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 <linux/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
);
388 * - overflow/wrap-around
390 if (lblock
+ len
<= lblock
)
392 return ext4_data_block_valid(EXT4_SB(inode
->i_sb
), block
, len
);
395 static int ext4_valid_extent_idx(struct inode
*inode
,
396 struct ext4_extent_idx
*ext_idx
)
398 ext4_fsblk_t block
= ext4_idx_pblock(ext_idx
);
400 return ext4_data_block_valid(EXT4_SB(inode
->i_sb
), block
, 1);
403 static int ext4_valid_extent_entries(struct inode
*inode
,
404 struct ext4_extent_header
*eh
,
407 unsigned short entries
;
408 if (eh
->eh_entries
== 0)
411 entries
= le16_to_cpu(eh
->eh_entries
);
415 struct ext4_extent
*ext
= EXT_FIRST_EXTENT(eh
);
416 struct ext4_super_block
*es
= EXT4_SB(inode
->i_sb
)->s_es
;
417 ext4_fsblk_t pblock
= 0;
418 ext4_lblk_t lblock
= 0;
419 ext4_lblk_t prev
= 0;
422 if (!ext4_valid_extent(inode
, ext
))
425 /* Check for overlapping extents */
426 lblock
= le32_to_cpu(ext
->ee_block
);
427 len
= ext4_ext_get_actual_len(ext
);
428 if ((lblock
<= prev
) && prev
) {
429 pblock
= ext4_ext_pblock(ext
);
430 es
->s_last_error_block
= cpu_to_le64(pblock
);
435 prev
= lblock
+ len
- 1;
438 struct ext4_extent_idx
*ext_idx
= EXT_FIRST_INDEX(eh
);
440 if (!ext4_valid_extent_idx(inode
, ext_idx
))
449 static int __ext4_ext_check(const char *function
, unsigned int line
,
450 struct inode
*inode
, struct ext4_extent_header
*eh
,
451 int depth
, ext4_fsblk_t pblk
)
453 const char *error_msg
;
454 int max
= 0, err
= -EFSCORRUPTED
;
456 if (unlikely(eh
->eh_magic
!= EXT4_EXT_MAGIC
)) {
457 error_msg
= "invalid magic";
460 if (unlikely(le16_to_cpu(eh
->eh_depth
) != depth
)) {
461 error_msg
= "unexpected eh_depth";
464 if (unlikely(eh
->eh_max
== 0)) {
465 error_msg
= "invalid eh_max";
468 max
= ext4_ext_max_entries(inode
, depth
);
469 if (unlikely(le16_to_cpu(eh
->eh_max
) > max
)) {
470 error_msg
= "too large eh_max";
473 if (unlikely(le16_to_cpu(eh
->eh_entries
) > le16_to_cpu(eh
->eh_max
))) {
474 error_msg
= "invalid eh_entries";
477 if (!ext4_valid_extent_entries(inode
, eh
, depth
)) {
478 error_msg
= "invalid extent entries";
481 if (unlikely(depth
> 32)) {
482 error_msg
= "too large eh_depth";
485 /* Verify checksum on non-root extent tree nodes */
486 if (ext_depth(inode
) != depth
&&
487 !ext4_extent_block_csum_verify(inode
, eh
)) {
488 error_msg
= "extent tree corrupted";
495 ext4_error_inode(inode
, function
, line
, 0,
496 "pblk %llu bad header/extent: %s - magic %x, "
497 "entries %u, max %u(%u), depth %u(%u)",
498 (unsigned long long) pblk
, error_msg
,
499 le16_to_cpu(eh
->eh_magic
),
500 le16_to_cpu(eh
->eh_entries
), le16_to_cpu(eh
->eh_max
),
501 max
, le16_to_cpu(eh
->eh_depth
), depth
);
505 #define ext4_ext_check(inode, eh, depth, pblk) \
506 __ext4_ext_check(__func__, __LINE__, (inode), (eh), (depth), (pblk))
508 int ext4_ext_check_inode(struct inode
*inode
)
510 return ext4_ext_check(inode
, ext_inode_hdr(inode
), ext_depth(inode
), 0);
513 static struct buffer_head
*
514 __read_extent_tree_block(const char *function
, unsigned int line
,
515 struct inode
*inode
, ext4_fsblk_t pblk
, int depth
,
518 struct buffer_head
*bh
;
521 bh
= sb_getblk_gfp(inode
->i_sb
, pblk
, __GFP_MOVABLE
| GFP_NOFS
);
523 return ERR_PTR(-ENOMEM
);
525 if (!bh_uptodate_or_lock(bh
)) {
526 trace_ext4_ext_load_extent(inode
, pblk
, _RET_IP_
);
527 err
= bh_submit_read(bh
);
531 if (buffer_verified(bh
) && !(flags
& EXT4_EX_FORCE_CACHE
))
533 err
= __ext4_ext_check(function
, line
, inode
,
534 ext_block_hdr(bh
), depth
, pblk
);
537 set_buffer_verified(bh
);
539 * If this is a leaf block, cache all of its entries
541 if (!(flags
& EXT4_EX_NOCACHE
) && depth
== 0) {
542 struct ext4_extent_header
*eh
= ext_block_hdr(bh
);
543 struct ext4_extent
*ex
= EXT_FIRST_EXTENT(eh
);
544 ext4_lblk_t prev
= 0;
547 for (i
= le16_to_cpu(eh
->eh_entries
); i
> 0; i
--, ex
++) {
548 unsigned int status
= EXTENT_STATUS_WRITTEN
;
549 ext4_lblk_t lblk
= le32_to_cpu(ex
->ee_block
);
550 int len
= ext4_ext_get_actual_len(ex
);
552 if (prev
&& (prev
!= lblk
))
553 ext4_es_cache_extent(inode
, prev
,
557 if (ext4_ext_is_unwritten(ex
))
558 status
= EXTENT_STATUS_UNWRITTEN
;
559 ext4_es_cache_extent(inode
, lblk
, len
,
560 ext4_ext_pblock(ex
), status
);
571 #define read_extent_tree_block(inode, pblk, depth, flags) \
572 __read_extent_tree_block(__func__, __LINE__, (inode), (pblk), \
576 * This function is called to cache a file's extent information in the
579 int ext4_ext_precache(struct inode
*inode
)
581 struct ext4_inode_info
*ei
= EXT4_I(inode
);
582 struct ext4_ext_path
*path
= NULL
;
583 struct buffer_head
*bh
;
584 int i
= 0, depth
, ret
= 0;
586 if (!ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
))
587 return 0; /* not an extent-mapped inode */
589 down_read(&ei
->i_data_sem
);
590 depth
= ext_depth(inode
);
592 path
= kzalloc(sizeof(struct ext4_ext_path
) * (depth
+ 1),
595 up_read(&ei
->i_data_sem
);
599 /* Don't cache anything if there are no external extent blocks */
602 path
[0].p_hdr
= ext_inode_hdr(inode
);
603 ret
= ext4_ext_check(inode
, path
[0].p_hdr
, depth
, 0);
606 path
[0].p_idx
= EXT_FIRST_INDEX(path
[0].p_hdr
);
609 * If this is a leaf block or we've reached the end of
610 * the index block, go up
613 path
[i
].p_idx
> EXT_LAST_INDEX(path
[i
].p_hdr
)) {
614 brelse(path
[i
].p_bh
);
619 bh
= read_extent_tree_block(inode
,
620 ext4_idx_pblock(path
[i
].p_idx
++),
622 EXT4_EX_FORCE_CACHE
);
629 path
[i
].p_hdr
= ext_block_hdr(bh
);
630 path
[i
].p_idx
= EXT_FIRST_INDEX(path
[i
].p_hdr
);
632 ext4_set_inode_state(inode
, EXT4_STATE_EXT_PRECACHED
);
634 up_read(&ei
->i_data_sem
);
635 ext4_ext_drop_refs(path
);
641 static void ext4_ext_show_path(struct inode
*inode
, struct ext4_ext_path
*path
)
643 int k
, l
= path
->p_depth
;
646 for (k
= 0; k
<= l
; k
++, path
++) {
648 ext_debug(" %d->%llu", le32_to_cpu(path
->p_idx
->ei_block
),
649 ext4_idx_pblock(path
->p_idx
));
650 } else if (path
->p_ext
) {
651 ext_debug(" %d:[%d]%d:%llu ",
652 le32_to_cpu(path
->p_ext
->ee_block
),
653 ext4_ext_is_unwritten(path
->p_ext
),
654 ext4_ext_get_actual_len(path
->p_ext
),
655 ext4_ext_pblock(path
->p_ext
));
662 static void ext4_ext_show_leaf(struct inode
*inode
, struct ext4_ext_path
*path
)
664 int depth
= ext_depth(inode
);
665 struct ext4_extent_header
*eh
;
666 struct ext4_extent
*ex
;
672 eh
= path
[depth
].p_hdr
;
673 ex
= EXT_FIRST_EXTENT(eh
);
675 ext_debug("Displaying leaf extents for inode %lu\n", inode
->i_ino
);
677 for (i
= 0; i
< le16_to_cpu(eh
->eh_entries
); i
++, ex
++) {
678 ext_debug("%d:[%d]%d:%llu ", le32_to_cpu(ex
->ee_block
),
679 ext4_ext_is_unwritten(ex
),
680 ext4_ext_get_actual_len(ex
), ext4_ext_pblock(ex
));
685 static void ext4_ext_show_move(struct inode
*inode
, struct ext4_ext_path
*path
,
686 ext4_fsblk_t newblock
, int level
)
688 int depth
= ext_depth(inode
);
689 struct ext4_extent
*ex
;
691 if (depth
!= level
) {
692 struct ext4_extent_idx
*idx
;
693 idx
= path
[level
].p_idx
;
694 while (idx
<= EXT_MAX_INDEX(path
[level
].p_hdr
)) {
695 ext_debug("%d: move %d:%llu in new index %llu\n", level
,
696 le32_to_cpu(idx
->ei_block
),
697 ext4_idx_pblock(idx
),
705 ex
= path
[depth
].p_ext
;
706 while (ex
<= EXT_MAX_EXTENT(path
[depth
].p_hdr
)) {
707 ext_debug("move %d:%llu:[%d]%d in new leaf %llu\n",
708 le32_to_cpu(ex
->ee_block
),
710 ext4_ext_is_unwritten(ex
),
711 ext4_ext_get_actual_len(ex
),
718 #define ext4_ext_show_path(inode, path)
719 #define ext4_ext_show_leaf(inode, path)
720 #define ext4_ext_show_move(inode, path, newblock, level)
723 void ext4_ext_drop_refs(struct ext4_ext_path
*path
)
729 depth
= path
->p_depth
;
730 for (i
= 0; i
<= depth
; i
++, path
++)
738 * ext4_ext_binsearch_idx:
739 * binary search for the closest index of the given block
740 * the header must be checked before calling this
743 ext4_ext_binsearch_idx(struct inode
*inode
,
744 struct ext4_ext_path
*path
, ext4_lblk_t block
)
746 struct ext4_extent_header
*eh
= path
->p_hdr
;
747 struct ext4_extent_idx
*r
, *l
, *m
;
750 ext_debug("binsearch for %u(idx): ", block
);
752 l
= EXT_FIRST_INDEX(eh
) + 1;
753 r
= EXT_LAST_INDEX(eh
);
756 if (block
< le32_to_cpu(m
->ei_block
))
760 ext_debug("%p(%u):%p(%u):%p(%u) ", l
, le32_to_cpu(l
->ei_block
),
761 m
, le32_to_cpu(m
->ei_block
),
762 r
, le32_to_cpu(r
->ei_block
));
766 ext_debug(" -> %u->%lld ", le32_to_cpu(path
->p_idx
->ei_block
),
767 ext4_idx_pblock(path
->p_idx
));
769 #ifdef CHECK_BINSEARCH
771 struct ext4_extent_idx
*chix
, *ix
;
774 chix
= ix
= EXT_FIRST_INDEX(eh
);
775 for (k
= 0; k
< le16_to_cpu(eh
->eh_entries
); k
++, ix
++) {
777 le32_to_cpu(ix
->ei_block
) <= le32_to_cpu(ix
[-1].ei_block
)) {
778 printk(KERN_DEBUG
"k=%d, ix=0x%p, "
780 ix
, EXT_FIRST_INDEX(eh
));
781 printk(KERN_DEBUG
"%u <= %u\n",
782 le32_to_cpu(ix
->ei_block
),
783 le32_to_cpu(ix
[-1].ei_block
));
785 BUG_ON(k
&& le32_to_cpu(ix
->ei_block
)
786 <= le32_to_cpu(ix
[-1].ei_block
));
787 if (block
< le32_to_cpu(ix
->ei_block
))
791 BUG_ON(chix
!= path
->p_idx
);
798 * ext4_ext_binsearch:
799 * binary search for closest extent of the given block
800 * the header must be checked before calling this
803 ext4_ext_binsearch(struct inode
*inode
,
804 struct ext4_ext_path
*path
, ext4_lblk_t block
)
806 struct ext4_extent_header
*eh
= path
->p_hdr
;
807 struct ext4_extent
*r
, *l
, *m
;
809 if (eh
->eh_entries
== 0) {
811 * this leaf is empty:
812 * we get such a leaf in split/add case
817 ext_debug("binsearch for %u: ", block
);
819 l
= EXT_FIRST_EXTENT(eh
) + 1;
820 r
= EXT_LAST_EXTENT(eh
);
824 if (block
< le32_to_cpu(m
->ee_block
))
828 ext_debug("%p(%u):%p(%u):%p(%u) ", l
, le32_to_cpu(l
->ee_block
),
829 m
, le32_to_cpu(m
->ee_block
),
830 r
, le32_to_cpu(r
->ee_block
));
834 ext_debug(" -> %d:%llu:[%d]%d ",
835 le32_to_cpu(path
->p_ext
->ee_block
),
836 ext4_ext_pblock(path
->p_ext
),
837 ext4_ext_is_unwritten(path
->p_ext
),
838 ext4_ext_get_actual_len(path
->p_ext
));
840 #ifdef CHECK_BINSEARCH
842 struct ext4_extent
*chex
, *ex
;
845 chex
= ex
= EXT_FIRST_EXTENT(eh
);
846 for (k
= 0; k
< le16_to_cpu(eh
->eh_entries
); k
++, ex
++) {
847 BUG_ON(k
&& le32_to_cpu(ex
->ee_block
)
848 <= le32_to_cpu(ex
[-1].ee_block
));
849 if (block
< le32_to_cpu(ex
->ee_block
))
853 BUG_ON(chex
!= path
->p_ext
);
859 int ext4_ext_tree_init(handle_t
*handle
, struct inode
*inode
)
861 struct ext4_extent_header
*eh
;
863 eh
= ext_inode_hdr(inode
);
866 eh
->eh_magic
= EXT4_EXT_MAGIC
;
867 eh
->eh_max
= cpu_to_le16(ext4_ext_space_root(inode
, 0));
868 ext4_mark_inode_dirty(handle
, inode
);
872 struct ext4_ext_path
*
873 ext4_find_extent(struct inode
*inode
, ext4_lblk_t block
,
874 struct ext4_ext_path
**orig_path
, int flags
)
876 struct ext4_extent_header
*eh
;
877 struct buffer_head
*bh
;
878 struct ext4_ext_path
*path
= orig_path
? *orig_path
: NULL
;
879 short int depth
, i
, ppos
= 0;
882 eh
= ext_inode_hdr(inode
);
883 depth
= ext_depth(inode
);
884 if (depth
< 0 || depth
> EXT4_MAX_EXTENT_DEPTH
) {
885 EXT4_ERROR_INODE(inode
, "inode has invalid extent depth: %d",
892 ext4_ext_drop_refs(path
);
893 if (depth
> path
[0].p_maxdepth
) {
895 *orig_path
= path
= NULL
;
899 /* account possible depth increase */
900 path
= kzalloc(sizeof(struct ext4_ext_path
) * (depth
+ 2),
903 return ERR_PTR(-ENOMEM
);
904 path
[0].p_maxdepth
= depth
+ 1;
910 /* walk through the tree */
912 ext_debug("depth %d: num %d, max %d\n",
913 ppos
, le16_to_cpu(eh
->eh_entries
), le16_to_cpu(eh
->eh_max
));
915 ext4_ext_binsearch_idx(inode
, path
+ ppos
, block
);
916 path
[ppos
].p_block
= ext4_idx_pblock(path
[ppos
].p_idx
);
917 path
[ppos
].p_depth
= i
;
918 path
[ppos
].p_ext
= NULL
;
920 bh
= read_extent_tree_block(inode
, path
[ppos
].p_block
, --i
,
927 eh
= ext_block_hdr(bh
);
929 path
[ppos
].p_bh
= bh
;
930 path
[ppos
].p_hdr
= eh
;
933 path
[ppos
].p_depth
= i
;
934 path
[ppos
].p_ext
= NULL
;
935 path
[ppos
].p_idx
= NULL
;
938 ext4_ext_binsearch(inode
, path
+ ppos
, block
);
939 /* if not an empty leaf */
940 if (path
[ppos
].p_ext
)
941 path
[ppos
].p_block
= ext4_ext_pblock(path
[ppos
].p_ext
);
943 ext4_ext_show_path(inode
, path
);
948 ext4_ext_drop_refs(path
);
956 * ext4_ext_insert_index:
957 * insert new index [@logical;@ptr] into the block at @curp;
958 * check where to insert: before @curp or after @curp
960 static int ext4_ext_insert_index(handle_t
*handle
, struct inode
*inode
,
961 struct ext4_ext_path
*curp
,
962 int logical
, ext4_fsblk_t ptr
)
964 struct ext4_extent_idx
*ix
;
967 err
= ext4_ext_get_access(handle
, inode
, curp
);
971 if (unlikely(logical
== le32_to_cpu(curp
->p_idx
->ei_block
))) {
972 EXT4_ERROR_INODE(inode
,
973 "logical %d == ei_block %d!",
974 logical
, le32_to_cpu(curp
->p_idx
->ei_block
));
975 return -EFSCORRUPTED
;
978 if (unlikely(le16_to_cpu(curp
->p_hdr
->eh_entries
)
979 >= le16_to_cpu(curp
->p_hdr
->eh_max
))) {
980 EXT4_ERROR_INODE(inode
,
981 "eh_entries %d >= eh_max %d!",
982 le16_to_cpu(curp
->p_hdr
->eh_entries
),
983 le16_to_cpu(curp
->p_hdr
->eh_max
));
984 return -EFSCORRUPTED
;
987 if (logical
> le32_to_cpu(curp
->p_idx
->ei_block
)) {
989 ext_debug("insert new index %d after: %llu\n", logical
, ptr
);
990 ix
= curp
->p_idx
+ 1;
993 ext_debug("insert new index %d before: %llu\n", logical
, ptr
);
997 len
= EXT_LAST_INDEX(curp
->p_hdr
) - ix
+ 1;
1000 ext_debug("insert new index %d: "
1001 "move %d indices from 0x%p to 0x%p\n",
1002 logical
, len
, ix
, ix
+ 1);
1003 memmove(ix
+ 1, ix
, len
* sizeof(struct ext4_extent_idx
));
1006 if (unlikely(ix
> EXT_MAX_INDEX(curp
->p_hdr
))) {
1007 EXT4_ERROR_INODE(inode
, "ix > EXT_MAX_INDEX!");
1008 return -EFSCORRUPTED
;
1011 ix
->ei_block
= cpu_to_le32(logical
);
1012 ext4_idx_store_pblock(ix
, ptr
);
1013 le16_add_cpu(&curp
->p_hdr
->eh_entries
, 1);
1015 if (unlikely(ix
> EXT_LAST_INDEX(curp
->p_hdr
))) {
1016 EXT4_ERROR_INODE(inode
, "ix > EXT_LAST_INDEX!");
1017 return -EFSCORRUPTED
;
1020 err
= ext4_ext_dirty(handle
, inode
, curp
);
1021 ext4_std_error(inode
->i_sb
, err
);
1028 * inserts new subtree into the path, using free index entry
1030 * - allocates all needed blocks (new leaf and all intermediate index blocks)
1031 * - makes decision where to split
1032 * - moves remaining extents and index entries (right to the split point)
1033 * into the newly allocated blocks
1034 * - initializes subtree
1036 static int ext4_ext_split(handle_t
*handle
, struct inode
*inode
,
1038 struct ext4_ext_path
*path
,
1039 struct ext4_extent
*newext
, int at
)
1041 struct buffer_head
*bh
= NULL
;
1042 int depth
= ext_depth(inode
);
1043 struct ext4_extent_header
*neh
;
1044 struct ext4_extent_idx
*fidx
;
1045 int i
= at
, k
, m
, a
;
1046 ext4_fsblk_t newblock
, oldblock
;
1048 ext4_fsblk_t
*ablocks
= NULL
; /* array of allocated blocks */
1050 size_t ext_size
= 0;
1052 /* make decision: where to split? */
1053 /* FIXME: now decision is simplest: at current extent */
1055 /* if current leaf will be split, then we should use
1056 * border from split point */
1057 if (unlikely(path
[depth
].p_ext
> EXT_MAX_EXTENT(path
[depth
].p_hdr
))) {
1058 EXT4_ERROR_INODE(inode
, "p_ext > EXT_MAX_EXTENT!");
1059 return -EFSCORRUPTED
;
1061 if (path
[depth
].p_ext
!= EXT_MAX_EXTENT(path
[depth
].p_hdr
)) {
1062 border
= path
[depth
].p_ext
[1].ee_block
;
1063 ext_debug("leaf will be split."
1064 " next leaf starts at %d\n",
1065 le32_to_cpu(border
));
1067 border
= newext
->ee_block
;
1068 ext_debug("leaf will be added."
1069 " next leaf starts at %d\n",
1070 le32_to_cpu(border
));
1074 * If error occurs, then we break processing
1075 * and mark filesystem read-only. index won't
1076 * be inserted and tree will be in consistent
1077 * state. Next mount will repair buffers too.
1081 * Get array to track all allocated blocks.
