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 extent_status
*newes
);
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
);
720 struct ext4_ext_path
*
721 ext4_ext_find_extent(struct inode
*inode
, ext4_lblk_t block
,
722 struct ext4_ext_path
*path
)
724 struct ext4_extent_header
*eh
;
725 struct buffer_head
*bh
;
726 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
);
759 if (!bh_uptodate_or_lock(bh
)) {
760 trace_ext4_ext_load_extent(inode
, block
,
762 ret
= bh_submit_read(bh
);
768 eh
= ext_block_hdr(bh
);
770 if (unlikely(ppos
> depth
)) {
772 EXT4_ERROR_INODE(inode
,
773 "ppos %d > depth %d", ppos
, depth
);
777 path
[ppos
].p_bh
= bh
;
778 path
[ppos
].p_hdr
= eh
;
781 ret
= ext4_ext_check_block(inode
, eh
, i
, bh
);
786 path
[ppos
].p_depth
= i
;
787 path
[ppos
].p_ext
= NULL
;
788 path
[ppos
].p_idx
= NULL
;
791 ext4_ext_binsearch(inode
, path
+ ppos
, block
);
792 /* if not an empty leaf */
793 if (path
[ppos
].p_ext
)
794 path
[ppos
].p_block
= ext4_ext_pblock(path
[ppos
].p_ext
);
796 ext4_ext_show_path(inode
, path
);
801 ext4_ext_drop_refs(path
);
808 * ext4_ext_insert_index:
809 * insert new index [@logical;@ptr] into the block at @curp;
810 * check where to insert: before @curp or after @curp
812 static int ext4_ext_insert_index(handle_t
*handle
, struct inode
*inode
,
813 struct ext4_ext_path
*curp
,
814 int logical
, ext4_fsblk_t ptr
)
816 struct ext4_extent_idx
*ix
;
819 err
= ext4_ext_get_access(handle
, inode
, curp
);
823 if (unlikely(logical
== le32_to_cpu(curp
->p_idx
->ei_block
))) {
824 EXT4_ERROR_INODE(inode
,
825 "logical %d == ei_block %d!",
826 logical
, le32_to_cpu(curp
->p_idx
->ei_block
));
830 if (unlikely(le16_to_cpu(curp
->p_hdr
->eh_entries
)
831 >= le16_to_cpu(curp
->p_hdr
->eh_max
))) {
832 EXT4_ERROR_INODE(inode
,
833 "eh_entries %d >= eh_max %d!",
834 le16_to_cpu(curp
->p_hdr
->eh_entries
),
835 le16_to_cpu(curp
->p_hdr
->eh_max
));
839 if (logical
> le32_to_cpu(curp
->p_idx
->ei_block
)) {
841 ext_debug("insert new index %d after: %llu\n", logical
, ptr
);
842 ix
= curp
->p_idx
+ 1;
845 ext_debug("insert new index %d before: %llu\n", logical
, ptr
);
849 len
= EXT_LAST_INDEX(curp
->p_hdr
) - ix
+ 1;
852 ext_debug("insert new index %d: "
853 "move %d indices from 0x%p to 0x%p\n",
854 logical
, len
, ix
, ix
+ 1);
855 memmove(ix
+ 1, ix
, len
* sizeof(struct ext4_extent_idx
));
858 if (unlikely(ix
> EXT_MAX_INDEX(curp
->p_hdr
))) {
859 EXT4_ERROR_INODE(inode
, "ix > EXT_MAX_INDEX!");
863 ix
->ei_block
= cpu_to_le32(logical
);
864 ext4_idx_store_pblock(ix
, ptr
);
865 le16_add_cpu(&curp
->p_hdr
->eh_entries
, 1);
867 if (unlikely(ix
> EXT_LAST_INDEX(curp
->p_hdr
))) {
868 EXT4_ERROR_INODE(inode
, "ix > EXT_LAST_INDEX!");
872 err
= ext4_ext_dirty(handle
, inode
, curp
);
873 ext4_std_error(inode
->i_sb
, err
);
880 * inserts new subtree into the path, using free index entry
882 * - allocates all needed blocks (new leaf and all intermediate index blocks)
883 * - makes decision where to split
884 * - moves remaining extents and index entries (right to the split point)
885 * into the newly allocated blocks
886 * - initializes subtree
888 static int ext4_ext_split(handle_t
*handle
, struct inode
*inode
,
890 struct ext4_ext_path
*path
,
891 struct ext4_extent
*newext
, int at
)
893 struct buffer_head
*bh
= NULL
;
894 int depth
= ext_depth(inode
);
895 struct ext4_extent_header
*neh
;
896 struct ext4_extent_idx
*fidx
;
898 ext4_fsblk_t newblock
, oldblock
;
900 ext4_fsblk_t
*ablocks
= NULL
; /* array of allocated blocks */
903 /* make decision: where to split? */
904 /* FIXME: now decision is simplest: at current extent */
906 /* if current leaf will be split, then we should use
907 * border from split point */
908 if (unlikely(path
[depth
].p_ext
> EXT_MAX_EXTENT(path
[depth
].p_hdr
))) {
909 EXT4_ERROR_INODE(inode
, "p_ext > EXT_MAX_EXTENT!");
912 if (path
[depth
].p_ext
!= EXT_MAX_EXTENT(path
[depth
].p_hdr
)) {
913 border
= path
[depth
].p_ext
[1].ee_block
;
914 ext_debug("leaf will be split."
915 " next leaf starts at %d\n",
916 le32_to_cpu(border
));
918 border
= newext
->ee_block
;
919 ext_debug("leaf will be added."
920 " next leaf starts at %d\n",
921 le32_to_cpu(border
));
925 * If error occurs, then we break processing
926 * and mark filesystem read-only. index won't
927 * be inserted and tree will be in consistent
928 * state. Next mount will repair buffers too.
932 * Get array to track all allocated blocks.
933 * We need this to handle errors and free blocks
936 ablocks
= kzalloc(sizeof(ext4_fsblk_t
) * depth
, GFP_NOFS
);
940 /* allocate all needed blocks */
941 ext_debug("allocate %d blocks for indexes/leaf\n", depth
- at
);
942 for (a
= 0; a
< depth
- at
; a
++) {
943 newblock
= ext4_ext_new_meta_block(handle
, inode
, path
,
944 newext
, &err
, flags
);
947 ablocks
[a
] = newblock
;
950 /* initialize new leaf */
951 newblock
= ablocks
[--a
];
952 if (unlikely(newblock
== 0)) {
953 EXT4_ERROR_INODE(inode
, "newblock == 0!");
957 bh
= sb_getblk(inode
->i_sb
, newblock
);
964 err
= ext4_journal_get_create_access(handle
, bh
);
968 neh
= ext_block_hdr(bh
);
970 neh
->eh_max
= cpu_to_le16(ext4_ext_space_block(inode
, 0));
971 neh
->eh_magic
= EXT4_EXT_MAGIC
;
974 /* move remainder of path[depth] to the new leaf */
975 if (unlikely(path
[depth
].p_hdr
->eh_entries
!=
976 path
[depth
].p_hdr
->eh_max
)) {
977 EXT4_ERROR_INODE(inode
, "eh_entries %d != eh_max %d!",
978 path
[depth
].p_hdr
->eh_entries
,
979 path
[depth
].p_hdr
->eh_max
);
983 /* start copy from next extent */
984 m
= EXT_MAX_EXTENT(path
[depth
].p_hdr
) - path
[depth
].p_ext
++;
985 ext4_ext_show_move(inode
, path
, newblock
, depth
);
987 struct ext4_extent
*ex
;
988 ex
= EXT_FIRST_EXTENT(neh
);
989 memmove(ex
, path
[depth
].p_ext
, sizeof(struct ext4_extent
) * m
);
990 le16_add_cpu(&neh
->eh_entries
, m
);
993 ext4_extent_block_csum_set(inode
, neh
);
994 set_buffer_uptodate(bh
);
997 err
= ext4_handle_dirty_metadata(handle
, inode
, bh
);
1003 /* correct old leaf */
1005 err
= ext4_ext_get_access(handle
, inode
, path
+ depth
);
1008 le16_add_cpu(&path
[depth
].p_hdr
->eh_entries
, -m
);
1009 err
= ext4_ext_dirty(handle
, inode
, path
+ depth
);
1015 /* create intermediate indexes */
1017 if (unlikely(k
< 0)) {
1018 EXT4_ERROR_INODE(inode
, "k %d < 0!", k
);
1023 ext_debug("create %d intermediate indices\n", k
);
1024 /* insert new index into current index block */
1025 /* current depth stored in i var */
1028 oldblock
= newblock
;
1029 newblock
= ablocks
[--a
];
1030 bh
= sb_getblk(inode
->i_sb
, newblock
);
1031 if (unlikely(!bh
)) {
1037 err
= ext4_journal_get_create_access(handle
, bh
);
1041 neh
= ext_block_hdr(bh
);
1042 neh
->eh_entries
= cpu_to_le16(1);
1043 neh
->eh_magic
= EXT4_EXT_MAGIC
;
1044 neh
->eh_max
= cpu_to_le16(ext4_ext_space_block_idx(inode
, 0));
1045 neh
->eh_depth
= cpu_to_le16(depth
- i
);
1046 fidx
= EXT_FIRST_INDEX(neh
);
1047 fidx
->ei_block
= border
;
1048 ext4_idx_store_pblock(fidx
, oldblock
);
1050 ext_debug("int.index at %d (block %llu): %u -> %llu\n",
1051 i
, newblock
, le32_to_cpu(border
), oldblock
);
1053 /* move remainder of path[i] to the new index block */
1054 if (unlikely(EXT_MAX_INDEX(path
[i
].p_hdr
) !=
1055 EXT_LAST_INDEX(path
[i
].p_hdr
))) {
1056 EXT4_ERROR_INODE(inode
,
1057 "EXT_MAX_INDEX != EXT_LAST_INDEX ee_block %d!",
1058 le32_to_cpu(path
[i
].p_ext
->ee_block
));
1062 /* start copy indexes */
1063 m
= EXT_MAX_INDEX(path
[i
].p_hdr
) - path
[i
].p_idx
++;
1064 ext_debug("cur 0x%p, last 0x%p\n", path
[i
].p_idx
,
1065 EXT_MAX_INDEX(path
[i
].p_hdr
));
1066 ext4_ext_show_move(inode
, path
, newblock
, i
);
1068 memmove(++fidx
, path
[i
].p_idx
,
1069 sizeof(struct ext4_extent_idx
) * m
);
1070 le16_add_cpu(&neh
->eh_entries
, m
);
1072 ext4_extent_block_csum_set(inode
, neh
);
1073 set_buffer_uptodate(bh
);
1076 err
= ext4_handle_dirty_metadata(handle
, inode
, bh
);
1082 /* correct old index */
1084 err
= ext4_ext_get_access(handle
, inode
, path
+ i
);
1087 le16_add_cpu(&path
[i
].p_hdr
->eh_entries
, -m
);
1088 err
= ext4_ext_dirty(handle
, inode
, path
+ i
);
1096 /* insert new index */
1097 err
= ext4_ext_insert_index(handle
, inode
, path
+ at
,
1098 le32_to_cpu(border
), newblock
);
1102 if (buffer_locked(bh
))
1108 /* free all allocated blocks in error case */
1109 for (i
= 0; i
< depth
; i
++) {
1112 ext4_free_blocks(handle
, inode
, NULL
, ablocks
[i
], 1,
1113 EXT4_FREE_BLOCKS_METADATA
);
1122 * ext4_ext_grow_indepth:
1123 * implements tree growing procedure:
1124 * - allocates new block
1125 * - moves top-level data (index block or leaf) into the new block
1126 * - initializes new top-level, creating index that points to the
1127 * just created block
1129 static int ext4_ext_grow_indepth(handle_t
*handle
, struct inode
*inode
,
1131 struct ext4_extent
*newext
)
1133 struct ext4_extent_header
*neh
;
1134 struct buffer_head
*bh
;
1135 ext4_fsblk_t newblock
;
1138 newblock
= ext4_ext_new_meta_block(handle
, inode
, NULL
,
1139 newext
, &err
, flags
);
1143 bh
= sb_getblk(inode
->i_sb
, newblock
);
1148 err
= ext4_journal_get_create_access(handle
, bh
);
1154 /* move top-level index/leaf into new block */
1155 memmove(bh
->b_data
, EXT4_I(inode
)->i_data
,
1156 sizeof(EXT4_I(inode
)->i_data
));
1158 /* set size of new block */
1159 neh
= ext_block_hdr(bh
);
1160 /* old root could have indexes or leaves
1161 * so calculate e_max right way */
1162 if (ext_depth(inode
))
1163 neh
->eh_max
= cpu_to_le16(ext4_ext_space_block_idx(inode
, 0));
1165 neh
->eh_max
= cpu_to_le16(ext4_ext_space_block(inode
, 0));
1166 neh
->eh_magic
= EXT4_EXT_MAGIC
;
1167 ext4_extent_block_csum_set(inode
, neh
);
1168 set_buffer_uptodate(bh
);
1171 err
= ext4_handle_dirty_metadata(handle
, inode
, bh
);
1175 /* Update top-level index: num,max,pointer */
1176 neh
= ext_inode_hdr(inode
);
1177 neh
->eh_entries
= cpu_to_le16(1);
1178 ext4_idx_store_pblock(EXT_FIRST_INDEX(neh
), newblock
);
1179 if (neh
->eh_depth
== 0) {
1180 /* Root extent block becomes index block */
1181 neh
->eh_max
= cpu_to_le16(ext4_ext_space_root_idx(inode
, 0));
1182 EXT_FIRST_INDEX(neh
)->ei_block
=
1183 EXT_FIRST_EXTENT(neh
)->ee_block
;
1185 ext_debug("new root: num %d(%d), lblock %d, ptr %llu\n",
1186 le16_to_cpu(neh
->eh_entries
), le16_to_cpu(neh
->eh_max
),
1187 le32_to_cpu(EXT_FIRST_INDEX(neh
)->ei_block
),
1188 ext4_idx_pblock(EXT_FIRST_INDEX(neh
)));
1190 le16_add_cpu(&neh
->eh_depth
, 1);
1191 ext4_mark_inode_dirty(handle
, inode
);
1199 * ext4_ext_create_new_leaf:
1200 * finds empty index and adds new leaf.
1201 * if no free index is found, then it requests in-depth growing.
1203 static int ext4_ext_create_new_leaf(handle_t
*handle
, struct inode
*inode
,
1205 struct ext4_ext_path
*path
,
1206 struct ext4_extent
*newext
)
1208 struct ext4_ext_path
*curp
;
1209 int depth
, i
, err
= 0;
1212 i
= depth
= ext_depth(inode
);
1214 /* walk up to the tree and look for free index entry */
1215 curp
= path
+ depth
;
1216 while (i
> 0 && !EXT_HAS_FREE_INDEX(curp
)) {
1221 /* we use already allocated block for index block,
1222 * so subsequent data blocks should be contiguous */
1223 if (EXT_HAS_FREE_INDEX(curp
)) {
1224 /* if we found index with free entry, then use that
1225 * entry: create all needed subtree and add new leaf */
1226 err
= ext4_ext_split(handle
, inode
, flags
, path
, newext
, i
);
1231 ext4_ext_drop_refs(path
);
1232 path
= ext4_ext_find_extent(inode
,
1233 (ext4_lblk_t
)le32_to_cpu(newext
->ee_block
),
1236 err
= PTR_ERR(path
);
1238 /* tree is full, time to grow in depth */
1239 err
= ext4_ext_grow_indepth(handle
, inode
, flags
, newext
);
1244 ext4_ext_drop_refs(path
);
1245 path
= ext4_ext_find_extent(inode
,
1246 (ext4_lblk_t
)le32_to_cpu(newext
->ee_block
),
1249 err
= PTR_ERR(path
);
1254 * only first (depth 0 -> 1) produces free space;
1255 * in all other cases we have to split the grown tree
1257 depth
= ext_depth(inode
);
1258 if (path
[depth
].p_hdr
->eh_entries
== path
[depth
].p_hdr
->eh_max
) {
1259 /* now we need to split */
1269 * search the closest allocated block to the left for *logical
1270 * and returns it at @logical + it's physical address at @phys
1271 * if *logical is the smallest allocated block, the function
1272 * returns 0 at @phys
1273 * return value contains 0 (success) or error code
1275 static int ext4_ext_search_left(struct inode
*inode
,
1276 struct ext4_ext_path
*path
,
1277 ext4_lblk_t
*logical
, ext4_fsblk_t
*phys
)
1279 struct ext4_extent_idx
*ix
;
1280 struct ext4_extent
*ex
;
1283 if (unlikely(path
== NULL
)) {
1284 EXT4_ERROR_INODE(inode
, "path == NULL *logical %d!", *logical
);
1287 depth
= path
->p_depth
;
1290 if (depth
== 0 && path
->p_ext
== NULL
)
1293 /* usually extent in the path covers blocks smaller
1294 * then *logical, but it can be that extent is the
1295 * first one in the file */
1297 ex
= path
[depth
].p_ext
;
1298 ee_len
= ext4_ext_get_actual_len(ex
);
1299 if (*logical
< le32_to_cpu(ex
->ee_block
)) {
1300 if (unlikely(EXT_FIRST_EXTENT(path
[depth
].p_hdr
) != ex
)) {
1301 EXT4_ERROR_INODE(inode
,
1302 "EXT_FIRST_EXTENT != ex *logical %d ee_block %d!",
1303 *logical
, le32_to_cpu(ex
->ee_block
));
1306 while (--depth
>= 0) {
1307 ix
= path
[depth
].p_idx
;
1308 if (unlikely(ix
!= EXT_FIRST_INDEX(path
[depth
].p_hdr
))) {
1309 EXT4_ERROR_INODE(inode
,
1310 "ix (%d) != EXT_FIRST_INDEX (%d) (depth %d)!",
1311 ix
!= NULL
? le32_to_cpu(ix
->ei_block
) : 0,
1312 EXT_FIRST_INDEX(path
[depth
].p_hdr
) != NULL
?
