->i_sb
, newblock
);
934 ext4_std_error(inode
->i_sb
, err
);
939 err
= ext4_journal_get_create_access(handle
, bh
);
945 /* move top-level index/leaf into new block */
946 memmove(bh
->b_data
, curp
->p_hdr
, sizeof(EXT4_I(inode
)->i_data
));
948 /* set size of new block */
949 neh
= ext_block_hdr(bh
);
950 /* old root could have indexes or leaves
951 * so calculate e_max right way */
952 if (ext_depth(inode
))
953 neh
->eh_max
= cpu_to_le16(ext4_ext_space_block_idx(inode
));
955 neh
->eh_max
= cpu_to_le16(ext4_ext_space_block(inode
));
956 neh
->eh_magic
= EXT4_EXT_MAGIC
;
957 set_buffer_uptodate(bh
);
960 err
= ext4_handle_dirty_metadata(handle
, inode
, bh
);
964 /* create index in new top-level index: num,max,pointer */
965 err
= ext4_ext_get_access(handle
, inode
, curp
);
969 curp
->p_hdr
->eh_magic
= EXT4_EXT_MAGIC
;
970 curp
->p_hdr
->eh_max
= cpu_to_le16(ext4_ext_space_root_idx(inode
));
971 curp
->p_hdr
->eh_entries
= cpu_to_le16(1);
972 curp
->p_idx
= EXT_FIRST_INDEX(curp
->p_hdr
);
974 if (path
[0].p_hdr
->eh_depth
)
975 curp
->p_idx
->ei_block
=
976 EXT_FIRST_INDEX(path
[0].p_hdr
)->ei_block
;
978 curp
->p_idx
->ei_block
=
979 EXT_FIRST_EXTENT(path
[0].p_hdr
)->ee_block
;
980 ext4_idx_store_pblock(curp
->p_idx
, newblock
);
982 neh
= ext_inode_hdr(inode
);
983 fidx
= EXT_FIRST_INDEX(neh
);
984 ext_debug("new root: num %d(%d), lblock %d, ptr %llu\n",
985 le16_to_cpu(neh
->eh_entries
), le16_to_cpu(neh
->eh_max
),
986 le32_to_cpu(fidx
->ei_block
), idx_pblock(fidx
));
988 neh
->eh_depth
= cpu_to_le16(path
->p_depth
+ 1);
989 err
= ext4_ext_dirty(handle
, inode
, curp
);
997 * ext4_ext_create_new_leaf:
998 * finds empty index and adds new leaf.
999 * if no free index is found, then it requests in-depth growing.
1001 static int ext4_ext_create_new_leaf(handle_t
*handle
, struct inode
*inode
,
1002 struct ext4_ext_path
*path
,
1003 struct ext4_extent
*newext
)
1005 struct ext4_ext_path
*curp
;
1006 int depth
, i
, err
= 0;
1009 i
= depth
= ext_depth(inode
);
1011 /* walk up to the tree and look for free index entry */
1012 curp
= path
+ depth
;
1013 while (i
> 0 && !EXT_HAS_FREE_INDEX(curp
)) {
1018 /* we use already allocated block for index block,
1019 * so subsequent data blocks should be contiguous */
1020 if (EXT_HAS_FREE_INDEX(curp
)) {
1021 /* if we found index with free entry, then use that
1022 * entry: create all needed subtree and add new leaf */
1023 err
= ext4_ext_split(handle
, inode
, path
, newext
, i
);
1028 ext4_ext_drop_refs(path
);
1029 path
= ext4_ext_find_extent(inode
,
1030 (ext4_lblk_t
)le32_to_cpu(newext
->ee_block
),
1033 err
= PTR_ERR(path
);
1035 /* tree is full, time to grow in depth */
1036 err
= ext4_ext_grow_indepth(handle
, inode
, path
, newext
);
1041 ext4_ext_drop_refs(path
);
1042 path
= ext4_ext_find_extent(inode
,
1043 (ext4_lblk_t
)le32_to_cpu(newext
->ee_block
),
1046 err
= PTR_ERR(path
);
1051 * only first (depth 0 -> 1) produces free space;
1052 * in all other cases we have to split the grown tree
1054 depth
= ext_depth(inode
);
1055 if (path
[depth
].p_hdr
->eh_entries
== path
[depth
].p_hdr
->eh_max
) {
1056 /* now we need to split */
1066 * search the closest allocated block to the left for *logical
1067 * and returns it at @logical + it's physical address at @phys
1068 * if *logical is the smallest allocated block, the function
1069 * returns 0 at @phys
1070 * return value contains 0 (success) or error code
1073 ext4_ext_search_left(struct inode
*inode
, struct ext4_ext_path
*path
,
1074 ext4_lblk_t
*logical
, ext4_fsblk_t
*phys
)
1076 struct ext4_extent_idx
*ix
;
1077 struct ext4_extent
*ex
;
1080 BUG_ON(path
== NULL
);
1081 depth
= path
->p_depth
;
1084 if (depth
== 0 && path
->p_ext
== NULL
)
1087 /* usually extent in the path covers blocks smaller
1088 * then *logical, but it can be that extent is the
1089 * first one in the file */
1091 ex
= path
[depth
].p_ext
;
1092 ee_len
= ext4_ext_get_actual_len(ex
);
1093 if (*logical
< le32_to_cpu(ex
->ee_block
)) {
1094 BUG_ON(EXT_FIRST_EXTENT(path
[depth
].p_hdr
) != ex
);
1095 while (--depth
>= 0) {
1096 ix
= path
[depth
].p_idx
;
1097 BUG_ON(ix
!= EXT_FIRST_INDEX(path
[depth
].p_hdr
));
1102 BUG_ON(*logical
< (le32_to_cpu(ex
->ee_block
) + ee_len
));
1104 *logical
= le32_to_cpu(ex
->ee_block
) + ee_len
- 1;
1105 *phys
= ext_pblock(ex
) + ee_len
- 1;
1110 * search the closest allocated block to the right for *logical
1111 * and returns it at @logical + it's physical address at @phys
1112 * if *logical is the smallest allocated block, the function
1113 * returns 0 at @phys
1114 * return value contains 0 (success) or error code
1117 ext4_ext_search_right(struct inode
*inode
, struct ext4_ext_path
*path
,
1118 ext4_lblk_t
*logical
, ext4_fsblk_t
*phys
)
1120 struct buffer_head
*bh
= NULL
;
1121 struct ext4_extent_header
*eh
;
1122 struct ext4_extent_idx
*ix
;
1123 struct ext4_extent
*ex
;
1125 int depth
; /* Note, NOT eh_depth; depth from top of tree */
1128 BUG_ON(path
== NULL
);
1129 depth
= path
->p_depth
;
1132 if (depth
== 0 && path
->p_ext
== NULL
)
1135 /* usually extent in the path covers blocks smaller
1136 * then *logical, but it can be that extent is the
1137 * first one in the file */
1139 ex
= path
[depth
].p_ext
;
1140 ee_len
= ext4_ext_get_actual_len(ex
);
1141 if (*logical
< le32_to_cpu(ex
->ee_block
)) {
1142 BUG_ON(EXT_FIRST_EXTENT(path
[depth
].p_hdr
) != ex
);
1143 while (--depth
>= 0) {
1144 ix
= path
[depth
].p_idx
;
1145 BUG_ON(ix
!= EXT_FIRST_INDEX(path
[depth
].p_hdr
));
1147 *logical
= le32_to_cpu(ex
->ee_block
);
1148 *phys
= ext_pblock(ex
);
1152 BUG_ON(*logical
< (le32_to_cpu(ex
->ee_block
) + ee_len
));
1154 if (ex
!= EXT_LAST_EXTENT(path
[depth
].p_hdr
)) {
1155 /* next allocated block in this leaf */
1157 *logical
= le32_to_cpu(ex
->ee_block
);
1158 *phys
= ext_pblock(ex
);
1162 /* go up and search for index to the right */
1163 while (--depth
>= 0) {
1164 ix
= path
[depth
].p_idx
;
1165 if (ix
!= EXT_LAST_INDEX(path
[depth
].p_hdr
))
1169 /* we've gone up to the root and found no index to the right */
1173 /* we've found index to the right, let's
1174 * follow it and find the closest allocated
1175 * block to the right */
1177 block
= idx_pblock(ix
);
1178 while (++depth
< path
->p_depth
) {
1179 bh
= sb_bread(inode
->i_sb
, block
);
1182 eh
= ext_block_hdr(bh
);
1183 /* subtract from p_depth to get proper eh_depth */
1184 if (ext4_ext_check_header(inode
, eh
, path
->p_depth
- depth
)) {
1188 ix
= EXT_FIRST_INDEX(eh
);
1189 block
= idx_pblock(ix
);
1193 bh
= sb_bread(inode
->i_sb
, block
);
1196 eh
= ext_block_hdr(bh
);
1197 if (ext4_ext_check_header(inode
, eh
, path
->p_depth
- depth
)) {
1201 ex
= EXT_FIRST_EXTENT(eh
);
1202 *logical
= le32_to_cpu(ex
->ee_block
);
1203 *phys
= ext_pblock(ex
);
1209 * ext4_ext_next_allocated_block:
1210 * returns allocated block in subsequent extent or EXT_MAX_BLOCK.
