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 License
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 <asm/uaccess.h>
41 #include <linux/fiemap.h>
42 #include <linux/backing-dev.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_UNWRIT1 0x2 /* mark first half unwritten */
55 #define EXT4_EXT_MARK_UNWRIT2 0x4 /* mark second half unwritten */
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_metadata_csum(inode
->i_sb
))
80 et
= find_ext4_extent_tail(eh
);
81 if (et
->et_checksum
!= ext4_extent_block_csum(inode
, eh
))
86 static void ext4_extent_block_csum_set(struct inode
*inode
,
87 struct ext4_extent_header
*eh
)
89 struct ext4_extent_tail
*et
;
91 if (!ext4_has_metadata_csum(inode
->i_sb
))
94 et
= find_ext4_extent_tail(eh
);
95 et
->et_checksum
= ext4_extent_block_csum(inode
, eh
);
98 static int ext4_split_extent(handle_t
*handle
,
100 struct ext4_ext_path
**ppath
,
101 struct ext4_map_blocks
*map
,
105 static int ext4_split_extent_at(handle_t
*handle
,
107 struct ext4_ext_path
**ppath
,
112 static int ext4_find_delayed_extent(struct inode
*inode
,
113 struct extent_status
*newes
);
115 static int ext4_ext_truncate_extend_restart(handle_t
*handle
,
121 if (!ext4_handle_valid(handle
))
123 if (handle
->h_buffer_credits
>= needed
)
126 * If we need to extend the journal get a few extra blocks
127 * while we're at it for efficiency's sake.
130 err
= ext4_journal_extend(handle
, needed
- handle
->h_buffer_credits
);
133 err
= ext4_truncate_restart_trans(handle
, inode
, needed
);
145 static int ext4_ext_get_access(handle_t
*handle
, struct inode
*inode
,
146 struct ext4_ext_path
*path
)
149 /* path points to block */
150 BUFFER_TRACE(path
->p_bh
, "get_write_access");
151 return ext4_journal_get_write_access(handle
, path
->p_bh
);
153 /* path points to leaf/index in inode body */
154 /* we use in-core data, no need to protect them */
164 int __ext4_ext_dirty(const char *where
, unsigned int line
, handle_t
*handle
,
165 struct inode
*inode
, struct ext4_ext_path
*path
)
169 WARN_ON(!rwsem_is_locked(&EXT4_I(inode
)->i_data_sem
));
171 ext4_extent_block_csum_set(inode
, ext_block_hdr(path
->p_bh
));
172 /* path points to block */
173 err
= __ext4_handle_dirty_metadata(where
, line
, handle
,
176 /* path points to leaf/index in inode body */
177 err
= ext4_mark_inode_dirty(handle
, inode
);
182 static ext4_fsblk_t
ext4_ext_find_goal(struct inode
*inode
,
183 struct ext4_ext_path
*path
,
187 int depth
= path
->p_depth
;
188 struct ext4_extent
*ex
;
191 * Try to predict block placement assuming that we are
192 * filling in a file which will eventually be
193 * non-sparse --- i.e., in the case of libbfd writing
194 * an ELF object sections out-of-order but in a way
195 * the eventually results in a contiguous object or
196 * executable file, or some database extending a table
197 * space file. However, this is actually somewhat
198 * non-ideal if we are writing a sparse file such as
199 * qemu or KVM writing a raw image file that is going
200 * to stay fairly sparse, since it will end up
201 * fragmenting the file system's free space. Maybe we
202 * should have some hueristics or some way to allow
203 * userspace to pass a hint to file system,
204 * especially if the latter case turns out to be
207 ex
= path
[depth
].p_ext
;
209 ext4_fsblk_t ext_pblk
= ext4_ext_pblock(ex
);
210 ext4_lblk_t ext_block
= le32_to_cpu(ex
->ee_block
);
212 if (block
> ext_block
)
213 return ext_pblk
+ (block
- ext_block
);
215 return ext_pblk
- (ext_block
- block
);
218 /* it looks like index is empty;
219 * try to find starting block from index itself */
220 if (path
[depth
].p_bh
)
221 return path
[depth
].p_bh
->b_blocknr
;
224 /* OK. use inode's group */
225 return ext4_inode_to_goal_block(inode
);
229 * Allocation for a meta data block
232 ext4_ext_new_meta_block(handle_t
*handle
, struct inode
*inode
,
233 struct ext4_ext_path
*path
,
234 struct ext4_extent
*ex
, int *err
, unsigned int flags
)
236 ext4_fsblk_t goal
, newblock
;
238 goal
= ext4_ext_find_goal(inode
, path
, le32_to_cpu(ex
->ee_block
));
239 newblock
= ext4_new_meta_blocks(handle
, inode
, goal
, flags
,
244 static inline int ext4_ext_space_block(struct inode
*inode
, int check
)
248 size
= (inode
->i_sb
->s_blocksize
- sizeof(struct ext4_extent_header
))
249 / sizeof(struct ext4_extent
);
250 #ifdef AGGRESSIVE_TEST
251 if (!check
&& size
> 6)
257 static inline int ext4_ext_space_block_idx(struct inode
*inode
, int check
)
261 size
= (inode
->i_sb
->s_blocksize
- sizeof(struct ext4_extent_header
))
262 / sizeof(struct ext4_extent_idx
);
263 #ifdef AGGRESSIVE_TEST
264 if (!check
&& size
> 5)
270 static inline int ext4_ext_space_root(struct inode
*inode
, int check
)
274 size
= sizeof(EXT4_I(inode
)->i_data
);
275 size
-= sizeof(struct ext4_extent_header
);
276 size
/= sizeof(struct ext4_extent
);
277 #ifdef AGGRESSIVE_TEST
278 if (!check
&& size
> 3)
284 static inline int ext4_ext_space_root_idx(struct inode
*inode
, int check
)
288 size
= sizeof(EXT4_I(inode
)->i_data
);
289 size
-= sizeof(struct ext4_extent_header
);
290 size
/= sizeof(struct ext4_extent_idx
);
291 #ifdef AGGRESSIVE_TEST
292 if (!check
&& size
> 4)
299 ext4_force_split_extent_at(handle_t
*handle
, struct inode
*inode
,
300 struct ext4_ext_path
**ppath
, ext4_lblk_t lblk
,
303 struct ext4_ext_path
*path
= *ppath
;
304 int unwritten
= ext4_ext_is_unwritten(path
[path
->p_depth
].p_ext
);
306 return ext4_split_extent_at(handle
, inode
, ppath
, lblk
, unwritten
?
307 EXT4_EXT_MARK_UNWRIT1
|EXT4_EXT_MARK_UNWRIT2
: 0,
308 EXT4_EX_NOCACHE
| EXT4_GET_BLOCKS_PRE_IO
|
309 (nofail
? EXT4_GET_BLOCKS_METADATA_NOFAIL
:0));
313 * Calculate the number of metadata blocks needed
314 * to allocate @blocks
315 * Worse case is one block per extent
317 int ext4_ext_calc_metadata_amount(struct inode
*inode
, ext4_lblk_t lblock
)
319 struct ext4_inode_info
*ei
= EXT4_I(inode
);
322 idxs
= ((inode
->i_sb
->s_blocksize
- sizeof(struct ext4_extent_header
))
323 / sizeof(struct ext4_extent_idx
));
326 * If the new delayed allocation block is contiguous with the
327 * previous da block, it can share index blocks with the
328 * previous block, so we only need to allocate a new index
329 * block every idxs leaf blocks. At ldxs**2 blocks, we need
330 * an additional index block, and at ldxs**3 blocks, yet
331 * another index blocks.
333 if (ei
->i_da_metadata_calc_len
&&
334 ei
->i_da_metadata_calc_last_lblock
+1 == lblock
) {
337 if ((ei
->i_da_metadata_calc_len
% idxs
) == 0)
339 if ((ei
->i_da_metadata_calc_len
% (idxs
*idxs
)) == 0)
341 if ((ei
->i_da_metadata_calc_len
% (idxs
*idxs
*idxs
)) == 0) {
343 ei
->i_da_metadata_calc_len
= 0;
345 ei
->i_da_metadata_calc_len
++;
346 ei
->i_da_metadata_calc_last_lblock
++;
351 * In the worst case we need a new set of index blocks at
352 * every level of the inode's extent tree.
354 ei
->i_da_metadata_calc_len
= 1;
355 ei
->i_da_metadata_calc_last_lblock
= lblock
;
356 return ext_depth(inode
) + 1;
360 ext4_ext_max_entries(struct inode
*inode
, int depth
)
364 if (depth
== ext_depth(inode
)) {
366 max
= ext4_ext_space_root(inode
, 1);
368 max
= ext4_ext_space_root_idx(inode
, 1);
371 max
= ext4_ext_space_block(inode
, 1);
373 max
= ext4_ext_space_block_idx(inode
, 1);
379 static int ext4_valid_extent(struct inode
*inode
, struct ext4_extent
*ext
)
381 ext4_fsblk_t block
= ext4_ext_pblock(ext
);
382 int len
= ext4_ext_get_actual_len(ext
);
383 ext4_lblk_t lblock
= le32_to_cpu(ext
->ee_block
);
384 ext4_lblk_t last
= lblock
+ len
- 1;
386 if (len
== 0 || lblock
> last
)
388 return ext4_data_block_valid(EXT4_SB(inode
->i_sb
), block
, len
);
391 static int ext4_valid_extent_idx(struct inode
*inode
,
392 struct ext4_extent_idx
*ext_idx
)
394 ext4_fsblk_t block
= ext4_idx_pblock(ext_idx
);
396 return ext4_data_block_valid(EXT4_SB(inode
->i_sb
), block
, 1);
399 static int ext4_valid_extent_entries(struct inode
*inode
,
400 struct ext4_extent_header
*eh
,
403 unsigned short entries
;
404 if (eh
->eh_entries
== 0)
407 entries
= le16_to_cpu(eh
->eh_entries
);
411 struct ext4_extent
*ext
= EXT_FIRST_EXTENT(eh
);
412 struct ext4_super_block
*es
= EXT4_SB(inode
->i_sb
)->s_es
;
413 ext4_fsblk_t pblock
= 0;
414 ext4_lblk_t lblock
= 0;
415 ext4_lblk_t prev
= 0;
418 if (!ext4_valid_extent(inode
, ext
))
421 /* Check for overlapping extents */
422 lblock
= le32_to_cpu(ext
->ee_block
);
423 len
= ext4_ext_get_actual_len(ext
);
424 if ((lblock
<= prev
) && prev
) {
425 pblock
= ext4_ext_pblock(ext
);
426 es
->s_last_error_block
= cpu_to_le64(pblock
);
431 prev
= lblock
+ len
- 1;
434 struct ext4_extent_idx
*ext_idx
= EXT_FIRST_INDEX(eh
);
436 if (!ext4_valid_extent_idx(inode
, ext_idx
))
445 static int __ext4_ext_check(const char *function
, unsigned int line
,
446 struct inode
*inode
, struct ext4_extent_header
*eh
,
447 int depth
, ext4_fsblk_t pblk
)
449 const char *error_msg
;
450 int max
= 0, err
= -EFSCORRUPTED
;
452 if (unlikely(eh
->eh_magic
!= EXT4_EXT_MAGIC
)) {
453 error_msg
= "invalid magic";
456 if (unlikely(le16_to_cpu(eh
->eh_depth
) != depth
)) {
457 error_msg
= "unexpected eh_depth";
460 if (unlikely(eh
->eh_max
== 0)) {
461 error_msg
= "invalid eh_max";
464 max
= ext4_ext_max_entries(inode
, depth
);
465 if (unlikely(le16_to_cpu(eh
->eh_max
) > max
)) {
466 error_msg
= "too large eh_max";
469 if (unlikely(le16_to_cpu(eh
->eh_entries
) > le16_to_cpu(eh
->eh_max
))) {
470 error_msg
= "invalid eh_entries";
473 if (!ext4_valid_extent_entries(inode
, eh
, depth
)) {
474 error_msg
= "invalid extent entries";
477 /* Verify checksum on non-root extent tree nodes */
478 if (ext_depth(inode
) != depth
&&
479 !ext4_extent_block_csum_verify(inode
, eh
)) {
480 error_msg
= "extent tree corrupted";
487 ext4_error_inode(inode
, function
, line
, 0,
488 "pblk %llu bad header/extent: %s - magic %x, "
489 "entries %u, max %u(%u), depth %u(%u)",
490 (unsigned long long) pblk
, error_msg
,
491 le16_to_cpu(eh
->eh_magic
),
492 le16_to_cpu(eh
->eh_entries
), le16_to_cpu(eh
->eh_max
),
493 max
, le16_to_cpu(eh
->eh_depth
), depth
);
497 #define ext4_ext_check(inode, eh, depth, pblk) \
498 __ext4_ext_check(__func__, __LINE__, (inode), (eh), (depth), (pblk))
500 int ext4_ext_check_inode(struct inode
*inode
)
502 return ext4_ext_check(inode
, ext_inode_hdr(inode
), ext_depth(inode
), 0);
505 static struct buffer_head
*
506 __read_extent_tree_block(const char *function
, unsigned int line
,
507 struct inode
*inode
, ext4_fsblk_t pblk
, int depth
,
510 struct buffer_head
*bh
;
513 bh
= sb_getblk_gfp(inode
->i_sb
, pblk
, __GFP_MOVABLE
| GFP_NOFS
);
515 return ERR_PTR(-ENOMEM
);
517 if (!bh_uptodate_or_lock(bh
)) {
518 trace_ext4_ext_load_extent(inode
, pblk
, _RET_IP_
);
519 err
= bh_submit_read(bh
);
523 if (buffer_verified(bh
) && !(flags
& EXT4_EX_FORCE_CACHE
))
525 err
= __ext4_ext_check(function
, line
, inode
,
526 ext_block_hdr(bh
), depth
, pblk
);
529 set_buffer_verified(bh
);
531 * If this is a leaf block, cache all of its entries
533 if (!(flags
& EXT4_EX_NOCACHE
) && depth
== 0) {
534 struct ext4_extent_header
*eh
= ext_block_hdr(bh
);
535 struct ext4_extent
*ex
= EXT_FIRST_EXTENT(eh
);
536 ext4_lblk_t prev
= 0;
539 for (i
= le16_to_cpu(eh
->eh_entries
); i
> 0; i
--, ex
++) {
540 unsigned int status
= EXTENT_STATUS_WRITTEN
;
541 ext4_lblk_t lblk
= le32_to_cpu(ex
->ee_block
);
542 int len
= ext4_ext_get_actual_len(ex
);
544 if (prev
&& (prev
!= lblk
))
545 ext4_es_cache_extent(inode
, prev
,
549 if (ext4_ext_is_unwritten(ex
))
550 status
= EXTENT_STATUS_UNWRITTEN
;
551 ext4_es_cache_extent(inode
, lblk
, len
,
552 ext4_ext_pblock(ex
), status
);
563 #define read_extent_tree_block(inode, pblk, depth, flags) \
564 __read_extent_tree_block(__func__, __LINE__, (inode), (pblk), \
568 * This function is called to cache a file's extent information in the
571 int ext4_ext_precache(struct inode
*inode
)
573 struct ext4_inode_info
*ei
= EXT4_I(inode
);
574 struct ext4_ext_path
*path
= NULL
;
575 struct buffer_head
*bh
;
576 int i
= 0, depth
, ret
= 0;
578 if (!ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
))
579 return 0; /* not an extent-mapped inode */
581 down_read(&ei
->i_data_sem
);
582 depth
= ext_depth(inode
);
584 path
= kzalloc(sizeof(struct ext4_ext_path
) * (depth
+ 1),
587 up_read(&ei
->i_data_sem
);
591 /* Don't cache anything if there are no external extent blocks */
594 path
[0].p_hdr
= ext_inode_hdr(inode
);
595 ret
= ext4_ext_check(inode
, path
[0].p_hdr
, depth
, 0);
598 path
[0].p_idx
= EXT_FIRST_INDEX(path
[0].p_hdr
);
601 * If this is a leaf block or we've reached the end of
602 * the index block, go up
605 path
[i
].p_idx
> EXT_LAST_INDEX(path
[i
].p_hdr
)) {
606 brelse(path
[i
].p_bh
);
611 bh
= read_extent_tree_block(inode
,
612 ext4_idx_pblock(path
[i
].p_idx
++),
614 EXT4_EX_FORCE_CACHE
);
621 path
[i
].p_hdr
= ext_block_hdr(bh
);
622 path
[i
].p_idx
= EXT_FIRST_INDEX(path
[i
].p_hdr
);
624 ext4_set_inode_state(inode
, EXT4_STATE_EXT_PRECACHED
);
626 up_read(&ei
->i_data_sem
);
627 ext4_ext_drop_refs(path
);
633 static void ext4_ext_show_path(struct inode
*inode
, struct ext4_ext_path
*path
)
635 int k
, l
= path
->p_depth
;
638 for (k
= 0; k
<= l
; k
++, path
++) {
640 ext_debug(" %d->%llu", le32_to_cpu(path
->p_idx
->ei_block
),
641 ext4_idx_pblock(path
->p_idx
));
642 } else if (path
->p_ext
) {
643 ext_debug(" %d:[%d]%d:%llu ",
644 le32_to_cpu(path
->p_ext
->ee_block
),
645 ext4_ext_is_unwritten(path
->p_ext
),
646 ext4_ext_get_actual_len(path
->p_ext
),
647 ext4_ext_pblock(path
->p_ext
));
654 static void ext4_ext_show_leaf(struct inode
*inode
, struct ext4_ext_path
*path
)
656 int depth
= ext_depth(inode
);
657 struct ext4_extent_header
*eh
;
658 struct ext4_extent
*ex
;
664 eh
= path
[depth
].p_hdr
;
665 ex
= EXT_FIRST_EXTENT(eh
);
667 ext_debug("Displaying leaf extents for inode %lu\n", inode
->i_ino
);
669 for (i
= 0; i
< le16_to_cpu(eh
->eh_entries
); i
++, ex
++) {
670 ext_debug("%d:[%d]%d:%llu ", le32_to_cpu(ex
->ee_block
),
671 ext4_ext_is_unwritten(ex
),
672 ext4_ext_get_actual_len(ex
), ext4_ext_pblock(ex
));
677 static void ext4_ext_show_move(struct inode
*inode
, struct ext4_ext_path
*path
,
678 ext4_fsblk_t newblock
, int level
)
680 int depth
= ext_depth(inode
);
681 struct ext4_extent
*ex
;
683 if (depth
!= level
) {
684 struct ext4_extent_idx
*idx
;
685 idx
= path
[level
].p_idx
;
686 while (idx
<= EXT_MAX_INDEX(path
[level
].p_hdr
)) {
687 ext_debug("%d: move %d:%llu in new index %llu\n", level
,
688 le32_to_cpu(idx
->ei_block
),
689 ext4_idx_pblock(idx
),
697 ex
= path
[depth
].p_ext
;
698 while (ex
<= EXT_MAX_EXTENT(path
[depth
].p_hdr
)) {
699 ext_debug("move %d:%llu:[%d]%d in new leaf %llu\n",
700 le32_to_cpu(ex
->ee_block
),
702 ext4_ext_is_unwritten(ex
),
703 ext4_ext_get_actual_len(ex
),
710 #define ext4_ext_show_path(inode, path)
711 #define ext4_ext_show_leaf(inode, path)
712 #define ext4_ext_show_move(inode, path, newblock, level)
715 void ext4_ext_drop_refs(struct ext4_ext_path
*path
)
721 depth
= path
->p_depth
;
722 for (i
= 0; i
<= depth
; i
++, path
++)
730 * ext4_ext_binsearch_idx:
731 * binary search for the closest index of the given block
732 * the header must be checked before calling this
735 ext4_ext_binsearch_idx(struct inode
*inode
,
736 struct ext4_ext_path
*path
, ext4_lblk_t block
)
738 struct ext4_extent_header
*eh
= path
->p_hdr
;
739 struct ext4_extent_idx
*r
, *l
, *m
;
742 ext_debug("binsearch for %u(idx): ", block
);
744 l
= EXT_FIRST_INDEX(eh
) + 1;
745 r
= EXT_LAST_INDEX(eh
);
748 if (block
< le32_to_cpu(m
->ei_block
))
752 ext_debug("%p(%u):%p(%u):%p(%u) ", l
, le32_to_cpu(l
->ei_block
),
753 m
, le32_to_cpu(m
->ei_block
),
754 r
, le32_to_cpu(r
->ei_block
));
758 ext_debug(" -> %u->%lld ", le32_to_cpu(path
->p_idx
->ei_block
),
759 ext4_idx_pblock(path
->p_idx
));
761 #ifdef CHECK_BINSEARCH
763 struct ext4_extent_idx
*chix
, *ix
;
766 chix
= ix
= EXT_FIRST_INDEX(eh
);
767 for (k
= 0; k
< le16_to_cpu(eh
->eh_entries
); k
++, ix
++) {
769 le32_to_cpu(ix
->ei_block
) <= le32_to_cpu(ix
[-1].ei_block
)) {
770 printk(KERN_DEBUG
"k=%d, ix=0x%p, "
772 ix
, EXT_FIRST_INDEX(eh
));
773 printk(KERN_DEBUG
"%u <= %u\n",
774 le32_to_cpu(ix
->ei_block
),
775 le32_to_cpu(ix
[-1].ei_block
));
777 BUG_ON(k
&& le32_to_cpu(ix
->ei_block
)
778 <= le32_to_cpu(ix
[-1].ei_block
));
779 if (block
< le32_to_cpu(ix
->ei_block
))
783 BUG_ON(chix
!= path
->p_idx
);
790 * ext4_ext_binsearch:
791 * binary search for closest extent of the given block
792 * the header must be checked before calling this
795 ext4_ext_binsearch(struct inode
*inode
,
796 struct ext4_ext_path
*path
, ext4_lblk_t block
)
798 struct ext4_extent_header
*eh
= path
->p_hdr
;
799 struct ext4_extent
*r
, *l
, *m
;
801 if (eh
->eh_entries
== 0) {
803 * this leaf is empty:
804 * we get such a leaf in split/add case
809 ext_debug("binsearch for %u: ", block
);
811 l
= EXT_FIRST_EXTENT(eh
) + 1;
812 r
= EXT_LAST_EXTENT(eh
);
816 if (block
< le32_to_cpu(m
->ee_block
))
820 ext_debug("%p(%u):%p(%u):%p(%u) ", l
, le32_to_cpu(l
->ee_block
),
821 m
, le32_to_cpu(m
->ee_block
),
822 r
, le32_to_cpu(r
->ee_block
));
826 ext_debug(" -> %d:%llu:[%d]%d ",
827 le32_to_cpu(path
->p_ext
->ee_block
),
828 ext4_ext_pblock(path
->p_ext
),
829 ext4_ext_is_unwritten(path
->p_ext
),
830 ext4_ext_get_actual_len(path
->p_ext
));
832 #ifdef CHECK_BINSEARCH
834 struct ext4_extent
*chex
, *ex
;
837 chex
= ex
= EXT_FIRST_EXTENT(eh
);
838 for (k
= 0; k
< le16_to_cpu(eh
->eh_entries
); k
++, ex
++) {
839 BUG_ON(k
&& le32_to_cpu(ex
->ee_block
)
840 <= le32_to_cpu(ex
[-1].ee_block
));
841 if (block
< le32_to_cpu(ex
->ee_block
))
845 BUG_ON(chex
!= path
->p_ext
);
851 int ext4_ext_tree_init(handle_t
*handle
, struct inode
*inode
)
853 struct ext4_extent_header
*eh
;
855 eh
= ext_inode_hdr(inode
);
858 eh
->eh_magic
= EXT4_EXT_MAGIC
;
859 eh
->eh_max
= cpu_to_le16(ext4_ext_space_root(inode
, 0));
860 ext4_mark_inode_dirty(handle
, inode
);
864 struct ext4_ext_path
*
865 ext4_find_extent(struct inode
*inode
, ext4_lblk_t block
,
866 struct ext4_ext_path
**orig_path
, int flags
)
868 struct ext4_extent_header
*eh
;
869 struct buffer_head
*bh
;
870 struct ext4_ext_path
*path
= orig_path
? *orig_path
: NULL
;
871 short int depth
, i
, ppos
= 0;
874 eh
= ext_inode_hdr(inode
);
875 depth
= ext_depth(inode
);
878 ext4_ext_drop_refs(path
);
879 if (depth
> path
[0].p_maxdepth
) {
881 *orig_path
= path
= NULL
;
885 /* account possible depth increase */
886 path
= kzalloc(sizeof(struct ext4_ext_path
) * (depth
+ 2),
889 return ERR_PTR(-ENOMEM
);
890 path
[0].p_maxdepth
= depth
+ 1;
896 /* walk through the tree */
898 ext_debug("depth %d: num %d, max %d\n",
899 ppos
, le16_to_cpu(eh
->eh_entries
), le16_to_cpu(eh
->eh_max
));
901 ext4_ext_binsearch_idx(inode
, path
+ ppos
, block
);
902 path
[ppos
].p_block
= ext4_idx_pblock(path
[ppos
].p_idx
);
903 path
[ppos
].p_depth
= i
;
904 path
[ppos
].p_ext
= NULL
;
906 bh
= read_extent_tree_block(inode
, path
[ppos
].p_block
, --i
,
913 eh
= ext_block_hdr(bh
);
915 path
[ppos
].p_bh
= bh
;
916 path
[ppos
].p_hdr
= eh
;
919 path
[ppos
].p_depth
= i
;
920 path
[ppos
].p_ext
= NULL
;
921 path
[ppos
].p_idx
= NULL
;
924 ext4_ext_binsearch(inode
, path
+ ppos
, block
);
925 /* if not an empty leaf */
926 if (path
[ppos
].p_ext
)
927 path
[ppos
].p_block
= ext4_ext_pblock(path
[ppos
].p_ext
);
929 ext4_ext_show_path(inode
, path
);
934 ext4_ext_drop_refs(path
);
942 * ext4_ext_insert_index:
943 * insert new index [@logical;@ptr] into the block at @curp;
944 * check where to insert: before @curp or after @curp
946 static int ext4_ext_insert_index(handle_t
*handle
, struct inode
*inode
,
947 struct ext4_ext_path
*curp
,
948 int logical
, ext4_fsblk_t ptr
)
950 struct ext4_extent_idx
*ix
;
953 err
= ext4_ext_get_access(handle
, inode
, curp
);
957 if (unlikely(logical
== le32_to_cpu(curp
->p_idx
->ei_block
))) {
958 EXT4_ERROR_INODE(inode
,
959 "logical %d == ei_block %d!",
960 logical
, le32_to_cpu(curp
->p_idx
->ei_block
));
961 return -EFSCORRUPTED
;
964 if (unlikely(le16_to_cpu(curp
->p_hdr
->eh_entries
)
965 >= le16_to_cpu(curp
->p_hdr
->eh_max
))) {
966 EXT4_ERROR_INODE(inode
,
967 "eh_entries %d >= eh_max %d!",
968 le16_to_cpu(curp
->p_hdr
->eh_entries
),
969 le16_to_cpu(curp
->p_hdr
->eh_max
));
970 return -EFSCORRUPTED
;
973 if (logical
> le32_to_cpu(curp
->p_idx
->ei_block
)) {
975 ext_debug("insert new index %d after: %llu\n", logical
, ptr
);
976 ix
= curp
->p_idx
+ 1;
979 ext_debug("insert new index %d before: %llu\n", logical
, ptr
);
983 len
= EXT_LAST_INDEX(curp
->p_hdr
) - ix
+ 1;
986 ext_debug("insert new index %d: "
987 "move %d indices from 0x%p to 0x%p\n",
988 logical
, len
, ix
, ix
+ 1);
989 memmove(ix
+ 1, ix
, len
* sizeof(struct ext4_extent_idx
));
992 if (unlikely(ix
> EXT_MAX_INDEX(curp
->p_hdr
))) {
993 EXT4_ERROR_INODE(inode
, "ix > EXT_MAX_INDEX!");
994 return -EFSCORRUPTED
;
997 ix
->ei_block
= cpu_to_le32(logical
);
998 ext4_idx_store_pblock(ix
, ptr
);
999 le16_add_cpu(&curp
->p_hdr
->eh_entries
, 1);
1001 if (unlikely(ix
> EXT_LAST_INDEX(curp
->p_hdr
))) {
1002 EXT4_ERROR_INODE(inode
, "ix > EXT_LAST_INDEX!");
1003 return -EFSCORRUPTED
;
1006 err
= ext4_ext_dirty(handle
, inode
, curp
);
1007 ext4_std_error(inode
->i_sb
, err
);
1014 * inserts new subtree into the path, using free index entry
1016 * - allocates all needed blocks (new leaf and all intermediate index blocks)
1017 * - makes decision where to split
1018 * - moves remaining extents and index entries (right to the split point)
1019 * into the newly allocated blocks
1020 * - initializes subtree
1022 static int ext4_ext_split(handle_t
*handle
, struct inode
*inode
,
1024 struct ext4_ext_path
*path
,
1025 struct ext4_extent
*newext
, int at
)
1027 struct buffer_head
*bh
= NULL
;
1028 int depth
= ext_depth(inode
);
1029 struct ext4_extent_header
*neh
;
1030 struct ext4_extent_idx
*fidx
;
1031 int i
= at
, k
, m
, a
;
1032 ext4_fsblk_t newblock
, oldblock
;
1034 ext4_fsblk_t
*ablocks
= NULL
; /* array of allocated blocks */
1037 /* make decision: where to split? */
1038 /* FIXME: now decision is simplest: at current extent */
1040 /* if current leaf will be split, then we should use
1041 * border from split point */
1042 if (unlikely(path
[depth
].p_ext
> EXT_MAX_EXTENT(path
[depth
].p_hdr
))) {
1043 EXT4_ERROR_INODE(inode
, "p_ext > EXT_MAX_EXTENT!");
1044 return -EFSCORRUPTED
;
1046 if (path
[depth
].p_ext
!= EXT_MAX_EXTENT(path
[depth
].p_hdr
)) {
1047 border
= path
[depth
].p_ext
[1].ee_block
;
1048 ext_debug("leaf will be split."
