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 if (!ext4_has_feature_journal(inode
->i_sb
) ||
535 le32_to_cpu(EXT4_SB(inode
->i_sb
)->s_es
->s_journal_inum
))) {
536 err
= __ext4_ext_check(function
, line
, inode
,
537 ext_block_hdr(bh
), depth
, pblk
);
541 set_buffer_verified(bh
);
543 * If this is a leaf block, cache all of its entries
545 if (!(flags
& EXT4_EX_NOCACHE
) && depth
== 0) {
546 struct ext4_extent_header
*eh
= ext_block_hdr(bh
);
547 struct ext4_extent
*ex
= EXT_FIRST_EXTENT(eh
);
548 ext4_lblk_t prev
= 0;
551 for (i
= le16_to_cpu(eh
->eh_entries
); i
> 0; i
--, ex
++) {
552 unsigned int status
= EXTENT_STATUS_WRITTEN
;
553 ext4_lblk_t lblk
= le32_to_cpu(ex
->ee_block
);
554 int len
= ext4_ext_get_actual_len(ex
);
556 if (prev
&& (prev
!= lblk
))
557 ext4_es_cache_extent(inode
, prev
,
561 if (ext4_ext_is_unwritten(ex
))
562 status
= EXTENT_STATUS_UNWRITTEN
;
563 ext4_es_cache_extent(inode
, lblk
, len
,
564 ext4_ext_pblock(ex
), status
);
575 #define read_extent_tree_block(inode, pblk, depth, flags) \
576 __read_extent_tree_block(__func__, __LINE__, (inode), (pblk), \
580 * This function is called to cache a file's extent information in the
583 int ext4_ext_precache(struct inode
*inode
)
585 struct ext4_inode_info
*ei
= EXT4_I(inode
);
586 struct ext4_ext_path
*path
= NULL
;
587 struct buffer_head
*bh
;
588 int i
= 0, depth
, ret
= 0;
590 if (!ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
))
591 return 0; /* not an extent-mapped inode */
593 down_read(&ei
->i_data_sem
);
594 depth
= ext_depth(inode
);
596 path
= kzalloc(sizeof(struct ext4_ext_path
) * (depth
+ 1),
599 up_read(&ei
->i_data_sem
);
603 /* Don't cache anything if there are no external extent blocks */
606 path
[0].p_hdr
= ext_inode_hdr(inode
);
607 ret
= ext4_ext_check(inode
, path
[0].p_hdr
, depth
, 0);
610 path
[0].p_idx
= EXT_FIRST_INDEX(path
[0].p_hdr
);
613 * If this is a leaf block or we've reached the end of
614 * the index block, go up
617 path
[i
].p_idx
> EXT_LAST_INDEX(path
[i
].p_hdr
)) {
618 brelse(path
[i
].p_bh
);
623 bh
= read_extent_tree_block(inode
,
624 ext4_idx_pblock(path
[i
].p_idx
++),
626 EXT4_EX_FORCE_CACHE
);
633 path
[i
].p_hdr
= ext_block_hdr(bh
);
634 path
[i
].p_idx
= EXT_FIRST_INDEX(path
[i
].p_hdr
);
636 ext4_set_inode_state(inode
, EXT4_STATE_EXT_PRECACHED
);
638 up_read(&ei
->i_data_sem
);
639 ext4_ext_drop_refs(path
);
645 static void ext4_ext_show_path(struct inode
*inode
, struct ext4_ext_path
*path
)
647 int k
, l
= path
->p_depth
;
650 for (k
= 0; k
<= l
; k
++, path
++) {
652 ext_debug(" %d->%llu", le32_to_cpu(path
->p_idx
->ei_block
),
653 ext4_idx_pblock(path
->p_idx
));
654 } else if (path
->p_ext
) {
655 ext_debug(" %d:[%d]%d:%llu ",
656 le32_to_cpu(path
->p_ext
->ee_block
),
657 ext4_ext_is_unwritten(path
->p_ext
),
658 ext4_ext_get_actual_len(path
->p_ext
),
659 ext4_ext_pblock(path
->p_ext
));
666 static void ext4_ext_show_leaf(struct inode
*inode
, struct ext4_ext_path
*path
)
668 int depth
= ext_depth(inode
);
669 struct ext4_extent_header
*eh
;
670 struct ext4_extent
*ex
;
676 eh
= path
[depth
].p_hdr
;
677 ex
= EXT_FIRST_EXTENT(eh
);
679 ext_debug("Displaying leaf extents for inode %lu\n", inode
->i_ino
);
681 for (i
= 0; i
< le16_to_cpu(eh
->eh_entries
); i
++, ex
++) {
682 ext_debug("%d:[%d]%d:%llu ", le32_to_cpu(ex
->ee_block
),
683 ext4_ext_is_unwritten(ex
),
684 ext4_ext_get_actual_len(ex
), ext4_ext_pblock(ex
));
689 static void ext4_ext_show_move(struct inode
*inode
, struct ext4_ext_path
*path
,
690 ext4_fsblk_t newblock
, int level
)
692 int depth
= ext_depth(inode
);
693 struct ext4_extent
*ex
;
695 if (depth
!= level
) {
696 struct ext4_extent_idx
*idx
;
697 idx
= path
[level
].p_idx
;
698 while (idx
<= EXT_MAX_INDEX(path
[level
].p_hdr
)) {
699 ext_debug("%d: move %d:%llu in new index %llu\n", level
,
700 le32_to_cpu(idx
->ei_block
),
701 ext4_idx_pblock(idx
),
709 ex
= path
[depth
].p_ext
;
710 while (ex
<= EXT_MAX_EXTENT(path
[depth
].p_hdr
)) {
711 ext_debug("move %d:%llu:[%d]%d in new leaf %llu\n",
712 le32_to_cpu(ex
->ee_block
),
714 ext4_ext_is_unwritten(ex
),
715 ext4_ext_get_actual_len(ex
),
722 #define ext4_ext_show_path(inode, path)
723 #define ext4_ext_show_leaf(inode, path)
724 #define ext4_ext_show_move(inode, path, newblock, level)
727 void ext4_ext_drop_refs(struct ext4_ext_path
*path
)
733 depth
= path
->p_depth
;
734 for (i
= 0; i
<= depth
; i
++, path
++)
742 * ext4_ext_binsearch_idx:
743 * binary search for the closest index of the given block
744 * the header must be checked before calling this
747 ext4_ext_binsearch_idx(struct inode
*inode
,
748 struct ext4_ext_path
*path
, ext4_lblk_t block
)
750 struct ext4_extent_header
*eh
= path
->p_hdr
;
751 struct ext4_extent_idx
*r
, *l
, *m
;
754 ext_debug("binsearch for %u(idx): ", block
);
756 l
= EXT_FIRST_INDEX(eh
) + 1;
757 r
= EXT_LAST_INDEX(eh
);
760 if (block
< le32_to_cpu(m
->ei_block
))
764 ext_debug("%p(%u):%p(%u):%p(%u) ", l
, le32_to_cpu(l
->ei_block
),
765 m
, le32_to_cpu(m
->ei_block
),
766 r
, le32_to_cpu(r
->ei_block
));
770 ext_debug(" -> %u->%lld ", le32_to_cpu(path
->p_idx
->ei_block
),
771 ext4_idx_pblock(path
->p_idx
));
773 #ifdef CHECK_BINSEARCH
775 struct ext4_extent_idx
*chix
, *ix
;
778 chix
= ix
= EXT_FIRST_INDEX(eh
);
779 for (k
= 0; k
< le16_to_cpu(eh
->eh_entries
); k
++, ix
++) {
781 le32_to_cpu(ix
->ei_block
) <= le32_to_cpu(ix
[-1].ei_block
)) {
782 printk(KERN_DEBUG
"k=%d, ix=0x%p, "
784 ix
, EXT_FIRST_INDEX(eh
));
785 printk(KERN_DEBUG
"%u <= %u\n",
786 le32_to_cpu(ix
->ei_block
),
787 le32_to_cpu(ix
[-1].ei_block
));
789 BUG_ON(k
&& le32_to_cpu(ix
->ei_block
)
790 <= le32_to_cpu(ix
[-1].ei_block
));
791 if (block
< le32_to_cpu(ix
->ei_block
))
795 BUG_ON(chix
!= path
->p_idx
);
802 * ext4_ext_binsearch:
803 * binary search for closest extent of the given block
804 * the header must be checked before calling this
807 ext4_ext_binsearch(struct inode
*inode
,
808 struct ext4_ext_path
*path
, ext4_lblk_t block
)
810 struct ext4_extent_header
*eh
= path
->p_hdr
;
811 struct ext4_extent
*r
, *l
, *m
;
813 if (eh
->eh_entries
== 0) {
815 * this leaf is empty:
816 * we get such a leaf in split/add case
821 ext_debug("binsearch for %u: ", block
);
823 l
= EXT_FIRST_EXTENT(eh
) + 1;
824 r
= EXT_LAST_EXTENT(eh
);
828 if (block
< le32_to_cpu(m
->ee_block
))
832 ext_debug("%p(%u):%p(%u):%p(%u) ", l
, le32_to_cpu(l
->ee_block
),
833 m
, le32_to_cpu(m
->ee_block
),
834 r
, le32_to_cpu(r
->ee_block
));
838 ext_debug(" -> %d:%llu:[%d]%d ",
839 le32_to_cpu(path
->p_ext
->ee_block
),
840 ext4_ext_pblock(path
->p_ext
),
841 ext4_ext_is_unwritten(path
->p_ext
),
842 ext4_ext_get_actual_len(path
->p_ext
));
844 #ifdef CHECK_BINSEARCH
846 struct ext4_extent
*chex
, *ex
;
849 chex
= ex
= EXT_FIRST_EXTENT(eh
);
850 for (k
= 0; k
< le16_to_cpu(eh
->eh_entries
); k
++, ex
++) {
851 BUG_ON(k
&& le32_to_cpu(ex
->ee_block
)
852 <= le32_to_cpu(ex
[-1].ee_block
));
853 if (block
< le32_to_cpu(ex
->ee_block
))
857 BUG_ON(chex
!= path
->p_ext
);
863 int ext4_ext_tree_init(handle_t
*handle
, struct inode
*inode
)
865 struct ext4_extent_header
*eh
;
867 eh
= ext_inode_hdr(inode
);
870 eh
->eh_magic
= EXT4_EXT_MAGIC
;
871 eh
->eh_max
= cpu_to_le16(ext4_ext_space_root(inode
, 0));
872 ext4_mark_inode_dirty(handle
, inode
);
876 struct ext4_ext_path
*
877 ext4_find_extent(struct inode
*inode
, ext4_lblk_t block
,
878 struct ext4_ext_path
**orig_path
, int flags
)
880 struct ext4_extent_header
*eh
;
881 struct buffer_head
*bh
;
882 struct ext4_ext_path
*path
= orig_path
? *orig_path
: NULL
;
883 short int depth
, i
, ppos
= 0;
886 eh
= ext_inode_hdr(inode
);
887 depth
= ext_depth(inode
);
888 if (depth
< 0 || depth
> EXT4_MAX_EXTENT_DEPTH
) {
889 EXT4_ERROR_INODE(inode
, "inode has invalid extent depth: %d",
896 ext4_ext_drop_refs(path
);
897 if (depth
> path
[0].p_maxdepth
) {
899 *orig_path
= path
= NULL
;
903 /* account possible depth increase */
904 path
= kzalloc(sizeof(struct ext4_ext_path
) * (depth
+ 2),
907 return ERR_PTR(-ENOMEM
);
908 path
[0].p_maxdepth
= depth
+ 1;
914 /* walk through the tree */
916 ext_debug("depth %d: num %d, max %d\n",
917 ppos
, le16_to_cpu(eh
->eh_entries
), le16_to_cpu(eh
->eh_max
));
919 ext4_ext_binsearch_idx(inode
, path
+ ppos
, block
);
920 path
[ppos
].p_block
= ext4_idx_pblock(path
[ppos
].p_idx
);
921 path
[ppos
].p_depth
= i
;
922 path
[ppos
].p_ext
= NULL
;
924 bh
= read_extent_tree_block(inode
, path
[ppos
].p_block
, --i
,
931 eh
= ext_block_hdr(bh
);
933 path
[ppos
].p_bh
= bh
;
934 path
[ppos
].p_hdr
= eh
;
937 path
[ppos
].p_depth
= i
;
938 path
[ppos
].p_ext
= NULL
;
939 path
[ppos
].p_idx
= NULL
;
942 ext4_ext_binsearch(inode
, path
+ ppos
, block
);
943 /* if not an empty leaf */
944 if (path
[ppos
].p_ext
)
945 path
[ppos
].p_block
= ext4_ext_pblock(path
[ppos
].p_ext
);
947 ext4_ext_show_path(inode
, path
);
952 ext4_ext_drop_refs(path
);
960 * ext4_ext_insert_index:
961 * insert new index [@logical;@ptr] into the block at @curp;
962 * check where to insert: before @curp or after @curp
964 static int ext4_ext_insert_index(handle_t
*handle
, struct inode
*inode
,
965 struct ext4_ext_path
*curp
,
966 int logical
, ext4_fsblk_t ptr
)
968 struct ext4_extent_idx
*ix
;
971 err
= ext4_ext_get_access(handle
, inode
, curp
);
975 if (unlikely(logical
== le32_to_cpu(curp
->p_idx
->ei_block
))) {
976 EXT4_ERROR_INODE(inode
,
977 "logical %d == ei_block %d!",
978 logical
, le32_to_cpu(curp
->p_idx
->ei_block
));
979 return -EFSCORRUPTED
;
982 if (unlikely(le16_to_cpu(curp
->p_hdr
->eh_entries
)
983 >= le16_to_cpu(curp
->p_hdr
->eh_max
))) {
984 EXT4_ERROR_INODE(inode
,
985 "eh_entries %d >= eh_max %d!",
986 le16_to_cpu(curp
->p_hdr
->eh_entries
),
987 le16_to_cpu(curp
->p_hdr
->eh_max
));
988 return -EFSCORRUPTED
;
991 if (logical
> le32_to_cpu(curp
->p_idx
->ei_block
)) {
993 ext_debug("insert new index %d after: %llu\n", logical
, ptr
);
994 ix
= curp
->p_idx
+ 1;
997 ext_debug("insert new index %d before: %llu\n", logical
, ptr
);
1001 len
= EXT_LAST_INDEX(curp
->p_hdr
) - ix
+ 1;
1004 ext_debug("insert new index %d: "
1005 "move %d indices from 0x%p to 0x%p\n",
1006 logical
, len
, ix
, ix
+ 1);
1007 memmove(ix
+ 1, ix
, len
* sizeof(struct ext4_extent_idx
));
1010 if (unlikely(ix
> EXT_MAX_INDEX(curp
->p_hdr
))) {
1011 EXT4_ERROR_INODE(inode
, "ix > EXT_MAX_INDEX!");
1012 return -EFSCORRUPTED
;
1015 ix
->ei_block
= cpu_to_le32(logical
);
1016 ext4_idx_store_pblock(ix
, ptr
);
1017 le16_add_cpu(&curp
->p_hdr
->eh_entries
, 1);
1019 if (unlikely(ix
> EXT_LAST_INDEX(curp
->p_hdr
))) {
1020 EXT4_ERROR_INODE(inode
, "ix > EXT_LAST_INDEX!");
1021 return -EFSCORRUPTED
;
1024 err
= ext4_ext_dirty(handle
, inode
, curp
);
1025 ext4_std_error(inode
->i_sb
, err
);
1032 * inserts new subtree into the path, using free index entry
1034 * - allocates all needed blocks (new leaf and all intermediate index blocks)
1035 * - makes decision where to split
1036 * - moves remaining extents and index entries (right to the split point)
1037 * into the newly allocated blocks
1038 * - initializes subtree
1040 static int ext4_ext_split(handle_t
*handle
, struct inode
*inode
,
1042 struct ext4_ext_path
*path
,
1043 struct ext4_extent
*newext
, int at
)
1045 struct buffer_head
*bh
= NULL
;
1046 int depth
= ext_depth(inode
);
1047 struct ext4_extent_header
*neh
;
1048 struct ext4_extent_idx
*fidx
;
1049 int i
= at
, k
, m
, a
;
1050 ext4_fsblk_t newblock
, oldblock
;
1052 ext4_fsblk_t
*ablocks
= NULL
; /* array of allocated blocks */
1054 size_t ext_size
= 0;
1056 /* make decision: where to split? */
1057 /* FIXME: now decision is simplest: at current extent */
1059 /* if current leaf will be split, then we should use
1060 * border from split point */
1061 if (unlikely(path
[depth
].p_ext
> EXT_MAX_EXTENT(path
[depth
].p_hdr
))) {
1062 EXT4_ERROR_INODE(inode
, "p_ext > EXT_MAX_EXTENT!");
1063 return -EFSCORRUPTED
;
1065 if (path
[depth
].p_ext
!= EXT_MAX_EXTENT(path
[depth
].p_hdr
)) {
1066 border
= path
[depth
].p_ext
[1].ee_block
;
1067 ext_debug("leaf will be split."
1068 " next leaf starts at %d\n",
1069 le32_to_cpu(border
));
1071 border
= newext
->ee_block
;
1072 ext_debug("leaf will be added."
1073 " next leaf starts at %d\n",
1074 le32_to_cpu(border
));
1078 * If error occurs, then we break processing
1079 * and mark filesystem read-only. index won't
1080 * be inserted and tree will be in consistent
1081 * state. Next mount will repair buffers too.
1085 * Get array to track all allocated blocks.
1086 * We need this to handle errors and free blocks
1089 ablocks
= kzalloc(sizeof(ext4_fsblk_t
) * depth
, GFP_NOFS
);
1093 /* allocate all needed blocks */
1094 ext_debug("allocate %d blocks for indexes/leaf\n", depth
- at
);
1095 for (a
= 0; a
< depth
- at
; a
++) {
1096 newblock
= ext4_ext_new_meta_block(handle
, inode
, path
,
1097 newext
, &err
, flags
);
1100 ablocks
[a
] = newblock
;
1103 /* initialize new leaf */
1104 newblock
= ablocks
[--a
];
1105 if (unlikely(newblock
== 0)) {
1106 EXT4_ERROR_INODE(inode
, "newblock == 0!");
1107 err
= -EFSCORRUPTED
;
1110 bh
= sb_getblk_gfp(inode
->i_sb
, newblock
, __GFP_MOVABLE
| GFP_NOFS
);
1111 if (unlikely(!bh
)) {
1117 err
= ext4_journal_get_create_access(handle
, bh
);
1121 neh
= ext_block_hdr(bh
);
1122 neh
->eh_entries
= 0;
1123 neh
->eh_max
= cpu_to_le16(ext4_ext_space_block(inode
, 0));
1124 neh
->eh_magic
= EXT4_EXT_MAGIC
;
1127 /* move remainder of path[depth] to the new leaf */
1128 if (unlikely(path
[depth
].p_hdr
->eh_entries
!=
1129 path
[depth
].p_hdr
->eh_max
)) {
1130 EXT4_ERROR_INODE(inode
, "eh_entries %d != eh_max %d!",
1131 path
[depth
].p_hdr
->eh_entries
,
1132 path
[depth
].p_hdr
->eh_max
);
1133 err
= -EFSCORRUPTED
;
1136 /* start copy from next extent */
1137 m
= EXT_MAX_EXTENT(path
[depth
].p_hdr
) - path
[depth
].p_ext
++;
1138 ext4_ext_show_move(inode
, path
, newblock
, depth
);
1140 struct ext4_extent
*ex
;
1141 ex
= EXT_FIRST_EXTENT(neh
);
1142 memmove(ex
, path
[depth
].p_ext
, sizeof(struct ext4_extent
) * m
);
1143 le16_add_cpu(&neh
->eh_entries
, m
);
1146 /* zero out unused area in the extent block */
1147 ext_size
= sizeof(struct ext4_extent_header
) +
1148 sizeof(struct ext4_extent
) * le16_to_cpu(neh
->eh_entries
);
1149 memset(bh
->b_data
+ ext_size
, 0, inode
->i_sb
->s_blocksize
- ext_size
);
1150 ext4_extent_block_csum_set(inode
, neh
);
1151 set_buffer_uptodate(bh
);
1154 err
= ext4_handle_dirty_metadata(handle
, inode
, bh
);
1160 /* correct old leaf */
1162 err
= ext4_ext_get_access(handle
, inode
, path
+ depth
);
1165 le16_add_cpu(&path
[depth
].p_hdr
->eh_entries
, -m
);
1166 err
= ext4_ext_dirty(handle
, inode
, path
+ depth
);
1172 /* create intermediate indexes */
1174 if (unlikely(k
< 0)) {
1175 EXT4_ERROR_INODE(inode
, "k %d < 0!", k
);
1176 err
= -EFSCORRUPTED
;
1180 ext_debug("create %d intermediate indices\n", k
);
1181 /* insert new index into current index block */
1182 /* current depth stored in i var */
1185 oldblock
= newblock
;
1186 newblock
= ablocks
[--a
];
1187 bh
= sb_getblk(inode
->i_sb
, newblock
);
1188 if (unlikely(!bh
)) {
1194 err
= ext4_journal_get_create_access(handle
, bh
);
1198 neh
= ext_block_hdr(bh
);
1199 neh
->eh_entries
= cpu_to_le16(1);
1200 neh
->eh_magic
= EXT4_EXT_MAGIC
;
1201 neh
->eh_max
= cpu_to_le16(ext4_ext_space_block_idx(inode
, 0));
1202 neh
->eh_depth
= cpu_to_le16(depth
- i
);
1203 fidx
= EXT_FIRST_INDEX(neh
);
1204 fidx
->ei_block
= border
;
1205 ext4_idx_store_pblock(fidx
, oldblock
);
1207 ext_debug("int.index at %d (block %llu): %u -> %llu\n",
1208 i
, newblock
, le32_to_cpu(border
), oldblock
);
1210 /* move remainder of path[i] to the new index block */
1211 if (unlikely(EXT_MAX_INDEX(path
[i
].p_hdr
) !=
1212 EXT_LAST_INDEX(path
[i
].p_hdr
))) {
1213 EXT4_ERROR_INODE(inode
,
1214 "EXT_MAX_INDEX != EXT_LAST_INDEX ee_block %d!",
1215 le32_to_cpu(path
[i
].p_ext
->ee_block
));
1216 err
= -EFSCORRUPTED
;
1219 /* start copy indexes */
1220 m
= EXT_MAX_INDEX(path
[i
].p_hdr
) - path
[i
].p_idx
++;
1221 ext_debug("cur 0x%p, last 0x%p\n", path
[i
].p_idx
,
1222 EXT_MAX_INDEX(path
[i
].p_hdr
));
1223 ext4_ext_show_move(inode
, path
, newblock
, i
);
1225 memmove(++fidx
, path
[i
].p_idx
,
1226 sizeof(struct ext4_extent_idx
) * m
);
1227 le16_add_cpu(&neh
->eh_entries
, m
);
1229 /* zero out unused area in the extent block */
1230 ext_size
= sizeof(struct ext4_extent_header
) +
1231 (sizeof(struct ext4_extent
) * le16_to_cpu(neh
->eh_entries
));
1232 memset(bh
->b_data
+ ext_size
, 0,
1233 inode
->i_sb
->s_blocksize
- ext_size
);
1234 ext4_extent_block_csum_set(inode
, neh
);
1235 set_buffer_uptodate(bh
);
1238 err
= ext4_handle_dirty_metadata(handle
, inode
, bh
);
1244 /* correct old index */
1246 err
= ext4_ext_get_access(handle
, inode
, path
+ i
);
1249 le16_add_cpu(&path
[i
].p_hdr
->eh_entries
, -m
);
1250 err
= ext4_ext_dirty(handle
, inode
, path
+ i
);
1258 /* insert new index */
1259 err
= ext4_ext_insert_index(handle
, inode
, path
+ at
,
1260 le32_to_cpu(border
), newblock
);
1264 if (buffer_locked(bh
))
1270 /* free all allocated blocks in error case */
1271 for (i
= 0; i
< depth
; i
++) {
1274 ext4_free_blocks(handle
, inode
, NULL
, ablocks
[i
], 1,
1275 EXT4_FREE_BLOCKS_METADATA
);
1284 * ext4_ext_grow_indepth:
1285 * implements tree growing procedure:
1286 * - allocates new block
1287 * - moves top-level data (index block or leaf) into the new block
1288 * - initializes new top-level, creating index that points to the
1289 * just created block
1291 static int ext4_ext_grow_indepth(handle_t
*handle
, struct inode
*inode
,
1294 struct ext4_extent_header
*neh
;
1295 struct buffer_head
*bh
;
1296 ext4_fsblk_t newblock
, goal
= 0;
1297 struct ext4_super_block
*es
= EXT4_SB(inode
->i_sb
)->s_es
;
1299 size_t ext_size
= 0;
1301 /* Try to prepend new index to old one */
1302 if (ext_depth(inode
))
1303 goal
= ext4_idx_pblock(EXT_FIRST_INDEX(ext_inode_hdr(inode
)));
1304 if (goal
> le32_to_cpu(es
->s_first_data_block
)) {
1305 flags
|= EXT4_MB_HINT_TRY_GOAL
;
1308 goal
= ext4_inode_to_goal_block(inode
);
1309 newblock
= ext4_new_meta_blocks(handle
, inode
, goal
, flags
,
1314 bh
= sb_getblk_gfp(inode
->i_sb
, newblock
, __GFP_MOVABLE
| GFP_NOFS
);
1319 err
= ext4_journal_get_create_access(handle
, bh
);
1325 ext_size
= sizeof(EXT4_I(inode
)->i_data
);
1326 /* move top-level index/leaf into new block */
1327 memmove(bh
->b_data
, EXT4_I(inode
)->i_data
, ext_size
);
1328 /* zero out unused area in the extent block */
1329 memset(bh
->b_data
+ ext_size
, 0, inode
->i_sb
->s_blocksize
- ext_size
);
1331 /* set size of new block */
1332 neh
= ext_block_hdr(bh
);
1333 /* old root could have indexes or leaves
1334 * so calculate e_max right way */
1335 if (ext_depth(inode
))
1336 neh
->eh_max
= cpu_to_le16(ext4_ext_space_block_idx(inode
, 0));
1338 neh
->eh_max
= cpu_to_le16(ext4_ext_space_block(inode
, 0));
1339 neh
->eh_magic
= EXT4_EXT_MAGIC
;
1340 ext4_extent_block_csum_set(inode
, neh
);
1341 set_buffer_uptodate(bh
);
1344 err
= ext4_handle_dirty_metadata(handle
, inode
, bh
);
1348 /* Update top-level index: num,max,pointer */
1349 neh
= ext_inode_hdr(inode
);
1350 neh
->eh_entries
= cpu_to_le16(1);
1351 ext4_idx_store_pblock(EXT_FIRST_INDEX(neh
), newblock
);
1352 if (neh
->eh_depth
== 0) {
1353 /* Root extent block becomes index block */
1354 neh
->eh_max
= cpu_to_le16(ext4_ext_space_root_idx(inode
, 0));
1355 EXT_FIRST_INDEX(neh
)->ei_block
=
1356 EXT_FIRST_EXTENT(neh
)->ee_block
;
1358 ext_debug("new root: num %d(%d), lblock %d, ptr %llu\n",
1359 le16_to_cpu(neh
->eh_entries
), le16_to_cpu(neh
->eh_max
),
1360 le32_to_cpu(EXT_FIRST_INDEX(neh
)->ei_block
),
1361 ext4_idx_pblock(EXT_FIRST_INDEX(neh
)));
1363 le16_add_cpu(&neh
->eh_depth
, 1);
1364 ext4_mark_inode_dirty(handle
, inode
);
1372 * ext4_ext_create_new_leaf:
1373 * finds empty index and adds new leaf.