1211 * NOTE: it considers block number from index entry as
1212 * allocated block. Thus, index entries have to be consistent
1216 ext4_ext_next_allocated_block(struct ext4_ext_path
*path
)
1220 BUG_ON(path
== NULL
);
1221 depth
= path
->p_depth
;
1223 if (depth
== 0 && path
->p_ext
== NULL
)
1224 return EXT_MAX_BLOCK
;
1226 while (depth
>= 0) {
1227 if (depth
== path
->p_depth
) {
1229 if (path
[depth
].p_ext
!=
1230 EXT_LAST_EXTENT(path
[depth
].p_hdr
))
1231 return le32_to_cpu(path
[depth
].p_ext
[1].ee_block
);
1234 if (path
[depth
].p_idx
!=
1235 EXT_LAST_INDEX(path
[depth
].p_hdr
))
1236 return le32_to_cpu(path
[depth
].p_idx
[1].ei_block
);
1241 return EXT_MAX_BLOCK
;
1245 * ext4_ext_next_leaf_block:
1246 * returns first allocated block from next leaf or EXT_MAX_BLOCK
1248 static ext4_lblk_t
ext4_ext_next_leaf_block(struct inode
*inode
,
1249 struct ext4_ext_path
*path
)
1253 BUG_ON(path
== NULL
);
1254 depth
= path
->p_depth
;
1256 /* zero-tree has no leaf blocks at all */
1258 return EXT_MAX_BLOCK
;
1260 /* go to index block */
1263 while (depth
>= 0) {
1264 if (path
[depth
].p_idx
!=
1265 EXT_LAST_INDEX(path
[depth
].p_hdr
))
1266 return (ext4_lblk_t
)
1267 le32_to_cpu(path
[depth
].p_idx
[1].ei_block
);
1271 return EXT_MAX_BLOCK
;
1275 * ext4_ext_correct_indexes:
1276 * if leaf gets modified and modified extent is first in the leaf,
1277 * then we have to correct all indexes above.
1278 * TODO: do we need to correct tree in all cases?
1280 static int ext4_ext_correct_indexes(handle_t
*handle
, struct inode
*inode
,
1281 struct ext4_ext_path
*path
)
1283 struct ext4_extent_header
*eh
;
1284 int depth
= ext_depth(inode
);
1285 struct ext4_extent
*ex
;
1289 eh
= path
[depth
].p_hdr
;
1290 ex
= path
[depth
].p_ext
;
1295 /* there is no tree at all */
1299 if (ex
!= EXT_FIRST_EXTENT(eh
)) {
1300 /* we correct tree if first leaf got modified only */
1305 * TODO: we need correction if border is smaller than current one
1308 border
= path
[depth
].p_ext
->ee_block
;
1309 err
= ext4_ext_get_access(handle
, inode
, path
+ k
);
1312 path
[k
].p_idx
->ei_block
= border
;
1313 err
= ext4_ext_dirty(handle
, inode
, path
+ k
);
1318 /* change all left-side indexes */
1319 if (path
[k
+1].p_idx
!= EXT_FIRST_INDEX(path
[k
+1].p_hdr
))
1321 err
= ext4_ext_get_access(handle
, inode
, path
+ k
);
1324 path
[k
].p_idx
->ei_block
= border
;
1325 err
= ext4_ext_dirty(handle
, inode
, path
+ k
);
1334 ext4_can_extents_be_merged(struct inode
*inode
, struct ext4_extent
*ex1
,
1335 struct ext4_extent
*ex2
)
1337 unsigned short ext1_ee_len
, ext2_ee_len
, max_len
;
1340 * Make sure that either both extents are uninitialized, or
1343 if (ext4_ext_is_uninitialized(ex1
) ^ ext4_ext_is_uninitialized(ex2
))
1346 if (ext4_ext_is_uninitialized(ex1
))
1347 max_len
= EXT_UNINIT_MAX_LEN
;
1349 max_len
= EXT_INIT_MAX_LEN
;
1351 ext1_ee_len
= ext4_ext_get_actual_len(ex1
);
1352 ext2_ee_len
= ext4_ext_get_actual_len(ex2
);
1354 if (le32_to_cpu(ex1
->ee_block
) + ext1_ee_len
!=
1355 le32_to_cpu(ex2
->ee_block
))
1359 * To allow future support for preallocated extents to be added
1360 * as an RO_COMPAT feature, refuse to merge to extents if
1361 * this can result in the top bit of ee_len being set.
1363 if (ext1_ee_len
+ ext2_ee_len
> max_len
)
1365 #ifdef AGGRESSIVE_TEST
1366 if (ext1_ee_len
>= 4)
1370 if (ext_pblock(ex1
) + ext1_ee_len
== ext_pblock(ex2
))
1376 * This function tries to merge the "ex" extent to the next extent in the tree.
1377 * It always tries to merge towards right. If you want to merge towards
1378 * left, pass "ex - 1" as argument instead of "ex".
1379 * Returns 0 if the extents (ex and ex+1) were _not_ merged and returns
1380 * 1 if they got merged.
1382 int ext4_ext_try_to_merge(struct inode
*inode
,
1383 struct ext4_ext_path
*path
,
1384 struct ext4_extent
*ex
)
1386 struct ext4_extent_header
*eh
;
1387 unsigned int depth
, len
;
1389 int uninitialized
= 0;
1391 depth
= ext_depth(inode
);
1392 BUG_ON(path
[depth
].p_hdr
== NULL
);
1393 eh
= path
[depth
].p_hdr
;
1395 while (ex
< EXT_LAST_EXTENT(eh
)) {
1396 if (!ext4_can_extents_be_merged(inode
, ex
, ex
+ 1))
1398 /* merge with next extent! */
1399 if (ext4_ext_is_uninitialized(ex
))
1401 ex
->ee_len
= cpu_to_le16(ext4_ext_get_actual_len(ex
)
1402 + ext4_ext_get_actual_len(ex
+ 1));
1404 ext4_ext_mark_uninitialized(ex
);
1406 if (ex
+ 1 < EXT_LAST_EXTENT(eh
)) {
1407 len
= (EXT_LAST_EXTENT(eh
) - ex
- 1)
1408 * sizeof(struct ext4_extent
);
1409 memmove(ex
+ 1, ex
+ 2, len
);
1411 le16_add_cpu(&eh
->eh_entries
, -1);
1413 WARN_ON(eh
->eh_entries
== 0);
1414 if (!eh
->eh_entries
)
1415 ext4_error(inode
->i_sb
, "ext4_ext_try_to_merge",
1416 "inode#%lu, eh->eh_entries = 0!", inode
->i_ino
);
1423 * check if a portion of the "newext" extent overlaps with an
1426 * If there is an overlap discovered, it updates the length of the newext
1427 * such that there will be no overlap, and then returns 1.
1428 * If there is no overlap found, it returns 0.
1430 unsigned int ext4_ext_check_overlap(struct inode
*inode
,
1431 struct ext4_extent
*newext
,
1432 struct ext4_ext_path
*path
)
1435 unsigned int depth
, len1
;
1436 unsigned int ret
= 0;
1438 b1
= le32_to_cpu(newext
->ee_block
);
1439 len1
= ext4_ext_get_actual_len(newext
);
1440 depth
= ext_depth(inode
);
1441 if (!path
[depth
].p_ext
)
1443 b2
= le32_to_cpu(path
[depth
].p_ext
->ee_block
);
1446 * get the next allocated block if the extent in the path
1447 * is before the requested block(s)
1450 b2
= ext4_ext_next_allocated_block(path
);
1451 if (b2
== EXT_MAX_BLOCK
)
1455 /* check for wrap through zero on extent logical start block*/
1456 if (b1
+ len1
< b1
) {
1457 len1
= EXT_MAX_BLOCK
- b1
;
1458 newext
->ee_len
= cpu_to_le16(len1
);
1462 /* check for overlap */
1463 if (b1
+ len1
> b2
) {
1464 newext
->ee_len
= cpu_to_le16(b2
- b1
);
1472 * ext4_ext_insert_extent:
1473 * tries to merge requsted extent into the existing extent or
1474 * inserts requested extent as new one into the tree,
1475 * creating new leaf in the no-space case.
1477 int ext4_ext_insert_extent(handle_t
*handle
, struct inode
*inode
,
1478 struct ext4_ext_path
*path
,
1479 struct ext4_extent
*newext
)
1481 struct ext4_extent_header
*eh
;
1482 struct ext4_extent
*ex
, *fex
;
1483 struct ext4_extent
*nearex
; /* nearest extent */
1484 struct ext4_ext_path
*npath
= NULL
;
1485 int depth
, len
, err
;
1487 unsigned uninitialized
= 0;
1489 BUG_ON(ext4_ext_get_actual_len(newext
) == 0);
1490 depth
= ext_depth(inode
);
1491 ex
= path
[depth
].p_ext
;
1492 BUG_ON(path
[depth
].p_hdr
== NULL
);
1494 /* try to insert block into found extent and return */
1495 if (ex
&& ext4_can_extents_be_merged(inode
, ex
, newext
)) {
1496 ext_debug("append %d block to %d:%d (from %llu)\n",
1497 ext4_ext_get_actual_len(newext
),
1498 le32_to_cpu(ex
->ee_block
),
1499 ext4_ext_get_actual_len(ex
), ext_pblock(ex
));
1500 err
= ext4_ext_get_access(handle
, inode
, path
+ depth
);
1505 * ext4_can_extents_be_merged should have checked that either
1506 * both extents are uninitialized, or both aren't. Thus we
1507 * need to check only one of them here.
