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"
44 #include "ext4_extents.h"
47 #include <trace/events/ext4.h>
50 * used by extent splitting.
52 #define EXT4_EXT_MAY_ZEROOUT 0x1 /* safe to zeroout if split fails \
54 #define EXT4_EXT_MARK_UNINIT1 0x2 /* mark first half uninitialized */
55 #define EXT4_EXT_MARK_UNINIT2 0x4 /* mark second half uninitialized */
57 #define EXT4_EXT_DATA_VALID1 0x8 /* first half contains valid data */
58 #define EXT4_EXT_DATA_VALID2 0x10 /* second half contains valid data */
60 static __le32
ext4_extent_block_csum(struct inode
*inode
,
61 struct ext4_extent_header
*eh
)
63 struct ext4_inode_info
*ei
= EXT4_I(inode
);
64 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
67 csum
= ext4_chksum(sbi
, ei
->i_csum_seed
, (__u8
*)eh
,
68 EXT4_EXTENT_TAIL_OFFSET(eh
));
69 return cpu_to_le32(csum
);
72 static int ext4_extent_block_csum_verify(struct inode
*inode
,
73 struct ext4_extent_header
*eh
)
75 struct ext4_extent_tail
*et
;
77 if (!EXT4_HAS_RO_COMPAT_FEATURE(inode
->i_sb
,
78 EXT4_FEATURE_RO_COMPAT_METADATA_CSUM
))
81 et
= find_ext4_extent_tail(eh
);
82 if (et
->et_checksum
!= ext4_extent_block_csum(inode
, eh
))
87 static void ext4_extent_block_csum_set(struct inode
*inode
,
88 struct ext4_extent_header
*eh
)
90 struct ext4_extent_tail
*et
;
92 if (!EXT4_HAS_RO_COMPAT_FEATURE(inode
->i_sb
,
93 EXT4_FEATURE_RO_COMPAT_METADATA_CSUM
))
96 et
= find_ext4_extent_tail(eh
);
97 et
->et_checksum
= ext4_extent_block_csum(inode
, eh
);
100 static int ext4_split_extent(handle_t
*handle
,
102 struct ext4_ext_path
*path
,
103 struct ext4_map_blocks
*map
,
107 static int ext4_split_extent_at(handle_t
*handle
,
109 struct ext4_ext_path
*path
,
114 static int ext4_find_delayed_extent(struct inode
*inode
,
115 struct ext4_ext_cache
*newex
);
117 static int ext4_ext_truncate_extend_restart(handle_t
*handle
,
123 if (!ext4_handle_valid(handle
))
125 if (handle
->h_buffer_credits
> needed
)
127 err
= ext4_journal_extend(handle
, needed
);
130 err
= ext4_truncate_restart_trans(handle
, inode
, needed
);
142 static int ext4_ext_get_access(handle_t
*handle
, struct inode
*inode
,
143 struct ext4_ext_path
*path
)
146 /* path points to block */
147 return ext4_journal_get_write_access(handle
, path
->p_bh
);
149 /* path points to leaf/index in inode body */
150 /* we use in-core data, no need to protect them */
160 #define ext4_ext_dirty(handle, inode, path) \
161 __ext4_ext_dirty(__func__, __LINE__, (handle), (inode), (path))
162 static int __ext4_ext_dirty(const char *where
, unsigned int line
,
163 handle_t
*handle
, struct inode
*inode
,
164 struct ext4_ext_path
*path
)
168 ext4_extent_block_csum_set(inode
, ext_block_hdr(path
->p_bh
));
169 /* path points to block */
170 err
= __ext4_handle_dirty_metadata(where
, line
, handle
,
173 /* path points to leaf/index in inode body */
174 err
= ext4_mark_inode_dirty(handle
, inode
);
179 static ext4_fsblk_t
ext4_ext_find_goal(struct inode
*inode
,
180 struct ext4_ext_path
*path
,
184 int depth
= path
->p_depth
;
185 struct ext4_extent
*ex
;
188 * Try to predict block placement assuming that we are
189 * filling in a file which will eventually be
190 * non-sparse --- i.e., in the case of libbfd writing
191 * an ELF object sections out-of-order but in a way
192 * the eventually results in a contiguous object or
193 * executable file, or some database extending a table
194 * space file. However, this is actually somewhat
195 * non-ideal if we are writing a sparse file such as
196 * qemu or KVM writing a raw image file that is going
197 * to stay fairly sparse, since it will end up
198 * fragmenting the file system's free space. Maybe we
199 * should have some hueristics or some way to allow
200 * userspace to pass a hint to file system,
201 * especially if the latter case turns out to be
204 ex
= path
[depth
].p_ext
;
206 ext4_fsblk_t ext_pblk
= ext4_ext_pblock(ex
);
207 ext4_lblk_t ext_block
= le32_to_cpu(ex
->ee_block
);
209 if (block
> ext_block
)
210 return ext_pblk
+ (block
- ext_block
);
212 return ext_pblk
- (ext_block
- block
);
215 /* it looks like index is empty;
216 * try to find starting block from index itself */
217 if (path
[depth
].p_bh
)
218 return path
[depth
].p_bh
->b_blocknr
;
221 /* OK. use inode's group */
222 return ext4_inode_to_goal_block(inode
);
226 * Allocation for a meta data block
229 ext4_ext_new_meta_block(handle_t
*handle
, struct inode
*inode
,
230 struct ext4_ext_path
*path
,
231 struct ext4_extent
*ex
, int *err
, unsigned int flags
)
233 ext4_fsblk_t goal
, newblock
;
235 goal
= ext4_ext_find_goal(inode
, path
, le32_to_cpu(ex
->ee_block
));
236 newblock
= ext4_new_meta_blocks(handle
, inode
, goal
, flags
,
241 static inline int ext4_ext_space_block(struct inode
*inode
, int check
)
245 size
= (inode
->i_sb
->s_blocksize
- sizeof(struct ext4_extent_header
))
246 / sizeof(struct ext4_extent
);
247 #ifdef AGGRESSIVE_TEST
248 if (!check
&& size
> 6)
254 static inline int ext4_ext_space_block_idx(struct inode
*inode
, int check
)
258 size
= (inode
->i_sb
->s_blocksize
- sizeof(struct ext4_extent_header
))
259 / sizeof(struct ext4_extent_idx
);
260 #ifdef AGGRESSIVE_TEST
261 if (!check
&& size
> 5)
267 static inline int ext4_ext_space_root(struct inode
*inode
, int check
)
271 size
= sizeof(EXT4_I(inode
)->i_data
);
272 size
-= sizeof(struct ext4_extent_header
);
273 size
/= sizeof(struct ext4_extent
);
274 #ifdef AGGRESSIVE_TEST
275 if (!check
&& size
> 3)
281 static inline int ext4_ext_space_root_idx(struct inode
*inode
, int check
)
285 size
= sizeof(EXT4_I(inode
)->i_data
);
286 size
-= sizeof(struct ext4_extent_header
);
287 size
/= sizeof(struct ext4_extent_idx
);
288 #ifdef AGGRESSIVE_TEST
289 if (!check
&& size
> 4)
296 * Calculate the number of metadata blocks needed
297 * to allocate @blocks
298 * Worse case is one block per extent
300 int ext4_ext_calc_metadata_amount(struct inode
*inode
, ext4_lblk_t lblock
)
302 struct ext4_inode_info
*ei
= EXT4_I(inode
);
305 idxs
= ((inode
->i_sb
->s_blocksize
- sizeof(struct ext4_extent_header
))
306 / sizeof(struct ext4_extent_idx
));
309 * If the new delayed allocation block is contiguous with the
310 * previous da block, it can share index blocks with the
311 * previous block, so we only need to allocate a new index
312 * block every idxs leaf blocks. At ldxs**2 blocks, we need
313 * an additional index block, and at ldxs**3 blocks, yet
314 * another index blocks.
316 if (ei
->i_da_metadata_calc_len
&&
317 ei
->i_da_metadata_calc_last_lblock
+1 == lblock
) {
320 if ((ei
->i_da_metadata_calc_len
% idxs
) == 0)
322 if ((ei
->i_da_metadata_calc_len
% (idxs
*idxs
)) == 0)
324 if ((ei
->i_da_metadata_calc_len
% (idxs
*idxs
*idxs
)) == 0) {
326 ei
->i_da_metadata_calc_len
= 0;
328 ei
->i_da_metadata_calc_len
++;
329 ei
->i_da_metadata_calc_last_lblock
++;
334 * In the worst case we need a new set of index blocks at
335 * every level of the inode's extent tree.
337 ei
->i_da_metadata_calc_len
= 1;
338 ei
->i_da_metadata_calc_last_lblock
= lblock
;
339 return ext_depth(inode
) + 1;
343 ext4_ext_max_entries(struct inode
*inode
, int depth
)
347 if (depth
== ext_depth(inode
)) {
349 max
= ext4_ext_space_root(inode
, 1);
351 max
= ext4_ext_space_root_idx(inode
, 1);
354 max
= ext4_ext_space_block(inode
, 1);
356 max
= ext4_ext_space_block_idx(inode
, 1);
362 static int ext4_valid_extent(struct inode
*inode
, struct ext4_extent
*ext
)
364 ext4_fsblk_t block
= ext4_ext_pblock(ext
);
365 int len
= ext4_ext_get_actual_len(ext
);
369 return ext4_data_block_valid(EXT4_SB(inode
->i_sb
), block
, len
);
372 static int ext4_valid_extent_idx(struct inode
*inode
,
373 struct ext4_extent_idx
*ext_idx
)
375 ext4_fsblk_t block
= ext4_idx_pblock(ext_idx
);
377 return ext4_data_block_valid(EXT4_SB(inode
->i_sb
), block
, 1);
380 static int ext4_valid_extent_entries(struct inode
*inode
,
381 struct ext4_extent_header
*eh
,
384 unsigned short entries
;
385 if (eh
->eh_entries
== 0)
388 entries
= le16_to_cpu(eh
->eh_entries
);
392 struct ext4_extent
*ext
= EXT_FIRST_EXTENT(eh
);
394 if (!ext4_valid_extent(inode
, ext
))
400 struct ext4_extent_idx
*ext_idx
= EXT_FIRST_INDEX(eh
);
402 if (!ext4_valid_extent_idx(inode
, ext_idx
))
411 static int __ext4_ext_check(const char *function
, unsigned int line
,
412 struct inode
*inode
, struct ext4_extent_header
*eh
,
415 const char *error_msg
;
418 if (unlikely(eh
->eh_magic
!= EXT4_EXT_MAGIC
)) {
419 error_msg
= "invalid magic";
422 if (unlikely(le16_to_cpu(eh
->eh_depth
) != depth
)) {
423 error_msg
= "unexpected eh_depth";
426 if (unlikely(eh
->eh_max
== 0)) {
427 error_msg
= "invalid eh_max";
430 max
= ext4_ext_max_entries(inode
, depth
);
431 if (unlikely(le16_to_cpu(eh
->eh_max
) > max
)) {
432 error_msg
= "too large eh_max";
435 if (unlikely(le16_to_cpu(eh
->eh_entries
) > le16_to_cpu(eh
->eh_max
))) {
436 error_msg
= "invalid eh_entries";
439 if (!ext4_valid_extent_entries(inode
, eh
, depth
)) {
440 error_msg
= "invalid extent entries";
443 /* Verify checksum on non-root extent tree nodes */
444 if (ext_depth(inode
) != depth
&&
445 !ext4_extent_block_csum_verify(inode
, eh
)) {
446 error_msg
= "extent tree corrupted";
452 ext4_error_inode(inode
, function
, line
, 0,
453 "bad header/extent: %s - magic %x, "
454 "entries %u, max %u(%u), depth %u(%u)",
455 error_msg
, le16_to_cpu(eh
->eh_magic
),
456 le16_to_cpu(eh
->eh_entries
), le16_to_cpu(eh
->eh_max
),
457 max
, le16_to_cpu(eh
->eh_depth
), depth
);
462 #define ext4_ext_check(inode, eh, depth) \
463 __ext4_ext_check(__func__, __LINE__, inode, eh, depth)
465 int ext4_ext_check_inode(struct inode
*inode
)
467 return ext4_ext_check(inode
, ext_inode_hdr(inode
), ext_depth(inode
));
470 static int __ext4_ext_check_block(const char *function
, unsigned int line
,
472 struct ext4_extent_header
*eh
,
474 struct buffer_head
*bh
)
478 if (buffer_verified(bh
))
480 ret
= ext4_ext_check(inode
, eh
, depth
);
483 set_buffer_verified(bh
);
487 #define ext4_ext_check_block(inode, eh, depth, bh) \
488 __ext4_ext_check_block(__func__, __LINE__, inode, eh, depth, bh)
491 static void ext4_ext_show_path(struct inode
*inode
, struct ext4_ext_path
*path
)
493 int k
, l
= path
->p_depth
;
496 for (k
= 0; k
<= l
; k
++, path
++) {
498 ext_debug(" %d->%llu", le32_to_cpu(path
->p_idx
->ei_block
),
499 ext4_idx_pblock(path
->p_idx
));
500 } else if (path
->p_ext
) {
501 ext_debug(" %d:[%d]%d:%llu ",
502 le32_to_cpu(path
->p_ext
->ee_block
),
503 ext4_ext_is_uninitialized(path
->p_ext
),
504 ext4_ext_get_actual_len(path
->p_ext
),
505 ext4_ext_pblock(path
->p_ext
));
512 static void ext4_ext_show_leaf(struct inode
*inode
, struct ext4_ext_path
*path
)
514 int depth
= ext_depth(inode
);
515 struct ext4_extent_header
*eh
;
516 struct ext4_extent
*ex
;
522 eh
= path
[depth
].p_hdr
;
523 ex
= EXT_FIRST_EXTENT(eh
);
525 ext_debug("Displaying leaf extents for inode %lu\n", inode
->i_ino
);
527 for (i
= 0; i
< le16_to_cpu(eh
->eh_entries
); i
++, ex
++) {
528 ext_debug("%d:[%d]%d:%llu ", le32_to_cpu(ex
->ee_block
),
529 ext4_ext_is_uninitialized(ex
),
530 ext4_ext_get_actual_len(ex
), ext4_ext_pblock(ex
));
535 static void ext4_ext_show_move(struct inode
*inode
, struct ext4_ext_path
*path
,
536 ext4_fsblk_t newblock
, int level
)
538 int depth
= ext_depth(inode
);
539 struct ext4_extent
*ex
;
541 if (depth
!= level
) {
542 struct ext4_extent_idx
*idx
;
543 idx
= path
[level
].p_idx
;
544 while (idx
<= EXT_MAX_INDEX(path
[level
].p_hdr
)) {
545 ext_debug("%d: move %d:%llu in new index %llu\n", level
,
546 le32_to_cpu(idx
->ei_block
),
547 ext4_idx_pblock(idx
),
555 ex
= path
[depth
].p_ext
;
556 while (ex
<= EXT_MAX_EXTENT(path
[depth
].p_hdr
)) {
557 ext_debug("move %d:%llu:[%d]%d in new leaf %llu\n",
558 le32_to_cpu(ex
->ee_block
),
560 ext4_ext_is_uninitialized(ex
),
561 ext4_ext_get_actual_len(ex
),
568 #define ext4_ext_show_path(inode, path)
569 #define ext4_ext_show_leaf(inode, path)
570 #define ext4_ext_show_move(inode, path, newblock, level)
573 void ext4_ext_drop_refs(struct ext4_ext_path
*path
)
575 int depth
= path
->p_depth
;
578 for (i
= 0; i
<= depth
; i
++, path
++)
586 * ext4_ext_binsearch_idx:
587 * binary search for the closest index of the given block
588 * the header must be checked before calling this
591 ext4_ext_binsearch_idx(struct inode
*inode
,
592 struct ext4_ext_path
*path
, ext4_lblk_t block
)
594 struct ext4_extent_header
*eh
= path
->p_hdr
;
595 struct ext4_extent_idx
*r
, *l
, *m
;
598 ext_debug("binsearch for %u(idx): ", block
);
600 l
= EXT_FIRST_INDEX(eh
) + 1;
601 r
= EXT_LAST_INDEX(eh
);
604 if (block
< le32_to_cpu(m
->ei_block
))
608 ext_debug("%p(%u):%p(%u):%p(%u) ", l
, le32_to_cpu(l
->ei_block
),
609 m
, le32_to_cpu(m
->ei_block
),
610 r
, le32_to_cpu(r
->ei_block
));
614 ext_debug(" -> %u->%lld ", le32_to_cpu(path
->p_idx
->ei_block
),
615 ext4_idx_pblock(path
->p_idx
));
617 #ifdef CHECK_BINSEARCH
619 struct ext4_extent_idx
*chix
, *ix
;
622 chix
= ix
= EXT_FIRST_INDEX(eh
);
623 for (k
= 0; k
< le16_to_cpu(eh
->eh_entries
); k
++, ix
++) {
625 le32_to_cpu(ix
->ei_block
) <= le32_to_cpu(ix
[-1].ei_block
)) {
626 printk(KERN_DEBUG
"k=%d, ix=0x%p, "
628 ix
, EXT_FIRST_INDEX(eh
));
629 printk(KERN_DEBUG
"%u <= %u\n",
630 le32_to_cpu(ix
->ei_block
),
631 le32_to_cpu(ix
[-1].ei_block
));
633 BUG_ON(k
&& le32_to_cpu(ix
->ei_block
)
634 <= le32_to_cpu(ix
[-1].ei_block
));
635 if (block
< le32_to_cpu(ix
->ei_block
))
639 BUG_ON(chix
!= path
->p_idx
);
646 * ext4_ext_binsearch:
647 * binary search for closest extent of the given block
648 * the header must be checked before calling this
651 ext4_ext_binsearch(struct inode
*inode
,
652 struct ext4_ext_path
*path
, ext4_lblk_t block
)
654 struct ext4_extent_header
*eh
= path
->p_hdr
;
655 struct ext4_extent
*r
, *l
, *m
;
657 if (eh
->eh_entries
== 0) {
659 * this leaf is empty:
660 * we get such a leaf in split/add case
665 ext_debug("binsearch for %u: ", block
);
667 l
= EXT_FIRST_EXTENT(eh
) + 1;
668 r
= EXT_LAST_EXTENT(eh
);
672 if (block
< le32_to_cpu(m
->ee_block
))
676 ext_debug("%p(%u):%p(%u):%p(%u) ", l
, le32_to_cpu(l
->ee_block
),
677 m
, le32_to_cpu(m
->ee_block
),
678 r
, le32_to_cpu(r
->ee_block
));
682 ext_debug(" -> %d:%llu:[%d]%d ",
683 le32_to_cpu(path
->p_ext
->ee_block
),
684 ext4_ext_pblock(path
->p_ext
),
685 ext4_ext_is_uninitialized(path
->p_ext
),
686 ext4_ext_get_actual_len(path
->p_ext
));
688 #ifdef CHECK_BINSEARCH
690 struct ext4_extent
*chex
, *ex
;
693 chex
= ex
= EXT_FIRST_EXTENT(eh
);
694 for (k
= 0; k
< le16_to_cpu(eh
->eh_entries
); k
++, ex
++) {
695 BUG_ON(k
&& le32_to_cpu(ex
->ee_block
)
696 <= le32_to_cpu(ex
[-1].ee_block
));
697 if (block
< le32_to_cpu(ex
->ee_block
))
701 BUG_ON(chex
!= path
->p_ext
);
707 int ext4_ext_tree_init(handle_t
*handle
, struct inode
*inode
)
709 struct ext4_extent_header
*eh
;
711 eh
= ext_inode_hdr(inode
);
714 eh
->eh_magic
= EXT4_EXT_MAGIC
;
715 eh
->eh_max
= cpu_to_le16(ext4_ext_space_root(inode
, 0));
716 ext4_mark_inode_dirty(handle
, inode
);
717 ext4_ext_invalidate_cache(inode
);
721 struct ext4_ext_path
*
722 ext4_ext_find_extent(struct inode
*inode
, ext4_lblk_t block
,
723 struct ext4_ext_path
*path
)
725 struct ext4_extent_header
*eh
;
726 struct buffer_head
*bh
;
727 short int depth
, i
, ppos
= 0, alloc
= 0;
730 eh
= ext_inode_hdr(inode
);
731 depth
= ext_depth(inode
);
733 /* account possible depth increase */
735 path
= kzalloc(sizeof(struct ext4_ext_path
) * (depth
+ 2),
738 return ERR_PTR(-ENOMEM
);
745 /* walk through the tree */
747 ext_debug("depth %d: num %d, max %d\n",
748 ppos
, le16_to_cpu(eh
->eh_entries
), le16_to_cpu(eh
->eh_max
));
750 ext4_ext_binsearch_idx(inode
, path
+ ppos
, block
);
751 path
[ppos
].p_block
= ext4_idx_pblock(path
[ppos
].p_idx
);
752 path
[ppos
].p_depth
= i
;
753 path
[ppos
].p_ext
= NULL
;
755 bh
= sb_getblk(inode
->i_sb
, path
[ppos
].p_block
);
760 if (!bh_uptodate_or_lock(bh
)) {
761 trace_ext4_ext_load_extent(inode
, block
,
763 ret
= bh_submit_read(bh
);
769 eh
= ext_block_hdr(bh
);
771 if (unlikely(ppos
> depth
)) {
773 EXT4_ERROR_INODE(inode
,
774 "ppos %d > depth %d", ppos
, depth
);
778 path
[ppos
].p_bh
= bh
;
779 path
[ppos
].p_hdr
= eh
;
782 ret
= ext4_ext_check_block(inode
, eh
, i
, bh
);
787 path
[ppos
].p_depth
= i
;
788 path
[ppos
].p_ext
= NULL
;
789 path
[ppos
].p_idx
= NULL
;
792 ext4_ext_binsearch(inode
, path
+ ppos
, block
);
793 /* if not an empty leaf */
794 if (path
[ppos
].p_ext
)
795 path
[ppos
].p_block
= ext4_ext_pblock(path
[ppos
].p_ext
);
797 ext4_ext_show_path(inode
, path
);
802 ext4_ext_drop_refs(path
);
809 * ext4_ext_insert_index:
810 * insert new index [@logical;@ptr] into the block at @curp;
811 * check where to insert: before @curp or after @curp
813 static int ext4_ext_insert_index(handle_t
*handle
, struct inode
*inode
,
814 struct ext4_ext_path
*curp
,
815 int logical
, ext4_fsblk_t ptr
)
817 struct ext4_extent_idx
*ix
;
820 err
= ext4_ext_get_access(handle
, inode
, curp
);
824 if (unlikely(logical
== le32_to_cpu(curp
->p_idx
->ei_block
))) {
825 EXT4_ERROR_INODE(inode
,
826 "logical %d == ei_block %d!",
827 logical
, le32_to_cpu(curp
->p_idx
->ei_block
));
831 if (unlikely(le16_to_cpu(curp
->p_hdr
->eh_entries
)
832 >= le16_to_cpu(curp
->p_hdr
->eh_max
))) {
833 EXT4_ERROR_INODE(inode
,
834 "eh_entries %d >= eh_max %d!",
835 le16_to_cpu(curp
->p_hdr
->eh_entries
),
836 le16_to_cpu(curp
->p_hdr
->eh_max
));
840 if (logical
> le32_to_cpu(curp
->p_idx
->ei_block
)) {
842 ext_debug("insert new index %d after: %llu\n", logical
, ptr
);
843 ix
= curp
->p_idx
+ 1;
846 ext_debug("insert new index %d before: %llu\n", logical
, ptr
);
850 len
= EXT_LAST_INDEX(curp
->p_hdr
) - ix
+ 1;
853 ext_debug("insert new index %d: "
854 "move %d indices from 0x%p to 0x%p\n",
855 logical
, len
, ix
, ix
+ 1);
856 memmove(ix
+ 1, ix
, len
* sizeof(struct ext4_extent_idx
));
859 if (unlikely(ix
> EXT_MAX_INDEX(curp
->p_hdr
))) {
860 EXT4_ERROR_INODE(inode
, "ix > EXT_MAX_INDEX!");
864 ix
->ei_block
= cpu_to_le32(logical
);
865 ext4_idx_store_pblock(ix
, ptr
);
866 le16_add_cpu(&curp
->p_hdr
->eh_entries
, 1);
868 if (unlikely(ix
> EXT_LAST_INDEX(curp
->p_hdr
))) {
869 EXT4_ERROR_INODE(inode
, "ix > EXT_LAST_INDEX!");
873 err
= ext4_ext_dirty(handle
, inode
, curp
);
874 ext4_std_error(inode
->i_sb
, err
);
881 * inserts new subtree into the path, using free index entry
883 * - allocates all needed blocks (new leaf and all intermediate index blocks)
884 * - makes decision where to split
885 * - moves remaining extents and index entries (right to the split point)
886 * into the newly allocated blocks
887 * - initializes subtree
889 static int ext4_ext_split(handle_t
*handle
, struct inode
*inode
,
891 struct ext4_ext_path
*path
,
892 struct ext4_extent
*newext
, int at
)
894 struct buffer_head
*bh
= NULL
;
895 int depth
= ext_depth(inode
);
896 struct ext4_extent_header
*neh
;
897 struct ext4_extent_idx
*fidx
;
899 ext4_fsblk_t newblock
, oldblock
;
901 ext4_fsblk_t
*ablocks
= NULL
; /* array of allocated blocks */
904 /* make decision: where to split? */
905 /* FIXME: now decision is simplest: at current extent */
907 /* if current leaf will be split, then we should use
908 * border from split point */
909 if (unlikely(path
[depth
].p_ext
> EXT_MAX_EXTENT(path
[depth
].p_hdr
))) {
910 EXT4_ERROR_INODE(inode
, "p_ext > EXT_MAX_EXTENT!");
913 if (path
[depth
].p_ext
!= EXT_MAX_EXTENT(path
[depth
].p_hdr
)) {
914 border
= path
[depth
].p_ext
[1].ee_block
;
915 ext_debug("leaf will be split."
