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;
729 eh
= ext_inode_hdr(inode
);
730 depth
= ext_depth(inode
);
732 /* account possible depth increase */
734 path
= kzalloc(sizeof(struct ext4_ext_path
) * (depth
+ 2),
737 return ERR_PTR(-ENOMEM
);
744 /* walk through the tree */
746 ext_debug("depth %d: num %d, max %d\n",
747 ppos
, le16_to_cpu(eh
->eh_entries
), le16_to_cpu(eh
->eh_max
));
749 ext4_ext_binsearch_idx(inode
, path
+ ppos
, block
);
750 path
[ppos
].p_block
= ext4_idx_pblock(path
[ppos
].p_idx
);
751 path
[ppos
].p_depth
= i
;
752 path
[ppos
].p_ext
= NULL
;
754 bh
= sb_getblk(inode
->i_sb
, path
[ppos
].p_block
);
757 if (!bh_uptodate_or_lock(bh
)) {
758 trace_ext4_ext_load_extent(inode
, block
,
760 if (bh_submit_read(bh
) < 0) {
765 eh
= ext_block_hdr(bh
);
767 if (unlikely(ppos
> depth
)) {
769 EXT4_ERROR_INODE(inode
,
770 "ppos %d > depth %d", ppos
, depth
);
773 path
[ppos
].p_bh
= bh
;
774 path
[ppos
].p_hdr
= eh
;
777 if (ext4_ext_check_block(inode
, eh
, i
, bh
))
781 path
[ppos
].p_depth
= i
;
782 path
[ppos
].p_ext
= NULL
;
783 path
[ppos
].p_idx
= NULL
;
786 ext4_ext_binsearch(inode
, path
+ ppos
, block
);
787 /* if not an empty leaf */
788 if (path
[ppos
].p_ext
)
789 path
[ppos
].p_block
= ext4_ext_pblock(path
[ppos
].p_ext
);
791 ext4_ext_show_path(inode
, path
);
796 ext4_ext_drop_refs(path
);
799 return ERR_PTR(-EIO
);
803 * ext4_ext_insert_index:
804 * insert new index [@logical;@ptr] into the block at @curp;
805 * check where to insert: before @curp or after @curp
807 static int ext4_ext_insert_index(handle_t
*handle
, struct inode
*inode
,
808 struct ext4_ext_path
*curp
,
809 int logical
, ext4_fsblk_t ptr
)
811 struct ext4_extent_idx
*ix
;
814 err
= ext4_ext_get_access(handle
, inode
, curp
);
818 if (unlikely(logical
== le32_to_cpu(curp
->p_idx
->ei_block
))) {
819 EXT4_ERROR_INODE(inode
,
820 "logical %d == ei_block %d!",
821 logical
, le32_to_cpu(curp
->p_idx
->ei_block
));
825 if (unlikely(le16_to_cpu(curp
->p_hdr
->eh_entries
)
826 >= le16_to_cpu(curp
->p_hdr
->eh_max
))) {
827 EXT4_ERROR_INODE(inode
,
828 "eh_entries %d >= eh_max %d!",
829 le16_to_cpu(curp
->p_hdr
->eh_entries
),
830 le16_to_cpu(curp
->p_hdr
->eh_max
));
834 if (logical
> le32_to_cpu(curp
->p_idx
->ei_block
)) {
836 ext_debug("insert new index %d after: %llu\n", logical
, ptr
);
837 ix
= curp
->p_idx
+ 1;
840 ext_debug("insert new index %d before: %llu\n", logical
, ptr
);
844 len
= EXT_LAST_INDEX(curp
->p_hdr
) - ix
+ 1;
847 ext_debug("insert new index %d: "
848 "move %d indices from 0x%p to 0x%p\n",
849 logical
, len
, ix
, ix
+ 1);
850 memmove(ix
+ 1, ix
, len
* sizeof(struct ext4_extent_idx
));
853 if (unlikely(ix
> EXT_MAX_INDEX(curp
->p_hdr
))) {
854 EXT4_ERROR_INODE(inode
, "ix > EXT_MAX_INDEX!");
858 ix
->ei_block
= cpu_to_le32(logical
);
859 ext4_idx_store_pblock(ix
, ptr
);
860 le16_add_cpu(&curp
->p_hdr
->eh_entries
, 1);
862 if (unlikely(ix
> EXT_LAST_INDEX(curp
->p_hdr
))) {
863 EXT4_ERROR_INODE(inode
, "ix > EXT_LAST_INDEX!");
867 err
= ext4_ext_dirty(handle
, inode
, curp
);
868 ext4_std_error(inode
->i_sb
, err
);
875 * inserts new subtree into the path, using free index entry
877 * - allocates all needed blocks (new leaf and all intermediate index blocks)
878 * - makes decision where to split
879 * - moves remaining extents and index entries (right to the split point)
880 * into the newly allocated blocks
881 * - initializes subtree
883 static int ext4_ext_split(handle_t
*handle
, struct inode
*inode
,
885 struct ext4_ext_path
*path
,
886 struct ext4_extent
*newext
, int at
)
888 struct buffer_head
*bh
= NULL
;
889 int depth
= ext_depth(inode
);
890 struct ext4_extent_header
*neh
;
891 struct ext4_extent_idx
*fidx
;
893 ext4_fsblk_t newblock
, oldblock
;
895 ext4_fsblk_t
*ablocks
= NULL
; /* array of allocated blocks */
898 /* make decision: where to split? */
899 /* FIXME: now decision is simplest: at current extent */
901 /* if current leaf will be split, then we should use
902 * border from split point */
903 if (unlikely(path
[depth
].p_ext
> EXT_MAX_EXTENT(path
[depth
].p_hdr
))) {
904 EXT4_ERROR_INODE(inode
, "p_ext > EXT_MAX_EXTENT!");
907 if (path
[depth
].p_ext
!= EXT_MAX_EXTENT(path
[depth
].p_hdr
)) {
908 border
= path
[depth
].p_ext
[1].ee_block
;
909 ext_debug("leaf will be split."
910 " next leaf starts at %d\n",
911 le32_to_cpu(border
));
913 border
= newext
->ee_block
;
914 ext_debug("leaf will be added."
915 " next leaf starts at %d\n",
916 le32_to_cpu(border
));
920 * If error occurs, then we break processing
921 * and mark filesystem read-only. index won't
922 * be inserted and tree will be in consistent
923 * state. Next mount will repair buffers too.
927 * Get array to track all allocated blocks.
928 * We need this to handle errors and free blocks
931 ablocks
= kzalloc(sizeof(ext4_fsblk_t
) * depth
, GFP_NOFS
);
935 /* allocate all needed blocks */
936 ext_debug("allocate %d blocks for indexes/leaf\n", depth
- at
);
937 for (a
= 0; a
< depth
- at
; a
++) {
938 newblock
= ext4_ext_new_meta_block(handle
, inode
, path
,
939 newext
, &err
, flags
);
942 ablocks
[a
] = newblock
;
945 /* initialize new leaf */
946 newblock
= ablocks
[--a
];
947 if (unlikely(newblock
== 0)) {
948 EXT4_ERROR_INODE(inode
, "newblock == 0!");
952 bh
= sb_getblk(inode
->i_sb
, newblock
);
959 err
= ext4_journal_get_create_access(handle
, bh
);
963 neh
= ext_block_hdr(bh
);
965 neh
->eh_max
= cpu_to_le16(ext4_ext_space_block(inode
, 0));
966 neh
->eh_magic
= EXT4_EXT_MAGIC
;
969 /* move remainder of path[depth] to the new leaf */
970 if (unlikely(path
[depth
].p_hdr
->eh_entries
!=
971 path
[depth
].p_hdr
->eh_max
)) {
972 EXT4_ERROR_INODE(inode
, "eh_entries %d != eh_max %d!",
973 path
[depth
].p_hdr
->eh_entries
,
974 path
[depth
].p_hdr
->eh_max
);
978 /* start copy from next extent */
979 m
= EXT_MAX_EXTENT(path
[depth
].p_hdr
) - path
[depth
].p_ext
++;
980 ext4_ext_show_move(inode
, path
, newblock
, depth
);
982 struct ext4_extent
*ex
;
983 ex
= EXT_FIRST_EXTENT(neh
);
984 memmove(ex
, path
[depth
].p_ext
, sizeof(struct ext4_extent
) * m
);
985 le16_add_cpu(&neh
->eh_entries
, m
);
988 ext4_extent_block_csum_set(inode
, neh
);
989 set_buffer_uptodate(bh
);
992 err
= ext4_handle_dirty_metadata(handle
, inode
, bh
);
998 /* correct old leaf */
1000 err
= ext4_ext_get_access(handle
, inode
, path
+ depth
);
1003 le16_add_cpu(&path
[depth
].p_hdr
->eh_entries
, -m
);
1004 err
= ext4_ext_dirty(handle
, inode
, path
+ depth
);
1010 /* create intermediate indexes */
1012 if (unlikely(k
< 0)) {
1013 EXT4_ERROR_INODE(inode
, "k %d < 0!", k
);
1018 ext_debug("create %d intermediate indices\n", k
);
1019 /* insert new index into current index block */
1020 /* current depth stored in i var */
1023 oldblock
= newblock
;
1024 newblock
= ablocks
[--a
];
1025 bh
= sb_getblk(inode
->i_sb
, newblock
);
1032 err
= ext4_journal_get_create_access(handle
, bh
);
1036 neh
= ext_block_hdr(bh
);
1037 neh
->eh_entries
= cpu_to_le16(1);
1038 neh
->eh_magic
= EXT4_EXT_MAGIC
;
1039 neh
->eh_max
= cpu_to_le16(ext4_ext_space_block_idx(inode
, 0));
1040 neh
->eh_depth
= cpu_to_le16(depth
- i
);
1041 fidx
= EXT_FIRST_INDEX(neh
);
1042 fidx
->ei_block
= border
;
1043 ext4_idx_store_pblock(fidx
, oldblock
);
1045 ext_debug("int.index at %d (block %llu): %u -> %llu\n",
1046 i
, newblock
, le32_to_cpu(border
), oldblock
);
1048 /* move remainder of path[i] to the new index block */
1049 if (unlikely(EXT_MAX_INDEX(path
[i
].p_hdr
) !=
1050 EXT_LAST_INDEX(path
[i
].p_hdr
))) {
1051 EXT4_ERROR_INODE(inode
,
1052 "EXT_MAX_INDEX != EXT_LAST_INDEX ee_block %d!",
1053 le32_to_cpu(path
[i
].p_ext
->ee_block
));
1057 /* start copy indexes */
1058 m
= EXT_MAX_INDEX(path
[i
].p_hdr
) - path
[i
].p_idx
++;
1059 ext_debug("cur 0x%p, last 0x%p\n", path
[i
].p_idx
,
1060 EXT_MAX_INDEX(path
[i
].p_hdr
));
1061 ext4_ext_show_move(inode
, path
, newblock
, i
);
1063 memmove(++fidx
, path
[i
].p_idx
,
1064 sizeof(struct ext4_extent_idx
) * m
);
1065 le16_add_cpu(&neh
->eh_entries
, m
);
1067 ext4_extent_block_csum_set(inode
, neh
);
1068 set_buffer_uptodate(bh
);
1071 err
= ext4_handle_dirty_metadata(handle
, inode
, bh
);
1077 /* correct old index */
1079 err
= ext4_ext_get_access(handle
, inode
, path
+ i
);
1082 le16_add_cpu(&path
[i
].p_hdr
->eh_entries
, -m
);
1083 err
= ext4_ext_dirty(handle
, inode
, path
+ i
);
1091 /* insert new index */
1092 err
= ext4_ext_insert_index(handle
, inode
, path
+ at
,
1093 le32_to_cpu(border
), newblock
);
1097 if (buffer_locked(bh
))
1103 /* free all allocated blocks in error case */
1104 for (i
= 0; i
< depth
; i
++) {
1107 ext4_free_blocks(handle
, inode
, NULL
, ablocks
[i
], 1,
1108 EXT4_FREE_BLOCKS_METADATA
);
1117 * ext4_ext_grow_indepth:
1118 * implements tree growing procedure:
1119 * - allocates new block
1120 * - moves top-level data (index block or leaf) into the new block
1121 * - initializes new top-level, creating index that points to the
1122 * just created block
1124 static int ext4_ext_grow_indepth(handle_t
*handle
, struct inode
*inode
,
1126 struct ext4_extent
*newext
)
1128 struct ext4_extent_header
*neh
;
1129 struct buffer_head
*bh
;
1130 ext4_fsblk_t newblock
;
1133 newblock
= ext4_ext_new_meta_block(handle
, inode
, NULL
,
1134 newext
, &err
, flags
);
1138 bh
= sb_getblk(inode
->i_sb
, newblock
);
1141 ext4_std_error(inode
->i_sb
, err
);
1146 err
= ext4_journal_get_create_access(handle
, bh
);
1152 /* move top-level index/leaf into new block */
1153 memmove(bh
->b_data
, EXT4_I(inode
)->i_data
,
1154 sizeof(EXT4_I(inode
)->i_data
));
1156 /* set size of new block */
1157 neh
= ext_block_hdr(bh
);
1158 /* old root could have indexes or leaves
1159 * so calculate e_max right way */
1160 if (ext_depth(inode
))
1161 neh
->eh_max
= cpu_to_le16(ext4_ext_space_block_idx(inode
, 0));
1163 neh
->eh_max
= cpu_to_le16(ext4_ext_space_block(inode
, 0));
1164 neh
->eh_magic
= EXT4_EXT_MAGIC
;
1165 ext4_extent_block_csum_set(inode
, neh
);
1166 set_buffer_uptodate(bh
);
1169 err
= ext4_handle_dirty_metadata(handle
, inode
, bh
);
1173 /* Update top-level index: num,max,pointer */
1174 neh
= ext_inode_hdr(inode
);
1175 neh
->eh_entries
= cpu_to_le16(1);
1176 ext4_idx_store_pblock(EXT_FIRST_INDEX(neh
), newblock
);
1177 if (neh
->eh_depth
== 0) {
1178 /* Root extent block becomes index block */
1179 neh
->eh_max
= cpu_to_le16(ext4_ext_space_root_idx(inode
, 0));
1180 EXT_FIRST_INDEX(neh
)->ei_block
=
1181 EXT_FIRST_EXTENT(neh
)->ee_block
;
1183 ext_debug("new root: num %d(%d), lblock %d, ptr %llu\n",
1184 le16_to_cpu(neh
->eh_entries
), le16_to_cpu(neh
->eh_max
),
1185 le32_to_cpu(EXT_FIRST_INDEX(neh
)->ei_block
),
1186 ext4_idx_pblock(EXT_FIRST_INDEX(neh
)));
1188 le16_add_cpu(&neh
->eh_depth
, 1);
1189 ext4_mark_inode_dirty(handle
, inode
);
1197 * ext4_ext_create_new_leaf:
1198 * finds empty index and adds new leaf.
1199 * if no free index is found, then it requests in-depth growing.
1201 static int ext4_ext_create_new_leaf(handle_t
*handle
, struct inode
*inode
,
1203 struct ext4_ext_path
*path
,
1204 struct ext4_extent
*newext
)
1206 struct ext4_ext_path
*curp
;
1207 int depth
, i
, err
= 0;
1210 i
= depth
= ext_depth(inode
);
1212 /* walk up to the tree and look for free index entry */
1213 curp
= path
+ depth
;
1214 while (i
> 0 && !EXT_HAS_FREE_INDEX(curp
)) {
1219 /* we use already allocated block for index block,
1220 * so subsequent data blocks should be contiguous */
1221 if (EXT_HAS_FREE_INDEX(curp
)) {
1222 /* if we found index with free entry, then use that
1223 * entry: create all needed subtree and add new leaf */
1224 err
= ext4_ext_split(handle
, inode
, flags
, path
, newext
, i
);
1229 ext4_ext_drop_refs(path
);
1230 path
= ext4_ext_find_extent(inode
,
1231 (ext4_lblk_t
)le32_to_cpu(newext
->ee_block
),
1234 err
= PTR_ERR(path
);
1236 /* tree is full, time to grow in depth */
1237 err
= ext4_ext_grow_indepth(handle
, inode
, flags
, newext
);
1242 ext4_ext_drop_refs(path
);
1243 path
= ext4_ext_find_extent(inode
,
1244 (ext4_lblk_t
)le32_to_cpu(newext
->ee_block
),
1247 err
= PTR_ERR(path
);
1252 * only first (depth 0 -> 1) produces free space;
1253 * in all other cases we have to split the grown tree
1255 depth
= ext_depth(inode
);
1256 if (path
[depth
].p_hdr
->eh_entries
== path
[depth
].p_hdr
->eh_max
) {
1257 /* now we need to split */
1267 * search the closest allocated block to the left for *logical
1268 * and returns it at @logical + it's physical address at @phys
1269 * if *logical is the smallest allocated block, the function
1270 * returns 0 at @phys
1271 * return value contains 0 (success) or error code
1273 static int ext4_ext_search_left(struct inode
*inode
,
1274 struct ext4_ext_path
*path
,
1275 ext4_lblk_t
*logical
, ext4_fsblk_t
*phys
)
1277 struct ext4_extent_idx
*ix
;
1278 struct ext4_extent
*ex
;
1281 if (unlikely(path
== NULL
)) {
1282 EXT4_ERROR_INODE(inode
, "path == NULL *logical %d!", *logical
);
1285 depth
= path
->p_depth
;
1288 if (depth
== 0 && path
->p_ext
== NULL
)
1291 /* usually extent in the path covers blocks smaller
1292 * then *logical, but it can be that extent is the
1293 * first one in the file */
1295 ex
= path
[depth
].p_ext
;
1296 ee_len
= ext4_ext_get_actual_len(ex
);
1297 if (*logical
< le32_to_cpu(ex
->ee_block
)) {
1298 if (unlikely(EXT_FIRST_EXTENT(path
[depth
].p_hdr
) != ex
)) {
1299 EXT4_ERROR_INODE(inode
,
1300 "EXT_FIRST_EXTENT != ex *logical %d ee_block %d!",
1301 *logical
, le32_to_cpu(ex
->ee_block
));
1304 while (--depth
>= 0) {
1305 ix
= path
[depth
].p_idx
;
1306 if (unlikely(ix
!= EXT_FIRST_INDEX(path
[depth
].p_hdr
))) {
1307 EXT4_ERROR_INODE(inode
,
1308 "ix (%d) != EXT_FIRST_INDEX (%d) (depth %d)!",
1309 ix
!= NULL
? le32_to_cpu(ix
->ei_block
) : 0,
1310 EXT_FIRST_INDEX(path
[depth
].p_hdr
) != NULL
?