1509 if (ext4_ext_is_uninitialized(ex
))
1511 ex
->ee_len
= cpu_to_le16(ext4_ext_get_actual_len(ex
)
1512 + ext4_ext_get_actual_len(newext
));
1514 ext4_ext_mark_uninitialized(ex
);
1515 eh
= path
[depth
].p_hdr
;
1521 depth
= ext_depth(inode
);
1522 eh
= path
[depth
].p_hdr
;
1523 if (le16_to_cpu(eh
->eh_entries
) < le16_to_cpu(eh
->eh_max
))
1526 /* probably next leaf has space for us? */
1527 fex
= EXT_LAST_EXTENT(eh
);
1528 next
= ext4_ext_next_leaf_block(inode
, path
);
1529 if (le32_to_cpu(newext
->ee_block
) > le32_to_cpu(fex
->ee_block
)
1530 && next
!= EXT_MAX_BLOCK
) {
1531 ext_debug("next leaf block - %d\n", next
);
1532 BUG_ON(npath
!= NULL
);
1533 npath
= ext4_ext_find_extent(inode
, next
, NULL
);
1535 return PTR_ERR(npath
);
1536 BUG_ON(npath
->p_depth
!= path
->p_depth
);
1537 eh
= npath
[depth
].p_hdr
;
1538 if (le16_to_cpu(eh
->eh_entries
) < le16_to_cpu(eh
->eh_max
)) {
1539 ext_debug("next leaf isnt full(%d)\n",
1540 le16_to_cpu(eh
->eh_entries
));
1544 ext_debug("next leaf has no free space(%d,%d)\n",
1545 le16_to_cpu(eh
->eh_entries
), le16_to_cpu(eh
->eh_max
));
1549 * There is no free space in the found leaf.
1550 * We're gonna add a new leaf in the tree.
1552 err
= ext4_ext_create_new_leaf(handle
, inode
, path
, newext
);
1555 depth
= ext_depth(inode
);
1556 eh
= path
[depth
].p_hdr
;
1559 nearex
= path
[depth
].p_ext
;
1561 err
= ext4_ext_get_access(handle
, inode
, path
+ depth
);
1566 /* there is no extent in this leaf, create first one */
1567 ext_debug("first extent in the leaf: %d:%llu:%d\n",
1568 le32_to_cpu(newext
->ee_block
),
1570 ext4_ext_get_actual_len(newext
));
1571 path
[depth
].p_ext
= EXT_FIRST_EXTENT(eh
);
1572 } else if (le32_to_cpu(newext
->ee_block
)
1573 > le32_to_cpu(nearex
->ee_block
)) {
1574 /* BUG_ON(newext->ee_block == nearex->ee_block); */
1575 if (nearex
!= EXT_LAST_EXTENT(eh
)) {
1576 len
= EXT_MAX_EXTENT(eh
) - nearex
;
1577 len
= (len
- 1) * sizeof(struct ext4_extent
);
1578 len
= len
< 0 ? 0 : len
;
1579 ext_debug("insert %d:%llu:%d after: nearest 0x%p, "
1580 "move %d from 0x%p to 0x%p\n",
1581 le32_to_cpu(newext
->ee_block
),
1583 ext4_ext_get_actual_len(newext
),
1584 nearex
, len
, nearex
+ 1, nearex
+ 2);
1585 memmove(nearex
+ 2, nearex
+ 1, len
);
1587 path
[depth
].p_ext
= nearex
+ 1;
1589 BUG_ON(newext
->ee_block
== nearex
->ee_block
);
1590 len
= (EXT_MAX_EXTENT(eh
) - nearex
) * sizeof(struct ext4_extent
);
1591 len
= len
< 0 ? 0 : len
;
1592 ext_debug("insert %d:%llu:%d before: nearest 0x%p, "
1593 "move %d from 0x%p to 0x%p\n",
1594 le32_to_cpu(newext
->ee_block
),
1596 ext4_ext_get_actual_len(newext
),
1597 nearex
, len
, nearex
+ 1, nearex
+ 2);
1598 memmove(nearex
+ 1, nearex
, len
);
1599 path
[depth
].p_ext
= nearex
;
1602 le16_add_cpu(&eh
->eh_entries
, 1);
1603 nearex
= path
[depth
].p_ext
;
1604 nearex
->ee_block
= newext
->ee_block
;
1605 ext4_ext_store_pblock(nearex
, ext_pblock(newext
));
1606 nearex
->ee_len
= newext
->ee_len
;
1609 /* try to merge extents to the right */
1610 ext4_ext_try_to_merge(inode
, path
, nearex
);
1612 /* try to merge extents to the left */
1614 /* time to correct all indexes above */
1615 err
= ext4_ext_correct_indexes(handle
, inode
, path
);
1619 err
= ext4_ext_dirty(handle
, inode
, path
+ depth
);
1623 ext4_ext_drop_refs(npath
);
1626 ext4_ext_invalidate_cache(inode
);
1630 int ext4_ext_walk_space(struct inode
*inode
, ext4_lblk_t block
,
1631 ext4_lblk_t num
, ext_prepare_callback func
,
1634 struct ext4_ext_path
*path
= NULL
;
1635 struct ext4_ext_cache cbex
;
1636 struct ext4_extent
*ex
;
1637 ext4_lblk_t next
, start
= 0, end
= 0;
1638 ext4_lblk_t last
= block
+ num
;
1639 int depth
, exists
, err
= 0;
1641 BUG_ON(func
== NULL
);
1642 BUG_ON(inode
== NULL
);
1644 while (block
< last
&& block
!= EXT_MAX_BLOCK
) {
1646 /* find extent for this block */
1647 path
= ext4_ext_find_extent(inode
, block
, path
);
1649 err
= PTR_ERR(path
);
1654 depth
= ext_depth(inode
);
1655 BUG_ON(path
[depth
].p_hdr
== NULL
);
1656 ex
= path
[depth
].p_ext
;
1657 next
= ext4_ext_next_allocated_block(path
);
1661 /* there is no extent yet, so try to allocate
1662 * all requested space */
1665 } else if (le32_to_cpu(ex
->ee_block
) > block
) {
1666 /* need to allocate space before found extent */
1668 end
= le32_to_cpu(ex
->ee_block
);
1669 if (block
+ num
< end
)
1671 } else if (block
>= le32_to_cpu(ex
->ee_block
)
1672 + ext4_ext_get_actual_len(ex
)) {
1673 /* need to allocate space after found extent */
1678 } else if (block
>= le32_to_cpu(ex
->ee_block
)) {
1680 * some part of requested space is covered
1684 end
= le32_to_cpu(ex
->ee_block
)
1685 + ext4_ext_get_actual_len(ex
);
1686 if (block
+ num
< end
)
1692 BUG_ON(end
<= start
);
1695 cbex
.ec_block
= start
;
1696 cbex
.ec_len
= end
- start
;
1698 cbex
.ec_type
= EXT4_EXT_CACHE_GAP
;
1700 cbex
.ec_block
= le32_to_cpu(ex
->ee_block
);
1701 cbex
.ec_len
= ext4_ext_get_actual_len(ex
);
1702 cbex
.ec_start
= ext_pblock(ex
);
1703 cbex
.ec_type
= EXT4_EXT_CACHE_EXTENT
;
1706 BUG_ON(cbex
.ec_len
== 0);
1707 err
= func(inode
, path
, &cbex
, ex
, cbdata
);
1708 ext4_ext_drop_refs(path
);
1713 if (err
== EXT_REPEAT
)
1715 else if (err
== EXT_BREAK
) {
1720 if (ext_depth(inode
) != depth
) {
1721 /* depth was changed. we have to realloc path */
1726 block
= cbex
.ec_block
+ cbex
.ec_len
;
1730 ext4_ext_drop_refs(path
);
1738 ext4_ext_put_in_cache(struct inode
*inode
, ext4_lblk_t block
,
1739 __u32 len
, ext4_fsblk_t start
, int type
)
1741 struct ext4_ext_cache
*cex
;
1743 cex
= &EXT4_I(inode
)->i_cached_extent
;
1744 cex
->ec_type
= type
;
1745 cex
->ec_block
= block
;
1747 cex
->ec_start
= start
;
1751 * ext4_ext_put_gap_in_cache:
1752 * calculate boundaries of the gap that the requested block fits into
1753 * and cache this gap
1756 ext4_ext_put_gap_in_cache(struct inode
*inode
, struct ext4_ext_path
*path
,
1759 int depth
= ext_depth(inode
);
1762 struct ext4_extent
*ex
;
1764 ex
= path
[depth
].p_ext
;
1766 /* there is no extent yet, so gap is [0;-] */
1768 len
= EXT_MAX_BLOCK
;
1769 ext_debug("cache gap(whole file):");
1770 } else if (block
< le32_to_cpu(ex
->ee_block
)) {
1772 len
= le32_to_cpu(ex
->ee_block
) - block
;
1773 ext_debug("cache gap(before): %u [%u:%u]",
1775 le32_to_cpu(ex
->ee_block
),
1776 ext4_ext_get_actual_len(ex
));
1777 } else if (block
>= le32_to_cpu(ex
->ee_block
)
1778 + ext4_ext_get_actual_len(ex
)) {
1780 lblock
= le32_to_cpu(ex
->ee_block
)
1781 + ext4_ext_get_actual_len(ex
);
1783 next
= ext4_ext_next_allocated_block(path
);
1784 ext_debug("cache gap(after): [%u:%u] %u",
1785 le32_to_cpu(ex
->ee_block
),
1786 ext4_ext_get_actual_len(ex
),
1788 BUG_ON(next
== lblock
);
1789 len
= next
- lblock
;
1795 ext_debug(" -> %u:%lu\n", lblock
, len
);
1796 ext4_ext_put_in_cache(inode
, lblock
, len
, 0, EXT4_EXT_CACHE_GAP
);
1800 ext4_ext_in_cache(struct inode
*inode
, ext4_lblk_t block
,
1801 struct ext4_extent
*ex
)
1803 struct ext4_ext_cache
*cex
;
1805 cex
= &EXT4_I(inode
)->i_cached_extent
;
1807 /* has cache valid data? */
1808 if (cex
->ec_type
== EXT4_EXT_CACHE_NO
)
1809 return EXT4_EXT_CACHE_NO
;
1811 BUG_ON(cex
->ec_type
!= EXT4_EXT_CACHE_GAP
&&
1812 cex
->ec_type
!= EXT4_EXT_CACHE_EXTENT
);
1813 if (block
>= cex
->ec_block
&& block
< cex
->ec_block
+ cex
->ec_len
) {
1814 ex
->ee_block
= cpu_to_le32(cex
->ec_block
);
1815 ext4_ext_store_pblock(ex
, cex
->ec_start
);
1816 ex
->ee_len
= cpu_to_le16(cex
->ec_len
);
1817 ext_debug("%u cached by %u:%u:%llu\n",
1819 cex
->ec_block
, cex
->ec_len
, cex
->ec_start
);
1820 return cex
->ec_type
;
1824 return EXT4_EXT_CACHE_NO
;
1829 * removes index from the index block.