1082 * We need this to handle errors and free blocks
1085 ablocks
= kzalloc(sizeof(ext4_fsblk_t
) * depth
, GFP_NOFS
);
1089 /* allocate all needed blocks */
1090 ext_debug("allocate %d blocks for indexes/leaf\n", depth
- at
);
1091 for (a
= 0; a
< depth
- at
; a
++) {
1092 newblock
= ext4_ext_new_meta_block(handle
, inode
, path
,
1093 newext
, &err
, flags
);
1096 ablocks
[a
] = newblock
;
1099 /* initialize new leaf */
1100 newblock
= ablocks
[--a
];
1101 if (unlikely(newblock
== 0)) {
1102 EXT4_ERROR_INODE(inode
, "newblock == 0!");
1103 err
= -EFSCORRUPTED
;
1106 bh
= sb_getblk_gfp(inode
->i_sb
, newblock
, __GFP_MOVABLE
| GFP_NOFS
);
1107 if (unlikely(!bh
)) {
1113 err
= ext4_journal_get_create_access(handle
, bh
);
1117 neh
= ext_block_hdr(bh
);
1118 neh
->eh_entries
= 0;
1119 neh
->eh_max
= cpu_to_le16(ext4_ext_space_block(inode
, 0));
1120 neh
->eh_magic
= EXT4_EXT_MAGIC
;
1123 /* move remainder of path[depth] to the new leaf */
1124 if (unlikely(path
[depth
].p_hdr
->eh_entries
!=
1125 path
[depth
].p_hdr
->eh_max
)) {
1126 EXT4_ERROR_INODE(inode
, "eh_entries %d != eh_max %d!",
1127 path
[depth
].p_hdr
->eh_entries
,
1128 path
[depth
].p_hdr
->eh_max
);
1129 err
= -EFSCORRUPTED
;
1132 /* start copy from next extent */
1133 m
= EXT_MAX_EXTENT(path
[depth
].p_hdr
) - path
[depth
].p_ext
++;
1134 ext4_ext_show_move(inode
, path
, newblock
, depth
);
1136 struct ext4_extent
*ex
;
1137 ex
= EXT_FIRST_EXTENT(neh
);
1138 memmove(ex
, path
[depth
].p_ext
, sizeof(struct ext4_extent
) * m
);
1139 le16_add_cpu(&neh
->eh_entries
, m
);
1142 /* zero out unused area in the extent block */
1143 ext_size
= sizeof(struct ext4_extent_header
) +
1144 sizeof(struct ext4_extent
) * le16_to_cpu(neh
->eh_entries
);
1145 memset(bh
->b_data
+ ext_size
, 0, inode
->i_sb
->s_blocksize
- ext_size
);
1146 ext4_extent_block_csum_set(inode
, neh
);
1147 set_buffer_uptodate(bh
);
1150 err
= ext4_handle_dirty_metadata(handle
, inode
, bh
);
1156 /* correct old leaf */
1158 err
= ext4_ext_get_access(handle
, inode
, path
+ depth
);
1161 le16_add_cpu(&path
[depth
].p_hdr
->eh_entries
, -m
);
1162 err
= ext4_ext_dirty(handle
, inode
, path
+ depth
);
1168 /* create intermediate indexes */
1170 if (unlikely(k
< 0)) {
1171 EXT4_ERROR_INODE(inode
, "k %d < 0!", k
);
1172 err
= -EFSCORRUPTED
;
1176 ext_debug("create %d intermediate indices\n", k
);
1177 /* insert new index into current index block */
1178 /* current depth stored in i var */
1181 oldblock
= newblock
;
1182 newblock
= ablocks
[--a
];
1183 bh
= sb_getblk(inode
->i_sb
, newblock
);
1184 if (unlikely(!bh
)) {
1190 err
= ext4_journal_get_create_access(handle
, bh
);
1194 neh
= ext_block_hdr(bh
);
1195 neh
->eh_entries
= cpu_to_le16(1);
1196 neh
->eh_magic
= EXT4_EXT_MAGIC
;
1197 neh
->eh_max
= cpu_to_le16(ext4_ext_space_block_idx(inode
, 0));
1198 neh
->eh_depth
= cpu_to_le16(depth
- i
);
1199 fidx
= EXT_FIRST_INDEX(neh
);
1200 fidx
->ei_block
= border
;
1201 ext4_idx_store_pblock(fidx
, oldblock
);
1203 ext_debug("int.index at %d (block %llu): %u -> %llu\n",
1204 i
, newblock
, le32_to_cpu(border
), oldblock
);
1206 /* move remainder of path[i] to the new index block */
1207 if (unlikely(EXT_MAX_INDEX(path
[i
].p_hdr
) !=
1208 EXT_LAST_INDEX(path
[i
].p_hdr
))) {
1209 EXT4_ERROR_INODE(inode
,
1210 "EXT_MAX_INDEX != EXT_LAST_INDEX ee_block %d!",
1211 le32_to_cpu(path
[i
].p_ext
->ee_block
));
1212 err
= -EFSCORRUPTED
;
1215 /* start copy indexes */
1216 m
= EXT_MAX_INDEX(path
[i
].p_hdr
) - path
[i
].p_idx
++;
1217 ext_debug("cur 0x%p, last 0x%p\n", path
[i
].p_idx
,
1218 EXT_MAX_INDEX(path
[i
].p_hdr
));
1219 ext4_ext_show_move(inode
, path
, newblock
, i
);
1221 memmove(++fidx
, path
[i
].p_idx
,
1222 sizeof(struct ext4_extent_idx
) * m
);
1223 le16_add_cpu(&neh
->eh_entries
, m
);
1225 /* zero out unused area in the extent block */
1226 ext_size
= sizeof(struct ext4_extent_header
) +
1227 (sizeof(struct ext4_extent
) * le16_to_cpu(neh
->eh_entries
));
1228 memset(bh
->b_data
+ ext_size
, 0,
1229 inode
->i_sb
->s_blocksize
- ext_size
);
1230 ext4_extent_block_csum_set(inode
, neh
);
1231 set_buffer_uptodate(bh
);
1234 err
= ext4_handle_dirty_metadata(handle
, inode
, bh
);
1240 /* correct old index */
1242 err
= ext4_ext_get_access(handle
, inode
, path
+ i
);
1245 le16_add_cpu(&path
[i
].p_hdr
->eh_entries
, -m
);
1246 err
= ext4_ext_dirty(handle
, inode
, path
+ i
);
1254 /* insert new index */
1255 err
= ext4_ext_insert_index(handle
, inode
, path
+ at
,
1256 le32_to_cpu(border
), newblock
);
1260 if (buffer_locked(bh
))
1266 /* free all allocated blocks in error case */
1267 for (i
= 0; i
< depth
; i
++) {
1270 ext4_free_blocks(handle
, inode
, NULL
, ablocks
[i
], 1,
1271 EXT4_FREE_BLOCKS_METADATA
);
1280 * ext4_ext_grow_indepth:
1281 * implements tree growing procedure:
1282 * - allocates new block
1283 * - moves top-level data (index block or leaf) into the new block
1284 * - initializes new top-level, creating index that points to the
1285 * just created block
1287 static int ext4_ext_grow_indepth(handle_t
*handle
, struct inode
*inode
,
1290 struct ext4_extent_header
*neh
;
1291 struct buffer_head
*bh
;
1292 ext4_fsblk_t newblock
, goal
= 0;
1293 struct ext4_super_block
*es
= EXT4_SB(inode
->i_sb
)->s_es
;
1295 size_t ext_size
= 0;
1297 /* Try to prepend new index to old one */
1298 if (ext_depth(inode
))
1299 goal
= ext4_idx_pblock(EXT_FIRST_INDEX(ext_inode_hdr(inode
)));
1300 if (goal
> le32_to_cpu(es
->s_first_data_block
)) {
1301 flags
|= EXT4_MB_HINT_TRY_GOAL
;
1304 goal
= ext4_inode_to_goal_block(inode
);
1305 newblock
= ext4_new_meta_blocks(handle
, inode
, goal
, flags
,
1310 bh
= sb_getblk_gfp(inode
->i_sb
, newblock
, __GFP_MOVABLE
| GFP_NOFS
);
1315 err
= ext4_journal_get_create_access(handle
, bh
);
1321 ext_size
= sizeof(EXT4_I(inode
)->i_data
);
1322 /* move top-level index/leaf into new block */
1323 memmove(bh
->b_data
, EXT4_I(inode
)->i_data
, ext_size
);
1324 /* zero out unused area in the extent block */
1325 memset(bh
->b_data
+ ext_size
, 0, inode
->i_sb
->s_blocksize
- ext_size
);
1327 /* set size of new block */
1328 neh
= ext_block_hdr(bh
);
1329 /* old root could have indexes or leaves
1330 * so calculate e_max right way */
1331 if (ext_depth(inode
))
1332 neh
->eh_max
= cpu_to_le16(ext4_ext_space_block_idx(inode
, 0));
1334 neh
->eh_max
= cpu_to_le16(ext4_ext_space_block(inode
, 0));
1335 neh
->eh_magic
= EXT4_EXT_MAGIC
;
1336 ext4_extent_block_csum_set(inode
, neh
);
1337 set_buffer_uptodate(bh
);
1340 err
= ext4_handle_dirty_metadata(handle
, inode
, bh
);
1344 /* Update top-level index: num,max,pointer */
1345 neh
= ext_inode_hdr(inode
);
1346 neh
->eh_entries
= cpu_to_le16(1);
1347 ext4_idx_store_pblock(EXT_FIRST_INDEX(neh
), newblock
);
1348 if (neh
->eh_depth
== 0) {
1349 /* Root extent block becomes index block */
1350 neh
->eh_max
= cpu_to_le16(ext4_ext_space_root_idx(inode
, 0));
1351 EXT_FIRST_INDEX(neh
)->ei_block
=
1352 EXT_FIRST_EXTENT(neh
)->ee_block
;
1354 ext_debug("new root: num %d(%d), lblock %d, ptr %llu\n",
1355 le16_to_cpu(neh
->eh_entries
), le16_to_cpu(neh
->eh_max
),
1356 le32_to_cpu(EXT_FIRST_INDEX(neh
)->ei_block
),
1357 ext4_idx_pblock(EXT_FIRST_INDEX(neh
)));
1359 le16_add_cpu(&neh
->eh_depth
, 1);
1360 ext4_mark_inode_dirty(handle
, inode
);
1368 * ext4_ext_create_new_leaf:
1369 * finds empty index and adds new leaf.
1370 * if no free index is found, then it requests in-depth growing.
1372 static int ext4_ext_create_new_leaf(handle_t
*handle
, struct inode
*inode
,
1373 unsigned int mb_flags
,
1374 unsigned int gb_flags
,
1375 struct ext4_ext_path
**ppath
,
1376 struct ext4_extent
*newext
)
1378 struct ext4_ext_path
*path
= *ppath
;
1379 struct ext4_ext_path
*curp
;
1380 int depth
, i
, err
= 0;
1383 i
= depth
= ext_depth(inode
);
1385 /* walk up to the tree and look for free index entry */
1386 curp
= path
+ depth
;
1387 while (i
> 0 && !EXT_HAS_FREE_INDEX(curp
)) {
1392 /* we use already allocated block for index block,
1393 * so subsequent data blocks should be contiguous */
1394 if (EXT_HAS_FREE_INDEX(curp
)) {
1395 /* if we found index with free entry, then use that
1396 * entry: create all needed subtree and add new leaf */
1397 err
= ext4_ext_split(handle
, inode
, mb_flags
, path
, newext
, i
);
1402 path
= ext4_find_extent(inode
,
1403 (ext4_lblk_t
)le32_to_cpu(newext
->ee_block
),
1406 err
= PTR_ERR(path
);
1408 /* tree is full, time to grow in depth */
1409 err
= ext4_ext_grow_indepth(handle
, inode
, mb_flags
);
1414 path
= ext4_find_extent(inode
,
1415 (ext4_lblk_t
)le32_to_cpu(newext
->ee_block
),
1418 err
= PTR_ERR(path
);
1423 * only first (depth 0 -> 1) produces free space;
1424 * in all other cases we have to split the grown tree
1426 depth
= ext_depth(inode
);
1427 if (path
[depth
].p_hdr
->eh_entries
== path
[depth
].p_hdr
->eh_max
) {
1428 /* now we need to split */
1438 * search the closest allocated block to the left for *logical
1439 * and returns it at @logical + it's physical address at @phys
1440 * if *logical is the smallest allocated block, the function
1441 * returns 0 at @phys
1442 * return value contains 0 (success) or error code
1444 static int ext4_ext_search_left(struct inode
*inode
,
1445 struct ext4_ext_path
*path
,
1446 ext4_lblk_t
*logical
, ext4_fsblk_t
*phys
)
1448 struct ext4_extent_idx
*ix
;
1449 struct ext4_extent
*ex
;
1452 if (unlikely(path
== NULL
)) {
1453 EXT4_ERROR_INODE(inode
, "path == NULL *logical %d!", *logical
);
1454 return -EFSCORRUPTED
;
1456 depth
= path
->p_depth
;
1459 if (depth
== 0 && path
->p_ext
== NULL
)
1462 /* usually extent in the path covers blocks smaller
1463 * then *logical, but it can be that extent is the
1464 * first one in the file */
1466 ex
= path
[depth
].p_ext
;
1467 ee_len
= ext4_ext_get_actual_len(ex
);
1468 if (*logical
< le32_to_cpu(ex
->ee_block
)) {
1469 if (unlikely(EXT_FIRST_EXTENT(path
[depth
].p_hdr
) != ex
)) {
1470 EXT4_ERROR_INODE(inode
,
1471 "EXT_FIRST_EXTENT != ex *logical %d ee_block %d!",
1472 *logical
, le32_to_cpu(ex
->ee_block
));
1473 return -EFSCORRUPTED
;
1475 while (--depth
>= 0) {
1476 ix
= path
[depth
].p_idx
;
1477 if (unlikely(ix
!= EXT_FIRST_INDEX(path
[depth
].p_hdr
))) {
1478 EXT4_ERROR_INODE(inode
,
1479 "ix (%d) != EXT_FIRST_INDEX (%d) (depth %d)!",
1480 ix
!= NULL
? le32_to_cpu(ix
->ei_block
) : 0,
1481 EXT_FIRST_INDEX(path
[depth
].p_hdr
) != NULL
?
1482 le32_to_cpu(EXT_FIRST_INDEX(path
[depth
].p_hdr
)->ei_block
) : 0,
1484 return -EFSCORRUPTED
;
1490 if (unlikely(*logical
< (le32_to_cpu(ex
->ee_block
) + ee_len
))) {
1491 EXT4_ERROR_INODE(inode
,
1492 "logical %d < ee_block %d + ee_len %d!",
1493 *logical
, le32_to_cpu(ex
->ee_block
), ee_len
);
1494 return -EFSCORRUPTED
;
1497 *logical
= le32_to_cpu(ex
->ee_block
) + ee_len
- 1;
1498 *phys
= ext4_ext_pblock(ex
) + ee_len
- 1;
1503 * search the closest allocated block to the right for *logical
1504 * and returns it at @logical + it's physical address at @phys
1505 * if *logical is the largest allocated block, the function
1506 * returns 0 at @phys
1507 * return value contains 0 (success) or error code
1509 static int ext4_ext_search_right(struct inode
*inode
,
1510 struct ext4_ext_path
*path
,
1511 ext4_lblk_t
*logical
, ext4_fsblk_t
*phys
,
1512 struct ext4_extent
**ret_ex
)
1514 struct buffer_head
*bh
= NULL
;
1515 struct ext4_extent_header
*eh
;
1516 struct ext4_extent_idx
*ix
;
1517 struct ext4_extent
*ex
;
1519 int depth
; /* Note, NOT eh_depth; depth from top of tree */
1522 if (unlikely(path
== NULL
)) {
1523 EXT4_ERROR_INODE(inode
, "path == NULL *logical %d!", *logical
);
1524 return -EFSCORRUPTED
;
1526 depth
= path
->p_depth
;
1529 if (depth
== 0 && path
->p_ext
== NULL
)
1532 /* usually extent in the path covers blocks smaller
1533 * then *logical, but it can be that extent is the
1534 * first one in the file */
1536 ex
= path
[depth
].p_ext
;
1537 ee_len
= ext4_ext_get_actual_len(ex
);
1538 if (*logical
< le32_to_cpu(ex
->ee_block
)) {
1539 if (unlikely(EXT_FIRST_EXTENT(path
[depth
].p_hdr
) != ex
)) {
1540 EXT4_ERROR_INODE(inode
,
1541 "first_extent(path[%d].p_hdr) != ex",
1543 return -EFSCORRUPTED
;
1545 while (--depth
>= 0) {
1546 ix
= path
[depth
].p_idx
;
1547 if (unlikely(ix
!= EXT_FIRST_INDEX(path
[depth
].p_hdr
))) {
1548 EXT4_ERROR_INODE(inode
,
1549 "ix != EXT_FIRST_INDEX *logical %d!",
1551 return -EFSCORRUPTED
;
1557 if (unlikely(*logical
< (le32_to_cpu(ex
->ee_block
) + ee_len
))) {
1558 EXT4_ERROR_INODE(inode
,
1559 "logical %d < ee_block %d + ee_len %d!",
1560 *logical
, le32_to_cpu(ex
->ee_block
), ee_len
);
1561 return -EFSCORRUPTED
;
1564 if (ex
!= EXT_LAST_EXTENT(path
[depth
].p_hdr
)) {
1565 /* next allocated block in this leaf */
1570 /* go up and search for index to the right */
1571 while (--depth
>= 0) {
1572 ix
= path
[depth
].p_idx
;
1573 if (ix
!= EXT_LAST_INDEX(path
[depth
].p_hdr
))
1577 /* we've gone up to the root and found no index to the right */
1581 /* we've found index to the right, let's
1582 * follow it and find the closest allocated
1583 * block to the right */
1585 block
= ext4_idx_pblock(ix
);
1586 while (++depth
< path
->p_depth
) {
1587 /* subtract from p_depth to get proper eh_depth */
1588 bh
= read_extent_tree_block(inode
, block
,
1589 path
->p_depth
- depth
, 0);
1592 eh
= ext_block_hdr(bh
);
1593 ix
= EXT_FIRST_INDEX(eh
);
1594 block
= ext4_idx_pblock(ix
);
1598 bh
= read_extent_tree_block(inode
, block
, path
->p_depth
- depth
, 0);
1601 eh
= ext_block_hdr(bh
);
1602 ex
= EXT_FIRST_EXTENT(eh
);
1604 *logical
= le32_to_cpu(ex
->ee_block
);
1605 *phys
= ext4_ext_pblock(ex
);
1613 * ext4_ext_next_allocated_block:
1614 * returns allocated block in subsequent extent or EXT_MAX_BLOCKS.
1615 * NOTE: it considers block number from index entry as
1616 * allocated block. Thus, index entries have to be consistent
1620 ext4_ext_next_allocated_block(struct ext4_ext_path
*path
)
1624 BUG_ON(path
== NULL
);
1625 depth
= path
->p_depth
;
1627 if (depth
== 0 && path
->p_ext
== NULL
)
1628 return EXT_MAX_BLOCKS
;
1630 while (depth
>= 0) {
1631 if (depth
== path
->p_depth
) {
1633 if (path
[depth
].p_ext
&&
1634 path
[depth
].p_ext
!=
1635 EXT_LAST_EXTENT(path
[depth
].p_hdr
))
1636 return le32_to_cpu(path
[depth
].p_ext
[1].ee_block
);
1639 if (path
[depth
].p_idx
!=
1640 EXT_LAST_INDEX(path
[depth
].p_hdr
))
1641 return le32_to_cpu(path
[depth
].p_idx
[1].ei_block
);
1646 return EXT_MAX_BLOCKS
;
1650 * ext4_ext_next_leaf_block:
1651 * returns first allocated block from next leaf or EXT_MAX_BLOCKS
1653 static ext4_lblk_t
ext4_ext_next_leaf_block(struct ext4_ext_path
*path
)
1657 BUG_ON(path
== NULL
);
1658 depth
= path
->p_depth
;
1660 /* zero-tree has no leaf blocks at all */
1662 return EXT_MAX_BLOCKS
;
1664 /* go to index block */
1667 while (depth
>= 0) {
1668 if (path
[depth
].p_idx
!=
1669 EXT_LAST_INDEX(path
[depth
].p_hdr
))
1670 return (ext4_lblk_t
)
1671 le32_to_cpu(path
[depth
].p_idx
[1].ei_block
);
1675 return EXT_MAX_BLOCKS
;
1679 * ext4_ext_correct_indexes:
1680 * if leaf gets modified and modified extent is first in the leaf,
1681 * then we have to correct all indexes above.
1682 * TODO: do we need to correct tree in all cases?
1684 static int ext4_ext_correct_indexes(handle_t
*handle
, struct inode
*inode
,
1685 struct ext4_ext_path
*path
)
1687 struct ext4_extent_header
*eh
;
1688 int depth
= ext_depth(inode
);
1689 struct ext4_extent
*ex
;
1693 eh
= path
[depth
].p_hdr
;
1694 ex
= path
[depth
].p_ext
;
1696 if (unlikely(ex
== NULL
|| eh
== NULL
)) {
1697 EXT4_ERROR_INODE(inode
,
1698 "ex %p == NULL or eh %p == NULL", ex
, eh
);
1699 return -EFSCORRUPTED
;
1703 /* there is no tree at all */
1707 if (ex
!= EXT_FIRST_EXTENT(eh
)) {
1708 /* we correct tree if first leaf got modified only */
1713 * TODO: we need correction if border is smaller than current one
1716 border
= path
[depth
].p_ext
->ee_block
;
1717 err
= ext4_ext_get_access(handle
, inode
, path
+ k
);
1720 path
[k
].p_idx
->ei_block
= border
;
1721 err
= ext4_ext_dirty(handle
, inode
, path
+ k
);
1726 /* change all left-side indexes */
1727 if (path
[k
+1].p_idx
!= EXT_FIRST_INDEX(path
[k
+1].p_hdr
))
1729 err
= ext4_ext_get_access(handle
, inode
, path
+ k
);
1732 path
[k
].p_idx
->ei_block
= border
;
1733 err
= ext4_ext_dirty(handle
, inode
, path
+ k
);
1742 ext4_can_extents_be_merged(struct inode
*inode
, struct ext4_extent
*ex1
,
1743 struct ext4_extent
*ex2
)
1745 unsigned short ext1_ee_len
, ext2_ee_len
;
1747 if (ext4_ext_is_unwritten(ex1
) != ext4_ext_is_unwritten(ex2
))
1750 ext1_ee_len
= ext4_ext_get_actual_len(ex1
);
1751 ext2_ee_len
= ext4_ext_get_actual_len(ex2
);
1753 if (le32_to_cpu(ex1
->ee_block
) + ext1_ee_len
!=
1754 le32_to_cpu(ex2
->ee_block
))
1758 * To allow future support for preallocated extents to be added
1759 * as an RO_COMPAT feature, refuse to merge to extents if
1760 * this can result in the top bit of ee_len being set.
1762 if (ext1_ee_len
+ ext2_ee_len
> EXT_INIT_MAX_LEN
)
1765 * The check for IO to unwritten extent is somewhat racy as we
1766 * increment i_unwritten / set EXT4_STATE_DIO_UNWRITTEN only after
1767 * dropping i_data_sem. But reserved blocks should save us in that
1770 if (ext4_ext_is_unwritten(ex1
) &&
1771 (ext4_test_inode_state(inode
, EXT4_STATE_DIO_UNWRITTEN
) ||
1772 atomic_read(&EXT4_I(inode
)->i_unwritten
) ||
1773 (ext1_ee_len
+ ext2_ee_len
> EXT_UNWRITTEN_MAX_LEN
)))
1775 #ifdef AGGRESSIVE_TEST
1776 if (ext1_ee_len
>= 4)
1780 if (ext4_ext_pblock(ex1
) + ext1_ee_len
== ext4_ext_pblock(ex2
))
1786 * This function tries to merge the "ex" extent to the next extent in the tree.
1787 * It always tries to merge towards right. If you want to merge towards
1788 * left, pass "ex - 1" as argument instead of "ex".
1789 * Returns 0 if the extents (ex and ex+1) were _not_ merged and returns
1790 * 1 if they got merged.
1792 static int ext4_ext_try_to_merge_right(struct inode
*inode
,
1793 struct ext4_ext_path
*path
,
1794 struct ext4_extent
*ex
)
1796 struct ext4_extent_header
*eh
;
1797 unsigned int depth
, len
;
1798 int merge_done
= 0, unwritten
;
1800 depth
= ext_depth(inode
);
1801 BUG_ON(path
[depth
].p_hdr
== NULL
);
1802 eh
= path
[depth
].p_hdr
;
1804 while (ex
< EXT_LAST_EXTENT(eh
)) {
1805 if (!ext4_can_extents_be_merged(inode
, ex
, ex
+ 1))
1807 /* merge with next extent! */
1808 unwritten
= ext4_ext_is_unwritten(ex
);
1809 ex
->ee_len
= cpu_to_le16(ext4_ext_get_actual_len(ex
)
1810 + ext4_ext_get_actual_len(ex
+ 1));
1812 ext4_ext_mark_unwritten(ex
);
1814 if (ex
+ 1 < EXT_LAST_EXTENT(eh
)) {
1815 len
= (EXT_LAST_EXTENT(eh
) - ex
- 1)
1816 * sizeof(struct ext4_extent
);
1817 memmove(ex
+ 1, ex
+ 2, len
);
1819 le16_add_cpu(&eh
->eh_entries
, -1);
1821 WARN_ON(eh
->eh_entries
== 0);
1822 if (!eh
->eh_entries
)
1823 EXT4_ERROR_INODE(inode
, "eh->eh_entries = 0!");
1830 * This function does a very simple check to see if we can collapse
1831 * an extent tree with a single extent tree leaf block into the inode.
1833 static void ext4_ext_try_to_merge_up(handle_t
*handle
,
1834 struct inode
*inode
,
1835 struct ext4_ext_path
*path
)
1838 unsigned max_root
= ext4_ext_space_root(inode
, 0);
1841 if ((path
[0].p_depth
!= 1) ||
1842 (le16_to_cpu(path
[0].p_hdr
->eh_entries
) != 1) ||
1843 (le16_to_cpu(path
[1].p_hdr
->eh_entries
) > max_root
))
1847 * We need to modify the block allocation bitmap and the block
1848 * group descriptor to release the extent tree block. If we
1849 * can't get the journal credits, give up.
