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 int __ext4_ext_dirty(const char *where
, unsigned int line
, handle_t
*handle
,
161 struct inode
*inode
, struct ext4_ext_path
*path
)
165 ext4_extent_block_csum_set(inode
, ext_block_hdr(path
->p_bh
));
166 /* path points to block */
167 err
= __ext4_handle_dirty_metadata(where
, line
, handle
,
170 /* path points to leaf/index in inode body */
171 err
= ext4_mark_inode_dirty(handle
, inode
);
176 static ext4_fsblk_t
ext4_ext_find_goal(struct inode
*inode
,
177 struct ext4_ext_path
*path
,
181 int depth
= path
->p_depth
;
182 struct ext4_extent
*ex
;
185 * Try to predict block placement assuming that we are
186 * filling in a file which will eventually be
187 * non-sparse --- i.e., in the case of libbfd writing
188 * an ELF object sections out-of-order but in a way
189 * the eventually results in a contiguous object or
190 * executable file, or some database extending a table
191 * space file. However, this is actually somewhat
192 * non-ideal if we are writing a sparse file such as
193 * qemu or KVM writing a raw image file that is going
194 * to stay fairly sparse, since it will end up
195 * fragmenting the file system's free space. Maybe we
196 * should have some hueristics or some way to allow
197 * userspace to pass a hint to file system,
198 * especially if the latter case turns out to be
201 ex
= path
[depth
].p_ext
;
203 ext4_fsblk_t ext_pblk
= ext4_ext_pblock(ex
);
204 ext4_lblk_t ext_block
= le32_to_cpu(ex
->ee_block
);
206 if (block
> ext_block
)
207 return ext_pblk
+ (block
- ext_block
);
209 return ext_pblk
- (ext_block
- block
);
212 /* it looks like index is empty;
213 * try to find starting block from index itself */
214 if (path
[depth
].p_bh
)
215 return path
[depth
].p_bh
->b_blocknr
;
218 /* OK. use inode's group */
219 return ext4_inode_to_goal_block(inode
);
223 * Allocation for a meta data block
226 ext4_ext_new_meta_block(handle_t
*handle
, struct inode
*inode
,
227 struct ext4_ext_path
*path
,
228 struct ext4_extent
*ex
, int *err
, unsigned int flags
)
230 ext4_fsblk_t goal
, newblock
;
232 goal
= ext4_ext_find_goal(inode
, path
, le32_to_cpu(ex
->ee_block
));
233 newblock
= ext4_new_meta_blocks(handle
, inode
, goal
, flags
,
238 static inline int ext4_ext_space_block(struct inode
*inode
, int check
)
242 size
= (inode
->i_sb
->s_blocksize
- sizeof(struct ext4_extent_header
))
243 / sizeof(struct ext4_extent
);
244 #ifdef AGGRESSIVE_TEST
245 if (!check
&& size
> 6)
251 static inline int ext4_ext_space_block_idx(struct inode
*inode
, int check
)
255 size
= (inode
->i_sb
->s_blocksize
- sizeof(struct ext4_extent_header
))
256 / sizeof(struct ext4_extent_idx
);
257 #ifdef AGGRESSIVE_TEST
258 if (!check
&& size
> 5)
264 static inline int ext4_ext_space_root(struct inode
*inode
, int check
)
268 size
= sizeof(EXT4_I(inode
)->i_data
);
269 size
-= sizeof(struct ext4_extent_header
);
270 size
/= sizeof(struct ext4_extent
);
271 #ifdef AGGRESSIVE_TEST
272 if (!check
&& size
> 3)
278 static inline int ext4_ext_space_root_idx(struct inode
*inode
, int check
)
282 size
= sizeof(EXT4_I(inode
)->i_data
);
283 size
-= sizeof(struct ext4_extent_header
);
284 size
/= sizeof(struct ext4_extent_idx
);
285 #ifdef AGGRESSIVE_TEST
286 if (!check
&& size
> 4)
293 * Calculate the number of metadata blocks needed
294 * to allocate @blocks
295 * Worse case is one block per extent
297 int ext4_ext_calc_metadata_amount(struct inode
*inode
, ext4_lblk_t lblock
)
299 struct ext4_inode_info
*ei
= EXT4_I(inode
);
302 idxs
= ((inode
->i_sb
->s_blocksize
- sizeof(struct ext4_extent_header
))
303 / sizeof(struct ext4_extent_idx
));
306 * If the new delayed allocation block is contiguous with the
307 * previous da block, it can share index blocks with the
308 * previous block, so we only need to allocate a new index
309 * block every idxs leaf blocks. At ldxs**2 blocks, we need
310 * an additional index block, and at ldxs**3 blocks, yet
311 * another index blocks.
313 if (ei
->i_da_metadata_calc_len
&&
314 ei
->i_da_metadata_calc_last_lblock
+1 == lblock
) {
317 if ((ei
->i_da_metadata_calc_len
% idxs
) == 0)
319 if ((ei
->i_da_metadata_calc_len
% (idxs
*idxs
)) == 0)
321 if ((ei
->i_da_metadata_calc_len
% (idxs
*idxs
*idxs
)) == 0) {
323 ei
->i_da_metadata_calc_len
= 0;
325 ei
->i_da_metadata_calc_len
++;
326 ei
->i_da_metadata_calc_last_lblock
++;
331 * In the worst case we need a new set of index blocks at
332 * every level of the inode's extent tree.
334 ei
->i_da_metadata_calc_len
= 1;
335 ei
->i_da_metadata_calc_last_lblock
= lblock
;
336 return ext_depth(inode
) + 1;
340 ext4_ext_max_entries(struct inode
*inode
, int depth
)
344 if (depth
== ext_depth(inode
)) {
346 max
= ext4_ext_space_root(inode
, 1);
348 max
= ext4_ext_space_root_idx(inode
, 1);
351 max
= ext4_ext_space_block(inode
, 1);
353 max
= ext4_ext_space_block_idx(inode
, 1);
359 static int ext4_valid_extent(struct inode
*inode
, struct ext4_extent
*ext
)
361 ext4_fsblk_t block
= ext4_ext_pblock(ext
);
362 int len
= ext4_ext_get_actual_len(ext
);
366 return ext4_data_block_valid(EXT4_SB(inode
->i_sb
), block
, len
);
369 static int ext4_valid_extent_idx(struct inode
*inode
,
370 struct ext4_extent_idx
*ext_idx
)
372 ext4_fsblk_t block
= ext4_idx_pblock(ext_idx
);
374 return ext4_data_block_valid(EXT4_SB(inode
->i_sb
), block
, 1);
377 static int ext4_valid_extent_entries(struct inode
*inode
,
378 struct ext4_extent_header
*eh
,
381 unsigned short entries
;
382 if (eh
->eh_entries
== 0)
385 entries
= le16_to_cpu(eh
->eh_entries
);
389 struct ext4_extent
*ext
= EXT_FIRST_EXTENT(eh
);
391 if (!ext4_valid_extent(inode
, ext
))
397 struct ext4_extent_idx
*ext_idx
= EXT_FIRST_INDEX(eh
);
399 if (!ext4_valid_extent_idx(inode
, ext_idx
))
408 static int __ext4_ext_check(const char *function
, unsigned int line
,
409 struct inode
*inode
, struct ext4_extent_header
*eh
,
412 const char *error_msg
;
415 if (unlikely(eh
->eh_magic
!= EXT4_EXT_MAGIC
)) {
416 error_msg
= "invalid magic";
419 if (unlikely(le16_to_cpu(eh
->eh_depth
) != depth
)) {
420 error_msg
= "unexpected eh_depth";
423 if (unlikely(eh
->eh_max
== 0)) {
424 error_msg
= "invalid eh_max";
427 max
= ext4_ext_max_entries(inode
, depth
);
428 if (unlikely(le16_to_cpu(eh
->eh_max
) > max
)) {
429 error_msg
= "too large eh_max";
432 if (unlikely(le16_to_cpu(eh
->eh_entries
) > le16_to_cpu(eh
->eh_max
))) {
433 error_msg
= "invalid eh_entries";
436 if (!ext4_valid_extent_entries(inode
, eh
, depth
)) {
437 error_msg
= "invalid extent entries";
440 /* Verify checksum on non-root extent tree nodes */
441 if (ext_depth(inode
) != depth
&&
442 !ext4_extent_block_csum_verify(inode
, eh
)) {
443 error_msg
= "extent tree corrupted";
449 ext4_error_inode(inode
, function
, line
, 0,
450 "bad header/extent: %s - magic %x, "
451 "entries %u, max %u(%u), depth %u(%u)",
452 error_msg
, le16_to_cpu(eh
->eh_magic
),
453 le16_to_cpu(eh
->eh_entries
), le16_to_cpu(eh
->eh_max
),
454 max
, le16_to_cpu(eh
->eh_depth
), depth
);
459 #define ext4_ext_check(inode, eh, depth) \
460 __ext4_ext_check(__func__, __LINE__, inode, eh, depth)
462 int ext4_ext_check_inode(struct inode
*inode
)
464 return ext4_ext_check(inode
, ext_inode_hdr(inode
), ext_depth(inode
));
467 static struct buffer_head
*
468 __read_extent_tree_block(const char *function
, unsigned int line
,
469 struct inode
*inode
, ext4_fsblk_t pblk
, int depth
)
471 struct buffer_head
*bh
;
474 bh
= sb_getblk(inode
->i_sb
, pblk
);
476 return ERR_PTR(-ENOMEM
);
478 if (!bh_uptodate_or_lock(bh
)) {
479 trace_ext4_ext_load_extent(inode
, pblk
, _RET_IP_
);
480 err
= bh_submit_read(bh
);
484 if (buffer_verified(bh
))
486 err
= __ext4_ext_check(function
, line
, inode
,
487 ext_block_hdr(bh
), depth
);
490 set_buffer_verified(bh
);
498 #define read_extent_tree_block(inode, pblk, depth) \
499 __read_extent_tree_block(__func__, __LINE__, (inode), (pblk), (depth))
502 static void ext4_ext_show_path(struct inode
*inode
, struct ext4_ext_path
*path
)
504 int k
, l
= path
->p_depth
;
507 for (k
= 0; k
<= l
; k
++, path
++) {
509 ext_debug(" %d->%llu", le32_to_cpu(path
->p_idx
->ei_block
),
510 ext4_idx_pblock(path
->p_idx
));
511 } else if (path
->p_ext
) {
512 ext_debug(" %d:[%d]%d:%llu ",
513 le32_to_cpu(path
->p_ext
->ee_block
),
514 ext4_ext_is_uninitialized(path
->p_ext
),
515 ext4_ext_get_actual_len(path
->p_ext
),
516 ext4_ext_pblock(path
->p_ext
));
523 static void ext4_ext_show_leaf(struct inode
*inode
, struct ext4_ext_path
*path
)
525 int depth
= ext_depth(inode
);
526 struct ext4_extent_header
*eh
;
527 struct ext4_extent
*ex
;
533 eh
= path
[depth
].p_hdr
;
534 ex
= EXT_FIRST_EXTENT(eh
);
536 ext_debug("Displaying leaf extents for inode %lu\n", inode
->i_ino
);
538 for (i
= 0; i
< le16_to_cpu(eh
->eh_entries
); i
++, ex
++) {
539 ext_debug("%d:[%d]%d:%llu ", le32_to_cpu(ex
->ee_block
),
540 ext4_ext_is_uninitialized(ex
),
541 ext4_ext_get_actual_len(ex
), ext4_ext_pblock(ex
));
546 static void ext4_ext_show_move(struct inode
*inode
, struct ext4_ext_path
*path
,
547 ext4_fsblk_t newblock
, int level
)
549 int depth
= ext_depth(inode
);
550 struct ext4_extent
*ex
;
552 if (depth
!= level
) {
553 struct ext4_extent_idx
*idx
;
554 idx
= path
[level
].p_idx
;
555 while (idx
<= EXT_MAX_INDEX(path
[level
].p_hdr
)) {
556 ext_debug("%d: move %d:%llu in new index %llu\n", level
,
557 le32_to_cpu(idx
->ei_block
),
558 ext4_idx_pblock(idx
),
566 ex
= path
[depth
].p_ext
;
567 while (ex
<= EXT_MAX_EXTENT(path
[depth
].p_hdr
)) {
568 ext_debug("move %d:%llu:[%d]%d in new leaf %llu\n",
569 le32_to_cpu(ex
->ee_block
),
571 ext4_ext_is_uninitialized(ex
),
572 ext4_ext_get_actual_len(ex
),
579 #define ext4_ext_show_path(inode, path)
580 #define ext4_ext_show_leaf(inode, path)
581 #define ext4_ext_show_move(inode, path, newblock, level)
584 void ext4_ext_drop_refs(struct ext4_ext_path
*path
)
586 int depth
= path
->p_depth
;
589 for (i
= 0; i
<= depth
; i
++, path
++)
597 * ext4_ext_binsearch_idx:
598 * binary search for the closest index of the given block
599 * the header must be checked before calling this
602 ext4_ext_binsearch_idx(struct inode
*inode
,
603 struct ext4_ext_path
*path
, ext4_lblk_t block
)
605 struct ext4_extent_header
*eh
= path
->p_hdr
;
606 struct ext4_extent_idx
*r
, *l
, *m
;
609 ext_debug("binsearch for %u(idx): ", block
);
611 l
= EXT_FIRST_INDEX(eh
) + 1;
612 r
= EXT_LAST_INDEX(eh
);
615 if (block
< le32_to_cpu(m
->ei_block
))
619 ext_debug("%p(%u):%p(%u):%p(%u) ", l
, le32_to_cpu(l
->ei_block
),
620 m
, le32_to_cpu(m
->ei_block
),
621 r
, le32_to_cpu(r
->ei_block
));
625 ext_debug(" -> %u->%lld ", le32_to_cpu(path
->p_idx
->ei_block
),
626 ext4_idx_pblock(path
->p_idx
));
628 #ifdef CHECK_BINSEARCH
630 struct ext4_extent_idx
*chix
, *ix
;
633 chix
= ix
= EXT_FIRST_INDEX(eh
);
634 for (k
= 0; k
< le16_to_cpu(eh
->eh_entries
); k
++, ix
++) {
636 le32_to_cpu(ix
->ei_block
) <= le32_to_cpu(ix
[-1].ei_block
)) {
637 printk(KERN_DEBUG
"k=%d, ix=0x%p, "
639 ix
, EXT_FIRST_INDEX(eh
));
640 printk(KERN_DEBUG
"%u <= %u\n",
641 le32_to_cpu(ix
->ei_block
),
642 le32_to_cpu(ix
[-1].ei_block
));
644 BUG_ON(k
&& le32_to_cpu(ix
->ei_block
)
645 <= le32_to_cpu(ix
[-1].ei_block
));
646 if (block
< le32_to_cpu(ix
->ei_block
))
650 BUG_ON(chix
!= path
->p_idx
);
657 * ext4_ext_binsearch:
658 * binary search for closest extent of the given block
659 * the header must be checked before calling this
662 ext4_ext_binsearch(struct inode
*inode
,
663 struct ext4_ext_path
*path
, ext4_lblk_t block
)
665 struct ext4_extent_header
*eh
= path
->p_hdr
;
666 struct ext4_extent
*r
, *l
, *m
;
668 if (eh
->eh_entries
== 0) {
670 * this leaf is empty:
671 * we get such a leaf in split/add case
676 ext_debug("binsearch for %u: ", block
);
678 l
= EXT_FIRST_EXTENT(eh
) + 1;
679 r
= EXT_LAST_EXTENT(eh
);
683 if (block
< le32_to_cpu(m
->ee_block
))
687 ext_debug("%p(%u):%p(%u):%p(%u) ", l
, le32_to_cpu(l
->ee_block
),
688 m
, le32_to_cpu(m
->ee_block
),
689 r
, le32_to_cpu(r
->ee_block
));
693 ext_debug(" -> %d:%llu:[%d]%d ",
694 le32_to_cpu(path
->p_ext
->ee_block
),
695 ext4_ext_pblock(path
->p_ext
),
696 ext4_ext_is_uninitialized(path
->p_ext
),
697 ext4_ext_get_actual_len(path
->p_ext
));
699 #ifdef CHECK_BINSEARCH
701 struct ext4_extent
*chex
, *ex
;
704 chex
= ex
= EXT_FIRST_EXTENT(eh
);
705 for (k
= 0; k
< le16_to_cpu(eh
->eh_entries
); k
++, ex
++) {
706 BUG_ON(k
&& le32_to_cpu(ex
->ee_block
)
707 <= le32_to_cpu(ex
[-1].ee_block
));
708 if (block
< le32_to_cpu(ex
->ee_block
))
712 BUG_ON(chex
!= path
->p_ext
);
718 int ext4_ext_tree_init(handle_t
*handle
, struct inode
*inode
)
720 struct ext4_extent_header
*eh
;
722 eh
= ext_inode_hdr(inode
);
725 eh
->eh_magic
= EXT4_EXT_MAGIC
;
726 eh
->eh_max
= cpu_to_le16(ext4_ext_space_root(inode
, 0));
727 ext4_mark_inode_dirty(handle
, inode
);
731 struct ext4_ext_path
*
732 ext4_ext_find_extent(struct inode
*inode
, ext4_lblk_t block
,
733 struct ext4_ext_path
*path
)
735 struct ext4_extent_header
*eh
;
736 struct buffer_head
*bh
;
737 short int depth
, i
, ppos
= 0, alloc
= 0;
740 eh
= ext_inode_hdr(inode
);
741 depth
= ext_depth(inode
);
743 /* account possible depth increase */
745 path
= kzalloc(sizeof(struct ext4_ext_path
) * (depth
+ 2),
748 return ERR_PTR(-ENOMEM
);
755 /* walk through the tree */
757 ext_debug("depth %d: num %d, max %d\n",
758 ppos
, le16_to_cpu(eh
->eh_entries
), le16_to_cpu(eh
->eh_max
));
760 ext4_ext_binsearch_idx(inode
, path
+ ppos
, block
);
761 path
[ppos
].p_block
= ext4_idx_pblock(path
[ppos
].p_idx
);
762 path
[ppos
].p_depth
= i
;
763 path
[ppos
].p_ext
= NULL
;
765 bh
= read_extent_tree_block(inode
, path
[ppos
].p_block
, --i
);
771 eh
= ext_block_hdr(bh
);
773 if (unlikely(ppos
> depth
)) {
775 EXT4_ERROR_INODE(inode
,
776 "ppos %d > depth %d", ppos
, depth
);
780 path
[ppos
].p_bh
= bh
;
781 path
[ppos
].p_hdr
= eh
;
784 path
[ppos
].p_depth
= i
;
785 path
[ppos
].p_ext
= NULL
;
786 path
[ppos
].p_idx
= NULL
;
789 ext4_ext_binsearch(inode
, path
+ ppos
, block
);
790 /* if not an empty leaf */
791 if (path
[ppos
].p_ext
)
792 path
[ppos
].p_block
= ext4_ext_pblock(path
[ppos
].p_ext
);
794 ext4_ext_show_path(inode
, path
);
799 ext4_ext_drop_refs(path
);
806 * ext4_ext_insert_index:
807 * insert new index [@logical;@ptr] into the block at @curp;
808 * check where to insert: before @curp or after @curp
810 static int ext4_ext_insert_index(handle_t
*handle
, struct inode
*inode
,
811 struct ext4_ext_path
*curp
,
812 int logical
, ext4_fsblk_t ptr
)
814 struct ext4_extent_idx
*ix
;
817 err
= ext4_ext_get_access(handle
, inode
, curp
);
821 if (unlikely(logical
== le32_to_cpu(curp
->p_idx
->ei_block
))) {
822 EXT4_ERROR_INODE(inode
,
823 "logical %d == ei_block %d!",
824 logical
, le32_to_cpu(curp
->p_idx
->ei_block
));
828 if (unlikely(le16_to_cpu(curp
->p_hdr
->eh_entries
)
829 >= le16_to_cpu(curp
->p_hdr
->eh_max
))) {
830 EXT4_ERROR_INODE(inode
,
831 "eh_entries %d >= eh_max %d!",
832 le16_to_cpu(curp
->p_hdr
->eh_entries
),
833 le16_to_cpu(curp
->p_hdr
->eh_max
));
837 if (logical
> le32_to_cpu(curp
->p_idx
->ei_block
)) {
839 ext_debug("insert new index %d after: %llu\n", logical
, ptr
);
840 ix
= curp
->p_idx
+ 1;
843 ext_debug("insert new index %d before: %llu\n", logical
, ptr
);
847 len
= EXT_LAST_INDEX(curp
->p_hdr
) - ix
+ 1;
850 ext_debug("insert new index %d: "
851 "move %d indices from 0x%p to 0x%p\n",
852 logical
, len
, ix
, ix
+ 1);
853 memmove(ix
+ 1, ix
, len
* sizeof(struct ext4_extent_idx
));
856 if (unlikely(ix
> EXT_MAX_INDEX(curp
->p_hdr
))) {
857 EXT4_ERROR_INODE(inode
, "ix > EXT_MAX_INDEX!");
861 ix
->ei_block
= cpu_to_le32(logical
);
862 ext4_idx_store_pblock(ix
, ptr
);
863 le16_add_cpu(&curp
->p_hdr
->eh_entries
, 1);
865 if (unlikely(ix
> EXT_LAST_INDEX(curp
->p_hdr
))) {
866 EXT4_ERROR_INODE(inode
, "ix > EXT_LAST_INDEX!");
870 err
= ext4_ext_dirty(handle
, inode
, curp
);
871 ext4_std_error(inode
->i_sb
, err
);
878 * inserts new subtree into the path, using free index entry
880 * - allocates all needed blocks (new leaf and all intermediate index blocks)
881 * - makes decision where to split
882 * - moves remaining extents and index entries (right to the split point)
883 * into the newly allocated blocks
884 * - initializes subtree
886 static int ext4_ext_split(handle_t
*handle
, struct inode
*inode
,
888 struct ext4_ext_path
*path
,
889 struct ext4_extent
*newext
, int at
)
891 struct buffer_head
*bh
= NULL
;
892 int depth
= ext_depth(inode
);
893 struct ext4_extent_header
*neh
;
894 struct ext4_extent_idx
*fidx
;
896 ext4_fsblk_t newblock
, oldblock
;
898 ext4_fsblk_t
*ablocks
= NULL
; /* array of allocated blocks */
901 /* make decision: where to split? */
902 /* FIXME: now decision is simplest: at current extent */
904 /* if current leaf will be split, then we should use
905 * border from split point */
906 if (unlikely(path
[depth
].p_ext
> EXT_MAX_EXTENT(path
[depth
].p_hdr
))) {
907 EXT4_ERROR_INODE(inode
, "p_ext > EXT_MAX_EXTENT!");
910 if (path
[depth
].p_ext
!= EXT_MAX_EXTENT(path
[depth
].p_hdr
)) {
911 border
= path
[depth
].p_ext
[1].ee_block
;
912 ext_debug("leaf will be split."