1830 * It's used in truncate case only, thus all requests are for
1831 * last index in the block only.
1833 static int ext4_ext_rm_idx(handle_t
*handle
, struct inode
*inode
,
1834 struct ext4_ext_path
*path
)
1836 struct buffer_head
*bh
;
1840 /* free index block */
1842 leaf
= idx_pblock(path
->p_idx
);
1843 BUG_ON(path
->p_hdr
->eh_entries
== 0);
1844 err
= ext4_ext_get_access(handle
, inode
, path
);
1847 le16_add_cpu(&path
->p_hdr
->eh_entries
, -1);
1848 err
= ext4_ext_dirty(handle
, inode
, path
);
1851 ext_debug("index is empty, remove it, free block %llu\n", leaf
);
1852 bh
= sb_find_get_block(inode
->i_sb
, leaf
);
1853 ext4_forget(handle
, 1, inode
, bh
, leaf
);
1854 ext4_free_blocks(handle
, inode
, leaf
, 1, 1);
1859 * ext4_ext_calc_credits_for_single_extent:
1860 * This routine returns max. credits that needed to insert an extent
1861 * to the extent tree.
1862 * When pass the actual path, the caller should calculate credits
1865 int ext4_ext_calc_credits_for_single_extent(struct inode
*inode
, int nrblocks
,
1866 struct ext4_ext_path
*path
)
1869 int depth
= ext_depth(inode
);
1872 /* probably there is space in leaf? */
1873 if (le16_to_cpu(path
[depth
].p_hdr
->eh_entries
)
1874 < le16_to_cpu(path
[depth
].p_hdr
->eh_max
)) {
1877 * There are some space in the leaf tree, no
1878 * need to account for leaf block credit
1880 * bitmaps and block group descriptor blocks
1881 * and other metadat blocks still need to be
1884 /* 1 bitmap, 1 block group descriptor */
1885 ret
= 2 + EXT4_META_TRANS_BLOCKS(inode
->i_sb
);
1889 return ext4_chunk_trans_blocks(inode
, nrblocks
);
1893 * How many index/leaf blocks need to change/allocate to modify nrblocks?
1895 * if nrblocks are fit in a single extent (chunk flag is 1), then
1896 * in the worse case, each tree level index/leaf need to be changed
1897 * if the tree split due to insert a new extent, then the old tree
1898 * index/leaf need to be updated too
1900 * If the nrblocks are discontiguous, they could cause
1901 * the whole tree split more than once, but this is really rare.
1903 int ext4_ext_index_trans_blocks(struct inode
*inode
, int nrblocks
, int chunk
)
1906 int depth
= ext_depth(inode
);
1916 static int ext4_remove_blocks(handle_t
*handle
, struct inode
*inode
,
1917 struct ext4_extent
*ex
,
1918 ext4_lblk_t from
, ext4_lblk_t to
)
1920 struct buffer_head
*bh
;
1921 unsigned short ee_len
= ext4_ext_get_actual_len(ex
);
1922 int i
, metadata
= 0;
1924 if (S_ISDIR(inode
->i_mode
) || S_ISLNK(inode
->i_mode
))
1926 #ifdef EXTENTS_STATS
1928 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
1929 spin_lock(&sbi
->s_ext_stats_lock
);
1930 sbi
->s_ext_blocks
+= ee_len
;
1931 sbi
->s_ext_extents
++;
1932 if (ee_len
< sbi
->s_ext_min
)
1933 sbi
->s_ext_min
= ee_len
;
1934 if (ee_len
> sbi
->s_ext_max
)
1935 sbi
->s_ext_max
= ee_len
;
1936 if (ext_depth(inode
) > sbi
->s_depth_max
)
1937 sbi
->s_depth_max
= ext_depth(inode
);
1938 spin_unlock(&sbi
->s_ext_stats_lock
);
1941 if (from
>= le32_to_cpu(ex
->ee_block
)
1942 && to
== le32_to_cpu(ex
->ee_block
) + ee_len
- 1) {
1947 num
= le32_to_cpu(ex
->ee_block
) + ee_len
- from
;
1948 start
= ext_pblock(ex
) + ee_len
- num
;
1949 ext_debug("free last %u blocks starting %llu\n", num
, start
);
1950 for (i
= 0; i
< num
; i
++) {
1951 bh
= sb_find_get_block(inode
->i_sb
, start
+ i
);
1952 ext4_forget(handle
, 0, inode
, bh
, start
+ i
);
1954 ext4_free_blocks(handle
, inode
, start
, num
, metadata
);
1955 } else if (from
== le32_to_cpu(ex
->ee_block
)
1956 && to
<= le32_to_cpu(ex
->ee_block
) + ee_len
- 1) {
1957 printk(KERN_INFO
"strange request: removal %u-%u from %u:%u\n",
1958 from
, to
, le32_to_cpu(ex
->ee_block
), ee_len
);
1960 printk(KERN_INFO
"strange request: removal(2) "
1961 "%u-%u from %u:%u\n",
1962 from
, to
, le32_to_cpu(ex
->ee_block
), ee_len
);
1968 ext4_ext_rm_leaf(handle_t
*handle
, struct inode
*inode
,
1969 struct ext4_ext_path
*path
, ext4_lblk_t start
)
1971 int err
= 0, correct_index
= 0;
1972 int depth
= ext_depth(inode
), credits
;
1973 struct ext4_extent_header
*eh
;
1974 ext4_lblk_t a
, b
, block
;
1976 ext4_lblk_t ex_ee_block
;
1977 unsigned short ex_ee_len
;
1978 unsigned uninitialized
= 0;
1979 struct ext4_extent
*ex
;
1981 /* the header must be checked already in ext4_ext_remove_space() */
1982 ext_debug("truncate since %u in leaf\n", start
);
1983 if (!path
[depth
].p_hdr
)
1984 path
[depth
].p_hdr
= ext_block_hdr(path
[depth
].p_bh
);
1985 eh
= path
[depth
].p_hdr
;
1988 /* find where to start removing */
1989 ex
= EXT_LAST_EXTENT(eh
);
1991 ex_ee_block
= le32_to_cpu(ex
->ee_block
);
1992 if (ext4_ext_is_uninitialized(ex
))
1994 ex_ee_len
= ext4_ext_get_actual_len(ex
);
1996 while (ex
>= EXT_FIRST_EXTENT(eh
) &&
1997 ex_ee_block
+ ex_ee_len
> start
) {
1998 ext_debug("remove ext %lu:%u\n", ex_ee_block
, ex_ee_len
);
1999 path
[depth
].p_ext
= ex
;
2001 a
= ex_ee_block
> start
? ex_ee_block
: start
;
2002 b
= ex_ee_block
+ ex_ee_len
- 1 < EXT_MAX_BLOCK
?
2003 ex_ee_block
+ ex_ee_len
- 1 : EXT_MAX_BLOCK
;
2005 ext_debug(" border %u:%u\n", a
, b
);
2007 if (a
!= ex_ee_block
&& b
!= ex_ee_block
+ ex_ee_len
- 1) {
2011 } else if (a
!= ex_ee_block
) {
2012 /* remove tail of the extent */
2013 block
= ex_ee_block
;
2015 } else if (b
!= ex_ee_block
+ ex_ee_len
- 1) {
2016 /* remove head of the extent */
2019 /* there is no "make a hole" API yet */
2022 /* remove whole extent: excellent! */
2023 block
= ex_ee_block
;
2025 BUG_ON(a
!= ex_ee_block
);
2026 BUG_ON(b
!= ex_ee_block
+ ex_ee_len
- 1);
2030 * 3 for leaf, sb, and inode plus 2 (bmap and group
2031 * descriptor) for each block group; assume two block
2032 * groups plus ex_ee_len/blocks_per_block_group for
2035 credits
= 7 + 2*(ex_ee_len
/EXT4_BLOCKS_PER_GROUP(inode
->i_sb
));
2036 if (ex
== EXT_FIRST_EXTENT(eh
)) {
2038 credits
+= (ext_depth(inode
)) + 1;
2040 credits
+= 2 * EXT4_QUOTA_TRANS_BLOCKS(inode
->i_sb
);
2042 err
= ext4_ext_journal_restart(handle
, credits
);
2046 err
= ext4_ext_get_access(handle
, inode
, path
+ depth
);
2050 err
= ext4_remove_blocks(handle
, inode
, ex
, a
, b
);
2055 /* this extent is removed; mark slot entirely unused */
2056 ext4_ext_store_pblock(ex
, 0);
2057 le16_add_cpu(&eh
->eh_entries
, -1);
2060 ex
->ee_block
= cpu_to_le32(block
);
2061 ex
->ee_len
= cpu_to_le16(num
);
2063 * Do not mark uninitialized if all the blocks in the
2064 * extent have been removed.