916 " next leaf starts at %d\n",
917 le32_to_cpu(border
));
919 border
= newext
->ee_block
;
920 ext_debug("leaf will be added."
921 " next leaf starts at %d\n",
922 le32_to_cpu(border
));
926 * If error occurs, then we break processing
927 * and mark filesystem read-only. index won't
928 * be inserted and tree will be in consistent
929 * state. Next mount will repair buffers too.
933 * Get array to track all allocated blocks.
934 * We need this to handle errors and free blocks
937 ablocks
= kzalloc(sizeof(ext4_fsblk_t
) * depth
, GFP_NOFS
);
941 /* allocate all needed blocks */
942 ext_debug("allocate %d blocks for indexes/leaf\n", depth
- at
);
943 for (a
= 0; a
< depth
- at
; a
++) {
944 newblock
= ext4_ext_new_meta_block(handle
, inode
, path
,
945 newext
, &err
, flags
);
948 ablocks
[a
] = newblock
;
951 /* initialize new leaf */
952 newblock
= ablocks
[--a
];
953 if (unlikely(newblock
== 0)) {
954 EXT4_ERROR_INODE(inode
, "newblock == 0!");
958 bh
= sb_getblk(inode
->i_sb
, newblock
);
965 err
= ext4_journal_get_create_access(handle
, bh
);
969 neh
= ext_block_hdr(bh
);
971 neh
->eh_max
= cpu_to_le16(ext4_ext_space_block(inode
, 0));
972 neh
->eh_magic
= EXT4_EXT_MAGIC
;
975 /* move remainder of path[depth] to the new leaf */
976 if (unlikely(path
[depth
].p_hdr
->eh_entries
!=
977 path
[depth
].p_hdr
->eh_max
)) {
978 EXT4_ERROR_INODE(inode
, "eh_entries %d != eh_max %d!",
979 path
[depth
].p_hdr
->eh_entries
,
980 path
[depth
].p_hdr
->eh_max
);
984 /* start copy from next extent */
985 m
= EXT_MAX_EXTENT(path
[depth
].p_hdr
) - path
[depth
].p_ext
++;
986 ext4_ext_show_move(inode
, path
, newblock
, depth
);
988 struct ext4_extent
*ex
;
989 ex
= EXT_FIRST_EXTENT(neh
);
990 memmove(ex
, path
[depth
].p_ext
, sizeof(struct ext4_extent
) * m
);
991 le16_add_cpu(&neh
->eh_entries
, m
);
994 ext4_extent_block_csum_set(inode
, neh
);
995 set_buffer_uptodate(bh
);
998 err
= ext4_handle_dirty_metadata(handle
, inode
, bh
);
1004 /* correct old leaf */
1006 err
= ext4_ext_get_access(handle
, inode
, path
+ depth
);
1009 le16_add_cpu(&path
[depth
].p_hdr
->eh_entries
, -m
);
1010 err
= ext4_ext_dirty(handle
, inode
, path
+ depth
);
1016 /* create intermediate indexes */
1018 if (unlikely(k
< 0)) {
1019 EXT4_ERROR_INODE(inode
, "k %d < 0!", k
);
1024 ext_debug("create %d intermediate indices\n", k
);
1025 /* insert new index into current index block */
1026 /* current depth stored in i var */
1029 oldblock
= newblock
;
1030 newblock
= ablocks
[--a
];
1031 bh
= sb_getblk(inode
->i_sb
, newblock
);
1032 if (unlikely(!bh
)) {
1038 err
= ext4_journal_get_create_access(handle
, bh
);
1042 neh
= ext_block_hdr(bh
);
1043 neh
->eh_entries
= cpu_to_le16(1);
1044 neh
->eh_magic
= EXT4_EXT_MAGIC
;
1045 neh
->eh_max
= cpu_to_le16(ext4_ext_space_block_idx(inode
, 0));
1046 neh
->eh_depth
= cpu_to_le16(depth
- i
);
1047 fidx
= EXT_FIRST_INDEX(neh
);
1048 fidx
->ei_block
= border
;
1049 ext4_idx_store_pblock(fidx
, oldblock
);
1051 ext_debug("int.index at %d (block %llu): %u -> %llu\n",
1052 i
, newblock
, le32_to_cpu(border
), oldblock
);
1054 /* move remainder of path[i] to the new index block */
1055 if (unlikely(EXT_MAX_INDEX(path
[i
].p_hdr
) !=
1056 EXT_LAST_INDEX(path
[i
].p_hdr
))) {
1057 EXT4_ERROR_INODE(inode
,
1058 "EXT_MAX_INDEX != EXT_LAST_INDEX ee_block %d!",
1059 le32_to_cpu(path
[i
].p_ext
->ee_block
));
1063 /* start copy indexes */
1064 m
= EXT_MAX_INDEX(path
[i
].p_hdr
) - path
[i
].p_idx
++;
1065 ext_debug("cur 0x%p, last 0x%p\n", path
[i
].p_idx
,
1066 EXT_MAX_INDEX(path
[i
].p_hdr
));
1067 ext4_ext_show_move(inode
, path
, newblock
, i
);
1069 memmove(++fidx
, path
[i
].p_idx
,
1070 sizeof(struct ext4_extent_idx
) * m
);
1071 le16_add_cpu(&neh
->eh_entries
, m
);
1073 ext4_extent_block_csum_set(inode
, neh
);
1074 set_buffer_uptodate(bh
);
1077 err
= ext4_handle_dirty_metadata(handle
, inode
, bh
);
1083 /* correct old index */
1085 err
= ext4_ext_get_access(handle
, inode
, path
+ i
);
1088 le16_add_cpu(&path
[i
].p_hdr
->eh_entries
, -m
);
1089 err
= ext4_ext_dirty(handle
, inode
, path
+ i
);
1097 /* insert new index */
1098 err
= ext4_ext_insert_index(handle
, inode
, path
+ at
,
1099 le32_to_cpu(border
), newblock
);
1103 if (buffer_locked(bh
))
1109 /* free all allocated blocks in error case */
1110 for (i
= 0; i
< depth
; i
++) {
1113 ext4_free_blocks(handle
, inode
, NULL
, ablocks
[i
], 1,
1114 EXT4_FREE_BLOCKS_METADATA
);
1123 * ext4_ext_grow_indepth:
1124 * implements tree growing procedure:
1125 * - allocates new block
1126 * - moves top-level data (index block or leaf) into the new block
1127 * - initializes new top-level, creating index that points to the
1128 * just created block
1130 static int ext4_ext_grow_indepth(handle_t
*handle
, struct inode
*inode
,
1132 struct ext4_extent
*newext
)
1134 struct ext4_extent_header
*neh
;
1135 struct buffer_head
*bh
;
1136 ext4_fsblk_t newblock
;
1139 newblock
= ext4_ext_new_meta_block(handle
, inode
, NULL
,
1140 newext
, &err
, flags
);
1144 bh
= sb_getblk(inode
->i_sb
, newblock
);
1149 err
= ext4_journal_get_create_access(handle
, bh
);
1155 /* move top-level index/leaf into new block */
1156 memmove(bh
->b_data
, EXT4_I(inode
)->i_data
,
1157 sizeof(EXT4_I(inode
)->i_data
));
1159 /* set size of new block */
1160 neh
= ext_block_hdr(bh
);
1161 /* old root could have indexes or leaves
1162 * so calculate e_max right way */
1163 if (ext_depth(inode
))
1164 neh
->eh_max
= cpu_to_le16(ext4_ext_space_block_idx(inode
, 0));
1166 neh
->eh_max
= cpu_to_le16(ext4_ext_space_block(inode
, 0));
1167 neh
->eh_magic
= EXT4_EXT_MAGIC
;
1168 ext4_extent_block_csum_set(inode
, neh
);
1169 set_buffer_uptodate(bh
);
1172 err
= ext4_handle_dirty_metadata(handle
, inode
, bh
);
1176 /* Update top-level index: num,max,pointer */
1177 neh
= ext_inode_hdr(inode
);
1178 neh
->eh_entries
= cpu_to_le16(1);
1179 ext4_idx_store_pblock(EXT_FIRST_INDEX(neh
), newblock
);
1180 if (neh
->eh_depth
== 0) {
1181 /* Root extent block becomes index block */
1182 neh
->eh_max
= cpu_to_le16(ext4_ext_space_root_idx(inode
, 0));
1183 EXT_FIRST_INDEX(neh
)->ei_block
=
1184 EXT_FIRST_EXTENT(neh
)->ee_block
;
1186 ext_debug("new root: num %d(%d), lblock %d, ptr %llu\n",
1187 le16_to_cpu(neh
->eh_entries
), le16_to_cpu(neh
->eh_max
),
1188 le32_to_cpu(EXT_FIRST_INDEX(neh
)->ei_block
),
1189 ext4_idx_pblock(EXT_FIRST_INDEX(neh
)));
1191 le16_add_cpu(&neh
->eh_depth
, 1);
1192 ext4_mark_inode_dirty(handle
, inode
);
1200 * ext4_ext_create_new_leaf:
1201 * finds empty index and adds new leaf.
1202 * if no free index is found, then it requests in-depth growing.
1204 static int ext4_ext_create_new_leaf(handle_t
*handle
, struct inode
*inode
,
1206 struct ext4_ext_path
*path
,
1207 struct ext4_extent
*newext
)
1209 struct ext4_ext_path
*curp
;
1210 int depth
, i
, err
= 0;
1213 i
= depth
= ext_depth(inode
);
1215 /* walk up to the tree and look for free index entry */
1216 curp
= path
+ depth
;
1217 while (i
> 0 && !EXT_HAS_FREE_INDEX(curp
)) {
1222 /* we use already allocated block for index block,
1223 * so subsequent data blocks should be contiguous */
1224 if (EXT_HAS_FREE_INDEX(curp
)) {
1225 /* if we found index with free entry, then use that
1226 * entry: create all needed subtree and add new leaf */
1227 err
= ext4_ext_split(handle
, inode
, flags
, path
, newext
, i
);
1232 ext4_ext_drop_refs(path
);
1233 path
= ext4_ext_find_extent(inode
,
1234 (ext4_lblk_t
)le32_to_cpu(newext
->ee_block
),
1237 err
= PTR_ERR(path
);
1239 /* tree is full, time to grow in depth */
1240 err
= ext4_ext_grow_indepth(handle
, inode
, flags
, newext
);
1245 ext4_ext_drop_refs(path
);
1246 path
= ext4_ext_find_extent(inode
,
1247 (ext4_lblk_t
)le32_to_cpu(newext
->ee_block
),
1250 err
= PTR_ERR(path
);
1255 * only first (depth 0 -> 1) produces free space;
1256 * in all other cases we have to split the grown tree
1258 depth
= ext_depth(inode
);
1259 if (path
[depth
].p_hdr
->eh_entries
== path
[depth
].p_hdr
->eh_max
) {
1260 /* now we need to split */
1270 * search the closest allocated block to the left for *logical
1271 * and returns it at @logical + it's physical address at @phys
1272 * if *logical is the smallest allocated block, the function
1273 * returns 0 at @phys
1274 * return value contains 0 (success) or error code
1276 static int ext4_ext_search_left(struct inode
*inode
,
1277 struct ext4_ext_path
*path
,
1278 ext4_lblk_t
*logical
, ext4_fsblk_t
*phys
)
1280 struct ext4_extent_idx
*ix
;
1281 struct ext4_extent
*ex
;
1284 if (unlikely(path
== NULL
)) {
1285 EXT4_ERROR_INODE(inode
, "path == NULL *logical %d!", *logical
);
1288 depth
= path
->p_depth
;
1291 if (depth
== 0 && path
->p_ext
== NULL
)
1294 /* usually extent in the path covers blocks smaller
1295 * then *logical, but it can be that extent is the
1296 * first one in the file */
1298 ex
= path
[depth
].p_ext
;
1299 ee_len
= ext4_ext_get_actual_len(ex
);
1300 if (*logical
< le32_to_cpu(ex
->ee_block
)) {
1301 if (unlikely(EXT_FIRST_EXTENT(path
[depth
].p_hdr
) != ex
)) {
1302 EXT4_ERROR_INODE(inode
,
1303 "EXT_FIRST_EXTENT != ex *logical %d ee_block %d!",
1304 *logical
, le32_to_cpu(ex
->ee_block
));
1307 while (--depth
>= 0) {
1308 ix
= path
[depth
].p_idx
;
1309 if (unlikely(ix
!= EXT_FIRST_INDEX(path
[depth
].p_hdr
))) {
1310 EXT4_ERROR_INODE(inode
,
1311 "ix (%d) != EXT_FIRST_INDEX (%d) (depth %d)!",
1312 ix
!= NULL
? le32_to_cpu(ix
->ei_block
) : 0,
1313 EXT_FIRST_INDEX(path
[depth
].p_hdr
) != NULL
?
1314 le32_to_cpu(EXT_FIRST_INDEX(path
[depth
].p_hdr
)->ei_block
) : 0,
1322 if (unlikely(*logical
< (le32_to_cpu(ex
->ee_block
) + ee_len
))) {
1323 EXT4_ERROR_INODE(inode
,
1324 "logical %d < ee_block %d + ee_len %d!",
1325 *logical
, le32_to_cpu(ex
->ee_block
), ee_len
);
1329 *logical
= le32_to_cpu(ex
->ee_block
) + ee_len
- 1;
1330 *phys
= ext4_ext_pblock(ex
) + ee_len
- 1;
1335 * search the closest allocated block to the right for *logical
1336 * and returns it at @logical + it's physical address at @phys
1337 * if *logical is the largest allocated block, the function
1338 * returns 0 at @phys
1339 * return value contains 0 (success) or error code
1341 static int ext4_ext_search_right(struct inode
*inode
,
1342 struct ext4_ext_path
*path
,
1343 ext4_lblk_t
*logical
, ext4_fsblk_t
*phys
,
1344 struct ext4_extent
**ret_ex
)
1346 struct buffer_head
*bh
= NULL
;
1347 struct ext4_extent_header
*eh
;
1348 struct ext4_extent_idx
*ix
;
1349 struct ext4_extent
*ex
;
1351 int depth
; /* Note, NOT eh_depth; depth from top of tree */
1354 if (unlikely(path
== NULL
)) {
1355 EXT4_ERROR_INODE(inode
, "path == NULL *logical %d!", *logical
);
1358 depth
= path
->p_depth
;
1361 if (depth
== 0 && path
->p_ext
== NULL
)
1364 /* usually extent in the path covers blocks smaller
1365 * then *logical, but it can be that extent is the
1366 * first one in the file */
1368 ex
= path
[depth
].p_ext
;
1369 ee_len
= ext4_ext_get_actual_len(ex
);
1370 if (*logical
< le32_to_cpu(ex
->ee_block
)) {
1371 if (unlikely(EXT_FIRST_EXTENT(path
[depth
].p_hdr
) != ex
)) {
1372 EXT4_ERROR_INODE(inode
,
1373 "first_extent(path[%d].p_hdr) != ex",
1377 while (--depth
>= 0) {
1378 ix
= path
[depth
].p_idx
;
1379 if (unlikely(ix
!= EXT_FIRST_INDEX(path
[depth
].p_hdr
))) {
1380 EXT4_ERROR_INODE(inode
,
1381 "ix != EXT_FIRST_INDEX *logical %d!",
1389 if (unlikely(*logical
< (le32_to_cpu(ex
->ee_block
) + ee_len
))) {
1390 EXT4_ERROR_INODE(inode
,
1391 "logical %d < ee_block %d + ee_len %d!",
1392 *logical
, le32_to_cpu(ex
->ee_block
), ee_len
);
1396 if (ex
!= EXT_LAST_EXTENT(path
[depth
].p_hdr
)) {
1397 /* next allocated block in this leaf */
1402 /* go up and search for index to the right */
1403 while (--depth
>= 0) {
1404 ix
= path
[depth
].p_idx
;
1405 if (ix
!= EXT_LAST_INDEX(path
[depth
].p_hdr
))
1409 /* we've gone up to the root and found no index to the right */
1413 /* we've found index to the right, let's
1414 * follow it and find the closest allocated
1415 * block to the right */
1417 block
= ext4_idx_pblock(ix
);
1418 while (++depth
< path
->p_depth
) {
1419 bh
= sb_bread(inode
->i_sb
, block
);
1422 eh
= ext_block_hdr(bh
);
1423 /* subtract from p_depth to get proper eh_depth */
1424 if (ext4_ext_check_block(inode
, eh
,
1425 path
->p_depth
- depth
, bh
)) {
1429 ix
= EXT_FIRST_INDEX(eh
);
1430 block
= ext4_idx_pblock(ix
);
1434 bh
= sb_bread(inode
->i_sb
, block
);
1437 eh
= ext_block_hdr(bh
);
1438 if (ext4_ext_check_block(inode
, eh
, path
->p_depth
- depth
, bh
)) {
1442 ex
= EXT_FIRST_EXTENT(eh
);
1444 *logical
= le32_to_cpu(ex
->ee_block
);
1445 *phys
= ext4_ext_pblock(ex
);
1453 * ext4_ext_next_allocated_block:
1454 * returns allocated block in subsequent extent or EXT_MAX_BLOCKS.