913 " next leaf starts at %d\n",
914 le32_to_cpu(border
));
916 border
= newext
->ee_block
;
917 ext_debug("leaf will be added."
918 " next leaf starts at %d\n",
919 le32_to_cpu(border
));
923 * If error occurs, then we break processing
924 * and mark filesystem read-only. index won't
925 * be inserted and tree will be in consistent
926 * state. Next mount will repair buffers too.
930 * Get array to track all allocated blocks.
931 * We need this to handle errors and free blocks
934 ablocks
= kzalloc(sizeof(ext4_fsblk_t
) * depth
, GFP_NOFS
);
938 /* allocate all needed blocks */
939 ext_debug("allocate %d blocks for indexes/leaf\n", depth
- at
);
940 for (a
= 0; a
< depth
- at
; a
++) {
941 newblock
= ext4_ext_new_meta_block(handle
, inode
, path
,
942 newext
, &err
, flags
);
945 ablocks
[a
] = newblock
;
948 /* initialize new leaf */
949 newblock
= ablocks
[--a
];
950 if (unlikely(newblock
== 0)) {
951 EXT4_ERROR_INODE(inode
, "newblock == 0!");
955 bh
= sb_getblk(inode
->i_sb
, newblock
);
962 err
= ext4_journal_get_create_access(handle
, bh
);
966 neh
= ext_block_hdr(bh
);
968 neh
->eh_max
= cpu_to_le16(ext4_ext_space_block(inode
, 0));
969 neh
->eh_magic
= EXT4_EXT_MAGIC
;
972 /* move remainder of path[depth] to the new leaf */
973 if (unlikely(path
[depth
].p_hdr
->eh_entries
!=
974 path
[depth
].p_hdr
->eh_max
)) {
975 EXT4_ERROR_INODE(inode
, "eh_entries %d != eh_max %d!",
976 path
[depth
].p_hdr
->eh_entries
,
977 path
[depth
].p_hdr
->eh_max
);
981 /* start copy from next extent */
982 m
= EXT_MAX_EXTENT(path
[depth
].p_hdr
) - path
[depth
].p_ext
++;
983 ext4_ext_show_move(inode
, path
, newblock
, depth
);
985 struct ext4_extent
*ex
;
986 ex
= EXT_FIRST_EXTENT(neh
);
987 memmove(ex
, path
[depth
].p_ext
, sizeof(struct ext4_extent
) * m
);
988 le16_add_cpu(&neh
->eh_entries
, m
);
991 ext4_extent_block_csum_set(inode
, neh
);
992 set_buffer_uptodate(bh
);
995 err
= ext4_handle_dirty_metadata(handle
, inode
, bh
);
1001 /* correct old leaf */
1003 err
= ext4_ext_get_access(handle
, inode
, path
+ depth
);
1006 le16_add_cpu(&path
[depth
].p_hdr
->eh_entries
, -m
);
1007 err
= ext4_ext_dirty(handle
, inode
, path
+ depth
);
1013 /* create intermediate indexes */
1015 if (unlikely(k
< 0)) {
1016 EXT4_ERROR_INODE(inode
, "k %d < 0!", k
);
1021 ext_debug("create %d intermediate indices\n", k
);
1022 /* insert new index into current index block */
1023 /* current depth stored in i var */
1026 oldblock
= newblock
;
1027 newblock
= ablocks
[--a
];
1028 bh
= sb_getblk(inode
->i_sb
, newblock
);
1029 if (unlikely(!bh
)) {
1035 err
= ext4_journal_get_create_access(handle
, bh
);
1039 neh
= ext_block_hdr(bh
);
1040 neh
->eh_entries
= cpu_to_le16(1);
1041 neh
->eh_magic
= EXT4_EXT_MAGIC
;
1042 neh
->eh_max
= cpu_to_le16(ext4_ext_space_block_idx(inode
, 0));
1043 neh
->eh_depth
= cpu_to_le16(depth
- i
);
1044 fidx
= EXT_FIRST_INDEX(neh
);
1045 fidx
->ei_block
= border
;
1046 ext4_idx_store_pblock(fidx
, oldblock
);
1048 ext_debug("int.index at %d (block %llu): %u -> %llu\n",
1049 i
, newblock
, le32_to_cpu(border
), oldblock
);
1051 /* move remainder of path[i] to the new index block */
1052 if (unlikely(EXT_MAX_INDEX(path
[i
].p_hdr
) !=
1053 EXT_LAST_INDEX(path
[i
].p_hdr
))) {
1054 EXT4_ERROR_INODE(inode
,
1055 "EXT_MAX_INDEX != EXT_LAST_INDEX ee_block %d!",
1056 le32_to_cpu(path
[i
].p_ext
->ee_block
));
1060 /* start copy indexes */
1061 m
= EXT_MAX_INDEX(path
[i
].p_hdr
) - path
[i
].p_idx
++;
1062 ext_debug("cur 0x%p, last 0x%p\n", path
[i
].p_idx
,
1063 EXT_MAX_INDEX(path
[i
].p_hdr
));
1064 ext4_ext_show_move(inode
, path
, newblock
, i
);
1066 memmove(++fidx
, path
[i
].p_idx
,
1067 sizeof(struct ext4_extent_idx
) * m
);
1068 le16_add_cpu(&neh
->eh_entries
, m
);
1070 ext4_extent_block_csum_set(inode
, neh
);
1071 set_buffer_uptodate(bh
);
1074 err
= ext4_handle_dirty_metadata(handle
, inode
, bh
);
1080 /* correct old index */
1082 err
= ext4_ext_get_access(handle
, inode
, path
+ i
);
1085 le16_add_cpu(&path
[i
].p_hdr
->eh_entries
, -m
);
1086 err
= ext4_ext_dirty(handle
, inode
, path
+ i
);
1094 /* insert new index */
1095 err
= ext4_ext_insert_index(handle
, inode
, path
+ at
,
1096 le32_to_cpu(border
), newblock
);
1100 if (buffer_locked(bh
))
1106 /* free all allocated blocks in error case */
1107 for (i
= 0; i
< depth
; i
++) {
1110 ext4_free_blocks(handle
, inode
, NULL
, ablocks
[i
], 1,
1111 EXT4_FREE_BLOCKS_METADATA
);
1120 * ext4_ext_grow_indepth:
1121 * implements tree growing procedure:
1122 * - allocates new block
1123 * - moves top-level data (index block or leaf) into the new block
1124 * - initializes new top-level, creating index that points to the
1125 * just created block
1127 static int ext4_ext_grow_indepth(handle_t
*handle
, struct inode
*inode
,
1129 struct ext4_extent
*newext
)
1131 struct ext4_extent_header
*neh
;
1132 struct buffer_head
*bh
;
1133 ext4_fsblk_t newblock
;
1136 newblock
= ext4_ext_new_meta_block(handle
, inode
, NULL
,
1137 newext
, &err
, flags
);
1141 bh
= sb_getblk(inode
->i_sb
, newblock
);
1146 err
= ext4_journal_get_create_access(handle
, bh
);
1152 /* move top-level index/leaf into new block */
1153 memmove(bh
->b_data
, EXT4_I(inode
)->i_data
,
1154 sizeof(EXT4_I(inode
)->i_data
));
1156 /* set size of new block */
1157 neh
= ext_block_hdr(bh
);
1158 /* old root could have indexes or leaves
1159 * so calculate e_max right way */
1160 if (ext_depth(inode
))
1161 neh
->eh_max
= cpu_to_le16(ext4_ext_space_block_idx(inode
, 0));
1163 neh
->eh_max
= cpu_to_le16(ext4_ext_space_block(inode
, 0));
1164 neh
->eh_magic
= EXT4_EXT_MAGIC
;
1165 ext4_extent_block_csum_set(inode
, neh
);
1166 set_buffer_uptodate(bh
);
1169 err
= ext4_handle_dirty_metadata(handle
, inode
, bh
);
1173 /* Update top-level index: num,max,pointer */
1174 neh
= ext_inode_hdr(inode
);
1175 neh
->eh_entries
= cpu_to_le16(1);
1176 ext4_idx_store_pblock(EXT_FIRST_INDEX(neh
), newblock
);
1177 if (neh
->eh_depth
== 0) {
1178 /* Root extent block becomes index block */
1179 neh
->eh_max
= cpu_to_le16(ext4_ext_space_root_idx(inode
, 0));
1180 EXT_FIRST_INDEX(neh
)->ei_block
=
1181 EXT_FIRST_EXTENT(neh
)->ee_block
;
1183 ext_debug("new root: num %d(%d), lblock %d, ptr %llu\n",
1184 le16_to_cpu(neh
->eh_entries
), le16_to_cpu(neh
->eh_max
),
1185 le32_to_cpu(EXT_FIRST_INDEX(neh
)->ei_block
),
1186 ext4_idx_pblock(EXT_FIRST_INDEX(neh
)));
1188 le16_add_cpu(&neh
->eh_depth
, 1);
1189 ext4_mark_inode_dirty(handle
, inode
);
1197 * ext4_ext_create_new_leaf:
1198 * finds empty index and adds new leaf.
1199 * if no free index is found, then it requests in-depth growing.
1201 static int ext4_ext_create_new_leaf(handle_t
*handle
, struct inode
*inode
,
1203 struct ext4_ext_path
*path
,
1204 struct ext4_extent
*newext
)
1206 struct ext4_ext_path
*curp
;
1207 int depth
, i
, err
= 0;
1210 i
= depth
= ext_depth(inode
);
1212 /* walk up to the tree and look for free index entry */
1213 curp
= path
+ depth
;
1214 while (i
> 0 && !EXT_HAS_FREE_INDEX(curp
)) {
1219 /* we use already allocated block for index block,
1220 * so subsequent data blocks should be contiguous */
1221 if (EXT_HAS_FREE_INDEX(curp
)) {
1222 /* if we found index with free entry, then use that
1223 * entry: create all needed subtree and add new leaf */
1224 err
= ext4_ext_split(handle
, inode
, flags
, path
, newext
, i
);
1229 ext4_ext_drop_refs(path
);
1230 path
= ext4_ext_find_extent(inode
,
1231 (ext4_lblk_t
)le32_to_cpu(newext
->ee_block
),
1234 err
= PTR_ERR(path
);
1236 /* tree is full, time to grow in depth */
1237 err
= ext4_ext_grow_indepth(handle
, inode
, flags
, newext
);
1242 ext4_ext_drop_refs(path
);
1243 path
= ext4_ext_find_extent(inode
,
1244 (ext4_lblk_t
)le32_to_cpu(newext
->ee_block
),
1247 err
= PTR_ERR(path
);
1252 * only first (depth 0 -> 1) produces free space;
1253 * in all other cases we have to split the grown tree
1255 depth
= ext_depth(inode
);
1256 if (path
[depth
].p_hdr
->eh_entries
== path
[depth
].p_hdr
->eh_max
) {
1257 /* now we need to split */
1267 * search the closest allocated block to the left for *logical
1268 * and returns it at @logical + it's physical address at @phys
1269 * if *logical is the smallest allocated block, the function
1270 * returns 0 at @phys
1271 * return value contains 0 (success) or error code
1273 static int ext4_ext_search_left(struct inode
*inode
,
1274 struct ext4_ext_path
*path
,
1275 ext4_lblk_t
*logical
, ext4_fsblk_t
*phys
)
1277 struct ext4_extent_idx
*ix
;
1278 struct ext4_extent
*ex
;
1281 if (unlikely(path
== NULL
)) {
1282 EXT4_ERROR_INODE(inode
, "path == NULL *logical %d!", *logical
);
1285 depth
= path
->p_depth
;
1288 if (depth
== 0 && path
->p_ext
== NULL
)
1291 /* usually extent in the path covers blocks smaller
1292 * then *logical, but it can be that extent is the
1293 * first one in the file */
1295 ex
= path
[depth
].p_ext
;
1296 ee_len
= ext4_ext_get_actual_len(ex
);
1297 if (*logical
< le32_to_cpu(ex
->ee_block
)) {
1298 if (unlikely(EXT_FIRST_EXTENT(path
[depth
].p_hdr
) != ex
)) {
1299 EXT4_ERROR_INODE(inode
,
1300 "EXT_FIRST_EXTENT != ex *logical %d ee_block %d!",
1301 *logical
, le32_to_cpu(ex
->ee_block
));
1304 while (--depth
>= 0) {
1305 ix
= path
[depth
].p_idx
;
1306 if (unlikely(ix
!= EXT_FIRST_INDEX(path
[depth
].p_hdr
))) {
1307 EXT4_ERROR_INODE(inode
,
1308 "ix (%d) != EXT_FIRST_INDEX (%d) (depth %d)!",
1309 ix
!= NULL
? le32_to_cpu(ix
->ei_block
) : 0,
1310 EXT_FIRST_INDEX(path
[depth
].p_hdr
) != NULL
?
1311 le32_to_cpu(EXT_FIRST_INDEX(path
[depth
].p_hdr
)->ei_block
) : 0,
1319 if (unlikely(*logical
< (le32_to_cpu(ex
->ee_block
) + ee_len
))) {
1320 EXT4_ERROR_INODE(inode
,
1321 "logical %d < ee_block %d + ee_len %d!",
1322 *logical
, le32_to_cpu(ex
->ee_block
), ee_len
);
1326 *logical
= le32_to_cpu(ex
->ee_block
) + ee_len
- 1;
1327 *phys
= ext4_ext_pblock(ex
) + ee_len
- 1;
1332 * search the closest allocated block to the right for *logical
1333 * and returns it at @logical + it's physical address at @phys
1334 * if *logical is the largest allocated block, the function
1335 * returns 0 at @phys
1336 * return value contains 0 (success) or error code
1338 static int ext4_ext_search_right(struct inode
*inode
,
1339 struct ext4_ext_path
*path
,
1340 ext4_lblk_t
*logical
, ext4_fsblk_t
*phys
,
1341 struct ext4_extent
**ret_ex
)
1343 struct buffer_head
*bh
= NULL
;
1344 struct ext4_extent_header
*eh
;
1345 struct ext4_extent_idx
*ix
;
1346 struct ext4_extent
*ex
;
1348 int depth
; /* Note, NOT eh_depth; depth from top of tree */
1351 if (unlikely(path
== NULL
)) {
1352 EXT4_ERROR_INODE(inode
, "path == NULL *logical %d!", *logical
);
1355 depth
= path
->p_depth
;
1358 if (depth
== 0 && path
->p_ext
== NULL
)
1361 /* usually extent in the path covers blocks smaller
1362 * then *logical, but it can be that extent is the
1363 * first one in the file */
1365 ex
= path
[depth
].p_ext
;
1366 ee_len
= ext4_ext_get_actual_len(ex
);
1367 if (*logical
< le32_to_cpu(ex
->ee_block
)) {
1368 if (unlikely(EXT_FIRST_EXTENT(path
[depth
].p_hdr
) != ex
)) {
1369 EXT4_ERROR_INODE(inode
,
1370 "first_extent(path[%d].p_hdr) != ex",
1374 while (--depth
>= 0) {
1375 ix
= path
[depth
].p_idx
;
1376 if (unlikely(ix
!= EXT_FIRST_INDEX(path
[depth
].p_hdr
))) {
1377 EXT4_ERROR_INODE(inode
,
1378 "ix != EXT_FIRST_INDEX *logical %d!",
1386 if (unlikely(*logical
< (le32_to_cpu(ex
->ee_block
) + ee_len
))) {
1387 EXT4_ERROR_INODE(inode
,
1388 "logical %d < ee_block %d + ee_len %d!",
1389 *logical
, le32_to_cpu(ex
->ee_block
), ee_len
);
1393 if (ex
!= EXT_LAST_EXTENT(path
[depth
].p_hdr
)) {
1394 /* next allocated block in this leaf */
1399 /* go up and search for index to the right */
1400 while (--depth
>= 0) {
1401 ix
= path
[depth
].p_idx
;
1402 if (ix
!= EXT_LAST_INDEX(path
[depth
].p_hdr
))
1406 /* we've gone up to the root and found no index to the right */
1410 /* we've found index to the right, let's
1411 * follow it and find the closest allocated
1412 * block to the right */
1414 block
= ext4_idx_pblock(ix
);
1415 while (++depth
< path
->p_depth
) {
1416 /* subtract from p_depth to get proper eh_depth */
1417 bh
= read_extent_tree_block(inode
, block
,
1418 path
->p_depth
- depth
);
1421 eh
= ext_block_hdr(bh
);
1422 ix
= EXT_FIRST_INDEX(eh
);
1423 block
= ext4_idx_pblock(ix
);
1427 bh
= read_extent_tree_block(inode
, block
, path
->p_depth
- depth
);
1430 eh
= ext_block_hdr(bh
);
1431 ex
= EXT_FIRST_EXTENT(eh
);
1433 *logical
= le32_to_cpu(ex
->ee_block
);
1434 *phys
= ext4_ext_pblock(ex
);
1442 * ext4_ext_next_allocated_block:
1443 * returns allocated block in subsequent extent or EXT_MAX_BLOCKS.