1311 le32_to_cpu(EXT_FIRST_INDEX(path
[depth
].p_hdr
)->ei_block
) : 0,
1319 if (unlikely(*logical
< (le32_to_cpu(ex
->ee_block
) + ee_len
))) {
1320 EXT4_ERROR_INODE(inode
,
1321 "logical %d < ee_block %d + ee_len %d!",
1322 *logical
, le32_to_cpu(ex
->ee_block
), ee_len
);
1326 *logical
= le32_to_cpu(ex
->ee_block
) + ee_len
- 1;
1327 *phys
= ext4_ext_pblock(ex
) + ee_len
- 1;
1332 * search the closest allocated block to the right for *logical
1333 * and returns it at @logical + it's physical address at @phys
1334 * if *logical is the largest allocated block, the function
1335 * returns 0 at @phys
1336 * return value contains 0 (success) or error code
1338 static int ext4_ext_search_right(struct inode
*inode
,
1339 struct ext4_ext_path
*path
,
1340 ext4_lblk_t
*logical
, ext4_fsblk_t
*phys
,
1341 struct ext4_extent
**ret_ex
)
1343 struct buffer_head
*bh
= NULL
;
1344 struct ext4_extent_header
*eh
;
1345 struct ext4_extent_idx
*ix
;
1346 struct ext4_extent
*ex
;
1348 int depth
; /* Note, NOT eh_depth; depth from top of tree */
1351 if (unlikely(path
== NULL
)) {
1352 EXT4_ERROR_INODE(inode
, "path == NULL *logical %d!", *logical
);
1355 depth
= path
->p_depth
;
1358 if (depth
== 0 && path
->p_ext
== NULL
)
1361 /* usually extent in the path covers blocks smaller
1362 * then *logical, but it can be that extent is the
1363 * first one in the file */
1365 ex
= path
[depth
].p_ext
;
1366 ee_len
= ext4_ext_get_actual_len(ex
);
1367 if (*logical
< le32_to_cpu(ex
->ee_block
)) {
1368 if (unlikely(EXT_FIRST_EXTENT(path
[depth
].p_hdr
) != ex
)) {
1369 EXT4_ERROR_INODE(inode
,
1370 "first_extent(path[%d].p_hdr) != ex",
1374 while (--depth
>= 0) {
1375 ix
= path
[depth
].p_idx
;
1376 if (unlikely(ix
!= EXT_FIRST_INDEX(path
[depth
].p_hdr
))) {
1377 EXT4_ERROR_INODE(inode
,
1378 "ix != EXT_FIRST_INDEX *logical %d!",
1386 if (unlikely(*logical
< (le32_to_cpu(ex
->ee_block
) + ee_len
))) {
1387 EXT4_ERROR_INODE(inode
,
1388 "logical %d < ee_block %d + ee_len %d!",
1389 *logical
, le32_to_cpu(ex
->ee_block
), ee_len
);
1393 if (ex
!= EXT_LAST_EXTENT(path
[depth
].p_hdr
)) {
1394 /* next allocated block in this leaf */
1399 /* go up and search for index to the right */
1400 while (--depth
>= 0) {
1401 ix
= path
[depth
].p_idx
;
1402 if (ix
!= EXT_LAST_INDEX(path
[depth
].p_hdr
))
1406 /* we've gone up to the root and found no index to the right */
1410 /* we've found index to the right, let's
1411 * follow it and find the closest allocated
1412 * block to the right */
1414 block
= ext4_idx_pblock(ix
);
1415 while (++depth
< path
->p_depth
) {
1416 bh
= sb_bread(inode
->i_sb
, block
);
1419 eh
= ext_block_hdr(bh
);
1420 /* subtract from p_depth to get proper eh_depth */
1421 if (ext4_ext_check_block(inode
, eh
,
1422 path
->p_depth
- depth
, bh
)) {
1426 ix
= EXT_FIRST_INDEX(eh
);
1427 block
= ext4_idx_pblock(ix
);
1431 bh
= sb_bread(inode
->i_sb
, block
);
1434 eh
= ext_block_hdr(bh
);
1435 if (ext4_ext_check_block(inode
, eh
, path
->p_depth
- depth
, bh
)) {
1439 ex
= EXT_FIRST_EXTENT(eh
);
1441 *logical
= le32_to_cpu(ex
->ee_block
);
1442 *phys
= ext4_ext_pblock(ex
);
1450 * ext4_ext_next_allocated_block:
1451 * returns allocated block in subsequent extent or EXT_MAX_BLOCKS.
1452 * NOTE: it considers block number from index entry as
1453 * allocated block. Thus, index entries have to be consistent
1457 ext4_ext_next_allocated_block(struct ext4_ext_path
*path
)
1461 BUG_ON(path
== NULL
);
1462 depth
= path
->p_depth
;
1464 if (depth
== 0 && path
->p_ext
== NULL
)
1465 return EXT_MAX_BLOCKS
;
1467 while (depth
>= 0) {
1468 if (depth
== path
->p_depth
) {
1470 if (path
[depth
].p_ext
&&
1471 path
[depth
].p_ext
!=
1472 EXT_LAST_EXTENT(path
[depth
].p_hdr
))
1473 return le32_to_cpu(path
[depth
].p_ext
[1].ee_block
);
1476 if (path
[depth
].p_idx
!=
1477 EXT_LAST_INDEX(path
[depth
].p_hdr
))
1478 return le32_to_cpu(path
[depth
].p_idx
[1].ei_block
);
1483 return EXT_MAX_BLOCKS
;
1487 * ext4_ext_next_leaf_block:
1488 * returns first allocated block from next leaf or EXT_MAX_BLOCKS
1490 static ext4_lblk_t
ext4_ext_next_leaf_block(struct ext4_ext_path
*path
)
1494 BUG_ON(path
== NULL
);
1495 depth
= path
->p_depth
;
1497 /* zero-tree has no leaf blocks at all */
1499 return EXT_MAX_BLOCKS
;
1501 /* go to index block */
1504 while (depth
>= 0) {
1505 if (path
[depth
].p_idx
!=
1506 EXT_LAST_INDEX(path
[depth
].p_hdr
))
1507 return (ext4_lblk_t
)
1508 le32_to_cpu(path
[depth
].p_idx
[1].ei_block
);
1512 return EXT_MAX_BLOCKS
;
1516 * ext4_ext_correct_indexes:
1517 * if leaf gets modified and modified extent is first in the leaf,
1518 * then we have to correct all indexes above.
1519 * TODO: do we need to correct tree in all cases?
1521 static int ext4_ext_correct_indexes(handle_t
*handle
, struct inode
*inode
,
1522 struct ext4_ext_path
*path
)
1524 struct ext4_extent_header
*eh
;
1525 int depth
= ext_depth(inode
);
1526 struct ext4_extent
*ex
;
1530 eh
= path
[depth
].p_hdr
;
1531 ex
= path
[depth
].p_ext
;
1533 if (unlikely(ex
== NULL
|| eh
== NULL
)) {
1534 EXT4_ERROR_INODE(inode
,
1535 "ex %p == NULL or eh %p == NULL", ex
, eh
);
1540 /* there is no tree at all */
1544 if (ex
!= EXT_FIRST_EXTENT(eh
)) {
1545 /* we correct tree if first leaf got modified only */
1550 * TODO: we need correction if border is smaller than current one
1553 border
= path
[depth
].p_ext
->ee_block
;
1554 err
= ext4_ext_get_access(handle
, inode
, path
+ k
);
1557 path
[k
].p_idx
->ei_block
= border
;
1558 err
= ext4_ext_dirty(handle
, inode
, path
+ k
);
1563 /* change all left-side indexes */
1564 if (path
[k
+1].p_idx
!= EXT_FIRST_INDEX(path
[k
+1].p_hdr
))
1566 err
= ext4_ext_get_access(handle
, inode
, path
+ k
);
1569 path
[k
].p_idx
->ei_block
= border
;
1570 err
= ext4_ext_dirty(handle
, inode
, path
+ k
);
1579 ext4_can_extents_be_merged(struct inode
*inode
, struct ext4_extent
*ex1
,
1580 struct ext4_extent
*ex2
)
1582 unsigned short ext1_ee_len
, ext2_ee_len
, max_len
;
1585 * Make sure that either both extents are uninitialized, or
1588 if (ext4_ext_is_uninitialized(ex1
) ^ ext4_ext_is_uninitialized(ex2
))
1591 if (ext4_ext_is_uninitialized(ex1
))
1592 max_len
= EXT_UNINIT_MAX_LEN
;
1594 max_len
= EXT_INIT_MAX_LEN
;
1596 ext1_ee_len
= ext4_ext_get_actual_len(ex1
);
1597 ext2_ee_len
= ext4_ext_get_actual_len(ex2
);
1599 if (le32_to_cpu(ex1
->ee_block
) + ext1_ee_len
!=
1600 le32_to_cpu(ex2
->ee_block
))
1604 * To allow future support for preallocated extents to be added
1605 * as an RO_COMPAT feature, refuse to merge to extents if
1606 * this can result in the top bit of ee_len being set.
1608 if (ext1_ee_len
+ ext2_ee_len
> max_len
)
1610 #ifdef AGGRESSIVE_TEST
1611 if (ext1_ee_len
>= 4)
1615 if (ext4_ext_pblock(ex1
) + ext1_ee_len
== ext4_ext_pblock(ex2
))
1621 * This function tries to merge the "ex" extent to the next extent in the tree.
1622 * It always tries to merge towards right. If you want to merge towards
1623 * left, pass "ex - 1" as argument instead of "ex".
1624 * Returns 0 if the extents (ex and ex+1) were _not_ merged and returns
1625 * 1 if they got merged.
1627 static int ext4_ext_try_to_merge_right(struct inode
*inode
,
1628 struct ext4_ext_path
*path
,
1629 struct ext4_extent
*ex
)
1631 struct ext4_extent_header
*eh
;
1632 unsigned int depth
, len
;
1634 int uninitialized
= 0;
1636 depth
= ext_depth(inode
);
1637 BUG_ON(path
[depth
].p_hdr
== NULL
);
1638 eh
= path
[depth
].p_hdr
;
1640 while (ex
< EXT_LAST_EXTENT(eh
)) {
1641 if (!ext4_can_extents_be_merged(inode
, ex
, ex
+ 1))
1643 /* merge with next extent! */
1644 if (ext4_ext_is_uninitialized(ex
))
1646 ex
->ee_len
= cpu_to_le16(ext4_ext_get_actual_len(ex
)
1647 + ext4_ext_get_actual_len(ex
+ 1));
1649 ext4_ext_mark_uninitialized(ex
);
1651 if (ex
+ 1 < EXT_LAST_EXTENT(eh
)) {
1652 len
= (EXT_LAST_EXTENT(eh
) - ex
- 1)
1653 * sizeof(struct ext4_extent
);
1654 memmove(ex
+ 1, ex
+ 2, len
);
1656 le16_add_cpu(&eh
->eh_entries
, -1);
1658 WARN_ON(eh
->eh_entries
== 0);
1659 if (!eh
->eh_entries
)
1660 EXT4_ERROR_INODE(inode
, "eh->eh_entries = 0!");
1667 * This function does a very simple check to see if we can collapse
1668 * an extent tree with a single extent tree leaf block into the inode.
1670 static void ext4_ext_try_to_merge_up(handle_t
*handle
,
1671 struct inode
*inode
,
1672 struct ext4_ext_path
*path
)
1675 unsigned max_root
= ext4_ext_space_root(inode
, 0);
1678 if ((path
[0].p_depth
!= 1) ||
1679 (le16_to_cpu(path
[0].p_hdr
->eh_entries
) != 1) ||
1680 (le16_to_cpu(path
[1].p_hdr
->eh_entries
) > max_root
))
1684 * We need to modify the block allocation bitmap and the block
1685 * group descriptor to release the extent tree block. If we
1686 * can't get the journal credits, give up.
1688 if (ext4_journal_extend(handle
, 2))
1692 * Copy the extent data up to the inode
1694 blk
= ext4_idx_pblock(path
[0].p_idx
);
1695 s
= le16_to_cpu(path
[1].p_hdr
->eh_entries
) *
1696 sizeof(struct ext4_extent_idx
);
1697 s
+= sizeof(struct ext4_extent_header
);
1699 memcpy(path
[0].p_hdr
, path
[1].p_hdr
, s
);
1700 path
[0].p_depth
= 0;
1701 path
[0].p_ext
= EXT_FIRST_EXTENT(path
[0].p_hdr
) +
1702 (path
[1].p_ext
- EXT_FIRST_EXTENT(path
[1].p_hdr
));
1703 path
[0].p_hdr
->eh_max
= cpu_to_le16(max_root
);
1705 brelse(path
[1].p_bh
);
1706 ext4_free_blocks(handle
, inode
, NULL
, blk
, 1,
1707 EXT4_FREE_BLOCKS_METADATA
| EXT4_FREE_BLOCKS_FORGET
);
1711 * This function tries to merge the @ex extent to neighbours in the tree.
1712 * return 1 if merge left else 0.
1714 static void ext4_ext_try_to_merge(handle_t
*handle
,
1715 struct inode
*inode
,
1716 struct ext4_ext_path
*path
,
1717 struct ext4_extent
*ex
) {
1718 struct ext4_extent_header
*eh
;
1722 depth
= ext_depth(inode
);
1723 BUG_ON(path
[depth
].p_hdr
== NULL
);
1724 eh
= path
[depth
].p_hdr
;
1726 if (ex
> EXT_FIRST_EXTENT(eh
))
1727 merge_done
= ext4_ext_try_to_merge_right(inode
, path
, ex
- 1);
1730 (void) ext4_ext_try_to_merge_right(inode
, path
, ex
);
1732 ext4_ext_try_to_merge_up(handle
, inode
, path
);
1736 * check if a portion of the "newext" extent overlaps with an
1739 * If there is an overlap discovered, it updates the length of the newext
1740 * such that there will be no overlap, and then returns 1.
1741 * If there is no overlap found, it returns 0.
1743 static unsigned int ext4_ext_check_overlap(struct ext4_sb_info
*sbi
,
1744 struct inode
*inode
,
1745 struct ext4_extent
*newext
,
1746 struct ext4_ext_path
*path
)
1749 unsigned int depth
, len1
;
1750 unsigned int ret
= 0;
1752 b1
= le32_to_cpu(newext
->ee_block
);
1753 len1
= ext4_ext_get_actual_len(newext
);
1754 depth
= ext_depth(inode
);
1755 if (!path
[depth
].p_ext
)
1757 b2
= le32_to_cpu(path
[depth
].p_ext
->ee_block
);
1758 b2
&= ~(sbi
->s_cluster_ratio
- 1);
1761 * get the next allocated block if the extent in the path
1762 * is before the requested block(s)
1765 b2
= ext4_ext_next_allocated_block(path
);
1766 if (b2
== EXT_MAX_BLOCKS
)
1768 b2
&= ~(sbi
->s_cluster_ratio
- 1);
1771 /* check for wrap through zero on extent logical start block*/
1772 if (b1
+ len1
< b1
) {
1773 len1
= EXT_MAX_BLOCKS
- b1
;
1774 newext
->ee_len
= cpu_to_le16(len1
);
1778 /* check for overlap */
1779 if (b1
+ len1
> b2
) {
1780 newext
->ee_len
= cpu_to_le16(b2
- b1
);
1788 * ext4_ext_insert_extent:
1789 * tries to merge requsted extent into the existing extent or
1790 * inserts requested extent as new one into the tree,
1791 * creating new leaf in the no-space case.
1793 int ext4_ext_insert_extent(handle_t
*handle
, struct inode
*inode
,
1794 struct ext4_ext_path
*path
,
1795 struct ext4_extent
*newext
, int flag
)
1797 struct ext4_extent_header
*eh
;
1798 struct ext4_extent
*ex
, *fex
;
1799 struct ext4_extent
*nearex
; /* nearest extent */
1800 struct ext4_ext_path
*npath
= NULL
;
1801 int depth
, len
, err
;
1803 unsigned uninitialized
= 0;
1806 if (unlikely(ext4_ext_get_actual_len(newext
) == 0)) {
1807 EXT4_ERROR_INODE(inode
, "ext4_ext_get_actual_len(newext) == 0");
1810 depth
= ext_depth(inode
);
1811 ex
= path
[depth
].p_ext
;
1812 if (unlikely(path
[depth
].p_hdr
== NULL
)) {
1813 EXT4_ERROR_INODE(inode
, "path[%d].p_hdr == NULL", depth
);
1817 /* try to insert block into found extent and return */
1818 if (ex
&& !(flag
& EXT4_GET_BLOCKS_PRE_IO
)
1819 && ext4_can_extents_be_merged(inode
, ex
, newext
)) {
1820 ext_debug("append [%d]%d block to %u:[%d]%d (from %llu)\n",
1821 ext4_ext_is_uninitialized(newext
),
1822 ext4_ext_get_actual_len(newext
),
1823 le32_to_cpu(ex
->ee_block
),
1824 ext4_ext_is_uninitialized(ex
),
1825 ext4_ext_get_actual_len(ex
),
1826 ext4_ext_pblock(ex
));
1827 err
= ext4_ext_get_access(handle
, inode
, path
+ depth
);
1832 * ext4_can_extents_be_merged should have checked that either
1833 * both extents are uninitialized, or both aren't. Thus we
1834 * need to check only one of them here.
