2 * Copyright (c) 2003-2006, Cluster File Systems, Inc, info@clusterfs.com
3 * Written by Alex Tomas <alex@clusterfs.com>
5 * Architecture independence:
6 * Copyright (c) 2005, Bull S.A.
7 * Written by Pierre Peiffer <pierre.peiffer@bull.net>
9 * This program is free software; you can redistribute it and/or modify
10 * it under the terms of the GNU General Public License version 2 as
11 * published by the Free Software Foundation.
13 * This program is distributed in the hope that it will be useful,
14 * but WITHOUT ANY WARRANTY; without even the implied warranty of
15 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
16 * GNU General Public License for more details.
18 * You should have received a copy of the GNU General Public Licens
19 * along with this program; if not, write to the Free Software
20 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-
24 * Extents support for EXT4
27 * - ext4*_error() should be used in some situations
28 * - analyze all BUG()/BUG_ON(), use -EIO where appropriate
29 * - smart tree reduction
33 #include <linux/time.h>
34 #include <linux/jbd2.h>
35 #include <linux/highuid.h>
36 #include <linux/pagemap.h>
37 #include <linux/quotaops.h>
38 #include <linux/string.h>
39 #include <linux/slab.h>
40 #include <linux/falloc.h>
41 #include <asm/uaccess.h>
42 #include <linux/fiemap.h>
43 #include "ext4_jbd2.h"
45 #include <trace/events/ext4.h>
48 * used by extent splitting.
50 #define EXT4_EXT_MAY_ZEROOUT 0x1 /* safe to zeroout if split fails \
52 #define EXT4_EXT_MARK_UNINIT1 0x2 /* mark first half uninitialized */
53 #define EXT4_EXT_MARK_UNINIT2 0x4 /* mark second half uninitialized */
55 #define EXT4_EXT_DATA_VALID1 0x8 /* first half contains valid data */
56 #define EXT4_EXT_DATA_VALID2 0x10 /* second half contains valid data */
58 static __le32
ext4_extent_block_csum(struct inode
*inode
,
59 struct ext4_extent_header
*eh
)
61 struct ext4_inode_info
*ei
= EXT4_I(inode
);
62 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
65 csum
= ext4_chksum(sbi
, ei
->i_csum_seed
, (__u8
*)eh
,
66 EXT4_EXTENT_TAIL_OFFSET(eh
));
67 return cpu_to_le32(csum
);
70 static int ext4_extent_block_csum_verify(struct inode
*inode
,
71 struct ext4_extent_header
*eh
)
73 struct ext4_extent_tail
*et
;
75 if (!EXT4_HAS_RO_COMPAT_FEATURE(inode
->i_sb
,
76 EXT4_FEATURE_RO_COMPAT_METADATA_CSUM
))
79 et
= find_ext4_extent_tail(eh
);
80 if (et
->et_checksum
!= ext4_extent_block_csum(inode
, eh
))
85 static void ext4_extent_block_csum_set(struct inode
*inode
,
86 struct ext4_extent_header
*eh
)
88 struct ext4_extent_tail
*et
;
90 if (!EXT4_HAS_RO_COMPAT_FEATURE(inode
->i_sb
,
91 EXT4_FEATURE_RO_COMPAT_METADATA_CSUM
))
94 et
= find_ext4_extent_tail(eh
);
95 et
->et_checksum
= ext4_extent_block_csum(inode
, eh
);
98 static int ext4_split_extent(handle_t
*handle
,
100 struct ext4_ext_path
*path
,
101 struct ext4_map_blocks
*map
,
105 static int ext4_split_extent_at(handle_t
*handle
,
107 struct ext4_ext_path
*path
,
112 static int ext4_ext_truncate_extend_restart(handle_t
*handle
,
118 if (!ext4_handle_valid(handle
))
120 if (handle
->h_buffer_credits
> needed
)
122 err
= ext4_journal_extend(handle
, needed
);
125 err
= ext4_truncate_restart_trans(handle
, inode
, needed
);
137 static int ext4_ext_get_access(handle_t
*handle
, struct inode
*inode
,
138 struct ext4_ext_path
*path
)
141 /* path points to block */
142 return ext4_journal_get_write_access(handle
, path
->p_bh
);
144 /* path points to leaf/index in inode body */
145 /* we use in-core data, no need to protect them */
155 #define ext4_ext_dirty(handle, inode, path) \
156 __ext4_ext_dirty(__func__, __LINE__, (handle), (inode), (path))
157 static int __ext4_ext_dirty(const char *where
, unsigned int line
,
158 handle_t
*handle
, struct inode
*inode
,
159 struct ext4_ext_path
*path
)
163 ext4_extent_block_csum_set(inode
, ext_block_hdr(path
->p_bh
));
164 /* path points to block */
165 err
= __ext4_handle_dirty_metadata(where
, line
, handle
,
168 /* path points to leaf/index in inode body */
169 err
= ext4_mark_inode_dirty(handle
, inode
);
174 static ext4_fsblk_t
ext4_ext_find_goal(struct inode
*inode
,
175 struct ext4_ext_path
*path
,
179 int depth
= path
->p_depth
;
180 struct ext4_extent
*ex
;
183 * Try to predict block placement assuming that we are
184 * filling in a file which will eventually be
185 * non-sparse --- i.e., in the case of libbfd writing
186 * an ELF object sections out-of-order but in a way
187 * the eventually results in a contiguous object or
188 * executable file, or some database extending a table
189 * space file. However, this is actually somewhat
190 * non-ideal if we are writing a sparse file such as
191 * qemu or KVM writing a raw image file that is going
192 * to stay fairly sparse, since it will end up
193 * fragmenting the file system's free space. Maybe we
194 * should have some hueristics or some way to allow
195 * userspace to pass a hint to file system,
196 * especially if the latter case turns out to be
199 ex
= path
[depth
].p_ext
;
201 ext4_fsblk_t ext_pblk
= ext4_ext_pblock(ex
);
202 ext4_lblk_t ext_block
= le32_to_cpu(ex
->ee_block
);
204 if (block
> ext_block
)
205 return ext_pblk
+ (block
- ext_block
);
207 return ext_pblk
- (ext_block
- block
);
210 /* it looks like index is empty;
211 * try to find starting block from index itself */
212 if (path
[depth
].p_bh
)
213 return path
[depth
].p_bh
->b_blocknr
;
216 /* OK. use inode's group */
217 return ext4_inode_to_goal_block(inode
);
221 * Allocation for a meta data block
224 ext4_ext_new_meta_block(handle_t
*handle
, struct inode
*inode
,
225 struct ext4_ext_path
*path
,
226 struct ext4_extent
*ex
, int *err
, unsigned int flags
)
228 ext4_fsblk_t goal
, newblock
;
230 goal
= ext4_ext_find_goal(inode
, path
, le32_to_cpu(ex
->ee_block
));
231 newblock
= ext4_new_meta_blocks(handle
, inode
, goal
, flags
,
236 static inline int ext4_ext_space_block(struct inode
*inode
, int check
)
240 size
= (inode
->i_sb
->s_blocksize
- sizeof(struct ext4_extent_header
))
241 / sizeof(struct ext4_extent
);
242 #ifdef AGGRESSIVE_TEST
243 if (!check
&& size
> 6)
249 static inline int ext4_ext_space_block_idx(struct inode
*inode
, int check
)
253 size
= (inode
->i_sb
->s_blocksize
- sizeof(struct ext4_extent_header
))
254 / sizeof(struct ext4_extent_idx
);
255 #ifdef AGGRESSIVE_TEST
256 if (!check
&& size
> 5)
262 static inline int ext4_ext_space_root(struct inode
*inode
, int check
)
266 size
= sizeof(EXT4_I(inode
)->i_data
);
267 size
-= sizeof(struct ext4_extent_header
);
268 size
/= sizeof(struct ext4_extent
);
269 #ifdef AGGRESSIVE_TEST
270 if (!check
&& size
> 3)
276 static inline int ext4_ext_space_root_idx(struct inode
*inode
, int check
)
280 size
= sizeof(EXT4_I(inode
)->i_data
);
281 size
-= sizeof(struct ext4_extent_header
);
282 size
/= sizeof(struct ext4_extent_idx
);
283 #ifdef AGGRESSIVE_TEST
284 if (!check
&& size
> 4)
291 * Calculate the number of metadata blocks needed
292 * to allocate @blocks
293 * Worse case is one block per extent
295 int ext4_ext_calc_metadata_amount(struct inode
*inode
, ext4_lblk_t lblock
)
297 struct ext4_inode_info
*ei
= EXT4_I(inode
);
300 idxs
= ((inode
->i_sb
->s_blocksize
- sizeof(struct ext4_extent_header
))
301 / sizeof(struct ext4_extent_idx
));
304 * If the new delayed allocation block is contiguous with the
305 * previous da block, it can share index blocks with the
306 * previous block, so we only need to allocate a new index
307 * block every idxs leaf blocks. At ldxs**2 blocks, we need
308 * an additional index block, and at ldxs**3 blocks, yet
309 * another index blocks.
311 if (ei
->i_da_metadata_calc_len
&&
312 ei
->i_da_metadata_calc_last_lblock
+1 == lblock
) {
315 if ((ei
->i_da_metadata_calc_len
% idxs
) == 0)
317 if ((ei
->i_da_metadata_calc_len
% (idxs
*idxs
)) == 0)
319 if ((ei
->i_da_metadata_calc_len
% (idxs
*idxs
*idxs
)) == 0) {
321 ei
->i_da_metadata_calc_len
= 0;
323 ei
->i_da_metadata_calc_len
++;
324 ei
->i_da_metadata_calc_last_lblock
++;
329 * In the worst case we need a new set of index blocks at
330 * every level of the inode's extent tree.
332 ei
->i_da_metadata_calc_len
= 1;
333 ei
->i_da_metadata_calc_last_lblock
= lblock
;
334 return ext_depth(inode
) + 1;
338 ext4_ext_max_entries(struct inode
*inode
, int depth
)
342 if (depth
== ext_depth(inode
)) {
344 max
= ext4_ext_space_root(inode
, 1);
346 max
= ext4_ext_space_root_idx(inode
, 1);
349 max
= ext4_ext_space_block(inode
, 1);
351 max
= ext4_ext_space_block_idx(inode
, 1);
357 static int ext4_valid_extent(struct inode
*inode
, struct ext4_extent
*ext
)
359 ext4_fsblk_t block
= ext4_ext_pblock(ext
);
360 int len
= ext4_ext_get_actual_len(ext
);
364 return ext4_data_block_valid(EXT4_SB(inode
->i_sb
), block
, len
);
367 static int ext4_valid_extent_idx(struct inode
*inode
,
368 struct ext4_extent_idx
*ext_idx
)
370 ext4_fsblk_t block
= ext4_idx_pblock(ext_idx
);
372 return ext4_data_block_valid(EXT4_SB(inode
->i_sb
), block
, 1);
375 static int ext4_valid_extent_entries(struct inode
*inode
,
376 struct ext4_extent_header
*eh
,
379 unsigned short entries
;
380 if (eh
->eh_entries
== 0)
383 entries
= le16_to_cpu(eh
->eh_entries
);
387 struct ext4_extent
*ext
= EXT_FIRST_EXTENT(eh
);
389 if (!ext4_valid_extent(inode
, ext
))
395 struct ext4_extent_idx
*ext_idx
= EXT_FIRST_INDEX(eh
);
397 if (!ext4_valid_extent_idx(inode
, ext_idx
))
406 static int __ext4_ext_check(const char *function
, unsigned int line
,
407 struct inode
*inode
, struct ext4_extent_header
*eh
,
410 const char *error_msg
;
413 if (unlikely(eh
->eh_magic
!= EXT4_EXT_MAGIC
)) {
414 error_msg
= "invalid magic";
417 if (unlikely(le16_to_cpu(eh
->eh_depth
) != depth
)) {
418 error_msg
= "unexpected eh_depth";
421 if (unlikely(eh
->eh_max
== 0)) {
422 error_msg
= "invalid eh_max";
425 max
= ext4_ext_max_entries(inode
, depth
);
426 if (unlikely(le16_to_cpu(eh
->eh_max
) > max
)) {
427 error_msg
= "too large eh_max";
430 if (unlikely(le16_to_cpu(eh
->eh_entries
) > le16_to_cpu(eh
->eh_max
))) {
431 error_msg
= "invalid eh_entries";
434 if (!ext4_valid_extent_entries(inode
, eh
, depth
)) {
435 error_msg
= "invalid extent entries";
438 /* Verify checksum on non-root extent tree nodes */
439 if (ext_depth(inode
) != depth
&&
440 !ext4_extent_block_csum_verify(inode
, eh
)) {
441 error_msg
= "extent tree corrupted";
447 ext4_error_inode(inode
, function
, line
, 0,
448 "bad header/extent: %s - magic %x, "
449 "entries %u, max %u(%u), depth %u(%u)",
450 error_msg
, le16_to_cpu(eh
->eh_magic
),
451 le16_to_cpu(eh
->eh_entries
), le16_to_cpu(eh
->eh_max
),
452 max
, le16_to_cpu(eh
->eh_depth
), depth
);
457 #define ext4_ext_check(inode, eh, depth) \
458 __ext4_ext_check(__func__, __LINE__, inode, eh, depth)
460 int ext4_ext_check_inode(struct inode
*inode
)
462 return ext4_ext_check(inode
, ext_inode_hdr(inode
), ext_depth(inode
));
465 static int __ext4_ext_check_block(const char *function
, unsigned int line
,
467 struct ext4_extent_header
*eh
,
469 struct buffer_head
*bh
)
473 if (buffer_verified(bh
))
475 ret
= ext4_ext_check(inode
, eh
, depth
);
478 set_buffer_verified(bh
);
482 #define ext4_ext_check_block(inode, eh, depth, bh) \
483 __ext4_ext_check_block(__func__, __LINE__, inode, eh, depth, bh)
486 static void ext4_ext_show_path(struct inode
*inode
, struct ext4_ext_path
*path
)
488 int k
, l
= path
->p_depth
;
491 for (k
= 0; k
<= l
; k
++, path
++) {
493 ext_debug(" %d->%llu", le32_to_cpu(path
->p_idx
->ei_block
),
494 ext4_idx_pblock(path
->p_idx
));
495 } else if (path
->p_ext
) {
496 ext_debug(" %d:[%d]%d:%llu ",
497 le32_to_cpu(path
->p_ext
->ee_block
),
498 ext4_ext_is_uninitialized(path
->p_ext
),
499 ext4_ext_get_actual_len(path
->p_ext
),
500 ext4_ext_pblock(path
->p_ext
));
507 static void ext4_ext_show_leaf(struct inode
*inode
, struct ext4_ext_path
*path
)
509 int depth
= ext_depth(inode
);
510 struct ext4_extent_header
*eh
;
511 struct ext4_extent
*ex
;
517 eh
= path
[depth
].p_hdr
;
518 ex
= EXT_FIRST_EXTENT(eh
);
520 ext_debug("Displaying leaf extents for inode %lu\n", inode
->i_ino
);
522 for (i
= 0; i
< le16_to_cpu(eh
->eh_entries
); i
++, ex
++) {
523 ext_debug("%d:[%d]%d:%llu ", le32_to_cpu(ex
->ee_block
),
524 ext4_ext_is_uninitialized(ex
),
525 ext4_ext_get_actual_len(ex
), ext4_ext_pblock(ex
));
530 static void ext4_ext_show_move(struct inode
*inode
, struct ext4_ext_path
*path
,
531 ext4_fsblk_t newblock
, int level
)
533 int depth
= ext_depth(inode
);
534 struct ext4_extent
*ex
;
536 if (depth
!= level
) {
537 struct ext4_extent_idx
*idx
;
538 idx
= path
[level
].p_idx
;
539 while (idx
<= EXT_MAX_INDEX(path
[level
].p_hdr
)) {
540 ext_debug("%d: move %d:%llu in new index %llu\n", level
,
541 le32_to_cpu(idx
->ei_block
),
542 ext4_idx_pblock(idx
),
550 ex
= path
[depth
].p_ext
;
551 while (ex
<= EXT_MAX_EXTENT(path
[depth
].p_hdr
)) {
552 ext_debug("move %d:%llu:[%d]%d in new leaf %llu\n",
553 le32_to_cpu(ex
->ee_block
),
555 ext4_ext_is_uninitialized(ex
),
556 ext4_ext_get_actual_len(ex
),
563 #define ext4_ext_show_path(inode, path)
564 #define ext4_ext_show_leaf(inode, path)
565 #define ext4_ext_show_move(inode, path, newblock, level)
568 void ext4_ext_drop_refs(struct ext4_ext_path
*path
)
570 int depth
= path
->p_depth
;
573 for (i
= 0; i
<= depth
; i
++, path
++)
581 * ext4_ext_binsearch_idx:
582 * binary search for the closest index of the given block
583 * the header must be checked before calling this
586 ext4_ext_binsearch_idx(struct inode
*inode
,
587 struct ext4_ext_path
*path
, ext4_lblk_t block
)
589 struct ext4_extent_header
*eh
= path
->p_hdr
;
590 struct ext4_extent_idx
*r
, *l
, *m
;
593 ext_debug("binsearch for %u(idx): ", block
);
595 l
= EXT_FIRST_INDEX(eh
) + 1;
596 r
= EXT_LAST_INDEX(eh
);
599 if (block
< le32_to_cpu(m
->ei_block
))
603 ext_debug("%p(%u):%p(%u):%p(%u) ", l
, le32_to_cpu(l
->ei_block
),
604 m
, le32_to_cpu(m
->ei_block
),
605 r
, le32_to_cpu(r
->ei_block
));
609 ext_debug(" -> %u->%lld ", le32_to_cpu(path
->p_idx
->ei_block
),
610 ext4_idx_pblock(path
->p_idx
));
612 #ifdef CHECK_BINSEARCH
614 struct ext4_extent_idx
*chix
, *ix
;
617 chix
= ix
= EXT_FIRST_INDEX(eh
);
618 for (k
= 0; k
< le16_to_cpu(eh
->eh_entries
); k
++, ix
++) {
620 le32_to_cpu(ix
->ei_block
) <= le32_to_cpu(ix
[-1].ei_block
)) {
621 printk(KERN_DEBUG
"k=%d, ix=0x%p, "
623 ix
, EXT_FIRST_INDEX(eh
));
624 printk(KERN_DEBUG
"%u <= %u\n",
625 le32_to_cpu(ix
->ei_block
),
626 le32_to_cpu(ix
[-1].ei_block
));
628 BUG_ON(k
&& le32_to_cpu(ix
->ei_block
)
629 <= le32_to_cpu(ix
[-1].ei_block
));
630 if (block
< le32_to_cpu(ix
->ei_block
))
634 BUG_ON(chix
!= path
->p_idx
);
641 * ext4_ext_binsearch:
642 * binary search for closest extent of the given block
643 * the header must be checked before calling this
646 ext4_ext_binsearch(struct inode
*inode
,
647 struct ext4_ext_path
*path
, ext4_lblk_t block
)
649 struct ext4_extent_header
*eh
= path
->p_hdr
;
650 struct ext4_extent
*r
, *l
, *m
;
652 if (eh
->eh_entries
== 0) {
654 * this leaf is empty:
655 * we get such a leaf in split/add case
660 ext_debug("binsearch for %u: ", block
);
662 l
= EXT_FIRST_EXTENT(eh
) + 1;
663 r
= EXT_LAST_EXTENT(eh
);
667 if (block
< le32_to_cpu(m
->ee_block
))
671 ext_debug("%p(%u):%p(%u):%p(%u) ", l
, le32_to_cpu(l
->ee_block
),
672 m
, le32_to_cpu(m
->ee_block
),
673 r
, le32_to_cpu(r
->ee_block
));
677 ext_debug(" -> %d:%llu:[%d]%d ",
678 le32_to_cpu(path
->p_ext
->ee_block
),
679 ext4_ext_pblock(path
->p_ext
),
680 ext4_ext_is_uninitialized(path
->p_ext
),
681 ext4_ext_get_actual_len(path
->p_ext
));
683 #ifdef CHECK_BINSEARCH
685 struct ext4_extent
*chex
, *ex
;
688 chex
= ex
= EXT_FIRST_EXTENT(eh
);
689 for (k
= 0; k
< le16_to_cpu(eh
->eh_entries
); k
++, ex
++) {
690 BUG_ON(k
&& le32_to_cpu(ex
->ee_block
)
691 <= le32_to_cpu(ex
[-1].ee_block
));
692 if (block
< le32_to_cpu(ex
->ee_block
))
696 BUG_ON(chex
!= path
->p_ext
);
702 int ext4_ext_tree_init(handle_t
*handle
, struct inode
*inode
)
704 struct ext4_extent_header
*eh
;
706 eh
= ext_inode_hdr(inode
);
709 eh
->eh_magic
= EXT4_EXT_MAGIC
;
710 eh
->eh_max
= cpu_to_le16(ext4_ext_space_root(inode
, 0));
711 ext4_mark_inode_dirty(handle
, inode
);
712 ext4_ext_invalidate_cache(inode
);
716 struct ext4_ext_path
*
717 ext4_ext_find_extent(struct inode
*inode
, ext4_lblk_t block
,
718 struct ext4_ext_path
*path
)
720 struct ext4_extent_header
*eh
;
721 struct buffer_head
*bh
;
722 short int depth
, i
, ppos
= 0, alloc
= 0;
724 eh
= ext_inode_hdr(inode
);
725 depth
= ext_depth(inode
);
727 /* account possible depth increase */
729 path
= kzalloc(sizeof(struct ext4_ext_path
) * (depth
+ 2),
732 return ERR_PTR(-ENOMEM
);
739 /* walk through the tree */
741 ext_debug("depth %d: num %d, max %d\n",
742 ppos
, le16_to_cpu(eh
->eh_entries
), le16_to_cpu(eh
->eh_max
));
744 ext4_ext_binsearch_idx(inode
, path
+ ppos
, block
);
745 path
[ppos
].p_block
= ext4_idx_pblock(path
[ppos
].p_idx
);
746 path
[ppos
].p_depth
= i
;
747 path
[ppos
].p_ext
= NULL
;
749 bh
= sb_getblk(inode
->i_sb
, path
[ppos
].p_block
);
752 if (!bh_uptodate_or_lock(bh
)) {
753 trace_ext4_ext_load_extent(inode
, block
,
755 if (bh_submit_read(bh
) < 0) {
760 eh
= ext_block_hdr(bh
);
762 if (unlikely(ppos
> depth
)) {
764 EXT4_ERROR_INODE(inode
,
765 "ppos %d > depth %d", ppos
, depth
);
768 path
[ppos
].p_bh
= bh
;
769 path
[ppos
].p_hdr
= eh
;
772 if (ext4_ext_check_block(inode
, eh
, i
, bh
))
776 path
[ppos
].p_depth
= i
;
777 path
[ppos
].p_ext
= NULL
;
778 path
[ppos
].p_idx
= NULL
;
781 ext4_ext_binsearch(inode
, path
+ ppos
, block
);
782 /* if not an empty leaf */
783 if (path
[ppos
].p_ext
)
784 path
[ppos
].p_block
= ext4_ext_pblock(path
[ppos
].p_ext
);
786 ext4_ext_show_path(inode
, path
);
791 ext4_ext_drop_refs(path
);
794 return ERR_PTR(-EIO
);
798 * ext4_ext_insert_index:
799 * insert new index [@logical;@ptr] into the block at @curp;
800 * check where to insert: before @curp or after @curp
802 static int ext4_ext_insert_index(handle_t
*handle
, struct inode
*inode
,
803 struct ext4_ext_path
*curp
,
804 int logical
, ext4_fsblk_t ptr
)
806 struct ext4_extent_idx
*ix
;
809 err
= ext4_ext_get_access(handle
, inode
, curp
);
813 if (unlikely(logical
== le32_to_cpu(curp
->p_idx
->ei_block
))) {
814 EXT4_ERROR_INODE(inode
,
815 "logical %d == ei_block %d!",
816 logical
, le32_to_cpu(curp
->p_idx
->ei_block
));
820 if (unlikely(le16_to_cpu(curp
->p_hdr
->eh_entries
)
821 >= le16_to_cpu(curp
->p_hdr
->eh_max
))) {
822 EXT4_ERROR_INODE(inode
,
823 "eh_entries %d >= eh_max %d!",
824 le16_to_cpu(curp
->p_hdr
->eh_entries
),
825 le16_to_cpu(curp
->p_hdr
->eh_max
));
829 if (logical
> le32_to_cpu(curp
->p_idx
->ei_block
)) {
831 ext_debug("insert new index %d after: %llu\n", logical
, ptr
);
832 ix
= curp
->p_idx
+ 1;
835 ext_debug("insert new index %d before: %llu\n", logical
, ptr
);
839 len
= EXT_LAST_INDEX(curp
->p_hdr
) - ix
+ 1;
842 ext_debug("insert new index %d: "
843 "move %d indices from 0x%p to 0x%p\n",
844 logical
, len
, ix
, ix
+ 1);
845 memmove(ix
+ 1, ix
, len
* sizeof(struct ext4_extent_idx
));
848 if (unlikely(ix
> EXT_MAX_INDEX(curp
->p_hdr
))) {
849 EXT4_ERROR_INODE(inode
, "ix > EXT_MAX_INDEX!");
853 ix
->ei_block
= cpu_to_le32(logical
);
854 ext4_idx_store_pblock(ix
, ptr
);
855 le16_add_cpu(&curp
->p_hdr
->eh_entries
, 1);
857 if (unlikely(ix
> EXT_LAST_INDEX(curp
->p_hdr
))) {
858 EXT4_ERROR_INODE(inode
, "ix > EXT_LAST_INDEX!");
862 err
= ext4_ext_dirty(handle
, inode
, curp
);
863 ext4_std_error(inode
->i_sb
, err
);
870 * inserts new subtree into the path, using free index entry
872 * - allocates all needed blocks (new leaf and all intermediate index blocks)
873 * - makes decision where to split
874 * - moves remaining extents and index entries (right to the split point)
875 * into the newly allocated blocks
876 * - initializes subtree
878 static int ext4_ext_split(handle_t
*handle
, struct inode
*inode
,
880 struct ext4_ext_path
*path
,
881 struct ext4_extent
*newext
, int at
)
883 struct buffer_head
*bh
= NULL
;
884 int depth
= ext_depth(inode
);
885 struct ext4_extent_header
*neh
;
886 struct ext4_extent_idx
*fidx
;
888 ext4_fsblk_t newblock
, oldblock
;
890 ext4_fsblk_t
*ablocks
= NULL
; /* array of allocated blocks */
893 /* make decision: where to split? */
894 /* FIXME: now decision is simplest: at current extent */
896 /* if current leaf will be split, then we should use
897 * border from split point */
898 if (unlikely(path
[depth
].p_ext
> EXT_MAX_EXTENT(path
[depth
].p_hdr
))) {
899 EXT4_ERROR_INODE(inode
, "p_ext > EXT_MAX_EXTENT!");
902 if (path
[depth
].p_ext
!= EXT_MAX_EXTENT(path
[depth
].p_hdr
)) {
903 border
= path
[depth
].p_ext
[1].ee_block
;
904 ext_debug("leaf will be split."