1444 * NOTE: it considers block number from index entry as
1445 * allocated block. Thus, index entries have to be consistent
1449 ext4_ext_next_allocated_block(struct ext4_ext_path
*path
)
1453 BUG_ON(path
== NULL
);
1454 depth
= path
->p_depth
;
1456 if (depth
== 0 && path
->p_ext
== NULL
)
1457 return EXT_MAX_BLOCKS
;
1459 while (depth
>= 0) {
1460 if (depth
== path
->p_depth
) {
1462 if (path
[depth
].p_ext
&&
1463 path
[depth
].p_ext
!=
1464 EXT_LAST_EXTENT(path
[depth
].p_hdr
))
1465 return le32_to_cpu(path
[depth
].p_ext
[1].ee_block
);
1468 if (path
[depth
].p_idx
!=
1469 EXT_LAST_INDEX(path
[depth
].p_hdr
))
1470 return le32_to_cpu(path
[depth
].p_idx
[1].ei_block
);
1475 return EXT_MAX_BLOCKS
;
1479 * ext4_ext_next_leaf_block:
1480 * returns first allocated block from next leaf or EXT_MAX_BLOCKS
1482 static ext4_lblk_t
ext4_ext_next_leaf_block(struct ext4_ext_path
*path
)
1486 BUG_ON(path
== NULL
);
1487 depth
= path
->p_depth
;
1489 /* zero-tree has no leaf blocks at all */
1491 return EXT_MAX_BLOCKS
;
1493 /* go to index block */
1496 while (depth
>= 0) {
1497 if (path
[depth
].p_idx
!=
1498 EXT_LAST_INDEX(path
[depth
].p_hdr
))
1499 return (ext4_lblk_t
)
1500 le32_to_cpu(path
[depth
].p_idx
[1].ei_block
);
1504 return EXT_MAX_BLOCKS
;
1508 * ext4_ext_correct_indexes:
1509 * if leaf gets modified and modified extent is first in the leaf,
1510 * then we have to correct all indexes above.
1511 * TODO: do we need to correct tree in all cases?
1513 static int ext4_ext_correct_indexes(handle_t
*handle
, struct inode
*inode
,
1514 struct ext4_ext_path
*path
)
1516 struct ext4_extent_header
*eh
;
1517 int depth
= ext_depth(inode
);
1518 struct ext4_extent
*ex
;
1522 eh
= path
[depth
].p_hdr
;
1523 ex
= path
[depth
].p_ext
;
1525 if (unlikely(ex
== NULL
|| eh
== NULL
)) {
1526 EXT4_ERROR_INODE(inode
,
1527 "ex %p == NULL or eh %p == NULL", ex
, eh
);
1532 /* there is no tree at all */
1536 if (ex
!= EXT_FIRST_EXTENT(eh
)) {
1537 /* we correct tree if first leaf got modified only */
1542 * TODO: we need correction if border is smaller than current one
1545 border
= path
[depth
].p_ext
->ee_block
;
1546 err
= ext4_ext_get_access(handle
, inode
, path
+ k
);
1549 path
[k
].p_idx
->ei_block
= border
;
1550 err
= ext4_ext_dirty(handle
, inode
, path
+ k
);
1555 /* change all left-side indexes */
1556 if (path
[k
+1].p_idx
!= EXT_FIRST_INDEX(path
[k
+1].p_hdr
))
1558 err
= ext4_ext_get_access(handle
, inode
, path
+ k
);
1561 path
[k
].p_idx
->ei_block
= border
;
1562 err
= ext4_ext_dirty(handle
, inode
, path
+ k
);
1571 ext4_can_extents_be_merged(struct inode
*inode
, struct ext4_extent
*ex1
,
1572 struct ext4_extent
*ex2
)
1574 unsigned short ext1_ee_len
, ext2_ee_len
, max_len
;
1577 * Make sure that both extents are initialized. We don't merge
1578 * uninitialized extents so that we can be sure that end_io code has
1579 * the extent that was written properly split out and conversion to
1580 * initialized is trivial.
1582 if (ext4_ext_is_uninitialized(ex1
) || ext4_ext_is_uninitialized(ex2
))
1585 if (ext4_ext_is_uninitialized(ex1
))
1586 max_len
= EXT_UNINIT_MAX_LEN
;
1588 max_len
= EXT_INIT_MAX_LEN
;
1590 ext1_ee_len
= ext4_ext_get_actual_len(ex1
);
1591 ext2_ee_len
= ext4_ext_get_actual_len(ex2
);
1593 if (le32_to_cpu(ex1
->ee_block
) + ext1_ee_len
!=
1594 le32_to_cpu(ex2
->ee_block
))
1598 * To allow future support for preallocated extents to be added
1599 * as an RO_COMPAT feature, refuse to merge to extents if
1600 * this can result in the top bit of ee_len being set.
1602 if (ext1_ee_len
+ ext2_ee_len
> max_len
)
1604 #ifdef AGGRESSIVE_TEST
1605 if (ext1_ee_len
>= 4)
1609 if (ext4_ext_pblock(ex1
) + ext1_ee_len
== ext4_ext_pblock(ex2
))
1615 * This function tries to merge the "ex" extent to the next extent in the tree.
1616 * It always tries to merge towards right. If you want to merge towards
1617 * left, pass "ex - 1" as argument instead of "ex".
1618 * Returns 0 if the extents (ex and ex+1) were _not_ merged and returns
1619 * 1 if they got merged.
1621 static int ext4_ext_try_to_merge_right(struct inode
*inode
,
1622 struct ext4_ext_path
*path
,
1623 struct ext4_extent
*ex
)
1625 struct ext4_extent_header
*eh
;
1626 unsigned int depth
, len
;
1628 int uninitialized
= 0;
1630 depth
= ext_depth(inode
);
1631 BUG_ON(path
[depth
].p_hdr
== NULL
);
1632 eh
= path
[depth
].p_hdr
;
1634 while (ex
< EXT_LAST_EXTENT(eh
)) {
1635 if (!ext4_can_extents_be_merged(inode
, ex
, ex
+ 1))
1637 /* merge with next extent! */
1638 if (ext4_ext_is_uninitialized(ex
))
1640 ex
->ee_len
= cpu_to_le16(ext4_ext_get_actual_len(ex
)
1641 + ext4_ext_get_actual_len(ex
+ 1));
1643 ext4_ext_mark_uninitialized(ex
);
1645 if (ex
+ 1 < EXT_LAST_EXTENT(eh
)) {
1646 len
= (EXT_LAST_EXTENT(eh
) - ex
- 1)
1647 * sizeof(struct ext4_extent
);
1648 memmove(ex
+ 1, ex
+ 2, len
);
1650 le16_add_cpu(&eh
->eh_entries
, -1);
1652 WARN_ON(eh
->eh_entries
== 0);
1653 if (!eh
->eh_entries
)
1654 EXT4_ERROR_INODE(inode
, "eh->eh_entries = 0!");
1661 * This function does a very simple check to see if we can collapse
1662 * an extent tree with a single extent tree leaf block into the inode.
1664 static void ext4_ext_try_to_merge_up(handle_t
*handle
,
1665 struct inode
*inode
,
1666 struct ext4_ext_path
*path
)
1669 unsigned max_root
= ext4_ext_space_root(inode
, 0);
1672 if ((path
[0].p_depth
!= 1) ||
1673 (le16_to_cpu(path
[0].p_hdr
->eh_entries
) != 1) ||
1674 (le16_to_cpu(path
[1].p_hdr
->eh_entries
) > max_root
))
1678 * We need to modify the block allocation bitmap and the block
1679 * group descriptor to release the extent tree block. If we
1680 * can't get the journal credits, give up.
1682 if (ext4_journal_extend(handle
, 2))
1686 * Copy the extent data up to the inode
1688 blk
= ext4_idx_pblock(path
[0].p_idx
);
1689 s
= le16_to_cpu(path
[1].p_hdr
->eh_entries
) *
1690 sizeof(struct ext4_extent_idx
);
1691 s
+= sizeof(struct ext4_extent_header
);
1693 memcpy(path
[0].p_hdr
, path
[1].p_hdr
, s
);
1694 path
[0].p_depth
= 0;
1695 path
[0].p_ext
= EXT_FIRST_EXTENT(path
[0].p_hdr
) +
1696 (path
[1].p_ext
- EXT_FIRST_EXTENT(path
[1].p_hdr
));
1697 path
[0].p_hdr
->eh_max
= cpu_to_le16(max_root
);
1699 brelse(path
[1].p_bh
);
1700 ext4_free_blocks(handle
, inode
, NULL
, blk
, 1,
1701 EXT4_FREE_BLOCKS_METADATA
| EXT4_FREE_BLOCKS_FORGET
);
1705 * This function tries to merge the @ex extent to neighbours in the tree.
1706 * return 1 if merge left else 0.
1708 static void ext4_ext_try_to_merge(handle_t
*handle
,
1709 struct inode
*inode
,
1710 struct ext4_ext_path
*path
,
1711 struct ext4_extent
*ex
) {
1712 struct ext4_extent_header
*eh
;
1716 depth
= ext_depth(inode
);
1717 BUG_ON(path
[depth
].p_hdr
== NULL
);
1718 eh
= path
[depth
].p_hdr
;
1720 if (ex
> EXT_FIRST_EXTENT(eh
))
1721 merge_done
= ext4_ext_try_to_merge_right(inode
, path
, ex
- 1);
1724 (void) ext4_ext_try_to_merge_right(inode
, path
, ex
);
1726 ext4_ext_try_to_merge_up(handle
, inode
, path
);
1730 * check if a portion of the "newext" extent overlaps with an
1733 * If there is an overlap discovered, it updates the length of the newext
1734 * such that there will be no overlap, and then returns 1.
1735 * If there is no overlap found, it returns 0.
1737 static unsigned int ext4_ext_check_overlap(struct ext4_sb_info
*sbi
,
1738 struct inode
*inode
,
1739 struct ext4_extent
*newext
,
1740 struct ext4_ext_path
*path
)
1743 unsigned int depth
, len1
;
1744 unsigned int ret
= 0;
1746 b1
= le32_to_cpu(newext
->ee_block
);
1747 len1
= ext4_ext_get_actual_len(newext
);
1748 depth
= ext_depth(inode
);
1749 if (!path
[depth
].p_ext
)
1751 b2
= le32_to_cpu(path
[depth
].p_ext
->ee_block
);
1752 b2
&= ~(sbi
->s_cluster_ratio
- 1);
1755 * get the next allocated block if the extent in the path
1756 * is before the requested block(s)
1759 b2
= ext4_ext_next_allocated_block(path
);
1760 if (b2
== EXT_MAX_BLOCKS
)
1762 b2
&= ~(sbi
->s_cluster_ratio
- 1);
1765 /* check for wrap through zero on extent logical start block*/
1766 if (b1
+ len1
< b1
) {
1767 len1
= EXT_MAX_BLOCKS
- b1
;
1768 newext
->ee_len
= cpu_to_le16(len1
);
1772 /* check for overlap */
1773 if (b1
+ len1
> b2
) {
1774 newext
->ee_len
= cpu_to_le16(b2
- b1
);
1782 * ext4_ext_insert_extent:
1783 * tries to merge requsted extent into the existing extent or
1784 * inserts requested extent as new one into the tree,
1785 * creating new leaf in the no-space case.
1787 int ext4_ext_insert_extent(handle_t
*handle
, struct inode
*inode
,
1788 struct ext4_ext_path
*path
,
1789 struct ext4_extent
*newext
, int flag
)
1791 struct ext4_extent_header
*eh
;
1792 struct ext4_extent
*ex
, *fex
;
1793 struct ext4_extent
*nearex
; /* nearest extent */
1794 struct ext4_ext_path
*npath
= NULL
;
1795 int depth
, len
, err
;
1797 unsigned uninitialized
= 0;
1800 if (unlikely(ext4_ext_get_actual_len(newext
) == 0)) {
1801 EXT4_ERROR_INODE(inode
, "ext4_ext_get_actual_len(newext) == 0");
1804 depth
= ext_depth(inode
);
1805 ex
= path
[depth
].p_ext
;
1806 eh
= path
[depth
].p_hdr
;
1807 if (unlikely(path
[depth
].p_hdr
== NULL
)) {
1808 EXT4_ERROR_INODE(inode
, "path[%d].p_hdr == NULL", depth
);
1812 /* try to insert block into found extent and return */
1813 if (ex
&& !(flag
& EXT4_GET_BLOCKS_PRE_IO
)) {
1816 * Try to see whether we should rather test the extent on
1817 * right from ex, or from the left of ex. This is because
1818 * ext4_ext_find_extent() can return either extent on the
1819 * left, or on the right from the searched position. This
1820 * will make merging more effective.
1822 if (ex
< EXT_LAST_EXTENT(eh
) &&
1823 (le32_to_cpu(ex
->ee_block
) +
1824 ext4_ext_get_actual_len(ex
) <
1825 le32_to_cpu(newext
->ee_block
))) {
1828 } else if ((ex
> EXT_FIRST_EXTENT(eh
)) &&
1829 (le32_to_cpu(newext
->ee_block
) +
1830 ext4_ext_get_actual_len(newext
) <
1831 le32_to_cpu(ex
->ee_block
)))
1834 /* Try to append newex to the ex */
1835 if (ext4_can_extents_be_merged(inode
, ex
, newext
)) {
1836 ext_debug("append [%d]%d block to %u:[%d]%d"
1838 ext4_ext_is_uninitialized(newext
),
1839 ext4_ext_get_actual_len(newext
),
1840 le32_to_cpu(ex
->ee_block
),
1841 ext4_ext_is_uninitialized(ex
),
1842 ext4_ext_get_actual_len(ex
),
1843 ext4_ext_pblock(ex
));
1844 err
= ext4_ext_get_access(handle
, inode
,
1850 * ext4_can_extents_be_merged should have checked
1851 * that either both extents are uninitialized, or
1852 * both aren't. Thus we need to check only one of
1855 if (ext4_ext_is_uninitialized(ex
))
1857 ex
->ee_len
= cpu_to_le16(ext4_ext_get_actual_len(ex
)
1858 + ext4_ext_get_actual_len(newext
));
1860 ext4_ext_mark_uninitialized(ex
);
1861 eh
= path
[depth
].p_hdr
;
1867 /* Try to prepend newex to the ex */
1868 if (ext4_can_extents_be_merged(inode
, newext
, ex
)) {
1869 ext_debug("prepend %u[%d]%d block to %u:[%d]%d"
1871 le32_to_cpu(newext
->ee_block
),
1872 ext4_ext_is_uninitialized(newext
),
1873 ext4_ext_get_actual_len(newext
),
1874 le32_to_cpu(ex
->ee_block
),
1875 ext4_ext_is_uninitialized(ex
),
1876 ext4_ext_get_actual_len(ex
),
1877 ext4_ext_pblock(ex
));
1878 err
= ext4_ext_get_access(handle
, inode
,
1884 * ext4_can_extents_be_merged should have checked
1885 * that either both extents are uninitialized, or
1886 * both aren't. Thus we need to check only one of
1889 if (ext4_ext_is_uninitialized(ex
))
1891 ex
->ee_block
= newext
->ee_block
;
1892 ext4_ext_store_pblock(ex
, ext4_ext_pblock(newext
));
1893 ex
->ee_len
= cpu_to_le16(ext4_ext_get_actual_len(ex
)
1894 + ext4_ext_get_actual_len(newext
));
1896 ext4_ext_mark_uninitialized(ex
);
1897 eh
= path
[depth
].p_hdr
;
1903 depth
= ext_depth(inode
);
1904 eh
= path
[depth
].p_hdr
;
1905 if (le16_to_cpu(eh
->eh_entries
) < le16_to_cpu(eh
->eh_max
))
1908 /* probably next leaf has space for us? */
1909 fex
= EXT_LAST_EXTENT(eh
);
1910 next
= EXT_MAX_BLOCKS
;
1911 if (le32_to_cpu(newext
->ee_block
) > le32_to_cpu(fex
->ee_block
))
1912 next
= ext4_ext_next_leaf_block(path
);
1913 if (next
!= EXT_MAX_BLOCKS
) {
1914 ext_debug("next leaf block - %u\n", next
);
1915 BUG_ON(npath
!= NULL
);
1916 npath
= ext4_ext_find_extent(inode
, next
, NULL
);
1918 return PTR_ERR(npath
);
1919 BUG_ON(npath
->p_depth
!= path
->p_depth
);
1920 eh
= npath
[depth
].p_hdr
;
1921 if (le16_to_cpu(eh
->eh_entries
) < le16_to_cpu(eh
->eh_max
)) {
1922 ext_debug("next leaf isn't full(%d)\n",
1923 le16_to_cpu(eh
->eh_entries
));
1927 ext_debug("next leaf has no free space(%d,%d)\n",
1928 le16_to_cpu(eh
->eh_entries
), le16_to_cpu(eh
->eh_max
));
1932 * There is no free space in the found leaf.
1933 * We're gonna add a new leaf in the tree.
1935 if (flag
& EXT4_GET_BLOCKS_METADATA_NOFAIL
)
1936 flags
= EXT4_MB_USE_RESERVED
;
1937 err
= ext4_ext_create_new_leaf(handle
, inode
, flags
, path
, newext
);
1940 depth
= ext_depth(inode
);
1941 eh
= path
[depth
].p_hdr
;
1944 nearex
= path
[depth
].p_ext
;
1946 err
= ext4_ext_get_access(handle
, inode
, path
+ depth
);
1951 /* there is no extent in this leaf, create first one */
1952 ext_debug("first extent in the leaf: %u:%llu:[%d]%d\n",
1953 le32_to_cpu(newext
->ee_block
),
1954 ext4_ext_pblock(newext
),
1955 ext4_ext_is_uninitialized(newext
),
1956 ext4_ext_get_actual_len(newext
));
1957 nearex
= EXT_FIRST_EXTENT(eh
);
1959 if (le32_to_cpu(newext
->ee_block
)
1960 > le32_to_cpu(nearex
->ee_block
)) {
1962 ext_debug("insert %u:%llu:[%d]%d before: "
1964 le32_to_cpu(newext
->ee_block
),
1965 ext4_ext_pblock(newext
),
1966 ext4_ext_is_uninitialized(newext
),
1967 ext4_ext_get_actual_len(newext
),
1972 BUG_ON(newext
->ee_block
== nearex
->ee_block
);
1973 ext_debug("insert %u:%llu:[%d]%d after: "
1975 le32_to_cpu(newext
->ee_block
),
1976 ext4_ext_pblock(newext
),
1977 ext4_ext_is_uninitialized(newext
),
1978 ext4_ext_get_actual_len(newext
),
1981 len
= EXT_LAST_EXTENT(eh
) - nearex
+ 1;
1983 ext_debug("insert %u:%llu:[%d]%d: "
1984 "move %d extents from 0x%p to 0x%p\n",
1985 le32_to_cpu(newext
->ee_block
),
1986 ext4_ext_pblock(newext
),
1987 ext4_ext_is_uninitialized(newext
),
1988 ext4_ext_get_actual_len(newext
),
1989 len
, nearex
, nearex
+ 1);
1990 memmove(nearex
+ 1, nearex
,
1991 len
* sizeof(struct ext4_extent
));
1995 le16_add_cpu(&eh
->eh_entries
, 1);
1996 path
[depth
].p_ext
= nearex
;
1997 nearex
->ee_block
= newext
->ee_block
;
1998 ext4_ext_store_pblock(nearex
, ext4_ext_pblock(newext
));
1999 nearex
->ee_len
= newext
->ee_len
;
2002 /* try to merge extents */
2003 if (!(flag
& EXT4_GET_BLOCKS_PRE_IO
))
2004 ext4_ext_try_to_merge(handle
, inode
, path
, nearex
);
2007 /* time to correct all indexes above */
2008 err
= ext4_ext_correct_indexes(handle
, inode
, path
);
2012 err
= ext4_ext_dirty(handle
, inode
, path
+ path
->p_depth
);
2016 ext4_ext_drop_refs(npath
);
2022 static int ext4_fill_fiemap_extents(struct inode
*inode
,
2023 ext4_lblk_t block
, ext4_lblk_t num
,
2024 struct fiemap_extent_info
*fieinfo
)
2026 struct ext4_ext_path
*path
= NULL
;
2027 struct ext4_extent
*ex
;
2028 struct extent_status es
;
2029 ext4_lblk_t next
, next_del
, start
= 0, end
= 0;
2030 ext4_lblk_t last
= block
+ num
;
2031 int exists
, depth
= 0, err
= 0;
2032 unsigned int flags
= 0;
2033 unsigned char blksize_bits
= inode
->i_sb
->s_blocksize_bits
;
2035 while (block
< last
&& block
!= EXT_MAX_BLOCKS
) {
2037 /* find extent for this block */
2038 down_read(&EXT4_I(inode
)->i_data_sem
);
2040 if (path
&& ext_depth(inode
) != depth
) {
2041 /* depth was changed. we have to realloc path */
2046 path
= ext4_ext_find_extent(inode
, block
, path
);
2048 up_read(&EXT4_I(inode
)->i_data_sem
);
2049 err
= PTR_ERR(path
);
2054 depth
= ext_depth(inode
);
2055 if (unlikely(path
[depth
].p_hdr
== NULL
)) {
2056 up_read(&EXT4_I(inode
)->i_data_sem
);
2057 EXT4_ERROR_INODE(inode
, "path[%d].p_hdr == NULL", depth
);
2061 ex
= path
[depth
].p_ext
;
2062 next
= ext4_ext_next_allocated_block(path
);
2063 ext4_ext_drop_refs(path
);
2068 /* there is no extent yet, so try to allocate
2069 * all requested space */
2072 } else if (le32_to_cpu(ex
->ee_block
) > block
) {
2073 /* need to allocate space before found extent */
2075 end
= le32_to_cpu(ex
->ee_block
);
2076 if (block
+ num
< end
)
2078 } else if (block
>= le32_to_cpu(ex
->ee_block
)
2079 + ext4_ext_get_actual_len(ex
)) {
2080 /* need to allocate space after found extent */
2085 } else if (block
>= le32_to_cpu(ex
->ee_block
)) {
2087 * some part of requested space is covered
2091 end
= le32_to_cpu(ex
->ee_block
)
2092 + ext4_ext_get_actual_len(ex
);
2093 if (block
+ num
< end
)
2099 BUG_ON(end
<= start
);
2103 es
.es_len
= end
- start
;
2106 es
.es_lblk
= le32_to_cpu(ex
->ee_block
);
2107 es
.es_len
= ext4_ext_get_actual_len(ex
);
2108 es
.es_pblk
= ext4_ext_pblock(ex
);
2109 if (ext4_ext_is_uninitialized(ex
))
2110 flags
|= FIEMAP_EXTENT_UNWRITTEN
;
2114 * Find delayed extent and update es accordingly. We call
2115 * it even in !exists case to find out whether es is the
2116 * last existing extent or not.