1836 if (ext4_ext_is_uninitialized(ex
))
1838 ex
->ee_len
= cpu_to_le16(ext4_ext_get_actual_len(ex
)
1839 + ext4_ext_get_actual_len(newext
));
1841 ext4_ext_mark_uninitialized(ex
);
1842 eh
= path
[depth
].p_hdr
;
1847 depth
= ext_depth(inode
);
1848 eh
= path
[depth
].p_hdr
;
1849 if (le16_to_cpu(eh
->eh_entries
) < le16_to_cpu(eh
->eh_max
))
1852 /* probably next leaf has space for us? */
1853 fex
= EXT_LAST_EXTENT(eh
);
1854 next
= EXT_MAX_BLOCKS
;
1855 if (le32_to_cpu(newext
->ee_block
) > le32_to_cpu(fex
->ee_block
))
1856 next
= ext4_ext_next_leaf_block(path
);
1857 if (next
!= EXT_MAX_BLOCKS
) {
1858 ext_debug("next leaf block - %u\n", next
);
1859 BUG_ON(npath
!= NULL
);
1860 npath
= ext4_ext_find_extent(inode
, next
, NULL
);
1862 return PTR_ERR(npath
);
1863 BUG_ON(npath
->p_depth
!= path
->p_depth
);
1864 eh
= npath
[depth
].p_hdr
;
1865 if (le16_to_cpu(eh
->eh_entries
) < le16_to_cpu(eh
->eh_max
)) {
1866 ext_debug("next leaf isn't full(%d)\n",
1867 le16_to_cpu(eh
->eh_entries
));
1871 ext_debug("next leaf has no free space(%d,%d)\n",
1872 le16_to_cpu(eh
->eh_entries
), le16_to_cpu(eh
->eh_max
));
1876 * There is no free space in the found leaf.
1877 * We're gonna add a new leaf in the tree.
1879 if (flag
& EXT4_GET_BLOCKS_PUNCH_OUT_EXT
)
1880 flags
= EXT4_MB_USE_ROOT_BLOCKS
;
1881 err
= ext4_ext_create_new_leaf(handle
, inode
, flags
, path
, newext
);
1884 depth
= ext_depth(inode
);
1885 eh
= path
[depth
].p_hdr
;
1888 nearex
= path
[depth
].p_ext
;
1890 err
= ext4_ext_get_access(handle
, inode
, path
+ depth
);
1895 /* there is no extent in this leaf, create first one */
1896 ext_debug("first extent in the leaf: %u:%llu:[%d]%d\n",
1897 le32_to_cpu(newext
->ee_block
),
1898 ext4_ext_pblock(newext
),
1899 ext4_ext_is_uninitialized(newext
),
1900 ext4_ext_get_actual_len(newext
));
1901 nearex
= EXT_FIRST_EXTENT(eh
);
1903 if (le32_to_cpu(newext
->ee_block
)
1904 > le32_to_cpu(nearex
->ee_block
)) {
1906 ext_debug("insert %u:%llu:[%d]%d before: "
1908 le32_to_cpu(newext
->ee_block
),
1909 ext4_ext_pblock(newext
),
1910 ext4_ext_is_uninitialized(newext
),
1911 ext4_ext_get_actual_len(newext
),
1916 BUG_ON(newext
->ee_block
== nearex
->ee_block
);
1917 ext_debug("insert %u:%llu:[%d]%d after: "
1919 le32_to_cpu(newext
->ee_block
),
1920 ext4_ext_pblock(newext
),
1921 ext4_ext_is_uninitialized(newext
),
1922 ext4_ext_get_actual_len(newext
),
1925 len
= EXT_LAST_EXTENT(eh
) - nearex
+ 1;
1927 ext_debug("insert %u:%llu:[%d]%d: "
1928 "move %d extents from 0x%p to 0x%p\n",
1929 le32_to_cpu(newext
->ee_block
),
1930 ext4_ext_pblock(newext
),
1931 ext4_ext_is_uninitialized(newext
),
1932 ext4_ext_get_actual_len(newext
),
1933 len
, nearex
, nearex
+ 1);
1934 memmove(nearex
+ 1, nearex
,
1935 len
* sizeof(struct ext4_extent
));
1939 le16_add_cpu(&eh
->eh_entries
, 1);
1940 path
[depth
].p_ext
= nearex
;
1941 nearex
->ee_block
= newext
->ee_block
;
1942 ext4_ext_store_pblock(nearex
, ext4_ext_pblock(newext
));
1943 nearex
->ee_len
= newext
->ee_len
;
1946 /* try to merge extents */
1947 if (!(flag
& EXT4_GET_BLOCKS_PRE_IO
))
1948 ext4_ext_try_to_merge(handle
, inode
, path
, nearex
);
1951 /* time to correct all indexes above */
1952 err
= ext4_ext_correct_indexes(handle
, inode
, path
);
1956 err
= ext4_ext_dirty(handle
, inode
, path
+ path
->p_depth
);
1960 ext4_ext_drop_refs(npath
);
1963 ext4_ext_invalidate_cache(inode
);
1967 static int ext4_fill_fiemap_extents(struct inode
*inode
,
1968 ext4_lblk_t block
, ext4_lblk_t num
,
1969 struct fiemap_extent_info
*fieinfo
)
1971 struct ext4_ext_path
*path
= NULL
;
1972 struct ext4_ext_cache newex
;
1973 struct ext4_extent
*ex
;
1974 ext4_lblk_t next
, next_del
, start
= 0, end
= 0;
1975 ext4_lblk_t last
= block
+ num
;
1976 int exists
, depth
= 0, err
= 0;
1977 unsigned int flags
= 0;
1978 unsigned char blksize_bits
= inode
->i_sb
->s_blocksize_bits
;
1980 while (block
< last
&& block
!= EXT_MAX_BLOCKS
) {
1982 /* find extent for this block */
1983 down_read(&EXT4_I(inode
)->i_data_sem
);
1985 if (path
&& ext_depth(inode
) != depth
) {
1986 /* depth was changed. we have to realloc path */
1991 path
= ext4_ext_find_extent(inode
, block
, path
);
1993 up_read(&EXT4_I(inode
)->i_data_sem
);
1994 err
= PTR_ERR(path
);
1999 depth
= ext_depth(inode
);
2000 if (unlikely(path
[depth
].p_hdr
== NULL
)) {
2001 up_read(&EXT4_I(inode
)->i_data_sem
);
2002 EXT4_ERROR_INODE(inode
, "path[%d].p_hdr == NULL", depth
);
2006 ex
= path
[depth
].p_ext
;
2007 next
= ext4_ext_next_allocated_block(path
);
2008 ext4_ext_drop_refs(path
);
2013 /* there is no extent yet, so try to allocate
2014 * all requested space */
2017 } else if (le32_to_cpu(ex
->ee_block
) > block
) {
2018 /* need to allocate space before found extent */
2020 end
= le32_to_cpu(ex
->ee_block
);
2021 if (block
+ num
< end
)
2023 } else if (block
>= le32_to_cpu(ex
->ee_block
)
2024 + ext4_ext_get_actual_len(ex
)) {
2025 /* need to allocate space after found extent */
2030 } else if (block
>= le32_to_cpu(ex
->ee_block
)) {
2032 * some part of requested space is covered
2036 end
= le32_to_cpu(ex
->ee_block
)
2037 + ext4_ext_get_actual_len(ex
);
2038 if (block
+ num
< end
)
2044 BUG_ON(end
<= start
);
2047 newex
.ec_block
= start
;
2048 newex
.ec_len
= end
- start
;
2051 newex
.ec_block
= le32_to_cpu(ex
->ee_block
);
2052 newex
.ec_len
= ext4_ext_get_actual_len(ex
);
2053 newex
.ec_start
= ext4_ext_pblock(ex
);
2054 if (ext4_ext_is_uninitialized(ex
))
2055 flags
|= FIEMAP_EXTENT_UNWRITTEN
;
2059 * Find delayed extent and update newex accordingly. We call
2060 * it even in !exists case to find out whether newex is the
2061 * last existing extent or not.
2063 next_del
= ext4_find_delayed_extent(inode
, &newex
);
2064 if (!exists
&& next_del
) {
2066 flags
|= FIEMAP_EXTENT_DELALLOC
;
2068 up_read(&EXT4_I(inode
)->i_data_sem
);
2070 if (unlikely(newex
.ec_len
== 0)) {
2071 EXT4_ERROR_INODE(inode
, "newex.ec_len == 0");
2076 /* This is possible iff next == next_del == EXT_MAX_BLOCKS */
2077 if (next
== next_del
) {
2078 flags
|= FIEMAP_EXTENT_LAST
;
2079 if (unlikely(next_del
!= EXT_MAX_BLOCKS
||
2080 next
!= EXT_MAX_BLOCKS
)) {
2081 EXT4_ERROR_INODE(inode
,
2082 "next extent == %u, next "
2083 "delalloc extent = %u",
2091 err
= fiemap_fill_next_extent(fieinfo
,
2092 (__u64
)newex
.ec_block
<< blksize_bits
,
2093 (__u64
)newex
.ec_start
<< blksize_bits
,
2094 (__u64
)newex
.ec_len
<< blksize_bits
,
2104 block
= newex
.ec_block
+ newex
.ec_len
;
2108 ext4_ext_drop_refs(path
);
2116 ext4_ext_put_in_cache(struct inode
*inode
, ext4_lblk_t block
,
2117 __u32 len
, ext4_fsblk_t start
)
2119 struct ext4_ext_cache
*cex
;
2121 spin_lock(&EXT4_I(inode
)->i_block_reservation_lock
);
2122 trace_ext4_ext_put_in_cache(inode
, block
, len
, start
);
2123 cex
= &EXT4_I(inode
)->i_cached_extent
;
2124 cex
->ec_block
= block
;
2126 cex
->ec_start
= start
;
2127 spin_unlock(&EXT4_I(inode
)->i_block_reservation_lock
);
2131 * ext4_ext_put_gap_in_cache:
2132 * calculate boundaries of the gap that the requested block fits into
2133 * and cache this gap
2136 ext4_ext_put_gap_in_cache(struct inode
*inode
, struct ext4_ext_path
*path
,
2139 int depth
= ext_depth(inode
);
2142 struct ext4_extent
*ex
;
2144 ex
= path
[depth
].p_ext
;
2146 /* there is no extent yet, so gap is [0;-] */
2148 len
= EXT_MAX_BLOCKS
;
2149 ext_debug("cache gap(whole file):");
2150 } else if (block
< le32_to_cpu(ex
->ee_block
)) {
2152 len
= le32_to_cpu(ex
->ee_block
) - block
;
2153 ext_debug("cache gap(before): %u [%u:%u]",
2155 le32_to_cpu(ex
->ee_block
),
2156 ext4_ext_get_actual_len(ex
));
2157 } else if (block
>= le32_to_cpu(ex
->ee_block
)
2158 + ext4_ext_get_actual_len(ex
)) {
2160 lblock
= le32_to_cpu(ex
->ee_block
)
2161 + ext4_ext_get_actual_len(ex
);
2163 next
= ext4_ext_next_allocated_block(path
);
2164 ext_debug("cache gap(after): [%u:%u] %u",
2165 le32_to_cpu(ex
->ee_block
),
2166 ext4_ext_get_actual_len(ex
),
2168 BUG_ON(next
== lblock
);
2169 len
= next
- lblock
;
2175 ext_debug(" -> %u:%lu\n", lblock
, len
);
2176 ext4_ext_put_in_cache(inode
, lblock
, len
, 0);
2180 * ext4_ext_in_cache()
2181 * Checks to see if the given block is in the cache.
2182 * If it is, the cached extent is stored in the given
2183 * cache extent pointer.
2185 * @inode: The files inode
2186 * @block: The block to look for in the cache
2187 * @ex: Pointer where the cached extent will be stored
2188 * if it contains block
2190 * Return 0 if cache is invalid; 1 if the cache is valid
2193 ext4_ext_in_cache(struct inode
*inode
, ext4_lblk_t block
,
2194 struct ext4_extent
*ex
)
2196 struct ext4_ext_cache
*cex
;
2197 struct ext4_sb_info
*sbi
;
2201 * We borrow i_block_reservation_lock to protect i_cached_extent
2203 spin_lock(&EXT4_I(inode
)->i_block_reservation_lock
);
2204 cex
= &EXT4_I(inode
)->i_cached_extent
;
2205 sbi
= EXT4_SB(inode
->i_sb
);
2207 /* has cache valid data? */
2208 if (cex
->ec_len
== 0)
2211 if (in_range(block
, cex
->ec_block
, cex
->ec_len
)) {
2212 ex
->ee_block
= cpu_to_le32(cex
->ec_block
);
2213 ext4_ext_store_pblock(ex
, cex
->ec_start
);
2214 ex
->ee_len
= cpu_to_le16(cex
->ec_len
);
2215 ext_debug("%u cached by %u:%u:%llu\n",
2217 cex
->ec_block
, cex
->ec_len
, cex
->ec_start
);
2221 trace_ext4_ext_in_cache(inode
, block
, ret
);
2222 spin_unlock(&EXT4_I(inode
)->i_block_reservation_lock
);
2228 * removes index from the index block.
2230 static int ext4_ext_rm_idx(handle_t
*handle
, struct inode
*inode
,
2231 struct ext4_ext_path
*path
)
2236 /* free index block */
2238 leaf
= ext4_idx_pblock(path
->p_idx
);
2239 if (unlikely(path
->p_hdr
->eh_entries
== 0)) {
2240 EXT4_ERROR_INODE(inode
, "path->p_hdr->eh_entries == 0");
2243 err
= ext4_ext_get_access(handle
, inode
, path
);
2247 if (path
->p_idx
!= EXT_LAST_INDEX(path
->p_hdr
)) {
2248 int len
= EXT_LAST_INDEX(path
->p_hdr
) - path
->p_idx
;
2249 len
*= sizeof(struct ext4_extent_idx
);
2250 memmove(path
->p_idx
, path
->p_idx
+ 1, len
);
2253 le16_add_cpu(&path
->p_hdr
->eh_entries
, -1);
2254 err
= ext4_ext_dirty(handle
, inode
, path
);
2257 ext_debug("index is empty, remove it, free block %llu\n", leaf
);
2258 trace_ext4_ext_rm_idx(inode
, leaf
);
2260 ext4_free_blocks(handle
, inode
, NULL
, leaf
, 1,
2261 EXT4_FREE_BLOCKS_METADATA
| EXT4_FREE_BLOCKS_FORGET
);
2266 * ext4_ext_calc_credits_for_single_extent:
2267 * This routine returns max. credits that needed to insert an extent
2268 * to the extent tree.
2269 * When pass the actual path, the caller should calculate credits
2272 int ext4_ext_calc_credits_for_single_extent(struct inode
*inode
, int nrblocks
,
2273 struct ext4_ext_path
*path
)
2276 int depth
= ext_depth(inode
);
2279 /* probably there is space in leaf? */
2280 if (le16_to_cpu(path
[depth
].p_hdr
->eh_entries
)
2281 < le16_to_cpu(path
[depth
].p_hdr
->eh_max
)) {
2284 * There are some space in the leaf tree, no
2285 * need to account for leaf block credit
2287 * bitmaps and block group descriptor blocks
2288 * and other metadata blocks still need to be
2291 /* 1 bitmap, 1 block group descriptor */
2292 ret
= 2 + EXT4_META_TRANS_BLOCKS(inode
->i_sb
);
2297 return ext4_chunk_trans_blocks(inode
, nrblocks
);
2301 * How many index/leaf blocks need to change/allocate to modify nrblocks?
2303 * if nrblocks are fit in a single extent (chunk flag is 1), then
2304 * in the worse case, each tree level index/leaf need to be changed
2305 * if the tree split due to insert a new extent, then the old tree
2306 * index/leaf need to be updated too
2308 * If the nrblocks are discontiguous, they could cause
2309 * the whole tree split more than once, but this is really rare.
2311 int ext4_ext_index_trans_blocks(struct inode
*inode
, int nrblocks
, int chunk
)
2316 /* If we are converting the inline data, only one is needed here. */
2317 if (ext4_has_inline_data(inode
))
2320 depth
= ext_depth(inode
);
2330 static int ext4_remove_blocks(handle_t
*handle
, struct inode
*inode
,
2331 struct ext4_extent
*ex
,
2332 ext4_fsblk_t
*partial_cluster
,
2333 ext4_lblk_t from
, ext4_lblk_t to
)
2335 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
2336 unsigned short ee_len
= ext4_ext_get_actual_len(ex
);
2340 if (S_ISDIR(inode
->i_mode
) || S_ISLNK(inode
->i_mode
))
2341 flags
|= EXT4_FREE_BLOCKS_METADATA
| EXT4_FREE_BLOCKS_FORGET
;
2342 else if (ext4_should_journal_data(inode
))
2343 flags
|= EXT4_FREE_BLOCKS_FORGET
;
2346 * For bigalloc file systems, we never free a partial cluster
2347 * at the beginning of the extent. Instead, we make a note
2348 * that we tried freeing the cluster, and check to see if we
2349 * need to free it on a subsequent call to ext4_remove_blocks,
2350 * or at the end of the ext4_truncate() operation.
2352 flags
|= EXT4_FREE_BLOCKS_NOFREE_FIRST_CLUSTER
;
2354 trace_ext4_remove_blocks(inode
, ex
, from
, to
, *partial_cluster
);
2356 * If we have a partial cluster, and it's different from the
2357 * cluster of the last block, we need to explicitly free the
2358 * partial cluster here.