1374 * if no free index is found, then it requests in-depth growing.
1376 static int ext4_ext_create_new_leaf(handle_t
*handle
, struct inode
*inode
,
1377 unsigned int mb_flags
,
1378 unsigned int gb_flags
,
1379 struct ext4_ext_path
**ppath
,
1380 struct ext4_extent
*newext
)
1382 struct ext4_ext_path
*path
= *ppath
;
1383 struct ext4_ext_path
*curp
;
1384 int depth
, i
, err
= 0;
1387 i
= depth
= ext_depth(inode
);
1389 /* walk up to the tree and look for free index entry */
1390 curp
= path
+ depth
;
1391 while (i
> 0 && !EXT_HAS_FREE_INDEX(curp
)) {
1396 /* we use already allocated block for index block,
1397 * so subsequent data blocks should be contiguous */
1398 if (EXT_HAS_FREE_INDEX(curp
)) {
1399 /* if we found index with free entry, then use that
1400 * entry: create all needed subtree and add new leaf */
1401 err
= ext4_ext_split(handle
, inode
, mb_flags
, path
, newext
, i
);
1406 path
= ext4_find_extent(inode
,
1407 (ext4_lblk_t
)le32_to_cpu(newext
->ee_block
),
1410 err
= PTR_ERR(path
);
1412 /* tree is full, time to grow in depth */
1413 err
= ext4_ext_grow_indepth(handle
, inode
, mb_flags
);
1418 path
= ext4_find_extent(inode
,
1419 (ext4_lblk_t
)le32_to_cpu(newext
->ee_block
),
1422 err
= PTR_ERR(path
);
1427 * only first (depth 0 -> 1) produces free space;
1428 * in all other cases we have to split the grown tree
1430 depth
= ext_depth(inode
);
1431 if (path
[depth
].p_hdr
->eh_entries
== path
[depth
].p_hdr
->eh_max
) {
1432 /* now we need to split */
1442 * search the closest allocated block to the left for *logical
1443 * and returns it at @logical + it's physical address at @phys
1444 * if *logical is the smallest allocated block, the function
1445 * returns 0 at @phys
1446 * return value contains 0 (success) or error code
1448 static int ext4_ext_search_left(struct inode
*inode
,
1449 struct ext4_ext_path
*path
,
1450 ext4_lblk_t
*logical
, ext4_fsblk_t
*phys
)
1452 struct ext4_extent_idx
*ix
;
1453 struct ext4_extent
*ex
;
1456 if (unlikely(path
== NULL
)) {
1457 EXT4_ERROR_INODE(inode
, "path == NULL *logical %d!", *logical
);
1458 return -EFSCORRUPTED
;
1460 depth
= path
->p_depth
;
1463 if (depth
== 0 && path
->p_ext
== NULL
)
1466 /* usually extent in the path covers blocks smaller
1467 * then *logical, but it can be that extent is the
1468 * first one in the file */
1470 ex
= path
[depth
].p_ext
;
1471 ee_len
= ext4_ext_get_actual_len(ex
);
1472 if (*logical
< le32_to_cpu(ex
->ee_block
)) {
1473 if (unlikely(EXT_FIRST_EXTENT(path
[depth
].p_hdr
) != ex
)) {
1474 EXT4_ERROR_INODE(inode
,
1475 "EXT_FIRST_EXTENT != ex *logical %d ee_block %d!",
1476 *logical
, le32_to_cpu(ex
->ee_block
));
1477 return -EFSCORRUPTED
;
1479 while (--depth
>= 0) {
1480 ix
= path
[depth
].p_idx
;
1481 if (unlikely(ix
!= EXT_FIRST_INDEX(path
[depth
].p_hdr
))) {
1482 EXT4_ERROR_INODE(inode
,
1483 "ix (%d) != EXT_FIRST_INDEX (%d) (depth %d)!",
1484 ix
!= NULL
? le32_to_cpu(ix
->ei_block
) : 0,
1485 EXT_FIRST_INDEX(path
[depth
].p_hdr
) != NULL
?
1486 le32_to_cpu(EXT_FIRST_INDEX(path
[depth
].p_hdr
)->ei_block
) : 0,
1488 return -EFSCORRUPTED
;
1494 if (unlikely(*logical
< (le32_to_cpu(ex
->ee_block
) + ee_len
))) {
1495 EXT4_ERROR_INODE(inode
,
1496 "logical %d < ee_block %d + ee_len %d!",
1497 *logical
, le32_to_cpu(ex
->ee_block
), ee_len
);
1498 return -EFSCORRUPTED
;
1501 *logical
= le32_to_cpu(ex
->ee_block
) + ee_len
- 1;
1502 *phys
= ext4_ext_pblock(ex
) + ee_len
- 1;
1507 * search the closest allocated block to the right for *logical
1508 * and returns it at @logical + it's physical address at @phys
1509 * if *logical is the largest allocated block, the function
1510 * returns 0 at @phys
1511 * return value contains 0 (success) or error code
1513 static int ext4_ext_search_right(struct inode
*inode
,
1514 struct ext4_ext_path
*path
,
1515 ext4_lblk_t
*logical
, ext4_fsblk_t
*phys
,
1516 struct ext4_extent
**ret_ex
)
1518 struct buffer_head
*bh
= NULL
;
1519 struct ext4_extent_header
*eh
;
1520 struct ext4_extent_idx
*ix
;
1521 struct ext4_extent
*ex
;
1523 int depth
; /* Note, NOT eh_depth; depth from top of tree */
1526 if (unlikely(path
== NULL
)) {
1527 EXT4_ERROR_INODE(inode
, "path == NULL *logical %d!", *logical
);
1528 return -EFSCORRUPTED
;
1530 depth
= path
->p_depth
;
1533 if (depth
== 0 && path
->p_ext
== NULL
)
1536 /* usually extent in the path covers blocks smaller
1537 * then *logical, but it can be that extent is the
1538 * first one in the file */
1540 ex
= path
[depth
].p_ext
;
1541 ee_len
= ext4_ext_get_actual_len(ex
);
1542 if (*logical
< le32_to_cpu(ex
->ee_block
)) {
1543 if (unlikely(EXT_FIRST_EXTENT(path
[depth
].p_hdr
) != ex
)) {
1544 EXT4_ERROR_INODE(inode
,
1545 "first_extent(path[%d].p_hdr) != ex",
1547 return -EFSCORRUPTED
;
1549 while (--depth
>= 0) {
1550 ix
= path
[depth
].p_idx
;
1551 if (unlikely(ix
!= EXT_FIRST_INDEX(path
[depth
].p_hdr
))) {
1552 EXT4_ERROR_INODE(inode
,
1553 "ix != EXT_FIRST_INDEX *logical %d!",
1555 return -EFSCORRUPTED
;
1561 if (unlikely(*logical
< (le32_to_cpu(ex
->ee_block
) + ee_len
))) {
1562 EXT4_ERROR_INODE(inode
,
1563 "logical %d < ee_block %d + ee_len %d!",
1564 *logical
, le32_to_cpu(ex
->ee_block
), ee_len
);
1565 return -EFSCORRUPTED
;
1568 if (ex
!= EXT_LAST_EXTENT(path
[depth
].p_hdr
)) {
1569 /* next allocated block in this leaf */
1574 /* go up and search for index to the right */
1575 while (--depth
>= 0) {
1576 ix
= path
[depth
].p_idx
;
1577 if (ix
!= EXT_LAST_INDEX(path
[depth
].p_hdr
))
1581 /* we've gone up to the root and found no index to the right */
1585 /* we've found index to the right, let's
1586 * follow it and find the closest allocated
1587 * block to the right */
1589 block
= ext4_idx_pblock(ix
);
1590 while (++depth
< path
->p_depth
) {
1591 /* subtract from p_depth to get proper eh_depth */
1592 bh
= read_extent_tree_block(inode
, block
,
1593 path
->p_depth
- depth
, 0);
1596 eh
= ext_block_hdr(bh
);
1597 ix
= EXT_FIRST_INDEX(eh
);
1598 block
= ext4_idx_pblock(ix
);
1602 bh
= read_extent_tree_block(inode
, block
, path
->p_depth
- depth
, 0);
1605 eh
= ext_block_hdr(bh
);
1606 ex
= EXT_FIRST_EXTENT(eh
);
1608 *logical
= le32_to_cpu(ex
->ee_block
);
1609 *phys
= ext4_ext_pblock(ex
);
1617 * ext4_ext_next_allocated_block:
1618 * returns allocated block in subsequent extent or EXT_MAX_BLOCKS.
1619 * NOTE: it considers block number from index entry as
1620 * allocated block. Thus, index entries have to be consistent
1624 ext4_ext_next_allocated_block(struct ext4_ext_path
*path
)
1628 BUG_ON(path
== NULL
);
1629 depth
= path
->p_depth
;
1631 if (depth
== 0 && path
->p_ext
== NULL
)
1632 return EXT_MAX_BLOCKS
;
1634 while (depth
>= 0) {
1635 if (depth
== path
->p_depth
) {
1637 if (path
[depth
].p_ext
&&
1638 path
[depth
].p_ext
!=
1639 EXT_LAST_EXTENT(path
[depth
].p_hdr
))
1640 return le32_to_cpu(path
[depth
].p_ext
[1].ee_block
);
1643 if (path
[depth
].p_idx
!=
1644 EXT_LAST_INDEX(path
[depth
].p_hdr
))
1645 return le32_to_cpu(path
[depth
].p_idx
[1].ei_block
);
1650 return EXT_MAX_BLOCKS
;
1654 * ext4_ext_next_leaf_block:
1655 * returns first allocated block from next leaf or EXT_MAX_BLOCKS
1657 static ext4_lblk_t
ext4_ext_next_leaf_block(struct ext4_ext_path
*path
)
1661 BUG_ON(path
== NULL
);
1662 depth
= path
->p_depth
;
1664 /* zero-tree has no leaf blocks at all */
1666 return EXT_MAX_BLOCKS
;
1668 /* go to index block */
1671 while (depth
>= 0) {
1672 if (path
[depth
].p_idx
!=
1673 EXT_LAST_INDEX(path
[depth
].p_hdr
))
1674 return (ext4_lblk_t
)
1675 le32_to_cpu(path
[depth
].p_idx
[1].ei_block
);
1679 return EXT_MAX_BLOCKS
;
1683 * ext4_ext_correct_indexes:
1684 * if leaf gets modified and modified extent is first in the leaf,
1685 * then we have to correct all indexes above.
1686 * TODO: do we need to correct tree in all cases?
1688 static int ext4_ext_correct_indexes(handle_t
*handle
, struct inode
*inode
,
1689 struct ext4_ext_path
*path
)
1691 struct ext4_extent_header
*eh
;
1692 int depth
= ext_depth(inode
);
1693 struct ext4_extent
*ex
;
1697 eh
= path
[depth
].p_hdr
;
1698 ex
= path
[depth
].p_ext
;
1700 if (unlikely(ex
== NULL
|| eh
== NULL
)) {
1701 EXT4_ERROR_INODE(inode
,
1702 "ex %p == NULL or eh %p == NULL", ex
, eh
);
1703 return -EFSCORRUPTED
;
1707 /* there is no tree at all */
1711 if (ex
!= EXT_FIRST_EXTENT(eh
)) {
1712 /* we correct tree if first leaf got modified only */
1717 * TODO: we need correction if border is smaller than current one
1720 border
= path
[depth
].p_ext
->ee_block
;
1721 err
= ext4_ext_get_access(handle
, inode
, path
+ k
);
1724 path
[k
].p_idx
->ei_block
= border
;
1725 err
= ext4_ext_dirty(handle
, inode
, path
+ k
);
1730 /* change all left-side indexes */
1731 if (path
[k
+1].p_idx
!= EXT_FIRST_INDEX(path
[k
+1].p_hdr
))
1733 err
= ext4_ext_get_access(handle
, inode
, path
+ k
);
1736 path
[k
].p_idx
->ei_block
= border
;
1737 err
= ext4_ext_dirty(handle
, inode
, path
+ k
);
1746 ext4_can_extents_be_merged(struct inode
*inode
, struct ext4_extent
*ex1
,
1747 struct ext4_extent
*ex2
)
1749 unsigned short ext1_ee_len
, ext2_ee_len
;
1751 if (ext4_ext_is_unwritten(ex1
) != ext4_ext_is_unwritten(ex2
))
1754 ext1_ee_len
= ext4_ext_get_actual_len(ex1
);
1755 ext2_ee_len
= ext4_ext_get_actual_len(ex2
);
1757 if (le32_to_cpu(ex1
->ee_block
) + ext1_ee_len
!=
1758 le32_to_cpu(ex2
->ee_block
))
1762 * To allow future support for preallocated extents to be added
1763 * as an RO_COMPAT feature, refuse to merge to extents if
1764 * this can result in the top bit of ee_len being set.
1766 if (ext1_ee_len
+ ext2_ee_len
> EXT_INIT_MAX_LEN
)
1769 * The check for IO to unwritten extent is somewhat racy as we
1770 * increment i_unwritten / set EXT4_STATE_DIO_UNWRITTEN only after
1771 * dropping i_data_sem. But reserved blocks should save us in that
1774 if (ext4_ext_is_unwritten(ex1
) &&
1775 (ext4_test_inode_state(inode
, EXT4_STATE_DIO_UNWRITTEN
) ||
1776 atomic_read(&EXT4_I(inode
)->i_unwritten
) ||
1777 (ext1_ee_len
+ ext2_ee_len
> EXT_UNWRITTEN_MAX_LEN
)))
1779 #ifdef AGGRESSIVE_TEST
1780 if (ext1_ee_len
>= 4)
1784 if (ext4_ext_pblock(ex1
) + ext1_ee_len
== ext4_ext_pblock(ex2
))
1790 * This function tries to merge the "ex" extent to the next extent in the tree.
1791 * It always tries to merge towards right. If you want to merge towards
1792 * left, pass "ex - 1" as argument instead of "ex".
1793 * Returns 0 if the extents (ex and ex+1) were _not_ merged and returns
1794 * 1 if they got merged.
1796 static int ext4_ext_try_to_merge_right(struct inode
*inode
,
1797 struct ext4_ext_path
*path
,
1798 struct ext4_extent
*ex
)
1800 struct ext4_extent_header
*eh
;
1801 unsigned int depth
, len
;
1802 int merge_done
= 0, unwritten
;
1804 depth
= ext_depth(inode
);
1805 BUG_ON(path
[depth
].p_hdr
== NULL
);
1806 eh
= path
[depth
].p_hdr
;
1808 while (ex
< EXT_LAST_EXTENT(eh
)) {
1809 if (!ext4_can_extents_be_merged(inode
, ex
, ex
+ 1))
1811 /* merge with next extent! */
1812 unwritten
= ext4_ext_is_unwritten(ex
);
1813 ex
->ee_len
= cpu_to_le16(ext4_ext_get_actual_len(ex
)
1814 + ext4_ext_get_actual_len(ex
+ 1));
1816 ext4_ext_mark_unwritten(ex
);
1818 if (ex
+ 1 < EXT_LAST_EXTENT(eh
)) {
1819 len
= (EXT_LAST_EXTENT(eh
) - ex
- 1)
1820 * sizeof(struct ext4_extent
);
1821 memmove(ex
+ 1, ex
+ 2, len
);
1823 le16_add_cpu(&eh
->eh_entries
, -1);
1825 WARN_ON(eh
->eh_entries
== 0);
1826 if (!eh
->eh_entries
)
1827 EXT4_ERROR_INODE(inode
, "eh->eh_entries = 0!");
1834 * This function does a very simple check to see if we can collapse
1835 * an extent tree with a single extent tree leaf block into the inode.
1837 static void ext4_ext_try_to_merge_up(handle_t
*handle
,
1838 struct inode
*inode
,
1839 struct ext4_ext_path
*path
)
1842 unsigned max_root
= ext4_ext_space_root(inode
, 0);
1845 if ((path
[0].p_depth
!= 1) ||
1846 (le16_to_cpu(path
[0].p_hdr
->eh_entries
) != 1) ||
1847 (le16_to_cpu(path
[1].p_hdr
->eh_entries
) > max_root
))
1851 * We need to modify the block allocation bitmap and the block
1852 * group descriptor to release the extent tree block. If we
1853 * can't get the journal credits, give up.
1855 if (ext4_journal_extend(handle
, 2))
1859 * Copy the extent data up to the inode
1861 blk
= ext4_idx_pblock(path
[0].p_idx
);
1862 s
= le16_to_cpu(path
[1].p_hdr
->eh_entries
) *
1863 sizeof(struct ext4_extent_idx
);
1864 s
+= sizeof(struct ext4_extent_header
);
1866 path
[1].p_maxdepth
= path
[0].p_maxdepth
;
1867 memcpy(path
[0].p_hdr
, path
[1].p_hdr
, s
);
1868 path
[0].p_depth
= 0;
1869 path
[0].p_ext
= EXT_FIRST_EXTENT(path
[0].p_hdr
) +
1870 (path
[1].p_ext
- EXT_FIRST_EXTENT(path
[1].p_hdr
));
1871 path
[0].p_hdr
->eh_max
= cpu_to_le16(max_root
);
1873 brelse(path
[1].p_bh
);
1874 ext4_free_blocks(handle
, inode
, NULL
, blk
, 1,
1875 EXT4_FREE_BLOCKS_METADATA
| EXT4_FREE_BLOCKS_FORGET
);
1879 * This function tries to merge the @ex extent to neighbours in the tree.
1880 * return 1 if merge left else 0.
1882 static void ext4_ext_try_to_merge(handle_t
*handle
,
1883 struct inode
*inode
,
1884 struct ext4_ext_path
*path
,
1885 struct ext4_extent
*ex
) {
1886 struct ext4_extent_header
*eh
;
1890 depth
= ext_depth(inode
);
1891 BUG_ON(path
[depth
].p_hdr
== NULL
);
1892 eh
= path
[depth
].p_hdr
;
1894 if (ex
> EXT_FIRST_EXTENT(eh
))
1895 merge_done
= ext4_ext_try_to_merge_right(inode
, path
, ex
- 1);
1898 (void) ext4_ext_try_to_merge_right(inode
, path
, ex
);
1900 ext4_ext_try_to_merge_up(handle
, inode
, path
);
1904 * check if a portion of the "newext" extent overlaps with an
1907 * If there is an overlap discovered, it updates the length of the newext
1908 * such that there will be no overlap, and then returns 1.
1909 * If there is no overlap found, it returns 0.
1911 static unsigned int ext4_ext_check_overlap(struct ext4_sb_info
*sbi
,
1912 struct inode
*inode
,
1913 struct ext4_extent
*newext
,
1914 struct ext4_ext_path
*path
)
1917 unsigned int depth
, len1
;
1918 unsigned int ret
= 0;
1920 b1
= le32_to_cpu(newext
->ee_block
);
1921 len1
= ext4_ext_get_actual_len(newext
);
1922 depth
= ext_depth(inode
);
1923 if (!path
[depth
].p_ext
)
1925 b2
= EXT4_LBLK_CMASK(sbi
, le32_to_cpu(path
[depth
].p_ext
->ee_block
));
1928 * get the next allocated block if the extent in the path
1929 * is before the requested block(s)
1932 b2
= ext4_ext_next_allocated_block(path
);
1933 if (b2
== EXT_MAX_BLOCKS
)
1935 b2
= EXT4_LBLK_CMASK(sbi
, b2
);
1938 /* check for wrap through zero on extent logical start block*/
1939 if (b1
+ len1
< b1
) {
1940 len1
= EXT_MAX_BLOCKS
- b1
;
1941 newext
->ee_len
= cpu_to_le16(len1
);
1945 /* check for overlap */
1946 if (b1
+ len1
> b2
) {
1947 newext
->ee_len
= cpu_to_le16(b2
- b1
);
1955 * ext4_ext_insert_extent:
1956 * tries to merge requsted extent into the existing extent or
1957 * inserts requested extent as new one into the tree,
1958 * creating new leaf in the no-space case.