1455 * NOTE: it considers block number from index entry as
1456 * allocated block. Thus, index entries have to be consistent
1460 ext4_ext_next_allocated_block(struct ext4_ext_path
*path
)
1464 BUG_ON(path
== NULL
);
1465 depth
= path
->p_depth
;
1467 if (depth
== 0 && path
->p_ext
== NULL
)
1468 return EXT_MAX_BLOCKS
;
1470 while (depth
>= 0) {
1471 if (depth
== path
->p_depth
) {
1473 if (path
[depth
].p_ext
&&
1474 path
[depth
].p_ext
!=
1475 EXT_LAST_EXTENT(path
[depth
].p_hdr
))
1476 return le32_to_cpu(path
[depth
].p_ext
[1].ee_block
);
1479 if (path
[depth
].p_idx
!=
1480 EXT_LAST_INDEX(path
[depth
].p_hdr
))
1481 return le32_to_cpu(path
[depth
].p_idx
[1].ei_block
);
1486 return EXT_MAX_BLOCKS
;
1490 * ext4_ext_next_leaf_block:
1491 * returns first allocated block from next leaf or EXT_MAX_BLOCKS
1493 static ext4_lblk_t
ext4_ext_next_leaf_block(struct ext4_ext_path
*path
)
1497 BUG_ON(path
== NULL
);
1498 depth
= path
->p_depth
;
1500 /* zero-tree has no leaf blocks at all */
1502 return EXT_MAX_BLOCKS
;
1504 /* go to index block */
1507 while (depth
>= 0) {
1508 if (path
[depth
].p_idx
!=
1509 EXT_LAST_INDEX(path
[depth
].p_hdr
))
1510 return (ext4_lblk_t
)
1511 le32_to_cpu(path
[depth
].p_idx
[1].ei_block
);
1515 return EXT_MAX_BLOCKS
;
1519 * ext4_ext_correct_indexes:
1520 * if leaf gets modified and modified extent is first in the leaf,
1521 * then we have to correct all indexes above.
1522 * TODO: do we need to correct tree in all cases?
1524 static int ext4_ext_correct_indexes(handle_t
*handle
, struct inode
*inode
,
1525 struct ext4_ext_path
*path
)
1527 struct ext4_extent_header
*eh
;
1528 int depth
= ext_depth(inode
);
1529 struct ext4_extent
*ex
;
1533 eh
= path
[depth
].p_hdr
;
1534 ex
= path
[depth
].p_ext
;
1536 if (unlikely(ex
== NULL
|| eh
== NULL
)) {
1537 EXT4_ERROR_INODE(inode
,
1538 "ex %p == NULL or eh %p == NULL", ex
, eh
);
1543 /* there is no tree at all */
1547 if (ex
!= EXT_FIRST_EXTENT(eh
)) {
1548 /* we correct tree if first leaf got modified only */
1553 * TODO: we need correction if border is smaller than current one
1556 border
= path
[depth
].p_ext
->ee_block
;
1557 err
= ext4_ext_get_access(handle
, inode
, path
+ k
);
1560 path
[k
].p_idx
->ei_block
= border
;
1561 err
= ext4_ext_dirty(handle
, inode
, path
+ k
);
1566 /* change all left-side indexes */
1567 if (path
[k
+1].p_idx
!= EXT_FIRST_INDEX(path
[k
+1].p_hdr
))
1569 err
= ext4_ext_get_access(handle
, inode
, path
+ k
);
1572 path
[k
].p_idx
->ei_block
= border
;
1573 err
= ext4_ext_dirty(handle
, inode
, path
+ k
);
1582 ext4_can_extents_be_merged(struct inode
*inode
, struct ext4_extent
*ex1
,
1583 struct ext4_extent
*ex2
)
1585 unsigned short ext1_ee_len
, ext2_ee_len
, max_len
;
1588 * Make sure that either both extents are uninitialized, or
1591 if (ext4_ext_is_uninitialized(ex1
) ^ ext4_ext_is_uninitialized(ex2
))
1594 if (ext4_ext_is_uninitialized(ex1
))
1595 max_len
= EXT_UNINIT_MAX_LEN
;
1597 max_len
= EXT_INIT_MAX_LEN
;
1599 ext1_ee_len
= ext4_ext_get_actual_len(ex1
);
1600 ext2_ee_len
= ext4_ext_get_actual_len(ex2
);
1602 if (le32_to_cpu(ex1
->ee_block
) + ext1_ee_len
!=
1603 le32_to_cpu(ex2
->ee_block
))
1607 * To allow future support for preallocated extents to be added
1608 * as an RO_COMPAT feature, refuse to merge to extents if
1609 * this can result in the top bit of ee_len being set.
1611 if (ext1_ee_len
+ ext2_ee_len
> max_len
)
1613 #ifdef AGGRESSIVE_TEST
1614 if (ext1_ee_len
>= 4)
1618 if (ext4_ext_pblock(ex1
) + ext1_ee_len
== ext4_ext_pblock(ex2
))
1624 * This function tries to merge the "ex" extent to the next extent in the tree.
1625 * It always tries to merge towards right. If you want to merge towards
1626 * left, pass "ex - 1" as argument instead of "ex".
1627 * Returns 0 if the extents (ex and ex+1) were _not_ merged and returns
1628 * 1 if they got merged.
1630 static int ext4_ext_try_to_merge_right(struct inode
*inode
,
1631 struct ext4_ext_path
*path
,
1632 struct ext4_extent
*ex
)
1634 struct ext4_extent_header
*eh
;
1635 unsigned int depth
, len
;
1637 int uninitialized
= 0;
1639 depth
= ext_depth(inode
);
1640 BUG_ON(path
[depth
].p_hdr
== NULL
);
1641 eh
= path
[depth
].p_hdr
;
1643 while (ex
< EXT_LAST_EXTENT(eh
)) {
1644 if (!ext4_can_extents_be_merged(inode
, ex
, ex
+ 1))
1646 /* merge with next extent! */
1647 if (ext4_ext_is_uninitialized(ex
))
1649 ex
->ee_len
= cpu_to_le16(ext4_ext_get_actual_len(ex
)
1650 + ext4_ext_get_actual_len(ex
+ 1));
1652 ext4_ext_mark_uninitialized(ex
);
1654 if (ex
+ 1 < EXT_LAST_EXTENT(eh
)) {
1655 len
= (EXT_LAST_EXTENT(eh
) - ex
- 1)
1656 * sizeof(struct ext4_extent
);
1657 memmove(ex
+ 1, ex
+ 2, len
);
1659 le16_add_cpu(&eh
->eh_entries
, -1);
1661 WARN_ON(eh
->eh_entries
== 0);
1662 if (!eh
->eh_entries
)
1663 EXT4_ERROR_INODE(inode
, "eh->eh_entries = 0!");
1670 * This function does a very simple check to see if we can collapse
1671 * an extent tree with a single extent tree leaf block into the inode.
1673 static void ext4_ext_try_to_merge_up(handle_t
*handle
,
1674 struct inode
*inode
,
1675 struct ext4_ext_path
*path
)
1678 unsigned max_root
= ext4_ext_space_root(inode
, 0);
1681 if ((path
[0].p_depth
!= 1) ||
1682 (le16_to_cpu(path
[0].p_hdr
->eh_entries
) != 1) ||
1683 (le16_to_cpu(path
[1].p_hdr
->eh_entries
) > max_root
))
1687 * We need to modify the block allocation bitmap and the block
1688 * group descriptor to release the extent tree block. If we
1689 * can't get the journal credits, give up.
1691 if (ext4_journal_extend(handle
, 2))
1695 * Copy the extent data up to the inode
1697 blk
= ext4_idx_pblock(path
[0].p_idx
);
1698 s
= le16_to_cpu(path
[1].p_hdr
->eh_entries
) *
1699 sizeof(struct ext4_extent_idx
);
1700 s
+= sizeof(struct ext4_extent_header
);
1702 memcpy(path
[0].p_hdr
, path
[1].p_hdr
, s
);
1703 path
[0].p_depth
= 0;
1704 path
[0].p_ext
= EXT_FIRST_EXTENT(path
[0].p_hdr
) +
1705 (path
[1].p_ext
- EXT_FIRST_EXTENT(path
[1].p_hdr
));
1706 path
[0].p_hdr
->eh_max
= cpu_to_le16(max_root
);
1708 brelse(path
[1].p_bh
);
1709 ext4_free_blocks(handle
, inode
, NULL
, blk
, 1,
1710 EXT4_FREE_BLOCKS_METADATA
| EXT4_FREE_BLOCKS_FORGET
);
1714 * This function tries to merge the @ex extent to neighbours in the tree.
1715 * return 1 if merge left else 0.
1717 static void ext4_ext_try_to_merge(handle_t
*handle
,
1718 struct inode
*inode
,
1719 struct ext4_ext_path
*path
,
1720 struct ext4_extent
*ex
) {
1721 struct ext4_extent_header
*eh
;
1725 depth
= ext_depth(inode
);
1726 BUG_ON(path
[depth
].p_hdr
== NULL
);
1727 eh
= path
[depth
].p_hdr
;
1729 if (ex
> EXT_FIRST_EXTENT(eh
))
1730 merge_done
= ext4_ext_try_to_merge_right(inode
, path
, ex
- 1);
1733 (void) ext4_ext_try_to_merge_right(inode
, path
, ex
);
1735 ext4_ext_try_to_merge_up(handle
, inode
, path
);
1739 * check if a portion of the "newext" extent overlaps with an
1742 * If there is an overlap discovered, it updates the length of the newext
1743 * such that there will be no overlap, and then returns 1.
1744 * If there is no overlap found, it returns 0.
1746 static unsigned int ext4_ext_check_overlap(struct ext4_sb_info
*sbi
,
1747 struct inode
*inode
,
1748 struct ext4_extent
*newext
,
1749 struct ext4_ext_path
*path
)
1752 unsigned int depth
, len1
;
1753 unsigned int ret
= 0;
1755 b1
= le32_to_cpu(newext
->ee_block
);
1756 len1
= ext4_ext_get_actual_len(newext
);
1757 depth
= ext_depth(inode
);
1758 if (!path
[depth
].p_ext
)
1760 b2
= le32_to_cpu(path
[depth
].p_ext
->ee_block
);
1761 b2
&= ~(sbi
->s_cluster_ratio
- 1);
1764 * get the next allocated block if the extent in the path
1765 * is before the requested block(s)
1768 b2
= ext4_ext_next_allocated_block(path
);
1769 if (b2
== EXT_MAX_BLOCKS
)
1771 b2
&= ~(sbi
->s_cluster_ratio
- 1);
1774 /* check for wrap through zero on extent logical start block*/
1775 if (b1
+ len1
< b1
) {
1776 len1
= EXT_MAX_BLOCKS
- b1
;
1777 newext
->ee_len
= cpu_to_le16(len1
);
1781 /* check for overlap */
1782 if (b1
+ len1
> b2
) {
1783 newext
->ee_len
= cpu_to_le16(b2
- b1
);
1791 * ext4_ext_insert_extent:
1792 * tries to merge requsted extent into the existing extent or
1793 * inserts requested extent as new one into the tree,
1794 * creating new leaf in the no-space case.
1796 int ext4_ext_insert_extent(handle_t
*handle
, struct inode
*inode
,
1797 struct ext4_ext_path
*path
,
1798 struct ext4_extent
*newext
, int flag
)
1800 struct ext4_extent_header
*eh
;
1801 struct ext4_extent
*ex
, *fex
;
1802 struct ext4_extent
*nearex
; /* nearest extent */
1803 struct ext4_ext_path
*npath
= NULL
;
1804 int depth
, len
, err
;
1806 unsigned uninitialized
= 0;
1809 if (unlikely(ext4_ext_get_actual_len(newext
) == 0)) {
1810 EXT4_ERROR_INODE(inode
, "ext4_ext_get_actual_len(newext) == 0");
1813 depth
= ext_depth(inode
);
1814 ex
= path
[depth
].p_ext
;
1815 if (unlikely(path
[depth
].p_hdr
== NULL
)) {
1816 EXT4_ERROR_INODE(inode
, "path[%d].p_hdr == NULL", depth
);
1820 /* try to insert block into found extent and return */
1821 if (ex
&& !(flag
& EXT4_GET_BLOCKS_PRE_IO
)
1822 && ext4_can_extents_be_merged(inode
, ex
, newext
)) {
1823 ext_debug("append [%d]%d block to %u:[%d]%d (from %llu)\n",
1824 ext4_ext_is_uninitialized(newext
),
1825 ext4_ext_get_actual_len(newext
),
1826 le32_to_cpu(ex
->ee_block
),
1827 ext4_ext_is_uninitialized(ex
),
1828 ext4_ext_get_actual_len(ex
),
1829 ext4_ext_pblock(ex
));
1830 err
= ext4_ext_get_access(handle
, inode
, path
+ depth
);
1835 * ext4_can_extents_be_merged should have checked that either
1836 * both extents are uninitialized, or both aren't. Thus we
1837 * need to check only one of them here.
1839 if (ext4_ext_is_uninitialized(ex
))
1841 ex
->ee_len
= cpu_to_le16(ext4_ext_get_actual_len(ex
)
1842 + ext4_ext_get_actual_len(newext
));
1844 ext4_ext_mark_uninitialized(ex
);
1845 eh
= path
[depth
].p_hdr
;
1850 depth
= ext_depth(inode
);
1851 eh
= path
[depth
].p_hdr
;
1852 if (le16_to_cpu(eh
->eh_entries
) < le16_to_cpu(eh
->eh_max
))
1855 /* probably next leaf has space for us? */
1856 fex
= EXT_LAST_EXTENT(eh
);
1857 next
= EXT_MAX_BLOCKS
;
1858 if (le32_to_cpu(newext
->ee_block
) > le32_to_cpu(fex
->ee_block
))
1859 next
= ext4_ext_next_leaf_block(path
);
1860 if (next
!= EXT_MAX_BLOCKS
) {
1861 ext_debug("next leaf block - %u\n", next
);
1862 BUG_ON(npath
!= NULL
);
1863 npath
= ext4_ext_find_extent(inode
, next
, NULL
);
1865 return PTR_ERR(npath
);
1866 BUG_ON(npath
->p_depth
!= path
->p_depth
);
1867 eh
= npath
[depth
].p_hdr
;
1868 if (le16_to_cpu(eh
->eh_entries
) < le16_to_cpu(eh
->eh_max
)) {
1869 ext_debug("next leaf isn't full(%d)\n",
1870 le16_to_cpu(eh
->eh_entries
));
1874 ext_debug("next leaf has no free space(%d,%d)\n",
1875 le16_to_cpu(eh
->eh_entries
), le16_to_cpu(eh
->eh_max
));
1879 * There is no free space in the found leaf.
1880 * We're gonna add a new leaf in the tree.
1882 if (flag
& EXT4_GET_BLOCKS_PUNCH_OUT_EXT
)
1883 flags
= EXT4_MB_USE_ROOT_BLOCKS
;
1884 err
= ext4_ext_create_new_leaf(handle
, inode
, flags
, path
, newext
);
1887 depth
= ext_depth(inode
);
1888 eh
= path
[depth
].p_hdr
;
1891 nearex
= path
[depth
].p_ext
;
1893 err
= ext4_ext_get_access(handle
, inode
, path
+ depth
);
1898 /* there is no extent in this leaf, create first one */
1899 ext_debug("first extent in the leaf: %u:%llu:[%d]%d\n",
1900 le32_to_cpu(newext
->ee_block
),
1901 ext4_ext_pblock(newext
),
1902 ext4_ext_is_uninitialized(newext
),
1903 ext4_ext_get_actual_len(newext
));
1904 nearex
= EXT_FIRST_EXTENT(eh
);
1906 if (le32_to_cpu(newext
->ee_block
)
1907 > le32_to_cpu(nearex
->ee_block
)) {
1909 ext_debug("insert %u:%llu:[%d]%d before: "
1911 le32_to_cpu(newext
->ee_block
),
1912 ext4_ext_pblock(newext
),
1913 ext4_ext_is_uninitialized(newext
),
1914 ext4_ext_get_actual_len(newext
),
1919 BUG_ON(newext
->ee_block
== nearex
->ee_block
);
1920 ext_debug("insert %u:%llu:[%d]%d after: "
1922 le32_to_cpu(newext
->ee_block
),
1923 ext4_ext_pblock(newext
),
1924 ext4_ext_is_uninitialized(newext
),
1925 ext4_ext_get_actual_len(newext
),
1928 len
= EXT_LAST_EXTENT(eh
) - nearex
+ 1;
1930 ext_debug("insert %u:%llu:[%d]%d: "
1931 "move %d extents from 0x%p to 0x%p\n",
1932 le32_to_cpu(newext
->ee_block
),
1933 ext4_ext_pblock(newext
),
1934 ext4_ext_is_uninitialized(newext
),
1935 ext4_ext_get_actual_len(newext
),
1936 len
, nearex
, nearex
+ 1);
1937 memmove(nearex
+ 1, nearex
,
1938 len
* sizeof(struct ext4_extent
));
1942 le16_add_cpu(&eh
->eh_entries
, 1);
1943 path
[depth
].p_ext
= nearex
;
1944 nearex
->ee_block
= newext
->ee_block
;
1945 ext4_ext_store_pblock(nearex
, ext4_ext_pblock(newext
));
1946 nearex
->ee_len
= newext
->ee_len
;
1949 /* try to merge extents */
1950 if (!(flag
& EXT4_GET_BLOCKS_PRE_IO
))
1951 ext4_ext_try_to_merge(handle
, inode
, path
, nearex
);
1954 /* time to correct all indexes above */
1955 err
= ext4_ext_correct_indexes(handle
, inode
, path
);
1959 err
= ext4_ext_dirty(handle
, inode
, path
+ path
->p_depth
);
1963 ext4_ext_drop_refs(npath
);
1966 ext4_ext_invalidate_cache(inode
);
1970 static int ext4_fill_fiemap_extents(struct inode
*inode
,
1971 ext4_lblk_t block
, ext4_lblk_t num
,
1972 struct fiemap_extent_info
*fieinfo
)
1974 struct ext4_ext_path
*path
= NULL
;
1975 struct ext4_ext_cache newex
;
1976 struct ext4_extent
*ex
;
1977 ext4_lblk_t next
, next_del
, start
= 0, end
= 0;
1978 ext4_lblk_t last
= block
+ num
;
1979 int exists
, depth
= 0, err
= 0;
1980 unsigned int flags
= 0;
1981 unsigned char blksize_bits
= inode
->i_sb
->s_blocksize_bits
;
1983 while (block
< last
&& block
!= EXT_MAX_BLOCKS
) {
1985 /* find extent for this block */
1986 down_read(&EXT4_I(inode
)->i_data_sem
);
1988 if (path
&& ext_depth(inode
) != depth
) {
1989 /* depth was changed. we have to realloc path */
1994 path
= ext4_ext_find_extent(inode
, block
, path
);
1996 up_read(&EXT4_I(inode
)->i_data_sem
);
1997 err
= PTR_ERR(path
);
2002 depth
= ext_depth(inode
);
2003 if (unlikely(path
[depth
].p_hdr
== NULL
)) {
2004 up_read(&EXT4_I(inode
)->i_data_sem
);
2005 EXT4_ERROR_INODE(inode
, "path[%d].p_hdr == NULL", depth
);
2009 ex
= path
[depth
].p_ext
;
2010 next
= ext4_ext_next_allocated_block(path
);
2011 ext4_ext_drop_refs(path
);
2016 /* there is no extent yet, so try to allocate
2017 * all requested space */
2020 } else if (le32_to_cpu(ex
->ee_block
) > block
) {
2021 /* need to allocate space before found extent */
2023 end
= le32_to_cpu(ex
->ee_block
);
2024 if (block
+ num
< end
)
2026 } else if (block
>= le32_to_cpu(ex
->ee_block
)
2027 + ext4_ext_get_actual_len(ex
)) {
2028 /* need to allocate space after found extent */
2033 } else if (block
>= le32_to_cpu(ex
->ee_block
)) {
2035 * some part of requested space is covered
2039 end
= le32_to_cpu(ex
->ee_block
)
2040 + ext4_ext_get_actual_len(ex
);
2041 if (block
+ num
< end
)
2047 BUG_ON(end
<= start
);
2050 newex
.ec_block
= start
;
2051 newex
.ec_len
= end
- start
;
2054 newex
.ec_block
= le32_to_cpu(ex
->ee_block
);
2055 newex
.ec_len
= ext4_ext_get_actual_len(ex
);
2056 newex
.ec_start
= ext4_ext_pblock(ex
);
2057 if (ext4_ext_is_uninitialized(ex
))
2058 flags
|= FIEMAP_EXTENT_UNWRITTEN
;
2062 * Find delayed extent and update newex accordingly. We call
2063 * it even in !exists case to find out whether newex is the
2064 * last existing extent or not.