1313 le32_to_cpu(EXT_FIRST_INDEX(path
[depth
].p_hdr
)->ei_block
) : 0,
1321 if (unlikely(*logical
< (le32_to_cpu(ex
->ee_block
) + ee_len
))) {
1322 EXT4_ERROR_INODE(inode
,
1323 "logical %d < ee_block %d + ee_len %d!",
1324 *logical
, le32_to_cpu(ex
->ee_block
), ee_len
);
1328 *logical
= le32_to_cpu(ex
->ee_block
) + ee_len
- 1;
1329 *phys
= ext4_ext_pblock(ex
) + ee_len
- 1;
1334 * search the closest allocated block to the right for *logical
1335 * and returns it at @logical + it's physical address at @phys
1336 * if *logical is the largest allocated block, the function
1337 * returns 0 at @phys
1338 * return value contains 0 (success) or error code
1340 static int ext4_ext_search_right(struct inode
*inode
,
1341 struct ext4_ext_path
*path
,
1342 ext4_lblk_t
*logical
, ext4_fsblk_t
*phys
,
1343 struct ext4_extent
**ret_ex
)
1345 struct buffer_head
*bh
= NULL
;
1346 struct ext4_extent_header
*eh
;
1347 struct ext4_extent_idx
*ix
;
1348 struct ext4_extent
*ex
;
1350 int depth
; /* Note, NOT eh_depth; depth from top of tree */
1353 if (unlikely(path
== NULL
)) {
1354 EXT4_ERROR_INODE(inode
, "path == NULL *logical %d!", *logical
);
1357 depth
= path
->p_depth
;
1360 if (depth
== 0 && path
->p_ext
== NULL
)
1363 /* usually extent in the path covers blocks smaller
1364 * then *logical, but it can be that extent is the
1365 * first one in the file */
1367 ex
= path
[depth
].p_ext
;
1368 ee_len
= ext4_ext_get_actual_len(ex
);
1369 if (*logical
< le32_to_cpu(ex
->ee_block
)) {
1370 if (unlikely(EXT_FIRST_EXTENT(path
[depth
].p_hdr
) != ex
)) {
1371 EXT4_ERROR_INODE(inode
,
1372 "first_extent(path[%d].p_hdr) != ex",
1376 while (--depth
>= 0) {
1377 ix
= path
[depth
].p_idx
;
1378 if (unlikely(ix
!= EXT_FIRST_INDEX(path
[depth
].p_hdr
))) {
1379 EXT4_ERROR_INODE(inode
,
1380 "ix != EXT_FIRST_INDEX *logical %d!",
1388 if (unlikely(*logical
< (le32_to_cpu(ex
->ee_block
) + ee_len
))) {
1389 EXT4_ERROR_INODE(inode
,
1390 "logical %d < ee_block %d + ee_len %d!",
1391 *logical
, le32_to_cpu(ex
->ee_block
), ee_len
);
1395 if (ex
!= EXT_LAST_EXTENT(path
[depth
].p_hdr
)) {
1396 /* next allocated block in this leaf */
1401 /* go up and search for index to the right */
1402 while (--depth
>= 0) {
1403 ix
= path
[depth
].p_idx
;
1404 if (ix
!= EXT_LAST_INDEX(path
[depth
].p_hdr
))
1408 /* we've gone up to the root and found no index to the right */
1412 /* we've found index to the right, let's
1413 * follow it and find the closest allocated
1414 * block to the right */
1416 block
= ext4_idx_pblock(ix
);
1417 while (++depth
< path
->p_depth
) {
1418 bh
= sb_bread(inode
->i_sb
, block
);
1421 eh
= ext_block_hdr(bh
);
1422 /* subtract from p_depth to get proper eh_depth */
1423 if (ext4_ext_check_block(inode
, eh
,
1424 path
->p_depth
- depth
, bh
)) {
1428 ix
= EXT_FIRST_INDEX(eh
);
1429 block
= ext4_idx_pblock(ix
);
1433 bh
= sb_bread(inode
->i_sb
, block
);
1436 eh
= ext_block_hdr(bh
);
1437 if (ext4_ext_check_block(inode
, eh
, path
->p_depth
- depth
, bh
)) {
1441 ex
= EXT_FIRST_EXTENT(eh
);
1443 *logical
= le32_to_cpu(ex
->ee_block
);
1444 *phys
= ext4_ext_pblock(ex
);
1452 * ext4_ext_next_allocated_block:
1453 * returns allocated block in subsequent extent or EXT_MAX_BLOCKS.
1454 * NOTE: it considers block number from index entry as
1455 * allocated block. Thus, index entries have to be consistent
1459 ext4_ext_next_allocated_block(struct ext4_ext_path
*path
)
1463 BUG_ON(path
== NULL
);
1464 depth
= path
->p_depth
;
1466 if (depth
== 0 && path
->p_ext
== NULL
)
1467 return EXT_MAX_BLOCKS
;
1469 while (depth
>= 0) {
1470 if (depth
== path
->p_depth
) {
1472 if (path
[depth
].p_ext
&&
1473 path
[depth
].p_ext
!=
1474 EXT_LAST_EXTENT(path
[depth
].p_hdr
))
1475 return le32_to_cpu(path
[depth
].p_ext
[1].ee_block
);
1478 if (path
[depth
].p_idx
!=
1479 EXT_LAST_INDEX(path
[depth
].p_hdr
))
1480 return le32_to_cpu(path
[depth
].p_idx
[1].ei_block
);
1485 return EXT_MAX_BLOCKS
;
1489 * ext4_ext_next_leaf_block:
1490 * returns first allocated block from next leaf or EXT_MAX_BLOCKS
1492 static ext4_lblk_t
ext4_ext_next_leaf_block(struct ext4_ext_path
*path
)
1496 BUG_ON(path
== NULL
);
1497 depth
= path
->p_depth
;
1499 /* zero-tree has no leaf blocks at all */
1501 return EXT_MAX_BLOCKS
;
1503 /* go to index block */
1506 while (depth
>= 0) {
1507 if (path
[depth
].p_idx
!=
1508 EXT_LAST_INDEX(path
[depth
].p_hdr
))
1509 return (ext4_lblk_t
)
1510 le32_to_cpu(path
[depth
].p_idx
[1].ei_block
);
1514 return EXT_MAX_BLOCKS
;
1518 * ext4_ext_correct_indexes:
1519 * if leaf gets modified and modified extent is first in the leaf,
1520 * then we have to correct all indexes above.
1521 * TODO: do we need to correct tree in all cases?
1523 static int ext4_ext_correct_indexes(handle_t
*handle
, struct inode
*inode
,
1524 struct ext4_ext_path
*path
)
1526 struct ext4_extent_header
*eh
;
1527 int depth
= ext_depth(inode
);
1528 struct ext4_extent
*ex
;
1532 eh
= path
[depth
].p_hdr
;
1533 ex
= path
[depth
].p_ext
;
1535 if (unlikely(ex
== NULL
|| eh
== NULL
)) {
1536 EXT4_ERROR_INODE(inode
,
1537 "ex %p == NULL or eh %p == NULL", ex
, eh
);
1542 /* there is no tree at all */
1546 if (ex
!= EXT_FIRST_EXTENT(eh
)) {
1547 /* we correct tree if first leaf got modified only */
1552 * TODO: we need correction if border is smaller than current one
1555 border
= path
[depth
].p_ext
->ee_block
;
1556 err
= ext4_ext_get_access(handle
, inode
, path
+ k
);
1559 path
[k
].p_idx
->ei_block
= border
;
1560 err
= ext4_ext_dirty(handle
, inode
, path
+ k
);
1565 /* change all left-side indexes */
1566 if (path
[k
+1].p_idx
!= EXT_FIRST_INDEX(path
[k
+1].p_hdr
))
1568 err
= ext4_ext_get_access(handle
, inode
, path
+ k
);
1571 path
[k
].p_idx
->ei_block
= border
;
1572 err
= ext4_ext_dirty(handle
, inode
, path
+ k
);
1581 ext4_can_extents_be_merged(struct inode
*inode
, struct ext4_extent
*ex1
,
1582 struct ext4_extent
*ex2
)
1584 unsigned short ext1_ee_len
, ext2_ee_len
, max_len
;
1587 * Make sure that either both extents are uninitialized, or
1590 if (ext4_ext_is_uninitialized(ex1
) ^ ext4_ext_is_uninitialized(ex2
))
1593 if (ext4_ext_is_uninitialized(ex1
))
1594 max_len
= EXT_UNINIT_MAX_LEN
;
1596 max_len
= EXT_INIT_MAX_LEN
;
1598 ext1_ee_len
= ext4_ext_get_actual_len(ex1
);
1599 ext2_ee_len
= ext4_ext_get_actual_len(ex2
);
1601 if (le32_to_cpu(ex1
->ee_block
) + ext1_ee_len
!=
1602 le32_to_cpu(ex2
->ee_block
))
1606 * To allow future support for preallocated extents to be added
1607 * as an RO_COMPAT feature, refuse to merge to extents if
1608 * this can result in the top bit of ee_len being set.
1610 if (ext1_ee_len
+ ext2_ee_len
> max_len
)
1612 #ifdef AGGRESSIVE_TEST
1613 if (ext1_ee_len
>= 4)
1617 if (ext4_ext_pblock(ex1
) + ext1_ee_len
== ext4_ext_pblock(ex2
))
1623 * This function tries to merge the "ex" extent to the next extent in the tree.
1624 * It always tries to merge towards right. If you want to merge towards
1625 * left, pass "ex - 1" as argument instead of "ex".
1626 * Returns 0 if the extents (ex and ex+1) were _not_ merged and returns
1627 * 1 if they got merged.
1629 static int ext4_ext_try_to_merge_right(struct inode
*inode
,
1630 struct ext4_ext_path
*path
,
1631 struct ext4_extent
*ex
)
1633 struct ext4_extent_header
*eh
;
1634 unsigned int depth
, len
;
1636 int uninitialized
= 0;
1638 depth
= ext_depth(inode
);
1639 BUG_ON(path
[depth
].p_hdr
== NULL
);
1640 eh
= path
[depth
].p_hdr
;
1642 while (ex
< EXT_LAST_EXTENT(eh
)) {
1643 if (!ext4_can_extents_be_merged(inode
, ex
, ex
+ 1))
1645 /* merge with next extent! */
1646 if (ext4_ext_is_uninitialized(ex
))
1648 ex
->ee_len
= cpu_to_le16(ext4_ext_get_actual_len(ex
)
1649 + ext4_ext_get_actual_len(ex
+ 1));
1651 ext4_ext_mark_uninitialized(ex
);
1653 if (ex
+ 1 < EXT_LAST_EXTENT(eh
)) {
1654 len
= (EXT_LAST_EXTENT(eh
) - ex
- 1)
1655 * sizeof(struct ext4_extent
);
1656 memmove(ex
+ 1, ex
+ 2, len
);
1658 le16_add_cpu(&eh
->eh_entries
, -1);
1660 WARN_ON(eh
->eh_entries
== 0);
1661 if (!eh
->eh_entries
)
1662 EXT4_ERROR_INODE(inode
, "eh->eh_entries = 0!");
1669 * This function does a very simple check to see if we can collapse
1670 * an extent tree with a single extent tree leaf block into the inode.
1672 static void ext4_ext_try_to_merge_up(handle_t
*handle
,
1673 struct inode
*inode
,
1674 struct ext4_ext_path
*path
)
1677 unsigned max_root
= ext4_ext_space_root(inode
, 0);
1680 if ((path
[0].p_depth
!= 1) ||
1681 (le16_to_cpu(path
[0].p_hdr
->eh_entries
) != 1) ||
1682 (le16_to_cpu(path
[1].p_hdr
->eh_entries
) > max_root
))
1686 * We need to modify the block allocation bitmap and the block
1687 * group descriptor to release the extent tree block. If we
1688 * can't get the journal credits, give up.
1690 if (ext4_journal_extend(handle
, 2))
1694 * Copy the extent data up to the inode
1696 blk
= ext4_idx_pblock(path
[0].p_idx
);
1697 s
= le16_to_cpu(path
[1].p_hdr
->eh_entries
) *
1698 sizeof(struct ext4_extent_idx
);
1699 s
+= sizeof(struct ext4_extent_header
);
1701 memcpy(path
[0].p_hdr
, path
[1].p_hdr
, s
);
1702 path
[0].p_depth
= 0;
1703 path
[0].p_ext
= EXT_FIRST_EXTENT(path
[0].p_hdr
) +
1704 (path
[1].p_ext
- EXT_FIRST_EXTENT(path
[1].p_hdr
));
1705 path
[0].p_hdr
->eh_max
= cpu_to_le16(max_root
);
1707 brelse(path
[1].p_bh
);
1708 ext4_free_blocks(handle
, inode
, NULL
, blk
, 1,
1709 EXT4_FREE_BLOCKS_METADATA
| EXT4_FREE_BLOCKS_FORGET
);
1713 * This function tries to merge the @ex extent to neighbours in the tree.
1714 * return 1 if merge left else 0.
1716 static void ext4_ext_try_to_merge(handle_t
*handle
,
1717 struct inode
*inode
,
1718 struct ext4_ext_path
*path
,
1719 struct ext4_extent
*ex
) {
1720 struct ext4_extent_header
*eh
;
1724 depth
= ext_depth(inode
);
1725 BUG_ON(path
[depth
].p_hdr
== NULL
);
1726 eh
= path
[depth
].p_hdr
;
1728 if (ex
> EXT_FIRST_EXTENT(eh
))
1729 merge_done
= ext4_ext_try_to_merge_right(inode
, path
, ex
- 1);
1732 (void) ext4_ext_try_to_merge_right(inode
, path
, ex
);
1734 ext4_ext_try_to_merge_up(handle
, inode
, path
);
1738 * check if a portion of the "newext" extent overlaps with an
1741 * If there is an overlap discovered, it updates the length of the newext
1742 * such that there will be no overlap, and then returns 1.
1743 * If there is no overlap found, it returns 0.
1745 static unsigned int ext4_ext_check_overlap(struct ext4_sb_info
*sbi
,
1746 struct inode
*inode
,
1747 struct ext4_extent
*newext
,
1748 struct ext4_ext_path
*path
)
1751 unsigned int depth
, len1
;
1752 unsigned int ret
= 0;
1754 b1
= le32_to_cpu(newext
->ee_block
);
1755 len1
= ext4_ext_get_actual_len(newext
);
1756 depth
= ext_depth(inode
);
1757 if (!path
[depth
].p_ext
)
1759 b2
= le32_to_cpu(path
[depth
].p_ext
->ee_block
);
1760 b2
&= ~(sbi
->s_cluster_ratio
- 1);
1763 * get the next allocated block if the extent in the path
1764 * is before the requested block(s)
1767 b2
= ext4_ext_next_allocated_block(path
);
1768 if (b2
== EXT_MAX_BLOCKS
)
1770 b2
&= ~(sbi
->s_cluster_ratio
- 1);
1773 /* check for wrap through zero on extent logical start block*/
1774 if (b1
+ len1
< b1
) {
1775 len1
= EXT_MAX_BLOCKS
- b1
;
1776 newext
->ee_len
= cpu_to_le16(len1
);
1780 /* check for overlap */
1781 if (b1
+ len1
> b2
) {
1782 newext
->ee_len
= cpu_to_le16(b2
- b1
);
1790 * ext4_ext_insert_extent:
1791 * tries to merge requsted extent into the existing extent or
1792 * inserts requested extent as new one into the tree,
1793 * creating new leaf in the no-space case.
1795 int ext4_ext_insert_extent(handle_t
*handle
, struct inode
*inode
,
1796 struct ext4_ext_path
*path
,
1797 struct ext4_extent
*newext
, int flag
)
1799 struct ext4_extent_header
*eh
;
1800 struct ext4_extent
*ex
, *fex
;
1801 struct ext4_extent
*nearex
; /* nearest extent */
1802 struct ext4_ext_path
*npath
= NULL
;
1803 int depth
, len
, err
;
1805 unsigned uninitialized
= 0;
1808 if (unlikely(ext4_ext_get_actual_len(newext
) == 0)) {
1809 EXT4_ERROR_INODE(inode
, "ext4_ext_get_actual_len(newext) == 0");
1812 depth
= ext_depth(inode
);
1813 ex
= path
[depth
].p_ext
;
1814 if (unlikely(path
[depth
].p_hdr
== NULL
)) {
1815 EXT4_ERROR_INODE(inode
, "path[%d].p_hdr == NULL", depth
);
1819 /* try to insert block into found extent and return */
1820 if (ex
&& !(flag
& EXT4_GET_BLOCKS_PRE_IO
)
1821 && ext4_can_extents_be_merged(inode
, ex
, newext
)) {
1822 ext_debug("append [%d]%d block to %u:[%d]%d (from %llu)\n",
1823 ext4_ext_is_uninitialized(newext
),
1824 ext4_ext_get_actual_len(newext
),
1825 le32_to_cpu(ex
->ee_block
),
1826 ext4_ext_is_uninitialized(ex
),
1827 ext4_ext_get_actual_len(ex
),
1828 ext4_ext_pblock(ex
));
1829 err
= ext4_ext_get_access(handle
, inode
, path
+ depth
);
1834 * ext4_can_extents_be_merged should have checked that either
1835 * both extents are uninitialized, or both aren't. Thus we
1836 * need to check only one of them here.