1851 if (ext4_journal_extend(handle
, 2))
1855 * Copy the extent data up to the inode
1857 blk
= ext4_idx_pblock(path
[0].p_idx
);
1858 s
= le16_to_cpu(path
[1].p_hdr
->eh_entries
) *
1859 sizeof(struct ext4_extent_idx
);
1860 s
+= sizeof(struct ext4_extent_header
);
1862 path
[1].p_maxdepth
= path
[0].p_maxdepth
;
1863 memcpy(path
[0].p_hdr
, path
[1].p_hdr
, s
);
1864 path
[0].p_depth
= 0;
1865 path
[0].p_ext
= EXT_FIRST_EXTENT(path
[0].p_hdr
) +
1866 (path
[1].p_ext
- EXT_FIRST_EXTENT(path
[1].p_hdr
));
1867 path
[0].p_hdr
->eh_max
= cpu_to_le16(max_root
);
1869 brelse(path
[1].p_bh
);
1870 ext4_free_blocks(handle
, inode
, NULL
, blk
, 1,
1871 EXT4_FREE_BLOCKS_METADATA
| EXT4_FREE_BLOCKS_FORGET
);
1875 * This function tries to merge the @ex extent to neighbours in the tree.
1876 * return 1 if merge left else 0.
1878 static void ext4_ext_try_to_merge(handle_t
*handle
,
1879 struct inode
*inode
,
1880 struct ext4_ext_path
*path
,
1881 struct ext4_extent
*ex
) {
1882 struct ext4_extent_header
*eh
;
1886 depth
= ext_depth(inode
);
1887 BUG_ON(path
[depth
].p_hdr
== NULL
);
1888 eh
= path
[depth
].p_hdr
;
1890 if (ex
> EXT_FIRST_EXTENT(eh
))
1891 merge_done
= ext4_ext_try_to_merge_right(inode
, path
, ex
- 1);
1894 (void) ext4_ext_try_to_merge_right(inode
, path
, ex
);
1896 ext4_ext_try_to_merge_up(handle
, inode
, path
);
1900 * check if a portion of the "newext" extent overlaps with an
1903 * If there is an overlap discovered, it updates the length of the newext
1904 * such that there will be no overlap, and then returns 1.
1905 * If there is no overlap found, it returns 0.
1907 static unsigned int ext4_ext_check_overlap(struct ext4_sb_info
*sbi
,
1908 struct inode
*inode
,
1909 struct ext4_extent
*newext
,
1910 struct ext4_ext_path
*path
)
1913 unsigned int depth
, len1
;
1914 unsigned int ret
= 0;
1916 b1
= le32_to_cpu(newext
->ee_block
);
1917 len1
= ext4_ext_get_actual_len(newext
);
1918 depth
= ext_depth(inode
);
1919 if (!path
[depth
].p_ext
)
1921 b2
= EXT4_LBLK_CMASK(sbi
, le32_to_cpu(path
[depth
].p_ext
->ee_block
));
1924 * get the next allocated block if the extent in the path
1925 * is before the requested block(s)
1928 b2
= ext4_ext_next_allocated_block(path
);
1929 if (b2
== EXT_MAX_BLOCKS
)
1931 b2
= EXT4_LBLK_CMASK(sbi
, b2
);
1934 /* check for wrap through zero on extent logical start block*/
1935 if (b1
+ len1
< b1
) {
1936 len1
= EXT_MAX_BLOCKS
- b1
;
1937 newext
->ee_len
= cpu_to_le16(len1
);
1941 /* check for overlap */
1942 if (b1
+ len1
> b2
) {
1943 newext
->ee_len
= cpu_to_le16(b2
- b1
);
1951 * ext4_ext_insert_extent:
1952 * tries to merge requsted extent into the existing extent or
1953 * inserts requested extent as new one into the tree,
1954 * creating new leaf in the no-space case.
1956 int ext4_ext_insert_extent(handle_t
*handle
, struct inode
*inode
,
1957 struct ext4_ext_path
**ppath
,
1958 struct ext4_extent
*newext
, int gb_flags
)
1960 struct ext4_ext_path
*path
= *ppath
;
1961 struct ext4_extent_header
*eh
;
1962 struct ext4_extent
*ex
, *fex
;
1963 struct ext4_extent
*nearex
; /* nearest extent */
1964 struct ext4_ext_path
*npath
= NULL
;
1965 int depth
, len
, err
;
1967 int mb_flags
= 0, unwritten
;
1969 if (gb_flags
& EXT4_GET_BLOCKS_DELALLOC_RESERVE
)
1970 mb_flags
|= EXT4_MB_DELALLOC_RESERVED
;
1971 if (unlikely(ext4_ext_get_actual_len(newext
) == 0)) {
1972 EXT4_ERROR_INODE(inode
, "ext4_ext_get_actual_len(newext) == 0");
1973 return -EFSCORRUPTED
;
1975 depth
= ext_depth(inode
);
1976 ex
= path
[depth
].p_ext
;
1977 eh
= path
[depth
].p_hdr
;
1978 if (unlikely(path
[depth
].p_hdr
== NULL
)) {
1979 EXT4_ERROR_INODE(inode
, "path[%d].p_hdr == NULL", depth
);
1980 return -EFSCORRUPTED
;
1983 /* try to insert block into found extent and return */
1984 if (ex
&& !(gb_flags
& EXT4_GET_BLOCKS_PRE_IO
)) {
1987 * Try to see whether we should rather test the extent on
1988 * right from ex, or from the left of ex. This is because
1989 * ext4_find_extent() can return either extent on the
1990 * left, or on the right from the searched position. This
1991 * will make merging more effective.
1993 if (ex
< EXT_LAST_EXTENT(eh
) &&
1994 (le32_to_cpu(ex
->ee_block
) +
1995 ext4_ext_get_actual_len(ex
) <
1996 le32_to_cpu(newext
->ee_block
))) {
1999 } else if ((ex
> EXT_FIRST_EXTENT(eh
)) &&
2000 (le32_to_cpu(newext
->ee_block
) +
2001 ext4_ext_get_actual_len(newext
) <
2002 le32_to_cpu(ex
->ee_block
)))
2005 /* Try to append newex to the ex */
2006 if (ext4_can_extents_be_merged(inode
, ex
, newext
)) {
2007 ext_debug("append [%d]%d block to %u:[%d]%d"
2009 ext4_ext_is_unwritten(newext
),
2010 ext4_ext_get_actual_len(newext
),
2011 le32_to_cpu(ex
->ee_block
),
2012 ext4_ext_is_unwritten(ex
),
2013 ext4_ext_get_actual_len(ex
),
2014 ext4_ext_pblock(ex
));
2015 err
= ext4_ext_get_access(handle
, inode
,
2019 unwritten
= ext4_ext_is_unwritten(ex
);
2020 ex
->ee_len
= cpu_to_le16(ext4_ext_get_actual_len(ex
)
2021 + ext4_ext_get_actual_len(newext
));
2023 ext4_ext_mark_unwritten(ex
);
2024 eh
= path
[depth
].p_hdr
;
2030 /* Try to prepend newex to the ex */
2031 if (ext4_can_extents_be_merged(inode
, newext
, ex
)) {
2032 ext_debug("prepend %u[%d]%d block to %u:[%d]%d"
2034 le32_to_cpu(newext
->ee_block
),
2035 ext4_ext_is_unwritten(newext
),
2036 ext4_ext_get_actual_len(newext
),
2037 le32_to_cpu(ex
->ee_block
),
2038 ext4_ext_is_unwritten(ex
),
2039 ext4_ext_get_actual_len(ex
),
2040 ext4_ext_pblock(ex
));
2041 err
= ext4_ext_get_access(handle
, inode
,
2046 unwritten
= ext4_ext_is_unwritten(ex
);
2047 ex
->ee_block
= newext
->ee_block
;
2048 ext4_ext_store_pblock(ex
, ext4_ext_pblock(newext
));
2049 ex
->ee_len
= cpu_to_le16(ext4_ext_get_actual_len(ex
)
2050 + ext4_ext_get_actual_len(newext
));
2052 ext4_ext_mark_unwritten(ex
);
2053 eh
= path
[depth
].p_hdr
;
2059 depth
= ext_depth(inode
);
2060 eh
= path
[depth
].p_hdr
;
2061 if (le16_to_cpu(eh
->eh_entries
) < le16_to_cpu(eh
->eh_max
))
2064 /* probably next leaf has space for us? */
2065 fex
= EXT_LAST_EXTENT(eh
);
2066 next
= EXT_MAX_BLOCKS
;
2067 if (le32_to_cpu(newext
->ee_block
) > le32_to_cpu(fex
->ee_block
))
2068 next
= ext4_ext_next_leaf_block(path
);
2069 if (next
!= EXT_MAX_BLOCKS
) {
2070 ext_debug("next leaf block - %u\n", next
);
2071 BUG_ON(npath
!= NULL
);
2072 npath
= ext4_find_extent(inode
, next
, NULL
, 0);
2074 return PTR_ERR(npath
);
2075 BUG_ON(npath
->p_depth
!= path
->p_depth
);
2076 eh
= npath
[depth
].p_hdr
;
2077 if (le16_to_cpu(eh
->eh_entries
) < le16_to_cpu(eh
->eh_max
)) {
2078 ext_debug("next leaf isn't full(%d)\n",
2079 le16_to_cpu(eh
->eh_entries
));
2083 ext_debug("next leaf has no free space(%d,%d)\n",
2084 le16_to_cpu(eh
->eh_entries
), le16_to_cpu(eh
->eh_max
));
2088 * There is no free space in the found leaf.
2089 * We're gonna add a new leaf in the tree.
2091 if (gb_flags
& EXT4_GET_BLOCKS_METADATA_NOFAIL
)
2092 mb_flags
|= EXT4_MB_USE_RESERVED
;
2093 err
= ext4_ext_create_new_leaf(handle
, inode
, mb_flags
, gb_flags
,
2097 depth
= ext_depth(inode
);
2098 eh
= path
[depth
].p_hdr
;
2101 nearex
= path
[depth
].p_ext
;
2103 err
= ext4_ext_get_access(handle
, inode
, path
+ depth
);
2108 /* there is no extent in this leaf, create first one */
2109 ext_debug("first extent in the leaf: %u:%llu:[%d]%d\n",
2110 le32_to_cpu(newext
->ee_block
),
2111 ext4_ext_pblock(newext
),
2112 ext4_ext_is_unwritten(newext
),
2113 ext4_ext_get_actual_len(newext
));
2114 nearex
= EXT_FIRST_EXTENT(eh
);
2116 if (le32_to_cpu(newext
->ee_block
)
2117 > le32_to_cpu(nearex
->ee_block
)) {
2119 ext_debug("insert %u:%llu:[%d]%d before: "
2121 le32_to_cpu(newext
->ee_block
),
2122 ext4_ext_pblock(newext
),
2123 ext4_ext_is_unwritten(newext
),
2124 ext4_ext_get_actual_len(newext
),
2129 BUG_ON(newext
->ee_block
== nearex
->ee_block
);
2130 ext_debug("insert %u:%llu:[%d]%d after: "
2132 le32_to_cpu(newext
->ee_block
),
2133 ext4_ext_pblock(newext
),
2134 ext4_ext_is_unwritten(newext
),
2135 ext4_ext_get_actual_len(newext
),
2138 len
= EXT_LAST_EXTENT(eh
) - nearex
+ 1;
2140 ext_debug("insert %u:%llu:[%d]%d: "
2141 "move %d extents from 0x%p to 0x%p\n",
2142 le32_to_cpu(newext
->ee_block
),
2143 ext4_ext_pblock(newext
),
2144 ext4_ext_is_unwritten(newext
),
2145 ext4_ext_get_actual_len(newext
),
2146 len
, nearex
, nearex
+ 1);
2147 memmove(nearex
+ 1, nearex
,
2148 len
* sizeof(struct ext4_extent
));
2152 le16_add_cpu(&eh
->eh_entries
, 1);
2153 path
[depth
].p_ext
= nearex
;
2154 nearex
->ee_block
= newext
->ee_block
;
2155 ext4_ext_store_pblock(nearex
, ext4_ext_pblock(newext
));
2156 nearex
->ee_len
= newext
->ee_len
;
2159 /* try to merge extents */
2160 if (!(gb_flags
& EXT4_GET_BLOCKS_PRE_IO
))
2161 ext4_ext_try_to_merge(handle
, inode
, path
, nearex
);
2164 /* time to correct all indexes above */
2165 err
= ext4_ext_correct_indexes(handle
, inode
, path
);
2169 err
= ext4_ext_dirty(handle
, inode
, path
+ path
->p_depth
);
2172 ext4_ext_drop_refs(npath
);
2177 static int ext4_fill_fiemap_extents(struct inode
*inode
,
2178 ext4_lblk_t block
, ext4_lblk_t num
,
2179 struct fiemap_extent_info
*fieinfo
)
2181 struct ext4_ext_path
*path
= NULL
;
2182 struct ext4_extent
*ex
;
2183 struct extent_status es
;
2184 ext4_lblk_t next
, next_del
, start
= 0, end
= 0;
2185 ext4_lblk_t last
= block
+ num
;
2186 int exists
, depth
= 0, err
= 0;
2187 unsigned int flags
= 0;
2188 unsigned char blksize_bits
= inode
->i_sb
->s_blocksize_bits
;
2190 while (block
< last
&& block
!= EXT_MAX_BLOCKS
) {
2192 /* find extent for this block */
2193 down_read(&EXT4_I(inode
)->i_data_sem
);
2195 path
= ext4_find_extent(inode
, block
, &path
, 0);
2197 up_read(&EXT4_I(inode
)->i_data_sem
);
2198 err
= PTR_ERR(path
);
2203 depth
= ext_depth(inode
);
2204 if (unlikely(path
[depth
].p_hdr
== NULL
)) {
2205 up_read(&EXT4_I(inode
)->i_data_sem
);
2206 EXT4_ERROR_INODE(inode
, "path[%d].p_hdr == NULL", depth
);
2207 err
= -EFSCORRUPTED
;
2210 ex
= path
[depth
].p_ext
;
2211 next
= ext4_ext_next_allocated_block(path
);
2216 /* there is no extent yet, so try to allocate
2217 * all requested space */
2220 } else if (le32_to_cpu(ex
->ee_block
) > block
) {
2221 /* need to allocate space before found extent */
2223 end
= le32_to_cpu(ex
->ee_block
);
2224 if (block
+ num
< end
)
2226 } else if (block
>= le32_to_cpu(ex
->ee_block
)
2227 + ext4_ext_get_actual_len(ex
)) {
2228 /* need to allocate space after found extent */
2233 } else if (block
>= le32_to_cpu(ex
->ee_block
)) {
2235 * some part of requested space is covered
2239 end
= le32_to_cpu(ex
->ee_block
)
2240 + ext4_ext_get_actual_len(ex
);
2241 if (block
+ num
< end
)
2247 BUG_ON(end
<= start
);
2251 es
.es_len
= end
- start
;
2254 es
.es_lblk
= le32_to_cpu(ex
->ee_block
);
2255 es
.es_len
= ext4_ext_get_actual_len(ex
);
2256 es
.es_pblk
= ext4_ext_pblock(ex
);
2257 if (ext4_ext_is_unwritten(ex
))
2258 flags
|= FIEMAP_EXTENT_UNWRITTEN
;
2262 * Find delayed extent and update es accordingly. We call
2263 * it even in !exists case to find out whether es is the
2264 * last existing extent or not.
2266 next_del
= ext4_find_delayed_extent(inode
, &es
);
2267 if (!exists
&& next_del
) {
2269 flags
|= (FIEMAP_EXTENT_DELALLOC
|
2270 FIEMAP_EXTENT_UNKNOWN
);
2272 up_read(&EXT4_I(inode
)->i_data_sem
);
2274 if (unlikely(es
.es_len
== 0)) {
2275 EXT4_ERROR_INODE(inode
, "es.es_len == 0");
2276 err
= -EFSCORRUPTED
;
2281 * This is possible iff next == next_del == EXT_MAX_BLOCKS.
2282 * we need to check next == EXT_MAX_BLOCKS because it is
2283 * possible that an extent is with unwritten and delayed
2284 * status due to when an extent is delayed allocated and
2285 * is allocated by fallocate status tree will track both of
2288 * So we could return a unwritten and delayed extent, and
2289 * its block is equal to 'next'.
2291 if (next
== next_del
&& next
== EXT_MAX_BLOCKS
) {
2292 flags
|= FIEMAP_EXTENT_LAST
;
2293 if (unlikely(next_del
!= EXT_MAX_BLOCKS
||
2294 next
!= EXT_MAX_BLOCKS
)) {
2295 EXT4_ERROR_INODE(inode
,
2296 "next extent == %u, next "
2297 "delalloc extent = %u",
2299 err
= -EFSCORRUPTED
;
2305 err
= fiemap_fill_next_extent(fieinfo
,
2306 (__u64
)es
.es_lblk
<< blksize_bits
,
2307 (__u64
)es
.es_pblk
<< blksize_bits
,
2308 (__u64
)es
.es_len
<< blksize_bits
,
2318 block
= es
.es_lblk
+ es
.es_len
;
2321 ext4_ext_drop_refs(path
);
2327 * ext4_ext_determine_hole - determine hole around given block
2328 * @inode: inode we lookup in
2329 * @path: path in extent tree to @lblk
2330 * @lblk: pointer to logical block around which we want to determine hole
2332 * Determine hole length (and start if easily possible) around given logical
2333 * block. We don't try too hard to find the beginning of the hole but @path
2334 * actually points to extent before @lblk, we provide it.
2336 * The function returns the length of a hole starting at @lblk. We update @lblk
2337 * to the beginning of the hole if we managed to find it.
2339 static ext4_lblk_t
ext4_ext_determine_hole(struct inode
*inode
,
2340 struct ext4_ext_path
*path
,
2343 int depth
= ext_depth(inode
);
2344 struct ext4_extent
*ex
;
2347 ex
= path
[depth
].p_ext
;
2349 /* there is no extent yet, so gap is [0;-] */
2351 len
= EXT_MAX_BLOCKS
;
2352 } else if (*lblk
< le32_to_cpu(ex
->ee_block
)) {
2353 len
= le32_to_cpu(ex
->ee_block
) - *lblk
;
2354 } else if (*lblk
>= le32_to_cpu(ex
->ee_block
)
2355 + ext4_ext_get_actual_len(ex
)) {
2358 *lblk
= le32_to_cpu(ex
->ee_block
) + ext4_ext_get_actual_len(ex
);
2359 next
= ext4_ext_next_allocated_block(path
);
2360 BUG_ON(next
== *lblk
);
2369 * ext4_ext_put_gap_in_cache:
2370 * calculate boundaries of the gap that the requested block fits into
2371 * and cache this gap
2374 ext4_ext_put_gap_in_cache(struct inode
*inode
, ext4_lblk_t hole_start
,
2375 ext4_lblk_t hole_len
)
2377 struct extent_status es
;
2379 ext4_es_find_delayed_extent_range(inode
, hole_start
,
2380 hole_start
+ hole_len
- 1, &es
);
2382 /* There's delayed extent containing lblock? */
2383 if (es
.es_lblk
<= hole_start
)
2385 hole_len
= min(es
.es_lblk
- hole_start
, hole_len
);
2387 ext_debug(" -> %u:%u\n", hole_start
, hole_len
);
2388 ext4_es_insert_extent(inode
, hole_start
, hole_len
, ~0,
2389 EXTENT_STATUS_HOLE
);
2394 * removes index from the index block.
2396 static int ext4_ext_rm_idx(handle_t
*handle
, struct inode
*inode
,
2397 struct ext4_ext_path
*path
, int depth
)
2402 /* free index block */
2404 path
= path
+ depth
;
2405 leaf
= ext4_idx_pblock(path
->p_idx
);
2406 if (unlikely(path
->p_hdr
->eh_entries
== 0)) {
2407 EXT4_ERROR_INODE(inode
, "path->p_hdr->eh_entries == 0");
2408 return -EFSCORRUPTED
;
2410 err
= ext4_ext_get_access(handle
, inode
, path
);
2414 if (path
->p_idx
!= EXT_LAST_INDEX(path
->p_hdr
)) {
2415 int len
= EXT_LAST_INDEX(path
->p_hdr
) - path
->p_idx
;
2416 len
*= sizeof(struct ext4_extent_idx
);
2417 memmove(path
->p_idx
, path
->p_idx
+ 1, len
);
2420 le16_add_cpu(&path
->p_hdr
->eh_entries
, -1);
2421 err
= ext4_ext_dirty(handle
, inode
, path
);
2424 ext_debug("index is empty, remove it, free block %llu\n", leaf
);
2425 trace_ext4_ext_rm_idx(inode
, leaf
);
2427 ext4_free_blocks(handle
, inode
, NULL
, leaf
, 1,
2428 EXT4_FREE_BLOCKS_METADATA
| EXT4_FREE_BLOCKS_FORGET
);
2430 while (--depth
>= 0) {
2431 if (path
->p_idx
!= EXT_FIRST_INDEX(path
->p_hdr
))
2434 err
= ext4_ext_get_access(handle
, inode
, path
);
2437 path
->p_idx
->ei_block
= (path
+1)->p_idx
->ei_block
;
2438 err
= ext4_ext_dirty(handle
, inode
, path
);
2446 * ext4_ext_calc_credits_for_single_extent:
2447 * This routine returns max. credits that needed to insert an extent
2448 * to the extent tree.
2449 * When pass the actual path, the caller should calculate credits
2452 int ext4_ext_calc_credits_for_single_extent(struct inode
*inode
, int nrblocks
,
2453 struct ext4_ext_path
*path
)
2456 int depth
= ext_depth(inode
);
2459 /* probably there is space in leaf? */
2460 if (le16_to_cpu(path
[depth
].p_hdr
->eh_entries
)
2461 < le16_to_cpu(path
[depth
].p_hdr
->eh_max
)) {
2464 * There are some space in the leaf tree, no
2465 * need to account for leaf block credit
2467 * bitmaps and block group descriptor blocks
2468 * and other metadata blocks still need to be
2471 /* 1 bitmap, 1 block group descriptor */
2472 ret
= 2 + EXT4_META_TRANS_BLOCKS(inode
->i_sb
);
2477 return ext4_chunk_trans_blocks(inode
, nrblocks
);
2481 * How many index/leaf blocks need to change/allocate to add @extents extents?
2483 * If we add a single extent, then in the worse case, each tree level
2484 * index/leaf need to be changed in case of the tree split.
2486 * If more extents are inserted, they could cause the whole tree split more
2487 * than once, but this is really rare.
2489 int ext4_ext_index_trans_blocks(struct inode
*inode
, int extents
)
2494 /* If we are converting the inline data, only one is needed here. */
2495 if (ext4_has_inline_data(inode
))
2498 depth
= ext_depth(inode
);
2508 static inline int get_default_free_blocks_flags(struct inode
*inode
)
2510 if (S_ISDIR(inode
->i_mode
) || S_ISLNK(inode
->i_mode
) ||
2511 ext4_test_inode_flag(inode
, EXT4_INODE_EA_INODE
))
2512 return EXT4_FREE_BLOCKS_METADATA
| EXT4_FREE_BLOCKS_FORGET
;
2513 else if (ext4_should_journal_data(inode
))
2514 return EXT4_FREE_BLOCKS_FORGET
;
2518 static int ext4_remove_blocks(handle_t
*handle
, struct inode
*inode
,
2519 struct ext4_extent
*ex
,
2520 long long *partial_cluster
,
2521 ext4_lblk_t from
, ext4_lblk_t to
)
2523 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
2524 unsigned short ee_len
= ext4_ext_get_actual_len(ex
);
2526 int flags
= get_default_free_blocks_flags(inode
);
2529 * For bigalloc file systems, we never free a partial cluster
2530 * at the beginning of the extent. Instead, we make a note
2531 * that we tried freeing the cluster, and check to see if we
2532 * need to free it on a subsequent call to ext4_remove_blocks,
2533 * or at the end of ext4_ext_rm_leaf or ext4_ext_remove_space.
2535 flags
|= EXT4_FREE_BLOCKS_NOFREE_FIRST_CLUSTER
;
2537 trace_ext4_remove_blocks(inode
, ex
, from
, to
, *partial_cluster
);
2539 * If we have a partial cluster, and it's different from the
2540 * cluster of the last block, we need to explicitly free the
2541 * partial cluster here.