2360 pblk
= ext4_ext_pblock(ex
) + ee_len
- 1;
2361 if (*partial_cluster
&& (EXT4_B2C(sbi
, pblk
) != *partial_cluster
)) {
2362 ext4_free_blocks(handle
, inode
, NULL
,
2363 EXT4_C2B(sbi
, *partial_cluster
),
2364 sbi
->s_cluster_ratio
, flags
);
2365 *partial_cluster
= 0;
2368 #ifdef EXTENTS_STATS
2370 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
2371 spin_lock(&sbi
->s_ext_stats_lock
);
2372 sbi
->s_ext_blocks
+= ee_len
;
2373 sbi
->s_ext_extents
++;
2374 if (ee_len
< sbi
->s_ext_min
)
2375 sbi
->s_ext_min
= ee_len
;
2376 if (ee_len
> sbi
->s_ext_max
)
2377 sbi
->s_ext_max
= ee_len
;
2378 if (ext_depth(inode
) > sbi
->s_depth_max
)
2379 sbi
->s_depth_max
= ext_depth(inode
);
2380 spin_unlock(&sbi
->s_ext_stats_lock
);
2383 if (from
>= le32_to_cpu(ex
->ee_block
)
2384 && to
== le32_to_cpu(ex
->ee_block
) + ee_len
- 1) {
2388 num
= le32_to_cpu(ex
->ee_block
) + ee_len
- from
;
2389 pblk
= ext4_ext_pblock(ex
) + ee_len
- num
;
2390 ext_debug("free last %u blocks starting %llu\n", num
, pblk
);
2391 ext4_free_blocks(handle
, inode
, NULL
, pblk
, num
, flags
);
2393 * If the block range to be freed didn't start at the
2394 * beginning of a cluster, and we removed the entire
2395 * extent, save the partial cluster here, since we
2396 * might need to delete if we determine that the
2397 * truncate operation has removed all of the blocks in
2400 if (pblk
& (sbi
->s_cluster_ratio
- 1) &&
2402 *partial_cluster
= EXT4_B2C(sbi
, pblk
);
2404 *partial_cluster
= 0;
2405 } else if (from
== le32_to_cpu(ex
->ee_block
)
2406 && to
<= le32_to_cpu(ex
->ee_block
) + ee_len
- 1) {
2412 start
= ext4_ext_pblock(ex
);
2414 ext_debug("free first %u blocks starting %llu\n", num
, start
);
2415 ext4_free_blocks(handle
, inode
, NULL
, start
, num
, flags
);
2418 printk(KERN_INFO
"strange request: removal(2) "
2419 "%u-%u from %u:%u\n",
2420 from
, to
, le32_to_cpu(ex
->ee_block
), ee_len
);
2427 * ext4_ext_rm_leaf() Removes the extents associated with the
2428 * blocks appearing between "start" and "end", and splits the extents
2429 * if "start" and "end" appear in the same extent
2431 * @handle: The journal handle
2432 * @inode: The files inode
2433 * @path: The path to the leaf
2434 * @start: The first block to remove
2435 * @end: The last block to remove
2438 ext4_ext_rm_leaf(handle_t
*handle
, struct inode
*inode
,
2439 struct ext4_ext_path
*path
, ext4_fsblk_t
*partial_cluster
,
2440 ext4_lblk_t start
, ext4_lblk_t end
)
2442 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
2443 int err
= 0, correct_index
= 0;
2444 int depth
= ext_depth(inode
), credits
;
2445 struct ext4_extent_header
*eh
;
2448 ext4_lblk_t ex_ee_block
;
2449 unsigned short ex_ee_len
;
2450 unsigned uninitialized
= 0;
2451 struct ext4_extent
*ex
;
2453 /* the header must be checked already in ext4_ext_remove_space() */
2454 ext_debug("truncate since %u in leaf to %u\n", start
, end
);
2455 if (!path
[depth
].p_hdr
)
2456 path
[depth
].p_hdr
= ext_block_hdr(path
[depth
].p_bh
);
2457 eh
= path
[depth
].p_hdr
;
2458 if (unlikely(path
[depth
].p_hdr
== NULL
)) {
2459 EXT4_ERROR_INODE(inode
, "path[%d].p_hdr == NULL", depth
);
2462 /* find where to start removing */
2463 ex
= EXT_LAST_EXTENT(eh
);
2465 ex_ee_block
= le32_to_cpu(ex
->ee_block
);
2466 ex_ee_len
= ext4_ext_get_actual_len(ex
);
2468 trace_ext4_ext_rm_leaf(inode
, start
, ex
, *partial_cluster
);
2470 while (ex
>= EXT_FIRST_EXTENT(eh
) &&
2471 ex_ee_block
+ ex_ee_len
> start
) {
2473 if (ext4_ext_is_uninitialized(ex
))
2478 ext_debug("remove ext %u:[%d]%d\n", ex_ee_block
,
2479 uninitialized
, ex_ee_len
);
2480 path
[depth
].p_ext
= ex
;
2482 a
= ex_ee_block
> start
? ex_ee_block
: start
;
2483 b
= ex_ee_block
+ex_ee_len
- 1 < end
?
2484 ex_ee_block
+ex_ee_len
- 1 : end
;
2486 ext_debug(" border %u:%u\n", a
, b
);
2488 /* If this extent is beyond the end of the hole, skip it */
2489 if (end
< ex_ee_block
) {
2491 ex_ee_block
= le32_to_cpu(ex
->ee_block
);
2492 ex_ee_len
= ext4_ext_get_actual_len(ex
);
2494 } else if (b
!= ex_ee_block
+ ex_ee_len
- 1) {
2495 EXT4_ERROR_INODE(inode
,
2496 "can not handle truncate %u:%u "
2498 start
, end
, ex_ee_block
,
2499 ex_ee_block
+ ex_ee_len
- 1);
2502 } else if (a
!= ex_ee_block
) {
2503 /* remove tail of the extent */
2504 num
= a
- ex_ee_block
;
2506 /* remove whole extent: excellent! */
2510 * 3 for leaf, sb, and inode plus 2 (bmap and group
2511 * descriptor) for each block group; assume two block
2512 * groups plus ex_ee_len/blocks_per_block_group for
2515 credits
= 7 + 2*(ex_ee_len
/EXT4_BLOCKS_PER_GROUP(inode
->i_sb
));
2516 if (ex
== EXT_FIRST_EXTENT(eh
)) {
2518 credits
+= (ext_depth(inode
)) + 1;
2520 credits
+= EXT4_MAXQUOTAS_TRANS_BLOCKS(inode
->i_sb
);
2522 err
= ext4_ext_truncate_extend_restart(handle
, inode
, credits
);
2526 err
= ext4_ext_get_access(handle
, inode
, path
+ depth
);
2530 err
= ext4_remove_blocks(handle
, inode
, ex
, partial_cluster
,
2536 /* this extent is removed; mark slot entirely unused */
2537 ext4_ext_store_pblock(ex
, 0);
2539 ex
->ee_len
= cpu_to_le16(num
);
2541 * Do not mark uninitialized if all the blocks in the
2542 * extent have been removed.
2544 if (uninitialized
&& num
)
2545 ext4_ext_mark_uninitialized(ex
);
2547 * If the extent was completely released,
2548 * we need to remove it from the leaf
2551 if (end
!= EXT_MAX_BLOCKS
- 1) {
2553 * For hole punching, we need to scoot all the
2554 * extents up when an extent is removed so that
2555 * we dont have blank extents in the middle
2557 memmove(ex
, ex
+1, (EXT_LAST_EXTENT(eh
) - ex
) *
2558 sizeof(struct ext4_extent
));
2560 /* Now get rid of the one at the end */
2561 memset(EXT_LAST_EXTENT(eh
), 0,
2562 sizeof(struct ext4_extent
));
2564 le16_add_cpu(&eh
->eh_entries
, -1);
2566 *partial_cluster
= 0;
2568 err
= ext4_ext_dirty(handle
, inode
, path
+ depth
);
2572 ext_debug("new extent: %u:%u:%llu\n", ex_ee_block
, num
,
2573 ext4_ext_pblock(ex
));
2575 ex_ee_block
= le32_to_cpu(ex
->ee_block
);
2576 ex_ee_len
= ext4_ext_get_actual_len(ex
);
2579 if (correct_index
&& eh
->eh_entries
)
2580 err
= ext4_ext_correct_indexes(handle
, inode
, path
);
2583 * If there is still a entry in the leaf node, check to see if
2584 * it references the partial cluster. This is the only place
2585 * where it could; if it doesn't, we can free the cluster.
2587 if (*partial_cluster
&& ex
>= EXT_FIRST_EXTENT(eh
) &&
2588 (EXT4_B2C(sbi
, ext4_ext_pblock(ex
) + ex_ee_len
- 1) !=
2589 *partial_cluster
)) {
2590 int flags
= EXT4_FREE_BLOCKS_FORGET
;
2592 if (S_ISDIR(inode
->i_mode
) || S_ISLNK(inode
->i_mode
))
2593 flags
|= EXT4_FREE_BLOCKS_METADATA
;
2595 ext4_free_blocks(handle
, inode
, NULL
,
2596 EXT4_C2B(sbi
, *partial_cluster
),
2597 sbi
->s_cluster_ratio
, flags
);
2598 *partial_cluster
= 0;
2601 /* if this leaf is free, then we should
2602 * remove it from index block above */
2603 if (err
== 0 && eh
->eh_entries
== 0 && path
[depth
].p_bh
!= NULL
)
2604 err
= ext4_ext_rm_idx(handle
, inode
, path
+ depth
);
2611 * ext4_ext_more_to_rm:
2612 * returns 1 if current index has to be freed (even partial)
2615 ext4_ext_more_to_rm(struct ext4_ext_path
*path
)
2617 BUG_ON(path
->p_idx
== NULL
);
2619 if (path
->p_idx
< EXT_FIRST_INDEX(path
->p_hdr
))
2623 * if truncate on deeper level happened, it wasn't partial,
2624 * so we have to consider current index for truncation
2626 if (le16_to_cpu(path
->p_hdr
->eh_entries
) == path
->p_block
)
2631 static int ext4_ext_remove_space(struct inode
*inode
, ext4_lblk_t start
,
2634 struct super_block
*sb
= inode
->i_sb
;
2635 int depth
= ext_depth(inode
);
2636 struct ext4_ext_path
*path
= NULL
;
2637 ext4_fsblk_t partial_cluster
= 0;
2641 ext_debug("truncate since %u to %u\n", start
, end
);
2643 /* probably first extent we're gonna free will be last in block */
2644 handle
= ext4_journal_start(inode
, depth
+ 1);
2646 return PTR_ERR(handle
);
2649 ext4_ext_invalidate_cache(inode
);
2651 trace_ext4_ext_remove_space(inode
, start
, depth
);
2654 * Check if we are removing extents inside the extent tree. If that
2655 * is the case, we are going to punch a hole inside the extent tree
2656 * so we have to check whether we need to split the extent covering
2657 * the last block to remove so we can easily remove the part of it
2658 * in ext4_ext_rm_leaf().
2660 if (end
< EXT_MAX_BLOCKS
- 1) {
2661 struct ext4_extent
*ex
;
2662 ext4_lblk_t ee_block
;
2664 /* find extent for this block */
2665 path
= ext4_ext_find_extent(inode
, end
, NULL
);
2667 ext4_journal_stop(handle
);
2668 return PTR_ERR(path
);
2670 depth
= ext_depth(inode
);
2671 /* Leaf not may not exist only if inode has no blocks at all */
2672 ex
= path
[depth
].p_ext
;
2675 EXT4_ERROR_INODE(inode
,
2676 "path[%d].p_hdr == NULL",
2683 ee_block
= le32_to_cpu(ex
->ee_block
);
2686 * See if the last block is inside the extent, if so split
2687 * the extent at 'end' block so we can easily remove the
2688 * tail of the first part of the split extent in
2689 * ext4_ext_rm_leaf().
2691 if (end
>= ee_block
&&
2692 end
< ee_block
+ ext4_ext_get_actual_len(ex
) - 1) {
2695 if (ext4_ext_is_uninitialized(ex
))
2696 split_flag
= EXT4_EXT_MARK_UNINIT1
|
2697 EXT4_EXT_MARK_UNINIT2
;
2700 * Split the extent in two so that 'end' is the last
2701 * block in the first new extent
2703 err
= ext4_split_extent_at(handle
, inode
, path
,
2704 end
+ 1, split_flag
,
2705 EXT4_GET_BLOCKS_PRE_IO
|
2706 EXT4_GET_BLOCKS_PUNCH_OUT_EXT
);
2713 * We start scanning from right side, freeing all the blocks
2714 * after i_size and walking into the tree depth-wise.
2716 depth
= ext_depth(inode
);
2721 le16_to_cpu(path
[k
].p_hdr
->eh_entries
)+1;
2723 path
= kzalloc(sizeof(struct ext4_ext_path
) * (depth
+ 1),
2726 ext4_journal_stop(handle
);
2729 path
[0].p_depth
= depth
;
2730 path
[0].p_hdr
= ext_inode_hdr(inode
);
2733 if (ext4_ext_check(inode
, path
[0].p_hdr
, depth
)) {
2740 while (i
>= 0 && err
== 0) {
2742 /* this is leaf block */
2743 err
= ext4_ext_rm_leaf(handle
, inode
, path
,
2744 &partial_cluster
, start
,
2746 /* root level has p_bh == NULL, brelse() eats this */
2747 brelse(path
[i
].p_bh
);
2748 path
[i
].p_bh
= NULL
;
2753 /* this is index block */
2754 if (!path
[i
].p_hdr
) {
2755 ext_debug("initialize header\n");
2756 path
[i
].p_hdr
= ext_block_hdr(path
[i
].p_bh
);
2759 if (!path
[i
].p_idx
) {
2760 /* this level hasn't been touched yet */
2761 path
[i
].p_idx
= EXT_LAST_INDEX(path
[i
].p_hdr
);
2762 path
[i
].p_block
= le16_to_cpu(path
[i
].p_hdr
->eh_entries
)+1;
2763 ext_debug("init index ptr: hdr 0x%p, num %d\n",
2765 le16_to_cpu(path
[i
].p_hdr
->eh_entries
));
2767 /* we were already here, see at next index */
2771 ext_debug("level %d - index, first 0x%p, cur 0x%p\n",
2772 i
, EXT_FIRST_INDEX(path
[i
].p_hdr
),
2774 if (ext4_ext_more_to_rm(path
+ i
)) {
2775 struct buffer_head
*bh
;
2776 /* go to the next level */
2777 ext_debug("move to level %d (block %llu)\n",
2778 i
+ 1, ext4_idx_pblock(path
[i
].p_idx
));
2779 memset(path
+ i
+ 1, 0, sizeof(*path
));
2780 bh
= sb_bread(sb
, ext4_idx_pblock(path
[i
].p_idx
));
2782 /* should we reset i_size? */
2786 if (WARN_ON(i
+ 1 > depth
)) {
2790 if (ext4_ext_check_block(inode
, ext_block_hdr(bh
),
2791 depth
- i
- 1, bh
)) {
2795 path
[i
+ 1].p_bh
= bh
;
2797 /* save actual number of indexes since this
2798 * number is changed at the next iteration */
2799 path
[i
].p_block
= le16_to_cpu(path
[i
].p_hdr
->eh_entries
);
2802 /* we finished processing this index, go up */
2803 if (path
[i
].p_hdr
->eh_entries
== 0 && i
> 0) {
2804 /* index is empty, remove it;
2805 * handle must be already prepared by the
2806 * truncatei_leaf() */
2807 err
= ext4_ext_rm_idx(handle
, inode
, path
+ i
);
2809 /* root level has p_bh == NULL, brelse() eats this */
2810 brelse(path
[i
].p_bh
);
2811 path
[i
].p_bh
= NULL
;
2813 ext_debug("return to level %d\n", i
);
2817 trace_ext4_ext_remove_space_done(inode
, start
, depth
, partial_cluster
,
2818 path
->p_hdr
->eh_entries
);
2820 /* If we still have something in the partial cluster and we have removed
2821 * even the first extent, then we should free the blocks in the partial
2822 * cluster as well. */
2823 if (partial_cluster
&& path
->p_hdr
->eh_entries
== 0) {
2824 int flags
= EXT4_FREE_BLOCKS_FORGET
;
2826 if (S_ISDIR(inode
->i_mode
) || S_ISLNK(inode
->i_mode
))
2827 flags
|= EXT4_FREE_BLOCKS_METADATA
;
2829 ext4_free_blocks(handle
, inode
, NULL
,
2830 EXT4_C2B(EXT4_SB(sb
), partial_cluster
),
2831 EXT4_SB(sb
)->s_cluster_ratio
, flags
);
2832 partial_cluster
= 0;
2835 /* TODO: flexible tree reduction should be here */
2836 if (path
->p_hdr
->eh_entries
== 0) {
2838 * truncate to zero freed all the tree,
2839 * so we need to correct eh_depth
2841 err
= ext4_ext_get_access(handle
, inode
, path
);
2843 ext_inode_hdr(inode
)->eh_depth
= 0;
2844 ext_inode_hdr(inode
)->eh_max
=
2845 cpu_to_le16(ext4_ext_space_root(inode
, 0));
2846 err
= ext4_ext_dirty(handle
, inode
, path
);
2850 ext4_ext_drop_refs(path
);
2852 if (err
== -EAGAIN
) {
2856 ext4_journal_stop(handle
);
2862 * called at mount time
2864 void ext4_ext_init(struct super_block
*sb
)
2867 * possible initialization would be here
2870 if (EXT4_HAS_INCOMPAT_FEATURE(sb
, EXT4_FEATURE_INCOMPAT_EXTENTS
)) {
2871 #if defined(AGGRESSIVE_TEST) || defined(CHECK_BINSEARCH) || defined(EXTENTS_STATS)
2872 printk(KERN_INFO
"EXT4-fs: file extents enabled"
2873 #ifdef AGGRESSIVE_TEST
2874 ", aggressive tests"
2876 #ifdef CHECK_BINSEARCH
2879 #ifdef EXTENTS_STATS
2884 #ifdef EXTENTS_STATS
2885 spin_lock_init(&EXT4_SB(sb
)->s_ext_stats_lock
);
2886 EXT4_SB(sb
)->s_ext_min
= 1 << 30;
2887 EXT4_SB(sb
)->s_ext_max
= 0;
2893 * called at umount time
2895 void ext4_ext_release(struct super_block
*sb
)
2897 if (!EXT4_HAS_INCOMPAT_FEATURE(sb
, EXT4_FEATURE_INCOMPAT_EXTENTS
))
2900 #ifdef EXTENTS_STATS
2901 if (EXT4_SB(sb
)->s_ext_blocks
&& EXT4_SB(sb
)->s_ext_extents
) {
2902 struct ext4_sb_info
*sbi
= EXT4_SB(sb
);
2903 printk(KERN_ERR
"EXT4-fs: %lu blocks in %lu extents (%lu ave)\n",
2904 sbi
->s_ext_blocks
, sbi
->s_ext_extents
,
2905 sbi
->s_ext_blocks
/ sbi
->s_ext_extents
);
2906 printk(KERN_ERR
"EXT4-fs: extents: %lu min, %lu max, max depth %lu\n",
2907 sbi
->s_ext_min
, sbi
->s_ext_max
, sbi
->s_depth_max
);
2912 /* FIXME!! we need to try to merge to left or right after zero-out */
2913 static int ext4_ext_zeroout(struct inode
*inode
, struct ext4_extent
*ex
)
2915 ext4_fsblk_t ee_pblock
;
2916 unsigned int ee_len
;
2919 ee_len
= ext4_ext_get_actual_len(ex
);
2920 ee_pblock
= ext4_ext_pblock(ex
);
2922 ret
= sb_issue_zeroout(inode
->i_sb
, ee_pblock
, ee_len
, GFP_NOFS
);
2930 * ext4_split_extent_at() splits an extent at given block.