1960 int ext4_ext_insert_extent(handle_t
*handle
, struct inode
*inode
,
1961 struct ext4_ext_path
**ppath
,
1962 struct ext4_extent
*newext
, int gb_flags
)
1964 struct ext4_ext_path
*path
= *ppath
;
1965 struct ext4_extent_header
*eh
;
1966 struct ext4_extent
*ex
, *fex
;
1967 struct ext4_extent
*nearex
; /* nearest extent */
1968 struct ext4_ext_path
*npath
= NULL
;
1969 int depth
, len
, err
;
1971 int mb_flags
= 0, unwritten
;
1973 if (gb_flags
& EXT4_GET_BLOCKS_DELALLOC_RESERVE
)
1974 mb_flags
|= EXT4_MB_DELALLOC_RESERVED
;
1975 if (unlikely(ext4_ext_get_actual_len(newext
) == 0)) {
1976 EXT4_ERROR_INODE(inode
, "ext4_ext_get_actual_len(newext) == 0");
1977 return -EFSCORRUPTED
;
1979 depth
= ext_depth(inode
);
1980 ex
= path
[depth
].p_ext
;
1981 eh
= path
[depth
].p_hdr
;
1982 if (unlikely(path
[depth
].p_hdr
== NULL
)) {
1983 EXT4_ERROR_INODE(inode
, "path[%d].p_hdr == NULL", depth
);
1984 return -EFSCORRUPTED
;
1987 /* try to insert block into found extent and return */
1988 if (ex
&& !(gb_flags
& EXT4_GET_BLOCKS_PRE_IO
)) {
1991 * Try to see whether we should rather test the extent on
1992 * right from ex, or from the left of ex. This is because
1993 * ext4_find_extent() can return either extent on the
1994 * left, or on the right from the searched position. This
1995 * will make merging more effective.
1997 if (ex
< EXT_LAST_EXTENT(eh
) &&
1998 (le32_to_cpu(ex
->ee_block
) +
1999 ext4_ext_get_actual_len(ex
) <
2000 le32_to_cpu(newext
->ee_block
))) {
2003 } else if ((ex
> EXT_FIRST_EXTENT(eh
)) &&
2004 (le32_to_cpu(newext
->ee_block
) +
2005 ext4_ext_get_actual_len(newext
) <
2006 le32_to_cpu(ex
->ee_block
)))
2009 /* Try to append newex to the ex */
2010 if (ext4_can_extents_be_merged(inode
, ex
, newext
)) {
2011 ext_debug("append [%d]%d block to %u:[%d]%d"
2013 ext4_ext_is_unwritten(newext
),
2014 ext4_ext_get_actual_len(newext
),
2015 le32_to_cpu(ex
->ee_block
),
2016 ext4_ext_is_unwritten(ex
),
2017 ext4_ext_get_actual_len(ex
),
2018 ext4_ext_pblock(ex
));
2019 err
= ext4_ext_get_access(handle
, inode
,
2023 unwritten
= ext4_ext_is_unwritten(ex
);
2024 ex
->ee_len
= cpu_to_le16(ext4_ext_get_actual_len(ex
)
2025 + ext4_ext_get_actual_len(newext
));
2027 ext4_ext_mark_unwritten(ex
);
2028 eh
= path
[depth
].p_hdr
;
2034 /* Try to prepend newex to the ex */
2035 if (ext4_can_extents_be_merged(inode
, newext
, ex
)) {
2036 ext_debug("prepend %u[%d]%d block to %u:[%d]%d"
2038 le32_to_cpu(newext
->ee_block
),
2039 ext4_ext_is_unwritten(newext
),
2040 ext4_ext_get_actual_len(newext
),
2041 le32_to_cpu(ex
->ee_block
),
2042 ext4_ext_is_unwritten(ex
),
2043 ext4_ext_get_actual_len(ex
),
2044 ext4_ext_pblock(ex
));
2045 err
= ext4_ext_get_access(handle
, inode
,
2050 unwritten
= ext4_ext_is_unwritten(ex
);
2051 ex
->ee_block
= newext
->ee_block
;
2052 ext4_ext_store_pblock(ex
, ext4_ext_pblock(newext
));
2053 ex
->ee_len
= cpu_to_le16(ext4_ext_get_actual_len(ex
)
2054 + ext4_ext_get_actual_len(newext
));
2056 ext4_ext_mark_unwritten(ex
);
2057 eh
= path
[depth
].p_hdr
;
2063 depth
= ext_depth(inode
);
2064 eh
= path
[depth
].p_hdr
;
2065 if (le16_to_cpu(eh
->eh_entries
) < le16_to_cpu(eh
->eh_max
))
2068 /* probably next leaf has space for us? */
2069 fex
= EXT_LAST_EXTENT(eh
);
2070 next
= EXT_MAX_BLOCKS
;
2071 if (le32_to_cpu(newext
->ee_block
) > le32_to_cpu(fex
->ee_block
))
2072 next
= ext4_ext_next_leaf_block(path
);
2073 if (next
!= EXT_MAX_BLOCKS
) {
2074 ext_debug("next leaf block - %u\n", next
);
2075 BUG_ON(npath
!= NULL
);
2076 npath
= ext4_find_extent(inode
, next
, NULL
, 0);
2078 return PTR_ERR(npath
);
2079 BUG_ON(npath
->p_depth
!= path
->p_depth
);
2080 eh
= npath
[depth
].p_hdr
;
2081 if (le16_to_cpu(eh
->eh_entries
) < le16_to_cpu(eh
->eh_max
)) {
2082 ext_debug("next leaf isn't full(%d)\n",
2083 le16_to_cpu(eh
->eh_entries
));
2087 ext_debug("next leaf has no free space(%d,%d)\n",
2088 le16_to_cpu(eh
->eh_entries
), le16_to_cpu(eh
->eh_max
));
2092 * There is no free space in the found leaf.
2093 * We're gonna add a new leaf in the tree.
2095 if (gb_flags
& EXT4_GET_BLOCKS_METADATA_NOFAIL
)
2096 mb_flags
|= EXT4_MB_USE_RESERVED
;
2097 err
= ext4_ext_create_new_leaf(handle
, inode
, mb_flags
, gb_flags
,
2101 depth
= ext_depth(inode
);
2102 eh
= path
[depth
].p_hdr
;
2105 nearex
= path
[depth
].p_ext
;
2107 err
= ext4_ext_get_access(handle
, inode
, path
+ depth
);
2112 /* there is no extent in this leaf, create first one */
2113 ext_debug("first extent in the leaf: %u:%llu:[%d]%d\n",
2114 le32_to_cpu(newext
->ee_block
),
2115 ext4_ext_pblock(newext
),
2116 ext4_ext_is_unwritten(newext
),
2117 ext4_ext_get_actual_len(newext
));
2118 nearex
= EXT_FIRST_EXTENT(eh
);
2120 if (le32_to_cpu(newext
->ee_block
)
2121 > le32_to_cpu(nearex
->ee_block
)) {
2123 ext_debug("insert %u:%llu:[%d]%d before: "
2125 le32_to_cpu(newext
->ee_block
),
2126 ext4_ext_pblock(newext
),
2127 ext4_ext_is_unwritten(newext
),
2128 ext4_ext_get_actual_len(newext
),
2133 BUG_ON(newext
->ee_block
== nearex
->ee_block
);
2134 ext_debug("insert %u:%llu:[%d]%d after: "
2136 le32_to_cpu(newext
->ee_block
),
2137 ext4_ext_pblock(newext
),
2138 ext4_ext_is_unwritten(newext
),
2139 ext4_ext_get_actual_len(newext
),
2142 len
= EXT_LAST_EXTENT(eh
) - nearex
+ 1;
2144 ext_debug("insert %u:%llu:[%d]%d: "
2145 "move %d extents from 0x%p to 0x%p\n",
2146 le32_to_cpu(newext
->ee_block
),
2147 ext4_ext_pblock(newext
),
2148 ext4_ext_is_unwritten(newext
),
2149 ext4_ext_get_actual_len(newext
),
2150 len
, nearex
, nearex
+ 1);
2151 memmove(nearex
+ 1, nearex
,
2152 len
* sizeof(struct ext4_extent
));
2156 le16_add_cpu(&eh
->eh_entries
, 1);
2157 path
[depth
].p_ext
= nearex
;
2158 nearex
->ee_block
= newext
->ee_block
;
2159 ext4_ext_store_pblock(nearex
, ext4_ext_pblock(newext
));
2160 nearex
->ee_len
= newext
->ee_len
;
2163 /* try to merge extents */
2164 if (!(gb_flags
& EXT4_GET_BLOCKS_PRE_IO
))
2165 ext4_ext_try_to_merge(handle
, inode
, path
, nearex
);
2168 /* time to correct all indexes above */
2169 err
= ext4_ext_correct_indexes(handle
, inode
, path
);
2173 err
= ext4_ext_dirty(handle
, inode
, path
+ path
->p_depth
);
2176 ext4_ext_drop_refs(npath
);
2181 static int ext4_fill_fiemap_extents(struct inode
*inode
,
2182 ext4_lblk_t block
, ext4_lblk_t num
,
2183 struct fiemap_extent_info
*fieinfo
)
2185 struct ext4_ext_path
*path
= NULL
;
2186 struct ext4_extent
*ex
;
2187 struct extent_status es
;
2188 ext4_lblk_t next
, next_del
, start
= 0, end
= 0;
2189 ext4_lblk_t last
= block
+ num
;
2190 int exists
, depth
= 0, err
= 0;
2191 unsigned int flags
= 0;
2192 unsigned char blksize_bits
= inode
->i_sb
->s_blocksize_bits
;
2194 while (block
< last
&& block
!= EXT_MAX_BLOCKS
) {
2196 /* find extent for this block */
2197 down_read(&EXT4_I(inode
)->i_data_sem
);
2199 path
= ext4_find_extent(inode
, block
, &path
, 0);
2201 up_read(&EXT4_I(inode
)->i_data_sem
);
2202 err
= PTR_ERR(path
);
2207 depth
= ext_depth(inode
);
2208 if (unlikely(path
[depth
].p_hdr
== NULL
)) {
2209 up_read(&EXT4_I(inode
)->i_data_sem
);
2210 EXT4_ERROR_INODE(inode
, "path[%d].p_hdr == NULL", depth
);
2211 err
= -EFSCORRUPTED
;
2214 ex
= path
[depth
].p_ext
;
2215 next
= ext4_ext_next_allocated_block(path
);
2220 /* there is no extent yet, so try to allocate
2221 * all requested space */
2224 } else if (le32_to_cpu(ex
->ee_block
) > block
) {
2225 /* need to allocate space before found extent */
2227 end
= le32_to_cpu(ex
->ee_block
);
2228 if (block
+ num
< end
)
2230 } else if (block
>= le32_to_cpu(ex
->ee_block
)
2231 + ext4_ext_get_actual_len(ex
)) {
2232 /* need to allocate space after found extent */
2237 } else if (block
>= le32_to_cpu(ex
->ee_block
)) {
2239 * some part of requested space is covered
2243 end
= le32_to_cpu(ex
->ee_block
)
2244 + ext4_ext_get_actual_len(ex
);
2245 if (block
+ num
< end
)
2251 BUG_ON(end
<= start
);
2255 es
.es_len
= end
- start
;
2258 es
.es_lblk
= le32_to_cpu(ex
->ee_block
);
2259 es
.es_len
= ext4_ext_get_actual_len(ex
);
2260 es
.es_pblk
= ext4_ext_pblock(ex
);
2261 if (ext4_ext_is_unwritten(ex
))
2262 flags
|= FIEMAP_EXTENT_UNWRITTEN
;
2266 * Find delayed extent and update es accordingly. We call
2267 * it even in !exists case to find out whether es is the
2268 * last existing extent or not.
2270 next_del
= ext4_find_delayed_extent(inode
, &es
);
2271 if (!exists
&& next_del
) {
2273 flags
|= (FIEMAP_EXTENT_DELALLOC
|
2274 FIEMAP_EXTENT_UNKNOWN
);
2276 up_read(&EXT4_I(inode
)->i_data_sem
);
2278 if (unlikely(es
.es_len
== 0)) {
2279 EXT4_ERROR_INODE(inode
, "es.es_len == 0");
2280 err
= -EFSCORRUPTED
;
2285 * This is possible iff next == next_del == EXT_MAX_BLOCKS.
2286 * we need to check next == EXT_MAX_BLOCKS because it is
2287 * possible that an extent is with unwritten and delayed
2288 * status due to when an extent is delayed allocated and
2289 * is allocated by fallocate status tree will track both of
2292 * So we could return a unwritten and delayed extent, and
2293 * its block is equal to 'next'.
2295 if (next
== next_del
&& next
== EXT_MAX_BLOCKS
) {
2296 flags
|= FIEMAP_EXTENT_LAST
;
2297 if (unlikely(next_del
!= EXT_MAX_BLOCKS
||
2298 next
!= EXT_MAX_BLOCKS
)) {
2299 EXT4_ERROR_INODE(inode
,
2300 "next extent == %u, next "
2301 "delalloc extent = %u",
2303 err
= -EFSCORRUPTED
;
2309 err
= fiemap_fill_next_extent(fieinfo
,
2310 (__u64
)es
.es_lblk
<< blksize_bits
,
2311 (__u64
)es
.es_pblk
<< blksize_bits
,
2312 (__u64
)es
.es_len
<< blksize_bits
,
2322 block
= es
.es_lblk
+ es
.es_len
;
2325 ext4_ext_drop_refs(path
);
2331 * ext4_ext_determine_hole - determine hole around given block
2332 * @inode: inode we lookup in
2333 * @path: path in extent tree to @lblk
2334 * @lblk: pointer to logical block around which we want to determine hole
2336 * Determine hole length (and start if easily possible) around given logical
2337 * block. We don't try too hard to find the beginning of the hole but @path
2338 * actually points to extent before @lblk, we provide it.
2340 * The function returns the length of a hole starting at @lblk. We update @lblk
2341 * to the beginning of the hole if we managed to find it.
2343 static ext4_lblk_t
ext4_ext_determine_hole(struct inode
*inode
,
2344 struct ext4_ext_path
*path
,
2347 int depth
= ext_depth(inode
);
2348 struct ext4_extent
*ex
;
2351 ex
= path
[depth
].p_ext
;
2353 /* there is no extent yet, so gap is [0;-] */
2355 len
= EXT_MAX_BLOCKS
;
2356 } else if (*lblk
< le32_to_cpu(ex
->ee_block
)) {
2357 len
= le32_to_cpu(ex
->ee_block
) - *lblk
;
2358 } else if (*lblk
>= le32_to_cpu(ex
->ee_block
)
2359 + ext4_ext_get_actual_len(ex
)) {
2362 *lblk
= le32_to_cpu(ex
->ee_block
) + ext4_ext_get_actual_len(ex
);
2363 next
= ext4_ext_next_allocated_block(path
);
2364 BUG_ON(next
== *lblk
);
2373 * ext4_ext_put_gap_in_cache:
2374 * calculate boundaries of the gap that the requested block fits into
2375 * and cache this gap
2378 ext4_ext_put_gap_in_cache(struct inode
*inode
, ext4_lblk_t hole_start
,
2379 ext4_lblk_t hole_len
)
2381 struct extent_status es
;
2383 ext4_es_find_delayed_extent_range(inode
, hole_start
,
2384 hole_start
+ hole_len
- 1, &es
);
2386 /* There's delayed extent containing lblock? */
2387 if (es
.es_lblk
<= hole_start
)
2389 hole_len
= min(es
.es_lblk
- hole_start
, hole_len
);
2391 ext_debug(" -> %u:%u\n", hole_start
, hole_len
);
2392 ext4_es_insert_extent(inode
, hole_start
, hole_len
, ~0,
2393 EXTENT_STATUS_HOLE
);
2398 * removes index from the index block.
2400 static int ext4_ext_rm_idx(handle_t
*handle
, struct inode
*inode
,
2401 struct ext4_ext_path
*path
, int depth
)
2406 /* free index block */
2408 path
= path
+ depth
;
2409 leaf
= ext4_idx_pblock(path
->p_idx
);
2410 if (unlikely(path
->p_hdr
->eh_entries
== 0)) {
2411 EXT4_ERROR_INODE(inode
, "path->p_hdr->eh_entries == 0");
2412 return -EFSCORRUPTED
;
2414 err
= ext4_ext_get_access(handle
, inode
, path
);
2418 if (path
->p_idx
!= EXT_LAST_INDEX(path
->p_hdr
)) {
2419 int len
= EXT_LAST_INDEX(path
->p_hdr
) - path
->p_idx
;
2420 len
*= sizeof(struct ext4_extent_idx
);
2421 memmove(path
->p_idx
, path
->p_idx
+ 1, len
);
2424 le16_add_cpu(&path
->p_hdr
->eh_entries
, -1);
2425 err
= ext4_ext_dirty(handle
, inode
, path
);
2428 ext_debug("index is empty, remove it, free block %llu\n", leaf
);
2429 trace_ext4_ext_rm_idx(inode
, leaf
);
2431 ext4_free_blocks(handle
, inode
, NULL
, leaf
, 1,
2432 EXT4_FREE_BLOCKS_METADATA
| EXT4_FREE_BLOCKS_FORGET
);
2434 while (--depth
>= 0) {
2435 if (path
->p_idx
!= EXT_FIRST_INDEX(path
->p_hdr
))
2438 err
= ext4_ext_get_access(handle
, inode
, path
);
2441 path
->p_idx
->ei_block
= (path
+1)->p_idx
->ei_block
;
2442 err
= ext4_ext_dirty(handle
, inode
, path
);
2450 * ext4_ext_calc_credits_for_single_extent:
2451 * This routine returns max. credits that needed to insert an extent
2452 * to the extent tree.
2453 * When pass the actual path, the caller should calculate credits
2456 int ext4_ext_calc_credits_for_single_extent(struct inode
*inode
, int nrblocks
,
2457 struct ext4_ext_path
*path
)
2460 int depth
= ext_depth(inode
);
2463 /* probably there is space in leaf? */
2464 if (le16_to_cpu(path
[depth
].p_hdr
->eh_entries
)
2465 < le16_to_cpu(path
[depth
].p_hdr
->eh_max
)) {
2468 * There are some space in the leaf tree, no
2469 * need to account for leaf block credit
2471 * bitmaps and block group descriptor blocks
2472 * and other metadata blocks still need to be
2475 /* 1 bitmap, 1 block group descriptor */
2476 ret
= 2 + EXT4_META_TRANS_BLOCKS(inode
->i_sb
);
2481 return ext4_chunk_trans_blocks(inode
, nrblocks
);
2485 * How many index/leaf blocks need to change/allocate to add @extents extents?
2487 * If we add a single extent, then in the worse case, each tree level
2488 * index/leaf need to be changed in case of the tree split.
2490 * If more extents are inserted, they could cause the whole tree split more
2491 * than once, but this is really rare.
2493 int ext4_ext_index_trans_blocks(struct inode
*inode
, int extents
)
2498 /* If we are converting the inline data, only one is needed here. */
2499 if (ext4_has_inline_data(inode
))
2502 depth
= ext_depth(inode
);
2512 static inline int get_default_free_blocks_flags(struct inode
*inode
)
2514 if (S_ISDIR(inode
->i_mode
) || S_ISLNK(inode
->i_mode
) ||
2515 ext4_test_inode_flag(inode
, EXT4_INODE_EA_INODE
))
2516 return EXT4_FREE_BLOCKS_METADATA
| EXT4_FREE_BLOCKS_FORGET
;
2517 else if (ext4_should_journal_data(inode
))
2518 return EXT4_FREE_BLOCKS_FORGET
;
2522 static int ext4_remove_blocks(handle_t
*handle
, struct inode
*inode
,
2523 struct ext4_extent
*ex
,
2524 long long *partial_cluster
,
2525 ext4_lblk_t from
, ext4_lblk_t to
)
2527 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
2528 unsigned short ee_len
= ext4_ext_get_actual_len(ex
);
2530 int flags
= get_default_free_blocks_flags(inode
);
2533 * For bigalloc file systems, we never free a partial cluster
2534 * at the beginning of the extent. Instead, we make a note
2535 * that we tried freeing the cluster, and check to see if we
2536 * need to free it on a subsequent call to ext4_remove_blocks,
2537 * or at the end of ext4_ext_rm_leaf or ext4_ext_remove_space.
2539 flags
|= EXT4_FREE_BLOCKS_NOFREE_FIRST_CLUSTER
;
2541 trace_ext4_remove_blocks(inode
, ex
, from
, to
, *partial_cluster
);
2543 * If we have a partial cluster, and it's different from the
2544 * cluster of the last block, we need to explicitly free the
2545 * partial cluster here.
2547 pblk
= ext4_ext_pblock(ex
) + ee_len
- 1;
2548 if (*partial_cluster
> 0 &&
2549 *partial_cluster
!= (long long) EXT4_B2C(sbi
, pblk
)) {
2550 ext4_free_blocks(handle
, inode
, NULL
,
2551 EXT4_C2B(sbi
, *partial_cluster
),
2552 sbi
->s_cluster_ratio
, flags
);
2553 *partial_cluster
= 0;
2556 #ifdef EXTENTS_STATS
2558 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
2559 spin_lock(&sbi
->s_ext_stats_lock
);
2560 sbi
->s_ext_blocks
+= ee_len
;
2561 sbi
->s_ext_extents
++;
2562 if (ee_len
< sbi
->s_ext_min
)
2563 sbi
->s_ext_min
= ee_len
;
2564 if (ee_len
> sbi
->s_ext_max
)
2565 sbi
->s_ext_max
= ee_len
;
2566 if (ext_depth(inode
) > sbi
->s_depth_max
)
2567 sbi
->s_depth_max
= ext_depth(inode
);
2568 spin_unlock(&sbi
->s_ext_stats_lock
);
2571 if (from
>= le32_to_cpu(ex
->ee_block
)
2572 && to
== le32_to_cpu(ex
->ee_block
) + ee_len
- 1) {
2575 long long first_cluster
;
2577 num
= le32_to_cpu(ex
->ee_block
) + ee_len
- from
;
2578 pblk
= ext4_ext_pblock(ex
) + ee_len
- num
;
2580 * Usually we want to free partial cluster at the end of the
2581 * extent, except for the situation when the cluster is still
2582 * used by any other extent (partial_cluster is negative).
2584 if (*partial_cluster
< 0 &&
2585 *partial_cluster
== -(long long) EXT4_B2C(sbi
, pblk
+num
-1))
2586 flags
|= EXT4_FREE_BLOCKS_NOFREE_LAST_CLUSTER
;
2588 ext_debug("free last %u blocks starting %llu partial %lld\n",
2589 num
, pblk
, *partial_cluster
);
2590 ext4_free_blocks(handle
, inode
, NULL
, pblk
, num
, flags
);
2592 * If the block range to be freed didn't start at the
2593 * beginning of a cluster, and we removed the entire
2594 * extent and the cluster is not used by any other extent,
2595 * save the partial cluster here, since we might need to
2596 * delete if we determine that the truncate or punch hole
2597 * operation has removed all of the blocks in the cluster.
2598 * If that cluster is used by another extent, preserve its
2599 * negative value so it isn't freed later on.
2601 * If the whole extent wasn't freed, we've reached the
2602 * start of the truncated/punched region and have finished
2603 * removing blocks. If there's a partial cluster here it's
2604 * shared with the remainder of the extent and is no longer
2605 * a candidate for removal.
2607 if (EXT4_PBLK_COFF(sbi
, pblk
) && ee_len
== num
) {
2608 first_cluster
= (long long) EXT4_B2C(sbi
, pblk
);
2609 if (first_cluster
!= -*partial_cluster
)
2610 *partial_cluster
= first_cluster
;
2612 *partial_cluster
= 0;
2615 ext4_error(sbi
->s_sb
, "strange request: removal(2) "
2617 from
, to
, le32_to_cpu(ex
->ee_block
), ee_len
);
2623 * ext4_ext_rm_leaf() Removes the extents associated with the
2624 * blocks appearing between "start" and "end". Both "start"
2625 * and "end" must appear in the same extent or EIO is returned.