1049 " next leaf starts at %d\n",
1050 le32_to_cpu(border
));
1052 border
= newext
->ee_block
;
1053 ext_debug("leaf will be added."
1054 " next leaf starts at %d\n",
1055 le32_to_cpu(border
));
1059 * If error occurs, then we break processing
1060 * and mark filesystem read-only. index won't
1061 * be inserted and tree will be in consistent
1062 * state. Next mount will repair buffers too.
1066 * Get array to track all allocated blocks.
1067 * We need this to handle errors and free blocks
1070 ablocks
= kzalloc(sizeof(ext4_fsblk_t
) * depth
, GFP_NOFS
);
1074 /* allocate all needed blocks */
1075 ext_debug("allocate %d blocks for indexes/leaf\n", depth
- at
);
1076 for (a
= 0; a
< depth
- at
; a
++) {
1077 newblock
= ext4_ext_new_meta_block(handle
, inode
, path
,
1078 newext
, &err
, flags
);
1081 ablocks
[a
] = newblock
;
1084 /* initialize new leaf */
1085 newblock
= ablocks
[--a
];
1086 if (unlikely(newblock
== 0)) {
1087 EXT4_ERROR_INODE(inode
, "newblock == 0!");
1088 err
= -EFSCORRUPTED
;
1091 bh
= sb_getblk_gfp(inode
->i_sb
, newblock
, __GFP_MOVABLE
| GFP_NOFS
);
1092 if (unlikely(!bh
)) {
1098 err
= ext4_journal_get_create_access(handle
, bh
);
1102 neh
= ext_block_hdr(bh
);
1103 neh
->eh_entries
= 0;
1104 neh
->eh_max
= cpu_to_le16(ext4_ext_space_block(inode
, 0));
1105 neh
->eh_magic
= EXT4_EXT_MAGIC
;
1108 /* move remainder of path[depth] to the new leaf */
1109 if (unlikely(path
[depth
].p_hdr
->eh_entries
!=
1110 path
[depth
].p_hdr
->eh_max
)) {
1111 EXT4_ERROR_INODE(inode
, "eh_entries %d != eh_max %d!",
1112 path
[depth
].p_hdr
->eh_entries
,
1113 path
[depth
].p_hdr
->eh_max
);
1114 err
= -EFSCORRUPTED
;
1117 /* start copy from next extent */
1118 m
= EXT_MAX_EXTENT(path
[depth
].p_hdr
) - path
[depth
].p_ext
++;
1119 ext4_ext_show_move(inode
, path
, newblock
, depth
);
1121 struct ext4_extent
*ex
;
1122 ex
= EXT_FIRST_EXTENT(neh
);
1123 memmove(ex
, path
[depth
].p_ext
, sizeof(struct ext4_extent
) * m
);
1124 le16_add_cpu(&neh
->eh_entries
, m
);
1127 ext4_extent_block_csum_set(inode
, neh
);
1128 set_buffer_uptodate(bh
);
1131 err
= ext4_handle_dirty_metadata(handle
, inode
, bh
);
1137 /* correct old leaf */
1139 err
= ext4_ext_get_access(handle
, inode
, path
+ depth
);
1142 le16_add_cpu(&path
[depth
].p_hdr
->eh_entries
, -m
);
1143 err
= ext4_ext_dirty(handle
, inode
, path
+ depth
);
1149 /* create intermediate indexes */
1151 if (unlikely(k
< 0)) {
1152 EXT4_ERROR_INODE(inode
, "k %d < 0!", k
);
1153 err
= -EFSCORRUPTED
;
1157 ext_debug("create %d intermediate indices\n", k
);
1158 /* insert new index into current index block */
1159 /* current depth stored in i var */
1162 oldblock
= newblock
;
1163 newblock
= ablocks
[--a
];
1164 bh
= sb_getblk(inode
->i_sb
, newblock
);
1165 if (unlikely(!bh
)) {
1171 err
= ext4_journal_get_create_access(handle
, bh
);
1175 neh
= ext_block_hdr(bh
);
1176 neh
->eh_entries
= cpu_to_le16(1);
1177 neh
->eh_magic
= EXT4_EXT_MAGIC
;
1178 neh
->eh_max
= cpu_to_le16(ext4_ext_space_block_idx(inode
, 0));
1179 neh
->eh_depth
= cpu_to_le16(depth
- i
);
1180 fidx
= EXT_FIRST_INDEX(neh
);
1181 fidx
->ei_block
= border
;
1182 ext4_idx_store_pblock(fidx
, oldblock
);
1184 ext_debug("int.index at %d (block %llu): %u -> %llu\n",
1185 i
, newblock
, le32_to_cpu(border
), oldblock
);
1187 /* move remainder of path[i] to the new index block */
1188 if (unlikely(EXT_MAX_INDEX(path
[i
].p_hdr
) !=
1189 EXT_LAST_INDEX(path
[i
].p_hdr
))) {
1190 EXT4_ERROR_INODE(inode
,
1191 "EXT_MAX_INDEX != EXT_LAST_INDEX ee_block %d!",
1192 le32_to_cpu(path
[i
].p_ext
->ee_block
));
1193 err
= -EFSCORRUPTED
;
1196 /* start copy indexes */
1197 m
= EXT_MAX_INDEX(path
[i
].p_hdr
) - path
[i
].p_idx
++;
1198 ext_debug("cur 0x%p, last 0x%p\n", path
[i
].p_idx
,
1199 EXT_MAX_INDEX(path
[i
].p_hdr
));
1200 ext4_ext_show_move(inode
, path
, newblock
, i
);
1202 memmove(++fidx
, path
[i
].p_idx
,
1203 sizeof(struct ext4_extent_idx
) * m
);
1204 le16_add_cpu(&neh
->eh_entries
, m
);
1206 ext4_extent_block_csum_set(inode
, neh
);
1207 set_buffer_uptodate(bh
);
1210 err
= ext4_handle_dirty_metadata(handle
, inode
, bh
);
1216 /* correct old index */
1218 err
= ext4_ext_get_access(handle
, inode
, path
+ i
);
1221 le16_add_cpu(&path
[i
].p_hdr
->eh_entries
, -m
);
1222 err
= ext4_ext_dirty(handle
, inode
, path
+ i
);
1230 /* insert new index */
1231 err
= ext4_ext_insert_index(handle
, inode
, path
+ at
,
1232 le32_to_cpu(border
), newblock
);
1236 if (buffer_locked(bh
))
1242 /* free all allocated blocks in error case */
1243 for (i
= 0; i
< depth
; i
++) {
1246 ext4_free_blocks(handle
, inode
, NULL
, ablocks
[i
], 1,
1247 EXT4_FREE_BLOCKS_METADATA
);
1256 * ext4_ext_grow_indepth:
1257 * implements tree growing procedure:
1258 * - allocates new block
1259 * - moves top-level data (index block or leaf) into the new block
1260 * - initializes new top-level, creating index that points to the
1261 * just created block
1263 static int ext4_ext_grow_indepth(handle_t
*handle
, struct inode
*inode
,
1266 struct ext4_extent_header
*neh
;
1267 struct buffer_head
*bh
;
1268 ext4_fsblk_t newblock
, goal
= 0;
1269 struct ext4_super_block
*es
= EXT4_SB(inode
->i_sb
)->s_es
;
1272 /* Try to prepend new index to old one */
1273 if (ext_depth(inode
))
1274 goal
= ext4_idx_pblock(EXT_FIRST_INDEX(ext_inode_hdr(inode
)));
1275 if (goal
> le32_to_cpu(es
->s_first_data_block
)) {
1276 flags
|= EXT4_MB_HINT_TRY_GOAL
;
1279 goal
= ext4_inode_to_goal_block(inode
);
1280 newblock
= ext4_new_meta_blocks(handle
, inode
, goal
, flags
,
1285 bh
= sb_getblk_gfp(inode
->i_sb
, newblock
, __GFP_MOVABLE
| GFP_NOFS
);
1290 err
= ext4_journal_get_create_access(handle
, bh
);
1296 /* move top-level index/leaf into new block */
1297 memmove(bh
->b_data
, EXT4_I(inode
)->i_data
,
1298 sizeof(EXT4_I(inode
)->i_data
));
1300 /* set size of new block */
1301 neh
= ext_block_hdr(bh
);
1302 /* old root could have indexes or leaves
1303 * so calculate e_max right way */
1304 if (ext_depth(inode
))
1305 neh
->eh_max
= cpu_to_le16(ext4_ext_space_block_idx(inode
, 0));
1307 neh
->eh_max
= cpu_to_le16(ext4_ext_space_block(inode
, 0));
1308 neh
->eh_magic
= EXT4_EXT_MAGIC
;
1309 ext4_extent_block_csum_set(inode
, neh
);
1310 set_buffer_uptodate(bh
);
1313 err
= ext4_handle_dirty_metadata(handle
, inode
, bh
);
1317 /* Update top-level index: num,max,pointer */
1318 neh
= ext_inode_hdr(inode
);
1319 neh
->eh_entries
= cpu_to_le16(1);
1320 ext4_idx_store_pblock(EXT_FIRST_INDEX(neh
), newblock
);
1321 if (neh
->eh_depth
== 0) {
1322 /* Root extent block becomes index block */
1323 neh
->eh_max
= cpu_to_le16(ext4_ext_space_root_idx(inode
, 0));
1324 EXT_FIRST_INDEX(neh
)->ei_block
=
1325 EXT_FIRST_EXTENT(neh
)->ee_block
;
1327 ext_debug("new root: num %d(%d), lblock %d, ptr %llu\n",
1328 le16_to_cpu(neh
->eh_entries
), le16_to_cpu(neh
->eh_max
),
1329 le32_to_cpu(EXT_FIRST_INDEX(neh
)->ei_block
),
1330 ext4_idx_pblock(EXT_FIRST_INDEX(neh
)));
1332 le16_add_cpu(&neh
->eh_depth
, 1);
1333 ext4_mark_inode_dirty(handle
, inode
);
1341 * ext4_ext_create_new_leaf:
1342 * finds empty index and adds new leaf.
1343 * if no free index is found, then it requests in-depth growing.
1345 static int ext4_ext_create_new_leaf(handle_t
*handle
, struct inode
*inode
,
1346 unsigned int mb_flags
,
1347 unsigned int gb_flags
,
1348 struct ext4_ext_path
**ppath
,
1349 struct ext4_extent
*newext
)
1351 struct ext4_ext_path
*path
= *ppath
;
1352 struct ext4_ext_path
*curp
;
1353 int depth
, i
, err
= 0;
1356 i
= depth
= ext_depth(inode
);
1358 /* walk up to the tree and look for free index entry */
1359 curp
= path
+ depth
;
1360 while (i
> 0 && !EXT_HAS_FREE_INDEX(curp
)) {
1365 /* we use already allocated block for index block,
1366 * so subsequent data blocks should be contiguous */
1367 if (EXT_HAS_FREE_INDEX(curp
)) {
1368 /* if we found index with free entry, then use that
1369 * entry: create all needed subtree and add new leaf */
1370 err
= ext4_ext_split(handle
, inode
, mb_flags
, path
, newext
, i
);
1375 path
= ext4_find_extent(inode
,
1376 (ext4_lblk_t
)le32_to_cpu(newext
->ee_block
),
1379 err
= PTR_ERR(path
);
1381 /* tree is full, time to grow in depth */
1382 err
= ext4_ext_grow_indepth(handle
, inode
, mb_flags
);
1387 path
= ext4_find_extent(inode
,
1388 (ext4_lblk_t
)le32_to_cpu(newext
->ee_block
),
1391 err
= PTR_ERR(path
);
1396 * only first (depth 0 -> 1) produces free space;
1397 * in all other cases we have to split the grown tree
1399 depth
= ext_depth(inode
);
1400 if (path
[depth
].p_hdr
->eh_entries
== path
[depth
].p_hdr
->eh_max
) {
1401 /* now we need to split */
1411 * search the closest allocated block to the left for *logical
1412 * and returns it at @logical + it's physical address at @phys
1413 * if *logical is the smallest allocated block, the function
1414 * returns 0 at @phys
1415 * return value contains 0 (success) or error code
1417 static int ext4_ext_search_left(struct inode
*inode
,
1418 struct ext4_ext_path
*path
,
1419 ext4_lblk_t
*logical
, ext4_fsblk_t
*phys
)
1421 struct ext4_extent_idx
*ix
;
1422 struct ext4_extent
*ex
;
1425 if (unlikely(path
== NULL
)) {
1426 EXT4_ERROR_INODE(inode
, "path == NULL *logical %d!", *logical
);
1427 return -EFSCORRUPTED
;
1429 depth
= path
->p_depth
;
1432 if (depth
== 0 && path
->p_ext
== NULL
)
1435 /* usually extent in the path covers blocks smaller
1436 * then *logical, but it can be that extent is the
1437 * first one in the file */
1439 ex
= path
[depth
].p_ext
;
1440 ee_len
= ext4_ext_get_actual_len(ex
);
1441 if (*logical
< le32_to_cpu(ex
->ee_block
)) {
1442 if (unlikely(EXT_FIRST_EXTENT(path
[depth
].p_hdr
) != ex
)) {
1443 EXT4_ERROR_INODE(inode
,
1444 "EXT_FIRST_EXTENT != ex *logical %d ee_block %d!",
1445 *logical
, le32_to_cpu(ex
->ee_block
));
1446 return -EFSCORRUPTED
;
1448 while (--depth
>= 0) {
1449 ix
= path
[depth
].p_idx
;
1450 if (unlikely(ix
!= EXT_FIRST_INDEX(path
[depth
].p_hdr
))) {
1451 EXT4_ERROR_INODE(inode
,
1452 "ix (%d) != EXT_FIRST_INDEX (%d) (depth %d)!",
1453 ix
!= NULL
? le32_to_cpu(ix
->ei_block
) : 0,
1454 EXT_FIRST_INDEX(path
[depth
].p_hdr
) != NULL
?
1455 le32_to_cpu(EXT_FIRST_INDEX(path
[depth
].p_hdr
)->ei_block
) : 0,
1457 return -EFSCORRUPTED
;
1463 if (unlikely(*logical
< (le32_to_cpu(ex
->ee_block
) + ee_len
))) {
1464 EXT4_ERROR_INODE(inode
,
1465 "logical %d < ee_block %d + ee_len %d!",
1466 *logical
, le32_to_cpu(ex
->ee_block
), ee_len
);
1467 return -EFSCORRUPTED
;
1470 *logical
= le32_to_cpu(ex
->ee_block
) + ee_len
- 1;
1471 *phys
= ext4_ext_pblock(ex
) + ee_len
- 1;
1476 * search the closest allocated block to the right for *logical
1477 * and returns it at @logical + it's physical address at @phys
1478 * if *logical is the largest allocated block, the function
1479 * returns 0 at @phys
1480 * return value contains 0 (success) or error code
1482 static int ext4_ext_search_right(struct inode
*inode
,
1483 struct ext4_ext_path
*path
,
1484 ext4_lblk_t
*logical
, ext4_fsblk_t
*phys
,
1485 struct ext4_extent
**ret_ex
)
1487 struct buffer_head
*bh
= NULL
;
1488 struct ext4_extent_header
*eh
;
1489 struct ext4_extent_idx
*ix
;
1490 struct ext4_extent
*ex
;
1492 int depth
; /* Note, NOT eh_depth; depth from top of tree */
1495 if (unlikely(path
== NULL
)) {
1496 EXT4_ERROR_INODE(inode
, "path == NULL *logical %d!", *logical
);
1497 return -EFSCORRUPTED
;
1499 depth
= path
->p_depth
;
1502 if (depth
== 0 && path
->p_ext
== NULL
)
1505 /* usually extent in the path covers blocks smaller
1506 * then *logical, but it can be that extent is the
1507 * first one in the file */
1509 ex
= path
[depth
].p_ext
;
1510 ee_len
= ext4_ext_get_actual_len(ex
);
1511 if (*logical
< le32_to_cpu(ex
->ee_block
)) {
1512 if (unlikely(EXT_FIRST_EXTENT(path
[depth
].p_hdr
) != ex
)) {
1513 EXT4_ERROR_INODE(inode
,
1514 "first_extent(path[%d].p_hdr) != ex",
1516 return -EFSCORRUPTED
;
1518 while (--depth
>= 0) {
1519 ix
= path
[depth
].p_idx
;
1520 if (unlikely(ix
!= EXT_FIRST_INDEX(path
[depth
].p_hdr
))) {
1521 EXT4_ERROR_INODE(inode
,
1522 "ix != EXT_FIRST_INDEX *logical %d!",
1524 return -EFSCORRUPTED
;
1530 if (unlikely(*logical
< (le32_to_cpu(ex
->ee_block
) + ee_len
))) {
1531 EXT4_ERROR_INODE(inode
,
1532 "logical %d < ee_block %d + ee_len %d!",
1533 *logical
, le32_to_cpu(ex
->ee_block
), ee_len
);
1534 return -EFSCORRUPTED
;
1537 if (ex
!= EXT_LAST_EXTENT(path
[depth
].p_hdr
)) {
1538 /* next allocated block in this leaf */
1543 /* go up and search for index to the right */
1544 while (--depth
>= 0) {
1545 ix
= path
[depth
].p_idx
;
1546 if (ix
!= EXT_LAST_INDEX(path
[depth
].p_hdr
))
1550 /* we've gone up to the root and found no index to the right */
1554 /* we've found index to the right, let's
1555 * follow it and find the closest allocated
1556 * block to the right */
1558 block
= ext4_idx_pblock(ix
);
1559 while (++depth
< path
->p_depth
) {
1560 /* subtract from p_depth to get proper eh_depth */
1561 bh
= read_extent_tree_block(inode
, block
,
1562 path
->p_depth
- depth
, 0);
1565 eh
= ext_block_hdr(bh
);
1566 ix
= EXT_FIRST_INDEX(eh
);
1567 block
= ext4_idx_pblock(ix
);
1571 bh
= read_extent_tree_block(inode
, block
, path
->p_depth
- depth
, 0);
1574 eh
= ext_block_hdr(bh
);
1575 ex
= EXT_FIRST_EXTENT(eh
);
1577 *logical
= le32_to_cpu(ex
->ee_block
);
1578 *phys
= ext4_ext_pblock(ex
);
1586 * ext4_ext_next_allocated_block:
1587 * returns allocated block in subsequent extent or EXT_MAX_BLOCKS.
1588 * NOTE: it considers block number from index entry as
1589 * allocated block. Thus, index entries have to be consistent
1593 ext4_ext_next_allocated_block(struct ext4_ext_path
*path
)
1597 BUG_ON(path
== NULL
);
1598 depth
= path
->p_depth
;
1600 if (depth
== 0 && path
->p_ext
== NULL
)
1601 return EXT_MAX_BLOCKS
;
1603 while (depth
>= 0) {
1604 if (depth
== path
->p_depth
) {
1606 if (path
[depth
].p_ext
&&
1607 path
[depth
].p_ext
!=
1608 EXT_LAST_EXTENT(path
[depth
].p_hdr
))
1609 return le32_to_cpu(path
[depth
].p_ext
[1].ee_block
);
1612 if (path
[depth
].p_idx
!=
1613 EXT_LAST_INDEX(path
[depth
].p_hdr
))
1614 return le32_to_cpu(path
[depth
].p_idx
[1].ei_block
);
1619 return EXT_MAX_BLOCKS
;
1623 * ext4_ext_next_leaf_block:
1624 * returns first allocated block from next leaf or EXT_MAX_BLOCKS
1626 static ext4_lblk_t
ext4_ext_next_leaf_block(struct ext4_ext_path
*path
)
1630 BUG_ON(path
== NULL
);
1631 depth
= path
->p_depth
;
1633 /* zero-tree has no leaf blocks at all */
1635 return EXT_MAX_BLOCKS
;
1637 /* go to index block */
1640 while (depth
>= 0) {
1641 if (path
[depth
].p_idx
!=
1642 EXT_LAST_INDEX(path
[depth
].p_hdr
))
1643 return (ext4_lblk_t
)
1644 le32_to_cpu(path
[depth
].p_idx
[1].ei_block
);
1648 return EXT_MAX_BLOCKS
;
1652 * ext4_ext_correct_indexes:
1653 * if leaf gets modified and modified extent is first in the leaf,
1654 * then we have to correct all indexes above.
1655 * TODO: do we need to correct tree in all cases?