905 " next leaf starts at %d\n",
906 le32_to_cpu(border
));
908 border
= newext
->ee_block
;
909 ext_debug("leaf will be added."
910 " next leaf starts at %d\n",
911 le32_to_cpu(border
));
915 * If error occurs, then we break processing
916 * and mark filesystem read-only. index won't
917 * be inserted and tree will be in consistent
918 * state. Next mount will repair buffers too.
922 * Get array to track all allocated blocks.
923 * We need this to handle errors and free blocks
926 ablocks
= kzalloc(sizeof(ext4_fsblk_t
) * depth
, GFP_NOFS
);
930 /* allocate all needed blocks */
931 ext_debug("allocate %d blocks for indexes/leaf\n", depth
- at
);
932 for (a
= 0; a
< depth
- at
; a
++) {
933 newblock
= ext4_ext_new_meta_block(handle
, inode
, path
,
934 newext
, &err
, flags
);
937 ablocks
[a
] = newblock
;
940 /* initialize new leaf */
941 newblock
= ablocks
[--a
];
942 if (unlikely(newblock
== 0)) {
943 EXT4_ERROR_INODE(inode
, "newblock == 0!");
947 bh
= sb_getblk(inode
->i_sb
, newblock
);
954 err
= ext4_journal_get_create_access(handle
, bh
);
958 neh
= ext_block_hdr(bh
);
960 neh
->eh_max
= cpu_to_le16(ext4_ext_space_block(inode
, 0));
961 neh
->eh_magic
= EXT4_EXT_MAGIC
;
964 /* move remainder of path[depth] to the new leaf */
965 if (unlikely(path
[depth
].p_hdr
->eh_entries
!=
966 path
[depth
].p_hdr
->eh_max
)) {
967 EXT4_ERROR_INODE(inode
, "eh_entries %d != eh_max %d!",
968 path
[depth
].p_hdr
->eh_entries
,
969 path
[depth
].p_hdr
->eh_max
);
973 /* start copy from next extent */
974 m
= EXT_MAX_EXTENT(path
[depth
].p_hdr
) - path
[depth
].p_ext
++;
975 ext4_ext_show_move(inode
, path
, newblock
, depth
);
977 struct ext4_extent
*ex
;
978 ex
= EXT_FIRST_EXTENT(neh
);
979 memmove(ex
, path
[depth
].p_ext
, sizeof(struct ext4_extent
) * m
);
980 le16_add_cpu(&neh
->eh_entries
, m
);
983 ext4_extent_block_csum_set(inode
, neh
);
984 set_buffer_uptodate(bh
);
987 err
= ext4_handle_dirty_metadata(handle
, inode
, bh
);
993 /* correct old leaf */
995 err
= ext4_ext_get_access(handle
, inode
, path
+ depth
);
998 le16_add_cpu(&path
[depth
].p_hdr
->eh_entries
, -m
);
999 err
= ext4_ext_dirty(handle
, inode
, path
+ depth
);
1005 /* create intermediate indexes */
1007 if (unlikely(k
< 0)) {
1008 EXT4_ERROR_INODE(inode
, "k %d < 0!", k
);
1013 ext_debug("create %d intermediate indices\n", k
);
1014 /* insert new index into current index block */
1015 /* current depth stored in i var */
1018 oldblock
= newblock
;
1019 newblock
= ablocks
[--a
];
1020 bh
= sb_getblk(inode
->i_sb
, newblock
);
1027 err
= ext4_journal_get_create_access(handle
, bh
);
1031 neh
= ext_block_hdr(bh
);
1032 neh
->eh_entries
= cpu_to_le16(1);
1033 neh
->eh_magic
= EXT4_EXT_MAGIC
;
1034 neh
->eh_max
= cpu_to_le16(ext4_ext_space_block_idx(inode
, 0));
1035 neh
->eh_depth
= cpu_to_le16(depth
- i
);
1036 fidx
= EXT_FIRST_INDEX(neh
);
1037 fidx
->ei_block
= border
;
1038 ext4_idx_store_pblock(fidx
, oldblock
);
1040 ext_debug("int.index at %d (block %llu): %u -> %llu\n",
1041 i
, newblock
, le32_to_cpu(border
), oldblock
);
1043 /* move remainder of path[i] to the new index block */
1044 if (unlikely(EXT_MAX_INDEX(path
[i
].p_hdr
) !=
1045 EXT_LAST_INDEX(path
[i
].p_hdr
))) {
1046 EXT4_ERROR_INODE(inode
,
1047 "EXT_MAX_INDEX != EXT_LAST_INDEX ee_block %d!",
1048 le32_to_cpu(path
[i
].p_ext
->ee_block
));
1052 /* start copy indexes */
1053 m
= EXT_MAX_INDEX(path
[i
].p_hdr
) - path
[i
].p_idx
++;
1054 ext_debug("cur 0x%p, last 0x%p\n", path
[i
].p_idx
,
1055 EXT_MAX_INDEX(path
[i
].p_hdr
));
1056 ext4_ext_show_move(inode
, path
, newblock
, i
);
1058 memmove(++fidx
, path
[i
].p_idx
,
1059 sizeof(struct ext4_extent_idx
) * m
);
1060 le16_add_cpu(&neh
->eh_entries
, m
);
1062 ext4_extent_block_csum_set(inode
, neh
);
1063 set_buffer_uptodate(bh
);
1066 err
= ext4_handle_dirty_metadata(handle
, inode
, bh
);
1072 /* correct old index */
1074 err
= ext4_ext_get_access(handle
, inode
, path
+ i
);
1077 le16_add_cpu(&path
[i
].p_hdr
->eh_entries
, -m
);
1078 err
= ext4_ext_dirty(handle
, inode
, path
+ i
);
1086 /* insert new index */
1087 err
= ext4_ext_insert_index(handle
, inode
, path
+ at
,
1088 le32_to_cpu(border
), newblock
);
1092 if (buffer_locked(bh
))
1098 /* free all allocated blocks in error case */
1099 for (i
= 0; i
< depth
; i
++) {
1102 ext4_free_blocks(handle
, inode
, NULL
, ablocks
[i
], 1,
1103 EXT4_FREE_BLOCKS_METADATA
);
1112 * ext4_ext_grow_indepth:
1113 * implements tree growing procedure:
1114 * - allocates new block
1115 * - moves top-level data (index block or leaf) into the new block
1116 * - initializes new top-level, creating index that points to the
1117 * just created block
1119 static int ext4_ext_grow_indepth(handle_t
*handle
, struct inode
*inode
,
1121 struct ext4_extent
*newext
)
1123 struct ext4_extent_header
*neh
;
1124 struct buffer_head
*bh
;
1125 ext4_fsblk_t newblock
;
1128 newblock
= ext4_ext_new_meta_block(handle
, inode
, NULL
,
1129 newext
, &err
, flags
);
1133 bh
= sb_getblk(inode
->i_sb
, newblock
);
1136 ext4_std_error(inode
->i_sb
, err
);
1141 err
= ext4_journal_get_create_access(handle
, bh
);
1147 /* move top-level index/leaf into new block */
1148 memmove(bh
->b_data
, EXT4_I(inode
)->i_data
,
1149 sizeof(EXT4_I(inode
)->i_data
));
1151 /* set size of new block */
1152 neh
= ext_block_hdr(bh
);
1153 /* old root could have indexes or leaves
1154 * so calculate e_max right way */
1155 if (ext_depth(inode
))
1156 neh
->eh_max
= cpu_to_le16(ext4_ext_space_block_idx(inode
, 0));
1158 neh
->eh_max
= cpu_to_le16(ext4_ext_space_block(inode
, 0));
1159 neh
->eh_magic
= EXT4_EXT_MAGIC
;
1160 ext4_extent_block_csum_set(inode
, neh
);
1161 set_buffer_uptodate(bh
);
1164 err
= ext4_handle_dirty_metadata(handle
, inode
, bh
);
1168 /* Update top-level index: num,max,pointer */
1169 neh
= ext_inode_hdr(inode
);
1170 neh
->eh_entries
= cpu_to_le16(1);
1171 ext4_idx_store_pblock(EXT_FIRST_INDEX(neh
), newblock
);
1172 if (neh
->eh_depth
== 0) {
1173 /* Root extent block becomes index block */
1174 neh
->eh_max
= cpu_to_le16(ext4_ext_space_root_idx(inode
, 0));
1175 EXT_FIRST_INDEX(neh
)->ei_block
=
1176 EXT_FIRST_EXTENT(neh
)->ee_block
;
1178 ext_debug("new root: num %d(%d), lblock %d, ptr %llu\n",
1179 le16_to_cpu(neh
->eh_entries
), le16_to_cpu(neh
->eh_max
),
1180 le32_to_cpu(EXT_FIRST_INDEX(neh
)->ei_block
),
1181 ext4_idx_pblock(EXT_FIRST_INDEX(neh
)));
1183 le16_add_cpu(&neh
->eh_depth
, 1);
1184 ext4_mark_inode_dirty(handle
, inode
);
1192 * ext4_ext_create_new_leaf:
1193 * finds empty index and adds new leaf.
1194 * if no free index is found, then it requests in-depth growing.
1196 static int ext4_ext_create_new_leaf(handle_t
*handle
, struct inode
*inode
,
1198 struct ext4_ext_path
*path
,
1199 struct ext4_extent
*newext
)
1201 struct ext4_ext_path
*curp
;
1202 int depth
, i
, err
= 0;
1205 i
= depth
= ext_depth(inode
);
1207 /* walk up to the tree and look for free index entry */
1208 curp
= path
+ depth
;
1209 while (i
> 0 && !EXT_HAS_FREE_INDEX(curp
)) {
1214 /* we use already allocated block for index block,
1215 * so subsequent data blocks should be contiguous */
1216 if (EXT_HAS_FREE_INDEX(curp
)) {
1217 /* if we found index with free entry, then use that
1218 * entry: create all needed subtree and add new leaf */
1219 err
= ext4_ext_split(handle
, inode
, flags
, path
, newext
, i
);
1224 ext4_ext_drop_refs(path
);
1225 path
= ext4_ext_find_extent(inode
,
1226 (ext4_lblk_t
)le32_to_cpu(newext
->ee_block
),
1229 err
= PTR_ERR(path
);
1231 /* tree is full, time to grow in depth */
1232 err
= ext4_ext_grow_indepth(handle
, inode
, flags
, newext
);
1237 ext4_ext_drop_refs(path
);
1238 path
= ext4_ext_find_extent(inode
,
1239 (ext4_lblk_t
)le32_to_cpu(newext
->ee_block
),
1242 err
= PTR_ERR(path
);
1247 * only first (depth 0 -> 1) produces free space;
1248 * in all other cases we have to split the grown tree
1250 depth
= ext_depth(inode
);
1251 if (path
[depth
].p_hdr
->eh_entries
== path
[depth
].p_hdr
->eh_max
) {
1252 /* now we need to split */
1262 * search the closest allocated block to the left for *logical
1263 * and returns it at @logical + it's physical address at @phys
1264 * if *logical is the smallest allocated block, the function
1265 * returns 0 at @phys
1266 * return value contains 0 (success) or error code
1268 static int ext4_ext_search_left(struct inode
*inode
,
1269 struct ext4_ext_path
*path
,
1270 ext4_lblk_t
*logical
, ext4_fsblk_t
*phys
)
1272 struct ext4_extent_idx
*ix
;
1273 struct ext4_extent
*ex
;
1276 if (unlikely(path
== NULL
)) {
1277 EXT4_ERROR_INODE(inode
, "path == NULL *logical %d!", *logical
);
1280 depth
= path
->p_depth
;
1283 if (depth
== 0 && path
->p_ext
== NULL
)
1286 /* usually extent in the path covers blocks smaller
1287 * then *logical, but it can be that extent is the
1288 * first one in the file */
1290 ex
= path
[depth
].p_ext
;
1291 ee_len
= ext4_ext_get_actual_len(ex
);
1292 if (*logical
< le32_to_cpu(ex
->ee_block
)) {
1293 if (unlikely(EXT_FIRST_EXTENT(path
[depth
].p_hdr
) != ex
)) {
1294 EXT4_ERROR_INODE(inode
,
1295 "EXT_FIRST_EXTENT != ex *logical %d ee_block %d!",
1296 *logical
, le32_to_cpu(ex
->ee_block
));
1299 while (--depth
>= 0) {
1300 ix
= path
[depth
].p_idx
;
1301 if (unlikely(ix
!= EXT_FIRST_INDEX(path
[depth
].p_hdr
))) {
1302 EXT4_ERROR_INODE(inode
,
1303 "ix (%d) != EXT_FIRST_INDEX (%d) (depth %d)!",
1304 ix
!= NULL
? le32_to_cpu(ix
->ei_block
) : 0,
1305 EXT_FIRST_INDEX(path
[depth
].p_hdr
) != NULL
?
1306 le32_to_cpu(EXT_FIRST_INDEX(path
[depth
].p_hdr
)->ei_block
) : 0,
1314 if (unlikely(*logical
< (le32_to_cpu(ex
->ee_block
) + ee_len
))) {
1315 EXT4_ERROR_INODE(inode
,
1316 "logical %d < ee_block %d + ee_len %d!",
1317 *logical
, le32_to_cpu(ex
->ee_block
), ee_len
);
1321 *logical
= le32_to_cpu(ex
->ee_block
) + ee_len
- 1;
1322 *phys
= ext4_ext_pblock(ex
) + ee_len
- 1;
1327 * search the closest allocated block to the right for *logical
1328 * and returns it at @logical + it's physical address at @phys
1329 * if *logical is the largest allocated block, the function
1330 * returns 0 at @phys
1331 * return value contains 0 (success) or error code
1333 static int ext4_ext_search_right(struct inode
*inode
,
1334 struct ext4_ext_path
*path
,
1335 ext4_lblk_t
*logical
, ext4_fsblk_t
*phys
,
1336 struct ext4_extent
**ret_ex
)
1338 struct buffer_head
*bh
= NULL
;
1339 struct ext4_extent_header
*eh
;
1340 struct ext4_extent_idx
*ix
;
1341 struct ext4_extent
*ex
;
1343 int depth
; /* Note, NOT eh_depth; depth from top of tree */
1346 if (unlikely(path
== NULL
)) {
1347 EXT4_ERROR_INODE(inode
, "path == NULL *logical %d!", *logical
);
1350 depth
= path
->p_depth
;
1353 if (depth
== 0 && path
->p_ext
== NULL
)
1356 /* usually extent in the path covers blocks smaller
1357 * then *logical, but it can be that extent is the
1358 * first one in the file */
1360 ex
= path
[depth
].p_ext
;
1361 ee_len
= ext4_ext_get_actual_len(ex
);
1362 if (*logical
< le32_to_cpu(ex
->ee_block
)) {
1363 if (unlikely(EXT_FIRST_EXTENT(path
[depth
].p_hdr
) != ex
)) {
1364 EXT4_ERROR_INODE(inode
,
1365 "first_extent(path[%d].p_hdr) != ex",
1369 while (--depth
>= 0) {
1370 ix
= path
[depth
].p_idx
;
1371 if (unlikely(ix
!= EXT_FIRST_INDEX(path
[depth
].p_hdr
))) {
1372 EXT4_ERROR_INODE(inode
,
1373 "ix != EXT_FIRST_INDEX *logical %d!",
1381 if (unlikely(*logical
< (le32_to_cpu(ex
->ee_block
) + ee_len
))) {
1382 EXT4_ERROR_INODE(inode
,
1383 "logical %d < ee_block %d + ee_len %d!",
1384 *logical
, le32_to_cpu(ex
->ee_block
), ee_len
);
1388 if (ex
!= EXT_LAST_EXTENT(path
[depth
].p_hdr
)) {
1389 /* next allocated block in this leaf */
1394 /* go up and search for index to the right */
1395 while (--depth
>= 0) {
1396 ix
= path
[depth
].p_idx
;
1397 if (ix
!= EXT_LAST_INDEX(path
[depth
].p_hdr
))
1401 /* we've gone up to the root and found no index to the right */
1405 /* we've found index to the right, let's
1406 * follow it and find the closest allocated
1407 * block to the right */
1409 block
= ext4_idx_pblock(ix
);
1410 while (++depth
< path
->p_depth
) {
1411 bh
= sb_bread(inode
->i_sb
, block
);
1414 eh
= ext_block_hdr(bh
);
1415 /* subtract from p_depth to get proper eh_depth */
1416 if (ext4_ext_check_block(inode
, eh
,
1417 path
->p_depth
- depth
, bh
)) {
1421 ix
= EXT_FIRST_INDEX(eh
);
1422 block
= ext4_idx_pblock(ix
);
1426 bh
= sb_bread(inode
->i_sb
, block
);
1429 eh
= ext_block_hdr(bh
);
1430 if (ext4_ext_check_block(inode
, eh
, path
->p_depth
- depth
, bh
)) {
1434 ex
= EXT_FIRST_EXTENT(eh
);
1436 *logical
= le32_to_cpu(ex
->ee_block
);
1437 *phys
= ext4_ext_pblock(ex
);
1445 * ext4_ext_next_allocated_block:
1446 * returns allocated block in subsequent extent or EXT_MAX_BLOCKS.