2627 * @handle: The journal handle
2628 * @inode: The files inode
2629 * @path: The path to the leaf
2630 * @partial_cluster: The cluster which we'll have to free if all extents
2631 * has been released from it. However, if this value is
2632 * negative, it's a cluster just to the right of the
2633 * punched region and it must not be freed.
2634 * @start: The first block to remove
2635 * @end: The last block to remove
2638 ext4_ext_rm_leaf(handle_t
*handle
, struct inode
*inode
,
2639 struct ext4_ext_path
*path
,
2640 long long *partial_cluster
,
2641 ext4_lblk_t start
, ext4_lblk_t end
)
2643 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
2644 int err
= 0, correct_index
= 0;
2645 int depth
= ext_depth(inode
), credits
;
2646 struct ext4_extent_header
*eh
;
2649 ext4_lblk_t ex_ee_block
;
2650 unsigned short ex_ee_len
;
2651 unsigned unwritten
= 0;
2652 struct ext4_extent
*ex
;
2655 /* the header must be checked already in ext4_ext_remove_space() */
2656 ext_debug("truncate since %u in leaf to %u\n", start
, end
);
2657 if (!path
[depth
].p_hdr
)
2658 path
[depth
].p_hdr
= ext_block_hdr(path
[depth
].p_bh
);
2659 eh
= path
[depth
].p_hdr
;
2660 if (unlikely(path
[depth
].p_hdr
== NULL
)) {
2661 EXT4_ERROR_INODE(inode
, "path[%d].p_hdr == NULL", depth
);
2662 return -EFSCORRUPTED
;
2664 /* find where to start removing */
2665 ex
= path
[depth
].p_ext
;
2667 ex
= EXT_LAST_EXTENT(eh
);
2669 ex_ee_block
= le32_to_cpu(ex
->ee_block
);
2670 ex_ee_len
= ext4_ext_get_actual_len(ex
);
2672 trace_ext4_ext_rm_leaf(inode
, start
, ex
, *partial_cluster
);
2674 while (ex
>= EXT_FIRST_EXTENT(eh
) &&
2675 ex_ee_block
+ ex_ee_len
> start
) {
2677 if (ext4_ext_is_unwritten(ex
))
2682 ext_debug("remove ext %u:[%d]%d\n", ex_ee_block
,
2683 unwritten
, ex_ee_len
);
2684 path
[depth
].p_ext
= ex
;
2686 a
= ex_ee_block
> start
? ex_ee_block
: start
;
2687 b
= ex_ee_block
+ex_ee_len
- 1 < end
?
2688 ex_ee_block
+ex_ee_len
- 1 : end
;
2690 ext_debug(" border %u:%u\n", a
, b
);
2692 /* If this extent is beyond the end of the hole, skip it */
2693 if (end
< ex_ee_block
) {
2695 * We're going to skip this extent and move to another,
2696 * so note that its first cluster is in use to avoid
2697 * freeing it when removing blocks. Eventually, the
2698 * right edge of the truncated/punched region will
2699 * be just to the left.
2701 if (sbi
->s_cluster_ratio
> 1) {
2702 pblk
= ext4_ext_pblock(ex
);
2704 -(long long) EXT4_B2C(sbi
, pblk
);
2707 ex_ee_block
= le32_to_cpu(ex
->ee_block
);
2708 ex_ee_len
= ext4_ext_get_actual_len(ex
);
2710 } else if (b
!= ex_ee_block
+ ex_ee_len
- 1) {
2711 EXT4_ERROR_INODE(inode
,
2712 "can not handle truncate %u:%u "
2714 start
, end
, ex_ee_block
,
2715 ex_ee_block
+ ex_ee_len
- 1);
2716 err
= -EFSCORRUPTED
;
2718 } else if (a
!= ex_ee_block
) {
2719 /* remove tail of the extent */
2720 num
= a
- ex_ee_block
;
2722 /* remove whole extent: excellent! */
2726 * 3 for leaf, sb, and inode plus 2 (bmap and group
2727 * descriptor) for each block group; assume two block
2728 * groups plus ex_ee_len/blocks_per_block_group for
2731 credits
= 7 + 2*(ex_ee_len
/EXT4_BLOCKS_PER_GROUP(inode
->i_sb
));
2732 if (ex
== EXT_FIRST_EXTENT(eh
)) {
2734 credits
+= (ext_depth(inode
)) + 1;
2736 credits
+= EXT4_MAXQUOTAS_TRANS_BLOCKS(inode
->i_sb
);
2738 err
= ext4_ext_truncate_extend_restart(handle
, inode
, credits
);
2742 err
= ext4_ext_get_access(handle
, inode
, path
+ depth
);
2746 err
= ext4_remove_blocks(handle
, inode
, ex
, partial_cluster
,
2752 /* this extent is removed; mark slot entirely unused */
2753 ext4_ext_store_pblock(ex
, 0);
2755 ex
->ee_len
= cpu_to_le16(num
);
2757 * Do not mark unwritten if all the blocks in the
2758 * extent have been removed.
2760 if (unwritten
&& num
)
2761 ext4_ext_mark_unwritten(ex
);
2763 * If the extent was completely released,
2764 * we need to remove it from the leaf
2767 if (end
!= EXT_MAX_BLOCKS
- 1) {
2769 * For hole punching, we need to scoot all the
2770 * extents up when an extent is removed so that
2771 * we dont have blank extents in the middle
2773 memmove(ex
, ex
+1, (EXT_LAST_EXTENT(eh
) - ex
) *
2774 sizeof(struct ext4_extent
));
2776 /* Now get rid of the one at the end */
2777 memset(EXT_LAST_EXTENT(eh
), 0,
2778 sizeof(struct ext4_extent
));
2780 le16_add_cpu(&eh
->eh_entries
, -1);
2783 err
= ext4_ext_dirty(handle
, inode
, path
+ depth
);
2787 ext_debug("new extent: %u:%u:%llu\n", ex_ee_block
, num
,
2788 ext4_ext_pblock(ex
));
2790 ex_ee_block
= le32_to_cpu(ex
->ee_block
);
2791 ex_ee_len
= ext4_ext_get_actual_len(ex
);
2794 if (correct_index
&& eh
->eh_entries
)
2795 err
= ext4_ext_correct_indexes(handle
, inode
, path
);
2798 * If there's a partial cluster and at least one extent remains in
2799 * the leaf, free the partial cluster if it isn't shared with the
2800 * current extent. If it is shared with the current extent
2801 * we zero partial_cluster because we've reached the start of the
2802 * truncated/punched region and we're done removing blocks.
2804 if (*partial_cluster
> 0 && ex
>= EXT_FIRST_EXTENT(eh
)) {
2805 pblk
= ext4_ext_pblock(ex
) + ex_ee_len
- 1;
2806 if (*partial_cluster
!= (long long) EXT4_B2C(sbi
, pblk
)) {
2807 ext4_free_blocks(handle
, inode
, NULL
,
2808 EXT4_C2B(sbi
, *partial_cluster
),
2809 sbi
->s_cluster_ratio
,
2810 get_default_free_blocks_flags(inode
));
2812 *partial_cluster
= 0;
2815 /* if this leaf is free, then we should
2816 * remove it from index block above */
2817 if (err
== 0 && eh
->eh_entries
== 0 && path
[depth
].p_bh
!= NULL
)
2818 err
= ext4_ext_rm_idx(handle
, inode
, path
, depth
);
2825 * ext4_ext_more_to_rm:
2826 * returns 1 if current index has to be freed (even partial)
2829 ext4_ext_more_to_rm(struct ext4_ext_path
*path
)
2831 BUG_ON(path
->p_idx
== NULL
);
2833 if (path
->p_idx
< EXT_FIRST_INDEX(path
->p_hdr
))
2837 * if truncate on deeper level happened, it wasn't partial,
2838 * so we have to consider current index for truncation
2840 if (le16_to_cpu(path
->p_hdr
->eh_entries
) == path
->p_block
)
2845 int ext4_ext_remove_space(struct inode
*inode
, ext4_lblk_t start
,
2848 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
2849 int depth
= ext_depth(inode
);
2850 struct ext4_ext_path
*path
= NULL
;
2851 long long partial_cluster
= 0;
2855 ext_debug("truncate since %u to %u\n", start
, end
);
2857 /* probably first extent we're gonna free will be last in block */
2858 handle
= ext4_journal_start(inode
, EXT4_HT_TRUNCATE
, depth
+ 1);
2860 return PTR_ERR(handle
);
2863 trace_ext4_ext_remove_space(inode
, start
, end
, depth
);
2866 * Check if we are removing extents inside the extent tree. If that
2867 * is the case, we are going to punch a hole inside the extent tree
2868 * so we have to check whether we need to split the extent covering
2869 * the last block to remove so we can easily remove the part of it
2870 * in ext4_ext_rm_leaf().
2872 if (end
< EXT_MAX_BLOCKS
- 1) {
2873 struct ext4_extent
*ex
;
2874 ext4_lblk_t ee_block
, ex_end
, lblk
;
2877 /* find extent for or closest extent to this block */
2878 path
= ext4_find_extent(inode
, end
, NULL
, EXT4_EX_NOCACHE
);
2880 ext4_journal_stop(handle
);
2881 return PTR_ERR(path
);
2883 depth
= ext_depth(inode
);
2884 /* Leaf not may not exist only if inode has no blocks at all */
2885 ex
= path
[depth
].p_ext
;
2888 EXT4_ERROR_INODE(inode
,
2889 "path[%d].p_hdr == NULL",
2891 err
= -EFSCORRUPTED
;
2896 ee_block
= le32_to_cpu(ex
->ee_block
);
2897 ex_end
= ee_block
+ ext4_ext_get_actual_len(ex
) - 1;
2900 * See if the last block is inside the extent, if so split
2901 * the extent at 'end' block so we can easily remove the
2902 * tail of the first part of the split extent in
2903 * ext4_ext_rm_leaf().
2905 if (end
>= ee_block
&& end
< ex_end
) {
2908 * If we're going to split the extent, note that
2909 * the cluster containing the block after 'end' is
2910 * in use to avoid freeing it when removing blocks.
2912 if (sbi
->s_cluster_ratio
> 1) {
2913 pblk
= ext4_ext_pblock(ex
) + end
- ee_block
+ 2;
2915 -(long long) EXT4_B2C(sbi
, pblk
);
2919 * Split the extent in two so that 'end' is the last
2920 * block in the first new extent. Also we should not
2921 * fail removing space due to ENOSPC so try to use
2922 * reserved block if that happens.
2924 err
= ext4_force_split_extent_at(handle
, inode
, &path
,
2929 } else if (sbi
->s_cluster_ratio
> 1 && end
>= ex_end
) {
2931 * If there's an extent to the right its first cluster
2932 * contains the immediate right boundary of the
2933 * truncated/punched region. Set partial_cluster to
2934 * its negative value so it won't be freed if shared
2935 * with the current extent. The end < ee_block case
2936 * is handled in ext4_ext_rm_leaf().
2939 err
= ext4_ext_search_right(inode
, path
, &lblk
, &pblk
,
2945 -(long long) EXT4_B2C(sbi
, pblk
);
2949 * We start scanning from right side, freeing all the blocks
2950 * after i_size and walking into the tree depth-wise.
2952 depth
= ext_depth(inode
);
2957 le16_to_cpu(path
[k
].p_hdr
->eh_entries
)+1;
2959 path
= kzalloc(sizeof(struct ext4_ext_path
) * (depth
+ 1),
2962 ext4_journal_stop(handle
);
2965 path
[0].p_maxdepth
= path
[0].p_depth
= depth
;
2966 path
[0].p_hdr
= ext_inode_hdr(inode
);
2969 if (ext4_ext_check(inode
, path
[0].p_hdr
, depth
, 0)) {
2970 err
= -EFSCORRUPTED
;
2976 while (i
>= 0 && err
== 0) {
2978 /* this is leaf block */
2979 err
= ext4_ext_rm_leaf(handle
, inode
, path
,
2980 &partial_cluster
, start
,
2982 /* root level has p_bh == NULL, brelse() eats this */
2983 brelse(path
[i
].p_bh
);
2984 path
[i
].p_bh
= NULL
;
2989 /* this is index block */
2990 if (!path
[i
].p_hdr
) {
2991 ext_debug("initialize header\n");
2992 path
[i
].p_hdr
= ext_block_hdr(path
[i
].p_bh
);
2995 if (!path
[i
].p_idx
) {
2996 /* this level hasn't been touched yet */
2997 path
[i
].p_idx
= EXT_LAST_INDEX(path
[i
].p_hdr
);
2998 path
[i
].p_block
= le16_to_cpu(path
[i
].p_hdr
->eh_entries
)+1;
2999 ext_debug("init index ptr: hdr 0x%p, num %d\n",
3001 le16_to_cpu(path
[i
].p_hdr
->eh_entries
));
3003 /* we were already here, see at next index */
3007 ext_debug("level %d - index, first 0x%p, cur 0x%p\n",
3008 i
, EXT_FIRST_INDEX(path
[i
].p_hdr
),
3010 if (ext4_ext_more_to_rm(path
+ i
)) {
3011 struct buffer_head
*bh
;
3012 /* go to the next level */
3013 ext_debug("move to level %d (block %llu)\n",
3014 i
+ 1, ext4_idx_pblock(path
[i
].p_idx
));
3015 memset(path
+ i
+ 1, 0, sizeof(*path
));
3016 bh
= read_extent_tree_block(inode
,
3017 ext4_idx_pblock(path
[i
].p_idx
), depth
- i
- 1,
3020 /* should we reset i_size? */
3024 /* Yield here to deal with large extent trees.
3025 * Should be a no-op if we did IO above. */
3027 if (WARN_ON(i
+ 1 > depth
)) {
3028 err
= -EFSCORRUPTED
;
3031 path
[i
+ 1].p_bh
= bh
;
3033 /* save actual number of indexes since this
3034 * number is changed at the next iteration */
3035 path
[i
].p_block
= le16_to_cpu(path
[i
].p_hdr
->eh_entries
);
3038 /* we finished processing this index, go up */
3039 if (path
[i
].p_hdr
->eh_entries
== 0 && i
> 0) {
3040 /* index is empty, remove it;
3041 * handle must be already prepared by the
3042 * truncatei_leaf() */
3043 err
= ext4_ext_rm_idx(handle
, inode
, path
, i
);
3045 /* root level has p_bh == NULL, brelse() eats this */
3046 brelse(path
[i
].p_bh
);
3047 path
[i
].p_bh
= NULL
;
3049 ext_debug("return to level %d\n", i
);
3053 trace_ext4_ext_remove_space_done(inode
, start
, end
, depth
,
3054 partial_cluster
, path
->p_hdr
->eh_entries
);
3057 * If we still have something in the partial cluster and we have removed
3058 * even the first extent, then we should free the blocks in the partial
3059 * cluster as well. (This code will only run when there are no leaves
3060 * to the immediate left of the truncated/punched region.)
3062 if (partial_cluster
> 0 && err
== 0) {
3063 /* don't zero partial_cluster since it's not used afterwards */
3064 ext4_free_blocks(handle
, inode
, NULL
,
3065 EXT4_C2B(sbi
, partial_cluster
),
3066 sbi
->s_cluster_ratio
,
3067 get_default_free_blocks_flags(inode
));
3070 /* TODO: flexible tree reduction should be here */
3071 if (path
->p_hdr
->eh_entries
== 0) {
3073 * truncate to zero freed all the tree,
3074 * so we need to correct eh_depth
3076 err
= ext4_ext_get_access(handle
, inode
, path
);
3078 ext_inode_hdr(inode
)->eh_depth
= 0;
3079 ext_inode_hdr(inode
)->eh_max
=
3080 cpu_to_le16(ext4_ext_space_root(inode
, 0));
3081 err
= ext4_ext_dirty(handle
, inode
, path
);
3085 ext4_ext_drop_refs(path
);
3090 ext4_journal_stop(handle
);
3096 * called at mount time
3098 void ext4_ext_init(struct super_block
*sb
)
3101 * possible initialization would be here
3104 if (ext4_has_feature_extents(sb
)) {
3105 #if defined(AGGRESSIVE_TEST) || defined(CHECK_BINSEARCH) || defined(EXTENTS_STATS)
3106 printk(KERN_INFO
"EXT4-fs: file extents enabled"
3107 #ifdef AGGRESSIVE_TEST
3108 ", aggressive tests"
3110 #ifdef CHECK_BINSEARCH
3113 #ifdef EXTENTS_STATS
3118 #ifdef EXTENTS_STATS
3119 spin_lock_init(&EXT4_SB(sb
)->s_ext_stats_lock
);
3120 EXT4_SB(sb
)->s_ext_min
= 1 << 30;
3121 EXT4_SB(sb
)->s_ext_max
= 0;
3127 * called at umount time
3129 void ext4_ext_release(struct super_block
*sb
)
3131 if (!ext4_has_feature_extents(sb
))
3134 #ifdef EXTENTS_STATS
3135 if (EXT4_SB(sb
)->s_ext_blocks
&& EXT4_SB(sb
)->s_ext_extents
) {
3136 struct ext4_sb_info
*sbi
= EXT4_SB(sb
);
3137 printk(KERN_ERR
"EXT4-fs: %lu blocks in %lu extents (%lu ave)\n",
3138 sbi
->s_ext_blocks
, sbi
->s_ext_extents
,
3139 sbi
->s_ext_blocks
/ sbi
->s_ext_extents
);
3140 printk(KERN_ERR
"EXT4-fs: extents: %lu min, %lu max, max depth %lu\n",
3141 sbi
->s_ext_min
, sbi
->s_ext_max
, sbi
->s_depth_max
);
3146 static int ext4_zeroout_es(struct inode
*inode
, struct ext4_extent
*ex
)
3148 ext4_lblk_t ee_block
;
3149 ext4_fsblk_t ee_pblock
;
3150 unsigned int ee_len
;
3152 ee_block
= le32_to_cpu(ex
->ee_block
);
3153 ee_len
= ext4_ext_get_actual_len(ex
);
3154 ee_pblock
= ext4_ext_pblock(ex
);
3159 return ext4_es_insert_extent(inode
, ee_block
, ee_len
, ee_pblock
,
3160 EXTENT_STATUS_WRITTEN
);
3163 /* FIXME!! we need to try to merge to left or right after zero-out */
3164 static int ext4_ext_zeroout(struct inode
*inode
, struct ext4_extent
*ex
)
3166 ext4_fsblk_t ee_pblock
;
3167 unsigned int ee_len
;
3169 ee_len
= ext4_ext_get_actual_len(ex
);
3170 ee_pblock
= ext4_ext_pblock(ex
);
3171 return ext4_issue_zeroout(inode
, le32_to_cpu(ex
->ee_block
), ee_pblock
,
3176 * ext4_split_extent_at() splits an extent at given block.
3178 * @handle: the journal handle
3179 * @inode: the file inode
3180 * @path: the path to the extent
3181 * @split: the logical block where the extent is splitted.
3182 * @split_flags: indicates if the extent could be zeroout if split fails, and
3183 * the states(init or unwritten) of new extents.
3184 * @flags: flags used to insert new extent to extent tree.
3187 * Splits extent [a, b] into two extents [a, @split) and [@split, b], states
3188 * of which are deterimined by split_flag.
3190 * There are two cases:
3191 * a> the extent are splitted into two extent.
3192 * b> split is not needed, and just mark the extent.
3194 * return 0 on success.
3196 static int ext4_split_extent_at(handle_t
*handle
,
3197 struct inode
*inode
,
3198 struct ext4_ext_path
**ppath
,
3203 struct ext4_ext_path
*path
= *ppath
;
3204 ext4_fsblk_t newblock
;
3205 ext4_lblk_t ee_block
;
3206 struct ext4_extent
*ex
, newex
, orig_ex
, zero_ex
;
3207 struct ext4_extent
*ex2
= NULL
;
3208 unsigned int ee_len
, depth
;
3211 BUG_ON((split_flag
& (EXT4_EXT_DATA_VALID1
| EXT4_EXT_DATA_VALID2
)) ==
3212 (EXT4_EXT_DATA_VALID1
| EXT4_EXT_DATA_VALID2
));
3214 ext_debug("ext4_split_extents_at: inode %lu, logical"
3215 "block %llu\n", inode
->i_ino
, (unsigned long long)split
);
3217 ext4_ext_show_leaf(inode
, path
);
3219 depth
= ext_depth(inode
);
3220 ex
= path
[depth
].p_ext
;
3221 ee_block
= le32_to_cpu(ex
->ee_block
);
3222 ee_len
= ext4_ext_get_actual_len(ex
);
3223 newblock
= split
- ee_block
+ ext4_ext_pblock(ex
);
3225 BUG_ON(split
< ee_block
|| split
>= (ee_block
+ ee_len
));
3226 BUG_ON(!ext4_ext_is_unwritten(ex
) &&
3227 split_flag
& (EXT4_EXT_MAY_ZEROOUT
|
3228 EXT4_EXT_MARK_UNWRIT1
|
3229 EXT4_EXT_MARK_UNWRIT2
));
3231 err
= ext4_ext_get_access(handle
, inode
, path
+ depth
);
3235 if (split
== ee_block
) {
3237 * case b: block @split is the block that the extent begins with
3238 * then we just change the state of the extent, and splitting
3241 if (split_flag
& EXT4_EXT_MARK_UNWRIT2
)
3242 ext4_ext_mark_unwritten(ex
);
3244 ext4_ext_mark_initialized(ex
);
3246 if (!(flags
& EXT4_GET_BLOCKS_PRE_IO
))
3247 ext4_ext_try_to_merge(handle
, inode
, path
, ex
);
3249 err
= ext4_ext_dirty(handle
, inode
, path
+ path
->p_depth
);
3254 memcpy(&orig_ex
, ex
, sizeof(orig_ex
));
3255 ex
->ee_len
= cpu_to_le16(split
- ee_block
);
3256 if (split_flag
& EXT4_EXT_MARK_UNWRIT1
)
3257 ext4_ext_mark_unwritten(ex
);
3260 * path may lead to new leaf, not to original leaf any more
3261 * after ext4_ext_insert_extent() returns,
3263 err
= ext4_ext_dirty(handle
, inode
, path
+ depth
);
3265 goto fix_extent_len
;
3268 ex2
->ee_block
= cpu_to_le32(split
);
3269 ex2
->ee_len
= cpu_to_le16(ee_len
- (split
- ee_block
));
3270 ext4_ext_store_pblock(ex2
, newblock
);
3271 if (split_flag
& EXT4_EXT_MARK_UNWRIT2
)
3272 ext4_ext_mark_unwritten(ex2
);
3274 err
= ext4_ext_insert_extent(handle
, inode
, ppath
, &newex
, flags
);
3275 if (err
== -ENOSPC
&& (EXT4_EXT_MAY_ZEROOUT
& split_flag
)) {
3276 if (split_flag
& (EXT4_EXT_DATA_VALID1
|EXT4_EXT_DATA_VALID2
)) {
3277 if (split_flag
& EXT4_EXT_DATA_VALID1
) {
3278 err
= ext4_ext_zeroout(inode
, ex2
);
3279 zero_ex
.ee_block
= ex2
->ee_block
;
3280 zero_ex
.ee_len
= cpu_to_le16(
3281 ext4_ext_get_actual_len(ex2
));
3282 ext4_ext_store_pblock(&zero_ex
,
3283 ext4_ext_pblock(ex2
));
3285 err
= ext4_ext_zeroout(inode
, ex
);
3286 zero_ex
.ee_block
= ex
->ee_block
;
3287 zero_ex
.ee_len
= cpu_to_le16(
3288 ext4_ext_get_actual_len(ex
));
3289 ext4_ext_store_pblock(&zero_ex
,
3290 ext4_ext_pblock(ex
));
3293 err
= ext4_ext_zeroout(inode
, &orig_ex
);
3294 zero_ex
.ee_block
= orig_ex
.ee_block
;
3295 zero_ex
.ee_len
= cpu_to_le16(
3296 ext4_ext_get_actual_len(&orig_ex
));
3297 ext4_ext_store_pblock(&zero_ex
,
3298 ext4_ext_pblock(&orig_ex
));
3302 goto fix_extent_len
;
3303 /* update the extent length and mark as initialized */
3304 ex
->ee_len
= cpu_to_le16(ee_len
);
3305 ext4_ext_try_to_merge(handle
, inode
, path
, ex
);
3306 err
= ext4_ext_dirty(handle
, inode
, path
+ path
->p_depth
);
3308 goto fix_extent_len
;
3310 /* update extent status tree */
3311 err
= ext4_zeroout_es(inode
, &zero_ex
);
3315 goto fix_extent_len
;
3318 ext4_ext_show_leaf(inode
, path
);
3322 ex
->ee_len
= orig_ex
.ee_len
;
3323 ext4_ext_dirty(handle
, inode
, path
+ path
->p_depth
);
3328 * ext4_split_extents() splits an extent and mark extent which is covered
3329 * by @map as split_flags indicates
3331 * It may result in splitting the extent into multiple extents (up to three)
3332 * There are three possibilities:
3333 * a> There is no split required
3334 * b> Splits in two extents: Split is happening at either end of the extent
3335 * c> Splits in three extents: Somone is splitting in middle of the extent
3338 static int ext4_split_extent(handle_t
*handle
,
3339 struct inode
*inode
,
3340 struct ext4_ext_path
**ppath
,
3341 struct ext4_map_blocks
*map
,
3345 struct ext4_ext_path
*path
= *ppath
;
3346 ext4_lblk_t ee_block
;
3347 struct ext4_extent
*ex
;
3348 unsigned int ee_len
, depth
;
3351 int split_flag1
, flags1
;
3352 int allocated
= map
->m_len
;
3354 depth
= ext_depth(inode
);
3355 ex
= path
[depth
].p_ext
;
3356 ee_block
= le32_to_cpu(ex
->ee_block
);
3357 ee_len
= ext4_ext_get_actual_len(ex
);
3358 unwritten
= ext4_ext_is_unwritten(ex
);
3360 if (map
->m_lblk
+ map
->m_len
< ee_block
+ ee_len
) {
3361 split_flag1
= split_flag
& EXT4_EXT_MAY_ZEROOUT
;
3362 flags1
= flags
| EXT4_GET_BLOCKS_PRE_IO
;
3364 split_flag1
|= EXT4_EXT_MARK_UNWRIT1
|
3365 EXT4_EXT_MARK_UNWRIT2
;
3366 if (split_flag
& EXT4_EXT_DATA_VALID2
)
3367 split_flag1
|= EXT4_EXT_DATA_VALID1
;
3368 err
= ext4_split_extent_at(handle
, inode
, ppath
,
3369 map
->m_lblk
+ map
->m_len
, split_flag1
, flags1
);
3373 allocated
= ee_len
- (map
->m_lblk
- ee_block
);
3376 * Update path is required because previous ext4_split_extent_at() may
3377 * result in split of original leaf or extent zeroout.