2066 next_del
= ext4_find_delayed_extent(inode
, &newex
);
2067 if (!exists
&& next_del
) {
2069 flags
|= FIEMAP_EXTENT_DELALLOC
;
2071 up_read(&EXT4_I(inode
)->i_data_sem
);
2073 if (unlikely(newex
.ec_len
== 0)) {
2074 EXT4_ERROR_INODE(inode
, "newex.ec_len == 0");
2079 /* This is possible iff next == next_del == EXT_MAX_BLOCKS */
2080 if (next
== next_del
) {
2081 flags
|= FIEMAP_EXTENT_LAST
;
2082 if (unlikely(next_del
!= EXT_MAX_BLOCKS
||
2083 next
!= EXT_MAX_BLOCKS
)) {
2084 EXT4_ERROR_INODE(inode
,
2085 "next extent == %u, next "
2086 "delalloc extent = %u",
2094 err
= fiemap_fill_next_extent(fieinfo
,
2095 (__u64
)newex
.ec_block
<< blksize_bits
,
2096 (__u64
)newex
.ec_start
<< blksize_bits
,
2097 (__u64
)newex
.ec_len
<< blksize_bits
,
2107 block
= newex
.ec_block
+ newex
.ec_len
;
2111 ext4_ext_drop_refs(path
);
2119 ext4_ext_put_in_cache(struct inode
*inode
, ext4_lblk_t block
,
2120 __u32 len
, ext4_fsblk_t start
)
2122 struct ext4_ext_cache
*cex
;
2124 spin_lock(&EXT4_I(inode
)->i_block_reservation_lock
);
2125 trace_ext4_ext_put_in_cache(inode
, block
, len
, start
);
2126 cex
= &EXT4_I(inode
)->i_cached_extent
;
2127 cex
->ec_block
= block
;
2129 cex
->ec_start
= start
;
2130 spin_unlock(&EXT4_I(inode
)->i_block_reservation_lock
);
2134 * ext4_ext_put_gap_in_cache:
2135 * calculate boundaries of the gap that the requested block fits into
2136 * and cache this gap
2139 ext4_ext_put_gap_in_cache(struct inode
*inode
, struct ext4_ext_path
*path
,
2142 int depth
= ext_depth(inode
);
2145 struct ext4_extent
*ex
;
2147 ex
= path
[depth
].p_ext
;
2149 /* there is no extent yet, so gap is [0;-] */
2151 len
= EXT_MAX_BLOCKS
;
2152 ext_debug("cache gap(whole file):");
2153 } else if (block
< le32_to_cpu(ex
->ee_block
)) {
2155 len
= le32_to_cpu(ex
->ee_block
) - block
;
2156 ext_debug("cache gap(before): %u [%u:%u]",
2158 le32_to_cpu(ex
->ee_block
),
2159 ext4_ext_get_actual_len(ex
));
2160 } else if (block
>= le32_to_cpu(ex
->ee_block
)
2161 + ext4_ext_get_actual_len(ex
)) {
2163 lblock
= le32_to_cpu(ex
->ee_block
)
2164 + ext4_ext_get_actual_len(ex
);
2166 next
= ext4_ext_next_allocated_block(path
);
2167 ext_debug("cache gap(after): [%u:%u] %u",
2168 le32_to_cpu(ex
->ee_block
),
2169 ext4_ext_get_actual_len(ex
),
2171 BUG_ON(next
== lblock
);
2172 len
= next
- lblock
;
2178 ext_debug(" -> %u:%lu\n", lblock
, len
);
2179 ext4_ext_put_in_cache(inode
, lblock
, len
, 0);
2183 * ext4_ext_in_cache()
2184 * Checks to see if the given block is in the cache.
2185 * If it is, the cached extent is stored in the given
2186 * cache extent pointer.
2188 * @inode: The files inode
2189 * @block: The block to look for in the cache
2190 * @ex: Pointer where the cached extent will be stored
2191 * if it contains block
2193 * Return 0 if cache is invalid; 1 if the cache is valid
2196 ext4_ext_in_cache(struct inode
*inode
, ext4_lblk_t block
,
2197 struct ext4_extent
*ex
)
2199 struct ext4_ext_cache
*cex
;
2203 * We borrow i_block_reservation_lock to protect i_cached_extent
2205 spin_lock(&EXT4_I(inode
)->i_block_reservation_lock
);
2206 cex
= &EXT4_I(inode
)->i_cached_extent
;
2208 /* has cache valid data? */
2209 if (cex
->ec_len
== 0)
2212 if (in_range(block
, cex
->ec_block
, cex
->ec_len
)) {
2213 ex
->ee_block
= cpu_to_le32(cex
->ec_block
);
2214 ext4_ext_store_pblock(ex
, cex
->ec_start
);
2215 ex
->ee_len
= cpu_to_le16(cex
->ec_len
);
2216 ext_debug("%u cached by %u:%u:%llu\n",
2218 cex
->ec_block
, cex
->ec_len
, cex
->ec_start
);
2222 trace_ext4_ext_in_cache(inode
, block
, ret
);
2223 spin_unlock(&EXT4_I(inode
)->i_block_reservation_lock
);
2229 * removes index from the index block.
2231 static int ext4_ext_rm_idx(handle_t
*handle
, struct inode
*inode
,
2232 struct ext4_ext_path
*path
, int depth
)
2237 /* free index block */
2239 path
= path
+ depth
;
2240 leaf
= ext4_idx_pblock(path
->p_idx
);
2241 if (unlikely(path
->p_hdr
->eh_entries
== 0)) {
2242 EXT4_ERROR_INODE(inode
, "path->p_hdr->eh_entries == 0");
2245 err
= ext4_ext_get_access(handle
, inode
, path
);
2249 if (path
->p_idx
!= EXT_LAST_INDEX(path
->p_hdr
)) {
2250 int len
= EXT_LAST_INDEX(path
->p_hdr
) - path
->p_idx
;
2251 len
*= sizeof(struct ext4_extent_idx
);
2252 memmove(path
->p_idx
, path
->p_idx
+ 1, len
);
2255 le16_add_cpu(&path
->p_hdr
->eh_entries
, -1);
2256 err
= ext4_ext_dirty(handle
, inode
, path
);
2259 ext_debug("index is empty, remove it, free block %llu\n", leaf
);
2260 trace_ext4_ext_rm_idx(inode
, leaf
);
2262 ext4_free_blocks(handle
, inode
, NULL
, leaf
, 1,
2263 EXT4_FREE_BLOCKS_METADATA
| EXT4_FREE_BLOCKS_FORGET
);
2265 while (--depth
>= 0) {
2266 if (path
->p_idx
!= EXT_FIRST_INDEX(path
->p_hdr
))
2269 err
= ext4_ext_get_access(handle
, inode
, path
);
2272 path
->p_idx
->ei_block
= (path
+1)->p_idx
->ei_block
;
2273 err
= ext4_ext_dirty(handle
, inode
, path
);
2281 * ext4_ext_calc_credits_for_single_extent:
2282 * This routine returns max. credits that needed to insert an extent
2283 * to the extent tree.
2284 * When pass the actual path, the caller should calculate credits
2287 int ext4_ext_calc_credits_for_single_extent(struct inode
*inode
, int nrblocks
,
2288 struct ext4_ext_path
*path
)
2291 int depth
= ext_depth(inode
);
2294 /* probably there is space in leaf? */
2295 if (le16_to_cpu(path
[depth
].p_hdr
->eh_entries
)
2296 < le16_to_cpu(path
[depth
].p_hdr
->eh_max
)) {
2299 * There are some space in the leaf tree, no
2300 * need to account for leaf block credit
2302 * bitmaps and block group descriptor blocks
2303 * and other metadata blocks still need to be
2306 /* 1 bitmap, 1 block group descriptor */
2307 ret
= 2 + EXT4_META_TRANS_BLOCKS(inode
->i_sb
);
2312 return ext4_chunk_trans_blocks(inode
, nrblocks
);
2316 * How many index/leaf blocks need to change/allocate to modify nrblocks?
2318 * if nrblocks are fit in a single extent (chunk flag is 1), then
2319 * in the worse case, each tree level index/leaf need to be changed
2320 * if the tree split due to insert a new extent, then the old tree
2321 * index/leaf need to be updated too
2323 * If the nrblocks are discontiguous, they could cause
2324 * the whole tree split more than once, but this is really rare.
2326 int ext4_ext_index_trans_blocks(struct inode
*inode
, int nrblocks
, int chunk
)
2331 /* If we are converting the inline data, only one is needed here. */
2332 if (ext4_has_inline_data(inode
))
2335 depth
= ext_depth(inode
);
2345 static int ext4_remove_blocks(handle_t
*handle
, struct inode
*inode
,
2346 struct ext4_extent
*ex
,
2347 ext4_fsblk_t
*partial_cluster
,
2348 ext4_lblk_t from
, ext4_lblk_t to
)
2350 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
2351 unsigned short ee_len
= ext4_ext_get_actual_len(ex
);
2355 if (S_ISDIR(inode
->i_mode
) || S_ISLNK(inode
->i_mode
))
2356 flags
|= EXT4_FREE_BLOCKS_METADATA
| EXT4_FREE_BLOCKS_FORGET
;
2357 else if (ext4_should_journal_data(inode
))
2358 flags
|= EXT4_FREE_BLOCKS_FORGET
;
2361 * For bigalloc file systems, we never free a partial cluster
2362 * at the beginning of the extent. Instead, we make a note
2363 * that we tried freeing the cluster, and check to see if we
2364 * need to free it on a subsequent call to ext4_remove_blocks,
2365 * or at the end of the ext4_truncate() operation.
2367 flags
|= EXT4_FREE_BLOCKS_NOFREE_FIRST_CLUSTER
;
2369 trace_ext4_remove_blocks(inode
, ex
, from
, to
, *partial_cluster
);
2371 * If we have a partial cluster, and it's different from the
2372 * cluster of the last block, we need to explicitly free the
2373 * partial cluster here.
2375 pblk
= ext4_ext_pblock(ex
) + ee_len
- 1;
2376 if (*partial_cluster
&& (EXT4_B2C(sbi
, pblk
) != *partial_cluster
)) {
2377 ext4_free_blocks(handle
, inode
, NULL
,
2378 EXT4_C2B(sbi
, *partial_cluster
),
2379 sbi
->s_cluster_ratio
, flags
);
2380 *partial_cluster
= 0;
2383 #ifdef EXTENTS_STATS
2385 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
2386 spin_lock(&sbi
->s_ext_stats_lock
);
2387 sbi
->s_ext_blocks
+= ee_len
;
2388 sbi
->s_ext_extents
++;
2389 if (ee_len
< sbi
->s_ext_min
)
2390 sbi
->s_ext_min
= ee_len
;
2391 if (ee_len
> sbi
->s_ext_max
)
2392 sbi
->s_ext_max
= ee_len
;
2393 if (ext_depth(inode
) > sbi
->s_depth_max
)
2394 sbi
->s_depth_max
= ext_depth(inode
);
2395 spin_unlock(&sbi
->s_ext_stats_lock
);
2398 if (from
>= le32_to_cpu(ex
->ee_block
)
2399 && to
== le32_to_cpu(ex
->ee_block
) + ee_len
- 1) {
2403 num
= le32_to_cpu(ex
->ee_block
) + ee_len
- from
;
2404 pblk
= ext4_ext_pblock(ex
) + ee_len
- num
;
2405 ext_debug("free last %u blocks starting %llu\n", num
, pblk
);
2406 ext4_free_blocks(handle
, inode
, NULL
, pblk
, num
, flags
);
2408 * If the block range to be freed didn't start at the
2409 * beginning of a cluster, and we removed the entire
2410 * extent, save the partial cluster here, since we
2411 * might need to delete if we determine that the
2412 * truncate operation has removed all of the blocks in
2415 if (pblk
& (sbi
->s_cluster_ratio
- 1) &&
2417 *partial_cluster
= EXT4_B2C(sbi
, pblk
);
2419 *partial_cluster
= 0;
2420 } else if (from
== le32_to_cpu(ex
->ee_block
)
2421 && to
<= le32_to_cpu(ex
->ee_block
) + ee_len
- 1) {
2427 start
= ext4_ext_pblock(ex
);
2429 ext_debug("free first %u blocks starting %llu\n", num
, start
);
2430 ext4_free_blocks(handle
, inode
, NULL
, start
, num
, flags
);
2433 printk(KERN_INFO
"strange request: removal(2) "
2434 "%u-%u from %u:%u\n",
2435 from
, to
, le32_to_cpu(ex
->ee_block
), ee_len
);
2442 * ext4_ext_rm_leaf() Removes the extents associated with the
2443 * blocks appearing between "start" and "end", and splits the extents
2444 * if "start" and "end" appear in the same extent
2446 * @handle: The journal handle
2447 * @inode: The files inode
2448 * @path: The path to the leaf
2449 * @start: The first block to remove
2450 * @end: The last block to remove
2453 ext4_ext_rm_leaf(handle_t
*handle
, struct inode
*inode
,
2454 struct ext4_ext_path
*path
, ext4_fsblk_t
*partial_cluster
,
2455 ext4_lblk_t start
, ext4_lblk_t end
)
2457 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
2458 int err
= 0, correct_index
= 0;
2459 int depth
= ext_depth(inode
), credits
;
2460 struct ext4_extent_header
*eh
;
2463 ext4_lblk_t ex_ee_block
;
2464 unsigned short ex_ee_len
;
2465 unsigned uninitialized
= 0;
2466 struct ext4_extent
*ex
;
2468 /* the header must be checked already in ext4_ext_remove_space() */
2469 ext_debug("truncate since %u in leaf to %u\n", start
, end
);
2470 if (!path
[depth
].p_hdr
)
2471 path
[depth
].p_hdr
= ext_block_hdr(path
[depth
].p_bh
);
2472 eh
= path
[depth
].p_hdr
;
2473 if (unlikely(path
[depth
].p_hdr
== NULL
)) {
2474 EXT4_ERROR_INODE(inode
, "path[%d].p_hdr == NULL", depth
);
2477 /* find where to start removing */
2478 ex
= EXT_LAST_EXTENT(eh
);
2480 ex_ee_block
= le32_to_cpu(ex
->ee_block
);
2481 ex_ee_len
= ext4_ext_get_actual_len(ex
);
2483 trace_ext4_ext_rm_leaf(inode
, start
, ex
, *partial_cluster
);
2485 while (ex
>= EXT_FIRST_EXTENT(eh
) &&
2486 ex_ee_block
+ ex_ee_len
> start
) {
2488 if (ext4_ext_is_uninitialized(ex
))
2493 ext_debug("remove ext %u:[%d]%d\n", ex_ee_block
,
2494 uninitialized
, ex_ee_len
);
2495 path
[depth
].p_ext
= ex
;
2497 a
= ex_ee_block
> start
? ex_ee_block
: start
;
2498 b
= ex_ee_block
+ex_ee_len
- 1 < end
?
2499 ex_ee_block
+ex_ee_len
- 1 : end
;
2501 ext_debug(" border %u:%u\n", a
, b
);
2503 /* If this extent is beyond the end of the hole, skip it */
2504 if (end
< ex_ee_block
) {
2506 ex_ee_block
= le32_to_cpu(ex
->ee_block
);
2507 ex_ee_len
= ext4_ext_get_actual_len(ex
);
2509 } else if (b
!= ex_ee_block
+ ex_ee_len
- 1) {
2510 EXT4_ERROR_INODE(inode
,
2511 "can not handle truncate %u:%u "
2513 start
, end
, ex_ee_block
,
2514 ex_ee_block
+ ex_ee_len
- 1);
2517 } else if (a
!= ex_ee_block
) {
2518 /* remove tail of the extent */
2519 num
= a
- ex_ee_block
;
2521 /* remove whole extent: excellent! */
2525 * 3 for leaf, sb, and inode plus 2 (bmap and group
2526 * descriptor) for each block group; assume two block
2527 * groups plus ex_ee_len/blocks_per_block_group for
2530 credits
= 7 + 2*(ex_ee_len
/EXT4_BLOCKS_PER_GROUP(inode
->i_sb
));
2531 if (ex
== EXT_FIRST_EXTENT(eh
)) {
2533 credits
+= (ext_depth(inode
)) + 1;
2535 credits
+= EXT4_MAXQUOTAS_TRANS_BLOCKS(inode
->i_sb
);
2537 err
= ext4_ext_truncate_extend_restart(handle
, inode
, credits
);
2541 err
= ext4_ext_get_access(handle
, inode
, path
+ depth
);
2545 err
= ext4_remove_blocks(handle
, inode
, ex
, partial_cluster
,
2551 /* this extent is removed; mark slot entirely unused */
2552 ext4_ext_store_pblock(ex
, 0);
2554 ex
->ee_len
= cpu_to_le16(num
);
2556 * Do not mark uninitialized if all the blocks in the
2557 * extent have been removed.
2559 if (uninitialized
&& num
)
2560 ext4_ext_mark_uninitialized(ex
);
2562 * If the extent was completely released,
2563 * we need to remove it from the leaf
2566 if (end
!= EXT_MAX_BLOCKS
- 1) {
2568 * For hole punching, we need to scoot all the
2569 * extents up when an extent is removed so that
2570 * we dont have blank extents in the middle
2572 memmove(ex
, ex
+1, (EXT_LAST_EXTENT(eh
) - ex
) *
2573 sizeof(struct ext4_extent
));
2575 /* Now get rid of the one at the end */
2576 memset(EXT_LAST_EXTENT(eh
), 0,
2577 sizeof(struct ext4_extent
));
2579 le16_add_cpu(&eh
->eh_entries
, -1);
2581 *partial_cluster
= 0;
2583 err
= ext4_ext_dirty(handle
, inode
, path
+ depth
);
2587 ext_debug("new extent: %u:%u:%llu\n", ex_ee_block
, num
,
2588 ext4_ext_pblock(ex
));
2590 ex_ee_block
= le32_to_cpu(ex
->ee_block
);
2591 ex_ee_len
= ext4_ext_get_actual_len(ex
);
2594 if (correct_index
&& eh
->eh_entries
)
2595 err
= ext4_ext_correct_indexes(handle
, inode
, path
);
2598 * If there is still a entry in the leaf node, check to see if
2599 * it references the partial cluster. This is the only place
2600 * where it could; if it doesn't, we can free the cluster.