1657 static int ext4_ext_correct_indexes(handle_t
*handle
, struct inode
*inode
,
1658 struct ext4_ext_path
*path
)
1660 struct ext4_extent_header
*eh
;
1661 int depth
= ext_depth(inode
);
1662 struct ext4_extent
*ex
;
1666 eh
= path
[depth
].p_hdr
;
1667 ex
= path
[depth
].p_ext
;
1669 if (unlikely(ex
== NULL
|| eh
== NULL
)) {
1670 EXT4_ERROR_INODE(inode
,
1671 "ex %p == NULL or eh %p == NULL", ex
, eh
);
1672 return -EFSCORRUPTED
;
1676 /* there is no tree at all */
1680 if (ex
!= EXT_FIRST_EXTENT(eh
)) {
1681 /* we correct tree if first leaf got modified only */
1686 * TODO: we need correction if border is smaller than current one
1689 border
= path
[depth
].p_ext
->ee_block
;
1690 err
= ext4_ext_get_access(handle
, inode
, path
+ k
);
1693 path
[k
].p_idx
->ei_block
= border
;
1694 err
= ext4_ext_dirty(handle
, inode
, path
+ k
);
1699 /* change all left-side indexes */
1700 if (path
[k
+1].p_idx
!= EXT_FIRST_INDEX(path
[k
+1].p_hdr
))
1702 err
= ext4_ext_get_access(handle
, inode
, path
+ k
);
1705 path
[k
].p_idx
->ei_block
= border
;
1706 err
= ext4_ext_dirty(handle
, inode
, path
+ k
);
1715 ext4_can_extents_be_merged(struct inode
*inode
, struct ext4_extent
*ex1
,
1716 struct ext4_extent
*ex2
)
1718 unsigned short ext1_ee_len
, ext2_ee_len
;
1720 if (ext4_ext_is_unwritten(ex1
) != ext4_ext_is_unwritten(ex2
))
1723 ext1_ee_len
= ext4_ext_get_actual_len(ex1
);
1724 ext2_ee_len
= ext4_ext_get_actual_len(ex2
);
1726 if (le32_to_cpu(ex1
->ee_block
) + ext1_ee_len
!=
1727 le32_to_cpu(ex2
->ee_block
))
1731 * To allow future support for preallocated extents to be added
1732 * as an RO_COMPAT feature, refuse to merge to extents if
1733 * this can result in the top bit of ee_len being set.
1735 if (ext1_ee_len
+ ext2_ee_len
> EXT_INIT_MAX_LEN
)
1738 * The check for IO to unwritten extent is somewhat racy as we
1739 * increment i_unwritten / set EXT4_STATE_DIO_UNWRITTEN only after
1740 * dropping i_data_sem. But reserved blocks should save us in that
1743 if (ext4_ext_is_unwritten(ex1
) &&
1744 (ext4_test_inode_state(inode
, EXT4_STATE_DIO_UNWRITTEN
) ||
1745 atomic_read(&EXT4_I(inode
)->i_unwritten
) ||
1746 (ext1_ee_len
+ ext2_ee_len
> EXT_UNWRITTEN_MAX_LEN
)))
1748 #ifdef AGGRESSIVE_TEST
1749 if (ext1_ee_len
>= 4)
1753 if (ext4_ext_pblock(ex1
) + ext1_ee_len
== ext4_ext_pblock(ex2
))
1759 * This function tries to merge the "ex" extent to the next extent in the tree.
1760 * It always tries to merge towards right. If you want to merge towards
1761 * left, pass "ex - 1" as argument instead of "ex".
1762 * Returns 0 if the extents (ex and ex+1) were _not_ merged and returns
1763 * 1 if they got merged.
1765 static int ext4_ext_try_to_merge_right(struct inode
*inode
,
1766 struct ext4_ext_path
*path
,
1767 struct ext4_extent
*ex
)
1769 struct ext4_extent_header
*eh
;
1770 unsigned int depth
, len
;
1771 int merge_done
= 0, unwritten
;
1773 depth
= ext_depth(inode
);
1774 BUG_ON(path
[depth
].p_hdr
== NULL
);
1775 eh
= path
[depth
].p_hdr
;
1777 while (ex
< EXT_LAST_EXTENT(eh
)) {
1778 if (!ext4_can_extents_be_merged(inode
, ex
, ex
+ 1))
1780 /* merge with next extent! */
1781 unwritten
= ext4_ext_is_unwritten(ex
);
1782 ex
->ee_len
= cpu_to_le16(ext4_ext_get_actual_len(ex
)
1783 + ext4_ext_get_actual_len(ex
+ 1));
1785 ext4_ext_mark_unwritten(ex
);
1787 if (ex
+ 1 < EXT_LAST_EXTENT(eh
)) {
1788 len
= (EXT_LAST_EXTENT(eh
) - ex
- 1)
1789 * sizeof(struct ext4_extent
);
1790 memmove(ex
+ 1, ex
+ 2, len
);
1792 le16_add_cpu(&eh
->eh_entries
, -1);
1794 WARN_ON(eh
->eh_entries
== 0);
1795 if (!eh
->eh_entries
)
1796 EXT4_ERROR_INODE(inode
, "eh->eh_entries = 0!");
1803 * This function does a very simple check to see if we can collapse
1804 * an extent tree with a single extent tree leaf block into the inode.
1806 static void ext4_ext_try_to_merge_up(handle_t
*handle
,
1807 struct inode
*inode
,
1808 struct ext4_ext_path
*path
)
1811 unsigned max_root
= ext4_ext_space_root(inode
, 0);
1814 if ((path
[0].p_depth
!= 1) ||
1815 (le16_to_cpu(path
[0].p_hdr
->eh_entries
) != 1) ||
1816 (le16_to_cpu(path
[1].p_hdr
->eh_entries
) > max_root
))
1820 * We need to modify the block allocation bitmap and the block
1821 * group descriptor to release the extent tree block. If we
1822 * can't get the journal credits, give up.
1824 if (ext4_journal_extend(handle
, 2))
1828 * Copy the extent data up to the inode
1830 blk
= ext4_idx_pblock(path
[0].p_idx
);
1831 s
= le16_to_cpu(path
[1].p_hdr
->eh_entries
) *
1832 sizeof(struct ext4_extent_idx
);
1833 s
+= sizeof(struct ext4_extent_header
);
1835 path
[1].p_maxdepth
= path
[0].p_maxdepth
;
1836 memcpy(path
[0].p_hdr
, path
[1].p_hdr
, s
);
1837 path
[0].p_depth
= 0;
1838 path
[0].p_ext
= EXT_FIRST_EXTENT(path
[0].p_hdr
) +
1839 (path
[1].p_ext
- EXT_FIRST_EXTENT(path
[1].p_hdr
));
1840 path
[0].p_hdr
->eh_max
= cpu_to_le16(max_root
);
1842 brelse(path
[1].p_bh
);
1843 ext4_free_blocks(handle
, inode
, NULL
, blk
, 1,
1844 EXT4_FREE_BLOCKS_METADATA
| EXT4_FREE_BLOCKS_FORGET
);
1848 * This function tries to merge the @ex extent to neighbours in the tree.
1849 * return 1 if merge left else 0.
1851 static void ext4_ext_try_to_merge(handle_t
*handle
,
1852 struct inode
*inode
,
1853 struct ext4_ext_path
*path
,
1854 struct ext4_extent
*ex
) {
1855 struct ext4_extent_header
*eh
;
1859 depth
= ext_depth(inode
);
1860 BUG_ON(path
[depth
].p_hdr
== NULL
);
1861 eh
= path
[depth
].p_hdr
;
1863 if (ex
> EXT_FIRST_EXTENT(eh
))
1864 merge_done
= ext4_ext_try_to_merge_right(inode
, path
, ex
- 1);
1867 (void) ext4_ext_try_to_merge_right(inode
, path
, ex
);
1869 ext4_ext_try_to_merge_up(handle
, inode
, path
);
1873 * check if a portion of the "newext" extent overlaps with an
1876 * If there is an overlap discovered, it updates the length of the newext
1877 * such that there will be no overlap, and then returns 1.
1878 * If there is no overlap found, it returns 0.
1880 static unsigned int ext4_ext_check_overlap(struct ext4_sb_info
*sbi
,
1881 struct inode
*inode
,
1882 struct ext4_extent
*newext
,
1883 struct ext4_ext_path
*path
)
1886 unsigned int depth
, len1
;
1887 unsigned int ret
= 0;
1889 b1
= le32_to_cpu(newext
->ee_block
);
1890 len1
= ext4_ext_get_actual_len(newext
);
1891 depth
= ext_depth(inode
);
1892 if (!path
[depth
].p_ext
)
1894 b2
= EXT4_LBLK_CMASK(sbi
, le32_to_cpu(path
[depth
].p_ext
->ee_block
));
1897 * get the next allocated block if the extent in the path
1898 * is before the requested block(s)
1901 b2
= ext4_ext_next_allocated_block(path
);
1902 if (b2
== EXT_MAX_BLOCKS
)
1904 b2
= EXT4_LBLK_CMASK(sbi
, b2
);
1907 /* check for wrap through zero on extent logical start block*/
1908 if (b1
+ len1
< b1
) {
1909 len1
= EXT_MAX_BLOCKS
- b1
;
1910 newext
->ee_len
= cpu_to_le16(len1
);
1914 /* check for overlap */
1915 if (b1
+ len1
> b2
) {
1916 newext
->ee_len
= cpu_to_le16(b2
- b1
);
1924 * ext4_ext_insert_extent:
1925 * tries to merge requsted extent into the existing extent or
1926 * inserts requested extent as new one into the tree,
1927 * creating new leaf in the no-space case.
1929 int ext4_ext_insert_extent(handle_t
*handle
, struct inode
*inode
,
1930 struct ext4_ext_path
**ppath
,
1931 struct ext4_extent
*newext
, int gb_flags
)
1933 struct ext4_ext_path
*path
= *ppath
;
1934 struct ext4_extent_header
*eh
;
1935 struct ext4_extent
*ex
, *fex
;
1936 struct ext4_extent
*nearex
; /* nearest extent */
1937 struct ext4_ext_path
*npath
= NULL
;
1938 int depth
, len
, err
;
1940 int mb_flags
= 0, unwritten
;
1942 if (gb_flags
& EXT4_GET_BLOCKS_DELALLOC_RESERVE
)
1943 mb_flags
|= EXT4_MB_DELALLOC_RESERVED
;
1944 if (unlikely(ext4_ext_get_actual_len(newext
) == 0)) {
1945 EXT4_ERROR_INODE(inode
, "ext4_ext_get_actual_len(newext) == 0");
1946 return -EFSCORRUPTED
;
1948 depth
= ext_depth(inode
);
1949 ex
= path
[depth
].p_ext
;
1950 eh
= path
[depth
].p_hdr
;
1951 if (unlikely(path
[depth
].p_hdr
== NULL
)) {
1952 EXT4_ERROR_INODE(inode
, "path[%d].p_hdr == NULL", depth
);
1953 return -EFSCORRUPTED
;
1956 /* try to insert block into found extent and return */
1957 if (ex
&& !(gb_flags
& EXT4_GET_BLOCKS_PRE_IO
)) {
1960 * Try to see whether we should rather test the extent on
1961 * right from ex, or from the left of ex. This is because
1962 * ext4_find_extent() can return either extent on the
1963 * left, or on the right from the searched position. This
1964 * will make merging more effective.
1966 if (ex
< EXT_LAST_EXTENT(eh
) &&
1967 (le32_to_cpu(ex
->ee_block
) +
1968 ext4_ext_get_actual_len(ex
) <
1969 le32_to_cpu(newext
->ee_block
))) {
1972 } else if ((ex
> EXT_FIRST_EXTENT(eh
)) &&
1973 (le32_to_cpu(newext
->ee_block
) +
1974 ext4_ext_get_actual_len(newext
) <
1975 le32_to_cpu(ex
->ee_block
)))
1978 /* Try to append newex to the ex */
1979 if (ext4_can_extents_be_merged(inode
, ex
, newext
)) {
1980 ext_debug("append [%d]%d block to %u:[%d]%d"
1982 ext4_ext_is_unwritten(newext
),
1983 ext4_ext_get_actual_len(newext
),
1984 le32_to_cpu(ex
->ee_block
),
1985 ext4_ext_is_unwritten(ex
),
1986 ext4_ext_get_actual_len(ex
),
1987 ext4_ext_pblock(ex
));
1988 err
= ext4_ext_get_access(handle
, inode
,
1992 unwritten
= ext4_ext_is_unwritten(ex
);
1993 ex
->ee_len
= cpu_to_le16(ext4_ext_get_actual_len(ex
)
1994 + ext4_ext_get_actual_len(newext
));
1996 ext4_ext_mark_unwritten(ex
);
1997 eh
= path
[depth
].p_hdr
;
2003 /* Try to prepend newex to the ex */
2004 if (ext4_can_extents_be_merged(inode
, newext
, ex
)) {
2005 ext_debug("prepend %u[%d]%d block to %u:[%d]%d"
2007 le32_to_cpu(newext
->ee_block
),
2008 ext4_ext_is_unwritten(newext
),
2009 ext4_ext_get_actual_len(newext
),
2010 le32_to_cpu(ex
->ee_block
),
2011 ext4_ext_is_unwritten(ex
),
2012 ext4_ext_get_actual_len(ex
),
2013 ext4_ext_pblock(ex
));
2014 err
= ext4_ext_get_access(handle
, inode
,
2019 unwritten
= ext4_ext_is_unwritten(ex
);
2020 ex
->ee_block
= newext
->ee_block
;
2021 ext4_ext_store_pblock(ex
, ext4_ext_pblock(newext
));
2022 ex
->ee_len
= cpu_to_le16(ext4_ext_get_actual_len(ex
)
2023 + ext4_ext_get_actual_len(newext
));
2025 ext4_ext_mark_unwritten(ex
);
2026 eh
= path
[depth
].p_hdr
;
2032 depth
= ext_depth(inode
);
2033 eh
= path
[depth
].p_hdr
;
2034 if (le16_to_cpu(eh
->eh_entries
) < le16_to_cpu(eh
->eh_max
))
2037 /* probably next leaf has space for us? */
2038 fex
= EXT_LAST_EXTENT(eh
);
2039 next
= EXT_MAX_BLOCKS
;
2040 if (le32_to_cpu(newext
->ee_block
) > le32_to_cpu(fex
->ee_block
))
2041 next
= ext4_ext_next_leaf_block(path
);
2042 if (next
!= EXT_MAX_BLOCKS
) {
2043 ext_debug("next leaf block - %u\n", next
);
2044 BUG_ON(npath
!= NULL
);
2045 npath
= ext4_find_extent(inode
, next
, NULL
, 0);
2047 return PTR_ERR(npath
);
2048 BUG_ON(npath
->p_depth
!= path
->p_depth
);
2049 eh
= npath
[depth
].p_hdr
;
2050 if (le16_to_cpu(eh
->eh_entries
) < le16_to_cpu(eh
->eh_max
)) {
2051 ext_debug("next leaf isn't full(%d)\n",
2052 le16_to_cpu(eh
->eh_entries
));
2056 ext_debug("next leaf has no free space(%d,%d)\n",
2057 le16_to_cpu(eh
->eh_entries
), le16_to_cpu(eh
->eh_max
));
2061 * There is no free space in the found leaf.
2062 * We're gonna add a new leaf in the tree.
2064 if (gb_flags
& EXT4_GET_BLOCKS_METADATA_NOFAIL
)
2065 mb_flags
|= EXT4_MB_USE_RESERVED
;
2066 err
= ext4_ext_create_new_leaf(handle
, inode
, mb_flags
, gb_flags
,
2070 depth
= ext_depth(inode
);
2071 eh
= path
[depth
].p_hdr
;
2074 nearex
= path
[depth
].p_ext
;
2076 err
= ext4_ext_get_access(handle
, inode
, path
+ depth
);
2081 /* there is no extent in this leaf, create first one */
2082 ext_debug("first extent in the leaf: %u:%llu:[%d]%d\n",
2083 le32_to_cpu(newext
->ee_block
),
2084 ext4_ext_pblock(newext
),
2085 ext4_ext_is_unwritten(newext
),
2086 ext4_ext_get_actual_len(newext
));
2087 nearex
= EXT_FIRST_EXTENT(eh
);
2089 if (le32_to_cpu(newext
->ee_block
)
2090 > le32_to_cpu(nearex
->ee_block
)) {
2092 ext_debug("insert %u:%llu:[%d]%d before: "
2094 le32_to_cpu(newext
->ee_block
),
2095 ext4_ext_pblock(newext
),
2096 ext4_ext_is_unwritten(newext
),
2097 ext4_ext_get_actual_len(newext
),
2102 BUG_ON(newext
->ee_block
== nearex
->ee_block
);
2103 ext_debug("insert %u:%llu:[%d]%d after: "
2105 le32_to_cpu(newext
->ee_block
),
2106 ext4_ext_pblock(newext
),
2107 ext4_ext_is_unwritten(newext
),
2108 ext4_ext_get_actual_len(newext
),
2111 len
= EXT_LAST_EXTENT(eh
) - nearex
+ 1;
2113 ext_debug("insert %u:%llu:[%d]%d: "
2114 "move %d extents from 0x%p to 0x%p\n",
2115 le32_to_cpu(newext
->ee_block
),
2116 ext4_ext_pblock(newext
),
2117 ext4_ext_is_unwritten(newext
),
2118 ext4_ext_get_actual_len(newext
),
2119 len
, nearex
, nearex
+ 1);
2120 memmove(nearex
+ 1, nearex
,
2121 len
* sizeof(struct ext4_extent
));
2125 le16_add_cpu(&eh
->eh_entries
, 1);
2126 path
[depth
].p_ext
= nearex
;
2127 nearex
->ee_block
= newext
->ee_block
;
2128 ext4_ext_store_pblock(nearex
, ext4_ext_pblock(newext
));
2129 nearex
->ee_len
= newext
->ee_len
;
2132 /* try to merge extents */
2133 if (!(gb_flags
& EXT4_GET_BLOCKS_PRE_IO
))
2134 ext4_ext_try_to_merge(handle
, inode
, path
, nearex
);
2137 /* time to correct all indexes above */
2138 err
= ext4_ext_correct_indexes(handle
, inode
, path
);
2142 err
= ext4_ext_dirty(handle
, inode
, path
+ path
->p_depth
);
2145 ext4_ext_drop_refs(npath
);
2150 static int ext4_fill_fiemap_extents(struct inode
*inode
,
2151 ext4_lblk_t block
, ext4_lblk_t num
,
2152 struct fiemap_extent_info
*fieinfo
)
2154 struct ext4_ext_path
*path
= NULL
;
2155 struct ext4_extent
*ex
;
2156 struct extent_status es
;
2157 ext4_lblk_t next
, next_del
, start
= 0, end
= 0;
2158 ext4_lblk_t last
= block
+ num
;
2159 int exists
, depth
= 0, err
= 0;
2160 unsigned int flags
= 0;
2161 unsigned char blksize_bits
= inode
->i_sb
->s_blocksize_bits
;
2163 while (block
< last
&& block
!= EXT_MAX_BLOCKS
) {
2165 /* find extent for this block */
2166 down_read(&EXT4_I(inode
)->i_data_sem
);
2168 path
= ext4_find_extent(inode
, block
, &path
, 0);
2170 up_read(&EXT4_I(inode
)->i_data_sem
);
2171 err
= PTR_ERR(path
);
2176 depth
= ext_depth(inode
);
2177 if (unlikely(path
[depth
].p_hdr
== NULL
)) {
2178 up_read(&EXT4_I(inode
)->i_data_sem
);
2179 EXT4_ERROR_INODE(inode
, "path[%d].p_hdr == NULL", depth
);
2180 err
= -EFSCORRUPTED
;
2183 ex
= path
[depth
].p_ext
;
2184 next
= ext4_ext_next_allocated_block(path
);
2189 /* there is no extent yet, so try to allocate
2190 * all requested space */
2193 } else if (le32_to_cpu(ex
->ee_block
) > block
) {
2194 /* need to allocate space before found extent */
2196 end
= le32_to_cpu(ex
->ee_block
);
2197 if (block
+ num
< end
)
2199 } else if (block
>= le32_to_cpu(ex
->ee_block
)
2200 + ext4_ext_get_actual_len(ex
)) {
2201 /* need to allocate space after found extent */
2206 } else if (block
>= le32_to_cpu(ex
->ee_block
)) {
2208 * some part of requested space is covered
2212 end
= le32_to_cpu(ex
->ee_block
)
2213 + ext4_ext_get_actual_len(ex
);
2214 if (block
+ num
< end
)
2220 BUG_ON(end
<= start
);
2224 es
.es_len
= end
- start
;
2227 es
.es_lblk
= le32_to_cpu(ex
->ee_block
);
2228 es
.es_len
= ext4_ext_get_actual_len(ex
);
2229 es
.es_pblk
= ext4_ext_pblock(ex
);
2230 if (ext4_ext_is_unwritten(ex
))
2231 flags
|= FIEMAP_EXTENT_UNWRITTEN
;
2235 * Find delayed extent and update es accordingly. We call
2236 * it even in !exists case to find out whether es is the
2237 * last existing extent or not.
2239 next_del
= ext4_find_delayed_extent(inode
, &es
);
2240 if (!exists
&& next_del
) {
2242 flags
|= (FIEMAP_EXTENT_DELALLOC
|
2243 FIEMAP_EXTENT_UNKNOWN
);
2245 up_read(&EXT4_I(inode
)->i_data_sem
);
2247 if (unlikely(es
.es_len
== 0)) {
2248 EXT4_ERROR_INODE(inode
, "es.es_len == 0");
2249 err
= -EFSCORRUPTED
;
2254 * This is possible iff next == next_del == EXT_MAX_BLOCKS.
2255 * we need to check next == EXT_MAX_BLOCKS because it is
2256 * possible that an extent is with unwritten and delayed
2257 * status due to when an extent is delayed allocated and
2258 * is allocated by fallocate status tree will track both of
2261 * So we could return a unwritten and delayed extent, and
2262 * its block is equal to 'next'.
2264 if (next
== next_del
&& next
== EXT_MAX_BLOCKS
) {
2265 flags
|= FIEMAP_EXTENT_LAST
;
2266 if (unlikely(next_del
!= EXT_MAX_BLOCKS
||
2267 next
!= EXT_MAX_BLOCKS
)) {
2268 EXT4_ERROR_INODE(inode
,
2269 "next extent == %u, next "
2270 "delalloc extent = %u",
2272 err
= -EFSCORRUPTED
;
2278 err
= fiemap_fill_next_extent(fieinfo
,
2279 (__u64
)es
.es_lblk
<< blksize_bits
,
2280 (__u64
)es
.es_pblk
<< blksize_bits
,
2281 (__u64
)es
.es_len
<< blksize_bits
,
2291 block
= es
.es_lblk
+ es
.es_len
;
2294 ext4_ext_drop_refs(path
);
2300 * ext4_ext_determine_hole - determine hole around given block
2301 * @inode: inode we lookup in
2302 * @path: path in extent tree to @lblk
2303 * @lblk: pointer to logical block around which we want to determine hole
2305 * Determine hole length (and start if easily possible) around given logical
2306 * block. We don't try too hard to find the beginning of the hole but @path
2307 * actually points to extent before @lblk, we provide it.
2309 * The function returns the length of a hole starting at @lblk. We update @lblk
2310 * to the beginning of the hole if we managed to find it.
2312 static ext4_lblk_t
ext4_ext_determine_hole(struct inode
*inode
,
2313 struct ext4_ext_path
*path
,
2316 int depth
= ext_depth(inode
);
2317 struct ext4_extent
*ex
;
2320 ex
= path
[depth
].p_ext
;
2322 /* there is no extent yet, so gap is [0;-] */
2324 len
= EXT_MAX_BLOCKS
;
2325 } else if (*lblk
< le32_to_cpu(ex
->ee_block
)) {
2326 len
= le32_to_cpu(ex
->ee_block
) - *lblk
;
2327 } else if (*lblk
>= le32_to_cpu(ex
->ee_block
)
2328 + ext4_ext_get_actual_len(ex
)) {
2331 *lblk
= le32_to_cpu(ex
->ee_block
) + ext4_ext_get_actual_len(ex
);
2332 next
= ext4_ext_next_allocated_block(path
);
2333 BUG_ON(next
== *lblk
);
2342 * ext4_ext_put_gap_in_cache:
2343 * calculate boundaries of the gap that the requested block fits into
2344 * and cache this gap
2347 ext4_ext_put_gap_in_cache(struct inode
*inode
, ext4_lblk_t hole_start
,
2348 ext4_lblk_t hole_len
)
2350 struct extent_status es
;
2352 ext4_es_find_delayed_extent_range(inode
, hole_start
,
2353 hole_start
+ hole_len
- 1, &es
);
2355 /* There's delayed extent containing lblock? */
2356 if (es
.es_lblk
<= hole_start
)
2358 hole_len
= min(es
.es_lblk
- hole_start
, hole_len
);
2360 ext_debug(" -> %u:%u\n", hole_start
, hole_len
);
2361 ext4_es_insert_extent(inode
, hole_start
, hole_len
, ~0,
2362 EXTENT_STATUS_HOLE
);
2367 * removes index from the index block.