2066 if (uninitialized
&& num
)
2067 ext4_ext_mark_uninitialized(ex
);
2069 err
= ext4_ext_dirty(handle
, inode
, path
+ depth
);
2073 ext_debug("new extent: %u:%u:%llu\n", block
, num
,
2076 ex_ee_block
= le32_to_cpu(ex
->ee_block
);
2077 ex_ee_len
= ext4_ext_get_actual_len(ex
);
2080 if (correct_index
&& eh
->eh_entries
)
2081 err
= ext4_ext_correct_indexes(handle
, inode
, path
);
2083 /* if this leaf is free, then we should
2084 * remove it from index block above */
2085 if (err
== 0 && eh
->eh_entries
== 0 && path
[depth
].p_bh
!= NULL
)
2086 err
= ext4_ext_rm_idx(handle
, inode
, path
+ depth
);
2093 * ext4_ext_more_to_rm:
2094 * returns 1 if current index has to be freed (even partial)
2097 ext4_ext_more_to_rm(struct ext4_ext_path
*path
)
2099 BUG_ON(path
->p_idx
== NULL
);
2101 if (path
->p_idx
< EXT_FIRST_INDEX(path
->p_hdr
))
2105 * if truncate on deeper level happened, it wasn't partial,
2106 * so we have to consider current index for truncation
2108 if (le16_to_cpu(path
->p_hdr
->eh_entries
) == path
->p_block
)
2113 static int ext4_ext_remove_space(struct inode
*inode
, ext4_lblk_t start
)
2115 struct super_block
*sb
= inode
->i_sb
;
2116 int depth
= ext_depth(inode
);
2117 struct ext4_ext_path
*path
;
2121 ext_debug("truncate since %u\n", start
);
2123 /* probably first extent we're gonna free will be last in block */
2124 handle
= ext4_journal_start(inode
, depth
+ 1);
2126 return PTR_ERR(handle
);
2128 ext4_ext_invalidate_cache(inode
);
2131 * We start scanning from right side, freeing all the blocks
2132 * after i_size and walking into the tree depth-wise.
2134 path
= kzalloc(sizeof(struct ext4_ext_path
) * (depth
+ 1), GFP_NOFS
);
2136 ext4_journal_stop(handle
);
2139 path
[0].p_hdr
= ext_inode_hdr(inode
);
2140 if (ext4_ext_check_header(inode
, path
[0].p_hdr
, depth
)) {
2144 path
[0].p_depth
= depth
;
2146 while (i
>= 0 && err
== 0) {
2148 /* this is leaf block */
2149 err
= ext4_ext_rm_leaf(handle
, inode
, path
, start
);
2150 /* root level has p_bh == NULL, brelse() eats this */
2151 brelse(path
[i
].p_bh
);
2152 path
[i
].p_bh
= NULL
;
2157 /* this is index block */
2158 if (!path
[i
].p_hdr
) {
2159 ext_debug("initialize header\n");
2160 path
[i
].p_hdr
= ext_block_hdr(path
[i
].p_bh
);
2163 if (!path
[i
].p_idx
) {
2164 /* this level hasn't been touched yet */
2165 path
[i
].p_idx
= EXT_LAST_INDEX(path
[i
].p_hdr
);
2166 path
[i
].p_block
= le16_to_cpu(path
[i
].p_hdr
->eh_entries
)+1;
2167 ext_debug("init index ptr: hdr 0x%p, num %d\n",
2169 le16_to_cpu(path
[i
].p_hdr
->eh_entries
));
2171 /* we were already here, see at next index */
2175 ext_debug("level %d - index, first 0x%p, cur 0x%p\n",
2176 i
, EXT_FIRST_INDEX(path
[i
].p_hdr
),
2178 if (ext4_ext_more_to_rm(path
+ i
)) {
2179 struct buffer_head
*bh
;
2180 /* go to the next level */
2181 ext_debug("move to level %d (block %llu)\n",
2182 i
+ 1, idx_pblock(path
[i
].p_idx
));
2183 memset(path
+ i
+ 1, 0, sizeof(*path
));
2184 bh
= sb_bread(sb
, idx_pblock(path
[i
].p_idx
));
2186 /* should we reset i_size? */
2190 if (WARN_ON(i
+ 1 > depth
)) {
2194 if (ext4_ext_check_header(inode
, ext_block_hdr(bh
),
2199 path
[i
+ 1].p_bh
= bh
;
2201 /* save actual number of indexes since this
2202 * number is changed at the next iteration */
2203 path
[i
].p_block
= le16_to_cpu(path
[i
].p_hdr
->eh_entries
);
2206 /* we finished processing this index, go up */
2207 if (path
[i
].p_hdr
->eh_entries
== 0 && i
> 0) {
2208 /* index is empty, remove it;
2209 * handle must be already prepared by the
2210 * truncatei_leaf() */
2211 err
= ext4_ext_rm_idx(handle
, inode
, path
+ i
);
2213 /* root level has p_bh == NULL, brelse() eats this */
2214 brelse(path
[i
].p_bh
);
2215 path
[i
].p_bh
= NULL
;
2217 ext_debug("return to level %d\n", i
);
2221 /* TODO: flexible tree reduction should be here */
2222 if (path
->p_hdr
->eh_entries
== 0) {
2224 * truncate to zero freed all the tree,
2225 * so we need to correct eh_depth
2227 err
= ext4_ext_get_access(handle
, inode
, path
);
2229 ext_inode_hdr(inode
)->eh_depth
= 0;
2230 ext_inode_hdr(inode
)->eh_max
=
2231 cpu_to_le16(ext4_ext_space_root(inode
));
2232 err
= ext4_ext_dirty(handle
, inode
, path
);
2236 ext4_ext_drop_refs(path
);
2238 ext4_journal_stop(handle
);
2244 * called at mount time
2246 void ext4_ext_init(struct super_block
*sb
)
2249 * possible initialization would be here
2252 if (EXT4_HAS_INCOMPAT_FEATURE(sb
, EXT4_FEATURE_INCOMPAT_EXTENTS
)) {
2253 printk(KERN_INFO
"EXT4-fs: file extents enabled");
2254 #ifdef AGGRESSIVE_TEST
2255 printk(", aggressive tests");
2257 #ifdef CHECK_BINSEARCH
2258 printk(", check binsearch");
2260 #ifdef EXTENTS_STATS
2264 #ifdef EXTENTS_STATS
2265 spin_lock_init(&EXT4_SB(sb
)->s_ext_stats_lock
);
2266 EXT4_SB(sb
)->s_ext_min
= 1 << 30;
2267 EXT4_SB(sb
)->s_ext_max
= 0;
2273 * called at umount time
2275 void ext4_ext_release(struct super_block
*sb
)
2277 if (!EXT4_HAS_INCOMPAT_FEATURE(sb
, EXT4_FEATURE_INCOMPAT_EXTENTS
))
2280 #ifdef EXTENTS_STATS
2281 if (EXT4_SB(sb
)->s_ext_blocks
&& EXT4_SB(sb
)->s_ext_extents
) {
2282 struct ext4_sb_info
*sbi
= EXT4_SB(sb
);
2283 printk(KERN_ERR
"EXT4-fs: %lu blocks in %lu extents (%lu ave)\n",
2284 sbi
->s_ext_blocks
, sbi
->s_ext_extents
,
2285 sbi
->s_ext_blocks
/ sbi
->s_ext_extents
);
2286 printk(KERN_ERR
"EXT4-fs: extents: %lu min, %lu max, max depth %lu\n",
2287 sbi
->s_ext_min
, sbi
->s_ext_max
, sbi
->s_depth_max
);
2292 static void bi_complete(struct bio
*bio
, int error
)
2294 complete((struct completion
*)bio
->bi_private
);
2297 /* FIXME!! we need to try to merge to left or right after zero-out */
2298 static int ext4_ext_zeroout(struct inode
*inode
, struct ext4_extent
*ex
)
2302 int blkbits
, blocksize
;
2304 struct completion event
;
2305 unsigned int ee_len
, len
, done
, offset
;
2308 blkbits
= inode
->i_blkbits
;
2309 blocksize
= inode
->i_sb
->s_blocksize
;
2310 ee_len
= ext4_ext_get_actual_len(ex
);
2311 ee_pblock
= ext_pblock(ex
);
2313 /* convert ee_pblock to 512 byte sectors */
2314 ee_pblock
= ee_pblock
<< (blkbits
- 9);
2316 while (ee_len
> 0) {
2318 if (ee_len
> BIO_MAX_PAGES
)
2319 len
= BIO_MAX_PAGES
;
2323 bio
= bio_alloc(GFP_NOIO
, len
);
2326 bio
->bi_sector
= ee_pblock
;
2327 bio
->bi_bdev
= inode
->i_sb
->s_bdev
;
2331 while (done
< len
) {
2332 ret
= bio_add_page(bio
, ZERO_PAGE(0),
2334 if (ret
!= blocksize
) {
2336 * We can't add any more pages because of
2337 * hardware limitations. Start a new bio.