1838 if (ext4_ext_is_uninitialized(ex
))
1840 ex
->ee_len
= cpu_to_le16(ext4_ext_get_actual_len(ex
)
1841 + ext4_ext_get_actual_len(newext
));
1843 ext4_ext_mark_uninitialized(ex
);
1844 eh
= path
[depth
].p_hdr
;
1849 depth
= ext_depth(inode
);
1850 eh
= path
[depth
].p_hdr
;
1851 if (le16_to_cpu(eh
->eh_entries
) < le16_to_cpu(eh
->eh_max
))
1854 /* probably next leaf has space for us? */
1855 fex
= EXT_LAST_EXTENT(eh
);
1856 next
= EXT_MAX_BLOCKS
;
1857 if (le32_to_cpu(newext
->ee_block
) > le32_to_cpu(fex
->ee_block
))
1858 next
= ext4_ext_next_leaf_block(path
);
1859 if (next
!= EXT_MAX_BLOCKS
) {
1860 ext_debug("next leaf block - %u\n", next
);
1861 BUG_ON(npath
!= NULL
);
1862 npath
= ext4_ext_find_extent(inode
, next
, NULL
);
1864 return PTR_ERR(npath
);
1865 BUG_ON(npath
->p_depth
!= path
->p_depth
);
1866 eh
= npath
[depth
].p_hdr
;
1867 if (le16_to_cpu(eh
->eh_entries
) < le16_to_cpu(eh
->eh_max
)) {
1868 ext_debug("next leaf isn't full(%d)\n",
1869 le16_to_cpu(eh
->eh_entries
));
1873 ext_debug("next leaf has no free space(%d,%d)\n",
1874 le16_to_cpu(eh
->eh_entries
), le16_to_cpu(eh
->eh_max
));
1878 * There is no free space in the found leaf.
1879 * We're gonna add a new leaf in the tree.
1881 if (flag
& EXT4_GET_BLOCKS_PUNCH_OUT_EXT
)
1882 flags
= EXT4_MB_USE_ROOT_BLOCKS
;
1883 err
= ext4_ext_create_new_leaf(handle
, inode
, flags
, path
, newext
);
1886 depth
= ext_depth(inode
);
1887 eh
= path
[depth
].p_hdr
;
1890 nearex
= path
[depth
].p_ext
;
1892 err
= ext4_ext_get_access(handle
, inode
, path
+ depth
);
1897 /* there is no extent in this leaf, create first one */
1898 ext_debug("first extent in the leaf: %u:%llu:[%d]%d\n",
1899 le32_to_cpu(newext
->ee_block
),
1900 ext4_ext_pblock(newext
),
1901 ext4_ext_is_uninitialized(newext
),
1902 ext4_ext_get_actual_len(newext
));
1903 nearex
= EXT_FIRST_EXTENT(eh
);
1905 if (le32_to_cpu(newext
->ee_block
)
1906 > le32_to_cpu(nearex
->ee_block
)) {
1908 ext_debug("insert %u:%llu:[%d]%d before: "
1910 le32_to_cpu(newext
->ee_block
),
1911 ext4_ext_pblock(newext
),
1912 ext4_ext_is_uninitialized(newext
),
1913 ext4_ext_get_actual_len(newext
),
1918 BUG_ON(newext
->ee_block
== nearex
->ee_block
);
1919 ext_debug("insert %u:%llu:[%d]%d after: "
1921 le32_to_cpu(newext
->ee_block
),
1922 ext4_ext_pblock(newext
),
1923 ext4_ext_is_uninitialized(newext
),
1924 ext4_ext_get_actual_len(newext
),
1927 len
= EXT_LAST_EXTENT(eh
) - nearex
+ 1;
1929 ext_debug("insert %u:%llu:[%d]%d: "
1930 "move %d extents from 0x%p to 0x%p\n",
1931 le32_to_cpu(newext
->ee_block
),
1932 ext4_ext_pblock(newext
),
1933 ext4_ext_is_uninitialized(newext
),
1934 ext4_ext_get_actual_len(newext
),
1935 len
, nearex
, nearex
+ 1);
1936 memmove(nearex
+ 1, nearex
,
1937 len
* sizeof(struct ext4_extent
));
1941 le16_add_cpu(&eh
->eh_entries
, 1);
1942 path
[depth
].p_ext
= nearex
;
1943 nearex
->ee_block
= newext
->ee_block
;
1944 ext4_ext_store_pblock(nearex
, ext4_ext_pblock(newext
));
1945 nearex
->ee_len
= newext
->ee_len
;
1948 /* try to merge extents */
1949 if (!(flag
& EXT4_GET_BLOCKS_PRE_IO
))
1950 ext4_ext_try_to_merge(handle
, inode
, path
, nearex
);
1953 /* time to correct all indexes above */
1954 err
= ext4_ext_correct_indexes(handle
, inode
, path
);
1958 err
= ext4_ext_dirty(handle
, inode
, path
+ path
->p_depth
);
1962 ext4_ext_drop_refs(npath
);
1968 static int ext4_fill_fiemap_extents(struct inode
*inode
,
1969 ext4_lblk_t block
, ext4_lblk_t num
,
1970 struct fiemap_extent_info
*fieinfo
)
1972 struct ext4_ext_path
*path
= NULL
;
1973 struct ext4_extent
*ex
;
1974 struct extent_status es
;
1975 ext4_lblk_t next
, next_del
, start
= 0, end
= 0;
1976 ext4_lblk_t last
= block
+ num
;
1977 int exists
, depth
= 0, err
= 0;
1978 unsigned int flags
= 0;
1979 unsigned char blksize_bits
= inode
->i_sb
->s_blocksize_bits
;
1981 while (block
< last
&& block
!= EXT_MAX_BLOCKS
) {
1983 /* find extent for this block */
1984 down_read(&EXT4_I(inode
)->i_data_sem
);
1986 if (path
&& ext_depth(inode
) != depth
) {
1987 /* depth was changed. we have to realloc path */
1992 path
= ext4_ext_find_extent(inode
, block
, path
);
1994 up_read(&EXT4_I(inode
)->i_data_sem
);
1995 err
= PTR_ERR(path
);
2000 depth
= ext_depth(inode
);
2001 if (unlikely(path
[depth
].p_hdr
== NULL
)) {
2002 up_read(&EXT4_I(inode
)->i_data_sem
);
2003 EXT4_ERROR_INODE(inode
, "path[%d].p_hdr == NULL", depth
);
2007 ex
= path
[depth
].p_ext
;
2008 next
= ext4_ext_next_allocated_block(path
);
2009 ext4_ext_drop_refs(path
);
2014 /* there is no extent yet, so try to allocate
2015 * all requested space */
2018 } else if (le32_to_cpu(ex
->ee_block
) > block
) {
2019 /* need to allocate space before found extent */
2021 end
= le32_to_cpu(ex
->ee_block
);
2022 if (block
+ num
< end
)
2024 } else if (block
>= le32_to_cpu(ex
->ee_block
)
2025 + ext4_ext_get_actual_len(ex
)) {
2026 /* need to allocate space after found extent */
2031 } else if (block
>= le32_to_cpu(ex
->ee_block
)) {
2033 * some part of requested space is covered
2037 end
= le32_to_cpu(ex
->ee_block
)
2038 + ext4_ext_get_actual_len(ex
);
2039 if (block
+ num
< end
)
2045 BUG_ON(end
<= start
);
2049 es
.es_len
= end
- start
;
2052 es
.es_lblk
= le32_to_cpu(ex
->ee_block
);
2053 es
.es_len
= ext4_ext_get_actual_len(ex
);
2054 es
.es_pblk
= ext4_ext_pblock(ex
);
2055 if (ext4_ext_is_uninitialized(ex
))
2056 flags
|= FIEMAP_EXTENT_UNWRITTEN
;
2060 * Find delayed extent and update es accordingly. We call
2061 * it even in !exists case to find out whether es is the
2062 * last existing extent or not.
2064 next_del
= ext4_find_delayed_extent(inode
, &es
);
2065 if (!exists
&& next_del
) {
2067 flags
|= FIEMAP_EXTENT_DELALLOC
;
2069 up_read(&EXT4_I(inode
)->i_data_sem
);
2071 if (unlikely(es
.es_len
== 0)) {
2072 EXT4_ERROR_INODE(inode
, "es.es_len == 0");
2078 * This is possible iff next == next_del == EXT_MAX_BLOCKS.
2079 * we need to check next == EXT_MAX_BLOCKS because it is
2080 * possible that an extent is with unwritten and delayed
2081 * status due to when an extent is delayed allocated and
2082 * is allocated by fallocate status tree will track both of
2085 * So we could return a unwritten and delayed extent, and
2086 * its block is equal to 'next'.
2088 if (next
== next_del
&& next
== EXT_MAX_BLOCKS
) {
2089 flags
|= FIEMAP_EXTENT_LAST
;
2090 if (unlikely(next_del
!= EXT_MAX_BLOCKS
||
2091 next
!= EXT_MAX_BLOCKS
)) {
2092 EXT4_ERROR_INODE(inode
,
2093 "next extent == %u, next "
2094 "delalloc extent = %u",
2102 err
= fiemap_fill_next_extent(fieinfo
,
2103 (__u64
)es
.es_lblk
<< blksize_bits
,
2104 (__u64
)es
.es_pblk
<< blksize_bits
,
2105 (__u64
)es
.es_len
<< blksize_bits
,
2115 block
= es
.es_lblk
+ es
.es_len
;
2119 ext4_ext_drop_refs(path
);
2127 * ext4_ext_put_gap_in_cache:
2128 * calculate boundaries of the gap that the requested block fits into
2129 * and cache this gap
2132 ext4_ext_put_gap_in_cache(struct inode
*inode
, struct ext4_ext_path
*path
,
2135 int depth
= ext_depth(inode
);
2138 struct ext4_extent
*ex
;
2140 ex
= path
[depth
].p_ext
;
2143 * there is no extent yet, so gap is [0;-] and we
2146 ext_debug("cache gap(whole file):");
2147 } else if (block
< le32_to_cpu(ex
->ee_block
)) {
2149 len
= le32_to_cpu(ex
->ee_block
) - block
;
2150 ext_debug("cache gap(before): %u [%u:%u]",
2152 le32_to_cpu(ex
->ee_block
),
2153 ext4_ext_get_actual_len(ex
));
2154 if (!ext4_find_delalloc_range(inode
, lblock
, lblock
+ len
- 1))
2155 ext4_es_insert_extent(inode
, lblock
, len
, ~0,
2156 EXTENT_STATUS_HOLE
);
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
;
2170 if (!ext4_find_delalloc_range(inode
, lblock
, lblock
+ len
- 1))
2171 ext4_es_insert_extent(inode
, lblock
, len
, ~0,
2172 EXTENT_STATUS_HOLE
);
2178 ext_debug(" -> %u:%lu\n", lblock
, len
);
2183 * removes index from the index block.
2185 static int ext4_ext_rm_idx(handle_t
*handle
, struct inode
*inode
,
2186 struct ext4_ext_path
*path
, int depth
)
2191 /* free index block */
2193 path
= path
+ depth
;
2194 leaf
= ext4_idx_pblock(path
->p_idx
);
2195 if (unlikely(path
->p_hdr
->eh_entries
== 0)) {
2196 EXT4_ERROR_INODE(inode
, "path->p_hdr->eh_entries == 0");
2199 err
= ext4_ext_get_access(handle
, inode
, path
);
2203 if (path
->p_idx
!= EXT_LAST_INDEX(path
->p_hdr
)) {
2204 int len
= EXT_LAST_INDEX(path
->p_hdr
) - path
->p_idx
;
2205 len
*= sizeof(struct ext4_extent_idx
);
2206 memmove(path
->p_idx
, path
->p_idx
+ 1, len
);
2209 le16_add_cpu(&path
->p_hdr
->eh_entries
, -1);
2210 err
= ext4_ext_dirty(handle
, inode
, path
);
2213 ext_debug("index is empty, remove it, free block %llu\n", leaf
);
2214 trace_ext4_ext_rm_idx(inode
, leaf
);
2216 ext4_free_blocks(handle
, inode
, NULL
, leaf
, 1,
2217 EXT4_FREE_BLOCKS_METADATA
| EXT4_FREE_BLOCKS_FORGET
);
2219 while (--depth
>= 0) {
2220 if (path
->p_idx
!= EXT_FIRST_INDEX(path
->p_hdr
))
2223 err
= ext4_ext_get_access(handle
, inode
, path
);
2226 path
->p_idx
->ei_block
= (path
+1)->p_idx
->ei_block
;
2227 err
= ext4_ext_dirty(handle
, inode
, path
);
2235 * ext4_ext_calc_credits_for_single_extent:
2236 * This routine returns max. credits that needed to insert an extent
2237 * to the extent tree.
2238 * When pass the actual path, the caller should calculate credits
2241 int ext4_ext_calc_credits_for_single_extent(struct inode
*inode
, int nrblocks
,
2242 struct ext4_ext_path
*path
)
2245 int depth
= ext_depth(inode
);
2248 /* probably there is space in leaf? */
2249 if (le16_to_cpu(path
[depth
].p_hdr
->eh_entries
)
2250 < le16_to_cpu(path
[depth
].p_hdr
->eh_max
)) {
2253 * There are some space in the leaf tree, no
2254 * need to account for leaf block credit
2256 * bitmaps and block group descriptor blocks
2257 * and other metadata blocks still need to be
2260 /* 1 bitmap, 1 block group descriptor */
2261 ret
= 2 + EXT4_META_TRANS_BLOCKS(inode
->i_sb
);
2266 return ext4_chunk_trans_blocks(inode
, nrblocks
);
2270 * How many index/leaf blocks need to change/allocate to modify nrblocks?
2272 * if nrblocks are fit in a single extent (chunk flag is 1), then
2273 * in the worse case, each tree level index/leaf need to be changed
2274 * if the tree split due to insert a new extent, then the old tree
2275 * index/leaf need to be updated too
2277 * If the nrblocks are discontiguous, they could cause
2278 * the whole tree split more than once, but this is really rare.
2280 int ext4_ext_index_trans_blocks(struct inode
*inode
, int nrblocks
, int chunk
)
2285 /* If we are converting the inline data, only one is needed here. */
2286 if (ext4_has_inline_data(inode
))
2289 depth
= ext_depth(inode
);
2299 static int ext4_remove_blocks(handle_t
*handle
, struct inode
*inode
,
2300 struct ext4_extent
*ex
,
2301 ext4_fsblk_t
*partial_cluster
,
2302 ext4_lblk_t from
, ext4_lblk_t to
)
2304 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
2305 unsigned short ee_len
= ext4_ext_get_actual_len(ex
);
2309 if (S_ISDIR(inode
->i_mode
) || S_ISLNK(inode
->i_mode
))
2310 flags
|= EXT4_FREE_BLOCKS_METADATA
| EXT4_FREE_BLOCKS_FORGET
;
2311 else if (ext4_should_journal_data(inode
))
2312 flags
|= EXT4_FREE_BLOCKS_FORGET
;
2315 * For bigalloc file systems, we never free a partial cluster
2316 * at the beginning of the extent. Instead, we make a note
2317 * that we tried freeing the cluster, and check to see if we
2318 * need to free it on a subsequent call to ext4_remove_blocks,
2319 * or at the end of the ext4_truncate() operation.
2321 flags
|= EXT4_FREE_BLOCKS_NOFREE_FIRST_CLUSTER
;
2323 trace_ext4_remove_blocks(inode
, ex
, from
, to
, *partial_cluster
);
2325 * If we have a partial cluster, and it's different from the
2326 * cluster of the last block, we need to explicitly free the
2327 * partial cluster here.