2369 static int ext4_ext_rm_idx(handle_t
*handle
, struct inode
*inode
,
2370 struct ext4_ext_path
*path
, int depth
)
2375 /* free index block */
2377 path
= path
+ depth
;
2378 leaf
= ext4_idx_pblock(path
->p_idx
);
2379 if (unlikely(path
->p_hdr
->eh_entries
== 0)) {
2380 EXT4_ERROR_INODE(inode
, "path->p_hdr->eh_entries == 0");
2381 return -EFSCORRUPTED
;
2383 err
= ext4_ext_get_access(handle
, inode
, path
);
2387 if (path
->p_idx
!= EXT_LAST_INDEX(path
->p_hdr
)) {
2388 int len
= EXT_LAST_INDEX(path
->p_hdr
) - path
->p_idx
;
2389 len
*= sizeof(struct ext4_extent_idx
);
2390 memmove(path
->p_idx
, path
->p_idx
+ 1, len
);
2393 le16_add_cpu(&path
->p_hdr
->eh_entries
, -1);
2394 err
= ext4_ext_dirty(handle
, inode
, path
);
2397 ext_debug("index is empty, remove it, free block %llu\n", leaf
);
2398 trace_ext4_ext_rm_idx(inode
, leaf
);
2400 ext4_free_blocks(handle
, inode
, NULL
, leaf
, 1,
2401 EXT4_FREE_BLOCKS_METADATA
| EXT4_FREE_BLOCKS_FORGET
);
2403 while (--depth
>= 0) {
2404 if (path
->p_idx
!= EXT_FIRST_INDEX(path
->p_hdr
))
2407 err
= ext4_ext_get_access(handle
, inode
, path
);
2410 path
->p_idx
->ei_block
= (path
+1)->p_idx
->ei_block
;
2411 err
= ext4_ext_dirty(handle
, inode
, path
);
2419 * ext4_ext_calc_credits_for_single_extent:
2420 * This routine returns max. credits that needed to insert an extent
2421 * to the extent tree.
2422 * When pass the actual path, the caller should calculate credits
2425 int ext4_ext_calc_credits_for_single_extent(struct inode
*inode
, int nrblocks
,
2426 struct ext4_ext_path
*path
)
2429 int depth
= ext_depth(inode
);
2432 /* probably there is space in leaf? */
2433 if (le16_to_cpu(path
[depth
].p_hdr
->eh_entries
)
2434 < le16_to_cpu(path
[depth
].p_hdr
->eh_max
)) {
2437 * There are some space in the leaf tree, no
2438 * need to account for leaf block credit
2440 * bitmaps and block group descriptor blocks
2441 * and other metadata blocks still need to be
2444 /* 1 bitmap, 1 block group descriptor */
2445 ret
= 2 + EXT4_META_TRANS_BLOCKS(inode
->i_sb
);
2450 return ext4_chunk_trans_blocks(inode
, nrblocks
);
2454 * How many index/leaf blocks need to change/allocate to add @extents extents?
2456 * If we add a single extent, then in the worse case, each tree level
2457 * index/leaf need to be changed in case of the tree split.
2459 * If more extents are inserted, they could cause the whole tree split more
2460 * than once, but this is really rare.
2462 int ext4_ext_index_trans_blocks(struct inode
*inode
, int extents
)
2467 /* If we are converting the inline data, only one is needed here. */
2468 if (ext4_has_inline_data(inode
))
2471 depth
= ext_depth(inode
);
2481 static inline int get_default_free_blocks_flags(struct inode
*inode
)
2483 if (S_ISDIR(inode
->i_mode
) || S_ISLNK(inode
->i_mode
))
2484 return EXT4_FREE_BLOCKS_METADATA
| EXT4_FREE_BLOCKS_FORGET
;
2485 else if (ext4_should_journal_data(inode
))
2486 return EXT4_FREE_BLOCKS_FORGET
;
2490 static int ext4_remove_blocks(handle_t
*handle
, struct inode
*inode
,
2491 struct ext4_extent
*ex
,
2492 long long *partial_cluster
,
2493 ext4_lblk_t from
, ext4_lblk_t to
)
2495 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
2496 unsigned short ee_len
= ext4_ext_get_actual_len(ex
);
2498 int flags
= get_default_free_blocks_flags(inode
);
2501 * For bigalloc file systems, we never free a partial cluster
2502 * at the beginning of the extent. Instead, we make a note
2503 * that we tried freeing the cluster, and check to see if we
2504 * need to free it on a subsequent call to ext4_remove_blocks,
2505 * or at the end of ext4_ext_rm_leaf or ext4_ext_remove_space.
2507 flags
|= EXT4_FREE_BLOCKS_NOFREE_FIRST_CLUSTER
;
2509 trace_ext4_remove_blocks(inode
, ex
, from
, to
, *partial_cluster
);
2511 * If we have a partial cluster, and it's different from the
2512 * cluster of the last block, we need to explicitly free the
2513 * partial cluster here.
2515 pblk
= ext4_ext_pblock(ex
) + ee_len
- 1;
2516 if (*partial_cluster
> 0 &&
2517 *partial_cluster
!= (long long) EXT4_B2C(sbi
, pblk
)) {
2518 ext4_free_blocks(handle
, inode
, NULL
,
2519 EXT4_C2B(sbi
, *partial_cluster
),
2520 sbi
->s_cluster_ratio
, flags
);
2521 *partial_cluster
= 0;
2524 #ifdef EXTENTS_STATS
2526 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
2527 spin_lock(&sbi
->s_ext_stats_lock
);
2528 sbi
->s_ext_blocks
+= ee_len
;
2529 sbi
->s_ext_extents
++;
2530 if (ee_len
< sbi
->s_ext_min
)
2531 sbi
->s_ext_min
= ee_len
;
2532 if (ee_len
> sbi
->s_ext_max
)
2533 sbi
->s_ext_max
= ee_len
;
2534 if (ext_depth(inode
) > sbi
->s_depth_max
)
2535 sbi
->s_depth_max
= ext_depth(inode
);
2536 spin_unlock(&sbi
->s_ext_stats_lock
);
2539 if (from
>= le32_to_cpu(ex
->ee_block
)
2540 && to
== le32_to_cpu(ex
->ee_block
) + ee_len
- 1) {
2543 long long first_cluster
;
2545 num
= le32_to_cpu(ex
->ee_block
) + ee_len
- from
;
2546 pblk
= ext4_ext_pblock(ex
) + ee_len
- num
;
2548 * Usually we want to free partial cluster at the end of the
2549 * extent, except for the situation when the cluster is still
2550 * used by any other extent (partial_cluster is negative).
2552 if (*partial_cluster
< 0 &&
2553 *partial_cluster
== -(long long) EXT4_B2C(sbi
, pblk
+num
-1))
2554 flags
|= EXT4_FREE_BLOCKS_NOFREE_LAST_CLUSTER
;
2556 ext_debug("free last %u blocks starting %llu partial %lld\n",
2557 num
, pblk
, *partial_cluster
);
2558 ext4_free_blocks(handle
, inode
, NULL
, pblk
, num
, flags
);
2560 * If the block range to be freed didn't start at the
2561 * beginning of a cluster, and we removed the entire
2562 * extent and the cluster is not used by any other extent,
2563 * save the partial cluster here, since we might need to
2564 * delete if we determine that the truncate or punch hole
2565 * operation has removed all of the blocks in the cluster.
2566 * If that cluster is used by another extent, preserve its
2567 * negative value so it isn't freed later on.
2569 * If the whole extent wasn't freed, we've reached the
2570 * start of the truncated/punched region and have finished
2571 * removing blocks. If there's a partial cluster here it's
2572 * shared with the remainder of the extent and is no longer
2573 * a candidate for removal.
2575 if (EXT4_PBLK_COFF(sbi
, pblk
) && ee_len
== num
) {
2576 first_cluster
= (long long) EXT4_B2C(sbi
, pblk
);
2577 if (first_cluster
!= -*partial_cluster
)
2578 *partial_cluster
= first_cluster
;
2580 *partial_cluster
= 0;
2583 ext4_error(sbi
->s_sb
, "strange request: removal(2) "
2585 from
, to
, le32_to_cpu(ex
->ee_block
), ee_len
);
2591 * ext4_ext_rm_leaf() Removes the extents associated with the
2592 * blocks appearing between "start" and "end". Both "start"
2593 * and "end" must appear in the same extent or EIO is returned.
2595 * @handle: The journal handle
2596 * @inode: The files inode
2597 * @path: The path to the leaf
2598 * @partial_cluster: The cluster which we'll have to free if all extents
2599 * has been released from it. However, if this value is
2600 * negative, it's a cluster just to the right of the
2601 * punched region and it must not be freed.
2602 * @start: The first block to remove
2603 * @end: The last block to remove
2606 ext4_ext_rm_leaf(handle_t
*handle
, struct inode
*inode
,
2607 struct ext4_ext_path
*path
,
2608 long long *partial_cluster
,
2609 ext4_lblk_t start
, ext4_lblk_t end
)
2611 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
2612 int err
= 0, correct_index
= 0;
2613 int depth
= ext_depth(inode
), credits
;
2614 struct ext4_extent_header
*eh
;
2617 ext4_lblk_t ex_ee_block
;
2618 unsigned short ex_ee_len
;
2619 unsigned unwritten
= 0;
2620 struct ext4_extent
*ex
;
2623 /* the header must be checked already in ext4_ext_remove_space() */
2624 ext_debug("truncate since %u in leaf to %u\n", start
, end
);
2625 if (!path
[depth
].p_hdr
)
2626 path
[depth
].p_hdr
= ext_block_hdr(path
[depth
].p_bh
);
2627 eh
= path
[depth
].p_hdr
;
2628 if (unlikely(path
[depth
].p_hdr
== NULL
)) {
2629 EXT4_ERROR_INODE(inode
, "path[%d].p_hdr == NULL", depth
);
2630 return -EFSCORRUPTED
;
2632 /* find where to start removing */
2633 ex
= path
[depth
].p_ext
;
2635 ex
= EXT_LAST_EXTENT(eh
);
2637 ex_ee_block
= le32_to_cpu(ex
->ee_block
);
2638 ex_ee_len
= ext4_ext_get_actual_len(ex
);
2640 trace_ext4_ext_rm_leaf(inode
, start
, ex
, *partial_cluster
);
2642 while (ex
>= EXT_FIRST_EXTENT(eh
) &&
2643 ex_ee_block
+ ex_ee_len
> start
) {
2645 if (ext4_ext_is_unwritten(ex
))
2650 ext_debug("remove ext %u:[%d]%d\n", ex_ee_block
,
2651 unwritten
, ex_ee_len
);
2652 path
[depth
].p_ext
= ex
;
2654 a
= ex_ee_block
> start
? ex_ee_block
: start
;
2655 b
= ex_ee_block
+ex_ee_len
- 1 < end
?
2656 ex_ee_block
+ex_ee_len
- 1 : end
;
2658 ext_debug(" border %u:%u\n", a
, b
);
2660 /* If this extent is beyond the end of the hole, skip it */
2661 if (end
< ex_ee_block
) {
2663 * We're going to skip this extent and move to another,
2664 * so note that its first cluster is in use to avoid
2665 * freeing it when removing blocks. Eventually, the
2666 * right edge of the truncated/punched region will
2667 * be just to the left.
2669 if (sbi
->s_cluster_ratio
> 1) {
2670 pblk
= ext4_ext_pblock(ex
);
2672 -(long long) EXT4_B2C(sbi
, pblk
);
2675 ex_ee_block
= le32_to_cpu(ex
->ee_block
);
2676 ex_ee_len
= ext4_ext_get_actual_len(ex
);
2678 } else if (b
!= ex_ee_block
+ ex_ee_len
- 1) {
2679 EXT4_ERROR_INODE(inode
,
2680 "can not handle truncate %u:%u "
2682 start
, end
, ex_ee_block
,
2683 ex_ee_block
+ ex_ee_len
- 1);
2684 err
= -EFSCORRUPTED
;
2686 } else if (a
!= ex_ee_block
) {
2687 /* remove tail of the extent */
2688 num
= a
- ex_ee_block
;
2690 /* remove whole extent: excellent! */
2694 * 3 for leaf, sb, and inode plus 2 (bmap and group
2695 * descriptor) for each block group; assume two block
2696 * groups plus ex_ee_len/blocks_per_block_group for
2699 credits
= 7 + 2*(ex_ee_len
/EXT4_BLOCKS_PER_GROUP(inode
->i_sb
));
2700 if (ex
== EXT_FIRST_EXTENT(eh
)) {
2702 credits
+= (ext_depth(inode
)) + 1;
2704 credits
+= EXT4_MAXQUOTAS_TRANS_BLOCKS(inode
->i_sb
);
2706 err
= ext4_ext_truncate_extend_restart(handle
, inode
, credits
);
2710 err
= ext4_ext_get_access(handle
, inode
, path
+ depth
);
2714 err
= ext4_remove_blocks(handle
, inode
, ex
, partial_cluster
,
2720 /* this extent is removed; mark slot entirely unused */
2721 ext4_ext_store_pblock(ex
, 0);
2723 ex
->ee_len
= cpu_to_le16(num
);
2725 * Do not mark unwritten if all the blocks in the
2726 * extent have been removed.
2728 if (unwritten
&& num
)
2729 ext4_ext_mark_unwritten(ex
);
2731 * If the extent was completely released,
2732 * we need to remove it from the leaf
2735 if (end
!= EXT_MAX_BLOCKS
- 1) {
2737 * For hole punching, we need to scoot all the
2738 * extents up when an extent is removed so that
2739 * we dont have blank extents in the middle
2741 memmove(ex
, ex
+1, (EXT_LAST_EXTENT(eh
) - ex
) *
2742 sizeof(struct ext4_extent
));
2744 /* Now get rid of the one at the end */
2745 memset(EXT_LAST_EXTENT(eh
), 0,
2746 sizeof(struct ext4_extent
));
2748 le16_add_cpu(&eh
->eh_entries
, -1);
2751 err
= ext4_ext_dirty(handle
, inode
, path
+ depth
);
2755 ext_debug("new extent: %u:%u:%llu\n", ex_ee_block
, num
,
2756 ext4_ext_pblock(ex
));
2758 ex_ee_block
= le32_to_cpu(ex
->ee_block
);
2759 ex_ee_len
= ext4_ext_get_actual_len(ex
);
2762 if (correct_index
&& eh
->eh_entries
)
2763 err
= ext4_ext_correct_indexes(handle
, inode
, path
);
2766 * If there's a partial cluster and at least one extent remains in
2767 * the leaf, free the partial cluster if it isn't shared with the
2768 * current extent. If it is shared with the current extent
2769 * we zero partial_cluster because we've reached the start of the
2770 * truncated/punched region and we're done removing blocks.
2772 if (*partial_cluster
> 0 && ex
>= EXT_FIRST_EXTENT(eh
)) {
2773 pblk
= ext4_ext_pblock(ex
) + ex_ee_len
- 1;
2774 if (*partial_cluster
!= (long long) EXT4_B2C(sbi
, pblk
)) {
2775 ext4_free_blocks(handle
, inode
, NULL
,
2776 EXT4_C2B(sbi
, *partial_cluster
),
2777 sbi
->s_cluster_ratio
,
2778 get_default_free_blocks_flags(inode
));
2780 *partial_cluster
= 0;
2783 /* if this leaf is free, then we should
2784 * remove it from index block above */
2785 if (err
== 0 && eh
->eh_entries
== 0 && path
[depth
].p_bh
!= NULL
)
2786 err
= ext4_ext_rm_idx(handle
, inode
, path
, depth
);
2793 * ext4_ext_more_to_rm:
2794 * returns 1 if current index has to be freed (even partial)
2797 ext4_ext_more_to_rm(struct ext4_ext_path
*path
)
2799 BUG_ON(path
->p_idx
== NULL
);
2801 if (path
->p_idx
< EXT_FIRST_INDEX(path
->p_hdr
))
2805 * if truncate on deeper level happened, it wasn't partial,
2806 * so we have to consider current index for truncation
2808 if (le16_to_cpu(path
->p_hdr
->eh_entries
) == path
->p_block
)
2813 int ext4_ext_remove_space(struct inode
*inode
, ext4_lblk_t start
,
2816 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
2817 int depth
= ext_depth(inode
);
2818 struct ext4_ext_path
*path
= NULL
;
2819 long long partial_cluster
= 0;
2823 ext_debug("truncate since %u to %u\n", start
, end
);
2825 /* probably first extent we're gonna free will be last in block */
2826 handle
= ext4_journal_start(inode
, EXT4_HT_TRUNCATE
, depth
+ 1);
2828 return PTR_ERR(handle
);
2831 trace_ext4_ext_remove_space(inode
, start
, end
, depth
);
2834 * Check if we are removing extents inside the extent tree. If that
2835 * is the case, we are going to punch a hole inside the extent tree
2836 * so we have to check whether we need to split the extent covering
2837 * the last block to remove so we can easily remove the part of it
2838 * in ext4_ext_rm_leaf().
2840 if (end
< EXT_MAX_BLOCKS
- 1) {
2841 struct ext4_extent
*ex
;
2842 ext4_lblk_t ee_block
, ex_end
, lblk
;
2845 /* find extent for or closest extent to this block */
2846 path
= ext4_find_extent(inode
, end
, NULL
, EXT4_EX_NOCACHE
);
2848 ext4_journal_stop(handle
);
2849 return PTR_ERR(path
);
2851 depth
= ext_depth(inode
);
2852 /* Leaf not may not exist only if inode has no blocks at all */
2853 ex
= path
[depth
].p_ext
;
2856 EXT4_ERROR_INODE(inode
,
2857 "path[%d].p_hdr == NULL",
2859 err
= -EFSCORRUPTED
;
2864 ee_block
= le32_to_cpu(ex
->ee_block
);
2865 ex_end
= ee_block
+ ext4_ext_get_actual_len(ex
) - 1;
2868 * See if the last block is inside the extent, if so split
2869 * the extent at 'end' block so we can easily remove the
2870 * tail of the first part of the split extent in
2871 * ext4_ext_rm_leaf().
2873 if (end
>= ee_block
&& end
< ex_end
) {
2876 * If we're going to split the extent, note that
2877 * the cluster containing the block after 'end' is
2878 * in use to avoid freeing it when removing blocks.
2880 if (sbi
->s_cluster_ratio
> 1) {
2881 pblk
= ext4_ext_pblock(ex
) + end
- ee_block
+ 2;
2883 -(long long) EXT4_B2C(sbi
, pblk
);
2887 * Split the extent in two so that 'end' is the last
2888 * block in the first new extent. Also we should not
2889 * fail removing space due to ENOSPC so try to use
2890 * reserved block if that happens.
2892 err
= ext4_force_split_extent_at(handle
, inode
, &path
,
2897 } else if (sbi
->s_cluster_ratio
> 1 && end
>= ex_end
) {
2899 * If there's an extent to the right its first cluster
2900 * contains the immediate right boundary of the
2901 * truncated/punched region. Set partial_cluster to
2902 * its negative value so it won't be freed if shared
2903 * with the current extent. The end < ee_block case
2904 * is handled in ext4_ext_rm_leaf().
2907 err
= ext4_ext_search_right(inode
, path
, &lblk
, &pblk
,
2913 -(long long) EXT4_B2C(sbi
, pblk
);
2917 * We start scanning from right side, freeing all the blocks
2918 * after i_size and walking into the tree depth-wise.
2920 depth
= ext_depth(inode
);
2925 le16_to_cpu(path
[k
].p_hdr
->eh_entries
)+1;
2927 path
= kzalloc(sizeof(struct ext4_ext_path
) * (depth
+ 1),
2930 ext4_journal_stop(handle
);
2933 path
[0].p_maxdepth
= path
[0].p_depth
= depth
;
2934 path
[0].p_hdr
= ext_inode_hdr(inode
);
2937 if (ext4_ext_check(inode
, path
[0].p_hdr
, depth
, 0)) {
2938 err
= -EFSCORRUPTED
;
2944 while (i
>= 0 && err
== 0) {
2946 /* this is leaf block */
2947 err
= ext4_ext_rm_leaf(handle
, inode
, path
,
2948 &partial_cluster
, start
,
2950 /* root level has p_bh == NULL, brelse() eats this */
2951 brelse(path
[i
].p_bh
);
2952 path
[i
].p_bh
= NULL
;
2957 /* this is index block */
2958 if (!path
[i
].p_hdr
) {
2959 ext_debug("initialize header\n");
2960 path
[i
].p_hdr
= ext_block_hdr(path
[i
].p_bh
);
2963 if (!path
[i
].p_idx
) {
2964 /* this level hasn't been touched yet */
2965 path
[i
].p_idx
= EXT_LAST_INDEX(path
[i
].p_hdr
);
2966 path
[i
].p_block
= le16_to_cpu(path
[i
].p_hdr
->eh_entries
)+1;
2967 ext_debug("init index ptr: hdr 0x%p, num %d\n",
2969 le16_to_cpu(path
[i
].p_hdr
->eh_entries
));
2971 /* we were already here, see at next index */
2975 ext_debug("level %d - index, first 0x%p, cur 0x%p\n",
2976 i
, EXT_FIRST_INDEX(path
[i
].p_hdr
),
2978 if (ext4_ext_more_to_rm(path
+ i
)) {
2979 struct buffer_head
*bh
;
2980 /* go to the next level */
2981 ext_debug("move to level %d (block %llu)\n",
2982 i
+ 1, ext4_idx_pblock(path
[i
].p_idx
));
2983 memset(path
+ i
+ 1, 0, sizeof(*path
));
2984 bh
= read_extent_tree_block(inode
,
2985 ext4_idx_pblock(path
[i
].p_idx
), depth
- i
- 1,
2988 /* should we reset i_size? */
2992 /* Yield here to deal with large extent trees.
2993 * Should be a no-op if we did IO above. */
2995 if (WARN_ON(i
+ 1 > depth
)) {
2996 err
= -EFSCORRUPTED
;
2999 path
[i
+ 1].p_bh
= bh
;
3001 /* save actual number of indexes since this
3002 * number is changed at the next iteration */
3003 path
[i
].p_block
= le16_to_cpu(path
[i
].p_hdr
->eh_entries
);
3006 /* we finished processing this index, go up */
3007 if (path
[i
].p_hdr
->eh_entries
== 0 && i
> 0) {
3008 /* index is empty, remove it;
3009 * handle must be already prepared by the
3010 * truncatei_leaf() */
3011 err
= ext4_ext_rm_idx(handle
, inode
, path
, i
);
3013 /* root level has p_bh == NULL, brelse() eats this */
3014 brelse(path
[i
].p_bh
);
3015 path
[i
].p_bh
= NULL
;
3017 ext_debug("return to level %d\n", i
);
3021 trace_ext4_ext_remove_space_done(inode
, start
, end
, depth
,
3022 partial_cluster
, path
->p_hdr
->eh_entries
);
3025 * If we still have something in the partial cluster and we have removed
3026 * even the first extent, then we should free the blocks in the partial
3027 * cluster as well. (This code will only run when there are no leaves
3028 * to the immediate left of the truncated/punched region.)
3030 if (partial_cluster
> 0 && err
== 0) {
3031 /* don't zero partial_cluster since it's not used afterwards */
3032 ext4_free_blocks(handle
, inode
, NULL
,
3033 EXT4_C2B(sbi
, partial_cluster
),
3034 sbi
->s_cluster_ratio
,
3035 get_default_free_blocks_flags(inode
));
3038 /* TODO: flexible tree reduction should be here */
3039 if (path
->p_hdr
->eh_entries
== 0) {
3041 * truncate to zero freed all the tree,
3042 * so we need to correct eh_depth
3044 err
= ext4_ext_get_access(handle
, inode
, path
);
3046 ext_inode_hdr(inode
)->eh_depth
= 0;
3047 ext_inode_hdr(inode
)->eh_max
=
3048 cpu_to_le16(ext4_ext_space_root(inode
, 0));
3049 err
= ext4_ext_dirty(handle
, inode
, path
);
3053 ext4_ext_drop_refs(path
);
3058 ext4_journal_stop(handle
);
3064 * called at mount time
3066 void ext4_ext_init(struct super_block
*sb
)
3069 * possible initialization would be here
3072 if (ext4_has_feature_extents(sb
)) {
3073 #if defined(AGGRESSIVE_TEST) || defined(CHECK_BINSEARCH) || defined(EXTENTS_STATS)
3074 printk(KERN_INFO
"EXT4-fs: file extents enabled"
3075 #ifdef AGGRESSIVE_TEST
3076 ", aggressive tests"
3078 #ifdef CHECK_BINSEARCH
3081 #ifdef EXTENTS_STATS
3086 #ifdef EXTENTS_STATS
3087 spin_lock_init(&EXT4_SB(sb
)->s_ext_stats_lock
);
3088 EXT4_SB(sb
)->s_ext_min
= 1 << 30;
3089 EXT4_SB(sb
)->s_ext_max
= 0;
3095 * called at umount time
3097 void ext4_ext_release(struct super_block
*sb
)
3099 if (!ext4_has_feature_extents(sb
))
3102 #ifdef EXTENTS_STATS
3103 if (EXT4_SB(sb
)->s_ext_blocks
&& EXT4_SB(sb
)->s_ext_extents
) {
3104 struct ext4_sb_info
*sbi
= EXT4_SB(sb
);
3105 printk(KERN_ERR
"EXT4-fs: %lu blocks in %lu extents (%lu ave)\n",
3106 sbi
->s_ext_blocks
, sbi
->s_ext_extents
,
3107 sbi
->s_ext_blocks
/ sbi
->s_ext_extents
);
3108 printk(KERN_ERR
"EXT4-fs: extents: %lu min, %lu max, max depth %lu\n",
3109 sbi
->s_ext_min
, sbi
->s_ext_max
, sbi
->s_depth_max
);
3114 static int ext4_zeroout_es(struct inode
*inode
, struct ext4_extent
*ex
)
3116 ext4_lblk_t ee_block
;
3117 ext4_fsblk_t ee_pblock
;
3118 unsigned int ee_len
;
3120 ee_block
= le32_to_cpu(ex
->ee_block
);
3121 ee_len
= ext4_ext_get_actual_len(ex
);
3122 ee_pblock
= ext4_ext_pblock(ex
);
3127 return ext4_es_insert_extent(inode
, ee_block
, ee_len
, ee_pblock
,
3128 EXTENT_STATUS_WRITTEN
);
3131 /* FIXME!! we need to try to merge to left or right after zero-out */
3132 static int ext4_ext_zeroout(struct inode
*inode
, struct ext4_extent
*ex
)
3134 ext4_fsblk_t ee_pblock
;
3135 unsigned int ee_len
;
3137 ee_len
= ext4_ext_get_actual_len(ex
);
3138 ee_pblock
= ext4_ext_pblock(ex
);
3139 return ext4_issue_zeroout(inode
, le32_to_cpu(ex
->ee_block
), ee_pblock
,
3144 * ext4_split_extent_at() splits an extent at given block.