2932 * @handle: the journal handle
2933 * @inode: the file inode
2934 * @path: the path to the extent
2935 * @split: the logical block where the extent is splitted.
2936 * @split_flags: indicates if the extent could be zeroout if split fails, and
2937 * the states(init or uninit) of new extents.
2938 * @flags: flags used to insert new extent to extent tree.
2941 * Splits extent [a, b] into two extents [a, @split) and [@split, b], states
2942 * of which are deterimined by split_flag.
2944 * There are two cases:
2945 * a> the extent are splitted into two extent.
2946 * b> split is not needed, and just mark the extent.
2948 * return 0 on success.
2950 static int ext4_split_extent_at(handle_t
*handle
,
2951 struct inode
*inode
,
2952 struct ext4_ext_path
*path
,
2957 ext4_fsblk_t newblock
;
2958 ext4_lblk_t ee_block
;
2959 struct ext4_extent
*ex
, newex
, orig_ex
;
2960 struct ext4_extent
*ex2
= NULL
;
2961 unsigned int ee_len
, depth
;
2964 BUG_ON((split_flag
& (EXT4_EXT_DATA_VALID1
| EXT4_EXT_DATA_VALID2
)) ==
2965 (EXT4_EXT_DATA_VALID1
| EXT4_EXT_DATA_VALID2
));
2967 ext_debug("ext4_split_extents_at: inode %lu, logical"
2968 "block %llu\n", inode
->i_ino
, (unsigned long long)split
);
2970 ext4_ext_show_leaf(inode
, path
);
2972 depth
= ext_depth(inode
);
2973 ex
= path
[depth
].p_ext
;
2974 ee_block
= le32_to_cpu(ex
->ee_block
);
2975 ee_len
= ext4_ext_get_actual_len(ex
);
2976 newblock
= split
- ee_block
+ ext4_ext_pblock(ex
);
2978 BUG_ON(split
< ee_block
|| split
>= (ee_block
+ ee_len
));
2980 err
= ext4_ext_get_access(handle
, inode
, path
+ depth
);
2984 if (split
== ee_block
) {
2986 * case b: block @split is the block that the extent begins with
2987 * then we just change the state of the extent, and splitting
2990 if (split_flag
& EXT4_EXT_MARK_UNINIT2
)
2991 ext4_ext_mark_uninitialized(ex
);
2993 ext4_ext_mark_initialized(ex
);
2995 if (!(flags
& EXT4_GET_BLOCKS_PRE_IO
))
2996 ext4_ext_try_to_merge(handle
, inode
, path
, ex
);
2998 err
= ext4_ext_dirty(handle
, inode
, path
+ path
->p_depth
);
3003 memcpy(&orig_ex
, ex
, sizeof(orig_ex
));
3004 ex
->ee_len
= cpu_to_le16(split
- ee_block
);
3005 if (split_flag
& EXT4_EXT_MARK_UNINIT1
)
3006 ext4_ext_mark_uninitialized(ex
);
3009 * path may lead to new leaf, not to original leaf any more
3010 * after ext4_ext_insert_extent() returns,
3012 err
= ext4_ext_dirty(handle
, inode
, path
+ depth
);
3014 goto fix_extent_len
;
3017 ex2
->ee_block
= cpu_to_le32(split
);
3018 ex2
->ee_len
= cpu_to_le16(ee_len
- (split
- ee_block
));
3019 ext4_ext_store_pblock(ex2
, newblock
);
3020 if (split_flag
& EXT4_EXT_MARK_UNINIT2
)
3021 ext4_ext_mark_uninitialized(ex2
);
3023 err
= ext4_ext_insert_extent(handle
, inode
, path
, &newex
, flags
);
3024 if (err
== -ENOSPC
&& (EXT4_EXT_MAY_ZEROOUT
& split_flag
)) {
3025 if (split_flag
& (EXT4_EXT_DATA_VALID1
|EXT4_EXT_DATA_VALID2
)) {
3026 if (split_flag
& EXT4_EXT_DATA_VALID1
)
3027 err
= ext4_ext_zeroout(inode
, ex2
);
3029 err
= ext4_ext_zeroout(inode
, ex
);
3031 err
= ext4_ext_zeroout(inode
, &orig_ex
);
3034 goto fix_extent_len
;
3035 /* update the extent length and mark as initialized */
3036 ex
->ee_len
= cpu_to_le16(ee_len
);
3037 ext4_ext_try_to_merge(handle
, inode
, path
, ex
);
3038 err
= ext4_ext_dirty(handle
, inode
, path
+ path
->p_depth
);
3041 goto fix_extent_len
;
3044 ext4_ext_show_leaf(inode
, path
);
3048 ex
->ee_len
= orig_ex
.ee_len
;
3049 ext4_ext_dirty(handle
, inode
, path
+ depth
);
3054 * ext4_split_extents() splits an extent and mark extent which is covered
3055 * by @map as split_flags indicates
3057 * It may result in splitting the extent into multiple extents (upto three)
3058 * There are three possibilities:
3059 * a> There is no split required
3060 * b> Splits in two extents: Split is happening at either end of the extent
3061 * c> Splits in three extents: Somone is splitting in middle of the extent
3064 static int ext4_split_extent(handle_t
*handle
,
3065 struct inode
*inode
,
3066 struct ext4_ext_path
*path
,
3067 struct ext4_map_blocks
*map
,
3071 ext4_lblk_t ee_block
;
3072 struct ext4_extent
*ex
;
3073 unsigned int ee_len
, depth
;
3076 int split_flag1
, flags1
;
3078 depth
= ext_depth(inode
);
3079 ex
= path
[depth
].p_ext
;
3080 ee_block
= le32_to_cpu(ex
->ee_block
);
3081 ee_len
= ext4_ext_get_actual_len(ex
);
3082 uninitialized
= ext4_ext_is_uninitialized(ex
);
3084 if (map
->m_lblk
+ map
->m_len
< ee_block
+ ee_len
) {
3085 split_flag1
= split_flag
& EXT4_EXT_MAY_ZEROOUT
;
3086 flags1
= flags
| EXT4_GET_BLOCKS_PRE_IO
;
3088 split_flag1
|= EXT4_EXT_MARK_UNINIT1
|
3089 EXT4_EXT_MARK_UNINIT2
;
3090 if (split_flag
& EXT4_EXT_DATA_VALID2
)
3091 split_flag1
|= EXT4_EXT_DATA_VALID1
;
3092 err
= ext4_split_extent_at(handle
, inode
, path
,
3093 map
->m_lblk
+ map
->m_len
, split_flag1
, flags1
);
3098 ext4_ext_drop_refs(path
);
3099 path
= ext4_ext_find_extent(inode
, map
->m_lblk
, path
);
3101 return PTR_ERR(path
);
3103 if (map
->m_lblk
>= ee_block
) {
3104 split_flag1
= split_flag
& (EXT4_EXT_MAY_ZEROOUT
|
3105 EXT4_EXT_DATA_VALID2
);
3107 split_flag1
|= EXT4_EXT_MARK_UNINIT1
;
3108 if (split_flag
& EXT4_EXT_MARK_UNINIT2
)
3109 split_flag1
|= EXT4_EXT_MARK_UNINIT2
;
3110 err
= ext4_split_extent_at(handle
, inode
, path
,
3111 map
->m_lblk
, split_flag1
, flags
);
3116 ext4_ext_show_leaf(inode
, path
);
3118 return err
? err
: map
->m_len
;
3122 * This function is called by ext4_ext_map_blocks() if someone tries to write
3123 * to an uninitialized extent. It may result in splitting the uninitialized
3124 * extent into multiple extents (up to three - one initialized and two
3126 * There are three possibilities:
3127 * a> There is no split required: Entire extent should be initialized
3128 * b> Splits in two extents: Write is happening at either end of the extent
3129 * c> Splits in three extents: Somone is writing in middle of the extent
3132 * - The extent pointed to by 'path' is uninitialized.
3133 * - The extent pointed to by 'path' contains a superset
3134 * of the logical span [map->m_lblk, map->m_lblk + map->m_len).
3136 * Post-conditions on success:
3137 * - the returned value is the number of blocks beyond map->l_lblk
3138 * that are allocated and initialized.
3139 * It is guaranteed to be >= map->m_len.
3141 static int ext4_ext_convert_to_initialized(handle_t
*handle
,
3142 struct inode
*inode
,
3143 struct ext4_map_blocks
*map
,
3144 struct ext4_ext_path
*path
)
3146 struct ext4_sb_info
*sbi
;
3147 struct ext4_extent_header
*eh
;
3148 struct ext4_map_blocks split_map
;
3149 struct ext4_extent zero_ex
;
3150 struct ext4_extent
*ex
;
3151 ext4_lblk_t ee_block
, eof_block
;
3152 unsigned int ee_len
, depth
;
3153 int allocated
, max_zeroout
= 0;
3157 ext_debug("ext4_ext_convert_to_initialized: inode %lu, logical"
3158 "block %llu, max_blocks %u\n", inode
->i_ino
,
3159 (unsigned long long)map
->m_lblk
, map
->m_len
);
3161 sbi
= EXT4_SB(inode
->i_sb
);
3162 eof_block
= (inode
->i_size
+ inode
->i_sb
->s_blocksize
- 1) >>
3163 inode
->i_sb
->s_blocksize_bits
;
3164 if (eof_block
< map
->m_lblk
+ map
->m_len
)
3165 eof_block
= map
->m_lblk
+ map
->m_len
;
3167 depth
= ext_depth(inode
);
3168 eh
= path
[depth
].p_hdr
;
3169 ex
= path
[depth
].p_ext
;
3170 ee_block
= le32_to_cpu(ex
->ee_block
);
3171 ee_len
= ext4_ext_get_actual_len(ex
);
3172 allocated
= ee_len
- (map
->m_lblk
- ee_block
);
3174 trace_ext4_ext_convert_to_initialized_enter(inode
, map
, ex
);
3176 /* Pre-conditions */
3177 BUG_ON(!ext4_ext_is_uninitialized(ex
));
3178 BUG_ON(!in_range(map
->m_lblk
, ee_block
, ee_len
));
3181 * Attempt to transfer newly initialized blocks from the currently
3182 * uninitialized extent to its left neighbor. This is much cheaper
3183 * than an insertion followed by a merge as those involve costly
3184 * memmove() calls. This is the common case in steady state for
3185 * workloads doing fallocate(FALLOC_FL_KEEP_SIZE) followed by append
3188 * Limitations of the current logic:
3189 * - L1: we only deal with writes at the start of the extent.
3190 * The approach could be extended to writes at the end
3191 * of the extent but this scenario was deemed less common.
3192 * - L2: we do not deal with writes covering the whole extent.
3193 * This would require removing the extent if the transfer
3195 * - L3: we only attempt to merge with an extent stored in the
3196 * same extent tree node.
3198 if ((map
->m_lblk
== ee_block
) && /*L1*/
3199 (map
->m_len
< ee_len
) && /*L2*/
3200 (ex
> EXT_FIRST_EXTENT(eh
))) { /*L3*/
3201 struct ext4_extent
*prev_ex
;
3202 ext4_lblk_t prev_lblk
;
3203 ext4_fsblk_t prev_pblk
, ee_pblk
;
3204 unsigned int prev_len
, write_len
;
3207 prev_lblk
= le32_to_cpu(prev_ex
->ee_block
);
3208 prev_len
= ext4_ext_get_actual_len(prev_ex
);
3209 prev_pblk
= ext4_ext_pblock(prev_ex
);
3210 ee_pblk
= ext4_ext_pblock(ex
);
3211 write_len
= map
->m_len
;
3214 * A transfer of blocks from 'ex' to 'prev_ex' is allowed
3215 * upon those conditions:
3216 * - C1: prev_ex is initialized,
3217 * - C2: prev_ex is logically abutting ex,
3218 * - C3: prev_ex is physically abutting ex,
3219 * - C4: prev_ex can receive the additional blocks without
3220 * overflowing the (initialized) length limit.
3222 if ((!ext4_ext_is_uninitialized(prev_ex
)) && /*C1*/
3223 ((prev_lblk
+ prev_len
) == ee_block
) && /*C2*/
3224 ((prev_pblk
+ prev_len
) == ee_pblk
) && /*C3*/
3225 (prev_len
< (EXT_INIT_MAX_LEN
- write_len
))) { /*C4*/
3226 err
= ext4_ext_get_access(handle
, inode
, path
+ depth
);
3230 trace_ext4_ext_convert_to_initialized_fastpath(inode
,
3233 /* Shift the start of ex by 'write_len' blocks */
3234 ex
->ee_block
= cpu_to_le32(ee_block
+ write_len
);
3235 ext4_ext_store_pblock(ex
, ee_pblk
+ write_len
);
3236 ex
->ee_len
= cpu_to_le16(ee_len
- write_len
);
3237 ext4_ext_mark_uninitialized(ex
); /* Restore the flag */
3239 /* Extend prev_ex by 'write_len' blocks */
3240 prev_ex
->ee_len
= cpu_to_le16(prev_len
+ write_len
);
3242 /* Mark the block containing both extents as dirty */
3243 ext4_ext_dirty(handle
, inode
, path
+ depth
);
3245 /* Update path to point to the right extent */
3246 path
[depth
].p_ext
= prev_ex
;
3248 /* Result: number of initialized blocks past m_lblk */
3249 allocated
= write_len
;
3254 WARN_ON(map
->m_lblk
< ee_block
);
3256 * It is safe to convert extent to initialized via explicit
3257 * zeroout only if extent is fully insde i_size or new_size.
3259 split_flag
|= ee_block
+ ee_len
<= eof_block
? EXT4_EXT_MAY_ZEROOUT
: 0;
3261 if (EXT4_EXT_MAY_ZEROOUT
& split_flag
)
3262 max_zeroout
= sbi
->s_extent_max_zeroout_kb
>>
3263 inode
->i_sb
->s_blocksize_bits
;
3265 /* If extent is less than s_max_zeroout_kb, zeroout directly */
3266 if (max_zeroout
&& (ee_len
<= max_zeroout
)) {
3267 err
= ext4_ext_zeroout(inode
, ex
);
3271 err
= ext4_ext_get_access(handle
, inode
, path
+ depth
);
3274 ext4_ext_mark_initialized(ex
);
3275 ext4_ext_try_to_merge(handle
, inode
, path
, ex
);
3276 err
= ext4_ext_dirty(handle
, inode
, path
+ path
->p_depth
);
3282 * 1. split the extent into three extents.
3283 * 2. split the extent into two extents, zeroout the first half.
3284 * 3. split the extent into two extents, zeroout the second half.
3285 * 4. split the extent into two extents with out zeroout.
3287 split_map
.m_lblk
= map
->m_lblk
;
3288 split_map
.m_len
= map
->m_len
;
3290 if (max_zeroout
&& (allocated
> map
->m_len
)) {
3291 if (allocated
<= max_zeroout
) {
3294 cpu_to_le32(map
->m_lblk
);
3295 zero_ex
.ee_len
= cpu_to_le16(allocated
);
3296 ext4_ext_store_pblock(&zero_ex
,
3297 ext4_ext_pblock(ex
) + map
->m_lblk
- ee_block
);
3298 err
= ext4_ext_zeroout(inode
, &zero_ex
);
3301 split_map
.m_lblk
= map
->m_lblk
;
3302 split_map
.m_len
= allocated
;
3303 } else if (map
->m_lblk
- ee_block
+ map
->m_len
< max_zeroout
) {
3305 if (map
->m_lblk
!= ee_block
) {
3306 zero_ex
.ee_block
= ex
->ee_block
;
3307 zero_ex
.ee_len
= cpu_to_le16(map
->m_lblk
-
3309 ext4_ext_store_pblock(&zero_ex
,
3310 ext4_ext_pblock(ex
));
3311 err
= ext4_ext_zeroout(inode
, &zero_ex
);
3316 split_map
.m_lblk
= ee_block
;
3317 split_map
.m_len
= map
->m_lblk
- ee_block
+ map
->m_len
;
3318 allocated
= map
->m_len
;
3322 allocated
= ext4_split_extent(handle
, inode
, path
,
3323 &split_map
, split_flag
, 0);
3328 return err
? err
: allocated
;
3332 * This function is called by ext4_ext_map_blocks() from
3333 * ext4_get_blocks_dio_write() when DIO to write
3334 * to an uninitialized extent.
3336 * Writing to an uninitialized extent may result in splitting the uninitialized
3337 * extent into multiple initialized/uninitialized extents (up to three)
3338 * There are three possibilities:
3339 * a> There is no split required: Entire extent should be uninitialized
3340 * b> Splits in two extents: Write is happening at either end of the extent
3341 * c> Splits in three extents: Somone is writing in middle of the extent
3343 * One of more index blocks maybe needed if the extent tree grow after
3344 * the uninitialized extent split. To prevent ENOSPC occur at the IO
3345 * complete, we need to split the uninitialized extent before DIO submit
3346 * the IO. The uninitialized extent called at this time will be split
3347 * into three uninitialized extent(at most). After IO complete, the part
3348 * being filled will be convert to initialized by the end_io callback function
3349 * via ext4_convert_unwritten_extents().