2329 pblk
= ext4_ext_pblock(ex
) + ee_len
- 1;
2330 if (*partial_cluster
&& (EXT4_B2C(sbi
, pblk
) != *partial_cluster
)) {
2331 ext4_free_blocks(handle
, inode
, NULL
,
2332 EXT4_C2B(sbi
, *partial_cluster
),
2333 sbi
->s_cluster_ratio
, flags
);
2334 *partial_cluster
= 0;
2337 #ifdef EXTENTS_STATS
2339 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
2340 spin_lock(&sbi
->s_ext_stats_lock
);
2341 sbi
->s_ext_blocks
+= ee_len
;
2342 sbi
->s_ext_extents
++;
2343 if (ee_len
< sbi
->s_ext_min
)
2344 sbi
->s_ext_min
= ee_len
;
2345 if (ee_len
> sbi
->s_ext_max
)
2346 sbi
->s_ext_max
= ee_len
;
2347 if (ext_depth(inode
) > sbi
->s_depth_max
)
2348 sbi
->s_depth_max
= ext_depth(inode
);
2349 spin_unlock(&sbi
->s_ext_stats_lock
);
2352 if (from
>= le32_to_cpu(ex
->ee_block
)
2353 && to
== le32_to_cpu(ex
->ee_block
) + ee_len
- 1) {
2357 num
= le32_to_cpu(ex
->ee_block
) + ee_len
- from
;
2358 pblk
= ext4_ext_pblock(ex
) + ee_len
- num
;
2359 ext_debug("free last %u blocks starting %llu\n", num
, pblk
);
2360 ext4_free_blocks(handle
, inode
, NULL
, pblk
, num
, flags
);
2362 * If the block range to be freed didn't start at the
2363 * beginning of a cluster, and we removed the entire
2364 * extent, save the partial cluster here, since we
2365 * might need to delete if we determine that the
2366 * truncate operation has removed all of the blocks in
2369 if (pblk
& (sbi
->s_cluster_ratio
- 1) &&
2371 *partial_cluster
= EXT4_B2C(sbi
, pblk
);
2373 *partial_cluster
= 0;
2374 } else if (from
== le32_to_cpu(ex
->ee_block
)
2375 && to
<= le32_to_cpu(ex
->ee_block
) + ee_len
- 1) {
2381 start
= ext4_ext_pblock(ex
);
2383 ext_debug("free first %u blocks starting %llu\n", num
, start
);
2384 ext4_free_blocks(handle
, inode
, NULL
, start
, num
, flags
);
2387 printk(KERN_INFO
"strange request: removal(2) "
2388 "%u-%u from %u:%u\n",
2389 from
, to
, le32_to_cpu(ex
->ee_block
), ee_len
);
2396 * ext4_ext_rm_leaf() Removes the extents associated with the
2397 * blocks appearing between "start" and "end", and splits the extents
2398 * if "start" and "end" appear in the same extent
2400 * @handle: The journal handle
2401 * @inode: The files inode
2402 * @path: The path to the leaf
2403 * @start: The first block to remove
2404 * @end: The last block to remove
2407 ext4_ext_rm_leaf(handle_t
*handle
, struct inode
*inode
,
2408 struct ext4_ext_path
*path
, ext4_fsblk_t
*partial_cluster
,
2409 ext4_lblk_t start
, ext4_lblk_t end
)
2411 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
2412 int err
= 0, correct_index
= 0;
2413 int depth
= ext_depth(inode
), credits
;
2414 struct ext4_extent_header
*eh
;
2417 ext4_lblk_t ex_ee_block
;
2418 unsigned short ex_ee_len
;
2419 unsigned uninitialized
= 0;
2420 struct ext4_extent
*ex
;
2422 /* the header must be checked already in ext4_ext_remove_space() */
2423 ext_debug("truncate since %u in leaf to %u\n", start
, end
);
2424 if (!path
[depth
].p_hdr
)
2425 path
[depth
].p_hdr
= ext_block_hdr(path
[depth
].p_bh
);
2426 eh
= path
[depth
].p_hdr
;
2427 if (unlikely(path
[depth
].p_hdr
== NULL
)) {
2428 EXT4_ERROR_INODE(inode
, "path[%d].p_hdr == NULL", depth
);
2431 /* find where to start removing */
2432 ex
= EXT_LAST_EXTENT(eh
);
2434 ex_ee_block
= le32_to_cpu(ex
->ee_block
);
2435 ex_ee_len
= ext4_ext_get_actual_len(ex
);
2437 trace_ext4_ext_rm_leaf(inode
, start
, ex
, *partial_cluster
);
2439 while (ex
>= EXT_FIRST_EXTENT(eh
) &&
2440 ex_ee_block
+ ex_ee_len
> start
) {
2442 if (ext4_ext_is_uninitialized(ex
))
2447 ext_debug("remove ext %u:[%d]%d\n", ex_ee_block
,
2448 uninitialized
, ex_ee_len
);
2449 path
[depth
].p_ext
= ex
;
2451 a
= ex_ee_block
> start
? ex_ee_block
: start
;
2452 b
= ex_ee_block
+ex_ee_len
- 1 < end
?
2453 ex_ee_block
+ex_ee_len
- 1 : end
;
2455 ext_debug(" border %u:%u\n", a
, b
);
2457 /* If this extent is beyond the end of the hole, skip it */
2458 if (end
< ex_ee_block
) {
2460 ex_ee_block
= le32_to_cpu(ex
->ee_block
);
2461 ex_ee_len
= ext4_ext_get_actual_len(ex
);
2463 } else if (b
!= ex_ee_block
+ ex_ee_len
- 1) {
2464 EXT4_ERROR_INODE(inode
,
2465 "can not handle truncate %u:%u "
2467 start
, end
, ex_ee_block
,
2468 ex_ee_block
+ ex_ee_len
- 1);
2471 } else if (a
!= ex_ee_block
) {
2472 /* remove tail of the extent */
2473 num
= a
- ex_ee_block
;
2475 /* remove whole extent: excellent! */
2479 * 3 for leaf, sb, and inode plus 2 (bmap and group
2480 * descriptor) for each block group; assume two block
2481 * groups plus ex_ee_len/blocks_per_block_group for
2484 credits
= 7 + 2*(ex_ee_len
/EXT4_BLOCKS_PER_GROUP(inode
->i_sb
));
2485 if (ex
== EXT_FIRST_EXTENT(eh
)) {
2487 credits
+= (ext_depth(inode
)) + 1;
2489 credits
+= EXT4_MAXQUOTAS_TRANS_BLOCKS(inode
->i_sb
);
2491 err
= ext4_ext_truncate_extend_restart(handle
, inode
, credits
);
2495 err
= ext4_ext_get_access(handle
, inode
, path
+ depth
);
2499 err
= ext4_remove_blocks(handle
, inode
, ex
, partial_cluster
,
2505 /* this extent is removed; mark slot entirely unused */
2506 ext4_ext_store_pblock(ex
, 0);
2508 ex
->ee_len
= cpu_to_le16(num
);
2510 * Do not mark uninitialized if all the blocks in the
2511 * extent have been removed.
2513 if (uninitialized
&& num
)
2514 ext4_ext_mark_uninitialized(ex
);
2516 * If the extent was completely released,
2517 * we need to remove it from the leaf
2520 if (end
!= EXT_MAX_BLOCKS
- 1) {
2522 * For hole punching, we need to scoot all the
2523 * extents up when an extent is removed so that
2524 * we dont have blank extents in the middle
2526 memmove(ex
, ex
+1, (EXT_LAST_EXTENT(eh
) - ex
) *
2527 sizeof(struct ext4_extent
));
2529 /* Now get rid of the one at the end */
2530 memset(EXT_LAST_EXTENT(eh
), 0,
2531 sizeof(struct ext4_extent
));
2533 le16_add_cpu(&eh
->eh_entries
, -1);
2535 *partial_cluster
= 0;
2537 err
= ext4_ext_dirty(handle
, inode
, path
+ depth
);
2541 ext_debug("new extent: %u:%u:%llu\n", ex_ee_block
, num
,
2542 ext4_ext_pblock(ex
));
2544 ex_ee_block
= le32_to_cpu(ex
->ee_block
);
2545 ex_ee_len
= ext4_ext_get_actual_len(ex
);
2548 if (correct_index
&& eh
->eh_entries
)
2549 err
= ext4_ext_correct_indexes(handle
, inode
, path
);
2552 * If there is still a entry in the leaf node, check to see if
2553 * it references the partial cluster. This is the only place
2554 * where it could; if it doesn't, we can free the cluster.
2556 if (*partial_cluster
&& ex
>= EXT_FIRST_EXTENT(eh
) &&
2557 (EXT4_B2C(sbi
, ext4_ext_pblock(ex
) + ex_ee_len
- 1) !=
2558 *partial_cluster
)) {
2559 int flags
= EXT4_FREE_BLOCKS_FORGET
;
2561 if (S_ISDIR(inode
->i_mode
) || S_ISLNK(inode
->i_mode
))
2562 flags
|= EXT4_FREE_BLOCKS_METADATA
;
2564 ext4_free_blocks(handle
, inode
, NULL
,
2565 EXT4_C2B(sbi
, *partial_cluster
),
2566 sbi
->s_cluster_ratio
, flags
);
2567 *partial_cluster
= 0;
2570 /* if this leaf is free, then we should
2571 * remove it from index block above */
2572 if (err
== 0 && eh
->eh_entries
== 0 && path
[depth
].p_bh
!= NULL
)
2573 err
= ext4_ext_rm_idx(handle
, inode
, path
, depth
);
2580 * ext4_ext_more_to_rm:
2581 * returns 1 if current index has to be freed (even partial)
2584 ext4_ext_more_to_rm(struct ext4_ext_path
*path
)
2586 BUG_ON(path
->p_idx
== NULL
);
2588 if (path
->p_idx
< EXT_FIRST_INDEX(path
->p_hdr
))
2592 * if truncate on deeper level happened, it wasn't partial,
2593 * so we have to consider current index for truncation
2595 if (le16_to_cpu(path
->p_hdr
->eh_entries
) == path
->p_block
)
2600 static int ext4_ext_remove_space(struct inode
*inode
, ext4_lblk_t start
,
2603 struct super_block
*sb
= inode
->i_sb
;
2604 int depth
= ext_depth(inode
);
2605 struct ext4_ext_path
*path
= NULL
;
2606 ext4_fsblk_t partial_cluster
= 0;
2610 ext_debug("truncate since %u to %u\n", start
, end
);
2612 /* probably first extent we're gonna free will be last in block */
2613 handle
= ext4_journal_start(inode
, EXT4_HT_TRUNCATE
, depth
+ 1);
2615 return PTR_ERR(handle
);
2618 trace_ext4_ext_remove_space(inode
, start
, depth
);
2621 * Check if we are removing extents inside the extent tree. If that
2622 * is the case, we are going to punch a hole inside the extent tree
2623 * so we have to check whether we need to split the extent covering
2624 * the last block to remove so we can easily remove the part of it
2625 * in ext4_ext_rm_leaf().
2627 if (end
< EXT_MAX_BLOCKS
- 1) {
2628 struct ext4_extent
*ex
;
2629 ext4_lblk_t ee_block
;
2631 /* find extent for this block */
2632 path
= ext4_ext_find_extent(inode
, end
, NULL
);
2634 ext4_journal_stop(handle
);
2635 return PTR_ERR(path
);
2637 depth
= ext_depth(inode
);
2638 /* Leaf not may not exist only if inode has no blocks at all */
2639 ex
= path
[depth
].p_ext
;
2642 EXT4_ERROR_INODE(inode
,
2643 "path[%d].p_hdr == NULL",
2650 ee_block
= le32_to_cpu(ex
->ee_block
);
2653 * See if the last block is inside the extent, if so split
2654 * the extent at 'end' block so we can easily remove the
2655 * tail of the first part of the split extent in
2656 * ext4_ext_rm_leaf().
2658 if (end
>= ee_block
&&
2659 end
< ee_block
+ ext4_ext_get_actual_len(ex
) - 1) {
2662 if (ext4_ext_is_uninitialized(ex
))
2663 split_flag
= EXT4_EXT_MARK_UNINIT1
|
2664 EXT4_EXT_MARK_UNINIT2
;
2667 * Split the extent in two so that 'end' is the last
2668 * block in the first new extent
2670 err
= ext4_split_extent_at(handle
, inode
, path
,
2671 end
+ 1, split_flag
,
2672 EXT4_GET_BLOCKS_PRE_IO
|
2673 EXT4_GET_BLOCKS_PUNCH_OUT_EXT
);
2680 * We start scanning from right side, freeing all the blocks
2681 * after i_size and walking into the tree depth-wise.
2683 depth
= ext_depth(inode
);
2688 le16_to_cpu(path
[k
].p_hdr
->eh_entries
)+1;
2690 path
= kzalloc(sizeof(struct ext4_ext_path
) * (depth
+ 1),
2693 ext4_journal_stop(handle
);
2696 path
[0].p_depth
= depth
;
2697 path
[0].p_hdr
= ext_inode_hdr(inode
);
2700 if (ext4_ext_check(inode
, path
[0].p_hdr
, depth
)) {
2707 while (i
>= 0 && err
== 0) {
2709 /* this is leaf block */
2710 err
= ext4_ext_rm_leaf(handle
, inode
, path
,
2711 &partial_cluster
, start
,
2713 /* root level has p_bh == NULL, brelse() eats this */
2714 brelse(path
[i
].p_bh
);
2715 path
[i
].p_bh
= NULL
;
2720 /* this is index block */
2721 if (!path
[i
].p_hdr
) {
2722 ext_debug("initialize header\n");
2723 path
[i
].p_hdr
= ext_block_hdr(path
[i
].p_bh
);
2726 if (!path
[i
].p_idx
) {
2727 /* this level hasn't been touched yet */
2728 path
[i
].p_idx
= EXT_LAST_INDEX(path
[i
].p_hdr
);
2729 path
[i
].p_block
= le16_to_cpu(path
[i
].p_hdr
->eh_entries
)+1;
2730 ext_debug("init index ptr: hdr 0x%p, num %d\n",
2732 le16_to_cpu(path
[i
].p_hdr
->eh_entries
));
2734 /* we were already here, see at next index */
2738 ext_debug("level %d - index, first 0x%p, cur 0x%p\n",
2739 i
, EXT_FIRST_INDEX(path
[i
].p_hdr
),
2741 if (ext4_ext_more_to_rm(path
+ i
)) {
2742 struct buffer_head
*bh
;
2743 /* go to the next level */
2744 ext_debug("move to level %d (block %llu)\n",
2745 i
+ 1, ext4_idx_pblock(path
[i
].p_idx
));
2746 memset(path
+ i
+ 1, 0, sizeof(*path
));
2747 bh
= sb_bread(sb
, ext4_idx_pblock(path
[i
].p_idx
));
2749 /* should we reset i_size? */
2753 if (WARN_ON(i
+ 1 > depth
)) {
2757 if (ext4_ext_check_block(inode
, ext_block_hdr(bh
),
2758 depth
- i
- 1, bh
)) {
2762 path
[i
+ 1].p_bh
= bh
;
2764 /* save actual number of indexes since this
2765 * number is changed at the next iteration */
2766 path
[i
].p_block
= le16_to_cpu(path
[i
].p_hdr
->eh_entries
);
2769 /* we finished processing this index, go up */
2770 if (path
[i
].p_hdr
->eh_entries
== 0 && i
> 0) {
2771 /* index is empty, remove it;
2772 * handle must be already prepared by the
2773 * truncatei_leaf() */
2774 err
= ext4_ext_rm_idx(handle
, inode
, path
, i
);
2776 /* root level has p_bh == NULL, brelse() eats this */
2777 brelse(path
[i
].p_bh
);
2778 path
[i
].p_bh
= NULL
;
2780 ext_debug("return to level %d\n", i
);
2784 trace_ext4_ext_remove_space_done(inode
, start
, depth
, partial_cluster
,
2785 path
->p_hdr
->eh_entries
);
2787 /* If we still have something in the partial cluster and we have removed
2788 * even the first extent, then we should free the blocks in the partial
2789 * cluster as well. */
2790 if (partial_cluster
&& path
->p_hdr
->eh_entries
== 0) {
2791 int flags
= EXT4_FREE_BLOCKS_FORGET
;
2793 if (S_ISDIR(inode
->i_mode
) || S_ISLNK(inode
->i_mode
))
2794 flags
|= EXT4_FREE_BLOCKS_METADATA
;
2796 ext4_free_blocks(handle
, inode
, NULL
,
2797 EXT4_C2B(EXT4_SB(sb
), partial_cluster
),
2798 EXT4_SB(sb
)->s_cluster_ratio
, flags
);
2799 partial_cluster
= 0;
2802 /* TODO: flexible tree reduction should be here */
2803 if (path
->p_hdr
->eh_entries
== 0) {
2805 * truncate to zero freed all the tree,
2806 * so we need to correct eh_depth
2808 err
= ext4_ext_get_access(handle
, inode
, path
);
2810 ext_inode_hdr(inode
)->eh_depth
= 0;
2811 ext_inode_hdr(inode
)->eh_max
=
2812 cpu_to_le16(ext4_ext_space_root(inode
, 0));
2813 err
= ext4_ext_dirty(handle
, inode
, path
);
2817 ext4_ext_drop_refs(path
);
2819 if (err
== -EAGAIN
) {
2823 ext4_journal_stop(handle
);
2829 * called at mount time
2831 void ext4_ext_init(struct super_block
*sb
)
2834 * possible initialization would be here
2837 if (EXT4_HAS_INCOMPAT_FEATURE(sb
, EXT4_FEATURE_INCOMPAT_EXTENTS
)) {
2838 #if defined(AGGRESSIVE_TEST) || defined(CHECK_BINSEARCH) || defined(EXTENTS_STATS)
2839 printk(KERN_INFO
"EXT4-fs: file extents enabled"
2840 #ifdef AGGRESSIVE_TEST
2841 ", aggressive tests"
2843 #ifdef CHECK_BINSEARCH
2846 #ifdef EXTENTS_STATS
2851 #ifdef EXTENTS_STATS
2852 spin_lock_init(&EXT4_SB(sb
)->s_ext_stats_lock
);
2853 EXT4_SB(sb
)->s_ext_min
= 1 << 30;
2854 EXT4_SB(sb
)->s_ext_max
= 0;
2860 * called at umount time
2862 void ext4_ext_release(struct super_block
*sb
)
2864 if (!EXT4_HAS_INCOMPAT_FEATURE(sb
, EXT4_FEATURE_INCOMPAT_EXTENTS
))
2867 #ifdef EXTENTS_STATS
2868 if (EXT4_SB(sb
)->s_ext_blocks
&& EXT4_SB(sb
)->s_ext_extents
) {
2869 struct ext4_sb_info
*sbi
= EXT4_SB(sb
);
2870 printk(KERN_ERR
"EXT4-fs: %lu blocks in %lu extents (%lu ave)\n",
2871 sbi
->s_ext_blocks
, sbi
->s_ext_extents
,
2872 sbi
->s_ext_blocks
/ sbi
->s_ext_extents
);
2873 printk(KERN_ERR
"EXT4-fs: extents: %lu min, %lu max, max depth %lu\n",
2874 sbi
->s_ext_min
, sbi
->s_ext_max
, sbi
->s_depth_max
);
2879 /* FIXME!! we need to try to merge to left or right after zero-out */
2880 static int ext4_ext_zeroout(struct inode
*inode
, struct ext4_extent
*ex
)
2882 ext4_fsblk_t ee_pblock
;
2883 unsigned int ee_len
;
2886 ee_len
= ext4_ext_get_actual_len(ex
);
2887 ee_pblock
= ext4_ext_pblock(ex
);
2889 ret
= sb_issue_zeroout(inode
->i_sb
, ee_pblock
, ee_len
, GFP_NOFS
);
2897 * ext4_split_extent_at() splits an extent at given block.