3146 * @handle: the journal handle
3147 * @inode: the file inode
3148 * @path: the path to the extent
3149 * @split: the logical block where the extent is splitted.
3150 * @split_flags: indicates if the extent could be zeroout if split fails, and
3151 * the states(init or unwritten) of new extents.
3152 * @flags: flags used to insert new extent to extent tree.
3155 * Splits extent [a, b] into two extents [a, @split) and [@split, b], states
3156 * of which are deterimined by split_flag.
3158 * There are two cases:
3159 * a> the extent are splitted into two extent.
3160 * b> split is not needed, and just mark the extent.
3162 * return 0 on success.
3164 static int ext4_split_extent_at(handle_t
*handle
,
3165 struct inode
*inode
,
3166 struct ext4_ext_path
**ppath
,
3171 struct ext4_ext_path
*path
= *ppath
;
3172 ext4_fsblk_t newblock
;
3173 ext4_lblk_t ee_block
;
3174 struct ext4_extent
*ex
, newex
, orig_ex
, zero_ex
;
3175 struct ext4_extent
*ex2
= NULL
;
3176 unsigned int ee_len
, depth
;
3179 BUG_ON((split_flag
& (EXT4_EXT_DATA_VALID1
| EXT4_EXT_DATA_VALID2
)) ==
3180 (EXT4_EXT_DATA_VALID1
| EXT4_EXT_DATA_VALID2
));
3182 ext_debug("ext4_split_extents_at: inode %lu, logical"
3183 "block %llu\n", inode
->i_ino
, (unsigned long long)split
);
3185 ext4_ext_show_leaf(inode
, path
);
3187 depth
= ext_depth(inode
);
3188 ex
= path
[depth
].p_ext
;
3189 ee_block
= le32_to_cpu(ex
->ee_block
);
3190 ee_len
= ext4_ext_get_actual_len(ex
);
3191 newblock
= split
- ee_block
+ ext4_ext_pblock(ex
);
3193 BUG_ON(split
< ee_block
|| split
>= (ee_block
+ ee_len
));
3194 BUG_ON(!ext4_ext_is_unwritten(ex
) &&
3195 split_flag
& (EXT4_EXT_MAY_ZEROOUT
|
3196 EXT4_EXT_MARK_UNWRIT1
|
3197 EXT4_EXT_MARK_UNWRIT2
));
3199 err
= ext4_ext_get_access(handle
, inode
, path
+ depth
);
3203 if (split
== ee_block
) {
3205 * case b: block @split is the block that the extent begins with
3206 * then we just change the state of the extent, and splitting
3209 if (split_flag
& EXT4_EXT_MARK_UNWRIT2
)
3210 ext4_ext_mark_unwritten(ex
);
3212 ext4_ext_mark_initialized(ex
);
3214 if (!(flags
& EXT4_GET_BLOCKS_PRE_IO
))
3215 ext4_ext_try_to_merge(handle
, inode
, path
, ex
);
3217 err
= ext4_ext_dirty(handle
, inode
, path
+ path
->p_depth
);
3222 memcpy(&orig_ex
, ex
, sizeof(orig_ex
));
3223 ex
->ee_len
= cpu_to_le16(split
- ee_block
);
3224 if (split_flag
& EXT4_EXT_MARK_UNWRIT1
)
3225 ext4_ext_mark_unwritten(ex
);
3228 * path may lead to new leaf, not to original leaf any more
3229 * after ext4_ext_insert_extent() returns,
3231 err
= ext4_ext_dirty(handle
, inode
, path
+ depth
);
3233 goto fix_extent_len
;
3236 ex2
->ee_block
= cpu_to_le32(split
);
3237 ex2
->ee_len
= cpu_to_le16(ee_len
- (split
- ee_block
));
3238 ext4_ext_store_pblock(ex2
, newblock
);
3239 if (split_flag
& EXT4_EXT_MARK_UNWRIT2
)
3240 ext4_ext_mark_unwritten(ex2
);
3242 err
= ext4_ext_insert_extent(handle
, inode
, ppath
, &newex
, flags
);
3243 if (err
== -ENOSPC
&& (EXT4_EXT_MAY_ZEROOUT
& split_flag
)) {
3244 if (split_flag
& (EXT4_EXT_DATA_VALID1
|EXT4_EXT_DATA_VALID2
)) {
3245 if (split_flag
& EXT4_EXT_DATA_VALID1
) {
3246 err
= ext4_ext_zeroout(inode
, ex2
);
3247 zero_ex
.ee_block
= ex2
->ee_block
;
3248 zero_ex
.ee_len
= cpu_to_le16(
3249 ext4_ext_get_actual_len(ex2
));
3250 ext4_ext_store_pblock(&zero_ex
,
3251 ext4_ext_pblock(ex2
));
3253 err
= ext4_ext_zeroout(inode
, ex
);
3254 zero_ex
.ee_block
= ex
->ee_block
;
3255 zero_ex
.ee_len
= cpu_to_le16(
3256 ext4_ext_get_actual_len(ex
));
3257 ext4_ext_store_pblock(&zero_ex
,
3258 ext4_ext_pblock(ex
));
3261 err
= ext4_ext_zeroout(inode
, &orig_ex
);
3262 zero_ex
.ee_block
= orig_ex
.ee_block
;
3263 zero_ex
.ee_len
= cpu_to_le16(
3264 ext4_ext_get_actual_len(&orig_ex
));
3265 ext4_ext_store_pblock(&zero_ex
,
3266 ext4_ext_pblock(&orig_ex
));
3270 goto fix_extent_len
;
3271 /* update the extent length and mark as initialized */
3272 ex
->ee_len
= cpu_to_le16(ee_len
);
3273 ext4_ext_try_to_merge(handle
, inode
, path
, ex
);
3274 err
= ext4_ext_dirty(handle
, inode
, path
+ path
->p_depth
);
3276 goto fix_extent_len
;
3278 /* update extent status tree */
3279 err
= ext4_zeroout_es(inode
, &zero_ex
);
3283 goto fix_extent_len
;
3286 ext4_ext_show_leaf(inode
, path
);
3290 ex
->ee_len
= orig_ex
.ee_len
;
3291 ext4_ext_dirty(handle
, inode
, path
+ path
->p_depth
);
3296 * ext4_split_extents() splits an extent and mark extent which is covered
3297 * by @map as split_flags indicates
3299 * It may result in splitting the extent into multiple extents (up to three)
3300 * There are three possibilities:
3301 * a> There is no split required
3302 * b> Splits in two extents: Split is happening at either end of the extent
3303 * c> Splits in three extents: Somone is splitting in middle of the extent
3306 static int ext4_split_extent(handle_t
*handle
,
3307 struct inode
*inode
,
3308 struct ext4_ext_path
**ppath
,
3309 struct ext4_map_blocks
*map
,
3313 struct ext4_ext_path
*path
= *ppath
;
3314 ext4_lblk_t ee_block
;
3315 struct ext4_extent
*ex
;
3316 unsigned int ee_len
, depth
;
3319 int split_flag1
, flags1
;
3320 int allocated
= map
->m_len
;
3322 depth
= ext_depth(inode
);
3323 ex
= path
[depth
].p_ext
;
3324 ee_block
= le32_to_cpu(ex
->ee_block
);
3325 ee_len
= ext4_ext_get_actual_len(ex
);
3326 unwritten
= ext4_ext_is_unwritten(ex
);
3328 if (map
->m_lblk
+ map
->m_len
< ee_block
+ ee_len
) {
3329 split_flag1
= split_flag
& EXT4_EXT_MAY_ZEROOUT
;
3330 flags1
= flags
| EXT4_GET_BLOCKS_PRE_IO
;
3332 split_flag1
|= EXT4_EXT_MARK_UNWRIT1
|
3333 EXT4_EXT_MARK_UNWRIT2
;
3334 if (split_flag
& EXT4_EXT_DATA_VALID2
)
3335 split_flag1
|= EXT4_EXT_DATA_VALID1
;
3336 err
= ext4_split_extent_at(handle
, inode
, ppath
,
3337 map
->m_lblk
+ map
->m_len
, split_flag1
, flags1
);
3341 allocated
= ee_len
- (map
->m_lblk
- ee_block
);
3344 * Update path is required because previous ext4_split_extent_at() may
3345 * result in split of original leaf or extent zeroout.
3347 path
= ext4_find_extent(inode
, map
->m_lblk
, ppath
, 0);
3349 return PTR_ERR(path
);
3350 depth
= ext_depth(inode
);
3351 ex
= path
[depth
].p_ext
;
3353 EXT4_ERROR_INODE(inode
, "unexpected hole at %lu",
3354 (unsigned long) map
->m_lblk
);
3355 return -EFSCORRUPTED
;
3357 unwritten
= ext4_ext_is_unwritten(ex
);
3360 if (map
->m_lblk
>= ee_block
) {
3361 split_flag1
= split_flag
& EXT4_EXT_DATA_VALID2
;
3363 split_flag1
|= EXT4_EXT_MARK_UNWRIT1
;
3364 split_flag1
|= split_flag
& (EXT4_EXT_MAY_ZEROOUT
|
3365 EXT4_EXT_MARK_UNWRIT2
);
3367 err
= ext4_split_extent_at(handle
, inode
, ppath
,
3368 map
->m_lblk
, split_flag1
, flags
);
3373 ext4_ext_show_leaf(inode
, path
);
3375 return err
? err
: allocated
;
3379 * This function is called by ext4_ext_map_blocks() if someone tries to write
3380 * to an unwritten extent. It may result in splitting the unwritten
3381 * extent into multiple extents (up to three - one initialized and two
3383 * There are three possibilities:
3384 * a> There is no split required: Entire extent should be initialized
3385 * b> Splits in two extents: Write is happening at either end of the extent
3386 * c> Splits in three extents: Somone is writing in middle of the extent
3389 * - The extent pointed to by 'path' is unwritten.
3390 * - The extent pointed to by 'path' contains a superset
3391 * of the logical span [map->m_lblk, map->m_lblk + map->m_len).
3393 * Post-conditions on success:
3394 * - the returned value is the number of blocks beyond map->l_lblk
3395 * that are allocated and initialized.
3396 * It is guaranteed to be >= map->m_len.
3398 static int ext4_ext_convert_to_initialized(handle_t
*handle
,
3399 struct inode
*inode
,
3400 struct ext4_map_blocks
*map
,
3401 struct ext4_ext_path
**ppath
,
3404 struct ext4_ext_path
*path
= *ppath
;
3405 struct ext4_sb_info
*sbi
;
3406 struct ext4_extent_header
*eh
;
3407 struct ext4_map_blocks split_map
;
3408 struct ext4_extent zero_ex
;
3409 struct ext4_extent
*ex
, *abut_ex
;
3410 ext4_lblk_t ee_block
, eof_block
;
3411 unsigned int ee_len
, depth
, map_len
= map
->m_len
;
3412 int allocated
= 0, max_zeroout
= 0;
3416 ext_debug("ext4_ext_convert_to_initialized: inode %lu, logical"
3417 "block %llu, max_blocks %u\n", inode
->i_ino
,
3418 (unsigned long long)map
->m_lblk
, map_len
);
3420 sbi
= EXT4_SB(inode
->i_sb
);
3421 eof_block
= (inode
->i_size
+ inode
->i_sb
->s_blocksize
- 1) >>
3422 inode
->i_sb
->s_blocksize_bits
;
3423 if (eof_block
< map
->m_lblk
+ map_len
)
3424 eof_block
= map
->m_lblk
+ map_len
;
3426 depth
= ext_depth(inode
);
3427 eh
= path
[depth
].p_hdr
;
3428 ex
= path
[depth
].p_ext
;
3429 ee_block
= le32_to_cpu(ex
->ee_block
);
3430 ee_len
= ext4_ext_get_actual_len(ex
);
3433 trace_ext4_ext_convert_to_initialized_enter(inode
, map
, ex
);
3435 /* Pre-conditions */
3436 BUG_ON(!ext4_ext_is_unwritten(ex
));
3437 BUG_ON(!in_range(map
->m_lblk
, ee_block
, ee_len
));
3440 * Attempt to transfer newly initialized blocks from the currently
3441 * unwritten extent to its neighbor. This is much cheaper
3442 * than an insertion followed by a merge as those involve costly
3443 * memmove() calls. Transferring to the left is the common case in
3444 * steady state for workloads doing fallocate(FALLOC_FL_KEEP_SIZE)
3445 * followed by append writes.
3447 * Limitations of the current logic:
3448 * - L1: we do not deal with writes covering the whole extent.
3449 * This would require removing the extent if the transfer
3451 * - L2: we only attempt to merge with an extent stored in the
3452 * same extent tree node.
3454 if ((map
->m_lblk
== ee_block
) &&
3455 /* See if we can merge left */
3456 (map_len
< ee_len
) && /*L1*/
3457 (ex
> EXT_FIRST_EXTENT(eh
))) { /*L2*/
3458 ext4_lblk_t prev_lblk
;
3459 ext4_fsblk_t prev_pblk
, ee_pblk
;
3460 unsigned int prev_len
;
3463 prev_lblk
= le32_to_cpu(abut_ex
->ee_block
);
3464 prev_len
= ext4_ext_get_actual_len(abut_ex
);
3465 prev_pblk
= ext4_ext_pblock(abut_ex
);
3466 ee_pblk
= ext4_ext_pblock(ex
);
3469 * A transfer of blocks from 'ex' to 'abut_ex' is allowed
3470 * upon those conditions:
3471 * - C1: abut_ex is initialized,
3472 * - C2: abut_ex is logically abutting ex,
3473 * - C3: abut_ex is physically abutting ex,
3474 * - C4: abut_ex can receive the additional blocks without
3475 * overflowing the (initialized) length limit.
3477 if ((!ext4_ext_is_unwritten(abut_ex
)) && /*C1*/
3478 ((prev_lblk
+ prev_len
) == ee_block
) && /*C2*/
3479 ((prev_pblk
+ prev_len
) == ee_pblk
) && /*C3*/
3480 (prev_len
< (EXT_INIT_MAX_LEN
- map_len
))) { /*C4*/
3481 err
= ext4_ext_get_access(handle
, inode
, path
+ depth
);
3485 trace_ext4_ext_convert_to_initialized_fastpath(inode
,
3488 /* Shift the start of ex by 'map_len' blocks */
3489 ex
->ee_block
= cpu_to_le32(ee_block
+ map_len
);
3490 ext4_ext_store_pblock(ex
, ee_pblk
+ map_len
);
3491 ex
->ee_len
= cpu_to_le16(ee_len
- map_len
);
3492 ext4_ext_mark_unwritten(ex
); /* Restore the flag */
3494 /* Extend abut_ex by 'map_len' blocks */
3495 abut_ex
->ee_len
= cpu_to_le16(prev_len
+ map_len
);
3497 /* Result: number of initialized blocks past m_lblk */
3498 allocated
= map_len
;
3500 } else if (((map
->m_lblk
+ map_len
) == (ee_block
+ ee_len
)) &&
3501 (map_len
< ee_len
) && /*L1*/
3502 ex
< EXT_LAST_EXTENT(eh
)) { /*L2*/
3503 /* See if we can merge right */
3504 ext4_lblk_t next_lblk
;
3505 ext4_fsblk_t next_pblk
, ee_pblk
;
3506 unsigned int next_len
;
3509 next_lblk
= le32_to_cpu(abut_ex
->ee_block
);
3510 next_len
= ext4_ext_get_actual_len(abut_ex
);
3511 next_pblk
= ext4_ext_pblock(abut_ex
);
3512 ee_pblk
= ext4_ext_pblock(ex
);
3515 * A transfer of blocks from 'ex' to 'abut_ex' is allowed
3516 * upon those conditions:
3517 * - C1: abut_ex is initialized,
3518 * - C2: abut_ex is logically abutting ex,
3519 * - C3: abut_ex is physically abutting ex,
3520 * - C4: abut_ex can receive the additional blocks without
3521 * overflowing the (initialized) length limit.
3523 if ((!ext4_ext_is_unwritten(abut_ex
)) && /*C1*/
3524 ((map
->m_lblk
+ map_len
) == next_lblk
) && /*C2*/
3525 ((ee_pblk
+ ee_len
) == next_pblk
) && /*C3*/
3526 (next_len
< (EXT_INIT_MAX_LEN
- map_len
))) { /*C4*/
3527 err
= ext4_ext_get_access(handle
, inode
, path
+ depth
);
3531 trace_ext4_ext_convert_to_initialized_fastpath(inode
,
3534 /* Shift the start of abut_ex by 'map_len' blocks */
3535 abut_ex
->ee_block
= cpu_to_le32(next_lblk
- map_len
);
3536 ext4_ext_store_pblock(abut_ex
, next_pblk
- map_len
);
3537 ex
->ee_len
= cpu_to_le16(ee_len
- map_len
);
3538 ext4_ext_mark_unwritten(ex
); /* Restore the flag */
3540 /* Extend abut_ex by 'map_len' blocks */
3541 abut_ex
->ee_len
= cpu_to_le16(next_len
+ map_len
);
3543 /* Result: number of initialized blocks past m_lblk */
3544 allocated
= map_len
;
3548 /* Mark the block containing both extents as dirty */
3549 ext4_ext_dirty(handle
, inode
, path
+ depth
);
3551 /* Update path to point to the right extent */
3552 path
[depth
].p_ext
= abut_ex
;
3555 allocated
= ee_len
- (map
->m_lblk
- ee_block
);
3557 WARN_ON(map
->m_lblk
< ee_block
);
3559 * It is safe to convert extent to initialized via explicit
3560 * zeroout only if extent is fully inside i_size or new_size.
3562 split_flag
|= ee_block
+ ee_len
<= eof_block
? EXT4_EXT_MAY_ZEROOUT
: 0;
3564 if (EXT4_EXT_MAY_ZEROOUT
& split_flag
)
3565 max_zeroout
= sbi
->s_extent_max_zeroout_kb
>>
3566 (inode
->i_sb
->s_blocksize_bits
- 10);
3568 if (ext4_encrypted_inode(inode
))
3571 /* If extent is less than s_max_zeroout_kb, zeroout directly */
3572 if (max_zeroout
&& (ee_len
<= max_zeroout
)) {
3573 err
= ext4_ext_zeroout(inode
, ex
);
3576 zero_ex
.ee_block
= ex
->ee_block
;
3577 zero_ex
.ee_len
= cpu_to_le16(ext4_ext_get_actual_len(ex
));
3578 ext4_ext_store_pblock(&zero_ex
, ext4_ext_pblock(ex
));
3580 err
= ext4_ext_get_access(handle
, inode
, path
+ depth
);
3583 ext4_ext_mark_initialized(ex
);
3584 ext4_ext_try_to_merge(handle
, inode
, path
, ex
);
3585 err
= ext4_ext_dirty(handle
, inode
, path
+ path
->p_depth
);
3591 * 1. split the extent into three extents.
3592 * 2. split the extent into two extents, zeroout the first half.
3593 * 3. split the extent into two extents, zeroout the second half.
3594 * 4. split the extent into two extents with out zeroout.
3596 split_map
.m_lblk
= map
->m_lblk
;
3597 split_map
.m_len
= map
->m_len
;
3599 if (max_zeroout
&& (allocated
> map
->m_len
)) {
3600 if (allocated
<= max_zeroout
) {
3603 cpu_to_le32(map
->m_lblk
);
3604 zero_ex
.ee_len
= cpu_to_le16(allocated
);
3605 ext4_ext_store_pblock(&zero_ex
,
3606 ext4_ext_pblock(ex
) + map
->m_lblk
- ee_block
);
3607 err
= ext4_ext_zeroout(inode
, &zero_ex
);
3610 split_map
.m_lblk
= map
->m_lblk
;
3611 split_map
.m_len
= allocated
;
3612 } else if (map
->m_lblk
- ee_block
+ map
->m_len
< max_zeroout
) {
3614 if (map
->m_lblk
!= ee_block
) {
3615 zero_ex
.ee_block
= ex
->ee_block
;
3616 zero_ex
.ee_len
= cpu_to_le16(map
->m_lblk
-
3618 ext4_ext_store_pblock(&zero_ex
,
3619 ext4_ext_pblock(ex
));
3620 err
= ext4_ext_zeroout(inode
, &zero_ex
);
3625 split_map
.m_lblk
= ee_block
;
3626 split_map
.m_len
= map
->m_lblk
- ee_block
+ map
->m_len
;
3627 allocated
= map
->m_len
;
3631 err
= ext4_split_extent(handle
, inode
, ppath
, &split_map
, split_flag
,
3636 /* If we have gotten a failure, don't zero out status tree */
3638 err
= ext4_zeroout_es(inode
, &zero_ex
);
3639 return err
? err
: allocated
;
3643 * This function is called by ext4_ext_map_blocks() from
3644 * ext4_get_blocks_dio_write() when DIO to write
3645 * to an unwritten extent.
3647 * Writing to an unwritten extent may result in splitting the unwritten
3648 * extent into multiple initialized/unwritten extents (up to three)
3649 * There are three possibilities:
3650 * a> There is no split required: Entire extent should be unwritten
3651 * b> Splits in two extents: Write is happening at either end of the extent
3652 * c> Splits in three extents: Somone is writing in middle of the extent
3654 * This works the same way in the case of initialized -> unwritten conversion.
3656 * One of more index blocks maybe needed if the extent tree grow after
3657 * the unwritten extent split. To prevent ENOSPC occur at the IO
3658 * complete, we need to split the unwritten extent before DIO submit
3659 * the IO. The unwritten extent called at this time will be split
3660 * into three unwritten extent(at most). After IO complete, the part
3661 * being filled will be convert to initialized by the end_io callback function
3662 * via ext4_convert_unwritten_extents().
3664 * Returns the size of unwritten extent to be written on success.
3666 static int ext4_split_convert_extents(handle_t
*handle
,
3667 struct inode
*inode
,
3668 struct ext4_map_blocks
*map
,
3669 struct ext4_ext_path
**ppath
,
3672 struct ext4_ext_path
*path
= *ppath
;
3673 ext4_lblk_t eof_block
;
3674 ext4_lblk_t ee_block
;
3675 struct ext4_extent
*ex
;
3676 unsigned int ee_len
;
3677 int split_flag
= 0, depth
;
3679 ext_debug("%s: inode %lu, logical block %llu, max_blocks %u\n",
3680 __func__
, inode
->i_ino
,
3681 (unsigned long long)map
->m_lblk
, map
->m_len
);
3683 eof_block
= (inode
->i_size
+ inode
->i_sb
->s_blocksize
- 1) >>
3684 inode
->i_sb
->s_blocksize_bits
;
3685 if (eof_block
< map
->m_lblk
+ map
->m_len
)
3686 eof_block
= map
->m_lblk
+ map
->m_len
;
3688 * It is safe to convert extent to initialized via explicit
3689 * zeroout only if extent is fully insde i_size or new_size.
3691 depth
= ext_depth(inode
);
3692 ex
= path
[depth
].p_ext
;
3693 ee_block
= le32_to_cpu(ex
->ee_block
);
3694 ee_len
= ext4_ext_get_actual_len(ex
);
3696 /* Convert to unwritten */
3697 if (flags
& EXT4_GET_BLOCKS_CONVERT_UNWRITTEN
) {
3698 split_flag
|= EXT4_EXT_DATA_VALID1
;
3699 /* Convert to initialized */
3700 } else if (flags
& EXT4_GET_BLOCKS_CONVERT
) {
3701 split_flag
|= ee_block
+ ee_len
<= eof_block
?