1447 * NOTE: it considers block number from index entry as
1448 * allocated block. Thus, index entries have to be consistent
1452 ext4_ext_next_allocated_block(struct ext4_ext_path
*path
)
1456 BUG_ON(path
== NULL
);
1457 depth
= path
->p_depth
;
1459 if (depth
== 0 && path
->p_ext
== NULL
)
1460 return EXT_MAX_BLOCKS
;
1462 while (depth
>= 0) {
1463 if (depth
== path
->p_depth
) {
1465 if (path
[depth
].p_ext
&&
1466 path
[depth
].p_ext
!=
1467 EXT_LAST_EXTENT(path
[depth
].p_hdr
))
1468 return le32_to_cpu(path
[depth
].p_ext
[1].ee_block
);
1471 if (path
[depth
].p_idx
!=
1472 EXT_LAST_INDEX(path
[depth
].p_hdr
))
1473 return le32_to_cpu(path
[depth
].p_idx
[1].ei_block
);
1478 return EXT_MAX_BLOCKS
;
1482 * ext4_ext_next_leaf_block:
1483 * returns first allocated block from next leaf or EXT_MAX_BLOCKS
1485 static ext4_lblk_t
ext4_ext_next_leaf_block(struct ext4_ext_path
*path
)
1489 BUG_ON(path
== NULL
);
1490 depth
= path
->p_depth
;
1492 /* zero-tree has no leaf blocks at all */
1494 return EXT_MAX_BLOCKS
;
1496 /* go to index block */
1499 while (depth
>= 0) {
1500 if (path
[depth
].p_idx
!=
1501 EXT_LAST_INDEX(path
[depth
].p_hdr
))
1502 return (ext4_lblk_t
)
1503 le32_to_cpu(path
[depth
].p_idx
[1].ei_block
);
1507 return EXT_MAX_BLOCKS
;
1511 * ext4_ext_correct_indexes:
1512 * if leaf gets modified and modified extent is first in the leaf,
1513 * then we have to correct all indexes above.
1514 * TODO: do we need to correct tree in all cases?
1516 static int ext4_ext_correct_indexes(handle_t
*handle
, struct inode
*inode
,
1517 struct ext4_ext_path
*path
)
1519 struct ext4_extent_header
*eh
;
1520 int depth
= ext_depth(inode
);
1521 struct ext4_extent
*ex
;
1525 eh
= path
[depth
].p_hdr
;
1526 ex
= path
[depth
].p_ext
;
1528 if (unlikely(ex
== NULL
|| eh
== NULL
)) {
1529 EXT4_ERROR_INODE(inode
,
1530 "ex %p == NULL or eh %p == NULL", ex
, eh
);
1535 /* there is no tree at all */
1539 if (ex
!= EXT_FIRST_EXTENT(eh
)) {
1540 /* we correct tree if first leaf got modified only */
1545 * TODO: we need correction if border is smaller than current one
1548 border
= path
[depth
].p_ext
->ee_block
;
1549 err
= ext4_ext_get_access(handle
, inode
, path
+ k
);
1552 path
[k
].p_idx
->ei_block
= border
;
1553 err
= ext4_ext_dirty(handle
, inode
, path
+ k
);
1558 /* change all left-side indexes */
1559 if (path
[k
+1].p_idx
!= EXT_FIRST_INDEX(path
[k
+1].p_hdr
))
1561 err
= ext4_ext_get_access(handle
, inode
, path
+ k
);
1564 path
[k
].p_idx
->ei_block
= border
;
1565 err
= ext4_ext_dirty(handle
, inode
, path
+ k
);
1574 ext4_can_extents_be_merged(struct inode
*inode
, struct ext4_extent
*ex1
,
1575 struct ext4_extent
*ex2
)
1577 unsigned short ext1_ee_len
, ext2_ee_len
, max_len
;
1580 * Make sure that either both extents are uninitialized, or
1583 if (ext4_ext_is_uninitialized(ex1
) ^ ext4_ext_is_uninitialized(ex2
))
1586 if (ext4_ext_is_uninitialized(ex1
))
1587 max_len
= EXT_UNINIT_MAX_LEN
;
1589 max_len
= EXT_INIT_MAX_LEN
;
1591 ext1_ee_len
= ext4_ext_get_actual_len(ex1
);
1592 ext2_ee_len
= ext4_ext_get_actual_len(ex2
);
1594 if (le32_to_cpu(ex1
->ee_block
) + ext1_ee_len
!=
1595 le32_to_cpu(ex2
->ee_block
))
1599 * To allow future support for preallocated extents to be added
1600 * as an RO_COMPAT feature, refuse to merge to extents if
1601 * this can result in the top bit of ee_len being set.
1603 if (ext1_ee_len
+ ext2_ee_len
> max_len
)
1605 #ifdef AGGRESSIVE_TEST
1606 if (ext1_ee_len
>= 4)
1610 if (ext4_ext_pblock(ex1
) + ext1_ee_len
== ext4_ext_pblock(ex2
))
1616 * This function tries to merge the "ex" extent to the next extent in the tree.
1617 * It always tries to merge towards right. If you want to merge towards
1618 * left, pass "ex - 1" as argument instead of "ex".
1619 * Returns 0 if the extents (ex and ex+1) were _not_ merged and returns
1620 * 1 if they got merged.
1622 static int ext4_ext_try_to_merge_right(struct inode
*inode
,
1623 struct ext4_ext_path
*path
,
1624 struct ext4_extent
*ex
)
1626 struct ext4_extent_header
*eh
;
1627 unsigned int depth
, len
;
1629 int uninitialized
= 0;
1631 depth
= ext_depth(inode
);
1632 BUG_ON(path
[depth
].p_hdr
== NULL
);
1633 eh
= path
[depth
].p_hdr
;
1635 while (ex
< EXT_LAST_EXTENT(eh
)) {
1636 if (!ext4_can_extents_be_merged(inode
, ex
, ex
+ 1))
1638 /* merge with next extent! */
1639 if (ext4_ext_is_uninitialized(ex
))
1641 ex
->ee_len
= cpu_to_le16(ext4_ext_get_actual_len(ex
)
1642 + ext4_ext_get_actual_len(ex
+ 1));
1644 ext4_ext_mark_uninitialized(ex
);
1646 if (ex
+ 1 < EXT_LAST_EXTENT(eh
)) {
1647 len
= (EXT_LAST_EXTENT(eh
) - ex
- 1)
1648 * sizeof(struct ext4_extent
);
1649 memmove(ex
+ 1, ex
+ 2, len
);
1651 le16_add_cpu(&eh
->eh_entries
, -1);
1653 WARN_ON(eh
->eh_entries
== 0);
1654 if (!eh
->eh_entries
)
1655 EXT4_ERROR_INODE(inode
, "eh->eh_entries = 0!");
1662 * This function does a very simple check to see if we can collapse
1663 * an extent tree with a single extent tree leaf block into the inode.
1665 static void ext4_ext_try_to_merge_up(handle_t
*handle
,
1666 struct inode
*inode
,
1667 struct ext4_ext_path
*path
)
1670 unsigned max_root
= ext4_ext_space_root(inode
, 0);
1673 if ((path
[0].p_depth
!= 1) ||
1674 (le16_to_cpu(path
[0].p_hdr
->eh_entries
) != 1) ||
1675 (le16_to_cpu(path
[1].p_hdr
->eh_entries
) > max_root
))
1679 * We need to modify the block allocation bitmap and the block
1680 * group descriptor to release the extent tree block. If we
1681 * can't get the journal credits, give up.
1683 if (ext4_journal_extend(handle
, 2))
1687 * Copy the extent data up to the inode
1689 blk
= ext4_idx_pblock(path
[0].p_idx
);
1690 s
= le16_to_cpu(path
[1].p_hdr
->eh_entries
) *
1691 sizeof(struct ext4_extent_idx
);
1692 s
+= sizeof(struct ext4_extent_header
);
1694 memcpy(path
[0].p_hdr
, path
[1].p_hdr
, s
);
1695 path
[0].p_depth
= 0;
1696 path
[0].p_ext
= EXT_FIRST_EXTENT(path
[0].p_hdr
) +
1697 (path
[1].p_ext
- EXT_FIRST_EXTENT(path
[1].p_hdr
));
1698 path
[0].p_hdr
->eh_max
= cpu_to_le16(max_root
);
1700 brelse(path
[1].p_bh
);
1701 ext4_free_blocks(handle
, inode
, NULL
, blk
, 1,
1702 EXT4_FREE_BLOCKS_METADATA
| EXT4_FREE_BLOCKS_FORGET
);
1706 * This function tries to merge the @ex extent to neighbours in the tree.
1707 * return 1 if merge left else 0.
1709 static void ext4_ext_try_to_merge(handle_t
*handle
,
1710 struct inode
*inode
,
1711 struct ext4_ext_path
*path
,
1712 struct ext4_extent
*ex
) {
1713 struct ext4_extent_header
*eh
;
1717 depth
= ext_depth(inode
);
1718 BUG_ON(path
[depth
].p_hdr
== NULL
);
1719 eh
= path
[depth
].p_hdr
;
1721 if (ex
> EXT_FIRST_EXTENT(eh
))
1722 merge_done
= ext4_ext_try_to_merge_right(inode
, path
, ex
- 1);
1725 (void) ext4_ext_try_to_merge_right(inode
, path
, ex
);
1727 ext4_ext_try_to_merge_up(handle
, inode
, path
);
1731 * check if a portion of the "newext" extent overlaps with an
1734 * If there is an overlap discovered, it updates the length of the newext
1735 * such that there will be no overlap, and then returns 1.
1736 * If there is no overlap found, it returns 0.
1738 static unsigned int ext4_ext_check_overlap(struct ext4_sb_info
*sbi
,
1739 struct inode
*inode
,
1740 struct ext4_extent
*newext
,
1741 struct ext4_ext_path
*path
)
1744 unsigned int depth
, len1
;
1745 unsigned int ret
= 0;
1747 b1
= le32_to_cpu(newext
->ee_block
);
1748 len1
= ext4_ext_get_actual_len(newext
);
1749 depth
= ext_depth(inode
);
1750 if (!path
[depth
].p_ext
)
1752 b2
= le32_to_cpu(path
[depth
].p_ext
->ee_block
);
1753 b2
&= ~(sbi
->s_cluster_ratio
- 1);
1756 * get the next allocated block if the extent in the path
1757 * is before the requested block(s)
1760 b2
= ext4_ext_next_allocated_block(path
);
1761 if (b2
== EXT_MAX_BLOCKS
)
1763 b2
&= ~(sbi
->s_cluster_ratio
- 1);
1766 /* check for wrap through zero on extent logical start block*/
1767 if (b1
+ len1
< b1
) {
1768 len1
= EXT_MAX_BLOCKS
- b1
;
1769 newext
->ee_len
= cpu_to_le16(len1
);
1773 /* check for overlap */
1774 if (b1
+ len1
> b2
) {
1775 newext
->ee_len
= cpu_to_le16(b2
- b1
);
1783 * ext4_ext_insert_extent:
1784 * tries to merge requsted extent into the existing extent or
1785 * inserts requested extent as new one into the tree,
1786 * creating new leaf in the no-space case.
1788 int ext4_ext_insert_extent(handle_t
*handle
, struct inode
*inode
,
1789 struct ext4_ext_path
*path
,
1790 struct ext4_extent
*newext
, int flag
)
1792 struct ext4_extent_header
*eh
;
1793 struct ext4_extent
*ex
, *fex
;
1794 struct ext4_extent
*nearex
; /* nearest extent */
1795 struct ext4_ext_path
*npath
= NULL
;
1796 int depth
, len
, err
;
1798 unsigned uninitialized
= 0;
1801 if (unlikely(ext4_ext_get_actual_len(newext
) == 0)) {
1802 EXT4_ERROR_INODE(inode
, "ext4_ext_get_actual_len(newext) == 0");
1805 depth
= ext_depth(inode
);
1806 ex
= path
[depth
].p_ext
;
1807 if (unlikely(path
[depth
].p_hdr
== NULL
)) {
1808 EXT4_ERROR_INODE(inode
, "path[%d].p_hdr == NULL", depth
);
1812 /* try to insert block into found extent and return */
1813 if (ex
&& !(flag
& EXT4_GET_BLOCKS_PRE_IO
)
1814 && ext4_can_extents_be_merged(inode
, ex
, newext
)) {
1815 ext_debug("append [%d]%d block to %u:[%d]%d (from %llu)\n",
1816 ext4_ext_is_uninitialized(newext
),
1817 ext4_ext_get_actual_len(newext
),
1818 le32_to_cpu(ex
->ee_block
),
1819 ext4_ext_is_uninitialized(ex
),
1820 ext4_ext_get_actual_len(ex
),
1821 ext4_ext_pblock(ex
));
1822 err
= ext4_ext_get_access(handle
, inode
, path
+ depth
);
1827 * ext4_can_extents_be_merged should have checked that either
1828 * both extents are uninitialized, or both aren't. Thus we
1829 * need to check only one of them here.
1831 if (ext4_ext_is_uninitialized(ex
))
1833 ex
->ee_len
= cpu_to_le16(ext4_ext_get_actual_len(ex
)
1834 + ext4_ext_get_actual_len(newext
));
1836 ext4_ext_mark_uninitialized(ex
);
1837 eh
= path
[depth
].p_hdr
;
1842 depth
= ext_depth(inode
);
1843 eh
= path
[depth
].p_hdr
;
1844 if (le16_to_cpu(eh
->eh_entries
) < le16_to_cpu(eh
->eh_max
))
1847 /* probably next leaf has space for us? */
1848 fex
= EXT_LAST_EXTENT(eh
);
1849 next
= EXT_MAX_BLOCKS
;
1850 if (le32_to_cpu(newext
->ee_block
) > le32_to_cpu(fex
->ee_block
))
1851 next
= ext4_ext_next_leaf_block(path
);
1852 if (next
!= EXT_MAX_BLOCKS
) {
1853 ext_debug("next leaf block - %u\n", next
);
1854 BUG_ON(npath
!= NULL
);
1855 npath
= ext4_ext_find_extent(inode
, next
, NULL
);
1857 return PTR_ERR(npath
);
1858 BUG_ON(npath
->p_depth
!= path
->p_depth
);
1859 eh
= npath
[depth
].p_hdr
;
1860 if (le16_to_cpu(eh
->eh_entries
) < le16_to_cpu(eh
->eh_max
)) {
1861 ext_debug("next leaf isn't full(%d)\n",
1862 le16_to_cpu(eh
->eh_entries
));
1866 ext_debug("next leaf has no free space(%d,%d)\n",
1867 le16_to_cpu(eh
->eh_entries
), le16_to_cpu(eh
->eh_max
));
1871 * There is no free space in the found leaf.
1872 * We're gonna add a new leaf in the tree.
1874 if (flag
& EXT4_GET_BLOCKS_PUNCH_OUT_EXT
)
1875 flags
= EXT4_MB_USE_ROOT_BLOCKS
;
1876 err
= ext4_ext_create_new_leaf(handle
, inode
, flags
, path
, newext
);
1879 depth
= ext_depth(inode
);
1880 eh
= path
[depth
].p_hdr
;
1883 nearex
= path
[depth
].p_ext
;
1885 err
= ext4_ext_get_access(handle
, inode
, path
+ depth
);
1890 /* there is no extent in this leaf, create first one */
1891 ext_debug("first extent in the leaf: %u:%llu:[%d]%d\n",
1892 le32_to_cpu(newext
->ee_block
),
1893 ext4_ext_pblock(newext
),
1894 ext4_ext_is_uninitialized(newext
),
1895 ext4_ext_get_actual_len(newext
));
1896 nearex
= EXT_FIRST_EXTENT(eh
);
1898 if (le32_to_cpu(newext
->ee_block
)
1899 > le32_to_cpu(nearex
->ee_block
)) {
1901 ext_debug("insert %u:%llu:[%d]%d before: "
1903 le32_to_cpu(newext
->ee_block
),
1904 ext4_ext_pblock(newext
),
1905 ext4_ext_is_uninitialized(newext
),
1906 ext4_ext_get_actual_len(newext
),
1911 BUG_ON(newext
->ee_block
== nearex
->ee_block
);
1912 ext_debug("insert %u:%llu:[%d]%d after: "
1914 le32_to_cpu(newext
->ee_block
),
1915 ext4_ext_pblock(newext
),
1916 ext4_ext_is_uninitialized(newext
),
1917 ext4_ext_get_actual_len(newext
),
1920 len
= EXT_LAST_EXTENT(eh
) - nearex
+ 1;
1922 ext_debug("insert %u:%llu:[%d]%d: "
1923 "move %d extents from 0x%p to 0x%p\n",
1924 le32_to_cpu(newext
->ee_block
),
1925 ext4_ext_pblock(newext
),
1926 ext4_ext_is_uninitialized(newext
),
1927 ext4_ext_get_actual_len(newext
),
1928 len
, nearex
, nearex
+ 1);
1929 memmove(nearex
+ 1, nearex
,
1930 len
* sizeof(struct ext4_extent
));
1934 le16_add_cpu(&eh
->eh_entries
, 1);
1935 path
[depth
].p_ext
= nearex
;
1936 nearex
->ee_block
= newext
->ee_block
;
1937 ext4_ext_store_pblock(nearex
, ext4_ext_pblock(newext
));
1938 nearex
->ee_len
= newext
->ee_len
;
1941 /* try to merge extents */
1942 if (!(flag
& EXT4_GET_BLOCKS_PRE_IO
))
1943 ext4_ext_try_to_merge(handle
, inode
, path
, nearex
);
1946 /* time to correct all indexes above */
1947 err
= ext4_ext_correct_indexes(handle
, inode
, path
);
1951 err
= ext4_ext_dirty(handle
, inode
, path
+ path
->p_depth
);
1955 ext4_ext_drop_refs(npath
);
1958 ext4_ext_invalidate_cache(inode
);
1962 static int ext4_ext_walk_space(struct inode
*inode
, ext4_lblk_t block
,
1963 ext4_lblk_t num
, ext_prepare_callback func
,
1966 struct ext4_ext_path
*path
= NULL
;
1967 struct ext4_ext_cache cbex
;
1968 struct ext4_extent
*ex
;
1969 ext4_lblk_t next
, start
= 0, end
= 0;
1970 ext4_lblk_t last
= block
+ num
;
1971 int depth
, exists
, err
= 0;
1973 BUG_ON(func
== NULL
);
1974 BUG_ON(inode
== NULL
);
1976 while (block
< last
&& block
!= EXT_MAX_BLOCKS
) {
1978 /* find extent for this block */
1979 down_read(&EXT4_I(inode
)->i_data_sem
);
1980 path
= ext4_ext_find_extent(inode
, block
, path
);
1981 up_read(&EXT4_I(inode
)->i_data_sem
);
1983 err
= PTR_ERR(path
);
1988 depth
= ext_depth(inode
);
1989 if (unlikely(path
[depth
].p_hdr
== NULL
)) {
1990 EXT4_ERROR_INODE(inode
, "path[%d].p_hdr == NULL", depth
);
1994 ex
= path
[depth
].p_ext
;
1995 next
= ext4_ext_next_allocated_block(path
);
1999 /* there is no extent yet, so try to allocate
2000 * all requested space */
2003 } else if (le32_to_cpu(ex
->ee_block
) > block
) {
2004 /* need to allocate space before found extent */
2006 end
= le32_to_cpu(ex
->ee_block
);
2007 if (block
+ num
< end
)
2009 } else if (block
>= le32_to_cpu(ex
->ee_block
)
2010 + ext4_ext_get_actual_len(ex
)) {
2011 /* need to allocate space after found extent */
2016 } else if (block
>= le32_to_cpu(ex
->ee_block
)) {
2018 * some part of requested space is covered
2022 end
= le32_to_cpu(ex
->ee_block
)
2023 + ext4_ext_get_actual_len(ex
);
2024 if (block
+ num
< end
)
2030 BUG_ON(end
<= start
);
2033 cbex
.ec_block
= start
;
2034 cbex
.ec_len
= end
- start
;
2037 cbex
.ec_block
= le32_to_cpu(ex
->ee_block
);
2038 cbex
.ec_len
= ext4_ext_get_actual_len(ex
);
2039 cbex
.ec_start
= ext4_ext_pblock(ex
);
2042 if (unlikely(cbex
.ec_len
== 0)) {
2043 EXT4_ERROR_INODE(inode
, "cbex.ec_len == 0");
2047 err
= func(inode
, next
, &cbex
, ex
, cbdata
);
2048 ext4_ext_drop_refs(path
);
2053 if (err
== EXT_REPEAT
)
2055 else if (err
== EXT_BREAK
) {
2060 if (ext_depth(inode
) != depth
) {
2061 /* depth was changed. we have to realloc path */
2066 block
= cbex
.ec_block
+ cbex
.ec_len
;
2070 ext4_ext_drop_refs(path
);
2078 ext4_ext_put_in_cache(struct inode
*inode
, ext4_lblk_t block
,
2079 __u32 len
, ext4_fsblk_t start
)
2081 struct ext4_ext_cache
*cex
;
2083 spin_lock(&EXT4_I(inode
)->i_block_reservation_lock
);
2084 trace_ext4_ext_put_in_cache(inode
, block
, len
, start
);
2085 cex
= &EXT4_I(inode
)->i_cached_extent
;
2086 cex
->ec_block
= block
;
2088 cex
->ec_start
= start
;
2089 spin_unlock(&EXT4_I(inode
)->i_block_reservation_lock
);
2093 * ext4_ext_put_gap_in_cache:
2094 * calculate boundaries of the gap that the requested block fits into
2095 * and cache this gap
2098 ext4_ext_put_gap_in_cache(struct inode
*inode
, struct ext4_ext_path
*path
,
2101 int depth
= ext_depth(inode
);
2104 struct ext4_extent
*ex
;
2106 ex
= path
[depth
].p_ext
;
2108 /* there is no extent yet, so gap is [0;-] */
2110 len
= EXT_MAX_BLOCKS
;
2111 ext_debug("cache gap(whole file):");
2112 } else if (block
< le32_to_cpu(ex
->ee_block
)) {
2114 len
= le32_to_cpu(ex
->ee_block
) - block
;
2115 ext_debug("cache gap(before): %u [%u:%u]",
2117 le32_to_cpu(ex
->ee_block
),
2118 ext4_ext_get_actual_len(ex
));
2119 } else if (block
>= le32_to_cpu(ex
->ee_block
)
2120 + ext4_ext_get_actual_len(ex
)) {
2122 lblock
= le32_to_cpu(ex
->ee_block
)
2123 + ext4_ext_get_actual_len(ex
);
2125 next
= ext4_ext_next_allocated_block(path
);
2126 ext_debug("cache gap(after): [%u:%u] %u",
2127 le32_to_cpu(ex
->ee_block
),
2128 ext4_ext_get_actual_len(ex
),
2130 BUG_ON(next
== lblock
);
2131 len
= next
- lblock
;
2137 ext_debug(" -> %u:%lu\n", lblock
, len
);
2138 ext4_ext_put_in_cache(inode
, lblock
, len
, 0);
2142 * ext4_ext_in_cache()
2143 * Checks to see if the given block is in the cache.