2899 * @handle: the journal handle
2900 * @inode: the file inode
2901 * @path: the path to the extent
2902 * @split: the logical block where the extent is splitted.
2903 * @split_flags: indicates if the extent could be zeroout if split fails, and
2904 * the states(init or uninit) of new extents.
2905 * @flags: flags used to insert new extent to extent tree.
2908 * Splits extent [a, b] into two extents [a, @split) and [@split, b], states
2909 * of which are deterimined by split_flag.
2911 * There are two cases:
2912 * a> the extent are splitted into two extent.
2913 * b> split is not needed, and just mark the extent.
2915 * return 0 on success.
2917 static int ext4_split_extent_at(handle_t
*handle
,
2918 struct inode
*inode
,
2919 struct ext4_ext_path
*path
,
2924 ext4_fsblk_t newblock
;
2925 ext4_lblk_t ee_block
;
2926 struct ext4_extent
*ex
, newex
, orig_ex
;
2927 struct ext4_extent
*ex2
= NULL
;
2928 unsigned int ee_len
, depth
;
2931 BUG_ON((split_flag
& (EXT4_EXT_DATA_VALID1
| EXT4_EXT_DATA_VALID2
)) ==
2932 (EXT4_EXT_DATA_VALID1
| EXT4_EXT_DATA_VALID2
));
2934 ext_debug("ext4_split_extents_at: inode %lu, logical"
2935 "block %llu\n", inode
->i_ino
, (unsigned long long)split
);
2937 ext4_ext_show_leaf(inode
, path
);
2939 depth
= ext_depth(inode
);
2940 ex
= path
[depth
].p_ext
;
2941 ee_block
= le32_to_cpu(ex
->ee_block
);
2942 ee_len
= ext4_ext_get_actual_len(ex
);
2943 newblock
= split
- ee_block
+ ext4_ext_pblock(ex
);
2945 BUG_ON(split
< ee_block
|| split
>= (ee_block
+ ee_len
));
2947 err
= ext4_ext_get_access(handle
, inode
, path
+ depth
);
2951 if (split
== ee_block
) {
2953 * case b: block @split is the block that the extent begins with
2954 * then we just change the state of the extent, and splitting
2957 if (split_flag
& EXT4_EXT_MARK_UNINIT2
)
2958 ext4_ext_mark_uninitialized(ex
);
2960 ext4_ext_mark_initialized(ex
);
2962 if (!(flags
& EXT4_GET_BLOCKS_PRE_IO
))
2963 ext4_ext_try_to_merge(handle
, inode
, path
, ex
);
2965 err
= ext4_ext_dirty(handle
, inode
, path
+ path
->p_depth
);
2970 memcpy(&orig_ex
, ex
, sizeof(orig_ex
));
2971 ex
->ee_len
= cpu_to_le16(split
- ee_block
);
2972 if (split_flag
& EXT4_EXT_MARK_UNINIT1
)
2973 ext4_ext_mark_uninitialized(ex
);
2976 * path may lead to new leaf, not to original leaf any more
2977 * after ext4_ext_insert_extent() returns,
2979 err
= ext4_ext_dirty(handle
, inode
, path
+ depth
);
2981 goto fix_extent_len
;
2984 ex2
->ee_block
= cpu_to_le32(split
);
2985 ex2
->ee_len
= cpu_to_le16(ee_len
- (split
- ee_block
));
2986 ext4_ext_store_pblock(ex2
, newblock
);
2987 if (split_flag
& EXT4_EXT_MARK_UNINIT2
)
2988 ext4_ext_mark_uninitialized(ex2
);
2990 err
= ext4_ext_insert_extent(handle
, inode
, path
, &newex
, flags
);
2991 if (err
== -ENOSPC
&& (EXT4_EXT_MAY_ZEROOUT
& split_flag
)) {
2992 if (split_flag
& (EXT4_EXT_DATA_VALID1
|EXT4_EXT_DATA_VALID2
)) {
2993 if (split_flag
& EXT4_EXT_DATA_VALID1
)
2994 err
= ext4_ext_zeroout(inode
, ex2
);
2996 err
= ext4_ext_zeroout(inode
, ex
);
2998 err
= ext4_ext_zeroout(inode
, &orig_ex
);
3001 goto fix_extent_len
;
3002 /* update the extent length and mark as initialized */
3003 ex
->ee_len
= cpu_to_le16(ee_len
);
3004 ext4_ext_try_to_merge(handle
, inode
, path
, ex
);
3005 err
= ext4_ext_dirty(handle
, inode
, path
+ path
->p_depth
);
3008 goto fix_extent_len
;
3011 ext4_ext_show_leaf(inode
, path
);
3015 ex
->ee_len
= orig_ex
.ee_len
;
3016 ext4_ext_dirty(handle
, inode
, path
+ depth
);
3021 * ext4_split_extents() splits an extent and mark extent which is covered
3022 * by @map as split_flags indicates
3024 * It may result in splitting the extent into multiple extents (upto three)
3025 * There are three possibilities:
3026 * a> There is no split required
3027 * b> Splits in two extents: Split is happening at either end of the extent
3028 * c> Splits in three extents: Somone is splitting in middle of the extent
3031 static int ext4_split_extent(handle_t
*handle
,
3032 struct inode
*inode
,
3033 struct ext4_ext_path
*path
,
3034 struct ext4_map_blocks
*map
,
3038 ext4_lblk_t ee_block
;
3039 struct ext4_extent
*ex
;
3040 unsigned int ee_len
, depth
;
3043 int split_flag1
, flags1
;
3045 depth
= ext_depth(inode
);
3046 ex
= path
[depth
].p_ext
;
3047 ee_block
= le32_to_cpu(ex
->ee_block
);
3048 ee_len
= ext4_ext_get_actual_len(ex
);
3049 uninitialized
= ext4_ext_is_uninitialized(ex
);
3051 if (map
->m_lblk
+ map
->m_len
< ee_block
+ ee_len
) {
3052 split_flag1
= split_flag
& EXT4_EXT_MAY_ZEROOUT
;
3053 flags1
= flags
| EXT4_GET_BLOCKS_PRE_IO
;
3055 split_flag1
|= EXT4_EXT_MARK_UNINIT1
|
3056 EXT4_EXT_MARK_UNINIT2
;
3057 if (split_flag
& EXT4_EXT_DATA_VALID2
)
3058 split_flag1
|= EXT4_EXT_DATA_VALID1
;
3059 err
= ext4_split_extent_at(handle
, inode
, path
,
3060 map
->m_lblk
+ map
->m_len
, split_flag1
, flags1
);
3065 ext4_ext_drop_refs(path
);
3066 path
= ext4_ext_find_extent(inode
, map
->m_lblk
, path
);
3068 return PTR_ERR(path
);
3070 if (map
->m_lblk
>= ee_block
) {
3071 split_flag1
= split_flag
& (EXT4_EXT_MAY_ZEROOUT
|
3072 EXT4_EXT_DATA_VALID2
);
3074 split_flag1
|= EXT4_EXT_MARK_UNINIT1
;
3075 if (split_flag
& EXT4_EXT_MARK_UNINIT2
)
3076 split_flag1
|= EXT4_EXT_MARK_UNINIT2
;
3077 err
= ext4_split_extent_at(handle
, inode
, path
,
3078 map
->m_lblk
, split_flag1
, flags
);
3083 ext4_ext_show_leaf(inode
, path
);
3085 return err
? err
: map
->m_len
;
3089 * This function is called by ext4_ext_map_blocks() if someone tries to write
3090 * to an uninitialized extent. It may result in splitting the uninitialized
3091 * extent into multiple extents (up to three - one initialized and two
3093 * There are three possibilities:
3094 * a> There is no split required: Entire extent should be initialized
3095 * b> Splits in two extents: Write is happening at either end of the extent
3096 * c> Splits in three extents: Somone is writing in middle of the extent
3099 * - The extent pointed to by 'path' is uninitialized.
3100 * - The extent pointed to by 'path' contains a superset
3101 * of the logical span [map->m_lblk, map->m_lblk + map->m_len).
3103 * Post-conditions on success:
3104 * - the returned value is the number of blocks beyond map->l_lblk
3105 * that are allocated and initialized.
3106 * It is guaranteed to be >= map->m_len.
3108 static int ext4_ext_convert_to_initialized(handle_t
*handle
,
3109 struct inode
*inode
,
3110 struct ext4_map_blocks
*map
,
3111 struct ext4_ext_path
*path
)
3113 struct ext4_sb_info
*sbi
;
3114 struct ext4_extent_header
*eh
;
3115 struct ext4_map_blocks split_map
;
3116 struct ext4_extent zero_ex
;
3117 struct ext4_extent
*ex
;
3118 ext4_lblk_t ee_block
, eof_block
;
3119 unsigned int ee_len
, depth
;
3120 int allocated
, max_zeroout
= 0;
3124 ext_debug("ext4_ext_convert_to_initialized: inode %lu, logical"
3125 "block %llu, max_blocks %u\n", inode
->i_ino
,
3126 (unsigned long long)map
->m_lblk
, map
->m_len
);
3128 sbi
= EXT4_SB(inode
->i_sb
);
3129 eof_block
= (inode
->i_size
+ inode
->i_sb
->s_blocksize
- 1) >>
3130 inode
->i_sb
->s_blocksize_bits
;
3131 if (eof_block
< map
->m_lblk
+ map
->m_len
)
3132 eof_block
= map
->m_lblk
+ map
->m_len
;
3134 depth
= ext_depth(inode
);
3135 eh
= path
[depth
].p_hdr
;
3136 ex
= path
[depth
].p_ext
;
3137 ee_block
= le32_to_cpu(ex
->ee_block
);
3138 ee_len
= ext4_ext_get_actual_len(ex
);
3139 allocated
= ee_len
- (map
->m_lblk
- ee_block
);
3141 trace_ext4_ext_convert_to_initialized_enter(inode
, map
, ex
);
3143 /* Pre-conditions */
3144 BUG_ON(!ext4_ext_is_uninitialized(ex
));
3145 BUG_ON(!in_range(map
->m_lblk
, ee_block
, ee_len
));
3148 * Attempt to transfer newly initialized blocks from the currently
3149 * uninitialized extent to its left neighbor. This is much cheaper
3150 * than an insertion followed by a merge as those involve costly
3151 * memmove() calls. This is the common case in steady state for
3152 * workloads doing fallocate(FALLOC_FL_KEEP_SIZE) followed by append
3155 * Limitations of the current logic:
3156 * - L1: we only deal with writes at the start of the extent.
3157 * The approach could be extended to writes at the end
3158 * of the extent but this scenario was deemed less common.
3159 * - L2: we do not deal with writes covering the whole extent.
3160 * This would require removing the extent if the transfer
3162 * - L3: we only attempt to merge with an extent stored in the
3163 * same extent tree node.
3165 if ((map
->m_lblk
== ee_block
) && /*L1*/
3166 (map
->m_len
< ee_len
) && /*L2*/
3167 (ex
> EXT_FIRST_EXTENT(eh
))) { /*L3*/
3168 struct ext4_extent
*prev_ex
;
3169 ext4_lblk_t prev_lblk
;
3170 ext4_fsblk_t prev_pblk
, ee_pblk
;
3171 unsigned int prev_len
, write_len
;
3174 prev_lblk
= le32_to_cpu(prev_ex
->ee_block
);
3175 prev_len
= ext4_ext_get_actual_len(prev_ex
);
3176 prev_pblk
= ext4_ext_pblock(prev_ex
);
3177 ee_pblk
= ext4_ext_pblock(ex
);
3178 write_len
= map
->m_len
;
3181 * A transfer of blocks from 'ex' to 'prev_ex' is allowed
3182 * upon those conditions:
3183 * - C1: prev_ex is initialized,
3184 * - C2: prev_ex is logically abutting ex,
3185 * - C3: prev_ex is physically abutting ex,
3186 * - C4: prev_ex can receive the additional blocks without
3187 * overflowing the (initialized) length limit.
3189 if ((!ext4_ext_is_uninitialized(prev_ex
)) && /*C1*/
3190 ((prev_lblk
+ prev_len
) == ee_block
) && /*C2*/
3191 ((prev_pblk
+ prev_len
) == ee_pblk
) && /*C3*/
3192 (prev_len
< (EXT_INIT_MAX_LEN
- write_len
))) { /*C4*/
3193 err
= ext4_ext_get_access(handle
, inode
, path
+ depth
);
3197 trace_ext4_ext_convert_to_initialized_fastpath(inode
,
3200 /* Shift the start of ex by 'write_len' blocks */
3201 ex
->ee_block
= cpu_to_le32(ee_block
+ write_len
);
3202 ext4_ext_store_pblock(ex
, ee_pblk
+ write_len
);
3203 ex
->ee_len
= cpu_to_le16(ee_len
- write_len
);
3204 ext4_ext_mark_uninitialized(ex
); /* Restore the flag */
3206 /* Extend prev_ex by 'write_len' blocks */
3207 prev_ex
->ee_len
= cpu_to_le16(prev_len
+ write_len
);
3209 /* Mark the block containing both extents as dirty */
3210 ext4_ext_dirty(handle
, inode
, path
+ depth
);
3212 /* Update path to point to the right extent */
3213 path
[depth
].p_ext
= prev_ex
;
3215 /* Result: number of initialized blocks past m_lblk */
3216 allocated
= write_len
;
3221 WARN_ON(map
->m_lblk
< ee_block
);
3223 * It is safe to convert extent to initialized via explicit
3224 * zeroout only if extent is fully insde i_size or new_size.
3226 split_flag
|= ee_block
+ ee_len
<= eof_block
? EXT4_EXT_MAY_ZEROOUT
: 0;
3228 if (EXT4_EXT_MAY_ZEROOUT
& split_flag
)
3229 max_zeroout
= sbi
->s_extent_max_zeroout_kb
>>
3230 inode
->i_sb
->s_blocksize_bits
;
3232 /* If extent is less than s_max_zeroout_kb, zeroout directly */
3233 if (max_zeroout
&& (ee_len
<= max_zeroout
)) {
3234 err
= ext4_ext_zeroout(inode
, ex
);
3238 err
= ext4_ext_get_access(handle
, inode
, path
+ depth
);
3241 ext4_ext_mark_initialized(ex
);
3242 ext4_ext_try_to_merge(handle
, inode
, path
, ex
);
3243 err
= ext4_ext_dirty(handle
, inode
, path
+ path
->p_depth
);
3249 * 1. split the extent into three extents.
3250 * 2. split the extent into two extents, zeroout the first half.
3251 * 3. split the extent into two extents, zeroout the second half.
3252 * 4. split the extent into two extents with out zeroout.
3254 split_map
.m_lblk
= map
->m_lblk
;
3255 split_map
.m_len
= map
->m_len
;
3257 if (max_zeroout
&& (allocated
> map
->m_len
)) {
3258 if (allocated
<= max_zeroout
) {
3261 cpu_to_le32(map
->m_lblk
);
3262 zero_ex
.ee_len
= cpu_to_le16(allocated
);
3263 ext4_ext_store_pblock(&zero_ex
,
3264 ext4_ext_pblock(ex
) + map
->m_lblk
- ee_block
);
3265 err
= ext4_ext_zeroout(inode
, &zero_ex
);
3268 split_map
.m_lblk
= map
->m_lblk
;
3269 split_map
.m_len
= allocated
;
3270 } else if (map
->m_lblk
- ee_block
+ map
->m_len
< max_zeroout
) {
3272 if (map
->m_lblk
!= ee_block
) {
3273 zero_ex
.ee_block
= ex
->ee_block
;
3274 zero_ex
.ee_len
= cpu_to_le16(map
->m_lblk
-
3276 ext4_ext_store_pblock(&zero_ex
,
3277 ext4_ext_pblock(ex
));
3278 err
= ext4_ext_zeroout(inode
, &zero_ex
);
3283 split_map
.m_lblk
= ee_block
;
3284 split_map
.m_len
= map
->m_lblk
- ee_block
+ map
->m_len
;
3285 allocated
= map
->m_len
;
3289 allocated
= ext4_split_extent(handle
, inode
, path
,
3290 &split_map
, split_flag
, 0);
3295 return err
? err
: allocated
;
3299 * This function is called by ext4_ext_map_blocks() from
3300 * ext4_get_blocks_dio_write() when DIO to write
3301 * to an uninitialized extent.
3303 * Writing to an uninitialized extent may result in splitting the uninitialized
3304 * extent into multiple initialized/uninitialized extents (up to three)
3305 * There are three possibilities:
3306 * a> There is no split required: Entire extent should be uninitialized
3307 * b> Splits in two extents: Write is happening at either end of the extent
3308 * c> Splits in three extents: Somone is writing in middle of the extent
3310 * One of more index blocks maybe needed if the extent tree grow after
3311 * the uninitialized extent split. To prevent ENOSPC occur at the IO
3312 * complete, we need to split the uninitialized extent before DIO submit
3313 * the IO. The uninitialized extent called at this time will be split
3314 * into three uninitialized extent(at most). After IO complete, the part
3315 * being filled will be convert to initialized by the end_io callback function
3316 * via ext4_convert_unwritten_extents().
3318 * Returns the size of uninitialized extent to be written on success.