2342 offset
+= blocksize
;
2343 if (offset
>= PAGE_CACHE_SIZE
)
2347 init_completion(&event
);
2348 bio
->bi_private
= &event
;
2349 bio
->bi_end_io
= bi_complete
;
2350 submit_bio(WRITE
, bio
);
2351 wait_for_completion(&event
);
2353 if (test_bit(BIO_UPTODATE
, &bio
->bi_flags
))
2361 ee_pblock
+= done
<< (blkbits
- 9);
2366 #define EXT4_EXT_ZERO_LEN 7
2369 * This function is called by ext4_ext_get_blocks() if someone tries to write
2370 * to an uninitialized extent. It may result in splitting the uninitialized
2371 * extent into multiple extents (upto three - one initialized and two
2373 * There are three possibilities:
2374 * a> There is no split required: Entire extent should be initialized
2375 * b> Splits in two extents: Write is happening at either end of the extent
2376 * c> Splits in three extents: Somone is writing in middle of the extent
2378 static int ext4_ext_convert_to_initialized(handle_t
*handle
,
2379 struct inode
*inode
,
2380 struct ext4_ext_path
*path
,
2382 unsigned int max_blocks
)
2384 struct ext4_extent
*ex
, newex
, orig_ex
;
2385 struct ext4_extent
*ex1
= NULL
;
2386 struct ext4_extent
*ex2
= NULL
;
2387 struct ext4_extent
*ex3
= NULL
;
2388 struct ext4_extent_header
*eh
;
2389 ext4_lblk_t ee_block
;
2390 unsigned int allocated
, ee_len
, depth
;
2391 ext4_fsblk_t newblock
;
2395 depth
= ext_depth(inode
);
2396 eh
= path
[depth
].p_hdr
;
2397 ex
= path
[depth
].p_ext
;
2398 ee_block
= le32_to_cpu(ex
->ee_block
);
2399 ee_len
= ext4_ext_get_actual_len(ex
);
2400 allocated
= ee_len
- (iblock
- ee_block
);
2401 newblock
= iblock
- ee_block
+ ext_pblock(ex
);
2403 orig_ex
.ee_block
= ex
->ee_block
;
2404 orig_ex
.ee_len
= cpu_to_le16(ee_len
);
2405 ext4_ext_store_pblock(&orig_ex
, ext_pblock(ex
));
2407 err
= ext4_ext_get_access(handle
, inode
, path
+ depth
);
2410 /* If extent has less than 2*EXT4_EXT_ZERO_LEN zerout directly */
2411 if (ee_len
<= 2*EXT4_EXT_ZERO_LEN
) {
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 /* zeroed the full extent */
2424 /* ex1: ee_block to iblock - 1 : uninitialized */
2425 if (iblock
> ee_block
) {
2427 ex1
->ee_len
= cpu_to_le16(iblock
- ee_block
);
2428 ext4_ext_mark_uninitialized(ex1
);
2432 * for sanity, update the length of the ex2 extent before
2433 * we insert ex3, if ex1 is NULL. This is to avoid temporary
2434 * overlap of blocks.
2436 if (!ex1
&& allocated
> max_blocks
)
2437 ex2
->ee_len
= cpu_to_le16(max_blocks
);
2438 /* ex3: to ee_block + ee_len : uninitialised */
2439 if (allocated
> max_blocks
) {
2440 unsigned int newdepth
;
2441 /* If extent has less than EXT4_EXT_ZERO_LEN zerout directly */
2442 if (allocated
<= EXT4_EXT_ZERO_LEN
) {
2444 * iblock == ee_block is handled by the zerouout
2446 * Mark first half uninitialized.
2447 * Mark second half initialized and zero out the
2448 * initialized extent
2450 ex
->ee_block
= orig_ex
.ee_block
;
2451 ex
->ee_len
= cpu_to_le16(ee_len
- allocated
);
2452 ext4_ext_mark_uninitialized(ex
);
2453 ext4_ext_store_pblock(ex
, ext_pblock(&orig_ex
));
2454 ext4_ext_dirty(handle
, inode
, path
+ depth
);
2457 ex3
->ee_block
= cpu_to_le32(iblock
);
2458 ext4_ext_store_pblock(ex3
, newblock
);
2459 ex3
->ee_len
= cpu_to_le16(allocated
);
2460 err
= ext4_ext_insert_extent(handle
, inode
, path
, ex3
);
2461 if (err
== -ENOSPC
) {
2462 err
= ext4_ext_zeroout(inode
, &orig_ex
);
2464 goto fix_extent_len
;
2465 ex
->ee_block
= orig_ex
.ee_block
;
2466 ex
->ee_len
= orig_ex
.ee_len
;
2467 ext4_ext_store_pblock(ex
, ext_pblock(&orig_ex
));
2468 ext4_ext_dirty(handle
, inode
, path
+ depth
);
2469 /* blocks available from iblock */
2473 goto fix_extent_len
;
2476 * We need to zero out the second half because
2477 * an fallocate request can update file size and
2478 * converting the second half to initialized extent
2479 * implies that we can leak some junk data to user
2482 err
= ext4_ext_zeroout(inode
, ex3
);
2485 * We should actually mark the
2486 * second half as uninit and return error
2487 * Insert would have changed the extent
2489 depth
= ext_depth(inode
);
2490 ext4_ext_drop_refs(path
);
2491 path
= ext4_ext_find_extent(inode
,
2494 err
= PTR_ERR(path
);
2497 /* get the second half extent details */
2498 ex
= path
[depth
].p_ext
;
2499 err
= ext4_ext_get_access(handle
, inode
,
2503 ext4_ext_mark_uninitialized(ex
);
2504 ext4_ext_dirty(handle
, inode
, path
+ depth
);
2508 /* zeroed the second half */
2512 ex3
->ee_block
= cpu_to_le32(iblock
+ max_blocks
);
2513 ext4_ext_store_pblock(ex3
, newblock
+ max_blocks
);
2514 ex3
->ee_len
= cpu_to_le16(allocated
- max_blocks
);
2515 ext4_ext_mark_uninitialized(ex3
);
2516 err
= ext4_ext_insert_extent(handle
, inode
, path
, ex3
);
2517 if (err
== -ENOSPC
) {
2518 err
= ext4_ext_zeroout(inode
, &orig_ex
);
2520 goto fix_extent_len
;
2521 /* update the extent length and mark as initialized */
2522 ex
->ee_block
= orig_ex
.ee_block
;
2523 ex
->ee_len
= orig_ex
.ee_len
;
2524 ext4_ext_store_pblock(ex
, ext_pblock(&orig_ex
));
2525 ext4_ext_dirty(handle
, inode
, path
+ depth
);
2526 /* zeroed the full extent */
2527 /* blocks available from iblock */
2531 goto fix_extent_len
;
2533 * The depth, and hence eh & ex might change
2534 * as part of the insert above.
2536 newdepth
= ext_depth(inode
);
2538 * update the extent length after successful insert of the
2541 orig_ex
.ee_len
= cpu_to_le16(ee_len
-
2542 ext4_ext_get_actual_len(ex3
));
2544 ext4_ext_drop_refs(path
);
2545 path
= ext4_ext_find_extent(inode
, iblock
, path
);
2547 err
= PTR_ERR(path
);
2550 eh
= path
[depth
].p_hdr
;
2551 ex
= path
[depth
].p_ext
;
2555 err
= ext4_ext_get_access(handle
, inode
, path
+ depth
);
2559 allocated
= max_blocks
;
2561 /* If extent has less than EXT4_EXT_ZERO_LEN and we are trying
2562 * to insert a extent in the middle zerout directly
2563 * otherwise give the extent a chance to merge to left
2565 if (le16_to_cpu(orig_ex
.ee_len
) <= EXT4_EXT_ZERO_LEN
&&
2566 iblock
!= ee_block
) {
2567 err
= ext4_ext_zeroout(inode
, &orig_ex
);
2569 goto fix_extent_len
;
2570 /* update the extent length and mark as initialized */
2571 ex
->ee_block
= orig_ex
.ee_block
;
2572 ex
->ee_len
= orig_ex
.ee_len
;
2573 ext4_ext_store_pblock(ex
, ext_pblock(&orig_ex
));
2574 ext4_ext_dirty(handle
, inode
, path
+ depth
);
2575 /* zero out the first half */
2576 /* blocks available from iblock */
2581 * If there was a change of depth as part of the
2582 * insertion of ex3 above, we need to update the length
2583 * of the ex1 extent again here
2585 if (ex1
&& ex1
!= ex
) {
2587 ex1
->ee_len
= cpu_to_le16(iblock
- ee_block
);
2588 ext4_ext_mark_uninitialized(ex1
);
2591 /* ex2: iblock to iblock + maxblocks-1 : initialised */
2592 ex2
->ee_block
= cpu_to_le32(iblock
);
2593 ext4_ext_store_pblock(ex2
, newblock
);
2594 ex2
->ee_len
= cpu_to_le16(allocated
);
2598 * New (initialized) extent starts from the first block
2599 * in the current extent. i.e., ex2 == ex
2600 * We have to see if it can be merged with the extent
2603 if (ex2
> EXT_FIRST_EXTENT(eh
)) {
2605 * To merge left, pass "ex2 - 1" to try_to_merge(),
2606 * since it merges towards right _only_.
2608 ret
= ext4_ext_try_to_merge(inode
, path
, ex2
- 1);
2610 err
= ext4_ext_correct_indexes(handle
, inode
, path
);
2613 depth
= ext_depth(inode
);
2618 * Try to Merge towards right. This might be required
2619 * only when the whole extent is being written to.
2620 * i.e. ex2 == ex and ex3 == NULL.
2623 ret
= ext4_ext_try_to_merge(inode
, path
, ex2
);
2625 err
= ext4_ext_correct_indexes(handle
, inode
, path
);
2630 /* Mark modified extent as dirty */
2631 err
= ext4_ext_dirty(handle
, inode
, path
+ depth
);
2634 err
= ext4_ext_insert_extent(handle
, inode
, path
, &newex
);
2635 if (err
== -ENOSPC
) {
2636 err
= ext4_ext_zeroout(inode
, &orig_ex
);
2638 goto fix_extent_len
;
2639 /* update the extent length and mark as initialized */
2640 ex
->ee_block
= orig_ex
.ee_block
;
2641 ex
->ee_len
= orig_ex
.ee_len
;
2642 ext4_ext_store_pblock(ex
, ext_pblock(&orig_ex
));
2643 ext4_ext_dirty(handle
, inode
, path
+ depth
);
2644 /* zero out the first half */
2647 goto fix_extent_len
;
2649 return err
? err
: allocated
;
2652 ex
->ee_block
= orig_ex
.ee_block
;
2653 ex
->ee_len
= orig_ex
.ee_len
;
2654 ext4_ext_store_pblock(ex
, ext_pblock(&orig_ex
));
2655 ext4_ext_mark_uninitialized(ex
);
2656 ext4_ext_dirty(handle
, inode
, path
+ depth
);
2661 * Block allocation/map/preallocation routine for extents based files
2664 * Need to be called with
2665 * down_read(&EXT4_I(inode)->i_data_sem) if not allocating file system block
2666 * (ie, create is zero). Otherwise down_write(&EXT4_I(inode)->i_data_sem)
2668 * return > 0, number of of blocks already mapped/allocated
2669 * if create == 0 and these are pre-allocated blocks
2670 * buffer head is unmapped
2671 * otherwise blocks are mapped
2673 * return = 0, if plain look up failed (blocks have not been allocated)
2674 * buffer head is unmapped
2676 * return < 0, error case.