3351 * Returns the size of uninitialized extent to be written on success.
3353 static int ext4_split_unwritten_extents(handle_t
*handle
,
3354 struct inode
*inode
,
3355 struct ext4_map_blocks
*map
,
3356 struct ext4_ext_path
*path
,
3359 ext4_lblk_t eof_block
;
3360 ext4_lblk_t ee_block
;
3361 struct ext4_extent
*ex
;
3362 unsigned int ee_len
;
3363 int split_flag
= 0, depth
;
3365 ext_debug("ext4_split_unwritten_extents: inode %lu, logical"
3366 "block %llu, max_blocks %u\n", inode
->i_ino
,
3367 (unsigned long long)map
->m_lblk
, map
->m_len
);
3369 eof_block
= (inode
->i_size
+ inode
->i_sb
->s_blocksize
- 1) >>
3370 inode
->i_sb
->s_blocksize_bits
;
3371 if (eof_block
< map
->m_lblk
+ map
->m_len
)
3372 eof_block
= map
->m_lblk
+ map
->m_len
;
3374 * It is safe to convert extent to initialized via explicit
3375 * zeroout only if extent is fully insde i_size or new_size.
3377 depth
= ext_depth(inode
);
3378 ex
= path
[depth
].p_ext
;
3379 ee_block
= le32_to_cpu(ex
->ee_block
);
3380 ee_len
= ext4_ext_get_actual_len(ex
);
3382 split_flag
|= ee_block
+ ee_len
<= eof_block
? EXT4_EXT_MAY_ZEROOUT
: 0;
3383 split_flag
|= EXT4_EXT_MARK_UNINIT2
;
3384 if (flags
& EXT4_GET_BLOCKS_CONVERT
)
3385 split_flag
|= EXT4_EXT_DATA_VALID2
;
3386 flags
|= EXT4_GET_BLOCKS_PRE_IO
;
3387 return ext4_split_extent(handle
, inode
, path
, map
, split_flag
, flags
);
3390 static int ext4_convert_unwritten_extents_endio(handle_t
*handle
,
3391 struct inode
*inode
,
3392 struct ext4_map_blocks
*map
,
3393 struct ext4_ext_path
*path
)
3395 struct ext4_extent
*ex
;
3396 ext4_lblk_t ee_block
;
3397 unsigned int ee_len
;
3401 depth
= ext_depth(inode
);
3402 ex
= path
[depth
].p_ext
;
3403 ee_block
= le32_to_cpu(ex
->ee_block
);
3404 ee_len
= ext4_ext_get_actual_len(ex
);
3406 ext_debug("ext4_convert_unwritten_extents_endio: inode %lu, logical"
3407 "block %llu, max_blocks %u\n", inode
->i_ino
,
3408 (unsigned long long)ee_block
, ee_len
);
3410 /* If extent is larger than requested then split is required */
3411 if (ee_block
!= map
->m_lblk
|| ee_len
> map
->m_len
) {
3412 err
= ext4_split_unwritten_extents(handle
, inode
, map
, path
,
3413 EXT4_GET_BLOCKS_CONVERT
);
3416 ext4_ext_drop_refs(path
);
3417 path
= ext4_ext_find_extent(inode
, map
->m_lblk
, path
);
3419 err
= PTR_ERR(path
);
3422 depth
= ext_depth(inode
);
3423 ex
= path
[depth
].p_ext
;
3426 err
= ext4_ext_get_access(handle
, inode
, path
+ depth
);
3429 /* first mark the extent as initialized */
3430 ext4_ext_mark_initialized(ex
);
3432 /* note: ext4_ext_correct_indexes() isn't needed here because
3433 * borders are not changed
3435 ext4_ext_try_to_merge(handle
, inode
, path
, ex
);
3437 /* Mark modified extent as dirty */
3438 err
= ext4_ext_dirty(handle
, inode
, path
+ path
->p_depth
);
3440 ext4_ext_show_leaf(inode
, path
);
3444 static void unmap_underlying_metadata_blocks(struct block_device
*bdev
,
3445 sector_t block
, int count
)
3448 for (i
= 0; i
< count
; i
++)
3449 unmap_underlying_metadata(bdev
, block
+ i
);
3453 * Handle EOFBLOCKS_FL flag, clearing it if necessary
3455 static int check_eofblocks_fl(handle_t
*handle
, struct inode
*inode
,
3457 struct ext4_ext_path
*path
,
3461 struct ext4_extent_header
*eh
;
3462 struct ext4_extent
*last_ex
;
3464 if (!ext4_test_inode_flag(inode
, EXT4_INODE_EOFBLOCKS
))
3467 depth
= ext_depth(inode
);
3468 eh
= path
[depth
].p_hdr
;
3471 * We're going to remove EOFBLOCKS_FL entirely in future so we
3472 * do not care for this case anymore. Simply remove the flag
3473 * if there are no extents.
3475 if (unlikely(!eh
->eh_entries
))
3477 last_ex
= EXT_LAST_EXTENT(eh
);
3479 * We should clear the EOFBLOCKS_FL flag if we are writing the
3480 * last block in the last extent in the file. We test this by
3481 * first checking to see if the caller to
3482 * ext4_ext_get_blocks() was interested in the last block (or
3483 * a block beyond the last block) in the current extent. If
3484 * this turns out to be false, we can bail out from this
3485 * function immediately.
3487 if (lblk
+ len
< le32_to_cpu(last_ex
->ee_block
) +
3488 ext4_ext_get_actual_len(last_ex
))
3491 * If the caller does appear to be planning to write at or
3492 * beyond the end of the current extent, we then test to see
3493 * if the current extent is the last extent in the file, by
3494 * checking to make sure it was reached via the rightmost node
3495 * at each level of the tree.
3497 for (i
= depth
-1; i
>= 0; i
--)
3498 if (path
[i
].p_idx
!= EXT_LAST_INDEX(path
[i
].p_hdr
))
3501 ext4_clear_inode_flag(inode
, EXT4_INODE_EOFBLOCKS
);
3502 return ext4_mark_inode_dirty(handle
, inode
);
3506 * ext4_find_delalloc_range: find delayed allocated block in the given range.
3508 * Return 1 if there is a delalloc block in the range, otherwise 0.
3510 static int ext4_find_delalloc_range(struct inode
*inode
,
3511 ext4_lblk_t lblk_start
,
3512 ext4_lblk_t lblk_end
)
3514 struct extent_status es
;
3516 es
.start
= lblk_start
;
3517 ext4_es_find_extent(inode
, &es
);
3519 return 0; /* there is no delay extent in this tree */
3520 else if (es
.start
<= lblk_start
&& lblk_start
< es
.start
+ es
.len
)
3522 else if (lblk_start
<= es
.start
&& es
.start
<= lblk_end
)
3528 int ext4_find_delalloc_cluster(struct inode
*inode
, ext4_lblk_t lblk
)
3530 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
3531 ext4_lblk_t lblk_start
, lblk_end
;
3532 lblk_start
= lblk
& (~(sbi
->s_cluster_ratio
- 1));
3533 lblk_end
= lblk_start
+ sbi
->s_cluster_ratio
- 1;
3535 return ext4_find_delalloc_range(inode
, lblk_start
, lblk_end
);
3539 * Determines how many complete clusters (out of those specified by the 'map')
3540 * are under delalloc and were reserved quota for.
3541 * This function is called when we are writing out the blocks that were
3542 * originally written with their allocation delayed, but then the space was
3543 * allocated using fallocate() before the delayed allocation could be resolved.
3544 * The cases to look for are:
3545 * ('=' indicated delayed allocated blocks
3546 * '-' indicates non-delayed allocated blocks)
3547 * (a) partial clusters towards beginning and/or end outside of allocated range
3548 * are not delalloc'ed.
3550 * |----c---=|====c====|====c====|===-c----|
3551 * |++++++ allocated ++++++|
3552 * ==> 4 complete clusters in above example
3554 * (b) partial cluster (outside of allocated range) towards either end is
3555 * marked for delayed allocation. In this case, we will exclude that
3558 * |----====c========|========c========|
3559 * |++++++ allocated ++++++|
3560 * ==> 1 complete clusters in above example
3563 * |================c================|
3564 * |++++++ allocated ++++++|
3565 * ==> 0 complete clusters in above example
3567 * The ext4_da_update_reserve_space will be called only if we
3568 * determine here that there were some "entire" clusters that span
3569 * this 'allocated' range.
3570 * In the non-bigalloc case, this function will just end up returning num_blks
3571 * without ever calling ext4_find_delalloc_range.
3574 get_reserved_cluster_alloc(struct inode
*inode
, ext4_lblk_t lblk_start
,
3575 unsigned int num_blks
)
3577 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
3578 ext4_lblk_t alloc_cluster_start
, alloc_cluster_end
;
3579 ext4_lblk_t lblk_from
, lblk_to
, c_offset
;
3580 unsigned int allocated_clusters
= 0;
3582 alloc_cluster_start
= EXT4_B2C(sbi
, lblk_start
);
3583 alloc_cluster_end
= EXT4_B2C(sbi
, lblk_start
+ num_blks
- 1);
3585 /* max possible clusters for this allocation */
3586 allocated_clusters
= alloc_cluster_end
- alloc_cluster_start
+ 1;
3588 trace_ext4_get_reserved_cluster_alloc(inode
, lblk_start
, num_blks
);
3590 /* Check towards left side */
3591 c_offset
= lblk_start
& (sbi
->s_cluster_ratio
- 1);
3593 lblk_from
= lblk_start
& (~(sbi
->s_cluster_ratio
- 1));
3594 lblk_to
= lblk_from
+ c_offset
- 1;
3596 if (ext4_find_delalloc_range(inode
, lblk_from
, lblk_to
))
3597 allocated_clusters
--;
3600 /* Now check towards right. */
3601 c_offset
= (lblk_start
+ num_blks
) & (sbi
->s_cluster_ratio
- 1);
3602 if (allocated_clusters
&& c_offset
) {
3603 lblk_from
= lblk_start
+ num_blks
;
3604 lblk_to
= lblk_from
+ (sbi
->s_cluster_ratio
- c_offset
) - 1;
3606 if (ext4_find_delalloc_range(inode
, lblk_from
, lblk_to
))
3607 allocated_clusters
--;
3610 return allocated_clusters
;
3614 ext4_ext_handle_uninitialized_extents(handle_t
*handle
, struct inode
*inode
,
3615 struct ext4_map_blocks
*map
,
3616 struct ext4_ext_path
*path
, int flags
,
3617 unsigned int allocated
, ext4_fsblk_t newblock
)
3621 ext4_io_end_t
*io
= ext4_inode_aio(inode
);
3623 ext_debug("ext4_ext_handle_uninitialized_extents: inode %lu, logical "
3624 "block %llu, max_blocks %u, flags %x, allocated %u\n",
3625 inode
->i_ino
, (unsigned long long)map
->m_lblk
, map
->m_len
,
3627 ext4_ext_show_leaf(inode
, path
);
3629 trace_ext4_ext_handle_uninitialized_extents(inode
, map
, flags
,
3630 allocated
, newblock
);
3632 /* get_block() before submit the IO, split the extent */
3633 if ((flags
& EXT4_GET_BLOCKS_PRE_IO
)) {
3634 ret
= ext4_split_unwritten_extents(handle
, inode
, map
,
3639 * Flag the inode(non aio case) or end_io struct (aio case)
3640 * that this IO needs to conversion to written when IO is
3644 ext4_set_io_unwritten_flag(inode
, io
);
3646 ext4_set_inode_state(inode
, EXT4_STATE_DIO_UNWRITTEN
);
3647 if (ext4_should_dioread_nolock(inode
))
3648 map
->m_flags
|= EXT4_MAP_UNINIT
;
3651 /* IO end_io complete, convert the filled extent to written */
3652 if ((flags
& EXT4_GET_BLOCKS_CONVERT
)) {
3653 ret
= ext4_convert_unwritten_extents_endio(handle
, inode
, map
,
3656 ext4_update_inode_fsync_trans(handle
, inode
, 1);
3657 err
= check_eofblocks_fl(handle
, inode
, map
->m_lblk
,
3663 /* buffered IO case */
3665 * repeat fallocate creation request
3666 * we already have an unwritten extent
3668 if (flags
& EXT4_GET_BLOCKS_UNINIT_EXT
)
3671 /* buffered READ or buffered write_begin() lookup */
3672 if ((flags
& EXT4_GET_BLOCKS_CREATE
) == 0) {
3674 * We have blocks reserved already. We
3675 * return allocated blocks so that delalloc
3676 * won't do block reservation for us. But
3677 * the buffer head will be unmapped so that
3678 * a read from the block returns 0s.
3680 map
->m_flags
|= EXT4_MAP_UNWRITTEN
;
3684 /* buffered write, writepage time, convert*/
3685 ret
= ext4_ext_convert_to_initialized(handle
, inode
, map
, path
);
3687 ext4_update_inode_fsync_trans(handle
, inode
, 1);
3694 map
->m_flags
|= EXT4_MAP_NEW
;
3696 * if we allocated more blocks than requested
3697 * we need to make sure we unmap the extra block
3698 * allocated. The actual needed block will get
3699 * unmapped later when we find the buffer_head marked
3702 if (allocated
> map
->m_len
) {
3703 unmap_underlying_metadata_blocks(inode
->i_sb
->s_bdev
,
3704 newblock
+ map
->m_len
,
3705 allocated
- map
->m_len
);
3706 allocated
= map
->m_len
;
3710 * If we have done fallocate with the offset that is already
3711 * delayed allocated, we would have block reservation
3712 * and quota reservation done in the delayed write path.
3713 * But fallocate would have already updated quota and block
3714 * count for this offset. So cancel these reservation
3716 if (flags
& EXT4_GET_BLOCKS_DELALLOC_RESERVE
) {
3717 unsigned int reserved_clusters
;
3718 reserved_clusters
= get_reserved_cluster_alloc(inode
,
3719 map
->m_lblk
, map
->m_len
);
3720 if (reserved_clusters
)
3721 ext4_da_update_reserve_space(inode
,
3727 map
->m_flags
|= EXT4_MAP_MAPPED
;
3728 if ((flags
& EXT4_GET_BLOCKS_KEEP_SIZE
) == 0) {
3729 err
= check_eofblocks_fl(handle
, inode
, map
->m_lblk
, path
,
3735 if (allocated
> map
->m_len
)
3736 allocated
= map
->m_len
;
3737 ext4_ext_show_leaf(inode
, path
);
3738 map
->m_pblk
= newblock
;
3739 map
->m_len
= allocated
;
3742 ext4_ext_drop_refs(path
);
3745 return err
? err
: allocated
;
3749 * get_implied_cluster_alloc - check to see if the requested
3750 * allocation (in the map structure) overlaps with a cluster already
3751 * allocated in an extent.
3752 * @sb The filesystem superblock structure
3753 * @map The requested lblk->pblk mapping
3754 * @ex The extent structure which might contain an implied
3755 * cluster allocation
3757 * This function is called by ext4_ext_map_blocks() after we failed to
3758 * find blocks that were already in the inode's extent tree. Hence,
3759 * we know that the beginning of the requested region cannot overlap
3760 * the extent from the inode's extent tree. There are three cases we
3761 * want to catch. The first is this case:
3763 * |--- cluster # N--|
3764 * |--- extent ---| |---- requested region ---|
3767 * The second case that we need to test for is this one:
3769 * |--------- cluster # N ----------------|
3770 * |--- requested region --| |------- extent ----|
3771 * |=======================|
3773 * The third case is when the requested region lies between two extents
3774 * within the same cluster:
3775 * |------------- cluster # N-------------|
3776 * |----- ex -----| |---- ex_right ----|
3777 * |------ requested region ------|
3778 * |================|
3780 * In each of the above cases, we need to set the map->m_pblk and
3781 * map->m_len so it corresponds to the return the extent labelled as
3782 * "|====|" from cluster #N, since it is already in use for data in
3783 * cluster EXT4_B2C(sbi, map->m_lblk). We will then return 1 to
3784 * signal to ext4_ext_map_blocks() that map->m_pblk should be treated
3785 * as a new "allocated" block region. Otherwise, we will return 0 and
3786 * ext4_ext_map_blocks() will then allocate one or more new clusters
3787 * by calling ext4_mb_new_blocks().
3789 static int get_implied_cluster_alloc(struct super_block
*sb
,
3790 struct ext4_map_blocks
*map
,
3791 struct ext4_extent
*ex
,
3792 struct ext4_ext_path
*path
)
3794 struct ext4_sb_info
*sbi
= EXT4_SB(sb
);
3795 ext4_lblk_t c_offset
= map
->m_lblk
& (sbi
->s_cluster_ratio
-1);
3796 ext4_lblk_t ex_cluster_start
, ex_cluster_end
;
3797 ext4_lblk_t rr_cluster_start
;
3798 ext4_lblk_t ee_block
= le32_to_cpu(ex
->ee_block
);
3799 ext4_fsblk_t ee_start
= ext4_ext_pblock(ex
);
3800 unsigned short ee_len
= ext4_ext_get_actual_len(ex
);
3802 /* The extent passed in that we are trying to match */
3803 ex_cluster_start
= EXT4_B2C(sbi
, ee_block
);
3804 ex_cluster_end
= EXT4_B2C(sbi
, ee_block
+ ee_len
- 1);
3806 /* The requested region passed into ext4_map_blocks() */
3807 rr_cluster_start
= EXT4_B2C(sbi
, map
->m_lblk
);
3809 if ((rr_cluster_start
== ex_cluster_end
) ||
3810 (rr_cluster_start
== ex_cluster_start
)) {
3811 if (rr_cluster_start
== ex_cluster_end
)
3812 ee_start
+= ee_len
- 1;
3813 map
->m_pblk
= (ee_start
& ~(sbi
->s_cluster_ratio
- 1)) +
3815 map
->m_len
= min(map
->m_len
,
3816 (unsigned) sbi
->s_cluster_ratio
- c_offset
);
3818 * Check for and handle this case:
3820 * |--------- cluster # N-------------|
3821 * |------- extent ----|
3822 * |--- requested region ---|
3826 if (map
->m_lblk
< ee_block
)
3827 map
->m_len
= min(map
->m_len
, ee_block
- map
->m_lblk
);
3830 * Check for the case where there is already another allocated
3831 * block to the right of 'ex' but before the end of the cluster.