3320 static int ext4_split_unwritten_extents(handle_t
*handle
,
3321 struct inode
*inode
,
3322 struct ext4_map_blocks
*map
,
3323 struct ext4_ext_path
*path
,
3326 ext4_lblk_t eof_block
;
3327 ext4_lblk_t ee_block
;
3328 struct ext4_extent
*ex
;
3329 unsigned int ee_len
;
3330 int split_flag
= 0, depth
;
3332 ext_debug("ext4_split_unwritten_extents: inode %lu, logical"
3333 "block %llu, max_blocks %u\n", inode
->i_ino
,
3334 (unsigned long long)map
->m_lblk
, map
->m_len
);
3336 eof_block
= (inode
->i_size
+ inode
->i_sb
->s_blocksize
- 1) >>
3337 inode
->i_sb
->s_blocksize_bits
;
3338 if (eof_block
< map
->m_lblk
+ map
->m_len
)
3339 eof_block
= map
->m_lblk
+ map
->m_len
;
3341 * It is safe to convert extent to initialized via explicit
3342 * zeroout only if extent is fully insde i_size or new_size.
3344 depth
= ext_depth(inode
);
3345 ex
= path
[depth
].p_ext
;
3346 ee_block
= le32_to_cpu(ex
->ee_block
);
3347 ee_len
= ext4_ext_get_actual_len(ex
);
3349 split_flag
|= ee_block
+ ee_len
<= eof_block
? EXT4_EXT_MAY_ZEROOUT
: 0;
3350 split_flag
|= EXT4_EXT_MARK_UNINIT2
;
3351 if (flags
& EXT4_GET_BLOCKS_CONVERT
)
3352 split_flag
|= EXT4_EXT_DATA_VALID2
;
3353 flags
|= EXT4_GET_BLOCKS_PRE_IO
;
3354 return ext4_split_extent(handle
, inode
, path
, map
, split_flag
, flags
);
3357 static int ext4_convert_unwritten_extents_endio(handle_t
*handle
,
3358 struct inode
*inode
,
3359 struct ext4_map_blocks
*map
,
3360 struct ext4_ext_path
*path
)
3362 struct ext4_extent
*ex
;
3363 ext4_lblk_t ee_block
;
3364 unsigned int ee_len
;
3368 depth
= ext_depth(inode
);
3369 ex
= path
[depth
].p_ext
;
3370 ee_block
= le32_to_cpu(ex
->ee_block
);
3371 ee_len
= ext4_ext_get_actual_len(ex
);
3373 ext_debug("ext4_convert_unwritten_extents_endio: inode %lu, logical"
3374 "block %llu, max_blocks %u\n", inode
->i_ino
,
3375 (unsigned long long)ee_block
, ee_len
);
3377 /* If extent is larger than requested then split is required */
3378 if (ee_block
!= map
->m_lblk
|| ee_len
> map
->m_len
) {
3379 err
= ext4_split_unwritten_extents(handle
, inode
, map
, path
,
3380 EXT4_GET_BLOCKS_CONVERT
);
3383 ext4_ext_drop_refs(path
);
3384 path
= ext4_ext_find_extent(inode
, map
->m_lblk
, path
);
3386 err
= PTR_ERR(path
);
3389 depth
= ext_depth(inode
);
3390 ex
= path
[depth
].p_ext
;
3393 err
= ext4_ext_get_access(handle
, inode
, path
+ depth
);
3396 /* first mark the extent as initialized */
3397 ext4_ext_mark_initialized(ex
);
3399 /* note: ext4_ext_correct_indexes() isn't needed here because
3400 * borders are not changed
3402 ext4_ext_try_to_merge(handle
, inode
, path
, ex
);
3404 /* Mark modified extent as dirty */
3405 err
= ext4_ext_dirty(handle
, inode
, path
+ path
->p_depth
);
3407 ext4_ext_show_leaf(inode
, path
);
3411 static void unmap_underlying_metadata_blocks(struct block_device
*bdev
,
3412 sector_t block
, int count
)
3415 for (i
= 0; i
< count
; i
++)
3416 unmap_underlying_metadata(bdev
, block
+ i
);
3420 * Handle EOFBLOCKS_FL flag, clearing it if necessary
3422 static int check_eofblocks_fl(handle_t
*handle
, struct inode
*inode
,
3424 struct ext4_ext_path
*path
,
3428 struct ext4_extent_header
*eh
;
3429 struct ext4_extent
*last_ex
;
3431 if (!ext4_test_inode_flag(inode
, EXT4_INODE_EOFBLOCKS
))
3434 depth
= ext_depth(inode
);
3435 eh
= path
[depth
].p_hdr
;
3438 * We're going to remove EOFBLOCKS_FL entirely in future so we
3439 * do not care for this case anymore. Simply remove the flag
3440 * if there are no extents.
3442 if (unlikely(!eh
->eh_entries
))
3444 last_ex
= EXT_LAST_EXTENT(eh
);
3446 * We should clear the EOFBLOCKS_FL flag if we are writing the
3447 * last block in the last extent in the file. We test this by
3448 * first checking to see if the caller to
3449 * ext4_ext_get_blocks() was interested in the last block (or
3450 * a block beyond the last block) in the current extent. If
3451 * this turns out to be false, we can bail out from this
3452 * function immediately.
3454 if (lblk
+ len
< le32_to_cpu(last_ex
->ee_block
) +
3455 ext4_ext_get_actual_len(last_ex
))
3458 * If the caller does appear to be planning to write at or
3459 * beyond the end of the current extent, we then test to see
3460 * if the current extent is the last extent in the file, by
3461 * checking to make sure it was reached via the rightmost node
3462 * at each level of the tree.
3464 for (i
= depth
-1; i
>= 0; i
--)
3465 if (path
[i
].p_idx
!= EXT_LAST_INDEX(path
[i
].p_hdr
))
3468 ext4_clear_inode_flag(inode
, EXT4_INODE_EOFBLOCKS
);
3469 return ext4_mark_inode_dirty(handle
, inode
);
3473 * ext4_find_delalloc_range: find delayed allocated block in the given range.
3475 * Return 1 if there is a delalloc block in the range, otherwise 0.
3477 int ext4_find_delalloc_range(struct inode
*inode
,
3478 ext4_lblk_t lblk_start
,
3479 ext4_lblk_t lblk_end
)
3481 struct extent_status es
;
3483 ext4_es_find_delayed_extent(inode
, lblk_start
, &es
);
3485 return 0; /* there is no delay extent in this tree */
3486 else if (es
.es_lblk
<= lblk_start
&&
3487 lblk_start
< es
.es_lblk
+ es
.es_len
)
3489 else if (lblk_start
<= es
.es_lblk
&& es
.es_lblk
<= lblk_end
)
3495 int ext4_find_delalloc_cluster(struct inode
*inode
, ext4_lblk_t lblk
)
3497 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
3498 ext4_lblk_t lblk_start
, lblk_end
;
3499 lblk_start
= lblk
& (~(sbi
->s_cluster_ratio
- 1));
3500 lblk_end
= lblk_start
+ sbi
->s_cluster_ratio
- 1;
3502 return ext4_find_delalloc_range(inode
, lblk_start
, lblk_end
);
3506 * Determines how many complete clusters (out of those specified by the 'map')
3507 * are under delalloc and were reserved quota for.
3508 * This function is called when we are writing out the blocks that were
3509 * originally written with their allocation delayed, but then the space was
3510 * allocated using fallocate() before the delayed allocation could be resolved.
3511 * The cases to look for are:
3512 * ('=' indicated delayed allocated blocks
3513 * '-' indicates non-delayed allocated blocks)
3514 * (a) partial clusters towards beginning and/or end outside of allocated range
3515 * are not delalloc'ed.
3517 * |----c---=|====c====|====c====|===-c----|
3518 * |++++++ allocated ++++++|
3519 * ==> 4 complete clusters in above example
3521 * (b) partial cluster (outside of allocated range) towards either end is
3522 * marked for delayed allocation. In this case, we will exclude that
3525 * |----====c========|========c========|
3526 * |++++++ allocated ++++++|
3527 * ==> 1 complete clusters in above example
3530 * |================c================|
3531 * |++++++ allocated ++++++|
3532 * ==> 0 complete clusters in above example
3534 * The ext4_da_update_reserve_space will be called only if we
3535 * determine here that there were some "entire" clusters that span
3536 * this 'allocated' range.
3537 * In the non-bigalloc case, this function will just end up returning num_blks
3538 * without ever calling ext4_find_delalloc_range.
3541 get_reserved_cluster_alloc(struct inode
*inode
, ext4_lblk_t lblk_start
,
3542 unsigned int num_blks
)
3544 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
3545 ext4_lblk_t alloc_cluster_start
, alloc_cluster_end
;
3546 ext4_lblk_t lblk_from
, lblk_to
, c_offset
;
3547 unsigned int allocated_clusters
= 0;
3549 alloc_cluster_start
= EXT4_B2C(sbi
, lblk_start
);
3550 alloc_cluster_end
= EXT4_B2C(sbi
, lblk_start
+ num_blks
- 1);
3552 /* max possible clusters for this allocation */
3553 allocated_clusters
= alloc_cluster_end
- alloc_cluster_start
+ 1;
3555 trace_ext4_get_reserved_cluster_alloc(inode
, lblk_start
, num_blks
);
3557 /* Check towards left side */
3558 c_offset
= lblk_start
& (sbi
->s_cluster_ratio
- 1);
3560 lblk_from
= lblk_start
& (~(sbi
->s_cluster_ratio
- 1));
3561 lblk_to
= lblk_from
+ c_offset
- 1;
3563 if (ext4_find_delalloc_range(inode
, lblk_from
, lblk_to
))
3564 allocated_clusters
--;
3567 /* Now check towards right. */
3568 c_offset
= (lblk_start
+ num_blks
) & (sbi
->s_cluster_ratio
- 1);
3569 if (allocated_clusters
&& c_offset
) {
3570 lblk_from
= lblk_start
+ num_blks
;
3571 lblk_to
= lblk_from
+ (sbi
->s_cluster_ratio
- c_offset
) - 1;
3573 if (ext4_find_delalloc_range(inode
, lblk_from
, lblk_to
))
3574 allocated_clusters
--;
3577 return allocated_clusters
;
3581 ext4_ext_handle_uninitialized_extents(handle_t
*handle
, struct inode
*inode
,
3582 struct ext4_map_blocks
*map
,
3583 struct ext4_ext_path
*path
, int flags
,
3584 unsigned int allocated
, ext4_fsblk_t newblock
)
3588 ext4_io_end_t
*io
= ext4_inode_aio(inode
);
3590 ext_debug("ext4_ext_handle_uninitialized_extents: inode %lu, logical "
3591 "block %llu, max_blocks %u, flags %x, allocated %u\n",
3592 inode
->i_ino
, (unsigned long long)map
->m_lblk
, map
->m_len
,
3594 ext4_ext_show_leaf(inode
, path
);
3596 trace_ext4_ext_handle_uninitialized_extents(inode
, map
, flags
,
3597 allocated
, newblock
);
3599 /* get_block() before submit the IO, split the extent */
3600 if ((flags
& EXT4_GET_BLOCKS_PRE_IO
)) {
3601 ret
= ext4_split_unwritten_extents(handle
, inode
, map
,
3606 * Flag the inode(non aio case) or end_io struct (aio case)
3607 * that this IO needs to conversion to written when IO is
3611 ext4_set_io_unwritten_flag(inode
, io
);
3613 ext4_set_inode_state(inode
, EXT4_STATE_DIO_UNWRITTEN
);
3614 map
->m_flags
|= EXT4_MAP_UNWRITTEN
;
3615 if (ext4_should_dioread_nolock(inode
))
3616 map
->m_flags
|= EXT4_MAP_UNINIT
;
3619 /* IO end_io complete, convert the filled extent to written */
3620 if ((flags
& EXT4_GET_BLOCKS_CONVERT
)) {
3621 ret
= ext4_convert_unwritten_extents_endio(handle
, inode
, map
,
3624 ext4_update_inode_fsync_trans(handle
, inode
, 1);
3625 err
= check_eofblocks_fl(handle
, inode
, map
->m_lblk
,
3631 /* buffered IO case */
3633 * repeat fallocate creation request
3634 * we already have an unwritten extent
3636 if (flags
& EXT4_GET_BLOCKS_UNINIT_EXT
) {
3637 map
->m_flags
|= EXT4_MAP_UNWRITTEN
;
3641 /* buffered READ or buffered write_begin() lookup */
3642 if ((flags
& EXT4_GET_BLOCKS_CREATE
) == 0) {
3644 * We have blocks reserved already. We
3645 * return allocated blocks so that delalloc
3646 * won't do block reservation for us. But
3647 * the buffer head will be unmapped so that
3648 * a read from the block returns 0s.
3650 map
->m_flags
|= EXT4_MAP_UNWRITTEN
;
3654 /* buffered write, writepage time, convert*/
3655 ret
= ext4_ext_convert_to_initialized(handle
, inode
, map
, path
);
3657 ext4_update_inode_fsync_trans(handle
, inode
, 1);
3664 map
->m_flags
|= EXT4_MAP_NEW
;
3666 * if we allocated more blocks than requested
3667 * we need to make sure we unmap the extra block
3668 * allocated. The actual needed block will get
3669 * unmapped later when we find the buffer_head marked
3672 if (allocated
> map
->m_len
) {
3673 unmap_underlying_metadata_blocks(inode
->i_sb
->s_bdev
,
3674 newblock
+ map
->m_len
,
3675 allocated
- map
->m_len
);
3676 allocated
= map
->m_len
;
3680 * If we have done fallocate with the offset that is already
3681 * delayed allocated, we would have block reservation
3682 * and quota reservation done in the delayed write path.
3683 * But fallocate would have already updated quota and block
3684 * count for this offset. So cancel these reservation
3686 if (flags
& EXT4_GET_BLOCKS_DELALLOC_RESERVE
) {
3687 unsigned int reserved_clusters
;
3688 reserved_clusters
= get_reserved_cluster_alloc(inode
,
3689 map
->m_lblk
, map
->m_len
);
3690 if (reserved_clusters
)
3691 ext4_da_update_reserve_space(inode
,
3697 map
->m_flags
|= EXT4_MAP_MAPPED
;
3698 if ((flags
& EXT4_GET_BLOCKS_KEEP_SIZE
) == 0) {
3699 err
= check_eofblocks_fl(handle
, inode
, map
->m_lblk
, path
,
3705 if (allocated
> map
->m_len
)
3706 allocated
= map
->m_len
;
3707 ext4_ext_show_leaf(inode
, path
);
3708 map
->m_pblk
= newblock
;
3709 map
->m_len
= allocated
;
3712 ext4_ext_drop_refs(path
);
3715 return err
? err
: allocated
;
3719 * get_implied_cluster_alloc - check to see if the requested
3720 * allocation (in the map structure) overlaps with a cluster already
3721 * allocated in an extent.
3722 * @sb The filesystem superblock structure
3723 * @map The requested lblk->pblk mapping
3724 * @ex The extent structure which might contain an implied
3725 * cluster allocation
3727 * This function is called by ext4_ext_map_blocks() after we failed to
3728 * find blocks that were already in the inode's extent tree. Hence,
3729 * we know that the beginning of the requested region cannot overlap
3730 * the extent from the inode's extent tree. There are three cases we
3731 * want to catch. The first is this case:
3733 * |--- cluster # N--|
3734 * |--- extent ---| |---- requested region ---|
3737 * The second case that we need to test for is this one:
3739 * |--------- cluster # N ----------------|
3740 * |--- requested region --| |------- extent ----|
3741 * |=======================|
3743 * The third case is when the requested region lies between two extents
3744 * within the same cluster:
3745 * |------------- cluster # N-------------|
3746 * |----- ex -----| |---- ex_right ----|
3747 * |------ requested region ------|
3748 * |================|
3750 * In each of the above cases, we need to set the map->m_pblk and
3751 * map->m_len so it corresponds to the return the extent labelled as
3752 * "|====|" from cluster #N, since it is already in use for data in
3753 * cluster EXT4_B2C(sbi, map->m_lblk). We will then return 1 to
3754 * signal to ext4_ext_map_blocks() that map->m_pblk should be treated
3755 * as a new "allocated" block region. Otherwise, we will return 0 and
3756 * ext4_ext_map_blocks() will then allocate one or more new clusters
3757 * by calling ext4_mb_new_blocks().
3759 static int get_implied_cluster_alloc(struct super_block
*sb
,
3760 struct ext4_map_blocks
*map
,
3761 struct ext4_extent
*ex
,
3762 struct ext4_ext_path
*path
)
3764 struct ext4_sb_info
*sbi
= EXT4_SB(sb
);
3765 ext4_lblk_t c_offset
= map
->m_lblk
& (sbi
->s_cluster_ratio
-1);
3766 ext4_lblk_t ex_cluster_start
, ex_cluster_end
;
3767 ext4_lblk_t rr_cluster_start
;
3768 ext4_lblk_t ee_block
= le32_to_cpu(ex
->ee_block
);
3769 ext4_fsblk_t ee_start
= ext4_ext_pblock(ex
);
3770 unsigned short ee_len
= ext4_ext_get_actual_len(ex
);
3772 /* The extent passed in that we are trying to match */
3773 ex_cluster_start
= EXT4_B2C(sbi
, ee_block
);
3774 ex_cluster_end
= EXT4_B2C(sbi
, ee_block
+ ee_len
- 1);
3776 /* The requested region passed into ext4_map_blocks() */
3777 rr_cluster_start
= EXT4_B2C(sbi
, map
->m_lblk
);
3779 if ((rr_cluster_start
== ex_cluster_end
) ||
3780 (rr_cluster_start
== ex_cluster_start
)) {
3781 if (rr_cluster_start
== ex_cluster_end
)
3782 ee_start
+= ee_len
- 1;
3783 map
->m_pblk
= (ee_start
& ~(sbi
->s_cluster_ratio
- 1)) +
3785 map
->m_len
= min(map
->m_len
,
3786 (unsigned) sbi
->s_cluster_ratio
- c_offset
);
3788 * Check for and handle this case:
3790 * |--------- cluster # N-------------|
3791 * |------- extent ----|
3792 * |--- requested region ---|
3796 if (map
->m_lblk
< ee_block
)
3797 map
->m_len
= min(map
->m_len
, ee_block
- map
->m_lblk
);
3800 * Check for the case where there is already another allocated
3801 * block to the right of 'ex' but before the end of the cluster.