3702 EXT4_EXT_MAY_ZEROOUT
: 0;
3703 split_flag
|= (EXT4_EXT_MARK_UNWRIT2
| EXT4_EXT_DATA_VALID2
);
3705 flags
|= EXT4_GET_BLOCKS_PRE_IO
;
3706 return ext4_split_extent(handle
, inode
, ppath
, map
, split_flag
, flags
);
3709 static int ext4_convert_unwritten_extents_endio(handle_t
*handle
,
3710 struct inode
*inode
,
3711 struct ext4_map_blocks
*map
,
3712 struct ext4_ext_path
**ppath
)
3714 struct ext4_ext_path
*path
= *ppath
;
3715 struct ext4_extent
*ex
;
3716 ext4_lblk_t ee_block
;
3717 unsigned int ee_len
;
3721 depth
= ext_depth(inode
);
3722 ex
= path
[depth
].p_ext
;
3723 ee_block
= le32_to_cpu(ex
->ee_block
);
3724 ee_len
= ext4_ext_get_actual_len(ex
);
3726 ext_debug("ext4_convert_unwritten_extents_endio: inode %lu, logical"
3727 "block %llu, max_blocks %u\n", inode
->i_ino
,
3728 (unsigned long long)ee_block
, ee_len
);
3730 /* If extent is larger than requested it is a clear sign that we still
3731 * have some extent state machine issues left. So extent_split is still
3733 * TODO: Once all related issues will be fixed this situation should be
3736 if (ee_block
!= map
->m_lblk
|| ee_len
> map
->m_len
) {
3738 ext4_warning("Inode (%ld) finished: extent logical block %llu,"
3739 " len %u; IO logical block %llu, len %u",
3740 inode
->i_ino
, (unsigned long long)ee_block
, ee_len
,
3741 (unsigned long long)map
->m_lblk
, map
->m_len
);
3743 err
= ext4_split_convert_extents(handle
, inode
, map
, ppath
,
3744 EXT4_GET_BLOCKS_CONVERT
);
3747 path
= ext4_find_extent(inode
, map
->m_lblk
, ppath
, 0);
3749 return PTR_ERR(path
);
3750 depth
= ext_depth(inode
);
3751 ex
= path
[depth
].p_ext
;
3754 err
= ext4_ext_get_access(handle
, inode
, path
+ depth
);
3757 /* first mark the extent as initialized */
3758 ext4_ext_mark_initialized(ex
);
3760 /* note: ext4_ext_correct_indexes() isn't needed here because
3761 * borders are not changed
3763 ext4_ext_try_to_merge(handle
, inode
, path
, ex
);
3765 /* Mark modified extent as dirty */
3766 err
= ext4_ext_dirty(handle
, inode
, path
+ path
->p_depth
);
3768 ext4_ext_show_leaf(inode
, path
);
3772 static void unmap_underlying_metadata_blocks(struct block_device
*bdev
,
3773 sector_t block
, int count
)
3776 for (i
= 0; i
< count
; i
++)
3777 unmap_underlying_metadata(bdev
, block
+ i
);
3781 * Handle EOFBLOCKS_FL flag, clearing it if necessary
3783 static int check_eofblocks_fl(handle_t
*handle
, struct inode
*inode
,
3785 struct ext4_ext_path
*path
,
3789 struct ext4_extent_header
*eh
;
3790 struct ext4_extent
*last_ex
;
3792 if (!ext4_test_inode_flag(inode
, EXT4_INODE_EOFBLOCKS
))
3795 depth
= ext_depth(inode
);
3796 eh
= path
[depth
].p_hdr
;
3799 * We're going to remove EOFBLOCKS_FL entirely in future so we
3800 * do not care for this case anymore. Simply remove the flag
3801 * if there are no extents.
3803 if (unlikely(!eh
->eh_entries
))
3805 last_ex
= EXT_LAST_EXTENT(eh
);
3807 * We should clear the EOFBLOCKS_FL flag if we are writing the
3808 * last block in the last extent in the file. We test this by
3809 * first checking to see if the caller to
3810 * ext4_ext_get_blocks() was interested in the last block (or
3811 * a block beyond the last block) in the current extent. If
3812 * this turns out to be false, we can bail out from this
3813 * function immediately.
3815 if (lblk
+ len
< le32_to_cpu(last_ex
->ee_block
) +
3816 ext4_ext_get_actual_len(last_ex
))
3819 * If the caller does appear to be planning to write at or
3820 * beyond the end of the current extent, we then test to see
3821 * if the current extent is the last extent in the file, by
3822 * checking to make sure it was reached via the rightmost node
3823 * at each level of the tree.
3825 for (i
= depth
-1; i
>= 0; i
--)
3826 if (path
[i
].p_idx
!= EXT_LAST_INDEX(path
[i
].p_hdr
))
3829 ext4_clear_inode_flag(inode
, EXT4_INODE_EOFBLOCKS
);
3830 return ext4_mark_inode_dirty(handle
, inode
);
3834 * ext4_find_delalloc_range: find delayed allocated block in the given range.
3836 * Return 1 if there is a delalloc block in the range, otherwise 0.
3838 int ext4_find_delalloc_range(struct inode
*inode
,
3839 ext4_lblk_t lblk_start
,
3840 ext4_lblk_t lblk_end
)
3842 struct extent_status es
;
3844 ext4_es_find_delayed_extent_range(inode
, lblk_start
, lblk_end
, &es
);
3846 return 0; /* there is no delay extent in this tree */
3847 else if (es
.es_lblk
<= lblk_start
&&
3848 lblk_start
< es
.es_lblk
+ es
.es_len
)
3850 else if (lblk_start
<= es
.es_lblk
&& es
.es_lblk
<= lblk_end
)
3856 int ext4_find_delalloc_cluster(struct inode
*inode
, ext4_lblk_t lblk
)
3858 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
3859 ext4_lblk_t lblk_start
, lblk_end
;
3860 lblk_start
= EXT4_LBLK_CMASK(sbi
, lblk
);
3861 lblk_end
= lblk_start
+ sbi
->s_cluster_ratio
- 1;
3863 return ext4_find_delalloc_range(inode
, lblk_start
, lblk_end
);
3867 * Determines how many complete clusters (out of those specified by the 'map')
3868 * are under delalloc and were reserved quota for.
3869 * This function is called when we are writing out the blocks that were
3870 * originally written with their allocation delayed, but then the space was
3871 * allocated using fallocate() before the delayed allocation could be resolved.
3872 * The cases to look for are:
3873 * ('=' indicated delayed allocated blocks
3874 * '-' indicates non-delayed allocated blocks)
3875 * (a) partial clusters towards beginning and/or end outside of allocated range
3876 * are not delalloc'ed.
3878 * |----c---=|====c====|====c====|===-c----|
3879 * |++++++ allocated ++++++|
3880 * ==> 4 complete clusters in above example
3882 * (b) partial cluster (outside of allocated range) towards either end is
3883 * marked for delayed allocation. In this case, we will exclude that
3886 * |----====c========|========c========|
3887 * |++++++ allocated ++++++|
3888 * ==> 1 complete clusters in above example
3891 * |================c================|
3892 * |++++++ allocated ++++++|
3893 * ==> 0 complete clusters in above example
3895 * The ext4_da_update_reserve_space will be called only if we
3896 * determine here that there were some "entire" clusters that span
3897 * this 'allocated' range.
3898 * In the non-bigalloc case, this function will just end up returning num_blks
3899 * without ever calling ext4_find_delalloc_range.
3902 get_reserved_cluster_alloc(struct inode
*inode
, ext4_lblk_t lblk_start
,
3903 unsigned int num_blks
)
3905 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
3906 ext4_lblk_t alloc_cluster_start
, alloc_cluster_end
;
3907 ext4_lblk_t lblk_from
, lblk_to
, c_offset
;
3908 unsigned int allocated_clusters
= 0;
3910 alloc_cluster_start
= EXT4_B2C(sbi
, lblk_start
);
3911 alloc_cluster_end
= EXT4_B2C(sbi
, lblk_start
+ num_blks
- 1);
3913 /* max possible clusters for this allocation */
3914 allocated_clusters
= alloc_cluster_end
- alloc_cluster_start
+ 1;
3916 trace_ext4_get_reserved_cluster_alloc(inode
, lblk_start
, num_blks
);
3918 /* Check towards left side */
3919 c_offset
= EXT4_LBLK_COFF(sbi
, lblk_start
);
3921 lblk_from
= EXT4_LBLK_CMASK(sbi
, lblk_start
);
3922 lblk_to
= lblk_from
+ c_offset
- 1;
3924 if (ext4_find_delalloc_range(inode
, lblk_from
, lblk_to
))
3925 allocated_clusters
--;
3928 /* Now check towards right. */
3929 c_offset
= EXT4_LBLK_COFF(sbi
, lblk_start
+ num_blks
);
3930 if (allocated_clusters
&& c_offset
) {
3931 lblk_from
= lblk_start
+ num_blks
;
3932 lblk_to
= lblk_from
+ (sbi
->s_cluster_ratio
- c_offset
) - 1;
3934 if (ext4_find_delalloc_range(inode
, lblk_from
, lblk_to
))
3935 allocated_clusters
--;
3938 return allocated_clusters
;
3942 convert_initialized_extent(handle_t
*handle
, struct inode
*inode
,
3943 struct ext4_map_blocks
*map
,
3944 struct ext4_ext_path
**ppath
,
3945 unsigned int allocated
)
3947 struct ext4_ext_path
*path
= *ppath
;
3948 struct ext4_extent
*ex
;
3949 ext4_lblk_t ee_block
;
3950 unsigned int ee_len
;
3955 * Make sure that the extent is no bigger than we support with
3958 if (map
->m_len
> EXT_UNWRITTEN_MAX_LEN
)
3959 map
->m_len
= EXT_UNWRITTEN_MAX_LEN
/ 2;
3961 depth
= ext_depth(inode
);
3962 ex
= path
[depth
].p_ext
;
3963 ee_block
= le32_to_cpu(ex
->ee_block
);
3964 ee_len
= ext4_ext_get_actual_len(ex
);
3966 ext_debug("%s: inode %lu, logical"
3967 "block %llu, max_blocks %u\n", __func__
, inode
->i_ino
,
3968 (unsigned long long)ee_block
, ee_len
);
3970 if (ee_block
!= map
->m_lblk
|| ee_len
> map
->m_len
) {
3971 err
= ext4_split_convert_extents(handle
, inode
, map
, ppath
,
3972 EXT4_GET_BLOCKS_CONVERT_UNWRITTEN
);
3975 path
= ext4_find_extent(inode
, map
->m_lblk
, ppath
, 0);
3977 return PTR_ERR(path
);
3978 depth
= ext_depth(inode
);
3979 ex
= path
[depth
].p_ext
;
3981 EXT4_ERROR_INODE(inode
, "unexpected hole at %lu",
3982 (unsigned long) map
->m_lblk
);
3983 return -EFSCORRUPTED
;
3987 err
= ext4_ext_get_access(handle
, inode
, path
+ depth
);
3990 /* first mark the extent as unwritten */
3991 ext4_ext_mark_unwritten(ex
);
3993 /* note: ext4_ext_correct_indexes() isn't needed here because
3994 * borders are not changed
3996 ext4_ext_try_to_merge(handle
, inode
, path
, ex
);
3998 /* Mark modified extent as dirty */
3999 err
= ext4_ext_dirty(handle
, inode
, path
+ path
->p_depth
);
4002 ext4_ext_show_leaf(inode
, path
);
4004 ext4_update_inode_fsync_trans(handle
, inode
, 1);
4005 err
= check_eofblocks_fl(handle
, inode
, map
->m_lblk
, path
, map
->m_len
);
4008 map
->m_flags
|= EXT4_MAP_UNWRITTEN
;
4009 if (allocated
> map
->m_len
)
4010 allocated
= map
->m_len
;
4011 map
->m_len
= allocated
;
4016 ext4_ext_handle_unwritten_extents(handle_t
*handle
, struct inode
*inode
,
4017 struct ext4_map_blocks
*map
,
4018 struct ext4_ext_path
**ppath
, int flags
,
4019 unsigned int allocated
, ext4_fsblk_t newblock
)
4021 struct ext4_ext_path
*path
= *ppath
;
4025 ext_debug("ext4_ext_handle_unwritten_extents: inode %lu, logical "
4026 "block %llu, max_blocks %u, flags %x, allocated %u\n",
4027 inode
->i_ino
, (unsigned long long)map
->m_lblk
, map
->m_len
,
4029 ext4_ext_show_leaf(inode
, path
);
4032 * When writing into unwritten space, we should not fail to
4033 * allocate metadata blocks for the new extent block if needed.
4035 flags
|= EXT4_GET_BLOCKS_METADATA_NOFAIL
;
4037 trace_ext4_ext_handle_unwritten_extents(inode
, map
, flags
,
4038 allocated
, newblock
);
4040 /* get_block() before submit the IO, split the extent */
4041 if (flags
& EXT4_GET_BLOCKS_PRE_IO
) {
4042 ret
= ext4_split_convert_extents(handle
, inode
, map
, ppath
,
4043 flags
| EXT4_GET_BLOCKS_CONVERT
);
4046 map
->m_flags
|= EXT4_MAP_UNWRITTEN
;
4049 /* IO end_io complete, convert the filled extent to written */
4050 if (flags
& EXT4_GET_BLOCKS_CONVERT
) {
4051 if (flags
& EXT4_GET_BLOCKS_ZERO
) {
4052 if (allocated
> map
->m_len
)
4053 allocated
= map
->m_len
;
4054 err
= ext4_issue_zeroout(inode
, map
->m_lblk
, newblock
,
4059 ret
= ext4_convert_unwritten_extents_endio(handle
, inode
, map
,
4062 ext4_update_inode_fsync_trans(handle
, inode
, 1);
4063 err
= check_eofblocks_fl(handle
, inode
, map
->m_lblk
,
4067 map
->m_flags
|= EXT4_MAP_MAPPED
;
4068 map
->m_pblk
= newblock
;
4069 if (allocated
> map
->m_len
)
4070 allocated
= map
->m_len
;
4071 map
->m_len
= allocated
;
4074 /* buffered IO case */
4076 * repeat fallocate creation request
4077 * we already have an unwritten extent
4079 if (flags
& EXT4_GET_BLOCKS_UNWRIT_EXT
) {
4080 map
->m_flags
|= EXT4_MAP_UNWRITTEN
;
4084 /* buffered READ or buffered write_begin() lookup */
4085 if ((flags
& EXT4_GET_BLOCKS_CREATE
) == 0) {
4087 * We have blocks reserved already. We
4088 * return allocated blocks so that delalloc
4089 * won't do block reservation for us. But
4090 * the buffer head will be unmapped so that
4091 * a read from the block returns 0s.
4093 map
->m_flags
|= EXT4_MAP_UNWRITTEN
;
4097 /* buffered write, writepage time, convert*/
4098 ret
= ext4_ext_convert_to_initialized(handle
, inode
, map
, ppath
, flags
);
4100 ext4_update_inode_fsync_trans(handle
, inode
, 1);
4107 map
->m_flags
|= EXT4_MAP_NEW
;
4109 * if we allocated more blocks than requested
4110 * we need to make sure we unmap the extra block
4111 * allocated. The actual needed block will get
4112 * unmapped later when we find the buffer_head marked
4115 if (allocated
> map
->m_len
) {
4116 unmap_underlying_metadata_blocks(inode
->i_sb
->s_bdev
,
4117 newblock
+ map
->m_len
,
4118 allocated
- map
->m_len
);
4119 allocated
= map
->m_len
;
4121 map
->m_len
= allocated
;
4124 * If we have done fallocate with the offset that is already
4125 * delayed allocated, we would have block reservation
4126 * and quota reservation done in the delayed write path.
4127 * But fallocate would have already updated quota and block
4128 * count for this offset. So cancel these reservation
4130 if (flags
& EXT4_GET_BLOCKS_DELALLOC_RESERVE
) {
4131 unsigned int reserved_clusters
;
4132 reserved_clusters
= get_reserved_cluster_alloc(inode
,
4133 map
->m_lblk
, map
->m_len
);
4134 if (reserved_clusters
)
4135 ext4_da_update_reserve_space(inode
,
4141 map
->m_flags
|= EXT4_MAP_MAPPED
;
4142 if ((flags
& EXT4_GET_BLOCKS_KEEP_SIZE
) == 0) {
4143 err
= check_eofblocks_fl(handle
, inode
, map
->m_lblk
, path
,
4149 if (allocated
> map
->m_len
)
4150 allocated
= map
->m_len
;
4151 ext4_ext_show_leaf(inode
, path
);
4152 map
->m_pblk
= newblock
;
4153 map
->m_len
= allocated
;
4155 return err
? err
: allocated
;
4159 * get_implied_cluster_alloc - check to see if the requested
4160 * allocation (in the map structure) overlaps with a cluster already
4161 * allocated in an extent.
4162 * @sb The filesystem superblock structure
4163 * @map The requested lblk->pblk mapping
4164 * @ex The extent structure which might contain an implied
4165 * cluster allocation
4167 * This function is called by ext4_ext_map_blocks() after we failed to
4168 * find blocks that were already in the inode's extent tree. Hence,
4169 * we know that the beginning of the requested region cannot overlap
4170 * the extent from the inode's extent tree. There are three cases we
4171 * want to catch. The first is this case:
4173 * |--- cluster # N--|
4174 * |--- extent ---| |---- requested region ---|
4177 * The second case that we need to test for is this one:
4179 * |--------- cluster # N ----------------|
4180 * |--- requested region --| |------- extent ----|
4181 * |=======================|
4183 * The third case is when the requested region lies between two extents
4184 * within the same cluster:
4185 * |------------- cluster # N-------------|
4186 * |----- ex -----| |---- ex_right ----|
4187 * |------ requested region ------|
4188 * |================|
4190 * In each of the above cases, we need to set the map->m_pblk and
4191 * map->m_len so it corresponds to the return the extent labelled as
4192 * "|====|" from cluster #N, since it is already in use for data in
4193 * cluster EXT4_B2C(sbi, map->m_lblk). We will then return 1 to
4194 * signal to ext4_ext_map_blocks() that map->m_pblk should be treated
4195 * as a new "allocated" block region. Otherwise, we will return 0 and
4196 * ext4_ext_map_blocks() will then allocate one or more new clusters
4197 * by calling ext4_mb_new_blocks().
4199 static int get_implied_cluster_alloc(struct super_block
*sb
,
4200 struct ext4_map_blocks
*map
,
4201 struct ext4_extent
*ex
,
4202 struct ext4_ext_path
*path
)
4204 struct ext4_sb_info
*sbi
= EXT4_SB(sb
);
4205 ext4_lblk_t c_offset
= EXT4_LBLK_COFF(sbi
, map
->m_lblk
);
4206 ext4_lblk_t ex_cluster_start
, ex_cluster_end
;
4207 ext4_lblk_t rr_cluster_start
;
4208 ext4_lblk_t ee_block
= le32_to_cpu(ex
->ee_block
);
4209 ext4_fsblk_t ee_start
= ext4_ext_pblock(ex
);
4210 unsigned short ee_len
= ext4_ext_get_actual_len(ex
);
4212 /* The extent passed in that we are trying to match */
4213 ex_cluster_start
= EXT4_B2C(sbi
, ee_block
);
4214 ex_cluster_end
= EXT4_B2C(sbi
, ee_block
+ ee_len
- 1);
4216 /* The requested region passed into ext4_map_blocks() */
4217 rr_cluster_start
= EXT4_B2C(sbi
, map
->m_lblk
);
4219 if ((rr_cluster_start
== ex_cluster_end
) ||
4220 (rr_cluster_start
== ex_cluster_start
)) {
4221 if (rr_cluster_start
== ex_cluster_end
)
4222 ee_start
+= ee_len
- 1;
4223 map
->m_pblk
= EXT4_PBLK_CMASK(sbi
, ee_start
) + c_offset
;
4224 map
->m_len
= min(map
->m_len
,
4225 (unsigned) sbi
->s_cluster_ratio
- c_offset
);
4227 * Check for and handle this case:
4229 * |--------- cluster # N-------------|
4230 * |------- extent ----|
4231 * |--- requested region ---|
4235 if (map
->m_lblk
< ee_block
)
4236 map
->m_len
= min(map
->m_len
, ee_block
- map
->m_lblk
);
4239 * Check for the case where there is already another allocated
4240 * block to the right of 'ex' but before the end of the cluster.
4242 * |------------- cluster # N-------------|
4243 * |----- ex -----| |---- ex_right ----|
4244 * |------ requested region ------|
4245 * |================|
4247 if (map
->m_lblk
> ee_block
) {
4248 ext4_lblk_t next
= ext4_ext_next_allocated_block(path
);
4249 map
->m_len
= min(map
->m_len
, next
- map
->m_lblk
);
4252 trace_ext4_get_implied_cluster_alloc_exit(sb
, map
, 1);
4256 trace_ext4_get_implied_cluster_alloc_exit(sb
, map
, 0);
4262 * Block allocation/map/preallocation routine for extents based files
4265 * Need to be called with
4266 * down_read(&EXT4_I(inode)->i_data_sem) if not allocating file system block
4267 * (ie, create is zero). Otherwise down_write(&EXT4_I(inode)->i_data_sem)
4269 * return > 0, number of of blocks already mapped/allocated
4270 * if create == 0 and these are pre-allocated blocks
4271 * buffer head is unmapped
4272 * otherwise blocks are mapped
4274 * return = 0, if plain look up failed (blocks have not been allocated)
4275 * buffer head is unmapped
4277 * return < 0, error case.
4279 int ext4_ext_map_blocks(handle_t
*handle
, struct inode
*inode
,
4280 struct ext4_map_blocks
*map
, int flags
)
4282 struct ext4_ext_path
*path
= NULL
;
4283 struct ext4_extent newex
, *ex
, *ex2
;
4284 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
4285 ext4_fsblk_t newblock
= 0;
4286 int free_on_err
= 0, err
= 0, depth
, ret
;
4287 unsigned int allocated
= 0, offset
= 0;
4288 unsigned int allocated_clusters
= 0;
4289 struct ext4_allocation_request ar
;
4290 ext4_lblk_t cluster_offset
;
4291 bool map_from_cluster
= false;
4293 ext_debug("blocks %u/%u requested for inode %lu\n",
4294 map
->m_lblk
, map
->m_len
, inode
->i_ino
);
4295 trace_ext4_ext_map_blocks_enter(inode
, map
->m_lblk
, map
->m_len
, flags
);
4297 /* find extent for this block */
4298 path
= ext4_find_extent(inode
, map
->m_lblk
, NULL
, 0);
4300 err
= PTR_ERR(path
);
4305 depth
= ext_depth(inode
);
4308 * consistent leaf must not be empty;
4309 * this situation is possible, though, _during_ tree modification;
4310 * this is why assert can't be put in ext4_find_extent()
4312 if (unlikely(path
[depth
].p_ext
== NULL
&& depth
!= 0)) {
4313 EXT4_ERROR_INODE(inode
, "bad extent address "
4314 "lblock: %lu, depth: %d pblock %lld",
4315 (unsigned long) map
->m_lblk
, depth
,
4316 path
[depth
].p_block
);
4317 err
= -EFSCORRUPTED
;
4321 ex
= path
[depth
].p_ext
;
4323 ext4_lblk_t ee_block
= le32_to_cpu(ex
->ee_block
);
4324 ext4_fsblk_t ee_start
= ext4_ext_pblock(ex
);
4325 unsigned short ee_len
;
4329 * unwritten extents are treated as holes, except that
4330 * we split out initialized portions during a write.
4332 ee_len
= ext4_ext_get_actual_len(ex
);
4334 trace_ext4_ext_show_extent(inode
, ee_block
, ee_start
, ee_len
);
4336 /* if found extent covers block, simply return it */
4337 if (in_range(map
->m_lblk
, ee_block
, ee_len
)) {
4338 newblock
= map
->m_lblk
- ee_block
+ ee_start
;
4339 /* number of remaining blocks in the extent */
4340 allocated
= ee_len
- (map
->m_lblk
- ee_block
);
4341 ext_debug("%u fit into %u:%d -> %llu\n", map
->m_lblk
,
4342 ee_block
, ee_len
, newblock
);
4345 * If the extent is initialized check whether the
4346 * caller wants to convert it to unwritten.
4348 if ((!ext4_ext_is_unwritten(ex
)) &&
4349 (flags
& EXT4_GET_BLOCKS_CONVERT_UNWRITTEN
)) {
4350 allocated
= convert_initialized_extent(
4351 handle
, inode
, map
, &path
,
4354 } else if (!ext4_ext_is_unwritten(ex
))
4357 ret
= ext4_ext_handle_unwritten_extents(
4358 handle
, inode
, map
, &path
, flags
,
4359 allocated
, newblock
);
4369 * requested block isn't allocated yet;
4370 * we couldn't try to create block if create flag is zero
4372 if ((flags
& EXT4_GET_BLOCKS_CREATE
) == 0) {
4373 ext4_lblk_t hole_start
, hole_len
;
4375 hole_start
= map
->m_lblk
;
4376 hole_len
= ext4_ext_determine_hole(inode
, path
, &hole_start
);
4378 * put just found gap into cache to speed up
4379 * subsequent requests
4381 ext4_ext_put_gap_in_cache(inode
, hole_start
, hole_len
);
4383 /* Update hole_len to reflect hole size after map->m_lblk */
4384 if (hole_start
!= map
->m_lblk
)
4385 hole_len
-= map
->m_lblk
- hole_start
;
4387 map
->m_len
= min_t(unsigned int, map
->m_len
, hole_len
);
4393 * Okay, we need to do block allocation.
4395 newex
.ee_block
= cpu_to_le32(map
->m_lblk
);
4396 cluster_offset
= EXT4_LBLK_COFF(sbi
, map
->m_lblk
);
4399 * If we are doing bigalloc, check to see if the extent returned
4400 * by ext4_find_extent() implies a cluster we can use.