2144 * If it is, the cached extent is stored in the given
2145 * cache extent pointer.
2147 * @inode: The files inode
2148 * @block: The block to look for in the cache
2149 * @ex: Pointer where the cached extent will be stored
2150 * if it contains block
2152 * Return 0 if cache is invalid; 1 if the cache is valid
2155 ext4_ext_in_cache(struct inode
*inode
, ext4_lblk_t block
,
2156 struct ext4_extent
*ex
)
2158 struct ext4_ext_cache
*cex
;
2159 struct ext4_sb_info
*sbi
;
2163 * We borrow i_block_reservation_lock to protect i_cached_extent
2165 spin_lock(&EXT4_I(inode
)->i_block_reservation_lock
);
2166 cex
= &EXT4_I(inode
)->i_cached_extent
;
2167 sbi
= EXT4_SB(inode
->i_sb
);
2169 /* has cache valid data? */
2170 if (cex
->ec_len
== 0)
2173 if (in_range(block
, cex
->ec_block
, cex
->ec_len
)) {
2174 ex
->ee_block
= cpu_to_le32(cex
->ec_block
);
2175 ext4_ext_store_pblock(ex
, cex
->ec_start
);
2176 ex
->ee_len
= cpu_to_le16(cex
->ec_len
);
2177 ext_debug("%u cached by %u:%u:%llu\n",
2179 cex
->ec_block
, cex
->ec_len
, cex
->ec_start
);
2183 trace_ext4_ext_in_cache(inode
, block
, ret
);
2184 spin_unlock(&EXT4_I(inode
)->i_block_reservation_lock
);
2190 * removes index from the index block.
2192 static int ext4_ext_rm_idx(handle_t
*handle
, struct inode
*inode
,
2193 struct ext4_ext_path
*path
)
2198 /* free index block */
2200 leaf
= ext4_idx_pblock(path
->p_idx
);
2201 if (unlikely(path
->p_hdr
->eh_entries
== 0)) {
2202 EXT4_ERROR_INODE(inode
, "path->p_hdr->eh_entries == 0");
2205 err
= ext4_ext_get_access(handle
, inode
, path
);
2209 if (path
->p_idx
!= EXT_LAST_INDEX(path
->p_hdr
)) {
2210 int len
= EXT_LAST_INDEX(path
->p_hdr
) - path
->p_idx
;
2211 len
*= sizeof(struct ext4_extent_idx
);
2212 memmove(path
->p_idx
, path
->p_idx
+ 1, len
);
2215 le16_add_cpu(&path
->p_hdr
->eh_entries
, -1);
2216 err
= ext4_ext_dirty(handle
, inode
, path
);
2219 ext_debug("index is empty, remove it, free block %llu\n", leaf
);
2220 trace_ext4_ext_rm_idx(inode
, leaf
);
2222 ext4_free_blocks(handle
, inode
, NULL
, leaf
, 1,
2223 EXT4_FREE_BLOCKS_METADATA
| EXT4_FREE_BLOCKS_FORGET
);
2228 * ext4_ext_calc_credits_for_single_extent:
2229 * This routine returns max. credits that needed to insert an extent
2230 * to the extent tree.
2231 * When pass the actual path, the caller should calculate credits
2234 int ext4_ext_calc_credits_for_single_extent(struct inode
*inode
, int nrblocks
,
2235 struct ext4_ext_path
*path
)
2238 int depth
= ext_depth(inode
);
2241 /* probably there is space in leaf? */
2242 if (le16_to_cpu(path
[depth
].p_hdr
->eh_entries
)
2243 < le16_to_cpu(path
[depth
].p_hdr
->eh_max
)) {
2246 * There are some space in the leaf tree, no
2247 * need to account for leaf block credit
2249 * bitmaps and block group descriptor blocks
2250 * and other metadata blocks still need to be
2253 /* 1 bitmap, 1 block group descriptor */
2254 ret
= 2 + EXT4_META_TRANS_BLOCKS(inode
->i_sb
);
2259 return ext4_chunk_trans_blocks(inode
, nrblocks
);
2263 * How many index/leaf blocks need to change/allocate to modify nrblocks?
2265 * if nrblocks are fit in a single extent (chunk flag is 1), then
2266 * in the worse case, each tree level index/leaf need to be changed
2267 * if the tree split due to insert a new extent, then the old tree
2268 * index/leaf need to be updated too
2270 * If the nrblocks are discontiguous, they could cause
2271 * the whole tree split more than once, but this is really rare.
2273 int ext4_ext_index_trans_blocks(struct inode
*inode
, int nrblocks
, int chunk
)
2276 int depth
= ext_depth(inode
);
2286 static int ext4_remove_blocks(handle_t
*handle
, struct inode
*inode
,
2287 struct ext4_extent
*ex
,
2288 ext4_fsblk_t
*partial_cluster
,
2289 ext4_lblk_t from
, ext4_lblk_t to
)
2291 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
2292 unsigned short ee_len
= ext4_ext_get_actual_len(ex
);
2296 if (S_ISDIR(inode
->i_mode
) || S_ISLNK(inode
->i_mode
))
2297 flags
|= EXT4_FREE_BLOCKS_METADATA
| EXT4_FREE_BLOCKS_FORGET
;
2298 else if (ext4_should_journal_data(inode
))
2299 flags
|= EXT4_FREE_BLOCKS_FORGET
;
2302 * For bigalloc file systems, we never free a partial cluster
2303 * at the beginning of the extent. Instead, we make a note
2304 * that we tried freeing the cluster, and check to see if we
2305 * need to free it on a subsequent call to ext4_remove_blocks,
2306 * or at the end of the ext4_truncate() operation.
2308 flags
|= EXT4_FREE_BLOCKS_NOFREE_FIRST_CLUSTER
;
2310 trace_ext4_remove_blocks(inode
, ex
, from
, to
, *partial_cluster
);
2312 * If we have a partial cluster, and it's different from the
2313 * cluster of the last block, we need to explicitly free the
2314 * partial cluster here.
2316 pblk
= ext4_ext_pblock(ex
) + ee_len
- 1;
2317 if (*partial_cluster
&& (EXT4_B2C(sbi
, pblk
) != *partial_cluster
)) {
2318 ext4_free_blocks(handle
, inode
, NULL
,
2319 EXT4_C2B(sbi
, *partial_cluster
),
2320 sbi
->s_cluster_ratio
, flags
);
2321 *partial_cluster
= 0;
2324 #ifdef EXTENTS_STATS
2326 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
2327 spin_lock(&sbi
->s_ext_stats_lock
);
2328 sbi
->s_ext_blocks
+= ee_len
;
2329 sbi
->s_ext_extents
++;
2330 if (ee_len
< sbi
->s_ext_min
)
2331 sbi
->s_ext_min
= ee_len
;
2332 if (ee_len
> sbi
->s_ext_max
)
2333 sbi
->s_ext_max
= ee_len
;
2334 if (ext_depth(inode
) > sbi
->s_depth_max
)
2335 sbi
->s_depth_max
= ext_depth(inode
);
2336 spin_unlock(&sbi
->s_ext_stats_lock
);
2339 if (from
>= le32_to_cpu(ex
->ee_block
)
2340 && to
== le32_to_cpu(ex
->ee_block
) + ee_len
- 1) {
2344 num
= le32_to_cpu(ex
->ee_block
) + ee_len
- from
;
2345 pblk
= ext4_ext_pblock(ex
) + ee_len
- num
;
2346 ext_debug("free last %u blocks starting %llu\n", num
, pblk
);
2347 ext4_free_blocks(handle
, inode
, NULL
, pblk
, num
, flags
);
2349 * If the block range to be freed didn't start at the
2350 * beginning of a cluster, and we removed the entire
2351 * extent, save the partial cluster here, since we
2352 * might need to delete if we determine that the
2353 * truncate operation has removed all of the blocks in
2356 if (pblk
& (sbi
->s_cluster_ratio
- 1) &&
2358 *partial_cluster
= EXT4_B2C(sbi
, pblk
);
2360 *partial_cluster
= 0;
2361 } else if (from
== le32_to_cpu(ex
->ee_block
)
2362 && to
<= le32_to_cpu(ex
->ee_block
) + ee_len
- 1) {
2368 start
= ext4_ext_pblock(ex
);
2370 ext_debug("free first %u blocks starting %llu\n", num
, start
);
2371 ext4_free_blocks(handle
, inode
, NULL
, start
, num
, flags
);
2374 printk(KERN_INFO
"strange request: removal(2) "
2375 "%u-%u from %u:%u\n",
2376 from
, to
, le32_to_cpu(ex
->ee_block
), ee_len
);
2383 * ext4_ext_rm_leaf() Removes the extents associated with the
2384 * blocks appearing between "start" and "end", and splits the extents
2385 * if "start" and "end" appear in the same extent
2387 * @handle: The journal handle
2388 * @inode: The files inode
2389 * @path: The path to the leaf
2390 * @start: The first block to remove
2391 * @end: The last block to remove
2394 ext4_ext_rm_leaf(handle_t
*handle
, struct inode
*inode
,
2395 struct ext4_ext_path
*path
, ext4_fsblk_t
*partial_cluster
,
2396 ext4_lblk_t start
, ext4_lblk_t end
)
2398 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
2399 int err
= 0, correct_index
= 0;
2400 int depth
= ext_depth(inode
), credits
;
2401 struct ext4_extent_header
*eh
;
2404 ext4_lblk_t ex_ee_block
;
2405 unsigned short ex_ee_len
;
2406 unsigned uninitialized
= 0;
2407 struct ext4_extent
*ex
;
2409 /* the header must be checked already in ext4_ext_remove_space() */
2410 ext_debug("truncate since %u in leaf to %u\n", start
, end
);
2411 if (!path
[depth
].p_hdr
)
2412 path
[depth
].p_hdr
= ext_block_hdr(path
[depth
].p_bh
);
2413 eh
= path
[depth
].p_hdr
;
2414 if (unlikely(path
[depth
].p_hdr
== NULL
)) {
2415 EXT4_ERROR_INODE(inode
, "path[%d].p_hdr == NULL", depth
);
2418 /* find where to start removing */
2419 ex
= EXT_LAST_EXTENT(eh
);
2421 ex_ee_block
= le32_to_cpu(ex
->ee_block
);
2422 ex_ee_len
= ext4_ext_get_actual_len(ex
);
2424 trace_ext4_ext_rm_leaf(inode
, start
, ex
, *partial_cluster
);
2426 while (ex
>= EXT_FIRST_EXTENT(eh
) &&
2427 ex_ee_block
+ ex_ee_len
> start
) {
2429 if (ext4_ext_is_uninitialized(ex
))
2434 ext_debug("remove ext %u:[%d]%d\n", ex_ee_block
,
2435 uninitialized
, ex_ee_len
);
2436 path
[depth
].p_ext
= ex
;
2438 a
= ex_ee_block
> start
? ex_ee_block
: start
;
2439 b
= ex_ee_block
+ex_ee_len
- 1 < end
?
2440 ex_ee_block
+ex_ee_len
- 1 : end
;
2442 ext_debug(" border %u:%u\n", a
, b
);
2444 /* If this extent is beyond the end of the hole, skip it */
2445 if (end
< ex_ee_block
) {
2447 ex_ee_block
= le32_to_cpu(ex
->ee_block
);
2448 ex_ee_len
= ext4_ext_get_actual_len(ex
);
2450 } else if (b
!= ex_ee_block
+ ex_ee_len
- 1) {
2451 EXT4_ERROR_INODE(inode
,
2452 "can not handle truncate %u:%u "
2454 start
, end
, ex_ee_block
,
2455 ex_ee_block
+ ex_ee_len
- 1);
2458 } else if (a
!= ex_ee_block
) {
2459 /* remove tail of the extent */
2460 num
= a
- ex_ee_block
;
2462 /* remove whole extent: excellent! */
2466 * 3 for leaf, sb, and inode plus 2 (bmap and group
2467 * descriptor) for each block group; assume two block
2468 * groups plus ex_ee_len/blocks_per_block_group for
2471 credits
= 7 + 2*(ex_ee_len
/EXT4_BLOCKS_PER_GROUP(inode
->i_sb
));
2472 if (ex
== EXT_FIRST_EXTENT(eh
)) {
2474 credits
+= (ext_depth(inode
)) + 1;
2476 credits
+= EXT4_MAXQUOTAS_TRANS_BLOCKS(inode
->i_sb
);
2478 err
= ext4_ext_truncate_extend_restart(handle
, inode
, credits
);
2482 err
= ext4_ext_get_access(handle
, inode
, path
+ depth
);
2486 err
= ext4_remove_blocks(handle
, inode
, ex
, partial_cluster
,
2492 /* this extent is removed; mark slot entirely unused */
2493 ext4_ext_store_pblock(ex
, 0);
2495 ex
->ee_len
= cpu_to_le16(num
);
2497 * Do not mark uninitialized if all the blocks in the
2498 * extent have been removed.
2500 if (uninitialized
&& num
)
2501 ext4_ext_mark_uninitialized(ex
);
2503 * If the extent was completely released,
2504 * we need to remove it from the leaf
2507 if (end
!= EXT_MAX_BLOCKS
- 1) {
2509 * For hole punching, we need to scoot all the
2510 * extents up when an extent is removed so that
2511 * we dont have blank extents in the middle
2513 memmove(ex
, ex
+1, (EXT_LAST_EXTENT(eh
) - ex
) *
2514 sizeof(struct ext4_extent
));
2516 /* Now get rid of the one at the end */
2517 memset(EXT_LAST_EXTENT(eh
), 0,
2518 sizeof(struct ext4_extent
));
2520 le16_add_cpu(&eh
->eh_entries
, -1);
2522 *partial_cluster
= 0;
2524 err
= ext4_ext_dirty(handle
, inode
, path
+ depth
);
2528 ext_debug("new extent: %u:%u:%llu\n", ex_ee_block
, num
,
2529 ext4_ext_pblock(ex
));
2531 ex_ee_block
= le32_to_cpu(ex
->ee_block
);
2532 ex_ee_len
= ext4_ext_get_actual_len(ex
);
2535 if (correct_index
&& eh
->eh_entries
)
2536 err
= ext4_ext_correct_indexes(handle
, inode
, path
);
2539 * If there is still a entry in the leaf node, check to see if
2540 * it references the partial cluster. This is the only place
2541 * where it could; if it doesn't, we can free the cluster.
2543 if (*partial_cluster
&& ex
>= EXT_FIRST_EXTENT(eh
) &&
2544 (EXT4_B2C(sbi
, ext4_ext_pblock(ex
) + ex_ee_len
- 1) !=
2545 *partial_cluster
)) {
2546 int flags
= EXT4_FREE_BLOCKS_FORGET
;
2548 if (S_ISDIR(inode
->i_mode
) || S_ISLNK(inode
->i_mode
))
2549 flags
|= EXT4_FREE_BLOCKS_METADATA
;
2551 ext4_free_blocks(handle
, inode
, NULL
,
2552 EXT4_C2B(sbi
, *partial_cluster
),
2553 sbi
->s_cluster_ratio
, flags
);
2554 *partial_cluster
= 0;
2557 /* if this leaf is free, then we should
2558 * remove it from index block above */
2559 if (err
== 0 && eh
->eh_entries
== 0 && path
[depth
].p_bh
!= NULL
)
2560 err
= ext4_ext_rm_idx(handle
, inode
, path
+ depth
);
2567 * ext4_ext_more_to_rm:
2568 * returns 1 if current index has to be freed (even partial)
2571 ext4_ext_more_to_rm(struct ext4_ext_path
*path
)
2573 BUG_ON(path
->p_idx
== NULL
);
2575 if (path
->p_idx
< EXT_FIRST_INDEX(path
->p_hdr
))
2579 * if truncate on deeper level happened, it wasn't partial,
2580 * so we have to consider current index for truncation
2582 if (le16_to_cpu(path
->p_hdr
->eh_entries
) == path
->p_block
)
2587 static int ext4_ext_remove_space(struct inode
*inode
, ext4_lblk_t start
,
2590 struct super_block
*sb
= inode
->i_sb
;
2591 int depth
= ext_depth(inode
);
2592 struct ext4_ext_path
*path
= NULL
;
2593 ext4_fsblk_t partial_cluster
= 0;
2597 ext_debug("truncate since %u to %u\n", start
, end
);
2599 /* probably first extent we're gonna free will be last in block */
2600 handle
= ext4_journal_start(inode
, depth
+ 1);
2602 return PTR_ERR(handle
);
2605 ext4_ext_invalidate_cache(inode
);
2607 trace_ext4_ext_remove_space(inode
, start
, depth
);
2610 * Check if we are removing extents inside the extent tree. If that
2611 * is the case, we are going to punch a hole inside the extent tree
2612 * so we have to check whether we need to split the extent covering
2613 * the last block to remove so we can easily remove the part of it
2614 * in ext4_ext_rm_leaf().
2616 if (end
< EXT_MAX_BLOCKS
- 1) {
2617 struct ext4_extent
*ex
;
2618 ext4_lblk_t ee_block
;
2620 /* find extent for this block */
2621 path
= ext4_ext_find_extent(inode
, end
, NULL
);
2623 ext4_journal_stop(handle
);
2624 return PTR_ERR(path
);
2626 depth
= ext_depth(inode
);
2627 /* Leaf not may not exist only if inode has no blocks at all */
2628 ex
= path
[depth
].p_ext
;
2631 EXT4_ERROR_INODE(inode
,
2632 "path[%d].p_hdr == NULL",
2639 ee_block
= le32_to_cpu(ex
->ee_block
);
2642 * See if the last block is inside the extent, if so split
2643 * the extent at 'end' block so we can easily remove the
2644 * tail of the first part of the split extent in
2645 * ext4_ext_rm_leaf().