3803 * |------------- cluster # N-------------|
3804 * |----- ex -----| |---- ex_right ----|
3805 * |------ requested region ------|
3806 * |================|
3808 if (map
->m_lblk
> ee_block
) {
3809 ext4_lblk_t next
= ext4_ext_next_allocated_block(path
);
3810 map
->m_len
= min(map
->m_len
, next
- map
->m_lblk
);
3813 trace_ext4_get_implied_cluster_alloc_exit(sb
, map
, 1);
3817 trace_ext4_get_implied_cluster_alloc_exit(sb
, map
, 0);
3823 * Block allocation/map/preallocation routine for extents based files
3826 * Need to be called with
3827 * down_read(&EXT4_I(inode)->i_data_sem) if not allocating file system block
3828 * (ie, create is zero). Otherwise down_write(&EXT4_I(inode)->i_data_sem)
3830 * return > 0, number of of blocks already mapped/allocated
3831 * if create == 0 and these are pre-allocated blocks
3832 * buffer head is unmapped
3833 * otherwise blocks are mapped
3835 * return = 0, if plain look up failed (blocks have not been allocated)
3836 * buffer head is unmapped
3838 * return < 0, error case.
3840 int ext4_ext_map_blocks(handle_t
*handle
, struct inode
*inode
,
3841 struct ext4_map_blocks
*map
, int flags
)
3843 struct ext4_ext_path
*path
= NULL
;
3844 struct ext4_extent newex
, *ex
, *ex2
;
3845 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
3846 ext4_fsblk_t newblock
= 0;
3847 int free_on_err
= 0, err
= 0, depth
;
3848 unsigned int allocated
= 0, offset
= 0;
3849 unsigned int allocated_clusters
= 0;
3850 struct ext4_allocation_request ar
;
3851 ext4_io_end_t
*io
= ext4_inode_aio(inode
);
3852 ext4_lblk_t cluster_offset
;
3853 int set_unwritten
= 0;
3855 ext_debug("blocks %u/%u requested for inode %lu\n",
3856 map
->m_lblk
, map
->m_len
, inode
->i_ino
);
3857 trace_ext4_ext_map_blocks_enter(inode
, map
->m_lblk
, map
->m_len
, flags
);
3859 /* find extent for this block */
3860 path
= ext4_ext_find_extent(inode
, map
->m_lblk
, NULL
);
3862 err
= PTR_ERR(path
);
3867 depth
= ext_depth(inode
);
3870 * consistent leaf must not be empty;
3871 * this situation is possible, though, _during_ tree modification;
3872 * this is why assert can't be put in ext4_ext_find_extent()
3874 if (unlikely(path
[depth
].p_ext
== NULL
&& depth
!= 0)) {
3875 EXT4_ERROR_INODE(inode
, "bad extent address "
3876 "lblock: %lu, depth: %d pblock %lld",
3877 (unsigned long) map
->m_lblk
, depth
,
3878 path
[depth
].p_block
);
3883 ex
= path
[depth
].p_ext
;
3885 ext4_lblk_t ee_block
= le32_to_cpu(ex
->ee_block
);
3886 ext4_fsblk_t ee_start
= ext4_ext_pblock(ex
);
3887 unsigned short ee_len
;
3890 * Uninitialized extents are treated as holes, except that
3891 * we split out initialized portions during a write.
3893 ee_len
= ext4_ext_get_actual_len(ex
);
3895 trace_ext4_ext_show_extent(inode
, ee_block
, ee_start
, ee_len
);
3897 /* if found extent covers block, simply return it */
3898 if (in_range(map
->m_lblk
, ee_block
, ee_len
)) {
3899 newblock
= map
->m_lblk
- ee_block
+ ee_start
;
3900 /* number of remaining blocks in the extent */
3901 allocated
= ee_len
- (map
->m_lblk
- ee_block
);
3902 ext_debug("%u fit into %u:%d -> %llu\n", map
->m_lblk
,
3903 ee_block
, ee_len
, newblock
);
3905 if (!ext4_ext_is_uninitialized(ex
))
3908 allocated
= ext4_ext_handle_uninitialized_extents(
3909 handle
, inode
, map
, path
, flags
,
3910 allocated
, newblock
);
3915 if ((sbi
->s_cluster_ratio
> 1) &&
3916 ext4_find_delalloc_cluster(inode
, map
->m_lblk
))
3917 map
->m_flags
|= EXT4_MAP_FROM_CLUSTER
;
3920 * requested block isn't allocated yet;
3921 * we couldn't try to create block if create flag is zero
3923 if ((flags
& EXT4_GET_BLOCKS_CREATE
) == 0) {
3925 * put just found gap into cache to speed up
3926 * subsequent requests
3928 if ((flags
& EXT4_GET_BLOCKS_NO_PUT_HOLE
) == 0)
3929 ext4_ext_put_gap_in_cache(inode
, path
, map
->m_lblk
);
3934 * Okay, we need to do block allocation.
3936 map
->m_flags
&= ~EXT4_MAP_FROM_CLUSTER
;
3937 newex
.ee_block
= cpu_to_le32(map
->m_lblk
);
3938 cluster_offset
= map
->m_lblk
& (sbi
->s_cluster_ratio
-1);
3941 * If we are doing bigalloc, check to see if the extent returned
3942 * by ext4_ext_find_extent() implies a cluster we can use.
3944 if (cluster_offset
&& ex
&&
3945 get_implied_cluster_alloc(inode
->i_sb
, map
, ex
, path
)) {
3946 ar
.len
= allocated
= map
->m_len
;
3947 newblock
= map
->m_pblk
;
3948 map
->m_flags
|= EXT4_MAP_FROM_CLUSTER
;
3949 goto got_allocated_blocks
;
3952 /* find neighbour allocated blocks */
3953 ar
.lleft
= map
->m_lblk
;
3954 err
= ext4_ext_search_left(inode
, path
, &ar
.lleft
, &ar
.pleft
);
3957 ar
.lright
= map
->m_lblk
;
3959 err
= ext4_ext_search_right(inode
, path
, &ar
.lright
, &ar
.pright
, &ex2
);
3963 /* Check if the extent after searching to the right implies a
3964 * cluster we can use. */
3965 if ((sbi
->s_cluster_ratio
> 1) && ex2
&&
3966 get_implied_cluster_alloc(inode
->i_sb
, map
, ex2
, path
)) {
3967 ar
.len
= allocated
= map
->m_len
;
3968 newblock
= map
->m_pblk
;
3969 map
->m_flags
|= EXT4_MAP_FROM_CLUSTER
;
3970 goto got_allocated_blocks
;
3974 * See if request is beyond maximum number of blocks we can have in
3975 * a single extent. For an initialized extent this limit is
3976 * EXT_INIT_MAX_LEN and for an uninitialized extent this limit is
3977 * EXT_UNINIT_MAX_LEN.
3979 if (map
->m_len
> EXT_INIT_MAX_LEN
&&
3980 !(flags
& EXT4_GET_BLOCKS_UNINIT_EXT
))
3981 map
->m_len
= EXT_INIT_MAX_LEN
;
3982 else if (map
->m_len
> EXT_UNINIT_MAX_LEN
&&
3983 (flags
& EXT4_GET_BLOCKS_UNINIT_EXT
))
3984 map
->m_len
= EXT_UNINIT_MAX_LEN
;
3986 /* Check if we can really insert (m_lblk)::(m_lblk + m_len) extent */
3987 newex
.ee_len
= cpu_to_le16(map
->m_len
);
3988 err
= ext4_ext_check_overlap(sbi
, inode
, &newex
, path
);
3990 allocated
= ext4_ext_get_actual_len(&newex
);
3992 allocated
= map
->m_len
;
3994 /* allocate new block */
3996 ar
.goal
= ext4_ext_find_goal(inode
, path
, map
->m_lblk
);
3997 ar
.logical
= map
->m_lblk
;
3999 * We calculate the offset from the beginning of the cluster
4000 * for the logical block number, since when we allocate a
4001 * physical cluster, the physical block should start at the
4002 * same offset from the beginning of the cluster. This is
4003 * needed so that future calls to get_implied_cluster_alloc()
4006 offset
= map
->m_lblk
& (sbi
->s_cluster_ratio
- 1);
4007 ar
.len
= EXT4_NUM_B2C(sbi
, offset
+allocated
);
4009 ar
.logical
-= offset
;
4010 if (S_ISREG(inode
->i_mode
))
4011 ar
.flags
= EXT4_MB_HINT_DATA
;
4013 /* disable in-core preallocation for non-regular files */
4015 if (flags
& EXT4_GET_BLOCKS_NO_NORMALIZE
)
4016 ar
.flags
|= EXT4_MB_HINT_NOPREALLOC
;
4017 newblock
= ext4_mb_new_blocks(handle
, &ar
, &err
);
4020 ext_debug("allocate new block: goal %llu, found %llu/%u\n",
4021 ar
.goal
, newblock
, allocated
);
4023 allocated_clusters
= ar
.len
;
4024 ar
.len
= EXT4_C2B(sbi
, ar
.len
) - offset
;
4025 if (ar
.len
> allocated
)
4028 got_allocated_blocks
:
4029 /* try to insert new extent into found leaf and return */
4030 ext4_ext_store_pblock(&newex
, newblock
+ offset
);
4031 newex
.ee_len
= cpu_to_le16(ar
.len
);
4032 /* Mark uninitialized */
4033 if (flags
& EXT4_GET_BLOCKS_UNINIT_EXT
){
4034 ext4_ext_mark_uninitialized(&newex
);
4035 map
->m_flags
|= EXT4_MAP_UNWRITTEN
;
4037 * io_end structure was created for every IO write to an
4038 * uninitialized extent. To avoid unnecessary conversion,
4039 * here we flag the IO that really needs the conversion.
4040 * For non asycn direct IO case, flag the inode state
4041 * that we need to perform conversion when IO is done.
4043 if ((flags
& EXT4_GET_BLOCKS_PRE_IO
))
4045 if (ext4_should_dioread_nolock(inode
))
4046 map
->m_flags
|= EXT4_MAP_UNINIT
;
4050 if ((flags
& EXT4_GET_BLOCKS_KEEP_SIZE
) == 0)
4051 err
= check_eofblocks_fl(handle
, inode
, map
->m_lblk
,
4054 err
= ext4_ext_insert_extent(handle
, inode
, path
,
4057 if (!err
&& set_unwritten
) {
4059 ext4_set_io_unwritten_flag(inode
, io
);
4061 ext4_set_inode_state(inode
,
4062 EXT4_STATE_DIO_UNWRITTEN
);
4065 if (err
&& free_on_err
) {
4066 int fb_flags
= flags
& EXT4_GET_BLOCKS_DELALLOC_RESERVE
?
4067 EXT4_FREE_BLOCKS_NO_QUOT_UPDATE
: 0;
4068 /* free data blocks we just allocated */
4069 /* not a good idea to call discard here directly,
4070 * but otherwise we'd need to call it every free() */
4071 ext4_discard_preallocations(inode
);
4072 ext4_free_blocks(handle
, inode
, NULL
, ext4_ext_pblock(&newex
),
4073 ext4_ext_get_actual_len(&newex
), fb_flags
);
4077 /* previous routine could use block we allocated */
4078 newblock
= ext4_ext_pblock(&newex
);
4079 allocated
= ext4_ext_get_actual_len(&newex
);
4080 if (allocated
> map
->m_len
)
4081 allocated
= map
->m_len
;
4082 map
->m_flags
|= EXT4_MAP_NEW
;
4085 * Update reserved blocks/metadata blocks after successful
4086 * block allocation which had been deferred till now.
4088 if (flags
& EXT4_GET_BLOCKS_DELALLOC_RESERVE
) {
4089 unsigned int reserved_clusters
;
4091 * Check how many clusters we had reserved this allocated range
4093 reserved_clusters
= get_reserved_cluster_alloc(inode
,
4094 map
->m_lblk
, allocated
);
4095 if (map
->m_flags
& EXT4_MAP_FROM_CLUSTER
) {
4096 if (reserved_clusters
) {
4098 * We have clusters reserved for this range.
4099 * But since we are not doing actual allocation
4100 * and are simply using blocks from previously
4101 * allocated cluster, we should release the
4102 * reservation and not claim quota.
4104 ext4_da_update_reserve_space(inode
,
4105 reserved_clusters
, 0);
4108 BUG_ON(allocated_clusters
< reserved_clusters
);
4109 /* We will claim quota for all newly allocated blocks.*/
4110 ext4_da_update_reserve_space(inode
, allocated_clusters
,
4112 if (reserved_clusters
< allocated_clusters
) {
4113 struct ext4_inode_info
*ei
= EXT4_I(inode
);
4114 int reservation
= allocated_clusters
-
4117 * It seems we claimed few clusters outside of
4118 * the range of this allocation. We should give
4119 * it back to the reservation pool. This can
4120 * happen in the following case:
4122 * * Suppose s_cluster_ratio is 4 (i.e., each
4123 * cluster has 4 blocks. Thus, the clusters
4124 * are [0-3],[4-7],[8-11]...
4125 * * First comes delayed allocation write for
4126 * logical blocks 10 & 11. Since there were no
4127 * previous delayed allocated blocks in the
4128 * range [8-11], we would reserve 1 cluster
4130 * * Next comes write for logical blocks 3 to 8.
4131 * In this case, we will reserve 2 clusters
4132 * (for [0-3] and [4-7]; and not for [8-11] as
4133 * that range has a delayed allocated blocks.
4134 * Thus total reserved clusters now becomes 3.
4135 * * Now, during the delayed allocation writeout
4136 * time, we will first write blocks [3-8] and
4137 * allocate 3 clusters for writing these
4138 * blocks. Also, we would claim all these
4139 * three clusters above.
4140 * * Now when we come here to writeout the
4141 * blocks [10-11], we would expect to claim
4142 * the reservation of 1 cluster we had made
4143 * (and we would claim it since there are no
4144 * more delayed allocated blocks in the range
4145 * [8-11]. But our reserved cluster count had
4146 * already gone to 0.
4148 * Thus, at the step 4 above when we determine
4149 * that there are still some unwritten delayed
4150 * allocated blocks outside of our current
4151 * block range, we should increment the
4152 * reserved clusters count so that when the
4153 * remaining blocks finally gets written, we
4156 dquot_reserve_block(inode
,
4157 EXT4_C2B(sbi
, reservation
));
4158 spin_lock(&ei
->i_block_reservation_lock
);
4159 ei
->i_reserved_data_blocks
+= reservation
;
4160 spin_unlock(&ei
->i_block_reservation_lock
);
4166 * Cache the extent and update transaction to commit on fdatasync only
4167 * when it is _not_ an uninitialized extent.
4169 if ((flags
& EXT4_GET_BLOCKS_UNINIT_EXT
) == 0)
4170 ext4_update_inode_fsync_trans(handle
, inode
, 1);
4172 ext4_update_inode_fsync_trans(handle
, inode
, 0);
4174 if (allocated
> map
->m_len
)
4175 allocated
= map
->m_len
;
4176 ext4_ext_show_leaf(inode
, path
);
4177 map
->m_flags
|= EXT4_MAP_MAPPED
;
4178 map
->m_pblk
= newblock
;
4179 map
->m_len
= allocated
;
4182 ext4_ext_drop_refs(path
);
4187 trace_ext4_ext_map_blocks_exit(inode
, map
, err
? err
: allocated
);
4189 return err
? err
: allocated
;
4192 void ext4_ext_truncate(struct inode
*inode
)
4194 struct address_space
*mapping
= inode
->i_mapping
;
4195 struct super_block
*sb
= inode
->i_sb
;
4196 ext4_lblk_t last_block
;
4202 * finish any pending end_io work so we won't run the risk of
4203 * converting any truncated blocks to initialized later
4205 ext4_flush_unwritten_io(inode
);
4208 * probably first extent we're gonna free will be last in block
4210 err
= ext4_writepage_trans_blocks(inode
);
4211 handle
= ext4_journal_start(inode
, EXT4_HT_TRUNCATE
, err
);
4215 if (inode
->i_size
% PAGE_CACHE_SIZE
!= 0) {
4216 page_len
= PAGE_CACHE_SIZE
-
4217 (inode
->i_size
& (PAGE_CACHE_SIZE
- 1));
4219 err
= ext4_discard_partial_page_buffers(handle
,
4220 mapping
, inode
->i_size
, page_len
, 0);
4226 if (ext4_orphan_add(handle
, inode
))
4229 down_write(&EXT4_I(inode
)->i_data_sem
);
4231 ext4_discard_preallocations(inode
);
4234 * TODO: optimization is possible here.
4235 * Probably we need not scan at all,
4236 * because page truncation is enough.
4239 /* we have to know where to truncate from in crash case */
4240 EXT4_I(inode
)->i_disksize
= inode
->i_size
;
4241 ext4_mark_inode_dirty(handle
, inode
);
4243 last_block
= (inode
->i_size
+ sb
->s_blocksize
- 1)
4244 >> EXT4_BLOCK_SIZE_BITS(sb
);
4245 err
= ext4_es_remove_extent(inode
, last_block
,
4246 EXT_MAX_BLOCKS
- last_block
);
4247 err
= ext4_ext_remove_space(inode
, last_block
, EXT_MAX_BLOCKS
- 1);
4249 /* In a multi-transaction truncate, we only make the final
4250 * transaction synchronous.