2543 pblk
= ext4_ext_pblock(ex
) + ee_len
- 1;
2544 if (*partial_cluster
> 0 &&
2545 *partial_cluster
!= (long long) EXT4_B2C(sbi
, pblk
)) {
2546 ext4_free_blocks(handle
, inode
, NULL
,
2547 EXT4_C2B(sbi
, *partial_cluster
),
2548 sbi
->s_cluster_ratio
, flags
);
2549 *partial_cluster
= 0;
2552 #ifdef EXTENTS_STATS
2554 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
2555 spin_lock(&sbi
->s_ext_stats_lock
);
2556 sbi
->s_ext_blocks
+= ee_len
;
2557 sbi
->s_ext_extents
++;
2558 if (ee_len
< sbi
->s_ext_min
)
2559 sbi
->s_ext_min
= ee_len
;
2560 if (ee_len
> sbi
->s_ext_max
)
2561 sbi
->s_ext_max
= ee_len
;
2562 if (ext_depth(inode
) > sbi
->s_depth_max
)
2563 sbi
->s_depth_max
= ext_depth(inode
);
2564 spin_unlock(&sbi
->s_ext_stats_lock
);
2567 if (from
>= le32_to_cpu(ex
->ee_block
)
2568 && to
== le32_to_cpu(ex
->ee_block
) + ee_len
- 1) {
2571 long long first_cluster
;
2573 num
= le32_to_cpu(ex
->ee_block
) + ee_len
- from
;
2574 pblk
= ext4_ext_pblock(ex
) + ee_len
- num
;
2576 * Usually we want to free partial cluster at the end of the
2577 * extent, except for the situation when the cluster is still
2578 * used by any other extent (partial_cluster is negative).
2580 if (*partial_cluster
< 0 &&
2581 *partial_cluster
== -(long long) EXT4_B2C(sbi
, pblk
+num
-1))
2582 flags
|= EXT4_FREE_BLOCKS_NOFREE_LAST_CLUSTER
;
2584 ext_debug("free last %u blocks starting %llu partial %lld\n",
2585 num
, pblk
, *partial_cluster
);
2586 ext4_free_blocks(handle
, inode
, NULL
, pblk
, num
, flags
);
2588 * If the block range to be freed didn't start at the
2589 * beginning of a cluster, and we removed the entire
2590 * extent and the cluster is not used by any other extent,
2591 * save the partial cluster here, since we might need to
2592 * delete if we determine that the truncate or punch hole
2593 * operation has removed all of the blocks in the cluster.
2594 * If that cluster is used by another extent, preserve its
2595 * negative value so it isn't freed later on.
2597 * If the whole extent wasn't freed, we've reached the
2598 * start of the truncated/punched region and have finished
2599 * removing blocks. If there's a partial cluster here it's
2600 * shared with the remainder of the extent and is no longer
2601 * a candidate for removal.
2603 if (EXT4_PBLK_COFF(sbi
, pblk
) && ee_len
== num
) {
2604 first_cluster
= (long long) EXT4_B2C(sbi
, pblk
);
2605 if (first_cluster
!= -*partial_cluster
)
2606 *partial_cluster
= first_cluster
;
2608 *partial_cluster
= 0;
2611 ext4_error(sbi
->s_sb
, "strange request: removal(2) "
2613 from
, to
, le32_to_cpu(ex
->ee_block
), ee_len
);
2619 * ext4_ext_rm_leaf() Removes the extents associated with the
2620 * blocks appearing between "start" and "end". Both "start"
2621 * and "end" must appear in the same extent or EIO is returned.
2623 * @handle: The journal handle
2624 * @inode: The files inode
2625 * @path: The path to the leaf
2626 * @partial_cluster: The cluster which we'll have to free if all extents
2627 * has been released from it. However, if this value is
2628 * negative, it's a cluster just to the right of the
2629 * punched region and it must not be freed.
2630 * @start: The first block to remove
2631 * @end: The last block to remove
2634 ext4_ext_rm_leaf(handle_t
*handle
, struct inode
*inode
,
2635 struct ext4_ext_path
*path
,
2636 long long *partial_cluster
,
2637 ext4_lblk_t start
, ext4_lblk_t end
)
2639 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
2640 int err
= 0, correct_index
= 0;
2641 int depth
= ext_depth(inode
), credits
;
2642 struct ext4_extent_header
*eh
;
2645 ext4_lblk_t ex_ee_block
;
2646 unsigned short ex_ee_len
;
2647 unsigned unwritten
= 0;
2648 struct ext4_extent
*ex
;
2651 /* the header must be checked already in ext4_ext_remove_space() */
2652 ext_debug("truncate since %u in leaf to %u\n", start
, end
);
2653 if (!path
[depth
].p_hdr
)
2654 path
[depth
].p_hdr
= ext_block_hdr(path
[depth
].p_bh
);
2655 eh
= path
[depth
].p_hdr
;
2656 if (unlikely(path
[depth
].p_hdr
== NULL
)) {
2657 EXT4_ERROR_INODE(inode
, "path[%d].p_hdr == NULL", depth
);
2658 return -EFSCORRUPTED
;
2660 /* find where to start removing */
2661 ex
= path
[depth
].p_ext
;
2663 ex
= EXT_LAST_EXTENT(eh
);
2665 ex_ee_block
= le32_to_cpu(ex
->ee_block
);
2666 ex_ee_len
= ext4_ext_get_actual_len(ex
);
2668 trace_ext4_ext_rm_leaf(inode
, start
, ex
, *partial_cluster
);
2670 while (ex
>= EXT_FIRST_EXTENT(eh
) &&
2671 ex_ee_block
+ ex_ee_len
> start
) {
2673 if (ext4_ext_is_unwritten(ex
))
2678 ext_debug("remove ext %u:[%d]%d\n", ex_ee_block
,
2679 unwritten
, ex_ee_len
);
2680 path
[depth
].p_ext
= ex
;
2682 a
= ex_ee_block
> start
? ex_ee_block
: start
;
2683 b
= ex_ee_block
+ex_ee_len
- 1 < end
?
2684 ex_ee_block
+ex_ee_len
- 1 : end
;
2686 ext_debug(" border %u:%u\n", a
, b
);
2688 /* If this extent is beyond the end of the hole, skip it */
2689 if (end
< ex_ee_block
) {
2691 * We're going to skip this extent and move to another,
2692 * so note that its first cluster is in use to avoid
2693 * freeing it when removing blocks. Eventually, the
2694 * right edge of the truncated/punched region will
2695 * be just to the left.
2697 if (sbi
->s_cluster_ratio
> 1) {
2698 pblk
= ext4_ext_pblock(ex
);
2700 -(long long) EXT4_B2C(sbi
, pblk
);
2703 ex_ee_block
= le32_to_cpu(ex
->ee_block
);
2704 ex_ee_len
= ext4_ext_get_actual_len(ex
);
2706 } else if (b
!= ex_ee_block
+ ex_ee_len
- 1) {
2707 EXT4_ERROR_INODE(inode
,
2708 "can not handle truncate %u:%u "
2710 start
, end
, ex_ee_block
,
2711 ex_ee_block
+ ex_ee_len
- 1);
2712 err
= -EFSCORRUPTED
;
2714 } else if (a
!= ex_ee_block
) {
2715 /* remove tail of the extent */
2716 num
= a
- ex_ee_block
;
2718 /* remove whole extent: excellent! */
2722 * 3 for leaf, sb, and inode plus 2 (bmap and group
2723 * descriptor) for each block group; assume two block
2724 * groups plus ex_ee_len/blocks_per_block_group for
2727 credits
= 7 + 2*(ex_ee_len
/EXT4_BLOCKS_PER_GROUP(inode
->i_sb
));
2728 if (ex
== EXT_FIRST_EXTENT(eh
)) {
2730 credits
+= (ext_depth(inode
)) + 1;
2732 credits
+= EXT4_MAXQUOTAS_TRANS_BLOCKS(inode
->i_sb
);
2734 err
= ext4_ext_truncate_extend_restart(handle
, inode
, credits
);
2738 err
= ext4_ext_get_access(handle
, inode
, path
+ depth
);
2742 err
= ext4_remove_blocks(handle
, inode
, ex
, partial_cluster
,
2748 /* this extent is removed; mark slot entirely unused */
2749 ext4_ext_store_pblock(ex
, 0);
2751 ex
->ee_len
= cpu_to_le16(num
);
2753 * Do not mark unwritten if all the blocks in the
2754 * extent have been removed.
2756 if (unwritten
&& num
)
2757 ext4_ext_mark_unwritten(ex
);
2759 * If the extent was completely released,
2760 * we need to remove it from the leaf
2763 if (end
!= EXT_MAX_BLOCKS
- 1) {
2765 * For hole punching, we need to scoot all the
2766 * extents up when an extent is removed so that
2767 * we dont have blank extents in the middle
2769 memmove(ex
, ex
+1, (EXT_LAST_EXTENT(eh
) - ex
) *
2770 sizeof(struct ext4_extent
));
2772 /* Now get rid of the one at the end */
2773 memset(EXT_LAST_EXTENT(eh
), 0,
2774 sizeof(struct ext4_extent
));
2776 le16_add_cpu(&eh
->eh_entries
, -1);
2779 err
= ext4_ext_dirty(handle
, inode
, path
+ depth
);
2783 ext_debug("new extent: %u:%u:%llu\n", ex_ee_block
, num
,
2784 ext4_ext_pblock(ex
));
2786 ex_ee_block
= le32_to_cpu(ex
->ee_block
);
2787 ex_ee_len
= ext4_ext_get_actual_len(ex
);
2790 if (correct_index
&& eh
->eh_entries
)
2791 err
= ext4_ext_correct_indexes(handle
, inode
, path
);
2794 * If there's a partial cluster and at least one extent remains in
2795 * the leaf, free the partial cluster if it isn't shared with the
2796 * current extent. If it is shared with the current extent
2797 * we zero partial_cluster because we've reached the start of the
2798 * truncated/punched region and we're done removing blocks.
2800 if (*partial_cluster
> 0 && ex
>= EXT_FIRST_EXTENT(eh
)) {
2801 pblk
= ext4_ext_pblock(ex
) + ex_ee_len
- 1;
2802 if (*partial_cluster
!= (long long) EXT4_B2C(sbi
, pblk
)) {
2803 ext4_free_blocks(handle
, inode
, NULL
,
2804 EXT4_C2B(sbi
, *partial_cluster
),
2805 sbi
->s_cluster_ratio
,
2806 get_default_free_blocks_flags(inode
));
2808 *partial_cluster
= 0;
2811 /* if this leaf is free, then we should
2812 * remove it from index block above */
2813 if (err
== 0 && eh
->eh_entries
== 0 && path
[depth
].p_bh
!= NULL
)
2814 err
= ext4_ext_rm_idx(handle
, inode
, path
, depth
);
2821 * ext4_ext_more_to_rm:
2822 * returns 1 if current index has to be freed (even partial)
2825 ext4_ext_more_to_rm(struct ext4_ext_path
*path
)
2827 BUG_ON(path
->p_idx
== NULL
);
2829 if (path
->p_idx
< EXT_FIRST_INDEX(path
->p_hdr
))
2833 * if truncate on deeper level happened, it wasn't partial,
2834 * so we have to consider current index for truncation
2836 if (le16_to_cpu(path
->p_hdr
->eh_entries
) == path
->p_block
)
2841 int ext4_ext_remove_space(struct inode
*inode
, ext4_lblk_t start
,
2844 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
2845 int depth
= ext_depth(inode
);
2846 struct ext4_ext_path
*path
= NULL
;
2847 long long partial_cluster
= 0;
2851 ext_debug("truncate since %u to %u\n", start
, end
);
2853 /* probably first extent we're gonna free will be last in block */
2854 handle
= ext4_journal_start(inode
, EXT4_HT_TRUNCATE
, depth
+ 1);
2856 return PTR_ERR(handle
);
2859 trace_ext4_ext_remove_space(inode
, start
, end
, depth
);
2862 * Check if we are removing extents inside the extent tree. If that
2863 * is the case, we are going to punch a hole inside the extent tree
2864 * so we have to check whether we need to split the extent covering
2865 * the last block to remove so we can easily remove the part of it
2866 * in ext4_ext_rm_leaf().
2868 if (end
< EXT_MAX_BLOCKS
- 1) {
2869 struct ext4_extent
*ex
;
2870 ext4_lblk_t ee_block
, ex_end
, lblk
;
2873 /* find extent for or closest extent to this block */
2874 path
= ext4_find_extent(inode
, end
, NULL
, EXT4_EX_NOCACHE
);
2876 ext4_journal_stop(handle
);
2877 return PTR_ERR(path
);
2879 depth
= ext_depth(inode
);
2880 /* Leaf not may not exist only if inode has no blocks at all */
2881 ex
= path
[depth
].p_ext
;
2884 EXT4_ERROR_INODE(inode
,
2885 "path[%d].p_hdr == NULL",
2887 err
= -EFSCORRUPTED
;
2892 ee_block
= le32_to_cpu(ex
->ee_block
);
2893 ex_end
= ee_block
+ ext4_ext_get_actual_len(ex
) - 1;
2896 * See if the last block is inside the extent, if so split
2897 * the extent at 'end' block so we can easily remove the
2898 * tail of the first part of the split extent in
2899 * ext4_ext_rm_leaf().
2901 if (end
>= ee_block
&& end
< ex_end
) {
2904 * If we're going to split the extent, note that
2905 * the cluster containing the block after 'end' is
2906 * in use to avoid freeing it when removing blocks.
2908 if (sbi
->s_cluster_ratio
> 1) {
2909 pblk
= ext4_ext_pblock(ex
) + end
- ee_block
+ 2;
2911 -(long long) EXT4_B2C(sbi
, pblk
);
2915 * Split the extent in two so that 'end' is the last
2916 * block in the first new extent. Also we should not
2917 * fail removing space due to ENOSPC so try to use
2918 * reserved block if that happens.
2920 err
= ext4_force_split_extent_at(handle
, inode
, &path
,
2925 } else if (sbi
->s_cluster_ratio
> 1 && end
>= ex_end
) {
2927 * If there's an extent to the right its first cluster
2928 * contains the immediate right boundary of the
2929 * truncated/punched region. Set partial_cluster to
2930 * its negative value so it won't be freed if shared
2931 * with the current extent. The end < ee_block case
2932 * is handled in ext4_ext_rm_leaf().
2935 err
= ext4_ext_search_right(inode
, path
, &lblk
, &pblk
,
2941 -(long long) EXT4_B2C(sbi
, pblk
);
2945 * We start scanning from right side, freeing all the blocks
2946 * after i_size and walking into the tree depth-wise.
2948 depth
= ext_depth(inode
);
2953 le16_to_cpu(path
[k
].p_hdr
->eh_entries
)+1;
2955 path
= kzalloc(sizeof(struct ext4_ext_path
) * (depth
+ 1),
2958 ext4_journal_stop(handle
);
2961 path
[0].p_maxdepth
= path
[0].p_depth
= depth
;
2962 path
[0].p_hdr
= ext_inode_hdr(inode
);
2965 if (ext4_ext_check(inode
, path
[0].p_hdr
, depth
, 0)) {
2966 err
= -EFSCORRUPTED
;
2972 while (i
>= 0 && err
== 0) {
2974 /* this is leaf block */
2975 err
= ext4_ext_rm_leaf(handle
, inode
, path
,
2976 &partial_cluster
, start
,
2978 /* root level has p_bh == NULL, brelse() eats this */
2979 brelse(path
[i
].p_bh
);
2980 path
[i
].p_bh
= NULL
;
2985 /* this is index block */
2986 if (!path
[i
].p_hdr
) {
2987 ext_debug("initialize header\n");
2988 path
[i
].p_hdr
= ext_block_hdr(path
[i
].p_bh
);
2991 if (!path
[i
].p_idx
) {
2992 /* this level hasn't been touched yet */
2993 path
[i
].p_idx
= EXT_LAST_INDEX(path
[i
].p_hdr
);
2994 path
[i
].p_block
= le16_to_cpu(path
[i
].p_hdr
->eh_entries
)+1;
2995 ext_debug("init index ptr: hdr 0x%p, num %d\n",
2997 le16_to_cpu(path
[i
].p_hdr
->eh_entries
));
2999 /* we were already here, see at next index */
3003 ext_debug("level %d - index, first 0x%p, cur 0x%p\n",
3004 i
, EXT_FIRST_INDEX(path
[i
].p_hdr
),
3006 if (ext4_ext_more_to_rm(path
+ i
)) {
3007 struct buffer_head
*bh
;
3008 /* go to the next level */
3009 ext_debug("move to level %d (block %llu)\n",
3010 i
+ 1, ext4_idx_pblock(path
[i
].p_idx
));
3011 memset(path
+ i
+ 1, 0, sizeof(*path
));
3012 bh
= read_extent_tree_block(inode
,
3013 ext4_idx_pblock(path
[i
].p_idx
), depth
- i
- 1,
3016 /* should we reset i_size? */
3020 /* Yield here to deal with large extent trees.
3021 * Should be a no-op if we did IO above. */
3023 if (WARN_ON(i
+ 1 > depth
)) {
3024 err
= -EFSCORRUPTED
;
3027 path
[i
+ 1].p_bh
= bh
;
3029 /* save actual number of indexes since this
3030 * number is changed at the next iteration */
3031 path
[i
].p_block
= le16_to_cpu(path
[i
].p_hdr
->eh_entries
);
3034 /* we finished processing this index, go up */
3035 if (path
[i
].p_hdr
->eh_entries
== 0 && i
> 0) {
3036 /* index is empty, remove it;
3037 * handle must be already prepared by the
3038 * truncatei_leaf() */
3039 err
= ext4_ext_rm_idx(handle
, inode
, path
, i
);
3041 /* root level has p_bh == NULL, brelse() eats this */
3042 brelse(path
[i
].p_bh
);
3043 path
[i
].p_bh
= NULL
;
3045 ext_debug("return to level %d\n", i
);
3049 trace_ext4_ext_remove_space_done(inode
, start
, end
, depth
,
3050 partial_cluster
, path
->p_hdr
->eh_entries
);
3053 * If we still have something in the partial cluster and we have removed
3054 * even the first extent, then we should free the blocks in the partial
3055 * cluster as well. (This code will only run when there are no leaves
3056 * to the immediate left of the truncated/punched region.)
3058 if (partial_cluster
> 0 && err
== 0) {
3059 /* don't zero partial_cluster since it's not used afterwards */
3060 ext4_free_blocks(handle
, inode
, NULL
,
3061 EXT4_C2B(sbi
, partial_cluster
),
3062 sbi
->s_cluster_ratio
,
3063 get_default_free_blocks_flags(inode
));
3066 /* TODO: flexible tree reduction should be here */
3067 if (path
->p_hdr
->eh_entries
== 0) {
3069 * truncate to zero freed all the tree,
3070 * so we need to correct eh_depth
3072 err
= ext4_ext_get_access(handle
, inode
, path
);
3074 ext_inode_hdr(inode
)->eh_depth
= 0;
3075 ext_inode_hdr(inode
)->eh_max
=
3076 cpu_to_le16(ext4_ext_space_root(inode
, 0));
3077 err
= ext4_ext_dirty(handle
, inode
, path
);
3081 ext4_ext_drop_refs(path
);
3086 ext4_journal_stop(handle
);
3092 * called at mount time
3094 void ext4_ext_init(struct super_block
*sb
)
3097 * possible initialization would be here
3100 if (ext4_has_feature_extents(sb
)) {
3101 #if defined(AGGRESSIVE_TEST) || defined(CHECK_BINSEARCH) || defined(EXTENTS_STATS)
3102 printk(KERN_INFO
"EXT4-fs: file extents enabled"
3103 #ifdef AGGRESSIVE_TEST
3104 ", aggressive tests"
3106 #ifdef CHECK_BINSEARCH
3109 #ifdef EXTENTS_STATS
3114 #ifdef EXTENTS_STATS
3115 spin_lock_init(&EXT4_SB(sb
)->s_ext_stats_lock
);
3116 EXT4_SB(sb
)->s_ext_min
= 1 << 30;
3117 EXT4_SB(sb
)->s_ext_max
= 0;
3123 * called at umount time
3125 void ext4_ext_release(struct super_block
*sb
)
3127 if (!ext4_has_feature_extents(sb
))
3130 #ifdef EXTENTS_STATS
3131 if (EXT4_SB(sb
)->s_ext_blocks
&& EXT4_SB(sb
)->s_ext_extents
) {
3132 struct ext4_sb_info
*sbi
= EXT4_SB(sb
);
3133 printk(KERN_ERR
"EXT4-fs: %lu blocks in %lu extents (%lu ave)\n",
3134 sbi
->s_ext_blocks
, sbi
->s_ext_extents
,
3135 sbi
->s_ext_blocks
/ sbi
->s_ext_extents
);
3136 printk(KERN_ERR
"EXT4-fs: extents: %lu min, %lu max, max depth %lu\n",
3137 sbi
->s_ext_min
, sbi
->s_ext_max
, sbi
->s_depth_max
);
3142 static int ext4_zeroout_es(struct inode
*inode
, struct ext4_extent
*ex
)
3144 ext4_lblk_t ee_block
;
3145 ext4_fsblk_t ee_pblock
;
3146 unsigned int ee_len
;
3148 ee_block
= le32_to_cpu(ex
->ee_block
);
3149 ee_len
= ext4_ext_get_actual_len(ex
);
3150 ee_pblock
= ext4_ext_pblock(ex
);
3155 return ext4_es_insert_extent(inode
, ee_block
, ee_len
, ee_pblock
,
3156 EXTENT_STATUS_WRITTEN
);
3159 /* FIXME!! we need to try to merge to left or right after zero-out */
3160 static int ext4_ext_zeroout(struct inode
*inode
, struct ext4_extent
*ex
)
3162 ext4_fsblk_t ee_pblock
;
3163 unsigned int ee_len
;
3165 ee_len
= ext4_ext_get_actual_len(ex
);
3166 ee_pblock
= ext4_ext_pblock(ex
);
3167 return ext4_issue_zeroout(inode
, le32_to_cpu(ex
->ee_block
), ee_pblock
,
3172 * ext4_split_extent_at() splits an extent at given block.
3174 * @handle: the journal handle
3175 * @inode: the file inode
3176 * @path: the path to the extent
3177 * @split: the logical block where the extent is splitted.
3178 * @split_flags: indicates if the extent could be zeroout if split fails, and
3179 * the states(init or unwritten) of new extents.
3180 * @flags: flags used to insert new extent to extent tree.
3183 * Splits extent [a, b] into two extents [a, @split) and [@split, b], states
3184 * of which are deterimined by split_flag.
3186 * There are two cases:
3187 * a> the extent are splitted into two extent.
3188 * b> split is not needed, and just mark the extent.
3190 * return 0 on success.