2118 next_del
= ext4_find_delayed_extent(inode
, &es
);
2119 if (!exists
&& next_del
) {
2121 flags
|= (FIEMAP_EXTENT_DELALLOC
|
2122 FIEMAP_EXTENT_UNKNOWN
);
2124 up_read(&EXT4_I(inode
)->i_data_sem
);
2126 if (unlikely(es
.es_len
== 0)) {
2127 EXT4_ERROR_INODE(inode
, "es.es_len == 0");
2133 * This is possible iff next == next_del == EXT_MAX_BLOCKS.
2134 * we need to check next == EXT_MAX_BLOCKS because it is
2135 * possible that an extent is with unwritten and delayed
2136 * status due to when an extent is delayed allocated and
2137 * is allocated by fallocate status tree will track both of
2140 * So we could return a unwritten and delayed extent, and
2141 * its block is equal to 'next'.
2143 if (next
== next_del
&& next
== EXT_MAX_BLOCKS
) {
2144 flags
|= FIEMAP_EXTENT_LAST
;
2145 if (unlikely(next_del
!= EXT_MAX_BLOCKS
||
2146 next
!= EXT_MAX_BLOCKS
)) {
2147 EXT4_ERROR_INODE(inode
,
2148 "next extent == %u, next "
2149 "delalloc extent = %u",
2157 err
= fiemap_fill_next_extent(fieinfo
,
2158 (__u64
)es
.es_lblk
<< blksize_bits
,
2159 (__u64
)es
.es_pblk
<< blksize_bits
,
2160 (__u64
)es
.es_len
<< blksize_bits
,
2170 block
= es
.es_lblk
+ es
.es_len
;
2174 ext4_ext_drop_refs(path
);
2182 * ext4_ext_put_gap_in_cache:
2183 * calculate boundaries of the gap that the requested block fits into
2184 * and cache this gap
2187 ext4_ext_put_gap_in_cache(struct inode
*inode
, struct ext4_ext_path
*path
,
2190 int depth
= ext_depth(inode
);
2193 struct ext4_extent
*ex
;
2195 ex
= path
[depth
].p_ext
;
2198 * there is no extent yet, so gap is [0;-] and we
2201 ext_debug("cache gap(whole file):");
2202 } else if (block
< le32_to_cpu(ex
->ee_block
)) {
2204 len
= le32_to_cpu(ex
->ee_block
) - block
;
2205 ext_debug("cache gap(before): %u [%u:%u]",
2207 le32_to_cpu(ex
->ee_block
),
2208 ext4_ext_get_actual_len(ex
));
2209 if (!ext4_find_delalloc_range(inode
, lblock
, lblock
+ len
- 1))
2210 ext4_es_insert_extent(inode
, lblock
, len
, ~0,
2211 EXTENT_STATUS_HOLE
);
2212 } else if (block
>= le32_to_cpu(ex
->ee_block
)
2213 + ext4_ext_get_actual_len(ex
)) {
2215 lblock
= le32_to_cpu(ex
->ee_block
)
2216 + ext4_ext_get_actual_len(ex
);
2218 next
= ext4_ext_next_allocated_block(path
);
2219 ext_debug("cache gap(after): [%u:%u] %u",
2220 le32_to_cpu(ex
->ee_block
),
2221 ext4_ext_get_actual_len(ex
),
2223 BUG_ON(next
== lblock
);
2224 len
= next
- lblock
;
2225 if (!ext4_find_delalloc_range(inode
, lblock
, lblock
+ len
- 1))
2226 ext4_es_insert_extent(inode
, lblock
, len
, ~0,
2227 EXTENT_STATUS_HOLE
);
2233 ext_debug(" -> %u:%lu\n", lblock
, len
);
2238 * removes index from the index block.
2240 static int ext4_ext_rm_idx(handle_t
*handle
, struct inode
*inode
,
2241 struct ext4_ext_path
*path
, int depth
)
2246 /* free index block */
2248 path
= path
+ depth
;
2249 leaf
= ext4_idx_pblock(path
->p_idx
);
2250 if (unlikely(path
->p_hdr
->eh_entries
== 0)) {
2251 EXT4_ERROR_INODE(inode
, "path->p_hdr->eh_entries == 0");
2254 err
= ext4_ext_get_access(handle
, inode
, path
);
2258 if (path
->p_idx
!= EXT_LAST_INDEX(path
->p_hdr
)) {
2259 int len
= EXT_LAST_INDEX(path
->p_hdr
) - path
->p_idx
;
2260 len
*= sizeof(struct ext4_extent_idx
);
2261 memmove(path
->p_idx
, path
->p_idx
+ 1, len
);
2264 le16_add_cpu(&path
->p_hdr
->eh_entries
, -1);
2265 err
= ext4_ext_dirty(handle
, inode
, path
);
2268 ext_debug("index is empty, remove it, free block %llu\n", leaf
);
2269 trace_ext4_ext_rm_idx(inode
, leaf
);
2271 ext4_free_blocks(handle
, inode
, NULL
, leaf
, 1,
2272 EXT4_FREE_BLOCKS_METADATA
| EXT4_FREE_BLOCKS_FORGET
);
2274 while (--depth
>= 0) {
2275 if (path
->p_idx
!= EXT_FIRST_INDEX(path
->p_hdr
))
2278 err
= ext4_ext_get_access(handle
, inode
, path
);
2281 path
->p_idx
->ei_block
= (path
+1)->p_idx
->ei_block
;
2282 err
= ext4_ext_dirty(handle
, inode
, path
);
2290 * ext4_ext_calc_credits_for_single_extent:
2291 * This routine returns max. credits that needed to insert an extent
2292 * to the extent tree.
2293 * When pass the actual path, the caller should calculate credits
2296 int ext4_ext_calc_credits_for_single_extent(struct inode
*inode
, int nrblocks
,
2297 struct ext4_ext_path
*path
)
2300 int depth
= ext_depth(inode
);
2303 /* probably there is space in leaf? */
2304 if (le16_to_cpu(path
[depth
].p_hdr
->eh_entries
)
2305 < le16_to_cpu(path
[depth
].p_hdr
->eh_max
)) {
2308 * There are some space in the leaf tree, no
2309 * need to account for leaf block credit
2311 * bitmaps and block group descriptor blocks
2312 * and other metadata blocks still need to be
2315 /* 1 bitmap, 1 block group descriptor */
2316 ret
= 2 + EXT4_META_TRANS_BLOCKS(inode
->i_sb
);
2321 return ext4_chunk_trans_blocks(inode
, nrblocks
);
2325 * How many index/leaf blocks need to change/allocate to add @extents extents?
2327 * If we add a single extent, then in the worse case, each tree level
2328 * index/leaf need to be changed in case of the tree split.
2330 * If more extents are inserted, they could cause the whole tree split more
2331 * than once, but this is really rare.
2333 int ext4_ext_index_trans_blocks(struct inode
*inode
, int extents
)
2338 /* If we are converting the inline data, only one is needed here. */
2339 if (ext4_has_inline_data(inode
))
2342 depth
= ext_depth(inode
);
2352 static inline int get_default_free_blocks_flags(struct inode
*inode
)
2354 if (S_ISDIR(inode
->i_mode
) || S_ISLNK(inode
->i_mode
))
2355 return EXT4_FREE_BLOCKS_METADATA
| EXT4_FREE_BLOCKS_FORGET
;
2356 else if (ext4_should_journal_data(inode
))
2357 return EXT4_FREE_BLOCKS_FORGET
;
2361 static int ext4_remove_blocks(handle_t
*handle
, struct inode
*inode
,
2362 struct ext4_extent
*ex
,
2363 long long *partial_cluster
,
2364 ext4_lblk_t from
, ext4_lblk_t to
)
2366 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
2367 unsigned short ee_len
= ext4_ext_get_actual_len(ex
);
2369 int flags
= get_default_free_blocks_flags(inode
);
2372 * For bigalloc file systems, we never free a partial cluster
2373 * at the beginning of the extent. Instead, we make a note
2374 * that we tried freeing the cluster, and check to see if we
2375 * need to free it on a subsequent call to ext4_remove_blocks,
2376 * or at the end of the ext4_truncate() operation.
2378 flags
|= EXT4_FREE_BLOCKS_NOFREE_FIRST_CLUSTER
;
2380 trace_ext4_remove_blocks(inode
, ex
, from
, to
, *partial_cluster
);
2382 * If we have a partial cluster, and it's different from the
2383 * cluster of the last block, we need to explicitly free the
2384 * partial cluster here.
2386 pblk
= ext4_ext_pblock(ex
) + ee_len
- 1;
2387 if ((*partial_cluster
> 0) &&
2388 (EXT4_B2C(sbi
, pblk
) != *partial_cluster
)) {
2389 ext4_free_blocks(handle
, inode
, NULL
,
2390 EXT4_C2B(sbi
, *partial_cluster
),
2391 sbi
->s_cluster_ratio
, flags
);
2392 *partial_cluster
= 0;
2395 #ifdef EXTENTS_STATS
2397 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
2398 spin_lock(&sbi
->s_ext_stats_lock
);
2399 sbi
->s_ext_blocks
+= ee_len
;
2400 sbi
->s_ext_extents
++;
2401 if (ee_len
< sbi
->s_ext_min
)
2402 sbi
->s_ext_min
= ee_len
;
2403 if (ee_len
> sbi
->s_ext_max
)
2404 sbi
->s_ext_max
= ee_len
;
2405 if (ext_depth(inode
) > sbi
->s_depth_max
)
2406 sbi
->s_depth_max
= ext_depth(inode
);
2407 spin_unlock(&sbi
->s_ext_stats_lock
);
2410 if (from
>= le32_to_cpu(ex
->ee_block
)
2411 && to
== le32_to_cpu(ex
->ee_block
) + ee_len
- 1) {
2414 unsigned int unaligned
;
2416 num
= le32_to_cpu(ex
->ee_block
) + ee_len
- from
;
2417 pblk
= ext4_ext_pblock(ex
) + ee_len
- num
;
2419 * Usually we want to free partial cluster at the end of the
2420 * extent, except for the situation when the cluster is still
2421 * used by any other extent (partial_cluster is negative).
2423 if (*partial_cluster
< 0 &&
2424 -(*partial_cluster
) == EXT4_B2C(sbi
, pblk
+ num
- 1))
2425 flags
|= EXT4_FREE_BLOCKS_NOFREE_LAST_CLUSTER
;
2427 ext_debug("free last %u blocks starting %llu partial %lld\n",
2428 num
, pblk
, *partial_cluster
);
2429 ext4_free_blocks(handle
, inode
, NULL
, pblk
, num
, flags
);
2431 * If the block range to be freed didn't start at the
2432 * beginning of a cluster, and we removed the entire
2433 * extent and the cluster is not used by any other extent,
2434 * save the partial cluster here, since we might need to
2435 * delete if we determine that the truncate operation has
2436 * removed all of the blocks in the cluster.
2438 * On the other hand, if we did not manage to free the whole
2439 * extent, we have to mark the cluster as used (store negative
2440 * cluster number in partial_cluster).
2442 unaligned
= pblk
& (sbi
->s_cluster_ratio
- 1);
2443 if (unaligned
&& (ee_len
== num
) &&
2444 (*partial_cluster
!= -((long long)EXT4_B2C(sbi
, pblk
))))
2445 *partial_cluster
= EXT4_B2C(sbi
, pblk
);
2447 *partial_cluster
= -((long long)EXT4_B2C(sbi
, pblk
));
2448 else if (*partial_cluster
> 0)
2449 *partial_cluster
= 0;
2451 ext4_error(sbi
->s_sb
, "strange request: removal(2) "
2452 "%u-%u from %u:%u\n",
2453 from
, to
, le32_to_cpu(ex
->ee_block
), ee_len
);
2459 * ext4_ext_rm_leaf() Removes the extents associated with the
2460 * blocks appearing between "start" and "end", and splits the extents
2461 * if "start" and "end" appear in the same extent
2463 * @handle: The journal handle
2464 * @inode: The files inode
2465 * @path: The path to the leaf
2466 * @partial_cluster: The cluster which we'll have to free if all extents
2467 * has been released from it. It gets negative in case
2468 * that the cluster is still used.
2469 * @start: The first block to remove
2470 * @end: The last block to remove
2473 ext4_ext_rm_leaf(handle_t
*handle
, struct inode
*inode
,
2474 struct ext4_ext_path
*path
,
2475 long long *partial_cluster
,
2476 ext4_lblk_t start
, ext4_lblk_t end
)
2478 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
2479 int err
= 0, correct_index
= 0;
2480 int depth
= ext_depth(inode
), credits
;
2481 struct ext4_extent_header
*eh
;
2484 ext4_lblk_t ex_ee_block
;
2485 unsigned short ex_ee_len
;
2486 unsigned uninitialized
= 0;
2487 struct ext4_extent
*ex
;
2490 /* the header must be checked already in ext4_ext_remove_space() */
2491 ext_debug("truncate since %u in leaf to %u\n", start
, end
);
2492 if (!path
[depth
].p_hdr
)
2493 path
[depth
].p_hdr
= ext_block_hdr(path
[depth
].p_bh
);
2494 eh
= path
[depth
].p_hdr
;
2495 if (unlikely(path
[depth
].p_hdr
== NULL
)) {
2496 EXT4_ERROR_INODE(inode
, "path[%d].p_hdr == NULL", depth
);
2499 /* find where to start removing */
2500 ex
= path
[depth
].p_ext
;
2502 ex
= EXT_LAST_EXTENT(eh
);
2504 ex_ee_block
= le32_to_cpu(ex
->ee_block
);
2505 ex_ee_len
= ext4_ext_get_actual_len(ex
);
2507 trace_ext4_ext_rm_leaf(inode
, start
, ex
, *partial_cluster
);
2509 while (ex
>= EXT_FIRST_EXTENT(eh
) &&
2510 ex_ee_block
+ ex_ee_len
> start
) {
2512 if (ext4_ext_is_uninitialized(ex
))
2517 ext_debug("remove ext %u:[%d]%d\n", ex_ee_block
,
2518 uninitialized
, ex_ee_len
);
2519 path
[depth
].p_ext
= ex
;
2521 a
= ex_ee_block
> start
? ex_ee_block
: start
;
2522 b
= ex_ee_block
+ex_ee_len
- 1 < end
?
2523 ex_ee_block
+ex_ee_len
- 1 : end
;
2525 ext_debug(" border %u:%u\n", a
, b
);
2527 /* If this extent is beyond the end of the hole, skip it */
2528 if (end
< ex_ee_block
) {
2530 * We're going to skip this extent and move to another,
2531 * so if this extent is not cluster aligned we have
2532 * to mark the current cluster as used to avoid
2533 * accidentally freeing it later on
2535 pblk
= ext4_ext_pblock(ex
);
2536 if (pblk
& (sbi
->s_cluster_ratio
- 1))
2538 -((long long)EXT4_B2C(sbi
, pblk
));
2540 ex_ee_block
= le32_to_cpu(ex
->ee_block
);
2541 ex_ee_len
= ext4_ext_get_actual_len(ex
);
2543 } else if (b
!= ex_ee_block
+ ex_ee_len
- 1) {
2544 EXT4_ERROR_INODE(inode
,
2545 "can not handle truncate %u:%u "
2547 start
, end
, ex_ee_block
,
2548 ex_ee_block
+ ex_ee_len
- 1);
2551 } else if (a
!= ex_ee_block
) {
2552 /* remove tail of the extent */
2553 num
= a
- ex_ee_block
;
2555 /* remove whole extent: excellent! */
2559 * 3 for leaf, sb, and inode plus 2 (bmap and group
2560 * descriptor) for each block group; assume two block
2561 * groups plus ex_ee_len/blocks_per_block_group for
2564 credits
= 7 + 2*(ex_ee_len
/EXT4_BLOCKS_PER_GROUP(inode
->i_sb
));
2565 if (ex
== EXT_FIRST_EXTENT(eh
)) {
2567 credits
+= (ext_depth(inode
)) + 1;
2569 credits
+= EXT4_MAXQUOTAS_TRANS_BLOCKS(inode
->i_sb
);
2571 err
= ext4_ext_truncate_extend_restart(handle
, inode
, credits
);
2575 err
= ext4_ext_get_access(handle
, inode
, path
+ depth
);
2579 err
= ext4_remove_blocks(handle
, inode
, ex
, partial_cluster
,
2585 /* this extent is removed; mark slot entirely unused */
2586 ext4_ext_store_pblock(ex
, 0);
2588 ex
->ee_len
= cpu_to_le16(num
);
2590 * Do not mark uninitialized if all the blocks in the
2591 * extent have been removed.
2593 if (uninitialized
&& num
)
2594 ext4_ext_mark_uninitialized(ex
);
2596 * If the extent was completely released,
2597 * we need to remove it from the leaf
2600 if (end
!= EXT_MAX_BLOCKS
- 1) {
2602 * For hole punching, we need to scoot all the
2603 * extents up when an extent is removed so that
2604 * we dont have blank extents in the middle
2606 memmove(ex
, ex
+1, (EXT_LAST_EXTENT(eh
) - ex
) *
2607 sizeof(struct ext4_extent
));
2609 /* Now get rid of the one at the end */
2610 memset(EXT_LAST_EXTENT(eh
), 0,
2611 sizeof(struct ext4_extent
));
2613 le16_add_cpu(&eh
->eh_entries
, -1);
2614 } else if (*partial_cluster
> 0)
2615 *partial_cluster
= 0;
2617 err
= ext4_ext_dirty(handle
, inode
, path
+ depth
);
2621 ext_debug("new extent: %u:%u:%llu\n", ex_ee_block
, num
,
2622 ext4_ext_pblock(ex
));
2624 ex_ee_block
= le32_to_cpu(ex
->ee_block
);
2625 ex_ee_len
= ext4_ext_get_actual_len(ex
);
2628 if (correct_index
&& eh
->eh_entries
)
2629 err
= ext4_ext_correct_indexes(handle
, inode
, path
);
2632 * Free the partial cluster only if the current extent does not
2633 * reference it. Otherwise we might free used cluster.