3833 * |------------- cluster # N-------------|
3834 * |----- ex -----| |---- ex_right ----|
3835 * |------ requested region ------|
3836 * |================|
3838 if (map
->m_lblk
> ee_block
) {
3839 ext4_lblk_t next
= ext4_ext_next_allocated_block(path
);
3840 map
->m_len
= min(map
->m_len
, next
- map
->m_lblk
);
3843 trace_ext4_get_implied_cluster_alloc_exit(sb
, map
, 1);
3847 trace_ext4_get_implied_cluster_alloc_exit(sb
, map
, 0);
3853 * Block allocation/map/preallocation routine for extents based files
3856 * Need to be called with
3857 * down_read(&EXT4_I(inode)->i_data_sem) if not allocating file system block
3858 * (ie, create is zero). Otherwise down_write(&EXT4_I(inode)->i_data_sem)
3860 * return > 0, number of of blocks already mapped/allocated
3861 * if create == 0 and these are pre-allocated blocks
3862 * buffer head is unmapped
3863 * otherwise blocks are mapped
3865 * return = 0, if plain look up failed (blocks have not been allocated)
3866 * buffer head is unmapped
3868 * return < 0, error case.
3870 int ext4_ext_map_blocks(handle_t
*handle
, struct inode
*inode
,
3871 struct ext4_map_blocks
*map
, int flags
)
3873 struct ext4_ext_path
*path
= NULL
;
3874 struct ext4_extent newex
, *ex
, *ex2
;
3875 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
3876 ext4_fsblk_t newblock
= 0;
3877 int free_on_err
= 0, err
= 0, depth
;
3878 unsigned int allocated
= 0, offset
= 0;
3879 unsigned int allocated_clusters
= 0;
3880 struct ext4_allocation_request ar
;
3881 ext4_io_end_t
*io
= ext4_inode_aio(inode
);
3882 ext4_lblk_t cluster_offset
;
3883 int set_unwritten
= 0;
3885 ext_debug("blocks %u/%u requested for inode %lu\n",
3886 map
->m_lblk
, map
->m_len
, inode
->i_ino
);
3887 trace_ext4_ext_map_blocks_enter(inode
, map
->m_lblk
, map
->m_len
, flags
);
3889 /* check in cache */
3890 if (ext4_ext_in_cache(inode
, map
->m_lblk
, &newex
)) {
3891 if (!newex
.ee_start_lo
&& !newex
.ee_start_hi
) {
3892 if ((sbi
->s_cluster_ratio
> 1) &&
3893 ext4_find_delalloc_cluster(inode
, map
->m_lblk
))
3894 map
->m_flags
|= EXT4_MAP_FROM_CLUSTER
;
3896 if ((flags
& EXT4_GET_BLOCKS_CREATE
) == 0) {
3898 * block isn't allocated yet and
3899 * user doesn't want to allocate it
3903 /* we should allocate requested block */
3905 /* block is already allocated */
3906 if (sbi
->s_cluster_ratio
> 1)
3907 map
->m_flags
|= EXT4_MAP_FROM_CLUSTER
;
3908 newblock
= map
->m_lblk
3909 - le32_to_cpu(newex
.ee_block
)
3910 + ext4_ext_pblock(&newex
);
3911 /* number of remaining blocks in the extent */
3912 allocated
= ext4_ext_get_actual_len(&newex
) -
3913 (map
->m_lblk
- le32_to_cpu(newex
.ee_block
));
3918 /* find extent for this block */
3919 path
= ext4_ext_find_extent(inode
, map
->m_lblk
, NULL
);
3921 err
= PTR_ERR(path
);
3926 depth
= ext_depth(inode
);
3929 * consistent leaf must not be empty;
3930 * this situation is possible, though, _during_ tree modification;
3931 * this is why assert can't be put in ext4_ext_find_extent()
3933 if (unlikely(path
[depth
].p_ext
== NULL
&& depth
!= 0)) {
3934 EXT4_ERROR_INODE(inode
, "bad extent address "
3935 "lblock: %lu, depth: %d pblock %lld",
3936 (unsigned long) map
->m_lblk
, depth
,
3937 path
[depth
].p_block
);
3942 ex
= path
[depth
].p_ext
;
3944 ext4_lblk_t ee_block
= le32_to_cpu(ex
->ee_block
);
3945 ext4_fsblk_t ee_start
= ext4_ext_pblock(ex
);
3946 unsigned short ee_len
;
3949 * Uninitialized extents are treated as holes, except that
3950 * we split out initialized portions during a write.
3952 ee_len
= ext4_ext_get_actual_len(ex
);
3954 trace_ext4_ext_show_extent(inode
, ee_block
, ee_start
, ee_len
);
3956 /* if found extent covers block, simply return it */
3957 if (in_range(map
->m_lblk
, ee_block
, ee_len
)) {
3958 newblock
= map
->m_lblk
- ee_block
+ ee_start
;
3959 /* number of remaining blocks in the extent */
3960 allocated
= ee_len
- (map
->m_lblk
- ee_block
);
3961 ext_debug("%u fit into %u:%d -> %llu\n", map
->m_lblk
,
3962 ee_block
, ee_len
, newblock
);
3965 * Do not put uninitialized extent
3968 if (!ext4_ext_is_uninitialized(ex
)) {
3969 ext4_ext_put_in_cache(inode
, ee_block
,
3973 allocated
= ext4_ext_handle_uninitialized_extents(
3974 handle
, inode
, map
, path
, flags
,
3975 allocated
, newblock
);
3980 if ((sbi
->s_cluster_ratio
> 1) &&
3981 ext4_find_delalloc_cluster(inode
, map
->m_lblk
))
3982 map
->m_flags
|= EXT4_MAP_FROM_CLUSTER
;
3985 * requested block isn't allocated yet;
3986 * we couldn't try to create block if create flag is zero
3988 if ((flags
& EXT4_GET_BLOCKS_CREATE
) == 0) {
3990 * put just found gap into cache to speed up
3991 * subsequent requests
3993 ext4_ext_put_gap_in_cache(inode
, path
, map
->m_lblk
);
3998 * Okay, we need to do block allocation.
4000 map
->m_flags
&= ~EXT4_MAP_FROM_CLUSTER
;
4001 newex
.ee_block
= cpu_to_le32(map
->m_lblk
);
4002 cluster_offset
= map
->m_lblk
& (sbi
->s_cluster_ratio
-1);
4005 * If we are doing bigalloc, check to see if the extent returned
4006 * by ext4_ext_find_extent() implies a cluster we can use.
4008 if (cluster_offset
&& ex
&&
4009 get_implied_cluster_alloc(inode
->i_sb
, map
, ex
, path
)) {
4010 ar
.len
= allocated
= map
->m_len
;
4011 newblock
= map
->m_pblk
;
4012 map
->m_flags
|= EXT4_MAP_FROM_CLUSTER
;
4013 goto got_allocated_blocks
;
4016 /* find neighbour allocated blocks */
4017 ar
.lleft
= map
->m_lblk
;
4018 err
= ext4_ext_search_left(inode
, path
, &ar
.lleft
, &ar
.pleft
);
4021 ar
.lright
= map
->m_lblk
;
4023 err
= ext4_ext_search_right(inode
, path
, &ar
.lright
, &ar
.pright
, &ex2
);
4027 /* Check if the extent after searching to the right implies a
4028 * cluster we can use. */
4029 if ((sbi
->s_cluster_ratio
> 1) && ex2
&&
4030 get_implied_cluster_alloc(inode
->i_sb
, map
, ex2
, path
)) {
4031 ar
.len
= allocated
= map
->m_len
;
4032 newblock
= map
->m_pblk
;
4033 map
->m_flags
|= EXT4_MAP_FROM_CLUSTER
;
4034 goto got_allocated_blocks
;
4038 * See if request is beyond maximum number of blocks we can have in
4039 * a single extent. For an initialized extent this limit is
4040 * EXT_INIT_MAX_LEN and for an uninitialized extent this limit is
4041 * EXT_UNINIT_MAX_LEN.
4043 if (map
->m_len
> EXT_INIT_MAX_LEN
&&
4044 !(flags
& EXT4_GET_BLOCKS_UNINIT_EXT
))
4045 map
->m_len
= EXT_INIT_MAX_LEN
;
4046 else if (map
->m_len
> EXT_UNINIT_MAX_LEN
&&
4047 (flags
& EXT4_GET_BLOCKS_UNINIT_EXT
))
4048 map
->m_len
= EXT_UNINIT_MAX_LEN
;
4050 /* Check if we can really insert (m_lblk)::(m_lblk + m_len) extent */
4051 newex
.ee_len
= cpu_to_le16(map
->m_len
);
4052 err
= ext4_ext_check_overlap(sbi
, inode
, &newex
, path
);
4054 allocated
= ext4_ext_get_actual_len(&newex
);
4056 allocated
= map
->m_len
;
4058 /* allocate new block */
4060 ar
.goal
= ext4_ext_find_goal(inode
, path
, map
->m_lblk
);
4061 ar
.logical
= map
->m_lblk
;
4063 * We calculate the offset from the beginning of the cluster
4064 * for the logical block number, since when we allocate a
4065 * physical cluster, the physical block should start at the
4066 * same offset from the beginning of the cluster. This is
4067 * needed so that future calls to get_implied_cluster_alloc()
4070 offset
= map
->m_lblk
& (sbi
->s_cluster_ratio
- 1);
4071 ar
.len
= EXT4_NUM_B2C(sbi
, offset
+allocated
);
4073 ar
.logical
-= offset
;
4074 if (S_ISREG(inode
->i_mode
))
4075 ar
.flags
= EXT4_MB_HINT_DATA
;
4077 /* disable in-core preallocation for non-regular files */
4079 if (flags
& EXT4_GET_BLOCKS_NO_NORMALIZE
)
4080 ar
.flags
|= EXT4_MB_HINT_NOPREALLOC
;
4081 newblock
= ext4_mb_new_blocks(handle
, &ar
, &err
);
4084 ext_debug("allocate new block: goal %llu, found %llu/%u\n",
4085 ar
.goal
, newblock
, allocated
);
4087 allocated_clusters
= ar
.len
;
4088 ar
.len
= EXT4_C2B(sbi
, ar
.len
) - offset
;
4089 if (ar
.len
> allocated
)
4092 got_allocated_blocks
:
4093 /* try to insert new extent into found leaf and return */
4094 ext4_ext_store_pblock(&newex
, newblock
+ offset
);
4095 newex
.ee_len
= cpu_to_le16(ar
.len
);
4096 /* Mark uninitialized */
4097 if (flags
& EXT4_GET_BLOCKS_UNINIT_EXT
){
4098 ext4_ext_mark_uninitialized(&newex
);
4100 * io_end structure was created for every IO write to an
4101 * uninitialized extent. To avoid unnecessary conversion,
4102 * here we flag the IO that really needs the conversion.
4103 * For non asycn direct IO case, flag the inode state
4104 * that we need to perform conversion when IO is done.
4106 if ((flags
& EXT4_GET_BLOCKS_PRE_IO
))
4108 if (ext4_should_dioread_nolock(inode
))
4109 map
->m_flags
|= EXT4_MAP_UNINIT
;
4113 if ((flags
& EXT4_GET_BLOCKS_KEEP_SIZE
) == 0)
4114 err
= check_eofblocks_fl(handle
, inode
, map
->m_lblk
,
4117 err
= ext4_ext_insert_extent(handle
, inode
, path
,
4120 if (!err
&& set_unwritten
) {
4122 ext4_set_io_unwritten_flag(inode
, io
);
4124 ext4_set_inode_state(inode
,
4125 EXT4_STATE_DIO_UNWRITTEN
);
4128 if (err
&& free_on_err
) {
4129 int fb_flags
= flags
& EXT4_GET_BLOCKS_DELALLOC_RESERVE
?
4130 EXT4_FREE_BLOCKS_NO_QUOT_UPDATE
: 0;
4131 /* free data blocks we just allocated */
4132 /* not a good idea to call discard here directly,
4133 * but otherwise we'd need to call it every free() */
4134 ext4_discard_preallocations(inode
);
4135 ext4_free_blocks(handle
, inode
, NULL
, ext4_ext_pblock(&newex
),
4136 ext4_ext_get_actual_len(&newex
), fb_flags
);
4140 /* previous routine could use block we allocated */
4141 newblock
= ext4_ext_pblock(&newex
);
4142 allocated
= ext4_ext_get_actual_len(&newex
);
4143 if (allocated
> map
->m_len
)
4144 allocated
= map
->m_len
;
4145 map
->m_flags
|= EXT4_MAP_NEW
;
4148 * Update reserved blocks/metadata blocks after successful
4149 * block allocation which had been deferred till now.
4151 if (flags
& EXT4_GET_BLOCKS_DELALLOC_RESERVE
) {
4152 unsigned int reserved_clusters
;
4154 * Check how many clusters we had reserved this allocated range
4156 reserved_clusters
= get_reserved_cluster_alloc(inode
,
4157 map
->m_lblk
, allocated
);
4158 if (map
->m_flags
& EXT4_MAP_FROM_CLUSTER
) {
4159 if (reserved_clusters
) {
4161 * We have clusters reserved for this range.
4162 * But since we are not doing actual allocation
4163 * and are simply using blocks from previously
4164 * allocated cluster, we should release the
4165 * reservation and not claim quota.
4167 ext4_da_update_reserve_space(inode
,
4168 reserved_clusters
, 0);
4171 BUG_ON(allocated_clusters
< reserved_clusters
);
4172 /* We will claim quota for all newly allocated blocks.*/
4173 ext4_da_update_reserve_space(inode
, allocated_clusters
,
4175 if (reserved_clusters
< allocated_clusters
) {
4176 struct ext4_inode_info
*ei
= EXT4_I(inode
);
4177 int reservation
= allocated_clusters
-
4180 * It seems we claimed few clusters outside of
4181 * the range of this allocation. We should give
4182 * it back to the reservation pool. This can
4183 * happen in the following case:
4185 * * Suppose s_cluster_ratio is 4 (i.e., each
4186 * cluster has 4 blocks. Thus, the clusters
4187 * are [0-3],[4-7],[8-11]...
4188 * * First comes delayed allocation write for
4189 * logical blocks 10 & 11. Since there were no
4190 * previous delayed allocated blocks in the
4191 * range [8-11], we would reserve 1 cluster
4193 * * Next comes write for logical blocks 3 to 8.
4194 * In this case, we will reserve 2 clusters
4195 * (for [0-3] and [4-7]; and not for [8-11] as
4196 * that range has a delayed allocated blocks.
4197 * Thus total reserved clusters now becomes 3.
4198 * * Now, during the delayed allocation writeout
4199 * time, we will first write blocks [3-8] and
4200 * allocate 3 clusters for writing these
4201 * blocks. Also, we would claim all these
4202 * three clusters above.
4203 * * Now when we come here to writeout the
4204 * blocks [10-11], we would expect to claim
4205 * the reservation of 1 cluster we had made
4206 * (and we would claim it since there are no
4207 * more delayed allocated blocks in the range
4208 * [8-11]. But our reserved cluster count had
4209 * already gone to 0.
4211 * Thus, at the step 4 above when we determine
4212 * that there are still some unwritten delayed
4213 * allocated blocks outside of our current
4214 * block range, we should increment the
4215 * reserved clusters count so that when the
4216 * remaining blocks finally gets written, we
4219 dquot_reserve_block(inode
,
4220 EXT4_C2B(sbi
, reservation
));
4221 spin_lock(&ei
->i_block_reservation_lock
);
4222 ei
->i_reserved_data_blocks
+= reservation
;
4223 spin_unlock(&ei
->i_block_reservation_lock
);
4229 * Cache the extent and update transaction to commit on fdatasync only
4230 * when it is _not_ an uninitialized extent.
4232 if ((flags
& EXT4_GET_BLOCKS_UNINIT_EXT
) == 0) {
4233 ext4_ext_put_in_cache(inode
, map
->m_lblk
, allocated
, newblock
);
4234 ext4_update_inode_fsync_trans(handle
, inode
, 1);
4236 ext4_update_inode_fsync_trans(handle
, inode
, 0);
4238 if (allocated
> map
->m_len
)
4239 allocated
= map
->m_len
;
4240 ext4_ext_show_leaf(inode
, path
);
4241 map
->m_flags
|= EXT4_MAP_MAPPED
;
4242 map
->m_pblk
= newblock
;
4243 map
->m_len
= allocated
;
4246 ext4_ext_drop_refs(path
);
4251 trace_ext4_ext_map_blocks_exit(inode
, map
, err
? err
: allocated
);
4253 return err
? err
: allocated
;
4256 void ext4_ext_truncate(struct inode
*inode
)
4258 struct address_space
*mapping
= inode
->i_mapping
;
4259 struct super_block
*sb
= inode
->i_sb
;
4260 ext4_lblk_t last_block
;
4266 * finish any pending end_io work so we won't run the risk of
4267 * converting any truncated blocks to initialized later
4269 ext4_flush_unwritten_io(inode
);
4272 * probably first extent we're gonna free will be last in block
4274 err
= ext4_writepage_trans_blocks(inode
);
4275 handle
= ext4_journal_start(inode
, err
);
4279 if (inode
->i_size
% PAGE_CACHE_SIZE
!= 0) {
4280 page_len
= PAGE_CACHE_SIZE
-
4281 (inode
->i_size
& (PAGE_CACHE_SIZE
- 1));
4283 err
= ext4_discard_partial_page_buffers(handle
,
4284 mapping
, inode
->i_size
, page_len
, 0);
4290 if (ext4_orphan_add(handle
, inode
))
4293 down_write(&EXT4_I(inode
)->i_data_sem
);
4294 ext4_ext_invalidate_cache(inode
);
4296 ext4_discard_preallocations(inode
);
4299 * TODO: optimization is possible here.