3192 static int ext4_split_extent_at(handle_t
*handle
,
3193 struct inode
*inode
,
3194 struct ext4_ext_path
**ppath
,
3199 struct ext4_ext_path
*path
= *ppath
;
3200 ext4_fsblk_t newblock
;
3201 ext4_lblk_t ee_block
;
3202 struct ext4_extent
*ex
, newex
, orig_ex
, zero_ex
;
3203 struct ext4_extent
*ex2
= NULL
;
3204 unsigned int ee_len
, depth
;
3207 BUG_ON((split_flag
& (EXT4_EXT_DATA_VALID1
| EXT4_EXT_DATA_VALID2
)) ==
3208 (EXT4_EXT_DATA_VALID1
| EXT4_EXT_DATA_VALID2
));
3210 ext_debug("ext4_split_extents_at: inode %lu, logical"
3211 "block %llu\n", inode
->i_ino
, (unsigned long long)split
);
3213 ext4_ext_show_leaf(inode
, path
);
3215 depth
= ext_depth(inode
);
3216 ex
= path
[depth
].p_ext
;
3217 ee_block
= le32_to_cpu(ex
->ee_block
);
3218 ee_len
= ext4_ext_get_actual_len(ex
);
3219 newblock
= split
- ee_block
+ ext4_ext_pblock(ex
);
3221 BUG_ON(split
< ee_block
|| split
>= (ee_block
+ ee_len
));
3222 BUG_ON(!ext4_ext_is_unwritten(ex
) &&
3223 split_flag
& (EXT4_EXT_MAY_ZEROOUT
|
3224 EXT4_EXT_MARK_UNWRIT1
|
3225 EXT4_EXT_MARK_UNWRIT2
));
3227 err
= ext4_ext_get_access(handle
, inode
, path
+ depth
);
3231 if (split
== ee_block
) {
3233 * case b: block @split is the block that the extent begins with
3234 * then we just change the state of the extent, and splitting
3237 if (split_flag
& EXT4_EXT_MARK_UNWRIT2
)
3238 ext4_ext_mark_unwritten(ex
);
3240 ext4_ext_mark_initialized(ex
);
3242 if (!(flags
& EXT4_GET_BLOCKS_PRE_IO
))
3243 ext4_ext_try_to_merge(handle
, inode
, path
, ex
);
3245 err
= ext4_ext_dirty(handle
, inode
, path
+ path
->p_depth
);
3250 memcpy(&orig_ex
, ex
, sizeof(orig_ex
));
3251 ex
->ee_len
= cpu_to_le16(split
- ee_block
);
3252 if (split_flag
& EXT4_EXT_MARK_UNWRIT1
)
3253 ext4_ext_mark_unwritten(ex
);
3256 * path may lead to new leaf, not to original leaf any more
3257 * after ext4_ext_insert_extent() returns,
3259 err
= ext4_ext_dirty(handle
, inode
, path
+ depth
);
3261 goto fix_extent_len
;
3264 ex2
->ee_block
= cpu_to_le32(split
);
3265 ex2
->ee_len
= cpu_to_le16(ee_len
- (split
- ee_block
));
3266 ext4_ext_store_pblock(ex2
, newblock
);
3267 if (split_flag
& EXT4_EXT_MARK_UNWRIT2
)
3268 ext4_ext_mark_unwritten(ex2
);
3270 err
= ext4_ext_insert_extent(handle
, inode
, ppath
, &newex
, flags
);
3271 if (err
== -ENOSPC
&& (EXT4_EXT_MAY_ZEROOUT
& split_flag
)) {
3272 if (split_flag
& (EXT4_EXT_DATA_VALID1
|EXT4_EXT_DATA_VALID2
)) {
3273 if (split_flag
& EXT4_EXT_DATA_VALID1
) {
3274 err
= ext4_ext_zeroout(inode
, ex2
);
3275 zero_ex
.ee_block
= ex2
->ee_block
;
3276 zero_ex
.ee_len
= cpu_to_le16(
3277 ext4_ext_get_actual_len(ex2
));
3278 ext4_ext_store_pblock(&zero_ex
,
3279 ext4_ext_pblock(ex2
));
3281 err
= ext4_ext_zeroout(inode
, ex
);
3282 zero_ex
.ee_block
= ex
->ee_block
;
3283 zero_ex
.ee_len
= cpu_to_le16(
3284 ext4_ext_get_actual_len(ex
));
3285 ext4_ext_store_pblock(&zero_ex
,
3286 ext4_ext_pblock(ex
));
3289 err
= ext4_ext_zeroout(inode
, &orig_ex
);
3290 zero_ex
.ee_block
= orig_ex
.ee_block
;
3291 zero_ex
.ee_len
= cpu_to_le16(
3292 ext4_ext_get_actual_len(&orig_ex
));
3293 ext4_ext_store_pblock(&zero_ex
,
3294 ext4_ext_pblock(&orig_ex
));
3298 goto fix_extent_len
;
3299 /* update the extent length and mark as initialized */
3300 ex
->ee_len
= cpu_to_le16(ee_len
);
3301 ext4_ext_try_to_merge(handle
, inode
, path
, ex
);
3302 err
= ext4_ext_dirty(handle
, inode
, path
+ path
->p_depth
);
3304 goto fix_extent_len
;
3306 /* update extent status tree */
3307 err
= ext4_zeroout_es(inode
, &zero_ex
);
3311 goto fix_extent_len
;
3314 ext4_ext_show_leaf(inode
, path
);
3318 ex
->ee_len
= orig_ex
.ee_len
;
3319 ext4_ext_dirty(handle
, inode
, path
+ path
->p_depth
);
3324 * ext4_split_extents() splits an extent and mark extent which is covered
3325 * by @map as split_flags indicates
3327 * It may result in splitting the extent into multiple extents (up to three)
3328 * There are three possibilities:
3329 * a> There is no split required
3330 * b> Splits in two extents: Split is happening at either end of the extent
3331 * c> Splits in three extents: Somone is splitting in middle of the extent
3334 static int ext4_split_extent(handle_t
*handle
,
3335 struct inode
*inode
,
3336 struct ext4_ext_path
**ppath
,
3337 struct ext4_map_blocks
*map
,
3341 struct ext4_ext_path
*path
= *ppath
;
3342 ext4_lblk_t ee_block
;
3343 struct ext4_extent
*ex
;
3344 unsigned int ee_len
, depth
;
3347 int split_flag1
, flags1
;
3348 int allocated
= map
->m_len
;
3350 depth
= ext_depth(inode
);
3351 ex
= path
[depth
].p_ext
;
3352 ee_block
= le32_to_cpu(ex
->ee_block
);
3353 ee_len
= ext4_ext_get_actual_len(ex
);
3354 unwritten
= ext4_ext_is_unwritten(ex
);
3356 if (map
->m_lblk
+ map
->m_len
< ee_block
+ ee_len
) {
3357 split_flag1
= split_flag
& EXT4_EXT_MAY_ZEROOUT
;
3358 flags1
= flags
| EXT4_GET_BLOCKS_PRE_IO
;
3360 split_flag1
|= EXT4_EXT_MARK_UNWRIT1
|
3361 EXT4_EXT_MARK_UNWRIT2
;
3362 if (split_flag
& EXT4_EXT_DATA_VALID2
)
3363 split_flag1
|= EXT4_EXT_DATA_VALID1
;
3364 err
= ext4_split_extent_at(handle
, inode
, ppath
,
3365 map
->m_lblk
+ map
->m_len
, split_flag1
, flags1
);
3369 allocated
= ee_len
- (map
->m_lblk
- ee_block
);
3372 * Update path is required because previous ext4_split_extent_at() may
3373 * result in split of original leaf or extent zeroout.
3375 path
= ext4_find_extent(inode
, map
->m_lblk
, ppath
, 0);
3377 return PTR_ERR(path
);
3378 depth
= ext_depth(inode
);
3379 ex
= path
[depth
].p_ext
;
3381 EXT4_ERROR_INODE(inode
, "unexpected hole at %lu",
3382 (unsigned long) map
->m_lblk
);
3383 return -EFSCORRUPTED
;
3385 unwritten
= ext4_ext_is_unwritten(ex
);
3388 if (map
->m_lblk
>= ee_block
) {
3389 split_flag1
= split_flag
& EXT4_EXT_DATA_VALID2
;
3391 split_flag1
|= EXT4_EXT_MARK_UNWRIT1
;
3392 split_flag1
|= split_flag
& (EXT4_EXT_MAY_ZEROOUT
|
3393 EXT4_EXT_MARK_UNWRIT2
);
3395 err
= ext4_split_extent_at(handle
, inode
, ppath
,
3396 map
->m_lblk
, split_flag1
, flags
);
3401 ext4_ext_show_leaf(inode
, path
);
3403 return err
? err
: allocated
;
3407 * This function is called by ext4_ext_map_blocks() if someone tries to write
3408 * to an unwritten extent. It may result in splitting the unwritten
3409 * extent into multiple extents (up to three - one initialized and two
3411 * There are three possibilities:
3412 * a> There is no split required: Entire extent should be initialized
3413 * b> Splits in two extents: Write is happening at either end of the extent
3414 * c> Splits in three extents: Somone is writing in middle of the extent
3417 * - The extent pointed to by 'path' is unwritten.
3418 * - The extent pointed to by 'path' contains a superset
3419 * of the logical span [map->m_lblk, map->m_lblk + map->m_len).
3421 * Post-conditions on success:
3422 * - the returned value is the number of blocks beyond map->l_lblk
3423 * that are allocated and initialized.
3424 * It is guaranteed to be >= map->m_len.
3426 static int ext4_ext_convert_to_initialized(handle_t
*handle
,
3427 struct inode
*inode
,
3428 struct ext4_map_blocks
*map
,
3429 struct ext4_ext_path
**ppath
,
3432 struct ext4_ext_path
*path
= *ppath
;
3433 struct ext4_sb_info
*sbi
;
3434 struct ext4_extent_header
*eh
;
3435 struct ext4_map_blocks split_map
;
3436 struct ext4_extent zero_ex1
, zero_ex2
;
3437 struct ext4_extent
*ex
, *abut_ex
;
3438 ext4_lblk_t ee_block
, eof_block
;
3439 unsigned int ee_len
, depth
, map_len
= map
->m_len
;
3440 int allocated
= 0, max_zeroout
= 0;
3442 int split_flag
= EXT4_EXT_DATA_VALID2
;
3444 ext_debug("ext4_ext_convert_to_initialized: inode %lu, logical"
3445 "block %llu, max_blocks %u\n", inode
->i_ino
,
3446 (unsigned long long)map
->m_lblk
, map_len
);
3448 sbi
= EXT4_SB(inode
->i_sb
);
3449 eof_block
= (inode
->i_size
+ inode
->i_sb
->s_blocksize
- 1) >>
3450 inode
->i_sb
->s_blocksize_bits
;
3451 if (eof_block
< map
->m_lblk
+ map_len
)
3452 eof_block
= map
->m_lblk
+ map_len
;
3454 depth
= ext_depth(inode
);
3455 eh
= path
[depth
].p_hdr
;
3456 ex
= path
[depth
].p_ext
;
3457 ee_block
= le32_to_cpu(ex
->ee_block
);
3458 ee_len
= ext4_ext_get_actual_len(ex
);
3459 zero_ex1
.ee_len
= 0;
3460 zero_ex2
.ee_len
= 0;
3462 trace_ext4_ext_convert_to_initialized_enter(inode
, map
, ex
);
3464 /* Pre-conditions */
3465 BUG_ON(!ext4_ext_is_unwritten(ex
));
3466 BUG_ON(!in_range(map
->m_lblk
, ee_block
, ee_len
));
3469 * Attempt to transfer newly initialized blocks from the currently
3470 * unwritten extent to its neighbor. This is much cheaper
3471 * than an insertion followed by a merge as those involve costly
3472 * memmove() calls. Transferring to the left is the common case in
3473 * steady state for workloads doing fallocate(FALLOC_FL_KEEP_SIZE)
3474 * followed by append writes.
3476 * Limitations of the current logic:
3477 * - L1: we do not deal with writes covering the whole extent.
3478 * This would require removing the extent if the transfer
3480 * - L2: we only attempt to merge with an extent stored in the
3481 * same extent tree node.
3483 if ((map
->m_lblk
== ee_block
) &&
3484 /* See if we can merge left */
3485 (map_len
< ee_len
) && /*L1*/
3486 (ex
> EXT_FIRST_EXTENT(eh
))) { /*L2*/
3487 ext4_lblk_t prev_lblk
;
3488 ext4_fsblk_t prev_pblk
, ee_pblk
;
3489 unsigned int prev_len
;
3492 prev_lblk
= le32_to_cpu(abut_ex
->ee_block
);
3493 prev_len
= ext4_ext_get_actual_len(abut_ex
);
3494 prev_pblk
= ext4_ext_pblock(abut_ex
);
3495 ee_pblk
= ext4_ext_pblock(ex
);
3498 * A transfer of blocks from 'ex' to 'abut_ex' is allowed
3499 * upon those conditions:
3500 * - C1: abut_ex is initialized,
3501 * - C2: abut_ex is logically abutting ex,
3502 * - C3: abut_ex is physically abutting ex,
3503 * - C4: abut_ex can receive the additional blocks without
3504 * overflowing the (initialized) length limit.
3506 if ((!ext4_ext_is_unwritten(abut_ex
)) && /*C1*/
3507 ((prev_lblk
+ prev_len
) == ee_block
) && /*C2*/
3508 ((prev_pblk
+ prev_len
) == ee_pblk
) && /*C3*/
3509 (prev_len
< (EXT_INIT_MAX_LEN
- map_len
))) { /*C4*/
3510 err
= ext4_ext_get_access(handle
, inode
, path
+ depth
);
3514 trace_ext4_ext_convert_to_initialized_fastpath(inode
,
3517 /* Shift the start of ex by 'map_len' blocks */
3518 ex
->ee_block
= cpu_to_le32(ee_block
+ map_len
);
3519 ext4_ext_store_pblock(ex
, ee_pblk
+ map_len
);
3520 ex
->ee_len
= cpu_to_le16(ee_len
- map_len
);
3521 ext4_ext_mark_unwritten(ex
); /* Restore the flag */
3523 /* Extend abut_ex by 'map_len' blocks */
3524 abut_ex
->ee_len
= cpu_to_le16(prev_len
+ map_len
);
3526 /* Result: number of initialized blocks past m_lblk */
3527 allocated
= map_len
;
3529 } else if (((map
->m_lblk
+ map_len
) == (ee_block
+ ee_len
)) &&
3530 (map_len
< ee_len
) && /*L1*/
3531 ex
< EXT_LAST_EXTENT(eh
)) { /*L2*/
3532 /* See if we can merge right */
3533 ext4_lblk_t next_lblk
;
3534 ext4_fsblk_t next_pblk
, ee_pblk
;
3535 unsigned int next_len
;
3538 next_lblk
= le32_to_cpu(abut_ex
->ee_block
);
3539 next_len
= ext4_ext_get_actual_len(abut_ex
);
3540 next_pblk
= ext4_ext_pblock(abut_ex
);
3541 ee_pblk
= ext4_ext_pblock(ex
);
3544 * A transfer of blocks from 'ex' to 'abut_ex' is allowed
3545 * upon those conditions:
3546 * - C1: abut_ex is initialized,
3547 * - C2: abut_ex is logically abutting ex,
3548 * - C3: abut_ex is physically abutting ex,
3549 * - C4: abut_ex can receive the additional blocks without
3550 * overflowing the (initialized) length limit.
3552 if ((!ext4_ext_is_unwritten(abut_ex
)) && /*C1*/
3553 ((map
->m_lblk
+ map_len
) == next_lblk
) && /*C2*/
3554 ((ee_pblk
+ ee_len
) == next_pblk
) && /*C3*/
3555 (next_len
< (EXT_INIT_MAX_LEN
- map_len
))) { /*C4*/
3556 err
= ext4_ext_get_access(handle
, inode
, path
+ depth
);
3560 trace_ext4_ext_convert_to_initialized_fastpath(inode
,
3563 /* Shift the start of abut_ex by 'map_len' blocks */
3564 abut_ex
->ee_block
= cpu_to_le32(next_lblk
- map_len
);
3565 ext4_ext_store_pblock(abut_ex
, next_pblk
- map_len
);
3566 ex
->ee_len
= cpu_to_le16(ee_len
- map_len
);
3567 ext4_ext_mark_unwritten(ex
); /* Restore the flag */
3569 /* Extend abut_ex by 'map_len' blocks */
3570 abut_ex
->ee_len
= cpu_to_le16(next_len
+ map_len
);
3572 /* Result: number of initialized blocks past m_lblk */
3573 allocated
= map_len
;
3577 /* Mark the block containing both extents as dirty */
3578 ext4_ext_dirty(handle
, inode
, path
+ depth
);
3580 /* Update path to point to the right extent */
3581 path
[depth
].p_ext
= abut_ex
;
3584 allocated
= ee_len
- (map
->m_lblk
- ee_block
);
3586 WARN_ON(map
->m_lblk
< ee_block
);
3588 * It is safe to convert extent to initialized via explicit
3589 * zeroout only if extent is fully inside i_size or new_size.
3591 split_flag
|= ee_block
+ ee_len
<= eof_block
? EXT4_EXT_MAY_ZEROOUT
: 0;
3593 if (EXT4_EXT_MAY_ZEROOUT
& split_flag
)
3594 max_zeroout
= sbi
->s_extent_max_zeroout_kb
>>
3595 (inode
->i_sb
->s_blocksize_bits
- 10);
3597 if (ext4_encrypted_inode(inode
))
3602 * 1. split the extent into three extents.
3603 * 2. split the extent into two extents, zeroout the head of the first
3605 * 3. split the extent into two extents, zeroout the tail of the second
3607 * 4. split the extent into two extents with out zeroout.
3608 * 5. no splitting needed, just possibly zeroout the head and / or the
3609 * tail of the extent.
3611 split_map
.m_lblk
= map
->m_lblk
;
3612 split_map
.m_len
= map
->m_len
;
3614 if (max_zeroout
&& (allocated
> split_map
.m_len
)) {
3615 if (allocated
<= max_zeroout
) {
3618 cpu_to_le32(split_map
.m_lblk
+
3621 cpu_to_le16(allocated
- split_map
.m_len
);
3622 ext4_ext_store_pblock(&zero_ex1
,
3623 ext4_ext_pblock(ex
) + split_map
.m_lblk
+
3624 split_map
.m_len
- ee_block
);
3625 err
= ext4_ext_zeroout(inode
, &zero_ex1
);
3628 split_map
.m_len
= allocated
;
3630 if (split_map
.m_lblk
- ee_block
+ split_map
.m_len
<
3633 if (split_map
.m_lblk
!= ee_block
) {
3634 zero_ex2
.ee_block
= ex
->ee_block
;
3635 zero_ex2
.ee_len
= cpu_to_le16(split_map
.m_lblk
-
3637 ext4_ext_store_pblock(&zero_ex2
,
3638 ext4_ext_pblock(ex
));
3639 err
= ext4_ext_zeroout(inode
, &zero_ex2
);
3644 split_map
.m_len
+= split_map
.m_lblk
- ee_block
;
3645 split_map
.m_lblk
= ee_block
;
3646 allocated
= map
->m_len
;
3650 err
= ext4_split_extent(handle
, inode
, ppath
, &split_map
, split_flag
,
3655 /* If we have gotten a failure, don't zero out status tree */
3657 err
= ext4_zeroout_es(inode
, &zero_ex1
);
3659 err
= ext4_zeroout_es(inode
, &zero_ex2
);
3661 return err
? err
: allocated
;
3665 * This function is called by ext4_ext_map_blocks() from
3666 * ext4_get_blocks_dio_write() when DIO to write
3667 * to an unwritten extent.
3669 * Writing to an unwritten extent may result in splitting the unwritten
3670 * extent into multiple initialized/unwritten extents (up to three)
3671 * There are three possibilities:
3672 * a> There is no split required: Entire extent should be unwritten
3673 * b> Splits in two extents: Write is happening at either end of the extent
3674 * c> Splits in three extents: Somone is writing in middle of the extent
3676 * This works the same way in the case of initialized -> unwritten conversion.
3678 * One of more index blocks maybe needed if the extent tree grow after
3679 * the unwritten extent split. To prevent ENOSPC occur at the IO
3680 * complete, we need to split the unwritten extent before DIO submit
3681 * the IO. The unwritten extent called at this time will be split
3682 * into three unwritten extent(at most). After IO complete, the part
3683 * being filled will be convert to initialized by the end_io callback function
3684 * via ext4_convert_unwritten_extents().
3686 * Returns the size of unwritten extent to be written on success.
3688 static int ext4_split_convert_extents(handle_t
*handle
,
3689 struct inode
*inode
,
3690 struct ext4_map_blocks
*map
,
3691 struct ext4_ext_path
**ppath
,
3694 struct ext4_ext_path
*path
= *ppath
;
3695 ext4_lblk_t eof_block
;
3696 ext4_lblk_t ee_block
;
3697 struct ext4_extent
*ex
;
3698 unsigned int ee_len
;
3699 int split_flag
= 0, depth
;
3701 ext_debug("%s: inode %lu, logical block %llu, max_blocks %u\n",
3702 __func__
, inode
->i_ino
,
3703 (unsigned long long)map
->m_lblk
, map
->m_len
);
3705 eof_block
= (inode
->i_size
+ inode
->i_sb
->s_blocksize
- 1) >>
3706 inode
->i_sb
->s_blocksize_bits
;
3707 if (eof_block
< map
->m_lblk
+ map
->m_len
)
3708 eof_block
= map
->m_lblk
+ map
->m_len
;
3710 * It is safe to convert extent to initialized via explicit
3711 * zeroout only if extent is fully insde i_size or new_size.
3713 depth
= ext_depth(inode
);
3714 ex
= path
[depth
].p_ext
;
3715 ee_block
= le32_to_cpu(ex
->ee_block
);
3716 ee_len
= ext4_ext_get_actual_len(ex
);
3718 /* Convert to unwritten */
3719 if (flags
& EXT4_GET_BLOCKS_CONVERT_UNWRITTEN
) {
3720 split_flag
|= EXT4_EXT_DATA_VALID1
;
3721 /* Convert to initialized */
3722 } else if (flags
& EXT4_GET_BLOCKS_CONVERT
) {
3723 split_flag
|= ee_block
+ ee_len
<= eof_block
?
3724 EXT4_EXT_MAY_ZEROOUT
: 0;
3725 split_flag
|= (EXT4_EXT_MARK_UNWRIT2
| EXT4_EXT_DATA_VALID2
);
3727 flags
|= EXT4_GET_BLOCKS_PRE_IO
;
3728 return ext4_split_extent(handle
, inode
, ppath
, map
, split_flag
, flags
);
3731 static int ext4_convert_unwritten_extents_endio(handle_t
*handle
,
3732 struct inode
*inode
,
3733 struct ext4_map_blocks
*map
,
3734 struct ext4_ext_path
**ppath
)
3736 struct ext4_ext_path
*path
= *ppath
;
3737 struct ext4_extent
*ex
;
3738 ext4_lblk_t ee_block
;
3739 unsigned int ee_len
;
3743 depth
= ext_depth(inode
);
3744 ex
= path
[depth
].p_ext
;
3745 ee_block
= le32_to_cpu(ex
->ee_block
);
3746 ee_len
= ext4_ext_get_actual_len(ex
);
3748 ext_debug("ext4_convert_unwritten_extents_endio: inode %lu, logical"
3749 "block %llu, max_blocks %u\n", inode
->i_ino
,
3750 (unsigned long long)ee_block
, ee_len
);
3752 /* If extent is larger than requested it is a clear sign that we still
3753 * have some extent state machine issues left. So extent_split is still
3755 * TODO: Once all related issues will be fixed this situation should be
3758 if (ee_block
!= map
->m_lblk
|| ee_len
> map
->m_len
) {
3760 ext4_warning("Inode (%ld) finished: extent logical block %llu,"
3761 " len %u; IO logical block %llu, len %u",
3762 inode
->i_ino
, (unsigned long long)ee_block
, ee_len
,
3763 (unsigned long long)map
->m_lblk
, map
->m_len
);
3765 err
= ext4_split_convert_extents(handle
, inode
, map
, ppath
,
3766 EXT4_GET_BLOCKS_CONVERT
);
3769 path
= ext4_find_extent(inode
, map
->m_lblk
, ppath
, 0);
3771 return PTR_ERR(path
);
3772 depth
= ext_depth(inode
);
3773 ex
= path
[depth
].p_ext
;
3776 err
= ext4_ext_get_access(handle
, inode
, path
+ depth
);
3779 /* first mark the extent as initialized */
3780 ext4_ext_mark_initialized(ex
);
3782 /* note: ext4_ext_correct_indexes() isn't needed here because
3783 * borders are not changed
3785 ext4_ext_try_to_merge(handle
, inode
, path
, ex
);
3787 /* Mark modified extent as dirty */
3788 err
= ext4_ext_dirty(handle
, inode
, path
+ path
->p_depth
);
3790 ext4_ext_show_leaf(inode
, path
);
3795 * Handle EOFBLOCKS_FL flag, clearing it if necessary
3797 static int check_eofblocks_fl(handle_t
*handle
, struct inode
*inode
,
3799 struct ext4_ext_path
*path
,
3803 struct ext4_extent_header
*eh
;
3804 struct ext4_extent
*last_ex
;
3806 if (!ext4_test_inode_flag(inode
, EXT4_INODE_EOFBLOCKS
))
3809 depth
= ext_depth(inode
);
3810 eh
= path
[depth
].p_hdr
;
3813 * We're going to remove EOFBLOCKS_FL entirely in future so we
3814 * do not care for this case anymore. Simply remove the flag
3815 * if there are no extents.
3817 if (unlikely(!eh
->eh_entries
))
3819 last_ex
= EXT_LAST_EXTENT(eh
);
3821 * We should clear the EOFBLOCKS_FL flag if we are writing the
3822 * last block in the last extent in the file. We test this by
3823 * first checking to see if the caller to
3824 * ext4_ext_get_blocks() was interested in the last block (or
3825 * a block beyond the last block) in the current extent. If
3826 * this turns out to be false, we can bail out from this
3827 * function immediately.
3829 if (lblk
+ len
< le32_to_cpu(last_ex
->ee_block
) +
3830 ext4_ext_get_actual_len(last_ex
))
3833 * If the caller does appear to be planning to write at or
3834 * beyond the end of the current extent, we then test to see
3835 * if the current extent is the last extent in the file, by
3836 * checking to make sure it was reached via the rightmost node
3837 * at each level of the tree.