2678 int ext4_ext_get_blocks(handle_t
*handle
, struct inode
*inode
,
2680 unsigned int max_blocks
, struct buffer_head
*bh_result
,
2681 int create
, int extend_disksize
)
2683 struct ext4_ext_path
*path
= NULL
;
2684 struct ext4_extent_header
*eh
;
2685 struct ext4_extent newex
, *ex
;
2686 ext4_fsblk_t newblock
;
2687 int err
= 0, depth
, ret
, cache_type
;
2688 unsigned int allocated
= 0;
2689 struct ext4_allocation_request ar
;
2692 __clear_bit(BH_New
, &bh_result
->b_state
);
2693 ext_debug("blocks %u/%u requested for inode %u\n",
2694 iblock
, max_blocks
, inode
->i_ino
);
2696 /* check in cache */
2697 cache_type
= ext4_ext_in_cache(inode
, iblock
, &newex
);
2699 if (cache_type
== EXT4_EXT_CACHE_GAP
) {
2702 * block isn't allocated yet and
2703 * user doesn't want to allocate it
2707 /* we should allocate requested block */
2708 } else if (cache_type
== EXT4_EXT_CACHE_EXTENT
) {
2709 /* block is already allocated */
2711 - le32_to_cpu(newex
.ee_block
)
2712 + ext_pblock(&newex
);
2713 /* number of remaining blocks in the extent */
2714 allocated
= ext4_ext_get_actual_len(&newex
) -
2715 (iblock
- le32_to_cpu(newex
.ee_block
));
2722 /* find extent for this block */
2723 path
= ext4_ext_find_extent(inode
, iblock
, NULL
);
2725 err
= PTR_ERR(path
);
2730 depth
= ext_depth(inode
);
2733 * consistent leaf must not be empty;
2734 * this situation is possible, though, _during_ tree modification;
2735 * this is why assert can't be put in ext4_ext_find_extent()
2737 BUG_ON(path
[depth
].p_ext
== NULL
&& depth
!= 0);
2738 eh
= path
[depth
].p_hdr
;
2740 ex
= path
[depth
].p_ext
;
2742 ext4_lblk_t ee_block
= le32_to_cpu(ex
->ee_block
);
2743 ext4_fsblk_t ee_start
= ext_pblock(ex
);
2744 unsigned short ee_len
;
2747 * Uninitialized extents are treated as holes, except that
2748 * we split out initialized portions during a write.
2750 ee_len
= ext4_ext_get_actual_len(ex
);
2751 /* if found extent covers block, simply return it */
2752 if (iblock
>= ee_block
&& iblock
< ee_block
+ ee_len
) {
2753 newblock
= iblock
- ee_block
+ ee_start
;
2754 /* number of remaining blocks in the extent */
2755 allocated
= ee_len
- (iblock
- ee_block
);
2756 ext_debug("%u fit into %lu:%d -> %llu\n", iblock
,
2757 ee_block
, ee_len
, newblock
);
2759 /* Do not put uninitialized extent in the cache */
2760 if (!ext4_ext_is_uninitialized(ex
)) {
2761 ext4_ext_put_in_cache(inode
, ee_block
,
2763 EXT4_EXT_CACHE_EXTENT
);
2766 if (create
== EXT4_CREATE_UNINITIALIZED_EXT
)
2770 * We have blocks reserved already. We
2771 * return allocated blocks so that delalloc
2772 * won't do block reservation for us. But
2773 * the buffer head will be unmapped so that
2774 * a read from the block returns 0s.
2776 if (allocated
> max_blocks
)
2777 allocated
= max_blocks
;
2778 set_buffer_unwritten(bh_result
);
2782 ret
= ext4_ext_convert_to_initialized(handle
, inode
,
2795 * requested block isn't allocated yet;
2796 * we couldn't try to create block if create flag is zero
2800 * put just found gap into cache to speed up
2801 * subsequent requests
2803 ext4_ext_put_gap_in_cache(inode
, path
, iblock
);
2807 * Okay, we need to do block allocation.
2810 /* find neighbour allocated blocks */
2812 err
= ext4_ext_search_left(inode
, path
, &ar
.lleft
, &ar
.pleft
);
2816 err
= ext4_ext_search_right(inode
, path
, &ar
.lright
, &ar
.pright
);
2821 * See if request is beyond maximum number of blocks we can have in
2822 * a single extent. For an initialized extent this limit is
2823 * EXT_INIT_MAX_LEN and for an uninitialized extent this limit is
2824 * EXT_UNINIT_MAX_LEN.
2826 if (max_blocks
> EXT_INIT_MAX_LEN
&&
2827 create
!= EXT4_CREATE_UNINITIALIZED_EXT
)
2828 max_blocks
= EXT_INIT_MAX_LEN
;
2829 else if (max_blocks
> EXT_UNINIT_MAX_LEN
&&
2830 create
== EXT4_CREATE_UNINITIALIZED_EXT
)
2831 max_blocks
= EXT_UNINIT_MAX_LEN
;
2833 /* Check if we can really insert (iblock)::(iblock+max_blocks) extent */
2834 newex
.ee_block
= cpu_to_le32(iblock
);
2835 newex
.ee_len
= cpu_to_le16(max_blocks
);
2836 err
= ext4_ext_check_overlap(inode
, &newex
, path
);
2838 allocated
= ext4_ext_get_actual_len(&newex
);
2840 allocated
= max_blocks
;
2842 /* allocate new block */
2844 ar
.goal
= ext4_ext_find_goal(inode
, path
, iblock
);
2845 ar
.logical
= iblock
;
2847 if (S_ISREG(inode
->i_mode
))
2848 ar
.flags
= EXT4_MB_HINT_DATA
;
2850 /* disable in-core preallocation for non-regular files */
2852 newblock
= ext4_mb_new_blocks(handle
, &ar
, &err
);
2855 ext_debug("allocate new block: goal %llu, found %llu/%lu\n",
2856 ar
.goal
, newblock
, allocated
);
2858 /* try to insert new extent into found leaf and return */
2859 ext4_ext_store_pblock(&newex
, newblock
);
2860 newex
.ee_len
= cpu_to_le16(ar
.len
);
2861 if (create
== EXT4_CREATE_UNINITIALIZED_EXT
) /* Mark uninitialized */
2862 ext4_ext_mark_uninitialized(&newex
);
2863 err
= ext4_ext_insert_extent(handle
, inode
, path
, &newex
);
2865 /* free data blocks we just allocated */
2866 /* not a good idea to call discard here directly,
2867 * but otherwise we'd need to call it every free() */
2868 ext4_discard_preallocations(inode
);
2869 ext4_free_blocks(handle
, inode
, ext_pblock(&newex
),
2870 ext4_ext_get_actual_len(&newex
), 0);
2874 /* previous routine could use block we allocated */
2875 newblock
= ext_pblock(&newex
);
2876 allocated
= ext4_ext_get_actual_len(&newex
);
2878 if (extend_disksize
) {
2879 disksize
= ((loff_t
) iblock
+ ar
.len
) << inode
->i_blkbits
;
2880 if (disksize
> i_size_read(inode
))
2881 disksize
= i_size_read(inode
);
2882 if (disksize
> EXT4_I(inode
)->i_disksize
)
2883 EXT4_I(inode
)->i_disksize
= disksize
;
2886 set_buffer_new(bh_result
);
2888 /* Cache only when it is _not_ an uninitialized extent */
2889 if (create
!= EXT4_CREATE_UNINITIALIZED_EXT
)
2890 ext4_ext_put_in_cache(inode
, iblock
, allocated
, newblock
,
2891 EXT4_EXT_CACHE_EXTENT
);
2893 if (allocated
> max_blocks
)
2894 allocated
= max_blocks
;
2895 ext4_ext_show_leaf(inode
, path
);
2896 set_buffer_mapped(bh_result
);
2897 bh_result
->b_bdev
= inode
->i_sb
->s_bdev
;
2898 bh_result
->b_blocknr
= newblock
;
2901 ext4_ext_drop_refs(path
);
2904 return err
? err
: allocated
;
2907 void ext4_ext_truncate(struct inode
*inode
)
2909 struct address_space
*mapping
= inode
->i_mapping
;
2910 struct super_block
*sb
= inode
->i_sb
;
2911 ext4_lblk_t last_block
;
2916 * probably first extent we're gonna free will be last in block
2918 err
= ext4_writepage_trans_blocks(inode
);
2919 handle
= ext4_journal_start(inode
, err
);
2923 if (inode
->i_size
& (sb
->s_blocksize
- 1))
2924 ext4_block_truncate_page(handle
, mapping
, inode
->i_size
);
2926 if (ext4_orphan_add(handle
, inode
))
2929 down_write(&EXT4_I(inode
)->i_data_sem
);
2930 ext4_ext_invalidate_cache(inode
);
2932 ext4_discard_preallocations(inode
);
2935 * TODO: optimization is possible here.
2936 * Probably we need not scan at all,
2937 * because page truncation is enough.