4300 * Probably we need not scan at all,
4301 * because page truncation is enough.
4304 /* we have to know where to truncate from in crash case */
4305 EXT4_I(inode
)->i_disksize
= inode
->i_size
;
4306 ext4_mark_inode_dirty(handle
, inode
);
4308 last_block
= (inode
->i_size
+ sb
->s_blocksize
- 1)
4309 >> EXT4_BLOCK_SIZE_BITS(sb
);
4310 err
= ext4_es_remove_extent(inode
, last_block
,
4311 EXT_MAX_BLOCKS
- last_block
);
4312 err
= ext4_ext_remove_space(inode
, last_block
, EXT_MAX_BLOCKS
- 1);
4314 /* In a multi-transaction truncate, we only make the final
4315 * transaction synchronous.
4318 ext4_handle_sync(handle
);
4320 up_write(&EXT4_I(inode
)->i_data_sem
);
4324 * If this was a simple ftruncate() and the file will remain alive,
4325 * then we need to clear up the orphan record which we created above.
4326 * However, if this was a real unlink then we were called by
4327 * ext4_delete_inode(), and we allow that function to clean up the
4328 * orphan info for us.
4331 ext4_orphan_del(handle
, inode
);
4333 inode
->i_mtime
= inode
->i_ctime
= ext4_current_time(inode
);
4334 ext4_mark_inode_dirty(handle
, inode
);
4335 ext4_journal_stop(handle
);
4338 static void ext4_falloc_update_inode(struct inode
*inode
,
4339 int mode
, loff_t new_size
, int update_ctime
)
4341 struct timespec now
;
4344 now
= current_fs_time(inode
->i_sb
);
4345 if (!timespec_equal(&inode
->i_ctime
, &now
))
4346 inode
->i_ctime
= now
;
4349 * Update only when preallocation was requested beyond
4352 if (!(mode
& FALLOC_FL_KEEP_SIZE
)) {
4353 if (new_size
> i_size_read(inode
))
4354 i_size_write(inode
, new_size
);
4355 if (new_size
> EXT4_I(inode
)->i_disksize
)
4356 ext4_update_i_disksize(inode
, new_size
);
4359 * Mark that we allocate beyond EOF so the subsequent truncate
4360 * can proceed even if the new size is the same as i_size.
4362 if (new_size
> i_size_read(inode
))
4363 ext4_set_inode_flag(inode
, EXT4_INODE_EOFBLOCKS
);
4369 * preallocate space for a file. This implements ext4's fallocate file
4370 * operation, which gets called from sys_fallocate system call.
4371 * For block-mapped files, posix_fallocate should fall back to the method
4372 * of writing zeroes to the required new blocks (the same behavior which is
4373 * expected for file systems which do not support fallocate() system call).
4375 long ext4_fallocate(struct file
*file
, int mode
, loff_t offset
, loff_t len
)
4377 struct inode
*inode
= file
->f_path
.dentry
->d_inode
;
4380 unsigned int max_blocks
;
4385 struct ext4_map_blocks map
;
4386 unsigned int credits
, blkbits
= inode
->i_blkbits
;
4389 * currently supporting (pre)allocate mode for extent-based
4392 if (!(ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
)))
4395 /* Return error if mode is not supported */
4396 if (mode
& ~(FALLOC_FL_KEEP_SIZE
| FALLOC_FL_PUNCH_HOLE
))
4399 if (mode
& FALLOC_FL_PUNCH_HOLE
)
4400 return ext4_punch_hole(file
, offset
, len
);
4402 trace_ext4_fallocate_enter(inode
, offset
, len
, mode
);
4403 map
.m_lblk
= offset
>> blkbits
;
4405 * We can't just convert len to max_blocks because
4406 * If blocksize = 4096 offset = 3072 and len = 2048
4408 max_blocks
= (EXT4_BLOCK_ALIGN(len
+ offset
, blkbits
) >> blkbits
)
4411 * credits to insert 1 extent into extent tree
4413 credits
= ext4_chunk_trans_blocks(inode
, max_blocks
);
4414 mutex_lock(&inode
->i_mutex
);
4415 ret
= inode_newsize_ok(inode
, (len
+ offset
));
4417 mutex_unlock(&inode
->i_mutex
);
4418 trace_ext4_fallocate_exit(inode
, offset
, max_blocks
, ret
);
4421 flags
= EXT4_GET_BLOCKS_CREATE_UNINIT_EXT
;
4422 if (mode
& FALLOC_FL_KEEP_SIZE
)
4423 flags
|= EXT4_GET_BLOCKS_KEEP_SIZE
;
4425 * Don't normalize the request if it can fit in one extent so
4426 * that it doesn't get unnecessarily split into multiple
4429 if (len
<= EXT_UNINIT_MAX_LEN
<< blkbits
)
4430 flags
|= EXT4_GET_BLOCKS_NO_NORMALIZE
;
4432 /* Prevent race condition between unwritten */
4433 ext4_flush_unwritten_io(inode
);
4435 while (ret
>= 0 && ret
< max_blocks
) {
4436 map
.m_lblk
= map
.m_lblk
+ ret
;
4437 map
.m_len
= max_blocks
= max_blocks
- ret
;
4438 handle
= ext4_journal_start(inode
, credits
);
4439 if (IS_ERR(handle
)) {
4440 ret
= PTR_ERR(handle
);
4443 ret
= ext4_map_blocks(handle
, inode
, &map
, flags
);
4447 printk(KERN_ERR
"%s: ext4_ext_map_blocks "
4448 "returned error inode#%lu, block=%u, "
4449 "max_blocks=%u", __func__
,
4450 inode
->i_ino
, map
.m_lblk
, max_blocks
);
4452 ext4_mark_inode_dirty(handle
, inode
);
4453 ret2
= ext4_journal_stop(handle
);
4456 if ((map
.m_lblk
+ ret
) >= (EXT4_BLOCK_ALIGN(offset
+ len
,
4457 blkbits
) >> blkbits
))
4458 new_size
= offset
+ len
;
4460 new_size
= ((loff_t
) map
.m_lblk
+ ret
) << blkbits
;
4462 ext4_falloc_update_inode(inode
, mode
, new_size
,
4463 (map
.m_flags
& EXT4_MAP_NEW
));
4464 ext4_mark_inode_dirty(handle
, inode
);
4465 if ((file
->f_flags
& O_SYNC
) && ret
>= max_blocks
)
4466 ext4_handle_sync(handle
);
4467 ret2
= ext4_journal_stop(handle
);
4471 if (ret
== -ENOSPC
&&
4472 ext4_should_retry_alloc(inode
->i_sb
, &retries
)) {
4476 mutex_unlock(&inode
->i_mutex
);
4477 trace_ext4_fallocate_exit(inode
, offset
, max_blocks
,
4478 ret
> 0 ? ret2
: ret
);
4479 return ret
> 0 ? ret2
: ret
;
4483 * This function convert a range of blocks to written extents
4484 * The caller of this function will pass the start offset and the size.
4485 * all unwritten extents within this range will be converted to
4488 * This function is called from the direct IO end io call back
4489 * function, to convert the fallocated extents after IO is completed.
4490 * Returns 0 on success.
4492 int ext4_convert_unwritten_extents(struct inode
*inode
, loff_t offset
,
4496 unsigned int max_blocks
;
4499 struct ext4_map_blocks map
;
4500 unsigned int credits
, blkbits
= inode
->i_blkbits
;
4502 map
.m_lblk
= offset
>> blkbits
;
4504 * We can't just convert len to max_blocks because
4505 * If blocksize = 4096 offset = 3072 and len = 2048
4507 max_blocks
= ((EXT4_BLOCK_ALIGN(len
+ offset
, blkbits
) >> blkbits
) -
4510 * credits to insert 1 extent into extent tree
4512 credits
= ext4_chunk_trans_blocks(inode
, max_blocks
);
4513 while (ret
>= 0 && ret
< max_blocks
) {
4515 map
.m_len
= (max_blocks
-= ret
);
4516 handle
= ext4_journal_start(inode
, credits
);
4517 if (IS_ERR(handle
)) {
4518 ret
= PTR_ERR(handle
);
4521 ret
= ext4_map_blocks(handle
, inode
, &map
,
4522 EXT4_GET_BLOCKS_IO_CONVERT_EXT
);
4525 ext4_msg(inode
->i_sb
, KERN_ERR
,
4526 "%s:%d: inode #%lu: block %u: len %u: "
4527 "ext4_ext_map_blocks returned %d",
4528 __func__
, __LINE__
, inode
->i_ino
, map
.m_lblk
,
4531 ext4_mark_inode_dirty(handle
, inode
);
4532 ret2
= ext4_journal_stop(handle
);
4533 if (ret
<= 0 || ret2
)
4536 return ret
> 0 ? ret2
: ret
;
4540 * If newex is not existing extent (newex->ec_start equals zero) find
4541 * delayed extent at start of newex and update newex accordingly and
4542 * return start of the next delayed extent.
4544 * If newex is existing extent (newex->ec_start is not equal zero)
4545 * return start of next delayed extent or EXT_MAX_BLOCKS if no delayed
4546 * extent found. Leave newex unmodified.
4548 static int ext4_find_delayed_extent(struct inode
*inode
,
4549 struct ext4_ext_cache
*newex
)
4551 struct extent_status es
;
4552 ext4_lblk_t next_del
;
4554 es
.start
= newex
->ec_block
;
4555 next_del
= ext4_es_find_extent(inode
, &es
);
4557 if (newex
->ec_start
== 0) {
4559 * No extent in extent-tree contains block @newex->ec_start,
4560 * then the block may stay in 1)a hole or 2)delayed-extent.
4566 if (es
.start
> newex
->ec_block
) {
4568 newex
->ec_len
= min(es
.start
- newex
->ec_block
,
4573 newex
->ec_len
= es
.start
+ es
.len
- newex
->ec_block
;
4578 /* fiemap flags we can handle specified here */
4579 #define EXT4_FIEMAP_FLAGS (FIEMAP_FLAG_SYNC|FIEMAP_FLAG_XATTR)
4581 static int ext4_xattr_fiemap(struct inode
*inode
,
4582 struct fiemap_extent_info
*fieinfo
)
4586 __u32 flags
= FIEMAP_EXTENT_LAST
;
4587 int blockbits
= inode
->i_sb
->s_blocksize_bits
;
4591 if (ext4_test_inode_state(inode
, EXT4_STATE_XATTR
)) {
4592 struct ext4_iloc iloc
;
4593 int offset
; /* offset of xattr in inode */
4595 error
= ext4_get_inode_loc(inode
, &iloc
);
4598 physical
= iloc
.bh
->b_blocknr
<< blockbits
;
4599 offset
= EXT4_GOOD_OLD_INODE_SIZE
+
4600 EXT4_I(inode
)->i_extra_isize
;
4602 length
= EXT4_SB(inode
->i_sb
)->s_inode_size
- offset
;
4603 flags
|= FIEMAP_EXTENT_DATA_INLINE
;
4605 } else { /* external block */
4606 physical
= EXT4_I(inode
)->i_file_acl
<< blockbits
;
4607 length
= inode
->i_sb
->s_blocksize
;
4611 error
= fiemap_fill_next_extent(fieinfo
, 0, physical
,
4613 return (error
< 0 ? error
: 0);
4617 * ext4_ext_punch_hole
4619 * Punches a hole of "length" bytes in a file starting
4622 * @inode: The inode of the file to punch a hole in
4623 * @offset: The starting byte offset of the hole
4624 * @length: The length of the hole
4626 * Returns the number of blocks removed or negative on err
4628 int ext4_ext_punch_hole(struct file
*file
, loff_t offset
, loff_t length
)
4630 struct inode
*inode
= file
->f_path
.dentry
->d_inode
;
4631 struct super_block
*sb
= inode
->i_sb
;
4632 ext4_lblk_t first_block
, stop_block
;
4633 struct address_space
*mapping
= inode
->i_mapping
;
4635 loff_t first_page
, last_page
, page_len
;
4636 loff_t first_page_offset
, last_page_offset
;
4637 int credits
, err
= 0;
4640 * Write out all dirty pages to avoid race conditions
4641 * Then release them.
4643 if (mapping
->nrpages
&& mapping_tagged(mapping
, PAGECACHE_TAG_DIRTY
)) {
4644 err
= filemap_write_and_wait_range(mapping
,
4645 offset
, offset
+ length
- 1);
4651 mutex_lock(&inode
->i_mutex
);
4652 /* It's not possible punch hole on append only file */
4653 if (IS_APPEND(inode
) || IS_IMMUTABLE(inode
)) {
4657 if (IS_SWAPFILE(inode
)) {
4662 /* No need to punch hole beyond i_size */
4663 if (offset
>= inode
->i_size
)
4667 * If the hole extends beyond i_size, set the hole
4668 * to end after the page that contains i_size
4670 if (offset
+ length
> inode
->i_size
) {
4671 length
= inode
->i_size
+
4672 PAGE_CACHE_SIZE
- (inode
->i_size
& (PAGE_CACHE_SIZE
- 1)) -
4676 first_page
= (offset
+ PAGE_CACHE_SIZE
- 1) >> PAGE_CACHE_SHIFT
;
4677 last_page
= (offset
+ length
) >> PAGE_CACHE_SHIFT
;
4679 first_page_offset
= first_page
<< PAGE_CACHE_SHIFT
;
4680 last_page_offset
= last_page
<< PAGE_CACHE_SHIFT
;
4682 /* Now release the pages */
4683 if (last_page_offset
> first_page_offset
) {
4684 truncate_pagecache_range(inode
, first_page_offset
,
4685 last_page_offset
- 1);
4688 /* Wait all existing dio workers, newcomers will block on i_mutex */
4689 ext4_inode_block_unlocked_dio(inode
);
4690 err
= ext4_flush_unwritten_io(inode
);
4693 inode_dio_wait(inode
);
4695 credits
= ext4_writepage_trans_blocks(inode
);
4696 handle
= ext4_journal_start(inode
, credits
);
4697 if (IS_ERR(handle
)) {
4698 err
= PTR_ERR(handle
);
4704 * Now we need to zero out the non-page-aligned data in the
4705 * pages at the start and tail of the hole, and unmap the buffer
4706 * heads for the block aligned regions of the page that were
4707 * completely zeroed.
4709 if (first_page
> last_page
) {
4711 * If the file space being truncated is contained within a page
4712 * just zero out and unmap the middle of that page
4714 err
= ext4_discard_partial_page_buffers(handle
,
4715 mapping
, offset
, length
, 0);
4721 * zero out and unmap the partial page that contains
4722 * the start of the hole
4724 page_len
= first_page_offset
- offset
;
4726 err
= ext4_discard_partial_page_buffers(handle
, mapping
,
4727 offset
, page_len
, 0);
4733 * zero out and unmap the partial page that contains
4734 * the end of the hole
4736 page_len
= offset
+ length
- last_page_offset
;
4738 err
= ext4_discard_partial_page_buffers(handle
, mapping
,
4739 last_page_offset
, page_len
, 0);
4746 * If i_size is contained in the last page, we need to
4747 * unmap and zero the partial page after i_size
4749 if (inode
->i_size
>> PAGE_CACHE_SHIFT
== last_page
&&
4750 inode
->i_size
% PAGE_CACHE_SIZE
!= 0) {
4752 page_len
= PAGE_CACHE_SIZE
-
4753 (inode
->i_size
& (PAGE_CACHE_SIZE
- 1));
4756 err
= ext4_discard_partial_page_buffers(handle
,
4757 mapping
, inode
->i_size
, page_len
, 0);
4764 first_block
= (offset
+ sb
->s_blocksize
- 1) >>
4765 EXT4_BLOCK_SIZE_BITS(sb
);
4766 stop_block
= (offset
+ length
) >> EXT4_BLOCK_SIZE_BITS(sb
);
4768 /* If there are no blocks to remove, return now */
4769 if (first_block
>= stop_block
)
4772 down_write(&EXT4_I(inode
)->i_data_sem
);
4773 ext4_ext_invalidate_cache(inode
);
4774 ext4_discard_preallocations(inode
);
4776 err
= ext4_es_remove_extent(inode
, first_block
,
4777 stop_block
- first_block
);
4778 err
= ext4_ext_remove_space(inode
, first_block
, stop_block
- 1);
4780 ext4_ext_invalidate_cache(inode
);
4781 ext4_discard_preallocations(inode
);
4784 ext4_handle_sync(handle
);
4786 up_write(&EXT4_I(inode
)->i_data_sem
);
4789 inode
->i_mtime
= inode
->i_ctime
= ext4_current_time(inode
);
4790 ext4_mark_inode_dirty(handle
, inode
);
4791 ext4_journal_stop(handle
);
4793 ext4_inode_resume_unlocked_dio(inode
);
4795 mutex_unlock(&inode
->i_mutex
);
4799 int ext4_fiemap(struct inode
*inode
, struct fiemap_extent_info
*fieinfo
,
4800 __u64 start
, __u64 len
)
4802 ext4_lblk_t start_blk
;
4805 /* fallback to generic here if not in extents fmt */
4806 if (!(ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
)))
4807 return generic_block_fiemap(inode
, fieinfo
, start
, len
,
4810 if (fiemap_check_flags(fieinfo
, EXT4_FIEMAP_FLAGS
))
4813 if (fieinfo
->fi_flags
& FIEMAP_FLAG_XATTR
) {
4814 error
= ext4_xattr_fiemap(inode
, fieinfo
);
4816 ext4_lblk_t len_blks
;
4819 start_blk
= start
>> inode
->i_sb
->s_blocksize_bits
;
4820 last_blk
= (start
+ len
- 1) >> inode
->i_sb
->s_blocksize_bits
;
4821 if (last_blk
>= EXT_MAX_BLOCKS
)
4822 last_blk
= EXT_MAX_BLOCKS
-1;
4823 len_blks
= ((ext4_lblk_t
) last_blk
) - start_blk
+ 1;
4826 * Walk the extent tree gathering extent information
4827 * and pushing extents back to the user.
4829 error
= ext4_fill_fiemap_extents(inode
, start_blk
,