3839 for (i
= depth
-1; i
>= 0; i
--)
3840 if (path
[i
].p_idx
!= EXT_LAST_INDEX(path
[i
].p_hdr
))
3843 ext4_clear_inode_flag(inode
, EXT4_INODE_EOFBLOCKS
);
3844 return ext4_mark_inode_dirty(handle
, inode
);
3848 * ext4_find_delalloc_range: find delayed allocated block in the given range.
3850 * Return 1 if there is a delalloc block in the range, otherwise 0.
3852 int ext4_find_delalloc_range(struct inode
*inode
,
3853 ext4_lblk_t lblk_start
,
3854 ext4_lblk_t lblk_end
)
3856 struct extent_status es
;
3858 ext4_es_find_delayed_extent_range(inode
, lblk_start
, lblk_end
, &es
);
3860 return 0; /* there is no delay extent in this tree */
3861 else if (es
.es_lblk
<= lblk_start
&&
3862 lblk_start
< es
.es_lblk
+ es
.es_len
)
3864 else if (lblk_start
<= es
.es_lblk
&& es
.es_lblk
<= lblk_end
)
3870 int ext4_find_delalloc_cluster(struct inode
*inode
, ext4_lblk_t lblk
)
3872 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
3873 ext4_lblk_t lblk_start
, lblk_end
;
3874 lblk_start
= EXT4_LBLK_CMASK(sbi
, lblk
);
3875 lblk_end
= lblk_start
+ sbi
->s_cluster_ratio
- 1;
3877 return ext4_find_delalloc_range(inode
, lblk_start
, lblk_end
);
3881 * Determines how many complete clusters (out of those specified by the 'map')
3882 * are under delalloc and were reserved quota for.
3883 * This function is called when we are writing out the blocks that were
3884 * originally written with their allocation delayed, but then the space was
3885 * allocated using fallocate() before the delayed allocation could be resolved.
3886 * The cases to look for are:
3887 * ('=' indicated delayed allocated blocks
3888 * '-' indicates non-delayed allocated blocks)
3889 * (a) partial clusters towards beginning and/or end outside of allocated range
3890 * are not delalloc'ed.
3892 * |----c---=|====c====|====c====|===-c----|
3893 * |++++++ allocated ++++++|
3894 * ==> 4 complete clusters in above example
3896 * (b) partial cluster (outside of allocated range) towards either end is
3897 * marked for delayed allocation. In this case, we will exclude that
3900 * |----====c========|========c========|
3901 * |++++++ allocated ++++++|
3902 * ==> 1 complete clusters in above example
3905 * |================c================|
3906 * |++++++ allocated ++++++|
3907 * ==> 0 complete clusters in above example
3909 * The ext4_da_update_reserve_space will be called only if we
3910 * determine here that there were some "entire" clusters that span
3911 * this 'allocated' range.
3912 * In the non-bigalloc case, this function will just end up returning num_blks
3913 * without ever calling ext4_find_delalloc_range.
3916 get_reserved_cluster_alloc(struct inode
*inode
, ext4_lblk_t lblk_start
,
3917 unsigned int num_blks
)
3919 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
3920 ext4_lblk_t alloc_cluster_start
, alloc_cluster_end
;
3921 ext4_lblk_t lblk_from
, lblk_to
, c_offset
;
3922 unsigned int allocated_clusters
= 0;
3924 alloc_cluster_start
= EXT4_B2C(sbi
, lblk_start
);
3925 alloc_cluster_end
= EXT4_B2C(sbi
, lblk_start
+ num_blks
- 1);
3927 /* max possible clusters for this allocation */
3928 allocated_clusters
= alloc_cluster_end
- alloc_cluster_start
+ 1;
3930 trace_ext4_get_reserved_cluster_alloc(inode
, lblk_start
, num_blks
);
3932 /* Check towards left side */
3933 c_offset
= EXT4_LBLK_COFF(sbi
, lblk_start
);
3935 lblk_from
= EXT4_LBLK_CMASK(sbi
, lblk_start
);
3936 lblk_to
= lblk_from
+ c_offset
- 1;
3938 if (ext4_find_delalloc_range(inode
, lblk_from
, lblk_to
))
3939 allocated_clusters
--;
3942 /* Now check towards right. */
3943 c_offset
= EXT4_LBLK_COFF(sbi
, lblk_start
+ num_blks
);
3944 if (allocated_clusters
&& c_offset
) {
3945 lblk_from
= lblk_start
+ num_blks
;
3946 lblk_to
= lblk_from
+ (sbi
->s_cluster_ratio
- c_offset
) - 1;
3948 if (ext4_find_delalloc_range(inode
, lblk_from
, lblk_to
))
3949 allocated_clusters
--;
3952 return allocated_clusters
;
3956 convert_initialized_extent(handle_t
*handle
, struct inode
*inode
,
3957 struct ext4_map_blocks
*map
,
3958 struct ext4_ext_path
**ppath
,
3959 unsigned int allocated
)
3961 struct ext4_ext_path
*path
= *ppath
;
3962 struct ext4_extent
*ex
;
3963 ext4_lblk_t ee_block
;
3964 unsigned int ee_len
;
3969 * Make sure that the extent is no bigger than we support with
3972 if (map
->m_len
> EXT_UNWRITTEN_MAX_LEN
)
3973 map
->m_len
= EXT_UNWRITTEN_MAX_LEN
/ 2;
3975 depth
= ext_depth(inode
);
3976 ex
= path
[depth
].p_ext
;
3977 ee_block
= le32_to_cpu(ex
->ee_block
);
3978 ee_len
= ext4_ext_get_actual_len(ex
);
3980 ext_debug("%s: inode %lu, logical"
3981 "block %llu, max_blocks %u\n", __func__
, inode
->i_ino
,
3982 (unsigned long long)ee_block
, ee_len
);
3984 if (ee_block
!= map
->m_lblk
|| ee_len
> map
->m_len
) {
3985 err
= ext4_split_convert_extents(handle
, inode
, map
, ppath
,
3986 EXT4_GET_BLOCKS_CONVERT_UNWRITTEN
);
3989 path
= ext4_find_extent(inode
, map
->m_lblk
, ppath
, 0);
3991 return PTR_ERR(path
);
3992 depth
= ext_depth(inode
);
3993 ex
= path
[depth
].p_ext
;
3995 EXT4_ERROR_INODE(inode
, "unexpected hole at %lu",
3996 (unsigned long) map
->m_lblk
);
3997 return -EFSCORRUPTED
;
4001 err
= ext4_ext_get_access(handle
, inode
, path
+ depth
);
4004 /* first mark the extent as unwritten */
4005 ext4_ext_mark_unwritten(ex
);
4007 /* note: ext4_ext_correct_indexes() isn't needed here because
4008 * borders are not changed
4010 ext4_ext_try_to_merge(handle
, inode
, path
, ex
);
4012 /* Mark modified extent as dirty */
4013 err
= ext4_ext_dirty(handle
, inode
, path
+ path
->p_depth
);
4016 ext4_ext_show_leaf(inode
, path
);
4018 ext4_update_inode_fsync_trans(handle
, inode
, 1);
4019 err
= check_eofblocks_fl(handle
, inode
, map
->m_lblk
, path
, map
->m_len
);
4022 map
->m_flags
|= EXT4_MAP_UNWRITTEN
;
4023 if (allocated
> map
->m_len
)
4024 allocated
= map
->m_len
;
4025 map
->m_len
= allocated
;
4030 ext4_ext_handle_unwritten_extents(handle_t
*handle
, struct inode
*inode
,
4031 struct ext4_map_blocks
*map
,
4032 struct ext4_ext_path
**ppath
, int flags
,
4033 unsigned int allocated
, ext4_fsblk_t newblock
)
4035 struct ext4_ext_path
*path
= *ppath
;
4039 ext_debug("ext4_ext_handle_unwritten_extents: inode %lu, logical "
4040 "block %llu, max_blocks %u, flags %x, allocated %u\n",
4041 inode
->i_ino
, (unsigned long long)map
->m_lblk
, map
->m_len
,
4043 ext4_ext_show_leaf(inode
, path
);
4046 * When writing into unwritten space, we should not fail to
4047 * allocate metadata blocks for the new extent block if needed.
4049 flags
|= EXT4_GET_BLOCKS_METADATA_NOFAIL
;
4051 trace_ext4_ext_handle_unwritten_extents(inode
, map
, flags
,
4052 allocated
, newblock
);
4054 /* get_block() before submit the IO, split the extent */
4055 if (flags
& EXT4_GET_BLOCKS_PRE_IO
) {
4056 ret
= ext4_split_convert_extents(handle
, inode
, map
, ppath
,
4057 flags
| EXT4_GET_BLOCKS_CONVERT
);
4060 map
->m_flags
|= EXT4_MAP_UNWRITTEN
;
4063 /* IO end_io complete, convert the filled extent to written */
4064 if (flags
& EXT4_GET_BLOCKS_CONVERT
) {
4065 if (flags
& EXT4_GET_BLOCKS_ZERO
) {
4066 if (allocated
> map
->m_len
)
4067 allocated
= map
->m_len
;
4068 err
= ext4_issue_zeroout(inode
, map
->m_lblk
, newblock
,
4073 ret
= ext4_convert_unwritten_extents_endio(handle
, inode
, map
,
4076 ext4_update_inode_fsync_trans(handle
, inode
, 1);
4077 err
= check_eofblocks_fl(handle
, inode
, map
->m_lblk
,
4081 map
->m_flags
|= EXT4_MAP_MAPPED
;
4082 map
->m_pblk
= newblock
;
4083 if (allocated
> map
->m_len
)
4084 allocated
= map
->m_len
;
4085 map
->m_len
= allocated
;
4088 /* buffered IO case */
4090 * repeat fallocate creation request
4091 * we already have an unwritten extent
4093 if (flags
& EXT4_GET_BLOCKS_UNWRIT_EXT
) {
4094 map
->m_flags
|= EXT4_MAP_UNWRITTEN
;
4098 /* buffered READ or buffered write_begin() lookup */
4099 if ((flags
& EXT4_GET_BLOCKS_CREATE
) == 0) {
4101 * We have blocks reserved already. We
4102 * return allocated blocks so that delalloc
4103 * won't do block reservation for us. But
4104 * the buffer head will be unmapped so that
4105 * a read from the block returns 0s.
4107 map
->m_flags
|= EXT4_MAP_UNWRITTEN
;
4111 /* buffered write, writepage time, convert*/
4112 ret
= ext4_ext_convert_to_initialized(handle
, inode
, map
, ppath
, flags
);
4114 ext4_update_inode_fsync_trans(handle
, inode
, 1);
4121 map
->m_flags
|= EXT4_MAP_NEW
;
4123 * if we allocated more blocks than requested
4124 * we need to make sure we unmap the extra block
4125 * allocated. The actual needed block will get
4126 * unmapped later when we find the buffer_head marked
4129 if (allocated
> map
->m_len
) {
4130 clean_bdev_aliases(inode
->i_sb
->s_bdev
, newblock
+ map
->m_len
,
4131 allocated
- map
->m_len
);
4132 allocated
= map
->m_len
;
4134 map
->m_len
= allocated
;
4137 * If we have done fallocate with the offset that is already
4138 * delayed allocated, we would have block reservation
4139 * and quota reservation done in the delayed write path.
4140 * But fallocate would have already updated quota and block
4141 * count for this offset. So cancel these reservation
4143 if (flags
& EXT4_GET_BLOCKS_DELALLOC_RESERVE
) {
4144 unsigned int reserved_clusters
;
4145 reserved_clusters
= get_reserved_cluster_alloc(inode
,
4146 map
->m_lblk
, map
->m_len
);
4147 if (reserved_clusters
)
4148 ext4_da_update_reserve_space(inode
,
4154 map
->m_flags
|= EXT4_MAP_MAPPED
;
4155 if ((flags
& EXT4_GET_BLOCKS_KEEP_SIZE
) == 0) {
4156 err
= check_eofblocks_fl(handle
, inode
, map
->m_lblk
, path
,
4162 if (allocated
> map
->m_len
)
4163 allocated
= map
->m_len
;
4164 ext4_ext_show_leaf(inode
, path
);
4165 map
->m_pblk
= newblock
;
4166 map
->m_len
= allocated
;
4168 return err
? err
: allocated
;
4172 * get_implied_cluster_alloc - check to see if the requested
4173 * allocation (in the map structure) overlaps with a cluster already
4174 * allocated in an extent.
4175 * @sb The filesystem superblock structure
4176 * @map The requested lblk->pblk mapping
4177 * @ex The extent structure which might contain an implied
4178 * cluster allocation
4180 * This function is called by ext4_ext_map_blocks() after we failed to
4181 * find blocks that were already in the inode's extent tree. Hence,
4182 * we know that the beginning of the requested region cannot overlap
4183 * the extent from the inode's extent tree. There are three cases we
4184 * want to catch. The first is this case:
4186 * |--- cluster # N--|
4187 * |--- extent ---| |---- requested region ---|
4190 * The second case that we need to test for is this one:
4192 * |--------- cluster # N ----------------|
4193 * |--- requested region --| |------- extent ----|
4194 * |=======================|
4196 * The third case is when the requested region lies between two extents
4197 * within the same cluster:
4198 * |------------- cluster # N-------------|
4199 * |----- ex -----| |---- ex_right ----|
4200 * |------ requested region ------|
4201 * |================|
4203 * In each of the above cases, we need to set the map->m_pblk and
4204 * map->m_len so it corresponds to the return the extent labelled as
4205 * "|====|" from cluster #N, since it is already in use for data in
4206 * cluster EXT4_B2C(sbi, map->m_lblk). We will then return 1 to
4207 * signal to ext4_ext_map_blocks() that map->m_pblk should be treated
4208 * as a new "allocated" block region. Otherwise, we will return 0 and
4209 * ext4_ext_map_blocks() will then allocate one or more new clusters
4210 * by calling ext4_mb_new_blocks().
4212 static int get_implied_cluster_alloc(struct super_block
*sb
,
4213 struct ext4_map_blocks
*map
,
4214 struct ext4_extent
*ex
,
4215 struct ext4_ext_path
*path
)
4217 struct ext4_sb_info
*sbi
= EXT4_SB(sb
);
4218 ext4_lblk_t c_offset
= EXT4_LBLK_COFF(sbi
, map
->m_lblk
);
4219 ext4_lblk_t ex_cluster_start
, ex_cluster_end
;
4220 ext4_lblk_t rr_cluster_start
;
4221 ext4_lblk_t ee_block
= le32_to_cpu(ex
->ee_block
);
4222 ext4_fsblk_t ee_start
= ext4_ext_pblock(ex
);
4223 unsigned short ee_len
= ext4_ext_get_actual_len(ex
);
4225 /* The extent passed in that we are trying to match */
4226 ex_cluster_start
= EXT4_B2C(sbi
, ee_block
);
4227 ex_cluster_end
= EXT4_B2C(sbi
, ee_block
+ ee_len
- 1);
4229 /* The requested region passed into ext4_map_blocks() */
4230 rr_cluster_start
= EXT4_B2C(sbi
, map
->m_lblk
);
4232 if ((rr_cluster_start
== ex_cluster_end
) ||
4233 (rr_cluster_start
== ex_cluster_start
)) {
4234 if (rr_cluster_start
== ex_cluster_end
)
4235 ee_start
+= ee_len
- 1;
4236 map
->m_pblk
= EXT4_PBLK_CMASK(sbi
, ee_start
) + c_offset
;
4237 map
->m_len
= min(map
->m_len
,
4238 (unsigned) sbi
->s_cluster_ratio
- c_offset
);
4240 * Check for and handle this case:
4242 * |--------- cluster # N-------------|
4243 * |------- extent ----|
4244 * |--- requested region ---|
4248 if (map
->m_lblk
< ee_block
)
4249 map
->m_len
= min(map
->m_len
, ee_block
- map
->m_lblk
);
4252 * Check for the case where there is already another allocated
4253 * block to the right of 'ex' but before the end of the cluster.
4255 * |------------- cluster # N-------------|
4256 * |----- ex -----| |---- ex_right ----|
4257 * |------ requested region ------|
4258 * |================|
4260 if (map
->m_lblk
> ee_block
) {
4261 ext4_lblk_t next
= ext4_ext_next_allocated_block(path
);
4262 map
->m_len
= min(map
->m_len
, next
- map
->m_lblk
);
4265 trace_ext4_get_implied_cluster_alloc_exit(sb
, map
, 1);
4269 trace_ext4_get_implied_cluster_alloc_exit(sb
, map
, 0);
4275 * Block allocation/map/preallocation routine for extents based files
4278 * Need to be called with
4279 * down_read(&EXT4_I(inode)->i_data_sem) if not allocating file system block
4280 * (ie, create is zero). Otherwise down_write(&EXT4_I(inode)->i_data_sem)
4282 * return > 0, number of of blocks already mapped/allocated
4283 * if create == 0 and these are pre-allocated blocks
4284 * buffer head is unmapped
4285 * otherwise blocks are mapped
4287 * return = 0, if plain look up failed (blocks have not been allocated)
4288 * buffer head is unmapped
4290 * return < 0, error case.
4292 int ext4_ext_map_blocks(handle_t
*handle
, struct inode
*inode
,
4293 struct ext4_map_blocks
*map
, int flags
)
4295 struct ext4_ext_path
*path
= NULL
;
4296 struct ext4_extent newex
, *ex
, *ex2
;
4297 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
4298 ext4_fsblk_t newblock
= 0;
4299 int free_on_err
= 0, err
= 0, depth
, ret
;
4300 unsigned int allocated
= 0, offset
= 0;
4301 unsigned int allocated_clusters
= 0;
4302 struct ext4_allocation_request ar
;
4303 ext4_lblk_t cluster_offset
;
4304 bool map_from_cluster
= false;
4306 ext_debug("blocks %u/%u requested for inode %lu\n",
4307 map
->m_lblk
, map
->m_len
, inode
->i_ino
);
4308 trace_ext4_ext_map_blocks_enter(inode
, map
->m_lblk
, map
->m_len
, flags
);
4310 /* find extent for this block */
4311 path
= ext4_find_extent(inode
, map
->m_lblk
, NULL
, 0);
4313 err
= PTR_ERR(path
);
4318 depth
= ext_depth(inode
);
4321 * consistent leaf must not be empty;
4322 * this situation is possible, though, _during_ tree modification;
4323 * this is why assert can't be put in ext4_find_extent()
4325 if (unlikely(path
[depth
].p_ext
== NULL
&& depth
!= 0)) {
4326 EXT4_ERROR_INODE(inode
, "bad extent address "
4327 "lblock: %lu, depth: %d pblock %lld",
4328 (unsigned long) map
->m_lblk
, depth
,
4329 path
[depth
].p_block
);
4330 err
= -EFSCORRUPTED
;
4334 ex
= path
[depth
].p_ext
;
4336 ext4_lblk_t ee_block
= le32_to_cpu(ex
->ee_block
);
4337 ext4_fsblk_t ee_start
= ext4_ext_pblock(ex
);
4338 unsigned short ee_len
;
4342 * unwritten extents are treated as holes, except that
4343 * we split out initialized portions during a write.
4345 ee_len
= ext4_ext_get_actual_len(ex
);
4347 trace_ext4_ext_show_extent(inode
, ee_block
, ee_start
, ee_len
);
4349 /* if found extent covers block, simply return it */
4350 if (in_range(map
->m_lblk
, ee_block
, ee_len
)) {
4351 newblock
= map
->m_lblk
- ee_block
+ ee_start
;
4352 /* number of remaining blocks in the extent */
4353 allocated
= ee_len
- (map
->m_lblk
- ee_block
);
4354 ext_debug("%u fit into %u:%d -> %llu\n", map
->m_lblk
,
4355 ee_block
, ee_len
, newblock
);
4358 * If the extent is initialized check whether the
4359 * caller wants to convert it to unwritten.
4361 if ((!ext4_ext_is_unwritten(ex
)) &&
4362 (flags
& EXT4_GET_BLOCKS_CONVERT_UNWRITTEN
)) {
4363 allocated
= convert_initialized_extent(
4364 handle
, inode
, map
, &path
,
4367 } else if (!ext4_ext_is_unwritten(ex
))
4370 ret
= ext4_ext_handle_unwritten_extents(
4371 handle
, inode
, map
, &path
, flags
,
4372 allocated
, newblock
);
4382 * requested block isn't allocated yet;
4383 * we couldn't try to create block if create flag is zero
4385 if ((flags
& EXT4_GET_BLOCKS_CREATE
) == 0) {
4386 ext4_lblk_t hole_start
, hole_len
;
4388 hole_start
= map
->m_lblk
;
4389 hole_len
= ext4_ext_determine_hole(inode
, path
, &hole_start
);
4391 * put just found gap into cache to speed up
4392 * subsequent requests
4394 ext4_ext_put_gap_in_cache(inode
, hole_start
, hole_len
);
4396 /* Update hole_len to reflect hole size after map->m_lblk */
4397 if (hole_start
!= map
->m_lblk
)
4398 hole_len
-= map
->m_lblk
- hole_start
;
4400 map
->m_len
= min_t(unsigned int, map
->m_len
, hole_len
);
4406 * Okay, we need to do block allocation.
4408 newex
.ee_block
= cpu_to_le32(map
->m_lblk
);
4409 cluster_offset
= EXT4_LBLK_COFF(sbi
, map
->m_lblk
);
4412 * If we are doing bigalloc, check to see if the extent returned
4413 * by ext4_find_extent() implies a cluster we can use.
4415 if (cluster_offset
&& ex
&&
4416 get_implied_cluster_alloc(inode
->i_sb
, map
, ex
, path
)) {
4417 ar
.len
= allocated
= map
->m_len
;
4418 newblock
= map
->m_pblk
;
4419 map_from_cluster
= true;
4420 goto got_allocated_blocks
;
4423 /* find neighbour allocated blocks */
4424 ar
.lleft
= map
->m_lblk
;
4425 err
= ext4_ext_search_left(inode
, path
, &ar
.lleft
, &ar
.pleft
);
4428 ar
.lright
= map
->m_lblk
;
4430 err
= ext4_ext_search_right(inode
, path
, &ar
.lright
, &ar
.pright
, &ex2
);
4434 /* Check if the extent after searching to the right implies a
4435 * cluster we can use. */
4436 if ((sbi
->s_cluster_ratio
> 1) && ex2
&&
4437 get_implied_cluster_alloc(inode
->i_sb
, map
, ex2
, path
)) {
4438 ar
.len
= allocated
= map
->m_len
;
4439 newblock
= map
->m_pblk
;
4440 map_from_cluster
= true;
4441 goto got_allocated_blocks
;
4445 * See if request is beyond maximum number of blocks we can have in
4446 * a single extent. For an initialized extent this limit is
4447 * EXT_INIT_MAX_LEN and for an unwritten extent this limit is
4448 * EXT_UNWRITTEN_MAX_LEN.