2602 if (*partial_cluster
&& ex
>= EXT_FIRST_EXTENT(eh
) &&
2603 (EXT4_B2C(sbi
, ext4_ext_pblock(ex
) + ex_ee_len
- 1) !=
2604 *partial_cluster
)) {
2605 int flags
= EXT4_FREE_BLOCKS_FORGET
;
2607 if (S_ISDIR(inode
->i_mode
) || S_ISLNK(inode
->i_mode
))
2608 flags
|= EXT4_FREE_BLOCKS_METADATA
;
2610 ext4_free_blocks(handle
, inode
, NULL
,
2611 EXT4_C2B(sbi
, *partial_cluster
),
2612 sbi
->s_cluster_ratio
, flags
);
2613 *partial_cluster
= 0;
2616 /* if this leaf is free, then we should
2617 * remove it from index block above */
2618 if (err
== 0 && eh
->eh_entries
== 0 && path
[depth
].p_bh
!= NULL
)
2619 err
= ext4_ext_rm_idx(handle
, inode
, path
, depth
);
2626 * ext4_ext_more_to_rm:
2627 * returns 1 if current index has to be freed (even partial)
2630 ext4_ext_more_to_rm(struct ext4_ext_path
*path
)
2632 BUG_ON(path
->p_idx
== NULL
);
2634 if (path
->p_idx
< EXT_FIRST_INDEX(path
->p_hdr
))
2638 * if truncate on deeper level happened, it wasn't partial,
2639 * so we have to consider current index for truncation
2641 if (le16_to_cpu(path
->p_hdr
->eh_entries
) == path
->p_block
)
2646 static int ext4_ext_remove_space(struct inode
*inode
, ext4_lblk_t start
,
2649 struct super_block
*sb
= inode
->i_sb
;
2650 int depth
= ext_depth(inode
);
2651 struct ext4_ext_path
*path
= NULL
;
2652 ext4_fsblk_t partial_cluster
= 0;
2656 ext_debug("truncate since %u to %u\n", start
, end
);
2658 /* probably first extent we're gonna free will be last in block */
2659 handle
= ext4_journal_start(inode
, depth
+ 1);
2661 return PTR_ERR(handle
);
2664 ext4_ext_invalidate_cache(inode
);
2666 trace_ext4_ext_remove_space(inode
, start
, depth
);
2669 * Check if we are removing extents inside the extent tree. If that
2670 * is the case, we are going to punch a hole inside the extent tree
2671 * so we have to check whether we need to split the extent covering
2672 * the last block to remove so we can easily remove the part of it
2673 * in ext4_ext_rm_leaf().
2675 if (end
< EXT_MAX_BLOCKS
- 1) {
2676 struct ext4_extent
*ex
;
2677 ext4_lblk_t ee_block
;
2679 /* find extent for this block */
2680 path
= ext4_ext_find_extent(inode
, end
, NULL
);
2682 ext4_journal_stop(handle
);
2683 return PTR_ERR(path
);
2685 depth
= ext_depth(inode
);
2686 /* Leaf not may not exist only if inode has no blocks at all */
2687 ex
= path
[depth
].p_ext
;
2690 EXT4_ERROR_INODE(inode
,
2691 "path[%d].p_hdr == NULL",
2698 ee_block
= le32_to_cpu(ex
->ee_block
);
2701 * See if the last block is inside the extent, if so split
2702 * the extent at 'end' block so we can easily remove the
2703 * tail of the first part of the split extent in
2704 * ext4_ext_rm_leaf().
2706 if (end
>= ee_block
&&
2707 end
< ee_block
+ ext4_ext_get_actual_len(ex
) - 1) {
2710 if (ext4_ext_is_uninitialized(ex
))
2711 split_flag
= EXT4_EXT_MARK_UNINIT1
|
2712 EXT4_EXT_MARK_UNINIT2
;
2715 * Split the extent in two so that 'end' is the last
2716 * block in the first new extent
2718 err
= ext4_split_extent_at(handle
, inode
, path
,
2719 end
+ 1, split_flag
,
2720 EXT4_GET_BLOCKS_PRE_IO
|
2721 EXT4_GET_BLOCKS_PUNCH_OUT_EXT
);
2728 * We start scanning from right side, freeing all the blocks
2729 * after i_size and walking into the tree depth-wise.
2731 depth
= ext_depth(inode
);
2736 le16_to_cpu(path
[k
].p_hdr
->eh_entries
)+1;
2738 path
= kzalloc(sizeof(struct ext4_ext_path
) * (depth
+ 1),
2741 ext4_journal_stop(handle
);
2744 path
[0].p_depth
= depth
;
2745 path
[0].p_hdr
= ext_inode_hdr(inode
);
2748 if (ext4_ext_check(inode
, path
[0].p_hdr
, depth
)) {
2755 while (i
>= 0 && err
== 0) {
2757 /* this is leaf block */
2758 err
= ext4_ext_rm_leaf(handle
, inode
, path
,
2759 &partial_cluster
, start
,
2761 /* root level has p_bh == NULL, brelse() eats this */
2762 brelse(path
[i
].p_bh
);
2763 path
[i
].p_bh
= NULL
;
2768 /* this is index block */
2769 if (!path
[i
].p_hdr
) {
2770 ext_debug("initialize header\n");
2771 path
[i
].p_hdr
= ext_block_hdr(path
[i
].p_bh
);
2774 if (!path
[i
].p_idx
) {
2775 /* this level hasn't been touched yet */
2776 path
[i
].p_idx
= EXT_LAST_INDEX(path
[i
].p_hdr
);
2777 path
[i
].p_block
= le16_to_cpu(path
[i
].p_hdr
->eh_entries
)+1;
2778 ext_debug("init index ptr: hdr 0x%p, num %d\n",
2780 le16_to_cpu(path
[i
].p_hdr
->eh_entries
));
2782 /* we were already here, see at next index */
2786 ext_debug("level %d - index, first 0x%p, cur 0x%p\n",
2787 i
, EXT_FIRST_INDEX(path
[i
].p_hdr
),
2789 if (ext4_ext_more_to_rm(path
+ i
)) {
2790 struct buffer_head
*bh
;
2791 /* go to the next level */
2792 ext_debug("move to level %d (block %llu)\n",
2793 i
+ 1, ext4_idx_pblock(path
[i
].p_idx
));
2794 memset(path
+ i
+ 1, 0, sizeof(*path
));
2795 bh
= sb_bread(sb
, ext4_idx_pblock(path
[i
].p_idx
));
2797 /* should we reset i_size? */
2801 if (WARN_ON(i
+ 1 > depth
)) {
2805 if (ext4_ext_check_block(inode
, ext_block_hdr(bh
),
2806 depth
- i
- 1, bh
)) {
2810 path
[i
+ 1].p_bh
= bh
;
2812 /* save actual number of indexes since this
2813 * number is changed at the next iteration */
2814 path
[i
].p_block
= le16_to_cpu(path
[i
].p_hdr
->eh_entries
);
2817 /* we finished processing this index, go up */
2818 if (path
[i
].p_hdr
->eh_entries
== 0 && i
> 0) {
2819 /* index is empty, remove it;
2820 * handle must be already prepared by the
2821 * truncatei_leaf() */
2822 err
= ext4_ext_rm_idx(handle
, inode
, path
, i
);
2824 /* root level has p_bh == NULL, brelse() eats this */
2825 brelse(path
[i
].p_bh
);
2826 path
[i
].p_bh
= NULL
;
2828 ext_debug("return to level %d\n", i
);
2832 trace_ext4_ext_remove_space_done(inode
, start
, depth
, partial_cluster
,
2833 path
->p_hdr
->eh_entries
);
2835 /* If we still have something in the partial cluster and we have removed
2836 * even the first extent, then we should free the blocks in the partial
2837 * cluster as well. */
2838 if (partial_cluster
&& path
->p_hdr
->eh_entries
== 0) {
2839 int flags
= EXT4_FREE_BLOCKS_FORGET
;
2841 if (S_ISDIR(inode
->i_mode
) || S_ISLNK(inode
->i_mode
))
2842 flags
|= EXT4_FREE_BLOCKS_METADATA
;
2844 ext4_free_blocks(handle
, inode
, NULL
,
2845 EXT4_C2B(EXT4_SB(sb
), partial_cluster
),
2846 EXT4_SB(sb
)->s_cluster_ratio
, flags
);
2847 partial_cluster
= 0;
2850 /* TODO: flexible tree reduction should be here */
2851 if (path
->p_hdr
->eh_entries
== 0) {
2853 * truncate to zero freed all the tree,
2854 * so we need to correct eh_depth
2856 err
= ext4_ext_get_access(handle
, inode
, path
);
2858 ext_inode_hdr(inode
)->eh_depth
= 0;
2859 ext_inode_hdr(inode
)->eh_max
=
2860 cpu_to_le16(ext4_ext_space_root(inode
, 0));
2861 err
= ext4_ext_dirty(handle
, inode
, path
);
2865 ext4_ext_drop_refs(path
);
2867 if (err
== -EAGAIN
) {
2871 ext4_journal_stop(handle
);
2877 * called at mount time
2879 void ext4_ext_init(struct super_block
*sb
)
2882 * possible initialization would be here
2885 if (EXT4_HAS_INCOMPAT_FEATURE(sb
, EXT4_FEATURE_INCOMPAT_EXTENTS
)) {
2886 #if defined(AGGRESSIVE_TEST) || defined(CHECK_BINSEARCH) || defined(EXTENTS_STATS)
2887 printk(KERN_INFO
"EXT4-fs: file extents enabled"
2888 #ifdef AGGRESSIVE_TEST
2889 ", aggressive tests"
2891 #ifdef CHECK_BINSEARCH
2894 #ifdef EXTENTS_STATS
2899 #ifdef EXTENTS_STATS
2900 spin_lock_init(&EXT4_SB(sb
)->s_ext_stats_lock
);
2901 EXT4_SB(sb
)->s_ext_min
= 1 << 30;
2902 EXT4_SB(sb
)->s_ext_max
= 0;
2908 * called at umount time
2910 void ext4_ext_release(struct super_block
*sb
)
2912 if (!EXT4_HAS_INCOMPAT_FEATURE(sb
, EXT4_FEATURE_INCOMPAT_EXTENTS
))
2915 #ifdef EXTENTS_STATS
2916 if (EXT4_SB(sb
)->s_ext_blocks
&& EXT4_SB(sb
)->s_ext_extents
) {
2917 struct ext4_sb_info
*sbi
= EXT4_SB(sb
);
2918 printk(KERN_ERR
"EXT4-fs: %lu blocks in %lu extents (%lu ave)\n",
2919 sbi
->s_ext_blocks
, sbi
->s_ext_extents
,
2920 sbi
->s_ext_blocks
/ sbi
->s_ext_extents
);
2921 printk(KERN_ERR
"EXT4-fs: extents: %lu min, %lu max, max depth %lu\n",
2922 sbi
->s_ext_min
, sbi
->s_ext_max
, sbi
->s_depth_max
);
2927 /* FIXME!! we need to try to merge to left or right after zero-out */
2928 static int ext4_ext_zeroout(struct inode
*inode
, struct ext4_extent
*ex
)
2930 ext4_fsblk_t ee_pblock
;
2931 unsigned int ee_len
;
2934 ee_len
= ext4_ext_get_actual_len(ex
);
2935 ee_pblock
= ext4_ext_pblock(ex
);
2937 ret
= sb_issue_zeroout(inode
->i_sb
, ee_pblock
, ee_len
, GFP_NOFS
);
2945 * ext4_split_extent_at() splits an extent at given block.
2947 * @handle: the journal handle
2948 * @inode: the file inode
2949 * @path: the path to the extent
2950 * @split: the logical block where the extent is splitted.
2951 * @split_flags: indicates if the extent could be zeroout if split fails, and
2952 * the states(init or uninit) of new extents.
2953 * @flags: flags used to insert new extent to extent tree.
2956 * Splits extent [a, b] into two extents [a, @split) and [@split, b], states
2957 * of which are deterimined by split_flag.
2959 * There are two cases:
2960 * a> the extent are splitted into two extent.
2961 * b> split is not needed, and just mark the extent.
2963 * return 0 on success.
2965 static int ext4_split_extent_at(handle_t
*handle
,
2966 struct inode
*inode
,
2967 struct ext4_ext_path
*path
,
2972 ext4_fsblk_t newblock
;
2973 ext4_lblk_t ee_block
;
2974 struct ext4_extent
*ex
, newex
, orig_ex
;
2975 struct ext4_extent
*ex2
= NULL
;
2976 unsigned int ee_len
, depth
;
2979 BUG_ON((split_flag
& (EXT4_EXT_DATA_VALID1
| EXT4_EXT_DATA_VALID2
)) ==
2980 (EXT4_EXT_DATA_VALID1
| EXT4_EXT_DATA_VALID2
));
2982 ext_debug("ext4_split_extents_at: inode %lu, logical"
2983 "block %llu\n", inode
->i_ino
, (unsigned long long)split
);
2985 ext4_ext_show_leaf(inode
, path
);
2987 depth
= ext_depth(inode
);
2988 ex
= path
[depth
].p_ext
;
2989 ee_block
= le32_to_cpu(ex
->ee_block
);
2990 ee_len
= ext4_ext_get_actual_len(ex
);
2991 newblock
= split
- ee_block
+ ext4_ext_pblock(ex
);
2993 BUG_ON(split
< ee_block
|| split
>= (ee_block
+ ee_len
));
2995 err
= ext4_ext_get_access(handle
, inode
, path
+ depth
);
2999 if (split
== ee_block
) {
3001 * case b: block @split is the block that the extent begins with
3002 * then we just change the state of the extent, and splitting
3005 if (split_flag
& EXT4_EXT_MARK_UNINIT2
)
3006 ext4_ext_mark_uninitialized(ex
);
3008 ext4_ext_mark_initialized(ex
);
3010 if (!(flags
& EXT4_GET_BLOCKS_PRE_IO
))
3011 ext4_ext_try_to_merge(handle
, inode
, path
, ex
);
3013 err
= ext4_ext_dirty(handle
, inode
, path
+ path
->p_depth
);
3018 memcpy(&orig_ex
, ex
, sizeof(orig_ex
));
3019 ex
->ee_len
= cpu_to_le16(split
- ee_block
);
3020 if (split_flag
& EXT4_EXT_MARK_UNINIT1
)
3021 ext4_ext_mark_uninitialized(ex
);
3024 * path may lead to new leaf, not to original leaf any more
3025 * after ext4_ext_insert_extent() returns,
3027 err
= ext4_ext_dirty(handle
, inode
, path
+ depth
);
3029 goto fix_extent_len
;
3032 ex2
->ee_block
= cpu_to_le32(split
);
3033 ex2
->ee_len
= cpu_to_le16(ee_len
- (split
- ee_block
));
3034 ext4_ext_store_pblock(ex2
, newblock
);
3035 if (split_flag
& EXT4_EXT_MARK_UNINIT2
)
3036 ext4_ext_mark_uninitialized(ex2
);
3038 err
= ext4_ext_insert_extent(handle
, inode
, path
, &newex
, flags
);
3039 if (err
== -ENOSPC
&& (EXT4_EXT_MAY_ZEROOUT
& split_flag
)) {
3040 if (split_flag
& (EXT4_EXT_DATA_VALID1
|EXT4_EXT_DATA_VALID2
)) {
3041 if (split_flag
& EXT4_EXT_DATA_VALID1
)
3042 err
= ext4_ext_zeroout(inode
, ex2
);
3044 err
= ext4_ext_zeroout(inode
, ex
);
3046 err
= ext4_ext_zeroout(inode
, &orig_ex
);
3049 goto fix_extent_len
;
3050 /* update the extent length and mark as initialized */
3051 ex
->ee_len
= cpu_to_le16(ee_len
);
3052 ext4_ext_try_to_merge(handle
, inode
, path
, ex
);
3053 err
= ext4_ext_dirty(handle
, inode
, path
+ path
->p_depth
);
3056 goto fix_extent_len
;
3059 ext4_ext_show_leaf(inode
, path
);
3063 ex
->ee_len
= orig_ex
.ee_len
;
3064 ext4_ext_dirty(handle
, inode
, path
+ depth
);
3069 * ext4_split_extents() splits an extent and mark extent which is covered
3070 * by @map as split_flags indicates
3072 * It may result in splitting the extent into multiple extents (upto three)
3073 * There are three possibilities:
3074 * a> There is no split required
3075 * b> Splits in two extents: Split is happening at either end of the extent
3076 * c> Splits in three extents: Somone is splitting in middle of the extent
3079 static int ext4_split_extent(handle_t
*handle
,
3080 struct inode
*inode
,
3081 struct ext4_ext_path
*path
,
3082 struct ext4_map_blocks
*map
,
3086 ext4_lblk_t ee_block
;
3087 struct ext4_extent
*ex
;
3088 unsigned int ee_len
, depth
;
3091 int split_flag1
, flags1
;
3093 depth
= ext_depth(inode
);
3094 ex
= path
[depth
].p_ext
;
3095 ee_block
= le32_to_cpu(ex
->ee_block
);
3096 ee_len
= ext4_ext_get_actual_len(ex
);
3097 uninitialized
= ext4_ext_is_uninitialized(ex
);
3099 if (map
->m_lblk
+ map
->m_len
< ee_block
+ ee_len
) {
3100 split_flag1
= split_flag
& EXT4_EXT_MAY_ZEROOUT
;
3101 flags1
= flags
| EXT4_GET_BLOCKS_PRE_IO
;
3103 split_flag1
|= EXT4_EXT_MARK_UNINIT1
|
3104 EXT4_EXT_MARK_UNINIT2
;
3105 if (split_flag
& EXT4_EXT_DATA_VALID2
)
3106 split_flag1
|= EXT4_EXT_DATA_VALID1
;
3107 err
= ext4_split_extent_at(handle
, inode
, path
,
3108 map
->m_lblk
+ map
->m_len
, split_flag1
, flags1
);
3113 ext4_ext_drop_refs(path
);
3114 path
= ext4_ext_find_extent(inode
, map
->m_lblk
, path
);
3116 return PTR_ERR(path
);
3118 if (map
->m_lblk
>= ee_block
) {
3119 split_flag1
= split_flag
& (EXT4_EXT_MAY_ZEROOUT
|
3120 EXT4_EXT_DATA_VALID2
);
3122 split_flag1
|= EXT4_EXT_MARK_UNINIT1
;
3123 if (split_flag
& EXT4_EXT_MARK_UNINIT2
)
3124 split_flag1
|= EXT4_EXT_MARK_UNINIT2
;
3125 err
= ext4_split_extent_at(handle
, inode
, path
,
3126 map
->m_lblk
, split_flag1
, flags
);
3131 ext4_ext_show_leaf(inode
, path
);
3133 return err
? err
: map
->m_len
;
3137 * This function is called by ext4_ext_map_blocks() if someone tries to write
3138 * to an uninitialized extent. It may result in splitting the uninitialized
3139 * extent into multiple extents (up to three - one initialized and two
3141 * There are three possibilities:
3142 * a> There is no split required: Entire extent should be initialized
3143 * b> Splits in two extents: Write is happening at either end of the extent
3144 * c> Splits in three extents: Somone is writing in middle of the extent
3147 * - The extent pointed to by 'path' is uninitialized.
3148 * - The extent pointed to by 'path' contains a superset
3149 * of the logical span [map->m_lblk, map->m_lblk + map->m_len).
3151 * Post-conditions on success:
3152 * - the returned value is the number of blocks beyond map->l_lblk
3153 * that are allocated and initialized.
3154 * It is guaranteed to be >= map->m_len.
3156 static int ext4_ext_convert_to_initialized(handle_t
*handle
,
3157 struct inode
*inode
,
3158 struct ext4_map_blocks
*map
,
3159 struct ext4_ext_path
*path
)
3161 struct ext4_sb_info
*sbi
;
3162 struct ext4_extent_header
*eh
;
3163 struct ext4_map_blocks split_map
;
3164 struct ext4_extent zero_ex
;
3165 struct ext4_extent
*ex
;
3166 ext4_lblk_t ee_block
, eof_block
;
3167 unsigned int ee_len
, depth
;
3168 int allocated
, max_zeroout
= 0;
3172 ext_debug("ext4_ext_convert_to_initialized: inode %lu, logical"
3173 "block %llu, max_blocks %u\n", inode
->i_ino
,
3174 (unsigned long long)map
->m_lblk
, map
->m_len
);
3176 sbi
= EXT4_SB(inode
->i_sb
);
3177 eof_block
= (inode
->i_size
+ inode
->i_sb
->s_blocksize
- 1) >>
3178 inode
->i_sb
->s_blocksize_bits
;
3179 if (eof_block
< map
->m_lblk
+ map
->m_len
)
3180 eof_block
= map
->m_lblk
+ map
->m_len
;
3182 depth
= ext_depth(inode
);
3183 eh
= path
[depth
].p_hdr
;
3184 ex
= path
[depth
].p_ext
;
3185 ee_block
= le32_to_cpu(ex
->ee_block
);
3186 ee_len
= ext4_ext_get_actual_len(ex
);
3187 allocated
= ee_len
- (map
->m_lblk
- ee_block
);
3189 trace_ext4_ext_convert_to_initialized_enter(inode
, map
, ex
);
3191 /* Pre-conditions */
3192 BUG_ON(!ext4_ext_is_uninitialized(ex
));
3193 BUG_ON(!in_range(map
->m_lblk
, ee_block
, ee_len
));
3196 * Attempt to transfer newly initialized blocks from the currently
3197 * uninitialized extent to its left neighbor. This is much cheaper
3198 * than an insertion followed by a merge as those involve costly
3199 * memmove() calls. This is the common case in steady state for
3200 * workloads doing fallocate(FALLOC_FL_KEEP_SIZE) followed by append
3203 * Limitations of the current logic:
3204 * - L1: we only deal with writes at the start of the extent.