2647 if (end
>= ee_block
&&
2648 end
< ee_block
+ ext4_ext_get_actual_len(ex
) - 1) {
2651 if (ext4_ext_is_uninitialized(ex
))
2652 split_flag
= EXT4_EXT_MARK_UNINIT1
|
2653 EXT4_EXT_MARK_UNINIT2
;
2656 * Split the extent in two so that 'end' is the last
2657 * block in the first new extent
2659 err
= ext4_split_extent_at(handle
, inode
, path
,
2660 end
+ 1, split_flag
,
2661 EXT4_GET_BLOCKS_PRE_IO
|
2662 EXT4_GET_BLOCKS_PUNCH_OUT_EXT
);
2669 * We start scanning from right side, freeing all the blocks
2670 * after i_size and walking into the tree depth-wise.
2672 depth
= ext_depth(inode
);
2677 le16_to_cpu(path
[k
].p_hdr
->eh_entries
)+1;
2679 path
= kzalloc(sizeof(struct ext4_ext_path
) * (depth
+ 1),
2682 ext4_journal_stop(handle
);
2685 path
[0].p_depth
= depth
;
2686 path
[0].p_hdr
= ext_inode_hdr(inode
);
2689 if (ext4_ext_check(inode
, path
[0].p_hdr
, depth
)) {
2696 while (i
>= 0 && err
== 0) {
2698 /* this is leaf block */
2699 err
= ext4_ext_rm_leaf(handle
, inode
, path
,
2700 &partial_cluster
, start
,
2702 /* root level has p_bh == NULL, brelse() eats this */
2703 brelse(path
[i
].p_bh
);
2704 path
[i
].p_bh
= NULL
;
2709 /* this is index block */
2710 if (!path
[i
].p_hdr
) {
2711 ext_debug("initialize header\n");
2712 path
[i
].p_hdr
= ext_block_hdr(path
[i
].p_bh
);
2715 if (!path
[i
].p_idx
) {
2716 /* this level hasn't been touched yet */
2717 path
[i
].p_idx
= EXT_LAST_INDEX(path
[i
].p_hdr
);
2718 path
[i
].p_block
= le16_to_cpu(path
[i
].p_hdr
->eh_entries
)+1;
2719 ext_debug("init index ptr: hdr 0x%p, num %d\n",
2721 le16_to_cpu(path
[i
].p_hdr
->eh_entries
));
2723 /* we were already here, see at next index */
2727 ext_debug("level %d - index, first 0x%p, cur 0x%p\n",
2728 i
, EXT_FIRST_INDEX(path
[i
].p_hdr
),
2730 if (ext4_ext_more_to_rm(path
+ i
)) {
2731 struct buffer_head
*bh
;
2732 /* go to the next level */
2733 ext_debug("move to level %d (block %llu)\n",
2734 i
+ 1, ext4_idx_pblock(path
[i
].p_idx
));
2735 memset(path
+ i
+ 1, 0, sizeof(*path
));
2736 bh
= sb_bread(sb
, ext4_idx_pblock(path
[i
].p_idx
));
2738 /* should we reset i_size? */
2742 if (WARN_ON(i
+ 1 > depth
)) {
2746 if (ext4_ext_check_block(inode
, ext_block_hdr(bh
),
2747 depth
- i
- 1, bh
)) {
2751 path
[i
+ 1].p_bh
= bh
;
2753 /* save actual number of indexes since this
2754 * number is changed at the next iteration */
2755 path
[i
].p_block
= le16_to_cpu(path
[i
].p_hdr
->eh_entries
);
2758 /* we finished processing this index, go up */
2759 if (path
[i
].p_hdr
->eh_entries
== 0 && i
> 0) {
2760 /* index is empty, remove it;
2761 * handle must be already prepared by the
2762 * truncatei_leaf() */
2763 err
= ext4_ext_rm_idx(handle
, inode
, path
+ i
);
2765 /* root level has p_bh == NULL, brelse() eats this */
2766 brelse(path
[i
].p_bh
);
2767 path
[i
].p_bh
= NULL
;
2769 ext_debug("return to level %d\n", i
);
2773 trace_ext4_ext_remove_space_done(inode
, start
, depth
, partial_cluster
,
2774 path
->p_hdr
->eh_entries
);
2776 /* If we still have something in the partial cluster and we have removed
2777 * even the first extent, then we should free the blocks in the partial
2778 * cluster as well. */
2779 if (partial_cluster
&& path
->p_hdr
->eh_entries
== 0) {
2780 int flags
= EXT4_FREE_BLOCKS_FORGET
;
2782 if (S_ISDIR(inode
->i_mode
) || S_ISLNK(inode
->i_mode
))
2783 flags
|= EXT4_FREE_BLOCKS_METADATA
;
2785 ext4_free_blocks(handle
, inode
, NULL
,
2786 EXT4_C2B(EXT4_SB(sb
), partial_cluster
),
2787 EXT4_SB(sb
)->s_cluster_ratio
, flags
);
2788 partial_cluster
= 0;
2791 /* TODO: flexible tree reduction should be here */
2792 if (path
->p_hdr
->eh_entries
== 0) {
2794 * truncate to zero freed all the tree,
2795 * so we need to correct eh_depth
2797 err
= ext4_ext_get_access(handle
, inode
, path
);
2799 ext_inode_hdr(inode
)->eh_depth
= 0;
2800 ext_inode_hdr(inode
)->eh_max
=
2801 cpu_to_le16(ext4_ext_space_root(inode
, 0));
2802 err
= ext4_ext_dirty(handle
, inode
, path
);
2806 ext4_ext_drop_refs(path
);
2808 if (err
== -EAGAIN
) {
2812 ext4_journal_stop(handle
);
2818 * called at mount time
2820 void ext4_ext_init(struct super_block
*sb
)
2823 * possible initialization would be here
2826 if (EXT4_HAS_INCOMPAT_FEATURE(sb
, EXT4_FEATURE_INCOMPAT_EXTENTS
)) {
2827 #if defined(AGGRESSIVE_TEST) || defined(CHECK_BINSEARCH) || defined(EXTENTS_STATS)
2828 printk(KERN_INFO
"EXT4-fs: file extents enabled"
2829 #ifdef AGGRESSIVE_TEST
2830 ", aggressive tests"
2832 #ifdef CHECK_BINSEARCH
2835 #ifdef EXTENTS_STATS
2840 #ifdef EXTENTS_STATS
2841 spin_lock_init(&EXT4_SB(sb
)->s_ext_stats_lock
);
2842 EXT4_SB(sb
)->s_ext_min
= 1 << 30;
2843 EXT4_SB(sb
)->s_ext_max
= 0;
2849 * called at umount time
2851 void ext4_ext_release(struct super_block
*sb
)
2853 if (!EXT4_HAS_INCOMPAT_FEATURE(sb
, EXT4_FEATURE_INCOMPAT_EXTENTS
))
2856 #ifdef EXTENTS_STATS
2857 if (EXT4_SB(sb
)->s_ext_blocks
&& EXT4_SB(sb
)->s_ext_extents
) {
2858 struct ext4_sb_info
*sbi
= EXT4_SB(sb
);
2859 printk(KERN_ERR
"EXT4-fs: %lu blocks in %lu extents (%lu ave)\n",
2860 sbi
->s_ext_blocks
, sbi
->s_ext_extents
,
2861 sbi
->s_ext_blocks
/ sbi
->s_ext_extents
);
2862 printk(KERN_ERR
"EXT4-fs: extents: %lu min, %lu max, max depth %lu\n",
2863 sbi
->s_ext_min
, sbi
->s_ext_max
, sbi
->s_depth_max
);
2868 /* FIXME!! we need to try to merge to left or right after zero-out */
2869 static int ext4_ext_zeroout(struct inode
*inode
, struct ext4_extent
*ex
)
2871 ext4_fsblk_t ee_pblock
;
2872 unsigned int ee_len
;
2875 ee_len
= ext4_ext_get_actual_len(ex
);
2876 ee_pblock
= ext4_ext_pblock(ex
);
2878 ret
= sb_issue_zeroout(inode
->i_sb
, ee_pblock
, ee_len
, GFP_NOFS
);
2886 * ext4_split_extent_at() splits an extent at given block.
2888 * @handle: the journal handle
2889 * @inode: the file inode
2890 * @path: the path to the extent
2891 * @split: the logical block where the extent is splitted.
2892 * @split_flags: indicates if the extent could be zeroout if split fails, and
2893 * the states(init or uninit) of new extents.
2894 * @flags: flags used to insert new extent to extent tree.
2897 * Splits extent [a, b] into two extents [a, @split) and [@split, b], states
2898 * of which are deterimined by split_flag.
2900 * There are two cases:
2901 * a> the extent are splitted into two extent.
2902 * b> split is not needed, and just mark the extent.
2904 * return 0 on success.
2906 static int ext4_split_extent_at(handle_t
*handle
,
2907 struct inode
*inode
,
2908 struct ext4_ext_path
*path
,
2913 ext4_fsblk_t newblock
;
2914 ext4_lblk_t ee_block
;
2915 struct ext4_extent
*ex
, newex
, orig_ex
;
2916 struct ext4_extent
*ex2
= NULL
;
2917 unsigned int ee_len
, depth
;
2920 BUG_ON((split_flag
& (EXT4_EXT_DATA_VALID1
| EXT4_EXT_DATA_VALID2
)) ==
2921 (EXT4_EXT_DATA_VALID1
| EXT4_EXT_DATA_VALID2
));
2923 ext_debug("ext4_split_extents_at: inode %lu, logical"
2924 "block %llu\n", inode
->i_ino
, (unsigned long long)split
);
2926 ext4_ext_show_leaf(inode
, path
);
2928 depth
= ext_depth(inode
);
2929 ex
= path
[depth
].p_ext
;
2930 ee_block
= le32_to_cpu(ex
->ee_block
);
2931 ee_len
= ext4_ext_get_actual_len(ex
);
2932 newblock
= split
- ee_block
+ ext4_ext_pblock(ex
);
2934 BUG_ON(split
< ee_block
|| split
>= (ee_block
+ ee_len
));
2936 err
= ext4_ext_get_access(handle
, inode
, path
+ depth
);
2940 if (split
== ee_block
) {
2942 * case b: block @split is the block that the extent begins with
2943 * then we just change the state of the extent, and splitting
2946 if (split_flag
& EXT4_EXT_MARK_UNINIT2
)
2947 ext4_ext_mark_uninitialized(ex
);
2949 ext4_ext_mark_initialized(ex
);
2951 if (!(flags
& EXT4_GET_BLOCKS_PRE_IO
))
2952 ext4_ext_try_to_merge(handle
, inode
, path
, ex
);
2954 err
= ext4_ext_dirty(handle
, inode
, path
+ path
->p_depth
);
2959 memcpy(&orig_ex
, ex
, sizeof(orig_ex
));
2960 ex
->ee_len
= cpu_to_le16(split
- ee_block
);
2961 if (split_flag
& EXT4_EXT_MARK_UNINIT1
)
2962 ext4_ext_mark_uninitialized(ex
);
2965 * path may lead to new leaf, not to original leaf any more
2966 * after ext4_ext_insert_extent() returns,
2968 err
= ext4_ext_dirty(handle
, inode
, path
+ depth
);
2970 goto fix_extent_len
;
2973 ex2
->ee_block
= cpu_to_le32(split
);
2974 ex2
->ee_len
= cpu_to_le16(ee_len
- (split
- ee_block
));
2975 ext4_ext_store_pblock(ex2
, newblock
);
2976 if (split_flag
& EXT4_EXT_MARK_UNINIT2
)
2977 ext4_ext_mark_uninitialized(ex2
);
2979 err
= ext4_ext_insert_extent(handle
, inode
, path
, &newex
, flags
);
2980 if (err
== -ENOSPC
&& (EXT4_EXT_MAY_ZEROOUT
& split_flag
)) {
2981 if (split_flag
& (EXT4_EXT_DATA_VALID1
|EXT4_EXT_DATA_VALID2
)) {
2982 if (split_flag
& EXT4_EXT_DATA_VALID1
)
2983 err
= ext4_ext_zeroout(inode
, ex2
);
2985 err
= ext4_ext_zeroout(inode
, ex
);
2987 err
= ext4_ext_zeroout(inode
, &orig_ex
);
2990 goto fix_extent_len
;
2991 /* update the extent length and mark as initialized */
2992 ex
->ee_len
= cpu_to_le16(ee_len
);
2993 ext4_ext_try_to_merge(handle
, inode
, path
, ex
);
2994 err
= ext4_ext_dirty(handle
, inode
, path
+ path
->p_depth
);
2997 goto fix_extent_len
;
3000 ext4_ext_show_leaf(inode
, path
);
3004 ex
->ee_len
= orig_ex
.ee_len
;
3005 ext4_ext_dirty(handle
, inode
, path
+ depth
);
3010 * ext4_split_extents() splits an extent and mark extent which is covered
3011 * by @map as split_flags indicates
3013 * It may result in splitting the extent into multiple extents (upto three)
3014 * There are three possibilities:
3015 * a> There is no split required
3016 * b> Splits in two extents: Split is happening at either end of the extent
3017 * c> Splits in three extents: Somone is splitting in middle of the extent
3020 static int ext4_split_extent(handle_t
*handle
,
3021 struct inode
*inode
,
3022 struct ext4_ext_path
*path
,
3023 struct ext4_map_blocks
*map
,
3027 ext4_lblk_t ee_block
;
3028 struct ext4_extent
*ex
;
3029 unsigned int ee_len
, depth
;
3032 int split_flag1
, flags1
;
3034 depth
= ext_depth(inode
);
3035 ex
= path
[depth
].p_ext
;
3036 ee_block
= le32_to_cpu(ex
->ee_block
);
3037 ee_len
= ext4_ext_get_actual_len(ex
);
3038 uninitialized
= ext4_ext_is_uninitialized(ex
);
3040 if (map
->m_lblk
+ map
->m_len
< ee_block
+ ee_len
) {
3041 split_flag1
= split_flag
& EXT4_EXT_MAY_ZEROOUT
;
3042 flags1
= flags
| EXT4_GET_BLOCKS_PRE_IO
;
3044 split_flag1
|= EXT4_EXT_MARK_UNINIT1
|
3045 EXT4_EXT_MARK_UNINIT2
;
3046 if (split_flag
& EXT4_EXT_DATA_VALID2
)
3047 split_flag1
|= EXT4_EXT_DATA_VALID1
;
3048 err
= ext4_split_extent_at(handle
, inode
, path
,
3049 map
->m_lblk
+ map
->m_len
, split_flag1
, flags1
);
3054 ext4_ext_drop_refs(path
);
3055 path
= ext4_ext_find_extent(inode
, map
->m_lblk
, path
);
3057 return PTR_ERR(path
);
3059 if (map
->m_lblk
>= ee_block
) {
3060 split_flag1
= split_flag
& (EXT4_EXT_MAY_ZEROOUT
|
3061 EXT4_EXT_DATA_VALID2
);
3063 split_flag1
|= EXT4_EXT_MARK_UNINIT1
;
3064 if (split_flag
& EXT4_EXT_MARK_UNINIT2
)
3065 split_flag1
|= EXT4_EXT_MARK_UNINIT2
;
3066 err
= ext4_split_extent_at(handle
, inode
, path
,
3067 map
->m_lblk
, split_flag1
, flags
);
3072 ext4_ext_show_leaf(inode
, path
);
3074 return err
? err
: map
->m_len
;
3078 * This function is called by ext4_ext_map_blocks() if someone tries to write
3079 * to an uninitialized extent. It may result in splitting the uninitialized
3080 * extent into multiple extents (up to three - one initialized and two
3082 * There are three possibilities:
3083 * a> There is no split required: Entire extent should be initialized
3084 * b> Splits in two extents: Write is happening at either end of the extent
3085 * c> Splits in three extents: Somone is writing in middle of the extent
3088 * - The extent pointed to by 'path' is uninitialized.
3089 * - The extent pointed to by 'path' contains a superset
3090 * of the logical span [map->m_lblk, map->m_lblk + map->m_len).
3092 * Post-conditions on success:
3093 * - the returned value is the number of blocks beyond map->l_lblk
3094 * that are allocated and initialized.
3095 * It is guaranteed to be >= map->m_len.
3097 static int ext4_ext_convert_to_initialized(handle_t
*handle
,
3098 struct inode
*inode
,
3099 struct ext4_map_blocks
*map
,
3100 struct ext4_ext_path
*path
)
3102 struct ext4_sb_info
*sbi
;
3103 struct ext4_extent_header
*eh
;
3104 struct ext4_map_blocks split_map
;
3105 struct ext4_extent zero_ex
;
3106 struct ext4_extent
*ex
;
3107 ext4_lblk_t ee_block
, eof_block
;
3108 unsigned int ee_len
, depth
;
3109 int allocated
, max_zeroout
= 0;
3113 ext_debug("ext4_ext_convert_to_initialized: inode %lu, logical"
3114 "block %llu, max_blocks %u\n", inode
->i_ino
,
3115 (unsigned long long)map
->m_lblk
, map
->m_len
);
3117 sbi
= EXT4_SB(inode
->i_sb
);
3118 eof_block
= (inode
->i_size
+ inode
->i_sb
->s_blocksize
- 1) >>
3119 inode
->i_sb
->s_blocksize_bits
;
3120 if (eof_block
< map
->m_lblk
+ map
->m_len
)
3121 eof_block
= map
->m_lblk
+ map
->m_len
;
3123 depth
= ext_depth(inode
);
3124 eh
= path
[depth
].p_hdr
;
3125 ex
= path
[depth
].p_ext
;
3126 ee_block
= le32_to_cpu(ex
->ee_block
);
3127 ee_len
= ext4_ext_get_actual_len(ex
);
3128 allocated
= ee_len
- (map
->m_lblk
- ee_block
);
3130 trace_ext4_ext_convert_to_initialized_enter(inode
, map
, ex
);
3132 /* Pre-conditions */
3133 BUG_ON(!ext4_ext_is_uninitialized(ex
));
3134 BUG_ON(!in_range(map
->m_lblk
, ee_block
, ee_len
));
3137 * Attempt to transfer newly initialized blocks from the currently
3138 * uninitialized extent to its left neighbor. This is much cheaper
3139 * than an insertion followed by a merge as those involve costly
3140 * memmove() calls. This is the common case in steady state for
3141 * workloads doing fallocate(FALLOC_FL_KEEP_SIZE) followed by append
3144 * Limitations of the current logic:
3145 * - L1: we only deal with writes at the start of the extent.
3146 * The approach could be extended to writes at the end
3147 * of the extent but this scenario was deemed less common.
3148 * - L2: we do not deal with writes covering the whole extent.
3149 * This would require removing the extent if the transfer
3151 * - L3: we only attempt to merge with an extent stored in the
3152 * same extent tree node.
3154 if ((map
->m_lblk
== ee_block
) && /*L1*/
3155 (map
->m_len
< ee_len
) && /*L2*/
3156 (ex
> EXT_FIRST_EXTENT(eh
))) { /*L3*/
3157 struct ext4_extent
*prev_ex
;
3158 ext4_lblk_t prev_lblk
;
3159 ext4_fsblk_t prev_pblk
, ee_pblk
;
3160 unsigned int prev_len
, write_len
;
3163 prev_lblk
= le32_to_cpu(prev_ex
->ee_block
);
3164 prev_len
= ext4_ext_get_actual_len(prev_ex
);
3165 prev_pblk
= ext4_ext_pblock(prev_ex
);
3166 ee_pblk
= ext4_ext_pblock(ex
);
3167 write_len
= map
->m_len
;
3170 * A transfer of blocks from 'ex' to 'prev_ex' is allowed
3171 * upon those conditions:
3172 * - C1: prev_ex is initialized,
3173 * - C2: prev_ex is logically abutting ex,
3174 * - C3: prev_ex is physically abutting ex,
3175 * - C4: prev_ex can receive the additional blocks without
3176 * overflowing the (initialized) length limit.
3178 if ((!ext4_ext_is_uninitialized(prev_ex
)) && /*C1*/
3179 ((prev_lblk
+ prev_len
) == ee_block
) && /*C2*/
3180 ((prev_pblk
+ prev_len
) == ee_pblk
) && /*C3*/
3181 (prev_len
< (EXT_INIT_MAX_LEN
- write_len
))) { /*C4*/
3182 err
= ext4_ext_get_access(handle
, inode
, path
+ depth
);
3186 trace_ext4_ext_convert_to_initialized_fastpath(inode
,
3189 /* Shift the start of ex by 'write_len' blocks */
3190 ex
->ee_block
= cpu_to_le32(ee_block
+ write_len
);
3191 ext4_ext_store_pblock(ex
, ee_pblk
+ write_len
);
3192 ex
->ee_len
= cpu_to_le16(ee_len
- write_len
);
3193 ext4_ext_mark_uninitialized(ex
); /* Restore the flag */
3195 /* Extend prev_ex by 'write_len' blocks */
3196 prev_ex
->ee_len
= cpu_to_le16(prev_len
+ write_len
);
3198 /* Mark the block containing both extents as dirty */
3199 ext4_ext_dirty(handle
, inode
, path
+ depth
);
3201 /* Update path to point to the right extent */
3202 path
[depth
].p_ext
= prev_ex
;
3204 /* Result: number of initialized blocks past m_lblk */
3205 allocated
= write_len
;
3210 WARN_ON(map
->m_lblk
< ee_block
);
3212 * It is safe to convert extent to initialized via explicit
3213 * zeroout only if extent is fully insde i_size or new_size.