4402 if (cluster_offset
&& ex
&&
4403 get_implied_cluster_alloc(inode
->i_sb
, map
, ex
, path
)) {
4404 ar
.len
= allocated
= map
->m_len
;
4405 newblock
= map
->m_pblk
;
4406 map_from_cluster
= true;
4407 goto got_allocated_blocks
;
4410 /* find neighbour allocated blocks */
4411 ar
.lleft
= map
->m_lblk
;
4412 err
= ext4_ext_search_left(inode
, path
, &ar
.lleft
, &ar
.pleft
);
4415 ar
.lright
= map
->m_lblk
;
4417 err
= ext4_ext_search_right(inode
, path
, &ar
.lright
, &ar
.pright
, &ex2
);
4421 /* Check if the extent after searching to the right implies a
4422 * cluster we can use. */
4423 if ((sbi
->s_cluster_ratio
> 1) && ex2
&&
4424 get_implied_cluster_alloc(inode
->i_sb
, map
, ex2
, path
)) {
4425 ar
.len
= allocated
= map
->m_len
;
4426 newblock
= map
->m_pblk
;
4427 map_from_cluster
= true;
4428 goto got_allocated_blocks
;
4432 * See if request is beyond maximum number of blocks we can have in
4433 * a single extent. For an initialized extent this limit is
4434 * EXT_INIT_MAX_LEN and for an unwritten extent this limit is
4435 * EXT_UNWRITTEN_MAX_LEN.
4437 if (map
->m_len
> EXT_INIT_MAX_LEN
&&
4438 !(flags
& EXT4_GET_BLOCKS_UNWRIT_EXT
))
4439 map
->m_len
= EXT_INIT_MAX_LEN
;
4440 else if (map
->m_len
> EXT_UNWRITTEN_MAX_LEN
&&
4441 (flags
& EXT4_GET_BLOCKS_UNWRIT_EXT
))
4442 map
->m_len
= EXT_UNWRITTEN_MAX_LEN
;
4444 /* Check if we can really insert (m_lblk)::(m_lblk + m_len) extent */
4445 newex
.ee_len
= cpu_to_le16(map
->m_len
);
4446 err
= ext4_ext_check_overlap(sbi
, inode
, &newex
, path
);
4448 allocated
= ext4_ext_get_actual_len(&newex
);
4450 allocated
= map
->m_len
;
4452 /* allocate new block */
4454 ar
.goal
= ext4_ext_find_goal(inode
, path
, map
->m_lblk
);
4455 ar
.logical
= map
->m_lblk
;
4457 * We calculate the offset from the beginning of the cluster
4458 * for the logical block number, since when we allocate a
4459 * physical cluster, the physical block should start at the
4460 * same offset from the beginning of the cluster. This is
4461 * needed so that future calls to get_implied_cluster_alloc()
4464 offset
= EXT4_LBLK_COFF(sbi
, map
->m_lblk
);
4465 ar
.len
= EXT4_NUM_B2C(sbi
, offset
+allocated
);
4467 ar
.logical
-= offset
;
4468 if (S_ISREG(inode
->i_mode
))
4469 ar
.flags
= EXT4_MB_HINT_DATA
;
4471 /* disable in-core preallocation for non-regular files */
4473 if (flags
& EXT4_GET_BLOCKS_NO_NORMALIZE
)
4474 ar
.flags
|= EXT4_MB_HINT_NOPREALLOC
;
4475 if (flags
& EXT4_GET_BLOCKS_DELALLOC_RESERVE
)
4476 ar
.flags
|= EXT4_MB_DELALLOC_RESERVED
;
4477 if (flags
& EXT4_GET_BLOCKS_METADATA_NOFAIL
)
4478 ar
.flags
|= EXT4_MB_USE_RESERVED
;
4479 newblock
= ext4_mb_new_blocks(handle
, &ar
, &err
);
4482 ext_debug("allocate new block: goal %llu, found %llu/%u\n",
4483 ar
.goal
, newblock
, allocated
);
4485 allocated_clusters
= ar
.len
;
4486 ar
.len
= EXT4_C2B(sbi
, ar
.len
) - offset
;
4487 if (ar
.len
> allocated
)
4490 got_allocated_blocks
:
4491 /* try to insert new extent into found leaf and return */
4492 ext4_ext_store_pblock(&newex
, newblock
+ offset
);
4493 newex
.ee_len
= cpu_to_le16(ar
.len
);
4494 /* Mark unwritten */
4495 if (flags
& EXT4_GET_BLOCKS_UNWRIT_EXT
){
4496 ext4_ext_mark_unwritten(&newex
);
4497 map
->m_flags
|= EXT4_MAP_UNWRITTEN
;
4501 if ((flags
& EXT4_GET_BLOCKS_KEEP_SIZE
) == 0)
4502 err
= check_eofblocks_fl(handle
, inode
, map
->m_lblk
,
4505 err
= ext4_ext_insert_extent(handle
, inode
, &path
,
4508 if (err
&& free_on_err
) {
4509 int fb_flags
= flags
& EXT4_GET_BLOCKS_DELALLOC_RESERVE
?
4510 EXT4_FREE_BLOCKS_NO_QUOT_UPDATE
: 0;
4511 /* free data blocks we just allocated */
4512 /* not a good idea to call discard here directly,
4513 * but otherwise we'd need to call it every free() */
4514 ext4_discard_preallocations(inode
);
4515 ext4_free_blocks(handle
, inode
, NULL
, newblock
,
4516 EXT4_C2B(sbi
, allocated_clusters
), fb_flags
);
4520 /* previous routine could use block we allocated */
4521 newblock
= ext4_ext_pblock(&newex
);
4522 allocated
= ext4_ext_get_actual_len(&newex
);
4523 if (allocated
> map
->m_len
)
4524 allocated
= map
->m_len
;
4525 map
->m_flags
|= EXT4_MAP_NEW
;
4528 * Update reserved blocks/metadata blocks after successful
4529 * block allocation which had been deferred till now.
4531 if (flags
& EXT4_GET_BLOCKS_DELALLOC_RESERVE
) {
4532 unsigned int reserved_clusters
;
4534 * Check how many clusters we had reserved this allocated range
4536 reserved_clusters
= get_reserved_cluster_alloc(inode
,
4537 map
->m_lblk
, allocated
);
4538 if (!map_from_cluster
) {
4539 BUG_ON(allocated_clusters
< reserved_clusters
);
4540 if (reserved_clusters
< allocated_clusters
) {
4541 struct ext4_inode_info
*ei
= EXT4_I(inode
);
4542 int reservation
= allocated_clusters
-
4545 * It seems we claimed few clusters outside of
4546 * the range of this allocation. We should give
4547 * it back to the reservation pool. This can
4548 * happen in the following case:
4550 * * Suppose s_cluster_ratio is 4 (i.e., each
4551 * cluster has 4 blocks. Thus, the clusters
4552 * are [0-3],[4-7],[8-11]...
4553 * * First comes delayed allocation write for
4554 * logical blocks 10 & 11. Since there were no
4555 * previous delayed allocated blocks in the
4556 * range [8-11], we would reserve 1 cluster
4558 * * Next comes write for logical blocks 3 to 8.
4559 * In this case, we will reserve 2 clusters
4560 * (for [0-3] and [4-7]; and not for [8-11] as
4561 * that range has a delayed allocated blocks.
4562 * Thus total reserved clusters now becomes 3.
4563 * * Now, during the delayed allocation writeout
4564 * time, we will first write blocks [3-8] and
4565 * allocate 3 clusters for writing these
4566 * blocks. Also, we would claim all these
4567 * three clusters above.
4568 * * Now when we come here to writeout the
4569 * blocks [10-11], we would expect to claim
4570 * the reservation of 1 cluster we had made
4571 * (and we would claim it since there are no
4572 * more delayed allocated blocks in the range
4573 * [8-11]. But our reserved cluster count had
4574 * already gone to 0.
4576 * Thus, at the step 4 above when we determine
4577 * that there are still some unwritten delayed
4578 * allocated blocks outside of our current
4579 * block range, we should increment the
4580 * reserved clusters count so that when the
4581 * remaining blocks finally gets written, we
4584 dquot_reserve_block(inode
,
4585 EXT4_C2B(sbi
, reservation
));
4586 spin_lock(&ei
->i_block_reservation_lock
);
4587 ei
->i_reserved_data_blocks
+= reservation
;
4588 spin_unlock(&ei
->i_block_reservation_lock
);
4591 * We will claim quota for all newly allocated blocks.
4592 * We're updating the reserved space *after* the
4593 * correction above so we do not accidentally free
4594 * all the metadata reservation because we might
4595 * actually need it later on.
4597 ext4_da_update_reserve_space(inode
, allocated_clusters
,
4603 * Cache the extent and update transaction to commit on fdatasync only
4604 * when it is _not_ an unwritten extent.
4606 if ((flags
& EXT4_GET_BLOCKS_UNWRIT_EXT
) == 0)
4607 ext4_update_inode_fsync_trans(handle
, inode
, 1);
4609 ext4_update_inode_fsync_trans(handle
, inode
, 0);
4611 if (allocated
> map
->m_len
)
4612 allocated
= map
->m_len
;
4613 ext4_ext_show_leaf(inode
, path
);
4614 map
->m_flags
|= EXT4_MAP_MAPPED
;
4615 map
->m_pblk
= newblock
;
4616 map
->m_len
= allocated
;
4618 ext4_ext_drop_refs(path
);
4621 trace_ext4_ext_map_blocks_exit(inode
, flags
, map
,
4622 err
? err
: allocated
);
4623 return err
? err
: allocated
;
4626 void ext4_ext_truncate(handle_t
*handle
, struct inode
*inode
)
4628 struct super_block
*sb
= inode
->i_sb
;
4629 ext4_lblk_t last_block
;
4633 * TODO: optimization is possible here.
4634 * Probably we need not scan at all,
4635 * because page truncation is enough.
4638 /* we have to know where to truncate from in crash case */
4639 EXT4_I(inode
)->i_disksize
= inode
->i_size
;
4640 ext4_mark_inode_dirty(handle
, inode
);
4642 last_block
= (inode
->i_size
+ sb
->s_blocksize
- 1)
4643 >> EXT4_BLOCK_SIZE_BITS(sb
);
4645 err
= ext4_es_remove_extent(inode
, last_block
,
4646 EXT_MAX_BLOCKS
- last_block
);
4647 if (err
== -ENOMEM
) {
4649 congestion_wait(BLK_RW_ASYNC
, HZ
/50);
4653 ext4_std_error(inode
->i_sb
, err
);
4656 err
= ext4_ext_remove_space(inode
, last_block
, EXT_MAX_BLOCKS
- 1);
4657 ext4_std_error(inode
->i_sb
, err
);
4660 static int ext4_alloc_file_blocks(struct file
*file
, ext4_lblk_t offset
,
4661 ext4_lblk_t len
, loff_t new_size
,
4662 int flags
, int mode
)
4664 struct inode
*inode
= file_inode(file
);
4670 struct ext4_map_blocks map
;
4671 unsigned int credits
;
4674 map
.m_lblk
= offset
;
4677 * Don't normalize the request if it can fit in one extent so
4678 * that it doesn't get unnecessarily split into multiple
4681 if (len
<= EXT_UNWRITTEN_MAX_LEN
)
4682 flags
|= EXT4_GET_BLOCKS_NO_NORMALIZE
;
4685 * credits to insert 1 extent into extent tree
4687 credits
= ext4_chunk_trans_blocks(inode
, len
);
4689 * We can only call ext_depth() on extent based inodes
4691 if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
))
4692 depth
= ext_depth(inode
);
4697 while (ret
>= 0 && len
) {
4699 * Recalculate credits when extent tree depth changes.
4701 if (depth
>= 0 && depth
!= ext_depth(inode
)) {
4702 credits
= ext4_chunk_trans_blocks(inode
, len
);
4703 depth
= ext_depth(inode
);
4706 handle
= ext4_journal_start(inode
, EXT4_HT_MAP_BLOCKS
,
4708 if (IS_ERR(handle
)) {
4709 ret
= PTR_ERR(handle
);
4712 ret
= ext4_map_blocks(handle
, inode
, &map
, flags
);
4714 ext4_debug("inode #%lu: block %u: len %u: "
4715 "ext4_ext_map_blocks returned %d",
4716 inode
->i_ino
, map
.m_lblk
,
4718 ext4_mark_inode_dirty(handle
, inode
);
4719 ret2
= ext4_journal_stop(handle
);
4723 map
.m_len
= len
= len
- ret
;
4724 epos
= (loff_t
)map
.m_lblk
<< inode
->i_blkbits
;
4725 inode
->i_ctime
= ext4_current_time(inode
);
4727 if (epos
> new_size
)
4729 if (ext4_update_inode_size(inode
, epos
) & 0x1)
4730 inode
->i_mtime
= inode
->i_ctime
;
4732 if (epos
> inode
->i_size
)
4733 ext4_set_inode_flag(inode
,
4734 EXT4_INODE_EOFBLOCKS
);
4736 ext4_mark_inode_dirty(handle
, inode
);
4737 ret2
= ext4_journal_stop(handle
);
4741 if (ret
== -ENOSPC
&&
4742 ext4_should_retry_alloc(inode
->i_sb
, &retries
)) {
4747 return ret
> 0 ? ret2
: ret
;
4750 static long ext4_zero_range(struct file
*file
, loff_t offset
,
4751 loff_t len
, int mode
)
4753 struct inode
*inode
= file_inode(file
);
4754 handle_t
*handle
= NULL
;
4755 unsigned int max_blocks
;
4756 loff_t new_size
= 0;
4760 int partial_begin
, partial_end
;
4763 unsigned int blkbits
= inode
->i_blkbits
;
4765 trace_ext4_zero_range(inode
, offset
, len
, mode
);
4767 if (!S_ISREG(inode
->i_mode
))
4770 /* Call ext4_force_commit to flush all data in case of data=journal. */
4771 if (ext4_should_journal_data(inode
)) {
4772 ret
= ext4_force_commit(inode
->i_sb
);
4778 * Round up offset. This is not fallocate, we neet to zero out
4779 * blocks, so convert interior block aligned part of the range to
4780 * unwritten and possibly manually zero out unaligned parts of the
4783 start
= round_up(offset
, 1 << blkbits
);
4784 end
= round_down((offset
+ len
), 1 << blkbits
);
4786 if (start
< offset
|| end
> offset
+ len
)
4788 partial_begin
= offset
& ((1 << blkbits
) - 1);
4789 partial_end
= (offset
+ len
) & ((1 << blkbits
) - 1);
4791 lblk
= start
>> blkbits
;
4792 max_blocks
= (end
>> blkbits
);
4793 if (max_blocks
< lblk
)
4801 * Indirect files do not support unwritten extnets
4803 if (!(ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
))) {
4808 if (!(mode
& FALLOC_FL_KEEP_SIZE
) &&
4809 offset
+ len
> i_size_read(inode
)) {
4810 new_size
= offset
+ len
;
4811 ret
= inode_newsize_ok(inode
, new_size
);
4816 flags
= EXT4_GET_BLOCKS_CREATE_UNWRIT_EXT
;
4817 if (mode
& FALLOC_FL_KEEP_SIZE
)
4818 flags
|= EXT4_GET_BLOCKS_KEEP_SIZE
;
4820 /* Wait all existing dio workers, newcomers will block on i_mutex */
4821 ext4_inode_block_unlocked_dio(inode
);
4822 inode_dio_wait(inode
);
4824 /* Preallocate the range including the unaligned edges */
4825 if (partial_begin
|| partial_end
) {
4826 ret
= ext4_alloc_file_blocks(file
,
4827 round_down(offset
, 1 << blkbits
) >> blkbits
,
4828 (round_up((offset
+ len
), 1 << blkbits
) -
4829 round_down(offset
, 1 << blkbits
)) >> blkbits
,
4830 new_size
, flags
, mode
);
4836 /* Zero range excluding the unaligned edges */
4837 if (max_blocks
> 0) {
4838 flags
|= (EXT4_GET_BLOCKS_CONVERT_UNWRITTEN
|
4842 * Prevent page faults from reinstantiating pages we have
4843 * released from page cache.
4845 down_write(&EXT4_I(inode
)->i_mmap_sem
);
4846 ret
= ext4_update_disksize_before_punch(inode
, offset
, len
);
4848 up_write(&EXT4_I(inode
)->i_mmap_sem
);
4851 /* Now release the pages and zero block aligned part of pages */
4852 truncate_pagecache_range(inode
, start
, end
- 1);
4853 inode
->i_mtime
= inode
->i_ctime
= ext4_current_time(inode
);
4855 ret
= ext4_alloc_file_blocks(file
, lblk
, max_blocks
, new_size
,
4857 up_write(&EXT4_I(inode
)->i_mmap_sem
);
4861 if (!partial_begin
&& !partial_end
)
4865 * In worst case we have to writeout two nonadjacent unwritten
4866 * blocks and update the inode
4868 credits
= (2 * ext4_ext_index_trans_blocks(inode
, 2)) + 1;
4869 if (ext4_should_journal_data(inode
))
4871 handle
= ext4_journal_start(inode
, EXT4_HT_MISC
, credits
);
4872 if (IS_ERR(handle
)) {
4873 ret
= PTR_ERR(handle
);
4874 ext4_std_error(inode
->i_sb
, ret
);
4878 inode
->i_mtime
= inode
->i_ctime
= ext4_current_time(inode
);
4880 ext4_update_inode_size(inode
, new_size
);
4883 * Mark that we allocate beyond EOF so the subsequent truncate
4884 * can proceed even if the new size is the same as i_size.
4886 if ((offset
+ len
) > i_size_read(inode
))
4887 ext4_set_inode_flag(inode
, EXT4_INODE_EOFBLOCKS
);
4889 ext4_mark_inode_dirty(handle
, inode
);
4891 /* Zero out partial block at the edges of the range */
4892 ret
= ext4_zero_partial_blocks(handle
, inode
, offset
, len
);
4894 if (file
->f_flags
& O_SYNC
)
4895 ext4_handle_sync(handle
);
4897 ext4_journal_stop(handle
);
4899 ext4_inode_resume_unlocked_dio(inode
);
4901 inode_unlock(inode
);
4906 * preallocate space for a file. This implements ext4's fallocate file
4907 * operation, which gets called from sys_fallocate system call.
4908 * For block-mapped files, posix_fallocate should fall back to the method
4909 * of writing zeroes to the required new blocks (the same behavior which is
4910 * expected for file systems which do not support fallocate() system call).
4912 long ext4_fallocate(struct file
*file
, int mode
, loff_t offset
, loff_t len
)
4914 struct inode
*inode
= file_inode(file
);
4915 loff_t new_size
= 0;
4916 unsigned int max_blocks
;
4920 unsigned int blkbits
= inode
->i_blkbits
;
4923 * Encrypted inodes can't handle collapse range or insert
4924 * range since we would need to re-encrypt blocks with a
4925 * different IV or XTS tweak (which are based on the logical
4928 * XXX It's not clear why zero range isn't working, but we'll
4929 * leave it disabled for encrypted inodes for now. This is a
4930 * bug we should fix....
4932 if (ext4_encrypted_inode(inode
) &&
4933 (mode
& (FALLOC_FL_COLLAPSE_RANGE
| FALLOC_FL_INSERT_RANGE
|
4934 FALLOC_FL_ZERO_RANGE
)))
4937 /* Return error if mode is not supported */
4938 if (mode
& ~(FALLOC_FL_KEEP_SIZE
| FALLOC_FL_PUNCH_HOLE
|
4939 FALLOC_FL_COLLAPSE_RANGE
| FALLOC_FL_ZERO_RANGE
|
4940 FALLOC_FL_INSERT_RANGE
))
4943 if (mode
& FALLOC_FL_PUNCH_HOLE
)
4944 return ext4_punch_hole(inode
, offset
, len
);
4946 ret
= ext4_convert_inline_data(inode
);
4950 if (mode
& FALLOC_FL_COLLAPSE_RANGE
)
4951 return ext4_collapse_range(inode
, offset
, len
);
4953 if (mode
& FALLOC_FL_INSERT_RANGE
)
4954 return ext4_insert_range(inode
, offset
, len
);
4956 if (mode
& FALLOC_FL_ZERO_RANGE
)
4957 return ext4_zero_range(file
, offset
, len
, mode
);
4959 trace_ext4_fallocate_enter(inode
, offset
, len
, mode
);
4960 lblk
= offset
>> blkbits
;
4962 * We can't just convert len to max_blocks because
4963 * If blocksize = 4096 offset = 3072 and len = 2048
4965 max_blocks
= (EXT4_BLOCK_ALIGN(len
+ offset
, blkbits
) >> blkbits
)
4968 flags
= EXT4_GET_BLOCKS_CREATE_UNWRIT_EXT
;
4969 if (mode
& FALLOC_FL_KEEP_SIZE
)
4970 flags
|= EXT4_GET_BLOCKS_KEEP_SIZE
;
4975 * We only support preallocation for extent-based files only
4977 if (!(ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
))) {
4982 if (!(mode
& FALLOC_FL_KEEP_SIZE
) &&
4983 offset
+ len
> i_size_read(inode
)) {
4984 new_size
= offset
+ len
;
4985 ret
= inode_newsize_ok(inode
, new_size
);
4990 /* Wait all existing dio workers, newcomers will block on i_mutex */
4991 ext4_inode_block_unlocked_dio(inode
);
4992 inode_dio_wait(inode
);
4994 ret
= ext4_alloc_file_blocks(file
, lblk
, max_blocks
, new_size
,
4996 ext4_inode_resume_unlocked_dio(inode
);
5000 if (file
->f_flags
& O_SYNC
&& EXT4_SB(inode
->i_sb
)->s_journal
) {
5001 ret
= jbd2_complete_transaction(EXT4_SB(inode
->i_sb
)->s_journal
,
5002 EXT4_I(inode
)->i_sync_tid
);
5005 inode_unlock(inode
);
5006 trace_ext4_fallocate_exit(inode
, offset
, max_blocks
, ret
);
5011 * This function convert a range of blocks to written extents
5012 * The caller of this function will pass the start offset and the size.
5013 * all unwritten extents within this range will be converted to
5016 * This function is called from the direct IO end io call back
5017 * function, to convert the fallocated extents after IO is completed.
5018 * Returns 0 on success.
5020 int ext4_convert_unwritten_extents(handle_t
*handle
, struct inode
*inode
,
5021 loff_t offset
, ssize_t len
)
5023 unsigned int max_blocks
;
5026 struct ext4_map_blocks map
;
5027 unsigned int credits
, blkbits
= inode
->i_blkbits
;
5029 map
.m_lblk
= offset
>> blkbits
;
5031 * We can't just convert len to max_blocks because
5032 * If blocksize = 4096 offset = 3072 and len = 2048
5034 max_blocks
= ((EXT4_BLOCK_ALIGN(len
+ offset
, blkbits
) >> blkbits
) -
5037 * This is somewhat ugly but the idea is clear: When transaction is
5038 * reserved, everything goes into it. Otherwise we rather start several
5039 * smaller transactions for conversion of each extent separately.
5042 handle
= ext4_journal_start_reserved(handle
,
5043 EXT4_HT_EXT_CONVERT
);
5045 return PTR_ERR(handle
);
5049 * credits to insert 1 extent into extent tree
5051 credits
= ext4_chunk_trans_blocks(inode
, max_blocks
);
5053 while (ret
>= 0 && ret
< max_blocks
) {
5055 map
.m_len
= (max_blocks
-= ret
);
5057 handle
= ext4_journal_start(inode
, EXT4_HT_MAP_BLOCKS
,
5059 if (IS_ERR(handle
)) {
5060 ret
= PTR_ERR(handle
);
5064 ret
= ext4_map_blocks(handle
, inode
, &map
,
5065 EXT4_GET_BLOCKS_IO_CONVERT_EXT
);
5067 ext4_warning(inode
->i_sb
,
5068 "inode #%lu: block %u: len %u: "
5069 "ext4_ext_map_blocks returned %d",
5070 inode
->i_ino
, map
.m_lblk
,
5072 ext4_mark_inode_dirty(handle
, inode
);
5074 ret2
= ext4_journal_stop(handle
);
5075 if (ret
<= 0 || ret2
)
5079 ret2
= ext4_journal_stop(handle
);
5080 return ret
> 0 ? ret2
: ret
;
5084 * If newes is not existing extent (newes->ec_pblk equals zero) find
5085 * delayed extent at start of newes and update newes accordingly and
5086 * return start of the next delayed extent.
5088 * If newes is existing extent (newes->ec_pblk is not equal zero)
5089 * return start of next delayed extent or EXT_MAX_BLOCKS if no delayed
5090 * extent found. Leave newes unmodified.
5092 static int ext4_find_delayed_extent(struct inode
*inode
,
5093 struct extent_status
*newes
)
5095 struct extent_status es
;
5096 ext4_lblk_t block
, next_del
;
5098 if (newes
->es_pblk
== 0) {
5099 ext4_es_find_delayed_extent_range(inode
, newes
->es_lblk
,
5100 newes
->es_lblk
+ newes
->es_len
- 1, &es
);
5103 * No extent in extent-tree contains block @newes->es_pblk,
5104 * then the block may stay in 1)a hole or 2)delayed-extent.