3205 * The approach could be extended to writes at the end
3206 * of the extent but this scenario was deemed less common.
3207 * - L2: we do not deal with writes covering the whole extent.
3208 * This would require removing the extent if the transfer
3210 * - L3: we only attempt to merge with an extent stored in the
3211 * same extent tree node.
3213 if ((map
->m_lblk
== ee_block
) && /*L1*/
3214 (map
->m_len
< ee_len
) && /*L2*/
3215 (ex
> EXT_FIRST_EXTENT(eh
))) { /*L3*/
3216 struct ext4_extent
*prev_ex
;
3217 ext4_lblk_t prev_lblk
;
3218 ext4_fsblk_t prev_pblk
, ee_pblk
;
3219 unsigned int prev_len
, write_len
;
3222 prev_lblk
= le32_to_cpu(prev_ex
->ee_block
);
3223 prev_len
= ext4_ext_get_actual_len(prev_ex
);
3224 prev_pblk
= ext4_ext_pblock(prev_ex
);
3225 ee_pblk
= ext4_ext_pblock(ex
);
3226 write_len
= map
->m_len
;
3229 * A transfer of blocks from 'ex' to 'prev_ex' is allowed
3230 * upon those conditions:
3231 * - C1: prev_ex is initialized,
3232 * - C2: prev_ex is logically abutting ex,
3233 * - C3: prev_ex is physically abutting ex,
3234 * - C4: prev_ex can receive the additional blocks without
3235 * overflowing the (initialized) length limit.
3237 if ((!ext4_ext_is_uninitialized(prev_ex
)) && /*C1*/
3238 ((prev_lblk
+ prev_len
) == ee_block
) && /*C2*/
3239 ((prev_pblk
+ prev_len
) == ee_pblk
) && /*C3*/
3240 (prev_len
< (EXT_INIT_MAX_LEN
- write_len
))) { /*C4*/
3241 err
= ext4_ext_get_access(handle
, inode
, path
+ depth
);
3245 trace_ext4_ext_convert_to_initialized_fastpath(inode
,
3248 /* Shift the start of ex by 'write_len' blocks */
3249 ex
->ee_block
= cpu_to_le32(ee_block
+ write_len
);
3250 ext4_ext_store_pblock(ex
, ee_pblk
+ write_len
);
3251 ex
->ee_len
= cpu_to_le16(ee_len
- write_len
);
3252 ext4_ext_mark_uninitialized(ex
); /* Restore the flag */
3254 /* Extend prev_ex by 'write_len' blocks */
3255 prev_ex
->ee_len
= cpu_to_le16(prev_len
+ write_len
);
3257 /* Mark the block containing both extents as dirty */
3258 ext4_ext_dirty(handle
, inode
, path
+ depth
);
3260 /* Update path to point to the right extent */
3261 path
[depth
].p_ext
= prev_ex
;
3263 /* Result: number of initialized blocks past m_lblk */
3264 allocated
= write_len
;
3269 WARN_ON(map
->m_lblk
< ee_block
);
3271 * It is safe to convert extent to initialized via explicit
3272 * zeroout only if extent is fully insde i_size or new_size.
3274 split_flag
|= ee_block
+ ee_len
<= eof_block
? EXT4_EXT_MAY_ZEROOUT
: 0;
3276 if (EXT4_EXT_MAY_ZEROOUT
& split_flag
)
3277 max_zeroout
= sbi
->s_extent_max_zeroout_kb
>>
3278 inode
->i_sb
->s_blocksize_bits
;
3280 /* If extent is less than s_max_zeroout_kb, zeroout directly */
3281 if (max_zeroout
&& (ee_len
<= max_zeroout
)) {
3282 err
= ext4_ext_zeroout(inode
, ex
);
3286 err
= ext4_ext_get_access(handle
, inode
, path
+ depth
);
3289 ext4_ext_mark_initialized(ex
);
3290 ext4_ext_try_to_merge(handle
, inode
, path
, ex
);
3291 err
= ext4_ext_dirty(handle
, inode
, path
+ path
->p_depth
);
3297 * 1. split the extent into three extents.
3298 * 2. split the extent into two extents, zeroout the first half.
3299 * 3. split the extent into two extents, zeroout the second half.
3300 * 4. split the extent into two extents with out zeroout.
3302 split_map
.m_lblk
= map
->m_lblk
;
3303 split_map
.m_len
= map
->m_len
;
3305 if (max_zeroout
&& (allocated
> map
->m_len
)) {
3306 if (allocated
<= max_zeroout
) {
3309 cpu_to_le32(map
->m_lblk
);
3310 zero_ex
.ee_len
= cpu_to_le16(allocated
);
3311 ext4_ext_store_pblock(&zero_ex
,
3312 ext4_ext_pblock(ex
) + map
->m_lblk
- ee_block
);
3313 err
= ext4_ext_zeroout(inode
, &zero_ex
);
3316 split_map
.m_lblk
= map
->m_lblk
;
3317 split_map
.m_len
= allocated
;
3318 } else if (map
->m_lblk
- ee_block
+ map
->m_len
< max_zeroout
) {
3320 if (map
->m_lblk
!= ee_block
) {
3321 zero_ex
.ee_block
= ex
->ee_block
;
3322 zero_ex
.ee_len
= cpu_to_le16(map
->m_lblk
-
3324 ext4_ext_store_pblock(&zero_ex
,
3325 ext4_ext_pblock(ex
));
3326 err
= ext4_ext_zeroout(inode
, &zero_ex
);
3331 split_map
.m_lblk
= ee_block
;
3332 split_map
.m_len
= map
->m_lblk
- ee_block
+ map
->m_len
;
3333 allocated
= map
->m_len
;
3337 allocated
= ext4_split_extent(handle
, inode
, path
,
3338 &split_map
, split_flag
, 0);
3343 return err
? err
: allocated
;
3347 * This function is called by ext4_ext_map_blocks() from
3348 * ext4_get_blocks_dio_write() when DIO to write
3349 * to an uninitialized extent.
3351 * Writing to an uninitialized extent may result in splitting the uninitialized
3352 * extent into multiple initialized/uninitialized extents (up to three)
3353 * There are three possibilities:
3354 * a> There is no split required: Entire extent should be uninitialized
3355 * b> Splits in two extents: Write is happening at either end of the extent
3356 * c> Splits in three extents: Somone is writing in middle of the extent
3358 * One of more index blocks maybe needed if the extent tree grow after
3359 * the uninitialized extent split. To prevent ENOSPC occur at the IO
3360 * complete, we need to split the uninitialized extent before DIO submit
3361 * the IO. The uninitialized extent called at this time will be split
3362 * into three uninitialized extent(at most). After IO complete, the part
3363 * being filled will be convert to initialized by the end_io callback function
3364 * via ext4_convert_unwritten_extents().
3366 * Returns the size of uninitialized extent to be written on success.
3368 static int ext4_split_unwritten_extents(handle_t
*handle
,
3369 struct inode
*inode
,
3370 struct ext4_map_blocks
*map
,
3371 struct ext4_ext_path
*path
,
3374 ext4_lblk_t eof_block
;
3375 ext4_lblk_t ee_block
;
3376 struct ext4_extent
*ex
;
3377 unsigned int ee_len
;
3378 int split_flag
= 0, depth
;
3380 ext_debug("ext4_split_unwritten_extents: inode %lu, logical"
3381 "block %llu, max_blocks %u\n", inode
->i_ino
,
3382 (unsigned long long)map
->m_lblk
, map
->m_len
);
3384 eof_block
= (inode
->i_size
+ inode
->i_sb
->s_blocksize
- 1) >>
3385 inode
->i_sb
->s_blocksize_bits
;
3386 if (eof_block
< map
->m_lblk
+ map
->m_len
)
3387 eof_block
= map
->m_lblk
+ map
->m_len
;
3389 * It is safe to convert extent to initialized via explicit
3390 * zeroout only if extent is fully insde i_size or new_size.
3392 depth
= ext_depth(inode
);
3393 ex
= path
[depth
].p_ext
;
3394 ee_block
= le32_to_cpu(ex
->ee_block
);
3395 ee_len
= ext4_ext_get_actual_len(ex
);
3397 split_flag
|= ee_block
+ ee_len
<= eof_block
? EXT4_EXT_MAY_ZEROOUT
: 0;
3398 split_flag
|= EXT4_EXT_MARK_UNINIT2
;
3399 if (flags
& EXT4_GET_BLOCKS_CONVERT
)
3400 split_flag
|= EXT4_EXT_DATA_VALID2
;
3401 flags
|= EXT4_GET_BLOCKS_PRE_IO
;
3402 return ext4_split_extent(handle
, inode
, path
, map
, split_flag
, flags
);
3405 static int ext4_convert_unwritten_extents_endio(handle_t
*handle
,
3406 struct inode
*inode
,
3407 struct ext4_map_blocks
*map
,
3408 struct ext4_ext_path
*path
)
3410 struct ext4_extent
*ex
;
3411 ext4_lblk_t ee_block
;
3412 unsigned int ee_len
;
3416 depth
= ext_depth(inode
);
3417 ex
= path
[depth
].p_ext
;
3418 ee_block
= le32_to_cpu(ex
->ee_block
);
3419 ee_len
= ext4_ext_get_actual_len(ex
);
3421 ext_debug("ext4_convert_unwritten_extents_endio: inode %lu, logical"
3422 "block %llu, max_blocks %u\n", inode
->i_ino
,
3423 (unsigned long long)ee_block
, ee_len
);
3425 /* If extent is larger than requested then split is required */
3426 if (ee_block
!= map
->m_lblk
|| ee_len
> map
->m_len
) {
3427 err
= ext4_split_unwritten_extents(handle
, inode
, map
, path
,
3428 EXT4_GET_BLOCKS_CONVERT
);
3431 ext4_ext_drop_refs(path
);
3432 path
= ext4_ext_find_extent(inode
, map
->m_lblk
, path
);
3434 err
= PTR_ERR(path
);
3437 depth
= ext_depth(inode
);
3438 ex
= path
[depth
].p_ext
;
3441 err
= ext4_ext_get_access(handle
, inode
, path
+ depth
);
3444 /* first mark the extent as initialized */
3445 ext4_ext_mark_initialized(ex
);
3447 /* note: ext4_ext_correct_indexes() isn't needed here because
3448 * borders are not changed
3450 ext4_ext_try_to_merge(handle
, inode
, path
, ex
);
3452 /* Mark modified extent as dirty */
3453 err
= ext4_ext_dirty(handle
, inode
, path
+ path
->p_depth
);
3455 ext4_ext_show_leaf(inode
, path
);
3459 static void unmap_underlying_metadata_blocks(struct block_device
*bdev
,
3460 sector_t block
, int count
)
3463 for (i
= 0; i
< count
; i
++)
3464 unmap_underlying_metadata(bdev
, block
+ i
);
3468 * Handle EOFBLOCKS_FL flag, clearing it if necessary
3470 static int check_eofblocks_fl(handle_t
*handle
, struct inode
*inode
,
3472 struct ext4_ext_path
*path
,
3476 struct ext4_extent_header
*eh
;
3477 struct ext4_extent
*last_ex
;
3479 if (!ext4_test_inode_flag(inode
, EXT4_INODE_EOFBLOCKS
))
3482 depth
= ext_depth(inode
);
3483 eh
= path
[depth
].p_hdr
;
3486 * We're going to remove EOFBLOCKS_FL entirely in future so we
3487 * do not care for this case anymore. Simply remove the flag
3488 * if there are no extents.
3490 if (unlikely(!eh
->eh_entries
))
3492 last_ex
= EXT_LAST_EXTENT(eh
);
3494 * We should clear the EOFBLOCKS_FL flag if we are writing the
3495 * last block in the last extent in the file. We test this by
3496 * first checking to see if the caller to
3497 * ext4_ext_get_blocks() was interested in the last block (or
3498 * a block beyond the last block) in the current extent. If
3499 * this turns out to be false, we can bail out from this
3500 * function immediately.
3502 if (lblk
+ len
< le32_to_cpu(last_ex
->ee_block
) +
3503 ext4_ext_get_actual_len(last_ex
))
3506 * If the caller does appear to be planning to write at or
3507 * beyond the end of the current extent, we then test to see
3508 * if the current extent is the last extent in the file, by
3509 * checking to make sure it was reached via the rightmost node
3510 * at each level of the tree.
3512 for (i
= depth
-1; i
>= 0; i
--)
3513 if (path
[i
].p_idx
!= EXT_LAST_INDEX(path
[i
].p_hdr
))
3516 ext4_clear_inode_flag(inode
, EXT4_INODE_EOFBLOCKS
);
3517 return ext4_mark_inode_dirty(handle
, inode
);
3521 * ext4_find_delalloc_range: find delayed allocated block in the given range.
3523 * Return 1 if there is a delalloc block in the range, otherwise 0.
3525 static int ext4_find_delalloc_range(struct inode
*inode
,
3526 ext4_lblk_t lblk_start
,
3527 ext4_lblk_t lblk_end
)
3529 struct extent_status es
;
3531 es
.start
= lblk_start
;
3532 ext4_es_find_extent(inode
, &es
);
3534 return 0; /* there is no delay extent in this tree */
3535 else if (es
.start
<= lblk_start
&& lblk_start
< es
.start
+ es
.len
)
3537 else if (lblk_start
<= es
.start
&& es
.start
<= lblk_end
)
3543 int ext4_find_delalloc_cluster(struct inode
*inode
, ext4_lblk_t lblk
)
3545 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
3546 ext4_lblk_t lblk_start
, lblk_end
;
3547 lblk_start
= lblk
& (~(sbi
->s_cluster_ratio
- 1));
3548 lblk_end
= lblk_start
+ sbi
->s_cluster_ratio
- 1;
3550 return ext4_find_delalloc_range(inode
, lblk_start
, lblk_end
);
3554 * Determines how many complete clusters (out of those specified by the 'map')
3555 * are under delalloc and were reserved quota for.
3556 * This function is called when we are writing out the blocks that were
3557 * originally written with their allocation delayed, but then the space was
3558 * allocated using fallocate() before the delayed allocation could be resolved.
3559 * The cases to look for are:
3560 * ('=' indicated delayed allocated blocks
3561 * '-' indicates non-delayed allocated blocks)
3562 * (a) partial clusters towards beginning and/or end outside of allocated range
3563 * are not delalloc'ed.
3565 * |----c---=|====c====|====c====|===-c----|
3566 * |++++++ allocated ++++++|
3567 * ==> 4 complete clusters in above example
3569 * (b) partial cluster (outside of allocated range) towards either end is
3570 * marked for delayed allocation. In this case, we will exclude that
3573 * |----====c========|========c========|
3574 * |++++++ allocated ++++++|
3575 * ==> 1 complete clusters in above example
3578 * |================c================|
3579 * |++++++ allocated ++++++|
3580 * ==> 0 complete clusters in above example
3582 * The ext4_da_update_reserve_space will be called only if we
3583 * determine here that there were some "entire" clusters that span
3584 * this 'allocated' range.
3585 * In the non-bigalloc case, this function will just end up returning num_blks
3586 * without ever calling ext4_find_delalloc_range.
3589 get_reserved_cluster_alloc(struct inode
*inode
, ext4_lblk_t lblk_start
,
3590 unsigned int num_blks
)
3592 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
3593 ext4_lblk_t alloc_cluster_start
, alloc_cluster_end
;
3594 ext4_lblk_t lblk_from
, lblk_to
, c_offset
;
3595 unsigned int allocated_clusters
= 0;
3597 alloc_cluster_start
= EXT4_B2C(sbi
, lblk_start
);
3598 alloc_cluster_end
= EXT4_B2C(sbi
, lblk_start
+ num_blks
- 1);
3600 /* max possible clusters for this allocation */
3601 allocated_clusters
= alloc_cluster_end
- alloc_cluster_start
+ 1;
3603 trace_ext4_get_reserved_cluster_alloc(inode
, lblk_start
, num_blks
);
3605 /* Check towards left side */
3606 c_offset
= lblk_start
& (sbi
->s_cluster_ratio
- 1);
3608 lblk_from
= lblk_start
& (~(sbi
->s_cluster_ratio
- 1));
3609 lblk_to
= lblk_from
+ c_offset
- 1;
3611 if (ext4_find_delalloc_range(inode
, lblk_from
, lblk_to
))
3612 allocated_clusters
--;
3615 /* Now check towards right. */
3616 c_offset
= (lblk_start
+ num_blks
) & (sbi
->s_cluster_ratio
- 1);
3617 if (allocated_clusters
&& c_offset
) {
3618 lblk_from
= lblk_start
+ num_blks
;
3619 lblk_to
= lblk_from
+ (sbi
->s_cluster_ratio
- c_offset
) - 1;
3621 if (ext4_find_delalloc_range(inode
, lblk_from
, lblk_to
))
3622 allocated_clusters
--;
3625 return allocated_clusters
;
3629 ext4_ext_handle_uninitialized_extents(handle_t
*handle
, struct inode
*inode
,
3630 struct ext4_map_blocks
*map
,
3631 struct ext4_ext_path
*path
, int flags
,
3632 unsigned int allocated
, ext4_fsblk_t newblock
)
3636 ext4_io_end_t
*io
= ext4_inode_aio(inode
);
3638 ext_debug("ext4_ext_handle_uninitialized_extents: inode %lu, logical "
3639 "block %llu, max_blocks %u, flags %x, allocated %u\n",
3640 inode
->i_ino
, (unsigned long long)map
->m_lblk
, map
->m_len
,
3642 ext4_ext_show_leaf(inode
, path
);
3644 trace_ext4_ext_handle_uninitialized_extents(inode
, map
, flags
,
3645 allocated
, newblock
);
3647 /* get_block() before submit the IO, split the extent */
3648 if ((flags
& EXT4_GET_BLOCKS_PRE_IO
)) {
3649 ret
= ext4_split_unwritten_extents(handle
, inode
, map
,
3654 * Flag the inode(non aio case) or end_io struct (aio case)
3655 * that this IO needs to conversion to written when IO is
3659 ext4_set_io_unwritten_flag(inode
, io
);
3661 ext4_set_inode_state(inode
, EXT4_STATE_DIO_UNWRITTEN
);
3662 if (ext4_should_dioread_nolock(inode
))
3663 map
->m_flags
|= EXT4_MAP_UNINIT
;
3666 /* IO end_io complete, convert the filled extent to written */
3667 if ((flags
& EXT4_GET_BLOCKS_CONVERT
)) {
3668 ret
= ext4_convert_unwritten_extents_endio(handle
, inode
, map
,
3671 ext4_update_inode_fsync_trans(handle
, inode
, 1);
3672 err
= check_eofblocks_fl(handle
, inode
, map
->m_lblk
,
3678 /* buffered IO case */
3680 * repeat fallocate creation request
3681 * we already have an unwritten extent
3683 if (flags
& EXT4_GET_BLOCKS_UNINIT_EXT
)
3686 /* buffered READ or buffered write_begin() lookup */
3687 if ((flags
& EXT4_GET_BLOCKS_CREATE
) == 0) {
3689 * We have blocks reserved already. We
3690 * return allocated blocks so that delalloc
3691 * won't do block reservation for us. But
3692 * the buffer head will be unmapped so that
3693 * a read from the block returns 0s.