3215 split_flag
|= ee_block
+ ee_len
<= eof_block
? EXT4_EXT_MAY_ZEROOUT
: 0;
3217 if (EXT4_EXT_MAY_ZEROOUT
& split_flag
)
3218 max_zeroout
= sbi
->s_extent_max_zeroout_kb
>>
3219 inode
->i_sb
->s_blocksize_bits
;
3221 /* If extent is less than s_max_zeroout_kb, zeroout directly */
3222 if (max_zeroout
&& (ee_len
<= max_zeroout
)) {
3223 err
= ext4_ext_zeroout(inode
, ex
);
3227 err
= ext4_ext_get_access(handle
, inode
, path
+ depth
);
3230 ext4_ext_mark_initialized(ex
);
3231 ext4_ext_try_to_merge(handle
, inode
, path
, ex
);
3232 err
= ext4_ext_dirty(handle
, inode
, path
+ path
->p_depth
);
3238 * 1. split the extent into three extents.
3239 * 2. split the extent into two extents, zeroout the first half.
3240 * 3. split the extent into two extents, zeroout the second half.
3241 * 4. split the extent into two extents with out zeroout.
3243 split_map
.m_lblk
= map
->m_lblk
;
3244 split_map
.m_len
= map
->m_len
;
3246 if (max_zeroout
&& (allocated
> map
->m_len
)) {
3247 if (allocated
<= max_zeroout
) {
3250 cpu_to_le32(map
->m_lblk
);
3251 zero_ex
.ee_len
= cpu_to_le16(allocated
);
3252 ext4_ext_store_pblock(&zero_ex
,
3253 ext4_ext_pblock(ex
) + map
->m_lblk
- ee_block
);
3254 err
= ext4_ext_zeroout(inode
, &zero_ex
);
3257 split_map
.m_lblk
= map
->m_lblk
;
3258 split_map
.m_len
= allocated
;
3259 } else if (map
->m_lblk
- ee_block
+ map
->m_len
< max_zeroout
) {
3261 if (map
->m_lblk
!= ee_block
) {
3262 zero_ex
.ee_block
= ex
->ee_block
;
3263 zero_ex
.ee_len
= cpu_to_le16(map
->m_lblk
-
3265 ext4_ext_store_pblock(&zero_ex
,
3266 ext4_ext_pblock(ex
));
3267 err
= ext4_ext_zeroout(inode
, &zero_ex
);
3272 split_map
.m_lblk
= ee_block
;
3273 split_map
.m_len
= map
->m_lblk
- ee_block
+ map
->m_len
;
3274 allocated
= map
->m_len
;
3278 allocated
= ext4_split_extent(handle
, inode
, path
,
3279 &split_map
, split_flag
, 0);
3284 return err
? err
: allocated
;
3288 * This function is called by ext4_ext_map_blocks() from
3289 * ext4_get_blocks_dio_write() when DIO to write
3290 * to an uninitialized extent.
3292 * Writing to an uninitialized extent may result in splitting the uninitialized
3293 * extent into multiple initialized/uninitialized extents (up to three)
3294 * There are three possibilities:
3295 * a> There is no split required: Entire extent should be uninitialized
3296 * b> Splits in two extents: Write is happening at either end of the extent
3297 * c> Splits in three extents: Somone is writing in middle of the extent
3299 * One of more index blocks maybe needed if the extent tree grow after
3300 * the uninitialized extent split. To prevent ENOSPC occur at the IO
3301 * complete, we need to split the uninitialized extent before DIO submit
3302 * the IO. The uninitialized extent called at this time will be split
3303 * into three uninitialized extent(at most). After IO complete, the part
3304 * being filled will be convert to initialized by the end_io callback function
3305 * via ext4_convert_unwritten_extents().
3307 * Returns the size of uninitialized extent to be written on success.
3309 static int ext4_split_unwritten_extents(handle_t
*handle
,
3310 struct inode
*inode
,
3311 struct ext4_map_blocks
*map
,
3312 struct ext4_ext_path
*path
,
3315 ext4_lblk_t eof_block
;
3316 ext4_lblk_t ee_block
;
3317 struct ext4_extent
*ex
;
3318 unsigned int ee_len
;
3319 int split_flag
= 0, depth
;
3321 ext_debug("ext4_split_unwritten_extents: inode %lu, logical"
3322 "block %llu, max_blocks %u\n", inode
->i_ino
,
3323 (unsigned long long)map
->m_lblk
, map
->m_len
);
3325 eof_block
= (inode
->i_size
+ inode
->i_sb
->s_blocksize
- 1) >>
3326 inode
->i_sb
->s_blocksize_bits
;
3327 if (eof_block
< map
->m_lblk
+ map
->m_len
)
3328 eof_block
= map
->m_lblk
+ map
->m_len
;
3330 * It is safe to convert extent to initialized via explicit
3331 * zeroout only if extent is fully insde i_size or new_size.
3333 depth
= ext_depth(inode
);
3334 ex
= path
[depth
].p_ext
;
3335 ee_block
= le32_to_cpu(ex
->ee_block
);
3336 ee_len
= ext4_ext_get_actual_len(ex
);
3338 split_flag
|= ee_block
+ ee_len
<= eof_block
? EXT4_EXT_MAY_ZEROOUT
: 0;
3339 split_flag
|= EXT4_EXT_MARK_UNINIT2
;
3340 if (flags
& EXT4_GET_BLOCKS_CONVERT
)
3341 split_flag
|= EXT4_EXT_DATA_VALID2
;
3342 flags
|= EXT4_GET_BLOCKS_PRE_IO
;
3343 return ext4_split_extent(handle
, inode
, path
, map
, split_flag
, flags
);
3346 static int ext4_convert_unwritten_extents_endio(handle_t
*handle
,
3347 struct inode
*inode
,
3348 struct ext4_map_blocks
*map
,
3349 struct ext4_ext_path
*path
)
3351 struct ext4_extent
*ex
;
3352 ext4_lblk_t ee_block
;
3353 unsigned int ee_len
;
3357 depth
= ext_depth(inode
);
3358 ex
= path
[depth
].p_ext
;
3359 ee_block
= le32_to_cpu(ex
->ee_block
);
3360 ee_len
= ext4_ext_get_actual_len(ex
);
3362 ext_debug("ext4_convert_unwritten_extents_endio: inode %lu, logical"
3363 "block %llu, max_blocks %u\n", inode
->i_ino
,
3364 (unsigned long long)ee_block
, ee_len
);
3366 /* If extent is larger than requested then split is required */
3367 if (ee_block
!= map
->m_lblk
|| ee_len
> map
->m_len
) {
3368 err
= ext4_split_unwritten_extents(handle
, inode
, map
, path
,
3369 EXT4_GET_BLOCKS_CONVERT
);
3372 ext4_ext_drop_refs(path
);
3373 path
= ext4_ext_find_extent(inode
, map
->m_lblk
, path
);
3375 err
= PTR_ERR(path
);
3378 depth
= ext_depth(inode
);
3379 ex
= path
[depth
].p_ext
;
3382 err
= ext4_ext_get_access(handle
, inode
, path
+ depth
);
3385 /* first mark the extent as initialized */
3386 ext4_ext_mark_initialized(ex
);
3388 /* note: ext4_ext_correct_indexes() isn't needed here because
3389 * borders are not changed
3391 ext4_ext_try_to_merge(handle
, inode
, path
, ex
);
3393 /* Mark modified extent as dirty */
3394 err
= ext4_ext_dirty(handle
, inode
, path
+ path
->p_depth
);
3396 ext4_ext_show_leaf(inode
, path
);
3400 static void unmap_underlying_metadata_blocks(struct block_device
*bdev
,
3401 sector_t block
, int count
)
3404 for (i
= 0; i
< count
; i
++)
3405 unmap_underlying_metadata(bdev
, block
+ i
);
3409 * Handle EOFBLOCKS_FL flag, clearing it if necessary
3411 static int check_eofblocks_fl(handle_t
*handle
, struct inode
*inode
,
3413 struct ext4_ext_path
*path
,
3417 struct ext4_extent_header
*eh
;
3418 struct ext4_extent
*last_ex
;
3420 if (!ext4_test_inode_flag(inode
, EXT4_INODE_EOFBLOCKS
))
3423 depth
= ext_depth(inode
);
3424 eh
= path
[depth
].p_hdr
;
3427 * We're going to remove EOFBLOCKS_FL entirely in future so we
3428 * do not care for this case anymore. Simply remove the flag
3429 * if there are no extents.
3431 if (unlikely(!eh
->eh_entries
))
3433 last_ex
= EXT_LAST_EXTENT(eh
);
3435 * We should clear the EOFBLOCKS_FL flag if we are writing the
3436 * last block in the last extent in the file. We test this by
3437 * first checking to see if the caller to
3438 * ext4_ext_get_blocks() was interested in the last block (or
3439 * a block beyond the last block) in the current extent. If
3440 * this turns out to be false, we can bail out from this
3441 * function immediately.
3443 if (lblk
+ len
< le32_to_cpu(last_ex
->ee_block
) +
3444 ext4_ext_get_actual_len(last_ex
))
3447 * If the caller does appear to be planning to write at or
3448 * beyond the end of the current extent, we then test to see
3449 * if the current extent is the last extent in the file, by
3450 * checking to make sure it was reached via the rightmost node
3451 * at each level of the tree.
3453 for (i
= depth
-1; i
>= 0; i
--)
3454 if (path
[i
].p_idx
!= EXT_LAST_INDEX(path
[i
].p_hdr
))
3457 ext4_clear_inode_flag(inode
, EXT4_INODE_EOFBLOCKS
);
3458 return ext4_mark_inode_dirty(handle
, inode
);
3462 * ext4_find_delalloc_range: find delayed allocated block in the given range.
3464 * Goes through the buffer heads in the range [lblk_start, lblk_end] and returns
3465 * whether there are any buffers marked for delayed allocation. It returns '1'
3466 * on the first delalloc'ed buffer head found. If no buffer head in the given
3467 * range is marked for delalloc, it returns 0.
3468 * lblk_start should always be <= lblk_end.
3469 * search_hint_reverse is to indicate that searching in reverse from lblk_end to
3470 * lblk_start might be more efficient (i.e., we will likely hit the delalloc'ed
3471 * block sooner). This is useful when blocks are truncated sequentially from
3472 * lblk_start towards lblk_end.
3474 static int ext4_find_delalloc_range(struct inode
*inode
,
3475 ext4_lblk_t lblk_start
,
3476 ext4_lblk_t lblk_end
,
3477 int search_hint_reverse
)
3479 struct address_space
*mapping
= inode
->i_mapping
;
3480 struct buffer_head
*head
, *bh
= NULL
;
3482 ext4_lblk_t i
, pg_lblk
;
3485 if (!test_opt(inode
->i_sb
, DELALLOC
))
3488 /* reverse search wont work if fs block size is less than page size */
3489 if (inode
->i_blkbits
< PAGE_CACHE_SHIFT
)
3490 search_hint_reverse
= 0;
3492 if (search_hint_reverse
)
3497 index
= i
>> (PAGE_CACHE_SHIFT
- inode
->i_blkbits
);
3499 while ((i
>= lblk_start
) && (i
<= lblk_end
)) {
3500 page
= find_get_page(mapping
, index
);
3504 if (!page_has_buffers(page
))
3507 head
= page_buffers(page
);
3512 pg_lblk
= index
<< (PAGE_CACHE_SHIFT
-
3515 if (unlikely(pg_lblk
< lblk_start
)) {
3517 * This is possible when fs block size is less
3518 * than page size and our cluster starts/ends in
3519 * middle of the page. So we need to skip the
3520 * initial few blocks till we reach the 'lblk'
3526 /* Check if the buffer is delayed allocated and that it
3527 * is not yet mapped. (when da-buffers are mapped during
3528 * their writeout, their da_mapped bit is set.)
3530 if (buffer_delay(bh
) && !buffer_da_mapped(bh
)) {
3531 page_cache_release(page
);
3532 trace_ext4_find_delalloc_range(inode
,
3533 lblk_start
, lblk_end
,
3534 search_hint_reverse
,
3538 if (search_hint_reverse
)
3542 } while ((i
>= lblk_start
) && (i
<= lblk_end
) &&
3543 ((bh
= bh
->b_this_page
) != head
));
3546 page_cache_release(page
);
3548 * Move to next page. 'i' will be the first lblk in the next
3551 if (search_hint_reverse
)
3555 i
= index
<< (PAGE_CACHE_SHIFT
- inode
->i_blkbits
);
3558 trace_ext4_find_delalloc_range(inode
, lblk_start
, lblk_end
,
3559 search_hint_reverse
, 0, 0);
3563 int ext4_find_delalloc_cluster(struct inode
*inode
, ext4_lblk_t lblk
,
3564 int search_hint_reverse
)
3566 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
3567 ext4_lblk_t lblk_start
, lblk_end
;
3568 lblk_start
= lblk
& (~(sbi
->s_cluster_ratio
- 1));
3569 lblk_end
= lblk_start
+ sbi
->s_cluster_ratio
- 1;
3571 return ext4_find_delalloc_range(inode
, lblk_start
, lblk_end
,
3572 search_hint_reverse
);
3576 * Determines how many complete clusters (out of those specified by the 'map')
3577 * are under delalloc and were reserved quota for.
3578 * This function is called when we are writing out the blocks that were
3579 * originally written with their allocation delayed, but then the space was
3580 * allocated using fallocate() before the delayed allocation could be resolved.
3581 * The cases to look for are:
3582 * ('=' indicated delayed allocated blocks
3583 * '-' indicates non-delayed allocated blocks)
3584 * (a) partial clusters towards beginning and/or end outside of allocated range
3585 * are not delalloc'ed.
3587 * |----c---=|====c====|====c====|===-c----|
3588 * |++++++ allocated ++++++|
3589 * ==> 4 complete clusters in above example
3591 * (b) partial cluster (outside of allocated range) towards either end is
3592 * marked for delayed allocation. In this case, we will exclude that
3595 * |----====c========|========c========|
3596 * |++++++ allocated ++++++|
3597 * ==> 1 complete clusters in above example
3600 * |================c================|
3601 * |++++++ allocated ++++++|
3602 * ==> 0 complete clusters in above example
3604 * The ext4_da_update_reserve_space will be called only if we
3605 * determine here that there were some "entire" clusters that span
3606 * this 'allocated' range.
3607 * In the non-bigalloc case, this function will just end up returning num_blks
3608 * without ever calling ext4_find_delalloc_range.
3611 get_reserved_cluster_alloc(struct inode
*inode
, ext4_lblk_t lblk_start
,
3612 unsigned int num_blks
)
3614 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
3615 ext4_lblk_t alloc_cluster_start
, alloc_cluster_end
;
3616 ext4_lblk_t lblk_from
, lblk_to
, c_offset
;
3617 unsigned int allocated_clusters
= 0;
3619 alloc_cluster_start
= EXT4_B2C(sbi
, lblk_start
);
3620 alloc_cluster_end
= EXT4_B2C(sbi
, lblk_start
+ num_blks
- 1);
3622 /* max possible clusters for this allocation */
3623 allocated_clusters
= alloc_cluster_end
- alloc_cluster_start
+ 1;
3625 trace_ext4_get_reserved_cluster_alloc(inode
, lblk_start
, num_blks
);
3627 /* Check towards left side */
3628 c_offset
= lblk_start
& (sbi
->s_cluster_ratio
- 1);
3630 lblk_from
= lblk_start
& (~(sbi
->s_cluster_ratio
- 1));
3631 lblk_to
= lblk_from
+ c_offset
- 1;
3633 if (ext4_find_delalloc_range(inode
, lblk_from
, lblk_to
, 0))
3634 allocated_clusters
--;
3637 /* Now check towards right. */
3638 c_offset
= (lblk_start
+ num_blks
) & (sbi
->s_cluster_ratio
- 1);
3639 if (allocated_clusters
&& c_offset
) {
3640 lblk_from
= lblk_start
+ num_blks
;
3641 lblk_to
= lblk_from
+ (sbi
->s_cluster_ratio
- c_offset
) - 1;
3643 if (ext4_find_delalloc_range(inode
, lblk_from
, lblk_to
, 0))
3644 allocated_clusters
--;
3647 return allocated_clusters
;
3651 ext4_ext_handle_uninitialized_extents(handle_t
*handle
, struct inode
*inode
,
3652 struct ext4_map_blocks
*map
,
3653 struct ext4_ext_path
*path
, int flags
,
3654 unsigned int allocated
, ext4_fsblk_t newblock
)
3658 ext4_io_end_t
*io
= ext4_inode_aio(inode
);
3660 ext_debug("ext4_ext_handle_uninitialized_extents: inode %lu, logical "
3661 "block %llu, max_blocks %u, flags %x, allocated %u\n",
3662 inode
->i_ino
, (unsigned long long)map
->m_lblk
, map
->m_len
,
3664 ext4_ext_show_leaf(inode
, path
);
3666 trace_ext4_ext_handle_uninitialized_extents(inode
, map
, allocated
,
3669 /* get_block() before submit the IO, split the extent */
3670 if ((flags
& EXT4_GET_BLOCKS_PRE_IO
)) {
3671 ret
= ext4_split_unwritten_extents(handle
, inode
, map
,
3676 * Flag the inode(non aio case) or end_io struct (aio case)
3677 * that this IO needs to conversion to written when IO is
3681 ext4_set_io_unwritten_flag(inode
, io
);
3683 ext4_set_inode_state(inode
, EXT4_STATE_DIO_UNWRITTEN
);
3684 if (ext4_should_dioread_nolock(inode
))
3685 map
->m_flags
|= EXT4_MAP_UNINIT
;
3688 /* IO end_io complete, convert the filled extent to written */
3689 if ((flags
& EXT4_GET_BLOCKS_CONVERT
)) {
3690 ret
= ext4_convert_unwritten_extents_endio(handle
, inode
, map
,
3693 ext4_update_inode_fsync_trans(handle
, inode
, 1);
3694 err
= check_eofblocks_fl(handle
, inode
, map
->m_lblk
,
3700 /* buffered IO case */
3702 * repeat fallocate creation request
3703 * we already have an unwritten extent
3705 if (flags
& EXT4_GET_BLOCKS_UNINIT_EXT
)
3708 /* buffered READ or buffered write_begin() lookup */
3709 if ((flags
& EXT4_GET_BLOCKS_CREATE
) == 0) {
3711 * We have blocks reserved already. We
3712 * return allocated blocks so that delalloc
3713 * won't do block reservation for us. But
3714 * the buffer head will be unmapped so that
3715 * a read from the block returns 0s.
3717 map
->m_flags
|= EXT4_MAP_UNWRITTEN
;
3721 /* buffered write, writepage time, convert*/
3722 ret
= ext4_ext_convert_to_initialized(handle
, inode
, map
, path
);
3724 ext4_update_inode_fsync_trans(handle
, inode
, 1);
3731 map
->m_flags
|= EXT4_MAP_NEW
;
3733 * if we allocated more blocks than requested
3734 * we need to make sure we unmap the extra block
3735 * allocated. The actual needed block will get
3736 * unmapped later when we find the buffer_head marked
3739 if (allocated
> map
->m_len
) {
3740 unmap_underlying_metadata_blocks(inode
->i_sb
->s_bdev
,
3741 newblock
+ map
->m_len
,
3742 allocated
- map
->m_len
);
3743 allocated
= map
->m_len
;
3747 * If we have done fallocate with the offset that is already
3748 * delayed allocated, we would have block reservation
3749 * and quota reservation done in the delayed write path.
3750 * But fallocate would have already updated quota and block
3751 * count for this offset. So cancel these reservation
3753 if (flags
& EXT4_GET_BLOCKS_DELALLOC_RESERVE
) {
3754 unsigned int reserved_clusters
;
3755 reserved_clusters
= get_reserved_cluster_alloc(inode
,
3756 map
->m_lblk
, map
->m_len
);
3757 if (reserved_clusters
)
3758 ext4_da_update_reserve_space(inode
,
3764 map
->m_flags
|= EXT4_MAP_MAPPED
;
3765 if ((flags
& EXT4_GET_BLOCKS_KEEP_SIZE
) == 0) {
3766 err
= check_eofblocks_fl(handle
, inode
, map
->m_lblk
, path
,
3772 if (allocated
> map
->m_len
)
3773 allocated
= map
->m_len
;
3774 ext4_ext_show_leaf(inode
, path
);
3775 map
->m_pblk
= newblock
;
3776 map
->m_len
= allocated
;
3779 ext4_ext_drop_refs(path
);
3782 return err
? err
: allocated
;
3786 * get_implied_cluster_alloc - check to see if the requested
3787 * allocation (in the map structure) overlaps with a cluster already
3788 * allocated in an extent.
3789 * @sb The filesystem superblock structure
3790 * @map The requested lblk->pblk mapping
3791 * @ex The extent structure which might contain an implied
3792 * cluster allocation
3794 * This function is called by ext4_ext_map_blocks() after we failed to
3795 * find blocks that were already in the inode's extent tree. Hence,
3796 * we know that the beginning of the requested region cannot overlap
3797 * the extent from the inode's extent tree. There are three cases we
3798 * want to catch. The first is this case:
3800 * |--- cluster # N--|
3801 * |--- extent ---| |---- requested region ---|
3804 * The second case that we need to test for is this one:
3806 * |--------- cluster # N ----------------|
3807 * |--- requested region --| |------- extent ----|
3808 * |=======================|
3810 * The third case is when the requested region lies between two extents
3811 * within the same cluster:
3812 * |------------- cluster # N-------------|
3813 * |----- ex -----| |---- ex_right ----|
3814 * |------ requested region ------|
3815 * |================|
3817 * In each of the above cases, we need to set the map->m_pblk and
3818 * map->m_len so it corresponds to the return the extent labelled as
3819 * "|====|" from cluster #N, since it is already in use for data in
3820 * cluster EXT4_B2C(sbi, map->m_lblk). We will then return 1 to
3821 * signal to ext4_ext_map_blocks() that map->m_pblk should be treated
3822 * as a new "allocated" block region. Otherwise, we will return 0 and
3823 * ext4_ext_map_blocks() will then allocate one or more new clusters
3824 * by calling ext4_mb_new_blocks().