4253 ext4_handle_sync(handle
);
4255 up_write(&EXT4_I(inode
)->i_data_sem
);
4259 * If this was a simple ftruncate() and the file will remain alive,
4260 * then we need to clear up the orphan record which we created above.
4261 * However, if this was a real unlink then we were called by
4262 * ext4_delete_inode(), and we allow that function to clean up the
4263 * orphan info for us.
4266 ext4_orphan_del(handle
, inode
);
4268 inode
->i_mtime
= inode
->i_ctime
= ext4_current_time(inode
);
4269 ext4_mark_inode_dirty(handle
, inode
);
4270 ext4_journal_stop(handle
);
4273 static void ext4_falloc_update_inode(struct inode
*inode
,
4274 int mode
, loff_t new_size
, int update_ctime
)
4276 struct timespec now
;
4279 now
= current_fs_time(inode
->i_sb
);
4280 if (!timespec_equal(&inode
->i_ctime
, &now
))
4281 inode
->i_ctime
= now
;
4284 * Update only when preallocation was requested beyond
4287 if (!(mode
& FALLOC_FL_KEEP_SIZE
)) {
4288 if (new_size
> i_size_read(inode
))
4289 i_size_write(inode
, new_size
);
4290 if (new_size
> EXT4_I(inode
)->i_disksize
)
4291 ext4_update_i_disksize(inode
, new_size
);
4294 * Mark that we allocate beyond EOF so the subsequent truncate
4295 * can proceed even if the new size is the same as i_size.
4297 if (new_size
> i_size_read(inode
))
4298 ext4_set_inode_flag(inode
, EXT4_INODE_EOFBLOCKS
);
4304 * preallocate space for a file. This implements ext4's fallocate file
4305 * operation, which gets called from sys_fallocate system call.
4306 * For block-mapped files, posix_fallocate should fall back to the method
4307 * of writing zeroes to the required new blocks (the same behavior which is
4308 * expected for file systems which do not support fallocate() system call).
4310 long ext4_fallocate(struct file
*file
, int mode
, loff_t offset
, loff_t len
)
4312 struct inode
*inode
= file_inode(file
);
4315 unsigned int max_blocks
;
4320 struct ext4_map_blocks map
;
4321 unsigned int credits
, blkbits
= inode
->i_blkbits
;
4323 /* Return error if mode is not supported */
4324 if (mode
& ~(FALLOC_FL_KEEP_SIZE
| FALLOC_FL_PUNCH_HOLE
))
4327 if (mode
& FALLOC_FL_PUNCH_HOLE
)
4328 return ext4_punch_hole(file
, offset
, len
);
4330 ret
= ext4_convert_inline_data(inode
);
4335 * currently supporting (pre)allocate mode for extent-based
4338 if (!(ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
)))
4341 trace_ext4_fallocate_enter(inode
, offset
, len
, mode
);
4342 map
.m_lblk
= offset
>> blkbits
;
4344 * We can't just convert len to max_blocks because
4345 * If blocksize = 4096 offset = 3072 and len = 2048
4347 max_blocks
= (EXT4_BLOCK_ALIGN(len
+ offset
, blkbits
) >> blkbits
)
4350 * credits to insert 1 extent into extent tree
4352 credits
= ext4_chunk_trans_blocks(inode
, max_blocks
);
4353 mutex_lock(&inode
->i_mutex
);
4354 ret
= inode_newsize_ok(inode
, (len
+ offset
));
4356 mutex_unlock(&inode
->i_mutex
);
4357 trace_ext4_fallocate_exit(inode
, offset
, max_blocks
, ret
);
4360 flags
= EXT4_GET_BLOCKS_CREATE_UNINIT_EXT
;
4361 if (mode
& FALLOC_FL_KEEP_SIZE
)
4362 flags
|= EXT4_GET_BLOCKS_KEEP_SIZE
;
4364 * Don't normalize the request if it can fit in one extent so
4365 * that it doesn't get unnecessarily split into multiple
4368 if (len
<= EXT_UNINIT_MAX_LEN
<< blkbits
)
4369 flags
|= EXT4_GET_BLOCKS_NO_NORMALIZE
;
4371 /* Prevent race condition between unwritten */
4372 ext4_flush_unwritten_io(inode
);
4374 while (ret
>= 0 && ret
< max_blocks
) {
4375 map
.m_lblk
= map
.m_lblk
+ ret
;
4376 map
.m_len
= max_blocks
= max_blocks
- ret
;
4377 handle
= ext4_journal_start(inode
, EXT4_HT_MAP_BLOCKS
,
4379 if (IS_ERR(handle
)) {
4380 ret
= PTR_ERR(handle
);
4383 ret
= ext4_map_blocks(handle
, inode
, &map
, flags
);
4386 ext4_warning(inode
->i_sb
,
4387 "inode #%lu: block %u: len %u: "
4388 "ext4_ext_map_blocks returned %d",
4389 inode
->i_ino
, map
.m_lblk
,
4392 ext4_mark_inode_dirty(handle
, inode
);
4393 ret2
= ext4_journal_stop(handle
);
4396 if ((map
.m_lblk
+ ret
) >= (EXT4_BLOCK_ALIGN(offset
+ len
,
4397 blkbits
) >> blkbits
))
4398 new_size
= offset
+ len
;
4400 new_size
= ((loff_t
) map
.m_lblk
+ ret
) << blkbits
;
4402 ext4_falloc_update_inode(inode
, mode
, new_size
,
4403 (map
.m_flags
& EXT4_MAP_NEW
));
4404 ext4_mark_inode_dirty(handle
, inode
);
4405 if ((file
->f_flags
& O_SYNC
) && ret
>= max_blocks
)
4406 ext4_handle_sync(handle
);
4407 ret2
= ext4_journal_stop(handle
);
4411 if (ret
== -ENOSPC
&&
4412 ext4_should_retry_alloc(inode
->i_sb
, &retries
)) {
4416 mutex_unlock(&inode
->i_mutex
);
4417 trace_ext4_fallocate_exit(inode
, offset
, max_blocks
,
4418 ret
> 0 ? ret2
: ret
);
4419 return ret
> 0 ? ret2
: ret
;
4423 * This function convert a range of blocks to written extents
4424 * The caller of this function will pass the start offset and the size.
4425 * all unwritten extents within this range will be converted to
4428 * This function is called from the direct IO end io call back
4429 * function, to convert the fallocated extents after IO is completed.
4430 * Returns 0 on success.
4432 int ext4_convert_unwritten_extents(struct inode
*inode
, loff_t offset
,
4436 unsigned int max_blocks
;
4439 struct ext4_map_blocks map
;
4440 unsigned int credits
, blkbits
= inode
->i_blkbits
;
4442 map
.m_lblk
= offset
>> blkbits
;
4444 * We can't just convert len to max_blocks because
4445 * If blocksize = 4096 offset = 3072 and len = 2048
4447 max_blocks
= ((EXT4_BLOCK_ALIGN(len
+ offset
, blkbits
) >> blkbits
) -
4450 * credits to insert 1 extent into extent tree
4452 credits
= ext4_chunk_trans_blocks(inode
, max_blocks
);
4453 while (ret
>= 0 && ret
< max_blocks
) {
4455 map
.m_len
= (max_blocks
-= ret
);
4456 handle
= ext4_journal_start(inode
, EXT4_HT_MAP_BLOCKS
, credits
);
4457 if (IS_ERR(handle
)) {
4458 ret
= PTR_ERR(handle
);
4461 ret
= ext4_map_blocks(handle
, inode
, &map
,
4462 EXT4_GET_BLOCKS_IO_CONVERT_EXT
);
4464 ext4_warning(inode
->i_sb
,
4465 "inode #%lu: block %u: len %u: "
4466 "ext4_ext_map_blocks returned %d",
4467 inode
->i_ino
, map
.m_lblk
,
4469 ext4_mark_inode_dirty(handle
, inode
);
4470 ret2
= ext4_journal_stop(handle
);
4471 if (ret
<= 0 || ret2
)
4474 return ret
> 0 ? ret2
: ret
;
4478 * If newes is not existing extent (newes->ec_pblk equals zero) find
4479 * delayed extent at start of newes and update newes accordingly and
4480 * return start of the next delayed extent.
4482 * If newes is existing extent (newes->ec_pblk is not equal zero)
4483 * return start of next delayed extent or EXT_MAX_BLOCKS if no delayed
4484 * extent found. Leave newes unmodified.
4486 static int ext4_find_delayed_extent(struct inode
*inode
,
4487 struct extent_status
*newes
)
4489 struct extent_status es
;
4490 ext4_lblk_t block
, next_del
;
4492 ext4_es_find_delayed_extent(inode
, newes
->es_lblk
, &es
);
4494 if (newes
->es_pblk
== 0) {
4496 * No extent in extent-tree contains block @newes->es_pblk,
4497 * then the block may stay in 1)a hole or 2)delayed-extent.
4503 if (es
.es_lblk
> newes
->es_lblk
) {
4505 newes
->es_len
= min(es
.es_lblk
- newes
->es_lblk
,
4510 newes
->es_len
= es
.es_lblk
+ es
.es_len
- newes
->es_lblk
;
4513 block
= newes
->es_lblk
+ newes
->es_len
;
4514 ext4_es_find_delayed_extent(inode
, block
, &es
);
4516 next_del
= EXT_MAX_BLOCKS
;
4518 next_del
= es
.es_lblk
;
4522 /* fiemap flags we can handle specified here */
4523 #define EXT4_FIEMAP_FLAGS (FIEMAP_FLAG_SYNC|FIEMAP_FLAG_XATTR)
4525 static int ext4_xattr_fiemap(struct inode
*inode
,
4526 struct fiemap_extent_info
*fieinfo
)
4530 __u32 flags
= FIEMAP_EXTENT_LAST
;
4531 int blockbits
= inode
->i_sb
->s_blocksize_bits
;
4535 if (ext4_test_inode_state(inode
, EXT4_STATE_XATTR
)) {
4536 struct ext4_iloc iloc
;
4537 int offset
; /* offset of xattr in inode */
4539 error
= ext4_get_inode_loc(inode
, &iloc
);
4542 physical
= iloc
.bh
->b_blocknr
<< blockbits
;
4543 offset
= EXT4_GOOD_OLD_INODE_SIZE
+
4544 EXT4_I(inode
)->i_extra_isize
;
4546 length
= EXT4_SB(inode
->i_sb
)->s_inode_size
- offset
;
4547 flags
|= FIEMAP_EXTENT_DATA_INLINE
;
4549 } else { /* external block */
4550 physical
= EXT4_I(inode
)->i_file_acl
<< blockbits
;
4551 length
= inode
->i_sb
->s_blocksize
;
4555 error
= fiemap_fill_next_extent(fieinfo
, 0, physical
,
4557 return (error
< 0 ? error
: 0);
4561 * ext4_ext_punch_hole
4563 * Punches a hole of "length" bytes in a file starting
4566 * @inode: The inode of the file to punch a hole in
4567 * @offset: The starting byte offset of the hole
4568 * @length: The length of the hole
4570 * Returns the number of blocks removed or negative on err
4572 int ext4_ext_punch_hole(struct file
*file
, loff_t offset
, loff_t length
)
4574 struct inode
*inode
= file_inode(file
);
4575 struct super_block
*sb
= inode
->i_sb
;
4576 ext4_lblk_t first_block
, stop_block
;
4577 struct address_space
*mapping
= inode
->i_mapping
;
4579 loff_t first_page
, last_page
, page_len
;
4580 loff_t first_page_offset
, last_page_offset
;
4581 int credits
, err
= 0;
4584 * Write out all dirty pages to avoid race conditions
4585 * Then release them.
4587 if (mapping
->nrpages
&& mapping_tagged(mapping
, PAGECACHE_TAG_DIRTY
)) {
4588 err
= filemap_write_and_wait_range(mapping
,
4589 offset
, offset
+ length
- 1);
4595 mutex_lock(&inode
->i_mutex
);
4596 /* It's not possible punch hole on append only file */
4597 if (IS_APPEND(inode
) || IS_IMMUTABLE(inode
)) {
4601 if (IS_SWAPFILE(inode
)) {
4606 /* No need to punch hole beyond i_size */
4607 if (offset
>= inode
->i_size
)
4611 * If the hole extends beyond i_size, set the hole
4612 * to end after the page that contains i_size
4614 if (offset
+ length
> inode
->i_size
) {
4615 length
= inode
->i_size
+
4616 PAGE_CACHE_SIZE
- (inode
->i_size
& (PAGE_CACHE_SIZE
- 1)) -
4620 first_page
= (offset
+ PAGE_CACHE_SIZE
- 1) >> PAGE_CACHE_SHIFT
;
4621 last_page
= (offset
+ length
) >> PAGE_CACHE_SHIFT
;
4623 first_page_offset
= first_page
<< PAGE_CACHE_SHIFT
;
4624 last_page_offset
= last_page
<< PAGE_CACHE_SHIFT
;
4626 /* Now release the pages */
4627 if (last_page_offset
> first_page_offset
) {
4628 truncate_pagecache_range(inode
, first_page_offset
,
4629 last_page_offset
- 1);
4632 /* Wait all existing dio workers, newcomers will block on i_mutex */
4633 ext4_inode_block_unlocked_dio(inode
);
4634 err
= ext4_flush_unwritten_io(inode
);
4637 inode_dio_wait(inode
);
4639 credits
= ext4_writepage_trans_blocks(inode
);
4640 handle
= ext4_journal_start(inode
, EXT4_HT_TRUNCATE
, credits
);
4641 if (IS_ERR(handle
)) {
4642 err
= PTR_ERR(handle
);
4648 * Now we need to zero out the non-page-aligned data in the
4649 * pages at the start and tail of the hole, and unmap the buffer
4650 * heads for the block aligned regions of the page that were
4651 * completely zeroed.
4653 if (first_page
> last_page
) {
4655 * If the file space being truncated is contained within a page
4656 * just zero out and unmap the middle of that page
4658 err
= ext4_discard_partial_page_buffers(handle
,
4659 mapping
, offset
, length
, 0);
4665 * zero out and unmap the partial page that contains
4666 * the start of the hole
4668 page_len
= first_page_offset
- offset
;
4670 err
= ext4_discard_partial_page_buffers(handle
, mapping
,
4671 offset
, page_len
, 0);
4677 * zero out and unmap the partial page that contains
4678 * the end of the hole
4680 page_len
= offset
+ length
- last_page_offset
;
4682 err
= ext4_discard_partial_page_buffers(handle
, mapping
,
4683 last_page_offset
, page_len
, 0);
4690 * If i_size is contained in the last page, we need to
4691 * unmap and zero the partial page after i_size
4693 if (inode
->i_size
>> PAGE_CACHE_SHIFT
== last_page
&&
4694 inode
->i_size
% PAGE_CACHE_SIZE
!= 0) {
4696 page_len
= PAGE_CACHE_SIZE
-
4697 (inode
->i_size
& (PAGE_CACHE_SIZE
- 1));
4700 err
= ext4_discard_partial_page_buffers(handle
,
4701 mapping
, inode
->i_size
, page_len
, 0);
4708 first_block
= (offset
+ sb
->s_blocksize
- 1) >>
4709 EXT4_BLOCK_SIZE_BITS(sb
);
4710 stop_block
= (offset
+ length
) >> EXT4_BLOCK_SIZE_BITS(sb
);
4712 /* If there are no blocks to remove, return now */
4713 if (first_block
>= stop_block
)
4716 down_write(&EXT4_I(inode
)->i_data_sem
);
4717 ext4_discard_preallocations(inode
);
4719 err
= ext4_es_remove_extent(inode
, first_block
,
4720 stop_block
- first_block
);
4721 err
= ext4_ext_remove_space(inode
, first_block
, stop_block
- 1);
4723 ext4_discard_preallocations(inode
);
4726 ext4_handle_sync(handle
);
4728 up_write(&EXT4_I(inode
)->i_data_sem
);
4731 inode
->i_mtime
= inode
->i_ctime
= ext4_current_time(inode
);
4732 ext4_mark_inode_dirty(handle
, inode
);
4733 ext4_journal_stop(handle
);
4735 ext4_inode_resume_unlocked_dio(inode
);
4737 mutex_unlock(&inode
->i_mutex
);
4741 int ext4_fiemap(struct inode
*inode
, struct fiemap_extent_info
*fieinfo
,
4742 __u64 start
, __u64 len
)
4744 ext4_lblk_t start_blk
;
4747 if (ext4_has_inline_data(inode
)) {
4750 error
= ext4_inline_data_fiemap(inode
, fieinfo
, &has_inline
);
4756 /* fallback to generic here if not in extents fmt */
4757 if (!(ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
)))
4758 return generic_block_fiemap(inode
, fieinfo
, start
, len
,
4761 if (fiemap_check_flags(fieinfo
, EXT4_FIEMAP_FLAGS
))
4764 if (fieinfo
->fi_flags
& FIEMAP_FLAG_XATTR
) {
4765 error
= ext4_xattr_fiemap(inode
, fieinfo
);
4767 ext4_lblk_t len_blks
;
4770 start_blk
= start
>> inode
->i_sb
->s_blocksize_bits
;
4771 last_blk
= (start
+ len
- 1) >> inode
->i_sb
->s_blocksize_bits
;
4772 if (last_blk
>= EXT_MAX_BLOCKS
)
4773 last_blk
= EXT_MAX_BLOCKS
-1;
4774 len_blks
= ((ext4_lblk_t
) last_blk
) - start_blk
+ 1;
4777 * Walk the extent tree gathering extent information
4778 * and pushing extents back to the user.
4780 error
= ext4_fill_fiemap_extents(inode
, start_blk
,