2635 if (*partial_cluster
> 0 &&
2636 (EXT4_B2C(sbi
, ext4_ext_pblock(ex
) + ex_ee_len
- 1) !=
2637 *partial_cluster
)) {
2638 int flags
= get_default_free_blocks_flags(inode
);
2640 ext4_free_blocks(handle
, inode
, NULL
,
2641 EXT4_C2B(sbi
, *partial_cluster
),
2642 sbi
->s_cluster_ratio
, flags
);
2643 *partial_cluster
= 0;
2646 /* if this leaf is free, then we should
2647 * remove it from index block above */
2648 if (err
== 0 && eh
->eh_entries
== 0 && path
[depth
].p_bh
!= NULL
)
2649 err
= ext4_ext_rm_idx(handle
, inode
, path
, depth
);
2656 * ext4_ext_more_to_rm:
2657 * returns 1 if current index has to be freed (even partial)
2660 ext4_ext_more_to_rm(struct ext4_ext_path
*path
)
2662 BUG_ON(path
->p_idx
== NULL
);
2664 if (path
->p_idx
< EXT_FIRST_INDEX(path
->p_hdr
))
2668 * if truncate on deeper level happened, it wasn't partial,
2669 * so we have to consider current index for truncation
2671 if (le16_to_cpu(path
->p_hdr
->eh_entries
) == path
->p_block
)
2676 int ext4_ext_remove_space(struct inode
*inode
, ext4_lblk_t start
,
2679 struct super_block
*sb
= inode
->i_sb
;
2680 int depth
= ext_depth(inode
);
2681 struct ext4_ext_path
*path
= NULL
;
2682 long long partial_cluster
= 0;
2686 ext_debug("truncate since %u to %u\n", start
, end
);
2688 /* probably first extent we're gonna free will be last in block */
2689 handle
= ext4_journal_start(inode
, EXT4_HT_TRUNCATE
, depth
+ 1);
2691 return PTR_ERR(handle
);
2694 trace_ext4_ext_remove_space(inode
, start
, end
, depth
);
2697 * Check if we are removing extents inside the extent tree. If that
2698 * is the case, we are going to punch a hole inside the extent tree
2699 * so we have to check whether we need to split the extent covering
2700 * the last block to remove so we can easily remove the part of it
2701 * in ext4_ext_rm_leaf().
2703 if (end
< EXT_MAX_BLOCKS
- 1) {
2704 struct ext4_extent
*ex
;
2705 ext4_lblk_t ee_block
;
2707 /* find extent for this block */
2708 path
= ext4_ext_find_extent(inode
, end
, NULL
);
2710 ext4_journal_stop(handle
);
2711 return PTR_ERR(path
);
2713 depth
= ext_depth(inode
);
2714 /* Leaf not may not exist only if inode has no blocks at all */
2715 ex
= path
[depth
].p_ext
;
2718 EXT4_ERROR_INODE(inode
,
2719 "path[%d].p_hdr == NULL",
2726 ee_block
= le32_to_cpu(ex
->ee_block
);
2729 * See if the last block is inside the extent, if so split
2730 * the extent at 'end' block so we can easily remove the
2731 * tail of the first part of the split extent in
2732 * ext4_ext_rm_leaf().
2734 if (end
>= ee_block
&&
2735 end
< ee_block
+ ext4_ext_get_actual_len(ex
) - 1) {
2738 if (ext4_ext_is_uninitialized(ex
))
2739 split_flag
= EXT4_EXT_MARK_UNINIT1
|
2740 EXT4_EXT_MARK_UNINIT2
;
2743 * Split the extent in two so that 'end' is the last
2744 * block in the first new extent. Also we should not
2745 * fail removing space due to ENOSPC so try to use
2746 * reserved block if that happens.
2748 err
= ext4_split_extent_at(handle
, inode
, path
,
2749 end
+ 1, split_flag
,
2750 EXT4_GET_BLOCKS_PRE_IO
|
2751 EXT4_GET_BLOCKS_METADATA_NOFAIL
);
2758 * We start scanning from right side, freeing all the blocks
2759 * after i_size and walking into the tree depth-wise.
2761 depth
= ext_depth(inode
);
2766 le16_to_cpu(path
[k
].p_hdr
->eh_entries
)+1;
2768 path
= kzalloc(sizeof(struct ext4_ext_path
) * (depth
+ 1),
2771 ext4_journal_stop(handle
);
2774 path
[0].p_depth
= depth
;
2775 path
[0].p_hdr
= ext_inode_hdr(inode
);
2778 if (ext4_ext_check(inode
, path
[0].p_hdr
, depth
)) {
2785 while (i
>= 0 && err
== 0) {
2787 /* this is leaf block */
2788 err
= ext4_ext_rm_leaf(handle
, inode
, path
,
2789 &partial_cluster
, start
,
2791 /* root level has p_bh == NULL, brelse() eats this */
2792 brelse(path
[i
].p_bh
);
2793 path
[i
].p_bh
= NULL
;
2798 /* this is index block */
2799 if (!path
[i
].p_hdr
) {
2800 ext_debug("initialize header\n");
2801 path
[i
].p_hdr
= ext_block_hdr(path
[i
].p_bh
);
2804 if (!path
[i
].p_idx
) {
2805 /* this level hasn't been touched yet */
2806 path
[i
].p_idx
= EXT_LAST_INDEX(path
[i
].p_hdr
);
2807 path
[i
].p_block
= le16_to_cpu(path
[i
].p_hdr
->eh_entries
)+1;
2808 ext_debug("init index ptr: hdr 0x%p, num %d\n",
2810 le16_to_cpu(path
[i
].p_hdr
->eh_entries
));
2812 /* we were already here, see at next index */
2816 ext_debug("level %d - index, first 0x%p, cur 0x%p\n",
2817 i
, EXT_FIRST_INDEX(path
[i
].p_hdr
),
2819 if (ext4_ext_more_to_rm(path
+ i
)) {
2820 struct buffer_head
*bh
;
2821 /* go to the next level */
2822 ext_debug("move to level %d (block %llu)\n",
2823 i
+ 1, ext4_idx_pblock(path
[i
].p_idx
));
2824 memset(path
+ i
+ 1, 0, sizeof(*path
));
2825 bh
= read_extent_tree_block(inode
,
2826 ext4_idx_pblock(path
[i
].p_idx
), depth
- i
- 1);
2828 /* should we reset i_size? */
2832 /* Yield here to deal with large extent trees.
2833 * Should be a no-op if we did IO above. */
2835 if (WARN_ON(i
+ 1 > depth
)) {
2839 path
[i
+ 1].p_bh
= bh
;
2841 /* save actual number of indexes since this
2842 * number is changed at the next iteration */
2843 path
[i
].p_block
= le16_to_cpu(path
[i
].p_hdr
->eh_entries
);
2846 /* we finished processing this index, go up */
2847 if (path
[i
].p_hdr
->eh_entries
== 0 && i
> 0) {
2848 /* index is empty, remove it;
2849 * handle must be already prepared by the
2850 * truncatei_leaf() */
2851 err
= ext4_ext_rm_idx(handle
, inode
, path
, i
);
2853 /* root level has p_bh == NULL, brelse() eats this */
2854 brelse(path
[i
].p_bh
);
2855 path
[i
].p_bh
= NULL
;
2857 ext_debug("return to level %d\n", i
);
2861 trace_ext4_ext_remove_space_done(inode
, start
, end
, depth
,
2862 partial_cluster
, path
->p_hdr
->eh_entries
);
2864 /* If we still have something in the partial cluster and we have removed
2865 * even the first extent, then we should free the blocks in the partial
2866 * cluster as well. */
2867 if (partial_cluster
> 0 && path
->p_hdr
->eh_entries
== 0) {
2868 int flags
= get_default_free_blocks_flags(inode
);
2870 ext4_free_blocks(handle
, inode
, NULL
,
2871 EXT4_C2B(EXT4_SB(sb
), partial_cluster
),
2872 EXT4_SB(sb
)->s_cluster_ratio
, flags
);
2873 partial_cluster
= 0;
2876 /* TODO: flexible tree reduction should be here */
2877 if (path
->p_hdr
->eh_entries
== 0) {
2879 * truncate to zero freed all the tree,
2880 * so we need to correct eh_depth
2882 err
= ext4_ext_get_access(handle
, inode
, path
);
2884 ext_inode_hdr(inode
)->eh_depth
= 0;
2885 ext_inode_hdr(inode
)->eh_max
=
2886 cpu_to_le16(ext4_ext_space_root(inode
, 0));
2887 err
= ext4_ext_dirty(handle
, inode
, path
);
2891 ext4_ext_drop_refs(path
);
2893 if (err
== -EAGAIN
) {
2897 ext4_journal_stop(handle
);
2903 * called at mount time
2905 void ext4_ext_init(struct super_block
*sb
)
2908 * possible initialization would be here
2911 if (EXT4_HAS_INCOMPAT_FEATURE(sb
, EXT4_FEATURE_INCOMPAT_EXTENTS
)) {
2912 #if defined(AGGRESSIVE_TEST) || defined(CHECK_BINSEARCH) || defined(EXTENTS_STATS)
2913 printk(KERN_INFO
"EXT4-fs: file extents enabled"
2914 #ifdef AGGRESSIVE_TEST
2915 ", aggressive tests"
2917 #ifdef CHECK_BINSEARCH
2920 #ifdef EXTENTS_STATS
2925 #ifdef EXTENTS_STATS
2926 spin_lock_init(&EXT4_SB(sb
)->s_ext_stats_lock
);
2927 EXT4_SB(sb
)->s_ext_min
= 1 << 30;
2928 EXT4_SB(sb
)->s_ext_max
= 0;
2934 * called at umount time
2936 void ext4_ext_release(struct super_block
*sb
)
2938 if (!EXT4_HAS_INCOMPAT_FEATURE(sb
, EXT4_FEATURE_INCOMPAT_EXTENTS
))
2941 #ifdef EXTENTS_STATS
2942 if (EXT4_SB(sb
)->s_ext_blocks
&& EXT4_SB(sb
)->s_ext_extents
) {
2943 struct ext4_sb_info
*sbi
= EXT4_SB(sb
);
2944 printk(KERN_ERR
"EXT4-fs: %lu blocks in %lu extents (%lu ave)\n",
2945 sbi
->s_ext_blocks
, sbi
->s_ext_extents
,
2946 sbi
->s_ext_blocks
/ sbi
->s_ext_extents
);
2947 printk(KERN_ERR
"EXT4-fs: extents: %lu min, %lu max, max depth %lu\n",
2948 sbi
->s_ext_min
, sbi
->s_ext_max
, sbi
->s_depth_max
);
2953 /* FIXME!! we need to try to merge to left or right after zero-out */
2954 static int ext4_ext_zeroout(struct inode
*inode
, struct ext4_extent
*ex
)
2956 ext4_fsblk_t ee_pblock
;
2957 unsigned int ee_len
;
2960 ee_len
= ext4_ext_get_actual_len(ex
);
2961 ee_pblock
= ext4_ext_pblock(ex
);
2963 ret
= sb_issue_zeroout(inode
->i_sb
, ee_pblock
, ee_len
, GFP_NOFS
);
2971 * ext4_split_extent_at() splits an extent at given block.
2973 * @handle: the journal handle
2974 * @inode: the file inode
2975 * @path: the path to the extent
2976 * @split: the logical block where the extent is splitted.
2977 * @split_flags: indicates if the extent could be zeroout if split fails, and
2978 * the states(init or uninit) of new extents.
2979 * @flags: flags used to insert new extent to extent tree.
2982 * Splits extent [a, b] into two extents [a, @split) and [@split, b], states
2983 * of which are deterimined by split_flag.
2985 * There are two cases:
2986 * a> the extent are splitted into two extent.
2987 * b> split is not needed, and just mark the extent.
2989 * return 0 on success.
2991 static int ext4_split_extent_at(handle_t
*handle
,
2992 struct inode
*inode
,
2993 struct ext4_ext_path
*path
,
2998 ext4_fsblk_t newblock
;
2999 ext4_lblk_t ee_block
;
3000 struct ext4_extent
*ex
, newex
, orig_ex
, zero_ex
;
3001 struct ext4_extent
*ex2
= NULL
;
3002 unsigned int ee_len
, depth
;
3005 BUG_ON((split_flag
& (EXT4_EXT_DATA_VALID1
| EXT4_EXT_DATA_VALID2
)) ==
3006 (EXT4_EXT_DATA_VALID1
| EXT4_EXT_DATA_VALID2
));
3008 ext_debug("ext4_split_extents_at: inode %lu, logical"
3009 "block %llu\n", inode
->i_ino
, (unsigned long long)split
);
3011 ext4_ext_show_leaf(inode
, path
);
3013 depth
= ext_depth(inode
);
3014 ex
= path
[depth
].p_ext
;
3015 ee_block
= le32_to_cpu(ex
->ee_block
);
3016 ee_len
= ext4_ext_get_actual_len(ex
);
3017 newblock
= split
- ee_block
+ ext4_ext_pblock(ex
);
3019 BUG_ON(split
< ee_block
|| split
>= (ee_block
+ ee_len
));
3020 BUG_ON(!ext4_ext_is_uninitialized(ex
) &&
3021 split_flag
& (EXT4_EXT_MAY_ZEROOUT
|
3022 EXT4_EXT_MARK_UNINIT1
|
3023 EXT4_EXT_MARK_UNINIT2
));
3025 err
= ext4_ext_get_access(handle
, inode
, path
+ depth
);
3029 if (split
== ee_block
) {
3031 * case b: block @split is the block that the extent begins with
3032 * then we just change the state of the extent, and splitting
3035 if (split_flag
& EXT4_EXT_MARK_UNINIT2
)
3036 ext4_ext_mark_uninitialized(ex
);
3038 ext4_ext_mark_initialized(ex
);
3040 if (!(flags
& EXT4_GET_BLOCKS_PRE_IO
))
3041 ext4_ext_try_to_merge(handle
, inode
, path
, ex
);
3043 err
= ext4_ext_dirty(handle
, inode
, path
+ path
->p_depth
);
3048 memcpy(&orig_ex
, ex
, sizeof(orig_ex
));
3049 ex
->ee_len
= cpu_to_le16(split
- ee_block
);
3050 if (split_flag
& EXT4_EXT_MARK_UNINIT1
)
3051 ext4_ext_mark_uninitialized(ex
);
3054 * path may lead to new leaf, not to original leaf any more
3055 * after ext4_ext_insert_extent() returns,
3057 err
= ext4_ext_dirty(handle
, inode
, path
+ depth
);
3059 goto fix_extent_len
;
3062 ex2
->ee_block
= cpu_to_le32(split
);
3063 ex2
->ee_len
= cpu_to_le16(ee_len
- (split
- ee_block
));
3064 ext4_ext_store_pblock(ex2
, newblock
);
3065 if (split_flag
& EXT4_EXT_MARK_UNINIT2
)
3066 ext4_ext_mark_uninitialized(ex2
);
3068 err
= ext4_ext_insert_extent(handle
, inode
, path
, &newex
, flags
);
3069 if (err
== -ENOSPC
&& (EXT4_EXT_MAY_ZEROOUT
& split_flag
)) {
3070 if (split_flag
& (EXT4_EXT_DATA_VALID1
|EXT4_EXT_DATA_VALID2
)) {
3071 if (split_flag
& EXT4_EXT_DATA_VALID1
) {
3072 err
= ext4_ext_zeroout(inode
, ex2
);
3073 zero_ex
.ee_block
= ex2
->ee_block
;
3074 zero_ex
.ee_len
= cpu_to_le16(
3075 ext4_ext_get_actual_len(ex2
));
3076 ext4_ext_store_pblock(&zero_ex
,
3077 ext4_ext_pblock(ex2
));
3079 err
= ext4_ext_zeroout(inode
, ex
);
3080 zero_ex
.ee_block
= ex
->ee_block
;
3081 zero_ex
.ee_len
= cpu_to_le16(
3082 ext4_ext_get_actual_len(ex
));
3083 ext4_ext_store_pblock(&zero_ex
,
3084 ext4_ext_pblock(ex
));
3087 err
= ext4_ext_zeroout(inode
, &orig_ex
);
3088 zero_ex
.ee_block
= orig_ex
.ee_block
;
3089 zero_ex
.ee_len
= cpu_to_le16(
3090 ext4_ext_get_actual_len(&orig_ex
));
3091 ext4_ext_store_pblock(&zero_ex
,
3092 ext4_ext_pblock(&orig_ex
));
3096 goto fix_extent_len
;
3097 /* update the extent length and mark as initialized */
3098 ex
->ee_len
= cpu_to_le16(ee_len
);
3099 ext4_ext_try_to_merge(handle
, inode
, path
, ex
);
3100 err
= ext4_ext_dirty(handle
, inode
, path
+ path
->p_depth
);
3102 goto fix_extent_len
;
3104 /* update extent status tree */
3105 err
= ext4_es_zeroout(inode
, &zero_ex
);
3109 goto fix_extent_len
;
3112 ext4_ext_show_leaf(inode
, path
);
3116 ex
->ee_len
= orig_ex
.ee_len
;
3117 ext4_ext_dirty(handle
, inode
, path
+ depth
);
3122 * ext4_split_extents() splits an extent and mark extent which is covered
3123 * by @map as split_flags indicates
3125 * It may result in splitting the extent into multiple extents (upto three)
3126 * There are three possibilities:
3127 * a> There is no split required
3128 * b> Splits in two extents: Split is happening at either end of the extent
3129 * c> Splits in three extents: Somone is splitting in middle of the extent
3132 static int ext4_split_extent(handle_t
*handle
,
3133 struct inode
*inode
,
3134 struct ext4_ext_path
*path
,
3135 struct ext4_map_blocks
*map
,
3139 ext4_lblk_t ee_block
;
3140 struct ext4_extent
*ex
;
3141 unsigned int ee_len
, depth
;
3144 int split_flag1
, flags1
;
3145 int allocated
= map
->m_len
;
3147 depth
= ext_depth(inode
);
3148 ex
= path
[depth
].p_ext
;
3149 ee_block
= le32_to_cpu(ex
->ee_block
);
3150 ee_len
= ext4_ext_get_actual_len(ex
);
3151 uninitialized
= ext4_ext_is_uninitialized(ex
);
3153 if (map
->m_lblk
+ map
->m_len
< ee_block
+ ee_len
) {
3154 split_flag1
= split_flag
& EXT4_EXT_MAY_ZEROOUT
;
3155 flags1
= flags
| EXT4_GET_BLOCKS_PRE_IO
;
3157 split_flag1
|= EXT4_EXT_MARK_UNINIT1
|
3158 EXT4_EXT_MARK_UNINIT2
;
3159 if (split_flag
& EXT4_EXT_DATA_VALID2
)
3160 split_flag1
|= EXT4_EXT_DATA_VALID1
;
3161 err
= ext4_split_extent_at(handle
, inode
, path
,
3162 map
->m_lblk
+ map
->m_len
, split_flag1
, flags1
);
3166 allocated
= ee_len
- (map
->m_lblk
- ee_block
);
3169 * Update path is required because previous ext4_split_extent_at() may
3170 * result in split of original leaf or extent zeroout.
3172 ext4_ext_drop_refs(path
);
3173 path
= ext4_ext_find_extent(inode
, map
->m_lblk
, path
);
3175 return PTR_ERR(path
);
3176 depth
= ext_depth(inode
);
3177 ex
= path
[depth
].p_ext
;
3178 uninitialized
= ext4_ext_is_uninitialized(ex
);
3181 if (map
->m_lblk
>= ee_block
) {
3182 split_flag1
= split_flag
& EXT4_EXT_DATA_VALID2
;
3183 if (uninitialized
) {
3184 split_flag1
|= EXT4_EXT_MARK_UNINIT1
;
3185 split_flag1
|= split_flag
& (EXT4_EXT_MAY_ZEROOUT
|
3186 EXT4_EXT_MARK_UNINIT2
);
3188 err
= ext4_split_extent_at(handle
, inode
, path
,
3189 map
->m_lblk
, split_flag1
, flags
);
3194 ext4_ext_show_leaf(inode
, path
);
3196 return err
? err
: allocated
;
3200 * This function is called by ext4_ext_map_blocks() if someone tries to write
3201 * to an uninitialized extent. It may result in splitting the uninitialized
3202 * extent into multiple extents (up to three - one initialized and two
3204 * There are three possibilities:
3205 * a> There is no split required: Entire extent should be initialized
3206 * b> Splits in two extents: Write is happening at either end of the extent
3207 * c> Splits in three extents: Somone is writing in middle of the extent
3210 * - The extent pointed to by 'path' is uninitialized.