3695 map
->m_flags
|= EXT4_MAP_UNWRITTEN
;
3699 /* buffered write, writepage time, convert*/
3700 ret
= ext4_ext_convert_to_initialized(handle
, inode
, map
, path
);
3702 ext4_update_inode_fsync_trans(handle
, inode
, 1);
3709 map
->m_flags
|= EXT4_MAP_NEW
;
3711 * if we allocated more blocks than requested
3712 * we need to make sure we unmap the extra block
3713 * allocated. The actual needed block will get
3714 * unmapped later when we find the buffer_head marked
3717 if (allocated
> map
->m_len
) {
3718 unmap_underlying_metadata_blocks(inode
->i_sb
->s_bdev
,
3719 newblock
+ map
->m_len
,
3720 allocated
- map
->m_len
);
3721 allocated
= map
->m_len
;
3725 * If we have done fallocate with the offset that is already
3726 * delayed allocated, we would have block reservation
3727 * and quota reservation done in the delayed write path.
3728 * But fallocate would have already updated quota and block
3729 * count for this offset. So cancel these reservation
3731 if (flags
& EXT4_GET_BLOCKS_DELALLOC_RESERVE
) {
3732 unsigned int reserved_clusters
;
3733 reserved_clusters
= get_reserved_cluster_alloc(inode
,
3734 map
->m_lblk
, map
->m_len
);
3735 if (reserved_clusters
)
3736 ext4_da_update_reserve_space(inode
,
3742 map
->m_flags
|= EXT4_MAP_MAPPED
;
3743 if ((flags
& EXT4_GET_BLOCKS_KEEP_SIZE
) == 0) {
3744 err
= check_eofblocks_fl(handle
, inode
, map
->m_lblk
, path
,
3750 if (allocated
> map
->m_len
)
3751 allocated
= map
->m_len
;
3752 ext4_ext_show_leaf(inode
, path
);
3753 map
->m_pblk
= newblock
;
3754 map
->m_len
= allocated
;
3757 ext4_ext_drop_refs(path
);
3760 return err
? err
: allocated
;
3764 * get_implied_cluster_alloc - check to see if the requested
3765 * allocation (in the map structure) overlaps with a cluster already
3766 * allocated in an extent.
3767 * @sb The filesystem superblock structure
3768 * @map The requested lblk->pblk mapping
3769 * @ex The extent structure which might contain an implied
3770 * cluster allocation
3772 * This function is called by ext4_ext_map_blocks() after we failed to
3773 * find blocks that were already in the inode's extent tree. Hence,
3774 * we know that the beginning of the requested region cannot overlap
3775 * the extent from the inode's extent tree. There are three cases we
3776 * want to catch. The first is this case:
3778 * |--- cluster # N--|
3779 * |--- extent ---| |---- requested region ---|
3782 * The second case that we need to test for is this one:
3784 * |--------- cluster # N ----------------|
3785 * |--- requested region --| |------- extent ----|
3786 * |=======================|
3788 * The third case is when the requested region lies between two extents
3789 * within the same cluster:
3790 * |------------- cluster # N-------------|
3791 * |----- ex -----| |---- ex_right ----|
3792 * |------ requested region ------|
3793 * |================|
3795 * In each of the above cases, we need to set the map->m_pblk and
3796 * map->m_len so it corresponds to the return the extent labelled as
3797 * "|====|" from cluster #N, since it is already in use for data in
3798 * cluster EXT4_B2C(sbi, map->m_lblk). We will then return 1 to
3799 * signal to ext4_ext_map_blocks() that map->m_pblk should be treated
3800 * as a new "allocated" block region. Otherwise, we will return 0 and
3801 * ext4_ext_map_blocks() will then allocate one or more new clusters
3802 * by calling ext4_mb_new_blocks().
3804 static int get_implied_cluster_alloc(struct super_block
*sb
,
3805 struct ext4_map_blocks
*map
,
3806 struct ext4_extent
*ex
,
3807 struct ext4_ext_path
*path
)
3809 struct ext4_sb_info
*sbi
= EXT4_SB(sb
);
3810 ext4_lblk_t c_offset
= map
->m_lblk
& (sbi
->s_cluster_ratio
-1);
3811 ext4_lblk_t ex_cluster_start
, ex_cluster_end
;
3812 ext4_lblk_t rr_cluster_start
;
3813 ext4_lblk_t ee_block
= le32_to_cpu(ex
->ee_block
);
3814 ext4_fsblk_t ee_start
= ext4_ext_pblock(ex
);
3815 unsigned short ee_len
= ext4_ext_get_actual_len(ex
);
3817 /* The extent passed in that we are trying to match */
3818 ex_cluster_start
= EXT4_B2C(sbi
, ee_block
);
3819 ex_cluster_end
= EXT4_B2C(sbi
, ee_block
+ ee_len
- 1);
3821 /* The requested region passed into ext4_map_blocks() */
3822 rr_cluster_start
= EXT4_B2C(sbi
, map
->m_lblk
);
3824 if ((rr_cluster_start
== ex_cluster_end
) ||
3825 (rr_cluster_start
== ex_cluster_start
)) {
3826 if (rr_cluster_start
== ex_cluster_end
)
3827 ee_start
+= ee_len
- 1;
3828 map
->m_pblk
= (ee_start
& ~(sbi
->s_cluster_ratio
- 1)) +
3830 map
->m_len
= min(map
->m_len
,
3831 (unsigned) sbi
->s_cluster_ratio
- c_offset
);
3833 * Check for and handle this case:
3835 * |--------- cluster # N-------------|
3836 * |------- extent ----|
3837 * |--- requested region ---|
3841 if (map
->m_lblk
< ee_block
)
3842 map
->m_len
= min(map
->m_len
, ee_block
- map
->m_lblk
);
3845 * Check for the case where there is already another allocated
3846 * block to the right of 'ex' but before the end of the cluster.
3848 * |------------- cluster # N-------------|
3849 * |----- ex -----| |---- ex_right ----|
3850 * |------ requested region ------|
3851 * |================|
3853 if (map
->m_lblk
> ee_block
) {
3854 ext4_lblk_t next
= ext4_ext_next_allocated_block(path
);
3855 map
->m_len
= min(map
->m_len
, next
- map
->m_lblk
);
3858 trace_ext4_get_implied_cluster_alloc_exit(sb
, map
, 1);
3862 trace_ext4_get_implied_cluster_alloc_exit(sb
, map
, 0);
3868 * Block allocation/map/preallocation routine for extents based files
3871 * Need to be called with
3872 * down_read(&EXT4_I(inode)->i_data_sem) if not allocating file system block
3873 * (ie, create is zero). Otherwise down_write(&EXT4_I(inode)->i_data_sem)
3875 * return > 0, number of of blocks already mapped/allocated
3876 * if create == 0 and these are pre-allocated blocks
3877 * buffer head is unmapped
3878 * otherwise blocks are mapped
3880 * return = 0, if plain look up failed (blocks have not been allocated)
3881 * buffer head is unmapped
3883 * return < 0, error case.
3885 int ext4_ext_map_blocks(handle_t
*handle
, struct inode
*inode
,
3886 struct ext4_map_blocks
*map
, int flags
)
3888 struct ext4_ext_path
*path
= NULL
;
3889 struct ext4_extent newex
, *ex
, *ex2
;
3890 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
3891 ext4_fsblk_t newblock
= 0;
3892 int free_on_err
= 0, err
= 0, depth
;
3893 unsigned int allocated
= 0, offset
= 0;
3894 unsigned int allocated_clusters
= 0;
3895 struct ext4_allocation_request ar
;
3896 ext4_io_end_t
*io
= ext4_inode_aio(inode
);
3897 ext4_lblk_t cluster_offset
;
3898 int set_unwritten
= 0;
3900 ext_debug("blocks %u/%u requested for inode %lu\n",
3901 map
->m_lblk
, map
->m_len
, inode
->i_ino
);
3902 trace_ext4_ext_map_blocks_enter(inode
, map
->m_lblk
, map
->m_len
, flags
);
3904 /* check in cache */
3905 if (ext4_ext_in_cache(inode
, map
->m_lblk
, &newex
)) {
3906 if (!newex
.ee_start_lo
&& !newex
.ee_start_hi
) {
3907 if ((sbi
->s_cluster_ratio
> 1) &&
3908 ext4_find_delalloc_cluster(inode
, map
->m_lblk
))
3909 map
->m_flags
|= EXT4_MAP_FROM_CLUSTER
;
3911 if ((flags
& EXT4_GET_BLOCKS_CREATE
) == 0) {
3913 * block isn't allocated yet and
3914 * user doesn't want to allocate it
3918 /* we should allocate requested block */
3920 /* block is already allocated */
3921 if (sbi
->s_cluster_ratio
> 1)
3922 map
->m_flags
|= EXT4_MAP_FROM_CLUSTER
;
3923 newblock
= map
->m_lblk
3924 - le32_to_cpu(newex
.ee_block
)
3925 + ext4_ext_pblock(&newex
);
3926 /* number of remaining blocks in the extent */
3927 allocated
= ext4_ext_get_actual_len(&newex
) -
3928 (map
->m_lblk
- le32_to_cpu(newex
.ee_block
));
3933 /* find extent for this block */
3934 path
= ext4_ext_find_extent(inode
, map
->m_lblk
, NULL
);
3936 err
= PTR_ERR(path
);
3941 depth
= ext_depth(inode
);
3944 * consistent leaf must not be empty;
3945 * this situation is possible, though, _during_ tree modification;
3946 * this is why assert can't be put in ext4_ext_find_extent()
3948 if (unlikely(path
[depth
].p_ext
== NULL
&& depth
!= 0)) {
3949 EXT4_ERROR_INODE(inode
, "bad extent address "
3950 "lblock: %lu, depth: %d pblock %lld",
3951 (unsigned long) map
->m_lblk
, depth
,
3952 path
[depth
].p_block
);
3957 ex
= path
[depth
].p_ext
;
3959 ext4_lblk_t ee_block
= le32_to_cpu(ex
->ee_block
);
3960 ext4_fsblk_t ee_start
= ext4_ext_pblock(ex
);
3961 unsigned short ee_len
;
3964 * Uninitialized extents are treated as holes, except that
3965 * we split out initialized portions during a write.
3967 ee_len
= ext4_ext_get_actual_len(ex
);
3969 trace_ext4_ext_show_extent(inode
, ee_block
, ee_start
, ee_len
);
3971 /* if found extent covers block, simply return it */
3972 if (in_range(map
->m_lblk
, ee_block
, ee_len
)) {
3973 newblock
= map
->m_lblk
- ee_block
+ ee_start
;
3974 /* number of remaining blocks in the extent */
3975 allocated
= ee_len
- (map
->m_lblk
- ee_block
);
3976 ext_debug("%u fit into %u:%d -> %llu\n", map
->m_lblk
,
3977 ee_block
, ee_len
, newblock
);
3980 * Do not put uninitialized extent
3983 if (!ext4_ext_is_uninitialized(ex
)) {
3984 ext4_ext_put_in_cache(inode
, ee_block
,
3988 allocated
= ext4_ext_handle_uninitialized_extents(
3989 handle
, inode
, map
, path
, flags
,
3990 allocated
, newblock
);
3995 if ((sbi
->s_cluster_ratio
> 1) &&
3996 ext4_find_delalloc_cluster(inode
, map
->m_lblk
))
3997 map
->m_flags
|= EXT4_MAP_FROM_CLUSTER
;
4000 * requested block isn't allocated yet;
4001 * we couldn't try to create block if create flag is zero
4003 if ((flags
& EXT4_GET_BLOCKS_CREATE
) == 0) {
4005 * put just found gap into cache to speed up
4006 * subsequent requests
4008 ext4_ext_put_gap_in_cache(inode
, path
, map
->m_lblk
);
4013 * Okay, we need to do block allocation.
4015 map
->m_flags
&= ~EXT4_MAP_FROM_CLUSTER
;
4016 newex
.ee_block
= cpu_to_le32(map
->m_lblk
);
4017 cluster_offset
= map
->m_lblk
& (sbi
->s_cluster_ratio
-1);
4020 * If we are doing bigalloc, check to see if the extent returned
4021 * by ext4_ext_find_extent() implies a cluster we can use.
4023 if (cluster_offset
&& ex
&&
4024 get_implied_cluster_alloc(inode
->i_sb
, map
, ex
, path
)) {
4025 ar
.len
= allocated
= map
->m_len
;
4026 newblock
= map
->m_pblk
;
4027 map
->m_flags
|= EXT4_MAP_FROM_CLUSTER
;
4028 goto got_allocated_blocks
;
4031 /* find neighbour allocated blocks */
4032 ar
.lleft
= map
->m_lblk
;
4033 err
= ext4_ext_search_left(inode
, path
, &ar
.lleft
, &ar
.pleft
);
4036 ar
.lright
= map
->m_lblk
;
4038 err
= ext4_ext_search_right(inode
, path
, &ar
.lright
, &ar
.pright
, &ex2
);
4042 /* Check if the extent after searching to the right implies a
4043 * cluster we can use. */
4044 if ((sbi
->s_cluster_ratio
> 1) && ex2
&&
4045 get_implied_cluster_alloc(inode
->i_sb
, map
, ex2
, path
)) {
4046 ar
.len
= allocated
= map
->m_len
;
4047 newblock
= map
->m_pblk
;
4048 map
->m_flags
|= EXT4_MAP_FROM_CLUSTER
;
4049 goto got_allocated_blocks
;
4053 * See if request is beyond maximum number of blocks we can have in
4054 * a single extent. For an initialized extent this limit is
4055 * EXT_INIT_MAX_LEN and for an uninitialized extent this limit is
4056 * EXT_UNINIT_MAX_LEN.
4058 if (map
->m_len
> EXT_INIT_MAX_LEN
&&
4059 !(flags
& EXT4_GET_BLOCKS_UNINIT_EXT
))
4060 map
->m_len
= EXT_INIT_MAX_LEN
;
4061 else if (map
->m_len
> EXT_UNINIT_MAX_LEN
&&
4062 (flags
& EXT4_GET_BLOCKS_UNINIT_EXT
))
4063 map
->m_len
= EXT_UNINIT_MAX_LEN
;
4065 /* Check if we can really insert (m_lblk)::(m_lblk + m_len) extent */
4066 newex
.ee_len
= cpu_to_le16(map
->m_len
);
4067 err
= ext4_ext_check_overlap(sbi
, inode
, &newex
, path
);
4069 allocated
= ext4_ext_get_actual_len(&newex
);
4071 allocated
= map
->m_len
;
4073 /* allocate new block */
4075 ar
.goal
= ext4_ext_find_goal(inode
, path
, map
->m_lblk
);
4076 ar
.logical
= map
->m_lblk
;
4078 * We calculate the offset from the beginning of the cluster
4079 * for the logical block number, since when we allocate a
4080 * physical cluster, the physical block should start at the
4081 * same offset from the beginning of the cluster. This is
4082 * needed so that future calls to get_implied_cluster_alloc()
4085 offset
= map
->m_lblk
& (sbi
->s_cluster_ratio
- 1);
4086 ar
.len
= EXT4_NUM_B2C(sbi
, offset
+allocated
);
4088 ar
.logical
-= offset
;
4089 if (S_ISREG(inode
->i_mode
))
4090 ar
.flags
= EXT4_MB_HINT_DATA
;
4092 /* disable in-core preallocation for non-regular files */
4094 if (flags
& EXT4_GET_BLOCKS_NO_NORMALIZE
)
4095 ar
.flags
|= EXT4_MB_HINT_NOPREALLOC
;
4096 newblock
= ext4_mb_new_blocks(handle
, &ar
, &err
);
4099 ext_debug("allocate new block: goal %llu, found %llu/%u\n",
4100 ar
.goal
, newblock
, allocated
);
4102 allocated_clusters
= ar
.len
;
4103 ar
.len
= EXT4_C2B(sbi
, ar
.len
) - offset
;
4104 if (ar
.len
> allocated
)
4107 got_allocated_blocks
:
4108 /* try to insert new extent into found leaf and return */
4109 ext4_ext_store_pblock(&newex
, newblock
+ offset
);
4110 newex
.ee_len
= cpu_to_le16(ar
.len
);
4111 /* Mark uninitialized */
4112 if (flags
& EXT4_GET_BLOCKS_UNINIT_EXT
){
4113 ext4_ext_mark_uninitialized(&newex
);
4115 * io_end structure was created for every IO write to an
4116 * uninitialized extent. To avoid unnecessary conversion,
4117 * here we flag the IO that really needs the conversion.
4118 * For non asycn direct IO case, flag the inode state
4119 * that we need to perform conversion when IO is done.
4121 if ((flags
& EXT4_GET_BLOCKS_PRE_IO
))
4123 if (ext4_should_dioread_nolock(inode
))
4124 map
->m_flags
|= EXT4_MAP_UNINIT
;
4128 if ((flags
& EXT4_GET_BLOCKS_KEEP_SIZE
) == 0)
4129 err
= check_eofblocks_fl(handle
, inode
, map
->m_lblk
,
4132 err
= ext4_ext_insert_extent(handle
, inode
, path
,
4135 if (!err
&& set_unwritten
) {
4137 ext4_set_io_unwritten_flag(inode
, io
);
4139 ext4_set_inode_state(inode
,
4140 EXT4_STATE_DIO_UNWRITTEN
);
4143 if (err
&& free_on_err
) {
4144 int fb_flags
= flags
& EXT4_GET_BLOCKS_DELALLOC_RESERVE
?
4145 EXT4_FREE_BLOCKS_NO_QUOT_UPDATE
: 0;
4146 /* free data blocks we just allocated */
4147 /* not a good idea to call discard here directly,
4148 * but otherwise we'd need to call it every free() */
4149 ext4_discard_preallocations(inode
);
4150 ext4_free_blocks(handle
, inode
, NULL
, ext4_ext_pblock(&newex
),
4151 ext4_ext_get_actual_len(&newex
), fb_flags
);
4155 /* previous routine could use block we allocated */
4156 newblock
= ext4_ext_pblock(&newex
);
4157 allocated
= ext4_ext_get_actual_len(&newex
);
4158 if (allocated
> map
->m_len
)
4159 allocated
= map
->m_len
;
4160 map
->m_flags
|= EXT4_MAP_NEW
;
4163 * Update reserved blocks/metadata blocks after successful
4164 * block allocation which had been deferred till now.
4166 if (flags
& EXT4_GET_BLOCKS_DELALLOC_RESERVE
) {
4167 unsigned int reserved_clusters
;
4169 * Check how many clusters we had reserved this allocated range
4171 reserved_clusters
= get_reserved_cluster_alloc(inode
,
4172 map
->m_lblk
, allocated
);
4173 if (map
->m_flags
& EXT4_MAP_FROM_CLUSTER
) {
4174 if (reserved_clusters
) {
4176 * We have clusters reserved for this range.
4177 * But since we are not doing actual allocation
4178 * and are simply using blocks from previously
4179 * allocated cluster, we should release the
4180 * reservation and not claim quota.
4182 ext4_da_update_reserve_space(inode
,
4183 reserved_clusters
, 0);
4186 BUG_ON(allocated_clusters
< reserved_clusters
);
4187 /* We will claim quota for all newly allocated blocks.*/
4188 ext4_da_update_reserve_space(inode
, allocated_clusters
,
4190 if (reserved_clusters
< allocated_clusters
) {
4191 struct ext4_inode_info
*ei
= EXT4_I(inode
);
4192 int reservation
= allocated_clusters
-
4195 * It seems we claimed few clusters outside of
4196 * the range of this allocation. We should give
4197 * it back to the reservation pool. This can
4198 * happen in the following case:
4200 * * Suppose s_cluster_ratio is 4 (i.e., each
4201 * cluster has 4 blocks. Thus, the clusters
4202 * are [0-3],[4-7],[8-11]...
4203 * * First comes delayed allocation write for
4204 * logical blocks 10 & 11. Since there were no
4205 * previous delayed allocated blocks in the
4206 * range [8-11], we would reserve 1 cluster
4208 * * Next comes write for logical blocks 3 to 8.
4209 * In this case, we will reserve 2 clusters
4210 * (for [0-3] and [4-7]; and not for [8-11] as
4211 * that range has a delayed allocated blocks.
4212 * Thus total reserved clusters now becomes 3.
4213 * * Now, during the delayed allocation writeout
4214 * time, we will first write blocks [3-8] and
4215 * allocate 3 clusters for writing these
4216 * blocks. Also, we would claim all these
4217 * three clusters above.
4218 * * Now when we come here to writeout the
4219 * blocks [10-11], we would expect to claim
4220 * the reservation of 1 cluster we had made
4221 * (and we would claim it since there are no
4222 * more delayed allocated blocks in the range
4223 * [8-11]. But our reserved cluster count had
4224 * already gone to 0.