4450 if (map
->m_len
> EXT_INIT_MAX_LEN
&&
4451 !(flags
& EXT4_GET_BLOCKS_UNWRIT_EXT
))
4452 map
->m_len
= EXT_INIT_MAX_LEN
;
4453 else if (map
->m_len
> EXT_UNWRITTEN_MAX_LEN
&&
4454 (flags
& EXT4_GET_BLOCKS_UNWRIT_EXT
))
4455 map
->m_len
= EXT_UNWRITTEN_MAX_LEN
;
4457 /* Check if we can really insert (m_lblk)::(m_lblk + m_len) extent */
4458 newex
.ee_len
= cpu_to_le16(map
->m_len
);
4459 err
= ext4_ext_check_overlap(sbi
, inode
, &newex
, path
);
4461 allocated
= ext4_ext_get_actual_len(&newex
);
4463 allocated
= map
->m_len
;
4465 /* allocate new block */
4467 ar
.goal
= ext4_ext_find_goal(inode
, path
, map
->m_lblk
);
4468 ar
.logical
= map
->m_lblk
;
4470 * We calculate the offset from the beginning of the cluster
4471 * for the logical block number, since when we allocate a
4472 * physical cluster, the physical block should start at the
4473 * same offset from the beginning of the cluster. This is
4474 * needed so that future calls to get_implied_cluster_alloc()
4477 offset
= EXT4_LBLK_COFF(sbi
, map
->m_lblk
);
4478 ar
.len
= EXT4_NUM_B2C(sbi
, offset
+allocated
);
4480 ar
.logical
-= offset
;
4481 if (S_ISREG(inode
->i_mode
))
4482 ar
.flags
= EXT4_MB_HINT_DATA
;
4484 /* disable in-core preallocation for non-regular files */
4486 if (flags
& EXT4_GET_BLOCKS_NO_NORMALIZE
)
4487 ar
.flags
|= EXT4_MB_HINT_NOPREALLOC
;
4488 if (flags
& EXT4_GET_BLOCKS_DELALLOC_RESERVE
)
4489 ar
.flags
|= EXT4_MB_DELALLOC_RESERVED
;
4490 if (flags
& EXT4_GET_BLOCKS_METADATA_NOFAIL
)
4491 ar
.flags
|= EXT4_MB_USE_RESERVED
;
4492 newblock
= ext4_mb_new_blocks(handle
, &ar
, &err
);
4495 ext_debug("allocate new block: goal %llu, found %llu/%u\n",
4496 ar
.goal
, newblock
, allocated
);
4498 allocated_clusters
= ar
.len
;
4499 ar
.len
= EXT4_C2B(sbi
, ar
.len
) - offset
;
4500 if (ar
.len
> allocated
)
4503 got_allocated_blocks
:
4504 /* try to insert new extent into found leaf and return */
4505 ext4_ext_store_pblock(&newex
, newblock
+ offset
);
4506 newex
.ee_len
= cpu_to_le16(ar
.len
);
4507 /* Mark unwritten */
4508 if (flags
& EXT4_GET_BLOCKS_UNWRIT_EXT
){
4509 ext4_ext_mark_unwritten(&newex
);
4510 map
->m_flags
|= EXT4_MAP_UNWRITTEN
;
4514 if ((flags
& EXT4_GET_BLOCKS_KEEP_SIZE
) == 0)
4515 err
= check_eofblocks_fl(handle
, inode
, map
->m_lblk
,
4518 err
= ext4_ext_insert_extent(handle
, inode
, &path
,
4521 if (err
&& free_on_err
) {
4522 int fb_flags
= flags
& EXT4_GET_BLOCKS_DELALLOC_RESERVE
?
4523 EXT4_FREE_BLOCKS_NO_QUOT_UPDATE
: 0;
4524 /* free data blocks we just allocated */
4525 /* not a good idea to call discard here directly,
4526 * but otherwise we'd need to call it every free() */
4527 ext4_discard_preallocations(inode
);
4528 ext4_free_blocks(handle
, inode
, NULL
, newblock
,
4529 EXT4_C2B(sbi
, allocated_clusters
), fb_flags
);
4533 /* previous routine could use block we allocated */
4534 newblock
= ext4_ext_pblock(&newex
);
4535 allocated
= ext4_ext_get_actual_len(&newex
);
4536 if (allocated
> map
->m_len
)
4537 allocated
= map
->m_len
;
4538 map
->m_flags
|= EXT4_MAP_NEW
;
4541 * Update reserved blocks/metadata blocks after successful
4542 * block allocation which had been deferred till now.
4544 if (flags
& EXT4_GET_BLOCKS_DELALLOC_RESERVE
) {
4545 unsigned int reserved_clusters
;
4547 * Check how many clusters we had reserved this allocated range
4549 reserved_clusters
= get_reserved_cluster_alloc(inode
,
4550 map
->m_lblk
, allocated
);
4551 if (!map_from_cluster
) {
4552 BUG_ON(allocated_clusters
< reserved_clusters
);
4553 if (reserved_clusters
< allocated_clusters
) {
4554 struct ext4_inode_info
*ei
= EXT4_I(inode
);
4555 int reservation
= allocated_clusters
-
4558 * It seems we claimed few clusters outside of
4559 * the range of this allocation. We should give
4560 * it back to the reservation pool. This can
4561 * happen in the following case:
4563 * * Suppose s_cluster_ratio is 4 (i.e., each
4564 * cluster has 4 blocks. Thus, the clusters
4565 * are [0-3],[4-7],[8-11]...
4566 * * First comes delayed allocation write for
4567 * logical blocks 10 & 11. Since there were no
4568 * previous delayed allocated blocks in the
4569 * range [8-11], we would reserve 1 cluster
4571 * * Next comes write for logical blocks 3 to 8.
4572 * In this case, we will reserve 2 clusters
4573 * (for [0-3] and [4-7]; and not for [8-11] as
4574 * that range has a delayed allocated blocks.
4575 * Thus total reserved clusters now becomes 3.
4576 * * Now, during the delayed allocation writeout
4577 * time, we will first write blocks [3-8] and
4578 * allocate 3 clusters for writing these
4579 * blocks. Also, we would claim all these
4580 * three clusters above.
4581 * * Now when we come here to writeout the
4582 * blocks [10-11], we would expect to claim
4583 * the reservation of 1 cluster we had made
4584 * (and we would claim it since there are no
4585 * more delayed allocated blocks in the range
4586 * [8-11]. But our reserved cluster count had
4587 * already gone to 0.
4589 * Thus, at the step 4 above when we determine
4590 * that there are still some unwritten delayed
4591 * allocated blocks outside of our current
4592 * block range, we should increment the
4593 * reserved clusters count so that when the
4594 * remaining blocks finally gets written, we
4597 dquot_reserve_block(inode
,
4598 EXT4_C2B(sbi
, reservation
));
4599 spin_lock(&ei
->i_block_reservation_lock
);
4600 ei
->i_reserved_data_blocks
+= reservation
;
4601 spin_unlock(&ei
->i_block_reservation_lock
);
4604 * We will claim quota for all newly allocated blocks.
4605 * We're updating the reserved space *after* the
4606 * correction above so we do not accidentally free
4607 * all the metadata reservation because we might
4608 * actually need it later on.
4610 ext4_da_update_reserve_space(inode
, allocated_clusters
,
4616 * Cache the extent and update transaction to commit on fdatasync only
4617 * when it is _not_ an unwritten extent.
4619 if ((flags
& EXT4_GET_BLOCKS_UNWRIT_EXT
) == 0)
4620 ext4_update_inode_fsync_trans(handle
, inode
, 1);
4622 ext4_update_inode_fsync_trans(handle
, inode
, 0);
4624 if (allocated
> map
->m_len
)
4625 allocated
= map
->m_len
;
4626 ext4_ext_show_leaf(inode
, path
);
4627 map
->m_flags
|= EXT4_MAP_MAPPED
;
4628 map
->m_pblk
= newblock
;
4629 map
->m_len
= allocated
;
4631 ext4_ext_drop_refs(path
);
4634 trace_ext4_ext_map_blocks_exit(inode
, flags
, map
,
4635 err
? err
: allocated
);
4636 return err
? err
: allocated
;
4639 int ext4_ext_truncate(handle_t
*handle
, struct inode
*inode
)
4641 struct super_block
*sb
= inode
->i_sb
;
4642 ext4_lblk_t last_block
;
4646 * TODO: optimization is possible here.
4647 * Probably we need not scan at all,
4648 * because page truncation is enough.
4651 /* we have to know where to truncate from in crash case */
4652 EXT4_I(inode
)->i_disksize
= inode
->i_size
;
4653 err
= ext4_mark_inode_dirty(handle
, inode
);
4657 last_block
= (inode
->i_size
+ sb
->s_blocksize
- 1)
4658 >> EXT4_BLOCK_SIZE_BITS(sb
);
4660 err
= ext4_es_remove_extent(inode
, last_block
,
4661 EXT_MAX_BLOCKS
- last_block
);
4662 if (err
== -ENOMEM
) {
4664 congestion_wait(BLK_RW_ASYNC
, HZ
/50);
4669 return ext4_ext_remove_space(inode
, last_block
, EXT_MAX_BLOCKS
- 1);
4672 static int ext4_alloc_file_blocks(struct file
*file
, ext4_lblk_t offset
,
4673 ext4_lblk_t len
, loff_t new_size
,
4676 struct inode
*inode
= file_inode(file
);
4682 struct ext4_map_blocks map
;
4683 unsigned int credits
;
4686 BUG_ON(!ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
));
4687 map
.m_lblk
= offset
;
4690 * Don't normalize the request if it can fit in one extent so
4691 * that it doesn't get unnecessarily split into multiple
4694 if (len
<= EXT_UNWRITTEN_MAX_LEN
)
4695 flags
|= EXT4_GET_BLOCKS_NO_NORMALIZE
;
4698 * credits to insert 1 extent into extent tree
4700 credits
= ext4_chunk_trans_blocks(inode
, len
);
4701 depth
= ext_depth(inode
);
4704 while (ret
>= 0 && len
) {
4706 * Recalculate credits when extent tree depth changes.
4708 if (depth
!= ext_depth(inode
)) {
4709 credits
= ext4_chunk_trans_blocks(inode
, len
);
4710 depth
= ext_depth(inode
);
4713 handle
= ext4_journal_start(inode
, EXT4_HT_MAP_BLOCKS
,
4715 if (IS_ERR(handle
)) {
4716 ret
= PTR_ERR(handle
);
4719 ret
= ext4_map_blocks(handle
, inode
, &map
, flags
);
4721 ext4_debug("inode #%lu: block %u: len %u: "
4722 "ext4_ext_map_blocks returned %d",
4723 inode
->i_ino
, map
.m_lblk
,
4725 ext4_mark_inode_dirty(handle
, inode
);
4726 ret2
= ext4_journal_stop(handle
);
4730 map
.m_len
= len
= len
- ret
;
4731 epos
= (loff_t
)map
.m_lblk
<< inode
->i_blkbits
;
4732 inode
->i_ctime
= current_time(inode
);
4734 if (epos
> new_size
)
4736 if (ext4_update_inode_size(inode
, epos
) & 0x1)
4737 inode
->i_mtime
= inode
->i_ctime
;
4739 if (epos
> inode
->i_size
)
4740 ext4_set_inode_flag(inode
,
4741 EXT4_INODE_EOFBLOCKS
);
4743 ext4_mark_inode_dirty(handle
, inode
);
4744 ext4_update_inode_fsync_trans(handle
, inode
, 1);
4745 ret2
= ext4_journal_stop(handle
);
4749 if (ret
== -ENOSPC
&&
4750 ext4_should_retry_alloc(inode
->i_sb
, &retries
)) {
4755 return ret
> 0 ? ret2
: ret
;
4758 static long ext4_zero_range(struct file
*file
, loff_t offset
,
4759 loff_t len
, int mode
)
4761 struct inode
*inode
= file_inode(file
);
4762 handle_t
*handle
= NULL
;
4763 unsigned int max_blocks
;
4764 loff_t new_size
= 0;
4768 int partial_begin
, partial_end
;
4771 unsigned int blkbits
= inode
->i_blkbits
;
4773 trace_ext4_zero_range(inode
, offset
, len
, mode
);
4775 if (!S_ISREG(inode
->i_mode
))
4778 /* Call ext4_force_commit to flush all data in case of data=journal. */
4779 if (ext4_should_journal_data(inode
)) {
4780 ret
= ext4_force_commit(inode
->i_sb
);
4786 * Round up offset. This is not fallocate, we neet to zero out
4787 * blocks, so convert interior block aligned part of the range to
4788 * unwritten and possibly manually zero out unaligned parts of the
4791 start
= round_up(offset
, 1 << blkbits
);
4792 end
= round_down((offset
+ len
), 1 << blkbits
);
4794 if (start
< offset
|| end
> offset
+ len
)
4796 partial_begin
= offset
& ((1 << blkbits
) - 1);
4797 partial_end
= (offset
+ len
) & ((1 << blkbits
) - 1);
4799 lblk
= start
>> blkbits
;
4800 max_blocks
= (end
>> blkbits
);
4801 if (max_blocks
< lblk
)
4809 * Indirect files do not support unwritten extnets
4811 if (!(ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
))) {
4816 if (!(mode
& FALLOC_FL_KEEP_SIZE
) &&
4817 (offset
+ len
> i_size_read(inode
) ||
4818 offset
+ len
> EXT4_I(inode
)->i_disksize
)) {
4819 new_size
= offset
+ len
;
4820 ret
= inode_newsize_ok(inode
, new_size
);
4825 flags
= EXT4_GET_BLOCKS_CREATE_UNWRIT_EXT
;
4826 if (mode
& FALLOC_FL_KEEP_SIZE
)
4827 flags
|= EXT4_GET_BLOCKS_KEEP_SIZE
;
4829 /* Wait all existing dio workers, newcomers will block on i_mutex */
4830 ext4_inode_block_unlocked_dio(inode
);
4831 inode_dio_wait(inode
);
4833 /* Preallocate the range including the unaligned edges */
4834 if (partial_begin
|| partial_end
) {
4835 ret
= ext4_alloc_file_blocks(file
,
4836 round_down(offset
, 1 << blkbits
) >> blkbits
,
4837 (round_up((offset
+ len
), 1 << blkbits
) -
4838 round_down(offset
, 1 << blkbits
)) >> blkbits
,
4845 /* Zero range excluding the unaligned edges */
4846 if (max_blocks
> 0) {
4847 flags
|= (EXT4_GET_BLOCKS_CONVERT_UNWRITTEN
|
4851 * Prevent page faults from reinstantiating pages we have
4852 * released from page cache.
4854 down_write(&EXT4_I(inode
)->i_mmap_sem
);
4855 ret
= ext4_update_disksize_before_punch(inode
, offset
, len
);
4857 up_write(&EXT4_I(inode
)->i_mmap_sem
);
4860 /* Now release the pages and zero block aligned part of pages */
4861 truncate_pagecache_range(inode
, start
, end
- 1);
4862 inode
->i_mtime
= inode
->i_ctime
= current_time(inode
);
4864 ret
= ext4_alloc_file_blocks(file
, lblk
, max_blocks
, new_size
,
4866 up_write(&EXT4_I(inode
)->i_mmap_sem
);
4870 if (!partial_begin
&& !partial_end
)
4874 * In worst case we have to writeout two nonadjacent unwritten
4875 * blocks and update the inode
4877 credits
= (2 * ext4_ext_index_trans_blocks(inode
, 2)) + 1;
4878 if (ext4_should_journal_data(inode
))
4880 handle
= ext4_journal_start(inode
, EXT4_HT_MISC
, credits
);
4881 if (IS_ERR(handle
)) {
4882 ret
= PTR_ERR(handle
);
4883 ext4_std_error(inode
->i_sb
, ret
);
4887 inode
->i_mtime
= inode
->i_ctime
= current_time(inode
);
4889 ext4_update_inode_size(inode
, new_size
);
4892 * Mark that we allocate beyond EOF so the subsequent truncate
4893 * can proceed even if the new size is the same as i_size.
4895 if ((offset
+ len
) > i_size_read(inode
))
4896 ext4_set_inode_flag(inode
, EXT4_INODE_EOFBLOCKS
);
4898 ext4_mark_inode_dirty(handle
, inode
);
4900 /* Zero out partial block at the edges of the range */
4901 ret
= ext4_zero_partial_blocks(handle
, inode
, offset
, len
);
4903 ext4_update_inode_fsync_trans(handle
, inode
, 1);
4905 if (file
->f_flags
& O_SYNC
)
4906 ext4_handle_sync(handle
);
4908 ext4_journal_stop(handle
);
4910 ext4_inode_resume_unlocked_dio(inode
);
4912 inode_unlock(inode
);
4917 * preallocate space for a file. This implements ext4's fallocate file
4918 * operation, which gets called from sys_fallocate system call.
4919 * For block-mapped files, posix_fallocate should fall back to the method
4920 * of writing zeroes to the required new blocks (the same behavior which is
4921 * expected for file systems which do not support fallocate() system call).
4923 long ext4_fallocate(struct file
*file
, int mode
, loff_t offset
, loff_t len
)
4925 struct inode
*inode
= file_inode(file
);
4926 loff_t new_size
= 0;
4927 unsigned int max_blocks
;
4931 unsigned int blkbits
= inode
->i_blkbits
;
4934 * Encrypted inodes can't handle collapse range or insert
4935 * range since we would need to re-encrypt blocks with a
4936 * different IV or XTS tweak (which are based on the logical
4939 * XXX It's not clear why zero range isn't working, but we'll
4940 * leave it disabled for encrypted inodes for now. This is a
4941 * bug we should fix....
4943 if (ext4_encrypted_inode(inode
) &&
4944 (mode
& (FALLOC_FL_COLLAPSE_RANGE
| FALLOC_FL_INSERT_RANGE
|
4945 FALLOC_FL_ZERO_RANGE
)))
4948 /* Return error if mode is not supported */
4949 if (mode
& ~(FALLOC_FL_KEEP_SIZE
| FALLOC_FL_PUNCH_HOLE
|
4950 FALLOC_FL_COLLAPSE_RANGE
| FALLOC_FL_ZERO_RANGE
|
4951 FALLOC_FL_INSERT_RANGE
))
4954 if (mode
& FALLOC_FL_PUNCH_HOLE
)
4955 return ext4_punch_hole(inode
, offset
, len
);
4957 ret
= ext4_convert_inline_data(inode
);
4961 if (mode
& FALLOC_FL_COLLAPSE_RANGE
)
4962 return ext4_collapse_range(inode
, offset
, len
);
4964 if (mode
& FALLOC_FL_INSERT_RANGE
)
4965 return ext4_insert_range(inode
, offset
, len
);
4967 if (mode
& FALLOC_FL_ZERO_RANGE
)
4968 return ext4_zero_range(file
, offset
, len
, mode
);
4970 trace_ext4_fallocate_enter(inode
, offset
, len
, mode
);
4971 lblk
= offset
>> blkbits
;
4973 max_blocks
= EXT4_MAX_BLOCKS(len
, offset
, blkbits
);
4974 flags
= EXT4_GET_BLOCKS_CREATE_UNWRIT_EXT
;
4975 if (mode
& FALLOC_FL_KEEP_SIZE
)
4976 flags
|= EXT4_GET_BLOCKS_KEEP_SIZE
;
4981 * We only support preallocation for extent-based files only
4983 if (!(ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
))) {
4988 if (!(mode
& FALLOC_FL_KEEP_SIZE
) &&
4989 (offset
+ len
> i_size_read(inode
) ||
4990 offset
+ len
> EXT4_I(inode
)->i_disksize
)) {
4991 new_size
= offset
+ len
;
4992 ret
= inode_newsize_ok(inode
, new_size
);
4997 /* Wait all existing dio workers, newcomers will block on i_mutex */
4998 ext4_inode_block_unlocked_dio(inode
);
4999 inode_dio_wait(inode
);
5001 ret
= ext4_alloc_file_blocks(file
, lblk
, max_blocks
, new_size
, flags
);
5002 ext4_inode_resume_unlocked_dio(inode
);
5006 if (file
->f_flags
& O_SYNC
&& EXT4_SB(inode
->i_sb
)->s_journal
) {
5007 ret
= jbd2_complete_transaction(EXT4_SB(inode
->i_sb
)->s_journal
,
5008 EXT4_I(inode
)->i_sync_tid
);
5011 inode_unlock(inode
);
5012 trace_ext4_fallocate_exit(inode
, offset
, max_blocks
, ret
);
5017 * This function convert a range of blocks to written extents
5018 * The caller of this function will pass the start offset and the size.
5019 * all unwritten extents within this range will be converted to
5022 * This function is called from the direct IO end io call back
5023 * function, to convert the fallocated extents after IO is completed.
5024 * Returns 0 on success.
5026 int ext4_convert_unwritten_extents(handle_t
*handle
, struct inode
*inode
,
5027 loff_t offset
, ssize_t len
)
5029 unsigned int max_blocks
;
5032 struct ext4_map_blocks map
;
5033 unsigned int credits
, blkbits
= inode
->i_blkbits
;
5035 map
.m_lblk
= offset
>> blkbits
;
5036 max_blocks
= EXT4_MAX_BLOCKS(len
, offset
, blkbits
);
5039 * This is somewhat ugly but the idea is clear: When transaction is
5040 * reserved, everything goes into it. Otherwise we rather start several
5041 * smaller transactions for conversion of each extent separately.
5044 handle
= ext4_journal_start_reserved(handle
,
5045 EXT4_HT_EXT_CONVERT
);
5047 return PTR_ERR(handle
);
5051 * credits to insert 1 extent into extent tree
5053 credits
= ext4_chunk_trans_blocks(inode
, max_blocks
);
5055 while (ret
>= 0 && ret
< max_blocks
) {
5057 map
.m_len
= (max_blocks
-= ret
);
5059 handle
= ext4_journal_start(inode
, EXT4_HT_MAP_BLOCKS
,
5061 if (IS_ERR(handle
)) {
5062 ret
= PTR_ERR(handle
);
5066 ret
= ext4_map_blocks(handle
, inode
, &map
,
5067 EXT4_GET_BLOCKS_IO_CONVERT_EXT
);
5069 ext4_warning(inode
->i_sb
,
5070 "inode #%lu: block %u: len %u: "
5071 "ext4_ext_map_blocks returned %d",
5072 inode
->i_ino
, map
.m_lblk
,
5074 ext4_mark_inode_dirty(handle
, inode
);
5076 ret2
= ext4_journal_stop(handle
);
5077 if (ret
<= 0 || ret2
)
5081 ret2
= ext4_journal_stop(handle
);
5082 return ret
> 0 ? ret2
: ret
;
5086 * If newes is not existing extent (newes->ec_pblk equals zero) find
5087 * delayed extent at start of newes and update newes accordingly and
5088 * return start of the next delayed extent.
5090 * If newes is existing extent (newes->ec_pblk is not equal zero)
5091 * return start of next delayed extent or EXT_MAX_BLOCKS if no delayed
5092 * extent found. Leave newes unmodified.
5094 static int ext4_find_delayed_extent(struct inode
*inode
,
5095 struct extent_status
*newes
)
5097 struct extent_status es
;
5098 ext4_lblk_t block
, next_del
;
5100 if (newes
->es_pblk
== 0) {
5101 ext4_es_find_delayed_extent_range(inode
, newes
->es_lblk
,
5102 newes
->es_lblk
+ newes
->es_len
- 1, &es
);
5105 * No extent in extent-tree contains block @newes->es_pblk,
5106 * then the block may stay in 1)a hole or 2)delayed-extent.
5112 if (es
.es_lblk
> newes
->es_lblk
) {
5114 newes
->es_len
= min(es
.es_lblk
- newes
->es_lblk
,
5119 newes
->es_len
= es
.es_lblk
+ es
.es_len
- newes
->es_lblk
;
5122 block
= newes
->es_lblk
+ newes
->es_len
;
5123 ext4_es_find_delayed_extent_range(inode
, block
, EXT_MAX_BLOCKS
, &es
);
5125 next_del
= EXT_MAX_BLOCKS
;
5127 next_del
= es
.es_lblk
;
5131 /* fiemap flags we can handle specified here */
5132 #define EXT4_FIEMAP_FLAGS (FIEMAP_FLAG_SYNC|FIEMAP_FLAG_XATTR)
5134 static int ext4_xattr_fiemap(struct inode
*inode
,
5135 struct fiemap_extent_info
*fieinfo
)
5139 __u32 flags
= FIEMAP_EXTENT_LAST
;
5140 int blockbits
= inode
->i_sb
->s_blocksize_bits
;
5144 if (ext4_test_inode_state(inode
, EXT4_STATE_XATTR
)) {
5145 struct ext4_iloc iloc
;
5146 int offset
; /* offset of xattr in inode */
5148 error
= ext4_get_inode_loc(inode
, &iloc
);
5151 physical
= (__u64
)iloc
.bh
->b_blocknr
<< blockbits
;
5152 offset
= EXT4_GOOD_OLD_INODE_SIZE
+
5153 EXT4_I(inode
)->i_extra_isize
;
5155 length
= EXT4_SB(inode
->i_sb
)->s_inode_size
- offset
;
5156 flags
|= FIEMAP_EXTENT_DATA_INLINE
;
5158 } else { /* external block */
5159 physical
= (__u64
)EXT4_I(inode
)->i_file_acl
<< blockbits
;
5160 length
= inode
->i_sb
->s_blocksize
;
5164 error
= fiemap_fill_next_extent(fieinfo
, 0, physical
,
5166 return (error
< 0 ? error
: 0);
5169 int ext4_fiemap(struct inode
*inode
, struct fiemap_extent_info
*fieinfo
,
5170 __u64 start
, __u64 len
)
5172 ext4_lblk_t start_blk
;
5175 if (ext4_has_inline_data(inode
)) {
5178 error
= ext4_inline_data_fiemap(inode
, fieinfo
, &has_inline
,
5185 if (fieinfo
->fi_flags
& FIEMAP_FLAG_CACHE
) {
5186 error
= ext4_ext_precache(inode
);
5191 /* fallback to generic here if not in extents fmt */
5192 if (!(ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
)))
5193 return generic_block_fiemap(inode
, fieinfo
, start
, len
,
5196 if (fiemap_check_flags(fieinfo
, EXT4_FIEMAP_FLAGS
))
5199 if (fieinfo
->fi_flags
& FIEMAP_FLAG_XATTR
) {
5200 error
= ext4_xattr_fiemap(inode
, fieinfo
);
5202 ext4_lblk_t len_blks
;
5205 start_blk
= start
>> inode
->i_sb
->s_blocksize_bits
;
5206 last_blk
= (start
+ len
- 1) >> inode
->i_sb
->s_blocksize_bits
;
5207 if (last_blk
>= EXT_MAX_BLOCKS
)
5208 last_blk
= EXT_MAX_BLOCKS
-1;
5209 len_blks
= ((ext4_lblk_t
) last_blk
) - start_blk
+ 1;
5212 * Walk the extent tree gathering extent information
5213 * and pushing extents back to the user.