3211 * - The extent pointed to by 'path' contains a superset
3212 * of the logical span [map->m_lblk, map->m_lblk + map->m_len).
3214 * Post-conditions on success:
3215 * - the returned value is the number of blocks beyond map->l_lblk
3216 * that are allocated and initialized.
3217 * It is guaranteed to be >= map->m_len.
3219 static int ext4_ext_convert_to_initialized(handle_t
*handle
,
3220 struct inode
*inode
,
3221 struct ext4_map_blocks
*map
,
3222 struct ext4_ext_path
*path
,
3225 struct ext4_sb_info
*sbi
;
3226 struct ext4_extent_header
*eh
;
3227 struct ext4_map_blocks split_map
;
3228 struct ext4_extent zero_ex
;
3229 struct ext4_extent
*ex
, *abut_ex
;
3230 ext4_lblk_t ee_block
, eof_block
;
3231 unsigned int ee_len
, depth
, map_len
= map
->m_len
;
3232 int allocated
= 0, max_zeroout
= 0;
3236 ext_debug("ext4_ext_convert_to_initialized: inode %lu, logical"
3237 "block %llu, max_blocks %u\n", inode
->i_ino
,
3238 (unsigned long long)map
->m_lblk
, map_len
);
3240 sbi
= EXT4_SB(inode
->i_sb
);
3241 eof_block
= (inode
->i_size
+ inode
->i_sb
->s_blocksize
- 1) >>
3242 inode
->i_sb
->s_blocksize_bits
;
3243 if (eof_block
< map
->m_lblk
+ map_len
)
3244 eof_block
= map
->m_lblk
+ map_len
;
3246 depth
= ext_depth(inode
);
3247 eh
= path
[depth
].p_hdr
;
3248 ex
= path
[depth
].p_ext
;
3249 ee_block
= le32_to_cpu(ex
->ee_block
);
3250 ee_len
= ext4_ext_get_actual_len(ex
);
3253 trace_ext4_ext_convert_to_initialized_enter(inode
, map
, ex
);
3255 /* Pre-conditions */
3256 BUG_ON(!ext4_ext_is_uninitialized(ex
));
3257 BUG_ON(!in_range(map
->m_lblk
, ee_block
, ee_len
));
3260 * Attempt to transfer newly initialized blocks from the currently
3261 * uninitialized extent to its neighbor. This is much cheaper
3262 * than an insertion followed by a merge as those involve costly
3263 * memmove() calls. Transferring to the left is the common case in
3264 * steady state for workloads doing fallocate(FALLOC_FL_KEEP_SIZE)
3265 * followed by append writes.
3267 * Limitations of the current logic:
3268 * - L1: we do not deal with writes covering the whole extent.
3269 * This would require removing the extent if the transfer
3271 * - L2: we only attempt to merge with an extent stored in the
3272 * same extent tree node.
3274 if ((map
->m_lblk
== ee_block
) &&
3275 /* See if we can merge left */
3276 (map_len
< ee_len
) && /*L1*/
3277 (ex
> EXT_FIRST_EXTENT(eh
))) { /*L2*/
3278 ext4_lblk_t prev_lblk
;
3279 ext4_fsblk_t prev_pblk
, ee_pblk
;
3280 unsigned int prev_len
;
3283 prev_lblk
= le32_to_cpu(abut_ex
->ee_block
);
3284 prev_len
= ext4_ext_get_actual_len(abut_ex
);
3285 prev_pblk
= ext4_ext_pblock(abut_ex
);
3286 ee_pblk
= ext4_ext_pblock(ex
);
3289 * A transfer of blocks from 'ex' to 'abut_ex' is allowed
3290 * upon those conditions:
3291 * - C1: abut_ex is initialized,
3292 * - C2: abut_ex is logically abutting ex,
3293 * - C3: abut_ex is physically abutting ex,
3294 * - C4: abut_ex can receive the additional blocks without
3295 * overflowing the (initialized) length limit.
3297 if ((!ext4_ext_is_uninitialized(abut_ex
)) && /*C1*/
3298 ((prev_lblk
+ prev_len
) == ee_block
) && /*C2*/
3299 ((prev_pblk
+ prev_len
) == ee_pblk
) && /*C3*/
3300 (prev_len
< (EXT_INIT_MAX_LEN
- map_len
))) { /*C4*/
3301 err
= ext4_ext_get_access(handle
, inode
, path
+ depth
);
3305 trace_ext4_ext_convert_to_initialized_fastpath(inode
,
3308 /* Shift the start of ex by 'map_len' blocks */
3309 ex
->ee_block
= cpu_to_le32(ee_block
+ map_len
);
3310 ext4_ext_store_pblock(ex
, ee_pblk
+ map_len
);
3311 ex
->ee_len
= cpu_to_le16(ee_len
- map_len
);
3312 ext4_ext_mark_uninitialized(ex
); /* Restore the flag */
3314 /* Extend abut_ex by 'map_len' blocks */
3315 abut_ex
->ee_len
= cpu_to_le16(prev_len
+ map_len
);
3317 /* Result: number of initialized blocks past m_lblk */
3318 allocated
= map_len
;
3320 } else if (((map
->m_lblk
+ map_len
) == (ee_block
+ ee_len
)) &&
3321 (map_len
< ee_len
) && /*L1*/
3322 ex
< EXT_LAST_EXTENT(eh
)) { /*L2*/
3323 /* See if we can merge right */
3324 ext4_lblk_t next_lblk
;
3325 ext4_fsblk_t next_pblk
, ee_pblk
;
3326 unsigned int next_len
;
3329 next_lblk
= le32_to_cpu(abut_ex
->ee_block
);
3330 next_len
= ext4_ext_get_actual_len(abut_ex
);
3331 next_pblk
= ext4_ext_pblock(abut_ex
);
3332 ee_pblk
= ext4_ext_pblock(ex
);
3335 * A transfer of blocks from 'ex' to 'abut_ex' is allowed
3336 * upon those conditions:
3337 * - C1: abut_ex is initialized,
3338 * - C2: abut_ex is logically abutting ex,
3339 * - C3: abut_ex is physically abutting ex,
3340 * - C4: abut_ex can receive the additional blocks without
3341 * overflowing the (initialized) length limit.
3343 if ((!ext4_ext_is_uninitialized(abut_ex
)) && /*C1*/
3344 ((map
->m_lblk
+ map_len
) == next_lblk
) && /*C2*/
3345 ((ee_pblk
+ ee_len
) == next_pblk
) && /*C3*/
3346 (next_len
< (EXT_INIT_MAX_LEN
- map_len
))) { /*C4*/
3347 err
= ext4_ext_get_access(handle
, inode
, path
+ depth
);
3351 trace_ext4_ext_convert_to_initialized_fastpath(inode
,
3354 /* Shift the start of abut_ex by 'map_len' blocks */
3355 abut_ex
->ee_block
= cpu_to_le32(next_lblk
- map_len
);
3356 ext4_ext_store_pblock(abut_ex
, next_pblk
- map_len
);
3357 ex
->ee_len
= cpu_to_le16(ee_len
- map_len
);
3358 ext4_ext_mark_uninitialized(ex
); /* Restore the flag */
3360 /* Extend abut_ex by 'map_len' blocks */
3361 abut_ex
->ee_len
= cpu_to_le16(next_len
+ map_len
);
3363 /* Result: number of initialized blocks past m_lblk */
3364 allocated
= map_len
;
3368 /* Mark the block containing both extents as dirty */
3369 ext4_ext_dirty(handle
, inode
, path
+ depth
);
3371 /* Update path to point to the right extent */
3372 path
[depth
].p_ext
= abut_ex
;
3375 allocated
= ee_len
- (map
->m_lblk
- ee_block
);
3377 WARN_ON(map
->m_lblk
< ee_block
);
3379 * It is safe to convert extent to initialized via explicit
3380 * zeroout only if extent is fully insde i_size or new_size.
3382 split_flag
|= ee_block
+ ee_len
<= eof_block
? EXT4_EXT_MAY_ZEROOUT
: 0;
3384 if (EXT4_EXT_MAY_ZEROOUT
& split_flag
)
3385 max_zeroout
= sbi
->s_extent_max_zeroout_kb
>>
3386 (inode
->i_sb
->s_blocksize_bits
- 10);
3388 /* If extent is less than s_max_zeroout_kb, zeroout directly */
3389 if (max_zeroout
&& (ee_len
<= max_zeroout
)) {
3390 err
= ext4_ext_zeroout(inode
, ex
);
3393 zero_ex
.ee_block
= ex
->ee_block
;
3394 zero_ex
.ee_len
= cpu_to_le16(ext4_ext_get_actual_len(ex
));
3395 ext4_ext_store_pblock(&zero_ex
, ext4_ext_pblock(ex
));
3397 err
= ext4_ext_get_access(handle
, inode
, path
+ depth
);
3400 ext4_ext_mark_initialized(ex
);
3401 ext4_ext_try_to_merge(handle
, inode
, path
, ex
);
3402 err
= ext4_ext_dirty(handle
, inode
, path
+ path
->p_depth
);
3408 * 1. split the extent into three extents.
3409 * 2. split the extent into two extents, zeroout the first half.
3410 * 3. split the extent into two extents, zeroout the second half.
3411 * 4. split the extent into two extents with out zeroout.
3413 split_map
.m_lblk
= map
->m_lblk
;
3414 split_map
.m_len
= map
->m_len
;
3416 if (max_zeroout
&& (allocated
> map
->m_len
)) {
3417 if (allocated
<= max_zeroout
) {
3420 cpu_to_le32(map
->m_lblk
);
3421 zero_ex
.ee_len
= cpu_to_le16(allocated
);
3422 ext4_ext_store_pblock(&zero_ex
,
3423 ext4_ext_pblock(ex
) + map
->m_lblk
- ee_block
);
3424 err
= ext4_ext_zeroout(inode
, &zero_ex
);
3427 split_map
.m_lblk
= map
->m_lblk
;
3428 split_map
.m_len
= allocated
;
3429 } else if (map
->m_lblk
- ee_block
+ map
->m_len
< max_zeroout
) {
3431 if (map
->m_lblk
!= ee_block
) {
3432 zero_ex
.ee_block
= ex
->ee_block
;
3433 zero_ex
.ee_len
= cpu_to_le16(map
->m_lblk
-
3435 ext4_ext_store_pblock(&zero_ex
,
3436 ext4_ext_pblock(ex
));
3437 err
= ext4_ext_zeroout(inode
, &zero_ex
);
3442 split_map
.m_lblk
= ee_block
;
3443 split_map
.m_len
= map
->m_lblk
- ee_block
+ map
->m_len
;
3444 allocated
= map
->m_len
;
3448 allocated
= ext4_split_extent(handle
, inode
, path
,
3449 &split_map
, split_flag
, flags
);
3454 /* If we have gotten a failure, don't zero out status tree */
3456 err
= ext4_es_zeroout(inode
, &zero_ex
);
3457 return err
? err
: allocated
;
3461 * This function is called by ext4_ext_map_blocks() from
3462 * ext4_get_blocks_dio_write() when DIO to write
3463 * to an uninitialized extent.
3465 * Writing to an uninitialized extent may result in splitting the uninitialized
3466 * extent into multiple initialized/uninitialized extents (up to three)
3467 * There are three possibilities:
3468 * a> There is no split required: Entire extent should be uninitialized
3469 * b> Splits in two extents: Write is happening at either end of the extent
3470 * c> Splits in three extents: Somone is writing in middle of the extent
3472 * One of more index blocks maybe needed if the extent tree grow after
3473 * the uninitialized extent split. To prevent ENOSPC occur at the IO
3474 * complete, we need to split the uninitialized extent before DIO submit
3475 * the IO. The uninitialized extent called at this time will be split
3476 * into three uninitialized extent(at most). After IO complete, the part
3477 * being filled will be convert to initialized by the end_io callback function
3478 * via ext4_convert_unwritten_extents().
3480 * Returns the size of uninitialized extent to be written on success.
3482 static int ext4_split_unwritten_extents(handle_t
*handle
,
3483 struct inode
*inode
,
3484 struct ext4_map_blocks
*map
,
3485 struct ext4_ext_path
*path
,
3488 ext4_lblk_t eof_block
;
3489 ext4_lblk_t ee_block
;
3490 struct ext4_extent
*ex
;
3491 unsigned int ee_len
;
3492 int split_flag
= 0, depth
;
3494 ext_debug("ext4_split_unwritten_extents: inode %lu, logical"
3495 "block %llu, max_blocks %u\n", inode
->i_ino
,
3496 (unsigned long long)map
->m_lblk
, map
->m_len
);
3498 eof_block
= (inode
->i_size
+ inode
->i_sb
->s_blocksize
- 1) >>
3499 inode
->i_sb
->s_blocksize_bits
;
3500 if (eof_block
< map
->m_lblk
+ map
->m_len
)
3501 eof_block
= map
->m_lblk
+ map
->m_len
;
3503 * It is safe to convert extent to initialized via explicit
3504 * zeroout only if extent is fully insde i_size or new_size.
3506 depth
= ext_depth(inode
);
3507 ex
= path
[depth
].p_ext
;
3508 ee_block
= le32_to_cpu(ex
->ee_block
);
3509 ee_len
= ext4_ext_get_actual_len(ex
);
3511 split_flag
|= ee_block
+ ee_len
<= eof_block
? EXT4_EXT_MAY_ZEROOUT
: 0;
3512 split_flag
|= EXT4_EXT_MARK_UNINIT2
;
3513 if (flags
& EXT4_GET_BLOCKS_CONVERT
)
3514 split_flag
|= EXT4_EXT_DATA_VALID2
;
3515 flags
|= EXT4_GET_BLOCKS_PRE_IO
;
3516 return ext4_split_extent(handle
, inode
, path
, map
, split_flag
, flags
);
3519 static int ext4_convert_unwritten_extents_endio(handle_t
*handle
,
3520 struct inode
*inode
,
3521 struct ext4_map_blocks
*map
,
3522 struct ext4_ext_path
*path
)
3524 struct ext4_extent
*ex
;
3525 ext4_lblk_t ee_block
;
3526 unsigned int ee_len
;
3530 depth
= ext_depth(inode
);
3531 ex
= path
[depth
].p_ext
;
3532 ee_block
= le32_to_cpu(ex
->ee_block
);
3533 ee_len
= ext4_ext_get_actual_len(ex
);
3535 ext_debug("ext4_convert_unwritten_extents_endio: inode %lu, logical"
3536 "block %llu, max_blocks %u\n", inode
->i_ino
,
3537 (unsigned long long)ee_block
, ee_len
);
3539 /* If extent is larger than requested it is a clear sign that we still
3540 * have some extent state machine issues left. So extent_split is still
3542 * TODO: Once all related issues will be fixed this situation should be
3545 if (ee_block
!= map
->m_lblk
|| ee_len
> map
->m_len
) {
3547 ext4_warning("Inode (%ld) finished: extent logical block %llu,"
3548 " len %u; IO logical block %llu, len %u\n",
3549 inode
->i_ino
, (unsigned long long)ee_block
, ee_len
,
3550 (unsigned long long)map
->m_lblk
, map
->m_len
);
3552 err
= ext4_split_unwritten_extents(handle
, inode
, map
, path
,
3553 EXT4_GET_BLOCKS_CONVERT
);
3556 ext4_ext_drop_refs(path
);
3557 path
= ext4_ext_find_extent(inode
, map
->m_lblk
, path
);
3559 err
= PTR_ERR(path
);
3562 depth
= ext_depth(inode
);
3563 ex
= path
[depth
].p_ext
;
3566 err
= ext4_ext_get_access(handle
, inode
, path
+ depth
);
3569 /* first mark the extent as initialized */
3570 ext4_ext_mark_initialized(ex
);
3572 /* note: ext4_ext_correct_indexes() isn't needed here because
3573 * borders are not changed
3575 ext4_ext_try_to_merge(handle
, inode
, path
, ex
);
3577 /* Mark modified extent as dirty */
3578 err
= ext4_ext_dirty(handle
, inode
, path
+ path
->p_depth
);
3580 ext4_ext_show_leaf(inode
, path
);
3584 static void unmap_underlying_metadata_blocks(struct block_device
*bdev
,
3585 sector_t block
, int count
)
3588 for (i
= 0; i
< count
; i
++)
3589 unmap_underlying_metadata(bdev
, block
+ i
);
3593 * Handle EOFBLOCKS_FL flag, clearing it if necessary
3595 static int check_eofblocks_fl(handle_t
*handle
, struct inode
*inode
,
3597 struct ext4_ext_path
*path
,
3601 struct ext4_extent_header
*eh
;
3602 struct ext4_extent
*last_ex
;
3604 if (!ext4_test_inode_flag(inode
, EXT4_INODE_EOFBLOCKS
))
3607 depth
= ext_depth(inode
);
3608 eh
= path
[depth
].p_hdr
;
3611 * We're going to remove EOFBLOCKS_FL entirely in future so we
3612 * do not care for this case anymore. Simply remove the flag
3613 * if there are no extents.
3615 if (unlikely(!eh
->eh_entries
))
3617 last_ex
= EXT_LAST_EXTENT(eh
);
3619 * We should clear the EOFBLOCKS_FL flag if we are writing the
3620 * last block in the last extent in the file. We test this by
3621 * first checking to see if the caller to
3622 * ext4_ext_get_blocks() was interested in the last block (or
3623 * a block beyond the last block) in the current extent. If
3624 * this turns out to be false, we can bail out from this
3625 * function immediately.
3627 if (lblk
+ len
< le32_to_cpu(last_ex
->ee_block
) +
3628 ext4_ext_get_actual_len(last_ex
))
3631 * If the caller does appear to be planning to write at or
3632 * beyond the end of the current extent, we then test to see
3633 * if the current extent is the last extent in the file, by
3634 * checking to make sure it was reached via the rightmost node
3635 * at each level of the tree.
3637 for (i
= depth
-1; i
>= 0; i
--)
3638 if (path
[i
].p_idx
!= EXT_LAST_INDEX(path
[i
].p_hdr
))
3641 ext4_clear_inode_flag(inode
, EXT4_INODE_EOFBLOCKS
);
3642 return ext4_mark_inode_dirty(handle
, inode
);
3646 * ext4_find_delalloc_range: find delayed allocated block in the given range.
3648 * Return 1 if there is a delalloc block in the range, otherwise 0.
3650 int ext4_find_delalloc_range(struct inode
*inode
,
3651 ext4_lblk_t lblk_start
,
3652 ext4_lblk_t lblk_end
)
3654 struct extent_status es
;
3656 ext4_es_find_delayed_extent_range(inode
, lblk_start
, lblk_end
, &es
);
3658 return 0; /* there is no delay extent in this tree */
3659 else if (es
.es_lblk
<= lblk_start
&&
3660 lblk_start
< es
.es_lblk
+ es
.es_len
)
3662 else if (lblk_start
<= es
.es_lblk
&& es
.es_lblk
<= lblk_end
)
3668 int ext4_find_delalloc_cluster(struct inode
*inode
, ext4_lblk_t lblk
)
3670 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
3671 ext4_lblk_t lblk_start
, lblk_end
;
3672 lblk_start
= lblk
& (~(sbi
->s_cluster_ratio
- 1));
3673 lblk_end
= lblk_start
+ sbi
->s_cluster_ratio
- 1;
3675 return ext4_find_delalloc_range(inode
, lblk_start
, lblk_end
);
3679 * Determines how many complete clusters (out of those specified by the 'map')
3680 * are under delalloc and were reserved quota for.