3379 path
= ext4_find_extent(inode
, map
->m_lblk
, ppath
, 0);
3381 return PTR_ERR(path
);
3382 depth
= ext_depth(inode
);
3383 ex
= path
[depth
].p_ext
;
3385 EXT4_ERROR_INODE(inode
, "unexpected hole at %lu",
3386 (unsigned long) map
->m_lblk
);
3387 return -EFSCORRUPTED
;
3389 unwritten
= ext4_ext_is_unwritten(ex
);
3392 if (map
->m_lblk
>= ee_block
) {
3393 split_flag1
= split_flag
& EXT4_EXT_DATA_VALID2
;
3395 split_flag1
|= EXT4_EXT_MARK_UNWRIT1
;
3396 split_flag1
|= split_flag
& (EXT4_EXT_MAY_ZEROOUT
|
3397 EXT4_EXT_MARK_UNWRIT2
);
3399 err
= ext4_split_extent_at(handle
, inode
, ppath
,
3400 map
->m_lblk
, split_flag1
, flags
);
3405 ext4_ext_show_leaf(inode
, path
);
3407 return err
? err
: allocated
;
3411 * This function is called by ext4_ext_map_blocks() if someone tries to write
3412 * to an unwritten extent. It may result in splitting the unwritten
3413 * extent into multiple extents (up to three - one initialized and two
3415 * There are three possibilities:
3416 * a> There is no split required: Entire extent should be initialized
3417 * b> Splits in two extents: Write is happening at either end of the extent
3418 * c> Splits in three extents: Somone is writing in middle of the extent
3421 * - The extent pointed to by 'path' is unwritten.
3422 * - The extent pointed to by 'path' contains a superset
3423 * of the logical span [map->m_lblk, map->m_lblk + map->m_len).
3425 * Post-conditions on success:
3426 * - the returned value is the number of blocks beyond map->l_lblk
3427 * that are allocated and initialized.
3428 * It is guaranteed to be >= map->m_len.
3430 static int ext4_ext_convert_to_initialized(handle_t
*handle
,
3431 struct inode
*inode
,
3432 struct ext4_map_blocks
*map
,
3433 struct ext4_ext_path
**ppath
,
3436 struct ext4_ext_path
*path
= *ppath
;
3437 struct ext4_sb_info
*sbi
;
3438 struct ext4_extent_header
*eh
;
3439 struct ext4_map_blocks split_map
;
3440 struct ext4_extent zero_ex1
, zero_ex2
;
3441 struct ext4_extent
*ex
, *abut_ex
;
3442 ext4_lblk_t ee_block
, eof_block
;
3443 unsigned int ee_len
, depth
, map_len
= map
->m_len
;
3444 int allocated
= 0, max_zeroout
= 0;
3446 int split_flag
= EXT4_EXT_DATA_VALID2
;
3448 ext_debug("ext4_ext_convert_to_initialized: inode %lu, logical"
3449 "block %llu, max_blocks %u\n", inode
->i_ino
,
3450 (unsigned long long)map
->m_lblk
, map_len
);
3452 sbi
= EXT4_SB(inode
->i_sb
);
3453 eof_block
= (inode
->i_size
+ inode
->i_sb
->s_blocksize
- 1) >>
3454 inode
->i_sb
->s_blocksize_bits
;
3455 if (eof_block
< map
->m_lblk
+ map_len
)
3456 eof_block
= map
->m_lblk
+ map_len
;
3458 depth
= ext_depth(inode
);
3459 eh
= path
[depth
].p_hdr
;
3460 ex
= path
[depth
].p_ext
;
3461 ee_block
= le32_to_cpu(ex
->ee_block
);
3462 ee_len
= ext4_ext_get_actual_len(ex
);
3463 zero_ex1
.ee_len
= 0;
3464 zero_ex2
.ee_len
= 0;
3466 trace_ext4_ext_convert_to_initialized_enter(inode
, map
, ex
);
3468 /* Pre-conditions */
3469 BUG_ON(!ext4_ext_is_unwritten(ex
));
3470 BUG_ON(!in_range(map
->m_lblk
, ee_block
, ee_len
));
3473 * Attempt to transfer newly initialized blocks from the currently
3474 * unwritten extent to its neighbor. This is much cheaper
3475 * than an insertion followed by a merge as those involve costly
3476 * memmove() calls. Transferring to the left is the common case in
3477 * steady state for workloads doing fallocate(FALLOC_FL_KEEP_SIZE)
3478 * followed by append writes.
3480 * Limitations of the current logic:
3481 * - L1: we do not deal with writes covering the whole extent.
3482 * This would require removing the extent if the transfer
3484 * - L2: we only attempt to merge with an extent stored in the
3485 * same extent tree node.
3487 if ((map
->m_lblk
== ee_block
) &&
3488 /* See if we can merge left */
3489 (map_len
< ee_len
) && /*L1*/
3490 (ex
> EXT_FIRST_EXTENT(eh
))) { /*L2*/
3491 ext4_lblk_t prev_lblk
;
3492 ext4_fsblk_t prev_pblk
, ee_pblk
;
3493 unsigned int prev_len
;
3496 prev_lblk
= le32_to_cpu(abut_ex
->ee_block
);
3497 prev_len
= ext4_ext_get_actual_len(abut_ex
);
3498 prev_pblk
= ext4_ext_pblock(abut_ex
);
3499 ee_pblk
= ext4_ext_pblock(ex
);
3502 * A transfer of blocks from 'ex' to 'abut_ex' is allowed
3503 * upon those conditions:
3504 * - C1: abut_ex is initialized,
3505 * - C2: abut_ex is logically abutting ex,
3506 * - C3: abut_ex is physically abutting ex,
3507 * - C4: abut_ex can receive the additional blocks without
3508 * overflowing the (initialized) length limit.
3510 if ((!ext4_ext_is_unwritten(abut_ex
)) && /*C1*/
3511 ((prev_lblk
+ prev_len
) == ee_block
) && /*C2*/
3512 ((prev_pblk
+ prev_len
) == ee_pblk
) && /*C3*/
3513 (prev_len
< (EXT_INIT_MAX_LEN
- map_len
))) { /*C4*/
3514 err
= ext4_ext_get_access(handle
, inode
, path
+ depth
);
3518 trace_ext4_ext_convert_to_initialized_fastpath(inode
,
3521 /* Shift the start of ex by 'map_len' blocks */
3522 ex
->ee_block
= cpu_to_le32(ee_block
+ map_len
);
3523 ext4_ext_store_pblock(ex
, ee_pblk
+ map_len
);
3524 ex
->ee_len
= cpu_to_le16(ee_len
- map_len
);
3525 ext4_ext_mark_unwritten(ex
); /* Restore the flag */
3527 /* Extend abut_ex by 'map_len' blocks */
3528 abut_ex
->ee_len
= cpu_to_le16(prev_len
+ map_len
);
3530 /* Result: number of initialized blocks past m_lblk */
3531 allocated
= map_len
;
3533 } else if (((map
->m_lblk
+ map_len
) == (ee_block
+ ee_len
)) &&
3534 (map_len
< ee_len
) && /*L1*/
3535 ex
< EXT_LAST_EXTENT(eh
)) { /*L2*/
3536 /* See if we can merge right */
3537 ext4_lblk_t next_lblk
;
3538 ext4_fsblk_t next_pblk
, ee_pblk
;
3539 unsigned int next_len
;
3542 next_lblk
= le32_to_cpu(abut_ex
->ee_block
);
3543 next_len
= ext4_ext_get_actual_len(abut_ex
);
3544 next_pblk
= ext4_ext_pblock(abut_ex
);
3545 ee_pblk
= ext4_ext_pblock(ex
);
3548 * A transfer of blocks from 'ex' to 'abut_ex' is allowed
3549 * upon those conditions:
3550 * - C1: abut_ex is initialized,
3551 * - C2: abut_ex is logically abutting ex,
3552 * - C3: abut_ex is physically abutting ex,
3553 * - C4: abut_ex can receive the additional blocks without
3554 * overflowing the (initialized) length limit.
3556 if ((!ext4_ext_is_unwritten(abut_ex
)) && /*C1*/
3557 ((map
->m_lblk
+ map_len
) == next_lblk
) && /*C2*/
3558 ((ee_pblk
+ ee_len
) == next_pblk
) && /*C3*/
3559 (next_len
< (EXT_INIT_MAX_LEN
- map_len
))) { /*C4*/
3560 err
= ext4_ext_get_access(handle
, inode
, path
+ depth
);
3564 trace_ext4_ext_convert_to_initialized_fastpath(inode
,
3567 /* Shift the start of abut_ex by 'map_len' blocks */
3568 abut_ex
->ee_block
= cpu_to_le32(next_lblk
- map_len
);
3569 ext4_ext_store_pblock(abut_ex
, next_pblk
- map_len
);
3570 ex
->ee_len
= cpu_to_le16(ee_len
- map_len
);
3571 ext4_ext_mark_unwritten(ex
); /* Restore the flag */
3573 /* Extend abut_ex by 'map_len' blocks */
3574 abut_ex
->ee_len
= cpu_to_le16(next_len
+ map_len
);
3576 /* Result: number of initialized blocks past m_lblk */
3577 allocated
= map_len
;
3581 /* Mark the block containing both extents as dirty */
3582 ext4_ext_dirty(handle
, inode
, path
+ depth
);
3584 /* Update path to point to the right extent */
3585 path
[depth
].p_ext
= abut_ex
;
3588 allocated
= ee_len
- (map
->m_lblk
- ee_block
);
3590 WARN_ON(map
->m_lblk
< ee_block
);
3592 * It is safe to convert extent to initialized via explicit
3593 * zeroout only if extent is fully inside i_size or new_size.
3595 split_flag
|= ee_block
+ ee_len
<= eof_block
? EXT4_EXT_MAY_ZEROOUT
: 0;
3597 if (EXT4_EXT_MAY_ZEROOUT
& split_flag
)
3598 max_zeroout
= sbi
->s_extent_max_zeroout_kb
>>
3599 (inode
->i_sb
->s_blocksize_bits
- 10);
3601 if (ext4_encrypted_inode(inode
))
3606 * 1. split the extent into three extents.
3607 * 2. split the extent into two extents, zeroout the head of the first
3609 * 3. split the extent into two extents, zeroout the tail of the second
3611 * 4. split the extent into two extents with out zeroout.
3612 * 5. no splitting needed, just possibly zeroout the head and / or the
3613 * tail of the extent.
3615 split_map
.m_lblk
= map
->m_lblk
;
3616 split_map
.m_len
= map
->m_len
;
3618 if (max_zeroout
&& (allocated
> split_map
.m_len
)) {
3619 if (allocated
<= max_zeroout
) {
3622 cpu_to_le32(split_map
.m_lblk
+
3625 cpu_to_le16(allocated
- split_map
.m_len
);
3626 ext4_ext_store_pblock(&zero_ex1
,
3627 ext4_ext_pblock(ex
) + split_map
.m_lblk
+
3628 split_map
.m_len
- ee_block
);
3629 err
= ext4_ext_zeroout(inode
, &zero_ex1
);
3632 split_map
.m_len
= allocated
;
3634 if (split_map
.m_lblk
- ee_block
+ split_map
.m_len
<
3637 if (split_map
.m_lblk
!= ee_block
) {
3638 zero_ex2
.ee_block
= ex
->ee_block
;
3639 zero_ex2
.ee_len
= cpu_to_le16(split_map
.m_lblk
-
3641 ext4_ext_store_pblock(&zero_ex2
,
3642 ext4_ext_pblock(ex
));
3643 err
= ext4_ext_zeroout(inode
, &zero_ex2
);
3648 split_map
.m_len
+= split_map
.m_lblk
- ee_block
;
3649 split_map
.m_lblk
= ee_block
;
3650 allocated
= map
->m_len
;
3654 err
= ext4_split_extent(handle
, inode
, ppath
, &split_map
, split_flag
,
3659 /* If we have gotten a failure, don't zero out status tree */
3661 err
= ext4_zeroout_es(inode
, &zero_ex1
);
3663 err
= ext4_zeroout_es(inode
, &zero_ex2
);
3665 return err
? err
: allocated
;
3669 * This function is called by ext4_ext_map_blocks() from
3670 * ext4_get_blocks_dio_write() when DIO to write
3671 * to an unwritten extent.
3673 * Writing to an unwritten extent may result in splitting the unwritten
3674 * extent into multiple initialized/unwritten extents (up to three)
3675 * There are three possibilities:
3676 * a> There is no split required: Entire extent should be unwritten
3677 * b> Splits in two extents: Write is happening at either end of the extent
3678 * c> Splits in three extents: Somone is writing in middle of the extent
3680 * This works the same way in the case of initialized -> unwritten conversion.
3682 * One of more index blocks maybe needed if the extent tree grow after
3683 * the unwritten extent split. To prevent ENOSPC occur at the IO
3684 * complete, we need to split the unwritten extent before DIO submit
3685 * the IO. The unwritten extent called at this time will be split
3686 * into three unwritten extent(at most). After IO complete, the part
3687 * being filled will be convert to initialized by the end_io callback function
3688 * via ext4_convert_unwritten_extents().
3690 * Returns the size of unwritten extent to be written on success.
3692 static int ext4_split_convert_extents(handle_t
*handle
,
3693 struct inode
*inode
,
3694 struct ext4_map_blocks
*map
,
3695 struct ext4_ext_path
**ppath
,
3698 struct ext4_ext_path
*path
= *ppath
;
3699 ext4_lblk_t eof_block
;
3700 ext4_lblk_t ee_block
;
3701 struct ext4_extent
*ex
;
3702 unsigned int ee_len
;
3703 int split_flag
= 0, depth
;
3705 ext_debug("%s: inode %lu, logical block %llu, max_blocks %u\n",
3706 __func__
, inode
->i_ino
,
3707 (unsigned long long)map
->m_lblk
, map
->m_len
);
3709 eof_block
= (inode
->i_size
+ inode
->i_sb
->s_blocksize
- 1) >>
3710 inode
->i_sb
->s_blocksize_bits
;
3711 if (eof_block
< map
->m_lblk
+ map
->m_len
)
3712 eof_block
= map
->m_lblk
+ map
->m_len
;
3714 * It is safe to convert extent to initialized via explicit
3715 * zeroout only if extent is fully insde i_size or new_size.
3717 depth
= ext_depth(inode
);
3718 ex
= path
[depth
].p_ext
;
3719 ee_block
= le32_to_cpu(ex
->ee_block
);
3720 ee_len
= ext4_ext_get_actual_len(ex
);
3722 /* Convert to unwritten */
3723 if (flags
& EXT4_GET_BLOCKS_CONVERT_UNWRITTEN
) {
3724 split_flag
|= EXT4_EXT_DATA_VALID1
;
3725 /* Convert to initialized */
3726 } else if (flags
& EXT4_GET_BLOCKS_CONVERT
) {
3727 split_flag
|= ee_block
+ ee_len
<= eof_block
?
3728 EXT4_EXT_MAY_ZEROOUT
: 0;
3729 split_flag
|= (EXT4_EXT_MARK_UNWRIT2
| EXT4_EXT_DATA_VALID2
);
3731 flags
|= EXT4_GET_BLOCKS_PRE_IO
;
3732 return ext4_split_extent(handle
, inode
, ppath
, map
, split_flag
, flags
);
3735 static int ext4_convert_unwritten_extents_endio(handle_t
*handle
,
3736 struct inode
*inode
,
3737 struct ext4_map_blocks
*map
,
3738 struct ext4_ext_path
**ppath
)
3740 struct ext4_ext_path
*path
= *ppath
;
3741 struct ext4_extent
*ex
;
3742 ext4_lblk_t ee_block
;
3743 unsigned int ee_len
;
3747 depth
= ext_depth(inode
);
3748 ex
= path
[depth
].p_ext
;
3749 ee_block
= le32_to_cpu(ex
->ee_block
);
3750 ee_len
= ext4_ext_get_actual_len(ex
);
3752 ext_debug("ext4_convert_unwritten_extents_endio: inode %lu, logical"
3753 "block %llu, max_blocks %u\n", inode
->i_ino
,
3754 (unsigned long long)ee_block
, ee_len
);
3756 /* If extent is larger than requested it is a clear sign that we still
3757 * have some extent state machine issues left. So extent_split is still
3759 * TODO: Once all related issues will be fixed this situation should be
3762 if (ee_block
!= map
->m_lblk
|| ee_len
> map
->m_len
) {
3763 #ifdef CONFIG_EXT4_DEBUG
3764 ext4_warning(inode
->i_sb
, "Inode (%ld) finished: extent logical block %llu,"
3765 " len %u; IO logical block %llu, len %u",
3766 inode
->i_ino
, (unsigned long long)ee_block
, ee_len
,
3767 (unsigned long long)map
->m_lblk
, map
->m_len
);
3769 err
= ext4_split_convert_extents(handle
, inode
, map
, ppath
,
3770 EXT4_GET_BLOCKS_CONVERT
);
3773 path
= ext4_find_extent(inode
, map
->m_lblk
, ppath
, 0);
3775 return PTR_ERR(path
);
3776 depth
= ext_depth(inode
);
3777 ex
= path
[depth
].p_ext
;
3780 err
= ext4_ext_get_access(handle
, inode
, path
+ depth
);
3783 /* first mark the extent as initialized */
3784 ext4_ext_mark_initialized(ex
);
3786 /* note: ext4_ext_correct_indexes() isn't needed here because
3787 * borders are not changed
3789 ext4_ext_try_to_merge(handle
, inode
, path
, ex
);
3791 /* Mark modified extent as dirty */
3792 err
= ext4_ext_dirty(handle
, inode
, path
+ path
->p_depth
);
3794 ext4_ext_show_leaf(inode
, path
);
3799 * Handle EOFBLOCKS_FL flag, clearing it if necessary
3801 static int check_eofblocks_fl(handle_t
*handle
, struct inode
*inode
,
3803 struct ext4_ext_path
*path
,
3807 struct ext4_extent_header
*eh
;
3808 struct ext4_extent
*last_ex
;
3810 if (!ext4_test_inode_flag(inode
, EXT4_INODE_EOFBLOCKS
))
3813 depth
= ext_depth(inode
);
3814 eh
= path
[depth
].p_hdr
;
3817 * We're going to remove EOFBLOCKS_FL entirely in future so we
3818 * do not care for this case anymore. Simply remove the flag
3819 * if there are no extents.
3821 if (unlikely(!eh
->eh_entries
))
3823 last_ex
= EXT_LAST_EXTENT(eh
);
3825 * We should clear the EOFBLOCKS_FL flag if we are writing the
3826 * last block in the last extent in the file. We test this by
3827 * first checking to see if the caller to
3828 * ext4_ext_get_blocks() was interested in the last block (or
3829 * a block beyond the last block) in the current extent. If
3830 * this turns out to be false, we can bail out from this
3831 * function immediately.
3833 if (lblk
+ len
< le32_to_cpu(last_ex
->ee_block
) +
3834 ext4_ext_get_actual_len(last_ex
))
3837 * If the caller does appear to be planning to write at or
3838 * beyond the end of the current extent, we then test to see
3839 * if the current extent is the last extent in the file, by
3840 * checking to make sure it was reached via the rightmost node
3841 * at each level of the tree.
3843 for (i
= depth
-1; i
>= 0; i
--)
3844 if (path
[i
].p_idx
!= EXT_LAST_INDEX(path
[i
].p_hdr
))
3847 ext4_clear_inode_flag(inode
, EXT4_INODE_EOFBLOCKS
);
3848 return ext4_mark_inode_dirty(handle
, inode
);
3852 * ext4_find_delalloc_range: find delayed allocated block in the given range.
3854 * Return 1 if there is a delalloc block in the range, otherwise 0.
3856 int ext4_find_delalloc_range(struct inode
*inode
,
3857 ext4_lblk_t lblk_start
,
3858 ext4_lblk_t lblk_end
)
3860 struct extent_status es
;
3862 ext4_es_find_delayed_extent_range(inode
, lblk_start
, lblk_end
, &es
);
3864 return 0; /* there is no delay extent in this tree */
3865 else if (es
.es_lblk
<= lblk_start
&&
3866 lblk_start
< es
.es_lblk
+ es
.es_len
)
3868 else if (lblk_start
<= es
.es_lblk
&& es
.es_lblk
<= lblk_end
)
3874 int ext4_find_delalloc_cluster(struct inode
*inode
, ext4_lblk_t lblk
)
3876 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
3877 ext4_lblk_t lblk_start
, lblk_end
;
3878 lblk_start
= EXT4_LBLK_CMASK(sbi
, lblk
);
3879 lblk_end
= lblk_start
+ sbi
->s_cluster_ratio
- 1;
3881 return ext4_find_delalloc_range(inode
, lblk_start
, lblk_end
);
3885 * Determines how many complete clusters (out of those specified by the 'map')
3886 * are under delalloc and were reserved quota for.