2940 /* we have to know where to truncate from in crash case */
2941 EXT4_I(inode
)->i_disksize
= inode
->i_size
;
2942 ext4_mark_inode_dirty(handle
, inode
);
2944 last_block
= (inode
->i_size
+ sb
->s_blocksize
- 1)
2945 >> EXT4_BLOCK_SIZE_BITS(sb
);
2946 err
= ext4_ext_remove_space(inode
, last_block
);
2948 /* In a multi-transaction truncate, we only make the final
2949 * transaction synchronous.
2952 ext4_handle_sync(handle
);
2955 up_write(&EXT4_I(inode
)->i_data_sem
);
2957 * If this was a simple ftruncate() and the file will remain alive,
2958 * then we need to clear up the orphan record which we created above.
2959 * However, if this was a real unlink then we were called by
2960 * ext4_delete_inode(), and we allow that function to clean up the
2961 * orphan info for us.
2964 ext4_orphan_del(handle
, inode
);
2966 inode
->i_mtime
= inode
->i_ctime
= ext4_current_time(inode
);
2967 ext4_mark_inode_dirty(handle
, inode
);
2968 ext4_journal_stop(handle
);
2971 static void ext4_falloc_update_inode(struct inode
*inode
,
2972 int mode
, loff_t new_size
, int update_ctime
)
2974 struct timespec now
;
2977 now
= current_fs_time(inode
->i_sb
);
2978 if (!timespec_equal(&inode
->i_ctime
, &now
))
2979 inode
->i_ctime
= now
;
2982 * Update only when preallocation was requested beyond
2985 if (!(mode
& FALLOC_FL_KEEP_SIZE
)) {
2986 if (new_size
> i_size_read(inode
))
2987 i_size_write(inode
, new_size
);
2988 if (new_size
> EXT4_I(inode
)->i_disksize
)
2989 ext4_update_i_disksize(inode
, new_size
);
2995 * preallocate space for a file. This implements ext4's fallocate inode
2996 * operation, which gets called from sys_fallocate system call.
2997 * For block-mapped files, posix_fallocate should fall back to the method
2998 * of writing zeroes to the required new blocks (the same behavior which is
2999 * expected for file systems which do not support fallocate() system call).
3001 long ext4_fallocate(struct inode
*inode
, int mode
, loff_t offset
, loff_t len
)
3006 unsigned int max_blocks
;
3010 struct buffer_head map_bh
;
3011 unsigned int credits
, blkbits
= inode
->i_blkbits
;
3014 * currently supporting (pre)allocate mode for extent-based
3017 if (!(EXT4_I(inode
)->i_flags
& EXT4_EXTENTS_FL
))
3020 /* preallocation to directories is currently not supported */
3021 if (S_ISDIR(inode
->i_mode
))
3024 block
= offset
>> blkbits
;
3026 * We can't just convert len to max_blocks because
3027 * If blocksize = 4096 offset = 3072 and len = 2048
3029 max_blocks
= (EXT4_BLOCK_ALIGN(len
+ offset
, blkbits
) >> blkbits
)
3032 * credits to insert 1 extent into extent tree
3034 credits
= ext4_chunk_trans_blocks(inode
, max_blocks
);
3035 mutex_lock(&inode
->i_mutex
);
3037 while (ret
>= 0 && ret
< max_blocks
) {
3038 block
= block
+ ret
;
3039 max_blocks
= max_blocks
- ret
;
3040 handle
= ext4_journal_start(inode
, credits
);
3041 if (IS_ERR(handle
)) {
3042 ret
= PTR_ERR(handle
);
3045 ret
= ext4_get_blocks_wrap(handle
, inode
, block
,
3046 max_blocks
, &map_bh
,
3047 EXT4_CREATE_UNINITIALIZED_EXT
, 0, 0);
3051 printk(KERN_ERR
"%s: ext4_ext_get_blocks "
3052 "returned error inode#%lu, block=%u, "
3053 "max_blocks=%u", __func__
,
3054 inode
->i_ino
, block
, max_blocks
);
3056 ext4_mark_inode_dirty(handle
, inode
);
3057 ret2
= ext4_journal_stop(handle
);
3060 if ((block
+ ret
) >= (EXT4_BLOCK_ALIGN(offset
+ len
,
3061 blkbits
) >> blkbits
))
3062 new_size
= offset
+ len
;
3064 new_size
= (block
+ ret
) << blkbits
;
3066 ext4_falloc_update_inode(inode
, mode
, new_size
,
3067 buffer_new(&map_bh
));
3068 ext4_mark_inode_dirty(handle
, inode
);
3069 ret2
= ext4_journal_stop(handle
);
3073 if (ret
== -ENOSPC
&&
3074 ext4_should_retry_alloc(inode
->i_sb
, &retries
)) {
3078 mutex_unlock(&inode
->i_mutex
);
3079 return ret
> 0 ? ret2
: ret
;
3083 * Callback function called for each extent to gather FIEMAP information.
3085 static int ext4_ext_fiemap_cb(struct inode
*inode
, struct ext4_ext_path
*path
,
3086 struct ext4_ext_cache
*newex
, struct ext4_extent
*ex
,
3089 struct fiemap_extent_info
*fieinfo
= data
;
3090 unsigned long blksize_bits
= inode
->i_sb
->s_blocksize_bits
;
3097 logical
= (__u64
)newex
->ec_block
<< blksize_bits
;
3099 if (newex
->ec_type
== EXT4_EXT_CACHE_GAP
) {
3102 struct buffer_head
*bh
= NULL
;
3104 offset
= logical
>> PAGE_SHIFT
;
3105 page
= find_get_page(inode
->i_mapping
, offset
);
3106 if (!page
|| !page_has_buffers(page
))
3107 return EXT_CONTINUE
;
3109 bh
= page_buffers(page
);
3112 return EXT_CONTINUE
;
3114 if (buffer_delay(bh
)) {
3115 flags
|= FIEMAP_EXTENT_DELALLOC
;
3116 page_cache_release(page
);
3118 page_cache_release(page
);
3119 return EXT_CONTINUE
;
3123 physical
= (__u64
)newex
->ec_start
<< blksize_bits
;
3124 length
= (__u64
)newex
->ec_len
<< blksize_bits
;
3126 if (ex
&& ext4_ext_is_uninitialized(ex
))
3127 flags
|= FIEMAP_EXTENT_UNWRITTEN
;
3130 * If this extent reaches EXT_MAX_BLOCK, it must be last.
3132 * Or if ext4_ext_next_allocated_block is EXT_MAX_BLOCK,
3133 * this also indicates no more allocated blocks.
3135 * XXX this might miss a single-block extent at EXT_MAX_BLOCK
3137 if (logical
+ length
- 1 == EXT_MAX_BLOCK
||
3138 ext4_ext_next_allocated_block(path
) == EXT_MAX_BLOCK
)
3139 flags
|= FIEMAP_EXTENT_LAST
;
3141 error
= fiemap_fill_next_extent(fieinfo
, logical
, physical
,
3148 return EXT_CONTINUE
;
3151 /* fiemap flags we can handle specified here */
3152 #define EXT4_FIEMAP_FLAGS (FIEMAP_FLAG_SYNC|FIEMAP_FLAG_XATTR)
3154 static int ext4_xattr_fiemap(struct inode
*inode
,
3155 struct fiemap_extent_info
*fieinfo
)
3159 __u32 flags
= FIEMAP_EXTENT_LAST
;
3160 int blockbits
= inode
->i_sb
->s_blocksize_bits
;
3164 if (EXT4_I(inode
)->i_state
& EXT4_STATE_XATTR
) {
3165 struct ext4_iloc iloc
;
3166 int offset
; /* offset of xattr in inode */
3168 error
= ext4_get_inode_loc(inode
, &iloc
);
3171 physical
= iloc
.bh
->b_blocknr
<< blockbits
;
3172 offset
= EXT4_GOOD_OLD_INODE_SIZE
+
3173 EXT4_I(inode
)->i_extra_isize
;
3175 length
= EXT4_SB(inode
->i_sb
)->s_inode_size
- offset
;
3176 flags
|= FIEMAP_EXTENT_DATA_INLINE
;
3177 } else { /* external block */
3178 physical
= EXT4_I(inode
)->i_file_acl
<< blockbits
;
3179 length
= inode
->i_sb
->s_blocksize
;
3183 error
= fiemap_fill_next_extent(fieinfo
, 0, physical
,
3185 return (error
< 0 ? error
: 0);
3188 int ext4_fiemap(struct inode
*inode
, struct fiemap_extent_info
*fieinfo
,
3189 __u64 start
, __u64 len
)
3191 ext4_lblk_t start_blk
;
3192 ext4_lblk_t len_blks
;
3195 /* fallback to generic here if not in extents fmt */
3196 if (!(EXT4_I(inode
)->i_flags
& EXT4_EXTENTS_FL
))
3197 return generic_block_fiemap(inode
, fieinfo
, start
, len
,
3200 if (fiemap_check_flags(fieinfo
, EXT4_FIEMAP_FLAGS
))
3203 if (fieinfo
->fi_flags
& FIEMAP_FLAG_XATTR
) {
3204 error
= ext4_xattr_fiemap(inode
, fieinfo
);
3206 start_blk
= start
>> inode
->i_sb
->s_blocksize_bits
;
3207 len_blks
= len
>> inode
->i_sb
->s_blocksize_bits
;
3210 * Walk the extent tree gathering extent information.
3211 * ext4_ext_fiemap_cb will push extents back to user.
3213 down_write(&EXT4_I(inode
)->i_data_sem
);
3214 error
= ext4_ext_walk_space(inode
, start_blk
, len_blks
,
3215 ext4_ext_fiemap_cb
, fieinfo
);
3216 up_write(&EXT4_I(inode
)->i_data_sem
);