3681 * This function is called when we are writing out the blocks that were
3682 * originally written with their allocation delayed, but then the space was
3683 * allocated using fallocate() before the delayed allocation could be resolved.
3684 * The cases to look for are:
3685 * ('=' indicated delayed allocated blocks
3686 * '-' indicates non-delayed allocated blocks)
3687 * (a) partial clusters towards beginning and/or end outside of allocated range
3688 * are not delalloc'ed.
3690 * |----c---=|====c====|====c====|===-c----|
3691 * |++++++ allocated ++++++|
3692 * ==> 4 complete clusters in above example
3694 * (b) partial cluster (outside of allocated range) towards either end is
3695 * marked for delayed allocation. In this case, we will exclude that
3698 * |----====c========|========c========|
3699 * |++++++ allocated ++++++|
3700 * ==> 1 complete clusters in above example
3703 * |================c================|
3704 * |++++++ allocated ++++++|
3705 * ==> 0 complete clusters in above example
3707 * The ext4_da_update_reserve_space will be called only if we
3708 * determine here that there were some "entire" clusters that span
3709 * this 'allocated' range.
3710 * In the non-bigalloc case, this function will just end up returning num_blks
3711 * without ever calling ext4_find_delalloc_range.
3714 get_reserved_cluster_alloc(struct inode
*inode
, ext4_lblk_t lblk_start
,
3715 unsigned int num_blks
)
3717 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
3718 ext4_lblk_t alloc_cluster_start
, alloc_cluster_end
;
3719 ext4_lblk_t lblk_from
, lblk_to
, c_offset
;
3720 unsigned int allocated_clusters
= 0;
3722 alloc_cluster_start
= EXT4_B2C(sbi
, lblk_start
);
3723 alloc_cluster_end
= EXT4_B2C(sbi
, lblk_start
+ num_blks
- 1);
3725 /* max possible clusters for this allocation */
3726 allocated_clusters
= alloc_cluster_end
- alloc_cluster_start
+ 1;
3728 trace_ext4_get_reserved_cluster_alloc(inode
, lblk_start
, num_blks
);
3730 /* Check towards left side */
3731 c_offset
= lblk_start
& (sbi
->s_cluster_ratio
- 1);
3733 lblk_from
= lblk_start
& (~(sbi
->s_cluster_ratio
- 1));
3734 lblk_to
= lblk_from
+ c_offset
- 1;
3736 if (ext4_find_delalloc_range(inode
, lblk_from
, lblk_to
))
3737 allocated_clusters
--;
3740 /* Now check towards right. */
3741 c_offset
= (lblk_start
+ num_blks
) & (sbi
->s_cluster_ratio
- 1);
3742 if (allocated_clusters
&& c_offset
) {
3743 lblk_from
= lblk_start
+ num_blks
;
3744 lblk_to
= lblk_from
+ (sbi
->s_cluster_ratio
- c_offset
) - 1;
3746 if (ext4_find_delalloc_range(inode
, lblk_from
, lblk_to
))
3747 allocated_clusters
--;
3750 return allocated_clusters
;
3754 ext4_ext_handle_uninitialized_extents(handle_t
*handle
, struct inode
*inode
,
3755 struct ext4_map_blocks
*map
,
3756 struct ext4_ext_path
*path
, int flags
,
3757 unsigned int allocated
, ext4_fsblk_t newblock
)
3761 ext4_io_end_t
*io
= ext4_inode_aio(inode
);
3763 ext_debug("ext4_ext_handle_uninitialized_extents: inode %lu, logical "
3764 "block %llu, max_blocks %u, flags %x, allocated %u\n",
3765 inode
->i_ino
, (unsigned long long)map
->m_lblk
, map
->m_len
,
3767 ext4_ext_show_leaf(inode
, path
);
3770 * When writing into uninitialized space, we should not fail to
3771 * allocate metadata blocks for the new extent block if needed.
3773 flags
|= EXT4_GET_BLOCKS_METADATA_NOFAIL
;
3775 trace_ext4_ext_handle_uninitialized_extents(inode
, map
, flags
,
3776 allocated
, newblock
);
3778 /* get_block() before submit the IO, split the extent */
3779 if ((flags
& EXT4_GET_BLOCKS_PRE_IO
)) {
3780 ret
= ext4_split_unwritten_extents(handle
, inode
, map
,
3785 * Flag the inode(non aio case) or end_io struct (aio case)
3786 * that this IO needs to conversion to written when IO is
3790 ext4_set_io_unwritten_flag(inode
, io
);
3792 ext4_set_inode_state(inode
, EXT4_STATE_DIO_UNWRITTEN
);
3793 map
->m_flags
|= EXT4_MAP_UNWRITTEN
;
3794 if (ext4_should_dioread_nolock(inode
))
3795 map
->m_flags
|= EXT4_MAP_UNINIT
;
3798 /* IO end_io complete, convert the filled extent to written */
3799 if ((flags
& EXT4_GET_BLOCKS_CONVERT
)) {
3800 ret
= ext4_convert_unwritten_extents_endio(handle
, inode
, map
,
3803 ext4_update_inode_fsync_trans(handle
, inode
, 1);
3804 err
= check_eofblocks_fl(handle
, inode
, map
->m_lblk
,
3808 map
->m_flags
|= EXT4_MAP_MAPPED
;
3809 if (allocated
> map
->m_len
)
3810 allocated
= map
->m_len
;
3811 map
->m_len
= allocated
;
3814 /* buffered IO case */
3816 * repeat fallocate creation request
3817 * we already have an unwritten extent
3819 if (flags
& EXT4_GET_BLOCKS_UNINIT_EXT
) {
3820 map
->m_flags
|= EXT4_MAP_UNWRITTEN
;
3824 /* buffered READ or buffered write_begin() lookup */
3825 if ((flags
& EXT4_GET_BLOCKS_CREATE
) == 0) {
3827 * We have blocks reserved already. We
3828 * return allocated blocks so that delalloc
3829 * won't do block reservation for us. But
3830 * the buffer head will be unmapped so that
3831 * a read from the block returns 0s.
3833 map
->m_flags
|= EXT4_MAP_UNWRITTEN
;
3837 /* buffered write, writepage time, convert*/
3838 ret
= ext4_ext_convert_to_initialized(handle
, inode
, map
, path
, flags
);
3840 ext4_update_inode_fsync_trans(handle
, inode
, 1);
3847 map
->m_flags
|= EXT4_MAP_NEW
;
3849 * if we allocated more blocks than requested
3850 * we need to make sure we unmap the extra block
3851 * allocated. The actual needed block will get
3852 * unmapped later when we find the buffer_head marked
3855 if (allocated
> map
->m_len
) {
3856 unmap_underlying_metadata_blocks(inode
->i_sb
->s_bdev
,
3857 newblock
+ map
->m_len
,
3858 allocated
- map
->m_len
);
3859 allocated
= map
->m_len
;
3861 map
->m_len
= allocated
;
3864 * If we have done fallocate with the offset that is already
3865 * delayed allocated, we would have block reservation
3866 * and quota reservation done in the delayed write path.
3867 * But fallocate would have already updated quota and block
3868 * count for this offset. So cancel these reservation
3870 if (flags
& EXT4_GET_BLOCKS_DELALLOC_RESERVE
) {
3871 unsigned int reserved_clusters
;
3872 reserved_clusters
= get_reserved_cluster_alloc(inode
,
3873 map
->m_lblk
, map
->m_len
);
3874 if (reserved_clusters
)
3875 ext4_da_update_reserve_space(inode
,
3881 map
->m_flags
|= EXT4_MAP_MAPPED
;
3882 if ((flags
& EXT4_GET_BLOCKS_KEEP_SIZE
) == 0) {
3883 err
= check_eofblocks_fl(handle
, inode
, map
->m_lblk
, path
,
3889 if (allocated
> map
->m_len
)
3890 allocated
= map
->m_len
;
3891 ext4_ext_show_leaf(inode
, path
);
3892 map
->m_pblk
= newblock
;
3893 map
->m_len
= allocated
;
3896 ext4_ext_drop_refs(path
);
3899 return err
? err
: allocated
;
3903 * get_implied_cluster_alloc - check to see if the requested
3904 * allocation (in the map structure) overlaps with a cluster already
3905 * allocated in an extent.
3906 * @sb The filesystem superblock structure
3907 * @map The requested lblk->pblk mapping
3908 * @ex The extent structure which might contain an implied
3909 * cluster allocation
3911 * This function is called by ext4_ext_map_blocks() after we failed to
3912 * find blocks that were already in the inode's extent tree. Hence,
3913 * we know that the beginning of the requested region cannot overlap
3914 * the extent from the inode's extent tree. There are three cases we
3915 * want to catch. The first is this case:
3917 * |--- cluster # N--|
3918 * |--- extent ---| |---- requested region ---|
3921 * The second case that we need to test for is this one:
3923 * |--------- cluster # N ----------------|
3924 * |--- requested region --| |------- extent ----|
3925 * |=======================|
3927 * The third case is when the requested region lies between two extents
3928 * within the same cluster:
3929 * |------------- cluster # N-------------|
3930 * |----- ex -----| |---- ex_right ----|
3931 * |------ requested region ------|
3932 * |================|
3934 * In each of the above cases, we need to set the map->m_pblk and
3935 * map->m_len so it corresponds to the return the extent labelled as
3936 * "|====|" from cluster #N, since it is already in use for data in
3937 * cluster EXT4_B2C(sbi, map->m_lblk). We will then return 1 to
3938 * signal to ext4_ext_map_blocks() that map->m_pblk should be treated
3939 * as a new "allocated" block region. Otherwise, we will return 0 and
3940 * ext4_ext_map_blocks() will then allocate one or more new clusters
3941 * by calling ext4_mb_new_blocks().
3943 static int get_implied_cluster_alloc(struct super_block
*sb
,
3944 struct ext4_map_blocks
*map
,
3945 struct ext4_extent
*ex
,
3946 struct ext4_ext_path
*path
)
3948 struct ext4_sb_info
*sbi
= EXT4_SB(sb
);
3949 ext4_lblk_t c_offset
= map
->m_lblk
& (sbi
->s_cluster_ratio
-1);
3950 ext4_lblk_t ex_cluster_start
, ex_cluster_end
;
3951 ext4_lblk_t rr_cluster_start
;
3952 ext4_lblk_t ee_block
= le32_to_cpu(ex
->ee_block
);
3953 ext4_fsblk_t ee_start
= ext4_ext_pblock(ex
);
3954 unsigned short ee_len
= ext4_ext_get_actual_len(ex
);
3956 /* The extent passed in that we are trying to match */
3957 ex_cluster_start
= EXT4_B2C(sbi
, ee_block
);
3958 ex_cluster_end
= EXT4_B2C(sbi
, ee_block
+ ee_len
- 1);
3960 /* The requested region passed into ext4_map_blocks() */
3961 rr_cluster_start
= EXT4_B2C(sbi
, map
->m_lblk
);
3963 if ((rr_cluster_start
== ex_cluster_end
) ||
3964 (rr_cluster_start
== ex_cluster_start
)) {
3965 if (rr_cluster_start
== ex_cluster_end
)
3966 ee_start
+= ee_len
- 1;
3967 map
->m_pblk
= (ee_start
& ~(sbi
->s_cluster_ratio
- 1)) +
3969 map
->m_len
= min(map
->m_len
,
3970 (unsigned) sbi
->s_cluster_ratio
- c_offset
);
3972 * Check for and handle this case:
3974 * |--------- cluster # N-------------|
3975 * |------- extent ----|
3976 * |--- requested region ---|
3980 if (map
->m_lblk
< ee_block
)
3981 map
->m_len
= min(map
->m_len
, ee_block
- map
->m_lblk
);
3984 * Check for the case where there is already another allocated
3985 * block to the right of 'ex' but before the end of the cluster.
3987 * |------------- cluster # N-------------|
3988 * |----- ex -----| |---- ex_right ----|
3989 * |------ requested region ------|
3990 * |================|
3992 if (map
->m_lblk
> ee_block
) {
3993 ext4_lblk_t next
= ext4_ext_next_allocated_block(path
);
3994 map
->m_len
= min(map
->m_len
, next
- map
->m_lblk
);
3997 trace_ext4_get_implied_cluster_alloc_exit(sb
, map
, 1);
4001 trace_ext4_get_implied_cluster_alloc_exit(sb
, map
, 0);
4007 * Block allocation/map/preallocation routine for extents based files
4010 * Need to be called with
4011 * down_read(&EXT4_I(inode)->i_data_sem) if not allocating file system block
4012 * (ie, create is zero). Otherwise down_write(&EXT4_I(inode)->i_data_sem)
4014 * return > 0, number of of blocks already mapped/allocated
4015 * if create == 0 and these are pre-allocated blocks
4016 * buffer head is unmapped
4017 * otherwise blocks are mapped
4019 * return = 0, if plain look up failed (blocks have not been allocated)
4020 * buffer head is unmapped
4022 * return < 0, error case.
4024 int ext4_ext_map_blocks(handle_t
*handle
, struct inode
*inode
,
4025 struct ext4_map_blocks
*map
, int flags
)
4027 struct ext4_ext_path
*path
= NULL
;
4028 struct ext4_extent newex
, *ex
, *ex2
;
4029 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
4030 ext4_fsblk_t newblock
= 0;
4031 int free_on_err
= 0, err
= 0, depth
;
4032 unsigned int allocated
= 0, offset
= 0;
4033 unsigned int allocated_clusters
= 0;
4034 struct ext4_allocation_request ar
;
4035 ext4_io_end_t
*io
= ext4_inode_aio(inode
);
4036 ext4_lblk_t cluster_offset
;
4037 int set_unwritten
= 0;
4039 ext_debug("blocks %u/%u requested for inode %lu\n",
4040 map
->m_lblk
, map
->m_len
, inode
->i_ino
);
4041 trace_ext4_ext_map_blocks_enter(inode
, map
->m_lblk
, map
->m_len
, flags
);
4043 /* find extent for this block */
4044 path
= ext4_ext_find_extent(inode
, map
->m_lblk
, NULL
);
4046 err
= PTR_ERR(path
);
4051 depth
= ext_depth(inode
);
4054 * consistent leaf must not be empty;
4055 * this situation is possible, though, _during_ tree modification;
4056 * this is why assert can't be put in ext4_ext_find_extent()
4058 if (unlikely(path
[depth
].p_ext
== NULL
&& depth
!= 0)) {
4059 EXT4_ERROR_INODE(inode
, "bad extent address "
4060 "lblock: %lu, depth: %d pblock %lld",
4061 (unsigned long) map
->m_lblk
, depth
,
4062 path
[depth
].p_block
);
4067 ex
= path
[depth
].p_ext
;
4069 ext4_lblk_t ee_block
= le32_to_cpu(ex
->ee_block
);
4070 ext4_fsblk_t ee_start
= ext4_ext_pblock(ex
);
4071 unsigned short ee_len
;
4074 * Uninitialized extents are treated as holes, except that
4075 * we split out initialized portions during a write.
4077 ee_len
= ext4_ext_get_actual_len(ex
);
4079 trace_ext4_ext_show_extent(inode
, ee_block
, ee_start
, ee_len
);
4081 /* if found extent covers block, simply return it */
4082 if (in_range(map
->m_lblk
, ee_block
, ee_len
)) {
4083 newblock
= map
->m_lblk
- ee_block
+ ee_start
;
4084 /* number of remaining blocks in the extent */
4085 allocated
= ee_len
- (map
->m_lblk
- ee_block
);
4086 ext_debug("%u fit into %u:%d -> %llu\n", map
->m_lblk
,
4087 ee_block
, ee_len
, newblock
);
4089 if (!ext4_ext_is_uninitialized(ex
))
4092 allocated
= ext4_ext_handle_uninitialized_extents(
4093 handle
, inode
, map
, path
, flags
,
4094 allocated
, newblock
);
4099 if ((sbi
->s_cluster_ratio
> 1) &&
4100 ext4_find_delalloc_cluster(inode
, map
->m_lblk
))
4101 map
->m_flags
|= EXT4_MAP_FROM_CLUSTER
;
4104 * requested block isn't allocated yet;
4105 * we couldn't try to create block if create flag is zero
4107 if ((flags
& EXT4_GET_BLOCKS_CREATE
) == 0) {
4109 * put just found gap into cache to speed up
4110 * subsequent requests
4112 if ((flags
& EXT4_GET_BLOCKS_NO_PUT_HOLE
) == 0)
4113 ext4_ext_put_gap_in_cache(inode
, path
, map
->m_lblk
);
4118 * Okay, we need to do block allocation.
4120 map
->m_flags
&= ~EXT4_MAP_FROM_CLUSTER
;
4121 newex
.ee_block
= cpu_to_le32(map
->m_lblk
);
4122 cluster_offset
= map
->m_lblk
& (sbi
->s_cluster_ratio
-1);
4125 * If we are doing bigalloc, check to see if the extent returned
4126 * by ext4_ext_find_extent() implies a cluster we can use.
4128 if (cluster_offset
&& ex
&&
4129 get_implied_cluster_alloc(inode
->i_sb
, map
, ex
, path
)) {
4130 ar
.len
= allocated
= map
->m_len
;
4131 newblock
= map
->m_pblk
;
4132 map
->m_flags
|= EXT4_MAP_FROM_CLUSTER
;
4133 goto got_allocated_blocks
;
4136 /* find neighbour allocated blocks */
4137 ar
.lleft
= map
->m_lblk
;
4138 err
= ext4_ext_search_left(inode
, path
, &ar
.lleft
, &ar
.pleft
);
4141 ar
.lright
= map
->m_lblk
;
4143 err
= ext4_ext_search_right(inode
, path
, &ar
.lright
, &ar
.pright
, &ex2
);
4147 /* Check if the extent after searching to the right implies a
4148 * cluster we can use. */
4149 if ((sbi
->s_cluster_ratio
> 1) && ex2
&&
4150 get_implied_cluster_alloc(inode
->i_sb
, map
, ex2
, path
)) {
4151 ar
.len
= allocated
= map
->m_len
;
4152 newblock
= map
->m_pblk
;
4153 map
->m_flags
|= EXT4_MAP_FROM_CLUSTER
;
4154 goto got_allocated_blocks
;
4158 * See if request is beyond maximum number of blocks we can have in
4159 * a single extent. For an initialized extent this limit is
4160 * EXT_INIT_MAX_LEN and for an uninitialized extent this limit is
4161 * EXT_UNINIT_MAX_LEN.
4163 if (map
->m_len
> EXT_INIT_MAX_LEN
&&
4164 !(flags
& EXT4_GET_BLOCKS_UNINIT_EXT
))
4165 map
->m_len
= EXT_INIT_MAX_LEN
;
4166 else if (map
->m_len
> EXT_UNINIT_MAX_LEN
&&
4167 (flags
& EXT4_GET_BLOCKS_UNINIT_EXT
))
4168 map
->m_len
= EXT_UNINIT_MAX_LEN
;
4170 /* Check if we can really insert (m_lblk)::(m_lblk + m_len) extent */
4171 newex
.ee_len
= cpu_to_le16(map
->m_len
);
4172 err
= ext4_ext_check_overlap(sbi
, inode
, &newex
, path
);
4174 allocated
= ext4_ext_get_actual_len(&newex
);
4176 allocated
= map
->m_len
;
4178 /* allocate new block */
4180 ar
.goal
= ext4_ext_find_goal(inode
, path
, map
->m_lblk
);
4181 ar
.logical
= map
->m_lblk
;
4183 * We calculate the offset from the beginning of the cluster
4184 * for the logical block number, since when we allocate a
4185 * physical cluster, the physical block should start at the
4186 * same offset from the beginning of the cluster. This is
4187 * needed so that future calls to get_implied_cluster_alloc()
4190 offset
= map
->m_lblk
& (sbi
->s_cluster_ratio
- 1);
4191 ar
.len
= EXT4_NUM_B2C(sbi
, offset
+allocated
);
4193 ar
.logical
-= offset
;
4194 if (S_ISREG(inode
->i_mode
))
4195 ar
.flags
= EXT4_MB_HINT_DATA
;
4197 /* disable in-core preallocation for non-regular files */
4199 if (flags
& EXT4_GET_BLOCKS_NO_NORMALIZE
)
4200 ar
.flags
|= EXT4_MB_HINT_NOPREALLOC
;
4201 newblock
= ext4_mb_new_blocks(handle
, &ar
, &err
);
4204 ext_debug("allocate new block: goal %llu, found %llu/%u\n",
4205 ar
.goal
, newblock
, allocated
);
4207 allocated_clusters
= ar
.len
;
4208 ar
.len
= EXT4_C2B(sbi
, ar
.len
) - offset
;
4209 if (ar
.len
> allocated
)
4212 got_allocated_blocks
:
4213 /* try to insert new extent into found leaf and return */
4214 ext4_ext_store_pblock(&newex
, newblock
+ offset
);
4215 newex
.ee_len
= cpu_to_le16(ar
.len
);
4216 /* Mark uninitialized */
4217 if (flags
& EXT4_GET_BLOCKS_UNINIT_EXT
){
4218 ext4_ext_mark_uninitialized(&newex
);
4219 map
->m_flags
|= EXT4_MAP_UNWRITTEN
;
4221 * io_end structure was created for every IO write to an
4222 * uninitialized extent. To avoid unnecessary conversion,
4223 * here we flag the IO that really needs the conversion.