5215 error
= ext4_fill_fiemap_extents(inode
, start_blk
,
5223 * Function to access the path buffer for marking it dirty.
5224 * It also checks if there are sufficient credits left in the journal handle
5228 ext4_access_path(handle_t
*handle
, struct inode
*inode
,
5229 struct ext4_ext_path
*path
)
5233 if (!ext4_handle_valid(handle
))
5237 * Check if need to extend journal credits
5238 * 3 for leaf, sb, and inode plus 2 (bmap and group
5239 * descriptor) for each block group; assume two block
5242 if (handle
->h_buffer_credits
< 7) {
5243 credits
= ext4_writepage_trans_blocks(inode
);
5244 err
= ext4_ext_truncate_extend_restart(handle
, inode
, credits
);
5245 /* EAGAIN is success */
5246 if (err
&& err
!= -EAGAIN
)
5250 err
= ext4_ext_get_access(handle
, inode
, path
);
5255 * ext4_ext_shift_path_extents:
5256 * Shift the extents of a path structure lying between path[depth].p_ext
5257 * and EXT_LAST_EXTENT(path[depth].p_hdr), by @shift blocks. @SHIFT tells
5258 * if it is right shift or left shift operation.
5261 ext4_ext_shift_path_extents(struct ext4_ext_path
*path
, ext4_lblk_t shift
,
5262 struct inode
*inode
, handle_t
*handle
,
5263 enum SHIFT_DIRECTION SHIFT
)
5266 struct ext4_extent
*ex_start
, *ex_last
;
5268 depth
= path
->p_depth
;
5270 while (depth
>= 0) {
5271 if (depth
== path
->p_depth
) {
5272 ex_start
= path
[depth
].p_ext
;
5274 return -EFSCORRUPTED
;
5276 ex_last
= EXT_LAST_EXTENT(path
[depth
].p_hdr
);
5278 err
= ext4_access_path(handle
, inode
, path
+ depth
);
5282 if (ex_start
== EXT_FIRST_EXTENT(path
[depth
].p_hdr
))
5285 while (ex_start
<= ex_last
) {
5286 if (SHIFT
== SHIFT_LEFT
) {
5287 le32_add_cpu(&ex_start
->ee_block
,
5289 /* Try to merge to the left. */
5291 EXT_FIRST_EXTENT(path
[depth
].p_hdr
))
5293 ext4_ext_try_to_merge_right(inode
,
5294 path
, ex_start
- 1))
5299 le32_add_cpu(&ex_last
->ee_block
, shift
);
5300 ext4_ext_try_to_merge_right(inode
, path
,
5305 err
= ext4_ext_dirty(handle
, inode
, path
+ depth
);
5309 if (--depth
< 0 || !update
)
5313 /* Update index too */
5314 err
= ext4_access_path(handle
, inode
, path
+ depth
);
5318 if (SHIFT
== SHIFT_LEFT
)
5319 le32_add_cpu(&path
[depth
].p_idx
->ei_block
, -shift
);
5321 le32_add_cpu(&path
[depth
].p_idx
->ei_block
, shift
);
5322 err
= ext4_ext_dirty(handle
, inode
, path
+ depth
);
5326 /* we are done if current index is not a starting index */
5327 if (path
[depth
].p_idx
!= EXT_FIRST_INDEX(path
[depth
].p_hdr
))
5338 * ext4_ext_shift_extents:
5339 * All the extents which lies in the range from @start to the last allocated
5340 * block for the @inode are shifted either towards left or right (depending
5341 * upon @SHIFT) by @shift blocks.
5342 * On success, 0 is returned, error otherwise.
5345 ext4_ext_shift_extents(struct inode
*inode
, handle_t
*handle
,
5346 ext4_lblk_t start
, ext4_lblk_t shift
,
5347 enum SHIFT_DIRECTION SHIFT
)
5349 struct ext4_ext_path
*path
;
5351 struct ext4_extent
*extent
;
5352 ext4_lblk_t stop
, *iterator
, ex_start
, ex_end
;
5354 /* Let path point to the last extent */
5355 path
= ext4_find_extent(inode
, EXT_MAX_BLOCKS
- 1, NULL
,
5358 return PTR_ERR(path
);
5360 depth
= path
->p_depth
;
5361 extent
= path
[depth
].p_ext
;
5365 stop
= le32_to_cpu(extent
->ee_block
);
5368 * For left shifts, make sure the hole on the left is big enough to
5369 * accommodate the shift. For right shifts, make sure the last extent
5370 * won't be shifted beyond EXT_MAX_BLOCKS.
5372 if (SHIFT
== SHIFT_LEFT
) {
5373 path
= ext4_find_extent(inode
, start
- 1, &path
,
5376 return PTR_ERR(path
);
5377 depth
= path
->p_depth
;
5378 extent
= path
[depth
].p_ext
;
5380 ex_start
= le32_to_cpu(extent
->ee_block
);
5381 ex_end
= le32_to_cpu(extent
->ee_block
) +
5382 ext4_ext_get_actual_len(extent
);
5388 if ((start
== ex_start
&& shift
> ex_start
) ||
5389 (shift
> start
- ex_end
)) {
5394 if (shift
> EXT_MAX_BLOCKS
-
5395 (stop
+ ext4_ext_get_actual_len(extent
))) {
5402 * In case of left shift, iterator points to start and it is increased
5403 * till we reach stop. In case of right shift, iterator points to stop
5404 * and it is decreased till we reach start.
5406 if (SHIFT
== SHIFT_LEFT
)
5412 * Its safe to start updating extents. Start and stop are unsigned, so
5413 * in case of right shift if extent with 0 block is reached, iterator
5414 * becomes NULL to indicate the end of the loop.
5416 while (iterator
&& start
<= stop
) {
5417 path
= ext4_find_extent(inode
, *iterator
, &path
,
5420 return PTR_ERR(path
);
5421 depth
= path
->p_depth
;
5422 extent
= path
[depth
].p_ext
;
5424 EXT4_ERROR_INODE(inode
, "unexpected hole at %lu",
5425 (unsigned long) *iterator
);
5426 return -EFSCORRUPTED
;
5428 if (SHIFT
== SHIFT_LEFT
&& *iterator
>
5429 le32_to_cpu(extent
->ee_block
)) {
5430 /* Hole, move to the next extent */
5431 if (extent
< EXT_LAST_EXTENT(path
[depth
].p_hdr
)) {
5432 path
[depth
].p_ext
++;
5434 *iterator
= ext4_ext_next_allocated_block(path
);
5439 if (SHIFT
== SHIFT_LEFT
) {
5440 extent
= EXT_LAST_EXTENT(path
[depth
].p_hdr
);
5441 *iterator
= le32_to_cpu(extent
->ee_block
) +
5442 ext4_ext_get_actual_len(extent
);
5444 extent
= EXT_FIRST_EXTENT(path
[depth
].p_hdr
);
5445 if (le32_to_cpu(extent
->ee_block
) > 0)
5446 *iterator
= le32_to_cpu(extent
->ee_block
) - 1;
5448 /* Beginning is reached, end of the loop */
5450 /* Update path extent in case we need to stop */
5451 while (le32_to_cpu(extent
->ee_block
) < start
)
5453 path
[depth
].p_ext
= extent
;
5455 ret
= ext4_ext_shift_path_extents(path
, shift
, inode
,
5461 ext4_ext_drop_refs(path
);
5467 * ext4_collapse_range:
5468 * This implements the fallocate's collapse range functionality for ext4
5469 * Returns: 0 and non-zero on error.
5471 int ext4_collapse_range(struct inode
*inode
, loff_t offset
, loff_t len
)
5473 struct super_block
*sb
= inode
->i_sb
;
5474 ext4_lblk_t punch_start
, punch_stop
;
5476 unsigned int credits
;
5477 loff_t new_size
, ioffset
;
5481 * We need to test this early because xfstests assumes that a
5482 * collapse range of (0, 1) will return EOPNOTSUPP if the file
5483 * system does not support collapse range.
5485 if (!ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
))
5488 /* Collapse range works only on fs block size aligned offsets. */
5489 if (offset
& (EXT4_CLUSTER_SIZE(sb
) - 1) ||
5490 len
& (EXT4_CLUSTER_SIZE(sb
) - 1))
5493 if (!S_ISREG(inode
->i_mode
))
5496 trace_ext4_collapse_range(inode
, offset
, len
);
5498 punch_start
= offset
>> EXT4_BLOCK_SIZE_BITS(sb
);
5499 punch_stop
= (offset
+ len
) >> EXT4_BLOCK_SIZE_BITS(sb
);
5501 /* Call ext4_force_commit to flush all data in case of data=journal. */
5502 if (ext4_should_journal_data(inode
)) {
5503 ret
= ext4_force_commit(inode
->i_sb
);
5510 * There is no need to overlap collapse range with EOF, in which case
5511 * it is effectively a truncate operation
5513 if (offset
+ len
>= i_size_read(inode
)) {
5518 /* Currently just for extent based files */
5519 if (!ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
)) {
5524 /* Wait for existing dio to complete */
5525 ext4_inode_block_unlocked_dio(inode
);
5526 inode_dio_wait(inode
);
5529 * Prevent page faults from reinstantiating pages we have released from
5532 down_write(&EXT4_I(inode
)->i_mmap_sem
);
5534 * Need to round down offset to be aligned with page size boundary
5535 * for page size > block size.
5537 ioffset
= round_down(offset
, PAGE_SIZE
);
5539 * Write tail of the last page before removed range since it will get
5540 * removed from the page cache below.
5542 ret
= filemap_write_and_wait_range(inode
->i_mapping
, ioffset
, offset
);
5546 * Write data that will be shifted to preserve them when discarding
5547 * page cache below. We are also protected from pages becoming dirty
5550 ret
= filemap_write_and_wait_range(inode
->i_mapping
, offset
+ len
,
5554 truncate_pagecache(inode
, ioffset
);
5556 credits
= ext4_writepage_trans_blocks(inode
);
5557 handle
= ext4_journal_start(inode
, EXT4_HT_TRUNCATE
, credits
);
5558 if (IS_ERR(handle
)) {
5559 ret
= PTR_ERR(handle
);
5563 down_write(&EXT4_I(inode
)->i_data_sem
);
5564 ext4_discard_preallocations(inode
);
5566 ret
= ext4_es_remove_extent(inode
, punch_start
,
5567 EXT_MAX_BLOCKS
- punch_start
);
5569 up_write(&EXT4_I(inode
)->i_data_sem
);
5573 ret
= ext4_ext_remove_space(inode
, punch_start
, punch_stop
- 1);
5575 up_write(&EXT4_I(inode
)->i_data_sem
);
5578 ext4_discard_preallocations(inode
);
5580 ret
= ext4_ext_shift_extents(inode
, handle
, punch_stop
,
5581 punch_stop
- punch_start
, SHIFT_LEFT
);
5583 up_write(&EXT4_I(inode
)->i_data_sem
);
5587 new_size
= i_size_read(inode
) - len
;
5588 i_size_write(inode
, new_size
);
5589 EXT4_I(inode
)->i_disksize
= new_size
;
5591 up_write(&EXT4_I(inode
)->i_data_sem
);
5593 ext4_handle_sync(handle
);
5594 inode
->i_mtime
= inode
->i_ctime
= current_time(inode
);
5595 ext4_mark_inode_dirty(handle
, inode
);
5596 ext4_update_inode_fsync_trans(handle
, inode
, 1);
5599 ext4_journal_stop(handle
);
5601 up_write(&EXT4_I(inode
)->i_mmap_sem
);
5602 ext4_inode_resume_unlocked_dio(inode
);
5604 inode_unlock(inode
);
5609 * ext4_insert_range:
5610 * This function implements the FALLOC_FL_INSERT_RANGE flag of fallocate.
5611 * The data blocks starting from @offset to the EOF are shifted by @len
5612 * towards right to create a hole in the @inode. Inode size is increased
5614 * Returns 0 on success, error otherwise.
5616 int ext4_insert_range(struct inode
*inode
, loff_t offset
, loff_t len
)
5618 struct super_block
*sb
= inode
->i_sb
;
5620 struct ext4_ext_path
*path
;
5621 struct ext4_extent
*extent
;
5622 ext4_lblk_t offset_lblk
, len_lblk
, ee_start_lblk
= 0;
5623 unsigned int credits
, ee_len
;
5624 int ret
= 0, depth
, split_flag
= 0;
5628 * We need to test this early because xfstests assumes that an
5629 * insert range of (0, 1) will return EOPNOTSUPP if the file
5630 * system does not support insert range.
5632 if (!ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
))
5635 /* Insert range works only on fs block size aligned offsets. */
5636 if (offset
& (EXT4_CLUSTER_SIZE(sb
) - 1) ||
5637 len
& (EXT4_CLUSTER_SIZE(sb
) - 1))
5640 if (!S_ISREG(inode
->i_mode
))
5643 trace_ext4_insert_range(inode
, offset
, len
);
5645 offset_lblk
= offset
>> EXT4_BLOCK_SIZE_BITS(sb
);
5646 len_lblk
= len
>> EXT4_BLOCK_SIZE_BITS(sb
);
5648 /* Call ext4_force_commit to flush all data in case of data=journal */
5649 if (ext4_should_journal_data(inode
)) {
5650 ret
= ext4_force_commit(inode
->i_sb
);
5656 /* Currently just for extent based files */
5657 if (!ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
)) {
5662 /* Check for wrap through zero */
5663 if (inode
->i_size
+ len
> inode
->i_sb
->s_maxbytes
) {
5668 /* Offset should be less than i_size */
5669 if (offset
>= i_size_read(inode
)) {
5674 /* Wait for existing dio to complete */
5675 ext4_inode_block_unlocked_dio(inode
);
5676 inode_dio_wait(inode
);
5679 * Prevent page faults from reinstantiating pages we have released from
5682 down_write(&EXT4_I(inode
)->i_mmap_sem
);
5684 * Need to round down to align start offset to page size boundary
5685 * for page size > block size.
5687 ioffset
= round_down(offset
, PAGE_SIZE
);
5688 /* Write out all dirty pages */
5689 ret
= filemap_write_and_wait_range(inode
->i_mapping
, ioffset
,
5693 truncate_pagecache(inode
, ioffset
);
5695 credits
= ext4_writepage_trans_blocks(inode
);
5696 handle
= ext4_journal_start(inode
, EXT4_HT_TRUNCATE
, credits
);
5697 if (IS_ERR(handle
)) {
5698 ret
= PTR_ERR(handle
);
5702 /* Expand file to avoid data loss if there is error while shifting */
5703 inode
->i_size
+= len
;
5704 EXT4_I(inode
)->i_disksize
+= len
;
5705 inode
->i_mtime
= inode
->i_ctime
= current_time(inode
);
5706 ret
= ext4_mark_inode_dirty(handle
, inode
);
5710 down_write(&EXT4_I(inode
)->i_data_sem
);
5711 ext4_discard_preallocations(inode
);
5713 path
= ext4_find_extent(inode
, offset_lblk
, NULL
, 0);
5715 up_write(&EXT4_I(inode
)->i_data_sem
);
5719 depth
= ext_depth(inode
);
5720 extent
= path
[depth
].p_ext
;
5722 ee_start_lblk
= le32_to_cpu(extent
->ee_block
);
5723 ee_len
= ext4_ext_get_actual_len(extent
);
5726 * If offset_lblk is not the starting block of extent, split
5727 * the extent @offset_lblk
5729 if ((offset_lblk
> ee_start_lblk
) &&
5730 (offset_lblk
< (ee_start_lblk
+ ee_len
))) {
5731 if (ext4_ext_is_unwritten(extent
))
5732 split_flag
= EXT4_EXT_MARK_UNWRIT1
|
5733 EXT4_EXT_MARK_UNWRIT2
;
5734 ret
= ext4_split_extent_at(handle
, inode
, &path
,
5735 offset_lblk
, split_flag
,
5737 EXT4_GET_BLOCKS_PRE_IO
|
5738 EXT4_GET_BLOCKS_METADATA_NOFAIL
);
5741 ext4_ext_drop_refs(path
);
5744 up_write(&EXT4_I(inode
)->i_data_sem
);
5748 ext4_ext_drop_refs(path
);
5752 ret
= ext4_es_remove_extent(inode
, offset_lblk
,
5753 EXT_MAX_BLOCKS
- offset_lblk
);
5755 up_write(&EXT4_I(inode
)->i_data_sem
);
5760 * if offset_lblk lies in a hole which is at start of file, use
5761 * ee_start_lblk to shift extents
5763 ret
= ext4_ext_shift_extents(inode
, handle
,
5764 ee_start_lblk
> offset_lblk
? ee_start_lblk
: offset_lblk
,
5765 len_lblk
, SHIFT_RIGHT
);
5767 up_write(&EXT4_I(inode
)->i_data_sem
);
5769 ext4_handle_sync(handle
);
5771 ext4_update_inode_fsync_trans(handle
, inode
, 1);
5774 ext4_journal_stop(handle
);
5776 up_write(&EXT4_I(inode
)->i_mmap_sem
);
5777 ext4_inode_resume_unlocked_dio(inode
);
5779 inode_unlock(inode
);
5784 * ext4_swap_extents - Swap extents between two inodes
5786 * @inode1: First inode
5787 * @inode2: Second inode
5788 * @lblk1: Start block for first inode
5789 * @lblk2: Start block for second inode
5790 * @count: Number of blocks to swap
5791 * @mark_unwritten: Mark second inode's extents as unwritten after swap
5792 * @erp: Pointer to save error value
5794 * This helper routine does exactly what is promise "swap extents". All other
5795 * stuff such as page-cache locking consistency, bh mapping consistency or
5796 * extent's data copying must be performed by caller.
5798 * i_mutex is held for both inodes
5799 * i_data_sem is locked for write for both inodes
5801 * All pages from requested range are locked for both inodes
5804 ext4_swap_extents(handle_t
*handle
, struct inode
*inode1
,
5805 struct inode
*inode2
, ext4_lblk_t lblk1
, ext4_lblk_t lblk2
,
5806 ext4_lblk_t count
, int unwritten
, int *erp
)
5808 struct ext4_ext_path
*path1
= NULL
;
5809 struct ext4_ext_path
*path2
= NULL
;
5810 int replaced_count
= 0;
5812 BUG_ON(!rwsem_is_locked(&EXT4_I(inode1
)->i_data_sem
));
5813 BUG_ON(!rwsem_is_locked(&EXT4_I(inode2
)->i_data_sem
));
5814 BUG_ON(!inode_is_locked(inode1
));
5815 BUG_ON(!inode_is_locked(inode2
));
5817 *erp
= ext4_es_remove_extent(inode1
, lblk1
, count
);
5820 *erp
= ext4_es_remove_extent(inode2
, lblk2
, count
);
5825 struct ext4_extent
*ex1
, *ex2
, tmp_ex
;
5826 ext4_lblk_t e1_blk
, e2_blk
;
5827 int e1_len
, e2_len
, len
;
5830 path1
= ext4_find_extent(inode1
, lblk1
, NULL
, EXT4_EX_NOCACHE
);
5831 if (IS_ERR(path1
)) {
5832 *erp
= PTR_ERR(path1
);
5838 path2
= ext4_find_extent(inode2
, lblk2
, NULL
, EXT4_EX_NOCACHE
);
5839 if (IS_ERR(path2
)) {
5840 *erp
= PTR_ERR(path2
);
5844 ex1
= path1
[path1
->p_depth
].p_ext
;
5845 ex2
= path2
[path2
->p_depth
].p_ext
;
5846 /* Do we have somthing to swap ? */
5847 if (unlikely(!ex2
|| !ex1
))
5850 e1_blk
= le32_to_cpu(ex1
->ee_block
);
5851 e2_blk
= le32_to_cpu(ex2
->ee_block
);
5852 e1_len
= ext4_ext_get_actual_len(ex1
);
5853 e2_len
= ext4_ext_get_actual_len(ex2
);
5856 if (!in_range(lblk1
, e1_blk
, e1_len
) ||
5857 !in_range(lblk2
, e2_blk
, e2_len
)) {
5858 ext4_lblk_t next1
, next2
;
5860 /* if hole after extent, then go to next extent */
5861 next1
= ext4_ext_next_allocated_block(path1
);
5862 next2
= ext4_ext_next_allocated_block(path2
);
5863 /* If hole before extent, then shift to that extent */
5868 /* Do we have something to swap */
5869 if (next1
== EXT_MAX_BLOCKS
|| next2
== EXT_MAX_BLOCKS
)
5871 /* Move to the rightest boundary */
5872 len
= next1
- lblk1
;
5873 if (len
< next2
- lblk2
)
5874 len
= next2
- lblk2
;
5883 /* Prepare left boundary */
5884 if (e1_blk
< lblk1
) {
5886 *erp
= ext4_force_split_extent_at(handle
, inode1
,
5891 if (e2_blk
< lblk2
) {
5893 *erp
= ext4_force_split_extent_at(handle
, inode2
,
5898 /* ext4_split_extent_at() may result in leaf extent split,
5899 * path must to be revalidated. */
5903 /* Prepare right boundary */
5905 if (len
> e1_blk
+ e1_len
- lblk1
)
5906 len
= e1_blk
+ e1_len
- lblk1
;
5907 if (len
> e2_blk
+ e2_len
- lblk2
)
5908 len
= e2_blk
+ e2_len
- lblk2
;
5910 if (len
!= e1_len
) {
5912 *erp
= ext4_force_split_extent_at(handle
, inode1
,
5913 &path1
, lblk1
+ len
, 0);
5917 if (len
!= e2_len
) {
5919 *erp
= ext4_force_split_extent_at(handle
, inode2
,
5920 &path2
, lblk2
+ len
, 0);
5924 /* ext4_split_extent_at() may result in leaf extent split,
5925 * path must to be revalidated. */
5929 BUG_ON(e2_len
!= e1_len
);
5930 *erp
= ext4_ext_get_access(handle
, inode1
, path1
+ path1
->p_depth
);
5933 *erp
= ext4_ext_get_access(handle
, inode2
, path2
+ path2
->p_depth
);
5937 /* Both extents are fully inside boundaries. Swap it now */
5939 ext4_ext_store_pblock(ex1
, ext4_ext_pblock(ex2
));
5940 ext4_ext_store_pblock(ex2
, ext4_ext_pblock(&tmp_ex
));
5941 ex1
->ee_len
= cpu_to_le16(e2_len
);
5942 ex2
->ee_len
= cpu_to_le16(e1_len
);
5944 ext4_ext_mark_unwritten(ex2
);
5945 if (ext4_ext_is_unwritten(&tmp_ex
))
5946 ext4_ext_mark_unwritten(ex1
);
5948 ext4_ext_try_to_merge(handle
, inode2
, path2
, ex2
);
5949 ext4_ext_try_to_merge(handle
, inode1
, path1
, ex1
);
5950 *erp
= ext4_ext_dirty(handle
, inode2
, path2
+
5954 *erp
= ext4_ext_dirty(handle
, inode1
, path1
+
5957 * Looks scarry ah..? second inode already points to new blocks,
5958 * and it was successfully dirtied. But luckily error may happen
5959 * only due to journal error, so full transaction will be
5966 replaced_count
+= len
;
5970 ext4_ext_drop_refs(path1
);
5972 ext4_ext_drop_refs(path2
);
5974 path1
= path2
= NULL
;
5976 return replaced_count
;