3826 static int get_implied_cluster_alloc(struct super_block
*sb
,
3827 struct ext4_map_blocks
*map
,
3828 struct ext4_extent
*ex
,
3829 struct ext4_ext_path
*path
)
3831 struct ext4_sb_info
*sbi
= EXT4_SB(sb
);
3832 ext4_lblk_t c_offset
= map
->m_lblk
& (sbi
->s_cluster_ratio
-1);
3833 ext4_lblk_t ex_cluster_start
, ex_cluster_end
;
3834 ext4_lblk_t rr_cluster_start
;
3835 ext4_lblk_t ee_block
= le32_to_cpu(ex
->ee_block
);
3836 ext4_fsblk_t ee_start
= ext4_ext_pblock(ex
);
3837 unsigned short ee_len
= ext4_ext_get_actual_len(ex
);
3839 /* The extent passed in that we are trying to match */
3840 ex_cluster_start
= EXT4_B2C(sbi
, ee_block
);
3841 ex_cluster_end
= EXT4_B2C(sbi
, ee_block
+ ee_len
- 1);
3843 /* The requested region passed into ext4_map_blocks() */
3844 rr_cluster_start
= EXT4_B2C(sbi
, map
->m_lblk
);
3846 if ((rr_cluster_start
== ex_cluster_end
) ||
3847 (rr_cluster_start
== ex_cluster_start
)) {
3848 if (rr_cluster_start
== ex_cluster_end
)
3849 ee_start
+= ee_len
- 1;
3850 map
->m_pblk
= (ee_start
& ~(sbi
->s_cluster_ratio
- 1)) +
3852 map
->m_len
= min(map
->m_len
,
3853 (unsigned) sbi
->s_cluster_ratio
- c_offset
);
3855 * Check for and handle this case:
3857 * |--------- cluster # N-------------|
3858 * |------- extent ----|
3859 * |--- requested region ---|
3863 if (map
->m_lblk
< ee_block
)
3864 map
->m_len
= min(map
->m_len
, ee_block
- map
->m_lblk
);
3867 * Check for the case where there is already another allocated
3868 * block to the right of 'ex' but before the end of the cluster.
3870 * |------------- cluster # N-------------|
3871 * |----- ex -----| |---- ex_right ----|
3872 * |------ requested region ------|
3873 * |================|
3875 if (map
->m_lblk
> ee_block
) {
3876 ext4_lblk_t next
= ext4_ext_next_allocated_block(path
);
3877 map
->m_len
= min(map
->m_len
, next
- map
->m_lblk
);
3880 trace_ext4_get_implied_cluster_alloc_exit(sb
, map
, 1);
3884 trace_ext4_get_implied_cluster_alloc_exit(sb
, map
, 0);
3890 * Block allocation/map/preallocation routine for extents based files
3893 * Need to be called with
3894 * down_read(&EXT4_I(inode)->i_data_sem) if not allocating file system block
3895 * (ie, create is zero). Otherwise down_write(&EXT4_I(inode)->i_data_sem)
3897 * return > 0, number of of blocks already mapped/allocated
3898 * if create == 0 and these are pre-allocated blocks
3899 * buffer head is unmapped
3900 * otherwise blocks are mapped
3902 * return = 0, if plain look up failed (blocks have not been allocated)
3903 * buffer head is unmapped
3905 * return < 0, error case.
3907 int ext4_ext_map_blocks(handle_t
*handle
, struct inode
*inode
,
3908 struct ext4_map_blocks
*map
, int flags
)
3910 struct ext4_ext_path
*path
= NULL
;
3911 struct ext4_extent newex
, *ex
, *ex2
;
3912 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
3913 ext4_fsblk_t newblock
= 0;
3914 int free_on_err
= 0, err
= 0, depth
, ret
;
3915 unsigned int allocated
= 0, offset
= 0;
3916 unsigned int allocated_clusters
= 0;
3917 struct ext4_allocation_request ar
;
3918 ext4_io_end_t
*io
= ext4_inode_aio(inode
);
3919 ext4_lblk_t cluster_offset
;
3920 int set_unwritten
= 0;
3922 ext_debug("blocks %u/%u requested for inode %lu\n",
3923 map
->m_lblk
, map
->m_len
, inode
->i_ino
);
3924 trace_ext4_ext_map_blocks_enter(inode
, map
->m_lblk
, map
->m_len
, flags
);
3926 /* check in cache */
3927 if (ext4_ext_in_cache(inode
, map
->m_lblk
, &newex
)) {
3928 if (!newex
.ee_start_lo
&& !newex
.ee_start_hi
) {
3929 if ((sbi
->s_cluster_ratio
> 1) &&
3930 ext4_find_delalloc_cluster(inode
, map
->m_lblk
, 0))
3931 map
->m_flags
|= EXT4_MAP_FROM_CLUSTER
;
3933 if ((flags
& EXT4_GET_BLOCKS_CREATE
) == 0) {
3935 * block isn't allocated yet and
3936 * user doesn't want to allocate it
3940 /* we should allocate requested block */
3942 /* block is already allocated */
3943 if (sbi
->s_cluster_ratio
> 1)
3944 map
->m_flags
|= EXT4_MAP_FROM_CLUSTER
;
3945 newblock
= map
->m_lblk
3946 - le32_to_cpu(newex
.ee_block
)
3947 + ext4_ext_pblock(&newex
);
3948 /* number of remaining blocks in the extent */
3949 allocated
= ext4_ext_get_actual_len(&newex
) -
3950 (map
->m_lblk
- le32_to_cpu(newex
.ee_block
));
3955 /* find extent for this block */
3956 path
= ext4_ext_find_extent(inode
, map
->m_lblk
, NULL
);
3958 err
= PTR_ERR(path
);
3963 depth
= ext_depth(inode
);
3966 * consistent leaf must not be empty;
3967 * this situation is possible, though, _during_ tree modification;
3968 * this is why assert can't be put in ext4_ext_find_extent()
3970 if (unlikely(path
[depth
].p_ext
== NULL
&& depth
!= 0)) {
3971 EXT4_ERROR_INODE(inode
, "bad extent address "
3972 "lblock: %lu, depth: %d pblock %lld",
3973 (unsigned long) map
->m_lblk
, depth
,
3974 path
[depth
].p_block
);
3979 ex
= path
[depth
].p_ext
;
3981 ext4_lblk_t ee_block
= le32_to_cpu(ex
->ee_block
);
3982 ext4_fsblk_t ee_start
= ext4_ext_pblock(ex
);
3983 unsigned short ee_len
;
3986 * Uninitialized extents are treated as holes, except that
3987 * we split out initialized portions during a write.
3989 ee_len
= ext4_ext_get_actual_len(ex
);
3991 trace_ext4_ext_show_extent(inode
, ee_block
, ee_start
, ee_len
);
3993 /* if found extent covers block, simply return it */
3994 if (in_range(map
->m_lblk
, ee_block
, ee_len
)) {
3995 newblock
= map
->m_lblk
- ee_block
+ ee_start
;
3996 /* number of remaining blocks in the extent */
3997 allocated
= ee_len
- (map
->m_lblk
- ee_block
);
3998 ext_debug("%u fit into %u:%d -> %llu\n", map
->m_lblk
,
3999 ee_block
, ee_len
, newblock
);
4002 * Do not put uninitialized extent
4005 if (!ext4_ext_is_uninitialized(ex
)) {
4006 ext4_ext_put_in_cache(inode
, ee_block
,
4010 ret
= ext4_ext_handle_uninitialized_extents(
4011 handle
, inode
, map
, path
, flags
,
4012 allocated
, newblock
);
4017 if ((sbi
->s_cluster_ratio
> 1) &&
4018 ext4_find_delalloc_cluster(inode
, map
->m_lblk
, 0))
4019 map
->m_flags
|= EXT4_MAP_FROM_CLUSTER
;
4022 * requested block isn't allocated yet;
4023 * we couldn't try to create block if create flag is zero
4025 if ((flags
& EXT4_GET_BLOCKS_CREATE
) == 0) {
4027 * put just found gap into cache to speed up
4028 * subsequent requests
4030 ext4_ext_put_gap_in_cache(inode
, path
, map
->m_lblk
);
4035 * Okay, we need to do block allocation.
4037 map
->m_flags
&= ~EXT4_MAP_FROM_CLUSTER
;
4038 newex
.ee_block
= cpu_to_le32(map
->m_lblk
);
4039 cluster_offset
= map
->m_lblk
& (sbi
->s_cluster_ratio
-1);
4042 * If we are doing bigalloc, check to see if the extent returned
4043 * by ext4_ext_find_extent() implies a cluster we can use.
4045 if (cluster_offset
&& ex
&&
4046 get_implied_cluster_alloc(inode
->i_sb
, map
, ex
, path
)) {
4047 ar
.len
= allocated
= map
->m_len
;
4048 newblock
= map
->m_pblk
;
4049 map
->m_flags
|= EXT4_MAP_FROM_CLUSTER
;
4050 goto got_allocated_blocks
;
4053 /* find neighbour allocated blocks */
4054 ar
.lleft
= map
->m_lblk
;
4055 err
= ext4_ext_search_left(inode
, path
, &ar
.lleft
, &ar
.pleft
);
4058 ar
.lright
= map
->m_lblk
;
4060 err
= ext4_ext_search_right(inode
, path
, &ar
.lright
, &ar
.pright
, &ex2
);
4064 /* Check if the extent after searching to the right implies a
4065 * cluster we can use. */
4066 if ((sbi
->s_cluster_ratio
> 1) && ex2
&&
4067 get_implied_cluster_alloc(inode
->i_sb
, map
, ex2
, path
)) {
4068 ar
.len
= allocated
= map
->m_len
;
4069 newblock
= map
->m_pblk
;
4070 map
->m_flags
|= EXT4_MAP_FROM_CLUSTER
;
4071 goto got_allocated_blocks
;
4075 * See if request is beyond maximum number of blocks we can have in
4076 * a single extent. For an initialized extent this limit is
4077 * EXT_INIT_MAX_LEN and for an uninitialized extent this limit is
4078 * EXT_UNINIT_MAX_LEN.
4080 if (map
->m_len
> EXT_INIT_MAX_LEN
&&
4081 !(flags
& EXT4_GET_BLOCKS_UNINIT_EXT
))
4082 map
->m_len
= EXT_INIT_MAX_LEN
;
4083 else if (map
->m_len
> EXT_UNINIT_MAX_LEN
&&
4084 (flags
& EXT4_GET_BLOCKS_UNINIT_EXT
))
4085 map
->m_len
= EXT_UNINIT_MAX_LEN
;
4087 /* Check if we can really insert (m_lblk)::(m_lblk + m_len) extent */
4088 newex
.ee_len
= cpu_to_le16(map
->m_len
);
4089 err
= ext4_ext_check_overlap(sbi
, inode
, &newex
, path
);
4091 allocated
= ext4_ext_get_actual_len(&newex
);
4093 allocated
= map
->m_len
;
4095 /* allocate new block */
4097 ar
.goal
= ext4_ext_find_goal(inode
, path
, map
->m_lblk
);
4098 ar
.logical
= map
->m_lblk
;
4100 * We calculate the offset from the beginning of the cluster
4101 * for the logical block number, since when we allocate a
4102 * physical cluster, the physical block should start at the
4103 * same offset from the beginning of the cluster. This is
4104 * needed so that future calls to get_implied_cluster_alloc()
4107 offset
= map
->m_lblk
& (sbi
->s_cluster_ratio
- 1);
4108 ar
.len
= EXT4_NUM_B2C(sbi
, offset
+allocated
);
4110 ar
.logical
-= offset
;
4111 if (S_ISREG(inode
->i_mode
))
4112 ar
.flags
= EXT4_MB_HINT_DATA
;
4114 /* disable in-core preallocation for non-regular files */
4116 if (flags
& EXT4_GET_BLOCKS_NO_NORMALIZE
)
4117 ar
.flags
|= EXT4_MB_HINT_NOPREALLOC
;
4118 newblock
= ext4_mb_new_blocks(handle
, &ar
, &err
);
4121 ext_debug("allocate new block: goal %llu, found %llu/%u\n",
4122 ar
.goal
, newblock
, allocated
);
4124 allocated_clusters
= ar
.len
;
4125 ar
.len
= EXT4_C2B(sbi
, ar
.len
) - offset
;
4126 if (ar
.len
> allocated
)
4129 got_allocated_blocks
:
4130 /* try to insert new extent into found leaf and return */
4131 ext4_ext_store_pblock(&newex
, newblock
+ offset
);
4132 newex
.ee_len
= cpu_to_le16(ar
.len
);
4133 /* Mark uninitialized */
4134 if (flags
& EXT4_GET_BLOCKS_UNINIT_EXT
){
4135 ext4_ext_mark_uninitialized(&newex
);
4137 * io_end structure was created for every IO write to an
4138 * uninitialized extent. To avoid unnecessary conversion,
4139 * here we flag the IO that really needs the conversion.
4140 * For non asycn direct IO case, flag the inode state
4141 * that we need to perform conversion when IO is done.
4143 if ((flags
& EXT4_GET_BLOCKS_PRE_IO
))
4145 if (ext4_should_dioread_nolock(inode
))
4146 map
->m_flags
|= EXT4_MAP_UNINIT
;
4150 if ((flags
& EXT4_GET_BLOCKS_KEEP_SIZE
) == 0)
4151 err
= check_eofblocks_fl(handle
, inode
, map
->m_lblk
,
4154 err
= ext4_ext_insert_extent(handle
, inode
, path
,
4157 if (!err
&& set_unwritten
) {
4159 ext4_set_io_unwritten_flag(inode
, io
);
4161 ext4_set_inode_state(inode
,
4162 EXT4_STATE_DIO_UNWRITTEN
);
4165 if (err
&& free_on_err
) {
4166 int fb_flags
= flags
& EXT4_GET_BLOCKS_DELALLOC_RESERVE
?
4167 EXT4_FREE_BLOCKS_NO_QUOT_UPDATE
: 0;
4168 /* free data blocks we just allocated */
4169 /* not a good idea to call discard here directly,
4170 * but otherwise we'd need to call it every free() */
4171 ext4_discard_preallocations(inode
);
4172 ext4_free_blocks(handle
, inode
, NULL
, ext4_ext_pblock(&newex
),
4173 ext4_ext_get_actual_len(&newex
), fb_flags
);
4177 /* previous routine could use block we allocated */
4178 newblock
= ext4_ext_pblock(&newex
);
4179 allocated
= ext4_ext_get_actual_len(&newex
);
4180 if (allocated
> map
->m_len
)
4181 allocated
= map
->m_len
;
4182 map
->m_flags
|= EXT4_MAP_NEW
;
4185 * Update reserved blocks/metadata blocks after successful
4186 * block allocation which had been deferred till now.
4188 if (flags
& EXT4_GET_BLOCKS_DELALLOC_RESERVE
) {
4189 unsigned int reserved_clusters
;
4191 * Check how many clusters we had reserved this allocated range
4193 reserved_clusters
= get_reserved_cluster_alloc(inode
,
4194 map
->m_lblk
, allocated
);
4195 if (map
->m_flags
& EXT4_MAP_FROM_CLUSTER
) {
4196 if (reserved_clusters
) {
4198 * We have clusters reserved for this range.
4199 * But since we are not doing actual allocation
4200 * and are simply using blocks from previously
4201 * allocated cluster, we should release the
4202 * reservation and not claim quota.
4204 ext4_da_update_reserve_space(inode
,
4205 reserved_clusters
, 0);
4208 BUG_ON(allocated_clusters
< reserved_clusters
);
4209 /* We will claim quota for all newly allocated blocks.*/
4210 ext4_da_update_reserve_space(inode
, allocated_clusters
,
4212 if (reserved_clusters
< allocated_clusters
) {
4213 struct ext4_inode_info
*ei
= EXT4_I(inode
);
4214 int reservation
= allocated_clusters
-
4217 * It seems we claimed few clusters outside of
4218 * the range of this allocation. We should give
4219 * it back to the reservation pool. This can
4220 * happen in the following case:
4222 * * Suppose s_cluster_ratio is 4 (i.e., each
4223 * cluster has 4 blocks. Thus, the clusters
4224 * are [0-3],[4-7],[8-11]...
4225 * * First comes delayed allocation write for
4226 * logical blocks 10 & 11. Since there were no
4227 * previous delayed allocated blocks in the
4228 * range [8-11], we would reserve 1 cluster
4230 * * Next comes write for logical blocks 3 to 8.
4231 * In this case, we will reserve 2 clusters
4232 * (for [0-3] and [4-7]; and not for [8-11] as
4233 * that range has a delayed allocated blocks.
4234 * Thus total reserved clusters now becomes 3.
4235 * * Now, during the delayed allocation writeout
4236 * time, we will first write blocks [3-8] and
4237 * allocate 3 clusters for writing these
4238 * blocks. Also, we would claim all these
4239 * three clusters above.
4240 * * Now when we come here to writeout the
4241 * blocks [10-11], we would expect to claim
4242 * the reservation of 1 cluster we had made
4243 * (and we would claim it since there are no
4244 * more delayed allocated blocks in the range
4245 * [8-11]. But our reserved cluster count had
4246 * already gone to 0.
4248 * Thus, at the step 4 above when we determine
4249 * that there are still some unwritten delayed
4250 * allocated blocks outside of our current
4251 * block range, we should increment the
4252 * reserved clusters count so that when the
4253 * remaining blocks finally gets written, we
4256 dquot_reserve_block(inode
,
4257 EXT4_C2B(sbi
, reservation
));
4258 spin_lock(&ei
->i_block_reservation_lock
);
4259 ei
->i_reserved_data_blocks
+= reservation
;
4260 spin_unlock(&ei
->i_block_reservation_lock
);
4266 * Cache the extent and update transaction to commit on fdatasync only
4267 * when it is _not_ an uninitialized extent.
4269 if ((flags
& EXT4_GET_BLOCKS_UNINIT_EXT
) == 0) {
4270 ext4_ext_put_in_cache(inode
, map
->m_lblk
, allocated
, newblock
);
4271 ext4_update_inode_fsync_trans(handle
, inode
, 1);
4273 ext4_update_inode_fsync_trans(handle
, inode
, 0);
4275 if (allocated
> map
->m_len
)
4276 allocated
= map
->m_len
;
4277 ext4_ext_show_leaf(inode
, path
);
4278 map
->m_flags
|= EXT4_MAP_MAPPED
;
4279 map
->m_pblk
= newblock
;
4280 map
->m_len
= allocated
;
4283 ext4_ext_drop_refs(path
);
4287 trace_ext4_ext_map_blocks_exit(inode
, map
->m_lblk
,
4288 newblock
, map
->m_len
, err
? err
: allocated
);
4290 return err
? err
: allocated
;
4293 void ext4_ext_truncate(struct inode
*inode
)
4295 struct address_space
*mapping
= inode
->i_mapping
;
4296 struct super_block
*sb
= inode
->i_sb
;
4297 ext4_lblk_t last_block
;
4303 * finish any pending end_io work so we won't run the risk of
4304 * converting any truncated blocks to initialized later
4306 ext4_flush_unwritten_io(inode
);
4309 * probably first extent we're gonna free will be last in block
4311 err
= ext4_writepage_trans_blocks(inode
);
4312 handle
= ext4_journal_start(inode
, err
);
4316 if (inode
->i_size
% PAGE_CACHE_SIZE
!= 0) {
4317 page_len
= PAGE_CACHE_SIZE
-
4318 (inode
->i_size
& (PAGE_CACHE_SIZE
- 1));
4320 err
= ext4_discard_partial_page_buffers(handle
,
4321 mapping
, inode
->i_size
, page_len
, 0);
4327 if (ext4_orphan_add(handle
, inode
))
4330 down_write(&EXT4_I(inode
)->i_data_sem
);
4331 ext4_ext_invalidate_cache(inode
);
4333 ext4_discard_preallocations(inode
);
4336 * TODO: optimization is possible here.
4337 * Probably we need not scan at all,
4338 * because page truncation is enough.
4341 /* we have to know where to truncate from in crash case */
4342 EXT4_I(inode
)->i_disksize
= inode
->i_size
;
4343 ext4_mark_inode_dirty(handle
, inode
);
4345 last_block
= (inode
->i_size
+ sb
->s_blocksize
- 1)
4346 >> EXT4_BLOCK_SIZE_BITS(sb
);
4347 err
= ext4_ext_remove_space(inode
, last_block
, EXT_MAX_BLOCKS
- 1);
4349 /* In a multi-transaction truncate, we only make the final
4350 * transaction synchronous.
4353 ext4_handle_sync(handle
);
4355 up_write(&EXT4_I(inode
)->i_data_sem
);
4359 * If this was a simple ftruncate() and the file will remain alive,
4360 * then we need to clear up the orphan record which we created above.
4361 * However, if this was a real unlink then we were called by
4362 * ext4_delete_inode(), and we allow that function to clean up the
4363 * orphan info for us.