4224 * For non asycn direct IO case, flag the inode state
4225 * that we need to perform conversion when IO is done.
4227 if ((flags
& EXT4_GET_BLOCKS_PRE_IO
))
4229 if (ext4_should_dioread_nolock(inode
))
4230 map
->m_flags
|= EXT4_MAP_UNINIT
;
4234 if ((flags
& EXT4_GET_BLOCKS_KEEP_SIZE
) == 0)
4235 err
= check_eofblocks_fl(handle
, inode
, map
->m_lblk
,
4238 err
= ext4_ext_insert_extent(handle
, inode
, path
,
4241 if (!err
&& set_unwritten
) {
4243 ext4_set_io_unwritten_flag(inode
, io
);
4245 ext4_set_inode_state(inode
,
4246 EXT4_STATE_DIO_UNWRITTEN
);
4249 if (err
&& free_on_err
) {
4250 int fb_flags
= flags
& EXT4_GET_BLOCKS_DELALLOC_RESERVE
?
4251 EXT4_FREE_BLOCKS_NO_QUOT_UPDATE
: 0;
4252 /* free data blocks we just allocated */
4253 /* not a good idea to call discard here directly,
4254 * but otherwise we'd need to call it every free() */
4255 ext4_discard_preallocations(inode
);
4256 ext4_free_blocks(handle
, inode
, NULL
, newblock
,
4257 EXT4_C2B(sbi
, allocated_clusters
), fb_flags
);
4261 /* previous routine could use block we allocated */
4262 newblock
= ext4_ext_pblock(&newex
);
4263 allocated
= ext4_ext_get_actual_len(&newex
);
4264 if (allocated
> map
->m_len
)
4265 allocated
= map
->m_len
;
4266 map
->m_flags
|= EXT4_MAP_NEW
;
4269 * Update reserved blocks/metadata blocks after successful
4270 * block allocation which had been deferred till now.
4272 if (flags
& EXT4_GET_BLOCKS_DELALLOC_RESERVE
) {
4273 unsigned int reserved_clusters
;
4275 * Check how many clusters we had reserved this allocated range
4277 reserved_clusters
= get_reserved_cluster_alloc(inode
,
4278 map
->m_lblk
, allocated
);
4279 if (map
->m_flags
& EXT4_MAP_FROM_CLUSTER
) {
4280 if (reserved_clusters
) {
4282 * We have clusters reserved for this range.
4283 * But since we are not doing actual allocation
4284 * and are simply using blocks from previously
4285 * allocated cluster, we should release the
4286 * reservation and not claim quota.
4288 ext4_da_update_reserve_space(inode
,
4289 reserved_clusters
, 0);
4292 BUG_ON(allocated_clusters
< reserved_clusters
);
4293 if (reserved_clusters
< allocated_clusters
) {
4294 struct ext4_inode_info
*ei
= EXT4_I(inode
);
4295 int reservation
= allocated_clusters
-
4298 * It seems we claimed few clusters outside of
4299 * the range of this allocation. We should give
4300 * it back to the reservation pool. This can
4301 * happen in the following case:
4303 * * Suppose s_cluster_ratio is 4 (i.e., each
4304 * cluster has 4 blocks. Thus, the clusters
4305 * are [0-3],[4-7],[8-11]...
4306 * * First comes delayed allocation write for
4307 * logical blocks 10 & 11. Since there were no
4308 * previous delayed allocated blocks in the
4309 * range [8-11], we would reserve 1 cluster
4311 * * Next comes write for logical blocks 3 to 8.
4312 * In this case, we will reserve 2 clusters
4313 * (for [0-3] and [4-7]; and not for [8-11] as
4314 * that range has a delayed allocated blocks.
4315 * Thus total reserved clusters now becomes 3.
4316 * * Now, during the delayed allocation writeout
4317 * time, we will first write blocks [3-8] and
4318 * allocate 3 clusters for writing these
4319 * blocks. Also, we would claim all these
4320 * three clusters above.
4321 * * Now when we come here to writeout the
4322 * blocks [10-11], we would expect to claim
4323 * the reservation of 1 cluster we had made
4324 * (and we would claim it since there are no
4325 * more delayed allocated blocks in the range
4326 * [8-11]. But our reserved cluster count had
4327 * already gone to 0.
4329 * Thus, at the step 4 above when we determine
4330 * that there are still some unwritten delayed
4331 * allocated blocks outside of our current
4332 * block range, we should increment the
4333 * reserved clusters count so that when the
4334 * remaining blocks finally gets written, we
4337 dquot_reserve_block(inode
,
4338 EXT4_C2B(sbi
, reservation
));
4339 spin_lock(&ei
->i_block_reservation_lock
);
4340 ei
->i_reserved_data_blocks
+= reservation
;
4341 spin_unlock(&ei
->i_block_reservation_lock
);
4344 * We will claim quota for all newly allocated blocks.
4345 * We're updating the reserved space *after* the
4346 * correction above so we do not accidentally free
4347 * all the metadata reservation because we might
4348 * actually need it later on.
4350 ext4_da_update_reserve_space(inode
, allocated_clusters
,
4356 * Cache the extent and update transaction to commit on fdatasync only
4357 * when it is _not_ an uninitialized extent.
4359 if ((flags
& EXT4_GET_BLOCKS_UNINIT_EXT
) == 0)
4360 ext4_update_inode_fsync_trans(handle
, inode
, 1);
4362 ext4_update_inode_fsync_trans(handle
, inode
, 0);
4364 if (allocated
> map
->m_len
)
4365 allocated
= map
->m_len
;
4366 ext4_ext_show_leaf(inode
, path
);
4367 map
->m_flags
|= EXT4_MAP_MAPPED
;
4368 map
->m_pblk
= newblock
;
4369 map
->m_len
= allocated
;
4372 ext4_ext_drop_refs(path
);
4377 trace_ext4_ext_map_blocks_exit(inode
, flags
, map
,
4378 err
? err
: allocated
);
4379 ext4_es_lru_add(inode
);
4380 return err
? err
: allocated
;
4383 void ext4_ext_truncate(handle_t
*handle
, struct inode
*inode
)
4385 struct super_block
*sb
= inode
->i_sb
;
4386 ext4_lblk_t last_block
;
4390 * TODO: optimization is possible here.
4391 * Probably we need not scan at all,
4392 * because page truncation is enough.
4395 /* we have to know where to truncate from in crash case */
4396 EXT4_I(inode
)->i_disksize
= inode
->i_size
;
4397 ext4_mark_inode_dirty(handle
, inode
);
4399 last_block
= (inode
->i_size
+ sb
->s_blocksize
- 1)
4400 >> EXT4_BLOCK_SIZE_BITS(sb
);
4402 err
= ext4_es_remove_extent(inode
, last_block
,
4403 EXT_MAX_BLOCKS
- last_block
);
4404 if (err
== -ENOMEM
) {
4406 congestion_wait(BLK_RW_ASYNC
, HZ
/50);
4410 ext4_std_error(inode
->i_sb
, err
);
4413 err
= ext4_ext_remove_space(inode
, last_block
, EXT_MAX_BLOCKS
- 1);
4414 ext4_std_error(inode
->i_sb
, err
);
4417 static void ext4_falloc_update_inode(struct inode
*inode
,
4418 int mode
, loff_t new_size
, int update_ctime
)
4420 struct timespec now
;
4423 now
= current_fs_time(inode
->i_sb
);
4424 if (!timespec_equal(&inode
->i_ctime
, &now
))
4425 inode
->i_ctime
= now
;
4428 * Update only when preallocation was requested beyond
4431 if (!(mode
& FALLOC_FL_KEEP_SIZE
)) {
4432 if (new_size
> i_size_read(inode
))
4433 i_size_write(inode
, new_size
);
4434 if (new_size
> EXT4_I(inode
)->i_disksize
)
4435 ext4_update_i_disksize(inode
, new_size
);
4438 * Mark that we allocate beyond EOF so the subsequent truncate
4439 * can proceed even if the new size is the same as i_size.
4441 if (new_size
> i_size_read(inode
))
4442 ext4_set_inode_flag(inode
, EXT4_INODE_EOFBLOCKS
);
4448 * preallocate space for a file. This implements ext4's fallocate file
4449 * operation, which gets called from sys_fallocate system call.
4450 * For block-mapped files, posix_fallocate should fall back to the method
4451 * of writing zeroes to the required new blocks (the same behavior which is
4452 * expected for file systems which do not support fallocate() system call).
4454 long ext4_fallocate(struct file
*file
, int mode
, loff_t offset
, loff_t len
)
4456 struct inode
*inode
= file_inode(file
);
4459 unsigned int max_blocks
;
4464 struct ext4_map_blocks map
;
4465 unsigned int credits
, blkbits
= inode
->i_blkbits
;
4467 /* Return error if mode is not supported */
4468 if (mode
& ~(FALLOC_FL_KEEP_SIZE
| FALLOC_FL_PUNCH_HOLE
))
4471 if (mode
& FALLOC_FL_PUNCH_HOLE
)
4472 return ext4_punch_hole(inode
, offset
, len
);
4474 ret
= ext4_convert_inline_data(inode
);
4479 * currently supporting (pre)allocate mode for extent-based
4482 if (!(ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
)))
4485 trace_ext4_fallocate_enter(inode
, offset
, len
, mode
);
4486 map
.m_lblk
= offset
>> blkbits
;
4488 * We can't just convert len to max_blocks because
4489 * If blocksize = 4096 offset = 3072 and len = 2048
4491 max_blocks
= (EXT4_BLOCK_ALIGN(len
+ offset
, blkbits
) >> blkbits
)
4494 * credits to insert 1 extent into extent tree
4496 credits
= ext4_chunk_trans_blocks(inode
, max_blocks
);
4497 mutex_lock(&inode
->i_mutex
);
4498 ret
= inode_newsize_ok(inode
, (len
+ offset
));
4500 mutex_unlock(&inode
->i_mutex
);
4501 trace_ext4_fallocate_exit(inode
, offset
, max_blocks
, ret
);
4504 flags
= EXT4_GET_BLOCKS_CREATE_UNINIT_EXT
;
4505 if (mode
& FALLOC_FL_KEEP_SIZE
)
4506 flags
|= EXT4_GET_BLOCKS_KEEP_SIZE
;
4508 * Don't normalize the request if it can fit in one extent so
4509 * that it doesn't get unnecessarily split into multiple
4512 if (len
<= EXT_UNINIT_MAX_LEN
<< blkbits
)
4513 flags
|= EXT4_GET_BLOCKS_NO_NORMALIZE
;
4516 while (ret
>= 0 && ret
< max_blocks
) {
4517 map
.m_lblk
= map
.m_lblk
+ ret
;
4518 map
.m_len
= max_blocks
= max_blocks
- ret
;
4519 handle
= ext4_journal_start(inode
, EXT4_HT_MAP_BLOCKS
,
4521 if (IS_ERR(handle
)) {
4522 ret
= PTR_ERR(handle
);
4525 ret
= ext4_map_blocks(handle
, inode
, &map
, flags
);
4528 ext4_warning(inode
->i_sb
,
4529 "inode #%lu: block %u: len %u: "
4530 "ext4_ext_map_blocks returned %d",
4531 inode
->i_ino
, map
.m_lblk
,
4534 ext4_mark_inode_dirty(handle
, inode
);
4535 ret2
= ext4_journal_stop(handle
);
4538 if ((map
.m_lblk
+ ret
) >= (EXT4_BLOCK_ALIGN(offset
+ len
,
4539 blkbits
) >> blkbits
))
4540 new_size
= offset
+ len
;
4542 new_size
= ((loff_t
) map
.m_lblk
+ ret
) << blkbits
;
4544 ext4_falloc_update_inode(inode
, mode
, new_size
,
4545 (map
.m_flags
& EXT4_MAP_NEW
));
4546 ext4_mark_inode_dirty(handle
, inode
);
4547 if ((file
->f_flags
& O_SYNC
) && ret
>= max_blocks
)
4548 ext4_handle_sync(handle
);
4549 ret2
= ext4_journal_stop(handle
);
4553 if (ret
== -ENOSPC
&&
4554 ext4_should_retry_alloc(inode
->i_sb
, &retries
)) {
4558 mutex_unlock(&inode
->i_mutex
);
4559 trace_ext4_fallocate_exit(inode
, offset
, max_blocks
,
4560 ret
> 0 ? ret2
: ret
);
4561 return ret
> 0 ? ret2
: ret
;
4565 * This function convert a range of blocks to written extents
4566 * The caller of this function will pass the start offset and the size.
4567 * all unwritten extents within this range will be converted to
4570 * This function is called from the direct IO end io call back
4571 * function, to convert the fallocated extents after IO is completed.
4572 * Returns 0 on success.
4574 int ext4_convert_unwritten_extents(handle_t
*handle
, struct inode
*inode
,
4575 loff_t offset
, ssize_t len
)
4577 unsigned int max_blocks
;
4580 struct ext4_map_blocks map
;
4581 unsigned int credits
, blkbits
= inode
->i_blkbits
;
4583 map
.m_lblk
= offset
>> blkbits
;
4585 * We can't just convert len to max_blocks because
4586 * If blocksize = 4096 offset = 3072 and len = 2048
4588 max_blocks
= ((EXT4_BLOCK_ALIGN(len
+ offset
, blkbits
) >> blkbits
) -
4591 * This is somewhat ugly but the idea is clear: When transaction is
4592 * reserved, everything goes into it. Otherwise we rather start several
4593 * smaller transactions for conversion of each extent separately.
4596 handle
= ext4_journal_start_reserved(handle
,
4597 EXT4_HT_EXT_CONVERT
);
4599 return PTR_ERR(handle
);
4603 * credits to insert 1 extent into extent tree
4605 credits
= ext4_chunk_trans_blocks(inode
, max_blocks
);
4607 while (ret
>= 0 && ret
< max_blocks
) {
4609 map
.m_len
= (max_blocks
-= ret
);
4611 handle
= ext4_journal_start(inode
, EXT4_HT_MAP_BLOCKS
,
4613 if (IS_ERR(handle
)) {
4614 ret
= PTR_ERR(handle
);
4618 ret
= ext4_map_blocks(handle
, inode
, &map
,
4619 EXT4_GET_BLOCKS_IO_CONVERT_EXT
);
4621 ext4_warning(inode
->i_sb
,
4622 "inode #%lu: block %u: len %u: "
4623 "ext4_ext_map_blocks returned %d",
4624 inode
->i_ino
, map
.m_lblk
,
4626 ext4_mark_inode_dirty(handle
, inode
);
4628 ret2
= ext4_journal_stop(handle
);
4629 if (ret
<= 0 || ret2
)
4633 ret2
= ext4_journal_stop(handle
);
4634 return ret
> 0 ? ret2
: ret
;
4638 * If newes is not existing extent (newes->ec_pblk equals zero) find
4639 * delayed extent at start of newes and update newes accordingly and
4640 * return start of the next delayed extent.
4642 * If newes is existing extent (newes->ec_pblk is not equal zero)
4643 * return start of next delayed extent or EXT_MAX_BLOCKS if no delayed
4644 * extent found. Leave newes unmodified.
4646 static int ext4_find_delayed_extent(struct inode
*inode
,
4647 struct extent_status
*newes
)
4649 struct extent_status es
;
4650 ext4_lblk_t block
, next_del
;
4652 if (newes
->es_pblk
== 0) {
4653 ext4_es_find_delayed_extent_range(inode
, newes
->es_lblk
,
4654 newes
->es_lblk
+ newes
->es_len
- 1, &es
);
4657 * No extent in extent-tree contains block @newes->es_pblk,
4658 * then the block may stay in 1)a hole or 2)delayed-extent.
4664 if (es
.es_lblk
> newes
->es_lblk
) {
4666 newes
->es_len
= min(es
.es_lblk
- newes
->es_lblk
,
4671 newes
->es_len
= es
.es_lblk
+ es
.es_len
- newes
->es_lblk
;
4674 block
= newes
->es_lblk
+ newes
->es_len
;
4675 ext4_es_find_delayed_extent_range(inode
, block
, EXT_MAX_BLOCKS
, &es
);
4677 next_del
= EXT_MAX_BLOCKS
;
4679 next_del
= es
.es_lblk
;
4683 /* fiemap flags we can handle specified here */
4684 #define EXT4_FIEMAP_FLAGS (FIEMAP_FLAG_SYNC|FIEMAP_FLAG_XATTR)
4686 static int ext4_xattr_fiemap(struct inode
*inode
,
4687 struct fiemap_extent_info
*fieinfo
)
4691 __u32 flags
= FIEMAP_EXTENT_LAST
;
4692 int blockbits
= inode
->i_sb
->s_blocksize_bits
;
4696 if (ext4_test_inode_state(inode
, EXT4_STATE_XATTR
)) {
4697 struct ext4_iloc iloc
;
4698 int offset
; /* offset of xattr in inode */
4700 error
= ext4_get_inode_loc(inode
, &iloc
);
4703 physical
= (__u64
)iloc
.bh
->b_blocknr
<< blockbits
;
4704 offset
= EXT4_GOOD_OLD_INODE_SIZE
+
4705 EXT4_I(inode
)->i_extra_isize
;
4707 length
= EXT4_SB(inode
->i_sb
)->s_inode_size
- offset
;
4708 flags
|= FIEMAP_EXTENT_DATA_INLINE
;
4710 } else { /* external block */
4711 physical
= (__u64
)EXT4_I(inode
)->i_file_acl
<< blockbits
;
4712 length
= inode
->i_sb
->s_blocksize
;
4716 error
= fiemap_fill_next_extent(fieinfo
, 0, physical
,
4718 return (error
< 0 ? error
: 0);
4721 int ext4_fiemap(struct inode
*inode
, struct fiemap_extent_info
*fieinfo
,
4722 __u64 start
, __u64 len
)
4724 ext4_lblk_t start_blk
;
4727 if (ext4_has_inline_data(inode
)) {
4730 error
= ext4_inline_data_fiemap(inode
, fieinfo
, &has_inline
);
4736 /* fallback to generic here if not in extents fmt */
4737 if (!(ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
)))
4738 return generic_block_fiemap(inode
, fieinfo
, start
, len
,
4741 if (fiemap_check_flags(fieinfo
, EXT4_FIEMAP_FLAGS
))
4744 if (fieinfo
->fi_flags
& FIEMAP_FLAG_XATTR
) {
4745 error
= ext4_xattr_fiemap(inode
, fieinfo
);
4747 ext4_lblk_t len_blks
;
4750 start_blk
= start
>> inode
->i_sb
->s_blocksize_bits
;
4751 last_blk
= (start
+ len
- 1) >> inode
->i_sb
->s_blocksize_bits
;
4752 if (last_blk
>= EXT_MAX_BLOCKS
)
4753 last_blk
= EXT_MAX_BLOCKS
-1;
4754 len_blks
= ((ext4_lblk_t
) last_blk
) - start_blk
+ 1;
4757 * Walk the extent tree gathering extent information
4758 * and pushing extents back to the user.
4760 error
= ext4_fill_fiemap_extents(inode
, start_blk
,