3887 * This function is called when we are writing out the blocks that were
3888 * originally written with their allocation delayed, but then the space was
3889 * allocated using fallocate() before the delayed allocation could be resolved.
3890 * The cases to look for are:
3891 * ('=' indicated delayed allocated blocks
3892 * '-' indicates non-delayed allocated blocks)
3893 * (a) partial clusters towards beginning and/or end outside of allocated range
3894 * are not delalloc'ed.
3896 * |----c---=|====c====|====c====|===-c----|
3897 * |++++++ allocated ++++++|
3898 * ==> 4 complete clusters in above example
3900 * (b) partial cluster (outside of allocated range) towards either end is
3901 * marked for delayed allocation. In this case, we will exclude that
3904 * |----====c========|========c========|
3905 * |++++++ allocated ++++++|
3906 * ==> 1 complete clusters in above example
3909 * |================c================|
3910 * |++++++ allocated ++++++|
3911 * ==> 0 complete clusters in above example
3913 * The ext4_da_update_reserve_space will be called only if we
3914 * determine here that there were some "entire" clusters that span
3915 * this 'allocated' range.
3916 * In the non-bigalloc case, this function will just end up returning num_blks
3917 * without ever calling ext4_find_delalloc_range.
3920 get_reserved_cluster_alloc(struct inode
*inode
, ext4_lblk_t lblk_start
,
3921 unsigned int num_blks
)
3923 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
3924 ext4_lblk_t alloc_cluster_start
, alloc_cluster_end
;
3925 ext4_lblk_t lblk_from
, lblk_to
, c_offset
;
3926 unsigned int allocated_clusters
= 0;
3928 alloc_cluster_start
= EXT4_B2C(sbi
, lblk_start
);
3929 alloc_cluster_end
= EXT4_B2C(sbi
, lblk_start
+ num_blks
- 1);
3931 /* max possible clusters for this allocation */
3932 allocated_clusters
= alloc_cluster_end
- alloc_cluster_start
+ 1;
3934 trace_ext4_get_reserved_cluster_alloc(inode
, lblk_start
, num_blks
);
3936 /* Check towards left side */
3937 c_offset
= EXT4_LBLK_COFF(sbi
, lblk_start
);
3939 lblk_from
= EXT4_LBLK_CMASK(sbi
, lblk_start
);
3940 lblk_to
= lblk_from
+ c_offset
- 1;
3942 if (ext4_find_delalloc_range(inode
, lblk_from
, lblk_to
))
3943 allocated_clusters
--;
3946 /* Now check towards right. */
3947 c_offset
= EXT4_LBLK_COFF(sbi
, lblk_start
+ num_blks
);
3948 if (allocated_clusters
&& c_offset
) {
3949 lblk_from
= lblk_start
+ num_blks
;
3950 lblk_to
= lblk_from
+ (sbi
->s_cluster_ratio
- c_offset
) - 1;
3952 if (ext4_find_delalloc_range(inode
, lblk_from
, lblk_to
))
3953 allocated_clusters
--;
3956 return allocated_clusters
;
3960 convert_initialized_extent(handle_t
*handle
, struct inode
*inode
,
3961 struct ext4_map_blocks
*map
,
3962 struct ext4_ext_path
**ppath
,
3963 unsigned int allocated
)
3965 struct ext4_ext_path
*path
= *ppath
;
3966 struct ext4_extent
*ex
;
3967 ext4_lblk_t ee_block
;
3968 unsigned int ee_len
;
3973 * Make sure that the extent is no bigger than we support with
3976 if (map
->m_len
> EXT_UNWRITTEN_MAX_LEN
)
3977 map
->m_len
= EXT_UNWRITTEN_MAX_LEN
/ 2;
3979 depth
= ext_depth(inode
);
3980 ex
= path
[depth
].p_ext
;
3981 ee_block
= le32_to_cpu(ex
->ee_block
);
3982 ee_len
= ext4_ext_get_actual_len(ex
);
3984 ext_debug("%s: inode %lu, logical"
3985 "block %llu, max_blocks %u\n", __func__
, inode
->i_ino
,
3986 (unsigned long long)ee_block
, ee_len
);
3988 if (ee_block
!= map
->m_lblk
|| ee_len
> map
->m_len
) {
3989 err
= ext4_split_convert_extents(handle
, inode
, map
, ppath
,
3990 EXT4_GET_BLOCKS_CONVERT_UNWRITTEN
);
3993 path
= ext4_find_extent(inode
, map
->m_lblk
, ppath
, 0);
3995 return PTR_ERR(path
);
3996 depth
= ext_depth(inode
);
3997 ex
= path
[depth
].p_ext
;
3999 EXT4_ERROR_INODE(inode
, "unexpected hole at %lu",
4000 (unsigned long) map
->m_lblk
);
4001 return -EFSCORRUPTED
;
4005 err
= ext4_ext_get_access(handle
, inode
, path
+ depth
);
4008 /* first mark the extent as unwritten */
4009 ext4_ext_mark_unwritten(ex
);
4011 /* note: ext4_ext_correct_indexes() isn't needed here because
4012 * borders are not changed
4014 ext4_ext_try_to_merge(handle
, inode
, path
, ex
);
4016 /* Mark modified extent as dirty */
4017 err
= ext4_ext_dirty(handle
, inode
, path
+ path
->p_depth
);
4020 ext4_ext_show_leaf(inode
, path
);
4022 ext4_update_inode_fsync_trans(handle
, inode
, 1);
4023 err
= check_eofblocks_fl(handle
, inode
, map
->m_lblk
, path
, map
->m_len
);
4026 map
->m_flags
|= EXT4_MAP_UNWRITTEN
;
4027 if (allocated
> map
->m_len
)
4028 allocated
= map
->m_len
;
4029 map
->m_len
= allocated
;
4034 ext4_ext_handle_unwritten_extents(handle_t
*handle
, struct inode
*inode
,
4035 struct ext4_map_blocks
*map
,
4036 struct ext4_ext_path
**ppath
, int flags
,
4037 unsigned int allocated
, ext4_fsblk_t newblock
)
4039 struct ext4_ext_path
*path
= *ppath
;
4043 ext_debug("ext4_ext_handle_unwritten_extents: inode %lu, logical "
4044 "block %llu, max_blocks %u, flags %x, allocated %u\n",
4045 inode
->i_ino
, (unsigned long long)map
->m_lblk
, map
->m_len
,
4047 ext4_ext_show_leaf(inode
, path
);
4050 * When writing into unwritten space, we should not fail to
4051 * allocate metadata blocks for the new extent block if needed.
4053 flags
|= EXT4_GET_BLOCKS_METADATA_NOFAIL
;
4055 trace_ext4_ext_handle_unwritten_extents(inode
, map
, flags
,
4056 allocated
, newblock
);
4058 /* get_block() before submit the IO, split the extent */
4059 if (flags
& EXT4_GET_BLOCKS_PRE_IO
) {
4060 ret
= ext4_split_convert_extents(handle
, inode
, map
, ppath
,
4061 flags
| EXT4_GET_BLOCKS_CONVERT
);
4064 map
->m_flags
|= EXT4_MAP_UNWRITTEN
;
4067 /* IO end_io complete, convert the filled extent to written */
4068 if (flags
& EXT4_GET_BLOCKS_CONVERT
) {
4069 if (flags
& EXT4_GET_BLOCKS_ZERO
) {
4070 if (allocated
> map
->m_len
)
4071 allocated
= map
->m_len
;
4072 err
= ext4_issue_zeroout(inode
, map
->m_lblk
, newblock
,
4077 ret
= ext4_convert_unwritten_extents_endio(handle
, inode
, map
,
4080 ext4_update_inode_fsync_trans(handle
, inode
, 1);
4081 err
= check_eofblocks_fl(handle
, inode
, map
->m_lblk
,
4085 map
->m_flags
|= EXT4_MAP_MAPPED
;
4086 map
->m_pblk
= newblock
;
4087 if (allocated
> map
->m_len
)
4088 allocated
= map
->m_len
;
4089 map
->m_len
= allocated
;
4092 /* buffered IO case */
4094 * repeat fallocate creation request
4095 * we already have an unwritten extent
4097 if (flags
& EXT4_GET_BLOCKS_UNWRIT_EXT
) {
4098 map
->m_flags
|= EXT4_MAP_UNWRITTEN
;
4102 /* buffered READ or buffered write_begin() lookup */
4103 if ((flags
& EXT4_GET_BLOCKS_CREATE
) == 0) {
4105 * We have blocks reserved already. We
4106 * return allocated blocks so that delalloc
4107 * won't do block reservation for us. But
4108 * the buffer head will be unmapped so that
4109 * a read from the block returns 0s.
4111 map
->m_flags
|= EXT4_MAP_UNWRITTEN
;
4115 /* buffered write, writepage time, convert*/
4116 ret
= ext4_ext_convert_to_initialized(handle
, inode
, map
, ppath
, flags
);
4118 ext4_update_inode_fsync_trans(handle
, inode
, 1);
4125 map
->m_flags
|= EXT4_MAP_NEW
;
4127 * if we allocated more blocks than requested
4128 * we need to make sure we unmap the extra block
4129 * allocated. The actual needed block will get
4130 * unmapped later when we find the buffer_head marked
4133 if (allocated
> map
->m_len
) {
4134 clean_bdev_aliases(inode
->i_sb
->s_bdev
, newblock
+ map
->m_len
,
4135 allocated
- map
->m_len
);
4136 allocated
= map
->m_len
;
4138 map
->m_len
= allocated
;
4141 * If we have done fallocate with the offset that is already
4142 * delayed allocated, we would have block reservation
4143 * and quota reservation done in the delayed write path.
4144 * But fallocate would have already updated quota and block
4145 * count for this offset. So cancel these reservation
4147 if (flags
& EXT4_GET_BLOCKS_DELALLOC_RESERVE
) {
4148 unsigned int reserved_clusters
;
4149 reserved_clusters
= get_reserved_cluster_alloc(inode
,
4150 map
->m_lblk
, map
->m_len
);
4151 if (reserved_clusters
)
4152 ext4_da_update_reserve_space(inode
,
4158 map
->m_flags
|= EXT4_MAP_MAPPED
;
4159 if ((flags
& EXT4_GET_BLOCKS_KEEP_SIZE
) == 0) {
4160 err
= check_eofblocks_fl(handle
, inode
, map
->m_lblk
, path
,
4166 if (allocated
> map
->m_len
)
4167 allocated
= map
->m_len
;
4168 ext4_ext_show_leaf(inode
, path
);
4169 map
->m_pblk
= newblock
;
4170 map
->m_len
= allocated
;
4172 return err
? err
: allocated
;
4176 * get_implied_cluster_alloc - check to see if the requested
4177 * allocation (in the map structure) overlaps with a cluster already
4178 * allocated in an extent.
4179 * @sb The filesystem superblock structure
4180 * @map The requested lblk->pblk mapping
4181 * @ex The extent structure which might contain an implied
4182 * cluster allocation
4184 * This function is called by ext4_ext_map_blocks() after we failed to
4185 * find blocks that were already in the inode's extent tree. Hence,
4186 * we know that the beginning of the requested region cannot overlap
4187 * the extent from the inode's extent tree. There are three cases we
4188 * want to catch. The first is this case:
4190 * |--- cluster # N--|
4191 * |--- extent ---| |---- requested region ---|
4194 * The second case that we need to test for is this one:
4196 * |--------- cluster # N ----------------|
4197 * |--- requested region --| |------- extent ----|
4198 * |=======================|
4200 * The third case is when the requested region lies between two extents
4201 * within the same cluster:
4202 * |------------- cluster # N-------------|
4203 * |----- ex -----| |---- ex_right ----|
4204 * |------ requested region ------|
4205 * |================|
4207 * In each of the above cases, we need to set the map->m_pblk and
4208 * map->m_len so it corresponds to the return the extent labelled as
4209 * "|====|" from cluster #N, since it is already in use for data in
4210 * cluster EXT4_B2C(sbi, map->m_lblk). We will then return 1 to
4211 * signal to ext4_ext_map_blocks() that map->m_pblk should be treated
4212 * as a new "allocated" block region. Otherwise, we will return 0 and
4213 * ext4_ext_map_blocks() will then allocate one or more new clusters
4214 * by calling ext4_mb_new_blocks().
4216 static int get_implied_cluster_alloc(struct super_block
*sb
,
4217 struct ext4_map_blocks
*map
,
4218 struct ext4_extent
*ex
,
4219 struct ext4_ext_path
*path
)
4221 struct ext4_sb_info
*sbi
= EXT4_SB(sb
);
4222 ext4_lblk_t c_offset
= EXT4_LBLK_COFF(sbi
, map
->m_lblk
);
4223 ext4_lblk_t ex_cluster_start
, ex_cluster_end
;
4224 ext4_lblk_t rr_cluster_start
;
4225 ext4_lblk_t ee_block
= le32_to_cpu(ex
->ee_block
);
4226 ext4_fsblk_t ee_start
= ext4_ext_pblock(ex
);
4227 unsigned short ee_len
= ext4_ext_get_actual_len(ex
);
4229 /* The extent passed in that we are trying to match */
4230 ex_cluster_start
= EXT4_B2C(sbi
, ee_block
);
4231 ex_cluster_end
= EXT4_B2C(sbi
, ee_block
+ ee_len
- 1);
4233 /* The requested region passed into ext4_map_blocks() */
4234 rr_cluster_start
= EXT4_B2C(sbi
, map
->m_lblk
);
4236 if ((rr_cluster_start
== ex_cluster_end
) ||
4237 (rr_cluster_start
== ex_cluster_start
)) {
4238 if (rr_cluster_start
== ex_cluster_end
)
4239 ee_start
+= ee_len
- 1;
4240 map
->m_pblk
= EXT4_PBLK_CMASK(sbi
, ee_start
) + c_offset
;
4241 map
->m_len
= min(map
->m_len
,
4242 (unsigned) sbi
->s_cluster_ratio
- c_offset
);
4244 * Check for and handle this case:
4246 * |--------- cluster # N-------------|
4247 * |------- extent ----|
4248 * |--- requested region ---|
4252 if (map
->m_lblk
< ee_block
)
4253 map
->m_len
= min(map
->m_len
, ee_block
- map
->m_lblk
);
4256 * Check for the case where there is already another allocated
4257 * block to the right of 'ex' but before the end of the cluster.
4259 * |------------- cluster # N-------------|
4260 * |----- ex -----| |---- ex_right ----|
4261 * |------ requested region ------|
4262 * |================|
4264 if (map
->m_lblk
> ee_block
) {
4265 ext4_lblk_t next
= ext4_ext_next_allocated_block(path
);
4266 map
->m_len
= min(map
->m_len
, next
- map
->m_lblk
);
4269 trace_ext4_get_implied_cluster_alloc_exit(sb
, map
, 1);
4273 trace_ext4_get_implied_cluster_alloc_exit(sb
, map
, 0);
4279 * Block allocation/map/preallocation routine for extents based files
4282 * Need to be called with
4283 * down_read(&EXT4_I(inode)->i_data_sem) if not allocating file system block
4284 * (ie, create is zero). Otherwise down_write(&EXT4_I(inode)->i_data_sem)
4286 * return > 0, number of of blocks already mapped/allocated
4287 * if create == 0 and these are pre-allocated blocks
4288 * buffer head is unmapped
4289 * otherwise blocks are mapped
4291 * return = 0, if plain look up failed (blocks have not been allocated)
4292 * buffer head is unmapped
4294 * return < 0, error case.
4296 int ext4_ext_map_blocks(handle_t
*handle
, struct inode
*inode
,
4297 struct ext4_map_blocks
*map
, int flags
)
4299 struct ext4_ext_path
*path
= NULL
;
4300 struct ext4_extent newex
, *ex
, *ex2
;
4301 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
4302 ext4_fsblk_t newblock
= 0;
4303 int free_on_err
= 0, err
= 0, depth
, ret
;
4304 unsigned int allocated
= 0, offset
= 0;
4305 unsigned int allocated_clusters
= 0;
4306 struct ext4_allocation_request ar
;
4307 ext4_lblk_t cluster_offset
;
4308 bool map_from_cluster
= false;
4310 ext_debug("blocks %u/%u requested for inode %lu\n",
4311 map
->m_lblk
, map
->m_len
, inode
->i_ino
);
4312 trace_ext4_ext_map_blocks_enter(inode
, map
->m_lblk
, map
->m_len
, flags
);
4314 /* find extent for this block */
4315 path
= ext4_find_extent(inode
, map
->m_lblk
, NULL
, 0);
4317 err
= PTR_ERR(path
);
4322 depth
= ext_depth(inode
);
4325 * consistent leaf must not be empty;
4326 * this situation is possible, though, _during_ tree modification;
4327 * this is why assert can't be put in ext4_find_extent()
4329 if (unlikely(path
[depth
].p_ext
== NULL
&& depth
!= 0)) {
4330 EXT4_ERROR_INODE(inode
, "bad extent address "
4331 "lblock: %lu, depth: %d pblock %lld",
4332 (unsigned long) map
->m_lblk
, depth
,
4333 path
[depth
].p_block
);
4334 err
= -EFSCORRUPTED
;
4338 ex
= path
[depth
].p_ext
;
4340 ext4_lblk_t ee_block
= le32_to_cpu(ex
->ee_block
);
4341 ext4_fsblk_t ee_start
= ext4_ext_pblock(ex
);
4342 unsigned short ee_len
;
4346 * unwritten extents are treated as holes, except that
4347 * we split out initialized portions during a write.
4349 ee_len
= ext4_ext_get_actual_len(ex
);
4351 trace_ext4_ext_show_extent(inode
, ee_block
, ee_start
, ee_len
);
4353 /* if found extent covers block, simply return it */
4354 if (in_range(map
->m_lblk
, ee_block
, ee_len
)) {
4355 newblock
= map
->m_lblk
- ee_block
+ ee_start
;
4356 /* number of remaining blocks in the extent */
4357 allocated
= ee_len
- (map
->m_lblk
- ee_block
);
4358 ext_debug("%u fit into %u:%d -> %llu\n", map
->m_lblk
,
4359 ee_block
, ee_len
, newblock
);
4362 * If the extent is initialized check whether the
4363 * caller wants to convert it to unwritten.
4365 if ((!ext4_ext_is_unwritten(ex
)) &&
4366 (flags
& EXT4_GET_BLOCKS_CONVERT_UNWRITTEN
)) {
4367 allocated
= convert_initialized_extent(
4368 handle
, inode
, map
, &path
,
4371 } else if (!ext4_ext_is_unwritten(ex
))
4374 ret
= ext4_ext_handle_unwritten_extents(
4375 handle
, inode
, map
, &path
, flags
,
4376 allocated
, newblock
);
4386 * requested block isn't allocated yet;
4387 * we couldn't try to create block if create flag is zero
4389 if ((flags
& EXT4_GET_BLOCKS_CREATE
) == 0) {
4390 ext4_lblk_t hole_start
, hole_len
;
4392 hole_start
= map
->m_lblk
;
4393 hole_len
= ext4_ext_determine_hole(inode
, path
, &hole_start
);
4395 * put just found gap into cache to speed up
4396 * subsequent requests
4398 ext4_ext_put_gap_in_cache(inode
, hole_start
, hole_len
);
4400 /* Update hole_len to reflect hole size after map->m_lblk */
4401 if (hole_start
!= map
->m_lblk
)
4402 hole_len
-= map
->m_lblk
- hole_start
;
4404 map
->m_len
= min_t(unsigned int, map
->m_len
, hole_len
);
4410 * Okay, we need to do block allocation.
4412 newex
.ee_block
= cpu_to_le32(map
->m_lblk
);
4413 cluster_offset
= EXT4_LBLK_COFF(sbi
, map
->m_lblk
);
4416 * If we are doing bigalloc, check to see if the extent returned
4417 * by ext4_find_extent() implies a cluster we can use.
4419 if (cluster_offset
&& ex
&&
4420 get_implied_cluster_alloc(inode
->i_sb
, map
, ex
, path
)) {
4421 ar
.len
= allocated
= map
->m_len
;
4422 newblock
= map
->m_pblk
;
4423 map_from_cluster
= true;
4424 goto got_allocated_blocks
;
4427 /* find neighbour allocated blocks */
4428 ar
.lleft
= map
->m_lblk
;
4429 err
= ext4_ext_search_left(inode
, path
, &ar
.lleft
, &ar
.pleft
);
4432 ar
.lright
= map
->m_lblk
;
4434 err
= ext4_ext_search_right(inode
, path
, &ar
.lright
, &ar
.pright
, &ex2
);
4438 /* Check if the extent after searching to the right implies a
4439 * cluster we can use. */
4440 if ((sbi
->s_cluster_ratio
> 1) && ex2
&&
4441 get_implied_cluster_alloc(inode
->i_sb
, map
, ex2
, path
)) {
4442 ar
.len
= allocated
= map
->m_len
;
4443 newblock
= map
->m_pblk
;
4444 map_from_cluster
= true;
4445 goto got_allocated_blocks
;
4449 * See if request is beyond maximum number of blocks we can have in
4450 * a single extent. For an initialized extent this limit is
4451 * EXT_INIT_MAX_LEN and for an unwritten extent this limit is
4452 * EXT_UNWRITTEN_MAX_LEN.
4454 if (map
->m_len
> EXT_INIT_MAX_LEN
&&
4455 !(flags
& EXT4_GET_BLOCKS_UNWRIT_EXT
))
4456 map
->m_len
= EXT_INIT_MAX_LEN
;
4457 else if (map
->m_len
> EXT_UNWRITTEN_MAX_LEN
&&
4458 (flags
& EXT4_GET_BLOCKS_UNWRIT_EXT
))
4459 map
->m_len
= EXT_UNWRITTEN_MAX_LEN
;
4461 /* Check if we can really insert (m_lblk)::(m_lblk + m_len) extent */
4462 newex
.ee_len
= cpu_to_le16(map
->m_len
);
4463 err
= ext4_ext_check_overlap(sbi
, inode
, &newex
, path
);
4465 allocated
= ext4_ext_get_actual_len(&newex
);
4467 allocated
= map
->m_len
;
4469 /* allocate new block */
4471 ar
.goal
= ext4_ext_find_goal(inode
, path
, map
->m_lblk
);
4472 ar
.logical
= map
->m_lblk
;
4474 * We calculate the offset from the beginning of the cluster
4475 * for the logical block number, since when we allocate a
4476 * physical cluster, the physical block should start at the
4477 * same offset from the beginning of the cluster. This is
4478 * needed so that future calls to get_implied_cluster_alloc()
4481 offset
= EXT4_LBLK_COFF(sbi
, map
->m_lblk
);
4482 ar
.len
= EXT4_NUM_B2C(sbi
, offset
+allocated
);
4484 ar
.logical
-= offset
;
4485 if (S_ISREG(inode
->i_mode
))
4486 ar
.flags
= EXT4_MB_HINT_DATA
;
4488 /* disable in-core preallocation for non-regular files */
4490 if (flags
& EXT4_GET_BLOCKS_NO_NORMALIZE
)
4491 ar
.flags
|= EXT4_MB_HINT_NOPREALLOC
;
4492 if (flags
& EXT4_GET_BLOCKS_DELALLOC_RESERVE
)
4493 ar
.flags
|= EXT4_MB_DELALLOC_RESERVED
;
4494 if (flags
& EXT4_GET_BLOCKS_METADATA_NOFAIL
)
4495 ar
.flags
|= EXT4_MB_USE_RESERVED
;
4496 newblock
= ext4_mb_new_blocks(handle
, &ar
, &err
);
4499 ext_debug("allocate new block: goal %llu, found %llu/%u\n",
4500 ar
.goal
, newblock
, allocated
);
4502 allocated_clusters
= ar
.len
;
4503 ar
.len
= EXT4_C2B(sbi
, ar
.len
) - offset
;
4504 if (ar
.len
> allocated
)
4507 got_allocated_blocks
:
4508 /* try to insert new extent into found leaf and return */
4509 ext4_ext_store_pblock(&newex
, newblock
+ offset
);
4510 newex
.ee_len
= cpu_to_le16(ar
.len
);
4511 /* Mark unwritten */
4512 if (flags
& EXT4_GET_BLOCKS_UNWRIT_EXT
){
4513 ext4_ext_mark_unwritten(&newex
);
4514 map
->m_flags
|= EXT4_MAP_UNWRITTEN
;
4518 if ((flags
& EXT4_GET_BLOCKS_KEEP_SIZE
) == 0)
4519 err
= check_eofblocks_fl(handle
, inode
, map
->m_lblk
,
4522 err
= ext4_ext_insert_extent(handle
, inode
, &path
,
4525 if (err
&& free_on_err
) {
4526 int fb_flags
= flags
& EXT4_GET_BLOCKS_DELALLOC_RESERVE
?