4226 * Thus, at the step 4 above when we determine
4227 * that there are still some unwritten delayed
4228 * allocated blocks outside of our current
4229 * block range, we should increment the
4230 * reserved clusters count so that when the
4231 * remaining blocks finally gets written, we
4234 dquot_reserve_block(inode
,
4235 EXT4_C2B(sbi
, reservation
));
4236 spin_lock(&ei
->i_block_reservation_lock
);
4237 ei
->i_reserved_data_blocks
+= reservation
;
4238 spin_unlock(&ei
->i_block_reservation_lock
);
4244 * Cache the extent and update transaction to commit on fdatasync only
4245 * when it is _not_ an uninitialized extent.
4247 if ((flags
& EXT4_GET_BLOCKS_UNINIT_EXT
) == 0) {
4248 ext4_ext_put_in_cache(inode
, map
->m_lblk
, allocated
, newblock
);
4249 ext4_update_inode_fsync_trans(handle
, inode
, 1);
4251 ext4_update_inode_fsync_trans(handle
, inode
, 0);
4253 if (allocated
> map
->m_len
)
4254 allocated
= map
->m_len
;
4255 ext4_ext_show_leaf(inode
, path
);
4256 map
->m_flags
|= EXT4_MAP_MAPPED
;
4257 map
->m_pblk
= newblock
;
4258 map
->m_len
= allocated
;
4261 ext4_ext_drop_refs(path
);
4266 trace_ext4_ext_map_blocks_exit(inode
, map
, err
? err
: allocated
);
4268 return err
? err
: allocated
;
4271 void ext4_ext_truncate(struct inode
*inode
)
4273 struct address_space
*mapping
= inode
->i_mapping
;
4274 struct super_block
*sb
= inode
->i_sb
;
4275 ext4_lblk_t last_block
;
4281 * finish any pending end_io work so we won't run the risk of
4282 * converting any truncated blocks to initialized later
4284 ext4_flush_unwritten_io(inode
);
4287 * probably first extent we're gonna free will be last in block
4289 err
= ext4_writepage_trans_blocks(inode
);
4290 handle
= ext4_journal_start(inode
, err
);
4294 if (inode
->i_size
% PAGE_CACHE_SIZE
!= 0) {
4295 page_len
= PAGE_CACHE_SIZE
-
4296 (inode
->i_size
& (PAGE_CACHE_SIZE
- 1));
4298 err
= ext4_discard_partial_page_buffers(handle
,
4299 mapping
, inode
->i_size
, page_len
, 0);
4305 if (ext4_orphan_add(handle
, inode
))
4308 down_write(&EXT4_I(inode
)->i_data_sem
);
4309 ext4_ext_invalidate_cache(inode
);
4311 ext4_discard_preallocations(inode
);
4314 * TODO: optimization is possible here.
4315 * Probably we need not scan at all,
4316 * because page truncation is enough.
4319 /* we have to know where to truncate from in crash case */
4320 EXT4_I(inode
)->i_disksize
= inode
->i_size
;
4321 ext4_mark_inode_dirty(handle
, inode
);
4323 last_block
= (inode
->i_size
+ sb
->s_blocksize
- 1)
4324 >> EXT4_BLOCK_SIZE_BITS(sb
);
4325 err
= ext4_es_remove_extent(inode
, last_block
,
4326 EXT_MAX_BLOCKS
- last_block
);
4327 err
= ext4_ext_remove_space(inode
, last_block
, EXT_MAX_BLOCKS
- 1);
4329 /* In a multi-transaction truncate, we only make the final
4330 * transaction synchronous.
4333 ext4_handle_sync(handle
);
4335 up_write(&EXT4_I(inode
)->i_data_sem
);
4339 * If this was a simple ftruncate() and the file will remain alive,
4340 * then we need to clear up the orphan record which we created above.
4341 * However, if this was a real unlink then we were called by
4342 * ext4_delete_inode(), and we allow that function to clean up the
4343 * orphan info for us.
4346 ext4_orphan_del(handle
, inode
);
4348 inode
->i_mtime
= inode
->i_ctime
= ext4_current_time(inode
);
4349 ext4_mark_inode_dirty(handle
, inode
);
4350 ext4_journal_stop(handle
);
4353 static void ext4_falloc_update_inode(struct inode
*inode
,
4354 int mode
, loff_t new_size
, int update_ctime
)
4356 struct timespec now
;
4359 now
= current_fs_time(inode
->i_sb
);
4360 if (!timespec_equal(&inode
->i_ctime
, &now
))
4361 inode
->i_ctime
= now
;
4364 * Update only when preallocation was requested beyond
4367 if (!(mode
& FALLOC_FL_KEEP_SIZE
)) {
4368 if (new_size
> i_size_read(inode
))
4369 i_size_write(inode
, new_size
);
4370 if (new_size
> EXT4_I(inode
)->i_disksize
)
4371 ext4_update_i_disksize(inode
, new_size
);
4374 * Mark that we allocate beyond EOF so the subsequent truncate
4375 * can proceed even if the new size is the same as i_size.
4377 if (new_size
> i_size_read(inode
))
4378 ext4_set_inode_flag(inode
, EXT4_INODE_EOFBLOCKS
);
4384 * preallocate space for a file. This implements ext4's fallocate file
4385 * operation, which gets called from sys_fallocate system call.
4386 * For block-mapped files, posix_fallocate should fall back to the method
4387 * of writing zeroes to the required new blocks (the same behavior which is
4388 * expected for file systems which do not support fallocate() system call).
4390 long ext4_fallocate(struct file
*file
, int mode
, loff_t offset
, loff_t len
)
4392 struct inode
*inode
= file
->f_path
.dentry
->d_inode
;
4395 unsigned int max_blocks
;
4400 struct ext4_map_blocks map
;
4401 unsigned int credits
, blkbits
= inode
->i_blkbits
;
4404 * currently supporting (pre)allocate mode for extent-based
4407 if (!(ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
)))
4410 /* Return error if mode is not supported */
4411 if (mode
& ~(FALLOC_FL_KEEP_SIZE
| FALLOC_FL_PUNCH_HOLE
))
4414 if (mode
& FALLOC_FL_PUNCH_HOLE
)
4415 return ext4_punch_hole(file
, offset
, len
);
4417 ret
= ext4_convert_inline_data(inode
);
4421 trace_ext4_fallocate_enter(inode
, offset
, len
, mode
);
4422 map
.m_lblk
= offset
>> blkbits
;
4424 * We can't just convert len to max_blocks because
4425 * If blocksize = 4096 offset = 3072 and len = 2048
4427 max_blocks
= (EXT4_BLOCK_ALIGN(len
+ offset
, blkbits
) >> blkbits
)
4430 * credits to insert 1 extent into extent tree
4432 credits
= ext4_chunk_trans_blocks(inode
, max_blocks
);
4433 mutex_lock(&inode
->i_mutex
);
4434 ret
= inode_newsize_ok(inode
, (len
+ offset
));
4436 mutex_unlock(&inode
->i_mutex
);
4437 trace_ext4_fallocate_exit(inode
, offset
, max_blocks
, ret
);
4440 flags
= EXT4_GET_BLOCKS_CREATE_UNINIT_EXT
;
4441 if (mode
& FALLOC_FL_KEEP_SIZE
)
4442 flags
|= EXT4_GET_BLOCKS_KEEP_SIZE
;
4444 * Don't normalize the request if it can fit in one extent so
4445 * that it doesn't get unnecessarily split into multiple
4448 if (len
<= EXT_UNINIT_MAX_LEN
<< blkbits
)
4449 flags
|= EXT4_GET_BLOCKS_NO_NORMALIZE
;
4451 /* Prevent race condition between unwritten */
4452 ext4_flush_unwritten_io(inode
);
4454 while (ret
>= 0 && ret
< max_blocks
) {
4455 map
.m_lblk
= map
.m_lblk
+ ret
;
4456 map
.m_len
= max_blocks
= max_blocks
- ret
;
4457 handle
= ext4_journal_start(inode
, credits
);
4458 if (IS_ERR(handle
)) {
4459 ret
= PTR_ERR(handle
);
4462 ret
= ext4_map_blocks(handle
, inode
, &map
, flags
);
4466 printk(KERN_ERR
"%s: ext4_ext_map_blocks "
4467 "returned error inode#%lu, block=%u, "
4468 "max_blocks=%u", __func__
,
4469 inode
->i_ino
, map
.m_lblk
, max_blocks
);
4471 ext4_mark_inode_dirty(handle
, inode
);
4472 ret2
= ext4_journal_stop(handle
);
4475 if ((map
.m_lblk
+ ret
) >= (EXT4_BLOCK_ALIGN(offset
+ len
,
4476 blkbits
) >> blkbits
))
4477 new_size
= offset
+ len
;
4479 new_size
= ((loff_t
) map
.m_lblk
+ ret
) << blkbits
;
4481 ext4_falloc_update_inode(inode
, mode
, new_size
,
4482 (map
.m_flags
& EXT4_MAP_NEW
));
4483 ext4_mark_inode_dirty(handle
, inode
);
4484 if ((file
->f_flags
& O_SYNC
) && ret
>= max_blocks
)
4485 ext4_handle_sync(handle
);
4486 ret2
= ext4_journal_stop(handle
);
4490 if (ret
== -ENOSPC
&&
4491 ext4_should_retry_alloc(inode
->i_sb
, &retries
)) {
4495 mutex_unlock(&inode
->i_mutex
);
4496 trace_ext4_fallocate_exit(inode
, offset
, max_blocks
,
4497 ret
> 0 ? ret2
: ret
);
4498 return ret
> 0 ? ret2
: ret
;
4502 * This function convert a range of blocks to written extents
4503 * The caller of this function will pass the start offset and the size.
4504 * all unwritten extents within this range will be converted to
4507 * This function is called from the direct IO end io call back
4508 * function, to convert the fallocated extents after IO is completed.
4509 * Returns 0 on success.
4511 int ext4_convert_unwritten_extents(struct inode
*inode
, loff_t offset
,
4515 unsigned int max_blocks
;
4518 struct ext4_map_blocks map
;
4519 unsigned int credits
, blkbits
= inode
->i_blkbits
;
4521 map
.m_lblk
= offset
>> blkbits
;
4523 * We can't just convert len to max_blocks because
4524 * If blocksize = 4096 offset = 3072 and len = 2048
4526 max_blocks
= ((EXT4_BLOCK_ALIGN(len
+ offset
, blkbits
) >> blkbits
) -
4529 * credits to insert 1 extent into extent tree
4531 credits
= ext4_chunk_trans_blocks(inode
, max_blocks
);
4532 while (ret
>= 0 && ret
< max_blocks
) {
4534 map
.m_len
= (max_blocks
-= ret
);
4535 handle
= ext4_journal_start(inode
, credits
);
4536 if (IS_ERR(handle
)) {
4537 ret
= PTR_ERR(handle
);
4540 ret
= ext4_map_blocks(handle
, inode
, &map
,
4541 EXT4_GET_BLOCKS_IO_CONVERT_EXT
);
4544 ext4_msg(inode
->i_sb
, KERN_ERR
,
4545 "%s:%d: inode #%lu: block %u: len %u: "
4546 "ext4_ext_map_blocks returned %d",
4547 __func__
, __LINE__
, inode
->i_ino
, map
.m_lblk
,
4550 ext4_mark_inode_dirty(handle
, inode
);
4551 ret2
= ext4_journal_stop(handle
);
4552 if (ret
<= 0 || ret2
)
4555 return ret
> 0 ? ret2
: ret
;
4559 * If newex is not existing extent (newex->ec_start equals zero) find
4560 * delayed extent at start of newex and update newex accordingly and
4561 * return start of the next delayed extent.
4563 * If newex is existing extent (newex->ec_start is not equal zero)
4564 * return start of next delayed extent or EXT_MAX_BLOCKS if no delayed
4565 * extent found. Leave newex unmodified.
4567 static int ext4_find_delayed_extent(struct inode
*inode
,
4568 struct ext4_ext_cache
*newex
)
4570 struct extent_status es
;
4571 ext4_lblk_t next_del
;
4573 es
.start
= newex
->ec_block
;
4574 next_del
= ext4_es_find_extent(inode
, &es
);
4576 if (newex
->ec_start
== 0) {
4578 * No extent in extent-tree contains block @newex->ec_start,
4579 * then the block may stay in 1)a hole or 2)delayed-extent.
4585 if (es
.start
> newex
->ec_block
) {
4587 newex
->ec_len
= min(es
.start
- newex
->ec_block
,
4592 newex
->ec_len
= es
.start
+ es
.len
- newex
->ec_block
;
4597 /* fiemap flags we can handle specified here */
4598 #define EXT4_FIEMAP_FLAGS (FIEMAP_FLAG_SYNC|FIEMAP_FLAG_XATTR)
4600 static int ext4_xattr_fiemap(struct inode
*inode
,
4601 struct fiemap_extent_info
*fieinfo
)
4605 __u32 flags
= FIEMAP_EXTENT_LAST
;
4606 int blockbits
= inode
->i_sb
->s_blocksize_bits
;
4610 if (ext4_test_inode_state(inode
, EXT4_STATE_XATTR
)) {
4611 struct ext4_iloc iloc
;
4612 int offset
; /* offset of xattr in inode */
4614 error
= ext4_get_inode_loc(inode
, &iloc
);
4617 physical
= iloc
.bh
->b_blocknr
<< blockbits
;
4618 offset
= EXT4_GOOD_OLD_INODE_SIZE
+
4619 EXT4_I(inode
)->i_extra_isize
;
4621 length
= EXT4_SB(inode
->i_sb
)->s_inode_size
- offset
;
4622 flags
|= FIEMAP_EXTENT_DATA_INLINE
;
4624 } else { /* external block */
4625 physical
= EXT4_I(inode
)->i_file_acl
<< blockbits
;
4626 length
= inode
->i_sb
->s_blocksize
;
4630 error
= fiemap_fill_next_extent(fieinfo
, 0, physical
,
4632 return (error
< 0 ? error
: 0);
4636 * ext4_ext_punch_hole
4638 * Punches a hole of "length" bytes in a file starting
4641 * @inode: The inode of the file to punch a hole in
4642 * @offset: The starting byte offset of the hole
4643 * @length: The length of the hole
4645 * Returns the number of blocks removed or negative on err
4647 int ext4_ext_punch_hole(struct file
*file
, loff_t offset
, loff_t length
)
4649 struct inode
*inode
= file
->f_path
.dentry
->d_inode
;
4650 struct super_block
*sb
= inode
->i_sb
;
4651 ext4_lblk_t first_block
, stop_block
;
4652 struct address_space
*mapping
= inode
->i_mapping
;
4654 loff_t first_page
, last_page
, page_len
;
4655 loff_t first_page_offset
, last_page_offset
;
4656 int credits
, err
= 0;
4659 * Write out all dirty pages to avoid race conditions
4660 * Then release them.
4662 if (mapping
->nrpages
&& mapping_tagged(mapping
, PAGECACHE_TAG_DIRTY
)) {
4663 err
= filemap_write_and_wait_range(mapping
,
4664 offset
, offset
+ length
- 1);
4670 mutex_lock(&inode
->i_mutex
);
4671 /* It's not possible punch hole on append only file */
4672 if (IS_APPEND(inode
) || IS_IMMUTABLE(inode
)) {
4676 if (IS_SWAPFILE(inode
)) {
4681 /* No need to punch hole beyond i_size */
4682 if (offset
>= inode
->i_size
)
4686 * If the hole extends beyond i_size, set the hole
4687 * to end after the page that contains i_size
4689 if (offset
+ length
> inode
->i_size
) {
4690 length
= inode
->i_size
+
4691 PAGE_CACHE_SIZE
- (inode
->i_size
& (PAGE_CACHE_SIZE
- 1)) -
4695 first_page
= (offset
+ PAGE_CACHE_SIZE
- 1) >> PAGE_CACHE_SHIFT
;
4696 last_page
= (offset
+ length
) >> PAGE_CACHE_SHIFT
;
4698 first_page_offset
= first_page
<< PAGE_CACHE_SHIFT
;
4699 last_page_offset
= last_page
<< PAGE_CACHE_SHIFT
;
4701 /* Now release the pages */
4702 if (last_page_offset
> first_page_offset
) {
4703 truncate_pagecache_range(inode
, first_page_offset
,
4704 last_page_offset
- 1);
4707 /* Wait all existing dio workers, newcomers will block on i_mutex */
4708 ext4_inode_block_unlocked_dio(inode
);
4709 err
= ext4_flush_unwritten_io(inode
);
4712 inode_dio_wait(inode
);
4714 credits
= ext4_writepage_trans_blocks(inode
);
4715 handle
= ext4_journal_start(inode
, credits
);
4716 if (IS_ERR(handle
)) {
4717 err
= PTR_ERR(handle
);
4723 * Now we need to zero out the non-page-aligned data in the
4724 * pages at the start and tail of the hole, and unmap the buffer
4725 * heads for the block aligned regions of the page that were
4726 * completely zeroed.
4728 if (first_page
> last_page
) {
4730 * If the file space being truncated is contained within a page
4731 * just zero out and unmap the middle of that page
4733 err
= ext4_discard_partial_page_buffers(handle
,
4734 mapping
, offset
, length
, 0);
4740 * zero out and unmap the partial page that contains
4741 * the start of the hole
4743 page_len
= first_page_offset
- offset
;
4745 err
= ext4_discard_partial_page_buffers(handle
, mapping
,
4746 offset
, page_len
, 0);
4752 * zero out and unmap the partial page that contains
4753 * the end of the hole
4755 page_len
= offset
+ length
- last_page_offset
;
4757 err
= ext4_discard_partial_page_buffers(handle
, mapping
,
4758 last_page_offset
, page_len
, 0);
4765 * If i_size is contained in the last page, we need to
4766 * unmap and zero the partial page after i_size
4768 if (inode
->i_size
>> PAGE_CACHE_SHIFT
== last_page
&&
4769 inode
->i_size
% PAGE_CACHE_SIZE
!= 0) {
4771 page_len
= PAGE_CACHE_SIZE
-
4772 (inode
->i_size
& (PAGE_CACHE_SIZE
- 1));
4775 err
= ext4_discard_partial_page_buffers(handle
,
4776 mapping
, inode
->i_size
, page_len
, 0);
4783 first_block
= (offset
+ sb
->s_blocksize
- 1) >>
4784 EXT4_BLOCK_SIZE_BITS(sb
);
4785 stop_block
= (offset
+ length
) >> EXT4_BLOCK_SIZE_BITS(sb
);
4787 /* If there are no blocks to remove, return now */
4788 if (first_block
>= stop_block
)
4791 down_write(&EXT4_I(inode
)->i_data_sem
);
4792 ext4_ext_invalidate_cache(inode
);
4793 ext4_discard_preallocations(inode
);
4795 err
= ext4_es_remove_extent(inode
, first_block
,
4796 stop_block
- first_block
);
4797 err
= ext4_ext_remove_space(inode
, first_block
, stop_block
- 1);
4799 ext4_ext_invalidate_cache(inode
);
4800 ext4_discard_preallocations(inode
);
4803 ext4_handle_sync(handle
);
4805 up_write(&EXT4_I(inode
)->i_data_sem
);
4808 inode
->i_mtime
= inode
->i_ctime
= ext4_current_time(inode
);
4809 ext4_mark_inode_dirty(handle
, inode
);
4810 ext4_journal_stop(handle
);
4812 ext4_inode_resume_unlocked_dio(inode
);
4814 mutex_unlock(&inode
->i_mutex
);
4818 int ext4_fiemap(struct inode
*inode
, struct fiemap_extent_info
*fieinfo
,
4819 __u64 start
, __u64 len
)
4821 ext4_lblk_t start_blk
;
4824 if (ext4_has_inline_data(inode
)) {
4827 error
= ext4_inline_data_fiemap(inode
, fieinfo
, &has_inline
);
4833 /* fallback to generic here if not in extents fmt */
4834 if (!(ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
)))
4835 return generic_block_fiemap(inode
, fieinfo
, start
, len
,
4838 if (fiemap_check_flags(fieinfo
, EXT4_FIEMAP_FLAGS
))
4841 if (fieinfo
->fi_flags
& FIEMAP_FLAG_XATTR
) {
4842 error
= ext4_xattr_fiemap(inode
, fieinfo
);
4844 ext4_lblk_t len_blks
;
4847 start_blk
= start
>> inode
->i_sb
->s_blocksize_bits
;
4848 last_blk
= (start
+ len
- 1) >> inode
->i_sb
->s_blocksize_bits
;
4849 if (last_blk
>= EXT_MAX_BLOCKS
)
4850 last_blk
= EXT_MAX_BLOCKS
-1;
4851 len_blks
= ((ext4_lblk_t
) last_blk
) - start_blk
+ 1;
4854 * Walk the extent tree gathering extent information
4855 * and pushing extents back to the user.
4857 error
= ext4_fill_fiemap_extents(inode
, start_blk
,