5110 if (es
.es_lblk
> newes
->es_lblk
) {
5112 newes
->es_len
= min(es
.es_lblk
- newes
->es_lblk
,
5117 newes
->es_len
= es
.es_lblk
+ es
.es_len
- newes
->es_lblk
;
5120 block
= newes
->es_lblk
+ newes
->es_len
;
5121 ext4_es_find_delayed_extent_range(inode
, block
, EXT_MAX_BLOCKS
, &es
);
5123 next_del
= EXT_MAX_BLOCKS
;
5125 next_del
= es
.es_lblk
;
5129 /* fiemap flags we can handle specified here */
5130 #define EXT4_FIEMAP_FLAGS (FIEMAP_FLAG_SYNC|FIEMAP_FLAG_XATTR)
5132 static int ext4_xattr_fiemap(struct inode
*inode
,
5133 struct fiemap_extent_info
*fieinfo
)
5137 __u32 flags
= FIEMAP_EXTENT_LAST
;
5138 int blockbits
= inode
->i_sb
->s_blocksize_bits
;
5142 if (ext4_test_inode_state(inode
, EXT4_STATE_XATTR
)) {
5143 struct ext4_iloc iloc
;
5144 int offset
; /* offset of xattr in inode */
5146 error
= ext4_get_inode_loc(inode
, &iloc
);
5149 physical
= (__u64
)iloc
.bh
->b_blocknr
<< blockbits
;
5150 offset
= EXT4_GOOD_OLD_INODE_SIZE
+
5151 EXT4_I(inode
)->i_extra_isize
;
5153 length
= EXT4_SB(inode
->i_sb
)->s_inode_size
- offset
;
5154 flags
|= FIEMAP_EXTENT_DATA_INLINE
;
5156 } else { /* external block */
5157 physical
= (__u64
)EXT4_I(inode
)->i_file_acl
<< blockbits
;
5158 length
= inode
->i_sb
->s_blocksize
;
5162 error
= fiemap_fill_next_extent(fieinfo
, 0, physical
,
5164 return (error
< 0 ? error
: 0);
5167 int ext4_fiemap(struct inode
*inode
, struct fiemap_extent_info
*fieinfo
,
5168 __u64 start
, __u64 len
)
5170 ext4_lblk_t start_blk
;
5173 if (ext4_has_inline_data(inode
)) {
5176 error
= ext4_inline_data_fiemap(inode
, fieinfo
, &has_inline
,
5183 if (fieinfo
->fi_flags
& FIEMAP_FLAG_CACHE
) {
5184 error
= ext4_ext_precache(inode
);
5189 /* fallback to generic here if not in extents fmt */
5190 if (!(ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
)))
5191 return generic_block_fiemap(inode
, fieinfo
, start
, len
,
5194 if (fiemap_check_flags(fieinfo
, EXT4_FIEMAP_FLAGS
))
5197 if (fieinfo
->fi_flags
& FIEMAP_FLAG_XATTR
) {
5198 error
= ext4_xattr_fiemap(inode
, fieinfo
);
5200 ext4_lblk_t len_blks
;
5203 start_blk
= start
>> inode
->i_sb
->s_blocksize_bits
;
5204 last_blk
= (start
+ len
- 1) >> inode
->i_sb
->s_blocksize_bits
;
5205 if (last_blk
>= EXT_MAX_BLOCKS
)
5206 last_blk
= EXT_MAX_BLOCKS
-1;
5207 len_blks
= ((ext4_lblk_t
) last_blk
) - start_blk
+ 1;
5210 * Walk the extent tree gathering extent information
5211 * and pushing extents back to the user.
5213 error
= ext4_fill_fiemap_extents(inode
, start_blk
,
5221 * Function to access the path buffer for marking it dirty.
5222 * It also checks if there are sufficient credits left in the journal handle
5226 ext4_access_path(handle_t
*handle
, struct inode
*inode
,
5227 struct ext4_ext_path
*path
)
5231 if (!ext4_handle_valid(handle
))
5235 * Check if need to extend journal credits
5236 * 3 for leaf, sb, and inode plus 2 (bmap and group
5237 * descriptor) for each block group; assume two block
5240 if (handle
->h_buffer_credits
< 7) {
5241 credits
= ext4_writepage_trans_blocks(inode
);
5242 err
= ext4_ext_truncate_extend_restart(handle
, inode
, credits
);
5243 /* EAGAIN is success */
5244 if (err
&& err
!= -EAGAIN
)
5248 err
= ext4_ext_get_access(handle
, inode
, path
);
5253 * ext4_ext_shift_path_extents:
5254 * Shift the extents of a path structure lying between path[depth].p_ext
5255 * and EXT_LAST_EXTENT(path[depth].p_hdr), by @shift blocks. @SHIFT tells
5256 * if it is right shift or left shift operation.
5259 ext4_ext_shift_path_extents(struct ext4_ext_path
*path
, ext4_lblk_t shift
,
5260 struct inode
*inode
, handle_t
*handle
,
5261 enum SHIFT_DIRECTION SHIFT
)
5264 struct ext4_extent
*ex_start
, *ex_last
;
5266 depth
= path
->p_depth
;
5268 while (depth
>= 0) {
5269 if (depth
== path
->p_depth
) {
5270 ex_start
= path
[depth
].p_ext
;
5272 return -EFSCORRUPTED
;
5274 ex_last
= EXT_LAST_EXTENT(path
[depth
].p_hdr
);
5276 err
= ext4_access_path(handle
, inode
, path
+ depth
);
5280 if (ex_start
== EXT_FIRST_EXTENT(path
[depth
].p_hdr
))
5283 while (ex_start
<= ex_last
) {
5284 if (SHIFT
== SHIFT_LEFT
) {
5285 le32_add_cpu(&ex_start
->ee_block
,
5287 /* Try to merge to the left. */
5289 EXT_FIRST_EXTENT(path
[depth
].p_hdr
))
5291 ext4_ext_try_to_merge_right(inode
,
5292 path
, ex_start
- 1))
5297 le32_add_cpu(&ex_last
->ee_block
, shift
);
5298 ext4_ext_try_to_merge_right(inode
, path
,
5303 err
= ext4_ext_dirty(handle
, inode
, path
+ depth
);
5307 if (--depth
< 0 || !update
)
5311 /* Update index too */
5312 err
= ext4_access_path(handle
, inode
, path
+ depth
);
5316 if (SHIFT
== SHIFT_LEFT
)
5317 le32_add_cpu(&path
[depth
].p_idx
->ei_block
, -shift
);
5319 le32_add_cpu(&path
[depth
].p_idx
->ei_block
, shift
);
5320 err
= ext4_ext_dirty(handle
, inode
, path
+ depth
);
5324 /* we are done if current index is not a starting index */
5325 if (path
[depth
].p_idx
!= EXT_FIRST_INDEX(path
[depth
].p_hdr
))
5336 * ext4_ext_shift_extents:
5337 * All the extents which lies in the range from @start to the last allocated
5338 * block for the @inode are shifted either towards left or right (depending
5339 * upon @SHIFT) by @shift blocks.
5340 * On success, 0 is returned, error otherwise.
5343 ext4_ext_shift_extents(struct inode
*inode
, handle_t
*handle
,
5344 ext4_lblk_t start
, ext4_lblk_t shift
,
5345 enum SHIFT_DIRECTION SHIFT
)
5347 struct ext4_ext_path
*path
;
5349 struct ext4_extent
*extent
;
5350 ext4_lblk_t stop
, *iterator
, ex_start
, ex_end
;
5352 /* Let path point to the last extent */
5353 path
= ext4_find_extent(inode
, EXT_MAX_BLOCKS
- 1, NULL
, 0);
5355 return PTR_ERR(path
);
5357 depth
= path
->p_depth
;
5358 extent
= path
[depth
].p_ext
;
5362 stop
= le32_to_cpu(extent
->ee_block
) +
5363 ext4_ext_get_actual_len(extent
);
5366 * In case of left shift, Don't start shifting extents until we make
5367 * sure the hole is big enough to accommodate the shift.
5369 if (SHIFT
== SHIFT_LEFT
) {
5370 path
= ext4_find_extent(inode
, start
- 1, &path
, 0);
5372 return PTR_ERR(path
);
5373 depth
= path
->p_depth
;
5374 extent
= path
[depth
].p_ext
;
5376 ex_start
= le32_to_cpu(extent
->ee_block
);
5377 ex_end
= le32_to_cpu(extent
->ee_block
) +
5378 ext4_ext_get_actual_len(extent
);
5384 if ((start
== ex_start
&& shift
> ex_start
) ||
5385 (shift
> start
- ex_end
)) {
5386 ext4_ext_drop_refs(path
);
5393 * In case of left shift, iterator points to start and it is increased
5394 * till we reach stop. In case of right shift, iterator points to stop
5395 * and it is decreased till we reach start.
5397 if (SHIFT
== SHIFT_LEFT
)
5402 /* Its safe to start updating extents */
5403 while (start
< stop
) {
5404 path
= ext4_find_extent(inode
, *iterator
, &path
, 0);
5406 return PTR_ERR(path
);
5407 depth
= path
->p_depth
;
5408 extent
= path
[depth
].p_ext
;
5410 EXT4_ERROR_INODE(inode
, "unexpected hole at %lu",
5411 (unsigned long) *iterator
);
5412 return -EFSCORRUPTED
;
5414 if (SHIFT
== SHIFT_LEFT
&& *iterator
>
5415 le32_to_cpu(extent
->ee_block
)) {
5416 /* Hole, move to the next extent */
5417 if (extent
< EXT_LAST_EXTENT(path
[depth
].p_hdr
)) {
5418 path
[depth
].p_ext
++;
5420 *iterator
= ext4_ext_next_allocated_block(path
);
5425 if (SHIFT
== SHIFT_LEFT
) {
5426 extent
= EXT_LAST_EXTENT(path
[depth
].p_hdr
);
5427 *iterator
= le32_to_cpu(extent
->ee_block
) +
5428 ext4_ext_get_actual_len(extent
);
5430 extent
= EXT_FIRST_EXTENT(path
[depth
].p_hdr
);
5431 *iterator
= le32_to_cpu(extent
->ee_block
) > 0 ?
5432 le32_to_cpu(extent
->ee_block
) - 1 : 0;
5433 /* Update path extent in case we need to stop */
5434 while (le32_to_cpu(extent
->ee_block
) < start
)
5436 path
[depth
].p_ext
= extent
;
5438 ret
= ext4_ext_shift_path_extents(path
, shift
, inode
,
5444 ext4_ext_drop_refs(path
);
5450 * ext4_collapse_range:
5451 * This implements the fallocate's collapse range functionality for ext4
5452 * Returns: 0 and non-zero on error.
5454 int ext4_collapse_range(struct inode
*inode
, loff_t offset
, loff_t len
)
5456 struct super_block
*sb
= inode
->i_sb
;
5457 ext4_lblk_t punch_start
, punch_stop
;
5459 unsigned int credits
;
5460 loff_t new_size
, ioffset
;
5464 * We need to test this early because xfstests assumes that a
5465 * collapse range of (0, 1) will return EOPNOTSUPP if the file
5466 * system does not support collapse range.
5468 if (!ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
))
5471 /* Collapse range works only on fs block size aligned offsets. */
5472 if (offset
& (EXT4_CLUSTER_SIZE(sb
) - 1) ||
5473 len
& (EXT4_CLUSTER_SIZE(sb
) - 1))
5476 if (!S_ISREG(inode
->i_mode
))
5479 trace_ext4_collapse_range(inode
, offset
, len
);
5481 punch_start
= offset
>> EXT4_BLOCK_SIZE_BITS(sb
);
5482 punch_stop
= (offset
+ len
) >> EXT4_BLOCK_SIZE_BITS(sb
);
5484 /* Call ext4_force_commit to flush all data in case of data=journal. */
5485 if (ext4_should_journal_data(inode
)) {
5486 ret
= ext4_force_commit(inode
->i_sb
);
5493 * There is no need to overlap collapse range with EOF, in which case
5494 * it is effectively a truncate operation
5496 if (offset
+ len
>= i_size_read(inode
)) {
5501 /* Currently just for extent based files */
5502 if (!ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
)) {
5507 /* Wait for existing dio to complete */
5508 ext4_inode_block_unlocked_dio(inode
);
5509 inode_dio_wait(inode
);
5512 * Prevent page faults from reinstantiating pages we have released from
5515 down_write(&EXT4_I(inode
)->i_mmap_sem
);
5517 * Need to round down offset to be aligned with page size boundary
5518 * for page size > block size.
5520 ioffset
= round_down(offset
, PAGE_SIZE
);
5522 * Write tail of the last page before removed range since it will get
5523 * removed from the page cache below.
5525 ret
= filemap_write_and_wait_range(inode
->i_mapping
, ioffset
, offset
);
5529 * Write data that will be shifted to preserve them when discarding
5530 * page cache below. We are also protected from pages becoming dirty
5533 ret
= filemap_write_and_wait_range(inode
->i_mapping
, offset
+ len
,
5537 truncate_pagecache(inode
, ioffset
);
5539 credits
= ext4_writepage_trans_blocks(inode
);
5540 handle
= ext4_journal_start(inode
, EXT4_HT_TRUNCATE
, credits
);
5541 if (IS_ERR(handle
)) {
5542 ret
= PTR_ERR(handle
);
5546 down_write(&EXT4_I(inode
)->i_data_sem
);
5547 ext4_discard_preallocations(inode
);
5549 ret
= ext4_es_remove_extent(inode
, punch_start
,
5550 EXT_MAX_BLOCKS
- punch_start
);
5552 up_write(&EXT4_I(inode
)->i_data_sem
);
5556 ret
= ext4_ext_remove_space(inode
, punch_start
, punch_stop
- 1);
5558 up_write(&EXT4_I(inode
)->i_data_sem
);
5561 ext4_discard_preallocations(inode
);
5563 ret
= ext4_ext_shift_extents(inode
, handle
, punch_stop
,
5564 punch_stop
- punch_start
, SHIFT_LEFT
);
5566 up_write(&EXT4_I(inode
)->i_data_sem
);
5570 new_size
= i_size_read(inode
) - len
;
5571 i_size_write(inode
, new_size
);
5572 EXT4_I(inode
)->i_disksize
= new_size
;
5574 up_write(&EXT4_I(inode
)->i_data_sem
);
5576 ext4_handle_sync(handle
);
5577 inode
->i_mtime
= inode
->i_ctime
= ext4_current_time(inode
);
5578 ext4_mark_inode_dirty(handle
, inode
);
5581 ext4_journal_stop(handle
);
5583 up_write(&EXT4_I(inode
)->i_mmap_sem
);
5584 ext4_inode_resume_unlocked_dio(inode
);
5586 inode_unlock(inode
);
5591 * ext4_insert_range:
5592 * This function implements the FALLOC_FL_INSERT_RANGE flag of fallocate.
5593 * The data blocks starting from @offset to the EOF are shifted by @len
5594 * towards right to create a hole in the @inode. Inode size is increased
5596 * Returns 0 on success, error otherwise.
5598 int ext4_insert_range(struct inode
*inode
, loff_t offset
, loff_t len
)
5600 struct super_block
*sb
= inode
->i_sb
;
5602 struct ext4_ext_path
*path
;
5603 struct ext4_extent
*extent
;
5604 ext4_lblk_t offset_lblk
, len_lblk
, ee_start_lblk
= 0;
5605 unsigned int credits
, ee_len
;
5606 int ret
= 0, depth
, split_flag
= 0;
5610 * We need to test this early because xfstests assumes that an
5611 * insert range of (0, 1) will return EOPNOTSUPP if the file
5612 * system does not support insert range.
5614 if (!ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
))
5617 /* Insert range works only on fs block size aligned offsets. */
5618 if (offset
& (EXT4_CLUSTER_SIZE(sb
) - 1) ||
5619 len
& (EXT4_CLUSTER_SIZE(sb
) - 1))
5622 if (!S_ISREG(inode
->i_mode
))
5625 trace_ext4_insert_range(inode
, offset
, len
);
5627 offset_lblk
= offset
>> EXT4_BLOCK_SIZE_BITS(sb
);
5628 len_lblk
= len
>> EXT4_BLOCK_SIZE_BITS(sb
);
5630 /* Call ext4_force_commit to flush all data in case of data=journal */
5631 if (ext4_should_journal_data(inode
)) {
5632 ret
= ext4_force_commit(inode
->i_sb
);
5638 /* Currently just for extent based files */
5639 if (!ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
)) {
5644 /* Check for wrap through zero */
5645 if (inode
->i_size
+ len
> inode
->i_sb
->s_maxbytes
) {
5650 /* Offset should be less than i_size */
5651 if (offset
>= i_size_read(inode
)) {
5656 /* Wait for existing dio to complete */
5657 ext4_inode_block_unlocked_dio(inode
);
5658 inode_dio_wait(inode
);
5661 * Prevent page faults from reinstantiating pages we have released from
5664 down_write(&EXT4_I(inode
)->i_mmap_sem
);
5666 * Need to round down to align start offset to page size boundary
5667 * for page size > block size.
5669 ioffset
= round_down(offset
, PAGE_SIZE
);
5670 /* Write out all dirty pages */
5671 ret
= filemap_write_and_wait_range(inode
->i_mapping
, ioffset
,
5675 truncate_pagecache(inode
, ioffset
);
5677 credits
= ext4_writepage_trans_blocks(inode
);
5678 handle
= ext4_journal_start(inode
, EXT4_HT_TRUNCATE
, credits
);
5679 if (IS_ERR(handle
)) {
5680 ret
= PTR_ERR(handle
);
5684 /* Expand file to avoid data loss if there is error while shifting */
5685 inode
->i_size
+= len
;
5686 EXT4_I(inode
)->i_disksize
+= len
;
5687 inode
->i_mtime
= inode
->i_ctime
= ext4_current_time(inode
);
5688 ret
= ext4_mark_inode_dirty(handle
, inode
);
5692 down_write(&EXT4_I(inode
)->i_data_sem
);
5693 ext4_discard_preallocations(inode
);
5695 path
= ext4_find_extent(inode
, offset_lblk
, NULL
, 0);
5697 up_write(&EXT4_I(inode
)->i_data_sem
);
5701 depth
= ext_depth(inode
);
5702 extent
= path
[depth
].p_ext
;
5704 ee_start_lblk
= le32_to_cpu(extent
->ee_block
);
5705 ee_len
= ext4_ext_get_actual_len(extent
);
5708 * If offset_lblk is not the starting block of extent, split
5709 * the extent @offset_lblk
5711 if ((offset_lblk
> ee_start_lblk
) &&
5712 (offset_lblk
< (ee_start_lblk
+ ee_len
))) {
5713 if (ext4_ext_is_unwritten(extent
))
5714 split_flag
= EXT4_EXT_MARK_UNWRIT1
|
5715 EXT4_EXT_MARK_UNWRIT2
;
5716 ret
= ext4_split_extent_at(handle
, inode
, &path
,
5717 offset_lblk
, split_flag
,
5719 EXT4_GET_BLOCKS_PRE_IO
|
5720 EXT4_GET_BLOCKS_METADATA_NOFAIL
);
5723 ext4_ext_drop_refs(path
);
5726 up_write(&EXT4_I(inode
)->i_data_sem
);
5731 ret
= ext4_es_remove_extent(inode
, offset_lblk
,
5732 EXT_MAX_BLOCKS
- offset_lblk
);
5734 up_write(&EXT4_I(inode
)->i_data_sem
);
5739 * if offset_lblk lies in a hole which is at start of file, use
5740 * ee_start_lblk to shift extents
5742 ret
= ext4_ext_shift_extents(inode
, handle
,
5743 ee_start_lblk
> offset_lblk
? ee_start_lblk
: offset_lblk
,
5744 len_lblk
, SHIFT_RIGHT
);
5746 up_write(&EXT4_I(inode
)->i_data_sem
);
5748 ext4_handle_sync(handle
);
5751 ext4_journal_stop(handle
);
5753 up_write(&EXT4_I(inode
)->i_mmap_sem
);
5754 ext4_inode_resume_unlocked_dio(inode
);
5756 inode_unlock(inode
);
5761 * ext4_swap_extents - Swap extents between two inodes
5763 * @inode1: First inode
5764 * @inode2: Second inode
5765 * @lblk1: Start block for first inode
5766 * @lblk2: Start block for second inode
5767 * @count: Number of blocks to swap
5768 * @mark_unwritten: Mark second inode's extents as unwritten after swap
5769 * @erp: Pointer to save error value
5771 * This helper routine does exactly what is promise "swap extents". All other
5772 * stuff such as page-cache locking consistency, bh mapping consistency or
5773 * extent's data copying must be performed by caller.
5775 * i_mutex is held for both inodes
5776 * i_data_sem is locked for write for both inodes
5778 * All pages from requested range are locked for both inodes
5781 ext4_swap_extents(handle_t
*handle
, struct inode
*inode1
,
5782 struct inode
*inode2
, ext4_lblk_t lblk1
, ext4_lblk_t lblk2
,
5783 ext4_lblk_t count
, int unwritten
, int *erp
)
5785 struct ext4_ext_path
*path1
= NULL
;
5786 struct ext4_ext_path
*path2
= NULL
;
5787 int replaced_count
= 0;
5789 BUG_ON(!rwsem_is_locked(&EXT4_I(inode1
)->i_data_sem
));
5790 BUG_ON(!rwsem_is_locked(&EXT4_I(inode2
)->i_data_sem
));
5791 BUG_ON(!inode_is_locked(inode1
));
5792 BUG_ON(!inode_is_locked(inode2
));
5794 *erp
= ext4_es_remove_extent(inode1
, lblk1
, count
);
5797 *erp
= ext4_es_remove_extent(inode2
, lblk2
, count
);
5802 struct ext4_extent
*ex1
, *ex2
, tmp_ex
;
5803 ext4_lblk_t e1_blk
, e2_blk
;
5804 int e1_len
, e2_len
, len
;
5807 path1
= ext4_find_extent(inode1
, lblk1
, NULL
, EXT4_EX_NOCACHE
);
5808 if (IS_ERR(path1
)) {
5809 *erp
= PTR_ERR(path1
);
5815 path2
= ext4_find_extent(inode2
, lblk2
, NULL
, EXT4_EX_NOCACHE
);
5816 if (IS_ERR(path2
)) {
5817 *erp
= PTR_ERR(path2
);
5821 ex1
= path1
[path1
->p_depth
].p_ext
;
5822 ex2
= path2
[path2
->p_depth
].p_ext
;
5823 /* Do we have somthing to swap ? */
5824 if (unlikely(!ex2
|| !ex1
))
5827 e1_blk
= le32_to_cpu(ex1
->ee_block
);
5828 e2_blk
= le32_to_cpu(ex2
->ee_block
);
5829 e1_len
= ext4_ext_get_actual_len(ex1
);
5830 e2_len
= ext4_ext_get_actual_len(ex2
);
5833 if (!in_range(lblk1
, e1_blk
, e1_len
) ||
5834 !in_range(lblk2
, e2_blk
, e2_len
)) {
5835 ext4_lblk_t next1
, next2
;
5837 /* if hole after extent, then go to next extent */
5838 next1
= ext4_ext_next_allocated_block(path1
);
5839 next2
= ext4_ext_next_allocated_block(path2
);
5840 /* If hole before extent, then shift to that extent */
5845 /* Do we have something to swap */
5846 if (next1
== EXT_MAX_BLOCKS
|| next2
== EXT_MAX_BLOCKS
)
5848 /* Move to the rightest boundary */
5849 len
= next1
- lblk1
;
5850 if (len
< next2
- lblk2
)
5851 len
= next2
- lblk2
;
5860 /* Prepare left boundary */
5861 if (e1_blk
< lblk1
) {
5863 *erp
= ext4_force_split_extent_at(handle
, inode1
,
5868 if (e2_blk
< lblk2
) {
5870 *erp
= ext4_force_split_extent_at(handle
, inode2
,
5875 /* ext4_split_extent_at() may result in leaf extent split,
5876 * path must to be revalidated. */
5880 /* Prepare right boundary */
5882 if (len
> e1_blk
+ e1_len
- lblk1
)
5883 len
= e1_blk
+ e1_len
- lblk1
;
5884 if (len
> e2_blk
+ e2_len
- lblk2
)
5885 len
= e2_blk
+ e2_len
- lblk2
;
5887 if (len
!= e1_len
) {
5889 *erp
= ext4_force_split_extent_at(handle
, inode1
,
5890 &path1
, lblk1
+ len
, 0);
5894 if (len
!= e2_len
) {
5896 *erp
= ext4_force_split_extent_at(handle
, inode2
,
5897 &path2
, lblk2
+ len
, 0);
5901 /* ext4_split_extent_at() may result in leaf extent split,
5902 * path must to be revalidated. */
5906 BUG_ON(e2_len
!= e1_len
);
5907 *erp
= ext4_ext_get_access(handle
, inode1
, path1
+ path1
->p_depth
);
5910 *erp
= ext4_ext_get_access(handle
, inode2
, path2
+ path2
->p_depth
);
5914 /* Both extents are fully inside boundaries. Swap it now */
5916 ext4_ext_store_pblock(ex1
, ext4_ext_pblock(ex2
));
5917 ext4_ext_store_pblock(ex2
, ext4_ext_pblock(&tmp_ex
));
5918 ex1
->ee_len
= cpu_to_le16(e2_len
);
5919 ex2
->ee_len
= cpu_to_le16(e1_len
);
5921 ext4_ext_mark_unwritten(ex2
);
5922 if (ext4_ext_is_unwritten(&tmp_ex
))
5923 ext4_ext_mark_unwritten(ex1
);
5925 ext4_ext_try_to_merge(handle
, inode2
, path2
, ex2
);
5926 ext4_ext_try_to_merge(handle
, inode1
, path1
, ex1
);
5927 *erp
= ext4_ext_dirty(handle
, inode2
, path2
+
5931 *erp
= ext4_ext_dirty(handle
, inode1
, path1
+
5934 * Looks scarry ah..? second inode already points to new blocks,
5935 * and it was successfully dirtied. But luckily error may happen
5936 * only due to journal error, so full transaction will be
5943 replaced_count
+= len
;
5947 ext4_ext_drop_refs(path1
);
5949 ext4_ext_drop_refs(path2
);
5951 path1
= path2
= NULL
;
5953 return replaced_count
;