4527 EXT4_FREE_BLOCKS_NO_QUOT_UPDATE
: 0;
4528 /* free data blocks we just allocated */
4529 /* not a good idea to call discard here directly,
4530 * but otherwise we'd need to call it every free() */
4531 ext4_discard_preallocations(inode
);
4532 ext4_free_blocks(handle
, inode
, NULL
, newblock
,
4533 EXT4_C2B(sbi
, allocated_clusters
), fb_flags
);
4537 /* previous routine could use block we allocated */
4538 newblock
= ext4_ext_pblock(&newex
);
4539 allocated
= ext4_ext_get_actual_len(&newex
);
4540 if (allocated
> map
->m_len
)
4541 allocated
= map
->m_len
;
4542 map
->m_flags
|= EXT4_MAP_NEW
;
4545 * Update reserved blocks/metadata blocks after successful
4546 * block allocation which had been deferred till now.
4548 if (flags
& EXT4_GET_BLOCKS_DELALLOC_RESERVE
) {
4549 unsigned int reserved_clusters
;
4551 * Check how many clusters we had reserved this allocated range
4553 reserved_clusters
= get_reserved_cluster_alloc(inode
,
4554 map
->m_lblk
, allocated
);
4555 if (!map_from_cluster
) {
4556 BUG_ON(allocated_clusters
< reserved_clusters
);
4557 if (reserved_clusters
< allocated_clusters
) {
4558 struct ext4_inode_info
*ei
= EXT4_I(inode
);
4559 int reservation
= allocated_clusters
-
4562 * It seems we claimed few clusters outside of
4563 * the range of this allocation. We should give
4564 * it back to the reservation pool. This can
4565 * happen in the following case:
4567 * * Suppose s_cluster_ratio is 4 (i.e., each
4568 * cluster has 4 blocks. Thus, the clusters
4569 * are [0-3],[4-7],[8-11]...
4570 * * First comes delayed allocation write for
4571 * logical blocks 10 & 11. Since there were no
4572 * previous delayed allocated blocks in the
4573 * range [8-11], we would reserve 1 cluster
4575 * * Next comes write for logical blocks 3 to 8.
4576 * In this case, we will reserve 2 clusters
4577 * (for [0-3] and [4-7]; and not for [8-11] as
4578 * that range has a delayed allocated blocks.
4579 * Thus total reserved clusters now becomes 3.
4580 * * Now, during the delayed allocation writeout
4581 * time, we will first write blocks [3-8] and
4582 * allocate 3 clusters for writing these
4583 * blocks. Also, we would claim all these
4584 * three clusters above.
4585 * * Now when we come here to writeout the
4586 * blocks [10-11], we would expect to claim
4587 * the reservation of 1 cluster we had made
4588 * (and we would claim it since there are no
4589 * more delayed allocated blocks in the range
4590 * [8-11]. But our reserved cluster count had
4591 * already gone to 0.
4593 * Thus, at the step 4 above when we determine
4594 * that there are still some unwritten delayed
4595 * allocated blocks outside of our current
4596 * block range, we should increment the
4597 * reserved clusters count so that when the
4598 * remaining blocks finally gets written, we
4601 dquot_reserve_block(inode
,
4602 EXT4_C2B(sbi
, reservation
));
4603 spin_lock(&ei
->i_block_reservation_lock
);
4604 ei
->i_reserved_data_blocks
+= reservation
;
4605 spin_unlock(&ei
->i_block_reservation_lock
);
4608 * We will claim quota for all newly allocated blocks.
4609 * We're updating the reserved space *after* the
4610 * correction above so we do not accidentally free
4611 * all the metadata reservation because we might
4612 * actually need it later on.
4614 ext4_da_update_reserve_space(inode
, allocated_clusters
,
4620 * Cache the extent and update transaction to commit on fdatasync only
4621 * when it is _not_ an unwritten extent.
4623 if ((flags
& EXT4_GET_BLOCKS_UNWRIT_EXT
) == 0)
4624 ext4_update_inode_fsync_trans(handle
, inode
, 1);
4626 ext4_update_inode_fsync_trans(handle
, inode
, 0);
4628 if (allocated
> map
->m_len
)
4629 allocated
= map
->m_len
;
4630 ext4_ext_show_leaf(inode
, path
);
4631 map
->m_flags
|= EXT4_MAP_MAPPED
;
4632 map
->m_pblk
= newblock
;
4633 map
->m_len
= allocated
;
4635 ext4_ext_drop_refs(path
);
4638 trace_ext4_ext_map_blocks_exit(inode
, flags
, map
,
4639 err
? err
: allocated
);
4640 return err
? err
: allocated
;
4643 int ext4_ext_truncate(handle_t
*handle
, struct inode
*inode
)
4645 struct super_block
*sb
= inode
->i_sb
;
4646 ext4_lblk_t last_block
;
4650 * TODO: optimization is possible here.
4651 * Probably we need not scan at all,
4652 * because page truncation is enough.
4655 /* we have to know where to truncate from in crash case */
4656 EXT4_I(inode
)->i_disksize
= inode
->i_size
;
4657 err
= ext4_mark_inode_dirty(handle
, inode
);
4661 last_block
= (inode
->i_size
+ sb
->s_blocksize
- 1)
4662 >> EXT4_BLOCK_SIZE_BITS(sb
);
4664 err
= ext4_es_remove_extent(inode
, last_block
,
4665 EXT_MAX_BLOCKS
- last_block
);
4666 if (err
== -ENOMEM
) {
4668 congestion_wait(BLK_RW_ASYNC
, HZ
/50);
4673 return ext4_ext_remove_space(inode
, last_block
, EXT_MAX_BLOCKS
- 1);
4676 static int ext4_alloc_file_blocks(struct file
*file
, ext4_lblk_t offset
,
4677 ext4_lblk_t len
, loff_t new_size
,
4680 struct inode
*inode
= file_inode(file
);
4686 struct ext4_map_blocks map
;
4687 unsigned int credits
;
4690 BUG_ON(!ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
));
4691 map
.m_lblk
= offset
;
4694 * Don't normalize the request if it can fit in one extent so
4695 * that it doesn't get unnecessarily split into multiple
4698 if (len
<= EXT_UNWRITTEN_MAX_LEN
)
4699 flags
|= EXT4_GET_BLOCKS_NO_NORMALIZE
;
4702 * credits to insert 1 extent into extent tree
4704 credits
= ext4_chunk_trans_blocks(inode
, len
);
4705 depth
= ext_depth(inode
);
4708 while (ret
>= 0 && len
) {
4710 * Recalculate credits when extent tree depth changes.
4712 if (depth
!= ext_depth(inode
)) {
4713 credits
= ext4_chunk_trans_blocks(inode
, len
);
4714 depth
= ext_depth(inode
);
4717 handle
= ext4_journal_start(inode
, EXT4_HT_MAP_BLOCKS
,
4719 if (IS_ERR(handle
)) {
4720 ret
= PTR_ERR(handle
);
4723 ret
= ext4_map_blocks(handle
, inode
, &map
, flags
);
4725 ext4_debug("inode #%lu: block %u: len %u: "
4726 "ext4_ext_map_blocks returned %d",
4727 inode
->i_ino
, map
.m_lblk
,
4729 ext4_mark_inode_dirty(handle
, inode
);
4730 ret2
= ext4_journal_stop(handle
);
4734 map
.m_len
= len
= len
- ret
;
4735 epos
= (loff_t
)map
.m_lblk
<< inode
->i_blkbits
;
4736 inode
->i_ctime
= current_time(inode
);
4738 if (epos
> new_size
)
4740 if (ext4_update_inode_size(inode
, epos
) & 0x1)
4741 inode
->i_mtime
= inode
->i_ctime
;
4743 if (epos
> inode
->i_size
)
4744 ext4_set_inode_flag(inode
,
4745 EXT4_INODE_EOFBLOCKS
);
4747 ext4_mark_inode_dirty(handle
, inode
);
4748 ext4_update_inode_fsync_trans(handle
, inode
, 1);
4749 ret2
= ext4_journal_stop(handle
);
4753 if (ret
== -ENOSPC
&&
4754 ext4_should_retry_alloc(inode
->i_sb
, &retries
)) {
4759 return ret
> 0 ? ret2
: ret
;
4762 static long ext4_zero_range(struct file
*file
, loff_t offset
,
4763 loff_t len
, int mode
)
4765 struct inode
*inode
= file_inode(file
);
4766 handle_t
*handle
= NULL
;
4767 unsigned int max_blocks
;
4768 loff_t new_size
= 0;
4772 int partial_begin
, partial_end
;
4775 unsigned int blkbits
= inode
->i_blkbits
;
4777 trace_ext4_zero_range(inode
, offset
, len
, mode
);
4779 if (!S_ISREG(inode
->i_mode
))
4782 /* Call ext4_force_commit to flush all data in case of data=journal. */
4783 if (ext4_should_journal_data(inode
)) {
4784 ret
= ext4_force_commit(inode
->i_sb
);
4790 * Round up offset. This is not fallocate, we neet to zero out
4791 * blocks, so convert interior block aligned part of the range to
4792 * unwritten and possibly manually zero out unaligned parts of the
4795 start
= round_up(offset
, 1 << blkbits
);
4796 end
= round_down((offset
+ len
), 1 << blkbits
);
4798 if (start
< offset
|| end
> offset
+ len
)
4800 partial_begin
= offset
& ((1 << blkbits
) - 1);
4801 partial_end
= (offset
+ len
) & ((1 << blkbits
) - 1);
4803 lblk
= start
>> blkbits
;
4804 max_blocks
= (end
>> blkbits
);
4805 if (max_blocks
< lblk
)
4813 * Indirect files do not support unwritten extnets
4815 if (!(ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
))) {
4820 if (!(mode
& FALLOC_FL_KEEP_SIZE
) &&
4821 (offset
+ len
> i_size_read(inode
) ||
4822 offset
+ len
> EXT4_I(inode
)->i_disksize
)) {
4823 new_size
= offset
+ len
;
4824 ret
= inode_newsize_ok(inode
, new_size
);
4829 flags
= EXT4_GET_BLOCKS_CREATE_UNWRIT_EXT
;
4830 if (mode
& FALLOC_FL_KEEP_SIZE
)
4831 flags
|= EXT4_GET_BLOCKS_KEEP_SIZE
;
4833 /* Wait all existing dio workers, newcomers will block on i_mutex */
4834 ext4_inode_block_unlocked_dio(inode
);
4835 inode_dio_wait(inode
);
4837 /* Preallocate the range including the unaligned edges */
4838 if (partial_begin
|| partial_end
) {
4839 ret
= ext4_alloc_file_blocks(file
,
4840 round_down(offset
, 1 << blkbits
) >> blkbits
,
4841 (round_up((offset
+ len
), 1 << blkbits
) -
4842 round_down(offset
, 1 << blkbits
)) >> blkbits
,
4849 /* Zero range excluding the unaligned edges */
4850 if (max_blocks
> 0) {
4851 flags
|= (EXT4_GET_BLOCKS_CONVERT_UNWRITTEN
|
4855 * Prevent page faults from reinstantiating pages we have
4856 * released from page cache.
4858 down_write(&EXT4_I(inode
)->i_mmap_sem
);
4859 ret
= ext4_update_disksize_before_punch(inode
, offset
, len
);
4861 up_write(&EXT4_I(inode
)->i_mmap_sem
);
4864 /* Now release the pages and zero block aligned part of pages */
4865 truncate_pagecache_range(inode
, start
, end
- 1);
4866 inode
->i_mtime
= inode
->i_ctime
= current_time(inode
);
4868 ret
= ext4_alloc_file_blocks(file
, lblk
, max_blocks
, new_size
,
4870 up_write(&EXT4_I(inode
)->i_mmap_sem
);
4874 if (!partial_begin
&& !partial_end
)
4878 * In worst case we have to writeout two nonadjacent unwritten
4879 * blocks and update the inode
4881 credits
= (2 * ext4_ext_index_trans_blocks(inode
, 2)) + 1;
4882 if (ext4_should_journal_data(inode
))
4884 handle
= ext4_journal_start(inode
, EXT4_HT_MISC
, credits
);
4885 if (IS_ERR(handle
)) {
4886 ret
= PTR_ERR(handle
);
4887 ext4_std_error(inode
->i_sb
, ret
);
4891 inode
->i_mtime
= inode
->i_ctime
= current_time(inode
);
4893 ext4_update_inode_size(inode
, new_size
);
4896 * Mark that we allocate beyond EOF so the subsequent truncate
4897 * can proceed even if the new size is the same as i_size.
4899 if ((offset
+ len
) > i_size_read(inode
))
4900 ext4_set_inode_flag(inode
, EXT4_INODE_EOFBLOCKS
);
4902 ext4_mark_inode_dirty(handle
, inode
);
4904 /* Zero out partial block at the edges of the range */
4905 ret
= ext4_zero_partial_blocks(handle
, inode
, offset
, len
);
4907 ext4_update_inode_fsync_trans(handle
, inode
, 1);
4909 if (file
->f_flags
& O_SYNC
)
4910 ext4_handle_sync(handle
);
4912 ext4_journal_stop(handle
);
4914 ext4_inode_resume_unlocked_dio(inode
);
4916 inode_unlock(inode
);
4921 * preallocate space for a file. This implements ext4's fallocate file
4922 * operation, which gets called from sys_fallocate system call.
4923 * For block-mapped files, posix_fallocate should fall back to the method
4924 * of writing zeroes to the required new blocks (the same behavior which is
4925 * expected for file systems which do not support fallocate() system call).
4927 long ext4_fallocate(struct file
*file
, int mode
, loff_t offset
, loff_t len
)
4929 struct inode
*inode
= file_inode(file
);
4930 loff_t new_size
= 0;
4931 unsigned int max_blocks
;
4935 unsigned int blkbits
= inode
->i_blkbits
;
4938 * Encrypted inodes can't handle collapse range or insert
4939 * range since we would need to re-encrypt blocks with a
4940 * different IV or XTS tweak (which are based on the logical
4943 * XXX It's not clear why zero range isn't working, but we'll
4944 * leave it disabled for encrypted inodes for now. This is a
4945 * bug we should fix....
4947 if (ext4_encrypted_inode(inode
) &&
4948 (mode
& (FALLOC_FL_COLLAPSE_RANGE
| FALLOC_FL_INSERT_RANGE
|
4949 FALLOC_FL_ZERO_RANGE
)))
4952 /* Return error if mode is not supported */
4953 if (mode
& ~(FALLOC_FL_KEEP_SIZE
| FALLOC_FL_PUNCH_HOLE
|
4954 FALLOC_FL_COLLAPSE_RANGE
| FALLOC_FL_ZERO_RANGE
|
4955 FALLOC_FL_INSERT_RANGE
))
4958 if (mode
& FALLOC_FL_PUNCH_HOLE
)
4959 return ext4_punch_hole(inode
, offset
, len
);
4961 ret
= ext4_convert_inline_data(inode
);
4965 if (mode
& FALLOC_FL_COLLAPSE_RANGE
)
4966 return ext4_collapse_range(inode
, offset
, len
);
4968 if (mode
& FALLOC_FL_INSERT_RANGE
)
4969 return ext4_insert_range(inode
, offset
, len
);
4971 if (mode
& FALLOC_FL_ZERO_RANGE
)
4972 return ext4_zero_range(file
, offset
, len
, mode
);
4974 trace_ext4_fallocate_enter(inode
, offset
, len
, mode
);
4975 lblk
= offset
>> blkbits
;
4977 max_blocks
= EXT4_MAX_BLOCKS(len
, offset
, blkbits
);
4978 flags
= EXT4_GET_BLOCKS_CREATE_UNWRIT_EXT
;
4979 if (mode
& FALLOC_FL_KEEP_SIZE
)
4980 flags
|= EXT4_GET_BLOCKS_KEEP_SIZE
;
4985 * We only support preallocation for extent-based files only
4987 if (!(ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
))) {
4992 if (!(mode
& FALLOC_FL_KEEP_SIZE
) &&
4993 (offset
+ len
> i_size_read(inode
) ||
4994 offset
+ len
> EXT4_I(inode
)->i_disksize
)) {
4995 new_size
= offset
+ len
;
4996 ret
= inode_newsize_ok(inode
, new_size
);
5001 /* Wait all existing dio workers, newcomers will block on i_mutex */
5002 ext4_inode_block_unlocked_dio(inode
);
5003 inode_dio_wait(inode
);
5005 ret
= ext4_alloc_file_blocks(file
, lblk
, max_blocks
, new_size
, flags
);
5006 ext4_inode_resume_unlocked_dio(inode
);
5010 if (file
->f_flags
& O_SYNC
&& EXT4_SB(inode
->i_sb
)->s_journal
) {
5011 ret
= jbd2_complete_transaction(EXT4_SB(inode
->i_sb
)->s_journal
,
5012 EXT4_I(inode
)->i_sync_tid
);
5015 inode_unlock(inode
);
5016 trace_ext4_fallocate_exit(inode
, offset
, max_blocks
, ret
);
5021 * This function convert a range of blocks to written extents
5022 * The caller of this function will pass the start offset and the size.
5023 * all unwritten extents within this range will be converted to
5026 * This function is called from the direct IO end io call back
5027 * function, to convert the fallocated extents after IO is completed.
5028 * Returns 0 on success.
5030 int ext4_convert_unwritten_extents(handle_t
*handle
, struct inode
*inode
,
5031 loff_t offset
, ssize_t len
)
5033 unsigned int max_blocks
;
5036 struct ext4_map_blocks map
;
5037 unsigned int credits
, blkbits
= inode
->i_blkbits
;
5039 map
.m_lblk
= offset
>> blkbits
;
5040 max_blocks
= EXT4_MAX_BLOCKS(len
, offset
, blkbits
);
5043 * This is somewhat ugly but the idea is clear: When transaction is
5044 * reserved, everything goes into it. Otherwise we rather start several
5045 * smaller transactions for conversion of each extent separately.
5048 handle
= ext4_journal_start_reserved(handle
,
5049 EXT4_HT_EXT_CONVERT
);
5051 return PTR_ERR(handle
);
5055 * credits to insert 1 extent into extent tree
5057 credits
= ext4_chunk_trans_blocks(inode
, max_blocks
);
5059 while (ret
>= 0 && ret
< max_blocks
) {
5061 map
.m_len
= (max_blocks
-= ret
);
5063 handle
= ext4_journal_start(inode
, EXT4_HT_MAP_BLOCKS
,
5065 if (IS_ERR(handle
)) {
5066 ret
= PTR_ERR(handle
);
5070 ret
= ext4_map_blocks(handle
, inode
, &map
,
5071 EXT4_GET_BLOCKS_IO_CONVERT_EXT
);
5073 ext4_warning(inode
->i_sb
,
5074 "inode #%lu: block %u: len %u: "
5075 "ext4_ext_map_blocks returned %d",
5076 inode
->i_ino
, map
.m_lblk
,
5078 ext4_mark_inode_dirty(handle
, inode
);
5080 ret2
= ext4_journal_stop(handle
);
5081 if (ret
<= 0 || ret2
)
5085 ret2
= ext4_journal_stop(handle
);
5086 return ret
> 0 ? ret2
: ret
;
5090 * If newes is not existing extent (newes->ec_pblk equals zero) find
5091 * delayed extent at start of newes and update newes accordingly and
5092 * return start of the next delayed extent.
5094 * If newes is existing extent (newes->ec_pblk is not equal zero)
5095 * return start of next delayed extent or EXT_MAX_BLOCKS if no delayed
5096 * extent found. Leave newes unmodified.
5098 static int ext4_find_delayed_extent(struct inode
*inode
,
5099 struct extent_status
*newes
)
5101 struct extent_status es
;
5102 ext4_lblk_t block
, next_del
;
5104 if (newes
->es_pblk
== 0) {
5105 ext4_es_find_delayed_extent_range(inode
, newes
->es_lblk
,
5106 newes
->es_lblk
+ newes
->es_len
- 1, &es
);
5109 * No extent in extent-tree contains block @newes->es_pblk,
5110 * then the block may stay in 1)a hole or 2)delayed-extent.
5116 if (es
.es_lblk
> newes
->es_lblk
) {
5118 newes
->es_len
= min(es
.es_lblk
- newes
->es_lblk
,
5123 newes
->es_len
= es
.es_lblk
+ es
.es_len
- newes
->es_lblk
;
5126 block
= newes
->es_lblk
+ newes
->es_len
;
5127 ext4_es_find_delayed_extent_range(inode
, block
, EXT_MAX_BLOCKS
, &es
);
5129 next_del
= EXT_MAX_BLOCKS
;
5131 next_del
= es
.es_lblk
;
5135 /* fiemap flags we can handle specified here */
5136 #define EXT4_FIEMAP_FLAGS (FIEMAP_FLAG_SYNC|FIEMAP_FLAG_XATTR)
5138 static int ext4_xattr_fiemap(struct inode
*inode
,
5139 struct fiemap_extent_info
*fieinfo
)
5143 __u32 flags
= FIEMAP_EXTENT_LAST
;
5144 int blockbits
= inode
->i_sb
->s_blocksize_bits
;
5148 if (ext4_test_inode_state(inode
, EXT4_STATE_XATTR
)) {
5149 struct ext4_iloc iloc
;
5150 int offset
; /* offset of xattr in inode */
5152 error
= ext4_get_inode_loc(inode
, &iloc
);
5155 physical
= (__u64
)iloc
.bh
->b_blocknr
<< blockbits
;
5156 offset
= EXT4_GOOD_OLD_INODE_SIZE
+
5157 EXT4_I(inode
)->i_extra_isize
;
5159 length
= EXT4_SB(inode
->i_sb
)->s_inode_size
- offset
;
5160 flags
|= FIEMAP_EXTENT_DATA_INLINE
;
5162 } else { /* external block */
5163 physical
= (__u64
)EXT4_I(inode
)->i_file_acl
<< blockbits
;
5164 length
= inode
->i_sb
->s_blocksize
;
5168 error
= fiemap_fill_next_extent(fieinfo
, 0, physical
,
5170 return (error
< 0 ? error
: 0);
5173 int ext4_fiemap(struct inode
*inode
, struct fiemap_extent_info
*fieinfo
,
5174 __u64 start
, __u64 len
)
5176 ext4_lblk_t start_blk
;
5179 if (ext4_has_inline_data(inode
)) {
5182 error
= ext4_inline_data_fiemap(inode
, fieinfo
, &has_inline
,
5189 if (fieinfo
->fi_flags
& FIEMAP_FLAG_CACHE
) {
5190 error
= ext4_ext_precache(inode
);
5195 /* fallback to generic here if not in extents fmt */
5196 if (!(ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
)))
5197 return generic_block_fiemap(inode
, fieinfo
, start
, len
,
5200 if (fiemap_check_flags(fieinfo
, EXT4_FIEMAP_FLAGS
))
5203 if (fieinfo
->fi_flags
& FIEMAP_FLAG_XATTR
) {
5204 error
= ext4_xattr_fiemap(inode
, fieinfo
);
5206 ext4_lblk_t len_blks
;
5209 start_blk
= start
>> inode
->i_sb
->s_blocksize_bits
;
5210 last_blk
= (start
+ len
- 1) >> inode
->i_sb
->s_blocksize_bits
;
5211 if (last_blk
>= EXT_MAX_BLOCKS
)
5212 last_blk
= EXT_MAX_BLOCKS
-1;
5213 len_blks
= ((ext4_lblk_t
) last_blk
) - start_blk
+ 1;
5216 * Walk the extent tree gathering extent information
5217 * and pushing extents back to the user.