4366 ext4_orphan_del(handle
, inode
);
4368 inode
->i_mtime
= inode
->i_ctime
= ext4_current_time(inode
);
4369 ext4_mark_inode_dirty(handle
, inode
);
4370 ext4_journal_stop(handle
);
4373 static void ext4_falloc_update_inode(struct inode
*inode
,
4374 int mode
, loff_t new_size
, int update_ctime
)
4376 struct timespec now
;
4379 now
= current_fs_time(inode
->i_sb
);
4380 if (!timespec_equal(&inode
->i_ctime
, &now
))
4381 inode
->i_ctime
= now
;
4384 * Update only when preallocation was requested beyond
4387 if (!(mode
& FALLOC_FL_KEEP_SIZE
)) {
4388 if (new_size
> i_size_read(inode
))
4389 i_size_write(inode
, new_size
);
4390 if (new_size
> EXT4_I(inode
)->i_disksize
)
4391 ext4_update_i_disksize(inode
, new_size
);
4394 * Mark that we allocate beyond EOF so the subsequent truncate
4395 * can proceed even if the new size is the same as i_size.
4397 if (new_size
> i_size_read(inode
))
4398 ext4_set_inode_flag(inode
, EXT4_INODE_EOFBLOCKS
);
4404 * preallocate space for a file. This implements ext4's fallocate file
4405 * operation, which gets called from sys_fallocate system call.
4406 * For block-mapped files, posix_fallocate should fall back to the method
4407 * of writing zeroes to the required new blocks (the same behavior which is
4408 * expected for file systems which do not support fallocate() system call).
4410 long ext4_fallocate(struct file
*file
, int mode
, loff_t offset
, loff_t len
)
4412 struct inode
*inode
= file
->f_path
.dentry
->d_inode
;
4415 unsigned int max_blocks
;
4420 struct ext4_map_blocks map
;
4421 unsigned int credits
, blkbits
= inode
->i_blkbits
;
4424 * currently supporting (pre)allocate mode for extent-based
4427 if (!(ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
)))
4430 /* Return error if mode is not supported */
4431 if (mode
& ~(FALLOC_FL_KEEP_SIZE
| FALLOC_FL_PUNCH_HOLE
))
4434 if (mode
& FALLOC_FL_PUNCH_HOLE
)
4435 return ext4_punch_hole(file
, offset
, len
);
4437 trace_ext4_fallocate_enter(inode
, offset
, len
, mode
);
4438 map
.m_lblk
= offset
>> blkbits
;
4440 * We can't just convert len to max_blocks because
4441 * If blocksize = 4096 offset = 3072 and len = 2048
4443 max_blocks
= (EXT4_BLOCK_ALIGN(len
+ offset
, blkbits
) >> blkbits
)
4446 * credits to insert 1 extent into extent tree
4448 credits
= ext4_chunk_trans_blocks(inode
, max_blocks
);
4449 mutex_lock(&inode
->i_mutex
);
4450 ret
= inode_newsize_ok(inode
, (len
+ offset
));
4452 mutex_unlock(&inode
->i_mutex
);
4453 trace_ext4_fallocate_exit(inode
, offset
, max_blocks
, ret
);
4456 flags
= EXT4_GET_BLOCKS_CREATE_UNINIT_EXT
;
4457 if (mode
& FALLOC_FL_KEEP_SIZE
)
4458 flags
|= EXT4_GET_BLOCKS_KEEP_SIZE
;
4460 * Don't normalize the request if it can fit in one extent so
4461 * that it doesn't get unnecessarily split into multiple
4464 if (len
<= EXT_UNINIT_MAX_LEN
<< blkbits
)
4465 flags
|= EXT4_GET_BLOCKS_NO_NORMALIZE
;
4467 /* Prevent race condition between unwritten */
4468 ext4_flush_unwritten_io(inode
);
4470 while (ret
>= 0 && ret
< max_blocks
) {
4471 map
.m_lblk
= map
.m_lblk
+ ret
;
4472 map
.m_len
= max_blocks
= max_blocks
- ret
;
4473 handle
= ext4_journal_start(inode
, credits
);
4474 if (IS_ERR(handle
)) {
4475 ret
= PTR_ERR(handle
);
4478 ret
= ext4_map_blocks(handle
, inode
, &map
, flags
);
4482 printk(KERN_ERR
"%s: ext4_ext_map_blocks "
4483 "returned error inode#%lu, block=%u, "
4484 "max_blocks=%u", __func__
,
4485 inode
->i_ino
, map
.m_lblk
, max_blocks
);
4487 ext4_mark_inode_dirty(handle
, inode
);
4488 ret2
= ext4_journal_stop(handle
);
4491 if ((map
.m_lblk
+ ret
) >= (EXT4_BLOCK_ALIGN(offset
+ len
,
4492 blkbits
) >> blkbits
))
4493 new_size
= offset
+ len
;
4495 new_size
= ((loff_t
) map
.m_lblk
+ ret
) << blkbits
;
4497 ext4_falloc_update_inode(inode
, mode
, new_size
,
4498 (map
.m_flags
& EXT4_MAP_NEW
));
4499 ext4_mark_inode_dirty(handle
, inode
);
4500 if ((file
->f_flags
& O_SYNC
) && ret
>= max_blocks
)
4501 ext4_handle_sync(handle
);
4502 ret2
= ext4_journal_stop(handle
);
4506 if (ret
== -ENOSPC
&&
4507 ext4_should_retry_alloc(inode
->i_sb
, &retries
)) {
4511 mutex_unlock(&inode
->i_mutex
);
4512 trace_ext4_fallocate_exit(inode
, offset
, max_blocks
,
4513 ret
> 0 ? ret2
: ret
);
4514 return ret
> 0 ? ret2
: ret
;
4518 * This function convert a range of blocks to written extents
4519 * The caller of this function will pass the start offset and the size.
4520 * all unwritten extents within this range will be converted to
4523 * This function is called from the direct IO end io call back
4524 * function, to convert the fallocated extents after IO is completed.
4525 * Returns 0 on success.
4527 int ext4_convert_unwritten_extents(struct inode
*inode
, loff_t offset
,
4531 unsigned int max_blocks
;
4534 struct ext4_map_blocks map
;
4535 unsigned int credits
, blkbits
= inode
->i_blkbits
;
4537 map
.m_lblk
= offset
>> blkbits
;
4539 * We can't just convert len to max_blocks because
4540 * If blocksize = 4096 offset = 3072 and len = 2048
4542 max_blocks
= ((EXT4_BLOCK_ALIGN(len
+ offset
, blkbits
) >> blkbits
) -
4545 * credits to insert 1 extent into extent tree
4547 credits
= ext4_chunk_trans_blocks(inode
, max_blocks
);
4548 while (ret
>= 0 && ret
< max_blocks
) {
4550 map
.m_len
= (max_blocks
-= ret
);
4551 handle
= ext4_journal_start(inode
, credits
);
4552 if (IS_ERR(handle
)) {
4553 ret
= PTR_ERR(handle
);
4556 ret
= ext4_map_blocks(handle
, inode
, &map
,
4557 EXT4_GET_BLOCKS_IO_CONVERT_EXT
);
4560 ext4_msg(inode
->i_sb
, KERN_ERR
,
4561 "%s:%d: inode #%lu: block %u: len %u: "
4562 "ext4_ext_map_blocks returned %d",
4563 __func__
, __LINE__
, inode
->i_ino
, map
.m_lblk
,
4566 ext4_mark_inode_dirty(handle
, inode
);
4567 ret2
= ext4_journal_stop(handle
);
4568 if (ret
<= 0 || ret2
)
4571 return ret
> 0 ? ret2
: ret
;
4575 * Callback function called for each extent to gather FIEMAP information.
4577 static int ext4_ext_fiemap_cb(struct inode
*inode
, ext4_lblk_t next
,
4578 struct ext4_ext_cache
*newex
, struct ext4_extent
*ex
,
4586 struct fiemap_extent_info
*fieinfo
= data
;
4587 unsigned char blksize_bits
;
4589 blksize_bits
= inode
->i_sb
->s_blocksize_bits
;
4590 logical
= (__u64
)newex
->ec_block
<< blksize_bits
;
4592 if (newex
->ec_start
== 0) {
4594 * No extent in extent-tree contains block @newex->ec_start,
4595 * then the block may stay in 1)a hole or 2)delayed-extent.
4597 * Holes or delayed-extents are processed as follows.
4598 * 1. lookup dirty pages with specified range in pagecache.
4599 * If no page is got, then there is no delayed-extent and
4600 * return with EXT_CONTINUE.
4601 * 2. find the 1st mapped buffer,
4602 * 3. check if the mapped buffer is both in the request range
4603 * and a delayed buffer. If not, there is no delayed-extent,
4605 * 4. a delayed-extent is found, the extent will be collected.
4607 ext4_lblk_t end
= 0;
4608 pgoff_t last_offset
;
4611 pgoff_t start_index
= 0;
4612 struct page
**pages
= NULL
;
4613 struct buffer_head
*bh
= NULL
;
4614 struct buffer_head
*head
= NULL
;
4615 unsigned int nr_pages
= PAGE_SIZE
/ sizeof(struct page
*);
4617 pages
= kmalloc(PAGE_SIZE
, GFP_KERNEL
);
4621 offset
= logical
>> PAGE_SHIFT
;
4623 last_offset
= offset
;
4625 ret
= find_get_pages_tag(inode
->i_mapping
, &offset
,
4626 PAGECACHE_TAG_DIRTY
, nr_pages
, pages
);
4628 if (!(flags
& FIEMAP_EXTENT_DELALLOC
)) {
4629 /* First time, try to find a mapped buffer. */
4632 for (index
= 0; index
< ret
; index
++)
4633 page_cache_release(pages
[index
]);
4636 return EXT_CONTINUE
;
4641 /* Try to find the 1st mapped buffer. */
4642 end
= ((__u64
)pages
[index
]->index
<< PAGE_SHIFT
) >>
4644 if (!page_has_buffers(pages
[index
]))
4646 head
= page_buffers(pages
[index
]);
4653 if (end
>= newex
->ec_block
+
4655 /* The buffer is out of
4656 * the request range.
4660 if (buffer_mapped(bh
) &&
4661 end
>= newex
->ec_block
) {
4662 start_index
= index
- 1;
4663 /* get the 1st mapped buffer. */
4664 goto found_mapped_buffer
;
4667 bh
= bh
->b_this_page
;
4669 } while (bh
!= head
);
4671 /* No mapped buffer in the range found in this page,
4672 * We need to look up next page.
4675 /* There is no page left, but we need to limit
4678 newex
->ec_len
= end
- newex
->ec_block
;
4683 /*Find contiguous delayed buffers. */
4684 if (ret
> 0 && pages
[0]->index
== last_offset
)
4685 head
= page_buffers(pages
[0]);
4691 found_mapped_buffer
:
4692 if (bh
!= NULL
&& buffer_delay(bh
)) {
4693 /* 1st or contiguous delayed buffer found. */
4694 if (!(flags
& FIEMAP_EXTENT_DELALLOC
)) {
4696 * 1st delayed buffer found, record
4697 * the start of extent.
4699 flags
|= FIEMAP_EXTENT_DELALLOC
;
4700 newex
->ec_block
= end
;
4701 logical
= (__u64
)end
<< blksize_bits
;
4703 /* Find contiguous delayed buffers. */
4705 if (!buffer_delay(bh
))
4706 goto found_delayed_extent
;
4707 bh
= bh
->b_this_page
;
4709 } while (bh
!= head
);
4711 for (; index
< ret
; index
++) {
4712 if (!page_has_buffers(pages
[index
])) {
4716 head
= page_buffers(pages
[index
]);
4722 if (pages
[index
]->index
!=
4723 pages
[start_index
]->index
+ index
4725 /* Blocks are not contiguous. */
4731 if (!buffer_delay(bh
))
4732 /* Delayed-extent ends. */
4733 goto found_delayed_extent
;
4734 bh
= bh
->b_this_page
;
4736 } while (bh
!= head
);
4738 } else if (!(flags
& FIEMAP_EXTENT_DELALLOC
))
4742 found_delayed_extent
:
4743 newex
->ec_len
= min(end
- newex
->ec_block
,
4744 (ext4_lblk_t
)EXT_INIT_MAX_LEN
);
4745 if (ret
== nr_pages
&& bh
!= NULL
&&
4746 newex
->ec_len
< EXT_INIT_MAX_LEN
&&
4748 /* Have not collected an extent and continue. */
4749 for (index
= 0; index
< ret
; index
++)
4750 page_cache_release(pages
[index
]);
4754 for (index
= 0; index
< ret
; index
++)
4755 page_cache_release(pages
[index
]);
4759 physical
= (__u64
)newex
->ec_start
<< blksize_bits
;
4760 length
= (__u64
)newex
->ec_len
<< blksize_bits
;
4762 if (ex
&& ext4_ext_is_uninitialized(ex
))
4763 flags
|= FIEMAP_EXTENT_UNWRITTEN
;
4765 if (next
== EXT_MAX_BLOCKS
)
4766 flags
|= FIEMAP_EXTENT_LAST
;
4768 ret
= fiemap_fill_next_extent(fieinfo
, logical
, physical
,
4774 return EXT_CONTINUE
;
4776 /* fiemap flags we can handle specified here */
4777 #define EXT4_FIEMAP_FLAGS (FIEMAP_FLAG_SYNC|FIEMAP_FLAG_XATTR)
4779 static int ext4_xattr_fiemap(struct inode
*inode
,
4780 struct fiemap_extent_info
*fieinfo
)
4784 __u32 flags
= FIEMAP_EXTENT_LAST
;
4785 int blockbits
= inode
->i_sb
->s_blocksize_bits
;
4789 if (ext4_test_inode_state(inode
, EXT4_STATE_XATTR
)) {
4790 struct ext4_iloc iloc
;
4791 int offset
; /* offset of xattr in inode */
4793 error
= ext4_get_inode_loc(inode
, &iloc
);
4796 physical
= iloc
.bh
->b_blocknr
<< blockbits
;
4797 offset
= EXT4_GOOD_OLD_INODE_SIZE
+
4798 EXT4_I(inode
)->i_extra_isize
;
4800 length
= EXT4_SB(inode
->i_sb
)->s_inode_size
- offset
;
4801 flags
|= FIEMAP_EXTENT_DATA_INLINE
;
4803 } else { /* external block */
4804 physical
= EXT4_I(inode
)->i_file_acl
<< blockbits
;
4805 length
= inode
->i_sb
->s_blocksize
;
4809 error
= fiemap_fill_next_extent(fieinfo
, 0, physical
,
4811 return (error
< 0 ? error
: 0);
4815 * ext4_ext_punch_hole
4817 * Punches a hole of "length" bytes in a file starting
4820 * @inode: The inode of the file to punch a hole in
4821 * @offset: The starting byte offset of the hole
4822 * @length: The length of the hole
4824 * Returns the number of blocks removed or negative on err
4826 int ext4_ext_punch_hole(struct file
*file
, loff_t offset
, loff_t length
)
4828 struct inode
*inode
= file
->f_path
.dentry
->d_inode
;
4829 struct super_block
*sb
= inode
->i_sb
;
4830 ext4_lblk_t first_block
, stop_block
;
4831 struct address_space
*mapping
= inode
->i_mapping
;
4833 loff_t first_page
, last_page
, page_len
;
4834 loff_t first_page_offset
, last_page_offset
;
4835 int credits
, err
= 0;
4838 * Write out all dirty pages to avoid race conditions
4839 * Then release them.
4841 if (mapping
->nrpages
&& mapping_tagged(mapping
, PAGECACHE_TAG_DIRTY
)) {
4842 err
= filemap_write_and_wait_range(mapping
,
4843 offset
, offset
+ length
- 1);
4849 mutex_lock(&inode
->i_mutex
);
4850 /* It's not possible punch hole on append only file */
4851 if (IS_APPEND(inode
) || IS_IMMUTABLE(inode
)) {
4855 if (IS_SWAPFILE(inode
)) {
4860 /* No need to punch hole beyond i_size */
4861 if (offset
>= inode
->i_size
)
4865 * If the hole extends beyond i_size, set the hole
4866 * to end after the page that contains i_size
4868 if (offset
+ length
> inode
->i_size
) {
4869 length
= inode
->i_size
+
4870 PAGE_CACHE_SIZE
- (inode
->i_size
& (PAGE_CACHE_SIZE
- 1)) -
4874 first_page
= (offset
+ PAGE_CACHE_SIZE
- 1) >> PAGE_CACHE_SHIFT
;
4875 last_page
= (offset
+ length
) >> PAGE_CACHE_SHIFT
;
4877 first_page_offset
= first_page
<< PAGE_CACHE_SHIFT
;
4878 last_page_offset
= last_page
<< PAGE_CACHE_SHIFT
;
4880 /* Now release the pages */
4881 if (last_page_offset
> first_page_offset
) {
4882 truncate_pagecache_range(inode
, first_page_offset
,
4883 last_page_offset
- 1);
4886 /* Wait all existing dio workers, newcomers will block on i_mutex */
4887 ext4_inode_block_unlocked_dio(inode
);
4888 err
= ext4_flush_unwritten_io(inode
);
4891 inode_dio_wait(inode
);
4893 credits
= ext4_writepage_trans_blocks(inode
);
4894 handle
= ext4_journal_start(inode
, credits
);
4895 if (IS_ERR(handle
)) {
4896 err
= PTR_ERR(handle
);
4902 * Now we need to zero out the non-page-aligned data in the
4903 * pages at the start and tail of the hole, and unmap the buffer
4904 * heads for the block aligned regions of the page that were
4905 * completely zeroed.
4907 if (first_page
> last_page
) {
4909 * If the file space being truncated is contained within a page
4910 * just zero out and unmap the middle of that page
4912 err
= ext4_discard_partial_page_buffers(handle
,
4913 mapping
, offset
, length
, 0);
4919 * zero out and unmap the partial page that contains
4920 * the start of the hole
4922 page_len
= first_page_offset
- offset
;
4924 err
= ext4_discard_partial_page_buffers(handle
, mapping
,
4925 offset
, page_len
, 0);
4931 * zero out and unmap the partial page that contains
4932 * the end of the hole
4934 page_len
= offset
+ length
- last_page_offset
;
4936 err
= ext4_discard_partial_page_buffers(handle
, mapping
,
4937 last_page_offset
, page_len
, 0);
4944 * If i_size is contained in the last page, we need to
4945 * unmap and zero the partial page after i_size
4947 if (inode
->i_size
>> PAGE_CACHE_SHIFT
== last_page
&&
4948 inode
->i_size
% PAGE_CACHE_SIZE
!= 0) {
4950 page_len
= PAGE_CACHE_SIZE
-
4951 (inode
->i_size
& (PAGE_CACHE_SIZE
- 1));
4954 err
= ext4_discard_partial_page_buffers(handle
,
4955 mapping
, inode
->i_size
, page_len
, 0);
4962 first_block
= (offset
+ sb
->s_blocksize
- 1) >>
4963 EXT4_BLOCK_SIZE_BITS(sb
);
4964 stop_block
= (offset
+ length
) >> EXT4_BLOCK_SIZE_BITS(sb
);
4966 /* If there are no blocks to remove, return now */
4967 if (first_block
>= stop_block
)
4970 down_write(&EXT4_I(inode
)->i_data_sem
);
4971 ext4_ext_invalidate_cache(inode
);
4972 ext4_discard_preallocations(inode
);
4974 err
= ext4_ext_remove_space(inode
, first_block
, stop_block
- 1);
4976 ext4_ext_invalidate_cache(inode
);
4977 ext4_discard_preallocations(inode
);
4980 ext4_handle_sync(handle
);
4982 up_write(&EXT4_I(inode
)->i_data_sem
);
4985 inode
->i_mtime
= inode
->i_ctime
= ext4_current_time(inode
);
4986 ext4_mark_inode_dirty(handle
, inode
);
4987 ext4_journal_stop(handle
);
4989 ext4_inode_resume_unlocked_dio(inode
);
4991 mutex_unlock(&inode
->i_mutex
);
4994 int ext4_fiemap(struct inode
*inode
, struct fiemap_extent_info
*fieinfo
,
4995 __u64 start
, __u64 len
)
4997 ext4_lblk_t start_blk
;
5000 /* fallback to generic here if not in extents fmt */
5001 if (!(ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
)))
5002 return generic_block_fiemap(inode
, fieinfo
, start
, len
,
5005 if (fiemap_check_flags(fieinfo
, EXT4_FIEMAP_FLAGS
))
5008 if (fieinfo
->fi_flags
& FIEMAP_FLAG_XATTR
) {
5009 error
= ext4_xattr_fiemap(inode
, fieinfo
);
5011 ext4_lblk_t len_blks
;
5014 start_blk
= start
>> inode
->i_sb
->s_blocksize_bits
;
5015 last_blk
= (start
+ len
- 1) >> inode
->i_sb
->s_blocksize_bits
;
5016 if (last_blk
>= EXT_MAX_BLOCKS
)
5017 last_blk
= EXT_MAX_BLOCKS
-1;
5018 len_blks
= ((ext4_lblk_t
) last_blk
) - start_blk
+ 1;
5021 * Walk the extent tree gathering extent information.
5022 * ext4_ext_fiemap_cb will push extents back to user.
5024 error
= ext4_ext_walk_space(inode
, start_blk
, len_blks
,
5025 ext4_ext_fiemap_cb
, fieinfo
);