5219 error
= ext4_fill_fiemap_extents(inode
, start_blk
,
5227 * Function to access the path buffer for marking it dirty.
5228 * It also checks if there are sufficient credits left in the journal handle
5232 ext4_access_path(handle_t
*handle
, struct inode
*inode
,
5233 struct ext4_ext_path
*path
)
5237 if (!ext4_handle_valid(handle
))
5241 * Check if need to extend journal credits
5242 * 3 for leaf, sb, and inode plus 2 (bmap and group
5243 * descriptor) for each block group; assume two block
5246 if (handle
->h_buffer_credits
< 7) {
5247 credits
= ext4_writepage_trans_blocks(inode
);
5248 err
= ext4_ext_truncate_extend_restart(handle
, inode
, credits
);
5249 /* EAGAIN is success */
5250 if (err
&& err
!= -EAGAIN
)
5254 err
= ext4_ext_get_access(handle
, inode
, path
);
5259 * ext4_ext_shift_path_extents:
5260 * Shift the extents of a path structure lying between path[depth].p_ext
5261 * and EXT_LAST_EXTENT(path[depth].p_hdr), by @shift blocks. @SHIFT tells
5262 * if it is right shift or left shift operation.
5265 ext4_ext_shift_path_extents(struct ext4_ext_path
*path
, ext4_lblk_t shift
,
5266 struct inode
*inode
, handle_t
*handle
,
5267 enum SHIFT_DIRECTION SHIFT
)
5270 struct ext4_extent
*ex_start
, *ex_last
;
5272 depth
= path
->p_depth
;
5274 while (depth
>= 0) {
5275 if (depth
== path
->p_depth
) {
5276 ex_start
= path
[depth
].p_ext
;
5278 return -EFSCORRUPTED
;
5280 ex_last
= EXT_LAST_EXTENT(path
[depth
].p_hdr
);
5282 err
= ext4_access_path(handle
, inode
, path
+ depth
);
5286 if (ex_start
== EXT_FIRST_EXTENT(path
[depth
].p_hdr
))
5289 while (ex_start
<= ex_last
) {
5290 if (SHIFT
== SHIFT_LEFT
) {
5291 le32_add_cpu(&ex_start
->ee_block
,
5293 /* Try to merge to the left. */
5295 EXT_FIRST_EXTENT(path
[depth
].p_hdr
))
5297 ext4_ext_try_to_merge_right(inode
,
5298 path
, ex_start
- 1))
5303 le32_add_cpu(&ex_last
->ee_block
, shift
);
5304 ext4_ext_try_to_merge_right(inode
, path
,
5309 err
= ext4_ext_dirty(handle
, inode
, path
+ depth
);
5313 if (--depth
< 0 || !update
)
5317 /* Update index too */
5318 err
= ext4_access_path(handle
, inode
, path
+ depth
);
5322 if (SHIFT
== SHIFT_LEFT
)
5323 le32_add_cpu(&path
[depth
].p_idx
->ei_block
, -shift
);
5325 le32_add_cpu(&path
[depth
].p_idx
->ei_block
, shift
);
5326 err
= ext4_ext_dirty(handle
, inode
, path
+ depth
);
5330 /* we are done if current index is not a starting index */
5331 if (path
[depth
].p_idx
!= EXT_FIRST_INDEX(path
[depth
].p_hdr
))
5342 * ext4_ext_shift_extents:
5343 * All the extents which lies in the range from @start to the last allocated
5344 * block for the @inode are shifted either towards left or right (depending
5345 * upon @SHIFT) by @shift blocks.
5346 * On success, 0 is returned, error otherwise.
5349 ext4_ext_shift_extents(struct inode
*inode
, handle_t
*handle
,
5350 ext4_lblk_t start
, ext4_lblk_t shift
,
5351 enum SHIFT_DIRECTION SHIFT
)
5353 struct ext4_ext_path
*path
;
5355 struct ext4_extent
*extent
;
5356 ext4_lblk_t stop
, *iterator
, ex_start
, ex_end
;
5358 /* Let path point to the last extent */
5359 path
= ext4_find_extent(inode
, EXT_MAX_BLOCKS
- 1, NULL
,
5362 return PTR_ERR(path
);
5364 depth
= path
->p_depth
;
5365 extent
= path
[depth
].p_ext
;
5369 stop
= le32_to_cpu(extent
->ee_block
);
5372 * For left shifts, make sure the hole on the left is big enough to
5373 * accommodate the shift. For right shifts, make sure the last extent
5374 * won't be shifted beyond EXT_MAX_BLOCKS.
5376 if (SHIFT
== SHIFT_LEFT
) {
5377 path
= ext4_find_extent(inode
, start
- 1, &path
,
5380 return PTR_ERR(path
);
5381 depth
= path
->p_depth
;
5382 extent
= path
[depth
].p_ext
;
5384 ex_start
= le32_to_cpu(extent
->ee_block
);
5385 ex_end
= le32_to_cpu(extent
->ee_block
) +
5386 ext4_ext_get_actual_len(extent
);
5392 if ((start
== ex_start
&& shift
> ex_start
) ||
5393 (shift
> start
- ex_end
)) {
5398 if (shift
> EXT_MAX_BLOCKS
-
5399 (stop
+ ext4_ext_get_actual_len(extent
))) {
5406 * In case of left shift, iterator points to start and it is increased
5407 * till we reach stop. In case of right shift, iterator points to stop
5408 * and it is decreased till we reach start.
5410 if (SHIFT
== SHIFT_LEFT
)
5416 * Its safe to start updating extents. Start and stop are unsigned, so
5417 * in case of right shift if extent with 0 block is reached, iterator
5418 * becomes NULL to indicate the end of the loop.
5420 while (iterator
&& start
<= stop
) {
5421 path
= ext4_find_extent(inode
, *iterator
, &path
,
5424 return PTR_ERR(path
);
5425 depth
= path
->p_depth
;
5426 extent
= path
[depth
].p_ext
;
5428 EXT4_ERROR_INODE(inode
, "unexpected hole at %lu",
5429 (unsigned long) *iterator
);
5430 return -EFSCORRUPTED
;
5432 if (SHIFT
== SHIFT_LEFT
&& *iterator
>
5433 le32_to_cpu(extent
->ee_block
)) {
5434 /* Hole, move to the next extent */
5435 if (extent
< EXT_LAST_EXTENT(path
[depth
].p_hdr
)) {
5436 path
[depth
].p_ext
++;
5438 *iterator
= ext4_ext_next_allocated_block(path
);
5443 if (SHIFT
== SHIFT_LEFT
) {
5444 extent
= EXT_LAST_EXTENT(path
[depth
].p_hdr
);
5445 *iterator
= le32_to_cpu(extent
->ee_block
) +
5446 ext4_ext_get_actual_len(extent
);
5448 extent
= EXT_FIRST_EXTENT(path
[depth
].p_hdr
);
5449 if (le32_to_cpu(extent
->ee_block
) > 0)
5450 *iterator
= le32_to_cpu(extent
->ee_block
) - 1;
5452 /* Beginning is reached, end of the loop */
5454 /* Update path extent in case we need to stop */
5455 while (le32_to_cpu(extent
->ee_block
) < start
)
5457 path
[depth
].p_ext
= extent
;
5459 ret
= ext4_ext_shift_path_extents(path
, shift
, inode
,
5465 ext4_ext_drop_refs(path
);
5471 * ext4_collapse_range:
5472 * This implements the fallocate's collapse range functionality for ext4
5473 * Returns: 0 and non-zero on error.
5475 int ext4_collapse_range(struct inode
*inode
, loff_t offset
, loff_t len
)
5477 struct super_block
*sb
= inode
->i_sb
;
5478 ext4_lblk_t punch_start
, punch_stop
;
5480 unsigned int credits
;
5481 loff_t new_size
, ioffset
;
5485 * We need to test this early because xfstests assumes that a
5486 * collapse range of (0, 1) will return EOPNOTSUPP if the file
5487 * system does not support collapse range.
5489 if (!ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
))
5492 /* Collapse range works only on fs block size aligned offsets. */
5493 if (offset
& (EXT4_CLUSTER_SIZE(sb
) - 1) ||
5494 len
& (EXT4_CLUSTER_SIZE(sb
) - 1))
5497 if (!S_ISREG(inode
->i_mode
))
5500 trace_ext4_collapse_range(inode
, offset
, len
);
5502 punch_start
= offset
>> EXT4_BLOCK_SIZE_BITS(sb
);
5503 punch_stop
= (offset
+ len
) >> EXT4_BLOCK_SIZE_BITS(sb
);
5505 /* Call ext4_force_commit to flush all data in case of data=journal. */
5506 if (ext4_should_journal_data(inode
)) {
5507 ret
= ext4_force_commit(inode
->i_sb
);
5514 * There is no need to overlap collapse range with EOF, in which case
5515 * it is effectively a truncate operation
5517 if (offset
+ len
>= i_size_read(inode
)) {
5522 /* Currently just for extent based files */
5523 if (!ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
)) {
5528 /* Wait for existing dio to complete */
5529 ext4_inode_block_unlocked_dio(inode
);
5530 inode_dio_wait(inode
);
5533 * Prevent page faults from reinstantiating pages we have released from
5536 down_write(&EXT4_I(inode
)->i_mmap_sem
);
5538 * Need to round down offset to be aligned with page size boundary
5539 * for page size > block size.
5541 ioffset
= round_down(offset
, PAGE_SIZE
);
5543 * Write tail of the last page before removed range since it will get
5544 * removed from the page cache below.
5546 ret
= filemap_write_and_wait_range(inode
->i_mapping
, ioffset
, offset
);
5550 * Write data that will be shifted to preserve them when discarding
5551 * page cache below. We are also protected from pages becoming dirty
5554 ret
= filemap_write_and_wait_range(inode
->i_mapping
, offset
+ len
,
5558 truncate_pagecache(inode
, ioffset
);
5560 credits
= ext4_writepage_trans_blocks(inode
);
5561 handle
= ext4_journal_start(inode
, EXT4_HT_TRUNCATE
, credits
);
5562 if (IS_ERR(handle
)) {
5563 ret
= PTR_ERR(handle
);
5567 down_write(&EXT4_I(inode
)->i_data_sem
);
5568 ext4_discard_preallocations(inode
);
5570 ret
= ext4_es_remove_extent(inode
, punch_start
,
5571 EXT_MAX_BLOCKS
- punch_start
);
5573 up_write(&EXT4_I(inode
)->i_data_sem
);
5577 ret
= ext4_ext_remove_space(inode
, punch_start
, punch_stop
- 1);
5579 up_write(&EXT4_I(inode
)->i_data_sem
);
5582 ext4_discard_preallocations(inode
);
5584 ret
= ext4_ext_shift_extents(inode
, handle
, punch_stop
,
5585 punch_stop
- punch_start
, SHIFT_LEFT
);
5587 up_write(&EXT4_I(inode
)->i_data_sem
);
5591 new_size
= i_size_read(inode
) - len
;
5592 i_size_write(inode
, new_size
);
5593 EXT4_I(inode
)->i_disksize
= new_size
;
5595 up_write(&EXT4_I(inode
)->i_data_sem
);
5597 ext4_handle_sync(handle
);
5598 inode
->i_mtime
= inode
->i_ctime
= current_time(inode
);
5599 ext4_mark_inode_dirty(handle
, inode
);
5600 ext4_update_inode_fsync_trans(handle
, inode
, 1);
5603 ext4_journal_stop(handle
);
5605 up_write(&EXT4_I(inode
)->i_mmap_sem
);
5606 ext4_inode_resume_unlocked_dio(inode
);
5608 inode_unlock(inode
);
5613 * ext4_insert_range:
5614 * This function implements the FALLOC_FL_INSERT_RANGE flag of fallocate.
5615 * The data blocks starting from @offset to the EOF are shifted by @len
5616 * towards right to create a hole in the @inode. Inode size is increased
5618 * Returns 0 on success, error otherwise.
5620 int ext4_insert_range(struct inode
*inode
, loff_t offset
, loff_t len
)
5622 struct super_block
*sb
= inode
->i_sb
;
5624 struct ext4_ext_path
*path
;
5625 struct ext4_extent
*extent
;
5626 ext4_lblk_t offset_lblk
, len_lblk
, ee_start_lblk
= 0;
5627 unsigned int credits
, ee_len
;
5628 int ret
= 0, depth
, split_flag
= 0;
5632 * We need to test this early because xfstests assumes that an
5633 * insert range of (0, 1) will return EOPNOTSUPP if the file
5634 * system does not support insert range.
5636 if (!ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
))
5639 /* Insert range works only on fs block size aligned offsets. */
5640 if (offset
& (EXT4_CLUSTER_SIZE(sb
) - 1) ||
5641 len
& (EXT4_CLUSTER_SIZE(sb
) - 1))
5644 if (!S_ISREG(inode
->i_mode
))
5647 trace_ext4_insert_range(inode
, offset
, len
);
5649 offset_lblk
= offset
>> EXT4_BLOCK_SIZE_BITS(sb
);
5650 len_lblk
= len
>> EXT4_BLOCK_SIZE_BITS(sb
);
5652 /* Call ext4_force_commit to flush all data in case of data=journal */
5653 if (ext4_should_journal_data(inode
)) {
5654 ret
= ext4_force_commit(inode
->i_sb
);
5660 /* Currently just for extent based files */
5661 if (!ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
)) {
5666 /* Check for wrap through zero */
5667 if (inode
->i_size
+ len
> inode
->i_sb
->s_maxbytes
) {
5672 /* Offset should be less than i_size */
5673 if (offset
>= i_size_read(inode
)) {
5678 /* Wait for existing dio to complete */
5679 ext4_inode_block_unlocked_dio(inode
);
5680 inode_dio_wait(inode
);
5683 * Prevent page faults from reinstantiating pages we have released from
5686 down_write(&EXT4_I(inode
)->i_mmap_sem
);
5688 * Need to round down to align start offset to page size boundary
5689 * for page size > block size.
5691 ioffset
= round_down(offset
, PAGE_SIZE
);
5692 /* Write out all dirty pages */
5693 ret
= filemap_write_and_wait_range(inode
->i_mapping
, ioffset
,
5697 truncate_pagecache(inode
, ioffset
);
5699 credits
= ext4_writepage_trans_blocks(inode
);
5700 handle
= ext4_journal_start(inode
, EXT4_HT_TRUNCATE
, credits
);
5701 if (IS_ERR(handle
)) {
5702 ret
= PTR_ERR(handle
);
5706 /* Expand file to avoid data loss if there is error while shifting */
5707 inode
->i_size
+= len
;
5708 EXT4_I(inode
)->i_disksize
+= len
;
5709 inode
->i_mtime
= inode
->i_ctime
= current_time(inode
);
5710 ret
= ext4_mark_inode_dirty(handle
, inode
);
5714 down_write(&EXT4_I(inode
)->i_data_sem
);
5715 ext4_discard_preallocations(inode
);
5717 path
= ext4_find_extent(inode
, offset_lblk
, NULL
, 0);
5719 up_write(&EXT4_I(inode
)->i_data_sem
);
5723 depth
= ext_depth(inode
);
5724 extent
= path
[depth
].p_ext
;
5726 ee_start_lblk
= le32_to_cpu(extent
->ee_block
);
5727 ee_len
= ext4_ext_get_actual_len(extent
);
5730 * If offset_lblk is not the starting block of extent, split
5731 * the extent @offset_lblk
5733 if ((offset_lblk
> ee_start_lblk
) &&
5734 (offset_lblk
< (ee_start_lblk
+ ee_len
))) {
5735 if (ext4_ext_is_unwritten(extent
))
5736 split_flag
= EXT4_EXT_MARK_UNWRIT1
|
5737 EXT4_EXT_MARK_UNWRIT2
;
5738 ret
= ext4_split_extent_at(handle
, inode
, &path
,
5739 offset_lblk
, split_flag
,
5741 EXT4_GET_BLOCKS_PRE_IO
|
5742 EXT4_GET_BLOCKS_METADATA_NOFAIL
);
5745 ext4_ext_drop_refs(path
);
5748 up_write(&EXT4_I(inode
)->i_data_sem
);
5752 ext4_ext_drop_refs(path
);
5756 ret
= ext4_es_remove_extent(inode
, offset_lblk
,
5757 EXT_MAX_BLOCKS
- offset_lblk
);
5759 up_write(&EXT4_I(inode
)->i_data_sem
);
5764 * if offset_lblk lies in a hole which is at start of file, use
5765 * ee_start_lblk to shift extents
5767 ret
= ext4_ext_shift_extents(inode
, handle
,
5768 ee_start_lblk
> offset_lblk
? ee_start_lblk
: offset_lblk
,
5769 len_lblk
, SHIFT_RIGHT
);
5771 up_write(&EXT4_I(inode
)->i_data_sem
);
5773 ext4_handle_sync(handle
);
5775 ext4_update_inode_fsync_trans(handle
, inode
, 1);
5778 ext4_journal_stop(handle
);
5780 up_write(&EXT4_I(inode
)->i_mmap_sem
);
5781 ext4_inode_resume_unlocked_dio(inode
);
5783 inode_unlock(inode
);
5788 * ext4_swap_extents - Swap extents between two inodes
5790 * @inode1: First inode
5791 * @inode2: Second inode
5792 * @lblk1: Start block for first inode
5793 * @lblk2: Start block for second inode
5794 * @count: Number of blocks to swap
5795 * @mark_unwritten: Mark second inode's extents as unwritten after swap
5796 * @erp: Pointer to save error value
5798 * This helper routine does exactly what is promise "swap extents". All other
5799 * stuff such as page-cache locking consistency, bh mapping consistency or
5800 * extent's data copying must be performed by caller.
5802 * i_mutex is held for both inodes
5803 * i_data_sem is locked for write for both inodes
5805 * All pages from requested range are locked for both inodes
5808 ext4_swap_extents(handle_t
*handle
, struct inode
*inode1
,
5809 struct inode
*inode2
, ext4_lblk_t lblk1
, ext4_lblk_t lblk2
,
5810 ext4_lblk_t count
, int unwritten
, int *erp
)
5812 struct ext4_ext_path
*path1
= NULL
;
5813 struct ext4_ext_path
*path2
= NULL
;
5814 int replaced_count
= 0;
5816 BUG_ON(!rwsem_is_locked(&EXT4_I(inode1
)->i_data_sem
));
5817 BUG_ON(!rwsem_is_locked(&EXT4_I(inode2
)->i_data_sem
));
5818 BUG_ON(!inode_is_locked(inode1
));
5819 BUG_ON(!inode_is_locked(inode2
));
5821 *erp
= ext4_es_remove_extent(inode1
, lblk1
, count
);
5824 *erp
= ext4_es_remove_extent(inode2
, lblk2
, count
);
5829 struct ext4_extent
*ex1
, *ex2
, tmp_ex
;
5830 ext4_lblk_t e1_blk
, e2_blk
;
5831 int e1_len
, e2_len
, len
;
5834 path1
= ext4_find_extent(inode1
, lblk1
, NULL
, EXT4_EX_NOCACHE
);
5835 if (IS_ERR(path1
)) {
5836 *erp
= PTR_ERR(path1
);
5842 path2
= ext4_find_extent(inode2
, lblk2
, NULL
, EXT4_EX_NOCACHE
);
5843 if (IS_ERR(path2
)) {
5844 *erp
= PTR_ERR(path2
);
5848 ex1
= path1
[path1
->p_depth
].p_ext
;
5849 ex2
= path2
[path2
->p_depth
].p_ext
;
5850 /* Do we have somthing to swap ? */
5851 if (unlikely(!ex2
|| !ex1
))
5854 e1_blk
= le32_to_cpu(ex1
->ee_block
);
5855 e2_blk
= le32_to_cpu(ex2
->ee_block
);
5856 e1_len
= ext4_ext_get_actual_len(ex1
);
5857 e2_len
= ext4_ext_get_actual_len(ex2
);
5860 if (!in_range(lblk1
, e1_blk
, e1_len
) ||
5861 !in_range(lblk2
, e2_blk
, e2_len
)) {
5862 ext4_lblk_t next1
, next2
;
5864 /* if hole after extent, then go to next extent */
5865 next1
= ext4_ext_next_allocated_block(path1
);
5866 next2
= ext4_ext_next_allocated_block(path2
);
5867 /* If hole before extent, then shift to that extent */
5872 /* Do we have something to swap */
5873 if (next1
== EXT_MAX_BLOCKS
|| next2
== EXT_MAX_BLOCKS
)
5875 /* Move to the rightest boundary */
5876 len
= next1
- lblk1
;
5877 if (len
< next2
- lblk2
)
5878 len
= next2
- lblk2
;
5887 /* Prepare left boundary */
5888 if (e1_blk
< lblk1
) {
5890 *erp
= ext4_force_split_extent_at(handle
, inode1
,
5895 if (e2_blk
< lblk2
) {
5897 *erp
= ext4_force_split_extent_at(handle
, inode2
,
5902 /* ext4_split_extent_at() may result in leaf extent split,
5903 * path must to be revalidated. */
5907 /* Prepare right boundary */
5909 if (len
> e1_blk
+ e1_len
- lblk1
)
5910 len
= e1_blk
+ e1_len
- lblk1
;
5911 if (len
> e2_blk
+ e2_len
- lblk2
)
5912 len
= e2_blk
+ e2_len
- lblk2
;
5914 if (len
!= e1_len
) {
5916 *erp
= ext4_force_split_extent_at(handle
, inode1
,
5917 &path1
, lblk1
+ len
, 0);
5921 if (len
!= e2_len
) {
5923 *erp
= ext4_force_split_extent_at(handle
, inode2
,
5924 &path2
, lblk2
+ len
, 0);
5928 /* ext4_split_extent_at() may result in leaf extent split,
5929 * path must to be revalidated. */
5933 BUG_ON(e2_len
!= e1_len
);
5934 *erp
= ext4_ext_get_access(handle
, inode1
, path1
+ path1
->p_depth
);
5937 *erp
= ext4_ext_get_access(handle
, inode2
, path2
+ path2
->p_depth
);
5941 /* Both extents are fully inside boundaries. Swap it now */
5943 ext4_ext_store_pblock(ex1
, ext4_ext_pblock(ex2
));
5944 ext4_ext_store_pblock(ex2
, ext4_ext_pblock(&tmp_ex
));
5945 ex1
->ee_len
= cpu_to_le16(e2_len
);
5946 ex2
->ee_len
= cpu_to_le16(e1_len
);
5948 ext4_ext_mark_unwritten(ex2
);
5949 if (ext4_ext_is_unwritten(&tmp_ex
))
5950 ext4_ext_mark_unwritten(ex1
);
5952 ext4_ext_try_to_merge(handle
, inode2
, path2
, ex2
);
5953 ext4_ext_try_to_merge(handle
, inode1
, path1
, ex1
);
5954 *erp
= ext4_ext_dirty(handle
, inode2
, path2
+
5958 *erp
= ext4_ext_dirty(handle
, inode1
, path1
+
5961 * Looks scarry ah..? second inode already points to new blocks,
5962 * and it was successfully dirtied. But luckily error may happen
5963 * only due to journal error, so full transaction will be
5970 replaced_count
+= len
;
5974 ext4_ext_drop_refs(path1
);
5976 ext4_ext_drop_refs(path2
);
5978 path1
= path2
= NULL
;
5980 return replaced_count
;